Being, Freedom, and Method

Being, Freedom,
and Method
Themes from the Philosophy of
Peter van Inwagen

 
John A. Keller

1
3
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© the several contributors 2017
The moral rights of the authors have been asserted
First Edition published in 2017
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2016940584
ISBN 978–0–19–871570–2
Printed in Great Britain by
Clays Ltd, St Ives plc
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
Sometimes [van Inwagen] is philosophically conservative, taking up positions
that have fallen out of fashion and giving them a new life by his clever and
original defense of them; his work on free will is a good example of this.
Sometimes, especially in his work opposing orthodox views about the compos-
ition of things out of parts, he takes up radically new positions, and again does a
very able job of arguing them. Either way, his aims are ambitious; and he makes
his case with the utmost precision, lucidity, and fairness. His work is a model of
how systematic metaphysics ought to be pursued. Seldom do I agree with his
conclusions; but whenever he upholds or attacks a position, he’s the one to beat.
I generally come away convinced that there’s absolutely no way I can grant his
premises and dodge his conclusions. He’s unsurpassed both for originality on
important matters and for tightness of argument.
David Lewis
April 7, 1994
Preface

Many thanks are due to Peter Momtchiloff and the contributors to this volume for
doing their jobs excellently and for patience when I faltered at mine. Very special
thanks, for oh so very many things, are due to Peter van Inwagen—my debt to
you will never be repaid. Finally, I thankfully acknowledge the assistance of Hannah
Zukoff, throughout the project, with formatting and other tedious nonsense.
Contents

Contributors xi

Introduction 1
John A. Keller

Part I. Being
1. Theories of Character 11
Michael J. Loux
2. A One Category Ontology 32
L. A. Paul
3. Properties as Parts of Ordinary Objects 62
Eric T. Olson
4. Time Travel and the Movable Present 80
Sara Bernstein

Part II. Freedom

5. The Disconnect Problem and the Influence Strategy 95
Mark Heller
6. Revisiting the Mind Argument 118
Alicia Finch
7.1. Symposium on the Fixity of the Past: Incompatibilism and
the Fixity of the Past 140
Neal A. Tognazzini and John Martin Fischer
7.2. Symposium on the Fixity of the Past: Freedom and Modality 149
Wesley H. Holliday

Part III. God

8. Defenseless: A Critique of Van Inwagen’s Response to the
Argument from Evil 159
Louise Antony
9. The Problem of Evil and Atonement 186
Eleonore Stump
10. Swing Vote 209
Frances Howard-Snyder
11. Theism and Allism 227
Alex Rosenberg
 x 

12. The Evolutionary Argument for Atheism 241

Daniel Howard-Snyder
13. Must Anselm be Interpreted as a Meinongian? 263
Lynne Rudder Baker

Part IV. Method

14. Why Isn’t There More Progress in Philosophy? 277
David J. Chalmers
15. Philosophical Individualism 299
John A. Keller
16. Are There Any Successful Philosophical Arguments? 324
Thomas Kelly and Sarah McGrath

Part V. Afterword
17. Concluding Meditation 343
Peter van Inwagen

The Complete Works of Peter van Inwagen (in English) 395

Index 401
Contributors

L OUISE A NTONY , University of Massachusetts, Amherst

L YNNE R UDDER B AKER , University of Massachusetts, Amherst
S ARA B ERNSTEIN , University of Notre Dame
D AVID J. C HALMERS , New York University
A LICIA F INCH , Northern Illinois University
J OHN M ARTIN F ISCHER , University of California, Riverside
M ARK H ELLER , Syracuse University
W ESLEY H. H OLLIDAY , University of California, Berkeley
D ANIEL H OWARD -S NYDER , Western Washington University
F RANCES H OWARD -S NYDER , Western Washington University
J OHN A. K ELLER , Niagara University
T HOMAS K ELLY , Princeton University
M ICHAEL J. L OUX , University of Notre Dame
S ARAH M C G RATH , Princeton University
E RIC T. O LSON , University of Sheffield
L. A. P AUL , University of North Carolina
A LEX R OSENBERG , Duke University
E LEONORE S TUMP , St Louis University
N EAL A. T OGNAZZINI , Western Washington University
P ETER VAN I NWAGEN , University of Notre Dame
 Introduction
John A. Keller

Reflective persons unswayed by wishful thinking can themselves now and again
have cause to wonder what, if anything, they are talking about.
Quine (1960: 242)1

Famously, or perhaps infamously, Peter van Inwagen doesn’t understand some

things. Of course, there are things that few or none of us understand. But van
Inwagen doesn’t understand some things that the rest of us seem to. For that reason,
many people find van Inwagen’s incomprehension—or at least his professed
incomprehension—to be off-putting, offensive, or insincere.
Whether people find it off-putting, offensive, or insincere to fail to understand
something seems to have very much to do with what that thing is. No one, or hardly
anyone, gets upset if, like van Inwagen, you say that certain unfashionable things like
agent causation, bare particulars, or hylomorphism make no sense. But if, also like
van Inwagen, you say that temporal parts, substitutional quantification, or tropes
make no sense, you are likely to be accused of deliberately not understanding things,
or of lying about what you understand, or worse. But why should this be? Most of us
think that many locutions that were once widespread in metaphysical theorizing are
ultimately incoherent. But then how can we be sure that locutions that are currently
widespread in metaphysical theorizing are not ultimately incoherent? How indeed?
We must admit, if we are being honest, that some of the locutions currently
employed by metaphysicians will turn out to be just as incoherent as some of the
locutions employed in the past. But then none of us understand them, at least if the
following principle is true:
The Nonsense Principle: If something doesn’t make sense, then no one under-
stands it.
Of course, as Annette Baier (1972) notes, there are many different things we call
“nonsense,” including mere blatant falsehoods.2 ‘I have 2.3 children’ is a kind of
nonsense—what we might call silly nonsense—but what it means is easily under-
stood, and understood to be false: that’s the problem with it. When van Inwagen calls

1
Reproduced with permission from Willard van Orman Quine, Word and Object, and The MIT Press.
2
See Diamond (1981) for discussion. Van Inwagen makes a similar observation—see note 6.
  . 

something nonsense, however, he typically means that it fails to have a meaning (at
least in context), and this is the kind of nonsense governed by The Nonsense
Principle. Call such nonsense semantic nonsense. One way for semantic nonsense
to arise is for a term to be introduced via a “failed introduction”: e.g., “let ‘Vulcan’
refer to the planet perturbing Mercury’s orbit.” Such terms simply don’t have
meanings, even if speakers think they do. But semantic nonsense also arises when a
term has been given more than one meaning (“multiple introductions”), and there is
nothing in the context (including speaker intentions) to differentiate between them.
In such contexts, these terms have, at best, “indeterminate meanings.” Herman
Cappelen (2013) argues that many philosophers’ uses of the word ‘intuition’ are
semantic nonsense, and that many of us speak and believe semantic nonsense.
Speaking and believing semantic nonsense has nothing to do with lack of education
or intelligence—philosophers lack neither. The problem is simply that we don’t know
we are speaking and believing semantic nonsense: both unnoticed failed introduc-
tions and unnoticed multiple introductions can give rise to what Gareth Evans (1982)
calls “illusions of thought.”3
If there is semantic nonsense, there are things that no one understands, and if there
are things no one understands, surely some metaphysical terms of art are among
them. But if none of us understand these expressions, why shouldn’t van Inwagen—
someone with, shall we say, rather stringent expectations when it comes to clarity—be
among the first to notice?4 And while no one likes to be found to have been talking
nonsense, we really shouldn’t be surprised or offended if we are. As van Inwagen says:
I do not understand what philosophers say . . . when they utter [certain sentences], and I think
the reason I do not understand them is that they have failed to explain what they mean . . . And
I think the reason they have failed to explain what they mean is that there is nothing, or
nothing coherent, that they do mean . . .
It is not my intention to insult anyone. It is very hard to make any sense at all when talking
about philosophical matters, and it is not necessarily a disgrace for a philosopher to lapse into
nonsense. It is not, or should not be, any more insulting to say to a philosopher, ‘What you say
is nonsense’, than it is to say, ‘What you say is false’. Very great philosophers, philosophers
whose shoes I am not fit to untie, have talked mostly nonsense . . . [N]onsense is as variable
in its quality as falsity . . . Descartes’ philosophy and what I wrote in my undergraduate

3
See Cappelen (2013), and also Recanati (1997), for further discussion. Of course, if there can be
“illusions of thought,” there can presumably also be “illusions of nonsense”—cases where something that
makes sense seems to be nonsense (e.g., “wave particle duality” and “there are as many even numbers as
numbers”).
4
In addition to his preference for strict “Chisholm-style” definitions, van Inwagen’s Wittgensteinian
streak is relevant, as is his skepticism about our ability to know interesting or substantive modal facts (see
van Inwagen (2014, 1988, and 1998b), respectively, for discussion). Many philosophers are drawn to
“desert landscapes”: austere ontologies that, e.g., reduce everything to physics and eliminate what cannot be
so reduced. Van Inwagen is drawn to austere methods that reduce the epistemology of philosophy to the
epistemology of non-philosophy (science, ordinary inquiry, etc.) and eliminate philosophical methods that
cannot be so reduced. Hence his skepticism about metaphysical explanation (see, e.g., his 2011) and his
requirement that all terms be ultimately definable using only antecedently understood bits of ordinary
language (see, e.g., his 2008).
  

philosophy papers are equally nonsense. But . . . [w]ould that I were capable of such nonsense
[as Descartes]! (Van Inwagen 1980: 285)

If the Nonsense principle is true, it is likely that there are currently fashionable
idioms that none of us understand. But is the Nonsense Principle true? It is at least
superficially plausible: ‘understand’ is a success term, and so while many people think
they understand things that they in fact do not understand, no one understands
something that cannot be understood. And isn’t it true that if something makes no
sense, then no one can understand it? What would it be to understand something that
makes no sense? ‘p makes no sense, but she understands it’ has a paradoxical air
about it, does it not? If p is semantic nonsense, there is just nothing there to
understand (in the relevant sense).5
If the Nonsense Principle is true, it follows that if, as some philosophers believe,
agent causation makes no sense, then no one understands agent causation. It also
follows that if, as some philosophers believe, the notion of a bare particular makes no
sense, no one understands the notion of a bare particular. Similarly, it follows that
if the idea of form—something that is supposed to be neither concrete nor
abstract—doesn’t make sense, then no one, not even Aristotle himself, understands
hylomorphic theories that rely on that notion. And of course it follows that if
substitutional quantification, temporal parts, tropes, and so on make no sense, then
no one understands them, even the very smart people who have devoted their lives to
their defense.6

5
There are many different senses of ‘understanding’, and one might worry that if the primary objects of
our understanding (in the relevant sense) are propositions, but the things that make sense (or not) are
sentences, the Nonsense Principle itself doesn’t make sense. (In fact, Neal Tognazzini did express this
worry to me.) But we can (at least derivatively) be said to understand a sentence (or term) s if we
understand the proposition (or concept) p expressed by s and know that p is expressed by s. Call this
sense of understanding ‘understandings’. The Nonsense Principle could then be expressed as follows:
The Nonsense Principles: If something is semantic nonsense (in context c), then no one understandss it.
Another worry about the Nonsense Principle is that understanding comes in degrees—to understands a
sentence one needn’t fully grasp the proposition it expresses, or to be able to narrow down the different
potential meanings to one. Rather, partially grasping the proposition it expresses, or being able to narrow
down the range to a few propositions, might count as partially understandings the sentence. This won’t
help with sentences that express too few (i.e., zero) propositions (due to failed introductions), but it may
help with sentences that can be used to express too many propositions (due to multiple introductions). If
one sees that a sentence s may be used to express (only) propositions p and q in context c, then it seems fair
to say that one has some understandings of the sentence (in c). On the other hand, if the speaker insists that
she doesn’t mean p or q, then I think that it is fair to say that one doesn’t understands her use of that
sentence (in c). In some such cases, it is plausible that s doesn’t express anything at all. Cf. van Inwagen
(1981) on substitutional quantification.
6
Of course, we can often transform semantic nonsense into silly nonsense via explicit definition, but
that doesn’t necessarily yield understanding. As van Inwagen says about David Lewis’s definition of
‘temporal part’:
Since I understand [the definition], I understand “Lewis-part” and I know what Lewis-parts
are. In a way. In the same way as the way in which I should understand talk of “propertyless
objects” if I were told that “propertyless object” meant “object of which nothing is true”; in
the same way as that in which I should understand talk of “two-dimensional cups” if I were
told that “two-dimensional cup” meant “cup that lies entirely in a plane.” These phrases
would not be . . . “semantical nonsense” . . . ; they would not be like “abracadabra” or . . . “Das
Nichts nichtet.” But I should hardly care to say that I understood what someone was talking
  . 

And so, I ask you again: given that a great many previously popular metaphysical
locutions make no sense, what are the chances that all currently popular metaphysical
locutions make sense? I think there is only one possible honest and self-aware answer
to this question.
But enough. It is not my purpose here to defend van Inwagen’s incomprehension.7
Rather, I want to point out the merits of his incomprehension, whatever its demerits.
Technical terms abound in philosophy, and they are not always introduced via
explicit definition. Technical terms without explicit definitions might be semantic
nonsense, but even when they aren’t, providing such definitions is a source of clarity,
and often results in unexpected benefits: a term turns out to be harder to define than
one expects, and different ways of defining it carry surprising philosophical com-
mitments (or render phrases making use of the term silly nonsense). Van Inwagen’s
insistence that we carefully define our terms does us a service, even if we don’t enjoy
it. We should be grateful for this treatment, and not judge van Inwagen “the way a
doctor would be judged by a jury of children if a pastry chef were to bring accusations
against him” (Gorgias: 521e2–4).8
But as valuable as clarity is, I think van Inwagen’s incomprehension has another
and perhaps more important benefit: it has forced him to approach philosophical
topics from new perspectives, thereby creating new questions and new debates
about how to answer them. Only six philosophers appear more than once on Kieran
Healy’s list of the “top fifty” most influential publications in the second half of the
twentieth century.9 They are David Lewis, Timothy Williamson, Donald Davidson,
Jerry Fodor, Hillary Putnam, and Peter van Inwagen.10 The list speaks for itself, but it
is worth calling attention to what it tells us about the benefits of rejecting fashionable
ways of framing philosophical problems and their solution sets. Over the course
of his career, van Inwagen’s incomprehension has led to immensely fruitful
questioning—questioning that opened up new and productive avenues of inquiry.
Failing to understand the standard options is a powerful impetus for discovering new
ones, and the landscape in the philosophy of action, metaphysics, meta-philosophy,
and philosophy of religion has been profoundly altered by van Inwagen’s work over

about . . . who talked of propertyless objects or two-dimensional cups, and who, moreover,
talked of them in a way that suggested that he supposed there were such things. For I cannot
see how there could be any such things. In fact, I think I can see clearly and distinctly why
there could not possibly be any propertyless objects or two-dimensional cups (so defined).
(Van Inwagen 2000)
7
I should, however, note that there are at least two cases where I thought I understood something—
where I thought I believed something—only to come to realize that what I “believed” made no sense. These
realizations came after years of (internal) eye rolling when van Inwagen professed not to be able to
understand me. Thanks Peter!
8
(Translation from Plato 1987.) For those who don’t know the reference, the gist of it is that children
are wont to choose what feels good rather than what is good: they’ll happily scarf down pastry after pastry
but won’t take their medicine or eat their vegetables. We should take our medicine.
9
Available here: <http://kieranhealy.org/blog/archives/2013/06/19/lewis-and-the-women/>.
10
David Lewis appears on the list eight times; the others appear twice each. Obviously, this is nowhere
near the whole story—Kripke and Quine, for example, only appear once on the list, but Naming and
Necessity (Kripke 1980) and Word and Object (Quine 1960) were each referenced more than the sum of
Putnam’s, Fodor’s, and van Inwagen’s top two publications.
  

the last four decades. As Timothy Williamson put it, van Inwagen’s work is among
the “liveliest, exactest, and most creative . . . of the final third of the 20th Century”
(Williamson 2007: 19). The essays in this volume reflect the impact of this work, and
aim to further it. Some of the essays are on themes from van Inwagen’s work that are
well explored, while others are on themes that are in their philosophical infancy. All
the themes will be important for years to come.
This volume is divided into five parts, the first four reflecting van Inwagen’s
primary areas of philosophical interest: Being (metaphysics), Freedom (philosophy
of action), God (philosophy of religion), and Method (meta-philosophy). The final
section contains reflections by van Inwagen on selected essays from the previous
sections.

Being
In the 1980s and 1990s, van Inwagen created, almost ex nihilo, a new domain of
metaphysical inquiry: questions about “material constitution.” Many philosophers
regard van Inwagen’s Material Beings (van Inwagen 1990) as the best book on
metaphysics published in the 1990s. It is, I think, certainly the most influential: the
debate about material constitution is now one of the most central and important
areas of contemporary metaphysics, and animalism has become a mainstream
position about the metaphysics of the human person. And of course van Inwagen
has also done extremely influential work in other areas of metaphysics, including
modality (where he defends an “abstractionist” modal ontology), modal epistemol-
ogy (where he defends a limited form of modal skepticism), the existence of abstracta
(defending platonism), and the ontology of fiction (defending the existence of
creatures of fiction).
The essays in this section are, however, focused on more recent themes in van
Inwagen’s work. While his best-known work in metaphysics is on material composi-
tion and the metaphysics of the human person, some of van Inwagen’s recent work is
transforming the literature on time travel and constituent ontologies. Sara Bernstein’s
contribution to this section explores the model of time travel proposed by van Inwagen
in ‘Changing the Past’ (van Inwagen 2010), and the papers by Michael Loux, Laurie
Paul, and Eric Olson all respond, in different ways, to van Inwagen’s arguments for
preferring a relational to a constituent ontology (in, e.g., van Inwagen 2011). Loux’s
contribution uses terminology made mainstream by van Inwagen to categorize diffe-
rent positions on what it is for a thing to have the character (or properties) that it does;
Olson follows van Inwagen in critiquing constituent ontologies; and Paul defends a
version of constituent ontology that is immune to many of van Inwagen’s (and
Olson’s) complaints.

Freedom
Van Inwagen’s work on free will in the 1970s and 1980s was instrumental in undoing
a near consensus amongst Anglophone philosophers in favor of some form of
compatibilism—a consensus that had reigned since roughly the time of Mill. In
conjunction with several important essays written in the 1970s, van Inwagen’s An
  . 

Essay on Free Will (van Inwagen 1983) cemented incompatibilism’s place at the
philosophical table, and remains widely regarded as one of the best and most
important books in the philosophy of action. The central argument of that book,
the Consequence Argument, is generally thought to be the single most important
argument for incompatibilism. Van Inwagen also instigated the contemporary debate
about the frequency of free actions, and has been one of the most forceful critics of
Harry Frankfurt’s arguments that moral responsibility is compatible with causal
determinism.
The essays in this section press van Inwagen on his views about freedom. Neal
Tognazzini and John Martin Fischer square off with Wesley Holliday concerning the
incompatibilist’s contention that we are unable to change the past and about the
relationship between freedom and modality more generally. Alicia Finch proposes a
new version of the Mind argument, utilizing the notion of grounding, which she
argues is valid if the Consequence Argument is. Finally, Mark Heller develops an
underexplored argument against libertarianism, hinging on the notion of degrees of
influence, and is led to endorse a form of contextualism about free will according to
which there are contexts such that it is correct to say ‘Compatibilism is true and we
have free will’ and others such that it is correct to say ‘Compatibilism is false and
we lack free will’, but no contexts such that it is correct to say ‘Compatibilism is false
and we have free will’.

God
After metaphysics and free will, van Inwagen is most famous for his work on the
philosophy of religion. He has written extensively on the problem of evil, the
Doctrine of the Trinity, and arguments for the existence of God. He is also one of
the primary defenders of “open theism” and one of the most prominent defenders of
the compatibility of evolutionary theory and Christianity.
Many of these themes and conclusions are explored in the essays in this section.
Alex Rosenberg argues, contra van Inwagen, that evolution and theism are incom-
patible, while Daniel Howard-Snyder argues that Paul Draper’s (2008) influential
argument for this incompatibility claim is a failure. Louise Antony critiques van
Inwagen’s reply to the argument from evil (in his 2006), while Frances Howard-
Snyder sympathetically explores some of the moral underpinnings of that reply.
Eleonore Stump lays the groundwork for a unified understanding of both the
problem of evil and the doctrine of the atonement. Finally, Lynne Baker pushes
back on van Inwagen’s dismissal of the ontological argument, arguing that Anselm’s
version of the argument does not hinge on an implausible Meinongian ontology.

Method
Given the extent of van Inwagen’s contributions to the philosophy of religion,
metaphysics, and the philosophy of action, it is perhaps not surprising that his
contributions to philosophical methodology are less well recognized that his contri-
butions to those sub-disciplines. But van Inwagen’s work on philosophical method
has been at least as important: his work on the philosophy of philosophy has been
  

both trailblazing and trendsetting. Van Inwagen wrote the foundational essay (van
Inwagen 1996) in the contemporary debate about the epistemic significance of disagree-
ment, and his work on meta-ontology (e.g., van Inwagen 1998a) is largely responsible
for the currently feverish level of interest in this topic.
While both meta-ontology and disagreement remain topics of tremendous import
and interest, there is no lack of good work on them. But van Inwagen has recently
ignited another meta-philosophical debate, giving a novel and fascinating account of
what is required for an argument to be a success, and arguing that there are no
successful arguments for substantive philosophical conclusions.11 The three essays in
this section all focus on this topic. David Chalmers investigates the connection
between success and progress, distinguishes three different questions that might be
expressed by ‘Why isn’t there more progress in philosophy?’, and explores some
competing answers to them. John Keller argues that there is no useful objective notion
of argumentative success—that many arguments (and most philosophical arguments)
can only be usefully evaluated for success relative to individuals. Finally, Thomas Kelly
and Sarah McGrath criticize van Inwagen’s account of success and failure and argue
that there are, in fact, successful philosophical arguments.

Concluding Meditation
In this last part of the volume, van Inwagen responds to the papers by Louise Antony,
David Chalmers, John Keller, Thomas Kelly and Sarah McGrath, Michael Loux,
Laurie Paul, and Alex Rosenberg. These responses give a nice indication of where the
next rounds of debate will be conducted.

Acknowledgments
Thanks to Lorraine Juliano Keller, Steve Petersen, Peter van Inwagen, and Dean Zimmerman
for helpful comments on this introduction.

References
Baier, A. (1972), ‘Nonsense’, in P. Edwards (ed.), The Encyclopedia of Philosophy (Macmillian).
Cappelen, H. (2013), ‘Nonsense and Illusions of Thought’, in Philosophical Perspectives:
Philosophy of Language 27: 22–50.
Diamond, C. (1981), ‘What Nonsense Might Be’, Philosophy 56: 5–22.
Draper, P. (2008), ‘Evolution and the Problem of Evil’, in L. Pojman and M. Rea (eds.),
Philosophy of Religion: An Anthology, 5th edn. (Wadsworth): 207–19.
Evans, G. (1982), Varieties of Reference (Oxford).
Kripke, S. (1980), Naming and Necessity (Harvard).
Plato (1987), Gorgias, translated by Zeyl, D. (Hackett).
Quine, W.O. (1960), Word and Object (MIT Press).
Recanati, F. (1997), ‘Can We Believe What We Do Not Understand?’, Mind & Language
12: 84–100.

11
In van Inwagen (2006).
  . 

Van Inwagen, P. (1980), ‘Philosophers and the Words “Human Body” ’, in P. van Inwagen
(ed.), Time and Cause: Essays Presented to Richard Taylor (Springer), 283–300.
Van Inwagen, P. (1981), ‘Why I Do Not Understand Substitutional Quantification’, Philosophical
Studies 39: 281–5. Reprinted in van Inwagen (2001).
Van Inwagen, P. (1983), An Essay on Free Will (Oxford).
Van Inwagen, P. (1988), ‘On Always Being Wrong’, Midwest Studies in Philosophy 12: 95–111.
Van Inwagen, P. (1990), Material Beings (Cornell).
Van Inwagen, P. (1996), ‘It is Wrong Everywhere, Always and for Anyone, to Believe Anything
Upon Insufficient Evidence’, in J. Jordan and D. Howard-Snyder (eds.), Faith, Freedom, and
Rationality (Rowman & Littlefield).
Van Inwagen, P. (1998a), ‘Meta-ontology’, Erkenntnis 48: 233–50. Reprinted in van Inwagen
(2001).
Van Inwagen, P. (1998b), ‘Modal Epistemology’, Philosophical Studies 92: 67–84. Reprinted in
van Inwagen (2001).
Van Inwagen, P. (2000), ‘Temporal Parts and Identity Across Time’, The Monist 83: 437–59.
Reprinted in van Inwagen (2001).
Van Inwagen, P. (2001), Ontology, Identity, and Modality (Cambridge).
Van Inwagen, P. (2006), The Problem of Evil (Oxford).
Van Inwagen, P. (2008), ‘How to Think about the Problem of Free Will’, Journal of Ethics
12: 327–41.
Van Inwagen, P. (2010), ‘Changing the Past’, in K. Bennett and D. Zimmerman (eds.), Oxford
Studies in Metaphysics, vol. 5 (Oxford), 3–28.
Van Inwagen, P. (2011), ‘Relational vs. Constituent Ontologies’, in Philosophical Perspectives:
Metaphysics.
Van Inwagen, P. (2014), ‘Five Questions’, in Existence: Essays in Ontology (Oxford).
Williamson, T. (2007), The Philosophy of Philosophy (Oxford).
PART I
Being
 1
Theories of Character
Michael J. Loux

The standard vocabulary we employ in characterizing debates over what is called the
problem of universals fails to do justice to the range of theoretical options in play in
those debates.1 We speak of nominalism, conceptualism, moderate realism, and
extreme realism. We learnt these labels in conjunction with the medieval debate,
but even there the labels fail us. They are supposed to be opposed; but, then, either
those who identified universals with the general terms of a mental language come out
nominalists, and there are few if any medieval conceptualists; or we restrict the term
‘nominalism’ to views that identify universals with merely conventional linguistic
items, and we have the result that there were no medieval nominalists except,
perhaps, the old standby, Roscelin of Compiègne. The labels are even less satisfactory
when used to characterize the contemporary debate. Who, after all, is a conceptualist
these days?2 And if we follow the traditional reading, moderate realism is supposed to
be the sort of view we meet in Chapter 3 of Aquinas’ De Ente et Essentia, but no one
today except the occasional Thomist endorses that account. Indeed, no one today
except the occasional Thomist claims even to understand the account. So if we stick

1
An earlier, shorter version of this paper was presented at a conference on the problem of universals at
the Scuola Normale Superiore in Pisa, Italy and was subsequently published in the proceedings of the
conference in the Scuola’s Documenti e Studi Sulla Tradizione Filosofica Medievale 18 (2007) under the title
“Perspectives on the Problem of Universals”; but as I try to show, what is called the problem of universals is
just a version of a more general problem, what I call the problem of character. Hence, the new title. Some
material from the earlier piece is reprinted here with the permission of the Scuola.
What led me to revise the earlier piece for publication in this volume was reading van Inwagen’s recent
work on ontological categories and the classification of ontological theories. See van Inwagen (2011) and
van Inwagen (2012). His taxonomy is broader than mine, but the two are complementary and, taken
together, they provide a nice picture of the dialectic at work in this important region of philosophy. Since
the earlier paper was “tucked away” in an Italian journal read almost exclusively by medievalists, I thought
it worth rewriting the paper.
I agree with most of what van Inwagen says about categories and ontological theories. My only serious
concern is his use of the framework of sets or classes in the definition of a category. Since sets or classes
have their memberships essentially, I do not see how van Inwagen’s definition can avoid the conclusion
that a category in one possible world is different from a category in another world merely in virtue of their
having different extensions in those worlds. The two could be composed exclusively of ontologically
indistinguishable objects. Since kinds (like properties and relations) do not have their extensions essen-
tially, I would have employed the notion of a kind in defining the concept of a category.
2
One possible exception here is Wilfrid Sellars. Sellars is typically taken to be a nominalist, but his view
could be read as a form of conceptualism. See Sellars (1963) and Sellars (1967).
  . 

with the traditional labels, we have the stark conclusion that there are only two
serious options in the contemporary debate over the problem of universals: extreme
realism and nominalism. It is difficult to believe that the philosophical imagination of
recent ontologists has been that impoverished!
We need, then, a new framework for characterizing the topography of debates in
this area. I want to suggest that we can develop the requisite framework if we invoke a
distinction Aristotle draws between opposing attempts to identify, as he puts it, the
substance of familiar sensible particulars. However, before we can see how Aristotle’s
distinction provides the foundation for a framework for characterizing both classical
and contemporary approaches to what we call the problem of universals, we need a
clearer understanding of just what that problem is supposed to be. So let us start
there.

1.1
The literature surrounding the so-called problem of universals does not seem to have
a single focus. Indeed, we are hard pressed to identify a single philosophical concern
here.3 We seem instead to have a number of distinct and different philosophical
issues in play. Philosophers who claim to be responding to the problem of universals
sometimes tell us that they are interested in what, following Russell, we might call the
character of familiar particulars,4 that is, their having the properties they do, their
belonging to the kinds they do, and their being related to each other in the ways they
are. This talk of character is meant to be understood prephilosophically; it is
supposed to be theory-neutral. Indeed, on this story, the problem of universals just
is that of providing the right theoretical account of the commonsense fact that a given
particular has this or that form of character.
But we often meet a somewhat different account of what the problem here is
supposed to be. We are told that the problem is one of providing the theoretical
underpinnings of the commonsense fact that numerically different concrete particu-
lars are similar or agree in attribute. Here, the phenomenon that is supposed to
require a theoretical explanation is the prephilosophical fact that distinct familiar
particulars, as we say, have the same property, belong to the same kind, or enter into
the same relation. This problem has been dubbed “the problem of the one over the
many,”5 and while it may appear otherwise, it is a problem distinct from our first
problem: the phenomenon central to the second problem always involves numeric-
ally different particulars; whereas the explanandum associated with the first problem
need not involve more than one particular; and if there are forms of character
necessarily unique to just one particular, then while our second question about

3
For general discussions of the problem of universals, see Oliver (1996), the introduction to Mellor and
Oliver (1997), and Chapters 1 and 2 of Loux (2006a). An important issue that I do not discuss is the
familiar set of difficulties of providing a nonquestion-begging account of the distinction between universals
and particulars. The difficulties are laid out in the classic Ramsey (1925), discussed in Chapters 5 and 6 of
Strawson (1959) and restated in MacBride (2005). I have little to add to Strawson’s defense of the
distinction except to point out that the sort of strategy that Strawson himself employed in the case of
the analytic/synthetic distinction seems applicable here. See Strawson and Grice (1957).
4 5
See Russell (1912: 92). See, for example, Armstrong (1980).
    

similarity cannot arise in their case, we can still ask for the metaphysical underpin-
nings of a particular’s having such a form of character.
And there are still other issues that are the focus of discussions in this neighbor-
hood. Philosophers confronting what they refer to as the problem of universals often
tell us that their concern is with the phenomenon of predication. But when they have
spoken of predication, these philosophers have not all had precisely the same set of
philosophical issues in mind. In all these discussions, the target phenomenon is
linguistic, but just which philosophical issue the linguistic data are supposed to
generate varies. Thus, we are sometimes told that the philosophical problem lies
with certain very basic subject/predicate sentences, sentences in which we apply a
predicate expression to a familiar particular. The claim is that such sentences call
out for analysis—a “logical” or “reductive” analysis6—in which the phenomenon of
predication gives way to a theoretically more perspicuous form. In other contexts, we
meet with the more modest demand that we provide truth conditions for sentences
of this form.7 Here, the project is that of identifying the kind of ontological config-
uration that must obtain if sentences of the subject/predicate form are to come out
true. In still other contexts, the demand is that we provide a semantical account of
terms suited to play the predicate role; and we are told that what needs to be
identified is the metaphysical ground of the fact that predicate terms with precisely
the same extension can, nevertheless, differ in meaning or intension.8 Finally, we
meet with the claim that it is the subject/predicate distinction itself that stands in
need of explanation.9 What is supposed to be needed is some sort of metaphysical
account of the fact that certain terms are suited to play the subject role and others the
predicate role.
Predication is not, however, the only linguistic phenomenon that has been at the
core of discussions about the problem of universals. In at least two ways, what we
might call the phenomenon of abstract reference has been central in these discus-
sions. First, there are sentences incorporating abstract singular terms—expressions
like ‘wisdom’, ‘triangularity’, and ‘mankind’.10 Many of these sentences express
beliefs all of us accept in our nonphilosophical moments; their incorporated abstract
singular terms appear to be playing genuinely referring roles; and in those roles they
appear to be picking out items other than familiar concrete particulars. Debate over
these sentences has focused on a pair of issues. Is it possible to paraphrase these
sentences in such a way that the appearance of singular reference disappears? If not,
just what do the relevant abstract terms pick out?
Second, there are sentences incorporating quantifiers that appear to range over
things other than familiar concrete particulars, sentences like ‘Some shapes have
never been named’.11 Again, many such sentences express beliefs we all accept in our
nonphilosophical moments. Such sentences, however, can be true only if the things
over which the quantifiers range satisfy the predicates the sentences apply to them.

6 7
Armstrong (1978a: 13). See, for example, Donagan (1963).
8 9
See, for example, Baylis (1951). See, for example, Frege (1892).
10
See Chapter 4 of Loux (1978) and Jackson (1977) for discussions of abstract singular terms.
11
See, again, Chapter 4 of Loux (1978). The classic discussion of the issue is, of course, found in Quine
(1948).
  . 

Again, debate over these sentences has focused on two issues. Can the sentences be
paraphrased in such a way that the apparent commitment to things other than
familiar concrete particulars disappears? If not, just which entities must we construe
their quantifiers as ranging over?
And still other issues have been at the forefront of discussions over universals. For
some philosophers, what takes center stage here are certain mental acts, acts in which
we ascribe or withhold attributes, and the challenge is to provide an account of the
nature and status of the objects of those acts.12 For others, questions about causal
powers provide the focus for a theory of universals. For these philosophers, the
challenge is to identify the metaphysical underpinnings for the active and passive
causal potentialities of familiar concrete particulars; and frequently that project gets
tied up with another project—that of providing a satisfactory account of the nomic
regularities displayed by familiar objects.13
So it seems that there is no such thing as the problem of universals. On the
contrary, we seem forced to conclude that what we call the debate over universals
unfolds into a plurality of independent debates, each focusing on a distinct and
different philosophical problem. Further reflection, however, might persuade us that
such a conclusion is premature. Consider the last two sets of issues we set out—the
issue of mental acts and the issue of causal powers and lawlike regularity. Pretty
clearly, these issues are just special cases of problems mentioned earlier. If we think
that mental acts of ascription are relevant to the debate over universals, it is because
we think that what we ascribe in such acts are forms of character, and we think that
the phenomenon of character is pivotal in discussions of universals. Likewise, if we
think that questions about causal powers and laws are relevant to debates over
universals, it is because we think that the causal powers a familiar particular displays
as well as the laws to which it is subject hinge on its character—on the properties it
has, on the kinds to which it belongs, and on the relations into which it enters; and we
think, again, that issues of character are among the core issues in debates surround-
ing universals.
And the idea that issues about mental acts and causal powers get subsumed under
the first issue we set out (the issue of character) seems to generalize. Pretty clearly,
our second issue (that of similarity or attribute agreement) is an issue of character—
the issue of shared character, and that explains its relevance to discussions about
universals. What makes the issue of predication relevant to such discussions is just
that predication represents the linguistic apparatus for expressing our ascriptions of
character; whereas what makes both abstract singular terms and what we might call
higher order quantifiers relevant is that they provide us with the linguistic tools for
making reference (both singular and general) to the various forms of character. The
upshot is the hypothesis that, despite its different expressions, the issue of character is
the central theme in debates surrounding what is called the problem of universals.
Most metaphysicians would endorse this hypothesis. Most, but not all. The
exceptions are philosophers who deny that the sorts of issues central in discussions

12
See Chisholm (1976) and Chisholm (1989b).
13
See Armstrong (1978b), Shoemaker (1980), Mellor (1991a). See also Putnam (1970).
    

of character, shared character, and predicated character represent serious philosoph-

ical concerns. These philosophers deny, for example, that there is any need to provide
a substantive explanation of the fact that a given individual or group of individuals
exhibits a given form of character. Indeed, they insist that we take facts of character to
be primitive or brute facts. As Quine puts it, “That . . . houses and roses and sunsets
are all of them red may be taken as ultimate and irreducible.”14 David Lewis makes a
similar point when he tells us that while we may need to provide an account of
commonsense facts of character and shared character, an acceptable account here is
one that takes the fact that one or more individuals are, say, red or human or
triangular, to be a primitive fact.15 And if there are no serious metaphysical problems
of character or shared character, it is difficult to see how the linguistic phenomenon of
predication could present us with a serious metaphysical problem. As philosophers of
this ilk want to claim, what makes a subject/predicate sentence of the form ‘x is F ’ true
is that the object named by ‘x’ is, in fact, F, and that they take to be metaphysically
basic.16
But while denying that our first three issues ought to give rise to any serious
metaphysical project, these philosophers have to confront the sorts of problems
presented by sentences incorporating the vehicles for abstract reference—abstract
singular terms and higher order quantifiers; they need to provide an account showing
how those sentences can manage to be true. Both Quine and Lewis agree: Quine,
implicitly, by dealing with sentences of both sorts;17 Lewis, explicitly, by telling us
that we need to provide substantive accounts of the two sorts of sentences.18

14 15
Quine (1948: 81). Lewis (1983).
16
I am inclined to identify these primitivists with what I call austere nominalists, philosophers who
deny the existence of anything other than concrete particulars and (possibly) sets of concrete particulars;
but one might challenge this identification on the grounds that one could deny that familiar facts of
character submit to any general substantive explanation while still being a realist about properties and
relations. Here, Peter van Inwagen comes to mind. Van Inwagen (2011) denies that it counts as a
substantive explanation of the fact that a concrete particular is, say, green to claim that it exemplifies the
color green. Whether he is right or not, most of those involved in the debate would take such a claim to
involve a genuine explanation; and despite his denial here, van Inwagen himself takes character to be
grounded in what he calls properties, and his properties include what have traditionally been called
universals. Thus, van Inwagen (2006a) tells us that, in our prephilosophical moments, we all believe that
there are anatomical features that insects and spiders share, and he argues that there is (likely) no way of
understanding how that belief could be true without conceding the existence of shared properties. But
surely this is to claim that the universals in question ground one or more facts about shared character.
Here, however, one might object that even an austere nominalist like David Lewis would agree that our
acceptance of the belief about spiders and insects commits us to the existence of properties. But if Lewis can
endorse this claim no less than a van Inwagen, then my suggestion that van Inwagen is committed to
conceding that his properties ground or explain shared character has to be wrong. But there is a difference
in the two cases. What Lewis calls properties do not, in any reasonable sense, serve to ground or explain
shared character. For Lewis, the property of being F is just the set of all possible F-objects. But clearly it is
not because a thing is a member of the set of all possible F-objects that it is F. The direction of
determination is precisely the reverse. The fact of character (that a thing is F) underlies what Lewis calls
its exhibiting the associated property (that the thing is a member of the set of all possible F-objects). For
van Inwagen, by contrast, properties are not sets. They are what he calls unsaturated assertibles, and, on his
view, it is because they share the relevant assertibles that spiders and insects agree anatomically.
17 18
See Quine (1948) and Quine (1960: 118–24, 241–3). Lewis (1983).
  . 

So there are philosophers who want to separate our problems; but it is important
to recognize that they are clearly in the minority. Most metaphysicians think that we
need to provide a substantive account of character, shared character, and predication—
and not just the sort of trivial account Lewis recommends—and they deny that our
accounts of character and shared character can come apart. They believe that what
makes the phenomenon of abstract reference a pressing concern is just that, in many
cases, abstract singular terms and higher order quantifiers take as their referents the
items underlying character, shared character, and predication. If a term like ‘wisdom’
picks out anything, it is the thing that provides a metaphysical ground for the fact
that familiar particulars are wise; and that is the thing that has to be involved in our
account of subject/predicate sentences of the form ‘x is wise’. Parallel points hold for
any quantifiers that range over the sorts of things picked out by abstract singular
terms like ‘wisdom’.

1.2
So, for all but a few philosophers, there is a single theme that ties together the
different strands in discussions of universals. It is the phenomenon of character.
Indeed, what we call the problem of universals might better be called the problem of
character; and we want to know what strategies are available to the metaphysician
seeking to provide a substantive account of this phenomenon. In a number of places
in the Metaphysics, Aristotle points to two very general approaches when he tells us
that in identifying the substance of a familiar sensible particular, we can make it
something immanent in the thing or something that exists “apart from” the thing.19
Aristotle’s concern here is with a more circumscribed phenomenon than our phe-
nomenon of character. To identify the substance of a thing is to identify that in virtue
of which the thing is what it is or belongs to the kind it does; but if we generalize and
take Aristotle’s suggestion to apply not just to accounts of a thing’s essence but to all
the dimensions of its character (including all of its accidental features), then we can
flesh out his suggestion in a straightforward way. The idea is that familiar sensible
particulars have their character dependently or derivatively. They, so to speak, inherit
their various forms of character; and the items from which they inherit their
character have their own distinctive character nonderivatively or independently. To
use Aristotle’s language, whereas sensible particulars have their character in virtue of
another (kat’allo), the items on which they depend have their proper character in
their own right (kath hauto). Aristotle’s two strategies differ in their accounts of just
how familiar sensible particulars derive their character from these privileged sources
or principles of character. Those who endorse the immanentist strategy tell us that
the things that have character in their own right are something like parts or
components of familiar particulars. So a familiar object has this or that form of
character because it has as some kind of part or component a thing that has an
appropriate character nonderivatively or in its own right. Those, by contrast, who
endorse the opposing, nonimmanentist strategy have the underived sources of

19
996a15, 998a21–23, 1080a37–1080b2.
    

character existing “apart from the sensibles”20 and tell us that it is in virtue of
standing in some nonmereological relation to those underived sources of character
that ordinary objects have the forms of character they do.
In characterizing precisely the same contrast in styles, Nicholas Wolterstorff
speaks of constituent ontologies and relational ontologies.21 He takes the term
‘constituent’ from one of the most committed practitioners of the immanentist
style of metaphysics—Gustav Bergmann.22 As Wolterstorff and Bergmann use the
term, talk of constituents points to the fact that, within an immanentist framework,
familiar particulars are composites, wholes, or complexes—things made up of meta-
physically prior and more fine-grained items. These items are to be distinguished
from a thing’s commonsense parts. As the constituent ontologist sees things, the
constituents making up a composite are responsible for its commonsense mereo-
logical structure (its being composed, say, of hands, feet, and the like) no less than
any other feature comprising the composite’s character.
So on the immanentist strategy familiar particulars have a mereological structure
over and above their commonsense mereological structure. They have two kinds of
parts or components. Both kinds of parts are less than or fall short of the wholes they
compose; but they fall short in different ways. A commonsense part is spatially less
than its whole: the place it occupies is a proper part of the place occupied by the
whole. By contrast, a constituent (what we might call a metaphysical part) is, as
Aristotle puts it, substantially less than its whole; it induces a form of being that is less
than the total form of being displayed by the whole.23
What Wolterstorff calls relational ontologists, by contrast, deny that familiar
particulars have any mereological structure other than their commonsense mereo-
logical structure. As they see it, the only parts or components ordinary objects have
are their commonsense, spatial parts. They insist that the items from which concrete
particulars derive their character are abstract entities, and they deny that anything
concrete can be made out or composed of abstract objects. Concrete entities can,
however, enter into ontologically significant nonmereological relations with abstract
entities; and as a result, they have whatever character they do.
Although I will follow Wolterstorff in using the term ‘relational’ as a label for the
nonimmanentist strategy, it is not the happiest of labels. For one thing, the constitu-
ent ontologist’s story of character derivation relies on relations no less than that of
what Wolterstorff calls the relational ontologist: it is in virtue of relations among its
constituents that a familiar particular has the structure and nature it does. For
another, few proponents of the nonimmanentist strategy want to use the term
‘relation’ in their account of just how particulars stand with respect to the various
underived sources of character. They think that talk of relations is just further talk of
character; it requires an account, and we seem to be off on a vicious regress.
Intriguingly, Wolterstorff himself defends the nonimmanentist strategy, and he is
the only nonimmanentist I know of who explicitly denies that there is a problem in

20
996a15. 21
See Wolterstorff (1970) and Wolterstorff (1991).
22
For Bergmann’s use of the term, see Chapters 1 and 2 of Bergmann (1967).
23
See Aristotle, 1034b32–4. David Lewis labels the immanentist’s constituents nonspatiotemporal parts.
See Lewis (1983).
  . 

supposing the connection between familiar objects and the underived sources of
character to be genuinely relational.24
Caveats are likewise in order with regard to my use of the term ‘mereological’ in
conjunction with the immanentist or constituent strategy. I have already denied that
the constituents of a thing are to be identified with its commonsense parts. The
constituent/whole relation must also be distinguished from the part/whole relation at
work in what is properly called mereology—the logic of parts and wholes. A complex
and its constituents are not related as a strictly mereological sum or fusion is related
to its parts. Nor should the relation of complex and constituent be assimilated to
either of two other familiar compositional relations—that of a set to its members or a
conjunctive property to its conjuncts. Both of these relations as well as the strictly
mereological relation between fusion/sum and proper part are normally taken to be
necessarily such that if it is possible for a plurality of items to make up or compose
the relevant sort of complex structure (whether a mereological sum, a set, or a
conjunctive property), then they do make it up.25 But the constituents of a complex
concrete object are regularly thought to compose that object only contingently: it is
possible for them to exist, we are told, without composing that object.
So the constituent/whole relation is a contingent relation distinct from other
familiar compositional relations. What more can be said about it? Well, immanent-
ists tell us that while it is only contingently that the constituents of a familiar
particular go together to compose that particular, the particular has those constitu-
ents necessarily or essentially. They endorse, then, what we might call the Principle of
Constituent Essentialism; and they supplement this principle with what we might
call, the Principle of Constituent Identity, the claim that it is impossible for numer-
ically distinct particulars to have all and only the same constituents arranged in
precisely the same way. And they defend their acceptance of these two principles by
telling us that since, on their approach, a familiar particular is nothing more than its
constituents or metaphysical parts arranged or configured in a particular way,
anyone who endorses the constituent strategy has to take these principles as con-
straints on the ontological enterprise itself.26

1.3
So we have two strategies for understanding how familiar concrete particulars derive
their character: the immanentist strategy where the underived sources of character
are constituents of those particulars and the nonimmanentist or relational strategy
where it is in virtue of some nonmereological relation to transcendent sources of
character that ordinary objects have their character. Examples of the two approaches
are easy to come by. Philosophers like David Armstrong and Gustav Bergmann are
paradigms of one strand in the constituent tradition. Here, familiar objects are
structured wholes whose constituents fall under two distinct categories: an under-
lying subject and the various universals predicated of it.27 A closely related strand

24 25
Wolterstorff (1973: 101–4). But see van Inwagen (2006b).
26
I discuss these principles in more detail in Loux (2006b).
27
See Chapter 5 of Armstrong (1989) and Chapter 2 of Bergmann (1967).
    

replaces the universals with tropes and speaks of those tropes as inhering in the
underlying subject. Locke is typically cited as an example of someone who endorses
that type of constituent ontology. Two other strands reject the idea of an underlying
subject or substratum and identify ordinary particulars with bundles of compresent
properties, one strand construing properties as universals and the other as tropes.
D. C. Williams and his followers endorse a trope bundle theory;28 whereas the Russell
of Inquiry into Meaning and Truth defends a realist version of the bundle theory (as
do Herbert Hochberg and Hector Castañeda).29 In earlier works such as The Problems
of Philosophy, however, Russell was a staunch defender of the relational strategy;30 and
in our own day, that strategy is defended by Roderick Chisholm, P. F. Strawson, Alvin
Plantinga, Peter van Inwagen, and, as I said, Nicholas Wolterstorff.31
The philosophers we have mentioned endorse either the constituent or relational
strategy, but not both. It is worth noting that a mixed ontology is possible. One
might, for example, hold that familiar particulars derive their character from tropes,
where the latter are constituents of the former; but, then, one might go on to claim
that these tropes have their character in virtue of standing in the relation or tie of
instantiation to trope-types, where these are universals that have their own character
nonderivatively.32 Such a theory manages to be both immanentist and relational, but
only because it is a two-step ontological theory. There is a genuine opposition
between our two strategies: at each step in one’s explanation of character one must
choose between the strategies. As our examples testify, when they seek to explain the
character of familiar particulars, ontologists typically provide accounts that take us,
in a single step, to some source of character, whether immanent or transcendent, that
has character nonderivatively or in its own right.
Apart from Locke, our examples of constituent and relational ontologists are
contemporary thinkers or figures from the recent past; but, of course, the classical
examples of our two approaches are supposed to be Aristotle himself and Plato.
Indeed, philosophers standardly use the adjectives ‘Aristotelian’ and ‘Platonic’ to
mark what they take to be a central divide in metaphysics. The divide they typically
have in mind, however, is not quite the contrast we have been discussing. It is a
contrast in which the distinction between constituent and relational approaches
figures, but it is a narrower distinction, one between two different theories about
the nature and status of universals. On the so-called Aristotelian view, universality is
ultimately a matter of overlap; whereas on what is called the Platonic view, having a
universal in common is a matter of standing in some sui generis nonmereological
relation or tie (‘exemplification’ is the standard label for the connection) to some
single separately existing abstract entity. What are typically called Aristotelian and
Platonic theories, then, are respectively, constituent and relational theories, but there

28 29
See Williams (1953). See Russell (1940), Castañeda (1974), and Hochberg (1960).
30
See Russell (1912), Chapter 9.
31
See Chisholm (1989b), Chapters 5 and 6 of Strawson (1959), Plantinga (1974), Wolterstorff (1973),
van Inwagen (2006a). See also Baylis (1951).
32
On a familiar reading of the Categories, where we have both first and second accidents, Aristotle
himself comes out endorsing such a two-step theory for the forms of character associated with the
accidental categories.
  . 

are both constituent and relational accounts of character derivation that are neither
Aristotelian nor Platonic—theories that refuse to identify the ultimate sources of
character with universals, however understood. The trope theoretic versions of
constituent ontology that we just mentioned are obvious examples, but there are
others. Some theorists who have agreed on the need to provide a substantive,
explanatory account of character derivation have refused to identify the sources of
character with properties, qualities, or attributes of any sort. Aristotle cites examples
of such theorists from both the constituent and relational camps. He construes his
materialist predecessors as what Wolterstorff calls constituent ontologists.33 For
them, the elementary material stuffs or particles are the underived sources of
character; and Aristotle points to Speusippus as someone who construed mathem-
atical objects as “separated substances” and took them to be that in virtue of which
everything else has whatever character it does;34 and if one takes as accurate a certain
interpretation of the Forms of at least some of the middle dialogues—that on which
they turn out to be nonsensible, immutable, paradigmatic particulars rather than
universals,35 then we have a Plato that, ironically, turns out to be a relationist who
rejects a Platonic theory of universals.
So there have been metaphysicians who have answered questions about character
without referring to universals; and we cannot forget that if there are forms of
character necessarily unique to a single concrete particular, then no theory of
character can restrict its underived sources of character to what have traditionally
been called universals, that is, multiply instantiable objects, objects capable of being
common to or shared by numerically different things.36 But while there certainly are
trope theorists on the contemporary metaphysical scene, most recent philosophers
who have thought that a substantive explanatory account of character is mandatory
have appealed to universals in that connection; and if we set aside, for the moment,
questions about individual concepts, haecceities, or individual essences, then we can
agree that, in the case of most recent metaphysicians, the distinction between constitu-
ent and relational theories, on the one hand, and the distinction between Aristotelian
and Platonic theories, on the other, come to much the same thing.
The upshot is that two models for understanding the phenomenon of character
have been dominant in recent analytic metaphysics: the Aristotelian model where we
have immanent universals and universality is a matter of distinct objects overlapping
and the Platonic model where we have transcendent universals and universality is a
matter of distinct objects standing in some sui generis relation or tie to a single
abstract object. Philosophers often characterize the contrast between these two
models in different ways. Thus, we are told that in an Aristotelian theory, universals
have spatial or spatiotemporal location, but that in a Platonic theory, they are, as the
Russell of Problems put it, “nowhere and nowhen.” In his jargon, universals do not
exist; they subsist.37 Likewise, we are told that although Aristotelians endorse some

33
See 998a21ff. 34
See 1076a21–2 and 1080b15ff.
35
Aristotle himself sometimes portrays Plato in these terms. See 997a34–997b12. The role of self
predication, if any, in the middle dialogues will be central in evaluating this reading of Plato.
36
For this definition of a universal, see Aristotle, De Interpretatione 7 (17a37–17b1).
37
Russell (1912), Chapter 9.
    

version of the so-called Principle of Instantiation, Platonists reject the principle in all
its versions.38 Finally, we meet with the idea that while, for the Platonist, universals
are one and all necessary beings, Aristotelians construe their immanent universals as
merely contingent beings.39
But these claims are not just alternative ways of drawing the distinction between
Aristotelian or constituent and Platonic or relational theories of universals. As
regards the issue of spatial or spatiotemporal location, it is certainly true that
Aristotelians have typically conceded that, in some sense, universals are located in
space or spacetime; but while there are theorists (like Alan Donagan) who have
endorsed this claim unconditionally,40 constituent ontologists have more typically
wanted to qualify or hedge the claim. Thus, we find Aristotle himself telling us that a
universal can be said to occupy a place only in the derivative sense that it is a
constituent of something—a material particular—that can be said to occupy that
place in the primary and proper sense.41 There is, of course, good reason for
constituent ontologists to be uncomfortable with the bald claim that universals
have spatial or spatiotemporal location. Although the idea that universals have spatial
or temporal location may make good sense within the context of an absolute theory
of space and time, most constituent ontologists have been relationists of one sort or
another about space and time. They have wanted to claim that if the framework of
material particulars is not actually prior to the frameworks of space, time, or space-
time, then at least the frameworks are mutually interdependent. But in either case,
the idea that there are places and times waiting to be occupied by the very universals
that constitute the items making up the framework of material particulars is prob-
lematic. Armstrong expresses this point when he claims that it is crude or improper to
speak of the spatiotemporal location of universals; what we should say, he tells us, is
that universals are central in the constitution of spacetime itself.42 But this is essen-
tially to deny that universals have a genuinely spatial or temporal location; and if it is
compatible with the constituent strategy for a proponent of universals to deny this,
then it is a mistake to claim that the issue of spatial or spatiotemporal location is what
is pivotal in the contrast between Aristotelian and Platonic theories of universals.
It is likewise a mistake to suppose that disagreement over the Principle of Instan-
tiation is what divides Aristotelian and Platonic theories of universals. In very rough
terms, the principle tells us that there are no uninstantiated universals. It has a
number of formulations. I suppose that one could hold that it is merely contingently
true; but those who reject uninstantiated universals standardly take their existence to
be impossible. Further, one’s views about time influence one’s formulation of the
principle. Those who (like David Armstrong, Gustav Bergmann, or D. H. Mellor)
take time to be just another dimension in which things are spread out, a dimension
on a par with the three spatial dimensions, will want to construe the principle as the
claim that necessarily every universal is instantiated at some time or other; whereas,
presentists (like Aristotle) will take the principle to express the much stronger claim
that necessarily every universal is currently instantiated. Now, it is certainly true that

38 39
See Loux (2006a: 40–3) and Chapter 5 of Armstrong (1989), Section I. See Mellor (1991b).
40 41
See Donagan (1963). 212b11–12. 42
Armstrong (1989), Chapter 5, Section VIII.
  . 

proponents of immanent universals have typically endorsed the principle; but it is

important to recognize that it is not mandatory that they do so. It is perfectly possible
for a constituent ontologist to hold that there are or might be universals that are
neither now nor, perhaps, ever instantiated. What distinguishes the Aristotelian from
the Platonist is just the view that what it is for a basic first order universal to be
instantiated is for it to be immanent in or a constituent of some contingent particular;
but one could hold that view while denying that the existence of any such universal
hinges on its being immanent in some contingent particular. In fact, Hector Castañeda,
a bundle theorist, seems to have held just such a combination of views.43 Nor does
denying that commit one to the idea that uninstantiated universals inhabit some
special world distinct from our world. That’s to assume that universals must exist
“somewhere,” and that, the Aristotelian can insist, is just a category mistake, a category
mistake encouraged by a Platonic picture of “two worlds.”
But not only is the Aristotelian not committed to any form of the Principle of
Instantiation, it is also the case that a Platonist can consistently endorse some version
of the principle. What distinguishes the Platonist is just the relational idea that
contingent particulars instantiate universals by standing in the appropriate sui
generis connection to them; but one could embrace that idea while holding that
either as a matter of actual fact or as a matter of necessity every universal is
instantiated. Indeed, a relational theory that endorsed a thesis of unrestricted self
predication or self instantiation for universals would be a theory that accepted
precisely such a combination of themes. We are often told that the Plato of the
middle dialogues endorsed a general thesis of self predication for his Forms. If he did,
then he held to a relational theory that accepted the Principle of Instantiation.44
And, finally, the idea that all constituent ontologists make universals contingent
entities and all relational ontologists make them necessary beings is problematic. If
the suggestion is that this contrast just follows from the contrast between theories
that endorse the Principle of Instantiation and those that reject it, then the fact that
the latter contrast does not map neatly onto the contrast between constituent
ontologies and relational ontologies calls into question the pairing of the contrast
between theories endorsing contingent universals and theories endorsing necessary
universals with the contrast between our two very general ontological strategies. But
the fact is that one can consistently endorse the Principle of Instantiation while
holding that universals are necessary beings. This is not, of course, the standard
move. The typical strategy among recent ontologists, at least, is to follow the lead of
David Armstrong and D. H. Mellor who endorse the principle and go on to make the
existence of the various individual universals the sort of contingent fact it is the
business of our various empirical sciences to establish for us.45 But we have noted
that the middle Plato might have endorsed the Principle of Instantiation, and it is
surely reasonable to think he would have construed his Forms as necessary beings.

43
Castañeda (1974).
44
A theory that endorses unrestricted self predication will, of course, have to deal with the problems
associated with the property version of Russell’s paradox. I discuss those problems in Section 1.4 of this
paper.
45
See Armstrong (1997: 36–7) and Mellor (1991b).
    

And there is the more straightforward example of Aristotle, where we have clear
textual evidence of someone who both endorsed the Principle of Instantiation and
held that universals are necessary beings.46
The case of Aristotle shows that there are costs associated with this combination of
views. As we have noted, Aristotle endorses a presentist version of the Principle of
Instantiation. He thinks that every basic universal is now instantiated; but since he
thinks that every basic universal is a necessary being, he is committed to the view that
every basic universal is eternally and necessarily instantiated. A contemporary the-
orist is certain to find this result problematic. The sheer implausibility of the idea that
necessarily every basic universal is instantiated at every moment in time should, by
itself, give us pause; but there is the further difficulty that if, like Aristotle, we take the
universals underlying the various biological species to be basic universals, then the
price of endorsing Aristotle’s view appears to be the repudiation of the core insights
of contemporary evolutionary theory in biology.
Now, relational ontologists invariably accept the idea that universals are necessary
beings. If, however, the middle Plato did endorse the Principle of Instantiation, he
would be unique among relational ontologists in that regard. The fact is that
relationists typically reject the principle. Accordingly, they are spared the difficulties
of Aristotle’s views; and they are able to enjoy its benefits. There are serious problems
with construing universals as merely contingent beings. For one thing, it is difficult to
accommodate the sort of proposition expressed by a sentence like ‘Courage is a
virtue.’ To all appearances, this is a proposition about the property of courage, so its
truth presupposes the existence of that property. It is, however, a necessary truth, a
proposition true in all possible worlds; but, then, it seems to follow that courage is a
necessary being, a thing that exists in all possible worlds; and, of course, for every
universal we are prepared to admit into our ontology, there will be necessarily true
propositions like this, so we have a perfectly general problem here. For another, if we
hold that universals are contingent beings, that is, if we hold that they are things that
exist in some worlds, but not in others, then we face the difficulty that Peter van
Inwagen has recently pointed to: we are committed to rejecting the seemingly irresist-
ible view that the accessibility relation between possible worlds is symmetrical.47

1.4
So the contrast between theories that grant and theories that deny universals spatial
or spatiotemporal location, the contrast between theories that endorse and theories
that reject the Principle of Instantiation, and the contrast between theories that make
universals contingent and those that construe them as necessary beings are all
different not only from the contrast between Aristotelian and Platonic theories of
universals but also from each other. Nonetheless, the contrasts typically tend to
march in tandem. Most Aristotelian theories endorse a picture where the existence
of universals is intimately bound up with the world of space and time; whereas
Platonists follow the Russell of The Problems of Philosophy in insisting that universals

46
See 14a6–10 and 1039b20–1040a5. 47
See van Inwagen (2006a).
  . 

are “nowhere” and “nowhen.” Likewise, although Aristotelians typically accept some
version of the Principle of Instantiation, Platonists usually insist that there are
uninstantiated universals. Finally, while most contemporary defenders of an imma-
nent theory of universals construe them as contingent entities, Platonists take their
transcendent universals to be necessary beings.
Another contrast that typically goes hand in hand with that between Aristotelian
and Platonic theories is one David Lewis points to—that between sparse and abun-
dant theories of universals.48 The distinction is best viewed as a matter of degree; and
Lewis’ labels convey the force of the contrast. We can get at the contrast if we fix on
the various predicate terms in our language and ask which of them are to be
correlated with universals. Every theory must set some limits here. There is, after
all, the predicate term ‘does not exemplify itself ’; and if we have it express a universal,
we find ourselves confronted straightaway with the attribute version of Russell’s
paradox. But some theories—the sparse theories—go much further; and as Lewis
points out, what we have been calling constituent theories are sparse theories.49
Constituent theorists typically take some version of the Principle of Simplicity to
constrain their characterization of the structure and constitution of familiar concrete
particulars. They do not want what we might call ontological overdetermination.
Accordingly, they insist that very few predicate terms carry ontological force—only
those expressing the contents required to invest familiar particulars with the char-
acter they have. They call those contents basic or fundamental universals; they tell us
that these universals are one and all fully determinate first order universals; and since
they endorse the Principle of Instantiation, they limit their basic or fundamental
universals to fully determinate first order universals that actually are (now or tense-
lessly) immanent in or constituents of some concrete particular.
But what universals are those? Well, it depends on one’s overarching metaphysical
prejudices. If one is, for example, an idealist, then one will hold that the basic or
fundamental universals are fully determinate mental universals of some sort. If, on
the other hand, one is a reductive physicalist, one will hold that what are fundamental
are the sorts of fully determinate universals posited by our best physics, and dualists
will likely posit universals of both sorts. In short, the universals posited by a
constituent theory will be limited to those fully determinate first order universals
that are required to yield what its practitioners take to be the ultimate nature or
character of familiar particulars. But while they have different conceptions of that
character, defenders of constituent theories all tend to understand character in causal
terms. Universals are constituents that confer character, and character is a matter of
causal power, where this involves both the active and passive causal potentialities of a
thing. The upshot is that, in constituent theories, we typically meet with the idea that
we can give identity conditions for universals in causal terms: universals are the same
that confer the same causal powers on the things whose constituents they are.50
The sparseness of a constituent theory may seem attractive, but it has its costs. We
seem committed to the existence of a variety of types of universals over and above

48 49 50
Lewis (1983). Ibid. Armstrong (1978b: 43–5). See also Shoemaker (1980).
    

those recognized by a typical Aristotelian or constituent ontologist.51 There are, first

of all, the generic universals whose species we take fully determinate first order
universals to be as well as all the still more general universals to which those generic
universals are subordinated. For philosophers who want to distinguish what are
called determinables (such as color) from genera,52 there will be another category
of universals that gets ruled out by the constituent ontologist’s restrictive ontology. In
addition, there are the various first order molecular universals that result from the
different forms of combination tying the universals the constituent ontologist takes
to be fundamental. One important case here is that of the so-called structural
universals, universals that apply to concrete particulars in virtue of their parts
instantiating certain other universals.53 Finally, there are the various higher order
universals, universals that are necessarily such that they are instantiated only by
universals.
Defenders of the sort of sparse theories we meet in constituent ontologies owe us a
story about these different cases. One obvious strategy here is a straightforwardly
reductivist approach that construes the recalcitrant universals as definable wholly
and completely in terms of universals recognized by the constituent ontologist.
A second approach is one that denies the possibility of a reduction of this sort and
opts instead for some sort of supervenience relation between fully determinate first
order universals and the problematic universals; and if there are universals that resist
treatment in either of these ways, a possible, if desperate, strategy is to endorse in
their case some sort of eliminativist account.
It is worth pointing to one other kind of universal that might appear problematic
for the constituent ontologist—the various relations that tie familiar concrete par-
ticulars to each other. If every universal is immanent in some concrete particular,
then we need a container for first order relations. It obviously will not do to make a
relation a constituent of just one of the particulars it relates; nor is making it a
constituent of each of them a much better option. That option makes relations
indistinguishable from nonrelational universals and leaves it a mystery just how
they succeed in relating distinct things. One might think that where we have an
n-adic relation, the n-tuple of related particulars is the thing of which the relation is a
constituent; but that option is not attractive either. Since n-tuples are set theoretical
entities, it is plausible to suppose that they are abstract objects; and while sets have
members, it is difficult to understand how abstract entities could be the sorts of things
that have what we have been calling constituents.54 One might try to reduce relations
to correlative relational properties and make those properties immanent in the
related particulars. Thus, the relation of being the father of might be replaced by
the correlative properties of being a father and being a child, and those properties, so
to speak, implanted, respectively, in the relevant father and the relevant son or
daughter. This is the strategy that is often identified with the Aristotelian tradition

51
See Armstrong (1989), Chapter 5, Sections II–III for a discussion of these issues.
52
See, for example, Prior (1949) and Searle (1959).
53
See Armstrong (1978b), where Armstrong endorses structural universals, at least tentatively. See
Lewis (1986a) and (1986b) for an argument against positing structural universals.
54
But see Lewis (1991).
  . 

and a strategy that Russell vehemently attacked.55 Alternatively, one might follow
philosophers like Armstrong and Bergmann and make facts or states of affairs
contingent particulars. The idea would be that facts too have an internal structure.
Some facts have as their constituents a particular and some monadic attribute; but
others—relational facts—have as their constituents not simply the related particulars
but the relation that ties them together as well. One could supplement this view with
the idea (found in both Armstrong and Bergmann) that what we call familiar
particulars just are very complex facts or states of affairs.56
The problems associated with relations are, of course, distinct from the problem
deriving from the sparseness of a theory. Pretty obviously, relations present no
special problems for defenders of a Platonic theory; nor do Platonists typically face
the constituent ontologist’s problems with generic, determinable, molecular, and
higher order universals. They typically endorse abundant theories where all those
kinds of universals have their place. Initially this might seem puzzling. No less than
constituent ontologists, we might think, relational ontologists are bound by some
version of the Principle of Simplicity, so that they too should be unwilling to posit
more universals than those required to give familiar particulars the character they
have. While some relational theorists might take this line and endorse a sparse theory
of transcendent universals, most proponents of a Platonic theory are impressed, first,
with the difficulties defenders of a sparse theory confront in their attempts to tell a
satisfactory story about what seem to be generic, determinable, molecular, and higher
order universals; and, second, they have thought that more is required of a theory of
character than an explanation of the fact that concrete particulars have the character
they do. Character, they think, is not merely something displayed by familiar
particulars; it is something that presents itself as a content for thought. We grasp
the various forms of character; we distinguish them from each other; we ascribe them
to objects; we withhold them from objects; and these facts hold not merely for fully
determinate first order universals but for the other cases as well, and not merely for
those universals that actually are instantiated at some time or other but for forms of
character that never get instantiated in the actual world and even, perhaps, for
contents that get instantiated in no possible world whatsoever. Indeed, relational
ontologists often take it to be as much a part of the essence of a form of character that
it be, in this way, a content for thought as that it be that in virtue of instantiating
which an ordinary object is as it is. The upshot is an abundant theory where the
constituent theorist’s causal identity conditions get supplanted by identity conditions
that make reference to both the instantiation of universals and the intentional acts
that make them their objects. On this view, a universal, U, is identical with a
universal, U', just in case U and U' are necessarily such that (1) a thing instantiates
U if and only if it instantiates U' and (2) a person conceives U if and only if that
person conceives U'.57

55
Russell (1912: 94–5).
56
And, of course, there remains the problem of relations between universals themselves.
57
This account is developed in Chisholm (1989b). Chisholm’s views about uninstantiated universals
have changed over the years. In Chisholm (1976), he denies that there are properties that are instantiated in
no possible world. In later accounts, he seems to allow for such properties.
    

A recent version of this sort of view is outlined in van Inwagen (2006a), where the
underived sources of character (the things that “play the property role”) are identified
with what van Inwagen calls unsaturated assertibles. They are things that can be
asserted of or ascribed to objects. So the view builds into the essence of the things that
ground character the fact that they are contents that can be attributed to things by
intellectual beings like ourselves; and since such beings can ascribe to objects forms of
character that never get instantiated, van Inwagen insists on an abundant theory.58
Accordingly, if we stick with the traditional account of universals as forms of
character possibly common to or instantiated by numerically different objects, then
we will want to say that Platonic theories like van Inwagen’s posit forms of character
that are not universals. But it is not just in dealing with forms of character that are
only possibly instantiated that such theories posit sources of character that are not
universals. Platonic theories often posit forms of character that, while actually
instantiated, are not universals. At one stage in his career, Roderick Chisholm
insisted that, for each concrete particular, we must posit an haecceity or individual
essence, a property essential and necessarily unique to that particular. Van Inwagen’s
theory likewise incorporates individual essences; and the theory of character at work
in Plantinga (1974) has as a central theme the idea that, for each concrete particular,
there are many such properties.59
Although Duns Scotus may be an Aristotelian who includes things like haecceities
among the constituents of a familiar particular, most Aristotelians have refused to do
so. The reason should be apparent. In general, Aristotelians have endorsed something
like a reductive picture of familiar particulars. Underlying their appeal to the constitu-
ent strategy is the intuition that a familiar particular is a kind of construction out of
ontologically more fundamental entities, the entities that are its constituents or meta-
physical parts. That intuition sits uncomfortably with the idea that, for each familiar
particular, there is an irreducibly basic form of being that is fully definitive of the
particular in the sense that it is both essential to and necessarily unique to the particular.

1.5
Now, if we put together the ideas we have been laying out, we have the materials for
what I promised at the beginning of the paper—a kind of topographical map of the
central views one might take in response to our problems. Let us continue to take the
issue of character derivation to be the entry point to the debate. Although we have met
philosophers like Quine and Lewis who deny the need to provide a substantive,
explanatory account of the character of familiar concrete particulars, we have found
that most metaphysicians take the project of providing such an account to be, as
Armstrong puts it, compulsory;60 and although a mixed ontology is possible, we have
found most ontologists employing either of two strategies in their attempts to carry out
the project: the constituent strategy and the relational strategy. Those who endorse the
constituent approach include both those who make the underived sources of character

58
See Part 5 of van Inwagen (2006a).
59
See Chisholm (1976), van Inwagen (2006a), and Plantinga (1974).
60
Armstrong (1980: 441).
  . 

something other than universals and those who make them universals. The former
group includes both those who, like the Greek materialists, want to make something
other than properties or attributes the underived principles of character and those who,
like contemporary trope theorists, insist that while not universals, the ultimate sources
of character are property-like objects. Those constituent ontologists who make uni-
versals the underived principles of character take universals to be immanent in familiar
concrete particulars. We called this view the Aristotelian theory of universals, and we
said that it is typically a sparse theory. We noted that while one could consistently
endorse the Aristotelian view while rejecting the Principle of Instantiation, most
defenders of immanent universals endorse the principle; and we saw that while
proponents of the Principle of Instantiation often take it to entail the view that
universals are contingent entities, it is possible, at some cost, to endorse the principle
and still construe fundamental first order universals as necessary beings.
We seem, however, to be committed to a variety of types of universals that are not
included among the constituent ontologist’s fundamental first order universals; and
we noted that the difficulties Aristotelians confront in accommodating these com-
mitments can provide motivation for adopting an abundant theory of universals.
Further motivation derives from reflection on the roles universals play as the objects
of a variety of intentional acts and attitudes. Now if we couple the desire for an
abundant theory of universals with doubts about the very idea that concrete entities
could, in any way, be made or composed of abstract objects, we are likely to find the
relational strategy attractive.
The relational theorist posits transcendent entities as the underived sources of
character and tells us that it is in virtue of sui generis relations or ties to these entities
that familiar particulars have the character they do. As we noted, one could consist-
ently endorse this strategy while denying the existence of universals; but most
relational ontologists have made universals the transcendent sources of character.
We called relationists of this sort defenders of a Platonic theory of universals, and we
noted that it is at least theoretically possible to couple a Platonic theory with
acceptance of the Principle of Instantiation, and we suggested the middle Plato as
someone who might have held this combination of views. We said, however, that if
he did, he is likely unique in this regard. What we have been calling Platonists
typically reject the Principle of Instantiation, and they construe universals as neces-
sary beings. While they endorse a roster of forms of character more abundant than
that of the Aristotelian, they do not all agree on just how abundant the theory should
be. Some Platonists want to posit forms of character only where we have contents
whose instantiation is possible; whereas others insist on a more generous ontology
including attributes that are instantiated in no possible world,61 and, we said,
Platonists sometimes add to the roster of underived sources of character forms of
character both essential and necessarily unique to familiar concrete particulars.
Here is a chart summarizing all of this, dividing things up along the typical fault
lines (i.e., assuming that relational ontologists will accept an abundant theory of
universals, etc.):

61
See Chisholm (1985: 157).
 Character of Familiar Concretes Particulars

No Substantive, Explanatory A Substantive, Explanatory

Account Required Account Required
(Quine, Lewis)

Mixed Strategy Relational Strategy

(Tropes are constituents that
instantiate trope-types)
Constituent Strategy
Ultimate Sources of Ultimate Sources
of Character Something of Character Include
Other than Universals Universals
Ultimate Sources of Ultimate Sources of (Speusippus, middle Plato (?)) (Platonic, Abundant Theory)
Character Something Other Character are Universals
Than Universals (Aristotelian, Sparse Theory)

Non-property-like Property-like
(Greek materialists) (trope theorists)
Accept Principle Reject Principle
Accept Principle Reject Principle of Instantiation of Instantiation
of Instantiation of Insantiation (middle Plato (?))
(Castañeda)

Universals are Necessary Universals are Every Form of Character Some Forms of Character
Beings Contingent Beings Exemplified in Some Possible Exemplified No Possible World
(Aristotle) (Armstrong, Mellor) World (early Chisholm) (Plantinga, later Chisholm, van
Inwagen)
  . 

Acknowledgments
This paper is dedicated to Peter van Inwagen and Lizette Bolduc; to Peter, for his contributions
to Philosophy; to Lizette, for making those contributions possible; and to both of them for the
warm and loyal friendship they have shown Dian and me over the past fifteen years.

References
Armstrong, D. (1978a), Nominalism and Realism: Universals and Scientific Realism (vol. I)
(Cambridge University Press).
Armstrong, D. (1978b), A Theory of Universals: Universals and Scientific Realism (vol. II)
(Cambridge University Press).
Armstrong, D. (1980), ‘Against Ostrich Nominalism’, in Pacific Philosophical Quarterly 61,
440–9.
Armstrong, D. (1989), On Universals (Westview).
Armstrong, D. (1997), A World of States of Affairs (Cambridge University Press).
Baylis, C. (1951), ‘Universals, Communicable Knowledge, and Metaphysics’, in Journal of
Philosophy 48; in Loux (1976).
Bergmann, G. (1967), Realism (University of Wisconsin Press).
Castañeda, H. (1974), ‘Thinking and the Structure of the World’, in Philosophia 4, 3–40.
Chisholm, R. (1976), Person and Object (Open Court).
Chisholm, R. (1985), ‘A Logical Characterization of the Psychological’, in Philosophy of Mind
and Philosophy of Psychology: Proceedings of the 9th International Wittgenstein Symposium
(Wien: Hölder-Pichler-Tempsky), 156–61.
Chisholm, R. (1989a), On Metaphysics (University of Minnesota Press).
Chisholm, R. (1989b), ‘Properties and States of Affairs Intentionally Considered’, in Chisholm
(1989a).
Donagan, A. (1963), ‘Universals and Metaphysical Realism’, in Monist; 47, 211–46; in Loux
(1976).
Frege, G. (1892), ‘Concept and Object’, in Geach and Black (1980), 427–9.
Geach, P., and Black, M. (1980), Translations from the Philosophical Writings of Gottlob Frege,
(3rd edn.) (Blackwell).
Hochberg, H. (1960), ‘Universals, Particulars, and Predication’, in Review of Metaphysics 14,
87–102.
Jackson, F. (1977), ‘Statements about Universals’, in Mind 86, 427–9.
Lewis, D. (1983), ‘New Work for a Theory of Universals’, in Australasian Journal of Philosophy
61, 25–46.
Lewis, D. (1986a), ‘Against Structural Universals’, in Australasian Journal of Philosophy 64.
Lewis, D. (1986b), On the Plurality of Worlds (Blackwell).
Lewis, D. (1991), Parts of Classes (Blackwell).
Loux, M. (1976), Universals and Particulars (2nd edn.) (University of Notre Dame Press).
Loux, M. (1978), Substance and Attribute (Reidel).
Loux, M. (2006a), Metaphysics (3rd edn.) (Routledge).
Loux, M. (2006b), ‘Aristotle’s Constitutent Ontology,’ in Oxford Studies in Metaphysics 2,
207–49.
MacBride, F. (2005), ‘The Universal-Particular Distinction: A Dogma of Metaphysics?’, in
Mind 114, 565–614.
Mellor, D. (1991a), Matters of Metaphysics (Cambridge University Press).
Mellor, D. (1991b), ‘Properties and Predicates’, in Mellor (1991a), 170–82.
Mellor, D., and Oliver, A. (1997), Properties (Oxford University Press).
    

Oliver, A. (1996), ‘The Metaphysics of Properties’, in Mind 105, 1–80.

Plantinga, A. (1974), The Nature of Necessity (Oxford University Press).
Prior, A. (1949), ‘Determinables, Determinates, and Determinants I and II’, in Mind 58.
Putnam, H. (1970), ‘On Properties’, in Rescher (1970).
Quine, W. (1948), ‘On What There Is’, in Review of Metaphysics 2; in Quine (1953), 1–19; Loux
(1978), and Mellor and Oliver (1997), 74–88.
Quine, W. (1953), From a Logical Point of View (Harvard University Press).
Quine, W. (1960), Word and Object (MIT Press).
Ramsey, F. P. (1925), ‘Universals’, in Mind 34, 401–7.
Rescher, N. (ed.) (1970), Essays in Honor of Carl G. Hempel (Reidel).
Russell, B. (1912), The Problems of Philosophy (Oxford University Press).
Russell, B. (1940), Inquiry into Meaning and Truth (Allen and Unwin).
Searle, J. (1959), ‘On Determinables and Resemblance’, in Aristotelian Society Supplementary
Volume 33.
Sellars, W. (1963), ‘Abstract Entities’, in Review of Metaphysics 16, in Loux (1976), 156–205.
Sellars, W. (1967), Science and Metaphysics (Routledge).
Shoemaker, S. (1980), ‘Causality and Properties’, in van Inwagen (1980).
Strawson, P. (1959), Individuals (Methuen).
Strawson, P., and Grice, H. P. (1957), ‘In Defense of a Dogma’, in Philosophical Review 65,
141–58.
Van Inwagen, P. (1980), Time and Cause (Reidel).
Van Inwagen, P. (2006a), ‘A Theory of Properties’, in Oxford Studies in Metaphysics 1, 107–38.
Van Inwagen, P. (2006b), ‘Can Mereological Sums Change Their Parts?’, in The Journal of
Philosophy 103, 614–30.
Van Inwagen, P. (2011), ‘Relational vs. Constituent Ontologies’, in Philosophical Perspectives
25, 389–406.
Van Inwagen, P. (2012), ‘What is an Ontological Category?’, in L. Novák, D. Novotný,
P. Sousedík, and D. Svoboda (eds.) Metaphysics: Aristotelian, Scholastic, Analytic (Ontos
Verlag).
Williams, D. (1953), ‘The Elements of Being: I’, in Review of Metaphysics 7, 3–18.
Wolterstorff, N. (1970), ‘Bergmann’s Constituent Ontology’, in Noûs 4, 109–34.
Wolterstorff, N. (1973), On Universals (University of Chicago Press).
Wolterstorff, N. (1991), ‘Divine Simplicity’, in Philosophical Perspectives 5, 531–52.
 2
A One Category Ontology
L. A. Paul

An ontology is defined by its fundamental categories. Following Peter van Inwagen

(2011), take divisions between fundamental categories to mark “real divisions
between things” that determine the basic categorical structure of the world. Speaking
metaphorically, the fundamental categorical structure of the world carves the world
at its fundamental joints.
The project of determining the fundamental categorical structure of the world
descends (arguably) from Aristotle, who started by dividing the realm of being into at
least two fundamental categories: the fundamental category of particulars, or the
present-in, and the fundamental category of universals, or the said-of. This gives the
world a certain sort of structure, built from things with two different natures, things
with the nature of being a bearer of properties or being a particular, and things with
the nature of being borne by particulars. This supports a two category ontology. The
idea is that the world exhibits this fundamental distinction between natures, a
distinction between individuals or bearers and the properties and relations these
things bear. The notion of understanding the ontological structure of the world in
terms of divisions between natures is also deeply embedded in the Lewisean notion of
naturalness, and is extended in Theodore Sider’s (2012) view that the world has
fundamental quantificational structure.1
I defend a one category ontology: an ontology that denies that we need more than
one fundamental category to support the ontological structure of the world. Cat-
egorical fundamentality is understood in terms of the metaphysically prior, as that in
which everything else in the world consists. One category ontologies are deeply
appealing, because their ontological simplicity gives them an unmatched elegance
and spareness. I’m a fan of a one category ontology that collapses the distinction
between particular and property, replacing it with a single fundamental category of
intrinsic characters or qualities. We may describe the qualities as qualitative characters

1
I take Sider’s (2012) view to be, in part, the view that the fundamental quantificational structure of the
world is such that there is only one category of being, as opposed to, say, the view that there are many ways
of existing and hence many categories of being. The fact that there is only one most natural interpretation
of the quantifier is the fact that there is only one fundamental way to exist, only one realm of being, and no
categorical distinctions corresponding to different ways of being. See McDaniel (2009) and Turner (2010)
for an explication of how to understand the connection between quantifier variance and ontological
pluralism.
     

or as modes, perhaps on the model of Aristotelian qualitative (nonsubstantial) kinds,

and I will use the term “properties” interchangeably with “qualities”. The qualities are
repeatable and reasonably sparse, although, as I discuss in section 2.6, there are
empirical reasons that may suggest, depending on one’s preferred fundamental
physical theory, that they include irreducibly intensive qualities. There are no unin-
stantiated qualities. I also assume that the fundamental qualitative natures are intrin-
sic, although physics may ultimately suggest that some of them are extrinsic.
On my view, matter, concrete objects, abstract objects, and perhaps even spacetime
are constructed from mereological fusions of qualities, so the world is simply a vast
mixture of qualities, including polyadic properties (i.e., relations).2 This means that
everything there is, including concrete objects like persons or stars, is a quality, a
qualitative fusion, or a portion of the extended qualitative fusion that is the world-
whole. I call my view mereological bundle theory.
In conjunction with offering my view as an appealing theory of what is funda-
mental, I shall argue that the distinctions made by the neo-Aristotelian polycatego-
rical theorist are metaphysically unfounded—there are no reasons to endorse them
apart from philosophical habit and prejudice. The assumption that there is a funda-
mental categorical distinction is an artifact of an outdated neo-Aristotelian worldview
according to which objects are substantial, chunky, extended, and complete, while
properties are abstract, unextended, and incomplete. We don’t need a fundamental
categorical division between particulars, individuals, or spacetime regions and their
properties, nor do we need a fundamental categorical division between things,
individuals, or bearers and the qualities “borne” by them. Polycategorical ontologies
that postulate multiple fundamental categories assign excess structure to the beast of
reality, making a mess of the carving. The one category ontologist hews the beast at its
ontological joints.
Mereological bundle theory is not just parsimonious in terms of the fundamental
categorical structure it assigns to the world. It is also parsimonious in the way it
builds the world from the constituents of the categories, as it builds the world from
one simple kind of relation: composition. Because it rejects the need for a funda-
mental division between object and property, it rejects any need for a primitive
connecting relation of “instantiation.” And instead of bringing in new primitive
bundling relations such as relations of “compresence” or “co-instantiation,” mereo-
logical bundle theory takes bundling to be a form of mereological composition,
developed using our antecedent notions of parthood (see Paul (2002, 2006a, and
2012) for extended treatments of these ideas).
The one category ontologist may grant, of course, that there exists additional
structure. She might hold that this further structure is merely ideological, that is, it
is a sort of non-ontological structure that is based on primitive expressions and
demarcations we make in our descriptions of the world.3 Or, she might grant that
there is a kind of derivative ontological structure, such as a mereological structure
where sums are constructed from simples. What matters is that none of this

2
I accept relationalism about spacetime, on broadly empirical grounds.
3
See Sider (2012) for discussion of adding to one’s ideology instead of one’s ontology.
 . . 

additional structure, whether ideological or ontological, involves endorsing the

existence of additional fundamental (or irreducibly emergent) categories in the
world.4 The appeal of a one category ontology is not that there is no structure in
the world, but rather that there is no fundamental division between real natures in the
world, and no new real natures are emergent in the world.
While my view is revisionary in some ways, it preserves many features of our
ordinary perspective on the world. It can take objects to be located in spacetime (by
being qualitative fusions that are fused with spatiotemporal relations or relational
properties), and can assume that there are certain primitive modal constraints on the
mixture that is the world, where such constraints are best thought of as de dicto
modal truths of the system. These de dicto modal truths endow the world with
various desirable features: they make it the case that composition is restricted in
certain ways, that the laws of thermodynamics are true, and so on. My qualitative
story fits well with the picture of the world given to us by quantum mechanics,
especially quantum field theory. It also fits well with the thought that properties,
especially located properties or “finegrained events,” are what are subsumed by causal
laws and are the best candidates for the causal relata (see Paul (2000) and Paul and
Hall (2013) for more on finegrained events as the causal relata).

2.1 Ontological Category Theory

Polycategorial ontologies distinguish the object from its properties. The intuitive
basis often given for the distinction is that objects are chunky, substantial, or concrete
things, whereas properties are somehow thinner and less substantial, or as some
would describe it, “abstract.” This neo-Aristotelian distinction is supposed to track a
real difference between what objects are and what properties are. Traditional sug-
gestions for the fundamental ontological categories that make such distinctions
include ontological categories for substances, substrata, particulars, universals, prop-
erties, objects, or spatiotemporal regions, and these suggestions have received renewed
attention in the work of philosophers like David Armstrong (2010), E. J. Lowe (2006),
and Peter van Inwagen (2011).
Substance theory takes objects to be, fundamentally, primitively unanalyzable or
irreducible substances of different sorts, and holds that substances have properties by
standing in some sort of relation to universals or other entities. As such, it is a
relational two category ontology. According to substance theory, what exists are
concrete objects (in one category) and abstracta like properties (in the other cat-
egory), and it is relational in that the concrete objects have their properties by bearing
relations to certain abstracta.5
Substratum theory also distinguishes objects from properties, taking objects to be
constructed from primitively individuated substrates and properties. As such, it is a
two category constitutive ontology where concrete objects have entities from the
property category and from the substratum category as constituents. Concrete

4
See van Inwagen (2011) for the division between primary (or fundamental) categories and secondary,
tertiary, etc. categories.
5
Van Inwagen (2011) provides an excellent account and justification of such a view.
     

objects are constructed from substrata and abstract objects like universals, or abstract
particulars such as tropes, so that the substratum has the properties. It is a constituent
ontology because objects in the world have these things as constituents.6
D. M. Armstrong (1989, 1997, 2010) and C. B. Martin (1980) endorse substratum
theories, and take properties to be ways objects are.
Both the substratum theory and the substance theory emphasize the differences
between particulars or particular things like substances and substrata, and properties.
As a trope theorist, Martin’s emphasis is on the difference between bearers and things
borne as opposed to an emphasis on a difference based purely on particularity.
Martin argues that tropes are co-dependent and need to be combined with a bearer
to “complete” the object (Martin 1980: 7–8).
There are polycategorical views that are related to these approaches but defy any
straightforward categorization, such as Kit Fine’s (1999) hylomorphic theory of
embodiment. Like substance theory, it takes its cue from Aristotle’s thought that
objects have a formal and a material element, but like substratum theory, objects have
constituents and structure. On Fine’s view, objects are constructed from an inten-
sional or conceptual relation R and material parts a and b. R has a certain structural
“form” that is “embodied” in matter such that aRb for material constituents a and b.
As described, Fine’s view is a two category constituent ontology. E. J. Lowe (2006)
defends a four category ontology where the world is constructed from substances,
substantial kinds, modes, and nonsubstantial universals and relations of instanti-
ation, characterization, and exemplification.
A somewhat different, but very popular, way to endorse a two category distinction
involves making a distinction between spacetime, in terms of points, regions, or
spatiotemporal substances of some sort, and n–adic properties (for example, Sider
2008). An ontology which takes spatiotemporal regions to be primitively individu-
ated substances instantiating properties, in line with relational ontologies, or takes
spatiotemporal regions to be substrata bearing properties, in line with constituent
ontologies, is a two category ontology. Many “supersubstanvialists” are, perhaps
implicitly, committed to a two category ontology of some kind. Again, the one
category ontologist rejects such theories because they assign excess structure to the
world. For the monocategorical theorist, spacetime consists of spatiotemporal rela-
tions, or properties, or of relational properties, suitably defined. It does not consist of
regions that are different in nature from n-adic properties and serve as their
“bearers.” Spacetime is not itself a distinct fundamental category, nor is it constructed
from substantial entities of any Aristotelian non-qualitative kinds. The one category
theorist holds that reality needs no such division between spatiotemporal regions and
nonspatiotemporal qualities.
Traditional bundle theory, such as the trope bundle theory of D. C. Williams
(1953) or Peter Simons (1994), or Russellian bundle theory, which takes objects to be
bundles of universals, is a constituent ontology that takes properties to be constitu-
ents of objects. As such, traditional bundle theory makes an ontological distinction

6
See Wolterstorff (1970) for the distinction between relational and constituent ontologies.
 . . 

between the bundle and its constituents.7 This distinction is not always as sharp as
that made by defenders of other constituent and relational ontologies. Traditional
bundle theory counts as a one category ontology in the sense that it constructs the
members of the object category from the members of the property category.8 But it
distinguishes between concrete objects and their abstract constituents, the properties
that are bundled together to make the object, and the distinction builds on the same
tradition of distinguishing between bearers of properties and the properties them-
selves that two category ontologies rely on.
This distinction is most obvious with Russellian bundle theories involving tran-
scendent universals, since such universals are not located in spacetime, while the
objects constructed from them are. (For reasons like this, van Inwagen (2011) argues
that these sorts of bundle theories are really two category ontologies.) But even if the
universals are immanent, they are still thought to be abstract in the sense of abstract
that picks out “less substantial.”
This sense of “abstract” (arguably derived from “abstraction”) is often the basis for
distinctions between properties and objects made by trope theorists. (We also see this
in Martin’s substratum theory.) Peter Simons is a traditional bundle theorist who
distinguishes the natures of tropes from the natures of objects by marking tropes off
as co-dependent entities that cannot exist by themselves: they need to be instantiated
with other tropes. Objects, on the other hand, are independent entities that can exist
without other objects (Simons 1994). Simon’s distinction has a deep historical basis,
and connects with the Lockean distinction between the object as the bearer and the
property as what is borne. G. F. Stout (1921) labels tropes as “abstract particulars,”
which also suggests this distinction.
Williams distinguishes between the concrete nature of objects and the abstract
nature of tropes in terms of ontological completeness: “the more special sort of
incompleteness which pertains to what we have called the ‘thin’ or ‘fine’ or ‘diffuse’
sort of constituent, like the color or shape of our lollipop, [is] in contrast with the
‘thick,’ ‘gross,’ or chunky sort of constituent, like the stick in it” (1953: 15). What is
explicit in Simons and in Russellian bundle theory, and implicit in Williams, is that
while we only need one category of thing, dependent tropes, to start with, we
somehow build a new kind of thing, independent objects, from the members of
this category. These theories in effect smuggle in two fundamental ontological
categories—or perhaps two irreducible ontological categories, the fundamental cat-
egory of properties with an emergent, irreducible category of objects—in an effort to
preserve a distinction between objects and properties.
One sort of one category ontology that competes with mereological bundle theory
in terms of parsimony is austere nominalism. Austere nominalism denies that there
are any such things as properties at all. There are only objects, taken to be concrete
particulars such as individual salt shakers, cats, and rocks (Cf. Loux 1998). These
concrete particulars occupy spatiotemporal locations and display features we expect

7
To their credit, Keith Campbell (1990) and John Bacon (1995) do not seem to make these sharp
ontological distinctions between object and property (although they make problematic categorical distinc-
tions elsewhere), and they are also clear that their bundling relations are non-mereological.
8
But see van Inwagen (2011) for an argument that this construction is unsuccessful.
     

from such objects, for example, if they are material objects they are impenetrable, and
more than one object of the very same kind cannot occupy the very same spatio-
temporal location.9 “The austere nominalist insists that at rock bottom there are only
structureless concrete particulars and that talk apparently about properties and the
like is simply disguised talk about the fundamental concrete particulars” (Loux 1998:
103) Strictly speaking, nothing has properties, although we may speak as though
objects really do have them.10
Austere nominalism is unacceptable, for it is insufficiently finegrained. Distinc-
tions between objects fail to capture all the distinctions we need to make. If every-
thing with a heart also has kidneys, or if everything human is also a featherless biped,
how are we to capture the distinction between having a heart and having kidneys
without endorsing radical views about the role of possibilia in making qualitative
distinctions within actual existents? Moreover, it suffers from empirical problems.
First, a theory of entities in the world as constructed from ontologically basic
concrete particulars, i.e., localized particles of some sort, is contradicted by standard
interpretations of quantum field theory (Cf. D. B. Malament 1996). Second, as
I describe in section 2.6, taking fundamental entities such as bosons to be anything
like “concrete particulars” is inconsistent with Received Views in quantum mechan-
ics. While parsimony is a virtue, it is unclear to me how to make austere nominalism
consistent with fundamental physics, and also unclear how to make it sufficiently
accommodating of the qualitative differences we find in our world.
A less austere version of nominalism, clement nominalism, takes the view that
everything there is is an object, and the fundamental objects are tropes: qualitative
located concrete particulars. Clement nominalism is a more interesting one category
view than austere nominalism. To accommodate the physics, clement nominalism
would need to dispense with its nominalistic overtones by rejecting any metaphys-
ically thick sense of “concrete particular” where such particulars are basic, located,
particle-like entities, making it unclear what sense of “object” the view relies upon. At
best, it could be thought of as a view where the fundamental entities were tropes.
However, for it to be a truly one category view, ultimately it would need to collapse
into something very like the versions of mereological bundle theory I describe in
section 2.4.
Other sorts of one category views are surely possible, and would have to be
evaluated on a case-by-case basis. My aim here is simply to question the current
dogma supporting the range of fundamental polycategorical structures put forward
by neo-Aristotelians, substance theorists, fans of propertied and related spacetimes,
and others, by defending a more appealing and parsimonious monocategorical
alternative.

9
Fine (2000a) argues that there are certain sorts of counterexamples to the principle that no more than
one kind of object can occupy a particular place.
10
I’m not quite sure how electrons and other fundamental particles get classed as “concrete particulars,”
since they aren’t really very concrete, but I think the austere nominalist usually wants to include them in his
ontology.
 . . 

2.2 A Qualitative Mereology

I take the derivative ontological structure of the world, the structure built from the
basic constituents of the world, to be mereological structure. Mereological structure is
based on relationships between parts and wholes. Such structure is not categorical:
sums of properties do not create new natures or real categories. (In this sense,
composition is like identity.) I take composition to be the basic building relation of
the world, and the individuals that are the basic parts are used to construct everything
else there is. What sorts of individuals are the fundamental constituents of the world,
the metaphysically prior simples that are fused to create the world-whole? This is the
delicate question. In my view, the fundamental constituents are properties, or
qualitative natures, and all else is mereologically composed from these.
So the world-whole is built by fusions of qualities. I shall take the basic notion of my
mereology to be the primitive notion of “proper part,” and assume that proper
parthood is analytically irreflexive, asymmetric, and transitive. With these notions,
along with a principle of supplementation and what I take to be uncontroversial
presuppositions about identity and existence, I capture the meaning of “part” with my
account of qualitative parts and define qualitative composition (Cf. Simons 1987).11
Hence I develop my qualitative mereology by starting with thin notions of parthood
and composition, ones which are perfectly well-defined mereologically and are also the
basis for classical extensional mereology. Of course, classical extensional mereology
takes parts and wholes to be spatiotemporal parts and wholes, where parts are individ-
uals that are—or are defined in terms of occupying—four-dimensional regions of
spacetime. But we can apply the basic notion of parthood to other sorts of constituents,
and define composition as a relation between these sorts of constituents, just as well as
we can develop these notions so that they apply to spatiotemporal regions.
My qualitative mereology is the basis for my mereological bundle theory: proper-
ties are literally objects and parts of objects, and properties are bundled using the
composition relation. Assuming an appropriate first-order predicate calculus with
identity, here are the basic axioms and definitions of my qualitative mereology
M (“qualitative parts” are property parts).
A1. For any x, x is not a proper qualitative part of itself. (Proper qualitative
parthood is irreflexive.)
A2. For all x and y, if x is a proper qualitative part of y, y is not a proper qualitative
part of x. (Proper qualitative parthood is asymmetric.)
A3. For all x, y, and z, if x is a proper qualitative part of y and y is a proper
qualitative part of z, x is a proper qualitative part of z. (Proper qualitative parthood
is transitive.)
A4. For all x and y, if x is a proper qualitative part of y, there is a z such that z is a
proper qualitative part of y, and z is qualitatively disjoint from x. (This is weak

11
Thus, I agree with Simons’s insightful comment: “If this is all there essentially is to the part-relation,
why can stronger principles sometimes apply? The answer lies not in the part-relation itself but in the
nature of the objects to which it applies” (Simons 1987: 363).
     

supplementation: if an individual has a proper qualitative part, it has at least one

other proper qualitative part.)
D1. For all x and y, x is a qualitative part of y iff x is a proper qualitative part of y or
x is identical to y. (An object’s improper qualitative part is just itself.)
D2: For all x and y, x qualitatively overlaps y iff x and y have a qualitative part in
common.
D3: For all x and y, x is qualitatively disjoint from y iff x and y have no qualitative
part in common.
D4: For all x and y, x is qualitatively composed of ys (or x is a qualitative fusion of
ys) iff x has all the ys as qualitative parts and has no qualitative part that is
qualitatively disjoint from each of the ys.12

Qualitative composition is neither covertly nor overtly spatiotemporal, nor is it

somehow tied to spatiotemporal location or occupation. Like many fans of mere-
ology, I take composition to be restricted, and I recognize the serious problems
associated with adequately determining the conditions under which composition
occurs. Hence I endorse a brute restriction and correspondingly reject a general
qualitative fusion axiom.13,14
I have described my properties as “qualitative natures,” and taken them to be a
kind of repeatable universal, perhaps akin to Aristotle’s nonsubstantial forms. Prop-
erties are located in virtue of being qualitatively fused to spatiotemporal relations or
relational properties. There are no uninstantiated universals. Not just any predicate
defines a property, so properties are sparse, and there are no negative properties,
merely negative predicates (if an object is ~F then it does not include F as a part).
Properties can be monadic or polyadic (a.k.a., relations).
My property mereology allows fundamental relations, if there are any, to fuse with
other properties in just the same way as monadic properties fuse with other proper-
ties. The fundamental relations have what we can metaphorically describe as “ends”
that fuse to n-adic properties. Now, there might not be any fundamental asymmetric
relations. If not, M could be made extensional (replacing the axiom of weak supple-
mentation with something stronger to give extensionality). If there are fundamental
asymmetric external relations, I take such relations to be relations with a certain sort
of intrinsic character: character that influences the structure of a fusion that includes
them. Metaphorically speaking, we can understand this as the view that the asym-
metric external relation has places, and which of these places other properties and

12
I suspect that qualitative parts are the only parts there are, but I include the designation here for
clarity. Note that haecceitistic and other impure properties can still be qualitative parts as I am using
“qualitative” and that the fusion relation is the composition relation.
13
Cf. Markosian (1998).
14
Note that since qualitative fusion may be restricted we have the resources to make sense of cases
where proper qualitative parts P, Q, and R are qualitatively fused together but there is no fusion of P and R,
and so no object that includes P and R. Imagine an object O that includes red, round, and squashed in its
fusion. Is there an object that is simply round and squashed? If so, then we grant that there can exist
incomplete objects, perhaps as long as such objects are part of a complete object. If not, then this is an
instance of restricted fusion.
 . . 

relations are fused with determines the overall character of the fusion that includes
the asymmetric relation. Less metaphorically speaking, the asymmetric external
relation has an intrinsic direction such that when it is fused to other properties, the
resulting fusion has a certain sort of structure. When asymmetric fundamental
external relation R is fused with properties P and Q, R is such that the fusion of
PRQ is different from the fusion of QRP. On this view, asymmetric external relations
provide fusions with structure via the mereological composition of properties with
relations that have places, so qualitative composition is not extensional. We might
describe the result as “neopredicational” fusing.15 For example, perhaps there is a
fundamental temporal relation of direction. If so, then the world will include an
asymmetric temporal structuring relation R, such that the fusion of PRQ has an
intrinsic direction because it includes the intrinsic character of R. If so, then the
fusion of QRP has a different intrinsic direction, even though it has the very same
proper parts. To mark such a difference, we may define primitive predicates D1 and
D2 that apply to PRQ and QRP, respectively.

2.3 Parthood and Constructing the World:

Traditional Bundle Theories
My denial of any distinction between properties and objects is drawn out by the way
that mereological bundle theory, when developed, differs from other bundle theories,
including classical trope bundle theories like that of Williams. The central difference
between my mereological bundle theory and traditional bundle theories is captured by
the notion of qualitative parthood that my view employs. Traditional bundle theorists
do not take properties or their instances to literally be parts of objects, while they do
take objects to have spatiotemporal parts. (They take properties to be constituents of
bundles, but not literally parts in the mereological sense.) This is made reasonably
clear simply by the fact that, instead of using qualitative composition as the bundling
relation, they use primitive relations like compresence or co-location, whose names
have overtly spatiotemporal overtones, to bundle objects.16 This difference costs them
in terms of parsimony, for such bundle theorists must introduce entirely new primi-
tive relations (like consubstantiation) to bundle properties and build the world, in
contrast to the mereological bundle theorist, who uses mereology to build the world,
and thus takes advantage of our intuitive grasp of parthood and composition as basic
building-relations. Moreover, as we shall see, the traditional bundle theorist’s way of
distinguishing between properties and parts leads to categorical difficulties.
Keith Campbell (1990), who is the clearest about his commitment to tropes as
entities that are ontologically concrete and independent (he specifies that tropes are
only abstract in an epistemic sense), is also clear about how tropes are bundled using
compresence, which is not mereological composition. One traditional bundle theorist,
Williams, has described the constituents of the bundle as “parts” of the bundle. But the
use of the term “part” here seems to be merely metaphorical, non-mereological, or

15
Related issues are taken up in Fine (2000b).
16
I recently discovered one exception to this rule: McDaniel (2001).
     

non-referentially descriptive in some other obscure sense. Or at least, if Williams and

other traditional bundle theorists do mean that properties are literally parts of objects,
their views are radically incomplete, perhaps to the point of absurdity.
It is worth developing this point, for exploring the distinction between traditional
bundle theory and mereological bundle theory will help to bring out the contours of
the latter. That traditional bundle theorists cannot literally be taking properties to be
parts of objects seems obvious when we recognize that, for a bundle theory to take
properties to be parts of objects, it must describe the basics of its property mereology
and explicitly fit it together with a spatiotemporal mereology. There is no such
description and no such fitting offered anywhere in the literature about traditional
bundle theory. Without an account of the basics of the property mereology and of the
fit between the qualitative and the spatiotemporal, the traditional bundle theorist
cannot make sense of how an object can have both property parts and spatiotemporal
parts, and so she cannot hope to answer obvious and damning objections.
In particular, parthood is transitive, and, I’d say, analytically so. But how could it
be so for those who take properties to be parts along with taking more material,
chunky concrete objects as spatiotemporal parts? I have my hand as a spatiotemporal
part, and my hand has being hand-shaped as a property part. If parthood is transitive,
it would seem that I have being hand-shaped as a part. But I am not hand-shaped,
and hence I do not have it as a part. How is the traditional bundle theorist to make
sense of this? Is the transitivity of parthood to be denied?
Or consider the way some philosophers think of ordinary material objects as
composed of smaller spatiotemporal parts. If objects are composed of smaller
spatiotemporal parts, how are they also composed of property parts? How can an
object like a chair be literally composed from tropes such as having shape s, having
mass m, having color b, etc., while also being composed of spatiotemporal parts
involving glue and pieces of wood and leather? The picture of objects as fusions of
properties makes no sense unless a relationship between property fusion and spatio-
temporal fusion has been clearly developed.
There is a further issue regarding composition that the traditional bundle theorist
has to take a stand on: if properties can be parts, do properties compose in all
circumstances? Is composition universal, as it is in classical extensional mereology?
If the traditional bundle theorist endorses universal composition, then she endorses
the existence of, say, fusions of the properties of being square and being round. Not a
good result.
If properties are bundled together using some non-mereological relation, we do
not need to develop the relationship between a property mereology and classical
extensional mereology or do the difficult work needed to define property compos-
ition, and so we can at least dodge the problems with transitivity. (That said, the
traditional bundle theorist should still explain just how bundling properties to build
objects is to be combined with building objects from spatiotemporal parts.) The idea
behind bundling is usually that, whatever the bundling relation is, e.g., compresence,
it is understood primitively, not as a relation developed using our antecedent notions
of parthood and other mereological ideas. Peter Simons sees this clearly, and as a
result argues that “Parts is one thing, properties another (and properties of parts
something else again)” (1994: 563).
 . . 

It isn’t really a surprise that traditional bundle theory does not blend property
parts together with spatiotemporal parts, for, as I noted above, traditional bundle
theories standardly retain a distinction between properties and spatiotemporal parts
based on the intuitive idea that properties exist in a way that spatiotemporal parts—
which are concrete objects—do not. Sometimes this is fleshed out as the idea that
properties are somehow less substantial or must co-occur with other properties,
while spatiotemporal parts, a species of concrete or material object, do not. “What
we standardly call ‘parts’ are a special kind, independent parts or pieces . . . ” (Simons
1994: 563). Williams also contrasts the way tropes are “fine” or “diffuse” parts with
material or spatiotemporal parthood. “To borrow now an old but pretty appropriate
term, a gross part, like the stick [of a lollipop], is ‘concrete,’ as the whole lollipop is,
while a fine or diffuse part, like the color component or shape component, is
‘abstract’ ” (Williams 1953: 6). In short, the ontological distinction between proper-
ties and objects is perpetrated by treating properties as entities or metaphysical
“parts” that are somehow different from the sorts of material objects that can be
spatiotemporal parts (proper or improper).
The thought seems to be that property parts are different in kind from chunky,
material-object-like spatiotemporal parts because properties are different from “con-
crete” objects like spatiotemporal parts, and so we must think of them and treat them
separately as some sort of non-mereological or metaphysical “part.” But this is
wrong. It is just a mistake to think properties cannot be chunky, concrete, complete,
or independent. They are chunky, and concrete, and complete, and independent—
because some of them are chunky, concrete, complete, and independent. In particular,
some of the properties that are ordinary objects are chunky, concrete, complete, and
independent. (As are some of the fundamental physical properties, such as field
intensities. See section 2.4.)
Do not be tempted by the fallacious idea that fusing is what somehow “makes” the
ordinary object (which is a fusion of properties) chunky or substantial. That’s not how
fusing works: it makes many into one, it doesn’t make non-substances into substances
or abstract things into concrete ones. And do not be tempted to argue that we know
that fusing properties makes the objects chunky or substantial on the grounds that we
can somehow see or detect that objects are chunky or substantial while detecting that
properties are not. For we cannot see or detect that fusing makes the properties into
chunky, substantial entities any more than we could see or detect that fusing a bunch
of appropriately arranged spatiotemporal parts makes a whole. There is no phenom-
enological difference between the fused and the un-fused.17 You can’t see or detect a
difference between a plurality of property instances and the fusion of them any more
than you could see or detect a difference between a plurality of spatiotemporal simples
(or any plurality of spatiotemporal parts) and the fusion of them.
Let me now turn to a final argument: that properties are somehow dependent in
a way that objects are not because of the fact that some properties, at least contin-
gently, always co-occur with others. The idea seems to be that properties and objects

17
Or at least, no obvious difference. I have to admit that I’m kind of partial to being a dogmatist, Pryor-
style, about our knowledge of composition. But that’s a different discussion.
     

have different natures: objects are independent objects while properties are
dependent objects. The view that properties are objects does not dispute the fact of
co-occurrence. It is true that some properties in the world are always fused with other
properties: as a matter of (contingent?) fact, some properties co-occur. But the fact
that there are certain modal relations in the world that ontologically entail that
certain properties are always fused (at least in the actual world, and maybe in all
possible worlds) does not mean that these properties are somehow ontically
dependent. It just means that there are certain facts about the universe that result
in certain connections: for example, that anything with mass also has extension.
Perhaps these connections are simply contingent facts about how the qualitative
profiles of objects are to be completed that are derived in some way from the physical
laws of the world. (I like this idea.)18 Or perhaps they derive from external de re
modal relations of necessity. In any case, they need not be explained by some sort of
internal or intrinsic ontological characteristic of “dependence” that we must ascribe
to the properties themselves. The flip side of this point answers the worry about how
to restrict composition. If there are deep modal facts about co-occurrence, surely
there are also deep modal facts about what cannot co-occur. The mereological bundle
theorist can thus deny universal composition to avoid the worry about the existence
of round squares and the like.
Recall that the traditional bundle theorist who describes properties as parts (really,
mostly Williams, as far as I can tell—others with developed bundle theories usually
eschew this language) avoids any clear statement about how a property mereology is
fitted to spatiotemporal mereology. If the claim about objects having property parts is
merely a metaphor, the traditional bundle theorist dodges the worries about transi-
tivity and avoids having to make sense of how, given qualitative parthood, extended
objects can be exhaustively constructed from smaller material parts.
But as my discussion above suggests, such a dodge deepens the divide between
property and object, endangering the purity of the bundle theorist’s claim to have a
truly single-category ontology,19 and undercutting its claim of superiority over
Lockean substratum theories. To the extent that there is an ontological distinction
drawn between the natures of the concrete objects and their concrete parts, and the
abstract particulars or universals that are their qualitative constituents, traditional
bundle theory starts to endorse the neo-Aristotelian’s distinction between properties
and individuals. And, as I have argued, to the extent that one holds that properties are
not literally parts, one holds that properties are not objects, which suggests, if not a
categorical difference between the real natures of these things, something very like
it.20 I suspect that traditional bundle theorists, in a misguided effort to preserve
philosophical intuitions about material objects that date at least from the days of
corpuscular philosophy and Cartesian mechanics, have unwittingly endorsed a
categorical distinction between property and object by creating a new category of
objects that somehow emerges from the bundling of properties.

18
Also see Schaffer (2003).
19
This supports the point made in van Inwagen (draft ms.) but from a different angle.
20
Only a bundle theorist like Campbell, who is the clearest about the concrete nature of tropes and takes
the bundling relation to be a primitive new relation of compresence can escape this charge.
 . . 

So there is no need for a categorical distinction between properties and objects.

Properties are not “abstract” or “dependent” or “incomplete” entities opposed to
“concrete” or “independent” or “complete” objects in any ontological sense at all.
Properties are not incomplete while objects are complete. Properties do not need
bearers to make objects. Properties and objects are the very same thing. We build
fusions of properties from properties, and some of these are what we describe as
“ordinary objects.”
Perhaps the one category ontologist denies our commonsense intuitions and
ordinary ways of speaking as a consequence: if so, what of it? Such commonsensical
intuitions and linguistic conventions are nothing more than a local, parochial effect
of the way we pre-theoretically approach the world, dispensable in favor of the
massive increase in ontological parsimony which a one category approach brings.
We might want to adopt an error theory of ordinary language to handle this fact,
following the approach of Sarah Moss (2012) to theories of persistence, or distinguish
between ordinary English and philosopher’s English, and look to an ideal interpreter
to interpret the relevant sentences of ordinary English in an appropriately philo-
sophical way.21 The translation manual used by the ideal interpreter can take
ordinary object terms to pick out fusions that include properties of having mass,
extension, and other properties that are distinctive of chunkiness or particularity, and
can distinguish, if need be, between different kinds of predication.

2.4 Parthood and Constructing the World:

Mereological Bundle Theory
It seems clear that traditional bundle theorists are unable to navigate the divide
between property parthood and spatiotemporal parthood without slipping into a
categorical crevasse. Can the mereological bundle theorist do any better? Yes.
She should start with spacetime: if she is to endorse a truly one category ontology,
she must start by being explicit that spatiotemporal locations are n-adic properties,
not primitive substances or individuals in a second fundamental category. She can do
this by taking spacetime to be relational, or by taking spacetime points to be
spatiotemporal tropes or trope-like equivalents. If we deny extensionality, we can
even recapture a kind of substantivalism about space where we have the same part, a
spacetime trope, endlessly repeated as a space of “points,” giving us a space with a
global cardinality but without haecceitistic difference.22 The charms of this substan-
tivalist picture notwithstanding, I shall assume spatiotemporal relationalism.
Once we have a suitably property-theoretic notion of spacetime, there are two
models the mereological bundle theorist might use to address the puzzles about
spatiotemporal composition. The first model accommodates most of the mainstream
metaphysical intuitions about spatiotemporal composition by endorsing two differ-
ent composition relations, one for qualitative parts and one for spatiotemporal parts,
and building the world up from quality-points plus spatiotemporal fusions. The
second model builds the world entirely from qualitative parts, and captures a kind

21 22
Jason Turner (2011: 15). Baker (2011), Paul (2013).
     

of holism that is congenial to certain sorts of fundamental physical theories. I’ll

discuss each model in more detail.
The first model for mereological bundle theory starts with properties qualitatively
fused together with locations (understood to be relations or relational properties of
having such-and-such locations) to create a mosaic-like lowest compositional level of
located, unextended qualitative fusions distributed through a network of spatiotem-
poral relations. I’ll call this model the mosaic model. M is the mereology that applies
to this level of composition. Extended objects are then created using a different
composition relation and hence a different mode of mereological construction.
This second composition relation is defined in the following way. Call each (max-
imal) located unextended qualitative fusion a “spatiotemporal part.” Proper spatio-
temporal parthood is defined in the usual way, as asymmetric and transitive, and one
can accept a strong supplementation principle to make the spatiotemporal mereology
extensional. Further axioms and definitions consistent with classical extensional
mereology can be accepted, including unrestricted composition. We might then
call our new composition relation “spatiotemporal composition.”
On this model, spatiotemporal parts and spatiotemporal composition are embed-
ded in a qualitative, one category ontology, and the relation between qualitative
and spatiotemporal composition is clear. The fundamental spatiotemporal parts or
“spatiotemporal simples” are qualitative, located fusions of properties, and larger
spatiotemporal parts are constructed from a relation defined on these spatiotemporal
simples.23 Rocks, persons, stars, and abstract objects are all fusions built from quality-
fusions then fused together by spatiotemporal composition. Such fusions, in addition
to being complex constructions of quality and spatiotemporal fusions, are also plain-
vanilla property fusions, where the properties fused are the whole (distributed) prop-
erties of the object. Take the spatiotemporal fusion of simples s1 and s2, where s1 has
the properties of having a mass of one gram and having a semi-circular shape and s2 has
the property of having a mass of one gram and having a semi-circular shape. When
there is a spatiotemporal fusion of s1 with s2, giving us an object with a mass of two
grams and the shape of a circle, this is also the fusion of the distributed property of
having a mass of two grams with the distributed property of having a circular shape.
Although there are two sorts of compositional structure in the world, the mosaic
theorist fiercely denies that fusing properties together to create located quality
bundles gives us an emergent or otherwise irreducible category of “objects,” or that
the different compositional structures demarcate different fundamental categories in
any way. The world is purely qualitative, and spatiotemporal parts are fusions of
properties (not emergent objects of any sort). We are simply building the world with
n-adic properties, albeit with different sorts of properties at different compositional
“levels.” The loss of parsimony here is a loss of parsimony with respect to the number
of composition relations, since there are two species of composition relation, but not
with fundamental categories, since there is still just one, and we still build the world
with one (generic) kind of relation, composition.

23
The mosaic model also makes very good sense of Goodman (1966), who takes qualitative parts to be
appearances of spatiotemporally located trope-like entities or patches of the overall phenomenological
quilt and builds a mereology of appearances in the spatiotemporal manifold.
 . . 

The problem with transitivity is also circumvented, for although each kind of
parthood is transitive, transitivity does not apply across different kinds of parthood.
Since the kind of parthood involved in qualitative parthood is different from the kind
of parthood involved in spatiotemporal parthood (qualitative parthood is not exten-
sional, and is defined using weak supplementation, while spatiotemporal parthood is
extensional, and is defined using strong supplementation), the worry about transi-
tivity simply fails to apply.
The view has appeal for those who like Lewis-style Humean mosaics, and indulges
our corpuscular intuitions and our attraction to classical-mechanical or particle-based
depictions of the world. If we understand fields in appropriately property-theoretic
terms, the model can even capture Barry Loewer’s (2004) Humean supervenience-
friendly account of the Lewisian mosaic, which Loewer designs in order to circumvent
worries about quantum nonlocality for the fan of mosaic-style views. It’s worth noting
that, understood this way, if we take clement nominalism and swap out primitively
located and individuated trope-instances, replacing them with repeatable qualitative
characters, and we define bundling in suitably mereological terms, clement nominal-
ism collapses into the mosaic model.24
The mereological bundle theorist might further develop the mosaic model. One
development goes fictionalist about spatiotemporal (or, alternatively, configuration-
space) composition. The fictionalist denies the existence of any sort of equivalent of
spatiotemporal composition after the level of the mosaic of located, unextended
qualitative fusions distributed through a network of spatiotemporal relations. On
this sort of (spatiotemporal compositional) fictionalist approach, one might describe
what seems to be a table as “some qualitative fusions-arranged-tablewise.”
A very different way of developing mereological bundle theory, the global model,
denies that strictly speaking, spatiotemporal composition is used to build the world.
Instead, the extended world is wholly and immediately constructed from a fusion of
n-adic properties, including spatiotemporal relations and perhaps a structuring law-
like relation, resulting in a distribution of properties across a spatiotemporal mani-
fold. It is the whole, structured world that results from the original fusion of
fundamental properties and relations. Although we can pick out portions of the
manifold and describe them as “spatiotemporal parts” or imagine them as the
products of the spatiotemporal composition of simples, all of this is merely a useful
fiction. The real parts of the world are the properties and relations that compose the
extended world-whole, and here, parthood is transitive.25
On this sort of fictionalist approach towards spatiotemporal composition, one
might describe what seems to be a spatiotemporal part of a table as “a portion of the
qualitative world-fusion that is distributed table-top-wise.” The fictionalism exactly
parallels that of the compositional nihilist, except instead of taking the phrase “this
table-top” to refer to a certain plurality of unextended simples arranged table-top-
wise, it takes it to refer to a certain region of the world-whole. One might also look to

24
Such a swap might be prohibited for the true nominalist: Loux (1998) defines the nominalist as one
who denies the existence of universals or multiply repeatable qualities.
25
I think this view has interesting connections to the holistic view Dasgupta’s (2009) describes as
“generalism”—although my view is still atomistic in his sense.
     

Horgan and Potrc (2009) for assistance with the semantics here.26 Another alterna-
tive would be to adopt a version of Jonathan Schaffer’s (2010) priority monism for
spatiotemporal parts (not monism in general, since the world is still built from
quality parts), taking spatiotemporal parts to be real but derivative. Here, again, we
have two different kinds of parts, and so transitivity would fail to apply. Such a view
has costs with regard to parsimony, but might be attractive overall: we just need to
remember that classical extensional mereology is either derivative or just a handy toy
model, and that the fundamental ontological basis for reality is a qualitative
mereology.
The global model has more physical plausibility than one might initially think:
consider the wave-function realist who takes the world-whole to be a wave function
(Cf. Ney 2012). On the GRW theory of the world, the world is a universal wave
function that evolves in accordance with the dynamical laws. Understood in terms of
mereological bundle theory, the wave function is the fusion of amplitude and phase
properties (along with any other properties of the system) with structuring properties
or relations, including the structuring relations described by Schrödinger’s equation
and by the collapse postulate. A variant of this view can fit the Everettian approach,
and one can also fit David Albert’s 1996 treatment of Bohmian mechanics by adding
a world-particle that is simply a fusion of properties to the plurality of things.27 For
this reason, I find the global model more appealing than the mosaic model. The
empirical facts about the world, especially given facts about the existence of entangled
states, just don’t seem to support the sort of atomistic world that the mosaic view
describes (although, admittedly, Loewer’s model is consistent with these facts).
However, the jury is still out on what the best fundamental physical theory will be,
and so for some, at least for the moment, the mosaic model retains its appeal.
There is much more to say about a one category ontology of qualitative characters,
but even more importantly, a standard objection must be addressed—one that has
been pressed many times and in many forms, and one which many take to definitely
refute bundle-theoretic ontologies. The objection concerns the way bundle theories
should handle questions concerning the identity and individuation of qualitative
duplicates. It needs to be discussed in detail.

2.5 The Individuation of Qualitative Duplicates

The importance of the distinctions between bundle theories and other theories
becomes clearer once we turn to questions about how objects are to be individuated.

26
Horgan and Portc (2009) defend the view that the world has no proper spatiotemporal parts and
develop a contextual semantics intended to accommodate our ordinary ways of speaking.
27
We can extend the global model in a way that is parallel to the first way we extended our first model of
mereological bundle theory. Instead of holding that there is only a single world-whole, or (as in the first
model) holding that there are many unextended fundamental qualitative fusions of the world, we might
hold that there are some or many extended fundamental qualitative fusions, where such fusions are
arranged as a mere plurality, that is, they do not spatiotemporally compose into a larger whole. If our
world is like this, then the fundamental entities are extended qualitative fusions (perhaps they are the
“spacetime states” of Wallace and Timpson (2009) or successive stages of the “world particle” of Albert
(1996)).
 . . 

The main issue for one category ontologists concerns counterexamples involving the
possibility of qualitative duplicates. Substance theorists and substratum theorists
individuate all objects, including duplicates, based on primitive differences founded
on haecceities: primitive facts about the identity or difference of objects. For this
reason, they can accommodate the possibility of qualitative duplicates.
Substance theorists, if they are willing to individuate at all, take concrete objects to
be primitively individuated in terms of being primitively different substances. One
way of cashing out this primitive difference is in terms of primitively different
instantiation relations, such that objects are different instantiations of their proper
kind in virtue of being primitively different substances that instantiate their sort
property. Another way to cash out the differences between objects is in terms of
primitively different individual essences or thisnesses, such that objects differ in
virtue of having different thisness properties. For example, Adams (1979) distin-
guishes between individuals or particular things and their suchnesses, and takes
particular things to be individuated by their thisnesses. In substance theory, however
thisnesses are to be understood, it is important that thisnesses are not taken to be
constituents of objects, since objects do not have deeper structure.
The method of individuation in substratum theory is similar in some ways to the
method of the substance theorist: objects are individuated in virtue of primitive
entities, in this case, substrata. As Sider points out, the substrata serve to primitively
individuate the objects they are constituents of. “Distinct particulars . . . have distinct
thin particulars, distinct non-universal ‘cores’ ” (2006: 387–8). The main difference
between the method of individuation of substance theory and the method of indi-
viduation of substratum theory consists in whether the individuating entity is a
constituent of the particular it individuates. For substratum theorists the entity is a
constituent of its particular, for substance theorists it isn’t. In both approaches,
however, the individuation is based on haecceitistic differences.
Endorsing haecceities, as I define it, involves endorsing the acceptance of some sort
of in-principle empirically undetectable primitive entity, such as a substratum, or
involves a broad commitment to primitive facts that determine identity and modal
character, as with substance theory. The basis for this endorsement is a commitment
to a version of the Principle of Sufficient Reason for identity claims: a commitment to
the need for a ground for identities and differences. As such, haecceitism holds that
primitive facts are (and explain) the ultimate ontological sources of individuation
and identity.
The way substratum and substance theorists individuate numerically distinct par-
ticulars is important, since advocates of both views claim that their views are superior
to bundle theories on individuative grounds. The claim is that bundle theories cannot
individuate numerically distinct indiscernible objects, and hence cannot accommodate
a range of situations involving symmetries that we take to be pre-theoretically possible.
For this reason, substance and substratum theorists argue that even though bundle
theory may enjoy the advantage of one category parsimony, their theories do a better
job of representing all the metaphysical possibilities.
But why think bundle theories can’t individuate numerically distinct indiscernible
objects? In brief, the objection based on indiscernible objects goes like this: a bundle
theory holds that objects are simply collections of properties. But there could exist
     

numerically distinct objects that are fusions of the same properties, that is, there
could exist numerically distinct but qualitatively indiscernible objects.
For example, consider world W. In W, all that exists are two perfectly homoge-
neous iron spheres that are two feet apart.28 By supposition, the spheres are perfect
duplicates, and they exist for exactly the same amount of time. Moreover, if we
stipulate that the spheres exist in a spacetime such that locations in the spacetime are
ontologically determined by which objects exist, then the spheres seem to share all of
their fundamental extrinsic properties, and hence all of their extrinsic properties. (By
stipulation, the extrinsic properties of having a certain location supervene in part on
the identity or difference of the spheres.) W seems to be possible. Note that by
granting the possibility of W, we take the spheres at different locations to be
numerically distinct spheres rather than a single bi-located sphere.29
The objection is developed by considering the possibility of W in conjunction with
the Principle of the Identity of Indiscernibles (PII). The PII claims that, necessarily, if
x and y are indiscernible, x is identical to y. There are many ways to interpret the
indiscernibility claim, and hence many interpretations of the PII. I will consider two
interpretations of the indiscernibility of x and y:
(i) x and y share all their properties, including properties such as being identical to x.
(ii) x and y share all of their pure intrinsic and extrinsic properties.
Pure properties are defined as properties that are not based on any sort of primitive
individuation of objects, i.e., they are not reducible, even in part, to any sort of
primitive identity or difference. They include properties such as being red, having a
mass of 2kg, etc. What pure properties exclude are primitive identity-based properties
like being this cup, being Alex, or being that blue shirt: these are impure properties.
The difference between interpretation (i) and interpretation (ii) captures the
worry. All parties agree that, if indiscernibility is interpreted as (i), the PII is true.
But what if indiscernibility is interpreted as (ii)? Is it true that, necessarily, if x and y
share all of their pure intrinsic and extrinsic properties, x is identical to y? If W is
possible, it seems that the PII under interpretation (ii) is not true.
The problem for bundle theory is that it is usually interpreted as entailing the truth
of the PII under (ii). Now, why would bundle theory entail any such thing? To get the
entailment, we need another premise, the Supervenience of Identity thesis: the
property of being identical to x reductively supervenes on x’s pure (intrinsic and
extrinsic) properties.
If the bundle theorist defends a pure bundle theory, and so holds that objects are
simply collections of pure properties, it would seem that she must endorse the
Supervenience of Identity thesis. And if she does, then the simple thought is that
bundle theory is committed to saying that the PII under interpretation (ii) is true.

28
Max Black (1952: 153–64) describes an example like this one as part of his discussion of the
individuation of objects.
29
Cover and (O’Leary-)Hawthorne (1998). It’s also important to note that the problem with symmetry
and location is not confined to worlds like W: as Hawthorne and Sider (2002) show, there are many
permutations of objects that can create trouble for bundle theory. The solutions to the problem of W can be
applied, mutatis mutandis, to these other problems.
 . . 

And this, many have thought, leads to disaster for the bundle theorist, since W seems
to be metaphysically possible.
The problem can be intensified by recognizing that W merely illustrates a more
general symmetry problem for the bundle theorist. Take any object which is perfectly
geometrically symmetrical with respect to the pattern of instantiation of its pure
properties. A two-dimensional object is intrinsically symmetrical in this way if it is
possible to divide it such that its instantiation of its pure properties is mirrored with
respect to an axis of symmetry, that is, it has two indistinguishable, numerically
distinct “halves.” (The properties instantiated at any two points lying on the perpen-
dicular of this object at equal distances from the axis of symmetry are indistinguish-
able.) For example, imagine an object whose left side is a perfect pure duplicate of its
right. The bundle theorist who accepts the Supervenience of Identity thesis is unable
to accommodate the possibility of any such symmetries: not only is the bundle
theorist unable to have a world like W, she is unable to grant the possibility of a
world with a single, extended, perfectly symmetrical object. The perfectly symmet-
rical spheres of W collapse into a single point.
Now, there is an obvious rejoinder the bundle theorist can make to objections
involving this sort of problem: reject the Supervenience of Identity thesis! There are
two ways to do this. First, one could deny the thesis by accepting haecceities or
impure properties as constituents of bundles. The substance theorist and the sub-
stratum theorist, in effect, deny the truth of the Supervenience of Identity in virtue of
their endorsement of haecceitism, and the bundle theorist could follow their lead.
But, second, one could follow the lead of the austere nominalist and deny the
Supervenience of Identity, while also rejecting any need for haecceities to perform the
individuative task. The austere nominalist simply holds that the thesis is false. It is
not “made false” by the existence of any primitive impure property, relation, or
substratum. Rather, identity facts simply supervene on the objects themselves: i.e., the
identity of x supervenes on x, and that’s all. The spheres of W are different. That’s it.
This move is simple, clean, and effective. It requires a kind of primitive individuation,
but one which is ungrounded in primitive facts. Call this sort of primitive difference
an “ungrounded difference.” (It amounts to a denial of any version of the principle of
sufficient reason that requires identity explanations.)
A version of the same move is made by the trope bundle theorist when she
embraces primitively individuated tropes, i.e., clement nominalism. According to
the (standard) trope bundle theorist, tropes are simply different, and this difference
does not supervene on any additional facts or properties. So tropes are distinguished
in virtue of ungrounded differences. Hence, trope bundle theory can accommodate
the possibility of W by holding that the tropes of each sphere are primitively
different, so the spheres are different objects. (Properties are sets of primitively
resembling tropes, so in the relevant sense the spheres can be said to include the
same properties even while they do not include the same tropes.)
But if the trope theorist and the austere nominalist can accommodate W, why can’t
the Russellian bundle theorist, even the Russellian who endorses transcendent
universals? As it turns out, the Russellian bundle theorist can accommodate W
(contra Benovsky 2008). Like the others, all she has to do is reject the Supervenience
of Identity thesis. I recommend that she do so by denying the PSR and accepting
     

ungrounded differences. On such a view, objects are constituted by nothing more

than compresent universals, but there can still be numerically distinct objects that are
bundles of the same properties. The objects are different, but not in virtue of
differences between their properties. Hence, in W, the spheres are different, but not
because of any primitive fact, property, or haecceity: they are barely different.
A variant on this view primitively individuates the primitive relations of compresence
that bundle the universals, on much the same model as the individuation of tropes.
Mereological bundle theory can follow the same lead. On the mosaic model of
mereological bundle theory, perhaps the fusions (the wholes) are barely different.
After all, she has already denied mereological extensionality. Or perhaps the fusion
relations are barely different, on the model of the difference between tropes. On the
global model of mereological bundle theory, we might brutely distinguish qualita-
tively symmetrical portions of the world-whole.
In any case, endorsing ungrounded differences allows the bundle theorist, whether
traditional or mereological, to accommodate the possibility of a purely symmetrical
universe. I find the move to reject the Supervenience of Identity thesis particularly
appropriate in the case of the Russellian bundle theorist and the mereological bundle
theorist, for they can then combine their reductive treatment of property resem-
blance (resemblance is explained in terms of sameness of properties) with the
possibilities for symmetry that we find intuitively plausible. (This move would
allow the Russellian to rebuff the arguments presented in Hawthorne and Sider
(2002).)
Now that we’ve seen how attractive ungrounded differences (or “bare differences”)
can be, it’s worth taking a little space to discuss a version of one option in particular
that I mentioned above, the one where one accepts impure properties or relations as
constituents of bundles. The move is relatively neglected, yet it has a distinguished
pedigree (e.g., we find it in Scotus) and there seem to be independent reasons why
one might find such a view plausible. I myself don’t want to endorse the move, but it
is interesting. The sorts of impure properties I have in mind would be derived from
ungrounded primitive differences between objects’ de re modal properties. (A more
traditional move would be to derive de re modal properties from primitive facts or
thisnesses.) One might hold that instances of de re modal properties are based on
bare differences.
Depending on one’s metaphysics of modality, such an individuation is not nearly
so ontologically heavy-duty as it might seem. For instance, one might hold that
instances of de re modal properties are individuated in the way other tropes are:
instances of being essentially such-and-such or accidentally so-and-so may primi-
tively resemble each other, yet remain numerically distinct. But even without a trope-
theoretic individuation (one might not like tropes), in the context of contemporary
reductionist, ersatzist accounts of de re modality, the ontological cost of this move is
very low. It is hard to get wound up about primitive differences between variables and
the like needed to ontologically distinguish representations, yet this is all the bundle
theorist needs to capture the postulated de re difference between indiscernibles.
Let me expand on this point a little. To make this move, I am assuming a certain
view of de re modality: an ersatzist treatment of de re modal properties in terms of
representation by abstract possibilia. On ersatzist accounts of de re modality, de re
 . . 

modal properties are reduced to ontologically thinner properties of being represented

in certain ways by abstract entities, such as sets of propositions. (Lest anyone think
that ersatzism requires antirealism about essence, note that it is perfectly possible for
a dyed-in-the-wool essentialist to be an ersatzist and a fan of counterpart theory. See
my (2006b).) To individuate de re modal properties, we simply need to primitively
individuate the representing done by abstract entities by allowing an abstract entity
to represent a pure situation twice over, i.e., to represent that there are two copies of a
pure situation. A way to do this as a linguistic ersatzer would be to represent that the
things represented are different by representing worlds quantificationally.30
To understand the intuitive motivation behind making primitive distinctions
between de re representations of purely indiscernible situations, consider Saul Kripke’s
famous defense of primitive distinctions between representations of possibility:
Two ordinary dice, (call them die A and die B) are thrown, displaying two numbers face up.
For each die, there are six possible results. Hence there are thirty-six possible states of the pair
of dice, as far as the numbers shown face-up are concerned, though only one of these states
corresponds to the way the dice actually will come out . . . The thirty-six possibilities, the one
that is actual included, are (abstract) states of the dice, not complex physical entities. . . . ‘How
do we know, in the state where die A is six and die B is five, whether it is A or B which is six?
Don’t we need a “criterion of transstate identity” to identify the die with a six—not the die with
a five—with our die A?’ The answer is, of course, that the state (die A, 6; die B, 5) is given as
such (and distinguished from the state (die B, 6, Die A, 5).) . . . The ‘possibilities’ simply are not
given purely qualitatively (as in: one die, 6, the other, 5). If they had been, there would have
been just twenty-one distinct possibilities, not thirty-six . . . Nor, when we regard such quali-
tatively identical states as (A, 6; B, 5) and (A, 5; B, 6) as distinct, need we suppose that A and
B are qualitatively distinct in some other respect, say, color. On the contrary, for the purposes of
the probability problem, the numerical face shown is thought of as if it were the only property of
each die. Finally, in setting up this innocent little exercise regarding the fall of the dice, with
possibilities that are not described purely qualitatively, we make no obscure metaphysical
commitment to dice as “bare particulars,” whatever that might mean. (1980: 16–18)

Kripke’s example brings out two points relevant to our discussion. First, it supports
the general point that ungrounded distinctions can make perfect intuitive sense:
I think we make ungrounded distinctions all the time when we make judgments
about probability, and we do not need haecceities in the form of bare particulars,
primitive thisnesses, or other ontological injections in order to do so.31 There’s no
primitive thing that the difference in possibilities must supervene upon. The possi-
bilities are simply barely different. Second, more specifically, it is intuitively natural
to primitively distinguish between the states or representations of possibility pro-
vided by possible outcomes of rolls of the dice even if we cannot purely distinguish
between the two different situations of an outcome of (A, 6; B, 5) and an outcome of

30
I am basically suggesting we follow Daniel Nolan’s (2002) ersatzist strategy for distinguishing alien
universals. Also see Sider (2002).
31
Kripke’s discussion doesn’t specify whether he thinks the primitive individuation has to be
ungrounded. I don’t think his example requires ungrounded primitive individuation to work, but it does
show how thinking of primitive individuation in this way makes good intuitive sense.
     

(A, 5; B, 6). I conclude that there are clear, intuitively plausible ways of primitively
individuating de re properties without heavy-duty ontological machinery.
All of this suggests that the primary ontological choice one must make, given the
seeming possibilities of various sorts of pure symmetries, is not between ontologies
but between whether or not we accommodate the possibility of these symmetries.
Only if one chooses to accommodate the possibilities must one then choose between
ontologies: between a universe with primitive grounded differences and a multiplicity
of categories, or a universe with primitive ungrounded differences and a single
category. In any case, if we assume that an intuitively appealing ontology requires
the accommodation of the possibilities of deeply pure symmetries, we need to reject
the Supervenience of Identity thesis.
Although many find the rejection of the Supervenience of Identity thesis and the
endorsement of ungrounded differences intuitively permissible, it is not without cost
to the bundle theorist. For what is the advantage of being a bundle theorist if one
endorses ungrounded difference? The worry comes to the following point: once one
endorses ungrounded differences, the appeal of bundle theory is lost, because it has
lost its purity. This objection has two parts. First, a universe of infinitely many
primitive differences gives us a messier ontology; a messier structure for reality.
I don’t find this part of the objection especially worrying. After all, if one accepts the
intuitions about differences, it reflects strong intuitions one has about the objects and
possibilities of this and other worlds. If we think our intuitions capture the onto-
logical facts, then we should endorse the structure they support. Moreover, substance
theorists and substratum theorists, and trope theorists of all varieties, already endorse
the existence of this sort of structure, albeit for independent reasons. The bundle
theorist simply joins the club in the ontological red-light district.
The second part of the objection is much more pressing, and I think it goes to the
root of why some are unimpressed by a bundle theory with ungrounded differences.
This part of the objection points out that it is no longer clear that a bundle theory
with ungrounded differences retains any overall ontological or categorical advantage
over substance theory or substratum theory. The thought is that the loss of purity
collapses the most obvious differences between bundle theory and its competitors.
The worry has some bite if one is a traditional bundle theorist, whether trope-
theoretic or Russellian. What are the differences between such a bundle theorist and
the substratum theorist? As we have seen, neither takes properties to be literally parts of
objects. Both endorse primitive individuation: the substratum theorist endorses primi-
tive differences grounded by substrata, while the bundle theorist endorses ungrounded
differences or differences grounded by thisnesses as constituents of the bundle. Bundle
theorists defend primitive bundling relations such as compresence, while substratum
theorists defend primitive substrata. Bundle theorists might argue that their
ungrounded primitivism is ontologically more minimalist than their competitors’
haecceitism, but the differences here seem rather small, even if they are not quite
terminological (Sider 2006, Benovsky 2010). Moreover, the substratum theorist can
argue that his view does a better job of accommodating the intuition that properties
need a bearer. Strictly speaking, because substratum theory distinguishes between
properties and their bearers, while bundle theory does not, substratum theory is a an
explicitly two category ontology while traditional bundle theory claims to be a one
 . . 

category ontology, but this is the only real point of difference between the two views,
and as I argued above, I’m inclined to hold that traditional bundle theory is a two
category ontology for other reasons (also see van Inwagen 2011).
But the mereological bundle theorist who accepts ungrounded individuation escapes
this problem virtually unscathed. She takes properties to be literally parts of objects,
and replaces bundling with a simpler, more familiar mereology of properties. (More
simple and more familiar because it is defined using our antecedent understanding of
“part.”) She denies any distinction between property and object, and builds this denial
into the foundation of her theory, thus retaining a significant measure of the simplicity
of her one category view, even if her mereology involves the denial of the uniqueness of
composition. Thus, if the Supervenience of Identity thesis is rejected, the real onto-
logical differences are between the mereological bundle theory on the one hand and the
substratum and traditional bundle theorists on the other.
I’ve been assuming that the possibilities of symmetry need to be addressed in one’s
ontology. But what if one thinks that we live in a fundamentally pure universe, one
where perfectly symmetric qualitative duplication does not occur? Call this sort of
person an “empiricist.” I find the appeal of the spareness and simplicity of such a
system, especially when coupled with mereological bundle theory, appealing.32 (Not
when applied to the case of asymmetric relations, since such relations impose quali-
tative asymmetry on a universe in virtue of their intrinsic character.) If we want to
preserve this purity and simplicity, we have two choices. We can endorse the PII
under interpretation (ii). Or we can reject the idea that a single ontology must suffice
for all possibilities, and endorse a contingent version of pure bundle theory. On this
view, it’s not analytic that “object” must pick out a bundle of properties. Other worlds
may have other ontologies. It’s just that pure mereological bundle theory is the best
theory of what objects are in the actual world, and worlds very like ours in the relevant
senses. This view either holds that the actual world contains no perfect symmetries
across locations, and so is not threatened by W-like cases, or holds that including the
right sort of spacetime in one’s ontology can accommodate all actual symmetries.
I find either option appealing. Personally, since I think metaphysical reasoning
involves the construction of models, and the acceptance of a metaphysical model as
true is based partly on a posteriori reasoning involving inference to the best explan-
ation, I prefer the contingent ontological option. So officially, I defend a contingent
version of pure mereological bundle theory. Those who find even the mere possibility
of primitive individuation abhorrent will prefer to embrace the PII.

2.6 Indiscernible Particles in Quantum Mechanics

Either view is plausible, except for one wrinkle: an objection concerning a widely
accepted empirical possibility of having multiple, purely indiscernible elementary

32
Dasgupta (2009) argues for the view that there are no primitively individuated individuals. This is not
quite the claim that there are no primitive individuals, nor is it the same as the claim that there is no
categorical difference between individuals and properties. However, his view indirectly supports a one
category ontology by undermining arguments that claim to have empirical support for ontologies that
include primitively individuated individuals.
     

particles at the very same spatiotemporal location.33 Call this the haecceitist objection.
According to this objection, empiricists must recognize and accommodate the real-
world fact of the existence of systems consisting of two or more purely indistinguish-
able elementary particles, such as 2-boson states, and this fact blocks both of the
empiricists’ options.34 Now, one might flat-out deny that, given contemporary
physics, we should even countenance the idea of anything like purely indiscernible
elementary particles, but let’s set this option aside for the purpose of the argument.
The haecceitist objection, restated, goes like this: it is a widely accepted empirical
truth that in the actual world there can exist states such that two or more perfectly
purely indistinguishable bosons occupy the very same location. In order to accom-
modate this truth, the objection goes, we must individuate the bosons impurely, for
there are two of them, they have the same location, and they are purely indistin-
guishable. The objection claims that the actual-world case both entails that the PII is
false and refutes the contingent empiricist’s pure bundle theory. In other words, the
objection claims that the empirical facts support haecceitism, or at the very least,
primitive individuation of some sort, and disprove the possibility of the purity of the
actual world.35
Or so one might think, if one is insufficiently clear about the metaphysical import
of the standard interpretation (the “Received View”) of these quantum systems. Let’s
look at the facts concerning these systems according to the Received View about
them. As will become clear, the possibility of multiple, purely indiscernible elemen-
tary particles poses no threat to the empiricist.
The decisive point is that, contrary to the assumptions behind the haecceitist
objection, the Received View of states involving indiscernible elementary particles
removes any basis for an argument for haecceitism. The Received View is that,
ontologically, the particles are not distinct entities, so no haecceities are needed to
distinguish “them.” The basis for this conclusion is the collection of empirical facts
derived from a standard interpretation of quantum statistics. According to this
interpretation of quantum statistics, facts about permutation invariance show us it
is wrong to think of quantum states, such as 2-boson states, as involving multiple
indiscernible bosons, even if they are usually described as “2-boson states” or as
“states involving multiple indiscernible particles.” This is because such quantum
states are “permutation invariant.” What this means is that the distribution of the
objective probabilities of the ways the states can occur refutes the possibility that, for
example, a 2-boson state is composed of two indiscernible bosons.
It’s worth saying a little more about how permutation invariance should be
understood. The idea is that it is ontologically (not merely epistemically) pointless
to try to permute the so-called indistinguishable particles in order to count, say, a
situation where particle A is in state S1 and particle B is in state S2 as different from

33
Della Rocca (2005) discusses an interesting variant of this possibility.
34
I’m ignoring options that involve moves like primitively individuating spatiotemporal locations or
primitively individuating instantiations of properties, since these are in effect impure solutions that such an
empiricist would reject.
35
Dasgupta (2009) disputes this, arguing that there is no possible empirical support for primitively
individuated individuals or bearers of properties and relations.
 . . 

a situation where B is in S2 and A is in S1. There is no ontological difference between

these situations: they are the same situation. The quantum statistical distribution of
the probabilities of the occurrence of the various ontologically possible situations
reflects this fact.
The thought is that, if there really were two particles, A and B, composing the
“two-particle” system, then they would be permutable, i.e., there would be four
different possible (and equiprobable) configurations. But the quantum statistical
facts do not support permutability: there are only three possible (equiprobable)
configurations. So the quantum statistics show us a fact about the ontology of these
states: it shows us that there is no ontological difference between the state where
particle A is in S1 and particle B is in S2 and the state where particle B is in S1 and
particle A is in S2. The Received View then imposes an interpretation that makes
sense of this: since there are no swappable (permutable) parts, the n-particle state is
not composed of numerically distinct particles after all. Such states are composed
some other way, which is suggested by their interesting, perhaps peculiar, lack of
quantitative structure. Although we use the phrase “2-boson-state” to refer to the
state or system in question, the state does not have the sort of structure that the name
suggests: in particular, it does not have internal quantitative structure involving
multiple repeating “particle units.”
So the Received View is that since there is no way to permute the “1-particle parts”
of a “2-particle” system, it is not divisible into 1-particle parts. A standard, but
linguistically unfortunate, way this claim is made is by saying that the indiscernible
particles, such as the two bosons of a 2-boson state, are “not individuals,” are “not
objects,” or “lack individuality” or “lose their identity.”
Metaphysicians must understand such claims in the correct way. Clearly, “objects”
here does not mean what metaphysicians often mean by “objects.” By their claim that
quantum particles are not objects, fans of the Received View mean only that quantum
particles do not physically resemble tiny billiard balls or atoms; or that whatever the
right ontology is here, to think of quantum particles under an atomistic conception is
mistaken. Quantum particles are not small, physically and spatiotemporally discrete
entities that we can count or measure; in this sense, they are fundamentally unlike
material objects, or unlike what we can call “thick objects.” Although the quantum
mechanical description of the ontology, at least initially, involves talk of “particles” and
of “states,” and of, say, “2-particle” or “3-particle” states, this description is fundamen-
tally misleading because it suggests the atomistic or ordinary conception of objecthood.
Whatever we are picking out by the phrase “quantum particle,” it is not anything even
remotely similar to what we ordinarily take a material object to be. Hence, the standard
way we use language to talk about “quantum particles” is misleading, presumably as
the result of our pre-theoretical or classical physics-based conceptual stance. What the
quantum statistics show is that this conceptual stance must be revised.
But if a 2-boson system is not divisible into two 1-particle parts, then what is it?
How can it make sense to postulate a partless entity of this sort?36 A natural way to

36
Under some sort of physically nonstandard interpretation, this could connect up to the possibility of
extended simples.
     

understand the Received View, based on the approach taken by Erwin Schrödinger,
takes these sorts of quantum states as being, fundamentally, instantiations of prop-
erties.37 Even though we speak of particles at the quantum level, such talk does not
refer to atom-like bits of matter whizzing around the void, or anything remotely like
this. We are simply referring to instantiations of fundamental properties in our talk
of n-particle systems. Such properties may or may not include instances of having
certain masses, locations, or forces, but we do not need a thick atom-like object to
“have” these properties for them to be instantiated.
We can take the property instances to be excitation strengths (“field excitations”),
or something very like. So a 2-boson state is just a field with an excitation strength of,
say, 2b. (Determining exactly what the 2b property is is one of the jobs of physics.)
This is how I interpret the Received View, and I believe this is how we should
understand comments like those of Saunders (2006): “We went wrong in thinking
the excitation numbers of the mode, because differing by integers, represented a
count of things; the real things are the modes” (p. 60).38
Now, this interpretation is mysterious in a certain way, because it opens up
ontological possibilities that do not seem to have been sufficiently appreciated.
What is mysterious is not that the fundamental entities are property instances. The
mystery attaches to the way we are inclined to think of the nature of these property
instances, which are what Armstrong labeled “fundamentally intensive properties.”
We are used to thinking of many types of properties as unstructured in a certain
sense; for example, the property of being sweet seems to lack internal structure. But
we are also used to thinking of other sorts of properties as having internal structure,
especially properties involving quantities, like having a mass of 2kg, or for that
matter, having a field excitation strength of 2b. We are inclined to think that
properties involving such quantities are constructed, somehow, from more funda-
mental quantities; perhaps unit quantities (although what exactly those unit quan-
tities might be is a matter of mystery). Having a mass of 2kg reductively supervenes
upon having two instances of having a mass of 1kg, and so on. But the possibility of
irreducibly intensive properties of quantity means that, at the very least, not all
properties of quantity are constructed this way. (Maybe none are.) Armstrong
(1997: 65) hoped that there were no irreducibly intensive properties of quantity,
since they create difficulties for his conception of universals.
The interpretation of the Received View entails that Armstrong is out of luck: there
are such intensives, for example, the property of 2-boson-ness. The mystery we are
left with is how to develop a metaphysics of irreducibly intensive properties, which is
a problem that metaphysicians seem to have been ignoring. The obvious patch is to
accept that there may be more fundamental properties than we might have thought
there were. In any case, the relevant point is that the irreducibly intensive properties
of these quantum states are internally unstructured in that they are not reducible to
conjunctions or co-instantiations of multiple instances of unit quantities such as
being a boson or being boson A. In other words, the intensive properties of the
quantum states we have been considering, like that of 2-boson-ness, are ontologically

37 38
See Teller (1983). Also see Teller (1983) and Schaffer (2001).
 . . 

basic. For example, the property of 3-boson-ness is ontologically basic, and so is not
constructed from 2-boson-ness and 1-boson-ness, or from a state of affairs involving
three indiscernible “atomistic” bosons. The advantage of moving to an interpretation
of a 2-boson state as an instantiation of 2-boson-ness is that, to many, it is much
more plausible to admit some irreducibly intensive properties into one’s ontology
than to admit, say, partless extended simples.
The bundle theoretic view, given the Received View, follows the standard inter-
pretation and holds that systems of indiscernible quantum particles are states that are
instances of irreducibly intensive properties. (Of course, according to the mereo-
logical bundle theorist, these are objects.) The proposal that property instances are
the fundamental quantum-mechanical entities can also characterize states in which
there is no definite number of particles or a superposition of particles, and where
such states or modes are representation-dependent.39 In such cases, the property
instances or states reflect the indeterminacy we find in the world. The property
instances here are relational or structural properties of having a certain (incomplete)
structure that can be represented in different ways.
Now that we have the Received View in hand, the consequences of the empirical
facts for the empiricist who desires a pure universe are obvious. Those who endorse
the PII can argue that n-boson states are not counterexamples to their list of what is
possible.40 Correspondingly, those who have independent reasons to endorse a
contingent pure ontology can argue that n-boson states are not counterexamples to
their claim that a pure bundle theory gives the best (i.e., true) model of the actual
world.41 For those who want to defend the view that claims supporting one category
ontologies are necessary truths, they might respond to the possibilities involving
symmetrical worlds by claiming that such symmetries involve single instantiations of
irreducibly intensive properties, not multiple instantiations of the same properties.
There is a further way, perhaps, in which the Received View supports mereological
bundle theory, for it provides a way to understand and motivate reasons for defend-
ing the view that there is no ontological difference between properties and objects.
Just to be clear: one can understand the Received View that quantum states are
fundamentally intensive property instantiations in a way that is consistent with many
different ontologies. Traditional bundle theories can endorse irreducibly intensive
properties just like mereological bundle theorists can. Substratum theorists can take
the field to be the bearer of the irreducibly intensive property, austere nominalists can
perhaps take a field to be an object in some sense and classify fields differently
depending on excitation strengths, etc. Substance theorists can take the field to be an
unstructured object standing in relations to irreducibly intensive universals.

39
I’m indebted to Richard Healey for pointing out that I need to address this issue.
40
The solution incorporating intensive properties may also be relevant to Della Rocca’s (2005)
discussion.
41
In fact, it seems to me that if one rejects the idea that the world is built from spatiotemporal parts, and
also takes the world to be a fusion of properties, with its structure determined by structural properties,
symmetry worries don’t even arise, since the world gets its structure from the structural property fused
with other properties, not from building up from smaller mixes of properties. Thus, mereological bundle
theory may avoid symmetry problems outright, since it does not build the world spatiotemporal hunk by
spatiotemporal hunk.
     

But there may be a more subtle way in which the interpretation of the Received
View supports mereological bundle theory. To see this, recall that the key move of the
interpretation is conceptual: it tells us that our knee-jerk way of thinking about the
things physicists describe as “objects” or “particles” as little material-like hunks of
stuff is fundamentally mistaken. We knew this already from quantum mechanics, but
the ontology of elementary boson states sharpens the need for ontological precision.
The case gives us a straightforward example of a situation where, for empirical
reasons, we must reconceive our approach to particle-talk and think in terms of
properties, not substances, with regard to talk of objects like particles. As such, we
have a nice, clear case of how it is possible and desirable to make the conceptual move
from substance to property in a particular case. Such a good example of collapsing
the artificial distinction between property and object—and making the fundamental
entities more property-like—lends support to the legitimacy of making this move in
the ontology of the mereological bundle theorist.42 Anybody who claims that such a
move involves some sort of category mistake or that it is unintelligible needs to
defend that claim using more than knee-jerk reflections. Moreover, the existence of
properties such as 2-boson-ness provides examples of actual properties that are
concrete or can exist independently, contra Simons and others’ claims about the
dependency of properties.
My personal preference is for a contingent, pure mereological bundle theory where
all things, including locations, are constructed from fusions of n-adic properties. As
stated, my view endorses the empiricist stance I described above: unless there is some
further empirical reason to make primitive distinctions, the simplicity and purity of
pure bundle theory should be preserved. However, even if we need to move to a
bundle theory that delimits the symmetrical possibilities by embracing ungrounded
differences, there is still a clear metaphysical advantage to embracing the one
category ontology that my mereological bundle theory gives us, as we retain the
clear distinction between bundle theory and substratum theory that traditional
bundle theorists lose. There is no need to look to any other ontology.

References
Adams, R. M. (1979), ‘Primitive Thisness and Primitive Identity’, in Journal of Philosophy 76/1:
5–26.
Albert, D. (1996), ‘Elementary Quantum Metaphysics’, in J. Cushing, A. Fine, and S. Goldstein
(eds.), Bohmian Mechanics Quantum Theory: An Appraisal (Springer Netherlands).
Armstrong, D. M. (1989), Universals: An Opinionated Introduction (Westview Press).
Armstrong, D. M. (1997), A World of States of Affairs (Cambridge University Press).
Armstrong, D. M. (2010), Sketch for a Systematic Metaphysics (Oxford University Press).
Bacon, J. (1995), Universals and Property Instances: The Alphabet of Being (Blackwell).
Baker, D. J. (2011), ‘Broken Symmetry and Spacetime’, in Philosophy of Science 78: 128–48.

42
The collapse of the distinction may also favor the austere nominalist, who could interpret the
Received View to endorse entities such as structureless 2-boson simples. I don’t think this is the usual
way the philosophers of physics are inclined to think of the situation, but it is coherent. See Katherine
Hawley (2009) for related discussion.
 . . 

Benovsky, J. (2008), ‘The Bundle Theory and the Substratum Theory: Deadly Enemies or Twin
Brothers?’, in Philosophical Studies 141: 175–90.
Benovsky, J. (2010), ‘Relational and Substantival Ontologies, and the Nature and the Role of
Primitives in Ontological Theories’, in Erkenntnis 73/1.
Black, M. (1952), ‘The Identity of Indiscernibles’, in Mind 61: 153–64.
Campbell, K. (1990), Abstract Particulars (Blackwell).
Cover, J., and (O’Leary-)Hawthorne, J. (1998), ‘A World of Universals’, in Philosophical Studies
91: 205–19.
Dasgupta, S. (2009), ‘Individuals: An Essay in Revisionary Metaphysics’, in Philosophical
Studies 145/1: 35–67.
Della Rocca, M. (2005), ‘Two Spheres, Twenty Spheres, and the Identity of Indiscernibles’, in
Pacific Philosophical Quarterly 86: 480–92.
Fine, K. (1999), ‘Things and Their Parts’, in P. A. French and H. K. Wettsetein (eds.), New
Directions in Philosophy: Midwest Studies in Philosophy 23 (Blackwell), 61–74.
Fine, K. (2000a), ‘A Counter-example to Locke’s Thesis’, in Monist 83/3: 357–61.
Fine, K. (2000b), ‘Neutral Relations’, in Philosophical Review 109/1: 1–33.
Goodman, N. (1966), The Structure of Appearance (The Bobbs-Merrill Company, Inc.).
Hawley, K. (2009), ‘Identity and Indiscernibility’, in Mind 118: 101–19.
Hawthorne, J., and Sider, T. (2002), ‘Locations’, in Philosophical Topics 30: 53–76.
Horgan, T., and Potrč, M. (2009), Austere Realism: Contextual Semantics Meets Minimal
Ontology (MIT Press).
Kripke, S. (1980), Naming and Necessity (Harvard University Press).
Loewer, B. (2004), ‘Humean Supervenience’, in J. Carroll (ed.), Readings on Laws of Nature
(Pittsburgh: University of Pittsburgh Press), 176–206.
Loux, M. J. (1998), Metaphysics: A Contemporary Introduction (Routledge).
Lowe, E. J. (2006), The Four-Category Ontology: A Metaphysical Foundation for Natural
Science (Oxford University Press).
Malament, D. B. (1996), ‘In Defense of Dogma: Why There Cannot Be a Relativistic Quantum
Mechanics of (Localizable) Particles’, in R. Clifton (ed.), Perspectives on Quantum Reality
(Kluwer), 1–10.
Markosian, N. (1998), ‘Brutal Composition’, in Philosophical Studies 92/3: 211–49.
Martin, C. B. (1980), ‘Substance Substantiated’, in Australasian Journal of Philosophy 58/1:
3–10.
McDaniel, K. (2001), ‘Tropes and Ordinary Physical Objects’, in Philosophical Studies 104/3:
269–90.
McDaniel, K. (2009), ‘Ways of Being’, in D. Chalmers, D. Manley, and R. Wasserman (eds.),
Metametaphysics (Oxford University Press), 290–319.
Moss, Sarah (2012), ‘Four-Dimensionalist Theories of Persistence’, in Australasian Journal of
Philosophy 90/4: 671–86.
Ney, A. (2012), ‘The Status of Our Ordinary Three Dimensions in a Quantum Universe’, in
Noûs 46/3.
Nolan, D. (2002), Topics in the Philosophy of Possible Worlds (Routledge).
Paul, L. A. (2000), ‘Aspect Causation’, in Journal of Philosophy XCVII(4), 235–56.
Paul, L. A. (2002), ‘Logical Parts’, in Noûs 36/4: 578–96.
Paul, L. A. (2006a), ‘Coincidence as Overlap’, in Noûs 40: 623–59.
Paul, L. A. (2006b), ‘In Defense of Essentialism’, in J. Hawthorne (ed.), Philosophical Perspec-
tives 20 (Metaphysics): 333–72.
Paul, L. A. (2012), ‘Building the World from its Fundamental Constituents’, in Philosophical
Studies 158: 221–56.
     

Paul, L. A. (2013), ‘Categorical Priority and Categorical Collapse’, in Aristotelian Society

Supplementary Volume 87: 89–113.
Paul, L. A., and Hall, Ned (2013), Causation: A User’s Guide (Oxford University Press).
Saunders, S. (2006), ‘Are Quantum Particles Objects?’, in Analysis 66: 52–63.
Schaffer, J. (2001), ‘The Individuation of Tropes’, in Australasian Journal of Philosophy 79:
247–57.
Schaffer, J. (2003), ‘The Problem of Free Mass: Must Properties Cluster?’, in Philosophy and
Phenomenological Research 66/1: 125–38.
Schaffer, J. (2010), ‘Monism: The Priority of the Whole’, in Philosophical Review 119: 31–76.
Scotus, J.D., Quaestiones in libros metaphysicorum V, n.d.
Sider, T. (2002), ‘The Ersatz Pluriverse’, in Journal of Philosophy 99: 279–315.
Sider, T. (2006), ‘Bare Particulars’, in Philosophical Perspectives 20: 387–97.
Sider, T. (2008), ‘Monism and Statespace Structure’, in R. Le Poidevin (ed.), Being: Develop-
ments in Contemporary Metaphysics (Cambridge), 129–50.
Sider, T. (2012), Writing the Book of the World (Oxford University Press).
Simons, P. (1987), Parts: A Study in Ontology (Oxford Clarendon Press).
Simons, P. (1994), ‘Particulars in Particular Clothing: Three Trope Theories of Substance’, in
Philosophy and Phenomenological Research 54: 553–75.
Stout, G. F. (1921), ‘The Nature of Universals and Propositions’, in Proceedings of the British
Academy, Vol. X.
Teller, P. (1983), ‘Quantum Physics, the Identity of Indiscernibles, and Some Unanswered
Questions’, in Philosophy of Science 50: 309–19.
Turner, J. (2010), ‘Ontological Pluralism’, in Journal of Philosophy 107.
Van Inwagen, P. (2011), ‘Relational vs. Constituent Ontologies’, in Philosophical Perspectives
25 (Metaphysics).
Wallace, D., and Timpson, C. G. (2009), ‘Quantum Mechanics on Spacetime I: Spacetime State
Realism’, in British Journal for the Philosophy of Science 61: 697–727.
Williams, D. C. (1953), ‘On the Elements of Being’, in The Review of Metaphysics VII: 3–18 &
171–92.
Wolterstorff, N. (1970), ‘Bergmann’s Constituent Ontology’, in Noûs 4/2: 109–34.
 3
Properties as Parts of
Ordinary Objects
Eric T. Olson

3.1 The Constituent Ontology

According to the so-called constituent ontology, concrete particulars are made up of
things other than concrete particulars. Dogs are made up of atoms. But the constitu-
ent ontology says that dogs also have parts that are nothing like atoms—so unlike
them as not to be concrete particulars at all.
The claim is not that dogs are made up partly of atoms and partly of these other
things, in the way that my desk is made up partly of metal pieces and partly of
wooden ones. A dog is made up entirely of atoms. Take away the atoms and nothing
is left. But it is also made up entirely of things that are neither atoms nor concrete
particulars of any other sort, so that taking away those things (whatever that might
amount to) would leave nothing. The atoms and the other things occupy different
levels of composition. On one level, a dog is made up of atoms, but on another level—
a metaphysically deeper one—it is made up of the other things.
The concrete particulars making up a dog and the way they relate to one another
are sometimes called its mereological structure. The things other than concrete
particulars making up a dog are called its constituents, and they and the way they
relate to one another are its logical or ontological structure.
The principal constituents of things, on the constituent ontology, are properties.
These might be shareable universals, so that the same property of being human is a
constituent of both you and me; or they might be nonshareable “tropes,” such as
my very own humanity: a thing numerically different from yours, though perhaps
exactly like it.
No one, to my knowledge, thinks that concrete particulars are made up of both
tropes and universals, or that some are made of tropes and others of universals. But
some say that they are made up of tropes or universals together with another
constituent—just one—that is not a property at all but rather a special sort of
particular: “the particular in abstraction from its properties” (Armstrong 1997:
123) or “thin particular” or “substratum.” It does not itself have properties as
constituents. It bears or supports properties, but does not exemplify them. The
properties “inhere in” without characterizing it, like feathers stuck into clay. It is
       

only because of their inhering in a common substratum, the idea goes, that they are
exemplified by a single concrete particular—though that concrete particular is not
the substratum itself but something made up of the substratum and the properties.
(There is dispute about whether a substratum could exist without supporting any
properties, as a “bare particular.”)
That makes four species of constituent ontology: that concrete particulars are
made up of universals, of tropes, of universals and a substratum, or of tropes and a
substratum.
The constituent ontology is not the view that properties are parts of certain
abstract objects: facts or states of affairs or propositions, say.1 Perhaps what makes
it true that the Eiger is steep is the existence of a fact composed of that mountain and
the property of steepness. Or maybe the proposition that the Eiger is steep has those
things as parts. But such an entity would not be visible or climbable or made of ice-
covered limestone. I am talking about the view that steepness is a part of the Eiger
itself—that large, cold, dangerous physical object.
So the constituent ontology says that concrete particulars are made up of proper-
ties (and perhaps substrata) in the way that they are made up of atoms, only on a
more fundamental level. The alternative is that concrete particulars are made up only
of concrete particulars. A dog is made up of atoms, and at a physically deeper level it
is made up of elementary particles. It is not in any sense made up of properties, or
these together with a substratum. There are no substrata at all. There may be properties,
but they are never parts, or anything like parts, of concrete particulars. Concrete
particulars simply have no ontological structure. (Armstrong says they would be
mere “blobs.”) This is sometimes called the relational ontology (Loux 2006: 208),
the idea being that properties are removed from the particulars exemplifying them,
and bear to them only an abstract and bloodless relation utterly unlike that of part to
whole. Because someone could reject the constituent ontology by denying the
existence of properties altogether, ‘relational ontology’ is not the best name for this
view. But good or bad, that’s the name it has.
Constituent and relational ontologies are not just two competing views, but
radically different ways of thinking about the metaphysics of concrete objects.
Relationalists are concerned with the way material things relate to their parts—
their concrete, particular parts, that is. They think about what changes of parts a
thing can survive, if any—think of the ship of Theseus, the puzzle of Dion and Theon
(or amputation paradox), and the problem of increase (or growing paradox). They
ask whether the same parts can compose two different “coinciding” objects at once,
and when smaller concrete particulars compose a larger one.2 For the most part,
constituent ontologists ignore these questions and ask entirely unrelated ones:
whether concrete particulars conform to the identity of indiscernibles, what it is for

1
Some object to my claim that states of affairs are abstract (Armstrong 1997; see also Wolterstorff 1970:
111, though this is not his view). They say that ordinary concrete objects such as dogs or mountains are
states of affairs. This is not a substantive disagreement, but merely a different use of the technical term
‘state of affairs.’
2
On the amputation paradox see van Inwagen (1981); on the problem of increase see Olson (2006); on
coinciding objects see Baker (2000), and on composition generally see van Inwagen (1990).
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

several properties to belong to the same thing, whether there is “more to” a particular
than its properties, and how to avoid Bradley’s regress, for instance3—questions of no
interest to most relationalists. The result is separate debates about the metaphysics of
concrete objects with little common ground. This can be frustrating, because parti-
cipants in these debates often presuppose a constituent or a relational ontology
without saying so, leaving readers to guess, on the basis of the moves they make,
which rules they’re playing by.
I cannot hide the fact that I was brought up as a relationalist. The idea that such
things as dogs have their properties as parts has always seemed to me deeply
confused—like the view, attractive to certain undergraduates, that dogs are properties.
But I won’t make that point here (see van Inwagen 2011). Let us suppose, if we can,
that the constituent ontology is coherent and intelligible. I want to explore problems
that it seems nevertheless to face—problems for the most part very different from the
ones constituent ontologists typically worry about. It may be that the problems are
only apparent and the appearance is based on a misunderstanding. In that case I hope
someone will be able to clear things up. Or maybe the problems are real but solvable.
In any event, I hope to say something useful to both sides.

3.2 Basic Principles

Let me try to state the basic principles of the constituent ontology. It may help to start
by asking how anyone could be led to suppose that ordinary things had properties as
parts. The usual answer is that it would tell us what it is for a thing to have a nature or
character (Loux 2006: 207, Wolterstorff 1970: 112). What is it, for instance, for the
book on my desk to be red? More precisely, what does its being red consist in? What
accounts for or explains it? What is its metaphysical ground? The question is not only
about being red: we could just as well ask what it is for the book to be rectangular or
solid or interesting. What accounts for any concrete particular’s being any way?
Constituent ontologists answer that a thing has a character by having or exempli-
fying properties. It may be too simple to say that a thing is red by having the property
of redness. There might be no such property, but only a lot of precise wavelengths,
and mixtures of different wavelengths, that make light red to human beings. But the
book’s being the way it is consists in its having some properties or other. This
consisting in or grounding is asymmetric: a thing’s properties account for its char-
acter and not vice versa. So:
1. Concrete particulars have their character because they have the properties they
have.4
In fact this looks like the only way of saying what accounts for a thing’s character.
What could account for a thing’s character if not its properties?
If a thing’s character consists in its having properties, then things with the same
properties will have the same character. If things could have a different character but

3
Rea (1997) consists of debates among relationalists. Some typical constituentist papers are collected in
the second half of Loux (1976); Simons (1994) gives a useful history of the view.
4
Relations may come into the story as well; see section 3.5.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

the same properties, their properties would not entirely account for their character,
contrary to 1. So:
2. Concrete particulars with the same properties have the same character.
(If the properties in question are tropes, their sameness will be exact resemblance.)
Constituent ontologists may also accept the converse principle:
2*. Concrete particulars with the same character have the same properties.
That is, when concrete particulars are exactly the same, this sameness consists in
their sharing all their properties. If those properties are universals, this would explain
qualitative sameness in terms of numerical sameness. It’s not clear whether constitu-
ent ontologists have to say this. Maybe different ontological structures could produce
the same character, in something like the way that different combinations of wave-
lengths can produce the same color or different neural properties can give rise to the
same mental property. There might be “multiple realizability” of character by
properties, so that a thing’s properties fix its character but not vice versa.
If a thing’s character consists in its properties, then just as things with all the same
properties will have exactly the same character, things sharing some properties will
have a similar character:
3. Concrete particulars that share properties are alike in some respect.
If the book and the chair share the property of being red, that makes them alike in
color.
Here the converse principle is far more doubtful:
3*. Concrete particulars that are alike in some respect share properties.5
Even if things’ properties fix their character, the book might be red by virtue of
having one property while the chair is red by having another. Or there might be
concrete particulars so different that they share no properties at all, yet still alike in
some respect: in being concrete particulars, for instance.
And some aspects of a thing’s character may consist not in its having certain
properties, but in its not having them. However different you and I may be, we are
alike in not being prime numbers. If this consisted in our sharing the property of not
being a prime number, it would make that property a common constituent of all
objects apart from prime numbers. And if a thing’s being distinct from the number 42
consisted in its having the property of not being 42, then every entity would be made
up of as many properties as there are numbers. Constituent ontologists generally deny
that there are any such properties as not being a prime number or not being the
number 42. They say that only an elite minority of predicates correspond to proper-
ties.6 A thing’s not being a prime number will presumably consist in its lacking
properties. So our being alike in this respect cannot consist in our sharing properties.

5
Though Armstrong appears to accept it, under the slogan ‘resemblance is partial identity’ (1989: 15).
6
This is another deep difference between the two schools. Relationalists typically see no reason to
suppose that properties are “sparse”: see e.g. van Inwagen (2004). The connection between properties’
being parts of ordinary objects and their being sparse is an underexplored topic.
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

Suppose all this is right: concrete particulars have their character because of the
properties they have (or lack). How does this suggest that particulars are made up of
properties? Well, if things have their character because of the properties they have—if
their character depends on their properties—then the relation between a thing and its
properties must be close and intimate. This intimacy is not a sort of inseparability,
since it applies to properties accidental to a thing as well as those essential to it. We
can get some idea of what it is by contrasting it with the Platonistic view that
properties are abstract objects not in space and time. This makes the relation between
a concrete particular and its properties as abstract and bloodless as the relation
between the Martian moons and the number two (van Inwagen 2011: 392). It may
be that a thing is red, necessarily, if and only if it relates in a certain way to a timeless
object in Plato’s heaven, just as the Martian moons are two, necessarily, if and only if
they relate in a certain way to the number two. But the character of a thing cannot be
grounded in that way. The book may exemplify a Platonic universal because it is red,
but it cannot be red because it exemplifies a Platonic universal. The book’s bearing a
Platonic universal can no more account for its being red than the truth of the
proposition that grass is green could explain why grass is green (van Inwagen
2011: 398).
If a thing is red because it exemplifies the property of being red and not vice versa,
then the property must somehow be, as the constituent ontologists say, in the thing.
It must be “down here” where the red things are, not up in Plato’s heaven with the
numbers and the propositions. This relation has to be something more than mere
spatial coincidence: even if the book coincided with a portion of a magnetic field, that
would not make it magnetic. The book’s properties, or at least its intrinsic properties,
must be somehow built into its structure. (Armstrong compares them to layers in
a cake.)
But what is it for a property to be in something? What is this close and intimate
relation between things and their properties, and how does it differ from the one that
features in the relational ontology? This is a notoriously hard question. The proposed
answer is that a thing’s properties are constituents of it—that is, parts. They are of
course not the same sort of parts as atoms. Maybe a thing’s properties and its atoms
stand in different parthood relations to it. Or maybe they stand in the same parthood
relation and the difference is simply that between properties and atoms. But a thing’s
properties are parts of it in some sense or other. Not just any part of a thing, or even
any part that is a property, is a constituent of it. I have electrons as parts, and each
electron has as a constituent the property of having a mass of approximately 9.11

1031kg. Since the parts of my parts are also my parts, this property too must be a
part of me. But not a constituent, as I don’t exemplify it: my mass (take my word for
it) is far greater.7 So we can say at least this:
4. A concrete particular has a property iff (and because) the property is a con-
stituent of it.
5. x is a constituent of y iff x is a part of y and x is in y.

7
As Chad Carmichael has pointed out to me, it follows that my skin color is in me but the masses of my
electrons are not, reminding us that ‘in’ is a technical term.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

Assuming that every concrete particular has some character or other, and that no
concrete particular can have a character by having no properties at all, it follows from
these principles that all concrete particulars have properties among their parts.
These are the basic principles of the constituent ontology. Relationalists, by con-
trast, may accept 2 and 3, but not 1, 4, and 5. They may say that for a thing to have a
certain character is for it to exemplify certain properties, but not that the second
fact grounds or explains or otherwise accounts for the first.

3.3 Constituency and Parthood

The constituent ontology says that properties are constituents of concrete particulars,
and I have taken constituency to be a sort of parthood. What could it mean to say that
dogs are bundles of properties if this did not imply that properties were parts of dogs?
(It can hardly mean that dogs are sets, in the mathematical sense, having properties
as members.) Constituent ontologists say that concrete particulars are composed of
properties (and perhaps substrata), and composition is defined in terms of parthood:
the xs compose y =df each of the xs is a part of y, and every part of y shares a part with
at least one of the xs.8 And constituency appears to share these characteristic features
of parthood:9
Inheritance of intrinsicness: If being F is intrinsic, then having a part (constitu-
ent) that is F is also intrinsic.
Inheritance of location: If x is a part (constituent) of y, then y is located where x is.
Transitivity: If x is a part (constituent) of y and y is a part (constituent) of z, then x
is a part (constituent) of z.
Weak supplementation: If x is a part (constituent) of y and x6¼y, then y has a part
(constituent) that does not overlap (i.e., share a part [constituent] with) x.
Strong supplementation: If y is not a part (constituent) of x, then y has a part
(constituent) that does not overlap x.
Dependence of whole on part: If all the parts (constituents) of a thing did not
exist and other things were equal, then that thing would not exist either.
If constituency is not a sort of parthood, we can only wonder what it might be.
But perhaps a thing’s constituents are not parts but only something like them, and
constitutive composition cannot be defined in terms of parthood.
Someone might say this on the grounds that constitutive composition violates
some of the principles of classical mereology (Armstrong 1989: 91f.). One such
principle is that composition is universal and unrestricted: any things whatever
compose something. But not just any things are constituents of something: nothing
has as constituents both the property of being human and that of being an electron.
Classical mereology also says that composition is unique—things can compose only

8
Most definitions also specify that no two of the xs share a part. Omitting this clause enables some of
the points in sections 3.7 and 3.10 to be made more simply.
9
The first two are from Sider (2007: 70); the third to the fifth are fairly uncontroversial principles of
classical mereology.
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

one object—whereas some constituent ontologists say that things can constitutively
compose two different objects at once (see section 3.5).
But this is a poor reason to deny that constituents are parts. Many non-
constituentists reject the uniqueness of composition, and even more reject universal
composition. They deny that there is any sense of ‘part’ in which things must always
be parts of something bigger. The mere existence of my left foot and the planet Mars
does not compel us to accept that there is also something composed of those two
objects—a large material thing with two parts a hundred million miles apart.
Whether these philosophers are right to reject the principles is of course disputed.
But if the very idea of parthood presupposed them, this rejection would amount to
denying that there are any parts. Yet these philosophers don’t deny that there are
parts. They are talking about the same parthood and composition that classical
mereologists speak of. They simply disagree with them about the truth of certain
metaphysical principles concerning parts, namely the universality and uniqueness of
composition. Even if those principles are true, they don’t appear to be true by the
definition of ‘part’. Constitutive composition need not conform to them in order to
count as a parthood relation.10
A better thought is that parts must be independent of wholes. The atoms com-
posing a dog can exist without being parts of that animal or any other ordinary
object. By contrast, constituent ontologists typically say that things’ constituents
depend for their existence on those things: a universal cannot exist without being a
constituent of something, and a trope cannot exist without being a constituent of the
particular thing it is in. Whether this is a good reason to deny that constituency is
parthood depends on whether a thing’s parts, just by being parts, must be able to exist
independently—that is, without being parts. Aristotelian hylomorphists say no: they
deny that a dog’s heart can exist without the dog, but not (usually) that it is a part of
the dog. They say only that hearts are not substances.
Or one might say that constituency cannot be parthood because parthood is
reflexive: everything is a part of itself, but not everything is a constituent of itself.
Otherwise every property would exemplify itself: the property of being red, being a
constituent of itself, would be red, just as my book is. But although the reflexivity of
parthood simplifies mereology (much as the principle that every set is a subset of
itself simplifies set theory), it may be negotiable.
I will return to this theme in section 3.11. Until then I will take constituents to be
parts.

3.4 Brute Character

Let me now examine the constituent ontology more critically. Its main virtue is
supposed to be that it tells us what it is for a concrete particular to have a character.11

10
Another assumption of classical mereology disputed by both relational and constituent ontologists is
that parthood is timeless: things have their parts without temporal qualification. For more on classical
mereology and its discontents, see Simons (1987).
11
Paul (2002) claims other virtues for her own constituent ontology, which differs in many ways from
“traditional” versions. I won’t discuss her view.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

This presupposes that there is something that a concrete particular’s having a

character consists in: a metaphysical ground or explanation. There has to be an
illuminating and nontrivial answer to the question of what it is for a thing to be F, an
answer having the same form for all values of ‘F ’. Why accept this? Why couldn’t a
thing’s having a character be a brute and unanalyzable fact, with no deeper meta-
physical ground? The constituent ontology is proposed as an answer—indeed the
only possible answer—to a certain metaphysical question. But why suppose that the
question has any answer?
The likely reply is that the constituent ontology would be a good answer if it were
true. If a good answer to a question is available, that makes it reasonable to suppose
that the question has an answer. But I doubt whether the constituent ontology is a
good answer.
For one thing, it doesn’t tell us what it is for things in general to have a character. It
does not enable us to complete the schema
x is F iff . . . .
The property of being red has a character. It’s a property, specifically a color. Its being
a color is as much an aspect of its character as being red is an aspect of the book’s
character. But the constituent ontology says nothing about what it is for red to be a
color. All it could say is that red’s being a color consists in its having the property of
being a color (or some other property or combination of properties) as a constituent.
But constituent ontologists don’t say this. If any thing’s having any character
consisted in its having a certain property as a constituent, all properties would be
made up of further properties. The property of being red would be made up of such
properties as being a color, which would in turn be made up of yet further properties;
and so it would go on. And the character of every property would be grounded in the
character of some other property. There would be no ultimate metaphysical ground
of the book’s being red: it would consist in something else, which in turn would
consist in something further, and so on without end, like a building with no lowest
floor. (Relationalists, by contrast, can happily say that every property has other
properties, generating an endless chain in which each link exemplifies the next,
because they don’t say that a thing’s character is grounded in its properties.)
Constituent ontologists say that the property of being red has no properties as
constituents, and thus no ontological structure.12 Either there are no “higher-order”
properties that a property can have, or else such properties are not in their instances,
and properties have properties in a different way from that in which concrete
particulars have them. Yet every property has a character. It follows that some things
have a character without having properties as parts. And because different such
things have different characters, their character cannot consist in their lacking such
parts either. (The character of a substratum might consist in its having no properties

12
Armstrong appears to say that complex properties have their character by virtue of being composed
of simpler properties, and thus resemble each other by having common constituents (1989: 106). Simple
properties, though, have brute character. (He says they never resemble each other, even though any two of
them are both properties, both simple, and more like each other than like a dog. This is an odd use of the
word ‘resemble.’)
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

as parts, but this is a special case.) So a thing’s character is not always grounded in its
ontological structure, or presumably in anything else. Some character is “brute.”
But then why suppose that concrete particulars have properties as parts? Why isn’t
their character brute? You might say that some things must have a brute character
because not everything can have its character derivatively (Loux 2006: 207f.). If my
book has its color by having redness as a part, it derives or “borrows” its character
from that property. And not everything can borrow its character. Borrowing has to
come to an end somewhere. Some things must have a character in their own right,
nonderivatively. But even if this explains why not all things have properties as parts,
it does nothing to explain why some do. Nor does it tell us which do and which don’t,
or why.

3.5 Relations
Not only does the constituent ontology give no account of the character of properties,
but it appears unable to account for the entire character even of concrete particulars.
Their having properties as constituents is normally taken to imply that every
property is a constituent of something: properties are in their instances, and every
property has to be somewhere (Armstrong 1997: 38, Loux 2006: 233–9). This will
hold for relations as well: if the book’s being red is grounded in its having as a
constituent the property of being red, then the book’s resting on the desk must be
grounded in something’s having as a constituent the relation of resting on. At least
this will be so for external relations, the holding of which is not grounded in the
properties of the relata.
This something—the thing of which the resting-on relation is a constituent—
cannot be either the book or the desk. Its being a constituent of the desk could only
account for some of the desk’s parts resting on others—its top resting on its legs, for
instance (and likewise for the book). The something has to include both book and
desk, taking up the whole of the book and the desk but not their surroundings: not
the carpet beneath the desk or the sandwich on top of the book. But it won’t be
composed only of the book and the desk either, as such a thing would not have the
resting-on relation as a constituent (though it might have that relation as a part, if it’s
a constituent of the desk). We need an object having the book, the desk, and the
resting-on relation as constituents—a concrete particular like the book and the desk.
This suggests an analog of principle 4:
4*. Concrete particular x bears an external relation to concrete particular y iff (and
because) there is a concrete particular having as constituents x, y, and the relation.
(Similar principles would hold for relations of more than two places.)
But the existence of a “relational complex”13 having as constituents the book, the
desk, and the resting-on relation cannot make it the case that the book rests on the
desk: it could just as easily make it the case that the desk rests on the book. It may
be that the book rests on the desk because the book bears the resting-on relation to

13
I take the term ‘complex’ from Price (1998: 24).
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

the desk. But this cannot consist in the resting-on relation’s being a constituent of
anything. The mere existence of something made up of the book, the desk, and the
resting-on relation cannot determine whether the book is on the desk or the desk on
the book.
Someone might say (as Armstrong does, 1997: 121) that the book’s resting on the
desk is grounded in one sort of object composed of the book, the desk, and the
resting-on relation, whereas the desk’s resting on the book (if I were to rearrange my
office furniture) would be grounded in a different sort of object composed of the
same three entities arranged differently. By analogy, we could glue a red Lego brick to
a blue one either with red on top and blue underneath or vice versa. If this would
create an object composed of the two bricks and the glue, it would create a qualita-
tively different object depending on how we arranged them. That looks like the only
way for the constituent ontology to account for things’ relational character.
But this works badly with relations that are neither symmetric (like touching) nor
asymmetric (like resting on). Suppose Alex loves Leo. The proposal is that this is
because there is a concrete particular composed of Alex, Leo, and the loving relation,
arranged, so to speak, in that order. Were those things arranged otherwise, they
would compose a different particular that would make it the case that Leo loves Alex.
Well, suppose Leo loves Alex too, so that the things are also arranged in this other
way. Then there are two objects—two concrete particulars—composed of Alex, Leo,
and the loving relation. One, the “Alex-to-Leo complex,” accounts for Alex’s loving
Leo; the other, the “Leo-to-Alex complex,” accounts for Leo’s loving Alex. They differ
in that their constituents are differently arranged. (Here the analogy fails: if bricks are
arranged with red above and blue below, they cannot at the same time have blue
above and red below.)
But this looks incoherent. The two complexes would be composed of the same
three constituents, and since each person loves the other, the constituents would be
arranged both Alex-to-Leo and Leo-to-Alex. If two objects are composed of the same
parts, and those parts are arranged in the same way, nothing about the way the parts
are arranged can make the objects qualitatively different. Of course, we know what it
is for Alex to love Leo, what it is for Leo to love Alex, and how these propositions
differ. But that does nothing to account for the difference between the Alex-to-Leo
and Leo-to-Alex complexes.
Someone might say that the Alex-to-Leo complex exists because Alex loves Leo,
whereas the Leo-to-Alex complex exists because Leo loves Alex. Were Alex to stop
loving Leo, there would cease to be an Alex-to-Leo complex, though a Leo-to-Alex
complex may remain. Despite the fact that they have the same parts arranged in the
same way, the two complexes differ by having different metaphysical grounds. But
this gets the grounding the wrong way round: the constituent ontology says that Alex
loves Leo because there is an Alex-to-Leo complex, not vice versa.
Now there are philosophers—“constitutionalists”—who think that concrete particu-
lars can and do compose qualitatively different objects at once: my atoms, for instance,
now compose both a person who is thinking about philosophy and an organism or
“body” that is not. Both have the same parts, arranged in the same way. What makes
them different is not the way their parts are arranged but something else. If this is so, it
may also be possible for concrete particulars and relations arranged in the same way to
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

compose qualitatively different objects. But others find this aspect of constitutionalism
deeply mysterious, and its advocates have done nothing to dispel the apparent mystery
(see Bennett 2004, Olson 2001). Apparently the constituent ontology can account for
the character of relational complexes (and the relational character of things generally)
only by embracing the same mystery.

3.6 How the Constituent Ontology Leads to

Substance Dualism
I have argued that the constituent ontology offers no good account of what gives
ordinary objects their character, undermining the main reason for accepting it.
(There may, of course, be others I haven’t considered.) I turn now to reasons to
think the view is false. They arise even if the worries I have been considering can be
answered. The problems have to do with objects that are more “abstract” than
ordinary concrete particulars in that they have fewer properties as constituents. We
might call them quasi-abstract objects. (‘Quasi-abstract things’ or ‘entities’ would be
an equally good name: I don’t mean anything special by ‘object’.)
Consider the thing composed of all my constituents except my physical properties:
shape, size, mass, temperature, atomic structure, and so on. (I ask in section 3.9
whether there has to be such a thing.) According to the constituent ontology
(principle 4), it will lack any physical properties. It will be a wholly nonphysical or
immaterial thing. Yet all my mental properties will be constituents of it, making it (by
4 again) psychologically indistinguishable from me. It will be an immaterial mind.
This is not quite Cartesian dualism, as it doesn’t imply that all thinking beings are
immaterial, or that physical and mental properties are incompatible. In a way it is less
mysterious than Cartesian dualism, since it allows that mental phenomena might
arise out of physical ones. But in another way it’s more mysterious: it implies that
even if all mental phenomena have a physical basis, some of their subjects—some
conscious, thinking beings—are wholly immaterial. It would mean that there are
both material and immaterial human thinkers, and that for every human being there
is one of each. It is an absurd amalgam of dualism and materialism.
The problem would not arise in this form if all mental properties were physical
properties. But it would arise in another form: consider the thing composed of all my
constituents except my nonmental properties. It will have only those of my physical
properties that are also mental properties. Since mass, shape, and color will not be
mental properties even if some physical properties are, it would be a massless,
shapeless, colorless mind.

3.7 Things Composed of Atoms

Maybe we should expect a thing composed of all my constituents except my
physical properties to be a strange sort of object. But the constituent ontology
would make any object composed of material things equally strange. Consider the
thing composed of my atoms: the thing such that each of my current atoms is now a
part of it, and every part of it now shares a part with one or more of those atoms.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

Call it O. O may have parts that are not atoms. The negative charge that is a
constituent of each of my atoms’ electrons might be one. My liver might be another,
since each of its parts would overlap one or more of my atoms. (Any atom that is a
part of my liver shares a part with it, since each is a part of itself.) But any part of O
other than an atom would have to be either a part of one of O’s atoms or else
composed of parts distributed across several of those atoms.
If O’s properties are parts of it, then each of them must also share a part with one
or more of O’s atoms. There are two cases. First, a property might be a part of one or
more of O’s individual atoms. Presumably any property that is a part of an atom is a
constituent of it, and thus a property it exemplifies. So any property of O’s that is a
part of one or more of its atoms will be a property it shares with those atoms. But O
will have many properties that it doesn’t share with any of its atoms: its human shape
and size, for instance, its atomic structure or chemical makeup, and its having a solid
surface.
Alternatively, a property might, like my liver, be composed of parts distributed
across several of O’s atoms without itself being a part of any of them. But O’s shape
could not be like this either. A small plaster figurine could have the same shape. So
could a thing composed of the figurine’s atoms. Its shape, like O’s, would have to be
composed of parts of its individual atoms. But its shape would have far fewer parts
than O’s shape has, or at any rate fewer atom-sized parts. So they could not be the
same shape (or qualitatively identical shape tropes).
It appears, then, that O’s shape could be neither a part of one of its atoms nor
composed of parts distributed across several of its atoms. But because O is composed
of atoms, every part of it must be either an atom, a part of an atom, or composed of
things that are parts of atoms. It follows that O’s shape cannot be a part of it. Yet the
constituent ontology requires a thing’s shape to be a part of it. We can only conclude
that O has no shape.
Likewise, O’s size or volume cannot be a part of any atom. Nor could it be
composed of parts of O’s atoms. If it were, those parts would presumably be the
atoms’ sizes: how could a size be composed of anything other than sizes? And if the
sizes of each of O’s atoms composed a size, it would be the sum of those sizes. But O’s
size is far greater than the sum of the sizes of its atoms, since most of its volume
consists of the empty space between the atoms. It follows that O cannot have any size
property as a part. If properties are parts, O can have no size.
Or consider O’s atomic structure. It will have to include relations among O’s
atoms. But those relations cannot be either parts of individual atoms or composed of
such parts. Because O is composed of atoms, such relations cannot be parts of it, and
thus on the constituent ontology it cannot have an atomic structure. But how could a
thing composed of close-knit atoms have no atomic structure?
The constituent ontology appears to imply that the object composed of my atoms
could have no shape or size or atomic structure. And this goes for all things
composed of atoms. But ordinary material things have a shape, size, and atomic
structure. It follows that ordinary material things—human beings, stones, pieces of
furniture—are not composed of atoms. They may have atoms among their parts, but
they need other parts as well, which don’t overlap their atoms. They have to be
composed of atoms together with certain properties (and probably relations). It is
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

metaphysically impossible for an ordinary material thing to be made up entirely of

smaller material things: atoms, bricks, pieces of yarn, or what have you.
You could still make a material thing out of bricks in the sense that you would not
literally need to “add” the properties to the bricks in order to construct one. All you
would have to do is arrange bricks, and that action would generate the necessary further
ingredients. But the resulting material thing would not be made up entirely of bricks.
This makes the nature of objects composed of material things mysterious. The
thing composed of my atoms could not be a person or a human being or even a
material thing. It would have no shape or size or atomic structure. What character it
would have is anyone’s guess.
This is trouble. For one thing, it puts the mystery in the wrong place: it makes what
ought to be familiar things mysterious, and what ought to be mysterious things
familiar and ordinary. We expect familiar things to be made up of atoms. If anything
is mysterious, it ought to be something made up partly of concrete and partly of
abstract entities—a compound of certain atoms and the number 42, say. Yet the
constituent ontology implies that things made up partly of atoms and partly of
properties and relations are the ordinary and familiar ones, and the weird things
are those made of atoms.
For another, it’s incompatible with the usual statement of the constituent ontol-
ogy. Constituentists say that ordinary objects are composed entirely of atoms, as well
as being entirely composed, on another level, of properties and relations. It now
appears that this cannot be right. At best familiar objects could be composed of atoms
together with properties and relations, much as a desk is composed of wooden pieces
together with metal ones.
But the real trouble is that it is impossible for a thing made up entirely of things
with shape and size to lack any shape and size of its own. How could things that
individually take up space add up to something that takes up no space? Consider a
thing composed of bricks. It will not typically have the shape of any of the bricks. Nor
could it have one composed of the shapes or other parts of the individual bricks, for
whatever shape it has could be shared with an object having a completely different
composition. If its shape would have to be a part of it, then it simply cannot have a
shape. Putting bricks together might produce something with a shape—a thing
composed of bricks together with a shape and perhaps other properties—but the
thing composed of the bricks would have none. For that matter, a thing composed of
the bricks and a certain shape would apparently have shape but no size, whereas a
thing composed of the bricks and a certain size would have size but no shape. If
ordinary things had properties as parts, there would be impossible objects.

3.8 Brute Character Again

Can the problem be solved? Someone might say that quasi-abstract objects have a
brute character not determined by their constituents. So the object composed of all
my parts except my physical properties, though it has no physical properties as
constituents, nevertheless has a physical character. Likewise, the thing now com-
posed of my atoms is physically just like me, even though my shape, size, and atomic
structure are not constituents of it. So there are no impossible objects. If this appears
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

to contradict the central claim of the constituent ontology, namely that the character
of a concrete particular consists in its constituent properties, we could make it
consistent by denying that quasi-abstract objects are concrete particulars.
But this would be like saying that if our theory of dogs doesn’t apply to terriers,
that only shows that terriers aren’t dogs. A thing composed of bricks is a paradigm
case of a concrete particular, just as a terrier is of a paradigm case of a dog. How could
a thing made of bricks not be concrete? How could it not be a particular?
You might say that concrete particulars are those things whose constituent prop-
erties account for their character: those satisfying principles 1–5. But that would
make the constituent ontology into the harmless tautology that those things with
properties as constituents have properties as constituents. It would remain an open
question whether there are any such things. If things composed of atoms, whether
concrete particulars or not, have a brute character not fixed by their constituents,
why not say that all things do?

3.9 Restricting Composition

Or one might deny the existence of quasi-abstract objects.14 My argument from the
constituent ontology to substance dualism assumed that there was something com-
posed of all my parts except my physical properties. And my argument that objects
composed of atoms could have no shape or size assumed that my atoms compose
something. But why assume this? Maybe there is nothing composed of all my parts
except my physical properties, and nothing composed of my atoms, or of any other
atoms. Maybe none of the problematic entities exist. There would then be no
problem of quasi-abstract objects. This would of course mean that composition is
restricted rather than universal: some things compose something bigger and others
don’t. But unrestricted composition is controversial, and there is no reason why
constituent ontologists must accept it (even if many do, e.g. Armstrong 1997: 13).
If quasi-abstract objects were always arbitrary and gerrymandered, so that we had
no reason to believe in their existence apart from the principle of unrestricted
composition, this would be a promising thought. Unless I believe that any things
whatever must compose something, it’s reasonable for me to doubt whether there is
anything composed of Peru, the Thirty Years War, and the key of B flat. Challenged
to say why those things don’t compose anything, I should be within my rights to
reply, “Why suppose that they do?”
But there is nothing arbitrary about a thing composed of my atoms. It seems
possible for atoms to compose something, even if not all things do, or even all atoms.
If composition is not universal, it would be surprising if there were a thing composed
of Peru, the Thirty Years War, and the key of B flat. It would not be surprising if there
were things composed of atoms. In fact it would be surprising if there were no such
things. If any things ever compose anything, surely certain atoms do.
It doesn’t help that the constituent ontology requires an otherwise generous
ontology of composite objects. Recall that whenever one concrete particular stands

14
Simply denying that they can exist independently of ordinary, “complete” objects is no help.
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

in an external relation to another, there must be a further concrete particular

composed of those things together with the relation—two of them if the relation is
nonsymmetric and each bears it to the other. Because my left foot is gravitationally
attracted to Mars, there must be something composed of those two things together
with the attraction relation, and perhaps also a second such object to account for my
foot’s attracting Mars. That makes a lot of objects besides the familiar ones that we
have names for. Combining this with the claim that atoms can never compose
anything looks unprincipled at best.

3.10 Mereological and Constitutive Composition

You may think the problems arise from conflating two different senses of parthood.
Constitutive parthood and composition are one thing; mereological parthood and
composition are something else. They are not only different, but independent, in that
a thing can be a constitutive part of something without being a mereological part, or a
mereological part without being a constitutive part. So I am mereologically composed
of certain atoms: each atom is a mereological part of me, and every mereological part
of me shares a mereological part with one or more of those atoms. But my atoms are
not constituents of me. I am not constitutively composed of atoms, but rather of
properties (set aside relations and substrata): each of my properties is a constituent of
me, and every constituent of me shares a constituent with one or more of those
properties. And a thing’s being mereologically composed of atoms is perfectly
compatible with its being constitutively composed of properties that share no
parts with those atoms. There is that there is nothing “quasi-abstract” or otherwise
problematic about objects mereologically composed of atoms: they are ordinary
things, with just the shape, size, and atomic structure we should expect.
Now if constituents are parts of any sort, there will be a more general notion
encompassing both constitutive and mereological parthood, and a similarly broad
notion of composition. We might call them  and , and
define them like this (where ‘part’ means ‘mereological part’):
x is a  of y =df x is a part or a constituent of y.
The xs  y =df each of the xs is a  of y, and every  of y shares a
 with one or more of the xs.
And the problem of quasi-abstract objects can be restated using these generic
notions. Consider the object now  of my atoms. By the definition of
, every  of it that is not an atom must be either a  of an
atom or  of  of several atoms. Yet that object’s shape and size could
be neither  of any individual atom nor  of  of several atoms. So
they could not be  of the object, and therefore not constituents of it. It could
have neither shape nor size, which is impossible.
But constituent ontologists will turn this reasoning on its head.15 They will deny
that my atoms  anything. Otherwise (by the definition of )

15
I thank Chad Carmichael for helping me to see this.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

everything that was either a part or a constituent of such a thing would have to share
a part or constituent with one or more of the atoms. That would rule out its having a
size or a shape, for the reason just given. Yet it would have to have a size and shape. It
follows that my atoms cannot  anything. At most the atoms together with
my size and shape could  something. Presumably atoms could never
 anything (except in the degenerate case where there is only one atom
that  itself). Yet atoms could still compose something. In fact the proposal
is compatible with universal composition, though not of course with universal
. But no one ever endorsed universal .
The proposal may seem confusing. A thing’s being a part of something (in the
mereological sense) entails its being a  of it; so it may appear that things’
composing something entails their  it. But the proposal denies this. It
says that my atoms compose a thing by being its parts (at a certain level of
decomposition). And they are also  of it. Yet they do not  that
thing. This is because not all its  share a  with one or more of the atoms.
Specifically, its shape and size are  of it—constituents—but do not share a
 with any of the atoms.
This looks like the best way for constituent ontologists to respond to the problem
of quasi-abstract objects. Remember, though, that mereological parthood is not a
special technical concept invented by metaphysicians. It’s just ordinary, unqualified
parthood. “Mereological” parts are simply parts. (As we saw earlier, this need not
entail conformity with all the principles of classical mereology.) So the proposal
amounts to denying that a thing’s constituents are parts of it. Constituency is not
parthood but only something like it. Ordinary objects are “made up” of properties,
but not in the sense of having them as parts.

3.11 What if Constituents Aren’t Parts?

Suppose, then, that constituents are not parts. Because this is an important departure
from the views we have considered so far, we had better call it the modified
constituent ontology.
The proposal will have to go beyond denying that constituency is a sort of
parthood. Being a constituent of something has to rule out its being a part of it.
Constituency and parthood must be incompatible. This is for at least two reasons.
First, suppose a constituent of something could be a part of it. Suppose my size were
a part of me (as well as a constituent). Then I could not be composed of atoms, but at
best of atoms together with a size property. But my atoms would still compose
something—mereologically compose, that is—or at least they would if atoms can ever
compose something. We might once again call that the thing they would compose O.
What would O be like? Seeing as it would be composed entirely of things with a size, it
too would have to have a size: the same as mine. But that size, once again, could not be
a part of it, as it would not be composed of parts distributed across O’s atoms. So there
would be two objects with the same size, and this size would be a part of one of them
(me) but not of the other (O—though it would be a constituent of O). Yet no difference
in our properties would account for the difference in how we relate to our sizes,
violating the claim that a thing’s character consists in its properties (principle 2).
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

  . 

Second, if a thing’s constituents could be parts of it but needn’t be, we should

expect some of things’ properties to be parts of them and some not to be. (It would be
surprising if properties could be parts of their instances, but they never are.) But
which? In what circumstances would a property of a thing be a part of it? Without
some answer to this question, however incomplete, there will be no solution to the
problem of quasi-abstract objects. And it’s hard to see what the answer might be.
These problems would not arise if constituency ruled out parthood. But this avoids
the trouble only by lapsing into obscurity. That a thing stands to its properties in a way
at least analogous to that of whole to part was the central claim of the constituent
ontology. Even if constituents aren’t parts exactly, they were supposed to be like parts.
If constituency is actually incompatible with parthood, what remains of this claim?
The point becomes more acute when we consider the thing composed of my atoms
together with my properties. Call it O*. Given that constituency rules out parthood,
O* will not have any of my properties: since those properties are parts of it, they
cannot be constituents, which (by principle 4) rules out its exemplifying them. And
things sharing no properties are utterly dissimilar. It will probably not even count as a
concrete particular. This will be due entirely to the difference between constituency
and parthood—between my properties’ being constituents of me rather than parts,
and parts of O* rather than constituents. That makes constituency and parthood very
different relations.
Someone might deny that there is anything composed of my atoms together with
my properties, rejecting the existence of O*. This would be a restriction on mereo-
logical composition. Suitably generalized, it would rule out properties’ being parts of
any ordinary object—even properties the object doesn’t exemplify. Not only are a
thing’s constituents never parts of it, but no properties could be, again reinforcing the
difference between constituency and parthood.
The urgent question at this point is whether the modified constituent ontology is
any different from the relational ontology. How does constituency differ from the
“abstract and bloodless” relation that properties bear to their instances according to
relationalism? What could it mean to say that dogs are composed of properties if no
property is anything like a part of a dog? What is there in the view for relationalists to
object to?
There is still the claim that a thing’s properties ground or explain its character
(with the exceptions noted in sections 3.4 and 3.5). This was supposed to rule out
things’ relating to their properties in the abstract way of the relational ontology: a
thing’s properties have to be constituents of it. But what is it to be a constituent? The
original answer was that a thing’s constituents are in it—built into its structure. And
to be in a thing was to be a sort of part of it, or something like a part. Though that
leaves many questions unanswered, it would clearly rule out the relational ontology.
It’s no help, though, if properties are not like parts. Relationalists are unlikely to
quarrel with the claim that properties are in their instances in a sense that is
indefinable and unrelated to parthood.
Until we’re told more about the modified constituent ontology, it’s hard to know
what to make of it.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

      

Acknowledgments
For invaluable comments on ancestors of this paper I thank John Keller, Jessica Leech, and
especially Chad Carmichael.

References
Armstrong, D. M. (1989), Universals: An Opinionated Introduction (Westview Press).
Armstrong, D. M. (1997), A World of States of Affairs (Cambridge University Press).
Baker, L. R. (2000), Persons and Bodies: A Constitution View (Cambridge University Press).
Bennett, K. (2004), ‘Spatio-temporal coincidence and the grounding problem’, in Philosophical
Studies 118: 339–71.
Loux, M. (ed.) (1976), Universals and Particulars (University of Notre Dame Press).
Loux, M. (2006), ‘Aristotle’s constituent ontology’, in D. Zimmerman (ed.), Oxford Studies in
Metaphysics, vol. 2 (Oxford University Press).
Olson, E. T. (2001), ‘Material coincidence and the indiscernibility problem’, in Philosophical
Quarterly 51: 337–55.
Olson, E. T. (2006), ‘The paradox of increase’, in Monist 89: 390–417.
Paul, L. A. (2002), ‘Logical parts’, in Noûs 36: 578–96.
Price, H. H. (1998), ‘Universals and resemblances’, in P. van Inwagen and D. Zimmerman
(eds.), Metaphysics: The Big Questions (Blackwell). (Originally published as chapter 1 of
Thinking and Experience, 1953.)
Rea, M. (ed.) (1997), Material Constitution (Rowman and Littlefield).
Sider, T. (2007), ‘Parthood’, in Philosophical Review 116: 51–91.
Simons, P. (1987), Parts: A Study in Ontology (Oxford University Press).
Simons, P. (1994), ‘Particulars in particular clothing: Three trope theories of substance’, in
Philosophy and Phenomenological Research 54: 553–75.
Van Inwagen, P. (1981), ‘The doctrine of arbitrary undetached parts’, in Pacific Philosophical
Quarterly 62: 123–37. Reprinted in Rea 1997.
Van Inwagen, P. (1990), Material Beings (Cornell University Press).
Van Inwagen, P. (2004), ‘A theory of properties’, in D. Zimmerman (ed.), Oxford Studies in
Metaphysics, vol. 1 (Oxford University Press).
Van Inwagen, P. (2011), ‘Relational vs. constituent ontologies’, in Philosophical Perspectives
25: Metaphysics: 389–405.
Wolterstorff, N. (1970), ‘Bergmann’s constituent ontology’, in Noûs 4: 109–34.
 4
Time Travel and the
Movable Present
Sara Bernstein

Science fiction has made time travel seem familiar and natural. Usually, time travel
goes something like the following: a time traveler enters a time machine, flips a
switch, and exits the time machine in, say, 1960. Then the time traveler embarks on
adventures that create vexing logical and metaphysical paradoxes: killing her own
grandfather, or throwing herself a tenth birthday party that didn’t occur in the first
place. Philosophical literature on time travel focuses largely on paradoxes generated
by time travel to the past (and more rarely to the future). But less attention is paid to
what the time travel itself consists in.
This paper develops one such conception by extracting a principle from a recently
proposed model of time travel, and building on it. In “Changing the Past” (2010),
Peter van Inwagen argues that a time traveler can change the past without paradox in
a growing block universe. After erasing the portion of past existence that generates
paradox, a new, non-paradox-generating block can be “grown” after the temporal
relocation of the time traveler.
This discussion articulates and expands on the underlying mechanism of van
Inwagen’s model: the time traveler’s control over the location of the objective present.
Van Inwagen’s discussion is aimed at preventing paradox by changing the past, but
I argue that it provides metaphysical tools for a new model of time travel that can be
generalized to most theories of time.1
Roadmap: in section 4.1, I lay out the requisite conceptual tools for discussing
time travel. I also give a brief summary of van Inwagen’s view. In section 4.2,
I distinguish between two different conceptions of time travel: (i) a traditional
conception in which the time traveler does not move the location of the objective
present, and (ii) time travel in which the time traveler shifts the location of the
objective present. I then develop and explore the latter model, which I call
movable objective present, or MOP. After generalizing MOP to presentist, moving

1
Hudson and Wasserman (2010) argue for a similar conclusion, but do not identify or utilize the more
general mechanism of the movable objective present. They argue that hypertime is the key feature that
allows the generalization of the model; here I argue that MOP is also essential to the model.
       

spotlight, and growing block theories of time, I present interesting possible errors
of MOP travel.

4.1 Theories of Time and Time Travel

Before delving into models of time travel, it will be helpful to discuss relevant views
on the metaphysics of time. Eternalism holds that all times are equally real. There is
no meaningful metaphysical distinction between the past, present, and future; Abra-
ham Lincoln is just as real as the iPhone 47. Whether or not one is located in the
present is merely a matter of perspective, akin to whether or not one is located “here”
in the spatial manifold.
Presentism holds that only the present is real. Everything that exists is present;
non-present things do not exist.
Growing block holds that the past and present are real. The growing blockist shares the
eternalist’s belief in the reality of the past, but stipulates that the present is the “growing
edge” of being. With every passing moment, the total amount of being expands.
There are additional questions about the existence and metaphysical status of the
objective present. Roughly, the existence of an “objective present” implies that a
present time slice is metaphysically privileged in some way. Eternalism, presentism,
and growing block vary with respect to the existence of a privileged or objective
present. Eternalism denies the existence of such a privileged present, though often
includes a subjective present that consists in the perspective of being present
without metaphysical privilege. Moving spotlight holds that the present is privileged
within the eternalist block by a kind of spotlight that “illuminates” temporally
successive slices of the universe. Presentism implies the existence of an objective
present by holding that what exists just is the objective present. Growing block
holds that the objective present corresponds with the forward-moving bleeding
edge of existence.
Whether or not there is an objective present has deep and important implications
for models of time travel, and I will not give an argument for its existence here. Nor
will I take a stand on which view of time is correct. I will only work with theories of
time that accept an objective present: presentism, growing block, and moving
spotlight theory. I will also assume that we can make sense of the idea of an objective
present, and that it can ground a certain type of time travel.
Onto van Inwagen’s model. Using the growing block theory of time, van Inwagen
proposes a model of time travel according to which a time traveler relocates herself to
an earlier temporal location in the existing block and annihilates the part of the block
between her temporal points of departure and arrival. From the point of arrival, a
new, paradox-free block is generated.
Let us examine this model in more detail. Consider a normal growing block
universe, represented by the following diagram:

(a) Growing block universe

t 2013
  

Let t represent the beginning of time, and 2013 be the present. Imagine that
Bianca, living in 2013, regrets not attending Woodstock in 1969. Here we have the
timeline as it occurred, without Bianca having attended Woodstock:
(b) No time travel

t 1969 2013
(no Bianca at Woodstock)

Now, according to van Inwagen’s model, a time traveler can travel to an earlier
point in the block and erase the portions of the block in between the temporal points
of departure and arrival. Suppose that Bianca travels from 2013 to 1969 (arriving very
shortly before Woodstock so that she can attend the event). In doing so, she
annihilates the portion of existence between 1969 and 2013, like so:
(c) Time travel

t 1969
(Bianca at Woodstock)

Post-time travel, existence is forty-four years smaller than it was before. Bianca
attends Woodstock, which she did not do the first time around. Whereas “normal”
time travel generates a paradox between Bianca’s initial non-attendance and later
attendance at Woodstock, van Inwagen’s scenario avoids such a problem: Bianca
erases the portion of the block in which her initial non-attendance at Woodstock
(plus all of the relevant aftereffects) occurs. After Bianca’s time travel, the portion of
the block beginning with her arrival is regenerated like so:
(d) Time travel plus regeneration

t 1969
(Bianca at Woodstock)

The year 1969 onward occurs a second time around. Only Bianca remembers the
“first” time 1969 occurred. But others do not, for she has removed their post-1969
time slices from existence in virtue of having removed all of post-first-time 1969 from
existence. And given that Bianca changes the events in second-time-around-1969
with her very presence, nothing else is guaranteed to go the way it did the first time.
The end of the Vietnam War, the invention of the microchip, sequencing the
human genome: they may not happen, given changes to the past and new regener-
ation of the block.
Van Inwagen’s model makes use of hypertime. Hypertime, roughly, is a temporal
manifold within which another temporal manifold exists. To make this idea clearer, van
Inwagen considers an Intelligence that exists outside of time but which sees events in
reality unfold in the order in which they occur in hypertime. However time is affected
by time travel, the Intelligence will see the unfolding of events hyperchronologically in
the passage of hypertime. For example, suppose that a time traveler “rewinds” reality
from t5 to t3. And suppose that the time traveler lives out her years happily in t3,
“regenerating” existence from t3 onwards. From within the temporal manifold, events
       

occur in the following order: t3, t4, t5, t3, t4, t5. From the hypertemporal perspective,
events occur in the following order: ht3, ht4, ht5, ht6, ht7, ht8. Events whose temporal
orders are shifted in time nonetheless have a linear progression in hypertime.
Without time travel, time and hypertime agree on the order of events. According
to van Inwagian time travel, time and hypertime disagree on the order of events.
Consequently, events can also have different temporal and hypertemporal tenses: the
van Inwagian time traveler can attend an event that is in the past and in the
hyperfuture, such as Woodstock.
Similarly, van Inwagian time travel results in a durational discrepancy between
time and hypertime. Suppose that hypertime is finite and that it lasts for 1000
temporal units. Normally, the quantity of existence of the growing block universe
grows steadily in such a way that time and hypertime agree on its duration: when
existence is 400 temporal units long, hyperexistence is also 400 temporal units long.
But time travel in which a portion of the growing block is erased generates hyper-
temporal duration that exceeds temporal duration. Consider our time traveler’s
erasure of the span of existence between 1969 and 2013. The duration of time extends
to 1969, but hypertime marches on past hyper 2013—forty-four more hyperyears
than the temporal manifold (assuming that is the only erasure).

4.2 The Traditional Model Versus Movable

Objective Present
4.2.1 The Traditional Model: Relocation
Suppose that Bianca travels to 1969 from 2013. According to the traditional model of
time travel, she removes herself from the present day and relocates herself to 1969,
leaving the extant temporal manifold intact. The traditional model conceptualizes travel
through time as similar to everyday travel through space: just as one does not change the
fundamental nature of space by changing spatial locations, one does not change the
fundamental nature of time by changing temporal locations.2 One space-travels by
moving through space, but not by changing space itself. On the traditional model of time
travel, one travels through time by moving through time either backwards or forwards,
but does not change the fundamental temporal manifold by doing so.
For example, when Bianca leaves 2013 for 1969, she leaves one temporal location
behind (2013) and arrives in another temporal location (1969).3 Like travel through
space, the mechanism of traditional time travel is simple relocation in which the time
traveler removes herself from one part of the temporal manifold and places herself
at another. Just as one does not destroy or affect the rest of the spatial manifold
by driving to a coffee shop, one does not destroy or affect the temporal manifold by
moving from one location to another.

2
One might change the size and shape of spacetime by time-traveling in this way, but not the
fundamental nature of time itself.
3
Here I set aside well-known issues involving “second time around” time travel, i.e., the idea that the
time traveler must have always been in 1969.
  

Importantly, Bianca also leaves others “on the ground” while she travels. Barring
well-known paradoxes involving parent and grandparent-killing, she does not
necessarily alter the existence of others in 2013 merely by traveling through time.
Rather, Bianca leaves others “behind” in 2013 as she relocates herself to 1969, much
as I leave my home behind when I head for work. This feature reflects the traditional
conceptualization of time travel as a kind of relocation of the time traveler similar to
spatial relocation.
According to Lewis (1976), time travel also involves a disagreement between
personal time and external time. Personal time, roughly speaking, is the time on
the time traveler’s watch. External time is the time elapsed on the basic timeline.
Suppose that travel from 2013 to 1969 takes Bianca exactly one year. One year of
personal time has elapsed for Bianca for the fifty-three external years she has traveled.
Similarly, the van Inwagian model involves a disagreement in the durations of time
and hypertime: the time traveler relocates the edge of time, while hypertime lurches
forward.

4.2.2 The New Model: Movable Objective Present (MOP)

I propose that the distinctive feature of van Inwagen’s model is the time traveler’s
control over the location of the objective present. When the time traveler flips the
switch in the time machine, she does not simply relocate herself to a different spot in
the temporal manifold, as on the traditional model. Rather, she changes the location
of the objective present for the entire temporal manifold: she shifts the location of the
objective present to the past of the temporal manifold, as hypertime marches on.
I call this sort of time travel movable objective present (hereafter: MOP). MOP is
unique because the mechanism of time travel involves a change in the location of the
objective present in addition to relocation of the time traveler.4
Before exploring MOP in further detail, it will be helpful to separate the key
features of MOP from van Inwagen’s model. The latter proposes that the time
traveler annihilate the time slices between 1969 and her initial moment of travel in
2013. But annihilation is not essential to MOP: MOP only necessitates the time
traveler’s control over the location of the objective present. A time travel machine
might also generate time slices into existence, as in scenarios involving time travel to
the future.5 The movement of the objective present doesn’t necessitate a change in the
duration of reality. Nor does MOP require the acceptance of the growing block theory
of time. The presentist MOP traveler travels in time by relocating the exclusive slice of
existence to her desired temporal destination in hypertime. The moving spotlight
MOP time traveler travels by relocating the spotlight of the present to her desired
destination in the eternalist block. I will explore these different conceptions later in

4
Technically, the location of the present need not be under the control of the time traveler herself. For
example, there could be a “jumpy” objective present that spontaneously switches locations in hypertime,
and is not under a time traveler’s control. Here I set aside such cases.
5
Hudson and Wasserman (2010) suggest this as well.
       

further detail, but for now the important points are that MOP requires neither
annihilation of existing time slices nor the growing block theory of time.
MOP requires the existence of an objective present. Presentism, moving spotlight,
and growing block agree on its existence, but disagree on its nature. I do not take a
stand on this important issue here, but I do make a few assumptions. First, I assume
that the objective present is causally receptive: those existing within it are able to
push, pull, move, and change features of reality. The present reality is dynamic and
changeable. Whatever distinguishes the objective present from other temporal loca-
tions, one of its distinctive features is its receptivity to modification and change.6
Looking down at the temporal manifold, the Intelligence thus sees at least one
temporal part of reality glimmering with change and activity. Second, I assume
that the objective present is maximally temporally thin: it is the smallest unit of
time rather than, e.g., a year-long temporal part of the block.
MOP requires the time traveler’s ability to “grasp” and control the location of the
objective present. Admittedly, the mechanism of such “grasping” is unclear: MOP
compounds the conceptual and metaphysical oddness of time travel by requiring that
the traveler have a handle on the entire present, not just on her own temporal
location. The MOP traveler’s machine is thus more powerful than the traditional
time travel machine: it must control the objective present in addition to the temporal
location of the time traveler.
MOP time travel has two key features: movability of the objective present, and
relocation of the time traveler. Grasping onto and relocating the objective present differs
from the traditional model in that the traveler shifts time with her rather than solely
shifting her location in time. According to the traditional model, the occurrence of time
travel doesn’t necessitate a wholesale change for the rest of reality. On MOP, the very
mechanism of time travel is a fundamental change in the location of the objective
present.
MOP must include relocation of the time traveler in addition to the shift of
the objective present. Merely “rewinding” the block doesn’t guarantee that the time
traveler is relocated, since the block rewinds to a location that might predate the
existence of the time traveler and the time machine without her inclusion. Moreover,
if the time traveler shifts the objective present without including herself, the
result is that reality “resets” to the prescribed location in the temporal manifold
and reality proceeds as it would have absent the time traveler and her machine.
Given determinism, it is the inclusion of the time traveler that regenerates reality
differently.
Let’s look at an example. Suppose that a MOP traveler who lives in 2013 sets her
time machine to return to 1969. She must flip two switches: first, the switch that
relocates the objective present from 2013 to 1969; second, the switch that preserves
the time traveler and her machine as the present is relocated. Why the latter? Because
relocating the objective present to 1969 only guarantees that the universe is “reset” to
that time—a time that might predate the existence of the time traveler and her time

6
Whoever is in the objective present, can change it. A time traveler can change the past, according to
MOP, because she can relocate the objective present in order to render it susceptible to change.
  

machine. For MOP to count as time travel rather than mere resetting, it must involve
inclusion of the time traveler.
Supposing that both switches have been flipped, the traveler moves herself and the
objective present to 1969. The MOP time traveler is selfish: she moves the location of
the objective present for everyone, rendering many of the other inhabitants of 2013
inert or non-existent. If time travel takes time, MOP entails that the traveler’s
personal time en route adds to the hypertemporal duration of the MOP traveler’s
trip. Suppose that the MOP machine takes one year to do its work, and that the time
traveler leaves 2013 for 1969. Upon the time traveler’s arrival, the hyperpresent will
be located in hyper 2014, even as the present is (re)located in 1969.
Three questions: What are the time traveler’s abilities in the relocated objective
present? What happens to the rest of the temporal manifold after MOP occurs? And
what does the Intelligence see? Answers to all three questions depend on the theory of
time in question. I now turn my attention to more specific models of MOP: presentist
MOP, moving spotlight MOP, and growing block MOP.

4.2.3 Presentist MOP

It is generally thought that time travel is incompatible with presentism, for the reason
that no destination exists to which the time traveler can depart.7 At best, she has no
destination to which to travel; at worst, the time traveler effectively kills herself by
flipping the switch in the time machine. This is commonly known as the “No
Destinations” problem for time travel in a presentist universe.8
MOP defies this dogma. For the presentist who accepts hypertime, MOP consists
in the time traveler’s relocation of the unique slice of existence to a different point in
time. Suppose that the time traveler hops into the time machine in 2013 and sets her
destination to 1969. According to presentist MOP, the time traveler “rewinds” the
objective present to 1969, allowing her to change events as she sees fit. As on van
Inwagen’s model, (i) only the time traveler’s memories postdate 1969, (ii) reality
progresses from the time traveler’s arrival. The presentist MOP traveler can change
past occurrences as she chooses without creating a paradox. Suppose that, remem-
bering her sadness over failing to attend Woodstock, she travels back to 1969 to
attend. Future events with which the “initial” Woodstock would conflict, such as the
consequent sadness, do not exist; Bianca “resets” reality at 1969, allowing it to
progress according to the causal results of her attendance at Woodstock.
The Intelligence sees individual slices of existence successively pop into existence
in hypertime. Temporal locations are repeated, but hypertemporal locations are
hyperchronological. The Intelligence views events in hyperchronological order. Sup-
pose that Woodstock occurs in 1969 (t1); that Bianca’s sadness over not attending
Woodstock begins in 1970 (t2); that her eventual time travel occurs in 2013 (t3); that
she attends Woodstock after MOP travel to 1969 (t1); and that she is happy in
second-time-around 1970 as a result (t2). The Intelligence views these events in the
following hyperchronological order: t1/ht1, t2/ht2, t3/ht3, t1/ht4, t2/ht5.

7 8
See Hales (2010). See Grey (1999) for one prominent version of this argument.
       

Since presentist MOP “resets” the objective present, its effect on the extant
temporal manifold bears similarity to van Inwagian time travel with annihilation:
the result of the time traveler’s voyage is the removal of the extant times from
existence, leaving events to regenerate differently the second time around. For
example, suppose that post-MOP Bianca cleverly picks up a lottery ticket 1970,
already aware of the winning numbers from her careful study of lottery history in
2013. Winning the lottery generates no paradox, since there are no existent events
(Bianca’s financial concern in 1972, for example) with which the event would
conflict.

4.2.4 Growing Block MOP

In a growing block universe, the MOP traveler sets her time machine to her desired
destination and moves herself and the objective present to a different temporal
location. Growing block MOP differs from van Inwagen’s view in that slices of
existence are not necessarily annihilated: the objective present is relocated, but the
quantity of temporal existence remains intact.
What happens to the growing edge of being when the present disagrees with the
hyperpresent? Here are several options. First option: Being stops its growth the
moment the time traveler leaves, with hypertime marching forward past the bleeding
edge of Being. Second option: Being is insensitive to time, such that a changeless,
dark portion of reality continues to grow in step with hypertime past the moment of
the traveler’s departure. Third option: contra MOP, the block is annihilated by the
time traveler’s relocation by necessity. Which option obtains depends on the rela-
tionships between the objective present, the hyperpresent, and the growing edge of
being. If the growing edge of being just is the objective present, then van Inwagen’s
picture (the third) is correct by metaphysical necessity. If hypertemporal location
directly following the objective present is only necessary for change, then option two
is correct: a changeless, dark reality continues to grow in step with hypertime,
unilluminated by the possibility of change conferred by the objective present. Finally,
if the objective present is what brings reality into existence, then the first option
obtains: moving the objective present behind the bleeding edge of being halts the
movement and generation of reality.
What are the time traveler’s abilities, given the existence of the past? A critical
assumption of MOP is that the objective present confers casual receptivity: what
exists within the objective present is changeable and dynamic. Suppose that Bianca
moves herself and the objective present from 2013 to 1969, causing hypertime to
creep past the bleeding edge of Being. And suppose that previously, 1969 included
neither her arrival via time machine nor her attendance at Woodstock. But if the
objective present is causally receptive, Bianca can attend Woodstock the second time
around. How can this occur given the existence of the past?
Growing block MOP requires that the time traveler regenerate a new block of
reality that “saves over” the existing temporal manifold. Given the acceptance of
hypertime, the idea is that there is a new block at every hypermoment, and that there
can be replacement blocks at successive hypermoments. Each new replacement block
determines all the facts about the past and the present. In giving a characterization
  

of changing the past in an eternalist universe, Hudson and Wasserman state this
idea well:
Suppose that at ticktock 100 an eternalist block is hyperpresent and determines all the facts
about the past and the future (relative to each of its time-slices). Further suppose that at
ticktock 101, with the sole exception of some slice, S, a new eternalist block has replaced the old
one [ . . . ]. The items that characterize the eternalist block at ticktock 101 determine all the facts
about the past and the future (relative to each of its time-slices, as well) [ . . . ].9

The key feature of the model is that the block generated by the time traveler
hyperfollows the existing block in which the past is fixed. In the eternalist model, the
new block fixes all facts about the past, present, and future. This feature can be
adapted to growing block by holding that each new block determines the facts about
the past and present, until the hypernext block is generated.
Here’s how it works. Call Bianca’s failure to attend Woodstock state W , and
Bianca’s attendance at Woodstock state W+. The Intelligence sees the block in state
W at hypertime 1, Bianca’s time travel at hypertime 2, and state W+ at hypertime 3.
At hypertime 2, a new block is generated in place of the old one. The Intelligence sees
the appearance of a new block hyperafter the existence of the hyperolder block that
contains state W-: at hypertime 2, all of reality is replaced by a new, maximally
determinant block. As Hudson and Wasserman put it:
. . . [At] ticktock 99, it was true both that you did not have the fabulous past you’d
been dreaming about and you will not have the glorious future you’d been hoping for, for at
ticktock 99, it is true that the machine will not work. [But] at ticktock 101, it was true both that
you did have the fabulous past you’d been dreaming about and you will have the glorious
future you’d been hoping for, for at ticktock 101 [ . . . ] the machine did work.

In other words: the block determines all of the past and present facts, but a
different hypersuccessive “replacement” block determines all of the past and present
facts at the next hypertemporal moment.
Above, I discussed how MOP compounds the conceptual oddity of time travel
by requiring that the time traveler’s machine have control over her temporal
relocation and over the location of the objective present itself. For growing block
MOP, the time traveler’s machine must also have powers of block regeneration:
broad powers indeed, but not necessarily a further stretch beyond traditional
time travel. What MOP loses in requirements on the abilities of the time traveler
and time machine, it gains in rejection of well-known paradoxes about changing
a fixed past.

4.2.5 Moving Spotlight MOP

Moving spotlight MOP consists in the time traveler’s control over and movement of
the “spotlight” to her desired temporal location. According to moving spotlight
MOP, the mechanism of time travel is the illumination or reillumination of a location

9
Hudson and Wasserman (2010: 6).
       

in the block. MOP causes a disruption in the successive temporal order of the
objective present.
To illustrate, let us return to Bianca, who desires to use her time machine in 2013
in order to attend Woodstock in 1969. Suppose, as before, that Woodstock occurs in
1969 (t1); that Bianca’s sadness over not attending Woodstock begins in 1970 (t2);
that her eventual time travel occurs in 2013 (t3); that she attends Woodstock after
MOP travel to 1969 (t1); and that she is happy in second-time-around 1970 as a
result (t2).
Without time travel, the location of the objective present is successively ordered.
For example, it is located at t1, at t2, at t3, and then at t4. With time travel, temporal
locations of the objective present are repeated: the objective present is located at
1969 several times over. The Intelligence views the location of the objective present
in the following hyperchronological order: t1/ht1, t2/ht2, t3/ht3, t1/ht4, t2/ht5. As
Bianca arrives in 1969, the Intelligence sees the objective present “jump” backwards
in the block.
As on the growing block model, moving spotlight MOP requires that the hyper-
present block regenerate and replace the old block in hypertemporal succession. At
t3, Bianca relocates the illuminating spotlight of the present to 1969. Upon her
arrival, a new hyperpresent block is generated that contains both her arrival and
her attendance at Woodstock: the shared slice between the hyperearlier block and the
block that replaces it.10 At each hypermoment, a spotlight illuminates a portion of the
block. With the addition of time travel, the block is replaced by a new block and,
presumably, a new spotlight: the spotlight shone on the time traveler.

4.3 MOP Errors and Extant Questions

The possible errors of the MOP mechanism are many. First, we can imagine the all-
important first switch making a mistake and taking too little or too much of present
reality with it. Suppose that Bianca jumps into her time machine in 2013 and flips
both switches. But unbeknownst to her, the temporal relocation switch malfunctions:
rather than transport Bianca and her time machine to 1969, it transports only the
time machine. Bianca is out of luck: the objective present has been relocated without
her and her time machine is missing. More seriously, if her malfunction occurs in a
presentist universe, she has taken her Woodstock-desiring self out of existence, since
the objective present is relocated to the past without her. If the malfunction occurs in
a growing block or a moving spotlight universe, then the slice of reality shared by the
hyperolder block and the hypernew block shares the time machine but lacks Bianca.
Either way, the past has been changed, but only by the presence of the time machine:
Bianca has lost her chance.
Or suppose that the temporal relocation switch malfunctions in the other direc-
tion, and brings, along with the time machine, the maximal spatial slice of January 7,
2013. That is, suppose that the malfunctioning relocation switch essentially

10
Since the goal is to apply MOP to various theories of time, I assume with Hudson and Wasserman
that we can make sense of multiple hyperpresent blocks. Elsewhere (Bernstein In Preparation), I explore
the coherence of this idea and suggest that it can be fleshed out using hyperhypertime.
  

duplicates the state of the universe on January 7, 2013 in 1969: the Vietnam War has
ended, the microchip has been invented, and the human genome has been fully
sequenced. If this occurs, then 2013-in-1969 progresses largely as hyper-2013 would
have progressed, save for the addition of the time traveler and her machine. Suppose
that Bianca sets her faulty time machine to 1969 in 2013 (t3), transports the spatially
maximal time slice of January 7, 2013 to 1969 (t2), and lives out the confusing reality
in 1969 (t1). The Intelligence views the unfortunate scenario in the following way:
1969 occurs the first time around; Bianca spatially transports the maximal slice of
January 7, 2013 to 1969; the world evolves largely as it was going to if the time travel
didn’t happen.
Another possible serious MOP error is the existence of competing time travelers
whose desired locations for the objective present conflict with each other. This is a
special problem for MOP, since the mechanism of time travel relies on the relocation
of the unique objective present rather than the relocation of the traveler alone. Here is
how such a scenario would go. Suppose that in addition to Bianca, there is a second
time traveler, David, in possession of a relocation-and-objective-present movement
device. On January 7, 2013, Bianca sets her device for 1969 and David sets his device
for 1950. One reading of the results involves multiple compossible objective presents:
multiple causally receptive time slices within which the time travelers can generate
change. Given the possibility of logically incompatible changes incurred by each time
traveler, I leave it as an extant question whether such a scenario is metaphysically
possible.
MOP raises additional extant questions about the relationships between the
objective present, the passage of time, existence, and hypertime. The conceptual
possibilities of MOP show that these features hang together and come apart in
interesting and intricate ways. To what extent does the passage of time determine
the location of the objective present? Does the objective present bring reality into
being, or simply march in step to its generation? MOP doesn’t answer these ques-
tions, but modeling it within different theories of time shows that the possibilities are
more subtle than previously imagined.

4.4 Conclusion
The underlying mechanism behind van Inwagen’s model is the movable objective
present. The idea is that the objective present is under the control of the time traveler
and that she can move it anywhere in the temporal manifold. In contrast to the
traditional model that conceptualizes time travel as a type of temporal relocation,
MOP involves the time traveler’s movement of the entire objective present in
addition to her temporal relocation. Given the acceptance of hypertime, MOP is
generalizable to presentist, growing block, and moving spotlight theories of time.
While MOP is slightly different for each, it allows the metaphysician to believe in the
possibility of time travel without assuming a particular theory of time.
MOP also reveals the intricate relationships between the objective present, the
passage of time, existence, and hypertime. While time travel is normally considered
an outré test case for theories of time, MOP reveals that time travel can be utilized as
a novel conceptual tool for modeling these relationships in further detail.
       

Acknowledgments
I owe thanks to audiences at the 2014 Pacific APA and the Gargnano Philosophy of Time
Conference for feedback on this paper. I am also grateful to Melissa Schumacher and Heather
Wallace for helpful comments. The work on this paper was completed at Duke University.

References
Bernstein, S. (In Preparation), ‘Time Travel and Hypertime’.
Grey, W. (1999), ‘Troubles with Time Travel, in Philosophy 74, 55–70.
Hales, S. (2010), ‘No Time Travel for Presentists’, in Logos and Episteme (2): 353–60.
Hudson, H., and Wasserman, R. (2010), ‘Van Inwagen on Time Travel and Changing the Past’,
in D. Zimmerman (ed.), Oxford Studies in Metaphysics, v.5 (Oxford University Press).
Lewis, D. (1976), ‘The Paradoxes of Time Travel’, in American Philosophical Quarterly 13 (2):
145–52.
Van Inwagen, P. (2010), ‘Changing the Past’, in D. Zimmerman (ed.), Oxford Studies in
Metaphysics, v.5 (Oxford University Press).
PART II
Freedom
 5
The Disconnect Problem and
the Influence Strategy
Mark Heller

In 1980 I had the extreme good fortune of taking a seminar from Peter van Inwagen
built around his manuscript on free will. Soon thereafter that manuscript was
published as An Essay on Free Will,1 reshaping the free will literature for the decades
to follow. I was also fortunate enough to later take a class from him on what became
Material Beings,2 which shaped a different literature and much of my career. Later,
van Inwagen was kind enough to supervise my dissertation. It is fair to say that I am
the philosopher I am in large part because of Peter van Inwagen, and I am very
grateful to him. In the present paper I return to my early free will roots. In particular,
I will focus mostly on a problem for libertarians, attempting to undermine one
common strategy that libertarians employ to answer the problem. But my closing
sections will explore broader conclusions.
There is a certain well-known problem facing libertarians—the so called “luck
problem.” If an event is undetermined, then it is random, and random events are not
within anyone’s control. A tad more carefully, if an event is undetermined then it is
not determined by the agent’s reasons, and this disconnection has the consequence
that it is just luck when an undetermined event appropriately corresponds to the
agent’s reasons. That, in turn, seems to have the consequence that the undetermined
event is not something that the agent controls or for which the agent is responsible.
There is a standard response to this problem—an agent can control her actions by
virtue of her reasons “influencing without determining” her decisions.3 I will propose
that if some influencing is good, then a little more influencing is always better. I will
further propose that that leaves the libertarian with no explanation for why influen-
cing is good but determining is bad. Thus, my initial conclusion: if you are going to
be a libertarian, you should reject the Influencing Without Determining strategy.
If we combine this conclusion with the premise that the Influencing Without
Determining strategy is the libertarian’s only plausible way to answer the luck

1 2
Van Inwagen (1983). Van Inwagen (1990).
3
Leibniz famously suggested that motives “incline without necessitating.” See Alexander (1956: 57).
I do not mean to impose any interpretation on Leibniz’s words. In particular, I do not mean to suggest that
Leibniz endorsed the “influencing without determining” strategy that is the focus of my paper.
  

problem, then we reach a bolder conclusion: libertarianism should be rejected. I will

not defend that additional premise here, but I will poke at it a bit in section 5.1 and
loosely explore the bolder conclusion and its consequences for free will in the later
parts of the paper. Libertarianism is an attempt to ground free will in indeterminism.
The attack on libertarianism is that in fact the indeterminism detracts from freedom
rather than supporting it. But this attack, it will turn out, depends on accepting van
Inwagen’s eminently plausible principle Beta, the very same principle that is at the
core of his attack on compatibilism. Thus we are led to the even bolder conclusion
that free will is incompatible with both determinism and indeterminism.
I will close the paper by proposing a contextualism about the term ‘free will’ which
will fend off the impossibilism suggested by the double incompatibility. I will suggest
that the term ‘free will,’ and related terms like ‘control’ and ‘up to us,’ are governed in
many contexts by principle Beta but that there are other contexts in which those
terms pick out a property that would violate Beta. However, be forewarned, this
contextualism is not the great savior it appears to be. The Beta-violating property that
is sometimes denoted is compatible with determinism, but it does not give us
everything we want from free will. In particular, that property does not ground
retributivist punishment or related reactive attitudes. In this regard, the view pro-
posed here is very much in the same family as Derk Pereboom’s.4 To the extent that
our current views about our place in the world and our current behavior are built
around retributivism, genuine change is called for.

5.1 The Freakish Demon

Demons and their ilk are not uncommon in the philosophical literature. In criticizing
libertarianism, R. E. Hobart has given us the freakish demon:
In proportion as an act of volition starts of itself without cause it is exactly, so far as the freedom
of the individual is concerned, as if it had been thrown into his mind from without . . . by a
freakish demon. (Italics removed.)5

There are many different elements present in this short quotation. It is tempting to
emphasize the freakishness of the demon. The charge of randomness that is often
leveled against libertarianism fits with such an emphasis. Mele’s version of the luck
objection underscores how freakish an agent’s actions would seem in a world in
which the agent acts contrary to her reasons.6 And, in “Free Will Remains a
Mystery,” van Inwagen provides an image of replaying an indeterministic sequence
over and over, and the resulting random pattern that would ensue.7
But I suggest that the freakishness is a red herring, distracting from Hobart’s
primary concern. What should be emphasized instead is the disconnect between the
agent’s volition and what Hobart calls her “character.” Having undetermined voli-
tions is like the demon case in the following respect: in neither case does the volition
arise from what the agent is like, cares for, or feels allegiance to; it does not come out

4
See Pereboom (2001). I find myself agreeing with much of what Pereboom says. My biggest disagree-
ment is that he is friendlier to the Agency Theory than I, though in the end he too rejects it.
5 6 7
Hobart (1934: 7). See chapter I, section 1, of Mele (2006). Van Inwagen (2000).
        

of her. The volition is disconnected from any reasons the agent might have for that
volition. It is “as if it had been thrown into [her] mind from without,” as Hobart says
in the quotation.
Van Inwagen does a nice job of emphasizing this disconnect with a slippery slope
argument that he offers on behalf of his opponent.8 Before presenting the slippery
slope argument, one caveat is needed. Van Inwagen is here considering an argument
with the conclusion that an undetermined movement is not an act at all. But the
conclusion I am interested in is that, conceding that it is an act, it is not a free act—it is
not within the agent’s control, it is not up to the agent, and the agent is not responsible
for it. I will present the argument as if that is the conclusion. I will also modify it
slightly to fit the present context, and I will abbreviate it down to its bare bones.
We start with a more detailed version of Hobart’s freakish demon case. Case 1: The
demon is playing a keyboard. There is a wire running from the keyboard, through the
agent’s skull, and into her brain. The wire is connected to the brain in such a way that
the input on the keyboard determines what will happen to particle P, and P’s
behavior either is the volition of which Hobart speaks or determines that volition.
It is clear that in this case the agent is not in control of the movement of particle P or
the action that results from that movement. Now we start down the slope. Case 2:
Add the stipulation that the demon’s actions themselves are not determined. Case 3:
Eliminate the demon and make the keyboard itself an indeterministic mechanism
sending signals down the wire. Case 4: Eliminate the keyboard and have the signals
indeterministically appear within the wire and then move into the brain. Case 5:
Shorten the indeterministic wire so that it is entirely internal to the brain. Case 6:
Make the indeterministic “wire” composed of brain cells. Case 7: Add that the
indeterministic “wire” is a natural part of the brain. The challenge to the libertarian
is that no difference between any pair of these cases is the sort of difference that could
introduce agential control into the picture. This is a good challenge.
Thus we have what I will call the “Freakish Demon argument” for the conclusion
that we have no control over undetermined events:
i. An undetermined event is like one caused by a freakish demon.
ii. Agent A has no control over an event that is caused by a freakish demon.
Therefore: Agent A has no control over an undetermined event.
The slippery slope supports premise i. Agent A could be any of us and one undeter-
mined event is like any other as far as the Freakish Demon argument goes.
What makes me especially fond of this slippery slope is that it points to discon-
nection as the root of the challenge. When we think about Cases 1–7, the feature of
indeterminism that seems to be doing the work is not the freakishness or randomness
of the resulting behavior. What is doing the work is that the agent’s action is divorced
from her character, deepest self, or reasons. In the initial Case 1, the freakish demon
may just happen, by pure chance, to have the same desires for the agent as she has for
herself. Her resulting behavior would look just like what she would have done on her

8
Van Inwagen’s presentation of the freakish demon argument runs from pp. 130–3 of van Inwagen
(1983), with the slippery slope variant beginning on the bottom of p.131.
  

own, but it would not be in her control because it was coming from “without.” It is
this externality to the self that is carried over through Case 7. If the start of a chain
leading to a volition is disconnected from Agent A’s character, then moving that start
inside the head will not stop it from being disconnected. I will henceforth use the title
“the disconnect problem” to refer to the charge that indeterminism produces dis-
connection and that disconnection conflicts with freedom.
Van Inwagen’s response to the slippery slope is to propose that the naturalness
introduced in Case 7 seems to him the right sort of difference to at least stop it from
being clear that there is no act. But he explicitly avoids taking a stand on whether that
act in Case 7 is a free act. And when, in his next section, he considers a similar
argument that is directed at freedom rather than action, his response is not to point
to a flaw but instead to point to a conflict with other beliefs of which he is more
certain.
Consider, for instance, the following argument against free will:
1. An agent A has no control over any event that is determined by an event that
A has no control over.
2. A has no control over any event that occurred before A was born.
3. A has no control over any undetermined event.
4. Every event is either undetermined or determined by some event that is
undetermined or determined by some event that occurred before A was born.9
Therefore: A has no control over any event.
Therefore: There is no free will.
The Freakish Demon argument supports premise 3. Premise 1 depends on van
Inwagen’s principle Beta.10
Np & N(p→q) ⊢ Nq
where ‘Np’ abbreviates “p, and no one has or ever had any choice about whether p”.
Without premise 1, or some other descendent of Beta, we could not get from 3 to
the conclusion that an agent doesn’t have any control of her actions in any of Cases 1
through 7. Van Inwagen is so confident that the conclusion of the anti-free will
argument is false, and so certain that Beta is valid, that it is reasonable for him to
reject the less certain premise 3, despite being able to point out no specific flaw in the
disconnect problem.11 However, I do not share his rankings. So, for the present,
I endorse the above as an argument against free will.
I close this section with the acknowledgment that Beta is not as central to the
incompatibilist program as it once was. Partly this is because Beta has been taken to

9
I am using ‘undetermined’ to mean not determined by prior events. Thus, if there are any agent-
caused events in the spirit of the Agency Theory, they would count as undetermined. If you disapprove of
this use of the term, the entire discussion could easily be rewritten to accommodate your use of the term.
10
Van Inwagen (1983: 94).
11
To be fair, my emphasis on the disconnection was not his emphasis, and he may well be able to point
to a specific flaw in my appeal to the disconnection. But he has supported premise 3 himself, or its
equivalent, and at least against his own defense he sees no specific flaw. See van Inwagen (2000).
        

be essentially committed to the Principle of Alternative Possibilities and that prin-

ciple has been taken to be threatened by Frankfurt-style counterexamples. Regardless
of what one thinks about the supposed counterexamples, this worry misses the
following point: variations on Beta that say nothing about alternative possibilities
are no less plausible than the original Beta. For instance, we could substitute for ‘Np’
the operator ‘p and no one has or ever had any control over p.’ This is only obviously
connected to PAP if we take for granted that ‘control’ should be understood in terms
of alternative possibilities, which should not be taken for granted. Or, when empha-
sizing responsibility, we might substitute for ‘Np’ the operator ‘p and it is not
anyone’s fault that p.’ I will not pursue these substitute principles here.12 In the
present context I will just say that any plausible argument for incompatibilism can be
adapted to fit the needs of the argument against free will that I presented above.13 The
bottom line is that those very reasons that give incompatibilism its plausibility, when
combined with the premise that no undetermined event is under anyone’s control,
lead to the conclusion that there is no free will at all.

5.2 Deliberation
If we have no free will, what else don’t we have? Well, it is old news that we don’t have
moral responsibility, at least not in a sense that would satisfy a retributivist.14 It is also
old news that we can preserve a lot of the other non-retributivist moral properties
that are in the moral responsibility family. But van Inwagen tells us that if we deny
free will we have to give up something more; we have to give up on deliberation.
Deliberation, he tells us, entails a belief in free will:
. . . all philosophers who have thought about deliberation agree on one point: one cannot
deliberate about whether to perform a certain act unless one believes it is possible for one to
perform it. (Anyone who doubts that this is indeed the case may find it instructive to imagine
that he is in a room with two doors and that he believes one of the doors to be unlocked and the
other to be locked and impassable, though he has no idea which is which; let him then attempt
to imagine himself deliberating about which door to leave by.)15

Then, describing Baron Holbach, who professes disbelief in free will, van Inwagen
insists that Holbach must believe in free will if he deliberates, which he certainly does:
A man who didn’t deliberate would either move about in random jerks and scuttles, or would
withdraw into catatonia. (That some catatonics are people who have ceased to believe in their
own free will is an interesting hypothesis.)16

12
It is worth mentioning that van Inwagen himself offers some substitute principles. One example is in
van Inwagen (1983: 104): “Np=df p and, in just the sense of having a choice that is relevant in debates about
moral responsibility, no one has, or ever had, any choice about whether p.”
13
In particular, I include Pereboom’s Manipulation Argument, beginning on p.112 of Pereboom
(2001).
14
I say that the dependence of retributivist responsibility on free will is old news. But that is not to say
that it is universally accepted, nor is it to say that I have an argument for it. It seems true to me, but I have
nothing non-obvious to say on its behalf. For the purposes of the present paper, I simply assert the
dependence.
15 16
Van Inwagen (1983: 154). Van Inwagen (1983: 157).
  

I say, though, that one can surrender one’s belief in free will without surrendering to
catatonia. One can deliberate, and rationally deliberate, without believing in free will.
We need to distinguish different ways in which one might lack free will—different
ways in which it might not be possible for one to perform an act. The difference
I have in mind has to do with the counterfactual effectiveness of my deliberation; it
has to do with what would happen if I were to give up on deliberating.17
If I were to believe that I lacked free will with respect to which door to exit through
on the grounds that I believe that one of the doors is locked, then I would not and
should not deliberate about which door to exit through. The deliberation would be
pointless. Regardless of the outcome of the deliberation, my exit would be through the
unlocked door. If I were to believe that I lacked free will on the grounds that I believed
that a freakish demon were controlling my volitions, then I would and should
stop deliberating. It would be pointless. No matter the outcome of my deliberations,
the demon would select a volition for me. If I were to believe I lacked free will on
the grounds that my decision is undetermined—wholly disconnected from my
deliberations—then I would and should stop deliberating. It would be pointless.
However, if I believed I lacked free will with respect to a particular action on the
grounds that I believed that incompatibilism and determinism are both true, I would
nonetheless deliberate about whether to perform that action, and I would be rational
to do so. The deliberation would not be pointless, because the deterministic chain
that produces that action produces it by way of the deliberation. Determinism is not
fatalism. If I were to deliberate to a different outcome or fail to deliberate at all, my
action would be different.18 As long as deliberation makes a difference it is not
pointless. And as long as I believe that deliberation makes a difference I have reason
to believe that it is not pointless. And as long as I have reason to believe it is not
pointless, I will and should continue to deliberate even if I believe that I have no free
will. And if my deliberating manifests anything about my beliefs, the most it
manifests is my belief that my deliberation will make a difference.
Agent A’s rationally deliberating about whether to do x does not require that she
believe it is possible for her to do x, in the sense of ‘possible’ that figures in the anti-
free will argument. That is, she does not have to believe that she has a choice about
whether to do x or it is up to her whether to do x or that it is in her control whether
she do x or that she has the ability to do x and the ability to refrain from doing x. We
are taking all of those central terms to be tied to principle Beta and hence to require
indeterminism for their instantiation (and also to require determinism according to
the anti-free will argument.) But at most a weaker kind of possibility is required for
rational deliberation. At most what is required is that A believe that x and not-x are
possible in the following sense: what she does makes a difference to whether x
happens. She does not have to believe she has the ability to bring about each of x

17
In grounding rational deliberation in counterfactual effectiveness I follow Pereboom (2008). He also
adds an ignorance condition, which I would be happy to endorse but I ignore for present purposes.
18
I suppose that if I were to fail to deliberate at all, I might, just by chance, end up with the same action
that I would get to by deliberating. But I am assuming here that if I fail to deliberate at all, the effect would
be a lack of action (in the spirit of van Inwagen’s catatonia claim). The relevant point, though, is that when
I do deliberate, my action tracks my deliberation.
        

and not-x. She has to believe she has counterfactual effectiveness with respect to each:
if she were to deliberate things would be different with respect to x than if she were
not to deliberate. More to the point, A should believe that if she were to deliberate
then her decision would reflect the outcome of that deliberation and that decision
would lead her to act in accordance with that decision. If she believes that, then she
has every reason to deliberate even if she doesn’t believe that she has the ability to
perform any action other than the one she is going to perform.
One might be tempted to raise a quite different worry about deliberation and free
will. This worry does not focus on the fact that the resulting action is not free
according to the anti-free will argument. Rather it focuses on the fact that the
deliberation itself is not free. One might worry that if A has no choice about whether
she deliberates or about how she deliberates, then it would be inappropriate for us to
apply normative terms like ‘rational’ or ‘reasonable’ to her act of deliberating. Just as
our moral judgments of actions as praiseworthy or blameworthy depend on there
being free will, so our normative judgments of actions, including the act of deliber-
ating, as being reasonable in the circumstances depend on their being free will, one
might suggest.
However, this worry is misguided. Giving up on free will does not force us to give
up on all normativity. Some states of affairs are still good and others bad, even if no
one is to blame for bringing about one of the lesser ones. We could distinguish people
who acted in accordance with the good from those who did not. And it is likewise for
intellectual norms. I want to say that it would be rational of me to keep deliberating as
long as my deliberation had counterfactual effectiveness. If you insist that the term
‘rational’ can no longer apply because it is too closely connected with permissibility
and, thereby, blameworthiness, then I surrender the word. It is still intellectually
better for me to deliberate if I believe in counterfactual effectiveness than if I don’t.
And that is the point of disagreement between me and van Inwagen.
And, so, I conclude this section. The price we pay for giving up free will need not
be as steep as it might first have appeared. Or at least, it need not be as steep for some
of us. Those who believe the world is deterministic in the way that the compatibilist
wants can comfortably accommodate deliberation. But, those who believe the world
is indeterministic in the way that the libertarian wants will have a harder time
accommodating deliberation. If there is a disconnect between the deliberation and
the volition, then deliberation does look pointless.

5.3 The Self

The compatibilist will reject Beta and premise 1 of the anti-free will argument.19 The
form of compatibilism to which I am most attracted is what Susan Wolf calls “Deep

19
Not all compatibilists are alike. In this paper, I will only be discussing compatibilists who deny Beta.
I do note that the only way for a compatibilist to accept Beta is to also accept the following claim: free
agents have a choice about either the past or the laws. David Lewis (1981), for instance, provides a weak
interpretation of ‘has a choice about’ that makes such a claim plausible. Of course, van Inwagen never
intended the weak interpretation, and he already knew that on Lewis’s weak interpretation we could have a
choice about the past or laws. The debate between van Inwagen and his opponents will be whether there is
  

Self Compatibilism,” according to which an action is free if it is properly caused by

the self—one’s self is the cause.20 Different Deep Self compatibilists will have
different accounts of what the self is. Hobart, somewhat simplistically, calls it the
“character,” which he, also simplistically, identifies with the agent’s dispositions to
act. It turns out to be no easy feat to fill in the correct details,21 but those details don’t
matter here, so long as we have a rough idea of the self as including things like the
agent’s desires (of the appropriate order), values, personality traits, etc. I am trying to
point towards those sorts of features that make us describe an agent as one sort of a
person rather than another. And I am assuming for the purposes of this paper that
these are the sorts of features of the agent that in normal cases ground reasons for
action. Henceforth, when I use the word ‘compatibilist,’ I will be speaking of the Deep
Self compatibilist.
This type of compatibilist will say that even if agent A has no control over having
the self she has, if that self is what determines her actions, then she has control over
her actions. She has control over the action, because what it is to have control over
the action is just for the action to be caused in the right way by the self, regardless of
how that self came about. She has no choice about whether she has that self rather
than a different sort, no choice about whether her having that self will lead her to that
action, but she has a choice about whether to perform that action. Thus, the Deep Self
compatibilist denies Beta. Though I find this general compatibilist approach to have a
good deal of plausibility, I am going to side with Beta and continue to endorse the
anti-free will argument, at least until I get to the contextualism section at the end of
this paper.
But, still, even if Beta wins over Deep Self compatibilism, it must be admitted that
the self-determination of which the compatibilist speaks is a pretty nice property to
have. Even if we have no free will, it is better to be a self-determining robot than a
robot whose actions are directly governed by a puppet master. Good people who have
the self-determination property will tend to do good things. And people who
have this property will tend to do the things they have most reason to do. Parents
who believe that their children have this property have good reason to try to shape
their children’s characters, and they have some reason for optimism that their
children will become self-fulfilled. If I can’t have free will, I’d still like to have self-
determination, and I’d like my child to have it too. And, as I argued in the preceding
section, believing that one has this property makes deliberation reasonable, even for
those who don’t believe in free will.
Just as compatibilists try to respond to the anti-free will argument by appeals to the
self, there is room for the libertarian to make a similar appeal in attempting to deny
premise 3 of the anti-free will argument. For instance, an agency theorist will start

a strong interpretation that still leaves Beta with its intuitive force while making it implausible that we have
a choice about the past or laws. All compatibilists must reject Beta under such a strong interpretation.
I thank a referee for pushing me to say more here.
20
See Wolf (2003: 372–87), and her earlier discussion of the idea in Wolf (1990), where she preferred
the term ‘real self.’ If you prefer the more general title “Actual Sequence Compatibilism,” that is fine with
me. I find the emphasis on the self to be more evocative.
21
Certainly we can learn from Frankfurt (1971) and from Fischer and Ravizza (1998).
        

with a different view of the self—a bare self.22 She can say that that undetermined
event is caused by the self. Let us set aside any disputes about whether this “causation
by the self” is a plausible kind of causation. I want to ask instead about the value of
this causation.
There are some senses of ‘random’ and ‘luck’ such that this sort of causation by the
self would seem to block the charges of randomness and luckiness. Agent A’s volition
is hers because it is self-caused. It does not just come upon her in a fit; she produces it.
It is not random that she chose one path over the other, even though it is undeter-
mined; it was her choice.23 It was not just lucky that she behaved the way she did;
it was her choice to do so. It was her choice because she was the one who made it
happen.
But none of this does anything to lessen the force of the disconnect problem.
Stipulating that A’s volitions are caused by the self in the kinds of ways we’ve been
considering does nothing to increase the connection between those volitions and A’s
reasons for action. It does nothing to lessen the force of the freakish demon’s slippery
slope. The causing of the volitions in the stipulated way might make the action more
certainly A’s own, but it does not decrease the sense that the volition is just as
external to the sort of person the agent is as it would be if it were caused by the
freakish demon. Imagine an agent who deliberates in accordance with her reasons
and then has no connection between that deliberation and her willing. Her willing is
entirely divorced from the careful deliberations she just made. There is simply no
point in such an agent deliberating, and, given the divorce, it is hard to see how we
could hold the agent responsible. Her willing may well have been caused by her, but
the type of causation does not endow the willing with any power to reflect the reasons
and deliberation that came before.24
The compatibilist’s kind of causation by self—what I called “self-determination”—
is different, because the reasons, or the sorts of things that ground reasons, were built
into the compatibilist’s conception of self. If I have the property of self-determination
that the compatibilist is fond of, I deliberate the way I do because of the sort of person
I am and the sort of reasons I have, and those reasons then play out, through my
deliberation, into my action. In the next section I will turn to the Influencing Without
Determining strategy, and that will allow the libertarian to provide a role for reasons
in the production of volitions. The present point, however, is that without that

22
The Agency Theorist that I describe in the present section is a bit of a fiction. The more realistic
characterization of the Agency Theorist would include an appeal to the Influencing Without Determining
strategy that I will discuss in the next section. For instance, Chisholm’s classic (1996: 11–44) makes an
attempt at the strategy (and re-popularized the Leibniz quotation cited in note 3).
23
Here I am trying to capture a certain libertarian intuition about how randomness is avoided by
claiming ownership of one’s acts by causing them. But there are also forceful reasons to insist that
randomness would still be a problem. Recall van Inwagen’s discussion of promising in van Inwagen
(2000: 17–18), developed further in his later van Inwagen (2011: 475–83).
24
Yishai Cohen has made me realize that the slippery slope would need to be reconstructed in non-
trivial ways in order to incorporate the Agency Theorist’s self into the series. However, it would not, say I,
alter the force of the slippery slope. As long as we see the freakish demon argument to be about
disconnection rather than about ownership of one’s acts, the Agency Theory can’t help. It is not enough
that the act be connected to a self. It must be connected to the features of an agent that make her the sort of
person she is.
  

strategy the disconnection is built into the indeterministic picture of how a self
causes. Whatever it is that makes A the source of her own volition, it is, so far, not
something that allows the reasons to reach across the indeterministic gap to affect the
volition. That is what is needed to solve the disconnect problem and to ground
deliberation.
I close this section by asserting the following conditional: If you thought the
disconnect problem was a serious threat to a libertarian account of freedom, then
you should continue to think so even after we concede a libertarian account of how
an undetermined event is caused by a self. And, just to be explicit, I am including
among the libertarian possibilities an account of self that is derived from the Agency
Theory. There is much that makes the Agency Theory metaphysically special in its
treatment of free will. But agent causation simply offers no assistance in attempting
to solve the disconnect problem.

5.4 Degrees of Influence

It is time to turn to the Influencing Without Determining strategy (henceforth, the
“Influence” strategy). Here we attempt on behalf of the libertarian to solve the
disconnect problem by asserting a connection. Though our reasons, or our other
character traits that ground our reasons, do not determine our decisions, they still
influence our decisions. The idea is to make it come out true that A acted for a reason
while still making it false that the reason determined her to act. A did it because of the
reason she had, but the reason wasn’t enough to make her do it.
The problem for this Influence strategy is that it seems that if some influence is
good, a little more influence would always be better. Here I highlight a different
element of that short quotation from Hobart:
In proportion as an act of volition starts of itself without cause it is exactly, so far as the freedom
of the individual is concerned, as if it had been thrown into his mind from without . . . by a
freakish demon. (Italics in original.)25

Hobart is making room in his metaphysics for a kind of influencing relation that falls
short of determinism. He is assuming that this influencing comes in degrees, and he is
asserting that the lower the degree of influence between A’s reasons and her volition,
the more it is like the freakish demon scenario. Likewise, he holds the inverse: the
higher the degree of influence, the less like the freakish demon. So, at least with respect
to the disconnect problem, the more influence the better. Earlier in his paper, he
asserted a proportionality thesis directly between determinism and free will:
I am not maintaining that determinism is true; only that it is true in so far as we have free
will. . . . [I]t is not here affirmed that there are no small exceptions, no slight undetermined
swervings. . . . All that is here said is that such absence of determination, if and so far as it exists,
is no gain to freedom, but sheer loss of it. . . . 26

I will now try to develop Hobart’s proportionality intuition.

25 26
Hobart (1934: 7). Hobart (1934: 2).
        

The indeterministic picture, as we have been imagining it so far, posits an event

E that is undetermined and is the first event in a deterministic chain that leads to A’s
action. Hobart is assuming that that event is A’s volition, but the disconnect problem
doesn’t depend on any assumption about what E is.27 We will continue talking as if it
is a volition. Prior to that undetermined volition there are all the events that ground
the sorts of properties that the compatibilist cares about: A’s character traits, her likes
and dislikes, her values, her beliefs—everything that determines how she will deliber-
ate on any given occasion. We’ll lump all of that together, and the resulting deliber-
ation process as well, under the description ‘A’s reasons.’ The disconnect problem can
be stated thus: if an event E is undetermined, then it is not sufficiently connected to
A’s reasons to qualify as being within her control, up to her, or something she has a
choice about. The Influence strategy answers this worry by holding that it is some-
times the case that indeed there is a sufficient connection between E and A’s reasons.
I will now offer the “Degree Argument” to undermine the Influence strategy.28
Let us describe the strength of connection between reasons R and volition E with a
scale from 0 to 100, with 100 being a deterministic connection. Now compare the
following two cases.
10. The connection between R and E is 10.
20. The connection between R and E is 20.
It seems to me clear that if you were driven to the Influence strategy by the initial
disconnect problem, then you will now agree that A has more control over the action
that results from E in case 20 than she does in case 10. That is, you might have
shrugged off the disconnect problem back in section 5.1, or you might have proposed
some answer to it other than the Influence strategy, but if not, then those same
intuitions that led you to the Influence strategy should have you favoring 20 over 10.
So far, I have been proposing a preference for 20 over 10 when the question on the
table is about the degree of one’s control over one’s actions. But it is worth taking a
moment to also consider the relative merits of 20 and 10 when the question is about
the degree to which deliberation is reasonable and the degree to which A is respon-
sible for her actions.
I argued above that the property that compatibilists focus on—the property I called
‘self-determination’ on their behalf—is a good property to have whether or not it is

27
It is worth being explicit about one particular example. Let us suppose, with Hobart, that an act is free
as long as the volition that produces it is caused by the agent’s character (which he identifies with the
“moral self”). Let us call that volition E. Suppose we consider an event that is part of the formation of the
agent’s character and call that event ‘F’. It seems that an incompatibilist can honor the desire to avoid
determinism while still avoiding the disconnect problem by making F the undetermined event. Because
F was prior to the character formation, the character is still fully connected to E and the resulting action.
However, in this story it is now F that is disconnected; F is disconnected from the agent’s earlier character.
Thus F was not within her control; it was not something she had a choice about. Moreover, the conditional
“if F were to occur, then E would occur” is not within her control. Thus, by Beta, the agent still has no
control over volition E or the consequent action. So, the strategy of moving the undetermined event further
back in the causal chain not only fails to help with that event in particular—event F—it only helps with
respect to event E if one is willing to deny Beta.
28
For an earlier version of the degree argument, see Markosian (1999).
  

free will. One reason it is a good property to have, I claimed, is because it provides a
ground for deliberation. There I was assuming that self-determination is all or
nothing. Agent A self-determines her actions when the connection between R and
E is 100. But we are now working under the assumption that the connection can vary
in strength. We know what to say about the extremes on the scale. When the
connection is 100, deliberation is reasonable. When the connection is zero, deliber-
ation is pointless. One such case is when a freakish demon is intervening. Another
such case is when A’s volitions are completely undetermined. The question I now
want to ask is: what should we say about the reasonableness of deliberation when the
connection between R and E is between 0 and 100?
It seems to me clear that a tiny amount of connectedness—say 0.01—is not enough
to make it reasonable to deliberate. And it seems to me clear that full determination
minus just a tiny bit—say 99.99—is enough to make it reasonable. Wherever the
(perhaps fuzzy) cutoff line is between unreasonable and reasonable, as the degree of
connectedness rises from zero, we get closer and closer to that cutoff line until we
cross it and then we get further and further beyond it. That strikes me as a pretty
good characterization of what it means to say that deliberation becomes more
reasonable as the connection becomes stronger. So, the proportionality that Hobart
asserts between determinism and free will I assert between connectedness and
reasonableness of deliberation.
The relation between connectedness and responsibility is murkier. Kane’s assassin
case suggests that increased connectedness does not increase responsibility.29 He
imagines an assassin who is subject to spasms that occur indeterministically. The
assassin is aiming at the victim, happens not to spasm, and so succeeds in killing the
victim, though if a spasm had occurred the shot would have gone wide and missed.
Despite the luck involved in not spasming, the assassin is fully responsible for the
murder. What is noteworthy is that the odds of spasming seem to be irrelevant to the
degree of responsibility. This suggests the following principle: If A tries to do x and
succeeds, A is fully responsible for x. Indeed, if the assassin has only one brief
moment of non-spasming within a lifetime of spasming, it is still true that if she
was trying to hit the victim and succeeded she is fully responsible.
This is an interesting case, and I do not at the moment wish to dispute Kane’s
intuitions about the case. However, I do want to warn against drawing the wrong
conclusion. I propose that this case should play no role in defending a libertarian
account of free will against the disconnect problem. The case shows that responsi-
bility is a trickier matter than we thought. There is something special about the
relation between trying and responsibility that is getting in the way of theorizing
about free will in this case. What the case shows is that sometimes one can be
responsible for actions that one is not wholly in control of. In fact, Kane goes out
of his way to separate responsibility from control, noting that indeterminism can
lessen control without lessening responsibility. But he does not emphasize just how
vast the separation is.

29
Kane (2002: 229). Kane’s fuller statement of his views can be found in Kane (1996).
        

To underscore just how far apart responsibility can come from control, note that
even if the assassin’s odds of success are 0.01 (and you can add in as many zeros as
you want between the decimal point and the 1), as long as she tried and succeeded
she would be held responsible. Mapping this onto the discussion above of degrees of
connectedness, this would be like saying that even if the connection between R and
E is only 0.01, A would still be responsible for the resulting action. I am not prepared
to concede that the moral of the assassin case does map onto the degree of connect-
edness discussion, but even if it does, I suggest that if responsibility is that independ-
ent of connectedness and control, we should be hesitant to draw any conclusions
about libertarianism and free will. If you are not convinced of that, perhaps you will
at least be convinced of this conditional: if you found the original disconnect problem
from section 5.1 to require an answer along the lines of the Influence strategy, then
you should set aside Kane’s assassin case. The connection we were looking for cannot
be as slight as it can be in the assassin case.
Let us now set aside deliberation and responsibility and return our focus to
control. I will henceforth work with the assumption that control and free will go
hand in hand, setting aside any question of how that assumption relates to Kane’s
attempted separation of control and responsibility.
As previously stated, it seems that case 20 provides agent A with more control than
does case 10. Likewise for each successive case:
10. The connection between R and E is 10.
20. The connection between R and E is 20.
30. The connection between R and E is 30.
40. The connection between R and E is 40.
50. The connection between R and E is 50.
60. The connection between R and E is 60.
Each time we compare one case to its neighbor we find that we always have the same
motivation for preferring the stronger connection over the weaker one. From this it
would seem to be a quick step to an Hobartian proportionality thesis: the degree of
control that an agent has over her action is proportional to the degree of connect-
edness between her reasons and her volition. Or, more roughly, the closer we get to
determinism, the more free will we have. And that would seem to lead us to
compatibilism: getting all the way to determinism provides total control.

5.5 The Transition Point

But this is to ignore Beta and the incompatibilist’s argument built around Beta. If we
choose to respect Beta, E cannot be determined and still be something that A has a
choice about. At this point we conclude that there is no free will. Determinism—or,
rather, the local relative of determinism that would be provided by a maximum
connection between R and E—is necessary for free will, but its denial is also necessary
for free will. But there is another option. We can deny that total connectedness is
required for control, thereby opening the door for libertarianism. The proposal is
that A has free will when both of the following hold: there is enough of a connection
between R and E to avoid the disconnect problem but there is also enough of a
  

disconnection to provide freedom from the past. As we increase the degree of

connection between R and E—as we move further along in the series above—we
get more and more free will, until we hit the point at which we are too close to
determinism. When we hit that transition point we start losing control with increased
connection.30
The difficulty for the proposal is telling a plausible story about the transition point.
No matter which point is selected, there is good reason to prefer the next stage in the
series.
If more of a connection is better for freedom at the lower end of the scale, then a
little bit more should still be better at any given point on the scale.
50. The connection between R and E is 50.
60. The connection between R and E is 60.
61. The connection between R and E is 61.
62. The connection between R and E is 62.
63. The connection between R and E is 63.
The incompatibilist has a reason why it would be bad to get all the way to determin-
ism, but she has no reason why it would be bad to move from 61 to 62, or from 62 to
63, or whichever. There is nothing about stage 62 that could be relevantly different
from 61. To be clear, the difficulty is not that it is hard to recognize the transition
point. It is that there is nothing that could make one point the transition point.
A little more connection is always better, and the risk of determinism is always some
distance ahead of us.31
The structural problem for the libertarian is that, whereas the disconnect problem
proclaims that it is always better to have more connection, at least as far as that
problem goes, the argument for incompatibilism only proclaims that it is better to
have some disconnect. We can respect both of these proclamations by having so
strong a connection that we fall infinitesimally short of determinism. Yet no liber-
tarian would be happy with that strong a connection. The challenge, then, is to tell us
enough about why we should shun determinism as to at least provide a rough idea of
how much disconnection we should want.
There is a sort of answer that has already been given by some libertarians,
including van Inwagen.32 The suggestion to consider is that A’s only free actions
are the ones that arise in a case of a tie between A’s conflicting reasons. And let us

30
Alternatively, but for all practical purposes equivalently, if the libertarian wants to break the link
between control and free will, she may say instead that when we hit the transition point, the control keeps
increasing as the degree of connection increases, but that is not increasing our free will and is even
decreasing it. I will continue to focus on the proposal that preserves the link between control and free will.
Everything I say can easily be translated to apply to the alternative.
31
It is tempting to toss this difficulty into the vagueness bin and leave it to philosophers of vagueness to
worry about. But the hallmark of vagueness is arbitrariness, and I doubt that the libertarian would be happy
to accept that it is arbitrary when influencing turns from a good thing to a bad thing. Perhaps the same
point, or perhaps a different point: what is going on in this case just doesn’t feel to me like what is going on
in the typical cases of vagueness. I will not pursue these issues further in this paper, and I would be
interested to see whether someone working on vagueness sees a potential response to my argument. I thank
a referee for pushing me to clarify an earlier version of this footnote.
32
Van Inwagen (1989); Balaguer (2002); and Kane (2002).
        

supplement this suggestion with Kane’s explicit statement that it is this internal
conflict between reasons that creates the indeterminism in the decision process.33
Now the answer to the question of how much disconnection is the right amount is
“however much results from a tie between conflicting reasons.” This amount might
be different in different cases, but in every case of a free action, there will be some
amount of disconnect that is present. The disconnect arises from A’s own reasons;
the disconnect does not intervene like a freakish demon to inhibit the force of her
reasons.
This is a promising suggestion, but I am unconvinced. One way for A’s reasons to
be tied is for her to have no reasons at all. This is not the situation that is being
suggested by the libertarians, but let us consider it anyway. This case is subject to the
initial disconnect problem from section 5.1. And, just to be explicit, since there are no
reasons in play at all, the Influence strategy cannot help. Now consider a case in
which A has reasons R and reasons R*, such that each has a miniscule strength of, say
0.01, but they pull in opposite directions, towards E and E* respectively. It is difficult
to accept that such little influence of reasons on volitions can make the difference
between free will and no free will. (Though, given what Kane says about the assassin
case, perhaps he would disagree with me right at this point. Even he might agree,
though, if he focuses on deliberation rather than responsibility.) But if we deny free
will in case 0.01, we are back into the degree argument. Having tied conflicting
reasons R and R* is not sufficient for free will when R and R* have miniscule strength;
so, then, there must be a transition point at which having tied conflicting reasons is
sufficient. And now we are back to the transition point problem, and the appeal to
tied conflicting reasons cannot help answer it. The transition point is not special with
respect to the existence of ties, because the cases on both sides of the transition point
are cases of conflicting reasons that are tied in strength. The appeal to ties might help
if every case of tied reasons occurs at the same stage in the series—for example, stage
62. But there is no reason to think that they would all occur at the same stage. Or the
appeal to ties might help if every case of a tie produced a transition point at the degree
of strength of the tied reasons. But the case of 0.01 strength shows that that does not
happen. So, I conclude that the appeal to ties does not help.
But Kane might object that I have misrepresented his position. I have ignored the
difference between, on the one hand, indeterminism that arises from the internal
conflict created by tied reasons, and, on the other hand, indeterminism that arises
from some other source and just happens to coincide with a case of tied competing
reasons. Kane might insist that it is only the indeterminism that arises from the
internal conflict that is compatible with freedom. We will call indeterminism that
comes about in the Kanean way ‘Kanean indeterminism,’ keeping in mind that this
does not refer to a special kind of indeterminism but only a special kind of root for
the indeterminism.
The proposal is that in a case in which this Kanean indeterminism is the only
indeterminism that is involved, the degree of strength of each reason will be 50—in a
case of pure Kanean indeterminism R* will exert a conflicting force against R until

33
Kane (2002: 229–30).
  

they balance out at the half way point. If we replace the Kanean indeterminism, or
add to it, with some other non-Kanean indeterminism, then, but only then, will we
end up with a case of tied reasons where the force of the reasons is other than 50.
Kane may be happy to agree that the non-Kanean indeterminism does lessen
freedom while still insisting that the indeterminism that arises in the way he describes
does not lessen freedom.
In support of there being an important distinction between Kanean and non-
Kanean indeterminism we can return to the idea of self-determination. Kanean
indeterminism is special in that it arises from the self. In a Kanean case, the fact
that R and R* are equally strong reasons for agent A says something about what sort
of person A is. The indeterminism that arises from the internal tie is not like the
intervention of a freakish demon because it is not external to the self the way
indeterminism is in non-Kanean cases.
Or so it might be thought. I am not yet satisfied. It is important to be careful about
what exactly it is that arises from the self in a Kanean case. The preceding consid-
erations point to the self-determination of the indeterminism—the sort of person
A is determines that volition E will be undetermined. However, this is very different
from the self-determination of the volition itself. What is self-determined in a
Kanean case is not the volition but only the volition’s being undetermined. Suppose
A forms E, and suppose that E is the volition that R points her towards. Because of
R*, R’s strength of support for E is 50. With respect to the relation between R and E,
R* is like an intervening freakish demon. The fact that the disconnect arises out of the
self does not lessen the disconnection between E and the self. Even though R* is part
of the self, or at least a reflection of the self, E is no more connected to the self by
virtue of R*’s influence on creating the indeterminism than it would be if the
indeterminism in the case were non-Kanean.
Once we see that the self-determination that is present in the Kanean case does not
have the right target, we see that Kanean indeterminism does not help the Influence
strategy to find a non-arbitrary transition point. It is tempting to think that a tie in
which both R and R* exert a 50 degree influence on the outcome and do so in the
Kanean way is non-arbitrarily different from a case in which R and R* each exert a
49 degree influence due either to non-Kanean indeterminism or due to a mix of Kanean
and non-Kanean indeterminism. But this is a mistake. The cases are non-arbitrarily
different along one scale, but not along the one that counts. They are non-arbitrarily
different when it comes to explaining why the indeterminism arises in the two cases.
But they are arbitrarily different when explaining why E happens in the two cases. In
the one case E is connected to R to degree 50, and in the other case E is connected to
degree 49, and in yet another case E is connected to R to degree 51. There is still
nothing that makes 50 degrees of connection the perfect amount. So, I repeat my
earlier claim; I conclude that the appeal to ties does not help.
Let us move, then, to a different approach to the question of the transition point.
Above I wrote,

The structural problem for the libertarian is that, whereas the disconnect problem proclaims
that it is always better to have more connection, at least as far as that problem goes, the
argument for incompatibilism only proclaims that it is better to have some disconnect.
        

But perhaps the incompatibilist will have her own proportionality principle—the
degree of connectedness reduces freedom to the degree to which the connectedness
holds. Any degree of connectedness means that, to that degree, agent A is being
swayed by her past. Perhaps she will assert that, with respect to this swaying problem,
it is always better to have less connection. There is no transition point. It is not that
the goodness of being influenced by our reasons eventually becomes outweighed by
the badness of being swayed by the past. Rather, these are two separate scales, and
more connection is good on one scale and bad on the other. With no transition point,
there is no need to tell a plausible story about why that particular point (even vaguely
specified) is the transition point.
The problem with this proposal is that it does not allow the two problems to
interact. It does not allow the goodness of the freedom from the past to cancel out the
badness of not having more influence by reasons. On the preceding picture, there was
the need for a transition point because a certain amount of connectedness was
supposed to be good, while too much was supposed to be bad. On the present
picture, however, any amount is both good and bad. Connectedness is good in that
it provides influence by reasons. It is bad in that it allows the past too much sway over
our actions. We can make it less bad in the second respect by reducing the degree of
connectedness, but in doing so we make things less good in the other respect.
Keeping that in mind, imagine starting at the low end of the scale and wondering
how much connectedness we should add to make things good. Wherever we stop,
whatever was bad about determinism is present to that degree, and whatever was
good about determinism is absent to the degree to which the connectedness is less
than complete. We never get more control by taking away connectedness.
Start with an event E1 that is out of A’s control. Say that that event influences event
E2 to degree 10. Degree 10 of connection is better than none from the perspective of
influence, but it is worse than none from the perspective of the past holding sway
over our actions. If you were bothered by determinism—if you were bothered by the
past holding sway over our actions—you should still feel bothered by that 10 degrees
of connectedness. And the extra 90 degrees of disconnectedness does nothing to
mitigate that badness that is already present because of the 10 degrees. It stops things
from being even worse from the perspective of the past’s sway, but it doesn’t reduce
the bad that was already present in the 10 degrees. And to the extent that the
90 degrees of disconnectedness is good, it is also bad from the perspective of the
connectedness problem. Compare this to a case of 20 degrees of connection. The
overall balance will still be the same. Case 10 is no better than case 20 simply because
the deterministic connection is weaker. The moral is that nothing gets better for A’s
control over E2 by bringing indeterminism into the story. Whatever degree of gain
there is along the freedom from the past dimension there is an equal degree of loss
along the influence dimension, and vice versa.
So, I end my discussion of the degree argument and accept the premise:
1*. An agent A has no control over any event that is influenced to any degree by
an event that A has no control over.
We can see this as a generalization of the original premise 1 in the anti-free will
argument. That is, premise 1 is an instance of 1*.
  

Our new anti-free will argument, rewritten to take account of the possibility of
influence without determinism, is:
1*. An agent A has no control over any event that is influenced to any degree by
an event that A has no control over.
2. A has no control over any event that occurred before A was born.
3. A has no control over a wholly undetermined event.
4*. Every event either is a wholly undetermined event or is influenced to some
degree by some event that is wholly undetermined or by some event that
occurred before A was born.34
Therefore: A has no control over any event.
Therefore: There is no free will.
The new anti-free will argument still rests on Beta. If I did not accept the validity of
Beta I would be a compatibilist. I would reject 1* in a Hobartian way, holding that the
degree of control that A has over an event E is proportional to the degree of
connectedness between E and A’s reasons. However, because I find Beta very
plausible, I find the anti-free will argument very plausible.
And so I conclude this section: libertarianism cannot survive the disconnect
problem. Or, at least I conclude with a conditional: if libertarianism could only
survive the disconnect problem by appealing to the Influence strategy, then libertar-
ianism cannot survive the disconnect problem. Still, I believe the stronger conclusion;
I am an anti-libertarian. In addition, because I find Beta extremely plausible, I find
the anti-free will argument very plausible. And yet, when I am not attending to Beta,
and when I am attending to the value of free will in ordinary life, I find compatibilism
very plausible. Perhaps this last is a weakness that I should just try to get over. But
I would like to at least explore a way to preserve everything I find plausible. That
takes us to the final section of this paper.

5.6 Contextualism
In this section I will be more speculative. I will make assertions without defense.
Consider this section to be exploratory. Also, though I will talk of the term ‘free will,’
what I say is really more applicable to more ordinary related terms like ‘has control
over,’ ‘is up to,’ and ‘has a choice about whether.’
Contextualism tells us that among the many properties that are in the free will
family, which one satisfies the concept free will or is the referent of the term ‘free will’
varies depending upon the context of the ascriber. (Henceforth, I will simplify by
talking only of terms and dropping talk of concepts.) I have, in effect, identified the
family of properties with those on the connectedness scale. The property of being
such that your reasons are connected to your volitions to degree 72.4, for instance, is

34
The truth of premise 4* depends on the assumption that being influenced to some degree is transitive.
That seems right to me, but if not, I could instead use a variant of 4* stated in terms of the ancestral of being
influenced to some degree.
        

a candidate for being the referent of the term ‘free will.’ This is only a rough start. For
one thing, I have been pretty loose about characterizing reasons. For another thing,
the properties that are really candidates aren’t the ones that are had on one particular
occasion but rather a summing over time of the ones that are had on particular
occasions. For instance, the property of having degree 59.7 of connectedness for one’s
first volition and degree 83.8 of connectedness for one’s second volition and degree
76.6 of connectedness for one’s third volition, and so on. And that isn’t right either,
because the properties corresponding to the connectedness scale leave out too much.
For instance, agency theorists will not find their favorite candidates on the list, at least
not explicitly. Likewise, compatibilists who analyze the ability to do otherwise as
conditionals will not find their favorites on the list either, at least not explicitly.
Despite all of these shortcomings and many others, for our present purposes I will
treat the connectedness scale as providing the candidates for being the referent of the
term ‘free will,’ because it will be useful to have a concrete example. Also, I am
confident that what I say about the low end of the scale will apply just as much to
Agency Theory variations and what I say about the high end of the scale will apply
just as much to conditional analysis variations.35
When R is fully connected to E—when A’s reasons determine her volitions—it is
an instance of what I earlier called ‘self-determination.’ Self-determination is a good
property to have. It grounds deliberation. And it provides a basis for forward looking
practices to encourage desirable behaviors, including, in particular, moral behavior.
Self-determination makes reasonable a significant chunk of our personal and inter-
personal attitudes and relations. At the other end of the connectedness scale is total
disconnection between R and E. Total disconnection is a bad property to have. It
makes deliberation pointless. It makes attempts at forward looking behavior modi-
fication pointless. It would leave our lives in shambles. Between these two extremes
on the connectedness scale there are perhaps an infinity of properties, one for each
possible degree of connectedness. These properties are more or less desirable in
proportion to their degree of connectedness.
None of the connectedness properties, low to high, can ground a kind of respon-
sibility that warrants retributivist reactions. If we have the properties at the low end,
our deeds are not our fault because we don’t have enough control over them. They
are not connected enough to our values and desires and the sorts of people we are.
And at the high end, our deeds are not our fault because we don’t have enough
freedom from the past. There was not much chance of avoiding the deed. In the
middle our deeds are not our fault because whatever freedom we get from the past we
get at the price of separating ourselves that same amount from our values and desires.
Any gain that comes by moving further away from one end is accompanied by a loss
by moving closer to the other.

35
In fact, the connectedness scale highlights only one dimension along which context might select.
I believe a proper account of our freedom language would involve the concept of normalcy (as in, “normal
upbringing” and “normal causes”), and that will provide another context-sensitive dimension. I also think
the proper analysis of freedom language will be modal, and that will provide another context-sensitive
dimension.
  

Everything I have said about the connectedness properties in the past two para-
graphs, if true, is true regardless of whether any of those properties gets the title ‘free
will.’ It is a good question which, if any, of the properties is the referent of the term
‘free will.’ But the answer to that question does not in the least affect the metaphysical
significance of the properties—does not affect what other properties are entailed by
those properties. That is a crucial feature of contextualism. Contextualism is not a
metaphysical thesis at all. It is a thesis about reference. The candidate referents have
their relations independent of what they are called. Contextualism is not in any way a
kind of relativism. That is, contextualism does not hold that the same proposition is
true relative to one speaker and false relative to another. It holds that the same
sentence can express a different proposition in one speaker’s mouth than it does in
another speaker’s mouth. Contextualism is trivial when applied to standard index-
icals like ‘I’ and ‘she’ but much less trivial when applied to central philosophical terms
like ‘knowledge’ and ‘free will.’
The claim that Beta is valid would seem to be a conceptual truth. As such, it would
seem to put a restriction on which properties can be the referent of the term ‘free
will.’ On the other hand, our actual use of the term would seem to conflict with Beta.
The latter fact, if it is a fact, might best be explained by a lack of philosophical
sophistication among typical speakers. But contextualism provides the possibility for
an alternative account. Perhaps there are contexts in which we use the term ‘free will’
in a way that respects Beta and other contexts in which we use it differently, and
perhaps both uses are correct.36
The idea is this. Many of the contexts in which we use the term ‘free will’ (or
related terms) are ones in which we are concerned with decision making and not at
all with blame or praise. If I am trying to decide what to do, or if I am discussing your
decision about what to do, with no thought about praise or blame, the property I’m
likely to care about, or at least one property that is likely to satisfy my concerns, is the
compatibilist’s self-determination. I think about the options before A, and I consider
whether her decision- making process will make a difference as to which option will
come about. I say things to her like, “If you don’t want to do it, don’t. It’s up to you.
Nobody’s making you do it.” Or “I don’t know why you’re even discussing this as if
you had a choice. You know your boss is going to make you do it whether you want to
or not.” In contexts such as these it is not implausible that the term we are using is
one that refers to the compatibilist’s self-determination or other compatibilist-
friendly properties on the high end of the connectedness scale. Relative to such
contexts compatibilism is true.
On the other hand, there are other contexts in which we are interested in praise
and blame. For instance, contexts in which the question of fault are particularly
relevant. I do not think that these contexts automatically invoke Beta. However, if we
are concerned about fault and we start actually filling in the details of what led to A’s
decision, whether it be those details that start to suggest determinism or those details

36
I take the contextualism I propose here to be a descendent of my (1996), which itself I take to be a
descendent of the Paradigm Case argument as discussed in van Inwagen (1983: 106–12). The common
thread is the attempt to separate conceptual truths like Beta from the ordinary denotation of the term ‘free
will.’
        

that suggest disconnectedness between A’s reasons and her decision, then Beta will
become operative. And, since there is no property that can satisfy the demands of
Beta, relative to these contexts there is nothing for ‘free will’ to refer to that will make
‘A has free will’ true.
Fault-centered contexts are not the only ones that can activate Beta. For instance,
explicitly considering Beta might have the effect of generating a context in which it is
valid. It would take more detailed work than I want to provide at the moment to
describe a plausible mechanism by which this effect would occur. But for compari-
son, contextualists in epistemology tend to say that explicitly considering skeptical
scenarios makes them relevant and explicitly considering the closure principle for
knowledge—Kp & K(p→q) ⊪- Kq—yields a context in which closure both is valid
and applies univocally. A contextualist about free will might look to that analogy for
some guidance.37
I have not said nearly enough here to convince anyone that contextualism is true.
But I put it on the table as a suggestion as to how to deal with competing pressures on
the word ‘free will.’ On the one hand, we have the connection between ‘free will’ and
responsibility—a connection so strong that it is now common in the literature to
merely stipulate that one is discussing the sense of ‘free will’ that is relevant to moral
responsibility. On the other hand there is the connection between ‘free will’ and
deliberation. If I am right, contra van Inwagen, that the compatibilist’s self-
determination provides grounds for deliberation, then we can preserve the link
between free will and deliberation by using ‘free will’ to refer to self-determination
(and near self-determination). If the property relevant for moral responsibility, at
least of the retributivist variety, is a different property (an impossible property, as it
seems to me), then that pulls us towards using ‘free will’ to refer to that different
property. Contextualism allows us to satisfy the pull in both directions, just not
within a single context.

5.7 Conclusion
The metaphysics, including the metaphysical underpinnings for retributive respon-
sibility, are not in the least affected by contextualism. If I become convinced that
contextualism is false, that will not lead me to change my mind about which
properties there are that are relevant to the general discussion of free will and
responsibility or about which of those properties we do or can instantiate. It will
not lead me to change my mind about which of the properties are well suited to doing
which jobs, and, hence, will not change my mind about which properties I wish I had.
In short, the sorts of properties the compatibilists tend to focus on are good
properties to have and most of us have them. They make it reasonable to deliberate,
but they do not provide a basis for retributivism. In contrast, the sorts of properties
the libertarians tend to focus on are not properties that I would want. They would

37
Earlier proponents or explorers of contextualism about free will include Horgan (1979) and
Hawthorne (2001). A more recent contextualist is Rieber (2006), and Horgan has returned to the subject
in his (2014).
  

undermine the significance of deliberation by increasing the disconnect between

reasons and action, and they would still not provide a basis for retributivism.
Am I, then, a compatibilist? An impossibilist? Or what? Well, those terms can map
onto the metaphysics that I have endorsed in different ways depending on which of
the various properties the term ‘free will’ denotes. The position I have gestured
towards here is a contextualism of the following sort. Relative to some contexts the
true thing to say is “compatibilism is true and we have free will.” Relative to other
contexts, the true thing to say is “Beta is valid, impossibilism is true, and we lack free
will.” I doubt that there is any context relative to which it is true to say “Beta is valid,
and we have free will by having control over undetermined events.” That claim,
I think, just incorporates a philosophical error. That is, even if contextualism is true,
I still think there is no context relative to which it is true to say “libertarianism is true.”

Acknowledgments
I presented a version of this paper to the Syracuse University Graduate Student conference in
April, 2012, and I thank that audience for their comments. I also presented a paper more
directly aimed at defending contextualism about the term ‘free will’ against objections and
distinguishing my version from Hawthorne’s at the Bled Philosophical Conference in 2006,
and I thank that audience as well. I am also grateful for conversations with Eric Barnes, Yishai
Cohen, Andre Gallois, Ned Markosian, Derk Pereboom, and Steve Sverdlik. And special
thanks to John Keller for putting together this volume, for his editorial assistance, and for
his patience.

References
Alexander, H. G. (1956), The Leibniz-Clarke Correspondence, ed. (Manchester University
Press).
Balaguer, M. (2002), ‘A Coherent, Naturalistic, and Plausible Formulation of Libertarian Free
Will’, in Noûs 36: 379–406.
Chisholm, R. (1996), ‘Freedom and Action’, in K. Lehrer (ed.), Free Will and Determinism
(Random House).
Fischer, J. M., and Ravizza, M. (1998), Responsibility and Control (Cambridge University
Press).
Frankfurt, H. (1971), ‘Freedom of the Will and the Concept of a Person’, in Journal of
Philosophy 68: 5–20.
Hawthorne, J. (2001), ‘Freedom in Context’, in Philosophical Studies 104: 63–79.
Heller, M. (1996), ‘The Mad Scientist Meets the Robot Cats’, in Philosophy and Phenomeno-
logical Research 56: 333–7.
Hobart, R. E. (1934), ‘Free Will as Involving Determination and Inconceivable Without It’, in
Mind 43: 1–27.
Horgan, T. (1979), ‘ “Could”, Possible Worlds, and Moral Responsibility’, in Southern Journal
of Philosophy 17: 345–58.
Horgan, T. (2014), ‘Injecting the Phenomenology of Agency into the Free Will Debate’,
Inaugural Marc Sanders Lecture, Pacific APA.
Kane, R. (1996), The Significance of Free Will (Oxford University Press).
        

Kane, R. (2002), ‘Free Will: New Directions for an Ancient Problem’, in Robert Kane (ed.), Free
Will (Blackwell).
Lewis, D. (1981), ‘Are We Free to Break the Laws?’, in Theoria 47.
Markosian, N. (1999), ‘A Compatibilist Version of the Theory of Agent Causation’, in Pacific
Philosophical Quarterly 80: 257–77.
Mele, A. (2006), Free Will and Luck (Oxford University Press).
Pereboom, D. (2001), Living Without Free Will (Cambridge University Press).
Pereboom, D. (2008), ‘A Compatibilist Account of the Epistemic Conditions on Rational
Deliberation’, in Journal of Ethics 12: 299.
Rieber, S. (2006), ‘Free Will and Contextualism,’ in Philosophical Studies 129: 223–52.
Van Inwagen, P. (1983), An Essay on Free Will (Oxford University Press).
Van Inwagen, P. (1989), ‘When is the Will Free’, in Philosophical Perspectives 3: 399–422.
Van Inwagen, P. (1990), Material Beings (Cornell University Press).
Van Inwagen, P. (2000), ‘Free Will Remains a Mystery’, in Philosophical Perspectives 14: 14–16.
Van Inwagen, P. (2011), ‘A Promising Argument’, in Robert Kane (ed.), The Oxford Handbook
of Free Will, 2nd edn. (Oxford University Press).
Wolf, S. (1990), Freedom Within Reason (Oxford University Press).
Wolf, S. (2003), ‘Sanity and the Metaphysics of Responsibility’, in G. Watson (ed.), Free Will
(Oxford University Press).
 6
Revisiting the Mind Argument
Alicia Finch

6.1 Introduction
In An Essay on Free Will,1 Peter van Inwagen defends libertarianism, the two-part
thesis that (i) incompatibilism is true and (ii) some agents act freely, where incompa-
tibilism is the thesis that, necessarily,2 agents act freely only if determinism3 is false.
His defense includes, on the one hand, the consequence argument for incompatibi-
lism and, on the other hand, responses to the Mind argument against libertarian-
ism.4,5 It is noteworthy that, as van Inwagen presents the dialectic, it seems that the
consequence argument is valid if only if the most compelling formulation of the
Mind argument is valid, too. Van Inwagen concludes his discussion of the Mind
argument by claiming that although he is rationally justified in believing the conse-
quence argument valid and the Mind argument unsound, he cannot see how the
premises of the Mind argument could be false. With this, one might very well wonder
why van Inwagen’s alleged defense of libertarianism should count as a defense at all:
if van Inwagen is correct about the libertarian’s dialectical position, this position is
very weak indeed.
In their (1996), McKay and Johnson prove that the consequence argument, as
formulated by van Inwagen, is invalid; they offer a new formulation of the conse-
quence argument in response. Finch and Warfield (1998) point out that, given the
proof in question, van Inwagen’s formulation of the Mind argument may be dis-
missed as invalid. Moreover, they argue, while the consequence argument may be
reformulated such that it is valid, the Mind argument cannot. As such, they conclude,
van Inwagen was mistaken when he claimed that, with respect to their validity, the
consequence and Mind arguments stand or fall together.

1
Hereafter, “The Essay”.
2
In what follows, “necessarily” should be read as “It is broadly logically necessary that.” Moreover,
I treat metaphysical necessity and broadly logical necessity as equivalent.
3
I will define determinism in the next section.
4
So named by van Inwagen because the argument has appeared so often on the pages of the journal
Mind. See Hobart (1934), Nowell-Smith (1948), and Smart (1961).
5
There is more to van Inwagen’s defense than what I discuss here. For instance, he argues for the thesis
that some agents sometimes act freely, and he develops objections to arguments for compatibilism (the
thesis that incompatibilism is false).
     

By the time that he writes “Free Will Remains a Mystery”(2000), van Inwagen
acknowledges that his earlier formulation of the consequence argument is invalid,
adding that, “Since [it] is invalid, the Mind Argument is invalid.”6 Rather than
considering whether a valid reformulation of the Mind argument is available, how-
ever, he offers a “very informal, intuitive form” of the Mind argument.
Whatever the success of the informal, intuitive form of the argument, van Inwa-
gen’s claim about the validity of the two arguments deserves to be reconsidered. If the
Mind argument can be reformulated such that it is valid if and only if the conse-
quence argument is valid, the libertarian’s dialectical situation is as van Inwagen
described in The Essay: either the libertarian must jettison the consequence argu-
ment, find a flaw in one of the premises of the Mind argument, or take van Inwagen’s
own position: there is something wrong with one of the premises of the Mind
argument, though she cannot see how that could be. This result would be as
significant now as it was when van Inwagen first argued for it. As such, this issue
ought not to be set to the side any longer.
Before I present van Inwagen’s formulations of both arguments, I will build a
foundation, so to speak, by laying out the terms of the debate. Once van Inwagen’s
arguments are in place, I will first discuss the work of McKay and Johnson, and then
turn to Finch and Warfield. With this, I will take up the task of offering my own
formulation of the consequence argument, emphasizing the inference principle on
which its validity rests. Finally, I will offer a formulation of the Mind argument that
seems to be valid if and only if the consequence argument is valid, too. This
formulation will depend on the notion of grounding, a notion that has rarely played
a part in any formulation the Mind argument.7 By invoking the notion of grounding,
the definition of free action, and theses that the consequence argument itself depends
on, I will arrive at the conclusion that the libertarian is in precisely the same
dialectical situation that van Inwagen said she was in.

6.2 Laying a Foundation

In The Essay, van Inwagen offers that, “When I say of a man that he ‘has free will’
I mean that very often, if not always, when he has to choose between two or more
mutually incompatible courses of action—that is, courses of action that it is impos-
sible for him to carry out more than one of—each of these courses of action is such
that he can, or is able to, or has it within his power to carry it out.”8 At the time that
he was writing, van Inwagen’s interlocutors took it for granted that he was properly
using terms like ‘free will’, ‘free action’, ‘up to’, and ‘having a choice about’. These
terms were intended as philosophical terms of art, used for the sake of facilitating
discussion of those cases in which agents were able to carry out more than one course
of action. Given that times have changed, and participants in the free will debate now
question what ‘free will’ means, it seems best that I stipulate that I will use them as

6
p. 11. He seems to mean the formulation of the Mind argument presented in The Essay.
7
The only exceptions I know of are my (2013a) and Tognazzini (2015). I note that when this paper
was accepted for publication, I did not know of Tognazzini’s piece, which was not yet published.
8
p. 8.
  

van Inwagen used them in The Essay. By extension, I will use them as they were used
by the respondents that I consider here. So:
(FA) S freely performs A at t =df. (i) S performs A at t and (ii) for some time
t- such that t- is earlier than t, S at t- is both (a) able to perform A at t and (b) able
to refrain from performing A at t.
Where:
(UP) It is up to S at t- whether S performs A at t =df. t- is earlier than t, and
S at t- is both (a) able to perform A at t and (b) able to refrain from performing A at t.
Moreover, “It is up to S at t-” may be used interchangeably with “S at t- has a choice
about” and “S at t- is able to perform A” may be used interchangeably with “S at t- can
perform A.”
Van Inwagen uses other terms of art in presenting both the consequence and Mind
arguments. For instance, he invokes the notions of propositions, possible worlds, states
of the entire physical world at an instant (hereafter, states of the world at an instant),
laws of nature, and, obviously, determinism. Propositions, according to van Inwagen,
are non-linguistic bearers of truth values expressed by declarative sentences. Possible
worlds may be construed as maximal possible state of affairs,9 and states of the world
at an instant should be treated as present-tense maximal states of affairs:
[A] state of affairs O is future-directed just in case either O’s obtaining entails that some
contingent thing will exist or O’s obtaining entails that no contingent thing will exist; [a] past-
directed state of affairs [is defined] in the obviously parallel way. Then a state of affairs O is
present-tense maximal if, and only if, for every atomic state of affairs O* that is neither future-
directed nor past-directed, either O includes O* or O precludes O*.10,11

According to van Inwagen, each such state corresponds to a proposition: “A prop-

osition expresses the state of the world at [time] t provided it is a true proposition
that asserts of some state that, at t, the world is in that state.”12
With respect to laws of nature, van Inwagen contends that they are propositions,
but acknowledges that he has “no idea how to explain this term, much less define
it.”13 However, he suggests, “The notion of a law of nature makes sense.”14 It is clear
enough, in the context at hand, what it is for a law of nature to be a law of nature.

9
Plantinga (1976) presents this account of possible worlds, according to which a possible state of
affairs is a way things might have been. A state of affairs O is maximal if and only if for every state of affairs
O*, O either includes or precludes O*, where (i) a state of O includes a state of affairs O* if it is not possible
for O to obtain and O* to fail to obtain; and (ii) a state of affairs O precludes O* if it is not possible that both
obtain. I note that while Plantinga uses ‘S’ to refer to an arbitrary state of affairs, I use ‘O’. I further note that
although Plantinga takes states of affairs to be abstract objects, nothing in what follows hinges on this thesis. In
his (1986), van Inwagen endorses this account of possible worlds, though he speaks in terms of propositions
rather than states of affairs.
10
Finch and Rea (2008), p. 10. Here I use the terms ‘O’ and ‘O*’ where Finch and Rea use ‘S’ and ‘S*’.
11
Van Inwagen himself says very little about this notion, contending that he may leave it “largely
unexplained,” given that “[His] argument is very nearly independent of its content” (p. 59). Nonetheless,
the few remarks that he makes indicate that he has something like this in mind.
12 13 14
p. 60. p. 60. p. 14.
     

At this point, van Inwagen is in a position to provide a definition of determinism.

Loosely speaking, it is the thesis that there is at any instance “exactly one nomo-
logically possible future.”15 To define the thesis more precisely, we may stipulate that:
‘t-’ designates some arbitrary time t-.
‘t’ designates some arbitrary time t such that t is later than t-.
‘Pt-’ designates the proposition that expresses the complete state of the world at t-.
‘Pt’ designates the proposition that expresses the complete state of the world at t.
and:
‘L’ represents the conjunction of the laws of nature into a single proposition.
In this case:
Determinism = df. The thesis that □((Pt- & L) → Pt)16
Moreover, if we add that:
‘p’ designates any true proposition whatever,
it is a consequence of determinism that:
□((Pt & L) → p)
With this foundation laid, we may consider van Inwagen’s presentation of the
consequence argument.

6.3 Van Inwagen and the Consequence Argument

Since determinism is a thesis about propositions, and since the conclusion of the
consequence argument is about actions, there must be some way to bridge the gap.
And so it is that the consequence argument is formulated, whether implicitly or
explicitly, by way of the notion of not having power over the truth value of a
proposition p or not being able to render a proposition p false. So, if a proposition
pA states that an agent S performs act A, and if S lacks power over the truth value of
pA, (i.e., is not able to render pA false), then S does not perform A freely.
From here, van Inwagen introduces the now famous ‘N-operator’, where:
“Np” is read as “p and no one has, or ever had, any choice about whether p.”
He then offers these inference principles:
(α) □p entails Np
(β) {N(p → q), Np} entails Nq
If one stipulates that:
‘PtBB’ designates the proposition that expresses the complete state of the world at
some time long before anyone’s birth,17

15
p. 3. 16
‘□p’ expresses the proposition that it is broadly logically necessary that p.
17
Van Inwagen does not use the term ‘PtBB’. I do so now to avoid confusion later.
  

one arrives at the formulation of the consequence argument that van Inwagen
presents in the Essay:
1. □((PtBB & L) → p) Consequence of Determinism
2. □(PtBB→ (L → p)) 1
3. N((PtBB→ (L → p)) 2, α
4. N PtBB premise, fixity of the distant past
5. N(L → p) 3, 4, β
6. NL premise, fixity of the laws of nature
7. Np 5, 6, β
Thus, van Inwagen concludes, if determinism is true, no one has, or ever had, any
choice about which actions she performs.
As far as an evaluation of the argument goes, van Inwagen contends that β is the
weakest link. Though it seems valid, its validity is not as obvious as (i) the validity of
α; (ii) the truth of the premise that no one has, or ever had, a choice about the distant
past; or (iii) the truth of the premise that no one has, or ever had, a choice about the
laws of nature.18
But for defenders of the consequence argument who are also libertarians, there
may be a far more serious problem with β: according to van Inwagen, the Mind
argument against libertarian seems to stand or fall with the validity of β. I turn now to
this argument.

6.4 Van Inwagen and the Mind Argument

In presenting what he takes to be the strongest formulation of the Mind argument,
van Inwagen begins by describing a scenario in which (i) a thief enters a church for
the sake of stealing a small sum of money from the poor box; (ii) after he arrives, the
thief suddenly begins to deliberate about whether to go through with the theft; (iii)
the outcome of the thief ’s deliberation is a belief-desire complex DB; and (iv) the
thief performs an act of repentance, R. As van Inwagen describes the case, it is
nomologically possible for DB to occur without R. There is a “truly undetermined
link in the chain that binds deliberation to action.” In order to facilitate discussion
van Inwagen suggests that the relation between DB and R is causal, so that (i) DB
causes R and (ii) it is nomologically possible for DB to occur without causing R.19
Van Inwagen emphasizes that, “Once DB has occurred, then everything relevant to
the question whether R is going to happen has occurred.”20 With these consider-
ations in place, van Inwagen endorses the general principle that no one has any

18
In his (1981), David Lewis challenges this thesis by suggesting that the claim that no one has a choice
about the laws of nature is ambiguous. He grants that no one can perform an act that either is or causes a
law-breaking event. However, he contends, one might nonetheless be able to do something such that, if one
were to do it, what is in fact a law (i.e., what is a law in the actual world) would not be a law. If determinism
is true and S performs act A freely at time t, there is a possible world in which the course of events diverges
from the actual course of events a little while before t. At this point of divergence, a divergence miracle
occurs: an event occurs such that it violates the actual laws of nature, but not the laws of the world in which
it occurs.
19
One might say that DB “indeterministically causes” R. 20
p. 144.
     

choice about the occurrence of an undetermined event. At this point, van Inwagen
takes himself to have established one premise of his formulation of the Mind
argument:
8. N (DB occurs → R occurs)
With Principle β in view, we ought to ask whether it was up to the thief that DB was
the outcome of his deliberation. In response, van Inwagen offers that:
It is unlikely that our thief had any choice about whether DB occurred; and even if he did, this
could only be because he had a choice about some earlier states of affairs of which DB was a
consequence; if so, the questions we are asking about DB could be asked about those earlier
states of affairs. . . . This process could, in principle, be carried on till we reached the thief ’s
‘initial state’ about which he currently has no choice.21

And so:
9. N (DB occurs)
Of course, given the validity of β, it follows that:
10. N (R occurs).
Because there is nothing special about the case of the thief, one may generalize from
(10) to the necessary falsity of libertarianism: unless an act is determined to occur, it
is not up to anyone whether she performs it.
Van Inwagen responds to this dialectical situation by asserting that the conse-
quence argument is valid, denying that the Mind argument is sound, and admitting
that he sees nothing wrong with the premises of the Mind argument. Of course, there
are other positions that a libertarian might take. However, given the way that van
Inwagen has laid out the dialectic, it seems that any libertarian who endorses the
consequence argument must either find something wrong with one of the premises of
the Mind argument or accept that her position, like van Inwagen’s, is precarious.

6.5 The Invalidity of the Consequence Argument

For the moment, I will set aside questions about the relationship between the
consequence and Mind arguments and focus on the invalidity of β. In particular,
I will address Thomas McKay and David Johnson’s proof that β, as formulated by
van Inwagen, is invalid.22
They begin by demonstrating that if β is valid, so is the Agglomeration Principle:
(Agglomeration) {Np, Nq} ├ N(p & q).
The demonstration is straightforward:
11. Np assumption
12. Nq assumption
13. □(p → (q → (p & q))) logical truth

21 22
p. 146. See their (1996).
  

14. N (p →(q → (p & q))) 13, α

15. N (q → (p & q)) 11, 14, β
16. N (p & q) 12, 15, β

And so they establish that β is valid only if Agglomeration is.

Next, they use the example of a fair coin toss to provide a counterexample to
Agglomeration. They stipulate that:
‘H’ represents the proposition that the coin does not land heads.23
‘T’ represents the proposition that the coin does not land tails.
And point out that given that it is not up to anyone what the outcome of a fair coin
toss turns out to be:
17. NH
18. NT
Since it follows via Agglomeration that:
19. N (H & T),
it is clear that Agglomeration is invalid. After all, there is something someone can do
that would render N (H & T) false: toss the coin.
Having presented their proof, they point out that the coin toss scenario is a
challenge to Agglomeration only if:
(Would) “Np” is read as “p and there is nothing anyone can do such that, if she
were to do it, p would be false.
Rather than:
(Might) “Np” is read as “p and there is nothing anyone can do such that, if she
were to do it, p might be false.
Their conclusion, then, is not that the consequence argument is invalid, but that the
N-operator should be read in terms of might rather than would.24

6.6 The Invalidity of the Mind Argument

As Finch and Warfield note in their (1998), if van Inwagen’s unrevised β is invalid, so
is the formulation of the Mind argument presented in The Essay. In considering the
significance of this claim, they argue that although the consequence argument can be
reformulated with valid β-like principles, the Mind argument cannot.
They begin by considering the β-like principle that they dub Beta 2:
(Beta 2) (Np & □(p & q)) implies Nq.

23
They use ‘p’ where I use ‘H’ and ‘q’ where I use ‘T’.
24
In what follows, I will use ‘β-might’ to refer to the β-like principle that includes the might-reading of
the N-operator. Because van Inwagen has indicated (personal correspondence) that he originally intended
the would-reading of the N-operator, I will continue to use ‘β’ to refer to the principle van Inwagen offered
in The Essay.
     

With Beta 2, they point out, the consequence argument may be reformulated such
that:
1. □((PtBB & L) → p)25 Consequence of Determinism
20. N(PtBB & L) premise
7. Np 1, 20, Beta 2
Moreover, Beta 2 is unsusceptible to the McKay-Johnson counterexample and, they
claim, is at least as plausible as the original β. As such, there is no reason to doubt the
validity of Beta 2.
In arguing for N(PtBB & L), they offer that:
The core intuition motivating the acceptance of van Inwagen’s premises likewise motivates the
acceptance of our premise. This core intuition is, we maintain, the intuition that the past is
fixed and beyond the power of human agents to affect in any way. [But the laws of nature] do
not change over time. Thus, the conjunction (PtBB & L) offers a description of what might be
called the “broad past”—the complete state of the world at a time in the distant past including
the laws of nature.

Having offered what they take to be a satisfactory formulation of the consequence

argument, they point out that any attempt to reformulate the Mind argument by
replacing β with Beta 2 necessarily fails. Such a reformulation would take this form:
9. N (DB occurs)
21. □ (DB occurs → R occurs)
10. N (R occurs)
Given that the Mind argument depends on the premise that it is nomologically
possible for DB to occur without R, this reformulation is of no use to the anti-
libertarian.
Having discussed Beta 2, Finch and Warfield consider other β-like principles. In
particular, they consider β-might, and explain that if the Mind argument were
formulated by way of this principle, it would clearly be unsound. They return to
the “conditional premise” of the Mind argument:
8. N (DB occurs → R occurs)
and point out that if indeterminism is true, there is indeed something the thief can do
such that, if he were to do it, R might not follow DB: he can do whatever he actually
does. After all, it is true by hypothesis that once DB has occurred the thief ’s
repentance might follow or it might not. Indeed, the conditional premise is alleged
to be true precisely because the repentance might or might not have followed
DB. Clearly, then, reformulating β by way of the might-reading of the N-operator
is of no use to the purveyor of the Mind argument.
In what follows, I will argue that, pace Finch and Warfield, the Mind argument can
be reformulated with a valid β-like principle; indeed, it can be formulated with what
I call Transfer, a β-like principle that is plausible if any β-like principle is. Once I have

25
‘PtBB’ is not their term. Once again, I am trying to avoid confusion.
  

defended this formulation, I will conclude that the libertarian’s dialectical situation is
just what van Inwagen said it was when he wrote The Essay.

6.7 Revisiting the Consequence Argument

In an attempt to show the connection between the consequence and Mind argu-
ments, I will begin by offering what I take to be the best formulation of the former: it
seems to me that it is sound if any formulation is.26
My presentation of the consequence argument begins with these stipulations:
‘W’ designates some possible world W.
‘S’ designates some agent S in world W.
‘LW’ designates the proposition that expresses the conjunction of all the laws of
nature that obtain in W.27
‘pW’ designates the proposition that p is true in world W.
And:
‘DW’ designates the thesis that determinism is true in world W.
In the light of these stipulations, the earlier definition of determinism may be
replaced by:
DW = df. □((Pt-W & LW) → Pt).
With this, I introduce a new N-operator:
‘□Ns,tpW’ is to be read as: □(pW and ∀S∀t(It is not up to S at t whether p)).
and a new β-like principle:
Transfer = {□Ns,t,pW, □(p → q)} ├ □Ns,t,qW.
Of course, Transfer is similar to Finch and Warfield’s Beta 2 in that the conditional is
flanked with ‘□’ rather than the N-operator.
As far as the premises of the consequence argument go, I appeal not to the notion
of the “broad past” but to what I call the law addition principle (LAP):
(LAP) □Ns,tpW ├ □ Ns,t(p & LW)W.
Moreover, instead of appealing to the fixity of the past, I appeal to the principle of the
fixity of the present (FP):
(FP) □ Ns,t(Pt)W
In other words, it is necessarily the case that there is nothing anyone at time t can do
such that, if she were to do it, the complete state of the world at t would not (or might

26
I present this formulation of the argument in my (2013a) and (2013b).
27
‘LW’ is a rigid designator. Moreover, I will assume that there is a maximality condition for proposi-
tions like LW, that is, propositions that express the conjunction of all the laws of nature that govern a
particular world W. If LW is such a proposition, then for any proposition pl that expresses a law of nature
that governs some possible world, either LW includes or precludes pl.
     

not) obtain. I note that even if this principle is not immediately obvious, it follows
trivially from the definition of free action.28
And now the consequence argument may be formulated such that:
22. DW & PtW & pW assumption
23. □ ((DW & PtW & pW) → □ ((Pt & LW) → p)) consequence of determinism
24. □ ((Pt & LW) → p) 22, 23
25. □ Ns,t(Pt)W FP
26. □ Ns,t(Pt & LW)W 25, LAP
27. □ Ns,tpW 24, 26, Transfer
28. □ (DW → Ns,tpW) 22, 27
The conclusion of this argument is, of course, the incompatibilist’s thesis: necessarily,
if determinism is true, then for any agent S, for any time t, and for any proposition p,
it is not up to S at t whether p.
As is always the case with the consequence argument, the premise regarding the
laws of nature is subject to challenge. In the present context, most of the challenges
may be ignored: after all, my purpose here is not to defend the consequence argument
but to make the case that if the consequence argument is sound, so is my formulation
of the Mind argument. Given that this is so, my purposes will be served if I can make
the case that any considerations that tell in favor of van Inwagen’s principle of the
fixity of the laws also tell in favor of (LAP).
In making my case, I begin by pointing out that when I introduced the N-operator,
I did not specify whether it should be read in terms of “might” or “would.” Because
I wish to remain neutral as long as possible, and because:
(M*) □(pW & ∀S∀t (There is nothing S at t can do such that, if S at t were to do it, p
might be false).
entails but is not entailed by:
(N*) □(pW & ∀S∀t (There is nothing S at t can do such that, if S at t were to do it,
p would be false).
I will proceed on the assumption that (N*) is the same proposition as:
(N) □(pW & ∀S∀t (It is not up to S at t whether p))
With respect to arguing for (LAP), I note that even if (N) and (N*) are the same
proposition, there is no reason not to introduce an M-operator according to which:
‘□Ms,tpW’ is to be read as: □(pW and ∀S∀t (There is nothing S at t can do such
that, if S were to do it, p might be false).
Moreover, McKay and Johnson’s coin toss case fails to undermine what I call the
Would-Might Agglomeration Principle, according to which:
(W-M) {□Ns,t pw, □Ms,t qW} ├ □Ns,t (p & q)W

28
I will discuss this point at length in what follows.
  

If (W-M) is true, and if the same considerations that support van Inwagen’s principle
of the fixity of the laws support this claim:
□Ms,t (LW)W
the case for (LAP) is indeed as good as van Inwagen’s case for the fixity of the laws.
As far as □Ms,t (LW)W goes, the first conjunct of the statement is obvious: it is
necessarily the case that the laws of nature that obtain in W obtain in W. But is it
necessarily the case that for any agent S in W, there is nothing S at t can do such that, if S
were to do it, some proposition l that is a law of nature in W might be false? First, let us
ask whether there is anything an agent in the actual world can do such that, if she were
to do it, l—a proposition that is actually a law of nature—might be false. In suggesting
that van Inwagen ought to adopt the might-reading of the N-operator, McKay and
Johnson seem to think that van Inwagen would answer in the negative. And surely they
are correct. After all, in The Essay, van Inwagen offers that “the laws of nature impose
limits on our abilities: they are partly determinative of what it is possible for us to do.
And indeed this conclusion is hardly more than a tautology.”29 Moreover, the laws of
nature in the actual world are not so special that (i) there is nothing anyone in the
actual world can do that might render a law of nature false, but (ii) in other possible
worlds, it is possible to falsify the laws. Anyone who accepts van Inwagen’s reasoning
on behalf of the fixity of the laws ought to accept that given what it is for a proposition
to be a law of nature governing a world W, if l is a law of nature in W, no one in W can
do anything that might render l false. But this is just to say: □Ms,t(LW)W.
Given that every version of the consequence argument depends on some variation
or another on the fixity of the laws of nature, I conclude that the consequence
argument is sound only if □Ms,t(LW)W is true. Furthermore, given that the Would-
Might Agglomeration Principle seems valid if any β-like principle is, I conclude that
the soundness of the consequence argument entails the truth of (LAP).

6.8 Reformulating the Mind Argument

6.8.1 Libertarianism
No formulation of the Mind argument will be convincing if it seems to depend on a
mistake about the libertarian’s commitments. As such, I begin my formulation of
the argument by presenting what I take these commitments to be. First, I introduce
the notion of the determiner of an agent S’s performing an act A at a time t, where a
determiner may be characterized, roughly, as the earliest time at which it is nomo-
logically necessary that S performs A at t.30
I acknowledge that it may seem that I am suggesting that whenever an agent S freely
performs an act A at a time t, there is some time t- such that (i) t- is earlier than t and
(ii) it is nomologically necessary, at t-, that S performs A at t. But my characterization
of a determiner does not imply that this is so. As a matter of fact, my argument will

29
p. 62.
30
A bit less roughly: time t- is the determiner of S’s performing A at t =df. (i) □((Pt- & L) → S performs
A at t); (ii) there is no time t-2 such that (a) t-2 is earlier than t- and (b) □((Pt-2 & L) → S performs A at t);
(iii) there is at least one time t-2 such that t-2 is earlier than t-.
     

concern what I call immediately free acts, where S performs an immediately free act A
at t only if t is the determiner of S’s performing A at t. For my purposes here, I need
not specify exactly what sort of act might be immediately free, but I assume that
immediately free acts are mental events, e.g., choices, decisions, acts of will, intentions,
formations of preferences, or the like.31
I contrast immediately free acts with so-called “derivatively free acts,” where an agent S
performs a derivatively free act A* at time t+ if (i) S performs an immediately free act A at
some time t such that t is earlier than t+ and (ii) S’s performing A* at t+ is “appropriately
related” to S’s performing A at t. In the present context, there is no need to consider what
the “appropriate relation” might be, or, indeed, whether there are any derivatively free
acts. I simply note in passing that, according to some self-proclaimed libertarians, there
are both immediately and derivatively free acts, and that if an act is derivatively free, its
occurrence is nomologically necessitated by an earlier, immediately free act.
With respect to libertarianism, what matters is that if libertarianism is true, then (i)
for every act that an agent performs freely, the agent performs an immediately free act
and (ii) every immediately free act occurs at its determiner. I take it that (i) is true
whether or not libertarianism is. In the present context, the relevant difference
between the libertarian and her opponent is her insistence that immediately free
acts occur at their own determiners.

6.8.2 The Definition of Free Action

Having addressed libertarianism, I turn my attention to free action per se. More
specifically, I turn my attention to what I have dubbed the Transition relation.
Loosely stated, this is the relation that obtains between agents who act freely and
the acts that they freely perform insofar as the acts in question are indeed free. In the
present context, this loose statement will not do: the Transition relation is at the heart
of my formulation of the Mind argument. So let me characterize this relation as
precisely as possible.
I begin with two reminders:
(FA) S freely performs A at t =df. (i) S performs A at t and (ii) for some time
t- such that t- is earlier than t, S at t- is both (a) able to perform A at t and (b) able
to refrain from performing A at t.
Where:
(UP) It is up to S at t- whether S performs A at t =df. t- is earlier than t, and S at
t- is both (a) able to perform A at t and (b) able to refrain from performing A at t.
I offer these reminders in order to bring into sharp relief that:
(I/W) An agent S freely performs an act A at time t if and only if (i) for some
time t- such that t- is earlier than t, S at t- is such that it is up to her whether to
perform A at t and (ii) S at t is such that it was up to her whether to perform A at t
(but is so no longer).

31
See Clarke (2003; 2011), Ekstrom (2001; 2003), Franklin (2011), Ginet (1990; 1997), Goetz (1997),
Kane (1996; 2011), McCann (1998), and O’Connor (2000; 2010) for various suggestions regarding what
sort of act an immediately free act might be.
  

And, as I see it, it is helpful to think of free action in terms of what is and what was up
to an agent.
In order to make the case that this is indeed helpful, I stipulate that:
‘pAt’ designates the proposition that an agent S performs act A at time t.
‘p-’ designates the proposition that S at t- is such that it is up to her whether pAt.
‘pt’ designates the proposition that S at t is such that it was up to her whether pAt.
‘At’ designates S’s performing A at t.
It ought to be clear by now that the obtaining of t is insufficient for the truth of pt: pt’s
being true depends not only on what is true at t but also on what is true at t-.
Moreover, the obtaining of t- is hardly sufficient for the truth of p-: if t- were the last
moment of time, nothing would be up to anyone anymore. So, p- and pt are both
propositions about more than one time, with the former pointing toward the future
and the latter pointing to the past. If one were metaphorically inclined, one might say
that “up to” connects (i) S at t- to (ii) At.
Speaking without metaphor, I take the “up to” idiom to be helpful for two reasons:
first, it draws attention to the temporal asymmetry of free action; second, it allows us
to see that for every free action, insofar as it is a free action, there is a diachronic
relation that obtains between (i) an agent S at a time t- and (ii) At, where t is later than
t-. For every free action, insofar as it is a free action, there is a transition from “it is up
to _ whether _” to “it was up to _ whether _.”
Within the free will debate, there is no language with which to refer to this relation.
As such, I will simply stipulate a name: insofar as an agent S freely performs an act A
at time t, there is a time t- such that the diachronic Transition relation obtains
between (i) S at t- and (ii) At:
(TA) An agent S freely performs A at t =df. ∃t- (The Transition relation obtains
between S at t- and At).
And, of course, (TA) might just as well be expressed as:
(TA*) The conjunction of (i) pAt and (ii) It is up to S at t- whether pAt =df. The
Transition relation obtains between S at t- and At.32
It is trivially true that the notions of free action, up to, and Transition relation are
inter-definable.
I should add that what I say here about its being up to an agent which action she
performs applies, mutatis mutandis, to any state of affairs such that it is up to an
agent whether that state of affairs obtains. If:
‘O’ designates the obtaining of some state of affairs O.

32
Of course, in most cases, there are many times at which it is up to an agent which action she performs.
As such, it may seem problematic that I am defining the Transition relation in terms of a single time, t-. I do
so only for ease of expression. If it is up to S at t0, t1, and t2 whether S performs A at t, this conjunction is
true: (The Transition relation obtains between S at t0 and At) & (The Transition relation obtains between S
at t1 and At) & (The Transition relation obtains between S at t2 and At). For every substantive point I make,
I could express it by using such a conjunction.
     

And:
‘PO’ designates the proposition that O obtains.
Then:
(TO*) The conjunction of (i) PO and (ii) It is up to S at t- whether PO =df. The
Transition relation obtains between S at t- and O.33
I note that if ever the Transition relation obtains—between, for instance, S at t- and
At—it follows of necessity34 that it might not have. After all, that it is up to S at
t- whether S performs A at t entails that S might not perform A at t. So, if S does not,
in fact, perform A at t, S at t- stands in the Transition relation to something other
than At.
At this point, one might wonder why I insisted that there is no language with
which to refer to the Transition relation and that it is thus necessary to stipulate to a
name. In response, I note that I am not making an interesting point about natural
language,35 but am simply acknowledging that within the free will debate, there is no
term of art with which to refer to the relation in question. Given that ‘free action’ is
itself a term of art, one needs a term of art with which to refer to the relation that an
agent bears to her free actions insofar as they are free.
Of course, one might ask about ‘choosing’ (or ‘willing’ or something similar). As
I already mentioned, it is typical to construe immediately free acts as choices; to ask
whether one bears the “choosing” relation to a choice would lead to all sorts of
confusion. More to the point, if a choice is a mental act that occurs at a time t, it is
certainly not trivially true that is up to the relevant agent at an earlier time t- whether
she performs the act that is the choice. Indeed, given that it is possible for someone to
perform the act of making a choice even if determinism is true, the libertarian cannot
contend that “choosing” is the relation that free agents qua free agents bear to free
acts qua free acts. In order to capture the triviality of the claim that an agent acts
freely only if she stands in the Transition relation to performing some action, it is
necessary to introduce a term of art.
With this discussion of the Transition relation in place, I am able to move forward
with my presentation of the Mind argument. I here shift my focus from that which is
trivial to that which is substantive. In particular, I will begin to consider accounts of
free action. To give an account of free action is to explain what it is in virtue of which
a free action is free. It ought to be clear by now that one cannot provide such an
account by simply describing the state of an agent before she acts freely or the free
action itself. To give an account of free action requires explaining the relation that
obtains between an agent and an action. More precisely: to provide an account of free

33
In what follows, I will use ‘PO’ and ‘O’ as I use them here. Moreover, I will use similar notation when
I speak of other states of affairs. If ‘O*’ designates some state of affairs O*, then ‘PO*’ designates the
proposition that O* obtains and ‘O*’ designates the obtaining of O*.
34
That is, broadly logical necessity.
35
Though I do not argue for this here, I do, in fact, believe that it is philosophically significant that
within the English language there is no way to refer to the Transition relation. Indeed, I suspect that not
only are other natural languages limited in this way but that these limitations are non-accidentally related
to the difficulty of providing a satisfactory account of free action.
  

action is to explain what it is in virtue of which an agent at an earlier time bears the
Transition relation to the act that she freely performs. I turn now to consider what
such an account might be.

6.8.3 Grounding
Though one might find the notion of “in virtue of ” obscure, it has lately received a
great deal of philosophical attention. To be more accurate, the notion of “grounding”
has recently generated an enormous philosophical literature, and participants in the
grounding debate regularly invoke the notion of “in virtue of” in order to convey
what the grounding relation is supposed to be.36 One might construe grounding in
terms of ontological (not temporal) priority and ontological (not temporal) poster-
iority: that which is ontologically prior grounds that which is ontologically posterior.
In explaining grounding, one might also invoke ontological (not causal) dependence
relations: if one state of affairs ontologically depends on another, the obtaining of the
latter grounds the obtaining of the former. Grounding relations are asymmetrical and
irreflexive; moreover, they are non-diachronic: insofar as distinct states of affairs
obtain in temporal succession, neither one grounds the other.37
If states of affairs are related non-diachronically, they might be related atemporally
or synchronically.38 When Socrates asked whether the pious is pious because the
gods love it, or whether the gods love the pious because it is pious, he asked a
question about atemporal ontological priority relations. Moreover, if p and <It is true
that p> are distinct propositions, and if each of them is true, the truth of the former
atemporally grounds the truth of the latter.
It is worthwhile to note that in the latter case, it seems that there is an infinite series
of distinct propositions, each of which is grounded in the truth of p. In such a case,
one might say that there is a non-vicious regress.
Indeed, it may be that the grounding relation is best explained in terms of the
distinction between vicious and non-vicious regresses. After all, it seems that the
defining feature of a non-vicious regress is that it obtains in virtue of the obtaining of
the first element in the series. The first element is ontologically prior to the other
elements; it “brings the series with it,” so to speak. In the case of a vicious regress,
however, the order of ontological dependence is reversed. The first element is
ontologically posterior to the second, the second element is ontologically posterior
to the third, the third element is ontologically posterior to the fourth, and so on. It is
logically impossible for the first element to “bring the series with it,” because the first
element does not exist until an infinite series is complete. And that, obviously, cannot
happen.
In any case, synchronic, rather than atemporal, relations are relevant in the present
context. Standard examples of synchronic grounding relations include relations
between mental and the physical events, and truthmakers and whatever they

36
See, for instance, Audi (2012).
37
However, this is not to suggest that if the obtaining of a state of affairs O* grounds the obtaining of a
state of affairs O, O* and O obtain at only one time. The point is that if a state of affairs O obtains across
times t and t+, O* also obtains across times t and t+.
38
Thanks to John Keller for bringing this distinction to my attention.
     

make true.39 Furthermore, it seems that the claim that “A thing has its dispositional
features in virtue of its categorical features”40 is also a claim about synchronic
grounding, as is the claim that, “Everything that happens can be explained in terms
of what happens on the ‘fundamental physical level.’ ”
As far as my presentation of the Mind argument goes, I am concerned with what
I call nomological grounding relations. If O* nomologically grounds O, (i) O* and O
are distinct states of affairs that obtain simultaneously and (ii) O* nomologically
necessitates O. In this case:
(NN) □ ((PO* & LW) → PO).
Of course, a similar statement was entailed by the definition of determinism. The
relevant distinction is that while nomological grounding relations obtain between
states of affairs that obtain simultaneously, determinism is a thesis about states of
affairs that obtain at different times. In order to highlight this distinction, it is useful
to use the term nomological determining to refer to nomological necessitation
relations that obtain in virtue of an earlier state of affairs’ nomologically necessitating
a later state of affairs.
If libertarianism is true, and if an agent performs an immediately free action, no
state of the world nomologically determines that the agent performs the act in
question. However, libertarianism is consistent with the thesis that an instance of
the Transition relation is nomologically grounded in O*, where O* is distinct from the
obtaining of this instance of the Transition relation. Indeed, I take it that in offering an
account of free action, a libertarian offers nothing other than an account of the sorts of
states of affairs that nomologically ground instances of the Transition relation.
This point is perhaps easiest to appreciate if one assumes the truth of libertarian-
ism and considers a specific case of an agent’s performing an immediately free act in
world W. For instance, one might consider Mary, an agent such that it is up to her, at
t-, whether to decide to marry Harry, and who performs, at t, the immediately free act
of deciding in his favor. In this case, an account of free action would explain what it is
in virtue of which Mary’s decision is free, or what it is in virtue of which the
Transition obtains between Mary at t- and her decision at t.
According to an event-causal account of free action,41 Mary’s decision is free in
virtue of the fact that some event e- that occurs at t- indeterministically causes some
event e that occurs at t. Different event-causal accounts obviously offer different
theses about what kinds of events the causes and effects must be, but in each case, the
obtaining of the indeterministic causal relation between some event e- that occurs at
t- and some event e that occurs at t is ontologically prior to the obtaining of the
Transition between the agent at t- and the agent’s immediately free act at t. Moreover,
while the event causal theorist is not committed to the thesis that e-’s causing e
metaphysically necessitates that Mary at t- bears the Transition relation to her
decision at t, she certainly seems to be committed to the thesis that the instance of
event causation nomologically necessitates the obtaining of the Transition.

39 40
Jonathan Schaffer offers these examples in his (2009: 364–5). Audi 101.
41
See Ekstrom (2001; 2003; 2011), Franklin (2011), and Kane (1996; 2011) for event-causal accounts of
free action.
  

In order to see that this is so, one might imagine that an indeterministic event-causal
account of free action is correct, and that for every state of the world at an instant that
obtains in W, the very same state of the world obtains in W*. So, Mary at t- in W* is just
as she is at t- in W; likewise, Mary at t in W* is in precisely the same state as Mary at t in
W, so Mary at t in W* is making the decision to marry Harry. Moreover, W*, like W, is
governed by laws of nature LW. Finally, e-’s causing e—the very state of affairs that, in
W, accounts for Mary’s deciding freely rather than not—also obtains in W*.
Given these features of the case, it seems impossible that Mary in W freely decides
to marry Harry, while Mary in W* makes the same decision non-freely. Unless this is
possible, the event-causal libertarian is committed to the thesis that the obtaining of
the causal relation between e- and e nomologically necessitates that the Transition
obtains between Mary at t- and her decision at t to marry Harry.
Proponents of agent-causal accounts of free action42 insist that it is impossible for a
causal relationship between events to account for a free act’s being free. Instead, agents
themselves—substances, persons—cause their free acts insofar as these acts are free. If
agent causation provides an account of free action, each instance of the Transition
relation is ontologically dependent on an instance of agent causation. Moreover, if an
agent causes an act, it is nomologically necessary that the act in question is free. Indeed,
one might reasonably suppose that it is metaphysically impossible for an agent to cause
an act without performing it freely. At a minimum, though, a nomological necessitation
relation obtains between an instance of agent causation and an instance of free action.
Another brand of libertarianism, simple indeterminism, offers that an immediately
free act’s being free may be accounted for by (i) the intrinsic qualities of the act itself
and (ii) the act’s occurring indeterministically.43 If one stipulates that Mary’s decision
at t has the relevant intrinsic features, the fact that Mary at t- bears the Transition
relation to her decision is grounded in nothing other than (i) the occurrence of the
event that is Mary’s decision at t and (ii) that the state of the world at t- does not
nomologically determine that this event occurs. If simple indeterminism is true, it
follows trivially that the obtaining of the Transition between Mary at t- and Mary’s
decision at t is nomologically necessitated by the state of affairs that grounds this
obtaining of the Transition relation: given that Mary makes the decision that she does,
the decision is free for no reason other than that the laws of nature are indeterministic.
At this point, it should be clear what I mean by the notion of nomological
grounding and why accounts of free action ought to be construed as theses about
what kinds of states of affairs ground the obtaining of the Transition relation.

6.8.4 No Grounding?
With this, I will consider the implications of the thesis that it is “basic” or “primitive”
or “ungrounded” that a free act is free.44 That is, I will consider the implications of

42
See Clarke (2003; 2011) and O’Connor (2000; 2010) for agent causal accounts of free action.
43
See Ginet (1990; 1997), Goetz (1997), and McCann (1998) for simple indeterministic accounts of free
action.
44
The argument of this section is inspired by conversations with van Inawgen as well as Colin McGinn’s
arguments against the thesis that some philosophically relevant entities are ontologically irreducible. See
McGinn (1993) for a thorough presentation of the arguments.
     

the thesis that (i) if an agent S performs an act A freely at time t, and (ii) if t- is the
time at which it is up to her whether to do so, then (iii) there is no state of affairs O*
such that O* nomologically grounds the obtaining of the Transition relation between
S at t- and At. Toward that end, I return to the case of Mary. I imagine that W, the
world in which she performs the immediately free act of deciding, at t, to marry
Harry, is empty except for whatever must exist for Mary at t- to bear the Transition
relation to her decision, at t, to marry Harry. I add that Mary does not act freely prior
to t, t is the last time that obtains in W, no state of affairs that is even partially
grounded in the obtaining of the Transition obtains, and the obtaining of the
Transition bears no diachronic necessitation relations to the obtaining of any state
of affairs in W. I further stipulate that the obtaining of the Transition between Mary
and her decision is nomologically ungrounded in W.
Given the definition of nomological grounding, it follows that there is a world W*
that is exactly like W, except that in W*, Mary at t- does not bear the Transition
relation to her decision, at t, to marry Harry. By hypothesis, both W and W* include
t- and t. So, with respect to the properties that Mary has at t-, there is no difference
between W and W*. As such, if Mary at t- in W has certain desires, beliefs, goals,
values, hopes, fears, and dreams, it is trivially true that she has them at t- in W*;
likewise, mutatis mutandis, for whatever properties she has at t. Moreover, with
respect to the diachronic relations that Mary stands in at t- and t, Mary in W and
Mary in W* are almost perfectly alike: not only does Mary at t in W make the same
decision as Mary at t in W*, she makes the decision after the same process of
deliberation, having been influenced by exactly the same factors, and subject to the
same laws of nature. Except that Mary in W stands in the Transition relation to her
decision at t to marry Harry, Mary in W and Mary in W* are similar in every way.45 Of
course, given the inter-definability of the Transition relation and an agent’s being both
(a) able to perform an act and (b) able to refrain from performing it, it follows that
even though Mary at t- in W is in exactly the same situation as Mary at t- in W*, Mary
at t- in W is able to refrain from deciding to marry Harry, but Mary at t- in W* is not.
Of course, this point about Mary in W and Mary in W* can be generalized.46 I take
it to be obvious that Mary in W acts freely if and only if Mary in W* acts freely as well.
As such, I affirm what I call the grounding thesis:
(GT) Necessarily, if some agent S at some time t- bears the Transition relation to
O, some state of affairs O* obtains across times t- and t, and O* nomologically
grounds this obtaining of the Transition between S at t- and O.
While I take this thesis to be obvious, the case of Mary in W and Mary in W* hardly
constitutes a proof: one may bite the proverbial bullet. If one does so, rejecting my

45
With one caveat: there may also be differences that are grounded in whether or not the Transition
obtains. For instance, while it is clearly impossible for Mary in W* to know that she acts freely, this may not
be true of Mary in W.
46
Of course, the same considerations that undermine the thesis that the obtaining of the Transition
relation is ungrounded also undermine the two-part thesis that (i) for an agent to cause an act “just is” for
the Transition to obtain between the agent and the act in question and (ii) the obtaining of the agent-causal
relation is “primitive” or “ungrounded.”
  

formulation of the Mind argument will be easy. For the rest of us, though, it may not
be as clear how to proceed.

6.8.5 A Dilemma
Once the grounding thesis is made explicit, a dilemma comes into view: if O*
nomologically grounds the obtaining of the Transition between an agent S at time
t- and At either it is up to S at t- whether O* obtains or it is not.
If one accepts the validity of the Transfer principle and the truth of (LAP), one
must avoid the second horn of the dilemma. In arguing that this is so, I stipulate that:
‘PTSAt’ designates the proposition that the Transition relation obtains between S at
t- and At
and:
‘GW (O*, TSAt)’ designates the proposition that, in W, O* nomologically grounds
the obtaining of the Transition relation between agent S at t- and At
Assuming the truth of GW (O*, TSAt), I offer that:
29. GW (O*, TSAt) assumption
30. □ ((PO* & LW) → PTSAt) 29, definition of nomological grounding
31. □ (PTSAt → pAt) definition of the Transition
32. □ ((PO* & LW) → pAt) 30, 31
33. □ Ns,t-(PO*)W assumption, second horn of dilemma
34. □ Ns,t-(PO* & LW)W 33, (LAP)
35. □ Ns,t-(pAt)W 32, 34 Transfer
So, if the grounding thesis and (LAP) are true, and Transfer is valid, the libertarian
must reject the second horn of the dilemma. That is, she must affirm that if O*
nomologically grounds the obtaining of the Transition between S at t- and At, it is up
to S at t- whether O* obtains.
What are the consequences of this affirmation? First, one must remember that it is
up to S at t- whether O* obtains if and only if the Transition obtains between S at
t- and O*. And now a new question arises: is this obtaining of the Transition between
S at t- and O* nomologically grounded in some O**? Given the grounding thesis, it
follows that it is. But now the second dilemma rears up once again: is it up to S at
t- whether O** obtains, or not? Let:
‘GW (O**, TSO*)’ designate the proposition that, in W, O** nomologically grounds
the obtaining of the Transition relation between agent S at time t- and O*
and
‘PTSO*’ designate the proposition that the Transition relation obtains between S at
t- and O*.
In that case:
36. GW (O**, TSO*) assumption
37. □ ((PO** & LW) → PTSO*) 36, definition of nomological grounding
     

38. □ (PTSO* → PO*) definition of the Transition

39. □ ((PO** & LW) → PO*) 37, 38
40. □ Ns,t-(PO**)W assumption, second horn of dilemma
41. □ Ns,t-(PO** & LW)W 40, (LAP)
42. □ Ns,t-(PO*)W 39, 41 Transfer

But, of course, (42) is identical to (33). So, if the consequence argument is valid and
(33) is true, it is not up to S at t- whether she performs A at t.
With this, a vicious regress comes into view. If an agent S freely performs A at time
t, the Transition obtains between S at t- and At. But:
T1. The Transition obtains between S at t- and At only if some ontologically prior
state of affairs O* obtains (Grounding thesis).
T2. O* obtains only if the Transition obtains between S at t- and O* (Argument
via Transfer).
T3. The Transition obtains between S at t- and O* only if some ontologically prior
state of affairs O** obtains (Grounding thesis).
T4. O** obtains only if the Transition obtains between S at t- and O** (Argument
via Transfer).
T5. The Transition obtains between S at t- and O**only if some ontologically
prior state of affairs O*** obtains (Grounding thesis).
T6. O*** obtains only if the Transition obtains between S at t- and O*** (Argu-
ment via Transfer).
And so on, ad infinitum. Although neither Transfer nor the grounding thesis
generates the regress on its own, it seems that the conjunction of them presents a
problem for the libertarian who endorses the consequence argument.

6.8.6 An Objection
Before I complete my discussion of the Mind argument, I should return to my
assumption that:
(N*) □(pW & ∀S∀t (There is nothing S at t can do such that, if S at t were to do it,
p would be false).
is the same proposition as:
(N) □(pW & ∀S∀t(It is not up to S at t whether p)).
After all, if (N) is equivalent not to (N*) but to:
(M*) □(pW & ∀S∀t (There is nothing S at t can do such that, if S at t were to do
it, p might be false),
the assumption with which I began my regress argument, □ Ns,t-(PO*)W, is false.
Indeed, a libertarian may point out that since, in W, (i) the obtaining of O*
nomologically grounds the obtaining of the Transition between S at t- and At and
(ii) t is the determiner of At , it is nomologically possible at t- that O* does not obtain.
  

Moreover, one might take “It is nomologically possible that O* does not obtain” to
express the proposition that PO* might be false. This echoes what Finch and Warfield
say about the premise that N (DB occurs → R occurs). As they explained, if
indeterminism is true, there is indeed something the thief can do such that, if he
were to do it, R (his act of repentance) might not follow DB (the relevant desire-belief
complex): he can do whatever he actually does. Likewise, if indeterminism is true in
W, no matter what S does at t-, O* might not obtain. In this case, S at t- in W can do
something such that, if she were to do it PO* might be false.
As I have argued in Finch (2013a), the problem with this response is that it has the
odd result that anyone who does anything at t- does something such that, if she were
to do it, PO* might be false. But then S has no more power over the truth value of PO
than does anyone else. After all, (PO* & LW) entails pAt, and, ex hypothesi, (PO* & LW).
Given Transfer, it follows that if it is up to S at t- whether pAt, it is up to S at t-
whether (PO* & LW). But it follows by way of (LAP) that if it is up to S at t- whether S
performs A at t, it is up to S at t- whether PO*. So, if the consequence argument is
sound, it is up to S at t- whether S performs A at t if and only if it is up to S at t-
whether PO*. But, according to the hypothesis under consideration, it is up to S at t-
whether PO* if and only if it is up to everyone who exists at t- whether PO*. So, then, it
is up to S at t- whether S performs A at t if and only if it is up to everyone who exists at
t- whether S performs A at t. There are many agents such that it is up to them at t-
whether S performs A at t, and S is just one of the many.47
While this conclusion is not, strictly speaking, absurd, it is implausible enough
that it seems reasonable to conclude that if the consequence argument is sound, so is
the regress argument.

6.9 Conclusion
With the argument for the grounding regress in place, my formulation of the Mind
argument is complete. Granted, the premises I offer are not van Inwagen’s. None-
theless, it seems that the dialectical situation is much the same as the situation that
van Inwagen described in The Essay: one may deny the validity of the consequence
argument, admit that the Mind argument is sound, or insist that one of the premises
of the Mind argument is false, though it seems for all the world to be true. In this
respect, The Essay was right all along.

Acknowledgments
I thank Baron Reed, Mike Rea, and the anonymous referees for this volume who provided me
with helpful commentary on earlier drafts. I am indebted to John Keller, whose careful reading
and suggestions for improvement made this paper significantly better than it would have been
otherwise. And, of course, I am grateful to Peter van Inwagen for all that he has taught me.

47
This paragraph is taken, almost verbatim, from Finch (2013a: 292).
     

Bibliography
Audi, P. (2012), ‘A Clarification and Defense of the Notion of Grounding’, in F. Correia and
B. Schneider (eds.), Metaphysics and Grounding: Understanding the Structure of Reality
(Cambridge University Press).
Clarke, R. (2003), Libertarian Accounts of Free Will (Oxford University Press).
Clarke, R. (2011), ‘Alternatives for Libertarians’, in R. Kane (ed.), The Oxford Handbook of Free
Will, 2nd edn. (Oxford University Press), 329–48.
Ekstrom, L. (2001), ‘Indeterminist Free Action’, in L. Ekstrom (ed.), Agency and Responsibility
(Westview).
Ekstrom, L. (2003), ‘Free Will, Chance, and Mystery’, in Philosophical Studies 113: 153–80.
Ekstrom, L. (2011), ‘Free Will is Not a Mystery’, in R. Kane (ed.), The Oxford Handbook of Free
Will, 2nd edn. (Oxford University Press), 366–80.
Finch, A. (2013a), ‘Against Libertarianism’, in Philosophical Studies 166: 475–93.
Finch, A. (2013b), ‘In Defense of the Consequence Argument: Time, Modality, and the Nature
of Free Action’, in Philosophical Studies 163: 151–71.
Finch, A., and Rea, M. (2008), ‘Presentism and Ockham’s Way Out’, in J. Kvanvig (ed.), Oxford
Studies in Philosophy of Religion (Oxford University Press).
Finch, A., and Warfield, T. (1998), ‘The Mind Argument and Libertarianism’, in Mind 107: 515–28.
Franklin, C. (2011), ‘Farewell to the Luck (and Mind) Argument’, in Philosophical Studies 156:
199–230.
Ginet, C. (1990), On Action (Cambridge University Press).
Ginet, C. (1997), ‘Freedom, Responsibility, and Agency’, in Journal of Ethics 1: 85–98.
Goetz, S. (1997), ‘Libertarian Choice’, in Faith and Philosophy 14: 195–211.
Hobart, R. (1934), ‘Free Will as Involving Determination and Inconceivable Without It’, in
Mind 43: 1–17.
Kane, R. (1996), The Significance of Free Will (Oxford University Press).
Kane, R. (2011), ‘Rethinking Free Will: New Perspectives on an Ancient Problem’, in R. Kane
(ed.), The Oxford Handbook of Free Will 2nd edn. (Oxford University Press), 381–404.
Lewis, D. (1981), ‘Are We Free to Break the Laws?’, in Theoria 47: 113–21.
McCann, H. J. (1998), The Works of Agency: On Human Action, Will, and Freedom (Cornell
University Press).
McGinn, C. (1993), Problems in Philosophy: The Limits of Inquiry (Oxford University Press).
McKay, T., and Johnson, D. (1996), ‘A Reconsideration of an Argument Against Compatibi-
lism’, in Philosophical Topics 24: 113–22.
Nowell-Smith, P. (1948), ‘Free Will and Moral Responsibility’, in Mind 55: 45–61.
O’Connor, T. (2000), Persons and Causes: The Metaphysics of Free Will (Oxford University
Press).
O’Connor, T. (2010), ‘Agent-Causal Theories of Freedom’, in R. Kane (ed.), Oxford Handbook
on Free Will 2nd edn. (Oxford University Press).
Plantinga, A. (1976), ‘Actualism and Possible Worlds’, in Theoria vol. 42: 139–60.
Schaffer, J. (2009), ‘On What Grounds What’, in D. Chalmers, D. Manley, and R. Wasserman
(eds.), Metametaphysics: New Essays on the Foundations of Ontology (Oxford University Press).
Smart, J. (1961), ‘Free-Will, Praise and Blame’, in Mind 70: 291–306.
Tognazzini, N. (2015), ‘Grounding the Luck Objection,’ in Australasian Journal of Philosophy
93: 127–38.
Van Inwagen, P. (1983), An Essay on Free Will (Clarendon Press).
Van Inwagen, P. (1986), ‘Two Concepts of Possible Worlds’, in Midwest Studies in Philosophy
11: 185–213.
Van Inwagen, P. (2000), ‘Free Will Remains a Mystery’, in Philosophical Perspectives 14: 1–20.
 7.1
Symposium on the Fixity
of the Past
Incompatibilism and the Fixity of the Past

Neal A. Tognazzini and John Martin Fischer

7.1.1
A style of argument that calls into question our freedom (in the sense that involves
freedom to do otherwise) has been around for millennia; it can be traced back to
Origen. The argument-form makes use of the crucial idea that the past is over-
and-done-with and thus fixed; we cannot now do anything about the distant past (or,
for that matter, the recent past)—it is now too late. In ancient and medieval times, the
argument was in service of “fatalism”—logical and theological. That is, the argument
purported to show that prior truth-values of statements about future human behav-
ior make it the case that the behavior is not free, or that God’s prior beliefs about
future human behavior similarly make it the case that the behavior is not free.
In the modern era, with the rise of science, a new way of filling in the venerable
argument-template has emerged, issuing in what Peter van Inwagen has called “The
Consequence Argument”.1 Van Inwagen gave it this name because, if causal deter-
minism is true, then all our behavior is the consequence of the past and laws of
nature. More specifically, if causal determinism is true, statements about the past
(intrinsically construed—the temporally nonrelational past), together with the laws
of nature, entail all truths about present and future human behavior. It can thus seem
mysterious how, if causal determinism were true, human beings could be free. It can
thus also seem unclear that we can legitimately be deemed morally responsible for
any of our behavior.
Various contemporary philosophers have sharpened these worries into an argu-
ment (with the same general form as the ancient argument stemming from the fixity
of the past).2 Peter van Inwagen’s great book, An Essay on Free Will, built on the
work of these philosophers by presenting and defending the argument in perhaps the

1
Van Inwagen (1983: 16).
2
See, for instance: Ginet (1966: 87–104); and Wiggins (1973: 31–62).
        

clearest and most forceful way ever. It has thus been an enormously important and
influential work, for which we have great admiration.
This is not to say that all philosophers have rushed to accept van Inwagen’s view
that the Consequence Argument is sound. Some philosophers have resisted its
validity, whereas others have called into question its soundness.3 In fact, there is a
strong argument to be made that the debate over whether determinism rules out the
ability to do otherwise is at a stalemate.4 Incompatibilists think that evaluating claims
about what agents can do requires holding fixed the past and the laws of nature;
compatibilists disagree. In particular, whereas the incompatibilist argument is motiv-
ated crucially by appeal to two “fixity” principles—the Principle of the Fixity of the
Laws and the Principle of the Fixity of the Past—various compatibilists simply think
we ought to reject one or the other of those principles. (“Multiple-pasts compatibi-
lists” think we need not hold fixed the past when evaluating claims about what an
agent can do, and “local miracle compatibilists” think we need not hold fixed the
laws.) There appears to be little that either side can say to convince the other, and it is
difficult to see how one might successfully argue for the relevant fixity principles that
drive the incompatibilist argument.
Recently, however, in a bold new paper, Wes Holliday has attempted to break this
seeming stalemate by presenting a new argument for the Principle of the Fixity of the
Past.5 Holliday’s argument is subtle and ingenious, and worthy of serious consider-
ation, especially given the promise it holds for genuinely advancing this old debate. In
what follows, however, we argue that despite its considerable ingenuity, Holliday’s
argument fails to convince, and the stalemate appears to remain.

7.1.2
First, let us take a look at the relevant argument for incompatibilism, in which the
Principle of the Fixity of the Past features so prominently. One version, stated
informally, runs as follows:
1. If determinism is true, then the past and the laws of nature together entail every
action that anyone ever actually performs.
2. So, if determinism is true, then if someone were to do anything other than what
they actually do, either the past would have to be different or the laws of nature
would have to be different.
3. But if, in order for someone to do otherwise, the past would have to be different,
then that person cannot do otherwise.
4. And if, in order for someone to do otherwise, the laws would have to be
different, then that person cannot do otherwise.

3
For developments and evaluations of such strategies, see, for example: Slote (1982: 5–24); Lewis (1981:
113–21); and Fischer (1994: Chapter 4).
4
John Martin Fischer makes a detailed case for this conclusion in The Metaphysics of Free Will, and van
Inwagen’s own considered view is that it is utterly mysterious how any of us can possibly have free will,
given the “seemingly unanswerable” arguments for its incompatibility with both determinism and inde-
terminism. See van Inwagen (2008: 327–41).
5
Holliday (2012: 179–207).
  .     

5. So, if determinism is true, then no one can perform any action other than the
actions they actually perform.
Premise 3 is an informal statement of the Principle of the Fixity of the Past, and
premise 4 is an informal statement of the Principle of the Fixity of the Laws. These
two premises are the controversial premises in the argument, and they are at the
heart of the alleged stalemate.
For the sake of his paper, Holliday presupposes that premise 4 is true, and instead
focuses his attention on premise 3, the Principle of the Fixity of the Past, offering a
new argument for it.6 His argument—the “Action-Type Argument for the Principle
of the Fixity of the Past” (p. 189)—relies crucially on introducing the action type
action that is inconsistent with the past, which he proposes to understand as follows:
an action such that if s were to do it, then the past would (have to) be different.7
Given the connection between this type of action and the Principle of the Fixity of the
Past, we might recast the latter principle for ease of exposition as follows:
(FP) An agent cannot perform an action that is inconsistent with the past.
Holliday’s argument for (FP), in a nutshell, is this: (i) necessarily, no one ever does
perform an action inconsistent with the past; (ii) if there is no world in which an
action of type X is ever performed, then necessarily, no one can perform an action of
type X; so, (iii) necessarily, no one can perform an action inconsistent with the past.8
That is, Holliday attempts to argue for a claim about what agents can do from a claim
about what it is metaphysically possible for agents to do. Since there is no possible
world in which an agent does perform an action that is inconsistent with the past (the
nonoccurrence of such actions is, after all, entailed by the past), no agent can perform
such an action. Again: (i) actions that are inconsistent with the past are impossible
(there is, after all, no possible world at which such an action is performed); (ii) no one
can do the impossible; so, (iii) no one can do anything inconsistent with the past.
And that’s just to say that (FP) is true.
It will be instructive to consider a quick compatibilist reply to this argument, because
it will bring out the need for an important clarification, which will then lead to a full-
blown statement of Holliday’s argument. The quick compatibilist reply is this:
We’re willing to grant that impossibility entails inability, but we deny that it is impossible to
perform an action that is inconsistent with the past. After all, actions that we are determined not
to perform are typically such that, had we wanted to perform them, we would have. In other
words: even if we are actually determined not to perform some action, there will still be plenty of
worlds in which we do perform that action, and those worlds will simply be ones with a different
past. So it’s not impossible to perform an action inconsistent with the actual past; it’s just that no
one ever will perform such an action. And will not manifestly does not entail cannot.

6
Since Holliday merely presupposes (rather than argues for) the truth of the Principle of the Fixity of
the Laws, he acknowledges that his argument does not address so-called local miracle compatibilists, who
reject this premise. See, for example, David Lewis, “Are We Free to Break the Laws?”.
7
Holliday (2012: 186–7). We have omitted the double time-indexing in Holliday’s original explication
of this action type because it is inessential for our purposes.
8
Holliday (2012: 201).
        

This is a typical compatibilist move and Holliday anticipates it. In response, Holliday
distinguishes between two senses in which an action might be said to be inconsistent
with the past, depending on which “past” is at issue. As Holliday puts it, an action
may either be of type F or of type I, where each type can be understood according
to the following functions on world-time pairs (and where ‘w@’ is the name of the
actual world):
F(w, t) = the set of actions inconsistent with the past relative to t of w@.
I(w, t) = the set of actions inconsistent with the past relative to t of w.9
What the quick compatibilist reply gets right is that actions of type F are not thereby
impossible: after all, just because an action is inconsistent with the past of the actual
world does not mean that there is no world in which that action is performed; it is just
that it will only be performed in worlds with pasts that differ in some way from the
actual past. But what the quick compatibilist reply overlooks, according to Holliday,
is that an action one is determined not to perform will also be an action of type I, and
actions of type I are impossible: in no world does anyone perform an action that is
inconsistent with the past of that world. The sense of ‘inconsistent with the past’ at
issue in Holliday’s argument, then, is the sense associated with actions of type I.
Indeed, Holliday maintains that it is precisely because the difference between type F
and type I actions has gone unnoticed that incompatibilists haven’t previously
noticed an argument for the Fixity of the Past to which they are entitled.
With this distinction in mind, we can informally restate Holliday’s argument as
follows: necessarily, no one ever performs an action of type I; if there is no world in
which an action of type X is ever performed, then necessarily, no one can perform an
action of type X; so, no one can perform an action of type I. That is: the Principle of
the Fixity of the Past is true. And then the incompatibilist argument can continue as
follows: actions we are determined not to perform are actions of type I (they fall
within the range of the above function that specifies how we are to understand type-I
actions); so, no one can perform an action that he is determined not to perform.
Holliday gives both an informal ‘natural language’ version of his argument, as well
as a formal version using symbols, and the worries we plan to raise about the
argument require that we have both versions in front of us. So, first, let us fix our
symbols (following Holliday):
D(s, y, w, t) = agent s does action y in world w at time t. (‘D’ for ‘does’.)
C(s, y, w, t) = agent s can, in world w at time t, do action y. (‘C’ for ‘can’.)
X(w, t) = the set of all actions that fall under type X in world w relative to time t.
y ∈ X(w, t) = action y falls under action type X in w relative to t.10
Now, we can state Holliday’s two-premise argument for the Principle of the Fixity of
the Past as follows (p. 201):

9
Holliday (2012: 191).
10
We have simplified Holliday’s formalization somewhat, but in ways that are inessential to the points
we wish to make.
  .     

(1) An agent cannot perform an action of type X if there is no possible world in

which an agent performs an action of type X.
∀X [¬∃s, w, t, y (y ∈ X(w, t) ∧ D(s, y, w, t)) → ∀s, w, t, y (y ∈ X(w, t) → ¬C(s, y, w, t))]
(Read from the symbols: for all action types, if there’s no world-time pair at which
anyone does perform an action which is of a certain type at that world-time pair,
then there’s no world-time pair at which anyone can perform an action which is of
that type at that world-time pair.)
(2) There is no possible world in which an agent performs an action that is
inconsistent with the past (an action of type I).
¬∃s, w, t, y (y ∈ I(w, t) ∧ D(s, y, w, t))
(Read from the symbols: there is no world-time pair at which any agent ever does
perform an action which is type I at that world-time pair.)
(3) Therefore, an agent cannot perform an action that is inconsistent with the past
(an action of type I).
∀s, w, t, y (y ∈ I(w, t) → ¬C(s, y, w, t))
(Read from the symbols: for every agent, action, and world-time pair, if the action
is type-I at that world-time pair, then the agent cannot perform it.)
The argument looks complicated, but in fact it is strikingly straightforward and
elegant: (1) impossibility implies inability; (2) type-I actions are impossible; so, (3)
no one is able to perform type-I actions (i.e. actions that are inconsistent with the
past). The argument is so straightforward, in fact, that it is difficult to see how it
might be resisted. Nevertheless, we think it ought to be resisted, and in the remainder
of the paper we explain how.

7.1.3
We actually have two related worries about the argument, and both are directed at
premise (1). The first worry is that premise (1) begs the question against the
compatibilist because it presupposes an understanding of ‘can’ that the compatibilist
has antecedent reason to reject. The second, related worry is that premise (1) does not
in fact adequately capture the intuitive idea that no one can do what is impossible,
and hence is unmotivated. We’ll begin with the charge of begging the question.
To begin, note that premise (1) is meant to be a generalized statement of the
seemingly uncontroversial thesis that if some action is impossible, then no one can
perform it. What makes it generalized is that it is stated not in terms of particular
actions but in terms of action types. So, at the level of action types, the claim is that if
some action falls under an impossible action type, then no one can perform actions of
that type, where an action type is impossible just in case there is no world in which
anyone ever performs an action of that type. But when we generalize the claim in this
way, we introduce an ambiguity. When we say that there is no world in which anyone
ever performs an action of a certain type, we might mean:
(a) There is no world in which anyone ever performs an action which is, in that
world, of type X.
        

¬∃s, w, t, y (y ∈ X(w, t) ∧ D(s, y, w, t))

or we might mean:
(b) There is no world in which anyone ever performs an action which is, as a
matter of actual fact, of type X.
¬∃s, w, t, y (y ∈ X(w@, t) ∧ D(s, y, w, t))
We are not claiming that Holliday’s first premise is ambiguous, though, because his
formalization makes it clear that he intends reading (a). The formalized version of
premise (1), again, is this:
(1) ∀X [¬∃s, w, t, y (y ∈ X(w, t) ∧ D(s, y, w, t)) → ∀s, w, t, y (y ∈ X(w, t) → ¬C(s,
y, w, t))]
His first premise, then, is a (universally quantified) conditional, which moves from
the claim that there is no world in which anyone does perform an action which is, in
that world, of type X, to the claim that there is no world in which anyone can perform
an action which is, in that world, of type X.
Now, we said earlier that we think this premise presupposes a question-begging
sense of ‘can’, and this is most easily seen by looking at its contrapositive, which gives
a necessary condition on an agent’s ability to perform an action:
(1C) ∀X [∃s, w, t, y (y ∈ X(w, t) ∧ C(s, y, w, t)) → (∃s, w, t, y (y ∈ X(w, t) ∧ D(s, y,
w, t))].
In other words:
(1C) An agent can, at some world-time pair, perform an action that is of type
X at that world-time pair only if there is some world-time pair at which an agent
does perform an action which is, at that world-time pair, of type X.
At first blush, this may seem to say no more than the innocuous and utterly
uncontroversial claim that an agent is only able to do those things that it is possible
to do. But in fact it says much more than that. In particular, it says that an agent is
able to perform an action which falls under a certain type only if there is a possible
world in which some agent performs an action of the same type. It is this restriction—
that an actual-world ability to perform a certain type of action requires an other-
worldly performance of an action of the same type—that causes trouble for Holliday’s
argument. Just as the compatibilist will deny that we need to hold fixed the past and
the laws when evaluating which other-worldly performances are relevant to actual-
world ability claims, so will they deny that we need to hold fixed the action type.
To see the problem with Holliday’s argument more clearly, consider a simple-
minded compatibilist view, according to which an agent can perform an action if and
only if the agent would perform the action were he to desire to perform it.11 Now
suppose, to take Holliday’s example, that in the actual world Themistocles is

11
This simple conditional analysis of ‘can’ is problematic for well-known reasons, but its simplicity
makes it well-suited for illustrating our point, which is a point about compatibilist views of ‘can’ in general,
even more sophisticated ones.
  .     

determined (i.e. causally determined) not to send his fleet to Corinth. The question is:
despite his being determined not to choose Corinth, can he? Holliday says no, since
choosing Corinth is inconsistent with the past, and in no world does anyone perform
an action that is inconsistent with the past. But our simple-minded compatibilist will
be puzzled why there needs to be such a world in order for Themistocles to be able to
choose Corinth. After all, on this compatibilist’s view, so long as there is some sphere
of nearby worlds in which Themistocles both desires to choose Corinth and does
choose Corinth, that will be sufficient for the truth of the claim that Themistocles can
choose Corinth. True, the compatibilist will admit, in those nearby worlds, choosing
Corinth is not an action that is inconsistent with the past (those worlds have different
pasts), so choosing Corinth is not, in those worlds, type I. It is, of course, type I in the
actual world, but our compatibilist will not be bothered by that, for his analysis of
‘can’ does not require that we hold fixed the type of action when evaluating claims
about what agents can do.
In his attempt to motivate premise (1), Holliday points out that when it is true that
an agent can perform an action, “there should be some possible world that ‘witnesses’
the truth of this can-claim” (p. 194). But our envisaged compatibilist does not deny
this; rather, the issue is whether the witnessing world needs to be one at which the
action is of the same type as it is in the world of the can-claim. Holliday’s premise (1)
requires that it is; but the compatibilist will maintain that this requirement stacks the
deck against him. Of course there is no world in which an action is performed which
is, in that world, inconsistent with the past. But all the compatibilist requires is that
there be some suitable world in which an action is performed which is inconsistent
with the past of our world (thus ‘witnessing’ the truth of the actual-world can-claim
despite the truth of determinism). And this means that our “quick compatibilist
reply” from above is actually, suitably embellished, the right reply to make to Holli-
day’s argument: just because an action is inconsistent with the past of the actual
world does not mean that there is no world in which that action is performed, and
just because an action falls under an impossible action type does not mean that there
is no world in which that action is performed; it’s just that such an action will only be
performed in worlds where it doesn’t fall under that type. And the existence of at least
some of these worlds will be sufficient for the truth of the claim that an agent can
perform the action despite his being determined not to.
This brings us to our second, related worry about Holliday’s first premise, which is
that it is not adequately supported by the intuitive claim that no one can do what is
impossible. Formulated in the way it needs to be for Holliday’s “Action-Type
Argument” to be valid, it says that no one can do things that fall under action
types that are necessarily uninstantiated. But to say that some action falls under an
action type that is necessarily uninstantiated is much different than saying that the
action is impossible. For although the type “inconsistent with the past” is indeed
necessarily uninstantiated, many actions that fall under this type are nevertheless
possibly performed; it is simply that in those worlds in which they are performed,
they do not fall under that type.
We have said that our two worries are related, so let us briefly explain what we
have in mind here. It is a tricky matter to determine whether an argument is
question-begging, but in order to substantiate that charge, it is clearly not enough
        

merely to point out that the argument contains some premise that an opponent of the
conclusion will reject.12 (If that were sufficient, then every valid argument would
count as question-begging.) We have made that point about Holliday’s argument—
that the compatibilist will reject premise (1) as it is formulated—but we have also said
something more, which is that premise (1) is given no independent motivation. It seems
at first to be motivated by the claim that no one can do the impossible, but we have
argued that this claim does not in fact support the much more robust premise (1).
Without an independent motivation to accept premise (1), then, its role in the
argument is no more than a bit of incompatibilist foot-stamping, and this is the
sense in which it is dialectically infelicitous.
It is controversial, of course, whether the existence of a non-actual possible world
in which some agent performs an action inconsistent with the actual past confers on
the agent the ability, in the actual world, to perform some action that is inconsistent
with the actual past. The compatibilist maintains that the existence of certain of those
other worlds (with alternative pasts) means that agents in the actual world never-
theless can perform actions they are determined not to perform; the incompatibilist
disagrees. But our point is simply that Holliday’s argument does not offer any new
reason for thinking we should go with the incompatibilist here. Instead, it simply
crystallizes the incompatibilist intuition.13

7.1.4
The argument for incompatibilism that relies on the fixity of the past and the laws—the
Consequence Argument—is highly contentious. Whereas many philosophers have
accepted it as sound, others hold that it is unsound. Given the roster of distinguished
philosophers who have vigorously disagreed about the argument, it would be a great
intellectual contribution if someone could establish one of its key premises. In his bold
and fascinating paper, Holliday offers a new argument for the fixity of the past premise.
This premise may well be true, but we have argued that, despite his inventive and
sophisticated argumentation, Holliday has not succeeded in establishing its truth.
Even if Holliday’s notable attempt is not entirely successful, we are all in debt to
Peter van Inwagen for articulating the Consequence Argument in a clear and sharp
way—a way that makes it possible to present and evaluate new and promising
defenses of its premises. It is a great virtue of van Inwagen’s regimentation of the
ancient argument that it provides a framework within which the basic issues can be
addressed in a fruitful way.

12
For an insightful discussion of begging the question, see Lecture 3 of van Inwagen (2006).
13
Moreover, since our critique is aimed at premise (1) of Holliday’s argument, the very same critique
would apply to the “direct” argument for incompatibilism that he presents at the end of his paper, whose
first premise is the same. See Holliday (2012: 202–5).
  .     

Acknowledgments
We are extremely grateful to Patrick Todd and John Keller for detailed comments on this
paper, and for helping us to see how best to articulate the ideas presented here. Thanks also to
the anonymous referees for their comments.

Bibliography
Fischer, J. M. (1994), The Metaphysics of Free Will (Blackwell), Chapter 4.
Ginet, C. (1966), ‘Might We Have No Choice?’, in K. Lehrer (ed.), Freedom and Determinism
(Random House), 87–104.
Holliday, W. (2012), ‘Freedom and the Fixity of the Past’, in The Philosophical Review 121.
Lewis, D. (1981), ‘Are We Free to Break the Laws?’, in Theoria 47: 113–21.
Slote, M. (1982), ‘Selective Necessity and the Free-Will Problem’, in Journal of Philosophy 79:
5–24.
Van Inwagen, P. (1983), An Essay on Free Will (Clarendon Press.)
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
Van Inwagen, P. (2008), ‘How to Think About the Problem of Free Will’, in The Journal of
Ethics 12: 327–41.
Wiggins, D. (1973), ‘Toward a Reasonable Libertarianism’, in T. Honderich (ed.), Essays on
Freedom of Action (Routledge and Kegan Paul), 31–62.
 7.2
Symposium on the Fixity
of the Past
Freedom and Modality

Wesley H. Holliday

7.2.1 ‘Can’ and ‘Possible’

What is the relation between freedom and modality? There is a long history of
attempts to answer this question, from philosophers of the Hellenistic period to
early modernity to the present. In recent history, Peter van Inwagen’s An Essay on
Free Will (1983) was a milestone contribution to the question. By presenting a modal
version of his now famous Consequence Argument, van Inwagen helped bring the
modern possible-worlds picture of modality to the fore in debates about the com-
patibility of freedom and determinism.
What is the relation between what one can do in the actual world and what one
does in other possible worlds? In “Freedom and the Fixity of the Past” (Holliday 2012,
hereafter ‘FFP’), I argued that there is at least (roughly) the following relation:
(1) For any action type X, an agent cannot perform an action of type X if there is
no possible world in which an agent performs an action of type X.1
In this paper, I will provide some further motivation for (1) and reply to objections
to it.
Reasoning from what no one does across possible worlds to what no one can do in
a particular world seems natural. Consider an epistemological example: in a world
where no one ever knows the complete description C of the world, is it nonetheless
the case that an agent can know or could have known C? Chalmers (2012: 49) reasons
as follows:
C . . . will have to be sufficiently encompassing that its truths are jointly true of this world and
this world alone. But then, assuming that no one actually knows C, it will be impossible to
know C. Any world in which someone knows C will differ from the actual world and will

1
To be more precise: an agent cannot in world w at time t perform an action y at t' (t  t') if it is settled
in w at t that y falls in w at t under any action type X for which there is no possible world w* and times t* and
t*' (t*  t*') such that an agent performs an action in w* at t*' that falls under X in w* at t*. For the definition
of ‘settled’, a notion due to John Perry, see FFP: 193.
  . 

therefore be a world in which C is false. But there are no worlds in which someone knows
C and C is false. So no one can know C.

In the last two sentences, Chalmers makes the natural reasoning step noted above.2
What distinguishes (1) from other principles relating freedom and modality is that
it is stated in terms of an action type (see FFP: 189f.). To see how this matters,
consider another example from FFP: imagine a siren so alluring that it is impossible
for anyone in her presence to move away from her; as a result, there is no possible
world in which someone who hears her song ever escapes her. Can someone who
hears the siren’s song perform an action such as, e.g., turning around and walking
away? I don’t think so. Can he run away instead? No. Changing the action doesn’t
help, as long as it is an action that moves the agent from within proximity of the siren
to a place away from her. For by hypothesis there is no world in which an agent
performs such an action.
To relate this reasoning to (1), let M be the action type action that moves the
agent from within proximity of the siren to a place away from her. As an instance of
(1), we have:
(1M) If there is no possible world in which an agent performs an action of
type M, then no agent can perform an action of type M.
Many actions fall under the action type M, including walking, running, skipping, and
jumping away from the siren. By (1M) and our hypothesis, it follows that no agent
can walk, run, skip, or jump away from the siren, in agreement with the previous
reasoning. Thus, the generality obtained by stating (1) in terms of an action type
allows us to draw a general conclusion about all actions falling under the type.
The fact that we can see a natural pattern of reasoning like that above as an instance
of the general form of reasoning according to (1) provides defeasible motivation for
(1). Of course, there are less general principles than (1) that we can see the above
reasoning as following, e.g., (1M) itself, but there are a range of examples accounted
for by (1) but not (1M), including the two examples to be given below (insulting the
gods and lifting the boulder). It is important that these examples, which can be
multiplied, provide motivation for (1) that is independent of any views about the
relation between freedom and determinism. I will return to this point below.
It is also important to preempt a general form of objection to (1). Consider another
mythological example: suppose there is no possible world in which a mortal performs
an action deemed insulting by the gods in that world. Further suppose that in the
actual world, the gods deem dressing up as Zeus an insult. 3 Two philosophers, A and
B, debate whether a mortal in the actual world can dress up as Zeus:

2
Spencer (2013) argues that an agent can know the likes of C and can know Fitch-paradoxical
propositions of the form ‘p and I do not know that p’, which may be in C, so he thinks there are
counterexamples to a principle in the spirit of (1). I side with those like Chalmers who think that C and
Fitch-paradoxical propositions cannot be known (see, e.g., Williamson 2002: Ch. 12), but I cannot do
justice to Spencer’s argument here.
3
We may even suppose that the gods deem choosing to dress up as Zeus an insult, so in the following
arguments, one may take the action in question to be choosing to dress up as Zeus, rather than dressing up
as Zeus.
        

: Given our assumptions, it is evident that no mortal can do so.

: Please explain.
: Since there is no possible world in which a mortal performs an action deemed
insulting by the gods in that world, a mortal cannot perform such an action;
and since dressing up as Zeus is such an action in our world, none of us can
do it.
: Not so fast. I agree that if a mortal can dress up as Zeus, then there is some
world in which a mortal does dress up at Zeus. But there are such worlds:
worlds in which the history of relations between gods and mortals is different,
so dressing up as Zeus is considered a compliment to the gods rather than an
insult! You agree that there are such worlds, do you not?
: I do. But surely a world in which a mortal compliments the gods by dressing up
as Zeus does not serve as a witness for the claim that in our world, where the
gods deem dressing up as Zeus an insult, a mortal can do so.
: Why not?
: According to your reasoning, in order to show that a mortal can perform an
action, such as dressing up as Zeus, which in our world is of some type, such as
action deemed insulting by the gods, we are allowed to find witnessing worlds in
which the action falls under a different type, such as action deemed compli-
mentary by the gods—even though the very reason for worrying that we cannot
dress up as Zeus was that the gods deem it an insult in our world. My friend,
you have committed the famous Fallacy of Switching Types!
(Well, it isn’t famous yet. But in FFP I suggested that the Fallacy of Switching Types,
although not by that name, is one of the mistakes made by philosophers who hold
that the freedom to do otherwise is compatible with determinism. But more on that
shortly.)
: You accuse me of fallacious reasoning. But we agree that there is no possible
world where a mortal performs an action deemed insulting by the gods there.
: We do.
: It follows, according to a possible-worlds analysis of counterfactuals (Lewis
1973), that the following counterfactual is true in our world, where the gods
deem dressing up as Zeus an insult: if a mortal were to dress up as Zeus, then
the gods would not deem it an insult!4 Then since there is a world in which a
mortal dresses up as Zeus and it is not an insult, I see no reason yet why we
cannot do so.
: I am not sure about your counterfactual. But you are assuming that the only
relevant action types are the types that the action would fall under if it were
performed—the types that it falls under in worlds whose “agential accessibility”
from the actual world is precisely what is under dispute—while the action type
the action does fall under is irrelevant. But this assumption will lead you astray.
: How so?

4
Since by assumption there are no worlds where an agent performs an action deemed insulting by the
gods in that world, all of the closest worlds where a mortal dresses up as Zeus are worlds where doing so is
not an insult.
  . 

: Suppose that in the actual world, lifting a particular boulder on Earth requires
at least n joules of energy.5 Also assume that there is no possible world in which
a human performs an action that requires at least n joules of energy. Then
surely a human in our world cannot lift the boulder. But you reason as follows:
by our assumption and the possible-worlds analysis of counterfactuals, if
someone were to lift the boulder, the action of lifting the boulder would require
fewer than n joules of energy; so when we ask ourselves whether a human can
lift the boulder, which in the actual world requires at least n joules of energy, we
should ask ourselves whether there is a possible world in which a human
performs an action that requires fewer than n joules of energy! But that is
absurd.
In FFP, I argued that some of the problems that A has identified with B’s reasoning
are analogous to problems with compatibilist reasoning in the debate over freedom
and determinism (see FFP: 202).
The main argument of FFP shows that a precise version of (1) (see footnote 1)
entails6 the incompatibilist view that in a deterministic world agents cannot do
anything except what they will do.7 I do not have room here to rehearse the
argument, so I refer the reader to FFP. As far as I know, no one has disputed the
validity of this argument for incompatibilism, but only the truth of (1). In my view,
the most promising form of compatibilist response to the argument is the following:
propose a modified version of (1) that restricts the quantification over action types in
(1) to a (perhaps context-dependent [cf. Horgan 1979]) subclass of action types,
identified in a principled way, such that the alternative version explains the initial
appeal of (1) and yet avoids the incompatibilist consequence.8 Another serious

5
This example is from section 4 of FFP.
6
Assuming a non-Humean view of the laws of nature. See section 5 of FFP.
7
Since the precise version only applies to actions and action types at times at which it is settled (see FFP:
193) that the action will fall under the action type—for otherwise the agent may do something so that the
action will not fall under the type—the principle does not have the fatalistic consequence that even in an
indeterministic world, agents cannot do anything except what they will do (see FFP: 197). But here is
another consequence of the principle: since there is no possible world in which a backward time traveler
performs an action that permanently prevents his own creation, if it is settled at a time that killing one of
his forebears falls under that action type, then the backward time traveler cannot do so at that time. Some
have claimed to the contrary that a backward time traveler can do so, but I do not have room to discuss the
issue here (see Vihvelin 1995, references therein, and the subsequent literature).
8
In the dialogue between A and B, we saw a problem with the following modification: only quantify
over the action types that the action would fall under if it were performed. Another problem is that this
restriction makes the principle trivial: for if X is an action type that the action would fall under if it were
performed (and the action itself is not impossible), then of course (according to possible-worlds analyses of
counterfactuals) there is a possible world where an agent performs an action of type X; so we would only
quantify over action types to which the antecedent of the conditional in the principle does not apply. Here
is another proposal that I find problematic: only quantify over the action types that the action would fall
under if the agent were to choose to perform the action. In its favor, this proposal does not commit the
compatibilist to the problem pointed out by A in the case of the boulder; for if the agent were to choose to
lift the boulder, then the action of lifting the boulder would still fall under the type action requiring at least
n joules of energy. Moreover, if this proposal were plausible, it would appear to block the incompatibilist
argument of FFP; for in the running example of FFP, if Themistocles were to choose to send the fleet to
Corinth, then the action of sending the fleet to Corinth might not fall under what I called action type I (see
FFP: 191). However, the argument of FFP can be reformulated in terms of the action of choosing to send the
        

response, which applies as much to compatibilist principles as to (1), is to argue that

there is no objective fact of the matter as to which position is correct, the (context-
ualist) compatibilist or (invariantist) incompatibilist, due to what Unger (1984: 54ff.)
calls the semantic relativity of ‘can’. However, in what follows I will consider two
objections to (1) along different lines.

7.2.2 Objections
In “Incompatibilism and the Fixity of the Past” (2017), Tognazzini and Fischer (hereafter
‘T&F’) raise two worries about (1): “The first worry is that premise (1) begs the question
against the compatibilist because it presupposes an understanding of ‘can’ that the
compatibilist has antecedent reason to reject. The second, related worry is that premise
(1) does not in fact adequately capture the intuitive idea that no one can do what is
impossible, and hence is unmotivated” (144). I will address these worries in reverse.

7.2.2.1 “No Independent Motivation”

FFP does not claim that (1) is supposed to capture the intuitive idea that no one can
do what is impossible, so I will take T&F’s second worry to be that FFP does not
provide, as they put it (147), “independent motivation” for (1). According to T&F,
the problem with (1) is that “it simply crystallizes the incompatibilist intuition” (147).
They seem to mean ‘simply’ in the sense of ‘merely’. But I would rather put the point
a different way: it crystallizes the incompatibilist intuition in a simple way, which is a
virtue. However, I do not agree that the only motivation for (1) comes from
incompatibilism itself. To the contrary, section 7.2.1 of this paper provides motivation
for (1) that is independent of the debate about freedom and determinism.9 The
motivation is of course defeasible, but (1) is not unmotivated.

7.2.2.2 “Begging the Question”

T&F do not explain what it means for the compatibilist to have “antecedent reason”
to reject (1). However, they correctly point out that (1) is inconsistent with standard
compatibilist possible-worlds analyses of ‘can’. Moreover, they say that the compa-
tibilist “will deny” it, “will be puzzled” by it, “will not be bothered by it,” and “will
maintain that it stacks the deck against him” (146), because the necessary condition
for ‘can’ in (1) is not a necessary condition according to compatibilist analyses.10

fleet to Corinth, with the conclusion that Themistocles cannot perform this action when (it is settled that) it
is of type I (indeed, Tognazzini and Fischer (2017) take choosing Corinth to be the action in the argument).
The compatibilist might reply that if Themistocles were to choose to perform the action of choosing to send
the fleet to Corinth, then the action of choosing to send the fleet to Corinth might not fall under type I, so
type I is not in the relevant subclass of action types. But this appeal to “choosing to choose” makes little
sense and leads to a regress.
9
Cf. van Inwagen (1983) on his principle (β): “the examples I gave in support of (β) [see 98] did not
presuppose the incompatibility of free will and determinism” (102).
10
Cf. van Inwagen (1983) on his principle (β): “there are almost certainly philosophers who would say
that the fact that determinism and compatibilism together entail that there are counterexamples to (β)
shows that my use of (β) in an argument for incompatibilism is question-begging. But if this accusation
were right, it’s hard to see how any argument could avoid begging the question” (102).
  . 

What A called in section 7.2.1 the Fallacy of Switching Types, the compatibilist
embraces: “Just as the compatibilist will deny that we need to hold fixed the past and
the laws when evaluating which other-worldly performances are relevant to actual-
world ability claims, so will they deny that we need to hold fixed the action type”
(T&F: 146). In short, compatibilists are happy with their own analysis,11 in terms of
which they can explain why they reject (1). Does this give them an “antecedent
reason” to reject (1)? In the sense that they had their analysis before they saw (1), it is
antecedent. Do they have an “antecedent reason” in a deeper sense?
The beauty of the charge that a premise begs the question is that it promises to
provide a way of undermining a valid argument without arguing that any of the
premises are false. To show that compatibilists have a good reason to reject (1), one
would presumably need to show that the analysis of ‘can’ that compatibilists use to
reject (1) is a good analysis, or give a good counterexample to (1) motivated
independently of compatibilism. But to show that compatibilists have an “antecedent
reason” to reject (1), this hard work is apparently not necessary. As T&F note (146,
note 11), the simple conditional analysis of ‘can’ that they consider is problematic for
well-known reasons. But an argument can still beg the question against a bad
theory.12 If it does, is that a bad thing?
It seems to me that we need a different way of thinking about the debate. In his
fascinating discussion of “Philosophical Failure,” van Inwagen (2006) suggests that we
should not think of a philosophical debate as an exchange between two philosophers
with opposing views, trying to convert each other. Instead, we should think of the
debate on the forensic model: the philosophers’ “purpose is not to convert each other
but rather to convert the audience—an audience whose members (in theory) bear no
initial allegiance to either position, although they regard the question ‘Which of these
two positions is correct?’ as an interesting and important one” (44). Viewing a
philosophical debate in this way has consequences for the idea of begging the question:
[O]n the model of debate I have endorsed, Norma the nominalist need not worry about
whether Ronald the realist will accept her premises. She is perfectly free to employ premises
she knows Ronald will reject; her only concern is whether the audience of agnostics will accept
these premises. Suppose, for example, that she uses the premise, ‘We can have knowledge only
of things that have the power to affect us.’ It may well be that no realist, certainly no realist who
had thought the matter through, would accept that premise. If Norma tried to use this premise
in a debate of the first sort, in an attempt to convert Ronald the realist to nominalism, Ronald
would very likely say, ‘Well, of course I don’t accept that; that just begs the question against my
position.’ But in a debate conceived on the forensic model, Ronald can’t make that response,

11
I am reminded of Nozick’s (1981) remark: “Though philosophy is carried out as a coercive activity, the
penalty philosophers wield is, after all, rather weak. If the other person is willing to bear the label of ‘irrational’
or ‘having the worse arguments,’ he can skip away happily maintaining his previous belief . . . ” (4).
12
Does the argument from FFP and this paper beg the question against the compatibilist? According to
Fischer and Pendergraft (2013), “an argument begs the question just in case the proponent of the argument
has no reason to accept the relevant premise, apart from a prior acceptance of the conclusion” where
‘reason’ does not mean decisive reason but rather a “reason that at least renders it plausible to accept the
relevant premise (apart from accepting the conclusion)” (584). But section 7.2.1 of this paper at least
renders it plausible to accept (1) apart from a prior acceptance of incompatibilism.
        

for the simple reason that what he thinks is quite irrelevant to the logic of the debate. If Ronald
thinks that there is any danger of the agnostics accepting this premise, it will do him no good to
tell the audience that of course no realist would accept this principle and that it therefore begs
the question against realism. He’ll have to get down to the business of convincing the agnostics
that they should reject, or at least not accept, this premise. (46)
I think the same points apply to T&F’s objection. Rather than claiming that (1) begs
the question against the compatibilist, the compatibilist needs to get down to the
business of convincing agnostics that they should not accept (1). One way to do so
would be to convince agnostics that the compatibilist analysis of ‘can’ is correct. T&F
acknowledge the problems with the simple conditional analysis of ‘can’, but perhaps
they have a revived analysis in mind? Or a counterexample to (1) motivated
independently of compatibilism?13 This may not be necessary for making the case
that (1) begs the question against the compatibilist, but it does seem to be necessary
for doing the real work of convincing agnostics that they should not accept (1).

7.2.3 Conclusion
I do not claim that the argument from FFP and this paper counts as a success
according to van Inwagen’s test: would it convince an audience of ideal agnostics?
Who knows? But I would not therefore conclude that the argument is a “philosoph-
ical failure.” Nor would I consider van Inwagen’s Consequence Argument a failure
for that reason. By deriving incompatibilism from more general modal principles,
these arguments clarify the logic and the costs of possible positions about the relation
between freedom and modality.

Acknowledgments
I wish to thank Mark Crimmins for helpful discussion on the topic of this paper and Timothy
Clarke for helpful comments on an earlier draft.

References
Chalmers, D. J. (2012), Constructing the World (Oxford University Press).
Fischer, J. M., and Pendergraft, G. (2013), ‘Does the Consequence Argument Beg the Ques-
tion?’, in Philosophical Studies, Vol. 166, No. 3: 575–95.
Holliday, W. H. (2012), ‘Freedom and the Fixity of the Past’, in Philosophical Review, Vol. 121,
No. 2: 179–207.
Horgan, T. (1979), ‘ ‘Could’, Possible Worlds, and Moral Responsibility’, in The Southern
Journal of Philosophy, Vol. 17, Issue 3: 345–58.
Lewis, D. (1973), Counterfactuals (Basil Blackwell).

13
Cf. van Inwagen (1983) on his modal argument: “If the compatibilist wishes to refute these
arguments—and, of course, nothing obliges him to do this—here is what he will have to do: he will have
to produce some set of propositions intuitively more plausible than the premises of these arguments and
show that these propositions entail compatibilism, or else he will have to devise arguments for the falsity of
some of the premises employed in the present chapter, arguments that can be evaluated and seen to be
sound independently of the question whether free will and determinism are compatible” (104).
  . 

Nozick, R. (1981), Philosophical Explanations (Harvard University Press).

Spencer, J. (2013), ‘Able to do the Impossible’, forthcoming in Mind.
Tognazzini, N. A., and Fischer, J. M. (2013), ‘Incompatibilism and the Fixity of the Past’, this
volume.
Unger, P. (1984), Philosophical Relativity (Oxford University Press).
Van Inwagen, P. (1983), An Essay on Free Will (Oxford University Press).
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
Vihvelin, K. (1995), ‘What Time Travelers Cannot Do’, in Philosophical Studies, Vol. 81, No. 2:
315–30.
Williamson, T. (2002), Knowledge and Its Limits (Oxford University Press).
PART III
God
 8
Defenseless
A Critique of Van Inwagen’s Response
to the Argument from Evil

Louise Antony

Many philosophers, including myself, think that the facts about suffering in our
world constitute a strong case against theism—the view that there exists a unique
eternal, omnipotent, omniscient, and perfectly benevolent being. Philosophical trad-
ition distinguishes two ways in which this case might be developed: there is the
“logical” argument from evil, and the “evidential” argument from evil. The logical
argument aims to show that theism cannot be true, given the facts about suffering,
while the evidential argument aims to show that theism is very unlikely to be true, given
the facts about suffering. There is a fairly wide consensus among today’s atheist
philosophers that the logical argument is unsuccessful,1 and so the evidential argument
has become the more popular strategy. Peter van Inwagen, however, doesn’t care
which argument the atheist deploys. He believes that he has developed a “defense” of
theism powerful enough to work against any appeal to suffering whatsoever. I dispute
this. I intend to show that van Inwagen’s “defense” is no defense at all.

8.1 Preliminaries—What is a Defense?

I will shortly offer a brief summary of van Inwagen’s approach to the argument from
evil. But before we proceed, let me make a few terminological stipulations, to aid
exposition. First, because “defense” is a technical term for van Inwagen,2 I’ll capitalize
the word to indicate when I have his technical sense in mind: “Defense.” Second,
I will follow van Inwagen’s convention of discussing his Defense in terms of two
hypothetical interlocutors: a proponent of the Defense, a man called “Theist”, and an
atheist critic of the Defense, a woman called “Atheist.”3 Third, I will occasionally use

1
But for an important exception, see Lewis (2006).
2
Van Inwagen, in his (1988), credits Plantinga with introducing the term, but says that the character-
ization of a “Defense” is van Inwagen’s own.
3
I will not follow van Inwagen in pretending that Theist and Atheist are ideal proponents of their
respective positions; I am not confident that I know the best moves for Theist to make, and Atheist is pretty
much going to be, for better or worse, the mouthpiece of the very non-ideal me.
  

the term “Suffering” as shorthand for the complex term “all of the suffering, with
respect to both quality and quantity, that we have reason to believe is present in the
actual world.” Finally, I will continue to speak of the “argument from evil,” even
though the issue is really the existence of suffering, whether or not that suffering is
caused by the malicious actions of some agent.
Van Inwagen’s treatment of the argument from evil is novel. Many theists attempt
to address the argument by constructing a theodicy: a positive account of God’s
purposes and actions that explains how an omnipotent and morally perfect being
could permit the amount and the types of suffering that obtain in our world. As van
Inwagen puts it, a theodicy is
. . . an attempt to state the real truth of the matter, or a large and significant part of it, about
why a just God allows evil to exist, evil that is, at least apparently, not distributed according to
desert. A theodicy is not simply an attempt to meet the charge that God’s ways are unjust: it is
an attempt to exhibit the justice of his ways. (Van Inwagen 2006: 6)
Van Inwagen says, however, that he is not in a position to offer a theodicy. Although
he has a story to tell that he thinks would, if true, account for the facts about suffering
in the world, he cannot show that this story is true, or even probably true. He does,
however, believe that his story can serve as what he calls a “Defense.” A Defense, in
van Inwagen’s technical sense, is
. . . a story that contains both God and all the evils that actually exist, a story that is put forward
not as true but as “true for all one knows”. (Van Inwagen 2006: xii)

The crucial strategic difference between offering a theodicy and offering a Defense is
that the apologist must argue for a theodicy—he must provide some reason for
thinking that the story he tells is true—whereas with a Defense, all the apologist need
do is expound the story. Van Inwagen contends that if Theist can advance a Defense,
then he has done all that is required to show that the argument from evil is a failure.4
But is this true? It seems clear that the provision of a Defense would constitute
an adequate response to what is called, traditionally, the logical argument from evil.
This argument purports to show that it’s literally impossible for there to be suffering
in a world that contains a being who is both morally good and able to do anything.
If someone could tell a story that explains how possibly a benevolent and all-powerful
being could tolerate suffering, in the amounts and kinds we see around us, that
would suffice to refute this argument. This story wouldn’t have to be plausible—
merely coherent.
The so-called evidential argument, however, is another matter. Here it seems we
really need probability. The evidential argument says that Suffering constitutes
strong evidence against theism, or at least that Suffering makes it much less likely
that theism is true than that it is false. What difference would a Defense make to this
argument? If we had reason to think the Defense was true, then we would have an

4
“Failure,” like “defense,” is a technical term for van Inwagen: an argument is a “failure” if it would not
“win assent from the members of a neutral audience who have listened to an ideal presentation of the
argument” (Van Inwagen 2006: xi). If an argument is a “failure” in this sense, van Inwagen contends, then
no one can be rationally obliged to accept it, and no one can be rationally faulted for rejecting it.
  

undermining defeater for the evidence we thought was provided by Suffering. But of
course we’ve not been given any reason to think the Defense is true—it is precisely
because van Inwagen cannot offer us this that his story is a Defense and not a
theodicy. But in that case, the Defense does not constitute a defeater, and the
argument stands.
Or so it seems. Van Inwagen thinks the reasoning I’ve just rehearsed is incorrect.
He thinks a Defense can be a kind of undermining defeater even in the absence of any
reason to think the Defense is true. Why? Because—and this is my gloss—a Defense
shows us how something that we thought made theism improbable (i.e., Suffering),
could, for all we know, be exactly what is to be expected in a theistic world. If we
cannot rule out our being in just such a theistic world—if such a world is ours, for all
we know—then we really do not know what Suffering signifies. In the face of a
Defense, therefore, we should become “modest.” We should admit that we simply
do not know the probabilities on which the evidential argument depends. We do not
know how likely it is that theism is true, given Suffering, and we do not know how
much more or less likely it is that atheism is true, given Suffering. The evidential
argument cannot get off the ground.
What should we make of this move of van Inwagen’s? On the one hand, the
considerations I raised earlier seem compelling: if we have no reason to think that a
Defense is true, then we have no reason to alter our view of the evidence, or to revise
our estimate of the likelihood that theism is true, given than evidence. Whenever we
engage in empirical inquiry, we must acknowledge, going in, that there is a possibility
that our conclusion is wrong, despite the evidence being as it is. And what is a
Defense anyway, other than a vividly drawn possibility? A Defense, if true, would
constitute an undermining defeater. But the fact that a story would constitute an
undermining defeater if it were true doesn’t mean that the story actually undermines
the evidence. The fact that evidence is always defeasible surely doesn’t entail that it
is always (or ever) defeated.5
On the other hand, the examples that van Inwagen marshals on behalf of his
strategy are also compelling. They show that Defense is a common argumentative
strategy, deployed in a wide variety of contexts. And in the cases van Inwagen cites—
as well as in others I was prompted to think of myself—the strategy seems both
legitimate and effective. So in the cases where Defense works, how does it work?
Here’s what I think. A Defense does not directly target a probability judgment;
rather, it targets the reasoning behind the judgment. The cases in which a Defense is
apt are cases in which there is, or appears to be, some compelling reason to doubt that
p is true, given that q is true. This reason creates what I am going to call a mystery;
given that reason, it is a mystery how p could be true, given that q is true. A successful
Defense removes the reason by dispelling the mystery—and only by dispelling
the mystery. If the story presented in the Defense is not tailored specifically to the
particular reasons in play, to the particular factors that generate the mystery, the
Defense will fail.

5
See the very illuminating back-and-forth between Paul Draper and van Inwagen on this point in
Howard-Snyder (1996): Draper (1996a), Van Inwagen (1991), Draper (1996b), and van Inwagen (1996).
  

Van Inwagen says that it is a condition of adequacy on a putative Defense that it is

“true for all we know.” Van Inwagen does not tell us exactly what he takes this to
mean, but from the fact that he thinks his Defense satisfies the condition, I presume
that it is pretty minimal. Presumably, the condition rules out only stories that involve
claims that are inconsistent with beliefs we have that we have good reason to think
are true. Maybe it also rules out propositions that are obviously entailed by things
that we have good reason to think are false.6 My contention is that this condition is
too weak to pick out the cases in which a Defense is, intuitively speaking, successful.
I want, therefore to add an additional constraint: what I’ll call the “No-Mystery”
Constraint. If you like, you can think of this constraint as a way of specifying the “for-
all-we-know” constraint that van Inwagen leaves vague: it focuses on the aspect of the
situation that, given “all we know”, is mysterious.
To evaluate van Inwagen’s Defense, then, we will need to know what mystery is
driving the argument from evil—we need to know, in other words, why Atheist
thinks that the probability of theism, given Suffering, is low. So here’s the plan. I’ll
begin by discussing some successful deployments of the Defense strategy, in order to
refine and support my analysis of the way Defense works, when it does work. I’ll then
give a rational reconstruction of Atheist’s reasoning about the relationship between
Suffering and theism. I’ll assume that Atheist begins with the logical argument. We’ll
see that there are considerations that force Atheist to abandon the logical argument
in favor of the evidential argument from evil. These considerations, it turns out, will
help articulate the mystery—or mysteries—that motivate the evidential argument,
and that warrant Atheist’s skepticism that a proper Defense can be constructed.
Finally, I’ll show why van Inwagen’s Defense fails to dispel these mysteries. I’ll argue
that the story he provides does not address the specific reasons why Atheist thinks
that Suffering makes theism improbable. It tells us nothing more than what Atheist
already conceded in giving up on the logical argument, namely that it is possible for
there to be a morally good, omnipotent being in a world that contains Suffering. Van
Inwagen’s Defense, I’ll conclude, is a failure (in the non-technical sense).

8.2 Successful Defense—Some Examples

One of the most familiar deployments of the strategy of Defense, van Inwagen points
out, is to be found in the public debate about evolution. When Atheist appeals to the
existence of Suffering to argue against theism, van Inwagen argues, she is doing much
the same thing that a fundamentalist (“Fundamentalist”) is doing when he appeals to
the appearance of design in nature to argue against the theory of evolution. Funda-
mentalist is making an evidential argument: he says that the orderliness and

6
One important issue is whether “all we know” is ever sufficient to support negative existential claims,
like “there are no centaurs.” I don’t know if the nonexistence of centaurs is deducible from contemporary
biological theory, but contemporary biological theory does, it seems to me, make it highly unlikely that a
creature with such an aberrant body-plan as a centaur’s could have emerged from any of the known
mammalian lineages. It seems similarly improbable, given contemporary science, that there could be a
period of human habitation in which our ancestors had the capacity to anticipate natural evils. Indeed, the
whole idea of “preternatural powers” seems, on its face, to be incompatible—in some sense—with the
import of natural science. But I will not press this point.
  

complexity that we observe in the world of nature (“Order” for short) is much more
likely to have been produced by an intelligent designer than to have been the product
of impersonal forces acting blindly on the phenotypic effects of random mutations.
Order is just not what we would predict if things happened the way the adaptationists
say; it is very hard to see, for example, how something as complex as the human eye
could have come into being if there was no designer.
Fundamentalist needn’t—and doesn’t—claim that it is literally impossible for the
eye (or some other equally complex structure) to have developed through natural
selection in order to make his case. It is enough for his purposes to make vivid how
highly improbable it is that this is so, relative to his own favored hypothesis. In
response to this sort of challenge, van Inwagen says, it is perfectly appropriate for the
defender of evolution to respond by offering a Defense, like so:
Professor Hawkins, an apologist for the Darwinian theory of evolution, tells a story according
to which the human eye . . . did come about as a result of the combined operation of [random
mutation and natural selection]. . . . Hawkins does not present her story as an account of the
actual course of evolution, and she does not take it to constitute a proof that the human eye is a
product of the interplay of random mutation and natural selection. Her story is intended
simply to refute an argument for the falsity of the Darwinian theory of evolution . . . . (Van
Inwagen 2006: 7)
Prof. Hawkins doesn’t claim that her story is actually, or even probably true. She
claims only that the story does not require us to deny anything that we believe to be a
fact, that it is “true for all we know.” But since Hawkins’s story, if true, would explain
Order as well as would the existence of an intelligent designer, Fundamentalist ought
to retract his claim that Order counts differentially against the theory of evolution,
and in favor of the thesis of intelligent design.
But now let us ask: what makes Hawkins’s Defense successful? Van Inwagen
doesn’t actually give us the content of the story Hawkins tells, but presumably, it
contains enough detail to address Fundamentalist’s particular reasons for thinking
that Order would be surprising if the theory of natural selection were true.
Fundamentalist, we may assume, understands the set of explanatory tools the
theory of natural selection provides. He understands how genetic variation within a
population can produce phenotypic variation, how different phenotypes can lead to
different rates of survival and reproductive success, and thus how the genes that
produce those phenotypes can become fixed in the population. So he understands, in
outline, anyway, what would have to be the case for the theory of natural selection to
be true. And he sees that there is a possible series of events that could have produced
the Order he sees around him in the way that the evolutionist says things happened.
But Fundamentalist might think that such a series of events would require a vast
amount of time—more time, surely, than the amount of time the planet has been in
existence (even substituting the evolutionist’s generous estimate of four billion years
for the Fundamentalist’s pious estimate of seven thousand or so).
If this is the nature of Fundamentalist’s concern, then it might be that all Hawkins
has to do to mount a successful Defense is to make vivid certain matters of scale: how
vastly many generations evolution had to work with, for example. Or, on the other
hand, Hawkins might show Fundamentalist how surprisingly quickly evolution can
  

occur. (Hawkins might use, for either of these purposes, a variety of computer
simulation games.)
But Fundamentalist might have in mind a more principled objection to the evolu-
tionist’s story; he might think that there are phenomena that can’t be explained by the
mechanisms posited by the theory of natural selection, even allowing for timescale.
Consider the human eye, the parts of which are functionally interconnected—the lens,
for example, is useless without the retina, but the complex structure of the retina’s
photosensitive surface would be useless without the focused image produced by the
lens. Since neither part does the organism any good without the other, it is hard to see
what series of gradual accumulations of adaptive alterations could have produced
the eye. If having a lens without a retina does an organism no good, and similarly for a
retina without a lens, how could either of those phenotypic properties have persisted
in the population long enough for chance to bring them together? Another oft-cited
case: wings. Wings in birds and mammals presumably evolved from forelegs; but the
deformations of the forelegs that constituted proto-wings would have been no good
for flying and probably not very good for walking. What pattern of selective pressure,
then, could have sustained the process of greater and greater deformation until finally
there emerged a creature who can actually fly?
Hawkins cannot respond to a challenge like this by simply positing a set of
(unspecified) selective pressures that are stipulated to have the appropriate results.
Fundamentalist is not making a “logical argument” against evolutionary theory: he is
not asserting that there could not be any selectional history that explains the eye or
the wing; he is saying only that the resources of the theory of evolution (as he
understands them) leave it a mystery what such a history would look like. For
Hawkins to make an apposite reply to this reasoning, to produce a successful
Defense, Hawkins is going to have to dispel the mystery. She is going to have to
display a selectional history that would, if it were true, account for the evolution of
structures like the eye or the wing.
That is, of course, just what many real-life Darwinians do. Dan-Erik Nilsson, for
example, has developed a gradualist adaptationist model that demonstrates how the
human eye could have developed from a single photo-sensitive spot on the skin of an
ancestor, by means of individually adaptive steps, within half a million years.7
Another Defensive strategy contemporary Darwinians have used is to reveal and
correct some mistaken assumption on which Fundamentalist’s challenge depends.
Darwin himself did this in response to the challenge to show what use a “half a wing”
might be: the mistaken assumption in this case being that there must be continuity of
function throughout the phylogenetic history of the structure in question.8 Funda-
mentalist, confronted by such stories, cannot say any longer that the existence of eyes
and wings is better evidence for intelligent design than for natural selection, because

7
Nilsson explains his model on a Public Broadcasting System program, Evolution: “Darwin’s Dangerous
Idea” (2001 WGBH Educational Foundation and Clear Blue Sky Productions, Inc.). There’s a clip available
here: http://www.pbs.org/wgbh/evolution/library/01/1/l_011_01.html.
8
Stephen Jay Gould discusses this challenge and presents several replies, including ones that in van
Inwagen’s terminology would count as Defenses. Some of these get transformed into “theodicies” through
reference to actual evidence. See http://www.stephenjaygould.org/library/gould_functionalshift.html.
  

his reasons for doubting the adequacy of natural selection in explaining Order have
been removed.
Another feature of a successful Defense, I submit, is that it is intelligible to the
challenger to whom it is offered: it does not trade mystery for mystery. Suppose that
one of those sticking points for Fundamentalist is the idea of genetic inheritance
itself; he just doesn’t see how brute causal interactions could support a transfer of
information from one generation to the next. If Hawkins wants to satisfy Funda-
mentalist, she will have to tell a story about the mechanism of genetic inheritance that
shows how inherently meaningless causal relations could track informational rela-
tions. But if Hawkins’s story appeals to hypothetical processes that are just as
mysterious to Fundamentalist as the process he doesn’t understand in the first
place, the Defense will fail. The point is not that Fundamentalist is stupid; I am not
presuming that. The requirement I have in mind is not that a successful Defense must
de facto persuade every person who demands one. Rather the requirement is that the
Defense must provide a response that does in fact address a challenger’s sticking
point, not that the challenger must recognize that it does. If the challenger cannot
appreciate the force of the Defense because it involves arcane theory or complex
technicalities, that is nothing against the Defense. This is very different from a
Defense’s relying on a mystery.
To bring home the point, let me offer another example—in this case, an actual
Defense—that illustrates how the intelligibility demand can be satisfied. I have in
mind a Defense of materialism against a dualist’s challenge. Descartes, as is well
known, made an a priori argument for mind–body dualism in the Meditations. But
he also made a different argument—an a posteriori argument—in the Discourse.
Briefly, the argument goes like this:
Human beings have the capacity to respond in an intelligent way to an unlimited variety of
circumstances—call this capacity “reason.” The infinitary character of reason is manifest in
human beings’ use of language—in the ability they have to combine and recombine words so as
to produce an unbounded number of distinct, meaningful utterances. Material “organs” can
only determine a fixed set of responses to a limited set of circumstances. Matter cannot be
divided infinitely; therefore no material being can possess an infinite number of material
organs. Therefore, this capacity that human beings have cannot be grounded in a material
organ; it must inhere in an immaterial substance.9
Turing, however, showed both that, and how, mere matter could support the kind of
capacity Descartes described. He offered, in effect, a Defense of materialism. Of
course, Turing’s work does not settle the mind–body question. It is clear that the
human brain does not have the architecture of a Turing machine, and so Turing’s way
cannot be the way that our minds are embodied, if in fact they are. But Turing’s work
does offer the materialist a Defense against the Cartesian charge that rationality is
more plausibly immaterial than material. It dispels the mystery about how cognitive
capacities could be embodied, by addressing Descartes’ questions how finite resources
could ground an infinitary capacity, and how the state transitions of a mechanical

9
My paraphrase. See René Descartes, Discourse on Method, Part V.
  

device could track rational relations. In light of Turing’s work, the evidential signifi-
cance of the fact that human beings use language has been altered.10
I find these real-life Defenses compelling (and not only, I hope, because I happen
to agree with the Defenders). But I think their success turns on a feature that is not
ensured by van Inwagen’s requirement that a Defense be “true for all we know.”
What I think makes these Defenses successful is that they address, and remove, the
particular sticking points advertised by the challenger. They provide substantive,
constructive answers to specific “how possibly” questions. I submit that it is only by
doing this that a Defense removes the challenger’s reason for assessing the probabil-
ities in the way that he or she does. Neither the Darwinian Defense nor the Turing
Defense would be compelling if it simply posited the existence of forces or processes
stipulated to possess the necessary properties, whatever those might be. To do that
would be, as I said above, to trade one mystery for another. So let me propose, then,
the following condition on an adequate Defense:
The No-Mystery Constraint: relative to the challenge at hand, an adequate
Defense must make no appeal to mysteries—it must not simply “pass the buck”,
trading in one mystery (or problem) for another.
With this constraint in mind, let us turn to the development of Atheist’s position on
the significance of Suffering.

8.3 From Logical to Evidential/From Apology

to Defense
The “Logical Argument” from evil purports to show that the existence of Suffering is
incompatible with the existence of a morally good, omnipotent being. Here is an oft-
quoted version of the argument, from Hume’s Dialogues Concerning Natural Religion:
Is he able, but not willing? Then he is malevolent. Is he willing, but not able? Then he is
impotent. Is he both able and willing? Whence then, is evil?
Atheist might find this argument persuasive. But Theist, if he is on his toes, will point
out that the formulation above is enthymematic. Here is a reformed version:
The Logical Argument from Evil
1. Some suffering exists.
2. No Tolerance: No morally good being would fail to prevent suffering if he or
she were able to prevent it.
3. No omnipotent being would be unable to prevent suffering.
4. Therefore, there is no being who is both morally good and omnipotent.
This version is valid, but there is clearly a problem with the second premise, the
principle that I am calling “No Tolerance.” The problem, as Theist will be quick to
point out, is that No Tolerance is false; everyday life suggests counterexamples.

10
Of course there are many other challenges to materialism that Turing’s work does not touch.
  

Consider a loving parent (hereafter “Parent”) who disciplines his child (“Child”)
by allowing the youngster to suffer (what are called in the parenting literature) the
“natural consequences” of her actions. Child, let us suppose, has ignored Parent’s
warning not to leave a certain favorite toy outside in the rain. Predictably, when
Child finds the toy the next day, it is rusted beyond repair. Parent, we may also
suppose, had seen the toy lying on the ground and could have rescued it before the
rain started. But if Parent had done this, then Child would not have taken to heart the
lesson that negligence can lead to loss. Child would have been left ignorant of an
important structural feature of her world—that if she wants to preserve that which
she values, she must be attentive. This lesson is so valuable that even a loving
parent—especially a loving parent—will allow his child to endure a brief period of
unhappiness in order that the lesson be imparted. Hence, No Tolerance is false, and
the logical argument fails.
Atheist, however, will object that Theist has dismissed the logical argument too
quickly. All that is shown by the case of Parent and Child is that the crucial principle
has been misformulated. No Tolerance is too simplistic. If we think just about the
human realm for a moment, we see the principle as stated fails to take account of a
couple of things: (1) the reasons a being may have for permitting suffering, and (2)
the system of constraints within which a being must act. In assessing Parent’s moral
responsibility, we must recognize first that Parent is acting on the basis of a morally
laudable goal, viz., sparing his Child suffering in the future. Second, we must note
that Parent is constrained to pursue this morally laudable goal against a background
that Parent did not choose and could not alter. Thus, the small amount of suffering
Parent was able to prevent is not all the suffering that Parent had to take into account.
Parent chooses the small, local suffering entailed by the loss of the toy over the
greater, future suffering to which Child is vulnerable if she doesn’t learn to take care
of her things. Parent cannot evade this choice; he has no control over the physical and
psychological contingencies that structure it. He cannot alter the conditions under
which material things decay, nor can he alter the laws of human psychology that
make harsh experience not only the best but often the only possible teacher. Thus,
while Parent can prevent one particular bit of suffering, he cannot prevent that bit of
suffering without exposing the child to greater suffering in the future. Parent is not
able to ensure that Child never suffers.
God, on the other hand, can alter the contingencies and the laws. (This is still
Atheist speaking.) He can suspend the laws of nature if he likes—indeed, it is up to
God what the laws of nature are. He could have designed a world where metal toys
did not rust when left out in the rain, or a world where children simply listened to
and abided by their parents’ advice. The human parent can shift the blame; God
cannot. Thus, while it may be true for limited beings that they have to permit some
suffering to occur in order to achieve goods of much greater value, omnipotent beings
cannot justify the toleration of suffering on the grounds that they have no choice.
(Atheist is finished now.)
Atheist’s point is well taken; an adequate formulation of the logical argument will
take account of the fact that a person does not necessarily do wrong in choosing not
to prevent a case of suffering that that person has the ability to prevent. So we need to
replace No Tolerance with a subtler principle, maybe something like this:
  

No Tolerance Unless: No morally good being would permit suffering to occur

unless either (a) permitting such suffering is necessary in order to realize some
greater moral good, or (b) preventing such suffering would entail even greater
suffering, or else the loss of some greater moral good.
This modified principle offers us two new parameters that enable us to take account
of the morally relevant context—one, the background against which the agent must
decide what to do, and two, the objectives of the agent. This emendation in the
principle gives us an appropriately adjustable bar—it allows, as it should, the
exculpation of Parent and other mortal beings in consideration of the difficult
trade-offs they are constantly faced with, but without giving much more powerful
beings a free pass.
Or maybe not. . . .
Theist is going to point out that once we’ve weakened Premise 2 in the original
logical argument, we will have to make a coordinate adjustment to Premise 3, with a
result something like this:
The Weakened “Logical” Argument from Evil
1. Suffering exists.
2. No Tolerance Unless: No morally good being would permit suffering to occur
unless either (a) permitting such suffering is necessary in order to realize some
greater moral good, or (b) preventing such suffering would entail even greater
suffering, or else the loss of some greater moral good.
3. There can be no surpassingly important moral goal such that not even an
omnipotent being could achieve it without permitting Suffering.
4. Therefore, there can be no being who is both morally good and omnipotent.
This is no good for Atheist. While the substitution of the more plausible “No
Tolerance Unless” for the crude “No Tolerance,” does protect Atheist’s argument
from mundane counterexamples, it also opens up a loophole. Premise 3 is needed if
the argument is to yield the conclusion that a morally good, omnipotent being is
impossible. The trouble is that Premise 3 gives Theist the design specs for a Defense.
What is needed, first of all, is a moral goal so compelling that it would warrant the
toleration of an enormous amount of suffering—as much as we observe in the world
around us.11 Second, that goal would need to be placed within a certain modal
environment, one that involves some form of necessity that structures God’s choices
in a way that is analogous to the way the harsh facts of the natural world structure
Parent’s choice. Since God is omnipotent, this form of necessity cannot be merely
nomological; it will have to be logical necessity, or something like it.
Now if Theist has a Defense on hand—a story that would satisfy all of the
conditions given above, a story that, if true, would “exhibit the justice of [God’s]

11
A more carefully formulated version of this principle would make it explicit that there are issues of
justice at stake as well as amounts and qualities of suffering. That is, the correct excusing condition will not
be a pure value-maximizing condition. (See Nozick on “side constraints” in Nozick (2013).) But I don’t
know how to formulate the correct principle precisely.
  

ways” within a world that contains Suffering—that would be quite compelling. But it
would also be overkill. Theist does not need to actually describe a goal that meets
these conditions in order to oppose the logical argument; Theist needs merely to
point out the possibility that such a goal exists. Atheist really cannot deny this:
Premise 3 is neither a logical nor (if this is different) a conceptual truth.
But now, what if Atheist changes tack? What if she tries to throw Theist on the
defensive? What if she gives up her claim that a world that contains Suffering could
not possibly have been created by a morally perfect, omnipotent being, and retreats
instead to the claim that it is very unlikely that such a being would create such a
world? She might try weakening the Logical Argument still further into an Evidential
Argument, like this:

An Evidential Argument from Evil

1. Suffering exists.
2. No Tolerance Unless: No morally good being would permit suffering to occur
unless either (a) permitting such suffering is necessary in order to realize some
greater moral good, or (b) preventing such suffering would entail even greater
suffering, or else the loss of some greater moral good.
3. Probably, there is no surpassingly important moral goal such that not even an
omnipotent being could achieve it without permitting Suffering.
4. Therefore, probably, there is no being who is both morally good and omnipotent.

In defense of the new Premise (3), Atheist can explain why she thinks that the
excusing conditions offered by No Tolerance Unless are going to be very difficult for
an omnipotent being to satisfy. It is hard to see, in the first place, how any moral goal
could be so compelling that it could warrant the toleration of such an enormous
amount of suffering as we observe in the world around us. Furthermore, even if such
a goal could be identified, the possibilities for achieving the goal would have to be
placed within a modal environment that (a) requires the toleration of suffering, and
(b) limits the possible means for its achievement. What kind of necessity, what kind
of limits can these be, that bind even an omnipotent being—a being who can arrange
even nature to suit its purposes? Although (as Atheist will now admit) there is
nothing to absolutely rule out the possibility of there being such a goal, or the
possibility of there being a suitable form of constraint, it is a mystery what such
things would look like.
Now Theist really needs a Defense. It is not enough for him to assert the possibility of
there being a way that these conditions could be met, he needs to counter Atheist’s
principled doubt about there being such a way. Methodologically speaking, Atheist is
now prepared to be instructed, to learn of a new kind of goal, or a new form of
constraint that she had not earlier considered. Learning about such things would reduce
her confidence in Premise 3—but she would really need to know what the goal is, and
how it necessitates the toleration of suffering. But so far, these things are mysterious.
Atheist, in other words, is prepared to accept the probity of a Defense—provided it
is a Defense that dispels the mysteries. It is instructive here to look at the way
Defenses can succeed in cases when there are moral issues at stake. So let us return
  

to my case of Parent disciplining Child. Suppose that you come upon the scene the
morning after the rainstorm. You see the child picking up the rusted remains of his
toy, and bursting into tears. Moved by Child’s obvious distress, you ask Parent what
happened—did he, Parent, realize that the poor child’s toy had been left out? Now
suppose that Parent answers, “Oh yes, I saw it out there when it started to rain, and
I could have brought it in, but I didn’t.” Without any more information, this reply
might well make you think that Parent is uncaring or irresponsible—or maybe even a
little sadistic. You can’t imagine why a good parent would have allowed the ruin of his
child’s prized possession. But then the following occurs to you: what if Parent had
earlier warned Child, repeatedly and clearly, not to leave the toy in the yard? And
what if Child ignored the warnings and left the toy exactly where Parent had said not
to leave it? In such circumstances, mightn’t even a good parent—especially a good
parent—decide to leave the toy out in order to teach the child a lesson? You don’t
know that this is what happened, of course, but once you realize that there is at least
one scenario on which a good parent might do what this parent did, the fact that
Child is suffering now seems altogether less probative. It now seems to say nothing,
in itself, about how good a parent Parent is.
But now suppose that you do a little more investigation. Child testifies, credibly,
that Parent never warned her about the consequences of leaving her toy out in the
rain, and that she, Child, didn’t know what those consequences would be. Maybe
Parent, himself, admits the truth of all this. Suppose, further, that you learn that there
was, in fact, no rainstorm. Parent, it turns out, deliberately activated the sprinkler
system and allowed it to run throughout the night. Discoveries of this sort continue,
and the more you learn, the more unsettled you become. It emerges that there have
been a host of incidents throughout Child’s short life in which Parent did things that
good parents never do. Once, Child was beaten severely for setting the table incor-
rectly. Another time, Parent arranged for a neighbor to offer Child a type of luscious
candy that Child had been specifically forbidden to eat; when Child (predictably)
succumbed to temptation, Parent punished her by locking her out of the house,
forcing her to sleep in a dangerous back alley where she was vulnerable to vermin and
desperate drug addicts. In light of all this new evidence, you conclude that you were
right in the first place: Parent is a very bad parent.
At this point, it would hardly be appropriate for someone who wanted to defend
Parent simply to point out that there is surely some possible reason a good parent
might have had for doing all the things that Parent has done, and that perhaps Parent
had no choice but to do the things he did. You don’t deny that it is possible that such a
reason exists—you cannot, at any rate, prove that it doesn’t. It is, however, a mystery
what such a reason could be. The details of the case raise particular sticking points,
such as: how could an etiquette mistake warrant a beating? How could it be that
Parent had no option but to expose Child to mortal danger? And since the danger is
mortal, what greater good could the punishment serve? A successful Defense would
have to dispel these mysteries. A successful Defender would have to actually produce
a reason that you could understand to justify the actions the parent performed. The
Defender has to tell a story that makes Parent’s actions morally intelligible. The story
doesn’t have to seem probable to you—but it does have to be a story that would make
you say: “well, for all I know, that’s true, and if that were the explanation for what
  

Parent has done, then I’d have to take it back—in that case, he wouldn’t be a bad
parent at all.” But if you are not given a story like that, you are going to stick to your
guns. You are still going to say that the evidence is quite sufficient to support a charge
of child abuse.12
Van Inwagen’s Defense of theism against the evidential argument from evil has the
same structure as the successful Defense of Parent in the first part of my story. Van
Inwagen tells us a story according to which God has a goal of great moral importance,
and it is a goal that can only be achieved if he tolerates Suffering. The substance of
van Inwagen’s story is also roughly the same as that of the story I tell: there is some
important lesson that God wants us to learn, for our own good, and the only
pedagogically effective means of imparting the lesson are ones that require his
tolerating Suffering.
I’ll argue, however, that when we examine van Inwagen’s story as a response to
Atheist’s reasons for judging the probability of Suffering, given theism, to be low, we
see that it does not constitute a Defense. Like the inadequate Defender in the second
part of my story, van Inwagen relies on mysteries. His story involves claims about
moral value and modal structure that are fundamentally stipulative, claims that do
nothing to address Atheist’s particular sticking points. He therefore gives her no
reason to reassess the significance of Suffering.

8.4 Atheist’s Sticking Points—What a Defense of

Theism Must Do
The logical argument from evil fails, and, I argued, we can see from the way that it
fails what has to be done by any theist who wishes to mount a Defense. Recall the
crucial premise in the (Weakened) Logical Argument:
No Tolerance Unless: No morally good being would permit suffering to occur
unless either (a) permitting such suffering is necessary in order to realize some
greater moral good, or (b) preventing such suffering would entail even greater
suffering, or else the loss of some greater moral good.
The atheist who propounds this logical argument may be supposing that God’s
omnipotence analytically guarantees that God’s choices could never be constrained
in the way that (a) or (b) specifies. So the first thing Theist must do is demonstrate
that this is not the case. There is a natural place for Theist to start—with the
reasonable and widely accepted principle that omnipotence entails only the power
to do what it is logically possible to do. This opens up some options.
The first option involves free will. Arguably, it is logically impossible for anyone
both to allow a person to choose freely and at the same time to ensure that that
person chooses one way rather than another. God’s “inability” to control the choices
of free agents would then constitute no restriction on his omnipotence, any more
than would God’s inability to make it be both raining and not raining in the same

12
I argue elsewhere that the information provided in the Bible about God’s treatment of his human
creatures amply supports this indictment. See Antony (2013).
  

place and at the same time. All that would then be needed to defeat the logical
argument is the possibility that free will is so great a moral good that its presence
warrants the toleration of whatever suffering might result from its exercise.
The second way out of the logical argument involves the gap between the formal
consistency of a sentence and the metaphysical possibility of the proposition that that
sentence expresses.13 This gap gives rise to what I think of as “modal illusions”—
things that appear to be possible but are not. Mathematical statements provide a good
example. Mathematical propositions are not contingent: if true, they are necessarily
true, and if false, they are necessarily false. But sentences expressing mathematical
statements needn’t be formally valid or formally inconsistent. (That is one reason
why we still don’t know if Goldbach’s Conjecture is true.) So-called “Frege cases”
provide another kind of example. The sentence “Hesperus has more nitrogen in its
atmosphere than Phosphorus does” does not have the form of a contradiction, and
might even be assented to by a rational person who simply doesn’t know that
“Hesperus” and “Phosphorus” both refer to the same planet.
There can be similar situations with sets of propositions: the set may harbor a
contradiction without our realizing it, and certainly without there being any evident
inconsistency among the sentences we use to express the propositions in question.
(Just ask Gottlob Frege!) I see, for example, no contradiction in supposing both that
(1) there exist creatures with free will and rationality and (2) no sentient creature
suffers. Yet, for all I know, I am mistaken about this. I may have laid down two
constraints that cannot, as a matter of metaphysical fact, be jointly satisfied.
The relevance of all this to the logical argument is probably clear: if it is true in
general that we have no way of knowing that a situation that seems possible actually
is possible, then we cannot be completely confident about the modal relations among
types of phenomena in the real world. It may appear that it is perfectly possible for
there to be flying pigs, but we cannot be confident that that appearance will persist
once we have a complete understanding of what it is to be a pig and what it is to fly.
Similarly, while it might appear to us that some instance or type of suffering is utterly
gratuitous, it might be in fact that there is a modal connection between that type of
suffering and some other vital feature of the world.
The logical argument from evil fails, then, because of two possibilities left open: the
normative possibility that there is a moral good that outweighs the moral disvalue of
the suffering we observe to exist in this world, and the modal possibility that there are
metaphysical connections among aspects of our world that not even an omnipotent
being can sunder. But now let us consider the evidential argument. Atheist, qua
proponent of the evidential argument, does not deny that there might be some story
according to which an omnipotent being permissibly tolerates the kinds, amounts,
and distribution of suffering that we observe in our world; she just thinks that it is a
mystery what such a story could be. And here is where a Defense would change the
picture. The provision of the sort of story that the atheist cannot imagine would make

13
I am happy to call this the difference between conceivability and possibility, but others are not, so I’ll
try not to use those terms.
  

the facts about suffering less telling, and might even justify the sort of agnosticism
van Inwagen calls for.
But now I want to bring to bear the lessons from the earlier discussion. Defenses, to
be successful, must be crafted so as to address the specific concerns of the challenger.
The fundamentalist critic of evolution, recall, is not saying that there is literally no
possible way that natural selection could have produced the mammalian eye; rather
he is saying that he cannot imagine any way it could have done so without there
having been a coincidence of cosmic proportions. The Defender of natural selection,
then, must tell a story that precisely meets this challenge—a story that presents an
alternative to intelligent design that does not depend on sheer coincidence, a story
that shows how an unthinking mechanism could reliably mimic the effects of
intelligent design. And there must be enough detail in the story that Fundamentalist
is satisfied.
Analogously, the theist who wants to offer a Defense against the evidential
argument must craft one that addresses the specific concerns of the atheist challen-
ger. What Atheist cannot imagine, and what she wants to see described in detail, is
the specific combination of moral goal and modal structure that would make it
possible for a morally perfect and omnipotent being to satisfy the conditions set out
in the principle No Tolerance Unless. Atheist is baffled, because there is no
apparent way of generalizing from the kinds of circumstances that make No Toler-
ance Unless applicable in the human realm to the situation of a morally perfect,
omnipotent being. Most obviously: human beings are, and an omnipotent being is
not, constrained by natural law. The fact that we are constrained by natural law has
normative consequences—it is the main reason why we must engage in moral
trading-off, and pretty much the only justification we have for so doing. If we did
not have to choose between allowing the trolley to kill five, and diverting it to kill only
one, if we could miraculously stop the trolley in its tracks before anyone was hurt,
then it would be irrelevant to the moral evaluation of our actions that the death of five
people is, prima facie, worse than the death of just one. As I pointed out in my
example, Parent’s justification for permitting Child to suffer depends in many
different ways on there being aspects of the circumstances that the parent cannot
control. The parent’s mortality, the structure of human character development, and
the law of entropy all played a role.
This is not to say that all normative necessities depend on the existence of physical
limitations for the agents involved. In my example, I assumed that the parent’s
concern for the child’s character development was rooted in a concern for the child’s
overall well-being, but perhaps it could be argued that it is simply morally better for a
child to have the kind of character fostered by this sort of training, whether it protects
the child from future suffering or not—a virtue theorist might take this position. If
the deontologists are right, there are moral duties that are independent of their
consequences, and so independent of the nomological structure that determines
those consequences. It might be, for example, that I am morally required to respect
a person’s right to life, even if the consequences of my doing so are, intuitively, worse
than would be the consequences of my killing them. But virtue theorists and
deontologists will generally allow that there are limits. If the development of some
virtue in a certain child required that the child be flogged senseless every day for a
  

month, and if possession of the virtue would be of dubious value to the child, it seems
implausible that the positive moral value of the virtue would outweigh the disvalue of
the child’s suffering. For the deontologist, there is the case of the innocent toddler
about to press the button that will ignite a thermonuclear holocaust, who can only be
stopped by gunshot. So even if there are non-consequentialist moral goods, the
existence of conflicts among them generally reflects natural necessities.
In sum: to satisfy Atheist’s specific concerns, the author of a Defense must eschew
two kinds of mysteries: normative mysteries and modal mysteries. It will not be
enough for Theist to posit moral goals or modal dependencies that have the requisite
formal properties; such bare posits are pertinent only to the logical argument. An
adequate Defense must provide substantive information—it must be a “real possi-
bility”, and not just a merely logical possibility.
With all this in mind, let us turn, finally, to van Inwagen’s Defense. The Defense
has two parts. The first part, a version of the “free will defense,” is meant to explain
God’s toleration of the suffering that is endured by human beings. This lot of
suffering includes, but is not limited to, the suffering that results causally from the
free choices of human beings. Here is a summary of the first part of van Inwagen’s
Defense, as presented in van Inwagen (2006: 85–7). (I will quote liberally, especially
in sections where it is important to have van Inwagen’s precise wording.)
After guiding biological evolution for millions of years, God singled out a small
breeding group of hominids and “miraculously raised them to rationality,” giving
them “the gifts of language, abstract thought, and disinterested love.” He also gave
them the “gift” of free will, “because free will is necessary for love.”
At the same time, God took human beings “into a kind of mystical union with
himself.” During this period of time, the newly-human beings “lived together in the
harmony of perfect love.” Also during this time, the human beings possessed
“preternatural powers” which enabled them “somehow to protect themselves from
wild beasts . . . , from disease . . . , and from random, destructive natural events. . . . ”
Despite this idyllic existence, the human beings “in some way that must be mysterious
to us . . . were not content with this paradisal state. They abused the gift of free will and
separated themselves from their union with God.” “The result was horrific: . . . they now
faced destruction by the random forces of nature, and were subject to old age and
death.” (Van Inwagen notes that, in this state of separation, human beings “were, as an
engineer might say, ‘not operating under design conditions.’”)
Although it would have been “just of [God] to leave human beings in the ruin they
had made of themselves and their world,” God mercifully “set in motion a rescue
operation,” the object of which is “bringing it about that human beings once more
love God.” But love must be freely chosen. In order for human beings to choose freely
to love God, they “must know that they need to be rescued,” which means knowing
“what it means to be separated from God,” viz., “to live in a world of horrors.” That’s
why God has to permit the horrors this world contains. If God miraculously canceled
them, “he would thereby frustrate his own plan of reconciliation. If he did that, we
should be content with our lot, and should see no reason to cooperate with him.”
The second part of the Defense is meant to account for God’s toleration of whatever
suffering is left unexplained by the first part, viz., that portion of suffering by non-
rational but sentient creatures—“beasts”—that cannot be ascribed to the actions of
human beings. Such creatures lack many (or all) of the morally significant capacities
  

that are cited in the first part of the Defense: they are incapable of making morally
significant choices, and thus incapable of the kind of guilt that might warrant
painful correction or punishment. Furthermore, they could not benefit from whatever
pedagogical or spiritual value such correction or punishment might hold for beings
like ourselves. So some different kind of story must be told.14
Part Two of the Defense consists of four hypotheses: it is short enough for me to
quote it in full:

1. Every world God could have made that contains higher-level sentient creatures
either contains patterns of suffering morally equivalent to those of the actual
world, or else is massively irregular.
2. Some important intrinsic or extrinsic good depends on the existence of higher-
level sentient creatures; this good is of sufficient magnitude that it outweighs
the patterns of suffering found in the actual world.
3. Being massively irregular is a defect in a world, a defect at least as great as the
defect of containing patterns of suffering morally equivalent to those found in
the actual world.15
4. The world—the cosmos, the physical universe—has been created by God.

That’s the story. Let’s look at the mysteries.

Van Inwagen does not deny that his story contains mysteries; indeed he freely
admits it. For example, he tells us that it is a mystery—actually, a “miracle”—how
God imbued our ancestors with rationality. It is also a mystery, he admits, what the
“preternatural” powers might have been that enabled our ancestors to avoid all the
natural causes of suffering. I do not object to these (what we might call) “natural”
mysteries; they are mysterious only relative to what we know about the natural world.
If the issue had to do with the mode of divine creation—if one were challenging the
theist to say “how possibly” an immaterial being could have created the material
world, for example—these mysteries would be a problem. Since the focus of the
evidential argument is meant to be on the normative question of how a being with
God’s powers (which themselves are mysterious) could be warranted in permitting
suffering, these natural mysteries are not a problem.
But these are not the only mysteries in van Inwagen’s story. His story depends on
modal mysteries and normative mysteries, and these are precisely the kinds of
mysteries that an adequate Defense must eschew. Van Inwagen merely posits, and
does not display, answers to the central questions posed by the proponent of the
evidential argument: what kind of necessity is it that constrains even an omnipotent

14
It is not clear to me why van Inwagen thinks that the first part accounts for any of the suffering of
beasts—the justification for God’s tolerating human suffering is, in general terms, because they deserve it—
they deserve being in a condition where they are vulnerable to all sorts of suffering. God can’t interfere with
the consequences of human actions partly because of his goal of getting human beings to want to return to
him. But this doesn’t explain why it is OK to allow beasts to suffer. It might be that part one explains why it
is often human beings’ fault that beasts suffer; but that doesn’t mean that the beasts deserve to suffer. And if
they don’t deserve to suffer, it is—prima facie—unjust to allow them to suffer.
15
There is something fishy in van Inwagen’s discussion of what a “massively irregular” world would
be—describing a world without bestial suffering that is massively irregular hardly shows that any world
without bestial suffering has to be massively irregular. But I won’t make anything of this.
  

being, and what kind is it that entails the toleration of suffering? And what moral goal
could warrant the toleration of suffering as vast and as various as we observe in the
world around us?
Let us start with part one of the Defense. As I have already noted, the structure of
this part of van Inwagen’s Defense mirrors the structure of my hypothetical case
involving a human parent and his child. Just like the human parent in my case, God
judges that he must, for the sake of some surpassingly important moral goal, permit
his “children” to suffer the painful consequences of their willful turn away from him.
But right away, we have, in God’s case, a modal mystery: in what sense is the horrific
aftermath of the human beings’ free choice a “consequence” of that choice?
One possibility is that the modality here is nomological, that the human beings’
plunge into insecurity and misery is supposed to be a causal consequence of the
choice they made. (This would be a new mystery in itself, since known causal laws
presumably do not subsume “preternatural” phenomena. But I set this aside with the
other “natural” mysteries.) The nomological interpretation may seem singularly
implausible, since mere causal constraint is supposed to be nothing for an omnipo-
tent being. We were willing to excuse the human parent because he must always
make his choices against a background of contingencies he cannot alter, but God
cannot make the same exculpatory appeal. He is not bound by natural law, and
natural law is anyway up to him.
But we should not be too fast in dismissing the nomological interpretation. What if
causal necessity can be turned into logical necessity? The first part of van Inwagen’s
Defense doesn’t provide any help here, but consider thesis (1) from the second part of
van Inwagen’s Defense:

(1) Every world God could have made that contains higher-level sentient crea-
tures either contains patterns of suffering morally equivalent to those of the actual
world, or else is massively irregular.
Van Inwagen’s idea here is that the nomological structure of the natural world is
exquisitely finely tuned; so much so that, contrary to casual reflection, it is literally
impossible for there to be a world that is both law-governed and inhabited by rational
creatures which does not contain any lower amount or degree of suffering than what
we observe in the actual world. The task of creating a law-governed world containing
higher sentient beings in which there is less suffering than in our world turns out to
be over-constrained.
This line of response echoes the one that God makes to long-suffering Job, when
Job complains that he has done nothing to deserve his sorry state. Van Inwagen
clearly intends the parallel, and quotes the pertinent passage:
Where wast thou when I laid the foundations of the earth? Declare if thou hast
understanding. . . .
Knowest thou the ordinances of heaven? Canst thou set the dominion thereof in the earth?
(Book of Job, 38:4 and 38:34; quoted in van Inwagen 2006: 123)

God’s answer to Job is clearly meant to be a conversation-stopper; it is meant to

chasten Job into silence. But van Inwagen cannot expect Atheist to give up so quickly.
  

Van Inwagen presumes that the moral of the biblical story, and thus of his own, is
that human beings should exhibit “modal humility.”16 I am all for modal humility—I
believe in the necessary a posteriori, after all—but it is impertinent of van Inwagen to
demand such humility in the context of evaluating a Defense. If Atheist were pressing
the logical argument from evil, then perhaps the epistemic possibility that no better
possible world is metaphysically possible should end the matter. But for the purposes
of a Defense, van Inwagen must do more than advert to such an epistemic possibility;
he must give us some serious idea how the system of constraints he envisions might
look. He must dispel the modal mystery.
Van Inwagen does offer a certain amount of detail in support of (1): he argues, for
example, that there are extremely strong constraints on the values that could be taken
by fundamental physical constants in order for life to emerge, and he points out that
it is very difficult to imagine a course of evolution through natural selection that did
not involve such sources of animal suffering as predation.17 But even if we accept
these particular points,18 we are still a long way from seeing why the creation of
creatures like us human beings literally necessitates a world in which Suffering is true.
Why, for example, could God not have simply created the world ex nihilo, precisely
as we find it (or as we found it at some earlier date), thereby eliminating at least all of
the suffering that sentient creatures actually suffered during the primeval period?
Why couldn’t there be a “sudden” world?
Van Inwagen considers the possibility of a sudden world, and has this to say about it:
A world that came into existence five minutes ago, complete with memories of an unreal past,
would be on that account alone massively irregular—if indeed such a world was metaphysically
possible. (Van Inwagen 2006: 115)

Van Inwagen then appeals to Thesis (3) to conclude that any sudden world would be
inferior to the actual world.
(3) Being massively irregular is a defect in a world, a defect at least as great as the
defect of containing patterns of suffering morally equivalent to those found in the
actual world.
But in appealing to Thesis (3), van Inwagen’s response trades in the modal mystery
we started with for a massive normative mystery.19 The kind of “defect” mentioned in
Thesis (3) has to be a moral defect in order to supply an answer to the sudden-world
challenge. But why think that there is any moral dimension to the matter of
nomological regularity? It is a familiar enough idea from epistemology and the
philosophy of science that human theorists have a bias toward regularity and
simplicity in their hypotheses about the empirical world, but if such a bias reflects

16
Van Inwagen (2006: 123). See also van Inwagen (1998).
17
Van Inwagen (2006: 119 and n.12).
18
Many scientists and philosophers of science have challenged the empirical assumptions behind
contemporary versions of the “fine-tuning” argument for intelligent design. See Sober (2009).
19
Not that I concede that the appeal to the defectiveness of an irregular world does make an adequate
response to the modal question. Even if we concede that it rules out sudden worlds, it doesn’t tell us why we
should think that there are not metaphysically possible nomological systems alternative to our own.
  

any kind of value at all, it would have to be aesthetic or pragmatic value, rather than
moral value. That a regular world is lovelier than an irregular world, or that it is easier
for a human being to comprehend—those suggestions make sense. But the idea that a
regular world is, ipso facto, morally better than an irregular one is bizarre. Is a
rectangle morally better than an irregular tetragon? Does a serial killer get moral
credit for sticking with the same modus operandi? To the extent that I can make any
sense of it at all, Thesis (3) seems patently false. Given the extreme implausibility of
this ranking, and absent any argument in its favor, I say that the probability of (3) is
quite low.
Not only does Thesis (3) raise a colossal normative mystery, it fails to resolve
completely the modal mystery with which we began. Suppose we grant van Inwagen
both Theses (1) and (3)—that is, let us grant for the sake of argument that the system
of natural law that governs the actual world is so tightly constrained that not even an
omnipotent being could alter it or replace it without engendering morally problem-
atic irregularities.20 This would take us some way toward solving the modal mystery.
It would explain (a) why God cannot alter the nomological background against
which he makes choices regarding his creatures, (b) why that background is neces-
sarily one in which predation, disease, and old age afflict sentient creatures, and thus,
(c) why the loss of human beings’ “preternatural powers” renders them subject to
these horrors. However, the original mystery remains: why did the human beings
have to lose their powers? How was this loss necessary?
Theses (1) and (3) give us nothing to say in response to this question, because a
world that contains creatures with “preternatural” powers is one that contains a
significant singularity. God had to intervene in the causal order to give the human
beings such powers in the first place; since their presence is contra-nomic, their loss
cannot be a causal consequence of the human beings’ choice, and thus the modal
strengthening afforded by (1) cannot explain why this consequence was necessary. So
we still need to know: in what sense could it be true that the human beings’ decision
to separate from God resulted in their losing their protective powers?
An answer is suggested by van Inwagen’s remark that “what it means to be
separated from God is to live in a life of horrors” (2006: 88). Perhaps van Inwagen
is suggesting that the connection between separating from God and losing one’s
protective powers is a matter of metaphysical necessity—the one just is the other. If
that is true, then the answer to the question why the human beings lost their
preternatural powers when they turned away from God is that, in turning away,
they ipso facto divested themselves of God’s protection. God could not have pre-
served the human beings’ powers, given that they had separated from him, any more
he could preserve a man’s bachelorhood, given that the man has married, or—here is
a better analogy—any more than God could have prevented Oedipus’s marrying his
mother, given that he married Jocasta.Of course it is going to matter to our under-
standing of the normative aspect of the story whether this necessity we are imaging is

20
Let us also put aside worries about the metric that determines what counts as a “massively” irregular
world, one that permits such singularities as God’s miraculous gift of rationality, and that would have
tolerated a single species’ being exempt from natural law for an indefinitely long period of time, as would
presumably have been the case if human beings had not chosen to separate themselves from God.
  

one that, like the connection between bachelorhood and marriage, is knowable a
priori, or rather is one that, as in the Oedipus case, can only be learned through
experience. Even if separation from God is in fact the same state of affairs as losing
divine protection, it doesn’t follow that in choosing to separate, the human beings
thereby chose to lose protection. If the human beings didn’t realize that separating
was losing divine protection, then they did not “choose” their fate in any normatively
significant sense.
Of course, van Inwagen would probably say that the human beings did understand
what they were choosing. As he tells the story (and it is, after all, his Defense), the
human beings either knew, or should have known, the facts about what their choice
entailed. Van Inwagen writes:
As regards physical suffering and untimely death, rebelling against God is like disregarding a
clearly worded notice, climbing a fence, and wandering about in a mine field. If someone does
that, it’s very close to a dead certainty that sooner or later something bad will happen to him.
(Van Inwagen 2006: 103)
Setting aside the matter of the “clearly worded” sign (where was it?), I am willing to
grant that, if it was really clear to the human beings that rebelling against God meant
exposing themselves to the suffering endured by the other sentient creatures they saw
around them (which suffering apparently did not discomfit them while they were
enjoying mystical union with God), then it would be correct to say that they chose a
life of vulnerability. (A minor mystery: these creatures are supposed to be rational?)
However, we still don’t have a complete solution to the modal mystery.
Supposing now that it would have been impossible for God to permit the human
beings to separate from him while preserving their preternatural powers (because
there is no such possible condition to permit), we can still ask why God had to permit
them to separate in the first place. Why, when his creatures announced that they
had decided to turn away from him, didn’t God just say no? As long as there is
any metaphysical distance at all between choosing to X, and successfully doing X,
there is room for divine intervention,21 and we are back to the modal mystery with
which we began.
Let it not be thought that, if the human beings chose to separate, then some of the
normative mystery will have been dispelled—that God would have been justified in
letting them have what they asked for. He would not. To see why not, let us return
again to the case of Parent and Child. I have allowed that, if Parent has made it clear
to Child what is likely to happen to her toy if she leaves it outside, then Parent would
be morally warranted in letting Child experience the “natural consequences” of her
negligence. But there are limits to what “natural consequences” a parent may
justifiably allow a child to suffer for the sake of teaching her a lesson. Even if the
parent has repeatedly warned the child against running into the street, the parent is
not justified in standing idly by, watching while the heedless kid rushes into the path

21
Or so it seems, unless we are to believe not just that separating is the same as losing protection, but that
choosing to separate is—that is, that the human beings had the power to separate just by choosing to do so.
But why would a benevolent creator have bestowed upon newly rational creatures a power like that? It is a
mystery.
  

of an oncoming car. It hardly matters whether or not the parent has explained the
danger to the child; the parent has a duty to keep the child out of mortal danger if he
can, no matter how many times he has tried to get the message across. (“Look,
officer—I warned her.”) Schooling, by whatever means, is meant to be for the child’s
own sake, and it cannot be that if the lessons themselves are deadly.
Here is another example, one that involves a battle that many real parents have to
fight. Children who possess a certain genetic anomaly lack the ability to properly
metabolize the amino acid phenylaline (Phe), a condition called phenylketonuria
(PKU). Such children are at risk for developing a catastrophic form of intellectual
disability, due to the build-up of Phe, which is toxic to the brain. This effect can
be prevented by eliminating from the child’s diet most foods that contain the amino
acid phenylalanine, replacing them with a special dietary supplement. But children
find the prescribed diet unpalatable and boring, and it can be quite a struggle for
parents to keep the child strictly to the diet. This is especially true when the children
are young, but the problem persists even when the children are old enough to
understand that it is terribly dangerous for them to eat certain foods.22 Unlike
children who have food allergies, who usually have had the experience of a serious
allergic reaction to condition their desire for the proscribed foods, children suscep-
tible to PKU will not experience any immediate adverse effects if they happen to
ingest one of the forbidden foods. Even if they believe what their parents tell them
about the risk of eating certain foods, they do not have the motivational benefit of
knowing what the danger “feels like.” And if such a child were to argue with her
parents, and insist that she is willing to take the risk of eating (say) a hot dog, that
she knows what she is doing—the parents must simply prevent her from doing what
she wants.
It might be objected, however, that the analogy I am drawing between a parent’s
duties to a child and God’s duties toward his human creatures is faulty, due to the fact
that the human beings in van Inwagen’s Defense are presumed to be adults, with the
normal cognitive and executive capacities typical of adults. To make my analogy
more apt, therefore, we should consider not the duties of a parent to a PKU-
susceptible child but rather the duties of adults toward PKU-susceptible adults.
With this emendation, the case looks different. Is it so clear that an adult friend
would be warranted in knocking a hot dog out of the hands of a PKU-susceptible
adult friend who, in full knowledge and equipped with a fully-functioning frontal
cortex, has undertaken to eat one? Would I support a law that forbade PKU-
susceptible adults from eating the disrecommended foods?23
A full response to this objection would require a discussion of the moral value of
autonomy, a discussion that I have neither the space nor the competence to provide.
But perhaps I can finesse that by pointing out that one important reason for valuing
autonomy in the human realm is the fact that no human being can be presumed to be
better placed than I am to determine what is in my best interest. But we can certainly
presume that God knows much better than any human being what is in that human

22
See “Phenylketonuria” at the Mayo Clinic information website: http://www.mayoclinic.com/health/
phenylketonuria/DS00514.
23
Thanks to John A. Keller for raising this objection.
  

being’s best interest—in fact, it is a presupposition of van Inwagen’s Defense that this
is the case. If the disagreement between God and the human beings is legitimate—if
the human beings have some right to their own view of things, then it becomes much
less clear that God has any right to impose his will on them.
All things considered then, I submit that the human beings in van Inwagen’s story
are very much in the condition of the PKU-susceptible children. They may know that
what they want is not good for them but want it anyway. It is a normative mystery
why a loving God would let them have it.
Now one might think that this is the point at which the concept of “free will”
begins to do some real work. But it is important to see what work needs to be done.
The question before us right now is not what justified God in permitting his creatures
to turn away but rather what required God to let his creatures turn away? Free will is
not going to give us an answer to that question.
Atheist is willing to grant that it is metaphysically impossible for God to grant an
agent free will and at the same time ensure that the agent acts in certain ways.24 Thus,
she agrees that in bestowing upon his human creatures freedom of will, God must
necessarily adopt a policy of non-interference, both with respect to what an agent
chooses to do, and with respect to the consequences of the agent’s having made the
choice he or she makes.25 The problem, though, is that all this argument gives us is
the necessity of a policy, and a policy needn’t be strict. God can (and does, according
to most theists) intervene now and then without destroying human freedom. So it
remains possible for God to intervene in any particular instance. Why not here, in this
extraordinarily important case, where a single intervention could have saved billions
of sentient creatures worlds of suffering?
All things considered, I am not sure that van Inwagen wants to take the line we
have been considering. Notwithstanding the suggestive rhetoric I quoted earlier, van
Inwagen says other things that seem to imply that it would have been possible for God to
have continued to protect human beings even after they chose to separate—that God
could have done so, but chose not to. This suggests a different answer to our modal
question: it was the divine will, not metaphysical necessity, that made loss of their
preternatural powers a consequence of the human beings’ choice to turn away from God.
If this is so, then we have once again exchanged a modal mystery for a normative
one. The principle of No Tolerance Unless tells us that it is morally permissible for an
agent to tolerate suffering that he or she is capable of preventing only if such suffering
is required by some surpassingly important moral goal. What, in this case, could that
surpassingly important moral goal be?
Van Inwagen answers this question quite explicitly. It is part of the story he tells in
the first part of his Defense that God had some options in the aftermath of the human
beings’ fateful choice. One option would have been to destroy them, as an act of
mercy. God could have “simply brought the story of humanity to an end at that point,
like a man who shoots a horse with a broken leg.” But, van Inwagen says, God is

24
Atheist has compatibilist leanings, but she will not make an issue out of van Inwagen’s libertarianism.
25
Instead of appealing to free will, Theist might appeal to Thesis (1): every time God intervenes in the
affairs of human beings, he introduces into the world an irregularity—so there cannot be too many such
interventions. But the same problem will arise.
  

“more than a God of mercy; he is a God of love” (2006: 87). As such, he desired to
restore the mystical union the human beings had once enjoyed with him, and so “he
set in motion a rescue operation . . . a plan to restore separated humanity to union
with himself” (2006: 87). Van Inwagen does not offer details of the plan, except to say
that it has one absolutely essential feature: “its object is to bring it about that human
beings once more love God” (2006: 87). Since loving God is the object of the plan, and
since “love essentially involves free will” (a stipulation that I will not challenge, but of
which I am suspicious), the plan must bring it about that the human beings “choose
freely to be united with God.” But for them to make such a choice, and thereby
cooperate with God in the “rescue operation,” they must
know that they need to be rescued. They must know what it means to be separated from him.
And what it means to be separated from God is to live in a world of horrors. If God simply
“canceled” all the horrors of this world by an endless series of miracles, he would thereby frustrate
his own plan of reconciliation. If he did that, we should be content with our lot and should see no
reason to cooperate with him. (Van Inwagen 2006: 88, my emphasis)

I quoted this passage in full because I find it stunning. Not only does it cancel the
implicature of the passage cited earlier, that the human beings at least knew what
they were in for when they chose to separate from God, it makes it clear that such
horrors as they are experiencing are not endemic to their choice. God could remove
them, but he chooses not to.
Furthermore, what kind of “free choice” are we talking about here? It is an old joke
by now to speak euphemistically of a threat as “an offer you can’t refuse.” But
seriously—if the idea is that only the prospect of gaining relief from their misery
can motivate human beings to return to God, it hardly seems apt to say that they are
doing it “of their own free choice.”
Now I am sorry if the following sounds flippant, but it must be said what van
Inwagen is describing here is a protection racket: the person offering “protection” is
the very same person who makes “protection” necessary. Bear in mind that it was
God who set up the universe in such a way that it requires divinely bestowed
“preternatural powers” for a sentient being to avoid suffering, and God who ordained
that a certain choice would result in human beings’ losing these powers. It is only true
that human beings “need to be rescued” because God has prearranged for them to be
endangered. There is here an enormous normative mystery: what justifies God in
establishing a set of contingencies like this?
Van Inwagen offers an analogy that is supposed to help us understand this. He
imagines a woman, Dorothy, who is suffering pain from a condition that can only
be cured by her making large lifestyle changes. There is a medicine that will alleviate
the pain, but without doing anything to mitigate the underlying conditions. The
doctor knows that if he treats the symptomatic pain, then Dorothy will lose the
stimulus she needs to make the broad changes she would need to make in order to
restore her health. Should he, nonetheless, give Dorothy the pain medication? Van
Inwagen writes:
[P]erhaps the answer is Yes—if that’s what Dorothy insists on . . . Perhaps it would be insuf-
ferably paternalistic to refuse to alleviate Dorothy’s pain in order to provide her with a
  

motivation to do what is to her own advantage . . . [O]ne might even say that someone who did
that was “playing God.” It is far from clear, however, whether there is anything wrong with
God’s behaving as if he were God. It is at least very plausible to suppose that it is morally
permissible for God to allow human beings to suffer if the inevitable result of suppressing the
suffering would be to deprive them of a very great good, one that far outweighs the suffering.
(Van Inwagen 2006: 88)

But this is not good enough. It is not good enough for a Defense to say that it is “at
least very plausible to suppose” that there is something that makes it all right for God
to behave in a way that would be morally wrong if a human being did it. The whole
point of the Defense is to show us what that difference might be. Just as the
fundamentalist wants an example of how a part of an eye or a proto-wing might be
adaptive, Atheist wants to know what kind of difference between God’s situation and
the situation of the human doctor would make it morally permissible for God to do
what it would not be permissible for the doctor to do.
There are differences, of course, but they all tend in the wrong direction. The
doctor cannot alter the contingencies that define Dorothy’s case; he cannot alter the
fact that her obesity puts a strain on her heart, he cannot alter the fact that the only
way for her to lose weight healthily is to eat less, and he cannot alter whatever facts
about her psychology make it difficult—perhaps, all things considered, nomologically
impossible—for her to eat less. It is because none of these factors are within the
doctor’s control that we might excuse him for “playing God,” for making Dorothy
suffer “in order to provide her with a motivation to do what is in her advantage.” But
what is God’s excuse?
There is another problem. In framing the No Tolerance Unless principle, we took
account of the possibility that some amount of suffering might be necessary for the
achievement of some morally important good. But we should also have considered
the matter of the sufficiency of the suffering. Let us consider Dorothy’s case. It might
be that, as the doctor fears, Dorothy will not be motivated to watch her weight
without the angina pain to spur her on. But suppose it is also the case—and the
doctor knows this—that Dorothy’s condition has deteriorated to the point where
weight loss, although necessary for her health to be restored, will not be enough. To
become healthy again, Dorothy would need a special, terribly unpalatable diet that
she would not stick to even if the alternative was having to bear the pain of angina. In
that circumstance, the doctor could not justify withholding the analgesic on the
grounds that doing so was necessary for Dorothy’s greater good—it needs to be the
case that the “pain protocol” has at least a good chance of success.
But now let us consider our miserable human beings. Even if we assume, as van
Inwagen asks us to, that their suffering is necessary for them to achieve this “very
great good,” why should we—and why should God—assume that this “pain protocol”
has a reasonable chance of success? In the first place, it seems likely that many human
beings will fail to learn the lesson that God assumes that they’ll learn. Some human
beings are not all that uncomfortable, after all—they might not consider that they live
“in a world of horrors.” And if they do realize that there are horrors out there—
slavery, the Holocaust, the bombing of Hiroshima—they might reasonably fail to
make any connection between those horrors and their own behavior. (God might not
  

know for certain what inferences a free agent will or won’t perform, but in that case,
he could not know for certain—indeed, it would not be true—that human beings
could not achieve the frame of mind necessary for restored union without living
in a world of horrors.)
In the second place, what makes God think that his expedient, even if successful, is
going to stick? That is, even if the human beings come to apprehend that their
miserable condition can only be alleviated if they turn back toward God, why expect
that they won’t just rebel again like they did the first time? Van Inwagen wants us to
believe that the human beings had a pretty clear idea of what they were facing the first
time they rebelled—remember the analogical story about the people who ignored the
“clearly worded notice”?—and yet they still chose to turn away. Why think that, once
they are removed from the world of horrors, restored to safe and blissful union with
God, that these human beings won’t just do the same damn thing again? Van
Inwagen makes a point of telling us that he has no idea why the human beings
chose to rebel in the first place: “somehow, in way that must be mysterious to us, they
were not content with this paradisal state.” In that case, why think that they’ll be
content the second time around? If God has some way of ensuring that there will not
be another rebellion—consistent with the human beings’ retaining their freedom of
will—then why didn’t he prevent it the first time?
In sum, van Inwagen’s Defense is a failure. It is replete with modal and normative
mysteries—mysteries that present themselves at exactly the points where Atheist
most needs detail. The story he tells says nothing more than what Atheist already
conceded in conceding that the logical argument was unsound—namely that it is
conceivable that there be some kind of necessity, and some kind of great moral good,
such that the achievement of that good necessitates the amount and kind of suffering
that we see all around us. That bare epistemic possibility is not enough to make us
doubt that Suffering means just what we thought it meant—that no benevolent
omnipotent being created or controls this world.

Acknowledgments
I would like to thank John A. Keller, Joseph Levine, and Christopher Meacham for helpful
discussion of the issues in this paper.

Bibliography
Antony, L. (2013), ‘Does God Love Us?’, in M. Bergmann, M. J. Murray, and M. C. Rea (eds.),
Divine Evil: The Moral Character of the God of Abraham (Oxford).
Descartes, R. Discourse on Method.
Draper, P. (1996a), ‘Pain and Pleasure: An Evidential Problem for Theists’, in D. Howard-
Snyder (ed.), The Evidential Argument from Evil (Indiana).
Draper, P. (1996b), ‘The Skeptical Theist’, in D. Howard-Snyder (ed.), The Evidential Argu-
ment from Evil (Indiana).
Howard-Snyder, D. (ed.) (1996), The Evidential Argument from Evil (Indiana).
Hume, D. (1998), Dialogues Concerning Natural Religion, R. H. Popkin (ed.), 2nd edn.
(Hackett Publishing).
  

Lewis, D. (2006), ‘Divine Evil’, in L. Antony (ed.), Philosophers Without Gods: Meditations on
Atheism and the Secular Life (Oxford University Press).
Nozick, R. (2013), Anarchy, State, and Utopia, 2nd edn. (Basic Books).
Sober, E. (2009), ‘Absence of Evidence and Evidence of Absence: Evidential Transitivity in
Connection with Fossils, Fishing, Fine-Tuning, and Firing Squads’, in Philosophical Studies
143: 63–90.
Van Inwagen, P. (1988), ‘The Magnitude, Duration, and Distribution of Evil: A Theodicy’, in
Philosophical Topics 26/2: 161–87.
Van Inwagen, P. (1991), ‘The Problem of Evil, the Problem of Air, and the Problem of Silence’,
in Philosophical Perspectives 5: 135–65.
Van Inwagen, P. (1996), ‘Reflections on Chapters by Draper, Russell, and Gale’, in D. Howard-
Snyder (ed.), The Evidential Argument from Evil (Indiana).
Van Inwagen, P. (1998), ‘Modal Epistemology’, in Philosophical Studies 92: 67–84.
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
 9
The Problem of Evil
and Atonement
Eleonore Stump

9.1 Introduction
In the course of his long and distinguished career, Peter van Inwagen has written
extensively about the problem of evil, carving out a position that has been widely
influential and is still the subject of much discussion today.1 Although my own views
on the problem of evil diverge from van Inwagen’s at various points, I have learned a
great deal from his work on this subject.2 And, in fact, underlying the disagreements,
there is considerable agreement about some fundamental claims. Among them are
the claims that the ultimate good for human beings is union with God, that this good
is imperiled by post-Fall human tendencies towards evil, and that suffering, gener-
ically considered, has some role to play in promoting that ultimate good.3
In this paper, I want to move forward from these general points of agreement by
exploring in a preliminary way the connection between attempted solutions to the
problem of evil and the doctrine of the atonement,4 specifically the view that there are
redemptive effects for human beings stemming from the passion and death of Christ.5
It is widely supposed, by both Christians and non-Christians alike, that the
doctrine of the atonement is the distinctive doctrine of Christianity; and Christians
often speak of the value of the atonement itself as infinite, or so great as to be
incommensurate with all other created goods.6 Of course, the doctrine of the

1
See van Inwagen (1988a; 1988b; 1991; 1996; 1997; 2000; 2004; 2005; 2006).
2
The theodicy that I myself favor I have explained and defended, at length, in Wandering in Darkness:
Narrative and the Problem of Suffering (2010).
3
In the convergence on these fundamental points, of course, there is nothing distinctive about my views
and van Inwagen’s. These claims have been espoused by the overwhelming majority of Christian thinkers
from the beginning of the Christian era to now.
4
For a more detailed explanation of the basic idea of atonement, see my “The Nature of the Atonement”
(2012).
5
Or, the incarnation, passion, death, and resurrection of Christ. Here I will focus on the passion and
death of Christ, but the main points would not be altered if the focus included the incarnation and
resurrection as well.
6
See, for example, Plantinga (2004). For some critical commentary on the theodicy Plantinga proposes
in this paper, see McCord Adams (2008). I share some of her concerns about attempts at theodicy based
largely or wholly on comparisons of the summed value of worlds.
       

atonement has been interpreted in varying ways in the history of Christian thought.7
And, traditionally, the passion and death of Christ have been thought to have not just
one effect but several. (Aquinas, for example, says in addition to its redemptive
effects, Christ’s passion operated as a source of merit, as a sacrifice, and as satisfaction
for human sins.8) Here, I will focus on the doctrine of the atonement with regard to
the redemptive effects of Christ’s passion and death; and I will try to do so in a way
that sidesteps differences among the varying interpretations of the doctrine and that
leaves aside considerations of any other effects of Christ’s passion and death.
My purpose in restricting the focus in this way is to explore the confluence of the
redemptive effects of the atonement and the role of suffering in God’s providential
plan for bringing human beings to union with God.9 There is reason to expect that
there would be some confluence between the doctrine of the atonement and a
solution to the problem of evil (assuming that there is one). In particular, we could
reasonably expect that in a consistent Christian theology there would be some
connection between the redemptive effects of the atonement, on the one hand, and
the good promoted by human suffering, on the other.
On one of the traditional Christian claims van Inwagen and I both accept,
something about suffering somehow conduces to the ultimate good for human
beings, which is union with God.10 But the atonement is also supposed to provide
this ultimate good. And so it seems that there should be some intrinsic connection
between the benefits for human beings postulated by the doctrine of the atonement
and the benefits supposed to justify God’s allowing suffering in the world. The
doctrine of the atonement and a theologically acceptable explanation of suffering
in the world ought to have a connected place in one grand unified theological theory
of everything in God’s providential plan of salvation.
To begin to explore what this connected place could be, it helps to think about
atonement more generally. ‘Atonement’ is an invented word composed of ‘at’ and
‘one’ jammed together with ‘ment’. It was once a neologism, coined to express the
nature of the solution to a problem. Traditionally, that problem has been under-
stood as the absence of oneness or closeness between God and human beings.11
However exactly we are to understand closeness and union between persons, it will
at least include harmony between their minds and wills.12 A person who does
something that is contrary to the will of a perfectly good God introduces distance
between himself and God. Understood in this way, distance between God and a

7
I have discussed some of these in Stump (2012a) and in “Conversion, Atonement, and Love,” (2013b).
8
Cf., e.g., ST III.48.
9
And even here I am limiting myself to those salvific effects that have to do with what I have elsewhere
called ‘the future problem of sin’, namely, the sinful dispositions in a human psyche. (See Chapter 15 of
Stump (2003).) There is also a past problem of sin, which I am leaving out here in the interests of brevity.
I am addressing that problem in this context in “The Atonement and the Problem of Shame,” (2016).
10
See Chapter 13 of Stump (2010) for a defense of the claim that allowing suffering is justified only if
the benefit brought about by the suffering goes primarily to the sufferer.
11
Of course, although this word is relatively new, religious rituals of atonement are old. They figure
prominently in the Hebrew Bible, as, for example, in the prescriptions for the day of atonement laid out in
Leviticus.
12
For a detailed discussion of the issue, see Chapter 6 of Stump (2010).
  

human person has at least a partial source in a human person’s failure to will what
God wills.13
In fact, as Augustine and the Christian tradition after him understood the absence
of unity between God and human beings, it is a function of the post-Fall human
proneness to sin, where sin is understood as something that is contrary to the will of a
perfectly good God.14 As Augustine himself helped to highlight,15 one hallmark of
this post-Fall defect in the will is the will’s intractability to itself, its proneness to
moral wrong even against its own desires for the good.
On this view, human beings will what they take to be good, but they can also
simultaneously will against it. Furthermore, although human beings can have a
second-order will to will the good, that higher-order will is frequently rendered
ineffective by contrary first-order willings of what is not good. A person in such a
condition wills the morally wrong thing he himself desires not to will. Worse yet, a
human being can be in so lamentable a moral condition that even his second-order
will is corrupted. He not only fails to will what is good but he also fails to will to will
what is good.
In addition, it is part of traditional Christian views that a human being cannot be
united around what is morally evil.16 The objective moral standard, at least in its
rudiments, is so accessible to ordinary reason that no human intellect is ever totally in
ignorance of it. Someone who takes something objectively evil, such as beating his
wife, to be morally acceptable or even morally good will also have some awareness,
however tacit or covert, that it is in fact morally reprehensible. Any wrongdoing
therefore generates a kind of double-mindedness in the wrongdoer. And double-
mindedness generates a correspondingly conflicted set of desires and volitions.17
Consequently, it is not possible for a person’s mind or will to be internally integrated
in evil; no one is completely single-minded or wholehearted in evil.18 For these
reasons, internal integration is possible only for a person single-mindedly under-
standing and wholeheartedly desiring the good.

13
For discussion of the issue and arguments for this claim, see Stump (2012b).
14
Whether one takes morality to be grounded in God’s will or in God’s nature, it remains the case that
willing what is morally bad is willing something contrary to the will of God.
15
So, for example, commenting on his own failures to will what he himself takes to be the good and
wants to will as the good, Augustine says,
the mind commands the body, and is presently obeyed: the mind commands itself, and is
resisted. . . . it commands that itself would will a thing; . . . [and it] never would give the
command, unless it willed it; yet it does not [will] . . . [what it] has commanded. . . . it
commands, . . . [because] it wills: and . . . the thing [is not] done which . . . [it] commanded, . . .
[because] it wills it not. . . . But it does not command fully, therefore is not the thing done,
which it commanded. For were the willing full, it never would command it to be, because it
would already be.
Augustine, Confessions VIII.9.
I like and have therefore used (with slight modifications) the translation by William Watts (Harvard
University Press, 1968).
16
For a discussion and defense of this position, see Stump (2010: Chapters 6 and 7).
17
For a discussion of the connection between intellect and will presupposed here, see Stump (2003).
18
See, for example, ST II-II q.45 a.4 and a.6. The relevant biblical claim can be found in Isaiah 48:22.
       

The problem to which the atonement is the solution therefore can be thought of as
the human tendency to fragmentation in the will, brought about by the human
proneness to sin. And, on orthodox Christian views, the atonement is a sufficient
solution for this problem. That is, the atonement offers redemption to all those who
will avail themselves of it; it makes union with God available to everyone who does
not reject it.19 Somehow20 the atonement brings it about that, in the end, those
human beings who do not reject the salvation offered in the atonement are united to
God. And, in union with God, they are internally integrated around the good. They
will only what God wills.
Given this way of understanding the doctrine of the atonement, however, what
role is left for suffering in the process of providing the ultimate good for human
beings? If the atonement makes union with God available to all human beings, then
what happens to the view that suffering also has a role to play in promoting union
with God? Is there something central to achieving union with God that is not
provided by the atonement but that suffering facilitates? Does suffering somehow
enable whatever the means are by which the atonement promotes union with God?
Or put the question the other way around. On the face of it, it seems in principle
possible to have a world in which there is the post-Fall human proneness to sin and
also the atonement to provide salvation from that post-Fall condition, but in which
no one suffers, because the proneness to sin is never successful in bringing harm to
others and there is no other source of suffering either. In such a world, what if
anything that would conduce to the ultimate good for human beings would be lost by
the absence of suffering?
Presenting and defending a grand unified theological theory that brings together
the doctrine of the atonement and any kind of attempted solution to the problem of
evil is obviously beyond the scope of a short paper. But even a preliminary answer to
these questions would give some insight into the connection between the doctrine of
the atonement and explanations for God’s allowing suffering; and so it would be
propaedeutic to such a grand unified theory. In this paper, I want to sketch such a
propaedeutic account.
To this end, I will first give a short characterization of the nature of suffering. Then
I will highlight the different routes by which suffering can move a person to alter
what he wills. Among these routes are two very different ways in which a sufferer can
try to harmonize his will with God’s will. I will argue that only one of these two ways
has any chance of success at bringing the sufferer either to peace or to harmony with
God’s will. Then I will change course and consider justification and sanctification,
two processes central to salvation, on orthodox Christian views; and I will examine
the way in which Christ’s passion and death function to promote each of these

19
These claims are true even for those who accept a doctrine of double predestination. On that doctrine,
those who reject the salvation provided by the atonement were predestined by God to reject it. In that
sense, the atonement was not intended for them, even if the value of the atonement was in itself so great
that it would have sufficed for salvation for everyone.
20
The various interpretations of the doctrine of the atonement differ in their understanding of this
‘somehow’. There is no creedal formula that effectively distinguishes among them and privileges just one as
orthodox. For some discussion of the main kinds of interpretation in the history of Christian thought, see
Stump (2012a).
  

processes. Finally, I will return to the paper’s main question and outline a way in
which both justification and sanctification are promoted by the only promising way
of willing what God wills in consequence of suffering.
The result will be an account of the way in which Christ’s suffering in his passion
and death can be interwoven with the suffering of a person Paula to work her salvation.
This account falls short of a full explanation of the confluence between the doctrine of
the atonement and an acceptable theodicy (assuming there is one). It is not sufficient
for a grand unified theological theory of God’s providential plan of salvation. On the
contrary, it is only programmatic. But, I hope, it will nonetheless be enough to suggest
directions for incorporating the views about the role of suffering in salvation that van
Inwagen and I both share into an account of the redemptive effects of the atonement.

9.2 The Nature of Suffering

If the problem to which the atonement is the solution consists in the post-Fall human
proneness to sin, then overcoming the human failure to will the good is central to the
redemptive work of the atonement. Overcoming it requires a person to be integrated
around the good. God wills only the good, however; and so a person’s proneness to
sin and fragmentation in will is healed to the extent to which she wills what God wills.
Consequently, if suffering also has a role to play in bringing about the ultimate good
for human beings of union with God, as attempted solutions to the problem of evil
have typically supposed, then something about suffering too should contribute to
bringing a person to the state of willing what God wills.
Now perhaps the most notable characteristic of most suffering is its aversive charac-
ter. Generally, a person in the midst of suffering is motivated to try to end his
suffering;21 and so suffering has a motive power. One way to explore the role of suffering
in the process of bringing a human person to willing what God wills is therefore to
consider the differing ways in which the aversiveness of suffering can move a person.22
In this regard, it is helpful to recognize that suffering itself is a function of the will.
Although it is tempting to think of suffering just as pain, a little reflection shows that
this way of characterizing suffering is mistaken. A human being can suffer even
without experiencing pain, as, for example, when a person suffers from an unjust
political system of which he approves and to which he is strongly committed on
ideological grounds.23 As I have argued elsewhere,24 a better way of thinking about
suffering is to see it as a function of what we care about.

21
The claim has to be qualified in this way, of course, because there are some kinds of pain or suffering
that human beings actually seek out. So-called natural childbirth is one example, and masochistic practices
are another. But it is enough for my purposes that most pain and suffering is aversive for most sufferers.
For a more detailed discussion of suffering, see Chapter 1 of Stump (2010).
22
I say ‘can have’ here because, of course, the aversiveness of suffering can also drive the sufferer into
psychic states that increase the fragmentation in his will and so alienate him from God. Insofar as a human
person’s will is free, suffering cannot determine any particular psychic state in that person. For a discussion
of these issues, see Stump (2010: Chapters 8 and 13).
23
In such a case the diminishment of flourishing can constitute an aversive stimulus even if pain
does not.
24
For detailed discussion of this claim, see Stump (2010: Chapter 1).
       

Considered in this way, there are two sides to suffering, an objective side and a
subjective side.
Because every human person has some care about what kind of person he is and
about his flourishing as that kind of person, part of what it is for him to suffer is for
him to be kept, to one degree or another, from flourishing. This is an objective side of
suffering, since there is an objective fact of the matter about what will make a human
person flourish.
On the other hand, however, what we care about has a subjective side too. This is
something to which a person is committed but which is not identical with his
flourishing and which may not even be compatible with it. What is at issue in this
subjective side of suffering can be thought of as the desires of the heart.25 When the
Psalmist says, “Delight yourself in the Lord, and he will give you the desires of your
heart,” we all have some idea of what the Psalmist is promising. Suffering also arises
when a human being fails to get a desire of his heart or has and then loses a desire of
his heart.
Human beings care about two kinds of things, then—their own objective flour-
ishing and also those things that are the subjective desires of the heart. Suffering
arises when something impedes or removes either of these kinds of things that people
care about.26 Another way to put this conclusion is that people suffer because they
lose or do not get what they care about.27 Consequently, as many thinkers in different
cultures and times have pointed out, human suffering arises because of human desire.
For this reason, also as noted in the history of philosophy and religion, there are
two different ways in which human suffering can be diminished or warded off. Either
(1) a human person can bring it about that she gets what she cares about and wills to
have, or else (2) she can will to accept what she gets and cease to care about what she
does not get.
Sometimes the first of these ways is efficacious. A person exerts himself or is lucky;
and the result is that, even in the face of initial setbacks causing suffering, he succeeds
after all in getting what he wants. But, of course, cases of this sort are in the minority
by comparison with cases in which the sufferer is helpless to change his circum-
stances and by that means get what he wants.
When a sufferer cannot get what he wants because the circumstances will not bend
to his efforts, then if he persists in wanting exactly what he cannot get, he will not
only perpetuate his suffering, he will add to it the frustration of his impotent efforts
and the anger or anxiety of his checked will. Insofar as suffering is aversive, this sort

25
The expression ‘the desire of the heart’ is ambiguous. It can mean either a particular kind of desire or
else the thing which is desired in that way. When we say, “the desire of his heart was to be a great
musician,” the expression refers to a desire; when we say, “In losing her, he lost the desire of his heart,” the
expression refers to the thing desired. I will not try to sort out this ambiguity; I will simply trust to the
context to disambiguate the expression.
26
To say that suffering is a function of what a person cares about is not to make all suffering a subjective
matter. To see the point here, consider Aristotle’s claim that all human beings desire happiness. Even
though this desire is something subjective, human happiness and the things conducive to it have an
objective character, as Aristotle shows.
27
There are, of course, cases in which a person is confused about what would make him flourish, or
cases in which what he cares about in regard to flourishing are incompossible things. For complications
involving such cases, see Stump (2010: Chapters 1, 13, and 14).
  

of reaction to the failure to get what one wants is unstable. A frustrated, angry, or
anxious sufferer will be motivated by his suffering and by his unhappy reactions to it
to seek some more peaceful psychic state.28
For these reasons, one might suppose that a sufferer would or should prefer the
second way to try to avoid suffering. That is, unless his efforts to get what he wants
are successful, the aversiveness of his suffering would or should move him to attempt
to will to accept what he gets and to cease to care about what he does not get.
In fact, one might suppose that this is the very reaction to suffering that orthodox
Christianity mandates. The ultimate good aimed at by the atonement is union with
God, which requires harmonization of the divine and the human wills. But, one
might reasonably suppose, on Christian doctrine no instance of suffering would
occur if God willed that it not occur. So it might seem that, from a Christian point
of view, any suffering that actually occurs does so only because it is in accord with the
will of God. For a person Jerome to will what God wills, then, it might seem that, with
regard to either his flourishing or his heart’s desires, Jerome must will to accept
whatever he gets when something contravenes his flourishing or his heart’s desires.
Consequently, insofar as willing what God wills is necessary for the ultimate good of
union with God, then it appears that the second approach to suffering has to be the
right one, on Christian views. A person should let go of his care for his own
flourishing or his heart’s desires, and he should will to accept what he gets.

9.3 Willing What God Wills

9.3.1 The No-Self View
But this conclusion is hasty. It rests on a particular interpretation of what it is for a
human person to will what God wills,29 and not all the major thinkers in the
Christian tradition have accepted it.
One medieval thinker who has accepted it is Meister Eckhart, and it will be helpful
to see his view in order to highlight the somewhat complicated alternative to it.
In discussing what he takes to be the right reaction to suffering, Eckhart says,
Seneca, a pagan philosopher, asks: ‘What is the best consolation in sorrow and in misfortune?’
And he says: ‘It is for a man to accept everything as if he had wished for it and had asked for it’;
for you would have wished for it, if you had known that everything happens by God’s will, with
his will and in his will. . . . [In] that alone, that it is God’s will that it should happen so, a good
man’s will ought to be so wholly one and united with God’s will that he and God have only one
will, though that should be for the man’s harm or even his damnation. . . . [A]ll his blessedness
consists in . . . willing and wanting to know nothing but God’s will.30

28
Of course, he might not act on that motivation. For some people, there is a kind of consolation in
chewing on anger, for example, which they may be loath to give up.
29
It also rests on an unsophisticated understanding of what God wills. See the discussion of the
difference between the antecedent and the consequent will of God in Chapter 16 of Stump (2003).
30
Eckhart (1981: 215–16).
       

There are many other passages in Eckhart’s work that one could cite to confirm the
impression of his views given by this comment on Seneca’s views. Elsewhere, for
example, Eckhart says,
[T]hrow all anxiety out of your heart, so that in your heart there be nothing but constant
joy. . . . [E]ven if I had to see with my own eyes my father and all my friends killed, my heart
would not be moved by it. . . . I have rightful joy only when neither sufferings nor torments can
ravish it from me. Then I am translated into the divine being where no suffering has a place. . . .
If you reach a state where you feel neither suffering nor vexation from whatever may happen,
so that suffering is not suffering for you and that all things are sheer joy for you, then the child
is truly born [that is, you have achieved spiritual regeneration.]31
Or, to take one last example, which makes the point in a more radical way, Eckhart
says,
it is not sufficient for us to have a detached attitude of mind at a specific point in time when we
wish to bind God to ourselves, but rather we should have a practiced detachment. . . . We must
learn to free ourselves of ourselves . . . , not holding on to what is our own or seeking anything,
either profit, pleasure, inwardness, sweetness, reward, heaven or our own will. God never gives
himself, or ever has given himself, to a will that is alien to himself, but only to his own will. . . .
[T]he more we cease to be in our own will, the more truly we begin to be in God’s will. . . . We
must train ourselves in self-abandonment until we retain nothing of our own. . . . We should
establish ourselves . . . in the best and most precious will of God through a pure ceasing-to-be of
our will and desire.32
As Eckhart sees it, then, for a person Jerome to will what God wills, Jerome can have a
second-order will to have a will that wills what God wills, but he must not have a first-
order will for anything at all. He must not desire, as Eckhart puts it, any profit or
pleasure or reward. He must not even desire union with God in heaven. His first-
order will must be empty, as it were, in order to enable him to accept whatever
happens.
In fact, Eckhart is willing to go even further. In some places, he writes as if he
thinks even the desire for willing what God wills has to be stamped out. In explaining
the spiritually excellent state of being poor in spirit, he says,
So long as a man has this particular wish to fulfill the ever beloved will of God . . . then this man
still has a will with which he wants to satisfy God’s will. . . . [A]s long as you have the will to
fulfill God’s will, . . . you are not poor [in spirit]; for he alone is a poor man who wills nothing
and desires nothing.33
As Eckhart sees it, then, even if Jerome has no first-order desires for anything in
particular, if he still has a second-order willing of his own, then what Jerome is
willing is his will, and not God’s will. On Eckhart’s way of thinking about the matter,
the only way really to will what God wills is for a person to have no will of his own at
all, on either the first-order or the second-order level, so that the will operative in a

31 32
Eckhart (1978: 135–6). Eckhart (1994: 41–2).
33
Schuermann (1978: 215).
  

human person is just the will that is God’s will (as distinct from the human will in
harmony with God’s will).
Since a person’s will is at least in part constitutive of his self, however, then if
Jerome has no will of his own, even at the second-order level, there is a sense in which
Jerome has so self either. In some places, Eckhart writes as if this is just the end he
supposes people should seek. So, for example, he says,
You might ask ‘When is the will a right will?’ The will is perfect and right when it has no
selfhood and when it has gone out of itself, having been taken up and transformed into the will
of God.34

If these texts represent Eckhart’s view accurately, then we would have to say that for
Eckhart a person who succeeds in willing what God wills simply has no will of his own
at all. The will operative in him is the will that is God’s and not his own, human will.
A person in this condition will certainly diminish his suffering. Insofar as suffering
results from the loss or undermining of what a person cares about, either as regards
his heart’s desires or as regards his flourishing, a person who cares about nothing will
not suffer either. As Eckhart claims, such a person will have only joy, even if he
should see his father being killed before his eyes.
The problem for Eckhart is that the response he advocates to suffering is destruc-
tive of the self, and the destruction of the self is itself destructive of the very possibility
of union with God.35 That is because union between persons requires that there be
two persons and thus two wills to unite.36 This fact is the reason why an internally
fragmented post-Fall will is an obstacle to the ultimate good for human beings of
union with God. There cannot be union between God’s will and a human will when
the internal divisions in the human will keep that will from being one will. By parity
of reasoning, there cannot be union between God and a human person if there is no
will at all in the human person.
And so in virtue of attempting to stamp out all will, or at least all first-order will, in
a human person, Eckhart’s recommended response to human suffering makes that
response at least as deadly as the original disorder in the will, for which the suffering
was supposed somehow to be therapeutic.
Furthermore, there is something inhuman about a person who tries to live out
Eckhart’s recommended response to suffering. Human beings are social animals, and
so part of what it is to be human is to be connected to others in social and caring
ways. To care for another person, however, is to desire both the good for that person
and union (of one or another appropriate kind) with that person.37 Eckhart might be
able to argue that in desiring nothing for another person, one is somehow willing
what God wills for that person and therefore willing the good for that person,

34
Eckhart (1994: 16). The German word translated ‘selfhood’ is ‘eigenschaft’.
35
For the distinction between the stamping out of the self and the denial or crucifixion of the self
recommended in the New Testament, see Stump (2010: Chapter 14).
36
On orthodox Christian views, there is only one will among the three persons of the Trinity. But the
unity among the persons of the Trinity is a metaphysical as-it-were substantial unity and not a coming
together into union of two separate entities. For metaphysical unity of the sort at issue in the Trinity, only
one will is in fact necessary.
37
See the discussion of love in Chapter 5 of Stump (2010).
       

although it seems to me that this argument would not be easy to make. But what one
cannot imagine is that a person who accepted Eckhart’s position could desire any
kind of union with another person. Someone who desires another person, as a
mother desires her child or a lover desires his beloved, cannot be correctly described
as a person who has no desires at all. And so to the extent to which one has stamped
out desire in oneself, one has also stamped out the desire for union with others. In
this lack of a desire for union with any particular others, there is also a lack of care for
them; and that lack of caring connection to others has something inhuman about it.
In addition, because there is something inhuman about it, ordinary human nature
will kick back against Eckhart’s position. And even for those people who do attempt
to adopt it, Eckhart’s recommended response is guaranteed to be unsuccessful, for
the same reasons that the will cannot be integrated in evil. No matter how much in
the grip of Eckhart’s theory a person is, some part of his intellect will recognize,
however tacitly or covertly, that the destruction of his flourishing or his heart’s
desires is not good. He will be unable not to see, for example, that the killing of his
father before his eyes is not good. And to that same extent he will be unable not to
desire that it not happen. So the attempt to stamp out in oneself all first-order desires
will result only in the fragmentation of the will that was the problem in the first place.
And, mutatis mutandis, similar things can be said about the attempt to stamp out any
desire in oneself even at the second-order level, which Eckhart sometimes seems also
to recommend.
Finally, insofar as Eckhart’s approach to suffering is unsuccessful, it will also not
alleviate the aversiveness of a person’s suffering.
For these reasons, Eckhart’s interpretation of willing what God wills and his
recommendation of his brand of willing what God wills will be at least as unsuccessful
at ameliorating suffering as the attempt to avert suffering by getting what one wills.

9.3.2 Having A Self To Deny

Against the background of Eckhart’s views, it is easier to understand the different
interpretation of willing what God wills that can be found in Aquinas’s work. In my
view, Aquinas’s interpretation and his recommendation of willing what God wills is a
much more promising approach to suffering, both as regards Christian theology and
as regards the amelioration of suffering. Like Eckhart, Aquinas maintains that the
goodness of the human will requires its conformity to the divine will;38 but his
understanding of this conformity is very different from that of Eckhart.
One way to begin to see the difference is to consider Aquinas’s response to
objections against the claim that human beings are obligated to will what God
wills. As Aquinas presents these objections, one objection argues that a human
person does not always know what God wills and that this ignorance militates against
the obligation to will what God wills.39 In reply to the objection, Aquinas says,
we know that whatever God wills, he wills it under the aspect of the good. Consequently,
whoever wills a thing under any aspect of the good has a will conformed to the Divine will, as to

38 39
ST I-II q.19 a.9. ST I-II q.19 a.10 obj.1.
  

the reason of the thing willed. But we know not what God wills in particular: and in this respect
we are not bound to conform our will to the divine will.40

In explaining his reply to the objection, Aquinas says,

the will tends to its object according as it is proposed by the reason. . . . And therefore if a man’s
will wills a thing to be according as it appears to be good, his will is good. . . . Now a thing may
happen to be good under a particular aspect, and yet not good under a universal aspect . . . and
therefore it comes to pass that a certain will is good from willing something considered under a
particular aspect, which thing God wills not [insofar as it is considered] under a universal
aspect. . . . 41

On this way of thinking about harmony between God’s will and a human will, there is
harmony when, like God, a human being wills the good because it is good. And this
harmony can exist even when the particular things wanted by the divine and the
human will are different. God grasps the good universally, that is, with all things
considered, whereas a human mind is limited in its apprehension of the good. So God
can reject as not the all-things-considered good something that a human being rightly
takes as good and wills because it is good; and yet it can nonetheless be true that the
human will is willing what God wills, in virtue of willing the good because it is good.
Aquinas thinks that this distinction between willing the good under a particular
aspect and willing it under a universal aspect contains the answer to yet another
objection to the claim that a person is obligated to will what God wills. Aquinas puts
that objection this way:
if a man were to will what God wills, this would sometimes be contrary to filial piety: for
instance, when God wills the death of a father: if his son were to will it also, it would be against
filial piety.42

Here Aquinas is addressing the sort of case discussed by Eckhart. Eckhart’s view is
that a person Jerome trying to will what God wills should accept the death of his
father as a joyful thing because it is in accordance with God’s will; and so, from
Eckhart’s point of view, the objection Aquinas is considering is just mistaken. From
Eckhart’s point of view, willing that his father die when his father does die is actually
required for Jerome to will rightly. But, as Aquinas sees it, the objection is correct in
supposing that it would be contrary to right willing if Jerome were to will the death of
his father when Jerome’s father does in fact die. And yet, Aquinas thinks, the
objection is confused in supposing that in order for Jerome to conform his will to
God’s will, Jerome needs to will the death of his father if this is what God wills.
On Aquinas’s view, Jerome’s will can be conformed to God’s will even when
Jerome’s will and God’s will are in opposition as regards the death of Jerome’s
father.43 To be in conformity with God’s will, Aquinas holds, it is enough for Jerome

40
ST I-II q.19 a.10 ad 1. The translation in this and the following quotations is from the standard
translation of the Fathers of the English Dominican Province, which I like and therefore have used, though
sometimes with slight modifications.
41 42
ST I-II q.19 a.10. ST I-II q. 19 a.10 Obj.3.
43
Aquinas’s position here depends crucially on his distinction between God’s antecedent and God’s
consequent will. Roughly put, God’s antecedent will is what God would have willed if everything in the
       

to will what he wills under the aspect of the good, for the sake of the good. To will
something in this way is to will what God wills, namely, the good, and also to will it as
God wills, namely, out of love of the good, that is, out of charity.44 For Aquinas, then,
a person Jerome can will the opposite of what God wills—say, with regard to the
death of Jerome’s father—and still be willing what God wills, as long as Jerome is
willing what he rightly takes to be a real good and willing it because it is good.45
For Aquinas, the model for willing what God wills and in fact the model for the
right response to suffering,46 is given by Jesus in his three-fold prayer in the Garden
of Gethsemane.47 In the first part of the prayer, Jesus asks that God would let the cup
pass from him. It is evident that in this part of the prayer Jesus is giving voice to his
desire not to die. So (in his human will48) Jesus is opposed to his death; his will does
not accept it as a good thing. On the other hand, Jesus finishes the prayer by saying to
God, “not my will but yours be done.”
In commenting on this prayer, Aquinas holds that in his prayer Jesus willed
something different from what God willed, because Jesus willed not to die when
God willed that Jesus die. Nonetheless, Aquinas thinks, in willing not to die, Jesus’
will was still in conformity to God’s will.49
Using Aquinas’s interpretation of willing what God wills, described above, we can
gain insight into Aquinas’s thought here by understanding Jesus’ will during his
prayer in terms of the hierarchical structure of the will.50 On this understanding (in
his human nature) Jesus had a first-order desire not to die. In this desire, he was
rightly taking not dying as a good and desiring it because it is good. But he also
understood that God’s grasp of the good could be different from his own, so that his
dying might be the good universally considered, as Aquinas puts it. And so (in his
human will) with regard to his dying, Jesus had a second-order desire for a will that
wills what God wills.
Even though his first-order desire was in opposition to God’s will as regards his
dying, Jesus’ first-order desire was in harmony with God’s will insofar as it willed

world had been up to him alone. God’s consequent will is what God actually does will, given what God’s
creatures will. For Aquinas, the will with which God assents to suffering is only his consequent will, not his
antecedent will. For a discussion of this distinction with regard to the problem of suffering, see Chapter 13
of Stump (2010).
44
ST I-II q.19 a.10.
45
More nuance is needed here in order to make this position consistent with Aquinas’s view that an
erring conscience binds. In cases where the conscience of a person Jerome is in error, then on Aquinas’s
views Jerome will be double-minded about the goodness of what he is willing, and he will not be whole-
hearted in willing it either. But these nuances, important as they are for other purposes, do not need to be
pursued for my purposes here.
46
See, for example, ST III q.21 a.3 and a.4, where Aquinas explains the different ways in which the
prayer in the Garden of Gethsemane instructs human beings and gives them a model to follow.
47
See Luke 22:42, Mark 14:36, and Matthew 26:39. For the remaining prayers in Gethsemane, see
Matthew 26:42–4.
48
This qualification is needed because, on the Chalcedonian formula for the incarnate Christ, there are
two wills in Christ. One of these is the human will of Christ, and the other is the will of God. Insofar as the
will of God cannot be opposed to the will of God, the only will of Christ that can be in any sense different
from the will of God is the human will of Christ.
49
ST III q.18 a.5.
50
For more discussion of this issue and Aquinas’s position, see Chapter 9 in Stump (2003).
  

what it willed as a good and willed it out of love of the good. In addition, Jesus’
second-order desire was straightforwardly in harmony with God’s will insofar as
Jesus wanted to will what God willed, even if God’s will should turn out to be in
opposition to Jesus’ first-order desires.
As Aquinas sees it, then, in this prayer in Gethsemane, (in his human will) Jesus
has a first-order desire for what he himself really does want, namely, not to die, and a
second-order volition51 for a will that wills what God wills, even if God’s will is that
Jesus die. Both the first-order desire and the second-order volition are part of Jesus’
will in this prayer.
As is made clear by the last part of the prayer he prays in Gethsemane and the rest
of the story, after the first part of his prayer, Jesus comes somehow to understand that
God’s will is that Jesus die then. Because Jesus has a second-order volition for a will
that wills what God wills, his second-order volition combines with his understanding
of God’s will to elicit in him a first-order desire to die then. But this new first-order
desire, to die then, does not stamp out the desire not to die. He does not go to his
death joyously, for example. And so the two opposed first-order desires exist in him
simultaneously.
Consequently, insofar as God’s will is that Jesus die then, and Jesus comes to
believe that it is, his first-order will is divided against itself. But because Jesus is
committed to the desirability of letting God’s desires take precedence over his own,
the first-order desire in himself which becomes the volition on which Jesus acts is the
first-order desire to die then. In this rank-ordering of God’s will over his own first-
order will not to die then, however, Jesus does not give up his desire not to die. He
still has that desire; he just acts counter to it, because his volition to will what God
wills is effective in determining which of his divided first-order desires he acts on. In
this sense, Jesus exhibits both his persistent maintaining of his own self (contrary to
the sort of position Eckhart recommends) and the denial of that self in the interest of
willing the good that God wills.
And so, as Aquinas sees it, in his prayers in the Garden of Gethsemane Jesus in effect
teaches people both that they should will what God wills and that they can count as
willing what God wills even when they desire something that God does not will.52
On Aquinas’s view, therefore, the right response to suffering is complex. It requires
a person Paula to continue to hold on to the desire for the thing that is lost or
withheld in the suffering; but at the same time it requires her to accept the loss
because she desires to have a will that wills what God wills. That is, she desires that
her father not die, even while she accepts the death of her father in virtue of willing to
will what God wills. This complex state in her will is matched by a complex state in
her intellect. She takes the death of her father to be in some sense a bad thing, but she

51
As I am using the term ‘volition’ here, a volition is a desire that would be effectual in bringing about
the thing it wills if nothing external to the will impeded it. When an agent’s second-order desire counts as a
volition, it is because that second-order desire is effectual in bringing the agent’s first-order desires into
harmony with it, so that the first-order desire which becomes the agent’s first-order volition is the first-
order desire that the agent wants to have. For more discussion of this distinction between desire, generically
considered, and volition as a species of desire, see Stump (1988).
52
ST III q.21 a.2.
       

also takes as a good thing accepting this bad thing, in virtue of the fact that she
takes God and God’s will to be good and she recognizes the death of her father as
God’s will.
If one thinks in terms of medical care, the difference between Eckhart’s view and
Aquinas’s view may become clearer. Suppose that Paula’s father is being subjected to
a painful bone marrow transplant in the hospital. Then in the grip of a spirit such as
Eckhart’s, Paula might simply accept with joy her father’s having this procedure, on
the grounds that it is what the doctors want. Or, in the Thomistic spirit, Paula might
both want and not want her father’s having this procedure. She might take the
medical things done to her father as bad, even while she accepts that agreeing to
them is good because it is in accordance with what the doctors want for her father. In
this condition, Paula’s desire that her father not have the transplant is overridden but
not extirpated by her desire to will what the doctors will. And Paula has this second-
order will because she takes the doctors and the will of the doctors to be good.
This complex state in the will in response to suffering can provide some of the
amelioration of suffering sought by the acceptance Eckhart advocates, but it does so
without the failure and inhumanity attaching to Eckhart’s approach. The persistence
of the thoroughly human first-order desires even in the face of suffering provides the
humanity of the Thomistic approach, and not having to try to stamp out those first-
order desires adds to the chance of success for those trying to adopt Aquinas’s
approach. On the other hand, the second-order desire to will what God wills governs
the will’s first-order desires and so produces some unity in the will. The sufferer
manages at the same time to have a will of her own and to ally her will with God’s will
in her suffering.
In what follows, for ease of reference, I will call this complex first- and second-order
state in the will ‘willing what God wills in the Thomistic sense’, or just ‘the Thomistic
will’ for short. In the Thomistic will, there is a surrender of the self (because of the
second-order desire for willing what God wills even when it is contrary to one’s own
first-order will) but not an abandonment of the self (because of the persisting first-
order desire for what is lost in suffering). Furthermore, although the Thomistic will is
still an internally divided state, it does have an integrated character. The will is not at
war with itself, because the second-order volition rules the first-order desires. And
insofar as the integration is around willing what God wills because God and God’s will
is good, the integration is an integration around the good.
In suffering, then, a person can simply war against his suffering and rail against
the forces that produced it; but by doing so he will exacerbate his suffering with the
pains of frustration and anger or anxiety, and he will feel that he does. But, in the
nature of things, the aversiveness of suffering will move a person to try to find some
diminishing of the suffering. Under the pressure of his suffering, a person can try
Eckhart’s approach. He can try to give up what he cares about and instead find
acceptable whatever happens to him. But, as I have argued, this move is self-
destructive and likely to fail. On the other hand, if he foregoes Eckhart’s option,
suffering can move him in the direction of the Thomistic will. If he forms the
Thomistic will, then he will come to some degree of peace, because there is some
acceptance of his suffering, but without the sacrifice of his humanity, since he does
not give up what he cares about.
  

To the extent to which the sufferer finds a place of peace, there will be for him some
amelioration of his suffering; and this place of peace will come when he is willing what
God wills, in the Thomistic sense. In this condition, and to the extent to which the
sufferer wills what God wills, the sufferer is also integrated around the good.
So one thing suffering can do is move a person in the direction of peace, which
comes with that integration around the good that is possible to have in a post-Fall
world, as modeled by Christ’s prayer in Gethsemane.

9.4 Suffering and Atonement

No one should mistake these thoughts about suffering for a theodicy, even if they do
elucidate one claim central to some theodicies, namely, that suffering has a role to play
in remedying the post-Fall human proclivity to fragmentation in the will brought about
by sinfulness. A theodicy attempts to provide a justification for God’s allowing
suffering; and some good stemming from suffering is not by itself sufficient to justify
anyone’s inflicting or allowing that suffering. In any event, my concern here is not with
theodicy. It is rather with the exploration of a connection between the atonement and
the role a theodicy can assign suffering in the process of salvation.
The first thing to notice in this regard is that there are two different ways in which
any such connection might be drawn.
On the one hand, we could suppose that something about Christ’s passion and
death is the primary provider of the means to union with God, and we could ask how
the Thomistic will elicited by suffering contributes to the efficacy of Christ’s passion
and death. In this case, the role of suffering in evoking the Thomistic will would be
instrumental to the work of the atonement.
On the other hand, however, it is not out of the question, that is, it is not
incompatible with any orthodox Christian doctrine, that the atonement provides
its redemptive effects at least in part by synergistically enhancing suffering’s ability to
elicit the Thomistic will from a sufferer. Seen this way, for at least some part of the
process of salvation, the primary work of salvation would be done by human
suffering; and Christ’s passion and death would somehow enhance or enable that
work. In that case, for that part of the process of salvation, the atonement would be
instrumental to the eliciting of the Thomistic will through suffering.
In my view, with the elucidation of the Thomistic will given above, it is possible to
discern a progression in which both of these connections have a place.
9.4.1 The Thomistic Will and Operative Grace
To see this progression, it is important to focus on one component of the Thomistic
will, namely, the second-order desire for a will to will what God wills. Where does
this second-order will come from?
On the account I have been arguing for above, when a person comes to the point of
framing his will into the configuration of the Thomistic will, the aversiveness of
suffering has played a major role in getting him there. But, on orthodox Christian
views, this role of suffering cannot be the whole story of the genesis of the Thomistic
will. The second-order desire in the Thomistic will is a will to will what God wills;
and since what God wills is only the good, this second-order desire can also be
       

understood as the will to will the good. It is therefore a good state of will. But
orthodox Christian theology is committed to eschewing Pelagianism. That is, it is
committed to rejecting the view that there is anything good in a human will that is
not infused into it by God’s grace.
In fact, the second-order desire of the Thomistic will is a part of, or is subsumed
under, the volitional component of faith, in which a person longs for God’s goodness
and repudiates the evil in himself. As traditionally understood, faith is the necessary
and sufficient condition for justification. Justification is the beginning of moral and
spiritual renewal in a person, and it leads to salvation if only it continues to its end.
On traditional Christian views, as long as he continues to have this state of will, a
person will eventually come to permanent and everlasting union with God.53
Although the rejection of Pelagianism implies that this act of will is infused into a
person Paula by God’s operative grace, at least from Augustine to Aquinas the
Christian tradition has also held that the act of will in question is a free act on the
part of the person being justified.54 Operative grace justifies Paula by producing in
her the second-order volition of the Thomistic will, and yet that second-order
volition is also supposed to be an act of free will on Paula’s part.
The claim that this second-order volition is free and yet also infused in a person by
God has been the subject of great controversy. Elsewhere, I have argued that the free will
in question can and should be taken in a libertarian sense.55 In my view, the traditional
account of grace and (libertarian) free will can be understood consistently this way.56
God is always willing to give operative grace to every person, as long as she does
not reject it. If a person Paula rejects it, God cannot give her operative grace without
violating her will. On the other hand, since God is always prepared to give this grace,
all that is needed for Paula to receive it is her ceasing to resist God. If she ceases
resisting and surrenders to God’s love, God will produce in her the acceptance of his
grace and the second-order volition for a will that wills the good, that is to say, the
second-order desire for a will that wills what God wills.
In giving grace to a person Paula, therefore, God is doing all the work, as it were, of
producing the relevant second-order will of faith in Paula. But in producing it God is
responsive to Paula, who is ultimately responsible for whether or not she has it since
it is ultimately up to Paula alone whether she resists God’s grace or ceases resisting it.
And so it is right to say that Paula’s will is free in a libertarian sense even when the
second-order desire of the Thomistic will is produced in Paula by the infusion of
God’s operative grace. Since it is entirely up to Paula whether she rejects God’s grace
or ceases rejecting God’s grace, it is entirely up to Paula whether or not she has the
second-order volition in question.57

53
Obviously, on this view, there will have to be some connection between the infusion of God’s grace
into a person and the passion and death of Christ. There is some contribution to that story in what follows
here. For further discussions, see also Stump (forthcoming).
54
See Chapter 13 in Stump (2003).
55
See Stump (2001). See also Chapters 9 and 12 in Stump (2003).
56
For more discussion of this position, see Chapter 12 in Stump (2003).
57
This story is compatible with the rejection of Pelagianism insofar as neither the rejection of God’s
grace nor the bare cessation of rejection counts as a good act of will. The first state is an act of will but not
good; the second state does not even count as an act of will.
  

It is important to be clear that, just as God cannot infuse grace in Paula while she is
rejecting it, without violating her free will, so he cannot single-handedly produce in
Paula a cessation of the resistance to God. If God single-handedly overrides Paula’s
will in order to produce a new state of will in Paula, then no matter whether that state
is a new second-order will or just a cessation of an old state of will, the resulting
condition in Paula is God’s will operative in her and not Paula’s will. Consequently, if
God acts on Paula’s will while Paula is resisting God, there will subsequently not be
two wills to unite but only one, namely God’s, which is in God but also in Paula. And
so the union sought between God and Paula will be precluded, not promoted, if God
acts on Paula’s will while Paula is resisting God. It is therefore crucial to Paula’s
justification, and consequently to Paula’s salvation, that she herself cease rejection of
God and God’s grace.

9.4.2 The Thomistic Will and the Atonement: Justification

The fact that it is entirely up to Paula whether or not she ceases resisting God’s grace
does not mean that there is nothing God can to do to help elicit this ceasing in Paula.
The aversiveness of Paula’s suffering can be an important part of what moves her
towards the Thomistic will and the ceasing to resist grace that is requisite for it.
Insofar as God permits the suffering Paula experiences, God plays a role in the
eliciting of the Thomistic will in Paula. And no doubt many experiences in a person’s
life, in addition to suffering, can contribute to readying a person for the transforma-
tive experience of surrender to God’s grace. Except for cases such as that of Paul on
the road to Damascus, perhaps all instances of conversion are preceded by many
experiences that soften the resistance of a person to this surrender to God’s love. And
God’s providential ordering of all these circumstances in Paula’s life can contribute to
softening her heart.
But in order for a person finally to cease resisting God’s grace, the heart of a person
must not only soften but crack and melt. The notion of a heart’s cracking or melting
is, of course, a metaphor (just as the notion of softening a heart is). To speak of
something’s cracking or melting is to describe something’s giving way to an external
influence after (or in spite of) some internal resistance or disinclination to it. To say
that a heart cracks or melts, then, is to imply that a will that previously was resistant
or disinclined towards something urged on it by someone (or something) else gives
over its opposition and leaves off its resistance.58
It is not implausible to suppose that the passion and death of Christ can be a
catalyst in bringing a person to this state. If Paula has responded to her suffering by
moving in the direction of ceasing to resist God’s grace, reflection on the passion and
death of Christ can be the catalyst that enables her to go all the rest of the way to
abandoning resistance entirely.

58
I have described the will’s state in such a case as if it consisted in ceasing to do an action, rather than
as performing the action of ceasing, both because the description of the will as passive seems to me truer to
the phenomena and also because Aquinas’s philosophical psychology allows for this possibility, which is
the basis for the position I am arguing for. For further discussion of this issue, see Chapter 13 in Stump
(2003).
       

Christ’s willingness to die for human beings, in their post-Fall condition with all its
defects, shows Paula God’s great love for her. In the passion and death of Christ, God
is not manifesting wrath or rejection of human beings. He is not displaying his regal
character, his almighty power, or his role as dreadful judge. On the contrary, in the
incarnate Christ God is allowing himself to be put to death in a painful and shaming
way because of his love for human beings and his desire to bring them to himself. If
anything can help Paula all the way to the cessation of resisting God’s grace and
surrender to God’s love, it does seem that Christ’s passion and death could do so.
A display of power can prompt fear and submission; but it takes deep love to prompt
the kind of surrender to love that precedes operative grace.59
For these reasons, one connection between the atonement and the Thomistic will
elicited through suffering can be explained in this way. Before a person Paula is
justified and God’s grace infuses the second-order desire of the Thomistic will into
her, Paula has a resistance towards God.60 When Paula has been readied by past
experience of suffering (which is itself a result of God’s providential working in
Paula’s life), the passion and death of Christ can be the final means for eliciting
Paula’s abandonment of resistance. The internal opposition to undergoing the
wholesale changes and the humbling entailed by the second-order desire of the
Thomistic will can give way in the face of the suffering of Christ and the love it shows.
The consequent cascade of events leads to Paula’s justification. When Paula’s
response to the passion and death of Christ results in her surrender to God’s love,
Paula’s giving over resistance to God is followed by God’s giving Paula operative
grace. In turn, that operative grace infuses into Paula the second-order will in which a
person longs for God and God’s goodness. This second-order will is in effect a desire
to will what God wills. Its presence in a person, together with the accompanying
states of intellect, constitutes justification.
In this way, the passion and death of Christ can be instrumental in eliciting from a
person the act of will of faith, which subsumes the Thomistic will, and in enabling
that person to come to justification, which will lead to union with God if it persists to
the end of life.

9.4.3 The Atonement and the Gifts and Fruits

of the Holy Spirit: Sanctification
Once justification has occurred, however, the order of instrumentality is reversed. To
see this part of the progression, it is important to understand that the doctrine of the
Trinity, which is the hallmark of Christianity, plays a role in the doctrine of the
atonement too. On the doctrine of the Trinity, there is only one God but three divine

59
For the distinction between submission and surrender, see Stump (2010: Chapter 8).
60
In his book Nagel (1997) Thomas Nagel gives a vivid description of the phenomenology of such
resistance. Describing the fear of religion, Nagel says,
I speak from experience, being strongly subject to this fear myself: I want atheism to be
true. . . . It isn’t just that I don’t believe in God and, naturally, hope that I’m right in my
belief. It’s that I hope there is no God! I don’t want there to be a God; I don’t want the
universe to be like that.
Nagel (1997: 130–1)
  

persons, each of whom is the same God while nonetheless being distinct from one
another.61 For my purposes, what is important to understand with regard to this
complicated theological doctrine is that, just as there are multiple persons in the
Trinity, analogously, if one might put it this way, there are multiple persons in any
person Paula who has come to justification.
That is because once Paula ceases resisting God’s grace, God infuses into her not
only operative grace but also the third person of the Trinity, the Holy Spirit.62 The
Holy Spirit comes to be within Paula’s psyche with the infusion of operative grace
and stays there as long as Paula does not return to rejecting God’s grace.
In addition, when the Holy Spirit is within Paula’s psyche, the Spirit brings into
Paula the dispositions and orientations called ‘the gifts and fruits of the Holy Spirit’.
The gifts are infused dispositions that enhance excellence in Paula’s will and intel-
lect.63 The fruits are the orientations that result from those dispositions and the
indwelling Spirit; they include love, joy, peace, patience, long-suffering, and seven
others.
A full explanation of the nature of the infusing of the Holy Spirit and the
accompanying gifts and fruits is a long story,64 but the part of that story relevant
here is this.
It is part of orthodox Christian doctrine that, in his passion and death, Christ
somehow took on the sin of all human beings. There are different interpretations of
this claim in the history of Christian thought. But one way to interpret it is as a claim
that, in his passion and death, Christ opened himself up to the psyches of all other
human beings, all at once, so that he somehow received in himself, in psychic union,
the sinful, fragmented, guilty, and shamed psyches of all other human beings,
without himself actually becoming guilty of any particular sin.65
Now union is mutual. In union, one might say, each united person is somehow
inside the other with whom he is united. There cannot be union unless the indwelling
of love is mutual. So to say that, in his passion and death, Christ opened himself to
the sinful psyches of other human beings is in effect to say that, in his passion and
death, Christ did his part of what is needed for his union with every human being.
His part was to open his psyche up to their indwelling in him, so to say. If Christ had
not in this way done what was needed from him for union with human beings, then
human beings could not unite with Christ.
But since in his passion and death Christ did his part of the uniting by letting sinful
human psyches indwell him, what is missing for union between him and any other

61
The doctrine of the Trinity is, of course, the subject of considerable discussion and controversy,
which is not directly relevant to my concerns in this paper. For my purposes, I will just take the doctrine as
data for developing the connection between the atonement and the role of suffering in salvation.
62
For a detailed discussion of the notion of the indwelling of the third person of the Trinity in a person
in faith, see Stump (2013: 63–87).
63
There are three gifts for the will (pietas, fortitude, and fear of the Lord), and four for the intellect
(wisdom, understanding, counsel, and knowledge). For a discussion of the gifts and fruits of the Holy Spirit
and their place in Aquinas’s ethics, see Stump (2011: 29–43).
64
For a helpful discussion of the gifts and fruits of the Holy Spirit, see Pinsent (2012).
65
For more discussion of this interpretation, see Stump (2012b).
       

human person in this life is just that she allow Christ into her psyche. And for that
possibility to be realized for Paula, Paula herself has to be open to Christ.
On Trinitarian lore, however, the Holy Spirit and Christ are the same God. There
is only one mind and one will in the Holy Spirit and in the second person of the
Trinity, for example; and the second person of the Trinity is also the only person in
the incarnate Christ. One way to summarize this complicated Trinitarian lore is to
say that the spirit of Christ is the Holy Spirit. In his passion and death on the cross,
Christ opened himself up to Paula’s psyche. In justification, when operative grace is
infused into Paula, the spirit of Christ comes into Paula with that grace. So when
Paula surrenders to God’s love and in justification is open to the indwelling Holy
Spirit, the other half of what was needed for mutual union between Paula and Christ
is begun.66
The gifts of the Holy Spirit that are dispositions for the will then begin the slow
work of the sanctification of Paula. That is, they work to bring Paula’s will, bit by bit,
more in harmony with God’s will as regards her first-order willing. They add strength
to the part of Paula’s first-order will that is in accord with her second-order will to
will what God wills. They help Paula to harmony between her first-order and second-
order desire, without taking away from her those first-order desires for what she cares
about and loses in suffering. And the gifts of the Holy Spirit having to do with the
intellect add excellence to Paula’s mind, too. They help her mind to be in harmony
with God’s mind, so that she knows what the will of God is with regard to particular
actions and occasions.67
More important than the dispositions that are infused with the Holy Spirit,
however, is just the fact that Paula has Christ intimately present to her.68 In the
condition of justification, when the Holy Spirit comes to indwell her, Paula has some
direct and immediate second-personal experience of Christ and Christ’s love for her,
however attenuated or tacit the awareness of that experience might be. In fact, the
orientations that are the fruits of the Holy Spirit stem from this second-personal
experience: love, because Christ who loves Paula is present to her; peace, because in
being united with Christ in this way, Paula somehow has her deepest heart’s desire
stilled; and joy, because Paula’s union with the God who loves her is a joyful thing for
her. Patience and long-suffering, the next two orientations on the list, arise in
consequence. In the presence of a loving God, one can endure suffering, which is
still really suffering, without losing love, peace, and joy.
So when the Holy Spirit is infused in Paula, her union with God is begun, not only
in the minimal sense that she has a second-order will to will what God wills but in the
much stronger sense that, in the person of the Holy Spirit, God himself is present to
her, with her, and, in fact, somehow indwelling in her.
This incipient union is brought about in Paula first because of the passion of Christ
on the cross, in which Christ opens himself to union with all human beings, and then

66
For defense of the claim that union between persons comes in degrees, see Stump (2010: Chapter 6).
67
For more discussion of the kind of second-personal knowledge involved in this connection between
God’s mind and the mind of a human person in faith, see Stump (2014).
68
‘Present to’ is a complicated notion, hard to spell out just in passing. For a detailed discussion of it, see
Chapter 6 of Stump (2010).
  

because the effects on Paula of her suffering and of Christ’s suffering for her elicit
from her a giving over of resistance to God’s grace. If only Paula does not return to
resisting grace, this beginning of union will continue in the process of sanctification.
And if Paula does not give up before the process of sanctification is complete, then
sanctification will ultimately eventuate in full union between her and God. In the
fullness of union, the presence of God to Paula will be greatly more powerful, and
Paula’s will will be completely integrated around the goodness of God.
The gifts and fruits of the Holy Spirit will help significantly in moving Paula from
the incipient uniting with which the process begins to full union with God. But so will
the inner, spiritual presence of Christ to Paula. Sanctification is inevitably accom-
panied by suffering, if in no other way, then because it requires Paula’s struggle
against the first-order desires that she herself wants not to act on. The indwelling
Holy Spirit in Paula, making available to her the loving presence of Christ, will
strengthen her for not giving up in the process. She does not have to endure that or
any other suffering in her life alone. Christ is Emmanuel for her: God with her. And
so she has a powerful help against returning to her former resistance to God.
We can sum up this part of the connection between the atonement and the
Thomistic will this way. Once justification has occurred, then the union made
possible by Christ’s passion and death enables Paula to enter into an incipient
union with Christ by means of the Holy Spirit indwelling in Paula. And so, because
of Christ’s passion and death, the second-order volition of the Thomistic will that is
effected in Paula in justification results in the Holy Spirit’s indwelling in her. And
this, together with the gifts and fruits of the Holy Spirit, helps to bring about Paula’s
sanctification, which results in her full union with God.

9.5 Conclusion
On the way of thinking about the connection sketched here between the role of
suffering in salvation and the redemptive effects of Christ’s passion and death, on the
doctrine of the atonement, there is a progression in which the suffering of a person
Paula and the passion and death of Christ work together to bring about Paula’s
salvation.
First, Paula’s reaction to her own suffering and her response to the passion and
death of Christ can work together to elicit in Paula the cessation of resistance to
God’s love and grace. That surrender is followed by the formation in Paula of the
second-order desire of faith, which subsumes the Thomistic will. Because Christ’s
passion and death can help to prompt in Paula the giving over of resistance to God,
without which God cannot infuse either the second-order will necessary for justifi-
cation or the Holy Spirit, the passion and death of Christ can be instrumental in
bringing a person to justification. In this way, the atonement plays a significant role
in synergistically furthering the effects that suffering can provoke in a human person.
But, secondly, once justification has occurred, and God has infused in Paula both
operative grace and the Holy Spirit, the process of sanctification starts. The uniting
begun by Christ in his passion and death then does its work through the indwelling
Holy Spirit together with the infused gifts and fruits of the Holy Spirit. In sanctifi-
cation, Paula’s will remains free, but God increasingly integrates Paula’s first-order
       

desires around the good, without destroying Paula’s own will or taking away from her
what she herself cares about. In consequence, Paula comes more and more to will
what God wills; and the incipient union between Paula and God characteristic of
Paula when she is first justified increases too.
In this way, the second-order desire of the Thomistic will that suffering helps to
elicit is instrumental to making possible for Paula the effects of sanctification
provided by the passion and death of Christ.
Furthermore, the work of sanctification would cease if at any point in the process
Paula returned to rejecting grace. For sanctification to continue and reach its end of
full union between Paula and God, Paula must not revert to rejecting God’s grace.
That is, she must continue to maintain the second-order desire of the Thomistic will.
So there is a complicated progression that interweaves Paula’s suffering and
Christ’s. Provided only that she does not return to her old habits of resistance,
through Paula’s suffering and Christ’s woven together in Paula’s life history, Paula’s
sanctification will result in her willing what God wills completely, in full union with
God. There is a role for Paula’s suffering in the progression leading to her union with
God, even though it is Christ’s passion and death that makes that salvation available.
In a world in which there is the atonement but no suffering, the chances of a
person’s availing himself of the salvific benefits of the atonement seem very small.
Every part of this story raises questions and calls out for further explanation. But
I hope that this sketch is at least suggestive for a way in which that further
explanation might go.

Acknowledgments
I am grateful to John Keller for helpful comments on an earlier draft of this paper.

Bibliography
Eckhart, M. (1978), Mystic and Philosopher, translations with commentary by Schuermann,
R., Sermon ‘See What Love’ (Indiana University Press), 135–6.
Eckhart, M. (1981), The Essential Sermons, Commentaries, Treatises, and Defense, in Colledge,
E. and McGinn, B. (tr.), ‘The Book of Divine Consolation’ (Paulist Press), 215–16.
Eckhart, M. (1994), Selected Writings, tr. Oliver Davies, ‘The Talks of Instruction’ (Penguin
Books), 41–2.
McCord Adams, M. (2008), ‘Plantinga on “Felix Culpa”: Analysis and Critique’, in Faith and
Philosophy 25: 123–39.
Nagel, T. (1997), The Last Word (Oxford University Press).
Pinsent, A. (2012), The Second Person Perspective in Aquinas’s Ethics: Virtues and Gifts
(Routledge).
Plantinga, A. (2004), ‘Supralapsarianism, or “O Felix Culpa” ’, in Van Inwagen, P. (ed.),
Christian Faith and the Problem of Evil (Eerdmans Publishing Co).
Schuermann (1978), Sermon ‘Blessed are the Poor’, 215.
Stump, E. (1988), ‘Sanctification, Hardening of the Heart, and Frankfurt’s Concept of Free Will’,
in Journal of Philosophy 85: 395–420, reprinted in (2003) Perspectives on Moral Responsibility,
John Martin Fischer and Mark Ravizza (eds.) (Cornell University Press), 211–34.
  

Stump, E. (2001), ‘Augustine on Free Will’, in Kretzmann, N. and Stump, E. (eds.), The
Cambridge Companion to Augustine (Cambridge University Press), 124–47.
Stump, E. (2003), Aquinas (Routledge).
Stump, E. (2010), Wandering in Darkness: Narrative and the Problem of Suffering (Oxford
University Press).
Stump, E. (2011), ‘The Non-Aristotelian Character of Aquinas’s Ethics: Aquinas on the
Passions’, in Faith and Philosophy 28.1: 29–43.
Stump, E. (2012a), ‘The Nature of the Atonement’, in Clark, K. and Rea, M. (eds.), Reason,
Metaphysics, and Mind: New Essays on the Philosophy of Alvin Plantinga (Oxford University
Press), 128–44.
Stump, E. (2012b), ‘Atonement and the Cry of Dereliction from the Cross’, in European
Journal for Philosophy of Religion 4.1: 1–17.
Stump, E. (2013a), ‘Omnipresence, Indwelling, and the Second-Personal’, in European Journal
for Philosophy of Religion 5.4: 63–87.
Stump, E. (2013b), ‘Conversion, Atonement, and Love’, in I. U. Dalferth and M. Rodgers (eds.),
Conversion (Mohr Siebeck).
Stump, E. (2014), ‘Faith, Wisdom, and the Transmission of Knowledge through Testimony’, in
O’Connor, T. and Goins, L. G. (eds.), Religious Faith and Intellectual Virtue (Oxford
University Press).
Stump, E. (2016), “The Atonement and the Problem of Shame,” in Senor, T. and DePaul, M.
(eds.), Selected Papers in Honor of William P. Alston, special supplement to Journal of
Philosophical Research 41: 111–29.
Van Inwagen, P. (1988a), ‘The Magnitude, Duration, and Distribution of Evil: A Theodicy’, in
Philosophical Topics XVI, no. 2: 161–87.
Van Inwagen, P. (1988b), ‘The Place of Chance in a World Sustained by God’, in Morris, T. V.
(ed.), Divine and Human Action: Essays in the Metaphysics of Theism (Cornell University
Press), 211–35.
Van Inwagen, P. (1991), ‘The Problem of Evil, the Problem of Air, and the Problem of Silence’,
in Philosophical Perspectives, vol.5: Philosophy of Religion: 135–65.
Van Inwagen, P. (1996), ‘Reflections on the Essays of Draper, Gale, and Russell’, in Howard-
Snyder, D. (ed.), The Evidential Argument from Evil (Indiana University Press), 219–43.
Van Inwagen, P. (1997), ‘Probability and Evil’, in Van Inwagen, P. (ed.), The Possibility of
Resurrection and Other Essays in Christian Apologetic (Westview Press), 69–87.
Van Inwagen, P. (2000), ‘The Argument from Particular Horrendous Evils’, in Proceedings of
the American Catholic Philosophical Association (annual supplement to The American
Catholic Philosophical Quarterly) Vol 74: 65–80.
Van Inwagen, P. (2004), ‘The Argument from Evil’, in Christian Faith and the Problem of Evil
(Wm. Eerdmans and Co.), 55–73.
Van Inwagen, P. (2005), ‘The Problem of Evil’, in Wainwright, W. (ed.), The Oxford Handbook
of Philosophy of Religion (Oxford University Press), 188–219.
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
 10
Swing Vote
Frances Howard-Snyder

10.1 Introduction
There is a persistent puzzle about why we should vote, since our vote almost certainly
will not make a difference to the outcome of the election. This is sometimes expressed
as a prudential problem: how is it in my interest to vote, given that my vote won’t make
a difference to the satisfaction of my goal of getting a government that serves my
interest; and sometimes as a moral problem: how is it the case that I ought to vote,
given that my vote won’t make a difference to achieving any morally worthy aim such
as universal health care, lower taxes, fewer wars, etc.? The reason my vote won’t make a
difference, of course, is that there are so many people voting, and that there is a
vanishingly tiny likelihood that any election will be decided by one vote. Similar
concerns can be raised with respect to other issues. Consider, for example, the question
of whether to become a vegetarian or to continue buying and consuming meat. For
many, the considerable suffering of pigs, cows, and chickens in factory farms is a good
moral reason to become a vegetarian, not to help the animal that already suffered and
died but to help future animals.1 The thought is that by buying and eating factory-farmed
meat one is causally implicated in the suffering of some animals, and that, by refraining
from doing so, one can prevent some suffering. Hud Hudson, following R. G. Frey,
points out that restaurants’ and supermarkets’ purchasing of meat is not sufficiently fine-
tuned to an individual’s food choices to make this connection. My supermarket will
order the same number of steaks whether I buy one or not. My choice to become a
vegetarian will not protect any animals.2 It is hard to imagine that similar points do not
apply to global warming. Your driving an SUV or using a cardboard cup or a plastic bag
is part of a big destructive picture, but it surely won’t make a difference by itself to
whether the ice cap melts and polar bears become extinct. And finally, Peter van Inwagen
has pointed out that almost any prison sentence could be reduced by a day without
untoward effects. Hence, most sentencing authorities appear to be causing gratuitous
suffering—a charge that proponents of the argument from evil levy against God.3
In what follows I shall focus on the voting issue.
Assume that sometimes the outcome of an election makes a big difference to
things that matter. Suppose that we are engaged in such an election—between two

1
Or perhaps it provides us with reasons to buy our meat from farms that treat their animals humanely.
2 3
See Hudson (1993); Frey (1983). See van Inwagen (2006: 100–1).
  -

candidates, whom I will call “Superior” and “Inferior”. Superior’s election would be
much better for the average citizen and non-citizen (according to the true value
theory—whatever that is) than Inferior’s. This fact seems to be a reason to vote for
Superior—particularly for those who recognize Superior’s superiority. But there are
some puzzles as to why this is so, especially given the fact that for any particular
agent, there is a virtual certainty that that person’s vote will not make the difference
between Superior’s getting elected and not. Now, some philosophers say that there
are reasons to vote for Superior that have nothing to do with getting him elected—
e.g., expressing our preference for him, taking a stand, taking part in the political
process, being seen by others to take a part in the political process, adding to
Superior’s mandate if he does get elected, etc.4 But it seems commonsensical that
the main reason to vote for Superior has to do with getting him elected so that he can
bring his values and skills to bear in making our country and the world better.
In this paper, I consider in detail three attempts to answer the question of why we
should vote that accept this assumption.5 The first is from Alvin Goldman, who argues
that someone who votes for Superior causes his election even if he would have been
elected without her vote; the second is from Carolina Sartorio, who argues that the voter
does not cause the result of the election but is nevertheless morally responsible for it;
and the third is from Derek Parfit, who argues that the voter increases the likelihood of
Superior being elected and hence, that the expected utility of voting for him is best.
I shall point out problems for Goldman’s and Sartorio’s view, and suggest that Parfit,
with slight modification, is correct. In section 10.5, I draw connections between my
conclusion about voting and van Inwagen’s views about vagueness and evil.

10.2 Alvin Goldman

Alvin Goldman argues that we are morally responsible for the outcome of the
election because of our causal role in producing that outcome.6 He recognizes that
there is a degree of overdetermination here, but argues that there are good models of
causation according to which overdetermining “causes” still count as causes. One
such model is Mackie’s INUS condition—according to which C is a cause of E iff C is
a necessary part of a sufficient condition of E.7 To take a simple model, imagine a
group of 100 voters who collectively decide an election. Sixty of them vote for
Superior and forty vote for Inferior. Obviously no individual vote by any of the
sixty was necessary. If Sally, say, had refrained from voting, Superior would still have
been elected. Goldman argues that we can consider a subset of the sixty—a group of
fifty-one voters, including some agent we are concerned with (Sally) and argue that
the votes of that group constitute a sufficient condition of the outcome; and that
Sally’s contribution was a necessary part of that sufficient condition. The thought

4
Brennan and Lomasky (1989) elaborate the “expressivist” justification of voting.
5
I do not mean to suggest that these three exhaust the possible or plausible answers to this challenge. In
particular, I shall not discuss, but I do not assume to be false, a sort of Kantian or rule consequentialist
argument that claims that each of us ought to vote, because if everyone (or most people) voted the result
would be good and if no one (or almost no one) voted the result would be bad.
6 7
Goldman (1999). Mackie (1965).
   

behind this slightly paradoxical claim is that sets have their members essentially. So,
Sally’s vote was an essential part of that set. The same can be said for each of the other
fifty-nine voters who voted for Superior. Goldman is not sure whether Mackie
himself would have endorsed this interpretation of his INUS condition, but Goldman
argues that if he didn’t, he should have. Goldman then goes on to give another
account of causation that has a similar result.
A vectorial causal system is a system in which states and state-changes result from the interplay
among forces that can be represented as vectors. . . . In tug-of-war, forces are exerted on a rope
in opposite directions, and movements of the rope—and of participants clinging to the rope—
are the results of the sum of the vectorial forces. When an element in a vectorial causal system
moves in a given direction, this is because the sum of the forces on that element are positive in
that direction. This sum is computed from three kinds of forces: (1) forces that are positive in
the direction of the movement, (2) forces that are negative in the direction of movement, and
(3) forces that are zero in the direction of movement. Finally, when thinking about the
causation of a given movement, we think of each force as a contributing factor in the
production of the movement, each negative force as a counteracting, or resisting, factor in
the production of the movement.8

So, according to Goldman, the rationale for voting is this:

[C]itizens have good reasons to vote because they bear partial responsibility for the electoral
outcome. Even if the individual’s vote is not decisive for a given candidate’s victory, such a vote
can still qualify as a partial cause of that victory. So a voter can earn moral or quasi-moral
credit for an electoral outcome even if he is not a swing voter.9
So, the general principle here seems to be: one has a moral reason to be a partial cause
of a good outcome O even if one makes no difference to whether O occurs or how
likely O is. This sounds plausible. Applying it to some of the other cases we
mentioned earlier—the behavior of someone who eats a hamburger is a morally
wrong causal factor in the suffering of (future) cows, even if exactly the same number
of cows will die whether or not she eats the hamburger; the behavior of some owner
of a gas-guzzling SUV is a morally wrong causal factor in global warming, even if his
behavior makes no difference to the seriousness of the ultimate crisis; just as the
behavior of a member of the firing squad is a morally wrong causal factor in the death
of the political prisoner, even though the prisoner would have died anyway. If a
prisoner is punished for a day that is not strictly necessary to the goods of reforming
him or deterring others, that extra day may have a value in virtue of being a causal
factor in producing those goods. Similar points apply to suffering that God allows
that is not strictly necessary for the great goods God aims at (as long as there is no
precise minimum of suffering that will accomplish these).
In spite of its apparent plausibility, Goldman’s view has the following troubling
implication. Suppose you see someone drowning. If you jump in and save her, you
will do something good and get “credit” for doing so; whereas if you allow Tom to
jump in and save her, he rather than you will have done the good, and will get the

8 9
Goldman (1999: 210). Goldman (1999: 217).
  -

credit. So, you’d better push Tom out of the way and get to her first. That’s a slight
parody, but the point is that Goldman’s reason for voting is not so much the good of
Superior’s getting elected as it is the good of your being the cause of that good upshot
and your getting the credit for it.10
I concede that if it is a matter of indifference whether you or Tom saves the
drowning victim and Tom is a little hesitant, you may earn some credit if you are the
one to save her, since you were the one to take the risk and the trouble. This may in
turn give you some reason to do the saving. But if there is something to be gained
from Tom’s doing the saving (he is slightly better trained, or a slightly stronger
swimmer) and he is willing to do it, it seems perverse for you to insist that you be the
one to do it. Moreover, consider examples from Parfit. Suppose that a rescue team is
needed to save 100 miners. Four members of the team are needed. You volunteer to
be the fifth. You and your colleagues succeed in saving the 100 miners. It is natural to
infer from Goldman’s theory that each of the five-membered team deserves an equal
share of the credit here. So, you did something as good as saving twenty miners. But
suppose that, at the time you joined the team, you had the option of saving ten other
miners all by yourself (who wouldn’t otherwise be saved). It seems that that would
have been the better choice, and yet Goldman’s theory implies that it is not, since
credit for saving twenty people is greater than credit for saving ten.11 One response
here might be to say that Goldman has indeed given you a reason to help save the 100
miners; but that it is a defeasible reason and that it is defeated by the stronger moral
reason to save ten all by oneself.12 But if we vary the case so that the alternative rescue
operation would involve saving one person’s finger instead of ten people’s lives, it still
seems that that is the right choice to make. Parfit has another example where an agent
has the option of adding his pint of water to a large vat that is used to provide succor
to desperately thirsty and injured soldiers. If the agent adds his pint, he will share in
helping the soldiers, but the catch is that the vat is perfectly full, so the extra pint will
cause (another) pint of water to spill over the edge. In other words, he will causally
contribute, but his contribution will make no difference. In this case, it would
seem better for the agent to drink the water himself rather than to waste it in this
way. Here we have a case where whatever moral weight Goldman’s causing the
good contributes, it can be outweighed by something trivial—such as the agent’s
(non-severe) thirst.
These results of Goldman’s view also apply to the voting scenario. I am faced, let us
suppose, with a choice between taking a tiny share of the credit for Superior’s
election, on the one hand, and, allowing others the full share of the credit, on
the other hand, and adding more good to the world by grading papers, working at
a soup kitchen or playing with my children. If it would be better for the rescuer

10
It is not entirely clear how “agent-centered” Goldman’s view is. Do I have a reason (moral reason) to
see to it that I do the saving rather than you? Or do I have a moral reason to maximize the number of
people (possibly including myself) who are causally implicated in good upshots? If his view is the latter,
then it would imply that there is no reason for me to save the victim rather than Tom. It is also not clear
whether this view requires a robust and morally significant distinction between doing and allowing. Some
will say that I cause the life saving by letting Tom jump in and save the victim.
11 12
Parfit (1984: 67–8). Thanks to Ryan Wasserman for this suggestion.
   

to save ten miners by herself, it will usually be better for me to do something other
than vote. It seems that Goldman has not given us a moral reason to vote.

10.3 Carolina Sartorio

The second account of the moral reason to vote is from Carolina Sartorio who argues
that we can be morally responsible for x even though we didn’t cause x; in this case,
someone who votes for Superior is morally responsible for Superior’s election
although she didn’t cause it.13 So, Sartorio differs from Goldman in saying that
voters are not causally responsible for the outcomes of elections, but she agrees with
him that voters are morally responsible for those outcomes. Here’s her argument. She
takes as her target a widely accepted principle called “Moral Entails Causal”:
Moral Entails Causal: If an agent is responsible for an outcome, it is in virtue of
the fact that he caused it (some action or omission of his caused it).
She then gives a counterexample, called Two Buttons, to this principle:
There was an accidental leak of a dangerous chemical at a high-risk chemical plant, which is on
the verge of causing an explosion. The explosion will occur unless the room containing the
chemical is immediately sealed. Suppose that sealing the room requires that two buttons—call
them “A” and “B” —be depressed at the same time t (say, two seconds from now.) You and
I work at the plant, in different rooms, and we are in charge of accident prevention. Button A is
in my room, and button B is in yours. We don’t have the time to get in touch with each other to
find out what the other is going to do; however, we are both aware of what we are supposed to
do. As it turns out, each of us independently decides to keep reading his magazine instead of
depressing the button. The explosion ensues.14

In this case, Sartorio argues, I didn’t cause the explosion but I am responsible for it.
She thinks I didn’t cause it because of another case. This case, Two-Buttons-One-
Stuck, is exactly like the first, except in this case, there is no other agent in the other
room. Instead there is a mechanism which is supposed to press the button, but which
gets stuck. She thinks it is obvious that I don’t cause the explosion in the second case,
and concludes that we should treat the two cases the same vis-à-vis causation.
She thinks I am responsible for the explosion in Two Buttons, because if I am not
responsible, then this is a case of moral luck, and since our situations are symmet-
rical, if I am not responsible, then neither are you. But she finds it unacceptably
paradoxical that neither of us is morally responsible. She contrasts this with the case
of Two-Buttons-One-Stuck, which she does think is a case of moral luck.
Her argument is:
1. I am not a cause of the explosion in Two Buttons. (Because of analogy with
Two-Buttons-One-Stuck.)

13
Sartorio (2004). Sartorio uses the word “responsible” as shorthand for “morally responsible.” I shall
sometimes do the same in what follows.
14
Sartorio (2004: 317).
  -

2. I am at least partly morally responsible for the explosion in Two Buttons

(otherwise we have an objectionable instance of moral luck).
3. So, Moral Entails Causal is false.
Sartorio connects her discussion of Two Buttons with the voting issue in a conclud-
ing paragraph of the paper. She writes, “On this view, if a citizen abstains and the
outcome of the election is bad, he is responsible for the outcome, not because he
causes it, but because he is responsible for something that causes it. If so, the view that
causation is the vehicle of responsibility could help us solve the voting problem.”15
Presumably, Sartorio’s account would apply mutatis mutandis to each meat-eater’s
responsibility for the suffering of animals in factory farms and to each SUV-owner’s
responsibility for global warming.
Let’s consider Sartorio’s moral luck argument for (2). Why is it so paradoxical that
neither you nor I are morally responsible for the explosion? Why would this be an
objectionable case of moral luck? Perhaps the thought is that I would have been
responsible for the explosion if I had failed to depress my button while you did depress
yours. If I am not responsible in the case where you also failed to depress your button,
then some factor totally outside of my control and my awareness makes the difference
between my being responsible and not. This is somewhat odd, since she thinks I am
responsible in Two Buttons but not in Two-Buttons-One-Stuck, and the difference
between these two scenarios is also (or could be) totally outside of my control and my
knowledge.
Perhaps the worry is that if neither of us is responsible, then nobody has acted
wrongly, and there is nobody to be “held responsible,” which means, nobody can
appropriately be blamed, or punished or required to pay compensation or to fix the
problem. But that seems false. One response is to say that although no individual is
responsible, we—the collective of you and me—is/are responsible.16 There still
remains a big question of what follows for individual members of such a collective.
If we are responsible, but I am not, why should I be blamed or punished? And,
although we can be blamed, it is not as though “we” can be punished or made to pay
compensation without me or you being punished or made to pay.17 A different
response would be to shift from talking about responsibility for the explosion (an
event or state of affairs) to responsibility for one’s own behavior. We can say, “You
are not responsible for the explosion. But you are responsible for your own behavior,
and that behavior was wrong and deserves blame and punishment.” You are respon-
sible for your behavior because it was freely chosen, you could have done otherwise,
and it was part and parcel of a set of behaviors that led to something very
bad. Perhaps Sartorio will raise a question here about how the behavior can be
wrong or blameworthy if it or its agent is not responsible for the bad upshot.
There are certainly moral theories that have an answer to this question. For example,
a subjective consequentialist might argue that I acted wrongly in doing what
I believed would cause the explosion, although I didn’t in fact cause the explosion
(and wasn’t responsible for the explosion). A rule consequentialist or Kantian would

15 16
Sartorio (2004: 332). Thanks to Hud Hudson for this suggestion.
17
There’s obviously a huge literature on this topic. I don’t mean to be taking a stand on it here.
   

argue that I acted wrongly in doing an act which was such that if everyone did the
same, a bad result would occur. A rule consequentialist or Kantian might deny (or
refuse to agree with the claim) that I am responsible for the explosion. So, much of
the basis for our intuitive agreement with her judgment here can be salvaged even if
we deny that I am morally responsible for the explosion.
I find it odd that Sartorio treats Two Buttons and Two-Buttons-One-Stuck so
differently vis-à-vis moral responsibility. In each case, I (our agent) act exactly the
same. Moreover, the result in each case is exactly the same and there’s nothing I can
do about it. Why say that the agent is morally responsible in one case but not the
other? I don’t share Sartorio’s intuition that there is such a difference, but I’d like to
consider variations on the two cases to see which feature that distinguishes them is
the one that is making the moral difference—at least for those who do agree with
Sartorio. Suppose we vary Two Buttons so that the button in the other room is
controlled by a human being who is a sort of automaton who is programmed to fail to
depress the button. We can also consider a variation in which the person in the other
room is a responsible agent, but at the time when I have to decide whether to depress
A, it is inevitable that B’s button-pusher has already failed to depress his—e.g.,
suppose at B needs to be depressed by T and A need to be depressed by T+1 minute,
and T has now passed. (Call these cases Two-Buttons-Deterministic I and II.)
The first of these seems indistinguishable from Two-Buttons-One-Stuck, since the
human being in the other room has ceased to be an agent in any interesting sense.
The fact that the mechanism for depressing the button involves a human being in the
one case and mechanical mechanism in the other doesn’t seem to make a difference.
The second of these looks to my eye like a case of preemption—like a case of shooting
a corpse. When I fail to depress my button, the explosion has already been set in
motion. Notice, that in this case, unlike in Sartorio’s, there is an asymmetry between
the two human agents. One agent is morally responsible for the bad upshot and the
other is the beneficiary of moral luck.
Now let’s vary the Two-Buttons-One-Stuck case. Suppose that the mechanism is
faulty. There is a 50 percent chance (somehow indeterministically realized) that the
button will stick. And let’s suppose that it does in fact get stuck. (Call this Two-
Buttons-One-Stuck-Indeterministic.) Note that at the time the agent has to decide
whether to depress her button, it is objectively undetermined whether the other
button will stick. In that case, by refraining from depressing her button, she has made
it certain that the explosion would occur; whereas if she had depressed her button,
there would have been a 50 percent chance that the explosion would be averted. In
other words, her behavior increased the likelihood of the explosion. This seems to me
to give her some responsibility for the explosion.
In our original understanding of Two-Buttons, perhaps we really think of it along
these lines. The other agent is like an indeterministic mechanism. He might or he
might not depress his button.
Could this be the explanation of the apparent difference between Two Buttons and
Two-Buttons-One-Stuck as Sartorio originally presents them? If so, then perhaps we
don’t have a genuine counterexample to Moral Entails Causal. In the deterministic
cases (Two-Buttons-One-Stuck and Two-Buttons-Deterministic I and II) I have
granted that there is no causal responsibility of the part of our agent, but I have
  -

argued that there is no moral responsibility in those cases either. With respect to the
indeterministic cases (in which I am tentatively including Two Buttons) I have
argued that our agent’s behavior is morally responsible for the explosion, but
I would also suggest that her behavior is also somewhat causally responsible. Pre-
sumably, my behavior is a causal factor of some upshot if it increases the likelihood
that the upshot will occur (and other conditions are in place).18
Perhaps Sartorio will object that even where it is certain that the other agent will
not depress his button (deterministic cases) I am responsible—because of her moral
luck argument. Here’s a reason to disagree with Sartorio on this point. Consider the
following principle:
PPP2: A person is morally responsible for a certain state of affairs only if (that
state of affairs obtains and) he could have prevented it from obtaining.
This principle, related both to the Principle of Alternate Possibilities and to “Ought”
Implies “Can”, has considerable appeal. Peter van Inwagen has ably defended it
against “Frankfurt-style” counterexamples.19
PPP2 implies that I am not morally responsible for the explosion in the deterministic
cases since I could not have prevented it; similarly, the other agent is not responsible
since he could not have prevented it. This powerful principle shouldn’t be given up so
lightly. (Especially since, as I suggested above, we can draw the sting of the moral luck
argument by pointing out that on some moral theories the agent may be responsible for
her behavior or inaction, in spite of not being morally responsible for the explosion.)
At the same time, someone else might object that PPP2 implies that I am not
morally responsible for the explosion in the two indeterministic cases described above.
I disagree. In Two-Buttons, and Two-Buttons-One-Stuck-Indeterministic, there is
something I could have done (namely depress the button) which is such that had
I done it it would have been only 50 percent likely that the explosion would occur. There
are two ways to proceed from here. First, we can distinguish two states of affairs: the
explosion’s occurring, and the explosion’s being certain to occur; and say that PPP2
implies that, although I am not responsible for the first, I am responsible for the second.
Alternatively, I can argue that I am responsible for the explosion (the explosion’s
occurring) since there is something I could have done which is such that had I done it,
it is not the case that the explosion would have occurred. This is not to say that there is
something I could have done which is such that if I had done it, the explosion would
not have occurred; i.e., both the “would” and the “would not” counterfactuals are false
in that scenario. If I had not done it, the explosion might have occurred. The important
point is that I have the power to make the relevant “would” counterfactual false.20 This
seems to be a slight variation on PPP2 but one that may well be close enough:

18
As long as other necessary conditions for causation obtain. Raising the probability isn’t enough if the
effect is preempted by a different causal chain which is itself indeterministic.
19
Van Inwagen (1978).
20
This follows from David Lewis’s semantics for counterfactuals. For more on this and its implications
for morality, see the discussion of MOE below.
   

PPP2*: A person P is morally responsible for a certain state of affairs S only if

S obtains and P could have behaved in such a way that if P had so behaved, it is not
the case that S would have obtained.
Either way, we can tell a story about why I am morally responsible for something in
the vicinity. I conclude that Sartorio has not succeeded in separating moral from
causal responsibility in the way she has attempted. What, if anything, follows about
voting? Well, presumably, if my voting or not voting makes a difference to the
relevant “would” counterfactuals, then I am (partially) morally responsible for the
outcome of the election. Let’s turn to a version of that idea.

10.4 Derek Parfit

Derek Parfit has a charmingly simple and elegant solution to the voting paradox.
If I vote, there may be a very small chance that my vote will make a difference. On one estimate,
if I am voting in one of the large and marginal states, which might go either way, the chance
that I shall make a difference would be about one in a hundred million. . . .
. . . Suppose that, if the next President is Superior, this will on average benefit Americans. . . . If
he is the one who is elected, average net benefit to Americans is this total sum divided by the
number of Americans. . . . If my vote has a chance of one in a hundred million of affecting the
result, the expected benefit of my voting is as shown below.
The average net benefit to Americans from Superior’s election x
The number of Americans
-————————————————————————-– – the costs to me and others of my voting
One hundred million
Since there are two hundred million Americans, this sum is likely to be positive. This will be so
if Superior’s election would on average bring to Americans a net benefit more than half as great
as the costs of my voting.21

A one in a hundred million chance of winning $500 for 200 million people has an
expected value of $1,000—much more than the cost of voting.22 For the act
consequentialist, who is required to act so as to benefit everyone, this seems
decisive. The idea is that there is some non-zero chance of your vote making the
difference. Multiply the difference (made by the election of the superior candidate)
by the small probability that your vote will make the difference, to determine the
expected utility of your voting. It will turn out to be greater than the expected utility
of not voting. Similar points can be made about the expected utility of refusing
to buy or eat meat, and the expected utility of recycling, and otherwise conserving
energy.
Parfit is here talking about “expected utility” or “expected value.” There are several
different ways in which expected value can factor into a decision about what we

21
Parfit (1984: 73–4).
22
Putting the gain to each citizen in terms of money is not meant to be taken literally. The gain to
citizens may be something more intangible and less obviously quantifiable, like justice, cleaner air, etc.
  -

ought to do or what is the right thing to do, even for the act consequentialist. For a
subjective consequentialist, the right act is the one with the greatest expected value.
For an objective consequentialist, the right act is the one with the greatest actual
value. For the objective consequentialist, expected value may be a factor to consider
as part of the best decision procedure for determining what is right (although this
seems to be a contingent and somewhat doubtful claim). Some philosophers, includ-
ing Parfit, hint that “right” and “wrong” and “ought” are ambiguous. We simply have
two different notions: objective rightness, wrongness, ought; and subjective rightness,
wrongness, ought. They will then say that the objectively right action is the one that
produces the greatest actual value, while the subjectively right action is the one that
produces the greatest expected value. In Reasons and Persons, Parfit describes (and
seems to endorse) a slight variation on this idea:
(C2) What each of us ought to do is whatever would make the outcome best, and
(C3) If someone does what he believes will make the outcome worse, he is acting
wrongly.23
Since an agent can have a false belief about what will make the outcome best, this
seems to imply that a single action can be both wrong and what one ought to do. He
continues by suggesting that wrongness is a matter of being such that we ought to be
blamed or punished. So, one way of making sense of Parfit here is to say that real
rightness and wrongness is a matter of actual consequences; while appropriate blame,
reward, and punishment is a matter of expected consequences.
If that’s right, then the voter in the large and marginal state who votes but then
discovers that her vote wasn’t needed, should think, “Gee, I thought I had an obligation
to vote. . . . But it turns out I didn’t, because all these other people voted. More than
enough of them supported Superior. I could and should have spent my time doing
something else instead of voting. I guess I’ve been guilty of blameless wrongdoing . . .
again.” On this view, the odds of doing wrong are very high if she votes. If Parfit’s
calculations are correct, then the odds of her doing wrong are 100 million to one, and a
reasonably well informed voter will know this. This is not to say that she was acting
irrationally or irresponsibly when she voted, of course, but nevertheless, it seems a bit
paradoxical to say that every voter who ever voted in a presidential election has not
acted as he or she ought. My own view is that this and other paradoxical implications of
objective consequentialism are reasons for someone attracted to the spirit of conse-
quentialism to embrace a subjective variation on consequentialism.24 However, I’d like
to explore a variation of objective consequentialism.
Suppose exactly 51, 475, 322 people (including me) vote for Superior, while exactly
51, 299, 941 people vote for Inferior. That’s a big gap—it looks like a 175, 381 margin
of error! That’s why my vote doesn’t seem to make a difference, and why my time
would have been better spent doing something else. But to call it a margin of error
assumes that if I had voted differently, everyone else would still have voted exactly the
same. In other words, that MOE is true.

23 24
Ibid. p. 24. See Howard-Snyder (2005).
   

MOE: If I had not voted for Superior, 51, 475, 321 other people would still have
voted for him.
Is MOE true? Standard accounts of counterfactuals tell us that a counterfactual, “If A,
then C would happen” (A>C) is true iff C is true in all the closest A-worlds.25 If A is
false, there will typically be several equally closest A-worlds. It will often be the case
that some of the closest A-worlds will be C-worlds and some will be not-C worlds.
A simple example to illustrate this involves an indeterministic coin toss. Suppose
I decline an offer to participate in some coin-tossing game. I might nevertheless
speculate about what would have happened if I had tossed the coin.
(H) If I had tossed the coin, it would have landed heads.
(T) If I had tossed the coin, it would have landed tails.
(H) and (T) have false antecedents. Given that the coin toss would have been decided
by an indeterministic mechanism, it seems that the coin might have landed heads and
it might have landed tails. Moreover, it seems (insofar as we understand closeness)
that some worlds where the coin lands heads are as close as some worlds where the
coin lands tails. It is not the case that there is a band of tails-worlds that is closer than
all heads-worlds. Which counterfactuals this phenomenon—the failure of “counter-
factual excluded middle”—affects depends on the correct theory of closeness of
worlds. There is a lot of controversy about this issue. Closeness of worlds is matter
of similarity, but do all similarities count? Do some similarities count for more than
others? To get truth conditions for counterfactuals that match our linguistic intu-
itions about counterfactuals, the answers seem to be, “No, it is not the case that all
similarities count,” and “Yes, some similarities count more than others.” Different
theorists disagree about how to prioritize different kinds of similarities. Are two
worlds with the same laws but radically different pasts more or less close than two
worlds with more or less exactly similar pasts and slight variations of the laws?
Jonathan Bennett and David Lewis say that the members of the latter pair are closer.
I find their arguments compelling, but will not rehearse them here. If that’s right,
then what determines the truth value of a counterfactual of the form A>C, where A is
false, is a band of A-worlds exactly like the actual world up to (or up to shortly
before) the time when C would be realized or not. If the actual world is indetermin-
istic in the appropriate way, these A-worlds may veer off from the actual world at
some node of indeterminacy in such a way that the A-worlds share all the laws of the
actual world. If the actual world is not indeterministic in the appropriate way, a small
miracle (relative to the actual laws) may be required to secure the veering off.

25
David Lewis (1973) writes, “L: A > C is non-vacuously true at I iff C is true at all the closest accessible
A-worlds to I”, and Jonathan Bennett writes, “All the analytic theories following the lead of Stalnaker and
Lewis have at their heart something like this:
A>C is true iff C obtains at every world of some class W of A-worlds such that every
member of W is closer to the actual world than is any A-world not in W” (Bennett 2003:
169).
Bennett puts it this way to accommodate cases where there is no closest A-world or A-worlds, because for
every such world there is a closer one.
  -

Now imagine that I do toss the coin and it lands heads. Here are some new
counterfactuals to consider:
(A) If I had tossed the coin, it would have landed heads.
(B) If I had tossed the coin with my left hand, it would have landed heads.
(C) If I had tossed the coin while you whistled, it would have landed heads.
(D) If I had tossed the coin while you made a bet in the other room, it would have
landed heads.
The antecedent of (A) is true; while the antecedents of (B)–(D) are false.
There is some disagreement amongst philosophers about the truth values of these
counterfactuals. For reasons similar to those I raised in connection with (H) and (T),
there is good reason to say that (B) is false; as is the counterfactual (B*): If I had
tossed the coin with my left hand, it would not have landed heads. This seems fairly
widely agreed to.
If one says that (A) is false, i.e., that A&C does not entail A>C, that assumes that
not all similarities count towards closeness, since no world can be as similar to the
actual world as the actual world is to itself. But if only some respects of similarity
count—e.g., if similarities with respect to the past count but similarities with respect
to the future do not—then some other world can be as close or as similar in those
respects to the actual world as the actual world is to itself.
Moreover, especially if the coin toss is indeterministic, it seems natural to interpret
A as requiring an imaginative retossing of the coin in worlds with the same past and
laws of nature as the actual world. In some of those worlds, presumably, the coin
lands tails. So, (A) is false. We get a similar result if we apply the same thought
experiment to (C) and (D). To assess the truth of (C) and (D) we imaginatively redo
the antecedent—adding whistling in one case, and a bet in the other. But, as with A, if
we redo the toss, the coin toss might end up heads and it might end up tails.26
It has been argued that these issues create difficulties for a version of consequen-
tialism that tells us to produce the best actual consequences.27 Even if we ignore the
word “actual” in the last sentence, notice that this view says that an act A is right if
and only if its consequences are or will be better than the consequences of all
alternative actions; i.e., if faced with a choice between actions A and B, our choice

26
Jonathan Bennett argues that (A) is false but that (C) and (D) are true. See Bennett (2003: 234–42).
27
Dennis Whitcomb raised a question about whether consequentialism is standardly interpreted in
terms of counterfactuals. My impression is that standard formulations do not always use counterfactual
terminology. Instead, they say things like, “Act consequentialism is the claim that an act is morally right if
and only if that act maximizes the net good for all, i.e., if and only if the net good for all of that act is greater
than this net amount for any incompatible act available to the agent on that occasion,” Walter Sinnott-
Armstrong, “Consequentialism,” Stanford Encyclopedia of Philosophy (Winter 2015 Edition). The phrase
“the net good for an alternative act, A” seems to refer to the good which would obtain if A were performed,
i.e., it seems to call for a counterfactual analysis. Fred Feldman has a slightly different way of interpreting
this. On his view, “a person, S, is obligated to see to the truth of a proposition, P, at a time, t, just in case
S sees to the truth of P in all of the best accessible worlds to S at t.” See Feldman (1986: Chapter 2).
Depending on how the concept of “accessibility” is interpreted—is a world where I roll a die and get a six
accessible to me—this view leads to different problems or inherits the problems counterfactual versions
have. See Vessel (2007) for more details.
   

of A is right if and only if it has better consequences than B would have. But
sometimes, as we have seen, the counterfactuals,
I do B> the consequences are better than those of A;
and
I do B > the consequences are worse than those of A
are both false or indeterminate, because some of the closest worlds where I do B are
better than those where I do A, and some of the closest worlds where I do B are as
good or worse.
This creates problems for objective consequentialism, because it seems that that
moral theory must say that there is no determinate moral fact here, or that every
option is permissible in such cases. Both options are unsatisfactory.28 Cases like this
arise fairly often. Consider an agent whose brakes fail as his car is descending a steep
hill. At the bottom of the hill is a busy intersection. His traffic light has just turned
red. Since he cannot stop or slow down, he must choose between continuing through
the intersection, on the one hand, and turning his car into a field adjacent to the road.
He takes the latter course. Unfortunately, he hits and kills a hiker who is sleeping
undetected under a bush in the field. Awful as this result is, if it is the least bad of the
options, then it is justified. But whether it is the least bad depends on what would
have happened if he had entered the intersection. But what would have happened had
he entered the intersection may well be subject to the indeterminacy we have been
discussing. Suppose there was another car containing a family of five entering the
intersection at right angles to where the agent’s car would have entered it. Maybe the
driver would have braked in time; maybe she wouldn’t have. It’s not clear that there is
a fact of the matter about this. (We don’t know . . . but the point that I am making is
that in some such cases, even an omniscient observer would have no answer to the
question of what would have happened, because there is no such answer.) Or
consider the homeless man with a sign that read “Moore Muny Four Fud,” who
I passed on my way to school this morning. If I had given him money, he might have
made better use of it than I would; but he might have made worse use of it.
Now, if we recognize that the consequences of an action continue far into the
future, and we recognize that there will be many future moments where the conse-
quences of my action will depend on how others choose to respond to it, the problem
is amplified dramatically.29 This would mean that most of our actions (including
some that seem clearly objectionably risky) would be permissible.
Now, let’s return to our questions about voting. Recall MOE, which seemed to be
the basis for an objective consequentialist to say that it wasn’t true that I ought to vote
(although perhaps I am blameworthy, etc. if I don’t):

28
I developed this argument in Howard-Snyder (2005). I subsequently discovered that Jean-Paul Vessel
(2003) had beaten me to it.
29
Jean-Paul Vessel argues persuasively that we get similar results with respect to this pair of counter-
factuals even if the world is (or we suppose it to be) deterministic. See Vessel (2007).
  -

MOE: If I had not voted for Superior, 51, 475, 321 others would still have voted
for him.
MOE presumably entails:
BETTER: If I had not voted for Superior (but had instead used the day earning
utility points somewhere else) the consequences of doing so would have been better.
Applying some of the standard machinery that we have just discussed, to determine
the truth value of MOE, we need to backtrack to a moment fairly shortly before my
vote, erase my vote with minimal or no changes to the laws of nature, and allow the
laws of nature to operate normally on the changed world. But notice, now, that just as
the coin toss could turn out differently if the coin is retossed, so other voters could
pull a different lever if asked to rethink the voting question, not all of them,
obviously, but maybe a percent or two of them. There is no guarantee that those
51, 475, 321 voters vote the same way. So, it seems that there are some closest I-don’t-
vote-worlds where fewer, even 176,000 fewer, voters vote for Superior. So, the
following counterfactual is not true:
If I had not voted for Superior, Superior would still have been elected.
This story relies on certain controversial assumptions about indeterminism and
closeness of worlds. Here’s a slightly different story that gets a similar result. Suppose
that voters’ voting behavior is determined. People vote because of their beliefs and
desires, these beliefs and desires are caused in the standard ways by environment,
genetics, experiences, reasoning, etc. In that case, when we consider a counterfactual
with the antecedent, “If I had not voted, . . . ” we need to consider worlds like the
actual world except that I don’t vote because some recent small miracle changed my
mind. It seems unlikely that there is a unique opportunity for such a miracle, but
leaving that potential source of indeterminacy aside, consider this one. Imagine a
world exactly like ours except that, in the presidential debate two weeks before the
election, Superior was caught on camera rolling his eyes at some inane remark made
by Inferior. This disturbed me. “Maybe Superior is not a true man of the people,”
I thought. “I still prefer his policies, but I’m not sure I like him personally. I think I’ll
refrain from voting for him.” If this small miracle changes my mind, it will likely
change other minds too. It is likely that many of my fellow voters will see the eye
rolling (maybe replayed on YouTube, CNN, or Fox) and feel and respond the same
way I did. In that case, MOE is false, and BETTER may well be false too.30
If all these counterfactuals are false, is there any hope for objective consequentialism?
Vessel makes an attractive suggestion when discussing a related issue—to embrace a
probabilistic version of objective consequentialism, modeled on the sort of decision
theoretic logic used by those who favor expected utility as the determinant of right

30
I don’t mean to suggest that either all votes are decided indeterministically at the last minute or that
all votes are determined by factors present in the world prior to all voters’ existence. Very likely some votes
are determined and some are undetermined, and those that are determined are settled at various different
times. Probably most voters go into the booth with their minds made up. These extra complications
shouldn’t affect the main point of my argument, however.
   

action, but replace subjective probabilities with objective probabilities, which are
determined by proportions of the set of closest A-worlds: e.g., it is more probable that
I roll two dice>I get a seven
than
I roll two dice>I get a twelve
because a higher proportion of the closest two-dice-rolling worlds are worlds where
I get a seven than worlds where I get a twelve. This in turn would have implications
for the rationality or morality of taking various bets on the results of the coin toss.
How about MOE? Well, consider the probabilities of
I don’t vote>enough voters vote for Superior to get him elected
and
I do vote>enough voters vote for Superior to get him elected.
I have argued that these are both false but it seems plausible that the probability of the
second is slightly higher than the probability of the first. If that’s right, then, even
those of an objectivist bent can accept a version of Parfit’s idea that by voting I raised
the probability of Superior’s being elected. Similarly, they can argue that by refraining
from buying and eating meat, I lowered the probability of some cows (maybe a whole
bunch or herd of cows) will suffer and die.

10.5 Application to van Inwagen on

the Problem of Evil
What Parfit says about a single voter (against the backdrop of millions of other
voters) might provide a useful analogue for the case of a single choice by an agent
against the backdrop of thousands or even millions of other choices she might make.
Peter van Inwagen writes “Suppose you are an official who has the power to release
anyone from prison at any time . . . Let’s suppose that the only good that results from
someone’s being in prison is the deterrence of crime. . . . Obviously 9 years and 364
days spent in prison is not going to have a significantly different power to deter
felonious assault than 10 years in prison.”31
So, one might think, if you have an obligation to eliminate gratuitous punishment,
you should release the prisoner a day early. But of course, the same is true of any n
and n 1 days in prison. One day doesn’t make a significant difference. But this leads
to the absurd result that no one should be sent to prison for any reason.
This example is, of course, supposed to model the situation of God vis-à-vis
gratuitous horrors—instances of horrific suffering—each of which is not necessary
to achieve God’s purposes. If God were to eliminate this particular evil, the same sorts
of consideration should lead him to eliminate the next and the next until there is no
evil. But, van Inwagen supposes, in that case, God’s purposes would not be served.

31
See van Inwagen (2006: 100–1).
  -

This story seems to involve various counterfactual claims analogous to those we

discussed in connection with voting:
If I hadn’t voted for Superior, the result of the election would be the same.
If the prisoner had been released a day early, the deterrent effects would be the same.
If God had prevented this instance of suffering, the result vis-à-vis His purposes
would have been the same.

But perhaps there is a degree of counterfactual indeterminacy about the realization

of God’s purposes; so that the last would-counterfactual is as questionable as the first.
Arguably, an extra horror increases the probability of God’s purposes being achieved.
Even if the increase is tiny, it can be enough to make the addition of the extra horror
have an expected utility at least as high as its alternative. (Note that van Inwagen says
it is “not significant”; he does not say it is non-existent.) There may be an element of
incommensurability here. The values at stake—minimizing suffering and realizing
the great goods of God’s purposes—may be such that fine-grained precise compari-
sons are impossible in borderline cases. However, the counterfactual point should be
enough to ensure that the choice to allow an extra instance of suffering in not
determinately worse than the alternative.

10.6 Concluding Remarks

In closing I’d like to return briefly to a discussion of Goldman and Sartorio. It may be
that Goldman’s idea is very similar to my revised version of Parfit. The fact that one is
causing some good state of affairs is generally regarded as a moral reason to act
because usually it is understood that the good state of affairs will not exist if one does
not act; or, at least, that it will be less likely without our causal input. So, for example,
we may add an extra member to the tug-of-war team “just to make sure.” Or we may
add a fifth member to the rescue team because of various uncertainties (or maybe
even objective indeterminacies) that make it unwise to get by with the bare min-
imum. If this is Goldman’s thought, then I have no quarrel with him. However, we
might consider cases where an agent’s behavior causes an upshot while lowering
the probability of that upshot. For example, there was an outbreak of smallpox in the
1770s in New England. The chances of catching the disease and dying from it were
(let’s say) between 20 percent and 30 percent. Now, some bold folk (including John
Adams’s wife) chose to have themselves inoculated with a weaker strain of smallpox.
This action lowered their chances of dying to, say, between 0.5 percent and 2 percent.
Obviously some people who were inoculated caused their own deaths in this way. But
it doesn’t seem that they acted wrongly or irrationally in doing so. It wouldn’t be a
good argument to say to one of them, “Sure, you’re improving your chances of
survival, but if you die because of the inoculation, you will be the cause of your own
death. So, that’s a reason not to do it.” This suggests that something similar to Parfit’s
line of thought is what makes Goldman’s view seem plausible.32

32
Thanks to Dennis Whitcomb for the ideas of this paragraph.
   

In my discussion of Sartorio, I endorsed the following principle:

PPP2*: A person is morally responsible for a certain state of affairs only if (that
state of affairs obtains and) he could have so behaved that it would not be the case
that if he had so behaved, the state of affairs would have occurred.
This represents only a necessary condition on moral responsibility. It doesn’t,
moreover, say anything about degrees of responsibility, or anything about how
facts about responsibility translate into facts about what one ought to do. If my
discussion in the last section of this paper is roughly correct, then consistently with
PPP2*, we can say that I am morally responsible for the outcome of the election
because that outcome obtains, and I could have so behaved (not voted) as to make the
following counterfactual false:
If I had not voted, the outcome would still have obtained.
An extension of this principle that will be attractive to some consequentialists is to
say that my degree of responsibility for the outcome of the election (and hence, the
strength of the moral reason for me to vote) is proportional to the difference I make
to the objective probability of that outcome multiplied by the value of that outcome.
I conclude that you should vote for Superior—even if act consequentialism is true,
yeah, even if objective act consequentialism is true.

Acknowledgments
Thanks to Hud Hudson, Daniel Howard-Snyder, Ryan Wasserman, and Dennis Whitcomb,
for comments on earlier drafts of this paper.

References
Bennett, J. (2003), A Philosophical Guide to Conditionals (Oxford University Press).
Brennan, G., and Lomasky, L. (1989), ‘Large Numbers, Small Costs: The Uneasy Formulation
of Democratic Rule’, in G. Brennan and L. Lomasky (eds.), Politics and Process: New Essays
in Democratic Thought (Cambridge University Press).
Feldman, F. (1986), Doing the Best We Can (D. Reidel Publishing Company).
Frey, R. G. (1983), ‘The Claim of Knowledge’, in Rights, Killing, and Suffering: Moral Vege-
tarianism and Applied Ethics (Basil Blackwell).
Goldman, A. I. (1999), ‘Why Citizens Should Vote: A Causal Responsibility Approach’, in
Social Philosophy and Policy 16/2: 201–17.
Howard-Snyder, F. (2005), ‘It’s the Thought that Counts’, in Utilitas 17/3: 265–81.
Hudson, H. (1993), ‘Collective Responsibility and Moral Vegetarianism’, in The Journal of
Social Philosophy 24/2 (Fall): 89–104.
Lewis, D. (1973), Counterfactuals (Harvard University Press).
Mackie, J. L. (1965), Causes and Conditions’, in American Philosophical Quarterly 2/4: 245–64.
Parfit, D. (1984), Reasons and Persons (Oxford University Press).
Sartorio, C. (2004), ‘How to be Responsible for Something without Causing it’, in Philosophical
Perspectives 18: 315–36.
Sinnott-Armstrong, W. (Winter 2015 Edition), ‘Consequentialism’, in E. N. Zalta (ed.), The
Stanford Encyclopedia of Philosophy, <http://plato.stanford.edu/archives/win2015/entries/
consequentialism/>.
  -

Van Inwagen, P. (1978), ‘Ability and Responsibility’, in The Philosophical Review LXXXVII/2.
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
Vessel, J. P. (2003), ‘Counterfactuals for Consequentialists’, in Philosophical Studies, 112:
103–25.
Vessel, J. P. (2007), ‘The Probabilistic Nature of Objective Consequentialism’, in Theoria
LXXIII.
 11
Theism and Allism
Alex Rosenberg

Peter van Inwagen holds that there is no incompatibility between theism and
Darwinism. Both theists and defenders of Darwinism should be pleased if van
Inwagen is correct. Theists should be pleased since the less logical conflict between
their views and science the better for the acceptance of theism. Mutatis mutandis,
biologists who teach and employ the theory in their work should be pleased because
they and their students would not have to choose between Darwinism and theism,
still less overcome funding obstacles motivated by religious objections to their
scientific views. Alas, such satisfaction would be specious.
I will argue that there are features of both Darwinian theory and theism that make
these two views incompatible, even as they are expounded by so acute a thinker as
van Inwagen.
Van Inwagen has explored the relationship between theism and Darwinism in
several different contexts.1 Here I shall deal mainly with the discussion of the theory
and its relation to theism advanced in “Weak Darwinism” and “The Place of Chance
in a World Sustained by God.”2 There are some significant changes in van Inwagen’s
views over the period in which he has written on this subject. But I have reason to
think that “Weak Darwinism” and “The Place of Chance” constitute van Inwagen’s
considered view of the adequacy of the theory of natural selection, of its logical
compatibility with theism, and of the compatibility of theism and objective chance.
By theism I mean the thesis that there is one and only one thing that has the
properties of omnipotence, omniscience, omnipresence, and benevolence. In the case
of a theist such as van Inwagen I am prepared to add a theodicy to theism owing to
van Inwagen’s defense of theism against the argument from evil by appeal to the role
and value of free will (which van Inwagen rightly takes to be incompatible with
determinism).3
The explananda of the Darwinian theory of natural selection are adaptation as well
as the complexity and the diversity that results from it. The theory also explains

1
In van Inwagen (1988b, 1994, 2003, 2009, and 2010).
2
Van Inwagen (2009 and 1988b).
3
Developing a theodicy is not the main task of van Inwagen (2006). There van Inwagen provides a
critique of the argument as “a failure.” Elsewhere van Inwagen famously defends the existence of free will
and the consequent falsity of determinism. His most explicit articulation of a theodicy is to be found in van
Inwagen (1988a).
  

particular adaptations when combined with boundary or initial conditions. But

owing to the fact that the boundary conditions required to provide detailed explan-
ations of particular adaptations are almost always indeterminable from present
evidence, and often large in number as well as complex in their interaction effects,
such detailed explanations are rare.4 By itself the theory only explains the fact that
there are adaptations in the universe. In fact we shall see that the process it describes
does not by itself even ensure that there are adaptations in the universe.5 It provides
an account of the necessary conditions for the appearance of adaptations, but these
conditions are not by themselves sufficient for the existence of any adaptations
whatever. This point will be important in what follows.
As van Inwagen rightly notes, the theory of natural selection is “almost unique
among scientific theories in that it is very hard to find a statement of it” (2009: 108).
Philosophers of biology and some biologists have pondered why this is so. Some have
offered a variety of candidates for a canonical statement of the theory, but no
explanation of its absence has secured agreement any more than any of the proposed
versions of the theory have.6 Van Inwagen therefore provides his own account of the
theory. In my view van Inwagen’s treatment adds to the theory some considerations
on which evolutionary biologists are in agreement but which are not parts of the
theory, and it omits a crucial component of the theory. It will become apparent
hereafter why these are consequential matters.
There are, van Inwagen contends, five parts to the theory. These are summarized
below with, in cases of my dissent, the reasons for dissenting from van Inwagen’s
formulation given in italics:
1. There have been living things on Earth for about 3.5 billion years and most taxa
are now extinct. This is a singular claim about a particular spatiotemporal
region and so not strictly a part of the theory, which consists only in general
claims without existential import.
2. Any two living things share common descent. It would be a quibble to note that
if living things have emerged on other planets this claim would not be true.
Nevertheless, as a claim about living things on Earth it should not be treated as
strictly a component of the theory.
3. The data to be explained by a theory of evolution are fourfold: two kinds of
adaptation (“apparent teleology or apparent design”) (2009: 109): organism to
environment, and components of organisms to one another; taxonomic diver-
sity; complexity and persistent increases in it over time. The explanation of
these data requires the theory be combined with boundary or initial conditions.

4
Some complexity and diversity may not be the result of natural selection alone, but of what Brandon
and McShea call a “Zero-force” law of evolutionary biology reflecting shear drift. See Brandon and McShea
(2010).
5
Van Inwagen may disagree with this claim about the theory. He writes that the theory “can be used to
explain a vast range of particular facts about living things, facts like the fact that the males of various species
of dabbling ducks have colorful plumage, strikingly different in different species . . . ” (Van Inwagen 2009:
108). Nothing hereafter will turn on whether this claim reflects a disagreement about the explanatory
resources of the theory itself, as opposed to its combination with boundary conditions to explain particular
matters of fact.
6
The best recent discussion of this matter is due to Godfrey-Smith (2007).
    

The theory of evolution purports that “the only purely biological part of the
explanation” of these data regarding “diversity and complexity and apparent
teleology is that provided by the operation of random mutation and natural
selection.” It is not a quibble to note that the theory of natural selection purports
to explain complexity and diversity only insofar as they are the result of adap-
tation. Diversity and complexity that is non-adaptive is a matter on which the
theory is silent (though it is not a matter on which other theories that “work with”
the theory of evolution are silent). The theory of evolution explains diversity and
complexity insofar as they are the results of adaptation. Van Inwagen also
introduces the term ‘mutation’ here to characterize what evolutionary biologists
identify as variation. More on this difference immediately below.
4. Mutations are mainly due to copying errors that occur during reproduction.
They have only biochemical causes. If the laws of chemistry will permit a certain
mutation to occur when a certain cell divides—if that mutation is a chemical
possibility—whether the mutation will occur and how probable its occurrence
is have nothing to do with whether its occurrence would be a “good thing” for
the descendants of that cell (or the descendants of the organism of which that
cell is a part). Strictly speaking the theory of evolution is silent on mutations. It
invokes ‘variation’ as a general label for all sources of hereditary novelty, but is
neutral on the nature of these sources. The sources of hereditary variations, we
are informed by a different theory, are all genetic ones, of which mutational
variation is an important kind. But there are other sources of variation—
recombination, crossing over, as well as some epigenetic processes, including
epigenetic variation (such as in patterns of nucleic acid methylation). Some
mutations are due to copying errors but others are the result of spontaneous base
changes, such as deamination of cytosine. Bearing in mind the variety of chemical
causes of changes in DNA sequence will be important hereafter. The key point
about which van Inwagen is correct is that variation is entirely independent of
whether it is a “good thing” for whatever bears it—gene, chromosome, cell,
organism, etc.
5. All apparent design in nature, all complexity, all diversity, is produced by the
gradual accumulation (directed by environmental pressure—that is, by natural
selection) of small hereditary differences, differences due to random mutations
or to the random recombination of genetic material. We may conveniently
cumulate some of the qualifications on van Inwagen’s account of the theory of
evolution here: Insofar as complexity and diversity are the result of adaptation
(and mainly they are), they along with adaptation are the result of variation of
hereditary differences and their selection. The theory does not take sides on
whether the variations must be all small or evolution gradual. As a matter of
fact for natural selection to cumulate into the degree of adaptation and conse-
quent complexity and diversity present on this planet, the variations will have
been mostly small and gradual.

Some of the differences between van Inwagen’s version of the theory and the version
which generates the dissentient remarks in italics above could be entirely terminological.
Van Inwagen describes the theory he summarizes as the “theory of evolution.” This
  

theory may not be viewed by him as synonymous with the theory of natural selection; it
may be that van Inwagen uses the label “theory of evolution” to include the theory of
natural selection and further claims widely endorsed among evolutionary biologists
about the boundary conditions and mechanisms of evolution on the Earth. In particular
“gradual accumulation of small hereditary differences” has been a matter of dispute
among exponents of the theory, and though I am confident it is required for the
accumulation of adaptations on this planet, natural selection may under other condi-
tions, in which differing mechanisms of heredity emerge, not proceed wholly in this way.
Van Inwagen’s account incorporates one key insight of Darwin and that is crucial:
variation is blind with respect to the usefulness of the variant by the creature that
bears it and the process whereby some variants are preserved long enough to be
represented in the next generation is a pure passive matter of filtering. The environ-
ment merely imposes “sieves”—as many as there are features of an organism’s
environment that may make a difference to its survival and reproduction.
There is one important lacuna in van Inwagen’s account of the process of natural
selection Darwin discovered: the inescapable role of chance, or objective probability
in evolutionary adaptation, a role reflected in the central place of “drift” in all latter-
day elaborations of the theory. As Darwin well knew, it is simply false that the fitter
always survive longer and reproduce more offspring than the less fit. Owing to
vicissitudes of nature, sometimes fitter variants perish through infrequent, irregular,
“exogenous” causes—earthquakes, avalanches, forest fires, lightning strikes, etc.
When the number of competing, and especially interbreeding, variants of a lineage
and the absolute number of fitter variants are small, these causes of death are more
likely to prevent the survival and reproduction of the fittest. Such outcomes are
labeled cases of “drift” or sometimes “genetic drift” to emphasize that the causal chain
that results in the survival of less fit individuals and the spread of their traits moves
through the channel of genes. As interbreeding populations increase in size, the scope
for “drift” is reduced. Were populations infinite in size (as population genetical models
assume), and environments constant long for enough periods, the distribution of traits
would more closely reflect the survival of the fittest. Note that the role of objective
chance in this part of the Darwinian process is distinct from any role it may have in
mutation. Darwin himself supposed that the sources of variation (now understood to
include a certain amount of probabilistic mutation) were entirely deterministic. He
knew nothing of quantum mechanics and assumed the universe was Newtonian.
What is the source of the objective chance involved in selection as opposed to
variation? Natural selection is a population-level process in which objective chances
play the same role as they do in the second law of thermodynamics, another
population-level process. For reasons that will become clear below this is no accident.
In both cases the source of objective chance remains a philosophically vexed ques-
tion. Attempts by philosophers, including the present author, to reduce drift to
subjective probability reflecting human ignorance, have failed, while biologists and
philosophers have offered powerful arguments for its irreducibility.7

7
See Brandon and McShea (2010). If it turns out that the only grounds for objective chance in the
universe, including as it figures in the second law of thermodynamics, are quantum mechanical, then its
existence may be held hostage to the correct interpretation of that theory. If, as I believe, the second law is
    

Subject to his acceptance of the role of objective chance in Darwinian processes, is

there any crucial issue on which van Inwagen dissents from the theory of natural
selection? He certainly does not dissent from the general claim that “only natural
causes have been at work in the history of life,” and explicitly rejects any role in
evolution for “intelligent design”.
Using the present author as an exponent of the view he does reject, van Inwagen
writes, “But then, what do Alex and I disagree about? Our disagreement is essentially
about the following proposition: the only explanation for all this diversity and
complexity is that provided by the operation of random natural selection.” But
note, the theory does not claim that there is no complexity and diversity except
that which adaptation produces. It is compatible with a very limited amount of each,
not much and of little significance once the first adaptation emerges. What van
Inwagen and I really disagree about is the claim that adaptation—the appearance of
design, the fittedness of parts to wholes and wholes to their environment—has any
other sources than natural selection. Following van Inwagen, call this thesis Allism—
“since it’s essentially the thesis that natural selection does it all.” Van Inwagen is an
agnostic about Allism. He contrasts Allism to “Weak Darwinism.” That theory
deletes the ‘only’ from the statement of the theory above, and qualifies it as follows:
The operation of random mutation and natural selection is at least a very important part of the
explanation of all this diversity, complexity and apparent teleology—perhaps it is the whole
explanation and perhaps not. (2009: 113, emphasis added)

Van Inwagen asks and answers the question, would the science of biology be
adversely affected were all biologists to accept only Weak Darwinism? “I don’t see
how it would be . . . One thing is certain. It wouldn’t prevent biologists from explain-
ing particular biological phenomena in terms of natural selection” (2009: 114).
Moreover, Weak Darwinism has all the same observable consequences as Allism,
but is a weaker theory. This, van Inwagen suggests is a mark in the favor of Weak
Darwinism: “when I was in graduate school . . . [the best philosophers of science] used
to say, if two theories have all the same observable consequences and one was
stronger than the other, you shouldn’t adopt the stronger one.” (2009: 114) Popper
would of course have held the contrary view.
It is perhaps arguable whether the biological evidence so far amassed more
strongly supports Allism as opposed to Weak Darwinism, although the same ques-
tion can be asked regarding any strict law in natural science and the same law with a
ceteris paribus qualification. I will not here review the overwhelming evidence for
Allism and the entire lack of evidence for exceptions to it that might substantiate
Weak Darwinism. Instead I will show why Allism is required by considerations of
consilience and the unity of science.
Physics imposes requirements on any theory of the emergence of adaptations—the
appearance of design. Obviously, the theory must be logically compatible with
physical theory. This means the theory can accord no role to design, purpose,

independent of quantum mechanics then objective chance of the sort Darwinian theory invokes is not
dependent for its existence on quantum indeterminism.
  

goals, ends, teleologies, entelechies, or other forces and processes ruled out by
physics. It also requires that the process described by the theory must be driven by
the second law of thermodynamics. For these reasons, physical theory provides the
strongest argument for Allism. For it turns out that there is only one way any account
whatever of adaption can satisfy physics. And that is the account Darwin hit upon.
These constraints operate to make the process that Darwin discovered the only way
adaptation—and its resultant complexity and diversity—can emerge in the physical
universe. Allism is compelled not by evidence biologists can amass but by consider-
ations from more basic science. Let us see why.
For about 400 years now physics has found itself able to explain more and more
and to greater and greater degrees of precision. Its ever-increasing explanatory power
and predictive precision have been the result of a history of theories that have
replaced one another in a succession that has resolutely excluded any role for the
sort of purposes, entelechies, teleologies—immanent or eminent goals, ends, or
designs—that physics from Aristotle to the Scholastics countenanced. No matter
what happens in physics hereafter, the weight of considerations against any role for
such explanatory factors is overwhelming. There is no room in physics for purpose or
any process in which ends play a role in shaping means.
If this is right, any process that builds adaptations—the appearance of means and
ends—must operate on a “substrate,” an initial condition, a starting point of zero
adaptation. A theory that explains the emergence of adaptations but does so by
helping itself to an initial extant adaptation and building further adaptations upon it
is ruled out by physics alone. Nothing that happens in the pre-biotic period, from the
big bang through the synthesis of all the elements and their combination into
inorganic and some organic molecules, can start from some configuration of matter
that is already a biological adaptation. One way to see why this is so is to consider
what is really wrong—conceptually, not factually, with Lamarck’s theory of adapta-
tion as the result of use and disuse in prior generations. The conceptual problem with
this hypothesis is that it postulates a wonderful adaptation: a mechanism that shapes
offspring’s inherited traits to meet needs that parents faced in the immediately
previous generation. Such a means to achieving ends, were it to exist, would be
highly adaptive itself and so requires an explanation of its emergence. But the same
problem daunts any account of adaptation that does not start with the zero-
adaptation baseline that physics imposes. It is not merely that a theory of the
emergence of adaptations that assumes some initial adaptation is question-begging.
It assumes something already ruled out by physics at the time Newton put an end to
Aristotelian Scholasticism.
Another fundamental fact that any theory of adaptation must accommodate is
that, as physics tells us, all the fundamental “laws of working” (a term introduced by
J. L. Mackie) operating in the universe are time-symmetrical: they do not by
themselves impose a direction of time on the universe. This is fairly clear in a
Newtonian world: film the motion of balls on a billiards table and then run it
backward as well as forward. The resulting sequences are equally well explained by
Newtonian theory. But the same goes for a world governed by the Schrödinger wave
equation. The only source of temporal asymmetry in the universe is the second law of
thermodynamics: the entropy of any closes system almost always increases over time,
    

where the “almost always” reflects the operation of extremely high objective prob-
abilities. The second law identifies a process—probable entropy increase—that is
time-asymmetrical, and it is the only fundamental physical law that does so. Indeed,
this is the basis for some arguments that the second law is not a law at all but owes its
operation to the combination of the time-symmetrical laws and the initial or bound-
ary conditions of extremely low entropy at the big bang. For present purposes we
may leave this dispute aside.
The evolution over 3.5 billion years of adaptations on Earth is evidently an
asymmetrical process. Indeed, there are substantial biological arguments that it is
an irreversible process (for example, von Baer’s laws of development). Insofar then as
the evolution of adaption is a physical process, one driven exclusively by forces that
figure in physical theory, it must harness the second law of thermodynamics. It must
be a process that almost invariably increases entropy, disorder, consumes more
energy than it produces.
Because the starting point of adaptational evolution must be a point of zero
adaptation and because the asymmetrical process that brings about the first and all
subsequent adaptations must respect the second law, indeed must be driven by the
process of entropy-increase, it follows that the initial and subsequent adaptations
must be extremely rare and energetically very costly. They must be rare since
adaptations are order-preserving: adaptations are structures or behaviors that have
a pay-off, a benefit, an advantage for the maintenance or improvement or spread or
persistence of some orderly system or other. As such, adaptations are decreases in, or
foster decreases in, entropy. The second law allows such departures from entropy
increase, but only under two conditions: either they must be extremely rare, indeed in
some cases vanishingly improbable; or they must be very expensive, that is, they must
extract more entropy from their surroundings than they save. The second law allows
a certain amount of trading off here: the less of a departure from entropic increase
one of these improbable events constitutes, the more frequently it will occur in the
100 billion year history of our universe. But for all that, the frequency of such
adaptation-making events can never be very high.
Thus, there are constraints imposed by physics on any account of the evolution of
adaptation: no teleology, a zero-adaptation stating point, initial adaptations must be
rare, and subsequent adaptations expensive. It will be easy to show not just that the
process Darwin discovered—blind variation and environmental filtration—honor
these requirements but that no other process can do so. This will suffice to establish
Allism, at least up to the assumption that the universe is governed by processes
physics alone mandates.
The only way the initial adaptation, on which all subsequent ones are built, can
honor the demands of physics is by being an objectively improbable event, the
extremely rare result of a simple thermodynamic random walk. The minimum
adaptation will be a feature of the earliest molecule that fortuitously combines
some degree of stability and replicability. The stability will have to be owing to its
being a local minimum free energy configuration, something well understood in
chemical thermodynamics. The replicability will have to be a matter of templating or
catalyzing or otherwise increasing, by its presence, the appearance of more copies of
the same structure. Chemical processes producing molecules with various degrees of
  

these two features are unlikely to occur anywhere or anywhen, but they are not
excluded by the second law. What is more, once they do occur, the long-term process
of chemical combination of atoms into molecules and molecules into larger ones, in
ways required by the second law, may continue to build more and more complicated
molecules, with greater degrees of stability and further capacities for replication over
time. But of course they may do so only at the expense of extracting more energy
from their environments, and increasing the rate of entropic increase.
So, the minimum adaptation must be improbable, subsequent adaptations must be
expensive and also improbable, and the cumulation of adaptations over time must be
a process driven by entropy increase. What could satisfy these requirements? Well,
evidently the process Darwin discovered does so. To see that any process which does
so must in fact be a Darwinian one, consider the creationist scenario of Bishop
Ussher. Early in the seventeenth century the good bishop famously calculated the age
of the universe by working backwards through the Old Testament. He was able to fix
the date of creation as October 23, 4004  (Julian calendar). Subsequent calculations
by Cambridge University theologians fixed the exact time of creation to 9:00 a.m.
(Greenwich mean time?). Suppose on that day the Earth and everything on it
appeared as a quantum singularity, with no cause at all. Call this “the Ussher
event.” Except for the fact that it has no cause, the Ussher event is just what biblical
inerrancy would lead us to believe happened. It is of course an event offered as a
radical alternative to the process Darwin discovered that honors the constraints of
physics. However, once we strip away from the “Special creation theory” those
features incompatible with physics, the Ussher event becomes just another case of
blind variation and natural selection, albeit a highly unlikely case of the process
Darwin described.
The second law tells us that the occurrence of the Ussher world does not have zero
probability. The probability of the spontaneous emergence of that much order in the
world is not absolutely ruled out by the second law, though nothing remotely like it
should be expected even in a universe many, many, many orders of magnitude
longer-lived than ours. But were the Ussher event to occur, the terrestrial environ-
ment would passively filter all the trait-tokens it creates for fitness and presumably
allow them all to continue to have the causal sequelae that biology, chemistry, and
physics would lead us to expect. The only things the constraints of physics impose on
this scenario are that the arrangements introduced by the Ussher event must incur
ever-increasing entropy increases, and that adaptations cannot appear owing to free-
floating purposes, their identity as means to any ends, or designs recorded prior to
the emergence of relatively complex brains in the universe, or any kind of future
causes asymmetrically reaching back into the past. In this respect the difference
between an Ussher event and its sequelae and the events which in fact brought
forth adaptations in the physical universe is only a matter of degree. The Ussher
event is simply far less probable. Otherwise, it is just another case of Darwinian
processes in action.
Though intended as an alternative to natural selection, the Ussher event isn’t one.
For this reason not only does physics make Darwinian processes possible but it also
makes them necessary, if adaptation of any kind is to emerge in the universe. Note,
the claim is not that physical processes make adaptation inevitable. Quite the
    

contrary, they make adaptation at most rare. Physical processes may go on forever
without producing the merest adaptation. Indeed, that is the most probable outcome.
Additionally, though they do allow for it, the appearance of adaptation will be rare
and accidental, and its persistence will be energetically expensive. It is therefore no
surprise we have found no evidence of adaptation anywhere else in the Universe.
So, the argument for Allism is simply that it is required by physics. This is a pretty
powerful argument since it adds all the evidential support of physics to the support of
at least the fundamental components of Darwinian theory. Of course, physics may be
mistaken, so the argument remains an inductive one. Furthermore, the argument for
Allism does not provide any interesting details about the course of evolution on Earth
or how the process of natural selection actually gets implemented here. It sheds no
real light on almost any substantive research question in biology, and doesn’t really
do more than underwrite the research program of the discipline. But it still provides a
great deal of support for Allism.
In this connection, it is worth noting the considerations van Inwagen offers against
adopting Allism. Some of them in fact tend to substantiate the support physics lends
to Allism. Van Inwagen offers two cautionary tales in favor of a methodological
lesson: “Don’t always assume that the mechanisms you have identified and can
describe and know are at work in the production of a certain phenomenon are all
the mechanisms that are at work in the production of that phenomenon” (2009: 116).
The example he cites, ironically enough, is an argument against Darwinian theory
advanced by Lord Kelvin in the nineteenth century. The case is ironical in the present
connection for it is Lord Kelvin that physics, and if I am right, Darwinian theory, has
to thank for the second law.
Kelvin rejected Darwin’s theory of natural selection on the grounds of physical
theory. Darwin had estimated that evolution on Earth had been going on for at
least 300 million years. He was off by an order of magnitude. Nevertheless, Kelvin
calculated that given the best available physics, the sun could not be expected to
radiate heat for more than 40 million years and that the Earth was no older than this
as well. Of course, quite apart from the fact that by itself Darwinian theory alone is
not committed to any particular amount of time being required for evolution to have
taken place on Earth, Kelvin’s physics was completely mistaken about the nature of
the sun’s energy production. It was only in the late 1940s that Hans Bethe recognized
that the sun was a thermonuclear fusion reaction lasting far longer than is required
even for the 3.5 billion years of the Earth’s age so far.
Van Inwagen suggests that Kelvin should have reasoned from the incompatibility
of his preferred physical theory and Darwinian theory that “there must be at least one
mechanism for producing heat that I don’t know about.” Kelvin should have rejected
Allism about his theory of heat production. By the same token Darwinians should
reject theirs.
The cases are very different in several respects. To begin with Darwin was deeply
disturbed by Kelvin’s argument and he was right to be disturbed by it. He rightly
recognized that physical theory had some sort of scientific priority over biological
theory. Insofar as some physical theory makes a large number of extremely well-
confirmed predictions, is widely employed in a variety of technological applications,
and explains in detail processes otherwise unaccountable, it should have greater
  

evidential weight than theories in other disciplines that lack these three features. But
physics, it turns out, supports Darwinian theory.
One reason Darwin and his supporters did not surrender the theory of natural
selection in the face of Kelvin’s arguments was owing to the role of highly imperfect
estimates of boundary conditions Kelvin required to estimate the age of the Earth.
The incompatibility between physics and Darwinian theory was not a logical incon-
sistency between generalizations or hypotheses but between bodies of hypothesis and
boundary conditions. This is a circumstance in which, as Duhem and Quine have
observed, there are many ways to reconcile inconsistencies.
The relationship between Allism and physical theory is not parallel to the details of
van Inwagen’s cautionary tale from Kelvin. Here the relationship is between the
theories themselves and not packages of theory and boundary conditions. When
theory in one discipline is logically incompatible with theories in other less funda-
mental disciplines, it is right to treat these latter theories as disconfirmed. By the same
token when fundamental physical theory entails a biological theory, it is right to treat
that theory as strongly confirmed. Moreover, the features of physical theory that
constrain adaptation to arise only through Darwinian processes are among the most
well-established parts of physics, not—as in Kelvin’s day, theories at the frontier of
the subject. There is nothing likely to happen in the foreseeable future of physics that
would lead to the surrender of the second law of thermodynamics, its probabilistic
prohibition against the persistence and increase of local order without global entropy
increase, or the exclusion of teleology from fundamental physical processes. This is
what makes Allism so secure an account of the emergence and persistence of
adaptation.
Besides over-enthusiastic induction, van Inwagen gives a second reason, or rather
motive, many writers have to endorse Allism:
A good many proponents of Darwinism think that Darwinism (if true) shows that there is no
God, that Darwinism is inconsistent with theism. And this is a conclusion may of them are
very happy with. If it is evident, however, that Darwinism is inconsistent with theism (they do
suppose this to be evident), it is at least much less evident that Weak Darwinism is inconsistent
with theism. Who knows what the unknown mechanism or mechanisms at work in the
development of life might be if there were such. (2009: 116)

Mutatis mutandis, a theist will have some considerable motivation to think that
Allism is false and to adopt Weak Darwinism. Motives, on either side of the question
whether Allism is true are not, are of course philosophically irrelevant. But what is
philosophically significant is that van Inwagen does not in fact believe that Allism
and theism are incompatible.
Van Inwagen writes:

. . . Darwinism [“understood to include Allism”] is not inconsistent with theism. The argument
is simple. I’ll present it as an . . . ad hominem argument, with homines in question being
Darwinists. “You Darwinists believe that the actual world is a Darwinian world—that is, a
world in which Darwin’s theory is true. But actuality implies possibility: anything that is actual
is possible. And God, if he exists, is by definition omnipotent. And an omnipotent being can
create any possible object, even if that object is a whole universe or cosmos. Well, this
    

Darwinian Earth of ours (as you believe it to be) is a possible object—since it exists. Therefore
an omnipotent being could create it—and could create the whole physical universe of which it is
a part. And if an omnipotent being could create a Darwinian world, then why should someone
who thinks that the actual world is a Darwinian world regard that feature of the actual world as
demonstrating that—as having even any tendency to show that—there is no God?” (2009: 119)
Here is an answer to van Inwagen’s rhetorical question, an argument to show that
Darwinism is incompatible with theism. Once the argument is expounded I will
consider some ways van Inwagen might respond to it.
The argument turns on the central element missing from van Inwagen’s account of
the theory—the indispensable role which it accords to objective chance. The argu-
ment for the incompatibility of theism and Darwinism also requires one theological
commitment. It will have to be an implication of theism that the omnipotent and
benevolent being “created man in his own image” (Genesis, 27): that is, at least beings
like us and possibly other creatures were the intended outcome of a process that God set
in motion, whether one that required six days or 3.5 billion years (the age of the Earth) or
13.7 billion years (the age of the universe) for that matter. I will not argue for this claim
here though I believe it follows from the combination of omnipotence, omniscience, and
complete benevolence which is supposed to characterize the theist’s God. While van
Inwagen has not endorsed the claim that Homo sapiens in particular is part of God’s plan
he has written that “ . . . the existence of animals made in God’s image—that is, rational
animals having free will and capable of love—is a part of God’s plan.”8
Assume that, as van Inwagen writes, the actual world is a Darwinian world. That is
assume that it is a world replete with all the adaptation currently extant, and that it
got that way by a process of blind variation and environmental filtration. The second
conjunct is crucial. It is not enough that, as Darwinians hold, many of the traits of
organisms are current adaptations. That is a matter on which all parties agree. What
is distinctively Darwinian is the historical thesis about how all this adaptation
emerged: from zero adaptation by a Darwinian process. This process must be one
in which a significant role is played by objective chance, both in producing variations
and, more importantly, in the population processes by which drift and selection
interact to produce evolutionary trajectories.
It is a fact that the role of drift has been significant in evolutionary change, and
more importantly that the role of small population size—where objective chance has
a magnified role—has been particularly critical to evolution on the Earth. For this
reason, Stephen J. Gould famously observed: “Wind back the tape of life to the early
days of the Burgess Shale; let it play again from an identical starting point, and the
chance becomes vanishingly small that anything like human intelligence would grace
the replay.”9 The same goes for every other trait adaptive or not that has emerged
through the process of blind variation and environmental filtration over the 3-billion-
year evolution of life on Earth. The process Darwin discovered is a chancy one. It
operates throughout the universe. The scant evidence we have about how it does so
suggests that the emergence and persistence of anything we would recognize as life is
an extremely rare thing in the universe, that the emergence of nucleic acid/amino

8 9
Van Inwagen (1988: 55–6). Gould (1989).
  

acid-based life is still rarer, that eukaryotic organisms rarer still, metazoans improb-
able in the extreme, and vertebrates, mammals, primates, the genus Homo, and
humanity successively less probable outcomes of Darwinian processes.
Another thing we can be equally confident about is that an omniscient being
would be aware of this fact about Darwinian processes, that owing to the role
objective chance plays in them, they are unlikely to eventuate in us, or indeed
anything much like us. An omniscient being who created the world and arranged
that it proceed in accordance with the process Darwin discovered would know that it
was unlikely to eventuate in us. He would know that Darwinian processes are
extremely unreliable means of making anything in his own image. A being combin-
ing this epistemic insight with omnipotence would be able to organize the creation
and history of the universe it created in accordance with any set of (presumably
logically consistent) laws of working, and arrange its initial conditions in such a way
as to produce human kind with probability 1.0 at the place and time of its choosing.
So, if we are an intended outcome of God’s handiwork, then he could not have
employed Darwinian processes to make us, and he had to have employed methods
incompatible with the operation of Darwinian processes.
If this line of argument is correct, then a Darwinian world is not compatible with the
existence of an omnipotent, omniscient, benevolent being who intended the existence of
our species or its particular members and took steps to achieve this outcome. For a
Darwinian world is one in which our existence is a highly improbable matter of objective
chance. An omnipotent omniscient being could have created us and used a process that
for all the world looks to us, or to any creature who is less than omnipotent, to be one in
which life is the result of the operation of an objectively chancy process of variation and
natural selection. It is compatible with theism to suppose that God employed a method
to bring us about so complex that we cannot understand it, and indeed one that on the
best evidence we mistakenly infer is the result of an objectively chancy Darwinian
process. What is more our Darwinian theory may also be unrivaled, even by the correct
non-chancy account, in application, where considerations of computational tractability
and simplicity may trump truth. But a theory with all these merits would still be false. It is
no logical reconciliation of Darwinian theory and theism to show that though knowably
false it is the best we can do by way of explaining life.
Let us consider three approaches that might circumvent this problem for someone
who, like van Inwagen, seeks to reconcile Darwinism and theism.
First, if the omnipotent, omniscient being can create a world in which objective
probabilities operate, then such a being can create indefinitely many such worlds, and
thereby create enough so that in at least one of these worlds humans emerge as a
matter of objective chance. But the theist will have to ask him- or herself several
questions about such a scenario. First, why would an omnipotent, omniscient God
use such a method to create humans? There is much to be said against such a method,
especially if indefinitely many of these worlds contain sapient creatures who will
suffer harms and make the problem of evil even more serious than it is.10 Second,

10
At the conclusion of ‘Weak Darwinism’ van Inwagen writes: “If anything we find in the world has any
tendency to show [that there is no God], it is the immense amount of suffering the world contains. This is
of course the so-called problem of evil. But the fact that the world contains an immense amount of
    

given that the omnipotent, omniscient being can produce a world with humans in it
by non-chancy means, why would he opt for a method that is chancy and requires
producing indefinitely many objectively chancy worlds? Surely, the answer that God
choose this method simply because he wanted to make Darwin’s theory true is not
one to be taken seriously.
A second potential route to reconciling omniscience with objective chances
invokes the “best systems” account of laws associated with J. S. Mill, Frank Ramsey,
and David Lewis: if all the systematizations optimally combining systematicity and
strength in their description of the history of the actual world include among their
underived fundamental laws ones that, like the second law, are probabilistic, then on
the Mill/Ramsey/Lewis account of laws the actual world will be one in which
objective chance obtains. The omnipotent, omniscient being could certainly arrange
a world so that the best system of laws included enough objective chance to make
Darwinian regularities among the (derived) laws in this system. Philosophical obs-
tacles to the adequacy of this approach are obvious. Indeed, they are so serious that
they undermine the Mill/Ramsey/Lewis theory as a metaphysical account of physical
necessity. The immediate theological question this reconciliation of chance and
omniscience raises is why an axiomatic system would maximize simplicity and
strength for an omniscient being? Such a being would need to make no compromises
in strength to attain simplicity since the being “already” knows everything that is
going to happen in the actual world. Since there is no scope in the epistemic economy
of the omniscient being for subjective probabilities, the objective probabilities
reflected in the best system play no predictive role for that being. Conferring an
explanatory role on such laws for the omniscient being will require a substantive and
controversial non-erotetic, non-pragmatic theory of explanation, for the omniscient
agent has no erotetic explanatory goals.
The problem facing any attempt to reconcile theism and Darwinian theory
parallels the problem of reconciling foreknowledge and free will. Indeed, it may
subsume the problem of foreknowledge and free will if the exercise of free will is a
matter of contracausal indeterminism, as van Inwagen has held.
There is a third approach to reconciling God and objective chance that van
Inwagen expounded long before “Weak Darwinism.” In “The Place of Chance in a
World Sustained by God”11 van Inwagen has argued that there are several sources of
chance in the world, including free will and quantum mechanical indeterminism.
These he reconciles with theism by holding that the omniscient God
. . . is capable of decreeing that a certain indefinite condition be satisfied without decreeing any
of the alternative states of affairs that would satisfy it . . . if there are alternative initial arrange-
ments of particles, any of which would have served God’s purposes for his creation equally well,
then certain features of the world must be due to mere chance. How pervasive these features
may be, and how important they might seem to us, are, of course, further questions . . . And this

suffering is not a discovery of science” (Van Inwagen 2009: 119). Insofar as the reconciliation of Darwinism
and theism might require the proliferation of worlds in which there is even more suffering than the
immense amount in this world, it makes the so-called problem of evil more pressing still.
11
Van Inwagen (1988b: n 2).
  

same result, the existence of states of affairs due to chance, follows from our consideration of
human freedom and natural indeterminism. I do not doubt all three sources of chance have in
fact been in operation, and that many features of the actual universe due to them—perhaps even
features as prominent as the human race.
In short, God’s plan does not include every detail of the history of the universe.
Though he is omniscient and omnipotent, he concerns himself with only some
features of the universe that he created and left many others to chance, including
which among the indefinitely many sorts of rational animals capable of free will and
loving God would in fact be realized.
The trouble is that, even if this is so, God could still not have “allowed” the
Darwinian process of blind variation and environmental filtration to “select” from
all the alternative kinds of rational, free-willed, God-loving creatures, one or more to
actualize. For owing to the substantial role of chance in Darwinian processes, they
might well have eventuated in no creatures of those kinds at all. And this is the
problem of reconciling theism and Darwinism, even for a conception of God that
reconciles omniscience, omnipotence, and chance.

References
Brandon, R., and McShea, D. (2010), Biology’s First Law: The Tendency for Diversity and
Complexity to Increase in Evolutionary Systems (University of Chicago Press).
Godfrey-Smith, P. (2007), ‘Conditions for Evolution by Natural Selection’, in Journal of
Philosophy 104: 489–516.
Gould, S. J. (1989), Wonderful Life: The Burgess Shale and the Nature of History (W. W. Norton
& Co.).
Van Inwagen, P. (1988a), ‘The Magnitude, Duration and Distrubtion of Evil: A Theodicy’, in
Philosophical Topics 16: 161–87.
Van Inwagen, P. (1988b), ‘The Place of Chance in a World Sustained by God’, in T. V. Morris
(ed.), Divine and Human Action: Essays in the Metaphysics of Theism (Cornell University
Press).
Van Inwagen, P. (1994), ‘Doubts About Darwinism’, in J. Buell and V. Hearne (eds.),
Darwinism: Science or Philosophy? (Foundation for Thought & Ethics), 177–91.
Van Inwagen, P. (2003), ‘The Compatibility of Darwinism and Design’, in N. A. Manson (ed.),
God and Design: The Teleological Argument and Modern Science (Routledge), 348–63.
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
Van Inwagen, P. (2009), ‘Weak Darwinism’, in L. Caruana (ed.), Darwin and Catholicism
(T&T Clark), 107–20.
Van Inwagen, P. (2010), ‘A Kind of Darwinism’, ‘Darwinism and Design’, ‘Science and
Scripture’, in M. Steward (ed.), Science and Religion in Dialogue, Vol. 2. (Wiley-Blackwell).
 12
The Evolutionary Argument
for Atheism
Daniel Howard-Snyder

It is commonly said that Darwinian evolution conflicts with theistic religion. Those
who say such things often have in mind claims that are peripheral to theistic
religion, for example, the claim that God created the earth about 6,000 years ago
or the claim that God directly created each species. I have no interest in the thesis
that Darwinian evolution conflicts with these peripheral claims. I do have an
interest, however, in the thesis that Darwinian evolution conflicts with claims that
are central to theistic religion. Among those who say that Darwinian evolution
conflicts with claims central to theistic religion, we find those who say that it is
incompatible with them and those who say that, although there is no incompatibil-
ity, evolution nevertheless provides significant evidence against them. In this essay,
I will focus on the second group.1 More narrowly, I will focus on Paul Draper’s
argument for this conclusion.2

12.1 Draper’s Evolutionary Argument for Atheism

Draper characterizes evolution as consisting of two theses:
Evolution. (1) The genealogical thesis: All complex organisms are the more or less
gradually modified descendants of a small number of simple unicellular organ-
isms. (2) The genetic thesis: All evolutionary change in populations of complex
organisms either is or is the result of trans-generational genetic change.3
Notice that evolution, so understood, is silent about the mechanisms of evolutionary
change. Darwinism is not silent. According to Draper, Darwinism consists of the
following thesis:
Darwinism. Natural selection operating on random genetic mutation is the prin-
cipal mechanism driving evolutionary change.4

1
For an extensive discussion of a variety of incompatibility claims, see van Inwagen (2010).
2 3 4
Draper (2008a [1997]). Draper (2008a [1997]: 209). Draper (2008a [1997]: 209).
  -

Darwinian evolution is the conjunction of Darwinism and evolution.

Draper asserts that evolutionary arguments for atheism that “appeal to Darwinian
evolution rather than just to evolution . . . overestimate the strength of the evidence
for Darwinism.”5 Draper does not say why he thinks that this is so, but for our
purposes it does not matter. What matters is that he appeals only to evolution, the
evidence for which, he says, “is overwhelming—so overwhelming that evolution can
legitimately be taken as fact rather than mere theory.” Draper has in mind the
evidence one finds in biology textbooks: “information about selective breeding and
other changes within populations of animals, as well as what we know about the
geographical distribution of living things, homologies, the fossil record, genetic and
biochemical evidence, imperfect adaptations, and vestigial organs.”6
I’m not so sure that the textbook evidence allows us to regard evolution, as Draper
defines it, as “fact rather than mere theory.” That’s because I’m not so sure about
whether the textbook evidence suffices for us to say that it is a fact that all complex
organisms are the more or less gradually modified descendents of a small number of
simple unicellular organisms. In this connection, we might consider those theorists
who in their theorizing seem to stretch the meaning of “more or less gradual
modification” to the breaking point. For example, do the products of the flurries of
adaptations within a population followed by long-term stability postulated by
Eldredge and Gould in their theory of punctuated equilibria still count as gradually
modified descendents of their predecessors? Or what about the distinct populations
that suddenly appear in the fossil record near the beginning of the Cambrian period?
Are they the more or less gradually modified descendents of their predecessors, even
on a geological time scale? Or consider the sort of change postulated by many
anthropologists and biolinguists in relation to human evolution. The fossil record,
some of them say, strongly suggests that anatomically modern humans, who were
behaviorally indistinguishable from their primitive neighbors, the Neanderthals and
homo erectus, gained the suite of cognitive capacities typical of behaviorally modern
humans, including the capacity for language, virtually “instantaneously.”7 Here are
Noam Chomsky’s speculations about the matter:
An elementary fact about the language faculty is that it is a system of discrete infinity. Any such
system is based on a primitive operation that takes n objects already constructed, and
constructs from them a new object: in the simplest case, the set of these n objects. Call that
operation Merge. Either Merge or some equivalent is a minimal requirement [for the posses-
sion of language]. With Merge available, we instantly have an unbounded system of hierarch-
ically structured expressions. The simplest account of the “Great Leap Forward” in the
evolution of humans would be that the brain was rewired, perhaps by some slight mutation,
to provide the operation Merge, at once laying a core part of the basis for what is found at that
dramatic “moment” of human evolution, at least in principle; to connect the dots is no trivial
problem. There are speculations about the evolution of language that postulate a far more
complex process: first some mutation that permits two-unit expressions (yielding selection
advantage in overcoming memory restrictions on lexical explosion), then mutations permit-
ting larger expressions, and finally the Great Leap that yields Merge. Perhaps the earlier steps

5 6 7
Draper (2008a [1997]: 209). Draper (2008a [1997]: 209). Diamond (1992).
      

really took place, but a more parsimonious speculation is that they did not, and that the Great
Leap was effectively instantaneous, in a single individual, who was instantly endowed with
intellectual capacities far superior to those of others, transmitted to offspring and coming to
predominate, perhaps linked as a secondary process to the [sensori-motor] system for exter-
nalization and interaction, including communication as a special case. At best a reasonable
guess, as are all speculations about such matters, but about the simplest one imaginable, and
not inconsistent with anything known or plausibly surmised. In fact, it is hard to see what
account of human evolution would not assume at least this much, in one or another form.8
If I understand him correctly, Chomsky speculates here that the transition from
wholly lacking to fully possessing the capacity to think with “an unbounded system
of hierarchically structured expressions,” and all that that entails, “was effectively
instantaneous, in a single individual, who was instantly endowed with intellectual
capacities far superior to those of others,” shortly thereafter “linked” in offspring “as
a secondary process to the [sensori-motor] system for,” among other things, com-
munication. He seems to be suggesting that the individual in question was virtually
capable of thinking in a language, but its parents were not. Put the parents, as infants,
in a modern English-speaking home, and they’d never learn to think in English. But
their child would, as would their grandchildren, great-grandchildren, and so on.
Indeed, he seems to suggest that their child, raised in an English-speaking home,
would have been able to speak English too if the evolution of her sensori-motor
system had kept apace with a similarly serendipitous “slight mutation” producing the
requisite cerebral “rewiring.”9
Like I said, I am unsure about whether the organisms resulting from these sorts of
postulated adaptations count as the more or less gradually modified descendents of
their predecessors. I would imagine, however, that those who postulate such things
are well aware of the textbook evidence Draper mentions. Will they regard their
postulations as conflicting with that evidence? I doubt it. Still, I think we can safely
say that they mean to push the semantic boundaries of “more or less gradually
modified,” especially “less.” That’s one reason why their “more” gradualist colleagues
raise a ruckus about their postulations. (Witness the highly publicized row between
Dawkins and Gould, for example.) Of course, we can easily imagine some maverick
graduate students really pushing the boundaries. That’s what grad students do. To be
sure, there is no sharp line between “more or less gradual” and “not gradual.”
Nevertheless, among the possible pathways of genetic space there exist many a
“great leap” from one point to another where, were they able, the keepers of the
guild would look at the offspring, and then back at its parents, and rightly say: “the
postulated mutation and subsequent adaptation is just too quick for that offspring
to be a more or less gradually modified descendent of theirs.” Were such a situation
to arise, would the keepers of the guild be able to correct the mavericks by pointing to
uncontroversial “information about selective breeding and other changes within

8
Chomsky (2005: 11–12).
9
Of course, friends of the alternative continuity hypothesis are rather down on the idea that some
human’s brain could be “rewired” by a “slight mutation” with the result Chomsky envisions. (See
McBrearty and Brooks 2000).
  -

populations of animals, as well as what we know about the geographical distribution

of living things, homologies, the fossil record, genetic and biochemical evidence,
imperfect adaptations, and vestigial organs”?
I suppose so. After all, the textbook evidence is impressive, to say the least. Still,
I’m not sure, and I’m unsure enough to refrain from joining Draper in his unsur-
passably high opinion of the textbook evidence. Nevertheless, let’s grant his opinion
and move on.
So then: how does Draper propose to move from the fact of evolution to atheism?
At the most general level, he argues that, antecedently, evolution is much more
probable on the assumption that naturalism is true than on the assumption that
theism is true; thus, since naturalism entails the denial of theism and naturalism is
no less intrinsically probable than theism, theism is very probably false—all other
evidence held equal. It will prove useful to state Draper’s argument explicitly.
1. Evolution is true.
2. Antecedently, evolution is much more probable on naturalism than on theism.
3. Naturalism is no less intrinsically probable than theism.
4. So, all other evidence held equal, naturalism is much more probable than
theism. (1–3)
5. Naturalism entails the denial of theism.
6. So, all other evidence held equal, theism is very probably false. (4, 5)
Before we assess Draper’s defense of his argument, we must get clearer on its central
terms.

12.2 Definitions
When he originally published his argument, Draper defined naturalism as the thesis
that “the physical universe is a ‘closed system’ in the sense that nothing that is neither
a part nor a product of it can affect it.”10 This is an inadequate definition of
naturalism. For consider the view that concrete reality, as opposed to abstract reality,
is exhausted by two realms, one material and one immaterial. Immaterial souls
populate the immaterial realm, one of whom is a perfect personal God. But since,
on the view in question, it is a matter of metaphysical necessity that there is no causal
interaction between these two realms, God cannot affect the material realm; nor can
lesser souls. On Draper’s definition, this view is a version of naturalism. That is the
wrong result.
Draper more recently defined naturalism by first defining “physical entities” and
“natural entities.” Physical entities, he says, are “entities of the sort studied in the
physical sciences (e.g. atoms) along with any other as yet undiscovered entities whose
behavior is governed by the fundamental laws of physics.”11 Natural entities, he says,
are “physical entities and entities that are either ontologically reducible to physical
entities or are caused by physical entities.”12 Given these definitions, Draper defines
naturalism as the thesis that there are natural entities and “[they] have only natural

10 11 12
Draper (2008a [1997]: 208). Draper (2008b: 150). Draper (2008b: 150).
      

causes.”13 This definition of naturalism is also inadequate. For what counts as

“entities of the sort studied in the physical sciences”? I should think that the correct
answer is this: whatever sort of entities the practitioners of the physical sciences
might appeal to in their theorizing as scientists. Of course, since some physical
scientists, whether past, present, future, or merely possible, aim to articulate, or at
least contribute to an articulation of, a Final Theory, a Theory of Everything, who’s
to say what they might appeal to? Moreover, given that what they in fact appeal to is
a highly contingent matter, dependent on culture, politics, personalities, incentives,
and so forth, who’s to say that they could not appeal to God, or something
sufficiently like him? Is that really so far-fetched? I don’t think so. Many scientists
are willing to posit the non-existence of God while doing their science. Why might
not some of them find it useful someday to return the favor? Surely that’s possible.
So, by Draper’s definition, a personal God qualifies as a physical entity, and so a
natural entity, and so not a supernatural entity, and so not ruled out by naturalism.
Again, this is the wrong result. (This definition also runs afoul of the worry for the
first definition.)
Most recently, Draper distinguishes “the mental world” from “the physical world,”
that is “the world of conscious experiences like thoughts, feelings, imaginings, and
sensations” from “the world of rocks, chemical reactions, galaxies, and neurons,” and
then says that “[n]aturalists claim that either the mental world does not exist or it
does exist but is asymmetrically dependent on the existence of the physical world
(i.e., were it not for a physical world, there would be no mental world, while there
would still be a physical world even if there were no mental world).”14 But again, this
is inadequate. For it implies that one is a naturalist if, but not only if, one thinks that
there is a mental world and a physical world and, in addition, that (i) if there were no
physical world, there would be no mental world and (ii) if there were no mental
world, there might still be a physical world. Imagine a theist who holds that God
exists necessarily and that it’s necessary that God creates a physical world.15 Natur-
ally, she will infer that it’s necessary both that there’s a mental world and a physical
world, in which case she will think that the antecedents of (i) and (ii) are absolutely
impossible. Further, imagine that she endorses the standard semantics for counter-
possibles, according to which every counterpossible is trivially true. Then she will
infer that (i) and (ii) are both true. Thus, on Draper’s definition, she is a naturalist.
But clearly she is not.
Draper is not the first person to face difficulty defining naturalism. In David
Papineau’s opinion, “[t]he term ‘naturalism’ has no very precise meaning in con-
temporary philosophy” and he suggests that “[i]t would be fruitless to try to adjudi-
cate some official way of understanding the term.”16 Of course, Draper does not need
an “official way of understanding the term.” He only needs one that both renders
premise (5) of his argument true—“Naturalism entails the denial of theism”—and
rules out what naturalists unanimously want to rule out, i.e. a personal God and
immaterial souls. Draper’s definitions do not do that.

13 14
Draper (2008b: 149–50). Draper and Dougherty (2013: 69–70).
15
Perhaps Aquinas fits this description. See Kretzmann (1998). 16
Papineau (2009).
  -

I propose that we adopt the well-worn adage that naturalism is the view that there
is no God or anything like him. Despite its imprecision, at least it satisfies our two
desiderata. As for theism, let’s go with Draper’s definition: there is a personal God,
perfect in power, knowledge, and moral goodness, who is the creator of the physical
universe and who can intervene in it.17
Draper concludes that evolution makes it very probable or likely that theism is
false, all other evidence held equal. But what does he mean by “probable” and
“likely”? There are many different ways to understand these terms.18 Draper says
that by “probable” and “likely” he means epistemic probability. Of course, that’s
helpful only if we know what that is. In the only published comment on the matter
that I know of, Draper says that epistemic probability is poorly understood. He’s
right. Still, he says that we can get by with the following formulation: “Relative to K, p
is epistemically more probable than q, where K is an epistemic situation and p and q
are propositions, just in case any fully rational person in K would have a higher
degree of belief in p than in q.”19
One difficulty with Draper’s formulation is that it presupposes that belief comes in
degrees. After all, if belief does not come in degrees, then, given his formulation, there
is no such thing as epistemic probability and so nothing bears that relation to
anything else, in which case three of the six statements that comprise his evolutionary
argument for atheism are false, including the conclusion. So those of us who deny
that belief comes in degrees, but who wish to engage that argument, must look
elsewhere to gain some purchase on epistemic probability.20
Fortunately, we need not look far. For when Draper argues for instances of ‘p is
more epistemically probable than q given K’ and the like, he argues for instances of
‘we have better reason to believe (or expect) p than q given K’ and their kin. In fact, he
never argues for instances of the former without arguing through instances of the
latter. Of course, those of us who deny that there are degrees of belief are perfectly
happy with better and worse reason. For example, I have much better reason to
believe that yellow jackets can fly than that hippos can fly; and I have much better
reason to expect to see a bald eagle soaring over Whatcom County’s Lake Padden in
Spring than a penguin waddling along its banks anytime. So in what follows I will
understand epistemic probability as follows: given K, p is more epistemically prob-
able than q for someone if and only if, given K, he has better reason to believe or
expect p than q, where K is his epistemic situation. Thought of in this way, epistemic
probability is a measure of the comparative goodness of one’s reason to believe p
rather than q, given K. In what follows, I will be so bold as to say that this is Draper’s
understanding of epistemic probability, even though I have neither heard nor seen
him say or write any such thing.21
Notice that epistemic probability, on Draper’s understanding of it, is relative to a
person and her epistemic situation at a time. Consequently, something might be
epistemically probable for a person at one time but not for the same person at

17 18
Draper (2008a [1997]: 207). For some of those ways, see Hájek (2012).
19 20
Draper (1989: 349 n2; 1996: 27 n2). Cp. van Inwagen (1997: 69–70).
21
For a very different understanding of epistemic probability and its relation to arguments framed in
the way Draper frames the argument under discussion, see van Inwagen (1997).
      

another time. Furthermore, something might be epistemically probable for one

person at one time but not for another person at the same time. This last fact raises
a pointed question: just whose epistemic probabilities are in question when Draper
frames his evolutionary argument for atheism? In what follows, I will deploy the
fiction of epistemic probability being relative to our epistemic situation, and I will
speak in terms of what we have reason to believe given our epistemic situation. At one
crucial point, however, I will dispense with that fiction and speak only of myself.
(This is also as good a place as any to call the reader’s attention to the fact that, in
much of what follows, I will leave reference to our epistemic situation implicit.)
Draper’s understanding of epistemic probability raises several more questions. For
example, exactly what is this “epistemic situation” of which he speaks? Does it
include nonmental items as well as mental items? If so, which ones? Among the
mental items, what does it include and what does it exclude? And what is it to “have”
good reason? And what does the goodness of good reason consist in? And what’s
meant by “reason” anyway? Is it just another word for evidence? If so, which of the
alternative conceptions of evidence is intended? And so on. It is clear that Draper
relies on the fact that the axioms of probability theory, and certain theorems as well,
are true of epistemic probability. I will grant his reliance on these things but, like him,
I am not going to answer the questions that remain, although it will be evident
enough that I will assume without argument partial answers to some of them at
certain points in what follows.
Draper also says things of the form ‘Antecedently, r is more probable on p than q,’
as when he states that, antecedently, evolution is much more probable on naturalism
than on theism. By “antecedently” in that statement Draper means “independent of
the observations and testimony that together constitute the primary evidence” upon
which our knowledge of evolution is based.22 Thus, given Draper’s understanding of
epistemic probability, Draper means that, if we “abstract from” our evidence for
evolution, we have much better reason to believe or expect evolution on naturalism
than on theism.
It is important for Draper’s purposes that what remains after abstraction includes
the fact that complex life exists and the fact that at one time there was no life (among
other facts). That is because it is no part of Draper’s aim to argue that those two facts
are more likely on naturalism than on theism. Evolution is his exclusive focus.
As for the comparative intrinsic probability of a pair of incompatible theses, the
rough idea is that if we abstract away from all (extrinsic) evidence for and against
them, and we simply focus on what they say, we will be able to tell which is more
probable on its own (if either is); that is, we will be able to tell which we have more
reason to believe or expect (if either). The very idea of intrinsic probability is vexing,
to say the least. More vexing still is its value for adjudicating incompatible meta-
physical theses and how it can be measured without bias. Of course, Draper’s argument
requires a stance on these issues, a stance that at least some will find problematic.
Rather than explore the matter further here, I will quickly pass over premise (3)—the
claim that naturalism is at least as intrinsically probable as theism—only pausing long

22
Draper (2008a [1997]: 209).
  -

enough to call our attention to the vexation it can arouse.23 Nor will I say anything
about Draper’s inference from (1)–(3) to (4), except that he assumes it is valid given
two theorems of the probability calculus that, he says, are true of epistemic probability.
I want to focus on premise (2).

12.3 Draper’s Defense of Premise 2: An Overview

Here is premise (2) again:
2. Antecedently, evolution is much more probable on naturalism than on theism.
That is, abstracting from our evidence for evolution, we have much better reason to
believe or expect evolution on naturalism than we do on theism. Draper’s defense of
(2) begins with another definition:
Special creationism. Some complex organisms are not the more or less gradually modified
descendants of a small number of simple unicellular organisms, but rather were created by the
direct intervention of God in the natural order.24

Obviously enough, since evolution includes the genealogical thesis, according to

which all complex organisms are the more or less gradually modified descendents
of a small number of simple unicellular organisms, “evolution entails that special
creationism is false.”25 In that case, says Draper, it follows from some basic theorems
of the probability calculus that are true of epistemic probability, that we can “estab-
lish” (2) if we establish that
special creationism is antecedently much more [epistemically] likely to be false on natural-
ism than on theism and that, even on the assumption that special creationism is false,
evolution is still antecedently at least as [epistemically] likely to be true on naturalism as it
is on theism.26

That is, to establish (2), Draper sets out to establish that, when we abstract from our
evidence for evolution and reflect on what remains, we will discover two things: (i) we
have much better reason to believe the denial of special creationism given naturalism
than we have given theism, and (ii) assuming the denial of special creationism, we
have at least as good reason to believe evolution given naturalism than we have given
theism. Draper puts these matters more formally. Let ‘T,’ ‘N,’ ‘E,’ and ‘S’ stand for
theism, naturalism, evolution, and special creationism respectively; let ‘>!’ stand for
‘is much greater than’; let ‘’ stand for ‘is equal to or greater than’; let ‘’ stand for ‘is
equal to or less than’; and let ‘P’ stand for ‘the antecedent epistemic probability of ’.

23
I should note that in his (2008a [1997]), Draper does not speak in terms of intrinsic probability but
rather in terms of plausibility. More recently, however, he describes the comparative plausibility of
naturalism and theism in terms of their intrinsic probability.
24 25
Draper (2008a [1997]: 209). Draper (2008a [1997]: 209–10).
26
Draper (2008a [1997]: 210).
      

(Notice how ‘P’ buries reference to particular persons in a particular epistemic

situation at a particular time.) Since S entails ~E, says Draper,
P(E/N) >! P(E/T) if and only if P(~S/N) x P(E/~S&N) >! P(~S/T) x P(E/~S&T).
To establish that the right-hand side of the biconditional is true, it suffices to establish
(i) P(~S/N) >! P(~S/T) and (ii) P(E/~S&N)  P(E/~S&T). In what follows, I will
focus on (i). Before I turn to Draper’s defense of (i), it will prove instructive to
consider Alvin Plantinga’s recent critique of Draper’s argument.

12.4 Plantinga on Draper’s Argument

In Where the Conflict Really Lies, Plantinga contends that Draper’s argument fails for
two reasons.
First, he says, if theism is noncontingent, “[Draper’s] argument won’t be rele-
vant.”27 Plantinga does not identify what it will not be relevant to, but we can gather
from the context that he means that it will not be relevant to Draper’s overall defense
of the claim that naturalism is more likely than theism, all other evidence held equal.
He writes:
Suppose . . . that theism is noncontingent: necessarily true or necessarily false. If so, [Draper’s
premise 2] doesn’t imply that naturalism is more likely than theism; instead [2] obviously
entails that theism is true. For if theism is noncontingent and false, then it is necessarily false;
the probability of a contingent proposition on a necessary falsehood is 1; hence P(E/T) is 1. But
if, as Draper claims, P(E/N) is greater than P(E/T), then P(E/T) is less than 1, hence T is not
necessarily false. If T is not necessarily false, however, then (given that it is noncontingent) it is
necessarily true. So if theism is noncontingent, and [Draper’s premise 2] is true, then theism is
true, and indeed necessarily true.28
What should we make of these words? So far as I can see, they contain a misunder-
standing. Draper thinks of probability as epistemic probability. Thinking of prob-
ability in this way, it is false that “the probability of a contingent proposition on a
necessary falsehood is 1.” Witness Dmitry. Dmitry thinks
B. Mt. Baker’s elevation is 10,781 feet,
which is a contingently true proposition. He also thinks
L. Lying whenever it suits him is morally acceptable,
which is a necessary falsehood. Should we conclude that the epistemic probability for
Dmitry of (B) on (L) is 1? That is, should we conclude that, given his epistemic
situation, he has maximally good reason to believe (B) on (L)? Of course we
shouldn’t. It appears, therefore, that Plantinga’s first objection to Draper’s evolution-
ary argument for atheism misses the mark: it is not true that “the probability of a
contingent proposition on a necessary falsehood is 1.”

27 28
Plantinga (2011: 50–1). Plantinga (2011: 50).
  -

Perhaps there is some other understanding of probability on which “the probabil-

ity of a contingent proposition on a necessary falsehood is 1.” For example, perhaps
there is some understanding of probability on which the following argument is
sound: for any proposition p and q, if p is necessarily false and q is true, then it’s
not possible for p to be true and q to be false; moreover, for any proposition p and q, if
it’s not possible for p to be true and q to be false, then the probability of q on p is 1; so,
for any proposition p and q, if p is necessarily false and q is true, then the probability
of q on p is 1. Of course, if we think of probability as a measure of the possible worlds
at which both p and q are true over those at which p is true, or if we think of it as a
measure of the area of logical space at which both p and q are true over that at which
p is true, then, since there is no world or region at which p is true—it is necessarily
false, after all—the probability of q on p is 0/0—which is not 1. So those understand-
ings of probability will not serve Plantinga’s purposes. But perhaps he has some other
understanding of probability in mind according to which our little argument, or
perhaps some other argument with the same conclusion, is sound. Whatever may be
the case on that score, it remains clear that on Draper’s understanding of epistemic
probability, it is false that “the probability of a contingent proposition on a necessary
falsehood is 1.”
Plantinga’s second objection seems to miss the mark as well. After summarizing
the basic structure of Draper’s argument for premise 2, he writes:
Suppose this [that is, premise 2] is true: how much does it really show? As he [Draper] says, if all
else is evidentially equal, theism is improbable. But of course all else is not evidentially equal. Aren’t
there a host of other probabilities in the neighborhood that favor theism at least as heavily?29

He then cites the probability of there being life on earth given theism, and the
probability of there being intelligent beings, and beings with a moral sense, and
creatures who worship God. He continues:
There will also be other “known facts” that are more probable on theism than on naturalism. If
so, however, the evidence favoring naturalism over theism that Draper cites will be more than
counterbalanced by evidence favoring theism over naturalism.30

Of course, Draper never denied any of this; in fact, he explicitly acknowledged

it. That was the point of the “all other evidence held equal” qualification. Draper’s
sole concern is the evidential status of evolution for atheism, not these other things.
I conclude that, if all that is to be said about Draper’s argument is what Plantinga
has to say about it, then—contrary to Plantinga’s repeated claim throughout Where
the Conflict Really Lies that there is no conflict between evolution and theism, whereby
“conflict” he means to include what he calls “probabilistic incompatibility”—evolution
does constitute significant evidence against theism.

12.5 Showing P(~S/N) >! P(~S/T)

As I said, Draper sets out to establish that P(E/N) >! P(E/T) by doing two things: (i)
showing that P(~S/N) >! P(~S/T) and (ii) showing that P(E/~S&N)  P(E/~S&T).

29 30
Plantinga (2011: 51). Plantinga (2011: 52).
      

I will focus on (i). Draper’s argument for P(~S/N) >! P(~S/T) moves in two steps.
Step 1 consists in an argument for the claim that P(~S/N) > P(~S/T). Step 2 consists
in an argument for the claim that “P(~S/T)  ½, which is to say P(S/T)  ½—that,
independent of the evidence for evolution, special creationism is at least twice as
likely as not on the assumption that theism is true.” In defense of this last claim,
Draper rejects reasons for thinking that God did not specially create and affirms
reasons for thinking that God did specially create, inferring that P(S/T)  ½. If both
steps are well-placed, it follows that P(~S/N) >! P(~S/T).
I will focus on Step 1: P(~S/N) > P(~S/T). That is to say: abstracting from the
evidence for evolution, given our epistemic situation, we have better reason to believe
the denial of special creationism on the assumption that naturalism is true than on
the assumption that theism is true. And what is that reason, exactly? Here are
Draper’s words:
Since naturalism entails that no God exists, it entails that special creationism is false. Thus, the
falsity of special creation is antecedently certain on naturalism: P(~S/N) = 1. But on theism
special creationism might, for all we know antecedently, be true: P(~S/T) < 1.31

The argument here seems to be that, first of all, abstracting from our evidence for
evolution, we have maximally good reason to believe the denial of special creationism
given naturalism. However, secondly, abstracting from our evidence from evolution,
for all we know, special creationism is true given theism, and so we do not have
maximally good reason to believe the denial of special creationism given theism.
Therefore, abstracting from the evidence for evolution, we have better reason to
believe the denial of special creationism on the assumption that naturalism is true
than on the assumption that theism is true. The first thought—the thought that,
abstracting from our evidence for evolution, we have maximally good reason to
believe the denial of special creationism given naturalism—seems to me to be right.
That is because, by definition, naturalism entails the denial of special creationism.
Therefore, we have just as good a reason to expect that special creationism is false
given naturalism as we have to expect that Joe is unmarried given that Joe is a
bachelor—which is to say we have maximally good reason to believe it, or as close to
maximally good reason to believe it as to make no difference for our purposes.
The second thought, however—the thought that, abstracting from our evidence for
evolution, for all we know, special creationism is true given theism, and so we do not
have maximally good reason to believe the denial of special creationism given
theism—does not seem clearly right. For it is not just obvious that, for all we
know, special creationism is true given theism. Whether that’s so depends on what
we know. Of course, we can’t say that, by definition, theism entails the denial of
special creationism. But, even so, theism might yet entail the denial of special
creationism. One proposition might non-trivially entail the denial of another without
doing so by definition. We need, therefore, to consider arguments for the conclusion
that P(~S/T) = 1. Now, one might think that this is a desperate maneuver. After all,
on theism, God is omnipotent and omniscient, and so he would have the power and

31
Draper (2008a [1997]: 210).
  -

wisdom to create directly by way of intervention in the natural order. So, abstracting
from the evidence for evolution, this gives us at least some reason to expect God to
specially create, and so we do not have maximally good reason to believe the denial of
special creationism given theism.
This argument contains a non sequitur. For, even though, on theism, God would
have the power and wisdom to create directly by way of intervention in the natural
order, we might have maximally good reason to think that, as a matter of necessity,
he would not exercise his power and wisdom toward that end. If that were the case,
then, abstracting from the evidence for evolution, we would have maximally good
reason to expect God not to specially create. So there is no avoiding the question: do
we have any such reason? I know of only one candidate worth considering.
Leibniz famously argued that, contrary to Newton and his followers, according to
whom God intervened occasionally to maintain the orbits of the planets, God would
not need to intervene since, at creation, he would have sufficient wisdom and power
to decree that the denizens of the natural world would possess those properties and
powers necessary to achieve his purposes in creating them in the first place, without
ever having to step in later to tinker here and there. “Whoever thinks otherwise must
needs have a very mean notion of the wisdom and power of God.”32 Leibniz’s claim is
relevant to our concerns. For if it is true, then, on the assumption that there is a God
whose purposes call for life to come into existence and for complex organisms to
exist, he would—right from the start, so to speak—use his power and wisdom to
decree initial conditions, laws, and whatever else he deemed necessary to ensure that
life arose in the natural order of things, in all its complexity and diversity, without
any need for later intervention on his part to keep things “on track.”
The point here is not just that God’s purposes in creating complex living things
would be better served by a law-governed world, and so without his fixing it now and
again, although that is a point worth making. Indeed, there seem to be great goods
that require nomic regularity, e.g. the mutual interdependence of living things, free
and effective choice, genuine love relationships, and the satisfaction of intellectual
investigation. Rather, the point is that if God were to achieve his purposes in creating
living things by bestowing on them those properties and powers necessary to achieve
his purposes, without his ever having to step in later and make adjustments—well,
that would be a great good in itself, in addition to whatever other goods might be
served by the arrangement. How might we think of this good? Wrapping up his
discussion of Leibniz, Malebranche, and others on the matter, Michael Murray
describes it like this:
[T]here is something grand, beautiful, and artful about a universe which contains within it
everything that is necessary in order for it to bring about the results God intends for it. God
could cause every event that we see in the natural world directly. But a powerful and rational
designer would . . . display his power and reason far more manifestly in a universe which is
itself a machine-making machine. A universe which achieves the ends God has for it in this
self-contained fashion does as much to express the glory of its creator as do the end-products
of the creative process.33

32 33
Leibniz (1956 [1715–16]: 11–12). Murray (2008: 146).
      

So a universe populated only with things whose properties and powers suffice to
render it unnecessary for God to tinker with it later is a world that is maximally
“grand, beautiful, and artful,” manifesting the gloriousness of its creator, his power,
and his wisdom. Let ‘a Leibnizian natural world’ designate a natural world populated
only with things whose properties and powers suffice to render it unnecessary
for God to tinker with them later in order to achieve his purposes. Now consider
the following argument, call it the Leibnizian argument for P(~S/T) = 1, or just the
Leibnizian argument, for short.
1. Necessarily, if God exists, then, at creation, God has the power and wisdom to
create a Leibnizian natural world.
2. Necessarily, if, at creation, God has the power and wisdom to create a Leibniz-
ian natural world, then he will exercise his power and wisdom toward that end,
provided that he has good enough reason to do so.
3. Necessarily, God has good enough reason to do so (if he exists).
4. So, necessarily, if God exists, then, at creation, he will exercise his power and
wisdom to create a Leibnizian natural world.
Of course, as a matter of necessity, if God exercises his power and wisdom in this
way, then that is how the natural world will be. Moreover, since what has just been
said about a natural world can with equal merit be said about any biological order
that is a part of it, it follows from (4) that,
5. Necessarily, if God exists, then God will not specially create.
No doubt, Leibniz would concur: “Whoever thinks otherwise must needs have a very
mean notion of the wisdom and power of God.”
The proponent of the Leibnizian argument will regard it as maximally good reason
for her to expect that God will not specially create given that theism is true. That is to
say, she will infer that, relative to her and her own epistemic situation, P(~S/T) = 1.
She will also point out that, as an added bonus, her expectation that God will not
specially create is somewhat confirmed—or at least not as yet disconfirmed—by
observed regularities in nature as well as the success of the physical sciences in
offering purely naturalistic explanations for the many processes and events in
the natural order it has come to understand, all independent of the evidence for
evolution.34 Draper considers something along these lines, specifically an argument
that he finds in the writings of Diogenes Allen and Howard J. Van Till, but he says he
doesn’t find it at all convincing. He writes:
Perhaps the idea is that, just as a perfectly rational car manufacturer would produce a car that
never needed its gas tank filled or its air filter replaced, a perfectly rational creator would make
a universe that ran on its own. But such a car would be preferable because filling up with gas or
replacing parts has a cost in terms of time, energy, and so on. An omnipotent and omniscient
creator wouldn’t have such worries. In general, what counts as a rational or perfect or defective
universe depends on the creator’s goals. What goal or plan of God would be better served by a
universe in which God never intervened?35

34
For more on the Leibnizian conception of God’s natural providence, see Murray (2003) and (2008).
35
Draper (2008a [1997]: 211).
  -

Whether these words suffice to dispose of Allen and Van Till’s argument need not
concern us, for they do nothing to tarnish the Leibnizian argument. The goal of God
that would be better served by a natural world in which he never intervened is, in
Murray’s words, a world that is maximally “grand, beautiful, and artful,” one that
better manifests God’s power and wisdom in achieving the other ends for the sake of
which he created it. Notice also that such ends as those mentioned earlier—the
mutual interdependence of living things, free and effective choice, genuine love
relationships, the satisfaction of intellectual investigation—would be ensured by a
Leibnizian natural world.
I anticipate the following objection to the Leibnizian argument. “Although we are
well-situated to tell whether premises (1) and (2) are true and we are well-situated to
discern the argument’s validity, we are in no position to tell whether premise (3) is
true. After all, what if there is some greater good that would be realized if God were to
intervene directly in the natural order and bring about his purposes in creation by
occasionally tinkering with the properties and powers of some of the inhabitants of
the natural world? It seems that, necessarily, God has good enough reason to exercise
his power and wisdom to create a Leibnizian natural world only if it is absolutely
impossible for there to be such a greater good. But we are in no position to say that it
is absolutely impossible for there to be such a greater good. So we are in no position
to say whether, necessarily, God has good enough reason to exercise his power and
wisdom to create a Leibnizian natural world. That is, we are in no position to say
whether premise (3) is true. Therefore, given our epistemic situation, the Leibnizian
argument is not a maximally good reason for anyone—including its proponent—to
believe the denial of special creationism given theism. Absent any other consider-
ation, we must conclude that P(~S/T) < 1”.
By way of reply, I agree with everything up to the ‘Therefore’. At that point,
however, the objection goes sour. For, even though we are in no position to say
whether premise (3) is true, it does not follow that, given our epistemic situation, the
Leibnizian argument is not a maximally good reason for us to believe the denial of
special creationism given theism. The most that we can infer from the fact that we are
in no position to say whether or not premise (3) is true is that we are in no position to
say whether, given our epistemic situation, the Leibnizian argument is a maximally
good reason for us to believe its conclusion. That is, the most that we can infer is that
we are in no position to say whether P(~S/T) = 1. Let me explain.
My explanation will be from my own point of view. During the course of my
explanation, I will dispense with the fiction that we share a common epistemic situation
that renders it appropriate for me to say things like “we are in no position to say whether
or not (3) is true” and “we are in no position to say whether, given our epistemic
situation, the Leibnizian argument is a maximally good reason for us to believe its
conclusion” and so on. I can’t speak for you. You’ll have to speak for yourself.
So consider the following proposition:
• I have maximally good reason to believe the conclusion of the Leibnizian
argument, given my epistemic situation.
Is it true? Well, what is my “epistemic situation”? Here are some salient items: my
understanding and awareness of the premises and inferences that constitute the
      

Leibnizian argument, my judgment that the inferences are valid, my judgment that
premises (1) and (2) are true, my judgment that premise (3) is the only questionable
thing about the argument, my judgment that I am in no position to tell whether or
not (3) is true, my belief that, for any noncircular sound argument, it is a maximally
good reason for me to believe its conclusion, and my judgment that, aside from
premise (3), the Leibnizian argument is a noncircular sound argument for its
conclusion. In light of the fact that my epistemic situation includes these items, is
the proposition indented just above true? Is it true that I have maximally good reason
to believe the conclusion of the Leibnizian argument, given my epistemic situation?
I can’t tell. That’s not because I can’t tell whether I have the Leibnizian argument.
Of course I have it. As I said, I understand each of the statements that constitute it,
and I’m aware of each of its premises and inferences. Moreover, I wrote it out a
couple of pages back, and I’m staring at it on a page next to me, and I have several
electronic copies of it in my two computers. So I have it. Rather, I can’t tell whether it
is a maximally good reason for me to believe its conclusion because my epistemic
situation puts me in no position to say whether it is such reason. For all I can tell, it is;
then again, for all I can tell, it isn’t. I am in the dark about the matter. I am in the dark
about whether P(~S/T) = 1.
Perhaps one will object. “We’re talking here about reasons for one to believe
something, given one’s epistemic situation. In general, if one of the premises of an
argument is inscrutable to a person, then, given her epistemic situation, how can it
provide her with a good reason, much less a maximally good reason, to believe its
conclusion? It can’t. And the same goes for you and your epistemic situation vis-à-vis
the Leibnizian argument. Given how you have described your epistemic situation—
and, more specifically, given the inscrutability of premise (3) to you—the Leibnizian
argument does not give you a maximally good reason to believe its conclusion.” By
way of reply, I never said that the Leibnizian argument gave me or provided me with a
maximally good reason to believe its conclusion. Rather, I said that I have it and that
my epistemic situation does not permit me to tell whether it is a maximally good
reason for me to believe its conclusion. Of course, if the Leibnizian argument did give
or provide me with a reason to believe its conclusion, then I’d be rational, justified,
warranted, or otherwise epistemically up to snuff in believing its conclusion on the
basis of it. But I’m not. So it neither gives nor provides me with any such thing. My
conceding that point, however, is right at home with my being in the dark about
whether it is a maximally good reason for me to believe its conclusion.
The objector might persist. “Wouldn’t the Leibnizian argument be much better if
you were able to tell whether premise (3) is true? Of course it would. But if it would be
better, then it’s not maximally good. Therefore, the Leibnizian argument is not a
maximally good reason for you to believe its conclusion.” I disagree. To suppose that
it—the Leibnizian argument—would be better if I was able to tell whether premise (3)
is true is like supposing that Mt. Adams basking in a late August alpenglow would be
more beautiful if Trevor Thomas had been able to see it when he walked its western
flank in 2010.36 It is like supposing that an instance of modus ponens would be

36
See <http://www.blindhikertrevorthomas.com/>.
  -

improved if your student who can’t distinguish it from the fallacy of affirming the
consequent gained the ability to do so. The goodness of an argument is not the least
bit affected by the deficiencies of those who entertain it, or the removal of those
deficiencies.
The objector might persist some more. “Look. Even if it is not the case that the
Leibnizian argument would be better if you were able to tell whether premise (3) is
true, it is not a maximally good reason for you to believe its conclusion. That’s
because, in general, an argument is a good reason for someone to believe its conclu-
sion only if she is in a position to discern the truth of all of its premises. But, as you
admit, you are in no such position with respect to premise (3) of the Leibnizian
argument. Thus, it is not a good reason for you to believe its conclusion, given that
your description of your epistemic situation is accurate. A fortiori, given your
epistemic situation, it is not a maximally good reason for you to believe its conclu-
sion.” Once again, I disagree. I doubt that, “in general, an argument is a good reason
for someone to believe its conclusion only if she is in a position to discern the truth of
all of its premises.” Whether an argument is a good reason for someone to believe its
conclusion has nothing at all to do with her. Rather, it has to do with it: with whether
the premises are true, with whether they adequately support its conclusion, and the
like. If an argument has good-making features such as these, features that do not have
anything to do with us, then it is a good reason for us to believe its conclusion—even if
some limitation of ours hinders us from discerning those features. In the same way, Mt.
Adams basking in a late August alpenglow is a beautiful sight for Trevor Thomas to see
even though he is blind. Likewise, the conditional corresponding to modus ponens is a
good thing for your student to understand even if she lacks the capacity to do so.
There are other objections along these lines that we might consider. But, so far as
I can see, they all fail to take into account what it is for an argument to be a good reason
for us to believe its conclusion. The bottom line is that, given our epistemic situation,
we can be in the dark about whether some argument is a good reason for us to believe
its conclusion. That’s how it is for us and the Leibnizian argument; or, at any rate, that’s
how it is for me. (That is also how it is for me and the best versions of the ontological
argument, where everything rests on the possibility premise, which—like premise (3)
of the Leibnizian argument—I am in no position to tell whether it is true or false.)
Earlier, I mentioned Draper’s argument for thinking that it is false that P(~S/T) = 1.
It went like this: abstracting from our evidence for evolution, for all we know, special
creationism is true given theism, and so we do not have maximally good reason to
believe the denial of special creationism given theism. I remarked that it is not just
obvious that, for all we know, special creationism is true given theism. Whether that’s
so depends on what we know. I then turned to the Leibnizian argument. As I have
indicated, however, that argument leaves intact Draper’s premise that, abstracting
from our evidence for evolution, for all we know, special creationism is true given
theism. So what should we make of Draper’s argument? I think we should reject the
inference. For, even if it is true that, abstracting from our evidence for evolution, for
all we know, God might specially create, we might yet possess, unbeknownst to us, a
maximally good reason for us to believe that he wouldn’t. That’s what our position
vis-à-vis the Leibnizian argument demonstrates; or, more accurately, that’s what my
position vis-à-vis that argument demonstrates.
      

To recapitulate, and to revert to the fiction that we share the same epistemic
situation: if we are in no position to tell whether, given our epistemic situation, the
Leibnizian argument is a maximally good reason for us to expect the denial of special
creationism given theism, then we are in no position to tell whether, given our epistemic
situation, our reason to expect the denial of special creationism given naturalism is
better than our reason to expect the denial of special creationism given theism. That is,
we are in no position to tell whether P(~S/N) > P(~S/T), which is the conclusion of Step
1—unless, of course, we have something else to settle the matter. Draper seems to think
we do have something else to settle the matter. At any rate, he offers an argument that,
by his lights, constitutes “very strong antecedent reason for believing that God did create
at least some complex life independently,” given that theism is true.37 Let’s take a look at
this “very strong reason.” Here it is in his own words:
Theism implies an extreme metaphysical dualism—a mind existed prior to the physical world
and was responsible for its existence. Thus on the assumption that theism is true, it is
antecedently [epistemically] likely that minds are fundamentally nonphysical entities and
hence that conscious life is fundamentally different from nonconscious life. But this in turn
makes it [epistemically] likely that conscious living things are not just the genetically modified
descendents of nonconscious living things—that conscious life was created independently.
And since special creationism is defined as the position that at least some complex life was
created independently, it follows that, on the assumption that theism is true, it is antecedently
[epistemically] likely that special creationism is true.
The dualism inherent in theism may explain why so many theists were drawn to the idea of
special creationism before (and in many cases after) the evidence for evolution was discovered.
For this dualism supports a dualistic view of human nature—a view that must have made the
idea that we are the effect of altering the nucleic acids of single-celled organisms seem
ludicrous. Offspring don’t have to be identical to their parents, but surely genetic change
can’t result in fundamental metaphysical lines being crossed! Thus, even if we know by past
experience that God, assuming he exists, generally doesn’t intervene in nature, the sort of
metaphysics presupposed by theism makes it antecedently [epistemically] likely that God did
intervene in the physical world in order to create a mental world within it. So it’s hardly
surprising that, before Darwin, many theists were special creationists.38

I think that we can fairly reconstruct the argument of this passage as follows:
1. Necessarily, if theism is true, then an immaterial mind existed prior to the
physical world and is responsible for its existence.
2. So, abstracting from the evidence for evolution, we have very good reason to
expect that, given theism, minds are fundamentally immaterial. (from 1)
3. Necessarily, if minds are fundamentally immaterial, then conscious living
things are fundamentally different from nonconscious living things.
4. So, abstracting from the evidence for evolution, we have very good reason to
expect that, given theism, nonconscious living things are fundamentally differ-
ent from conscious living things. (from 2 and 3)

37 38
Draper (2008a [1997]: 212). Draper (2008a [1997]: 212).
  -

5. Necessarily, if nonconscious living things are fundamentally different from

conscious living things, then conscious living things are not just the genetically
modified descendents of nonconscious living things but rather were created
independently by God.
6. Necessarily, if conscious living things are not just the genetically modified
descendents of nonconscious living things but rather were created by God
independently, then special creationism is true.
7. So, abstracting from the evidence for evolution, we have very good reason to
expect that, given theism, special creationism is true. (from 4–6)
What should we make of this line of thought? Well, premises (1) and (6) are true by
definition, and (4) and (7) seem to follow from the indicated premises. That leaves
(2), (3), and (5). Two ideas are lurking in these three premises, the idea of something
of one sort being “fundamentally” of another sort and the idea of something of one
sort being “fundamentally different” from something of another sort. Draper does
not expound on these ideas, so we cannot be sure what he means by these premises.
Nevertheless, what he has in mind seems clear enough.
Draper seems to have in mind two principles. The first principle is that
• for some pairs of categories, F and G, the Fs are fundamentally Gs just in the case
that, necessarily, if something is an F, then it is a G;
that is, nothing can be an F unless it is a G. Dogs are, in this sense, fundamentally
animals; nothing can be a dog unless it’s an animal. Numbers, in this sense, are
fundamentally immaterial; nothing can be a number unless it’s immaterial. The
second principle is that, for some pairs of categories, F and G, the Fs are funda-
mentally different from the Gs just in the case that it is absolutely impossible for
there to be a series of modifications of something that is an F such that, by virtue of
the cumulative effect of that series of modifications alone, something that is a
G comes to be. And the modifications in the series can be as gradual as you like
and take as much time as you like and involve as many things as you like. For the
sake of brevity, let’s stipulate that the necessary and sufficient condition just laid
down for the Fs being fundamentally different from the Gs holds between the Fs
and the Gs if and only if nothing can bridge the gap between them. This seems to
be in the spirit of what Draper says when he speaks of “fundamental metaphysical
lines” that “can’t be crossed,” only I prefer the metaphor of a metaphysical gap to
that of a metaphysical line. Thus, the second principle can be put more succinctly
as follows:
• for some pairs of categories, F and G, the Fs are fundamentally different from the
Gs just in the case that nothing can bridge the gap between the Fs and the Gs.

Perhaps the categories of abstracta and concreta are uncontroversial instances of this
principle. With all this in hand, we can see that premise (5) of Draper’s argument is
true. It is necessary that, if nothing can bridge the gap between nonconscious living
things and conscious living things, then a series of genetic modifications can’t bridge
the gap either. But is it true that nothing can bridge the gap between nonconscious
living things and conscious living things?
      

Draper seems to think so—at least if nothing can be a mind unless it is immaterial.
That’s just his premise (3), which, given the second of the two principles I have
imputed to him, is to be understood as this:
3’. Necessarily, if nothing can be a mind unless it is immaterial, then nothing can
bridge the gap between nonconscious living things and conscious living things.
What should we make of (3’)? I think it is false. For even if nothing can be a mind
unless it is immaterial, some things bridge the gap between nonconscious living
things and conscious living things. For example, some living things are awake and
then sleep soundly, after which time they wake up; and some living things get
knocked out, after which time they regain consciousness. In these cases, there is a
nonconscious living thing and there is a series of modifications of it—the soundly
sleeping or knocked-out organism—such that, by virtue of the cumulative effect of that
series of modifications alone, a conscious living thing comes to be. And this can be the
case even on the assumption that nothing can be a mind unless it is immaterial.
Perhaps Draper didn’t mean (3’) by (3). Perhaps he meant
3’’. Necessarily, if nothing can be a mind unless it is immaterial, then nothing can
bridge the gap between living things that lack the capacity for consciousness and
living things that possess the capacity for consciousness.
(3’’) avoids the counterexamples posed by sound sleepers and the knocked-out. Unfor-
tunately, however, it is false. For even if nothing can be a mind unless it is immaterial,
some living things that lack the capacity for consciousness naturally develop in such a way
that they come to possess that capacity. You, for example, are a good illustration of this
fact. That’s because you are an animal and, as such, you were once a fetus that lacked the
capacity for consciousness. True enough, at that time, you had the capacity to develop
into something that has the capacity for consciousness. But we must not confuse the
capacity to develop into something that has the capacity for thus-and-so with the capacity
for thus-and-so. Now I grant that the claim that you are an animal is contentious, perhaps
even offensive. But the same point could be made with respect to some livings things
whose status as animals is indisputable, for example, the Olympic marmot.39 For any
Olympic marmot pup that survives until it emerges from its birth burrow (typically, at
about eight weeks after conception), there was a time when it was a fetus and lacked the
capacity for consciousness. That time was followed by a series of modifications of that
fetus such that, by virtue of the cumulative effect of that series of modifications alone,
something that possessed the capacity for consciousness came to be. Gap bridged.
Perhaps Draper did not mean (3’’) by (3). Perhaps he meant
3’’’. Necessarily, if nothing can be a mind unless it is immaterial, then nothing can
bridge the gap between one species of living thing whose members lack any capacity
for consciousness and another species of living thing whose members tend to possess
a capacity for consciousness.

39
See <http://en.wikipedia.org/wiki/Olympic_marmot>.
  -

(3’’’) avoids the counterexamples that demonstrate the falsity of (3’) and (3’’);
moreover, it is not susceptible to easy counterexample in the way the latter are, so
far as I can see. Nevertheless, worries stemming from the doctrine of the essentiality
of origins aside,40 I do not see why it is not possible for an Olympic marmot to be a
product of a series of modifications of a large enough puddle of aerobacter aerogenes.
Of course, we are presently incapable of creating any such series ourselves; but since,
at bottom, the specialized cells, tissues, and organs of a marmot are nothing but
elementary particles arranged in a certain fashion, while undergoing the processes
that unify them and what they constitute into a marmot life, I don’t see why there
could not be a series of modifications of a puddle of aerobacter aerogenes the end
result of which was a marmot, or at least a marmot-like living thing.41
But whether I’m right about that or not, the deepest difficulty with Draper’s
argument for believing that, on the assumption that theism is true, God directly
created at least some complex life, is not premise (3) but premise (2). Given the first
principle above that I imputed to Draper, premise (2) reads as follows:
2’. So, abstracting from the evidence for evolution, we have good reason to expect
that, given theism, nothing can be a mind unless it is immaterial.
What should we make of (2’)? Naturally, we will agree that, given Draper’s definition
of “theism,” it is necessary that, if theism is true, then an immaterial mind existed
prior to the physical world and is responsible for its existence. That’s what premise
(1) states. And Draper explicitly infers (2) from (1): that’s what the word “thus”
indicates in the passage quoted at length above. But, how does (1) even begin to
suggest that (2’) is true? How does the fact that, necessarily, if theism is true, then an
immaterial mind existed prior to the physical world and is responsible for its
existence, even begin to suggest that, abstracting from the evidence for evolution,
we have good reason to expect that, given theism, nothing can be a mind unless it is
immaterial?
Suppose we grant that minds are things that are capable of consciousness—
including a subjective point of view, as Thomas Nagel never tires of reminding
us—and that they are capable of rationality, intention, and action. In that sense of
the word “mind,” you and I are both minds, as is God (if God exists). Furthermore,
let us suppose that God exists and that, therefore, at least one mind is immaterial.
Okay. So far so good. Should we infer that, therefore, abstracting from the evidence
for evolution, all minds—including ourselves—are immaterial? Is that the inference
Draper wishes us to draw? Of course, we’re all familiar with arguments for the
conclusion that nothing could be a mind unless it is immaterial; and we’re all equally
familiar with arguments for the conclusion that we are merely material beings. But if
Draper meant to take a stand on this debate, he would have mentioned it. No, I rather
think that the suggestion he means to put forward—and I see no hermeneutical
option here—is that if an immaterial mind shows up in the first scene of the first

40
See Kripke (1972: 114 n56), for an argument for the doctrine. For critical discussion, see Cameron
(2005).
41
There appears to be progress, however. See Pennisi (2010), Venter et al. (2010), and the publications
at <http://www.jcvi.org/cms/research/projects/first-self-replicating-synthetic-bacterial-cell/publications/>.
      

act of the only play on the ontological stage, then that fact all by itself gives us “very
good reason” to expect that material minds will not show up in any scene of any act
in that play.
How are we to make sense of this inference? Perhaps it’s an inference by analogy.
In the Game of the Nature of Mind, the opponents are those who think that human
minds are material and those who think that minds in general, including human
minds, are immaterial: the materialists against the immaterialists. If we go into the
game spotting the immaterialists a divine mind, then the score is 1–0 in favor of the
immaterialists before the materialists even go to bat. Of course, as everyone knows,
68.9 percent of the time, baseball teams that score first win the game. Thus, by
analogy, we have good reason to believe that the immaterialist will win. That is, we
have good reason to believe that, abstracting from the evidence for evolution, all
minds—including human minds—are immaterial given theism. Is that Draper’s
“very good reason”?
Of course, I jest. But I do not jest without a point. I find it enormously difficult to
see how it can be that the mere existence of an immaterial mind that existed prior to
the physical world and that is responsible for its existence constitutes any reason at
all to suppose that, abstracting from the evidence for evolution, nothing can be a
mind unless it is immaterial. And so I not only have the difficulty of accepting
premise (3), I have the difficulty of accepting premise (2) as well.

12.6 Conclusion
To establish premise (2) of his evolutionary argument for atheism, Draper set out to
do two things: show that P(~S/N) >! P(~S/T) and show that P(E/~S&N)  P(E/
~S&T). Draper’s argument for P(~S/N) >! P(~S/T) included the claim that P(~S/N) >
P(~S/T). While we rightly agree that P(~S/N) = 1, Draper has not established that P
(~S/T) < 1. Indeed, by my lights, given the Leibnizian argument, we are in no position
to say whether P(~S/T) = 1, absent any further consideration. What further consid-
eration Draper has to offer, however, is a failure in my opinion. So far as I can see,
therefore, Draper’s evolutionary argument for atheism is not good reason for us to
think that evolution provides significant evidence for atheism. Of course, I may have
interpreted Draper’s argument incorrectly. Thus, I grant that what he in fact said,
unlike what I understood him to say, might well constitute good reason for us to
think that evolution provides significant evidence for atheism. And, of course, even if
I have understood and assessed his argument correctly, evolution might yet provide
significant evidence against theism. But if it does, then, if I have understood and
assessed Draper’s argument correctly, we need to look elsewhere to see how evolution
provides significant evidence—or any evidence at all—for atheism.

Acknowledgments
For comments on earlier drafts, I thank Frances Howard-Snyder, Hud Hudson, John A. Keller,
and Dennis Whitcomb.
  -

References
Cameron, R. (2005), ‘A Note on Kripke’s Foonote 56 Argument for the Essentiality of Origins’,
in Ratio 18: 262–75.
Chomsky, N. (2005), ‘Three Factors in Language Design’, in Linguistic Inquiry 36: 1–22.
Diamond, J. (1992), The Third Chimpanzee (Harper & Row).
Draper, P. (1989), ‘Pain and Pleasure: An Evidential Problem for Theists’, in Noûs 23: 331–50.
Reprinted in D. Howard-Snyder (ed.), The Evidential Argument from Evil (Indiana University
Press) (1996): 12–29.
Draper, P. (2008a [1997]), ‘Evolution and the Problem of Evil’, in L. Pojman and M. Rea (eds.),
Philosophy of Religion: An Anthology, 5th edn. (Wadsworth): 207–19. (Originally published
in 1997.)
Draper, P. (2008b), ‘The Argument from Evil’, in P. Copan and C. Meister (eds.), Philosophy of
Religion: Classic and Contemporary Issues (Blackwell Publishing), 142–55.
Draper, P., and Dougherty, T. (2013), ‘Explanation and the Problem of Evil’, in J. McBrayer
and D. Howard-Snyder (eds.), The Blackwell Companion to the Problem of Evil (Wiley
Blackwell), 67–82.
Hájek, A. (2012), ‘Interpretations of Probability’, in E. N. Zalta (ed.), The Stanford Encyclopedia
of Philosophy (Winter 2012 Edition), <http://plato.stanford.edu/archives/win2012/en
tries/probability-interpret/>.
Kretzmann, N. (1998), The Metaphysics of Creation: Aquinas’s Natural Theology in Summa
Contra Gentiles II (Oxford University Press).
Kripke, S. (1972), Naming and Necessity (Harvard University Press).
Leibniz, G. W. (1956 [1715–1716]), in H. G. Alexander (ed.), The Leibniz-Clarke Correspond-
ence (Manchester University Press).
McBrearty, S., and Brooks, A. (2000), ‘The Revolution that Wasn’t: A New Interpretation of
the Origin of Modern Human Behavior’, in Journal of Human Evolution 39: 453–63.
Murray, M. (2003), ‘Natural Providence (or Design Trouble)’, in Faith and Philosophy
20: 307–27.
Murray, M. (2008), Nature Red in Tooth and Claw: Theism and the Problem of Animal
Suffering (Oxford University Press).
Papineau, D. (2009), ‘Naturalism’, in E. N. Zalta (ed.), The Stanford Encyclopedia of Philosophy,
<http://plato.stanford.edu/archives/spr2009/entries/naturalism/>.
Pennisi, E. (2010), ‘Synthetic Genome Appears to Bring New Life to Bacterium’, in Science 328/
5981: 958–9, <http://www.sciencemag.org/content/328/5981/958.full.pdf>.
Plantinga, A. (2011), Where the Conflict Really Lies: Science, Religion, and Naturalism (Oxford
University Press).
Van Inwagen, P. (1997), ‘Probability and Evil’, in The Possibility of Resurrection and Other
Essays in Christian Apologetics (Westview Press).
Van Inwagen, P. (2010), ‘A Kind of Darwinism’, ‘Darwinism and Design’, ‘Science and
Scripture’, in M. Stewart (ed.), Science and Religion in Dialogue, Vol. 2 (Wiley Blackwell),
813–46.
Venter, J. C. et al. (2010), ‘Creation of a Bacterial Cell Controlled by a Chemically Synthesized
Genome’, in Science 329/5987: 52–6, <http://www.sciencemag.org/content/329/5987/52>.
 13
Must Anselm be Interpreted as
a Meinongian?
Lynne Rudder Baker

Peter Van Inwagen recently published an essay entitled “Three Versions of the Onto-
logical Argument” (Van Inwagen 2012). The three versions he labeled “The Meinongian
Version,” “The Conceptual Version,” and “The Modal Version,” respectively. I want to
propose a fourth version, which, for want of a better label, I’ll call “The Cognitive-Ability
Version.”
What distinguishes the Cognitive-Ability version (hereafter, the “CA-version”) is
that, unlike all the other versions, the CA-version requires an empirical presuppos-
ition that the participants in the argument have a peculiar cognitive ability. The
cognitive ability is that human beings can think and talk about things independently
of whether they exist.1 The ability is exercised in everyday activities, linguistic and
otherwise: children anticipate the arrival of Santa Claus and leave out cookies for
him; a young reader visits London and seeks Sherlock Holmes’s residence; Ponce de
León sails to find the Fountain of Youth; a naive scholar decides what to do if she
found Gyges’s ring. And on and on.
Human beings clearly do have the ability to think and talk about putative things
that do not exist. Along with van Inwagen, I assume that there is some non-
Meinongian account of how this works. However, my argument does not depend
on any metaphysical analysis of the CA-assumption; it depends only on the truth of
claims like ‘Children leave out cookies for Santa Claus’ or ‘The Fountain of Youth is
greater than the Fountain of Boredom.’ At the level of ordinary language, it is difficult
to see how such claims could be false.
My aim here is to show that the CA-assumption makes room for an Anselmian
argument that is sound, is non-question-begging, and avoids any Meinongian
ontology. Van Inwagen says that “Anselm’s argument presupposes, and essentially

1
I intend the phrase ‘talks about things independently of whether they exist’ to be a unit, so that ‘things’
carries no ontological commitments; we may prefer to use the phrase ‘putative things.’ Similar consider-
ations apply to ‘something’ contained in the phrase ‘talks about something independently of whether it
exists.’ Both phrases are to be neutral with respect to the existence of what is talked about. When
‘something’ is replaced by a name or description, the issue arises of whether the name or description has
a denotation.
   

presupposes, an ontology that is . . . Meinongian” (Van Inwagen 2012: 8). I shall argue
otherwise. If I am right, the CA-version of Anselm’s argument in Proslogion II is not
captured by any of van Inwagen’s versions of the ontological argument.
To begin, I’ll explain the CA-assumption and its roles in my interpretation of
Anselm’s argument; then, I’ll state the argument and show the CA-assumption
affords a non-Meinongian interpretation of Anselm’s argument.

13.1 The CA-Assumption

The CA-assumption is that human beings have the ability to think and talk about
putative things independently of whether they exist.2 As I mentioned earlier, we are
very familiar with this ability, although it is little understood.3 In order to argue non-
question-beggingly about the existence of God, all disputants must be able to talk
about God independently of whether he exists. The expression ‘that than which
none greater can be conceived’ is used to pick out the putative entity in question.
(Gareth Matthews calls this expression a “referential peg”. Matthews 2005: 88) The
CA-assumption allows Anselm and the Fool to talk about that than which none
greater can be conceived without any commitment to the existence of that than
which none greater can be conceived.
One way to see how to achieve the needed neutrality with respect to that than
which none greater can be conceived is to note that Anselm and the Fool are in
circumstances in which it is undetermined whether there is anything that satisfies
‘that than which nothing greater can be conceived.’ Since we have countless
instances of thinking or talking about (putative) things whether they exist or
not—Homer, Snow White, Hamlet, Lancelot—we can take it to be a datum that ‘S
speaks or thinks about the F’ does not entail ‘There exists an F about which S speaks
or thinks.’4 Any correct and comprehensive semantic theory must account for this
obvious fact.
Hence, Anselm and the Fool each can talk about that than which nothing greater
can be conceived without either’s begging the question against the other. If there
exists something than which nothing greater can be conceived, then God exists; and if
not, then God does not exist. Moreover, Anselm can mount his argument without
any presuppositions about a concept or an idea of God. The argument does not
concern words, or concepts or ideas. It is ontological: it concerns the existence of that
than which none greater can be conceived (Baker and Matthews 2010: 31).

2
This section is prompted by an email correspondence with Peter van Inwagen. Van Inwagen uses the
word ‘item’—the most “ontologically neutral” word that he can think of—as “the most general count-
noun.” In his use, “everything without qualification, is an item; an item is anything that can be referred to
with a pronoun” (Van Inwagen 2012: 4). Van Inwagen, however, like me, thinks that not all apparently
referring pronouns do so refer.
3
I do not believe that other species have this cognitive ability to think about items independently of
whether they exist. The ability is a conceptual ability, and nonlinguistic animals lack the requisite concepts
to think about things independently of whether they exist.
4
So, exportation is not allowed. We talk and think about putative things regardless of whether or not
they exist in the absence of any kind of ontological argument.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   ? 

Now we can see how disputants can talk about something independently of
whether it exists. Speakers are talking about something independently of whether it
exists when they need to leave open the existence or nonexistence of whatever is
under discussion. Suppose that S uses ‘the F’ to speak about something independently
of whether the F exists; in that case, if ‘the F’ has a denotation, then S refers to it, and
if ‘the F’ has no denotation, then S refers to nothing. Nonetheless, even if S refers to
nothing, we can consider ‘the F’ still to have meaning; and S could still use ‘the F’ to
single out a topic for discussion. So,
S speaks about something independently of whether it exists only if 5 S uses a
description, ‘the F’, and either:
(i) 9x!Fx & S refers to the entity that is F
or
(ii) ~9xFx & S refers to nothing, but use of ‘the F’ may introduce a topic of
discussion
We can talk about something independently of whether it exists regardless of
whether we believe that it does exist, or we believe that it does not exist, or we are
unsure one way or the other.
Let us apply the above schema to Anselm’s argument. I’ll assume, as I believe that
Anselm and the Fool would assume, that if there exists that than which none
greater can be conceived, then that being is God. Since ‘that than which none
greater can be conceived’ has been described as a “quasi-definition” of ‘God’
(Matthews 2005: 84) I’ll use the name and the description interchangeably for
convenience. Then,
S speaks about that than which none greater can be conceived, or of God,
independently of whether he exists only if S uses ‘that than which none greater can
be conceived’ and either
(i) there exists that than which none greater can be conceived and S refers to it,
or
(ii) it is not the case that there exists that than which none greater can be
conceived, and S refers to nothing; but use of ‘that than which none greater
can be conceived’ may introduce a topic for discussion.
Now we can say that S talks about the F independently of whether it exists only if
S uses a description in a way that satisfies either (i) or (ii). If there exists that than
which none greater can be conceived, then both Anselm and the Fool are referring to
God; but if there exists no such being, then both Anselm and the Fool fail to refer at
all. They are both using the description and the name with the same referent (God or
nothing, depending on whether God exists), and neither is begging the question
against the other.

5
We could turn this condition into a necessary and sufficient condition by confining its application to
contexts in which either there was no agreement about whether the putative thing talked about existed, or it
was agreed that the the putative thing talked about did not exist (e.g., was a fictional object).
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   

13.2 The Roles of the CA-Assumption in Anselm’s

Argument
The CA-assumption has two roles in my version of Anselm’s argument in Proslogion
II.6 First, it ensures that Anselm and the Fool can argue about the existence of that
than which none greater can be conceived, without either one’s begging the question
against the other. In the context of the argument, by talking about God independ-
ently of whether he exists, they both in effect suspend judgment on God’s existence.
Second, the CA-assumption supports the notion of unmediated causal powers—a
notion that can be used to formulate a Great-Making Principle needed for my
interpretation of Anselm’s argument.
Every particular object that exists in reality has what I’ll call ‘unmediated causal
powers.’ An unmediated causal power is the capacity to have an effect, to make
something happen, without the intervention of any (other) mind. Something has
unmediated causal powers if and only if it manifests causal powers without the
mediation of other people’s thoughts, depictions, and literature in which it is portrayed.
For example, a rock has the unmediated causal power to disturb the ambient air
current as it falls. By way of contrast to the rock, Pegasus has no unmediated causal
powers, but, by means of the mediation of people’s thoughts, depictions, and stories, he
can have effects on reality. People who think about Pegasus and portray him in
paintings and stories have unmediated causal powers. The contrast is underwritten
by the CA-assumption: even though Pegasus does not exist, we can meaningfully say
that he has no unmediated causal powers (or that Pegasus flies), because we can talk
about Pegasus independently of whether he exists. The CA-assumption allows us to
speak about putative things with no unmediated causal powers—e.g., putative things
that do not exist. It is a psychological fact that people attribute causal powers to what
they believe exists. These are mediated causal powers.
The distinction between putative things with unmediated causal powers and
things—or putative things—with only mediated causal powers allows formulation
of a Great-Making Principle. But first we need what I’ll call a (partial) definition:
(D) If x is a putative thing with only mediated causal powers, then y is an
otherwise similar thing to x if and only if y has unmediated causal powers and y has
the all the qualitative properties that are attributed to x.7
The Fool, presumably, would agree about which properties believers attribute to God,
but the Fool would insist that God has only mediated causal powers—mediated by
the unmediated causal powers of his believers. Now the Great-Making Principle:

6
Shortly before my friend and colleague Gareth Matthews died in 2011, he and I published several
papers on Anselm’s “ontological argument” for the existence of God in Proslogion II (Matthews and Baker
2010; Baker and Matthews 2010; Matthews and Baker 2011). Our collaboration launched me into further
thought about the ontological argument.
7
Or at least the most important properties as those attributed to x—but this qualification seems
irrelevant to the ontological argument.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   ? 

(G) For any putative thing with only mediated causal powers, an otherwise
similar thing is greater.8
Note that (G) applies to putative objects that do not exist. To take a nontheistic
example of (G), Pegasus has no unmediated causal powers, but if there were a flying
horse in reality, it would be greater than Pegasus. An otherwise similar thing to
Pegasus would be an actual entity that resembles Pegasus as described in myth. By
(G), such an actual flying horse would be greater than Pegasus.
There may be a question about (G). How can we quantify over things that don’t exist
without being a Meinongian or being a substitutionalist about quantification? Well,
ordinary language permits quantification over nonexistent objects (from the Seven
Dwarves to shadows). So, whatever “strict and literal” strictly philosophical language
we end up with must accommodate this fact somehow. (For my purposes, it doesn’t
matter how.) So, we can (and should) interpret (G) in terms of ordinary language.9 This
will remove the danger of making (G) trivially true, by virtue of a false antecedent.
Ontologically speaking, there are no entities with only mediated causal powers, but the
phenomenon of attributing causal powers to putative things that do not exist is familiar.
One more preliminary point: the definition of ‘an otherwise similar thing’ has an
important implication for the ontological argument. If b has only mediated causal
powers and can be conceived, then an otherwise similar thing to b that has unme-
diated causal powers can also be conceived. So, if the Fool’s conception of God is of a
being with only mediated causal powers, then an otherwise similar thing to the Fool’s
conception of God likewise can be conceived and hence can be thought of independ-
ently of whether he exists. This is so because if some putative thing can be conceived
(thought of) with certain mediated causal powers, an otherwise similar thing with the
same, but unmediated, causal powers can also be conceived (thought of). So, the idea
of ‘an otherwise similar thing to x’ brings with it a corollary to (G):
(C) If x has only mediated causal powers and can be conceived (thought of
independently of whether it exists), then an otherwise similar thing to x, which
has unmediated causal powers, can be conceived (thought of independently of
whether it exists).
To sum up: in addition to rendering Anselm’s argument non-question-begging, the
CA-assumption allows the formulation of (C), and a Great-Making Principle like
(G), which are required for the soundness of Anselm’s argument.

13.3 The CA-Argument

The structure of Anselm’s argument is a reductio ad absurdum. Anselm imagines an
opponent, the Fool, who says, ‘There is no God,’ and imagines engaging the Fool in a
dispute about whether God—that than which none greater can be conceived—exists

8
In the context of the ontological argument, ‘is greater’ always means ‘is greater in reality’; what is
merely thought to be greater is irrelevant to the argument. If the item in question does not exist, then an
otherwise similar item that has unmediated causal powers is greater than it.
9
This was suggested to me by John Keller.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   

(in reality or in re).10 Anselm and the Fool agree that that than which none greater
can be conceived exists in intellectu or in the understanding. ‘Exists in intellectu’ may
be interpreted as an instance of the the CA-assumption, not as some kind of existing.
We may take the phrase ‘exists-in-intellectu’ to be a unit from which the term ‘exists’
cannot be detached. To say that something exists in intellectu is a way of saying that
we have something in mind, as in “You are in my thoughts every day.” To exist in
intellectu is ipso facto to be conceived.
Anselm construes the Fool’s position to be that God exists in intellectu—i.e., that
the Fool can think and talk about God (independently of whether he exists)—but that
God does not exist in reality; and Anselm shows that the Fool’s position is contra-
dictory.11 The argument progresses from (i) speaking of God independently of
whether he exists to (ii) assuming that he does not exist in reality to (iii) deducing
a contradiction from (i) and (ii).
With the CA-Assumption on hand, if someone thinks about God independently of
whether God exists, then God exists in intellectu: it is a conceptual truth that to exist
in intellectu just is to be thought about or conceived independently of existence in
reality.
I shall present a pair of arguments, the Auxiliary Argument, with an instance of the
CA-Assumption as a premise, and the Main Argument. As we shall see, the conclu-
sion of the Auxiliary Argument plays a crucial role in the Main Argument. So, the
Auxiliary Argument, and hence the CA-Assumption, are not dispensable.
The Auxiliary Argument

(a) Anselm and the Fool agree to use ‘that than which none greater can be
conceived’ independently of whether he exists in reality. (Instance of CA-
Assumption)
(b) If (a), then that than which none greater can be conceived exists in intellectu.
(conceptual truth; see two paragraphs up)
∴ (c) That than which none greater can be conceived exists in intellectu. (1,2 MP)

The Main Argument

Let: b = that than which none greater can be conceived.
Oab = a is an otherwise similar thing to b.
(1) b exists in intellectu and b does not exist in re. (Fool’s reductio premise)
(2) If (1), then b has only mediated causal powers. (Conceptual truth)

10
I am following van Inwagen’s use of the Latin phrases.
11
Let me mention that Gaunilo’s “lost island” argument is not analogous to Anselm’s argument.
Nothing that is the island than which none greater can be conceived can exist in reality. If Gaunilo’s
island existed in reality, it would be a complete object, with a complete set of properties (with so many palm
trees, such dimensions, etc.). In that case, we can always conceive of a greater island—slightly larger,
perhaps. Hence, we cannot say—on pain of contradiction—that an island than which none greater can be
conceived exists in reality. Gaunilo’s case, in which the contradiction arises from supposing that the
“perfect island” exists in reality, is a kind of converse of Anselm’s, in which the contradiction arises from
supposing that that God does not exist in reality (Baker and Matthews 2010).
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   ? 

∴ (3) b has only mediated causal powers. (1,2 MP)

(4) If (3), then a can be conceived of and Oab. (meaning of ‘Oab’, instance of (C))
(5) a can be conceived of and Oab. (3,4, MP)
(6) If a can be conceived of and Oab, then a > b. (meaning of ‘Oab’, instance of (G))
∴ (7) a > b12 (5,6 MP)
(8) ~(a > b) (5 (1st conjunct), meaning of ‘b’)
∴ (9) a > b & ~(a > b) (7,8, Conj.) Contradiction!
∴ (10) It is not the case that: God (= b)) exists in intellectu and does not exist in re.
(reductio ad absurdum (1)–(9)).
∴ (11) God exists in re. ((10), DeMorgan’s rule; (c) of auxiliary argument; disjunctive
syllogism)
The auxiliary and main arguments are both valid and, I believe, sound. Each of the
premises is justified.
Moreover, the argument is non-question-begging since throughout the argument, all
mentions of God are made independently of whether God exists (the CA-assumption).
To talk about God independently of whether he exists is to imply nothing about
whether he exists. If he does exist, then both Anselm and the Fool are referring to
him (the Fool, unknowingly); if he does not exist, then neither is referring to anything
(Anselm, unknowingly). Since both the Fool and Anselm are talking about God
independently of whether he exists, they are on even ground. Neither begs the question
against the other.
So, the first two desiderata (no begging the question and soundness) are satisfied.
What about the third desideratum—that Anselm’s argument is not committed to
Meinongianism?

13.4 Must Anselm be Interpreted as a Meinongian?

Whether Anselm must be interpreted as a Meinongian depends on the philosophical
account of a key distinction in Anselm’s argument. The key distinction in Anselm’s
argument (which is not a distinction in ordinary language) is between existing in
intellectu and existing in re.13,14 A philosophical account of this distinction will

12
On the CA-assumption, we can use ‘a’, ‘b’, and ‘Oab’ to talk about putative things whether they exist
or not. Nevertheless, ‘is greater’ in the context of Anselm’s argument always means ‘is greater in reality.’
One may be tempted to object that if an otherwise similar thing to b with unmediated causal powers does
not exist, then it is not greater in reality than that than which none greater can be conceived. The
temptation should be resisted. Line (7) is a conclusion of an apparently sound non-question-begging
reductio argument. The contradiction is that if that than which none greater can be conceived has only
mediated causal powers, we can conceive of something that is greater in reality—something with unme-
diated causal powers. It is a later conclusion—line (11), not line (7)—that the putative thing with
unmediated causal powers does exist in reality.
13
Some things—God, as Anselm thinks—both exist in intellectu and exist in re.
14
Van Inwagen makes an important point that is often overlooked. He explains that when Anselm uses
the phrase ‘exists in solo intellectu’, he does not mean that Anselm is speaking of something that is “in any
sense a mental object or thing or item” (Van Inwagen 2012: 5). In particular, he is not talking about
representations, mental or otherwise. Representations of a painting are not paintings.
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   

incorporate a philosophical account of existing in intellectu. To exist in intellectu—to

be spoken or thought of independently of whether a putative thing exists—is be an
object of thought.15 The distinction between existing in intellectu and existing in re is
undoubtedly Anselm’s, and it is undoubtedly crucial to his argument in Proslogion
II. But the ontology is a different matter.
Call an ontology that recognizes two modes of existence, a weaker mode (e.g.,
existence in thought or in intellectu) and a stronger, more demanding mode (exist-
ence in reality or in re), a (broadly) ‘Meinongian ontology’ (Van Inwagen 2012: 2).
Although the distinction between existing in intellectu and existing in re is crucial to
Anselm’s argument, the distinction between existing in intellectu and existing in re
itself does not signal any particular ontology. Van Inwagen says that Anselm’s
argument presupposes an “unreflective philosophy of being,” which van Inwagen
takes to be Meinongian: Anselm’s argument “presupposes, and essentially presup-
poses, an ontology that is . . . Meinongian” (Van Inwagen 2012: 8). And any argument
that “does not presuppose two modes of existence,” van Inwagen says, is “simply not
Anselm’s” (Van Inwagen 2012: 8).
Even if van Inwagen is right to ascribe “tacit acceptance” of a Meinongian ontology
to Anselm, Anselm’s crucial distinction is still susceptible to a non-Meinongian
interpretation. A disputant who tacitly accepts a Meinonginan ontology may well
accept a non-Meinongian ontology on reflection.16 I propose interpreting Anselm in
terms of the CA-distinction, which, as we have seen, does not presuppose a Mei-
nongian ontology. The reason to interpret Anselm in a non-Meinongian way is that
many people find Meinongian ontologies incredible—van Inwagen calls them
‘wrong’ (Van Inwagen 2012: 8)—and I think it desirable to find a way for Anselm’s
argument to be credible and correct.
It may seem natural to ally the distinction between existing in intellectu and
existing in re, taken at face value, with a Meinongian ontology. And this is just
what van Inwagen does. He builds a Meinongian distinction between two modes of
existence right into his interpretation of Anselm. It is uncontroversial, he says, that
on Anselm’s view, God exists in the weaker of these two modes, “in the weaker sense
of ‘is’/‘exists.’ ” Anselm’s argument concerns whether God “also enjoy[s] being in the
more demanding mode” (Van Inwagen 2012: 2). So, van Inwagen brings the Mei-
nongian ontology right into the description of the argument.
However, we need not, and should not, interpret ‘x exists in intellectu’ to imply the
existence of x in any sense. Existence in intellectu and existence in re are not two
modes of existence. When we say of something that it exists in intellectu and in re, we

15
Edward Wierenga pressed on me the need for a philosophical account of objects of thought in “The
Ontological Argument and Objects of Thought” at the Memorial Session for Gareth B. Matthews,
sponsored by the Society of Christian Philosophers, APA Chicago, February 2012 (Wierenga 2011).
I hope that my present paper is an adequate response to the questions that Wierenga raised. For a different
criticism of an earlier draft of my paper, see Mann (2012).
16
Van Inwagen says that the Meinongian ontology “whose tacit acceptance I have ascribed to Anselm
seems to me to be inextricably bound up in his reasoning” (Van Inwagen 2012: 8). However, Anselm’s
reasoning can be understood without any Meinongian ontology. There is a precedent for the independence
of reasoning from ontology: Van Inwagen has argued that we can reason about tables and chairs, no matter
how we understand their ontology (Van Inwagen 1990).
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   ? 

are not implying that there are two ways of existing—any more than saying, “You are
in my chair and you are in my thoughts” implies that there are two ways of being “in”
something. Existence in thought is not a kind of existence.
As we see from the explication of the CA-assumption, talking-about-something-
independently-of-whether-it-exists is a way of speaking of things that are thought
about and talked about, regardless of whether they exist (e.g., Pegasus, Hamlet,
Moses). Similarly, the expression ‘exists in intellectu’ should be interpreted as a
manner of speaking about things that we can think about or talk about independently
of whether they exist.
In light of the CA-assumption, something is an object of thought if and only if
someone talks about or thinks about it independently of whether it exists. For
anything that exists in intellectu (or is an object of thought): if it exists in reality,
then it is an object; otherwise, there is no object at all.17 To exist in solo intellectu, in
thought alone, is not to exist. There need be no object generally associated with what
exists in intellectu.
We may put Anselm’s distinction between existing in intellectu and existing in re
like this: to exist in intellectu is to be thought about or talked about independently of
existence. (So, Anselm and the Fool can agree that God exists in intellectu, without
begging any questions.) To exist in re is to exist regardless of anyone’s thinking about
the existent, or talking about it.18 So, on the CA-assumption, we can understand the
distinction between ‘existence in intellectu’ and ‘existence in re’ without appeal to two
kinds of existence.
We can account for the two stages of my interpretation of Anselm’s argument in
terms of the CA-assumption: The Auxiliary Argument uses the notion of talking
about a thing independently of whether it exists to establish that God exists in
intellectu. The Main Argument starts with the reductio premise, which follows
from the Fool’s belief that when he uses ‘God’ or ‘that than which nothing greater
can be conceived,’ he picks out only an object of thought that does not exist in reality.
The remainder of the Main Argument is Anselm’s showing that the Fool’s position is
self-contradictory. Hence, pace the Fool, God does exist in reality. No two modes of
existing are needed.19
What favors my non-Meinongian interpretation, as I suggested earlier, is that it
makes Anselm’s argument more credible, or so it seems to me. Here is an example of
how the CA-interpretation can deflect challenges to Anselm. In “The One Fatal
Flaw in Anselm’s Argument”, Peter Millican argues that Anselm’s argument has
two stages (Millican 2004). At Stage 1, reference is made to an object of thought that
is denoted by ‘that than which nothing greater can be conceived.’ At Stage 2, it is

17
We might say that a putative thing that fails to exist has mediated causal powers, and define ‘mediated
causal powers’ in terms of unmediated causal powers that existing things have. A putative thing that fails to
exist has mediated causal powers if and only if there is someone with unmediated causal powers who talks
about it independently of whether it exists and whose thinking (or speaking) of it has some effect in reality.
18
In my work on material constitution, I argue that there are “intention-dependent” objects—objects
(e.g., flags) that could not exist in a world without entities with intentions and other attitudes. Even so, no
one has to think about or talk about a particular flag in order for it to exist.
19
Since whatever exists in solo intellectu does not exist in re, we can interpret Anselm’s point like this:
Anselm shows that being such that none greater can be conceived is incompatible with nonexistence.
OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   

argued that the object of thought in question must exist in reality as well. Millican
argues that all versions of the ontological argument fail at one of the two stages: the
merely-thought-of object of thought (Stage 1) cannot guarantee the reality of God
(Stage 2) without begging the question against the atheist. The atheist, says Millican,
should just deny that there is any object of thought to be thought of. However, the
CA-assumption allows that calling something an ‘object of thought’ does not entail
that there is an object to be thought of; an object of thought is what-is-thought-
about-and-referred-to-independently-of-whether- it-exists. So, the CA-version of
Anselm’s argument does not fall to Millican’s objection.
An objection to my non-Meinongian interpretation of Anselm’s argument would
likely be to the CA-assumption that we have the ability to refer to things independ-
ently of whether they exist. However, to reject the CA-assumption would have
consequences more dire than overthrowing my interpretation of Anselm. Not only
would such rejection fly in the face of common practices—like ordinary language
quantification over nonexistent objects—but also without the CA-assumption, we
would lack resources to argue non-question-beggingly about the existence of any
individual. In order to discuss whether a certain individual—Homer, or Lancelot, or
Adam, say—really existed, we must be able to talk about Homer, Lancelot, or Adam—
independently of whether he existed. So, repudiating the CA-assumption—besides
flying in the face of ordinary practice—would impair our ability to make commonplace
existential inquiry across the board.

13.5 Conclusion
I hope to have shown that the CA-Argument is indeed a new interpretation of
Anselm. If it fit any of the categories in van Inwagen’s article, it would fit The
Meinongian Version. However, I have argued that the CA-Argument is non-
Meinongian. The CA-Argument is nevertheless Anselmian. It follows Anselm’s
reductio form, and it concerns the existence of God himself (rather than the concept,
nature, or essence of God); but also it provides a philosophical account of Anselm’s
argument without any Meinongian coloration.
The novelty of the CA-argument is that it takes Anselm’s argument to have an
unnoticed presupposition that is both empirical and cognitive: only beings with the
ability to think about and talk about putative things independently of whether they
exist can formulate a sound and non-question-begging version of Anselm’s argument.
Not only has this presupposition been unnoticed but it is an asset to Anselm: making
this presupposition explicit allows a non-Meinongian interpretation of Anselm’s
argument that results in a more tenable argument than the one he is usually given
credit for. This non-Meinongian interpretation seems like a worthwhile new wrinkle
on an old argument.

Acknowledgments
I am especially grateful to John Keller for many detailed comments. Thanks are also due to
Joseph Jedwab for comments on an earlier draft, to Peter van Inwagen for comments on several
drafts, and to William E. Mann and Edward Wierenga for comments on a more recent draft.
All critics have been very helpful.
 OUP CORRECTED PROOF – FINAL, 9/11/2016, SPi

   ? 

References
Baker, L., and Matthews, G. (2010), ‘Anselm’s Argument Reconsidered’, in Review of Metaphysics
64 (1): 31–54.
Mann, W. (2012), ‘Locating the Lost Island’, in Review of Metaphysics 66: 295–316.
Matthews, G. (2005), ‘The Ontological Argument’, in W. E. Mann (ed.), The Blackwell Guide to
the Philosophy of Religion (Blackwell Publishing), 81–102.
Matthews, G., and Baker, L. (2010), ‘The Ontological Argument Simplified’, in Analysis 70 (2):
210–12.
Matthews, G., and Baker, L. (2011), ‘Reply to Oppy’s Fool’, in Analysis 71 (2): 303.
Millican, P. (2004), ‘The One Fatal Flaw in Anselm’s Argument’, in Mind 113: 437–76.
Van Inwagen, P. (1990), Material Beings, (Cornell University Press).
Van Inwagen, P. (2012), ‘Three Versions of the Ontological Argument’, in M. Szatkowski (ed.),
Ontological Proofs Today (Ontos Verlag), 1–17.
Wierenga, E. (2011), ‘The Ontological Argument and Objects of Thought’, in The Annual
Proceedings of the Center for Philosophic Exchange 42.
PART IV
Method
 14
Why Isn’t There More Progress
in Philosophy?
David J. Chalmers

Is there progress in philosophy? I have two reactions to this question. First, the
answer is obviously yes. Second, it is the wrong question. The right question is not “Is
there progress?” but “Why isn’t there more?”.
We can distinguish three questions about philosophical progress. The Existence
Question: is there progress in philosophy? The Comparison Question: is there as
much progress in philosophy as in science? The Explanation Question (which tends
to presuppose a negative answer to at least one of these two questions): why isn’t
there more progress in philosophy?
What we might call a glass-half-full view of philosophical progress is that there is
some progress in philosophy. The glass-half-empty view is that there is not as much
as we would like. In effect, the glass-half-full view consists in a positive answer to the
Existence Question, while the glass-half-empty view (or at least one salient version of
it) consists in a negative answer to the Comparison Question. These views fall between
the extremes of a glass-empty view which answers no to the Existence Question, saying
there is no progress in philosophy, and a glass-full thesis which answers yes to the
Comparison Question, saying there is as much progress in philosophy as in science (or
as much as we would like).
Of course the glass-half-full thesis and the glass-half-empty thesis are consistent
with one another. I think for almost anyone deeply involved with the practice of
philosophy, both theses will ring true. In discussions of progress in philosophy, my
experience is that most people focus on the Existence Question: pessimists about
philosophical progress (e.g., Dietrich 2011; Nielsen 1987; McGinn 1993) argue for the
glass-empty thesis, and optimists (e.g., Stoljar forthcoming) respond by defending
the glass-half-full thesis. I will focus instead on the Comparison and Explanation
Questions. I will articulate a version of the glass-half-empty thesis, argue for it, and
then address the crucial question of what explains it.
I should say this this paper is as much an exercise in the sociology of philosophy as
in philosophy. For the most part I have abstracted away from my own philosophical
and metaphilosophical views in order to take an “outside view” of philosophical
progress from a sociological perspective. For much of the paper I am largely saying
the obvious, but sometimes the obvious is worth saying so that less obvious things
  . 

can be said from there. Only toward the end will I bring in my own views, which lean
a little more toward the optimistic, and see how the question of philosophical
progress stands in light of them.

14.1 The Central Thesis

The form of a glass-half-empty thesis is: there is less progress in philosophy than
some benchmark. To articulate such a thesis more precisely, one needs to articulate a
measure of progress and a benchmark. The measure of progress I will use is collective
convergence to the truth. The benchmark I will use is comparison to the hard sciences.
Here I take inspiration from Peter van Inwagen (2004: 332), who writes:
Disagreement in philosophy is pervasive and irresoluble. There is almost no thesis in philoso-
phy about which philosophers agree. If there is any philosophical thesis that all or most
philosophers affirm, it is a negative thesis: that formalism is not the right philosophy of
mathematics, for example, or that knowledge is not (simply) justified, true belief.
That is not how things are in the physical sciences. I concede that the “cutting edge” of
elementary-particle physics looks a lot like philosophy in point of pervasive and fundamental
disagreement among its respected practitioners. But there is in physics a large body of settled,
usable, uncontroversial theory and of measurements known to be accurate within limits that
have been specified. The cutting edge of philosophy, however, is pretty much the whole of it.

Van Inwagen’s thesis is not explicitly about progress, and the general tenor of his
discussion suggests something closer to a glass-empty thesis than a glass-half-empty
thesis. I think that at least once the issue is precisified a little, however, a glass-half-
empty thesis is more defensible.
Here is my central thesis: There has not been large collective convergence to the
truth on the big questions of philosophy.
Here the big questions of philosophy are questions like: What is the relationship
between mind and body? How do we know about the external world? What are the
fundamental principles of morality? Is there a god? Do we have free will? I will not try
to provide a more precise list than this, but any philosopher can come up with a list of
ten or so big questions fairly easily, and I suspect that there would be a lot of overlap
between these lists. We could even use these lists to operationally define the big
questions: the big questions of a field at time t are those that members of that field
would count as the big questions of the field at time t. For purposes of comparison, we
may want to impose some regimentation on the form of the big questions, for example
formulating them all as choices between a small number of mutually exhaustive options.
We can define collective convergence on an answer over a period as the increase in
degree of agreement on that answer from the start of the period to the end of the
period. Degree of agreement can be defined using one of various mathematical
measures of agreement across a group of people on a set of issues.1 Collective

1
I leave open the question of just what measure of agreement is best for present purposes. One useful
measure is Krippendorff ’s alpha (Krippendorff 2013: 221–50), equal to 1-(Do/De), where Do is the observed
incidence of disagreement between respondents (summed over all pairs of respondents and all questions) and
De is the expected incidence through chance alone. This measure can be applied to communities of different
  ’     ? 

convergence (simpliciter) over a period is defined as the collective convergence on the

dominant answer at the end of that period over the period.
Collective convergence on the big questions over a period is stipulated to be
convergence on the questions that were big at the start of the period. What matters
here is convergence on the big questions of the past rather than on big questions of
the present. Convergence on the latter can be expected to be low, as disagreement on
the answer to a question is highly relevant to making it a big question. Convergence
on the former provides a more discriminating measure which can reasonably be
expected to be high in some cases and low in others. For our purposes we could
choose an arbitrary period (say, from 200 years ago to today), or perhaps better,
choose many different periods and take the average convergence over those periods.
We can say that large collective convergence over a period requires as much
convergence as there has been over big questions in the hard sciences in the same
period. Here I will take the hard sciences to include at least mathematics and
the natural sciences: paradigmatically physics, chemistry, and biology. I set aside
the cognitive and social sciences, which have arguably seen less convergence on the
answers to their big questions. For current purposes I do not need to take a stand on
how philosophy fares relative to these.
Large collective convergence to the truth in a period requires large collective
convergence to correct answers to the big questions over that period. That is, we
must have as much increased agreement on correct answers to the big questions as in
the hard sciences. Because of the reference to truth or correctness, large collective
convergence to the truth requires a degree of realism about the domains in question.
But something like convergence to the truth is required in order that the convergence
constitutes progress and not regress.

14.2 Argument for the Central Thesis

Here is an argument for the central thesis. It has two premises, an empirical premise
and a bridging premise.
(1) Empirical premise: There has not been large collective convergence on the big
questions of philosophy.

sizes (not all of whose members need have a view on a given issue) and to questions whose answers have
many different sorts of structure. Disagreement is weighted by a measure of “distance” between any two
answers, which makes alpha particularly helpful in comparing questions with different numbers of answers.
With such a metric in hand, one can use a version of alpha to measure communal degree of agreement with a
specific answer. For our purposes some rescaling may be useful (e.g., imposing a lower bound of zero and
then squaring).
The definition of convergence here does not invoke the standard mathematical notion of convergence,
which applies to infinite series or infinite times rather than discrete periods and which would require that
the community eventually come arbitrarily close to universal consensus on an issue. But an element of its
flavor will be present if our measure of agreement has some bias toward universal agreement (as the
rescaled version of Krippendorff ’s alpha does), so that for example that a shift from 79% to 99% of the
community agreeing on an answer to a binary question counts as greater convergence than a shift from
40% to 60%.
  . 

(2) Bridging premise: If there has not been large collective convergence on the big
questions of philosophy, there has not been large collective convergence to the
truth on the big questions of philosophy.
(3) Conclusion: There has not been large collective convergence to the truth on
the big questions of philosophy.
The argument is valid. The bridging premise may look like a logical truth, but
given how large collective convergence is defined above, it is not. The antecedent of
this conditional premise says that there is less convergence on dominant answers to
the big questions in philosophy than in the hard sciences, while the consequent says
that there is less convergence on correct answers to those questions. These may come
apart in a way that renders the conditional false if there has been strong convergence
to the false (or strong convergence uncorrelated with truth) in the sciences along with
weak convergence to the truth in philosophy. Still, given that convergence in science
is largely convergence to the truth, or merely that convergence in science at least as
likely to be convergence to the truth as convergence in philosophy, then the premise
is plausible.2
One may worry that because most scientific theories eventually turn out to be false,
most convergence in science will be convergence to the false. This worry is less pressing
if we formulate questions in terms of a small number of mutually exhaustive choices, as
suggested earlier. This way, convergence to the truth will require only convergence on a
correct coarse-grained class of theories, rendering it much more plausible that much
convergence in science has been convergence to the truth. Alternatively, if questions
allow an open-ended range of answers, then invoking a metric for distance between
answers (also suggested earlier) will allow that agreement on strictly speaking false
theory that is relatively close to the truth constitutes a sort of convergence to the truth.
The main work in the argument is done by the empirical premise. I take it that it
will be plausible to those with passing familiarity with philosophical and scientific
practice. Still, it makes a sociological claim and cannot be decisively settled from the
armchair. We do not have all the empirical data required for a systematic investigation
of the premise, but we have some of it.
The 2009 PhilPapers Survey (Bourget and Chalmers 2014) surveyed professional
philosophers on answers to thirty important questions in philosophy. The survey was
sent to the members of 99 leading departments of philosophy (largely specializing in
analytic/Anglocentric philosophy) in North America, Europe, and Australasia. About
47 percent of the 2000 or so recipients of the survey filled out and returned the survey.
Questions were posed as a choice between two, three, or four options. Respondents
could indicate that they “accept” or “lean” toward one option, or give a variety of
“other” answers (e.g., unfamiliar with the issue, the question is too ambiguous to
answer, there is no fact of the matter, accept another option, and so on). The results
(collapsing “accept” and “lean toward” answers, and collapsing “other” answers) were
as follows.

2
Thanks to Hedda Hassel Morch and Rory Madden for pointing out ways in which the bridging
premise could turn out to be false.
  ’     ? 

(1) A priori knowledge: yes 71%, no 18%, other 11%.

(2) Abstract objects: Platonism 39%, nominalism 38%, other 23 %.
(3) Aesthetic value: objective 41%, subjective 35%, other 24%.
(4) Analytic/synthetic distinction: yes 65%, no 27%, other 8%.
(5) Epistemic justification: externalism 43%, internalism 26%, other 1%.
(6) External world: non-skeptical realism 82%, skepticism 5%, idealism 4%, other
9%.
(7) Free will: compatibilism 59%, libertarianism 14%, no free will 12 %, other
15 %.
(8) God: atheism 73%, theism 15%, other 13%.
(9) Knowledge claims: contextualism 40%, invariantism 31%, relativism 3%, other
26%.
(10) Knowledge: empiricism 35%, rationalism 28%, other 37%.
(11) Laws of nature: non-Humean 57%, Humean 25%, other 18 %.
(12) Logic: classical 52%, non-classical 15%, other 33%.
(13) Mental content: externalism 51%, internalism 20%, other 29 %.
(14) Meta-ethics: moral realism 56%, moral anti-realism 28%, other 16%.
(15) Metaphilosophy: naturalism 50%, non-naturalism 26%, other 24 %.
(16) Mind: physicalism 57%, non-physicalism 27%, other 16%.
(17) Moral judgment: cognitivism 66%, non-cognitivism 17%, other 17%.
(18) Moral motivation: internalism 35%, externalism 30%, other 35 %.
(19) Newcomb’s problem: two boxes 31%, one box 21%, other 47 %.
(20) Normative ethics: deontology 26%, consequentialism 24%, virtue ethics 18 %,
other 32%.
(21) Perceptual experience: representationalism 32%, qualia theory 12 %,
disjunctivism 11%, sense-datum theory 3%, other 42 %.
(22) Personal identity: psychological view 34%, biological view 17%, further-fact
view 12%, other 37%.
(23) Politics: egalitarianism 35%, communitarianism 14%, libertarianism 10%,
other 41%.
(24) Proper names: Millian 34%, Fregean 29%, other 37%.
(25) Science: scientific realism 75%, scientific anti-realism 12%, other 13%.
(26) Teletransporter: survival 36%, death 31%, other 33%.
(27) Time: B-theory 26%, A-theory 16%, other 58%.
(28) Trolley problem: switch 68%, don’t switch 8%, other 24%.
(29) Truth: correspondence 51%, deflationary 25%, epistemic 7 %, other 17%.
(30) Zombies: conceivable but not metaphysically possible 36%, metaphysically
possible 23%, inconceivable 16%, other 25%.

The degree of disagreement here is striking, if unsurprising. Only one view (non-
skeptical realism about the external world) attracts over 80 percent support. Three
views (a priori knowledge, atheism, scientific realism) attract over 70 percent sup-
port, with significant dissent, and three more views attract over 60 percent support.
On the other 23 questions, the leading view has less than 60 percent support.
  . 

Admittedly, not all of questions are among the “big questions” of the past, but
certainly some are: the questions about the external world, free will, god, knowledge,
meta-ethics, metaphilosophy, mind, and normative ethics, for example. Only two of
these (external world, god) have views with over 60 percent support (and in the case
of the external world question, the consensus is somewhat misleading, since arguably
the biggest question is how we know about the external world).
For fuller data to adjudicate the central thesis, we would need the results of the
PhilPapers Survey not just in 2009 but at regular intervals in the past: 1909, 1809,
and so on. At each point we would need to ask members of the philosophical
community first, what they take to be the big questions of philosophy, and second,
what they take to be the answers to those questions as well as to big questions from
past surveys. We would also need to have analogous longitudinal surveys in other
fields: the MathPapers Survey, the PhysPapers survey, the ChemPapers Survey, the
BioPapers Survey, and so on. And we would need a reasonable measure of
agreement at a time. I predict that if we had such surveys and measures, we
would find much less convergence on answers to the big questions suggested by
past surveys of philosophers than we would find for corresponding answers in
other fields.
Some partial data is given by the twenty-three problems that David Hilbert posed
for mathematics in 1900 (Hilbert 1902; Yandell 2002). Around ten of these twenty-
three problems have been clearly solved, leading to universal consensus, and seven
have been partially solved, leading to partial consensus. A similar pattern could
reasonably be expected in physics, chemistry, and biology. We can compare these
results to the problems in Bertrand Russell’s (1912) The Problems of Philosophy.3
None of these have led to universal convergence and almost none have led to
anything close.
Of course one can object to the thesis in various ways. One could argue that
there has been more convergence on the big philosophical questions of the past
than these case studies suggest. Alternatively, one could argue that there has been
less convergence on the big scientific questions of the past than they suggest.
A version of the first objection springs from the observation that disciplines
such as physics were once considered part of philosophy. If we go back to a time
before the split, then insofar as the big questions of physics are among the big
questions of philosophy at that point, high convergence on the former will lead to
significant convergence on the latter. Still, insofar as physics was just a proper part of
philosophy, and one more susceptible to convergence than the other parts, one would
still expect convergence on the former to produce less convergence on the latter. It is
also not entirely clear that philosophy as we understand it should get the credit for
the convergence in physics: what was called philosophy in the past was arguably a
different and broader field.
In any case, one can bypass this objection by focusing on a point after the split
between physics and philosophy: 1809 or 1909, say. The objector might respond

3
Thanks to Jeremy Goodman for suggesting the Hilbert/Russell comparison.
  ’     ? 

that now there will be less convergence in philosophy only because we have split off
the parts of it that have made most progress. But this is to concede the central thesis
and argue for a certain explanation of it, one that I consider later in the paper.
Another response is that there were further splits after this point: psychology, logic,
linguistics, and economics, for example. Still, I think that the questions resolved
by these areas constitute a small enough fraction of the big questions of philosophy
in 1809 or 1909 that even if philosophy gets credit for them, this will not bring
the level of convergence in philosophy close to the corresponding level in the
hard sciences.
Another version of the first objection suggests that some big philosophical ques-
tions of the past have reached consensus and so have dropped off the list of big
questions even without their own disciplines branching off. Perhaps something like
this is plausible for some moral and political questions, for example, such as the
question of whether all people are equal, where convergence within philosophy
reflects convergence in society more generally. And there may have been questions
that were regarded as truly important in a period (about the viability of certain
versions of idealism, say) on which there is now a consensus view. The existence of
questions like this helps to make a case against a glass-empty thesis. But where a
glass-half-empty thesis is concerned, it suffices to note that the proportion of
questions like these is lower in philosophy than in the hard sciences.
As for the second objection, one could argue that many of the big questions of the
hard sciences are themselves philosophical questions and have seen low convergence:
questions about the interpretation of quantum mechanics, for example, or about the
locus of natural selection. But as long as some of the big questions of the hard
sciences are not philosophical questions, as is surely plausible, and as long as these are
more susceptible to convergence, then we would still expect the central thesis to be
true. Furthermore, insofar as this objection relies on a contrast between philosophical
and nonphilosophical questions, it tends to reinforce the underlying contrast in
convergence.
It could also be suggested that numerous big nonphilosophical questions in the
hard sciences have met with low convergence: questions about the origins of life, for
example. This is surely right, but it remains plausible that enough have seen major
convergence that there is still a significant difference between the nonphilosophical
and the philosophical. The case of the Hilbert problems for mathematics brings this
out. The problems are mostly nonphilosophical, and although some of them are
unsolved, the overall convergence on them has been quite dramatic. Something
similar plausibly applies in physics, chemistry, and biology.
A final objection is that even though there is more agreement now in the hard
sciences than in philosophy, these sciences may also start from a position of more
agreement, resulting in a smaller increase in agreement in the sciences than in
philosophy. This hypothesis is mathematically consistent, but I do not think it is
especially plausible. Especially given a measure of agreement that is biased toward
universal agreement, as discussed earlier, and given that there is considerable dis-
agreement over the big questions of a time at that time, then the various cases in
which the hard sciences (unlike philosophy) approach universal consensus will tend
to yield greater overall convergence as well.
  . 

14.3 The Varieties of Progress

Despite this lack of convergence, it is hard to deny that the insights of Plato and
Aristotle, Hume and Kant, Frege and Russell, Kripke and Lewis have involved
significant philosophical progress. Correspondingly, my glass-half-empty thesis is
compatible with many different glass-half-full theses, asserting the existence of
various forms of progress in philosophy. We can systematize various such theses
by dropping the central requirements of my central thesis one at a time.
Drop “large”: There has been (non-large) collective convergence to the truth on
big questions of philosophy. It is plausible that there has been major convergence on
answers to a small number of the big questions of philosophy. In questions about
god there appears to have been major convergence toward atheism, for example. It is
also plausible that there has been minor convergence on answers to many other
questions, such as toward physicalism about the mind. Of course whether one
counts this convergence as convergence to the truth will depend on one’s own
philosophical views. Theists and dualists will hold that the convergence constitutes
regress rather than progress. But if we assume optimistically that the convergence is
indeed convergence to the truth, it may be that 10–20 percent more philosophers
have true beliefs about the answers to the big questions of 1809 in 2009 than in 1809.
If so, that is a sort of progress. Still, it remains plausible that convergence is greater in
other areas.
Drop “collective”: There has been large (non-collective) convergence to the truth
on the big questions of philosophy. The central thesis is consistent with the claim
that various individuals or subcommunities have themselves had large convergence
to true answers on the big questions. For example, on my more optimistic days
I can convince myself that over time I have converged to the truth on many of
these questions. But if so, sadly, it has not led to collective convergence to the truth.
Likewise, perhaps groups such as the logical empiricists or the Oxford realists
have converged on the truth. But again, community-wide convergence has not
ensued. Perhaps there has even been a large amount of community-wide conver-
gence at certain local temporal periods, but if so, this convergence has not persisted
over time.
Drop “big”: There has been large collective convergence to the truth on (non-big)
questions of philosophy. There has been large convergence on various smaller theses:
the thesis that knowledge is not justified true belief, for example, and the thesis that
conditional probabilities are not probabilities of conditionals. As van Inwagen
suggests in the passage above, we are especially good at converging on negative
theses that rule out certain specific views. There is also often convergence on
conditional theses, asserting conditional connections between views. But I take it
that these are not really answers to the big questions of philosophy.
Drop “convergence to the truth”: There have been large collective advances (not
involving convergence to the truth) on the big questions of philosophy. Certainly
there are many forms of philosophical progress that do not involve convergence to
the truth. It is plausible that we have a greatly increased understanding of the issues
underlying the big questions. We better understand the reasons for accepting and
  ’     ? 

rejecting key philosophical theses. We have come to explore new views and new areas
of philosophical space that we had not even conceived of earlier. We have developed
new methods and better arguments. In some cases we have applied philosophy to the
world. These are all certainly forms of progress. I simply note that they have not been
accompanied by large collective convergence to the truth.
I want to stress that I am not simply equating progress with convergence to the
truth. I am a pluralist about progress: there are many values that can be realized
through philosophy, and there are many ways of advancing and realizing those values.
Attaining the truth is certainly not the only such value. Still, it is certainly one such
value. It follows that progress toward the truth is one form of philosophical progress.
More strongly, I think a case can be made that attaining the truth is the primary
aim at least of many parts of philosophy, such as analytic philosophy. After all, most
philosophy, or at least most analytic philosophy, consists in putting forward theses as
true and arguing for their truth. I suspect that for the majority of philosophers, the
primary motivation in doing philosophy is to figure out the truth about the relevant
subject areas: What is the relation between mind and body? What is the nature of
reality and how can we know about it? Certainly this is the primary motivation in my
own case. So I am sympathetic with the claim that progress toward the truth has a
certain primacy among the forms of philosophical progress. But even if one denies
this, it is hard to deny that it is among those forms.
It is sometimes said that an obsession with truth reflects an overly scientistic
conception of philosophy. We should not think of philosophy as a quest for the
answers. Instead it is a quest for something else: understanding, clarity, enlighten-
ment. I agree that these are goals worth pursuing, and that philosophy can help us
pursue them. And I can see why, in the absence of answers to philosophical puzzles, it
might seem especially appealing to focus on these goals instead. Still, I think we
should acknowledge that this reaction involves something of a lowering of our sights
for philosophy. At least pretheoretically, many of us get into philosophy looking for
truth and looking for answers. One can argue that this hope is naive: truth and
knowledge are not to be had in philosophy, and one should settle for something
different. But even if so, both the hope and its naivete are worth marking.
Why is convergence to the truth important, and why should we be concerned about
its absence? One obvious answer is that we value knowledge, agreement is required
for knowledge, and convergence goes along with increases in knowledge. A strong
version of this view, suggested by van Inwagen’s discussion, is that where there is
sufficient disagreement among experts, no individuals can be said to know the truth.
Even if some individuals have hit on good arguments for true conclusions, how can
they have justified confidence that these are good arguments, when so many of their
peers disagree? I am not so sure: I think that at least in some cases, a good argument
can ground an individual’s knowledge of a conclusion even when peers reject it. For
example, I think that the presence of any number of peers who deny the existence of
consciousness would not undermine my knowledge that I am conscious. Likewise, it
would not undermine arguments that take this claim as a premise.
But even if agreement is not required for individual knowledge, some degree of
agreement is plausibly required for collective knowledge. If the community of experts
  . 

on a question has serious disagreement over the answer to that question, then that
community cannot be said to collectively know the answer to that question, and nor
can the broader community of which they are a part. Even when some individuals
know the answer to a question, this individual knowledge will not usually suffice for
collective knowledge, except perhaps in special circumstances such as when the
community defers to these individuals.
Furthermore, we value collective knowledge. One reason that progress of the hard
sciences has been so impressive is that it has plausibly enabled us—the community of
inquirers—to collectively know the answers to those questions. But in the absence of
sufficient agreement on the answers to philosophical questions, we cannot be said to
have collective knowledge of those answers.
Of course one can argue over just what degree and pattern of agreement is
required for collective knowledge. But it is highly plausible that the kind of
disagreement that we observe over the answers to the big questions of philosophy
suffices to undermine any claims of collective knowledge of the answers to most of
those questions. Perhaps one could argue that in the survey above, a few views
(non-skeptical realism about the external world, atheism, a priori knowledge)
display the sort of consensus that allows collective knowledge. But even that
claim would be bold, and the extension to claims with less consensus (physicalism
and compatibilism, say) seems so bold as to be implausible. So I take it that the
difference in agreement on the big questions in science and philosophy reflects a
significant difference in the collective knowledge that we have attained. Likewise,
the difference in convergence on the big questions reflects a significant difference in
the increase of collective knowledge over time.
This is not to deny that we have attained a great deal of collective knowledge in
philosophy. As Timothy Williamson (2006) has said, we knew much more in 2004
than in 1964, much more in 1964 than in 1924, and so on. But this collective
knowledge typically does not involve answers to the big questions. It is mainly
knowledge of the answers to smaller questions, of negative and conditional theses,
of frameworks available to answer questions, of connections between ideas, of the
way that arguments bear for and against conclusions, and so on. In the absence of
convergence on the big questions, collective knowledge of the answers to those
questions eludes us.

14.4 Philosophical Argument

I now turn to the central question: why isn’t there more progress in philosophy? And
in particular: why is there less convergence in philosophy than in the hard sciences?
An initial explanation, though perhaps this is merely an articulation of the
phenomenon, lies in the relative power of the methods used in these domains. The
hard sciences have methods—proof in the case of mathematics, and the observa-
tional/experimental method in physics, chemistry, and biology—that have the power
to compel agreement on the answers to the big questions. Philosophy has a method—
the method of argument—that does not.
  ’     ? 

What is the difference between these methods? One difference in that the methods
of experiment and proof start from widely agreed premises—observations in science,
axioms in mathematics—and proceed from there to strong and surprising conclu-
sions. We aspire to do this in philosophy too: witness Russell’s remark that the point
of philosophy is to start with something so simple as not to seem worth stating, and
to end with something so paradoxical that no one will believe it (Russell 1918). But in
practice, widely agreed premises rarely suffice to ground strong and surprising
conclusions in philosophy.
There are certainly many arguments for strong conclusions in philosophy. But in
the great majority of cases, they have premises that opponents can deny without too
much cost, or inferences that opponents can reject without too much cost. (I focus
mainly on premises, but everything I say also applies to inferences, for example
by turning nondeductive inferences into tacit premises of deductive arguments.)
Sometimes the denied premise is antecedently plausible, and the denial somewhat
surprising. But even then the denial rarely has the implausibility of denying
a mathematical axiom, or of denying a well-replicated experimental observation.
So these denials are usually tenable, at least in a broadly sociological sense of
tenability.
Let us say that consensus premises (and inferences) are those that are regarded by
the community as undeniable, or at least as incurring enormous cost for anyone who
denies them. A consensus premise might be denied by a few outliers, but it cannot be
subject to widespread disagreement within the community. Let us say that an
argument that uses only consensus premises and inferences is a decisive argument.
(Note that consensus premises and decisive arguments are both defined in socio-
logical terms.) Then the claim is that while there are decisive arguments for strong
conclusions in the sciences, there are relatively few such arguments in philosophy.
There are certainly some consensus premises in philosophical arguments. After
all, these premises can include mathematical axioms and empirical observations
themselves, as well as the theorems and theories that are grounded in them. But
mathematical premises and empirical observations alone almost never suffice to
draw strong philosophical conclusions. Further premises or inferences are required
to bridge from science and mathematics to philosophy, and these premises and
inferences are typically deniable.
In addition to mathematical axioms and empirical observations, there are some
philosophical intuitions that are extremely difficult to deny. But these intuitions are
not so common (many antecedently plausible intuitions turn out to be deniable), and
where they exist, it is typically difficult to draw strong philosophical conclusions from
them. There are some cases where these intuitions, perhaps in conjunction with
mathematical and empirical claims, allow us to draw strong and surprising conclu-
sions. This works particularly well for negative theses, where intuitions and formal
models can generate counterexamples to positive theses or other reasons to reject
them. Gettier’s argument from an intuition about a case to the conclusion that
knowledge is not justified true belief is one example. Lewis’s formal argument that
conditional probabilities are not probabilities of conditionals is perhaps another.
But it is notable that these negative conclusions fall far short of answers to the big
  . 

questions of philosophy. Almost any argument for a positive answer to these

questions involves deniable premises.4
For most practitioners of philosophy, the phenomenon of premise deniability is
familiar from both sides. As the old saying goes: one person’s modus ponens is
another person’s modus tollens. When we give arguments for our views, we are
frustrated to find opponents biting the bullet by rejecting what we took to be a plausible
premise, without this serving as any sign of defeat. When we address arguments against
our views, we sometimes work backwards from our rejection of the conclusion to see
which premises we have to deny, and we deny them. In the best cases, we learn
something from this, and we take on commitments that we might have antecedently
found surprising. But these commitments are rarely untenable to maintain.
As a result, philosophical arguments typically lead not to agreement but to
sophisticated disagreement. Advocates of a view learn what extra commitments
they need to take on to avoid the arguments. Bad versions of a view are rejected
and sophisticated versions are developed in their place. This leads to a sort of
eliminative progress where areas of philosophical space are eliminated, but only in
small fragments at a time. It is rare for a major general view (materialism or dualism,
compatibilism or incompatibilism, utilitarianism or deontology) to be eliminated in
this way. Instead, there are large surviving fragments containing the views needed to
avoid the arguments (type-B materialism with the phenomenal concept strategy,
source incompatibilism, two-level utilitarianism, theism without unrestricted ben-
evolence or omnipotence). The same sort of elimination, fragmentation, and refine-
ment often recurs at these lower levels. The views that survive yield a sort of fractal
structure in philosophical space, akin to the Mandelbrot set with its intricate com-
plexities at all levels, but in which large regions of space are rarely eliminated entirely.
This phenomenon might strike one as a philosophical analog of the Duhem-Quine
thesis, in a version saying that any scientific theory can be made compatible with any
evidence by appropriate adjustment to the background assumptions that bridge

4
To gather data here, I ran an informal Internet survey of philosophers, asking for arguments that are
near-universally regarded by philosophers as establishing their conclusions. Further candidates included
the forcible-organ donation argument against simple versions of utilitarianism, Kripke’s argument that
necessity comes apart from apriority, Gödel’s argument against versions of mathematical formalism, the
argument from evil against theism, the model-theoretic argument against global descriptivism, the perfect
actor argument against logical behaviorism, the multiple-realizability argument against the identity theory,
Goodman’s argument against purely formal inductive logic, arguments from relativity against presentism,
Frankfurt’s argument that moral responsibility does not require the ability to do otherwise, Hart’s
argument against Austin’s command theory of laws, Russell’s refutation of Frege’s Basic Law V, Moore’s
open question argument against analytic naturalism, Putnam’s argument for externalism about meaning,
Descartes’ cogito, and many others.
It is striking that the great majority of these arguments are naturally regarded as arguments for negative
conclusions, in that they are arguments against fairly specific views. Of course the negative/positive
distinction is not entirely clear, but we have a reasonable intuitive grasp on it. A few conclusions have a
positive flavor: one’s existence (the cogito), externalism (Putnam), and perhaps the necessary a posteriori
(Kripke) and atheism (the argument from evil). But the first three are at best marginal candidates for
answers to big questions, and the survey data suggests that the second and fourth are at best marginal cases
of near-universal agreement. All this reinforces the point that decisive arguments in philosophy are rare,
that decisive arguments for positive views are even rarer, and decisive arguments for positive answers to the
big questions are so rare as to be almost nonexistent.
  ’     ? 

between theory and evidence. But in practice, scientific theories are often decisively
rejected in light of evidence, with revised consistent versions of the theories being
rejected as untenable. In effect, some (nondeductive) inferences from evidence to
scientific theory have consensus status. Theories are ruled out not by consensus
evidence alone but by consensus evidence plus consensus inferences. In the philo-
sophical case, however, consensus evidence plus consensus inferences are much less
powerful. When someone argues against a philosophical theory, there is usually at
least a revision of the theory that is not just consistent but tenable in light of the
consensus evidence. Those who argue against a philosophical view sometimes accuse
their resourceful opponents of holding onto a degenerating research program, but it
is typically much harder to make this charge stick in philosophy than in science. This
could be because philosophers apply laxer standards to their theories, so that
inferences that have consensus status among scientists do not have it among philo-
sophers, but more plausibly it is because inferences that meet scientific standards
provide weaker reasons to accept philosophical conclusions than to accept scientific
conclusions.
It might also be objected that in science, positive theories are not usually estab-
lished by single experiments but by many experiments collectively. By parity, we
might hope that even if positive philosophical theories are not established by single
arguments, they might be established by a number of arguments collectively. There
are perhaps a few cases of negative theses being established this way: the rejection of
sense-datum theories of perception may be an example. But even these cases are rare,
and positive cases are even rarer. In practice, if an opponent can reject individual
arguments for a thesis without too much cost, they can usually reject collections of
arguments without much cost too.
Does this mean that all philosophical arguments for positive theses are unsuccess-
ful, as van Inwagen (2006) has suggested? (Van Inwagen talks about substantive
theses, but his discussion suggests that these are required to be positive theses.) This
depends on what one means by success. If one defines success in sociological terms,
so that success requires convincing almost everyone in a community, then we have
seen that at best very few philosophical arguments for positive theses have been
successful in our community. Van Inwagen defines success in idealized epistemo-
logical terms: a successful argument for a proposition p is one that would convince an
audience of ideal reasoners who are initially agnostic concerning p, in the presence of
an ideal opponent of p. I do not think that the sociological observations above (or the
sociological observations that van Inwagen appeals to) come close to establishing that
no philosophical arguments are successful in that sense. (See Kelly and McGrath
(2017) for more on this theme.) Human beings are simply too far from ideal for that
conclusion to follow.
It also does not follow from anything I have said that all philosophical arguments
are question-begging, or that they are dialectically powerless. Even when arguments
have deniable premises, they often have dialectical power, in that their premises have
antecedent support that does not rest on considerations too close to the conclusion.
In such a case the argument does not beg the question. Even though a sophisticated
and committed opponent will deny the premise, the argument might well move
an agnostic observer to accept the conclusion. In practice, we often use this sort of
  . 

dialectical power as a criterion for a good argument that many philosophers can
agree on, even if they disagree on the argument’s ultimate persuasiveness.
I am also not saying that these arguments cannot produce knowledge. Deniable
premises may nevertheless be known by many people to be true. As before, while too
much disagreement over a claim may undermine collective knowledge of that claim,
it need not undermine individual knowledge of that claim. Likewise, an argument can
ground individual knowledge even when peers reject it. This applies all the more where
non-peers are concerned. For all I have said, some arguments may have premises and
inferences that can only be denied unreasonably, or by nonideal reasoners. If so, these
arguments may well produce knowledge in beings more reasonable than the deniers.
So it is not straightforward to draw conclusions about lack of normative force from
premises about lack of sociological success. There is perhaps an intermediate nor-
mative notion, defining a successful argument as one with the power to persuade all
competent agnostics, where competence is some reasonably high but nonideal stand-
ard of rationality that many human philosophers meet. There is good reason to think
that few philosophical arguments for positive conclusions persuade all competent
philosophers, or even all competent agnostics. On the face of it, disagreement on big
questions among the most able philosophers (by any reasonably neutral measure) is
about as rife as disagreement among philosophers more generally. This suggests that
most philosophical arguments are not successful in the normative sense tied to
competence, even if they are successful in the other normative senses.
Burton Dreben once memorably said to me (on the only occasion that I met him,
in St. Louis around 1994): “Great philosophers don’t argue.” He went on to elaborate
that none of Frege, Russell, Wittgenstein, Carnap, or Quine really give arguments for
their views. Of course this is not strictly true, but I think his point was that for these
philosophers, the real work is not done by arguments for a thesis but by the thesis
itself, or the framework it is embedded in. A refined version of his claim (suggested to
me by Gene Callahan) might say: great philosophers may argue, but their arguments
are not what makes them great. A part of Dreben’s thought, as I understood it, was
that since arguments are so easily rebutted, giving arguments is a sign of weakness. It
is better to simply assert and develop a thesis. Then one’s readers have to engage with
the thesis itself, without the cheap distraction of rebutting arguments for the thesis.
(Rawls (2001) elaborates on Dreben’s views in a somewhat different direction:
“Burt would not, of course, deny the plain fact that philosophers make many
complicated arguments. But he thinks that at bottom there are no arguments one
philosopher can use to convince another of a metaphysical point. At the basic level,
philosophers simply rely on and appeal to different ‘data.’ It is a standoff with no
resolution by argument. Burt has said that Quine is a metaphysician, a metaphysician
of science. By that he means that Quine doesn’t argue for physicalism, or scientific
realism. He assumes it and works out his view from there.”)
I have found it impossible to follow Dreben’s advice myself. In my work I am a
compulsive arguer, which no doubt leaves me subject to a modus tollens from
Dreben’s thesis. But certainly it is rare that these arguments bring a large sector of
the population around. This is especially so when many of the people already have
firm commitments, as on issues such as the mind–body problem and the theory of
meaning: here it is hard to do more than bringing around a few people here and
  ’     ? 

there. On issues where people are initially agnostic or their commitments are weak,
there can be more movement.5 Even here the fact that such movement is limited
reinforces the basic point.
The upshot is that consensus in philosophy is as hard to obtain as it ever was,
and decisive arguments are as rare as they ever were. To me, this is the largest
disappointment in the practice of philosophy. Once one has been doing philosophy
for a while, one no longer expects arguments to produce agreement, and one deems
an argument good when it merely has some dialectical power. But this is an
adjustment of expectations in response to a disappointing reality. Antecedently to
doing philosophy, one might have hoped for something more.

14.5 New Philosophical Methods

Faced with the failure of traditional philosophical methods, we might look to new
methods. Occasionally, new methods developed by thinkers who considered them-
selves philosophers have helped to resolve questions once considered philosophical:
witness the development of logic, physics, psychology, and so on. It is natural to hope
that new methods might produce further progress.
In the last century or so, many new philosophical methods have been developed
and many old methods have been refined, in order to help reach philosophical
conclusions. Empirical philosophy draws on empirical science. Formal philosophy
draws on formal reasoning. Linguistic philosophy draws on the analysis of language.
Phenomenology draws on phenomenological reflection. Feminist philosophy draws
on analysis of the role of gender. Crosscultural philosophy draws on multiple cultural
traditions in philosophy. Experimental philosophy draws on the empirical study of
philosophical judgments.6
All of these methods have led to new insights and to philosophical progress. All
have led to new arguments for interesting conclusions. But manifestly, none of these
methods have led to recent convergence on answers to the big questions of philoso-
phy. In the wake of each of these methods, philosophical disagreement is as rife as it
ever was. Even within a tradition, there are few cases where the big questions are
regarded as settled. Instead, these methods have led us again to more sophisticated
versions of old disagreements.
In many cases, the basic problem is that of premise deniability. In the case of
empirical and formal philosophy, we have already seen that empirical and formal

5
To indulge in autobiography: I have the sense that my arguments with Andy Clark for the extended
mind thesis (an area where prior commitments were relatively weak) may have brought more people
around than my arguments against physicalism or for two-dimensional semantics (areas where prior
commitments are strong). Even there I suspect that the thesis and the framework have brought around as
many people as the arguments. Perhaps most effective of all has been the argument in “The Matrix as
Metaphysics,” which brings many people around to the view that if we are in a matrix scenario or that if we
are brains in vats, most of our beliefs are true. (At least it does this in lecture presentations and informal
discussions; there has been relatively little discussion of the argument in print.) Although people find this
view initially counterintuitive, it turns out that their antecedent commitment was weak.
6
Then there are many other methods that I am not competent to discuss. For example, Nielsen (1987)
suggests that the one hope for progress in philosophy is critical theory.
  . 

results must be combined with further bridging premises to settle a philosophical

question. In most cases, it turns out that these bridging premises can be denied. Often
they are about as controversial as the conclusions they aim to establish. In some
cases, empirical and formal results help to settle relatively small questions, as well as
introducing and addressing new important questions. But when they are brought to
bear on the big questions, it is rare that they do much to produce consensus.
There are some partial exceptions: perhaps the bearing of evolution on theism, the
bearing of relativity on presentism, and the bearing of Gödel’s theorem on mathem-
atical formalism. But these exceptions are not especially common, and even in these
cases, there are modified versions of the relevant views that have retained numerous
serious adherents. What exceptions there are seem mainly to fall into two classes.
First, there are cases where empirical methods bear strongly on areas of philosophy
that focus on concrete reality, such as subfields of metaphysics and the philosophy of
science. Second, there are cases where formal methods bear strongly on areas that
deal with formal questions, such as the philosophy of mathematics and logic. Of
course both sorts of method are often brought to bear on other areas—normative
areas such as ethics and epistemology, for example—but cases in which they produce
consensus are much rarer.
One might think that the philosophy of mind would be an exception, given the
obvious connections to neuroscience and psychology. But even here, these sciences
seem to have left the big questions—the problems of consciousness and intentional-
ity, of mental causation and free will—wide open. Certainly there have been argu-
ments from neuroscience and psychology to views about these problems, but in most
cases the bridging premises required have been as controversial as most other
philosophical claims. What has resulted is a greatly increased sophistication with
scientifically informed versions of the relevant views, but not much more in the way
of consensus. Perhaps the greatest bearing on these big questions about the mind has
come not from these fields but from physics, where the evidence for causal closure at
the microphysical level has put serious pressure on views such as interactionist
dualism. But even here many have resisted the pressure, and in any case the denial
of interactionism does not really amount to a positive view.
Other new methods do not even offer premises with the relative security of
empirical and formal premises. In phenomenology, for example, the key phenom-
enological premises are typically as deniable as any other philosophical premises.
Something similar may apply to feminist and crosscultural philosophy, while other
methods may exhibit a mix of the two patterns above.7 Some methods, such as
feminist philosophy and experimental philosophy, have played a crucial critical role,
but the upshot has been largely to lessen our confidence about the answers to the big
questions rather than to strengthen it.
Of course, new methods are always being developed. It still happens that issues
gradually migrate from philosophy to science as methods become more rigorous and
decisive: two recent examples include the development of formal semantics and the

7
For my take on the power and limits of experimental and linguistic philosophy, see <http://consc.net/
papers/xphi.pdf> and <http://consc.net/papers/langphil.pdf> respectively.
  ’     ? 

ongoing development of a science of consciousness. Still, even in these cases it would

be hard to say that the new methods have led to consensus on the biggest philo-
sophical questions that preceded their development. So while we can hope for further
methods that produce convergence on the big questions, these methods will have to
go well beyond what we have seen over the last century or so.

14.6 Explanations
So far I have given a very partial explanation of the relative lack of convergence in
philosophy. There is less convergence in philosophy because the philosophical
method has less power to compel agreement, and it has less power because of the
phenomenon of premise deniability: arguments for strong conclusions in philosophy
(unlike science and mathematics) almost always have premises or inferences that can
be rejected without too much cost.
Still, this explanation stays fairly close to the surface of the phenomenon. It is
natural to ask for a deeper explanation. Why are arguments from consensus premises
relatively powerless to settle the big questions of philosophy? And more generally,
why is there so little convergence in philosophy?
(1) Disciplinary speciation. The most popular answer to this question, at least
among philosophers, is that the field is subject to a sort of disciplinary speciation. As
I have discussed already, many new disciplines have sprung forth from philosophy
over the years: physics, psychology, logic, linguistics, economics, and so on. In each
case, these fields have sprung forth as tools have been developed to address questions
more precisely and more decisively. The key thesis is that when we develop methods
for conclusively answering philosophical questions, those methods come to consti-
tute a new field and the questions are no longer deemed philosophical. So it is only to
be expected that the questions that remain are subject to less agreement than those in
other disciplines.
There is certainly something to this explanation. The key thesis is plausible and the
central thesis seems to follow from it. Still, I think there are some limits on this
explanation.
I have already noted one limit: the fields that have split off have not always
answered the big philosophical questions that preceded them. Psychology has not
done much to settle the mind–body problem, for example, and linguistics has not
really settled the deepest philosophical questions about meaning. Logic and physics
have come closer, but even here it is arguable that they have not settled some of the
biggest antecedent philosophical questions. Now, it might be said that the smaller
philosophical questions that these fields settle nevertheless correspond to the big
questions in the new fields, thereby explaining the central thesis about relative
convergence. Still, one wants an explanation of why the antecedent big questions in
philosophy have been so hard to answer. Insofar as these big questions have not been
resolved by disciplinary speciation, then speciation cannot answer that question.
A more general objection is that although the speciation thesis may explain,
de dicto, why there is less convergence on big questions in philosophy than on big
questions elsewhere, we also want a de re explanation, concerning those big
  . 

questions, of why they in particular receive so little convergence. One hypothesis is

that this is simply a matter of luck: all questions are equally apt for convergence, and
through random luck some have received convergence (and thereby speciated)
sooner than others. But setting aside this implausible hypothesis, the relative lack
of convergence on these questions is presumably explained by something distinctive
about those questions and their relation to us. We can then ask just what distinctive
feature (or features) of these questions explains the lack of convergence. Speciation
does not have the power to answer this question, so a further answer is required.
(2) Anti-realism. One answer is that there is no convergence to the truth because
there are no objective truths to be had in the relevant domains. Where there is
objective truth, it serves as a sort of magnet for convergence, but in its absence there
is simply an unruly body of opinion which we should not expect to converge. Many
philosophers will have sympathy to this line in some areas. I have sympathies with
anti-realism about ethics and some questions in ontology. Still, this leaves plenty to
be realist about. And even accepting moral anti-realism, say, leaves open why there is
so little convergence on the question of moral realism itself. Of course there remains
the possibility of global philosophical anti-realism, but this is not an especially
plausible or attractive view.
(3) Verbal disputes. Another answer is that there is little convergence because
participants are talking past each other. Each side is using key terms in different ways
and each is correct where their own use of the term is concerned. In “Verbal
Disputes” (2011) I argued that verbal disputes are common in philosophy. For
example, I think many debates in the philosophy of free will and the philosophy of
language have a significant verbal element. And I think that resolving verbal disputes
can lead to philosophical progress. Still, often when we clarify the key terms in a
partly verbal dispute, we find that a substantive dispute remains. And there is a core
of fundamental questions (including many normative questions, as well the mind–
body problem and other issues involving “bedrock” philosophical concepts, in the
terms of “Verbal Disputes”) for which the diagnosis of a verbal dispute seems quite
implausible.
(4) Greater distance from data. An answer naturally suggested by the discussion of
decisive arguments is that there is less convergence in philosophy than in science
because philosophy tends to concern domains that are remote from clear data. To put
this in a Quinean mode, philosophical theses are a long way from the periphery in the
network of belief. Still, on the face of it, the same goes for many highly theoretical
claims in science, for example concerning the distant past and the very small. And
plausibly the same goes for mathematics. In that case one might point to mathem-
atical axioms and intuitions as data, but this then raises the question of why we don’t
have analogous philosophical data to settle philosophical questions. So this option
tends to relabel the problem rather than to solve it.8

8
It is also worth noting (as Larry Solum suggested to me) that the social sciences have much less
convergence than the hard sciences despite being less remote from data than philosophy. An interesting
general question is whether the lack of convergence in social sciences and in philosophy should receive
different explanations or a uniform explanation. My suspicion is the former: for example, the complexity
  ’     ? 

(5) Sociological explanations. It is natural to suppose that sociological factors play a

role in preventing convergence. When our arguments are not universally accepted,
we often chalk this up partly to our opponents’ professional background, or to false
assumptions that are widespread in the profession, or to professional or emotional
attachment to alternative views.
I think there is no denying that sociological factors play a major role in determin-
ing which philosophical views are widely accepted at a time. The unpopularity of the
analytic–synthetic distinction in the decades after “Two Dogmas of Empiricism” was
certainly not uncorrelated with Quine’s position of power in the profession and his
impact on graduate students. Still, many of the relevant sociological factors are also at
play in the sciences. So to explain a difference with the sciences, one has to either
point to relevant sociological differences, or combine the sociological explanation
with other distinctive features of philosophical questions.
Some potential sociological differences include the hypothesis that philosophers
are rewarded for disagreement more than in the sciences, that they are more
tolerant of dissent, or that they have been trained to have higher standards for
acceptance of views. One could also point to differences in funding, training, and
research structures. Still, it is hard to believe that the difference in convergence
between, say, the human genome project and the mind–body problem merely
comes down to these sociological factors. So sociological explanations work best
when they are combined with further theses about the distinctiveness of philo-
sophical questions. For example, one could suggest that the greater distance
between data and philosophical theses makes it easier for sociologically grounded
resistance to a thesis to get a grip.
(6) Psychological explanations. Closely related are psychological explanations,
holding that there is something distinctive about human minds or about philo-
sophers’ minds that prevents convergence on philosophical questions. Perhaps
there is some psychological flaw that prevents us from recognizing philosophical
truth, for example. At some level some explanation like this must be part of the story:
at least if ideal reasoners could converge on the truth, then our failure can be marked
down to the nonideality of our reasoning. But now the crucial questions will be: what
are the respects in which our reasoning is nonideal, and what are the respects in
which philosophical questions are distinctive, such that this nonideality of our
reasoning prevents us from converging to the truth on philosophical questions?
(7) Evolutionary explanations. It is sometimes suggested (e.g., by McGinn 1993)
that there is a Darwinian explanation for the lack of progress in philosophy. The
rough idea is that we did not evolve to be good at philosophy, since in our
evolutionary environment there were no selection pressures that favored philosoph-
ical ability or anything that strongly correlates with it. Perhaps there is something to
this, though it would take some work to explain why the same does not apply to the
ability to do abstract mathematics or highly theoretical science. In any case this sort

and messiness of social systems seem especially relevant in the social sciences but less relevant in
philosophy.
  . 

of explanation will work best in conjunction with a psychological explanation, and

raises the same crucial questions discussed under that topic.
I think all seven of these explanations may be partially correct. I do not think that
they collectively provide a full explanation of the phenomenon as they stand, though.
To do that, many of the details would need to be fleshed out. In particular, we still
need a good account just what is distinctive about philosophical questions such that
they lead to lack of convergence. Only (2) and (4) really address this, but (global anti-
realism aside) (2) only applies in some cases, while (4) is too close to a restatement of
the phenomenon. It ought to be possible to provide an account of this distinctiveness
that meshes with psychological, evolutionary, and perhaps sociological explanations
to provide a full account of the lack of convergence. But for now I think this remains
an open question.

14.7 The Prospects for Further Progress

Finally: what are the prospects for further philosophical progress? Is it possible that
we may eventually converge to the truth on the big questions of philosophy?
To get a grip on this, we need to address the question of whether the answers to
these questions are even knowable in principle, by sufficiently ideal reasoners. Here
I will just flag my own positive view on this question. In Constructing the World,
I argued for a scrutability thesis (called Fundamental Scrutability in the book)
holding that all truths are a priori entailed by fundamental empirical truths concern-
ing fundamental natural properties and laws. It follows (roughly) that if someone
could know all the fundamental empirical truths and reason ideally, they could know
all the truths, including all the philosophical truths.
Of course the scrutability thesis can be denied. If it is false, then even ideal
reasoning from fundamental empirical truths may not enable us to know the
philosophical truth. One could preserve a modified version of the thesis by expanding
the fundamental truths in the base to include certain philosophical truths: funda-
mental normative and ontological principles, say. But then the fundamental truths
themselves may lie beyond an ideal epistemological grasp. Either way, if philosoph-
ical truths are not scrutable from an appropriate basis, we should not expect
convergence to the truth even in a community of ideal reasoners.
If the scrutability thesis is true, on the other hand, a more optimistic view ensues.
The thesis does not entail that we can know all the philosophical truths but it provides
a useful way to classify the cases where we fall short, and more generally to classify
cases where we fail to converge. First, there are cases of anti-realism about a domain,
where there is no philosophical truth to know. Second, there are cases where multiple
parties all know philosophical truths, but where verbal disputes get in the way of their
recognizing their agreements. Third, there are cases where we are ignorant of relevant
fundamental empirical truths. Fourth, there are cases where our reasoning is nonideal.
I think that many of the hardest cases in philosophy fall into the last category:
questions whose answers are knowable by ideal reasoners, but not (yet) known by us.
This then raises the key question: are the answers knowable or unknowable by humans?
  ’     ? 

McGinn (1993) and van Inwagen (2009) have advocated unknowability: humans
are just not smart enough to answer the big questions. The idea is that there is some
level of intelligence or aptitude that would suffice to answer these questions, but that
humans fall below that level.
Van Inwagen argues for this conclusion as follows. He suggests that it is
implausible that we are much above that level, given the lack of progress to date,
and that it is antecedently improbable that we should be just barely at that level.
So it is much more likely that the level lies above us. I am not so sure about
this argument. I think we already know that for a vast range of questions, humans
are just barely at the level for doing them well: scientific and mathematical
questions, for example. Because of this, it is arguable that we lie at a special
intelligence threshold at which an extraordinarily wide range of questions come
to be within our grasp over time. It is not obvious whether philosophical questions
fall within that range, but it is not obviously more likely that they do not than
that they do.
If McGinn and van Inwagen are right, it remains open that we could answer
philosophical questions by first improving our intelligence level, perhaps by cognitive
enhancement or extension. Alternatively, we could construct artificial beings more
intelligent than us, who will then be able to construct artificial beings more intelligent
than them, and so on. The resulting intelligence explosion might lead to creatures
who could finally answer the big philosophical questions.
If McGinn and van Inwagen are wrong, on the other hand, then we may eventually
answer philosophical questions without radical cognitive enhancement. We may
need new methods, increased discipline, new sorts of insights, and perhaps a con-
ceptual revolution or two, but none of this will lie outside human capacity. It may
turn out that there is a curve of increasing philosophical sophistication such that past
a certain point on the curve, major progress is possible. We are not there yet, but we
are working our way toward it.
It is not obvious whether McGinn and van Inwagen are right or wrong. The
question of whether the big philosophical questions are humanly solvable is itself a
big metaphilosophical question. Like other big questions in philosophy, it is one we
do not currently know the answer to. Both answers to this metaphilosophical
question seem to be open, and we do not currently have strong reasons to favor
either one.
If we don’t know which of these two options obtains, then I think to do
philosophy we can make the working assumption that it is the second: the ques-
tions are answerable by us but as yet unsolved. Then we can simply do philosophy
as well as we can, doing our best to come up with those new insights, methods, and
concepts that might finally lead to us answering the questions. After all, we are still
learning to do philosophy well. To see how far it can take us, we have to keep doing
philosophy.

Acknowledgments
I first gave a brief version of this paper at the Harvard-Australia conference on Progress in
Philosophy at Harvard University in 2011. I thank that audience and those on subsequent
  . 

occasions in Arizona, Cambridge, Fordham, Liverpool, Oslo, Rio, Santiago, and at the Royal
Institute for Philosophy lecture in London. I am also grateful to participants in a number of
useful Internet discussions. For comments on the written version, thanks to Russell Blackford,
Melissa Ebbers, Alan Hájek, Robin Hanson, John Keller, Mark Lance, Seth Lazar, Christian
List, Luke Muehlhauser, Rick Repetti, and Joshua Weisberg. And thanks to Peter van Inwagen
for interesting conversations about these issues over the years.

References
Bourget, D., and Chalmers, D. J. (2014), ‘What Do Philosophers Believe?’, in Philosophical
Studies 170: 465–500.
Chalmers, D. J. (2011), ‘Verbal Disputes’, in Philosophical Review 120: 515–66.
Chalmers, D. J. (2012), Constructing the World (Oxford University Press).
Dietrich, E. (2011), ‘There Is No Progress in Philosophy’, in Essays in Philosophy 12: 329–44.
Hilbert, D. (1902), ‘Mathematical Problems’, in Bulletin of the American Mathematical Society
8: 437–79.
Kelly, T., and McGrath, S. (2017), ‘Are There Any Successful Philosophical Arguments?’, in
J. Keller (ed.) Being, Freedom, and Method: Themes from van Inwagen (Oxford University
Press).
Krippendorff, K. (2013), Content Analysis: An Introduction to its Methodology (3rd edn.) (Sage
Publishing).
McGinn, C. (1993), Problems in Philosophy (Oxford University Press).
Moody, T. (1986), ‘Progress in Philosophy’, in American Philosophical Quarterly 23: 35–46.
Nielsen, K. (1987), ‘Can There Be Progress in Philosophy?’, in Metaphilosophy 18: 1–30.
Rawls, J. (2001), ‘Afterword’, in J. Floyd and S. Shieh (eds.), Future Pasts: The Analytic
Tradition in Twentieth-Century Philosophy (Oxford University Press).
Russell, B. (1912), The Problems of Philosophy (Williams and Norgate).
Russell, B. (1918), ‘The Philosophy of Logical Atomism’, in The Monist 28: 295–527. Reprinted
in 2009 as The Philosophy of Logical Atomism (Taylor and Francis).
Stoljar, D. (forthcoming), Philosophical Progress: In Defense of a Reasonable Optimism.
Van Inwagen, P. (2004), ‘Freedom to Break the Laws’, in Midwest Studies in Philosophy 28:
334–50.
Van Inwagen, P. (2006), The Problem of Evil: The Gifford Lectures (Oxford University Press).
Van Inwagen, P. (2009), Metaphysics (3rd edn.) (Westview Press).
Williamson, T. (2006), ‘Must Do Better’, in P. Greenough and M. Lynch (eds.), Truth and
Realism (Oxford University Press).
Yandell, B. H. (2002), The Honors Class: Hilbert’s Problems and their Solvers (A. K. Peters).
 15
Philosophical Individualism
John A. Keller

If two people agree, one of them isn’t a philosopher.1

15.1 Introduction
It is natural to think that successful philosophical arguments are ideally publicly
recognizable: recognizable by all rational, informed, and fair-minded inquirers. Peter
van Inwagen, for example, has defended a view according to which successful
philosophical arguments are those that would be convincing—when ideally pre-
sented in the company of an ideal opponent—to an audience of ideal neutral
agnostics about the disputed question. Whether an argument meets this criterion is
ideally publicly recognizable: insofar as the merits of an argument are not publicly
recognized, it is because of ignorance, irrationality, or bias.2
Of course, things are not ideal, and the actual public is bound to misclassify some
arguments. Still, we might feel comfortable formulating coercive public policy on the
basis of arguments that meet van Inwagen’s criterion, since anyone that rejects the
conclusion of such an argument is suffering (perhaps non-culpably) from one of
the aforementioned maladies. The hope that such arguments provide the means to
peaceably settle public disputes has been pervasive and influential. John Rawls, for
example, holds that:
. . . citizens are to conduct their public political discussions of constitutional essentials and
matters of basic justice within the framework of what each sincerely regards as a reasonable
political conception of justice, a conception that expresses political values that others as free
and equal also might reasonably be expected reasonably to endorse. (Rawls 1993: xlvii, emphasis
added)
Along related lines, Timothy Williamson notes that:

1
This quip appears on p.3 of Rescher (1985). I have been unable to determine its provenance.
2
The idealization of the audience ensures that they respond rationally; the presence of an ideal
opponent ensures that the audience is informed; and the audience’s neutral agnosticism ensures that
they are not biased for or against the conclusion. See van Inwagen (2006: Lecture 3).
  . 

. . . we might hope that whether a proposition constitutes evidence is in principle unconten-

tiously decidable, in the sense that a community of inquirers can always in principle achieve
common knowledge as to whether any given proposition constitutes evidence . . . 3 (Williamson
2007: 210)

But alas, according to van Inwagen, this hope is in vain: there are no arguments for
substantive philosophical conclusions that meet his criterion, or any other publicly
recognizable criterion, even in ideal circumstances. If this is correct, we face a
dilemma: we can conclude that there are no successful arguments for substantive
philosophical theses, or adopt a non-ideally publicly recognizable conception of
success. But how could successful arguments fail to be even ideally publicly recogniz-
able? Is such a watered-down notion of “success” really success at all? My aim in this
essay is to argue that it is. There are good independent reasons to adopt a (non-
publicly recognizable) “private” or “individualistic” conception of success: the fact that
we thereby gain a few successful arguments for substantive philosophical conclusions
is just icing on the cake.

15.2 Philosophical Success

Most philosophers do not think that all philosophical arguments are failures. Most of
us find some philosophical arguments convincing, our own if no others. But how
many of the arguments we deem successful are for substantive conclusions? How
many are constructive, as opposed to destructive, telling us the way the world
is, rather than how it isn’t? And if we think that certain constructive arguments are
successful, what proportion of other philosophers will agree with us? Since (appar-
ently rational) philosophers disagree about almost everything, every argument for a
substantive philosophical thesis can be rejected without manifest irrationality.4 If
successful philosophical arguments must convince all rational comers, it would
appear that there are none. According to van Inwagen, however, this sets the bar
too high. He holds that an argument that does not convince all rational comers can
be successful if it would convince all ideal rational comers. More carefully, he holds
that an argument for conclusion c is successful just if it would be convincing—when
ideally presented in the company of an ideal opponent—to an audience of ideal
neutral agnostics about c. Ideal neutral agnostics about c have no opinion or leanings
concerning the truth or falsity of c, and are as intellectually virtuous as is humanly
possible. Since it is easier to convince an audience of neutral agnostics than it is to
convince an audience of die-hard opponents, this account sets the bar for philosoph-
ical success lower than an account that requires successful arguments to convince all
rational comers.

3
Note that common knowledge about what the evidence is wouldn’t necessarily lead to common
knowledge about what the evidence supports. In any case, Williamson argues that it is not even commonly
knowable what the evidence is. See section 15.3.4.
4
The PhilPapers survey and common experience both testify to the lack of agreement amongst
philosophers. See Bourget and Chalmers (2014) for a summary of the results of the survey. What it
means for a conclusion to be substantive is trickier than it might seem: see section 16.2 of Kelly and
McGrath (2017) for a nice discussion.
   

The primary motivation that van Inwagen gives for his view hinges on this insight.
Van Inwagen argues that any plausible account of success other than his own is
bound to run afoul of:
The Desideratum: A criterion of success should not entail that there are no
successful arguments for substantive philosophical conclusions.
The Desideratum expresses a plausible constraint on theories of success. But there is an
even stronger—indeed, non-negotiable—constraint, which may be expressed as follows:
The Weak Desideratum: A theory of success must not entail that paradigmatically
successful arguments—arguments for established scientific, historical, and non-
substantive philosophical conclusions—are not successful.
Any account of success that runs afoul of The Weak Desideratum is clearly
inadequate. Another such non-negotiable desideratum on a theory of success is:
The Requirement: A criterion of success must entail that those in possession of a
successful argument for conclusion c are, ceteris paribus, not rationally criticizable
for accepting c.
The Requirement is, I think, the core of our concept of success. When someone
believes something outré, we think they had better have an argument for it. And not
just a clever or interesting argument. (This is especially clear if their belief is both outré
and objectionable.) Whatever else successful arguments must do, then, they must make
it rational, or at least not irrational, to accept their conclusions.5 For reasons that will
become clear in section 15.4.3, however, I do not think that satisfying The Requirement
is in fact sufficient for success. But it is certainly necessaary.

15.2.1 Objective Success

Van Inwagen’s conception of success is dialectical—it measures success in terms of
its ability to convince some audience. But one might wonder why we shouldn’t adopt
an objective (attitude independent) conception of success. Why not hold that
successful arguments are sound? Call this the Soundness Account. Isn’t the Sound-
ness Account what we teach our students? Since sound arguments for substantive
philosophical conclusions obviously exist, the Soundness Account does not run afoul
of the Desideratum. It also satisfies the Weak Desideratum. The problem is with The
Requirement: sound arguments can have premises that we do not know or even
believe to be true. We know of sound arguments for substantive conclusions about
free will, morality, metaphysics, epistemology, etc. The problem is that we know of
unsound arguments too, and we don’t know which arguments are the sound ones.
For all we know, the zombie argument for property dualism is sound.6 But for all we
know it isn’t. For all we know, the consequence argument for incompatibilism is
sound,7 and for all we know it isn’t. And so on.

5
See sections 15.4 and 15.5 for further discussion. Note that, for reasons discussed in sections 15.2.2.1
and 15.2.3 (and in Keller 2015b), we should not require successful arguments to be such that one could
come to know their conclusions on the basis of their premises.
6 7
Chalmers (1996). Van Inwagen (1983).
  . 

It isn’t just that we don’t recognize the goodness of some sound arguments. Some
sound arguments are bad. Consider the following arguments:
1. The Peano axioms 1. The Peano axioms
2. The Peano axioms ⊃ Goldbach’s 2. The Peano axioms ⊃ Goldbach’s
Conjecture Conjecture is false
3. Therefore, Goldbach’s Conjecture 3. Therefore, Goldbach’s Conjecture
is false
One of these arguments is sound. Call that the Bad Sound Argument (BSA). The
BSA is not just sound: it is a transparently valid sound argument. But it isn’t a
successful argument, in any interesting sense of ‘successful’. In fact, it seems to be
a manifestly bad argument—if the BSA is the only argument I have in support of its
conclusion, I should not believe its conclusion. The reason for this is simple: I have
no reason to accept its second premise.8 The Soundness Account, then, runs afoul of
The Requirement. Having a sound argument for a conclusion c does not make it
rationally defensible to accept c if one does not believe, much less justifiably believe,
the premises of that argument. Some sound arguments have premises that are not
believed, others have premises that are literally unbelievable, and some have premises
that no human could be justified in believing.9 So the Soundness Account cannot
be correct.
Can the BSA be ruled out as fallacious? If so, it would not run afoul of the
Soundness+ Account, according to which non-fallacious sound arguments are suc-
cessful. But what fallacy might the BSA illustrate? The only plausible candidate is the
fallacy of begging the question (if there is such a fallacy), but if there is some subtle
sense in which the BSA is question-begging, that sense will almost certainly not be
objective: it will have to do with how one could know the premises and conclusion of
the argument, or with the context in which an argument is used. For example,
Sinnott-Armstrong (1999) argues that “to avoid begging the question one’s reason
to believe the premise must be independent of both (a) one’s belief in the conclusion
and also (b) one’s reason to believe the conclusion.” Perhaps some uses of the BSA
beg the question in this sense, but this sense of begging the question is not objective: it
is relativized to the beliefs and reasons possessed by the individuals assessing the
argument. So while objective criteria like soundness may be necessary for success,
they are not sufficient. To satisfy The Requirement, successful arguments must at
least have premises that are rationally believable. But then it is obvious why we need a
dialectical conception of success: different things are rationally believable to different
people. A sound argument may fail to convince me—it may be rational for it to fail to
convince me—if I have no reason to believe its premises. It follows that since even
paradigmatically non-deductive arguments can be easily formalized, a theory of
success is primarily a theory about the class of individuals that need to be convinced

8
Given that Goldbach’s Conjecture has been neither proven nor disproven, what could that reason be?
9
Trivially, any premise that is literally unbelievable is a premise that we could not justifiably believe, but
there are believable truths for which we could not have justification: e.g., many truths about events outside
of our light cone.
   

by an argument in order for it to be a success. In other words, a theory of success is

primarily a theory about the proper audience for philosophical arguments.10

15.2.2 Dialectical Success

One could take a number of different views about the nature of this audience.

... 
One view is that the proper audience for an argument is everybody, or perhaps
everybody rational. As noted above, however, no argument for a substantive philo-
sophical conclusion has convinced all comers. The problem isn’t just that this
criterion fails to satisfy The Desideratum—it also fails to satisfy the Weak Desider-
atum. As Williamson notes,
Having good evidence for a belief does not require being able to persuade all comers, however
strange their views, that you have such good evidence. No human beliefs pass that test.
(Williamson 2007: 212)
Logical heretics reject standard mathematical proofs, but this does not show that
those proofs are not successful arguments. This is especially obvious if the heretics
are wrong. But a correct proof is a paradigmatically successful argument, and (even
heterodox) logicians are paradigmatically rational. So successful arguments cannot
be defined as arguments that everyone (or everyone rational) will accept.11
Indeed, as the sociologists of science delight in pointing out, scientific theories often
become dominant only because their opponents die. It is much easier to convince
(initially agnostic) graduate students of some theory t than it is to convince those whose
careers have been defined by their opposition to t. Einstein famously (or infamously)
never fully endorsed quantum mechanics, but that hardly entails that there weren’t
successful arguments for the standard theory of elementary particles prior to Einstein’s
death. Ernst Mach went to his grave dissenting from the atomic theory of matter, etc.
This fact—that successful arguments do not always convince everybody, or even
everybody rational—is recognized in certain special cases. It is widely conceded, for
example, that no one has devised an anti-skeptical argument that is convincing to
skeptics. But that does not mean that there are no successful responses to skepticism.
As Williamson (2007) notes,
[I]f one uses only premises and forms of inference that a skeptic about perception will allow
one . . . one has little prospect of reaching the conclusion that one has hands. But that does not
show that we should not be confident that we have hands. (238)
Of course, skeptics will say that . . . claims about our environment [like “dreams with the
sustained coherence of waking life are very rare”] merely beg the question . . . But the claims
were not addressed to skeptics, in a futile attempt to persuade them out of skepticism. Instead,

10
The idea that argumentative success is objective seems to be a legacy of the unfortunate idea that
successful arguments are proofs: transparently (at least step-wise) valid arguments with communally
apodictic premises. But if there are premises that are communally apodictic, we are obviously unable to
prove much of scientific or philosophical interest from them.
11
See Williamson (2007: ch.4) for a discussion of related issues.
  . 

they figure in our appraisal of skeptical arguments, from our current non-skeptical point
of view.12 (249)

On this view, non-skeptics can successfully defend their knowledge by demonstrating

that the cost of skepticism is unacceptably high, even if skeptics themselves couldn’t
come to know that we have knowledge on the basis of those arguments, since those
arguments presuppose that we have knowledge (e.g., knowledge that “dreams with the
sustained coherence of waking life are very rare”).13 Many non-skeptics know
that they have such knowledge, so this presupposition is (at least according to
Williamson) a legitimate one. Still, it is not a presupposition skeptics can coherently
accept. Such anti-skeptical arguments are essentially defensive, but they do satisfy The
Requirement. This, in conjunction with our stipulation that the non-skeptics know
the premises of their arguments to be true, strongly suggests that these non-skeptical
argument are successful.
Now, one might think that those who hold onto their pet theories in the face of
successful arguments to the contrary are either irrational or uninformed, and hence
that the view that successful arguments must convince idealizations of one’s oppon-
ents does not run afoul of our desiderata. But however much we might wish it were
so, skeptics are not always irrational or uninformed. Similarly, Einstein certainly
wasn’t uninformed about quantum mechanics, and it is far from obvious that his
resistance to it was irrational.
In any case, there is an obvious reason why different people will be convinced by
different arguments. An argument for c is a presentation of some reasons or evidence
for c. But whether an argument gives one a compelling reason to accept c depends on
one’s other evidence, one’s priors, etc. If, as some people believe, no consistent set of
priors is more rational than any other, it follows that no argument is guaranteed to
make it irrational for one’s interlocutors to reject c. For if one’s interlocutor’s priors
related to c can be arbitrarily low, it will take an argument of arbitrary strength to
make it irrational for her to reject c. Similarly, since one’s interlocutor might have
arbitrarily strong evidence (however misleading) against c, it will take an argument of
arbitrary strength to make it irrational for her to deny c.14 If this is correct, The
Requirement and The Weak Desideratum jointly entail that successful arguments
cannot be those that convince all rational comers.

... 
And so we are led to a conception of success like van Inwagen’s, where a successful
argument for c must only be convincing to ideal neutral agnostics about c, even if it is
not convincing to everyone, including rational opponents of c. But does this criterion
do any better than the others with respect to our desiderata? It plausibly satisfies The
Requirement and The Weak Desideratum, but are there arguments for substantive

12
See Pryor (2000; 2004) and Kelly (2005) for a nice discussion of related issues.
13
Cf. the Introduction to Lewis (1983) on arguments being tools for “measuring the cost” of denying
their conclusions. See also Keller (2015b).
14
I take belief and knowledge to be more fundamental than credences and priors, but talk of credences
and priors is sometimes illuminating. Having “high” or “low priors” may be paraphrased as being inclined
to believe or disbelieve for those who prefer sticking to a single idiom.
   

philosophical conclusions that are convincing, when ideally presented in the pres-
ence of an ideal opponent, to rational agnostics? It is hard to believe that there are,
since, over time, one would think that the existence of such arguments would lead to
significantly more convergence on substantive philosophical theses than we in fact
observe. As van Inwagen puts it,
If any reasonably well-known philosophical argument for a substantive conclusion had the
power to convert an unbiased ideal audience to its conclusion (given that it was presented to
the audience under ideal conditions), then, to a high probability, assent to the conclusion of
that argument would be more widespread among philosophers than assent to any substantive
philosophical thesis actually is.15 (Van Inwagen 2006: 53)
But if van Inwagen’s criterion fails to satisfy The Desideratum, why doesn’t he adopt
a more liberal one? “Alas,” van Inwagen writes, “there is no more liberal criterion.
The criterion I have proposed is the most liberal possible criterion” (p.160, n.5). If all
criteria of philosophical success are partners in guilt when it comes to The Desider-
atum, however, one might wonder why we should favor van Inwagen’s criterion over
any other.
There are two aspects of van Inwagen’s criterion that make it preferable to the
others we have discussed so far: as noted above, it seems to satisfy The Requirement
and The Weak Desideratum, and it also does a better job of assigning the “burden of
proof” than these other criteria. According to van Inwagen, arguments for c may
properly rely on premises that opponents of c reject, as long as those premises would
be acceptable to agnostics about c. As van Inwagen says,
Norma the nominalist need not worry about whether Ronald the realist will accept her
premises. She is perfectly free to employ premises she knows Ronald will reject; her only
concern is whether the audience of agnostics will accept these premises. Suppose, for
example, that she uses the premise, “We can have knowledge only of things that have the
power to affect us.” It may well be that no realist . . . would accept that premise . . . [But if]
Ronald thinks that there is any danger of the agnostics accepting this premise, it will do him
no good to tell the audience that of course no realist would accept this principle and that it
therefore begs the question against realism. He’ll have to get down to the business of
convincing the agnostics that they should reject, or at least not accept, this premise. (Van
Inwagen 2006: 46)
The idea that we may properly use premises that our opponents reject is an
important insight that an adequate account of success should preserve. But this
reason for preferring van Inwagen’s account is beside the point, since van Inwagen’s
is not the most liberal possible criterion. Philosophical Individualism—roughly, the
view that an argument A for conclusion c is successful for individual i if and only if
A is convincing to i—is more liberal, in that it allows some arguments for substantive
philosophical conclusions to be successes. And so, it would seem, The Desideratum

15
See Kelly and McGrath (2017) for a critique of this line of reasoning, and Ballantyne (2014) for an
argument that there are “knockdown arguments” for substantive philosophical conclusions. (But see Keller
(2015b) for a critique of Ballantyne’s argument.)
  . 

supports philosophical individualism over both van Inwagen’s view and the other
more demanding views he rejects.16

15.2.3 Philosophical Individualism

What makes philosophical individualism “individualistic” is that it relativizes success
to individuals, holding that ‘x is a successful argument’, like ‘x is in motion’, has
implicitly relativistic truth-conditions—an argument might be a success for you
without being a success for me. That is, philosophical individualism relativizes the
success of an argument to the person evaluating it. There are several different flavors
to choose from:
The Belief Account: an argument is successful for an individual just if she believes
it is sound (and non-fallacious)
The True Belief Account: an argument is successful for an individual just if
she correctly believes it is sound (and non-fallacious)
The Justification Account: an argument is successful for an individual just if she
justifiably believes it is sound (and non-fallacious)
The Knowledge Account: an argument is successful for an individual just if
she knows it is sound (and non-fallacious).17
One way to think about these criteria is as specifying the type of epistemic gain
typically produced by successful arguments: new beliefs, new true beliefs, new
justified beliefs, or new knowledge. An argument that one, e.g., knows to be sound
(and non-fallacious) typically could be used to gain knowledge of its conclusion. But
I want to remain neutral about whether successful arguments must always be able to
provide one with some epistemic gain, since that would rule out essentially defensive
arguments—arguments that can only be used to defend conclusions one already
accepts (see section 15.2.2.1 and Keller 2015b), and arguments for things we know
essentially, such as, perhaps, that we exist. While it isn’t clear that there is anything
we know essentially, there could be an essentially omniscient being, or at least a
maximally epistemically good being—an epistemic god—for whom no argument

16
Fischer and Tognazzini (2007) suggests a liberalization of van Inwagen’s view that would only require
successful arguments to sway, rather than convince, an idealized neutral audience—to get them to change
their credences, even if they don’t change their minds. A similar account is proposed in Chalmers
(forthcoming), and van Inwagen (2017: 385) now endorses this modification of his 2006 view. There are
problems, however: if the account merely requires successful arguments to raise the credences of ideal
neutral agnostics at all—say, from .50 to .51—it runs afoul of The Requirement, since not all such
arguments articulate satisfactory reasons to accept their conclusions. But if the account requires successful
arguments to raise the credences of neutral agnostics to the threshold of rational belief, however, it satisfies
The Requirement but collapses back into van Inwagen’s original account. See section 15.3 for further
reasons to be skeptical of this approach.
17
Recall that we are focusing on deductive arguments, and note that when I say that i knows/believes/
etc. an argument A to be sound, this should be read as shorthand for “i knows/believes/etc. A’s premises
and i either knows/believes/etc. that A is valid, or A is transparently valid.” This is important since someone
could know, by testimony, that A was sound without having a clue what the argument was. Since such a
person may not even believe A’s premises, they do not know A to be sound in the relevant sense.
   

could provide anything of epistemic value. Still, an epistemic god could have suc-
cessful arguments—if anyone has successful arguments, epistemic gods do! Those
arguments could not provide the god with any epistemic gain, but they would
still satisfy The Requirement—they would still demonstrate the rationality of the
god’s beliefs.
In any case, satisfying the Knowledge Account seems clearly sufficient for success.
This account is more liberal than van Inwagen’s—it satisfies The Desideratum—since
there are some arguments for substantive philosophical conclusions that are known
by some people to be sound. For example, since I know that there are anatomical
properties shared by spiders and insects, and I know that there are anatomical
properties shared by spiders and insects entails there are properties, I have a successful
argument for the existence of properties. Successful for me, that is. It isn’t generally
successful: most nominalists don’t believe, and so don’t know, that that entailment
holds. And since I know this argument to be sound, it is clearly rational for me to
accept its conclusion—thus satisfying The Requirement.
In section 15.4 I argue that the Knowledge Account is correct: that knowing an
argument to be sound is also necessary for it to be successful (and thus that it satisfies
the Weak Desideratum). But determining which properties of individuals success is
relativized to is less important than realizing that success is relativized to individuals
in the first place. The idea that correctly evaluating an argument does not depend on
anyone’s attitude towards the argument but one’s own is a significant departure
from—and improvement upon—both our pre-theoretic way of thinking about suc-
cess and the dominant trend in the literature. An account of success that relativizes it
to something else about individuals—say, their justified beliefs—would still be better
than one that relativizes success to the properties of some third party.
The picture of philosophy as the first-person pursuit of understanding, found in the
work of philosophers like Aristotle, Descartes, and Marcus Aurelius, aligns well with
such individualistic conceptions of success. We are often assisted in this pursuit by
interaction with and input from others, but it is our own epistemic position that is the
arbiter of success. Even if we typically hope that our arguments will be of use to other
people struggling with the same or similar problems, the success of our arguments is
to be measured in terms of their ability to solve our problems. As Robert Nozick says,
Some of the things the skeptic says or points out . . . I accept; these are or become part of my
own belief system. My problem is that I don’t see (or no longer see, after the skeptic has
spoken) how these things go along with yet other things in my belief system . . . My task here is
to remove the conflict, to put my own beliefs in alignment, to show how those of the things the
skeptic says which I accept can be fit in with other things I accept.18 (Nozick 1981: 16)
All that is required to be a philosophical individualist is to think that the dialectical
standards Nozick claims are appropriate for arguing with the skeptic apply in general
(recall section 15.2.2.1). Our goal is not—or should not be—to convince our oppon-
ents that they are wrong. It is to see for ourselves why they are wrong—to uncover

18
Here Nozick seems to be assuming that believing an argument to be sound is sufficient for success.
This assumption is widespread: see, e.g., Sinnott-Armstrong (1999).
  . 

the flaws in their arguments. Arguments can be successful for some people and
not others, and the aim of philosophical inquiry is producing arguments that are
successful for us.

15.3 In Support of Philosophical Individualism

We saw above that The Desideratum and The Requirement seem to support (some
species of ) philosophical individualism. (And philosophical individualism is cer-
tainly compatible with the Weak Desideratum.) But there are a number of additional
reasons for thinking that success is individualistic.

15.3.1 Essentially First-Person Arguments

Sometimes I give arguments of the form “I remember that p, so p”. In many such
cases I couldn’t have a better argument for p: vividly remembering that p trumps—
rationally trumps—almost any degree of external objective evidence that not-p.
Sometimes my “memory arguments” are convincing to others, but there are cases
where others ought not be convinced, even though it is rational for me to be. For
example, let p be the claim that I was not at the scene of Jones’s murder, and suppose
I have been exquisitely framed for Jones’s murder. It might then be rational for me to
maintain my innocence in the face of the objective evidence to the contrary, since
I vividly remember doing something else at the time of Jones’s murder. But it
might not be rational for a jury to agree with me. Memory arguments are much
more compelling when it is one’s own memory that is being appealed to—in the first-
person case they are almost always sufficient to produce knowledge.19 Other such
essentially first-person arguments seem relevant to debates about phenomenology,
motivational internalism/externalism, the significance of religious experience, and
one’s own existence. Even if Descartes didn’t mean to be giving an argument for his
own existence (since he was suspending belief about the reliability of his reasoning),
we can formulate Cogito-esque arguments—essentially first person arguments for our
own existence—and those arguments are clearly successful. ‘I think, therefore I am’ is
a superb argument for my existence: indeed, it is rationally compelling. Rationally
compelling for me that is. It is much less compelling to others, including neutral
agnostics about my existence.

15.3.2 Against Neutrality

The primary appeal of criteria like van Inwagen’s over philosophical individualism is
that, by removing the requirement that successful arguments be convincing to a
neutral audience, philosophical individualism allows bias to (partially) determine
argumentative success or failure.20 While an agnostic about p neither accepts nor
rejects p, a neutral agnostic about p assigns p and not-p equal probability. There are,

19
This is not to say that knowledge based on memory is typically formed via memory arguments:
normally such knowledge is non-inferential.
20
Where by ‘bias’ I mean simply non-neutrality. Such bias needn’t be irrational: it might be rational, or
it might be non-rational without being irrational. (E.g., according to subjective Bayesians, having consistent
but loaded priors makes one non-rationally but not irrationally biased.)
   

however, problems with thinking that successful arguments must convince neutral
agnostics. First, neutral agnosticism is infectious. If neutral agnostics about p only
had to be agnostic about p itself, some such agnostics might firmly believe that all
arguments for or against p are sophistical. Such agnostics would be unlikely to be
convinced by any argument for or against p, successful or not.21
But once we see that neutral agnostics about p must be agnostic about more than p
itself, it is hard to know where to stop. Would neutral agnostics about the existence of
free will have to be compatibilists? Moral realists? Consequentialists? Physicalists?
Theists? Scientific realists? Even if our agnostics were strictly neutral about the
existence of free will, consider how differently “primed” towards free will someone
would be who was an incompatibilist, moral nihilist, atheist, physicalist, and scien-
tific realist compared with someone who was a compatibilist, moral realist, theist,
dualist, and scientific anti-realist. Of course, one can believe in free will as an
incompatibilist, moral nihilist, atheist, physicalist, and scientific realist without
manifest irrationality: Mark Balaguer accepts all of those except moral anti-realism,22
and there is no reason to think that adding moral anti-realism to the mix would make
his position inconsistent. But convincing Balaguer to reject free will seems much
easier than convincing someone who is agnostic about free will but who is a
compatibilist, moral realist, theist, dualist, and scientific anti-realist. Explicit oppos-
ition to a claim doesn’t guarantee that the totality of one’s views doesn’t “mesh”
better with that claim than does another totality that is neutral about the claim.
Of course, to be truly neutral with regard to free will one would also have to be
agnostic about determinism, consequentialism, reductionism, and much else. If the
members of a jury of “ideal neutral agnostics” were not agnostic about these theses,
they might come to different verdicts on the basis of their differing opinions. But an
audience that was agnostic about all of these topics would believe so little that it is
hard to imagine convincing them of anything at all.
An additional problem is that agnostics are not neutral in debates about agnosti-
cism itself, religious or otherwise. Pyrrhonian skeptics aim only to get us to suspend
belief about whether there is an external world, so someone who is agnostic about
whether there is an external world already accepts the skeptic’s position.23 When it
comes to arguments that are skeptical in this sense, there is no neutral vantage point.
But if anti-skeptical arguments do not need to appeal only to premises acceptable to
the skeptic, and they do not need to appeal only to premises acceptable to someone
agnostic about skepticism, what is left? The only plausible answer is that they can
appeal to premises that non-skeptics themselves accept. But that supports philo-
sophical individualism over views that claim that successful arguments must utilize
premises acceptable to everyone, a neutral audience, etc.24

21
Thanks to John Hawthorne for this point.
22
See, e.g., Balaguer (2010). I should note that Balaguer’s acceptance of moral realism, free will, and
incompatibilism is hedged.
23
Can one coherently be an agnostic about whether we can know the truth-value of x without being an
agnostic about x? If so, one could coherently claim to know that, say, God exists while being agnostic about
whether we can know if God exists. And that does not seem coherent.
24
Note that it is not clear to what extent bias accounts for disagreement in philosophy anyway, since
there is persistent apparently rational disagreement about philosophical questions about which no one (or
  . 

15.3.3 Against Idealization

There are also problems with the idea that the success of an argument depends on
what some idealized group of people think or feel. Most obviously, it is unclear how
the effect of an argument on such an audience would be relevant to the argument’s
actual satisfaction of The Requirement. Whether an argument produces some
epistemic gain (justification, knowledge, etc.) depends crucially on the beliefs of the
actual persons to whom it is addressed. An idealized version of myself might not have
the problem that the argument I am considering is attempting to remedy. Much
philosophy aims at solving problems that arise only because of our non-ideality, after
all. Philosophers aim to discover the entailments of claims like ‘Aquinas admired
Aristotle’, ‘Lois believes that Superman can fly’, and ‘Joe freely chose to lie to Mary’.
Does the first entail that there are objects that are not present? Does the second entail
that Lois believes that Clark Kent can fly? Does the third entail that Joe was not
determined? These are paradigmatically philosophical questions that a logically ideal
audience would have no reason to ask.25
A logically ideal audience would not only fail to face such philosophical problems,
such an audience could not be neutral about the solutions to such problems, since
they would know the solutions. Combining ideality and neutrality generates prob-
lems that ideality and neutrality do not face individually.26 Finally, note that many
people are persuaded by arguments aiming to show that there are no ideal persons.
The inability of those arguments to convince an audience of ideal persons is clearly
beside the point.27

15.3.4 Non-Neutrality
Further support for philosophical individualism may be derived from “evidence non-
neutrality.” Williamson (2007) argues that we should not expect all (rational) parties
to be able to agree on what the evidence is, or what it supports. On Williamson’s
view, one’s evidence is one’s knowledge, and so, since different people know different
things, an argument might be compelling evidence for someone who knows its
premises but not for someone who doesn’t. The non-neutrality of evidence is not
tied to Williamson’s account of evidence, however: taking one’s evidence to be one’s
justified beliefs makes evidence no more neutral than taking it to be knowledge.28
And on any account where evidence is non-neutral, there are striking implications
for how we evaluate arguments—implications that lend support to individualistic
conceptions of success:

hardly anyone) has a preconceived opinion: Newcomb’s Paradox, the nature of mental representation,
quantifier variance, whether properties are parts of their instances, and so on.
25
I earlier glossed van Inwagen’s ideal audience as being as rational as is humanly possible, rather than
logically omniscient. This may help with the objection here, but it opens space for the reaction of such an
audience to be the result of logical mistakes or shortcomings.
26
Kelly and McGrath (2017) argue along similar lines that ideal agnostics are not the relevant audience
for determining success.
27
Thanks to John Hawthorne for this point.
28
If, as Williamson (2000) argues, no or virtually no conditions are luminous, there is nothing capable
of playing the evidence role that we can expect all parties to be in a position to recognize.
   

How much do failures of Evidence Neutrality threaten the conduct of philosophy? From an
internal perspective, they make consensus harder. Each of many conflicting theories may be
the one best supported by the evidence by its own lights. The role of evidence as a neutral
arbiter is undermined. From an external perspective, both the good fortune of being right and
the misfortune of being wrong are magnified. If your theory is true, so are its consequences for
which propositions constitute evidence . . . If your theory is false, it may have false consequence
for which propositions constitute evidence . . . (Williamson 2007: 213)

A related view is defended by Michael Bergmann, who argues that believed defeaters
are defeaters: that one cannot know that p if one believes, even incorrectly, that one
has a defeater for p.29 If we think of defeaters for p as a form of evidence against p, this
view also entails that evidence is non-neutral, since different people—even people
who have the same “objective” defeaters for p—may believe they have, and so have,
different defeaters for p.30
Robert Nozick has argued for a similar position regarding the resolution of
antinomies: apparently inconsistent sets of apparently true sentences. We resolve
an antinomy when we come to see that one of those appearances is illusory—that the
apparently inconsistent sentences are not inconsistent after all, or that the apparent
truth of one or more of the sentences is an illusion. As Nozick puts it,
Given the (apparent) incompatibility between the apparent [truths] and p, there are two ways
to continue to maintain . . . p. First, one of the apparent [truths] can be denied, or there can be a
denial of their conjunction all together . . . Second, each of the apparent [truths] can continue to
be maintained, while their apparent incompatibility with p is removed, either by close scrutiny
showing the reasoning from them to not-p to be defective, or by embedding them in a wider
context or theory that specified how p holds in the face of these apparent [truths].31 (Nozick
1981: 10)

Since different apparent truths appear true to different people, and different claims
appear incompatible to different people, this way of thinking about philosophical
inquiry supports an individualistic conception of success. It is the apparent (to me)
truths that appear (to me) to be inconsistent with my other beliefs that I have to
worry about—you may have a different set of worries, or no worries in the neigh-
borhood at all. An argument that alleviates my worries might not alleviate yours, and
vice versa.
15.3.5 Bayesianism
Bayesians hold that what is rational to conclude on the basis of new arguments or
evidence depends on one’s priors. But different people have different priors, and
according to (subjective) Bayesian orthodoxy no set of coherent priors is more
rational than another. Hence, two people exposed to the same evidence and argu-
ments may reach different conclusions. Indeed, if they are rational, they must reach

29
See, e.g., Bergmann (2005).
30
What I say here applies most straightforwardly to rebutting defeaters, but a similar (albeit more
complicated) story could be told about undermining defeaters.
31
Nozick’s embedding strategy seems like an instance of the close scrutiny strategy, but since the two
strategies can have different “feels” to them I have left the passage intact.
  . 

different conclusions, each in accordance with her priors. But then there can be no
objective fact about whether an argument makes acceptance of its conclusion
rational, and hence whether the argument is successful. For this will depend on
one’s priors, and priors are individualistic. Hence, so is argumentative success.32

15.3.6 Reflective Equilibrium

It is widely thought that philosophy aims at bringing our judgments about principles
and cases into reflective equilibrium. Since our initial judgments about cases and
principles often differ, we will often come to different points of reflective equilibrium.
As David Lewis says,
The reader in search of knock-down arguments in favor of my theories will go away disap-
pointed . . . when all is said and done, and all the tricky arguments and distinctions and
counterexamples have been discovered, presumably we will still face the question which prices
are worth paying . . . On this question we may still differ. And if all is indeed said and done,
there will be no hope of discovering still further arguments to settle our differences . . . Our
“intuitions” are simply opinions; our philosophical theories are the same. Some are common-
sensical, some are sophisticated; some are particular, some general; some are more firmly held,
some less. But they are all opinions, and a reasonable goal for a philosopher is to bring them
into equilibrium. Our common task is to find out what equilibria there are that can withstand
examination, but it remains for each of us to come to rest at one or another of them . . .
If you say flatly that there is no god, and I say that there are countless gods but none of them
are our worldmates, then it may be that neither of us is making any mistake of method. We
may each be bringing our opinions to equilibrium in the most careful possible way, taking
account of all the arguments, distinctions, and counterexamples. But one of us, at least, is
making a mistake of fact. (Lewis 1983: x)33

If the goal of philosophical inquiry is to reach reflective equilibrium, we succeed in

philosophy if we reach reflective equilibrium. Different people can come to different
points of reflective equilibrium, however, even if they are familiar with the same
arguments. But if it is possible for you and me to know the same arguments and
distinctions, and think they support different conclusions, and be making no mistake
of method, then the success must be relativized to individuals.
Some object to the method of reflective equilibrium because it seems insufficient to
produce justification. Van Inwagen, for example, agrees with Lewis that equally
competent and informed philosophers often reach different states of equilibrium.
But he takes this to be prima facie incompatible with their being justified in holding
those positions, and suggests that the fact that many of one’s peers do not share one’s
point of reflective equilibrium should knock one out of it, at least if one is aware of

32
Yes, priors can often be “swamped” by the evidence. (See, e.g., Doob (1971) and Gaifman and Snir
(1982).) But they are not always swamped, and so as long as we hold that any consistent set of priors is
rational—or even a wide range of consistent priors—exposure to the same arguments will not necessarily
lead to agreement, even between people who respond to evidence perfectly.
33
Cappelen (2012) has a nice discussion of (van Inwagen and) Lewis’s view that what we call
“intuitions” are just things we believe. This view seems to support the idea that an argument being a
success for i simply requires i to believe it to be sound. See section 15.4.1.
   

this disagreement.34 But it is not clear why we should think that knowing that one’s
peers do not share one’s point of reflective equilibrium should knock one out of it.
Reaching different states of reflective equilibrium is a species of peer disagreement,
and it is far from clear that rational peer disagreement is impossible.35
Of course, that doesn’t entail that being in reflective equilibrium is sufficient for
justification. But that is only important if successful arguments are required to
provide justification for believing their conclusions—if satisfying The Requirement
is actually necessary for success. It is to this question that we now turn.

15.4 Alternative Forms of Individualism

I claimed above that i’s knowing that A is sound (and non-fallacious) is sufficient
for A’s being a success for i. According to the Knowledge Account, it is also necessary
condition. Knowledge is paradigmatically epistemically valuable, so it is easy to see
why arguments that can (typically) be used to produce knowledge are successes. But a
number of things other than knowledge are also sometimes held to have epistemic
value: belief, true belief, justified belief, etc. So let us consider accounts of success
framed in terms of an argument’s potential to produce these.
15.4.1 The Belief Account
The weakest plausible account of success holds that an argument is successful for
someone just if they believe it to be sound. On this account, successful arguments will
often have the power to produce belief in their conclusions. There are two reasons
why we might be drawn to such an account. First, the Belief Account delivers more
successful arguments than the Knowledge Account. Second, the Belief Account
seems phenomenologically correct: when I evaluate an argument, I ask myself
whether I believe it is sound. If I do, then, all else being equal,36 I count it a success
and accept its conclusion.
There are four main problems with the Belief Account. First, it runs afoul of The
Requirement: simply believing an argument to be sound does not justify one in
accepting its conclusion. Second, merely acquiring new beliefs is not, except perhaps
incidentally, a form of epistemic gain—gaining new false and unjustified beliefs is not
epistemically valuable. Third, the Belief Account makes successful arguments
implausibly easy to come by. And fourth, it allows there to be successful arguments
for false or contradictory conclusions.
15.4.2 The True Belief Account
So let us turn to the True Belief Account, according to which an argument is
successful for i just if i correctly believes it to be sound. According to the True Belief
Account, successful arguments will (typically) be able to produce true beliefs, and
true beliefs are (plausibly) epistemically valuable. The True Belief Account also fixes
some other problems with the Belief Account: the True Belief Account does not make

34
In van Inwagen (2004). See also Kelly and McGrath (2010).
35
Partly for reasons outlined by van Inwagen himself: see, e.g., van Inwagen (1996) and Kelly (2010).
36
If the conclusion is less plausible than the conjunction of the premises, all else is not equal.
  . 

successful arguments overly easy to come by, nor would it allow successful arguments
for false or contradictory conclusions. Since there is no special phenomenology
attached to true beliefs, phenomenological considerations do not give us a reason
to prefer the Belief Account to the True Belief Account. Finally, the True Belief
Account, like the Belief Account, makes successful arguments for substantive philo-
sophical conclusions more common than the Knowledge Account. Despite all of
these advantages, however, the True Belief Account runs afoul of The Requirement.
Since The Requirement is non-negotiable, the True Belief Account cannot be correct.

15.4.3 The Justified Belief Account

The Justified Belief Account (according to which an argument is successful for i just if
i justifiably believes it to be sound) seems tailor-made to satisfy The Requirement,
and will satisfy The Desideratum and The Weak Desideratum at least as well as the
Knowledge Account. Is the Justified Belief Account then preferable to the Knowledge
Account? There are three main problems. First, if the Justified Belief Account is not
to collapse into the Knowledge Account, we must assume that justification is fallible.
But then the Justified Belief Account entails that there could be successful arguments
for contradictory or false conclusions. As noted earlier, this seems like a serious
problem. Indeed, it suggests an additional desideratum on a theory of success:
The Constraint: A theory of success should not entail that there are successful
arguments for contradictory/false conclusions.
Of course, the Justified Belief Account wouldn’t allow arguments for contradictory
conclusions to be successful for the same people (at the same time), at least if we
assume that one can’t be simultaneously justified in believing both p and not-p.37 But
this only partially ameliorates the problem: it is at least somewhat implausible that
I could have a successful argument for p while you have a successful argument for
not-p. And of course, the Justified Belief Account still allows arguments for false
conclusions to be successes.
The second problem with the Justified Belief Account is that, if there is a class of
beliefs in no need of justification—one way of understanding the notion of a
“foundational” belief—then the Justified Belief Account will misclassify arguments
based on such foundational beliefs. If my beliefs about the reliability of sense
perception, the validity of modus ponens, and the reflexivity of identity need no
justification to be knowledge, I can successfully argue to other conclusions on the
basis of such premises, contra the Justified Belief Account.38
Finally, for reasons related to the paradox of the preface, the Justified Belief
Account doesn’t seem to satisfy The Requirement: I may be justified in believing

37
This will be false if propositions are individuated coarsely—e.g., if propositions are sets of worlds.
38
Recall n 17 on how to understand locutions like ‘i justifiably believes A to be sound’. Of course, we
might say that foundational beliefs are immediately or non-inferentially justified, rather than unjustified, so
this point is not decisive. However, it is natural to describe foundational beliefs as those that need no
justification. See, e.g., Sinnott-Armstrong (1999), “sometimes one has no reason at all to believe [p],
because [p] is self-evident and needs no justification.”
   

each claim on a long list, and I may be justified in believing that conjunction
introduction is valid, but not be justified in believing the conjunction of those
claims.39 Perhaps this falls under the ceteris paribus clause of The Requirement,
but it is an unlovely result nonetheless.

15.4.4 Summing Up
In light of the above, I see no reason to reject the Knowledge Account. I should stress,
however, that the primary aim of this paper is to argue that philosophical individu-
alism is true, not to defend any particular species of the view.

15.5 Success and Progress

It is natural to think that successful arguments should lead to agreement about the
truth of their conclusions, and that such agreement is required for progress.40 So
progress, agreement, and success appear to be closely related, if not the same topic
under different guises. Philosophical individualism, however, holds that agreement is
fundamentally distinct from success: that an argument can be successful for someone
who knows it to be sound even if no one else agrees.
But while the separation of agreement and success can be jarring, it is hard to see a
reason why they must be linked. What about success and progress? Once we realize
that it is possible for someone to know premises that others reject, it is difficult to see
how the production of a sound argument for some interesting philosophical conclu-
sion using those premises could fail to be a kind of progress for that person, at least if
producing such an argument was her goal. Isn’t progress just getting closer to one’s
goal? And so, if success is individualistic, so is progress. Agreement and convergence,
on the other hand, are essentially social.

15.5.1 Essentially Social Success

Some might insist, however, that success is essentially social—that success in phil-
osophy (say) must be recognized by the philosophical community. In certain con-
texts, such as high school debate, success is essentially social. One’s goal in a high
school debate is to convince the judges of one’s conclusion—or at least to convince
them that one’s argument is better than one’s opponents. A debater’s opinion of her
own arguments is irrelevant. Likewise, one might think, with philosophical argu-
ments and the philosophical community: successful arguments are those that the
community, or some privileged subset of the community, deems successful. Call this
the Communitarian Account of success.

39
Indeed, I might be justified in believing the negation of that conjunction! Note that the True Belief
Account is not afflicted by this kind of worry. If I correctly believe an argument to be sound, then, assuming
I believe the conclusion, I correctly believe it. It is unclear that the paradox threatens the Knowledge
Account. If I know the premises of an argument, and I know it to be valid and non-fallacious, then it is very
plausible that, assuming I believe the conclusion, I know it. See Hawthorne (2004: 49) for discussion.
40
As noted by van Inwagen in his discussion of philosophical failure. See also van Inwagen (2004),
Wilson (2013), and Chalmers (2017).
  . 

Note, however, that if we are willing to attribute propositional attitudes to collec-

tions of individuals, then philosophical individualists can give a nice account of social
success: an argument A is successful for collection of individuals c just in case c knows
that A is sound.41 Call collections of individuals that can bear propositional attitudes
‘social individuals’. There is no reason that a philosophical individualist cannot count
such social individuals as individuals. But she still can and must relativize success to
proper individuals as well. If, for example, I were shipwrecked on a deserted island,
I might spend my time thinking about the Sorites paradox. While on the island, I might
solve the puzzle, coming to know that an argument with, say, epistemicism as its
conclusion was sound. If I no longer feel puzzled, and have a better understanding of
vagueness, and know that epistemicism is true, then by any reasonable standard I have
made progress on the problem of vagueness, and my argument is a success.
Now suppose that after I am rescued from the island I cannot find anyone willing
to publish my argument. Indeed, suppose that no one else finds it convincing. Would
that mean that the argument that was successful is now a failure? How could that be?
A sound argument for epistemicism, which I know to be sound, is successful for me
even if it convinces no others. While philosophy is social in many important ways, it
is not essentially social. It can be practiced on a deserted island, and it can be
practiced there successfully.42
It is worth noting that while it is possible to rob a person p of her knowledge that an
argument A is sound—if, say, p’s interlocutors refuse to take the idea that that A is sound
seriously—is it much easier to rob someone of her due influence in the philosophical
community. While this latter form of epistemic injustice is an important one, I take it to
be a virtue of the Knowledge Account that it allows marginalized philosophers—
philosophers to whom few listen—to produce successful arguments. If we thought
(perhaps implausibly) that successful philosophical research involved the production
of successful arguments, a social conception of success would entail that by refusing to
take the arguments of marginalized philosophers seriously we thereby make it the case
that the research of those philosophers is not successful. That is surely not true.43
In any case, there are paradigmatically successful arguments that are, in principle,
successful only for the person giving them. Even if the Cogito does not qualify,
consider:
1. Only John knows that he is tired.
2. John is not wealthy.
3. Therefore, there are things that no wealthy person knows.

41
Everyone, I suppose, would be willing to attribute knowledge of p to a collection c (at least in a loose
sense) if everyone in c knows that p. The more interesting question is whether a collection c can know that p
even if some members of c don’t know that p. It is often claimed, for example, that the Greeks knew that the
Earth was round, but obviously not every Greek knew that the Earth was round. Perhaps this example can
be accommodated by treating ‘the Greeks’ as a generic (see, e.g., Leslie 2007), but other examples
attributing knowledge to Congress, one’s students, etc. may not be so easily handled.
42
This is of course consistent with the claim that truth is best pursued in an intellectual community.
43
See, e.g., Fricker (2007) for a nice discussion of epistemic injustice. While this essay is pushing back a
bit against the “social turn” in epistemology—a turn for which Fricker is the standard-bearer—it is worth
noting that much or all of what Fricker says makes sense within a classical epistemological framework, and
that much of what I say here is compatible with her view.
   

This is a great argument: assuming that there are no omniscient beings, I know
that it is sound. But it is not an argument that could be used to (rationally) convince
anyone else of its conclusion, including social individuals, since no one other than
I can know its first premise.44

15.6 Success Pluralism

The arguments of the previous section aimed to show that there must be an
individualistic notion of success. But some who grant that might still insist that in
practice philosophy is almost always social, and that when we argue in a social
context our arguments have social success conditions. After all, we often want our
arguments to convince others, or at least to be well received. So some might hold that
we should be success pluralists, holding that there are (at least) two kinds of success,
corresponding to two different goals we might have: personal success, which is
individualistic, and public success, which is not.
We have already seen the problems with several versions of non-individualism,
and the Communitarian Account, introduced in the previous section, fares no better.
One problem is that it runs afoul of The Desideratum: no arguments for substantive
philosophical conclusions have convinced the philosophical community. A second
problem is that, to make the account plausible, we are almost certainly going to need
to put some constraints on who belongs to the philosophical community, and weigh
the opinions of different members differently—convincing “the community” had
better not require convincing everyone who happens to belong to the APA or teach
philosophy, else it will run afoul of The Weak Desideratum. A natural thought would
be to give extra weight to the opinions of experts, but that requires a criterion for
expertise. And in any case, there is often more disagreement amongst the experts
than there is in the general philosophical population: the opinions of logicians are
more divided about the validity of modus ponens than are the opinions of philo-
sophers as a whole (cf. section 15.2.2.1).
A third problem is that people can belong to more than one philosophical
community. If success were a matter convincing one’s philosophical community,
contradictions would arise whenever one belonged to philosophical communities
that evaluated arguments differently. Since this happens, the Communitarian
Account flirts with incoherence. If we avoid incoherence by explicitly relativizing
success to philosophical communities, two other problems arise.
First, if the evaluator of an argument A belongs to communities that disagree about
the merits of A, she will not be able to say whether A is successful. She will be able to
say that it is successful relative to one of her communities and a failure relative to
another. But she will not be able to univocally say that the argument is, or is not, a
success. This lack of a unique result is for most practical purposes equivalent to
arriving at no result at all. Second, by relativizing success to multiple communities,

44
Premise 1 is what we might call a “Fitch-style” proposition, but rather than being true and
unknowable, it is something that only I can know. (Thanks to John Hawthorne for pointing out the
connection between such propositions and philosophical individualism.) As noted in section 15.3.1, there
are a variety of other arguments that are (plausibly) essentially first-personal.
  . 

we have taken a big step towards philosophical individualism. Once we have reached
this point, we should ask whether it is preferable to relativize success to multiple
communities or to individuals. General reasons for relativizing to individuals
were given in section 15.3, but relevant here is that it is possible for one to know
that the evaluation of one’s community is wrong. This can happen in ordinary cases,
I think, but even if not it is clearly possible for there to be exceptionally good
philosophers—people who are significantly better at philosophy than everyone else
in their community. Such a philosopher might be able to formulate arguments that
her co-workers cannot fully understand or appreciate. But it wouldn’t follow that her
arguments were not successful.45
Finally, note that philosophical individualism is perfectly compatible with the view
that an argument A can be a success relative to one community and not to another—
i.e., if the members of the first community know that A is sound while the second do
not. The philosophical individualist merely insists that one’s arguments can be
successes—for oneself—even if they are not successful relative to one’s philosophical
community, or indeed any (interesting) philosophical community at all.

15.6.1 Pluralism Revisited

Even if the Communitarian Account is unsatisfactory, one still might think that there
are different legitimate notions of success, only one of which is individualistic. While
we may have reason to think that some success is individualistic, why think that this
is the only kind of success? Indeed, given our proclivity to speak as if being successful
is a property of successful arguments, as opposed to a relation they stand in, and given
our proclivity to speak as if all and only sound (non-fallacious) arguments are
successful, one might think that we have a powerful reason to think that some
objective or at least non-relativistic notion of success is the primary one.46
To answer this objection we must distinguish between descriptive theses about
how ‘success’ is used from normative theses about how it should be used. Philosoph-
ical individualism is fundamentally a normative thesis: it is about the proper stand-
ards for evaluating arguments, the standards to which arguments should be subjected.
So it cannot be refuted simply by noting that it contradicts certain entrenched
patterns of use. But if the account of success provided by philosophical individualism
does not fit with our use of ‘success’, one might wonder how it can claim to be an
account of success at all? The short answer is: just as the account of addition provided
by Peano arithmetic is the correct account of adding, despite its failing to perfectly fit
our (imperfect) use of ‘add’. The long answer would require a lengthy discussion of
semantic externalism and meaning magnetism that would be inappropriate here, but
let me gesture at the shape that this answer would take. Semantic externalism
plausibly entails that if the normative claim that ‘success’ should be used individu-
alistically is correct, it follows that philosophical individualism is also an account of
the standard of success that we have (largely unwittingly) been appealing to all along.

45
Does this undermine our reason for thinking that van Inwagen’s criterion runs afoul of the
Desideratum? To some extent, yes. But there are other reasons for preferring philosophical individualism
to van Inwagen’s account: see section 15.3.
46
I thank David Chalmers, Lorraine Juliano Keller, and Steve Petersen for pressing me on this issue.
   

Since the appeal will have been largely unwitting, this account will not fit perfectly
with use: it will be a case where use is being trumped by other meaning-determining
considerations.47 But actual use of ‘success’ is a mess, with the most common use of
‘success’ in philosophical contexts assuming that successful arguments are proofs, or
at least valid arguments with premises that are commonly known. These standards
are untenable.
This, by itself, doesn’t refute success pluralism. Even if there is a sense in which
success is individualistic, there could be others. Why not say that (non-fallacious)
sound arguments are objectively successful, that arguments that convince our oppon-
ents are dialectically successful, that arguments that would convince an idealized
audience are ideally successful, that arguments with apodictic premises are demon-
strably successful, etc.? The short answer is that this multiplication of senses of
‘success’ is more trouble than it is worth. Consider the notion of “objective success”,
according to which sound arguments are successful. Is this any kind of success at all?
Is there any sense in which the BSA is successful? It is difficult to see what that sense
could be. If someone is asked why she believes an unproven arithmetical theorem,
giving the BSA is a wholly unsatisfactory response—and satisfying the Requirement
seems like a non-negotiable condition on being a successful argument.
What about dialectical success: convincing some audience, whether one’s oppon-
ents or a neutral or idealized audience? Note first that philosophical individualism
can accommodate the idea that successful arguments convince one’s audience, since
it entails that an argument is successful for one’s audience just if one’s audience
knows that the argument is sound. The advocate of dialectical success says that this is
not enough: that we need two species of success, “audience success” and “arguer
success.”48 But doesn’t it make more sense to analyze audience success and arguer
success in terms of success simpliciter? A philosophical individualist can say that an
argument is an audience success just if it is a success for the audience, and an
argument is an arguer success just if it is a success for the arguer. This seems more
parsimonious than, and hence preferable to, thinking that there are two distinct
notions of success in play.
Of course, there are a plurality of kinds of ‘success’ in the following sense:
convincing one’s opponents, or the philosophical community more generally, are
all goals someone might have, even if it is not the goal by which we measure the
success of an argument qua argument. Convincing people other than oneself might
be a professional goal, or a personal goal, or some other kind of goal. That is all true,
but irrelevant: whether an argument achieves such non-epistemic goals is distinct
from whether it is successful as an argument. It is not impossible, or even uncom-
mon, to use bad arguments because they will be convincing to others: that would be
one way to define ‘propaganda’. But we shouldn’t dignify propaganda by requiring
successful arguments to be useful propaganda tools. Successful arguments don’t

47
See Sider (2012) for a general discussion of such cases, and Keller (2015a: section 4.2) for a discussion
of how use can be trumped.
48
Cf. Sinnott-Armstrong (1999) on “audience justification” and “arguer justification.” But aren’t
arguers part of their own audience, at least if they are arguing in earnest? Yes: this is another reason to
accept philosophical individualism, with a univocal notion of “evaluator success.”
  . 

always make for successful propaganda, and vice versa. An argument’s lack of ability
to produce converts may point to a flaw in the argument as a political tool, but it
doesn’t necessarily point to any flaw in the argument itself. The idea that convincing
others is a kind of epistemic goodness for arguments presupposes that the people one
is trying to convince will respond rationally to the argument, and are in a position to
recognize the truth of its premises. But alas, that is not always the case.
What, finally, about van Inwagian ideal success? This kind of success would avoid
the problem afflicting non-ideal species of dialectical success: an ideal audience will
presumably respond rationally to arguments, and may know any knowable premises.
But there are other problems with this view, as we saw in section 15.3. Once we have
idealized away the limitations that undermined the other dialectal conceptions of
success, we may well have idealized away the limitations that give rise to philosoph-
ical problems in the first place. An audience that could just see that Goldbach’s
Conjecture followed from the Peano axioms might find the BSA convincing. That
doesn’t mean that the BSA is good, it just means that such idealized audiences aren’t
a barometer of success.

15.6.2 Arguments and Proofs

A final argument for pluralism about success comes from the fact that successful
mathematical arguments are proofs, while successful arguments in philosophy and
other disciplines may fall short of this standard. This would seem to indicate that
there are at least two notions of argumentative success.
It is non-trivial to give a satisfactory account of mathematical proof, but we can
bypass this complication by noting that there is a glaring difference between math-
ematical proofs and successful philosophical arguments: the axioms (and theorems)
that mathematicians use in their proofs are almost universally agreed upon (within
the relevant mathematical community) and perfectly precise. Virtually every math-
ematician knows that the proofs (in her sub-discipline, at least) are sound, and knows
that with something like certainty. Does this mean that there is a different standard
for success when it comes to mathematics? No it does not. If mathematical axioms
and theorems are commonly known, there is a de facto “public” and apodictic
conception of success in mathematics (since validity is publicly recognizable). But
there is no need to treat this as a different kind of success than the success we have in
in other disciplines. Rather, we can attribute the public nature of successful math-
ematical arguments, and the certainty of their conclusions, to the public and certain
knowledge of their premises. It is perfectly consistent with philosophical individual-
ism that there are some, indeed many, publicly recognizably successful arguments,
and even publicly recognizable demonstrations. And if we can account for the
differences between math and philosophy in this way, considerations of parsimony
suggest that we should account for them this way. There is simply no need to
postulate two kinds of success when we already know that premises can be publicly
or merely privately known, with or without certainty.49

49
Similarly, one difference between mathematical and scientific arguments is that scientific arguments
contain premises that are less universally known, and which are not known by anyone with the kind of
certainty with which the mathematical community knows its premises. See Chalmers (2017) for a
discussion of related issues.
   

15.6.3 Against Success Pluralism

The fundamental argument against success pluralism, then, is that there just aren’t a
plurality of viable notions of success. Every account of success we have considered,
other than van Inwagen’s Account, the Knowledge Account, and Justification
Account, runs afoul of The Weak Desideratum, or The Requirement, or both. Van
Inwagen’s account is incompatible with the existence of essentially first-person
arguments, evidence non-neutrality, and subjective Bayesianism. Furthermore, com-
plications arising from the ideality and neutrality of his audience threaten its
satisfaction of The Weak Desideratum and The Requirement. This still leaves
room for a weak form of pluralism that countenances just two kinds of success, as
outlined by the Knowledge and Justification Accounts. But these accounts are so
close to one another that it would be absurd to accept them both—they clearly have
the same target, and both are forms of individualism. In any case, The Constraint
gives us a principled reason to prefer the Knowledge Account to the Justification
Account. There is no reason, then, to be a success pluralist.

15.7 Conclusion
According to philosophical individualism, the success of an argument is relativized to
individuals. Seeing that this is the case is an important improvement upon other ways
of thinking about success, regardless of the species of philosophical individualism we
accept. I have defended the version of individualism that requires someone to know
an argument to be sound in order for it to be a success for her, but those with
different philosophical temperaments might be inclined to accept another.
It goes without saying that philosophical individualism has radical implications for
how we should evaluate the success of philosophical arguments. Van Inwagen
suggests that successful arguments would convince a substantial portion of the
philosophical community, and we often teach our students that arguments should
have uncontroversial premises, and in particular premises that opponents of the
argument’s conclusion can accept. If philosophical individualism is true, none of this
is correct. If we accept the Knowledge Account, we should rather ask whether we
know the argument to be sound and, perhaps, whether the person giving it does.
Nothing else matters.
Of course, it follows from philosophical individualism that success is not ideally
publicly recognizable. But this is not as bad as it seems: it does not, for example, entail
that success is subjective in any pejorative sense. For the most plausible species of
philosophical individualism hold that successful arguments must be sound. And the
soundness of our arguments is something that is (typically) independent of what we
think or feel. Still, philosophical individualism is not as good as we might have hoped.
For if success is individualistic, we cannot always use reason to peaceably settle public
disputes.50 This is disappointing, but it is unavoidable: it is the human condition. Of

50
At least in the most straightforward sense. Reason may dictate that, if success is individualistic, we
should accept a democratic conception of success for arguments in the public square. This would not,
however, be using reason to settle public disputes. Rather, it would be using reason to find the best non-
rational method for settling public disputes.
  . 

course, we should continue to use reason to settle as much as possible. Uncertainty,

too, is part of the human condition, but we should try to be as certain as possible
about important public policies. Likewise, even if we cannot hope to rationally
convince everyone to accept some policy, we should try to rationally convince
everyone we can. We just need to remember that this is a political goal, rather than
an epistemic requirement.

Acknowledgments
Thanks to Nathan Ballantyne, David Chalmers, John Hawthorne, Lorraine Juliano Keller,
Stephen Kershnar, Steve Petersen, Joshua Smart, and Peter van Inwagen for helpful comments
and discussion.

References
Balaguer, M. (2010), Free Will as an Open Scientific Problem (MIT Press).
Ballantyne, N. (2014), ‘Knockdown Arguments’, in Erkenntnis 79.
Bergmann, M. (2005), ‘Defeaters and Higher-Level Requirements’, in The Philosophical Quar-
terly, 55(220).
Bourget, D., and Chalmers, D. (2014), ‘What Do Philosophers Believe?’, in Philosophical
Studies 170(3): 465–500.
Cappelen, H. (2012), Philosophy without Intuitions (Oxford University Press).
Chalmers, D. (1996), The Conscious Mind (Oxford University Press).
Chalmers, D. (forthcoming), ‘On the Limits of Progress in Philosophy’, in J. A. Keller (ed.),
Being, Freedom, and Method: Themes from the Philosophy of Peter van Inwagen (Oxford
University Press).
Doob, J. L. (1971), ‘What is a Martingale?’, in American Mathematical Monthly 78: 451–62.
Fischer, J. M., and Tognazzini, N. A. (2007), ‘Exploring Evil and Philosophical Failure:
A Critical Notice of Peter van Inwagen’s The Problem of Evil’, in Faith and Philosophy
24(4): 458–74.
Fricker, M. (2007), Epistemic Injustice: Power and the Ethics of Knowing (Oxford University
Press).
Gaifman, H., and Snir, M. (1982), ‘Probabilities Over Rich Languages’, in Journal of Symbolic
Logic, 47: 495–548.
Hawthorne, J. (2004), Knowledge and Lotteries (Oxford University Press).
Keller, J. A. (2015a), ‘Paraphrase, Semantics, and Ontology’, in K. Bennett and D. Zimmerman
(eds.), Oxford Studies in Metaphysics, Vol. 9 (Oxford University Press).
Keller, J. A. (2015b), ‘On Knockdown Arguments’, in Erkenntnis 80.
Kelly, T. (2005), ‘Moorean Facts and Belief Revision, or Can the Skeptic Win?’, in Philosophical
Perspectives: Epistemology 19.
Kelly, T. (2010), ‘Peer Disagreement and Higher Order Evidence’, in R. Feldman and
T. Warfield (eds.), Disagreement (Oxford University Press).
Kelly, T., and McGrath, S. (2010), ‘Is Reflective Equilibrium Enough?’, in Philosophical
Perspectives 24(1): 325–59.
Kelly, T., and McGrath, S. (forthcoming), ‘Are There any Successful Philosophical Argu-
ments?’, in J. A. Keller (ed.), Being, Freedom, and Method: Themes from the Philosophy of
Peter van Inwagen (Oxford University Press).
Leslie, S.-J. (2007), ‘Generics and the Structure of the Mind’, in Philosophical Perspectives:
Philosophy of Mind 21: 375–403.
   

Lewis, D. (1983), Philosophical Papers, Vol. 1 (Oxford University Press).

Nozick, R. (1981), Philosophical Explanations (Belknap Press).
Pryor, J. (2000), ‘The Skeptic and the Dogmatist’, in Noûs 34: 517–49.
Pryor, J. (2004), ‘What’s Wrong with Moore’s Argument?’, in Philosophical Issues 14(1):
349–78.
Rawls, J. (1993), Political Liberalism (Columbia University Press).
Rescher, N. (1985), The Strife of Systems: An Essay on the Grounds and Implications of
Philosophical Diversity (University of Pittsburgh Press).
Sider, T. (2012), Writing the Book of the World (Oxford University Press).
Sinnott-Armstrong, W. (1999), ‘Begging the Question’, in Australasian Journal of Philosophy
77(2): 174–91.
Van Inwagen, P. (1983), An Essay on Free Will (Oxford University Press).
Van Inwagen, P. (1996), ‘It is Wrong, Everywhere, Always, and for Anyone, to Believe
Anything upon Insufficient Evidence’, in E. Stump and M. J. Murray (eds.), Philosophy of
Religion: The Big Questions (Blackwell).
Van Inwagen, P. (2004), ‘Freedom to Break the Laws’, in Midwest Studies in Philosophy,
XXVIII.
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
Van Inwagen, P. (2017), ‘Concluding Meditation’, in J. A. Keller (ed.), Being, Freedom, and
Method: Themes from van Inwagen (Oxford University Press).
Williamson, T. (2000), Knowledge and Its Limits (Oxford University Press).
Williamson, T. (2007), The Philosophy of Philosophy (Blackwell Publishers).
Wilson, J. (2013), ‘Three Dogmas of Metaphysical Methodology’, in M. Haug (ed.), Philosoph-
ical Methodology: The Armchair or the Laboratory? (Routledge).
 16
Are There any Successful
Philosophical Arguments?

Thomas Kelly and Sarah McGrath

16.1 Introduction: A Pessimistic Verdict

There is no shortage of bad arguments in philosophy. On this point, even philosophers
who agree about little else will find common ground. Indeed, given the traditional
conception of philosophy as an argument-driven discipline, the frequency with which
philosophers have produced and consumed bad arguments can seem like a depressing
fact about the subject. Still, one might wonder: just how bleak is the situation?
If Peter van Inwagen (2006: 52) is correct, the situation is bleak indeed:
There are certainly successful arguments, both in everyday life and in the sciences. But are
there any successful philosophical arguments? I know of none. (That is, I know of none for any
substantive philosophical thesis.)
Although couched as a disavowal of knowledge, it is clear from the context that what
van Inwagen says here is not intended as a confession of idiosyncratic ignorance on
his part. Rather, it is offered as an expression of skepticism about whether there are
any arguments of the relevant kind. Let’s call this
The Pessimistic Verdict: There are no successful philosophical arguments for
substantive philosophical theses.1
In this paper, we offer some critical reflections on van Inwagen’s case for The
Pessimistic Verdict and his stimulating discussion of related issues, as presented in
the third of his 2003 Gifford Lectures, “Philosophical Failure.”2 Central to van
Inwagen’s case for The Pessimistic Verdict is his development and defense of a
“criterion of philosophical success”: that is, an account of the conditions that a
philosophical argument must satisfy in order to qualify as a successful argument.

1
It is clear enough that van Inwagen believes that The Pessimistic Verdict is true. For example, at (2006:
53) he reports his belief that “no philosophical argument for a substantive conclusion has the power to
convert every member of an ideal and initially neutral audience to its conclusion.” And as we will see in
section 16.3, van Inwagen holds that an argument is a success if and only if it satisfies this condition.
2
Ultimately published as The Problem of Evil (Oxford University Press, 2006). In what follows, any page
references refer to this published text, unless otherwise noted.
      ? 

With this account in hand, he proceeds to argue that, plausibly, no philosophical

argument for a substantive conclusion meets the relevant standard. We will proceed
as follows. In section 16.2, we raise some preliminary questions about The Pessimistic
Verdict, focusing in particular on how the crucial notion of a substantive philosoph-
ical thesis should be understood. In section 16.3, we explore van Inwagen’s criterion
of philosophical success and develop an objection to that account. In section 16.4, we
argue that even if we are wrong and van Inwagen’s account of philosophical success is
correct, his case for The Pessimistic Verdict is not compelling. We close with some
further remarks in section 16.5.
A certain skepticism about the potency of philosophical argument is a recurring
theme in van Inwagen’s writings (see, e.g., 2004: 334–40; 1990: 115). We have chosen to
focus on “Philosophical Failure” because it represents his most sustained discussion of
the issue. Because it will play some role in what follows, a few remarks about the
context of that discussion are in order. The overarching project of van Inwagen’s
Gifford Lectures is to show that “the argument from evil”—roughly, the argument that
purports to establish the non-existence of God on the basis of the existence of evil—is a
failure. In this context, “Philosophical Failure” stands as a kind of metaphilosophical
interlude in a work otherwise devoted to the philosophy of religion. As such, it plays
two distinct roles in the larger argument of the lectures. First, it serves to clarify their
main thesis. Thus, van Inwagen opens the lecture with the following words:
I have said that my project in these lectures is to defend the conclusion that the argument from
evil is a failure. My purpose in the present lecture is to explain what I mean by calling this
argument, or any philosophical argument, a failure. (37)

But van Inwagen’s metaphilosophical discussion is also intended to play another role
in the larger argument of the lectures. Specifically, and although van Inwagen
willingly accepts the burden of showing where exactly the argument from evil fails,
he believes that once we are clear about what it would be for a philosophical
argument to succeed, we will have good reason to regard it as highly improbable or
unlikely that this particular argument is successful. Thus, immediately after present-
ing his argument for The Pessimistic Verdict (and while temporarily bracketing the
argument from evil from its scope), van Inwagen offers the following:
Is it plausible to hold that philosophy can provide a successful argument for the non-existence
of God, even though philosophy is unable to provide a successful argument for any other
substantive thesis? I have to say that this seems implausible to me. It seems antecedently highly
improbable that philosophy, in whose house there have been debated scores (at least) of
important questions, should be able to provide a decisive answer to exactly one of them. (54)
And he concludes the lecture with the following remark:
My hope is that my reflections on the topic of philosophical argument will lead you to the
conclusion that it would be a very odd thing if the argument from evil were a success. (55)
While acknowledging the relevance of van Inwagen’s metaphilosophical reflections
to the special case of arguments concerning the existence of God, we believe that they
are of quite general philosophical interest. In any case, that is the spirit in which we
will engage with them.
     

16.2 On Counterexamples and Philosophical Substance

Are there any successful philosophical arguments for substantive theses? Let’s begin
with a simpler and more straightforward question, albeit one that is still not perfectly
transparent: Are there any successful philosophical arguments at all?
Here are three quick candidates, philosophical arguments that at least some
philosophers have thought successfully established their conclusions. The first
belongs to metaphysics; the second to epistemology; the third to ethics.
(i) In his seminal treatise Material Beings (1990), van Inwagen offers the following
argument against compositional nihilism, the view that everything that exists is
simple.
(1) I exist.
(2) If I exist, I have proper parts.
(3) Therefore, compositional nihilism is false.3
(ii) Kelly and McGrath (2010: 332) endorse the following argument:
(1) We know that not everyone who is alive today will still be alive fifty years from
now.
(2) Therefore, we have some knowledge of the future.
(3) Therefore, inductive skepticism (the view that we know nothing about the
future) is false.
(iii) Consider a standard version of act consequentialism, according to which one is
under a standing moral obligation to act so that fewer rather than more people die, in
the absence of any further morally significant consequences. Against this theory,
someone might offer the following argument:
(1) Even if one knows with certainty that (i) one can save the lives of two
hospitalized patients dying from organ failure by forcibly overpowering and
harvesting the organs of an innocent bystander and that (ii) there is no other
way of saving the two patients, one is not morally required to sacrifice the
innocent bystander.
(2) If act consequentialism is true, one is morally required to sacrifice the
bystander.
(3) Therefore, act consequentialism is false.
Each of these arguments would be regarded as a compelling argument for its
conclusion by at least some philosophers; of course, in each case, there are other
philosophers who would deny that it is. (We assume that the mere fact that an
argument’s status is controversial among philosophers is not enough to settle the
question in favor of its detractors.) It is not our purpose to defend any of these

3
We should note that even in the work in which this argument is put forward, van Inwagen’s own
opinion of it seems relatively low. There, he includes it in a group of arguments of which he says the
following: “These arguments are perhaps rather weak, but I do not think that they are entirely worthless”
(1990: 115). Nevertheless, it is still true that there are philosophers who would regard the argument as
compelling, even if van Inwagen is not among their number. See, e.g., Schaffer (2009: 358).
      ? 

arguments in particular; rather, we want to call attention to what they have in

common. Each argument purports to refute a substantive philosophical thesis (com-
positional nihilism, inductive skepticism, act consequentialism) in virtue of provid-
ing a counterexample to that thesis. More generally, any putative counterexample to
a philosophical thesis can be presented as an argument against that thesis. The claim
that there are no successful philosophical arguments thus amounts to the claim that
there are no successful counterexamples to philosophical theses. Put the other way
around, anyone who thinks that philosophers at least occasionally devise successful
counterexamples should think that philosophers at least occasionally produce suc-
cessful arguments. (Although of course, the status of specific cases might very well be
controversial.) Since we believe that there are some successful counterexamples in
philosophy, we believe that there are some successful philosophical arguments.
Of course, The Pessimistic Verdict is not that there are no successful philosophical
arguments; rather, it is that there are no successful philosophical arguments for
substantive philosophical theses. And indeed, the inclusion of this important qualifi-
cation seems intended to allow for the successful refutation of philosophical theses,
either via the provision of a counterexample or via some other means. Here is van
Inwagen’s most extended discussion of the qualification:
I say “substantive philosophical thesis” because I concede that there are, so to call them,
minor philosophical theses—such as the thesis that, whatever knowledge may be, it is not
simply justified true belief—for which there are arguments that should convert any rational
person. I call this thesis minor not because I think that the problem of the analysis of
knowledge is unimportant, but precisely because the thesis does not constitute an analysis of
knowledge; its message is only that a certain proposed analysis is a failure. Or suppose, as
many have supposed, that Gödel’s incompleteness results show, establish that the formalists
were wrong about the nature of mathematics. The thesis that formalism is false may in one
way be an important philosophical thesis, but only because a lot of people had thought
that formalism was true. It is not a substantive philosophical thesis in the way formalism
itself is. (39)

In a related context elsewhere, van Inwagen writes of “negative” philosophical theses,

with reference to the same examples: “If there is any philosophical thesis that all or
most philosophers affirm, it is a negative thesis: that formalism is not the right
philosophy of mathematics, for example, or that knowledge is not (simply) justified,
true belief ” (2004: 334–5).
The idea then, seems to be this: even though Gettier provided successful counter-
examples to the justified true belief account of knowledge, and any such counter-
example can of course be presented in the form of a successful philosophical
argument, the conclusion of that argument is that knowledge is not (simply) justified
true belief. And that conclusion is a negative (or “minor”) thesis as opposed to a
substantive philosophical thesis. In this respect, it contrasts with the thesis that
knowledge is simply justified true belief, which is a philosophically substantive thesis
(albeit one that is generally taken to be false in the post-Gettier era). Thus, the fact
that there are successful counterexamples in philosophy does not pose a threat to The
Pessimistic Verdict, so long as the arguments corresponding to the counterexamples
establish only negative or minor philosophical theses.
     

However, we think that the way in which the ideology of philosophical substantiality
is being employed here leads to difficulties. Consider, for example, van Inwagen’s
earlier argument against compositional nihilism given above. We assume that com-
positional nihilism is an example of a substantive philosophical thesis in ontology; in
this respect, it is like formalism in the philosophy of mathematics or the thesis that
knowledge is justified true belief in epistemology. There is some initial temptation then
to say that an argument like van Inwagen’s, which purports to refute compositional
nihilism by means of a counterexample, is not itself an argument for a substantive
philosophical thesis: even if it is successful, it would establish only the negative or
minor philosophical thesis that compositional nihilism is false, as opposed to some
philosophically substantial conclusion. When construed in this way, such arguments
pose no threat to The Pessimistic Verdict, for they fall outside of its scope.
The problem, however, is this. There is an obvious sense in which the thesis of
compositional nihilism is itself a “negative” thesis: it is the denial of the claim that the
universe contains objects with proper parts. An argument that successfully refuted
compositional nihilism, either by means of a counterexample or in some other way,
would ipso facto be an argument that successfully established a positive thesis to the
effect that the universe contains objects of a certain ontological type, viz. objects that
are composed of (proper) parts. And why wouldn’t that ontological thesis count as a
substantive philosophical thesis? Indeed, that the universe contains objects of such
and such an ontological type would seem to be a paradigm of a substantive meta-
physical thesis. Given this, if van Inwagen’s argument against compositional nihilism
is a good one, it is not merely an example of a successful philosophical argument; it is
an example of a successful philosophical argument for a substantive philosophical
conclusion—and therefore, a counterexample to The Pessimistic Verdict.
A similar situation obtains with respect to the thesis of inductive skepticism in
epistemology. On the one hand, inductive skepticism is clearly a substantive philo-
sophical thesis. On the other hand, inductive skepticism is also a negative philosoph-
ical thesis: it is the denial of the claim that we possess a certain kind of knowledge.
Any successful argument against this negative thesis, via counterexample or some
other means, would ipso facto be a successful existence proof that we do have a
certain kind of knowledge, knowledge of the future. But the claim that we possess a
certain kind of knowledge is itself a substantive claim of epistemology. Thus, any
such argument would itself be a successful argument for a substantive philosophical
conclusion, and thus, a counterexample to The Pessimistic Verdict.4
One lesson that has emerged so far is the following. Although in some cases, the
negation of a substantive philosophical thesis is not a substantive philosophical
thesis, in other cases, the negation of a substantive philosophical thesis is itself a
substantive philosophical thesis. (Perhaps act consequentialism and the thesis that

4
Is it a relevant difference that while some competent philosophers would deny that the arguments
against compositional nihilism and inductive skepticism mentioned above succeed in refuting those views
via counterexample, no competent philosophers deny that Gettier succeeded in refuting the justified true
belief account of knowledge via counterexample? However, it is not true that all competent philosophers
agree that Gettier succeeded in refuting the justified true belief account via counterexample: dissenters
include Hetherington (2001) and Weatherson (2003).
      ? 

knowledge is justified true belief are examples of substantive philosophical theses of

the former kind, while compositional nihilism and inductive skepticism are examples
of the latter.) Thus, one way in which a philosopher could offer a successful argument
for a substantive conclusion (and thereby refute The Pessimistic Verdict) is by
offering a counterexample to a philosophical thesis that is itself the negation of a
substantive philosophical thesis. It is widely thought that, whatever their deficiencies
as a group, analytic philosophers are quite good at coming up with counterexamples
to each other’s claims. (Indeed, this point is often conceded even by detractors of
analytic philosophy, in the course of lamenting its practitioners’ lack of allegedly
more valuable intellectual skills.) Is every case in which this has occurred one in
which the negation of the refuted thesis is “minor” in van Inwagen’s sense? Perhaps,
but we see no reason to think this.
Indeed, when one recalls van Inwagen’s larger dialectical purposes, it seems that
there will be significant pressure on him to be extremely inclusive about what counts
as a substantive philosophical thesis. Recall from above that one of van Inwagen’s
purposes in arguing that philosophical failure is ubiquitous is to increase the plausi-
bility of the claim that the “argument from evil” is itself a failure. Consider, for
example, the passage that immediately follows van Inwagen’s argument for The
Pessimistic Verdict:
Now if it is indeed true that no philosophical argument for any substantive conclusion is
successful in the sense that I have proposed, it immediately follows that the argument from evil
is not a success in that sense—given, at any rate, two premises that I don’t think anyone would
deny: that the argument from evil is a philosophical argument and that the non-existence of
God is a substantive philosophical thesis. (53)
Here, van Inwagen assumes that everyone would agree that the non-existence of God is
a substantive philosophical thesis. In fact, however, we are inclined to deny this claim,
given his earlier remarks about what it is for something to count as a substantive
philosophical thesis in his sense. After all, the mere denial that God exists does not
seem to have much in common with formalism (understood as a positive account of
the nature of mathematics) or the justified true belief account of knowledge. One can,
after all, deny that God exists without having much in the way of a positive account
about the origins or nature of the universe. In fact, one who argues for the non-
existence of God on the basis of evil and then declines to say anything more seems in
some respects rather like someone who refutes a proposed analysis of knowledge and
then stops before offering an alternative account in its place. (“Look, I might not know
what the true story is here, but I know that it’s not that . . . ”)
Of course, there is an obvious sense in which the non-existence of God is a
substantive thesis. Notably, the opposite view has been held (in some times and
places virtually without exception) by many millions of people, including many
individuals of the highest intellectual caliber. So the thesis that God does not exist
is hardly a trivial one. But as we have seen, van Inwagen explicitly denies that the
popularity of a philosophical thesis (or even its popularity among a select group of
people) suffices to make the negation of that thesis a philosophically substantive
claim. (As we read him, even if all of the most sophisticated epistemologists had been
fervent advocates of the view that knowledge is justified true belief prior to Gettier,
     

Gettier’s thesis that knowledge is not justified true belief would still be a “minor” or
“negative” thesis as opposed to a substantive one.)
The dilemma that we see for van Inwagen then, is this. Unless he employs some
relatively inclusive sense of “substantive,” the negative thesis that God does not exist will
fail to count as substantive; rather, like the thesis that knowledge is not justified true
belief or that formalism is false, it will be classified as a “minor” philosophical thesis
(notwithstanding the considerable violence that this does to the ear). In that case, even
if The Pessimistic Verdict turns out to be true, this would have no bearing on the
plausibility that the argument from evil is a success; for the truth of The Pessimistic
Verdict is consistent with philosophers having produced arguments that successfully
establish any number of “minor” philosophical theses (the reference class to which the
thesis that God does not exist belongs). On the other hand, if van Inwagen employs a
relatively inclusive or liberal sense of “substantive” in order to allow the thesis that God
does not exist to qualify (despite its seemingly “negative” content), the harder it is to insist
that people like Gettier and Gödel failed to show anything of philosophical substance—
an admission which would, of course, immediately refute The Pessimistic Verdict.

16.3 Van Inwagen’s Criterion of Philosophical Success

The official statement of van Inwagen’s “criterion of philosophical success” runs as
follows:
An argument for p is a success just in case it can be used, under ideal circumstances, to convert
an audience of ideal agnostics (agnostics with respect to p) to belief in p—in the presence of an
ideal opponent of p. (47)
The criterion invokes idealized notions in three distinct places: in its appeal to ideal
circumstances, ideal agnostics, and an ideal opponent of p. Following the example of
van Inwagen’s own discussion, we will focus primarily on the role played by ideal
agnostics.
As characterized by van Inwagen, an ideal agnostic with respect to p has no initial
opinion about whether p is true, and moreover, “no predilection, emotional or
otherwise” to accept either p or not-p (44). However, the ideal agnostic “would very
much like to come to some sort of reasoned opinion—in fact, to achieve knowledge on
the matter if possible” (44). Thus, ideal agnostics are disinterested but not uninterested
with respect to the target proposition. In addition to this orientation, ideal agnostics are
“ideally rational” individuals (53), who possess “unlimited leisure and superhuman
patience” as well as “the highest possible intelligence and the highest degree of logical
and philosophical acumen” (44). On the other hand, an ideal opponent of p, while
sharing the same general intellectual virtues having to do with rationality, intelligence,
and so on, differs in his or her orientation towards the target proposition p: the ideal
opponent “will employ every rational means possible, at every stage of the debate, to
block the . . . attempt at conversion” (45).5

5
Although van Inwagen says relatively little about what is packed into “ideal circumstances,” the
general kind of thing he has in mind seems clear enough from his scattered remarks about the subject. For
example, we should not imagine that the attempt at conversion might be short-circuited by running out of
      ? 

On one natural reading, van Inwagen’s criterion demands a certain kind of

unanimity: a successful philosophical argument would be potent enough to convert
all of the members of an audience composed of ideal agnostics. One might worry that
this sets the bar too high. For example, imagine an argument that converts virtually
all of the ideal agnostics to its conclusion (perhaps there is a single, eccentric
holdout): wouldn’t it be too harsh to declare such an argument a failure? However,
van Inwagen explicitly assumes that the ideal agnostics will respond to any argument
in the same way (51–2). Given this assumption, an argument that is strong enough to
convert any ideal agnostic will be strong enough to convert them all.6
While further questions of detail could be raised about van Inwagen’s criterion, we
wish to develop a quite general objection to the project of characterizing successful
arguments in terms of their ability to convert idealized agnostics to their conclusions.
Our contention will be that this general framework breaks down when applied to
certain core cases, cases that any account of argumentative success should be able to
cleanly handle.
Consider first what would seem to be an undeniable case in which a philosopher
offers a successful argument. Imagine a philosopher who takes as premises a number
of distinct propositions, each one of which is an item of common knowledge, or at
least, something that is generally known to be true. From this basis of generally
known propositions, she proceeds to prove, by a series of transparently valid steps,
some philosophically substantive conclusion—in whatever sense of “philosophically
substantive” one prefers. (We can also imagine that each of the premises plays an
essential role in the argument: if any one of them were removed, the argument would
no longer be valid.) We aver that if a philosopher actually succeeded in doing this—
that is, in producing a transparently valid argument for a philosophically substantive
conclusion from generally known premises—then that argument would be an abso-
lute paradigm of a successful philosophical argument.7 Thus, it is a condition of
adequacy on any proposed criterion of philosophical success that it return the verdict

time or chalk, or that it is taking place in an uncomfortable or loud environment (cf. p. 43). In any case,
nothing in what follows will depend crucially on how exactly this notion is understood.
6
Compare the issue that arises for “ideal observer” theories in ethics, about whether anyone who
satisfied the requirements for being an ideal observer would converge on the same moral views (Kagan
1998: 274–5). For a recent exchange on the question of whether there is a single, ideally rational response to
any body of evidence and arguments, see Kelly (2013) and White (2013).
7
To be clear, we do not claim that any actual philosophical argument instantiates this general structure,
only that any argument that did would be a success. Indeed, we think that an argument with these
characteristics would be something like the holy grail when it comes to philosophical success, at least on
one historically prominent model: philosophers have clearly aspired to give valid arguments for philo-
sophically substantive conclusions, and moreover, in the course of doing so they have often deliberately
eschewed arguments employing esoteric knowledge as premises in favor of arguments employing premises
ostensibly knowable to anyone. (Consider in this context the traditional aspiration to provide valid
“proofs” of the existence of God employing only “self-evident” premises, as well as Bertrand Russell’s
oft-quoted remark that “The point of philosophy is to begin with something so obvious as not to seem
worth stating, and to end with something so paradoxical that no one will believe it” (Russell 1986).) We
ourselves do not think that it is a necessary condition on a successful philosophical argument that that
argument employ only premises that are common knowledge or generally known, only that it would
certainly be acceptable to start with such premises.
     

that an argument with these characteristics would be a successful philosophical

argument.
However, consider the problem that this kind of case poses for an account which
seeks to analyze philosophical success in terms of the ability to convert an audience
composed of idealized agnostics. Because the conclusion of the hypothetical argu-
ment follows validly from propositions that are generally known, the conclusion
follows validly from propositions that are known to members of the audience. Recall
next that an essential part of what it is to be an ideal agnostic is to be ideally rational.
Now, on standard accounts of ideal rationality, ideally rational agents possess a kind
of logical omniscience, and as we have seen, van Inwagen explicitly builds into his
characterization of ideal agnostics that they possess “the highest degree of logical
acumen.” Given this, it seems that ideal agnostics would not fail to believe proposi-
tions that are logically entailed by things that they know. Thus, given that the
logically valid argument proceeds from generally known premises, an ideal agnostic
would already believe the conclusion of the argument before the argument was ever
presented to her. But this of course contradicts the assumption that the ideal agnostics
are actually agnostics, or initially neutral with regard to the truth of the target
proposition p. In short, the problem is this: when one attempts to apply the criterion
to the kind of paradigm case that we have described, the apparatus of “ideal
agnostics” seems to break down, because there is no coherent way of making the
members of the target audience both agnostic and ideal. Any member of the audience
who was initially agnostic with respect to p would not be ideal, for she would have
failed to believe that which is logically entailed by what she knows to be true. Thus, at
least on first inspection, the attempt to explicate the notion of argumentative success
in terms of the ability to convert idealized agnostics seems to break down when
applied to a core case.
One might try to get around the objection by refining the proposal in various ways.
For example, even though the propositions employed as premises in the hypothetical
argument are stipulated to be generally known, one might take the relevant test to be
how idealized individuals would react to the argument in the absence of any prior
knowledge of its premises. Or perhaps, one requires that the ideal agnostics bracket
their knowledge of the premises, and thus provisionally suspend judgment about
whether p. (Compare Descartes’ professed procedure in the Meditations.) One then
asks: would the subsequent presentation of the argument be enough to convince such
individuals that p is true? If it would, then the argument counts as a success; if it
would not, then the argument counts as a failure. However, modified in that way the
criterion is clearly unsound: once knowledge of the propositions employed as
premises is removed or bracketed, the presentation of the argument might very
well fail to persuade otherwise ideal individuals of the conclusion, but that hardly
shows (or even provides reasons to think) that the argument is a failure.
Alternatively, one might try to preserve both the agnosticism and the cognitive
ideality of the audience to whom the argument is presented by modifying what is
packed into their cognitive ideality. Perhaps we should not assume that the ideal
agnostics are logically omniscient, or that they believe all of the logical consequences
of what they know. For example, Harman (1986) denies that a fully rational person
would believe all of the logical consequences of what she knows, on the grounds that
      ? 

a fully rational person would not clutter her mind by believing things that are
logically entailed by what she knows but which are of no interest or importance to
her. Notice, however, that qualifying ideality along these lines is unhelpful to van
Inwagen’s account. For as noted above, van Inwagen builds into his characterization
of the ideal agnostics that they would “very much like to come to some sort of
reasoned opinion—in fact, to achieve knowledge if possible” with respect to the target
proposition. Thus, a case in which a member of the audience does not believe the
conclusion of the argument prior to being presented with it would be a case in which
she has failed to believe a proposition that answers a question about which she is
concerned to know the truth, despite the fact that the proposition is logically entailed
by things that she knows. For this reason, the kind of pragmatic restrictions on full
rationality suggested by Harman are not helpful to van Inwagen’s account.
More promisingly, one might insist that the kind of full rationality possessed by
the ideal agnostics is consistent with their failing to believe propositions that are
logically entailed by what they know, even when the propositions concern matters
of interest. For example, perhaps full rationality does not require one to believe
logical consequences of what one knows that are extremely difficult to recognize
(cf. Cherniak 1986, McGrath 2010: 66). A given derivation might require one to
bring together a relatively large number of seemingly disparate and unrelated
pieces of prior knowledge; on this view of rationality, the mere fact that one has
failed to believe the conclusion of such an argument before being presented with
the argument is consistent with one’s being fully rational. A philosophical argu-
ment might thus qualify as successful by converting the ideal agnostics in virtue of
making manifest previously unrecognized connections. However, this maneuver
seems unpromising as well. For imagine that someone devises a philosophical
argument that provides a relatively short or simple proof of a substantive conclu-
sion, one that employs a relatively small number of generally known premises.
Surely that argument would qualify as a success. (Indeed, in many ways it would
be a more impressive intellectual product—compare the preference among math-
ematicians for relatively short, elegant proofs where such proofs are available.) If
we weaken the kind of rationality possessed by the ideal agnostics enough so that
it is consistent with their being initially ignorant of the conclusion even in this
kind of case, why suppose that, in some other case in which they are presented
with an argument that fails to convert them to its conclusion, this reflects badly on
the argument, as opposed to being a manifestation of their own cognitive
limitations?
No doubt other maneuvers might be suggested. Indeed, despite what we have said
thus far, we do not doubt that, given sufficient fiddling with the notion of cognitive
ideality, one could come up with some account that allowed the target audience to be
both agnostic and cognitively ideal in that sense, and which is not subject to the kind
of obvious problems that afflict the proposals that we have considered here. Still, how
much confidence one should invest in the resulting proposal in that event is open to
doubt. After all, the original idea of analyzing philosophically successful arguments as
those that have the power to convert ideally rational audiences of agnostics is an
intuitively attractive one, and an idea that would be illuminating if true. Once one
insists on characterizing the audience in terms of some non-standard account of ideal
     

rationality or cognitive ideality specially introduced for present purposes, it is

doubtful that those attractions are retained.
Moreover, we think that the problem that we have raised for the proposal reflects
something significant about the practice of attempting to construct compelling
philosophical arguments, and why success in the endeavor is likely to resist analysis
in terms of the conversion of idealized intellects. In paradigmatic cases in which
someone becomes rationally persuaded of a conclusion in virtue of being presented
with a philosophical argument, that process typically involves calling their attention
to connections between propositions (for example, that the relation of entailment
obtains between premises and conclusion) of which they were previously unaware, as
opposed to supplying them with justification or evidence for believing the argument’s
premises. Of course, we can certainly imagine exceptional cases in which the epi-
stemic efficacy of the presentation of the argument consists in providing justification
or warrant for believing its premises, justification that the members of the audience
previously lacked. For example, imagine a context in which the prevailing norms
dictate that someone will present an argument only if she takes it be sound (so there
is no question of someone’s playing “devil’s advocate,” etc.). Even if a member of the
audience was not previously in a position to justifiably believe some premise of the
argument, she might know that the person presenting the argument is in a good
position to know about such things. In these circumstances, the fact that the person
presenting the argument has endorsed it as sound might provide a kind of testimo-
nial justification or warrant for believing the premises; in this way, the audience
member might come to justifiably believe the argument’s conclusion. But this is
clearly not the usual case, nor the typical aspiration of philosophical argumentation.
Rather, in devising a philosophical argument, one selects one’s premises with an eye
towards what one’s audience already believes, or is disposed to believe, or already has
good reason to believe, and then attempts to show how these propositions support
the target conclusion. (Of course, one might at some point realize that some
proposition among one’s original premises stands in need of bolstering and attempt
to supply reasons for believing it; but at this point, one has strictly speaking begun
treating the relevant proposition not as a premise but rather as a lemma in a more
complex version of the argument.) In paradigmatic cases in which the presentation of
a philosophical argument rationally persuades the audience, the epistemic efficacy of
the presentation consists in making manifest the logical or rational connection
between premises and conclusion. But that philosophical arguments sometimes
play this role presupposes that those to whom they are addressed are less than
ideal intellects, for ideal intellects would in effect regard the information conveyed
by the presentation of the argument as old news.
On the other hand, standard philosophical practice also presupposes that we
resemble ideal intellects in at least this respect: that our beliefs are generally respon-
sive to reasons, at least when the reasons are called to attention. Just as the practice of
offering philosophical arguments would have little if any point among the ideally
rational, so too it would have little if any point among the pathologically irrational,
who respond with shrugs when it is pointed out that things that they currently believe
entail or support things that they do not currently believe. Thus, the ordinary practice
of employing philosophical arguments as tools of rational persuasion seems to
      ? 

presuppose that we occupy some intermediate point on the spectrum of rationality,

somewhere well above the pathologically irrational but well below the ideally
rational.

16.4 Philosophical Disagreement and

the Pessimistic Verdict
Recall The Pessimistic Verdict:
There are no successful arguments for substantive philosophical theses.
Let’s waive the line of objection offered in the previous section to van Inwagen’s
criterion of philosophical success in order to see how he employs that criterion in
arguing for The Pessimistic Verdict. We take van Inwagen’s argument for The
Pessimistic Verdict to be contained in passages such as the following:
All philosophical arguments, or at any rate all philosophical arguments that have attracted the
attention of the philosophical community, have been tested under circumstances that approxi-
mate sufficiently to the circumstances of an ideal debate, that it is reasonable to conclude that
they would fail the “ideal debate” test. If any reasonably well-known philosophical argument
for a substantive conclusion had the power to convert an unbiased ideal audience to its
conclusion (given that it was presented to the audience under ideal conditions), then, to a
high probability, assent to the conclusion of that argument would be more widespread among
philosophers than assent to any substantive philosophical thesis actually is. (52–3)
...
Leaving aside those philosophical theses that almost everyone would accept without argument,
there are no philosophical theses that are both substantive and uncontroversial. If the argu-
ment is a success by the terms of my definition, why has it not got the power to produce
considerably greater uniformity of opinion among philosophers in the matter of its conclusion
than in fact exists? Or if it has got that power, why has this power not been exercised? These
questions, I believe, have no good answers. (54)

In these passages, the fact that the conclusion of a philosophical argument is typically
denied by a significant number of philosophers is treated as compelling evidence that
the argument would not convince the ideal agnostics. We concede that how actual
philosophers react to an argument is at least some evidence as to how the ideal
agnostics would react. Of course, how strong that evidence is depends on how closely
actual philosophers (and the actual circumstances in which they consider arguments)
approximate idealized agnostics (and their idealized circumstances). On the assump-
tion that actual philosophers approximate their idealized counterparts very closely,
then the reactions of the former will be comparatively strong evidence about the
reactions of the latter; on the other hand, the greater the divergence between actual
philosophers and their idealized counterparts, the weaker the evidence is.
Notice that, given van Inwagen’s criterion of success and the level of controversy
that we find among actual philosophers, there will tend to be an inverse relationship
between (i) one’s optimism about the intellectual qualities of actual philosophers and
their circumstances on the one hand, and (ii) one’s optimism about the number of
     

successful philosophical arguments on the other. For example, if one is very opti-
mistic about actual philosophers and their circumstances, and takes them to approxi-
mate their idealized counterparts very closely, then one will see the failure of actual
philosophers to reach agreement as strong evidence that the ideal agnostics would
not respond to an argument by converging on its conclusion; one will thus give
relatively high credence to The Pessimistic Verdict. If, on the other hand, one is
pessimistic about actual philosophers and their circumstances, and tends to see them
as a class as falling well short of their idealized counterparts, then one is free to be
more optimistic about the kind of convergence that would obtain among the
idealized agnostics.
Against this background, a natural worry is that van Inwagen’s pessimism about
actual philosophical arguments is an artifact of being overly optimistic about actual
philosophers and their circumstances. Offhand, one might think that even the most
rational philosophers fall well short of the kind of ideal rationality possessed by the
ideal agnostics (however exactly “ideal rationality” is understood). Moreover, even
when maximally charitable assumptions are made about the rationality and other
intellectual virtues of actual philosophers, there is still the pressing question of how
well actual philosophers approximate the initial neutrality of the idealized agnostics.
After all, in many cases in which a philosopher is presented with an argument for a
substantive conclusion, he or she will already have an opinion about that conclusion.
(Notice that in many cases, her having an opinion about the conclusion prior to being
presented with the argument will be perfectly reasonable; indeed, in many cases it will
be unreasonable for her not to already have an opinion, given her epistemic circum-
stances.) And in many cases in which she lacks an opinion, she will nevertheless have
some “predilection” one way or the other. But anyone who has such an opinion or
predilection in a given case already differs from an ideal agnostic in a potentially
crucial respect.
Indeed, we think that it can be shown, using assumptions that van Inwagen
himself accepts, that the class of actual philosophers is sufficiently dissimilar to the
class of ideal agnostics to undermine the kind of plausibility argument that he offers.
Consider first philosophical arguments for substantive theses that have in fact
convinced large numbers of philosophers of their conclusions, and which are
regarded by these philosophers as rationally compelling. We might think here, for
example, of Kripke’s case that ordinary proper names are rigid designators
(1980), Putnam’s case for semantic externalism (1975), Burge’s argument for “anti-
individualism” (1979), or van Inwagen’s arguments for incompatibilism about free
will.8 Or consider the philosophical argument that is van Inwagen’s primary concern
in the Gifford Lectures, the argument from evil. (Here we will assume, with van
Inwagen, that the thesis that God does not exist is in fact a “substantive” one.) Many

8
As van Inwagen (2004: 350) notes: “My arguments for incompatibilism . . . have in fact demonstrated
that they have the power to form philosophical opinion: they have convinced some philosophers who were
trying to decide whether to be compatibilists or incompatibilists to become incompatibilists.” In an
attached footnote, he adds that “I think it is very probable that ‘they have convinced some philosophers’
is a gross understatement. I think it is very probable that they have convinced a great many philosophers”
(emphasis added). We agree.
      ? 

philosophers believe that the argument from evil is successful. Here, for example,
is David Lewis, in the posthumously published essay “Divine Evil” (2007):
In my view, even the most ambitious version [of the argument from evil] succeeds conclu-
sively. There is no evasion, unless the standards of success are set unreasonably high. (231)

But of course, there are many philosophers who disagree with this assessment,
including van Inwagen himself.
Recall that van Inwagen explicitly assumes that there will be no disagreement
among the idealized agnostics: a given philosophical argument will either persuade
all of them or else none of them. Now, if Lewis is right that the argument from evil
“succeeds conclusively” and can be resisted only if “the standards of success are set
unreasonably high,” then clearly the ideal agnostics would be persuaded by the
argument. (Presumably, the ideal agnostics would not insist on unreasonably high
standards of success before coming to believe the conclusion.) And again, there are
many generally reasonable actual philosophers who share Lewis’ assessment. On the
other hand, if van Inwagen’s assessment of the argument from evil is correct, then the
ideal agnostics would remain unpersuaded. (And of course, there are many generally
reasonable philosophers who share van Inwagen’s assessment of the argument.)
What we can conclude from this is the following: given that there is some fact of
the matter about what the ideal agnostics would think about the argument from evil,
there are many generally reasonable philosophers whose own opinion about the
argument is a very bad guide to what that fact of the matter is. Moreover, the same is
true, mutatis mutandis, of Kripke’s arguments that ordinary proper names are rigid
designators, Putnam’s arguments for semantic externalism, van Inwagen’s arguments
for incompatibilism, and any number of other arguments known to the philosophical
community. In view of this, the level of optimism as to how closely actual philo-
sophers approximate their idealized counterparts required by Inwagen’s argument
seems unwarranted: if there is one thing that we can be sure of in this vicinity, it is
that there is a tremendous amount of slack between the assessments of actual
philosophers (given the kinds of disagreements we find about whether particular
arguments are compelling) and the assessments of their idealized counterparts (given
their complete lack of disagreement).
Consider next the class of philosophical arguments of which the following is true:
informed philosophical opinion is substantially divided between those who regard
the argument as rationally compelling, and those who deny that it is. In these cases,
is there any general reason to think that someone who makes the substantive
judgment that an argument is not compelling is more likely to be correct, or tracking
the view of the ideal agnostics, than someone who makes the opposite substantive
judgment? As far as we can see, there is not. Still less is there any reason to suppose
that those who make the substantive judgment that an argument is not compelling
are tracking what the ideal agnostics think in every case, which is what The Pessim-
istic Verdict requires. It is not, after all, as though Lewis’ substantive judgment that
the argument from evil is rationally compelling (and that one who responds to the
argument by believing its conclusion is responding in the most reasonable way) is
more likely to be wrong than van Inwagen’s opposite substantive judgment simply
because of the type of judgment that it is. To the extent that we have good reasons to
     

believe that the argument from evil is a failure, this is due to the kind of direct
engagement with, and meticulous, sustained critique of the argument that van
Inwagen provides in the rest of the lectures, as opposed to the kind of general
considerations about philosophical failure considered here.9

16.5 Conclusion
It is occasionally remarked that there is something distinctively odd about the genre
of philosophical festschrift. In order to honor an unusually distinguished living
member of the discipline, a group of that person’s scholarly peers, past and current
colleagues, former students, and admirers are invited to contribute essays on some
aspect of his or her life’s work. In their essays, the contributors typically set out to
show just how wrongheaded the distinguished philosopher’s views on some topic
really are, or at least, that his or her arguments on that topic are unsound and ought
not to convince anyone. In an obvious way, our own contribution to the current
volume is in keeping with this rather questionable tradition. For we have attempted
to show that, with respect to the topic of philosophical success and failure, van
Inwagen’s arguments are unpersuasive.
There is another respect, however, in which our contribution can be viewed as
more in keeping with the true, underlying spirit of the occasion. For decades, Peter
van Inwagen has been at the forefront of philosophy. Few living philosophers can
match the power, depth, and influence of his writings across a comparable range of
areas. Moreover, unlike many philosophers of significant influence—and we take this
to be greatly to his credit—van Inwagen’s influence is due in no small measure to the
perceived power of his arguments. Now, as van Inwagen is well aware, if his
metaphilosophical views about success and failure in philosophy are correct, then
it is overwhelmingly likely that all of these arguments are failures. As he writes:
. . . are there any successful philosophical arguments? I know of none . . . I hate to admit this, if
only because I should like to think that some of the arguments associated with my name are
successes. But I have to admit that it’s at best highly improbable that they are. (52)
However, if we are correct, and van Inwagen’s metaphilosophical views are mistaken,
then it is a distinct possibility that some of the arguments associated with his name
are, in fact, genuine successes, notwithstanding the fact that some among us remain
stubbornly unconvinced.

Acknowledgments
Earlier versions of some of these ideas were presented in a graduate seminar co-taught with
John Hawthorne at Princeton University in the spring of 2013; we would like to thank John
and the students for their feedback on that occasion. Thanks also to David Chalmers for
reading and commenting on an earlier draft, to John Keller for extending an invitation to

9
For more general reflections on the epistemic significance of persistent disagreement among actual
philosophers, see Kelly (2016).
      ? 

contribute to this volume, and (last but not least) to Peter van Inwagen for his stimulating
writings over the years, from which both of us have learned much.

References
Burge, T. (1979), ‘Individualism and the Mental’, in Midwest Studies in Philosophy 4 (1):
73–122.
Cherniak, C. (1986), Minimal Rationality (MIT Press).
Harman, G. (1986), Change in View (MIT Press).
Hetherington, S. (2001), Good Knowledge, Bad Knowledge: On Two Dogmas of Epistemology
(Oxford University Press).
Kagan, S. (1998), Normative Ethics (Westview).
Kelly, T. (2013), ‘Evidence Can Be Permissive’, in M. Steup, J. Turri, and E. Sosa (eds.),
Contemporary Debates in Epistemology (Blackwell Publishers, 2nd edn.): 298–313.
Kelly, T. (2016), ‘Disagreement in Philosophy: Its Epistemic Significance’, in H. Cappelen,
T. Szabó Gendler, and J. Hawthorne (eds.), The Oxford Handbook of Philosophical Meth-
odology (Oxford University Press).
Kelly, T., and McGrath, S. (2010), ‘Is Reflective Equilibrium Enough?’, in Philosophical
Perspectives 24 (1): 325–59.
Kripke, S. (1980), Naming and Necessity (Harvard University Press).
Lewis, D. (2007), ‘Divine Evil’, in L. Anthony (ed.), Philosophers Without Gods (Oxford
University Press), 231–42.
McGrath, S. (2010), ‘Moral Realism Without Convergence’, in Philosophical Topics 38(2):
59–90.
Putnam, H. (1975), Mind, Language, and Reality (Cambridge University Press).
Russell, B. (1986), The Philosophy of Logical Atomism and Other Essays, 1914–19 (Allen &
Unwin).
Schaffer, J. (2009), ‘On What Grounds What’, in Chalmers, Manley, and Wasserman (eds.),
Metametaphysics (Oxford University Press), 347–83.
Van Inwagen, P. (1990), Material Beings (Cornell University Press).
Van Inwagen, P. (2004), ‘Freedom to Break the Laws’, in Midwest Studies in Philosophy 28 (1):
334–50.
Van Inwagen, P. (2006), The Problem of Evil: The Gifford Lectures Delivered in the University of
St. Andrews in 2003 (Oxford University Press).
Weatherson, B. (2003), ‘What Good Are Counterexamples?’, in Philosophical Studies 115 (1):
1–31.
White, R. (2013), ‘Evidence Cannot Be Permissive’, in Steup, Turri, and Sosa (eds.), Contem-
porary Debates in Epistemology (Blackwell Publishers, 2nd edn.).
PART V
Afterword
 17
Concluding Meditation
Peter van Inwagen

Introduction
There can be no greater honor for a scholar than to be the recipient of a Festschrift.
My gratitude to John Keller and the (other) contributors is so great that, in attempt-
ing to write this opening paragraph, I discover that I am able to find no words that are
adequate to express it. I will only say, thank you all. Thank you all so very much.
It is perhaps unusual for a Festschrift to contain responses by the recipient to the
essays it contains. (A Festschrift, after all, is not a “Schilpp volume.”) Apparently,
however, the editors at Oxford thought that such replies would be advisable. In any
case, the purpose of this preface is to say a few words about my replies.
I have not responded to all the essays. The reason is twofold yet simple: there was
not enough time and there would not have been enough space. I am rarely able to
write a short, simple reply to or comment on a philosophical essay. I may be able
to do this if I agree almost entirely with the author’s arguments and need only make a
few minor points. This is the case with my reply to Michael Loux’s essay. Or I may be
able to do this if the essay employs a system of philosophical concepts that is radically
different from my own—in which case I can often find something brief but helpful to
say about the nature of the gulf that separates the concepts the author employs from
the concepts I employ. This is the case with my reply to Laurie Paul’s essay. It even
occasionally happens that I agree entirely with an author—that the author’s essay is
one that, matters of style apart, I might have written myself. This is the case with Eric
Olson’s essay. In such a case, no reply is needed beyond what Alvin Plantinga once
called “a brief celebratory ceremony.” I will accordingly enact a suitable brief
celebratory ceremony: I hereby declare that Eric’s essay is a fine piece of work.
But cases like these are the exceptions. If I am commenting on or replying to an essay
that is addressed either to my own work or to some problem or topic that deeply
interests me, I almost always have a very hard time keeping my responses and comments
within reasonable bounds. However much I say, there always seems to be more.
For this reason, if I were to attempt to reply to all the essays this volume contains, it
would almost certainly be published only after my death—and with some of the
responses still unwritten. And it would probably not be a volume. It would probably
comprise at least two volumes—assuming that anyone was willing to publish such a
monster.
   

When I began to write replies to the essays in this book, I started with the ones with
whose authors I am in most fundamental disagreement—but authors whose ideology
(in Quine’s sense) and ontology are sufficiently similar to my own that a useful and
extended discussion of the matters on which we disagree is possible. Accordingly, the
first replies I wrote were to the essays of Louise Antony and Alex Rosenberg. No one,
I think, would dispute the statement that their essays represent points of view that are
fundamentally opposed to my own. But Louise’s and Alex’s points of view are not
“fundamentally opposed” to my own point of view in the way Laurie Paul’s is. It’s
evident that Louise and Alex’s opinions concerning God, Freedom, and Immortality
(or to use Douglas Adams’s tripartite list, Life, the Universe, and Everything) are
radically different from mine. Nevertheless, these disagreements, radical though they
are, can be framed using a common vocabulary and a common system of metaphys-
ical concepts—and this is not so in the case of my radical disagreements with Laurie.
My work on the replies to Louise and Alex’s essays illustrated the basic soundness
of Hofstadter’s Law: Any project always takes longer than you expect, even when you
take Hofstadter’s Law into account. I next wrote the replies to the essays of Michael
Loux and Laurie Paul, which went more quickly (although, of course, less quickly
than I expected), and then went on to the essays on “Method”—a topic that is of deep
and abiding interest to me. (The essays, that is, of David Chalmers, John Keller, and
Thomas Kelly and Sarah McGrath.) When I had finally finished these three replies,
I had written replies to only seven of the seventeen essays this volume contains
(counting the “Symposium on the Fixity of the Past” as two essays), and had written
almost 30,000 words. At that point, I was simply out of time and out of space.
I am deeply saddened by the fact that I have not written replies to many essays that
are by old students and old friends of mine—in several cases, both. (Sara Bernstein is too
young to be an old friend, but she is a very good friend.) I am deeply saddened by the fact
that I have not written replies to many excellent essays—ten of them, to be exact, for
every essay in the volume is an excellent essay. In only one case—the case of Eric Olson’s
essay, for the reasons I have given—was the failure to reply to an essay a matter of
deliberate choice. Setting that case aside, that I wrote replies to the seven essays I did
write replies to and not to the other equally deserving nine essays is to a very large degree
a matter of chance: an unintended consequence of the point at which I chose to begin,
my inability to write short replies, and the amount of time and space available to me. In
most of the possible worlds closest to actuality, I wrote replies to the members of some
other proper subset of the seventeen essays. I will close by thanking those contributors to
whose essays I did not write a reply—Eric Olson, Sara Bernstein, Mark Heller, Alicia
Finch, Neal Tognazzini, John Martin Fischer, Wes Holliday, Eleonore Stump, Frances
Howard-Snyder, Daniel Howard-Snyder, and Lynne Rudder Baker—for their contribu-
tions, which I read with pleasure and gratitude. All but two of these eleven philosophers
are close friends of mine, and five of them are former students. I wish I had been able to
do more. I hope my friends and students will forgive me.

17.1 Reply to Michael J. Loux

I turn now to Michael Loux’s rich and rewarding paper. I can think of no paper that
provides a better overview of what is going in ancient, medieval, recent, and present-day
   

discussions of the problem of universals—or, better, discussions in which the concept

“universal” plays a central role.1 In these brief remarks, I will touch on only two
things that Loux says. Both these things are statements about my own work, but,
given the nature of this book, that is perhaps excusable. The first of these statements
concerns a very minor matter. The second, however, is of central metaphysical
importance. Both statements occur in footnotes. The first is in note 1. In this note,
Loux expresses concern about my
. . . use of the framework of sets or classes in the definition of a category. Since sets or classes
have their memberships essentially, I do not see how van Inwagen’s definition can avoid the
conclusion that a category in one possible world is different from a category in another world
merely in virtue of their having different extensions in those worlds. The two could be
composed exclusively of ontologically indistinguishable objects. Since kinds (like properties
and relations) do not have their extensions essentially, I would have employed the notion of a
kind in defining the concept of a category.

On this matter, I must refer the reader to note 6 to “What is an Ontological

Category?”,2 which reads in part:
The “classes” that figure in this essay are—or are if they really exist—much more like biological
taxa than they are like sets. . . . Like taxa, and unlike sets, they can change their membership
with the passage of time and the membership of a class in one possible world may not even
overlap its membership in another. Like taxa, and unlike sets, moreover, they may have
“borderline members”. . . . I am not, however, seriously asserting that there really are things
that have the properties I have ascribed to classes. I issue this promissory note: I could—the
result would be rather awkward, I concede—eliminate the apparent reference to and quanti-
fication over classes in the sequel by paraphrase.
Here are two related points. First, my official ontology does not contain sets or
classes. “In the ontology room,” I eliminate (or at least I have committed myself to
being able to eliminate) by paraphrase all reference to and quantification over sets in
favor of quantification over properties. Secondly, my ontology contains “kinds” only
insofar as kinds can be identified with properties—the properties that, as it were,
correspond to them. Thus, for example, I would identify the kind “horse” (supposing
horses to constitute a kind) with the property equinity or horsehood or “being a
horse.” And if, in the course of a philosophical discussion, I am told that properties
have the wrong, well, properties to be kinds, I will simply make the philosopher who

1
I heartily agree with Loux’s contention (p. 12) that there is no such thing as the problem of
universals—and with his arguments in support of this contention. I would add that in my view, this
contention is a special case of what I take to be a general law of philosophical terminology. Some other
instances of this law: there are no such things as the mind–body problem, the problem of free will (and
determinism), the problem of evil, the problem of other minds, the problem of knowledge (and skepticism),
the (hard) problem of consciousness. . . . In general: no phrase of the form ‘the problem of X’ that is in
general philosophical use is actually a proper definite description. In every case, careful analysis of the texts
in which treatments of the supposed problem are commonly supposed to be found shows that these texts
comprise discussions of a family of problems that are in many cases closely interrelated but certainly
distinct. (And, for any such family, it is rare for a given work that discusses some of them to discuss all of
them.)
2
Van Inwagen (2014).
   

has advanced that thesis a present of the word ‘kind’: I will stop using it and frame my
theses by using words like ‘property’ and ‘attribute’.
The second point pertains to Loux’s note 16. It is a very long note indeed, and I will
quote only a part of it. (The reader should study the whole note carefully before
proceeding.)
Van Inwagen . . . denies that it counts as a substantive explanation of the fact that a concrete
particular is, say, green to claim that it exemplifies the color green. Whether he is right or not,
most of those involved in the debate would take such a claim to involve a genuine explanation;
and despite his denial here, van Inwagen himself takes character to be grounded in what he
calls properties, and his properties include what have traditionally been called universals. Thus,
van Inwagen . . . tells us that, in our prephilosophical moments, we all believe that there are
anatomical features that insects and spiders share, and he argues that there is (likely) no way of
understanding how that belief could be true without conceding the existence of shared
properties. But surely this is to claim that the universals in question ground one or more
facts about shared character . . . For van Inwagen . . . properties are . . . what he calls unsaturated
assertibles, and, on his view it is because they share the relevant assertibles that spiders and
insects agree anatomically.

I certainly agree that “most of those involved in the debate would take such a claim to
involve a genuine explanation.” They are, I say, wrong, wrong, wrong.3 At any rate,
they are wrong if they take properties or universals to be anything like what I take
them to be: unsaturated assertibles. And I’m not going to argue about whether
unsaturated assertibles should be called properties or universals. Once more I find
myself in a generous mood: I’m happy to make a present of the words ‘property’ and
‘universal’ to anyone who has strong feelings about their proper use and who thinks
that I am using them improperly. (“But who sees not that all the dispute is about a
word?”)
Let’s look at what an unsaturated assertible is. As a saturated assertible (a prop-
osition) is something one can say—and something such that if one says it what one
thereby says is true or false4—an unsaturated assertible is something one can say of
something. (And if one says an unsaturated assertible of something, it will either be
true of that thing or false of that thing.5) Suppose, for example, that I ask you what
color this apple is, and you reply, “It’s green.” And suppose Alice asks Jerry what
color his copy of A Theory of Justice is, and he replies, “It’s green.” Then there is one
thing, one unsaturated assertible, such that you said it of the apple and Jerry said it of
his copy of A Theory of Justice. Call it “that it is green.” (‘ “That it is green” ’ is
supposed to be a proper noun—or noun-phrase. If you want words that sound more
proper-noun-phraseish, call it ‘what one says of something when one says of it that it
is green’. Or if you want a perfect nominal, call it ‘the thing one says of something
when one says that it is green’.) Now it happens that what one says of Jerry’s book

3
Or, better, in Wolfgang Pauli’s fine phrase, they are not even wrong.
4
Or of indeterminate truth-value: neither determinately true nor determinately false.
5
Or of indeterminate application to that thing: neither determinately true nor determinately false of
that thing.
   

when one says that it is green is true of that book. But would anyone assent to the
following statement?
It counts as a substantive explanation of the fact that Jerry’s copy of A Theory of Justice is green
to claim that the unsaturated assertible “that it is green” is true of it.
Or this one?
It is because the unsaturated assertibles “that it is bilaterally symmetrical” and “that it has a
segmented body” and “that it has an exoskeleton” are true both of every spider and of every
insect that spiders and insects agree anatomically.

I certainly hope not. If anything, the order of explanation is the other way round:
That Jerry’s copy of A Theory of Justice is green counts as a substantive explanation of the fact
that the unsaturated assertible “that it is green” is true of it.
It is because spiders and insects agree anatomically (in that every member of the order Araneae
and every member of the class Insecta is bilaterally symmetrical, has a segmented body, and
has an exoskeleton) that the unsaturated assertibles “that it is bilaterally symmetrical” and
“that it has a segmented body” and “that it has an exoskeleton” are true both of every spider
and of every insect.
I refer anyone who disagrees to The Philosopher: It is because you are pale that we
who say that you are pale have the truth (Metaphysics IX 10).
We should remember that unsaturated assertibles, like numbers, are abstract objects
and can figure in explanations only in ways analogous to the ways in which numbers
can figure in explanations. (If every member of the US Supreme Court votes on an
issue, the vote cannot result in a tie because the Court has an odd number of members;
if you throw a pair of dice, you have one chance in 18 of throwing 11 because there are
36 ways the dice can fall and only two of them add up to 11 and 2 divided by 36 is equal
to 1 divided by 18.) Anyone who accepts the existence of numbers will accept the
following biconditional: ‘A planet has two moons only if and only if the number 2 is the
number of its moons’. But who would say that the fact that the number 2 is the number
of the moons of Mars counts as a substantive explanation (or any sort of explanation at
all) of the fact that Mars has two moons? Or, again, anyone who accepts the existence
of (singular) propositions will accept the biconditional ‘A planet has two moons if and
only if the proposition that it has two moons is true’. But who would say that the fact
that the proposition that Mars has two moons is true counts as a substantive explan-
ation of the fact that Mars has two moons?
In my view, the idea of a metaphysical explanation of the fact that (say) this copy of
A Theory of Justice is green is meaningless—that is to say, there is no such idea, there
are only the words ‘metaphysical explanation of the fact that this copy of A Theory of
Justice is green’, words that darken metaphysical counsel. There are, of course,
efficient-causal explanations of the fact that the book is green (explanations involving
the manufacturer’s use of green ink), and there are formal-causal explanations of fact
that the book is green (its surface absorbs photons and then, the book being more or
less in thermal equilibrium with its surroundings, emits photons; the photons it emits
have wavelengths predominantly in the 520–570 nm range). And that is all the
explanations of the book’s being green that there are. No meaningful sequence of
   

words in any possible language counts as a metaphysical explanation of the book’s

being green.

17.2 Reply to L. A. Paul

Laurie Paul and I stand in a relation that I stand in with almost every other
metaphysician whom I admire: near total disagreement. I find my root-and-branch
disagreement with her immensely profitable, however, because I learn so much from
it. I will give an example of the way I have learned from our disagreements that
I think is particularly important. She and I once arranged to have a lunch during
which we would try to get clear about this “property” business. I knew that she
advocated a one-category ontology, an ontology according to which everything is a
property. The Pauline ontology, as I shall call it, tells us that there are nothing but
properties and sums or fusions of properties—and that a fusion of properties is itself
a property. (According to the Pauline ontology, its author, Laurie Paul herself, is a
property—as are we all.)6
Now Paul and I both say we believe in objects we call properties, but (I supposed)
we couldn’t mean anything like the same thing by the word ‘property’. Given what
I mean by ‘property’ to say, “I am a property” is nonsense. Not nonsense like ‘Gubble
buggle guggle’, of course, and not (exactly) nonsense like ‘Das Nichts nichet’ or ‘The
world is a progressively realized community of interpretation’, but nonsense like ‘The
shadow of Caldwell Hall itself casts a shadow’ or ‘If you want to extract a cube root,
you will need to use a number 18L dental forceps’ or ‘Sara Bernstein is drinking
coffee from a two-dimensional cup’.
At some point during our luncheon conversation, I asked Paul why she believed
that the things she called ‘properties’ existed—why she supposed that there were
things that had the properties (in my sense of ‘property’) that she ascribed to
properties (in her sense of ‘property’). Her answer transfixed me: she told me that
she thought that one of the strongest arguments for the existence of propertiesLP7 was
that one could see some of them.
Just look at this Granny Smith apple, for example. You can see the property greenness
right there before you; it’s in the apple. Some would use some such conveniently obscure
phrase as ‘it inheres in the apple’, but the admirably forthright Pauline ontology has no
use for convenient obscurity: according to the Pauline ontology, greenness is a con-
stituent, and, in fact, a part—a part in the strict and mereological sense—of the apple.8
The apple is a fusion of many properties, greenness among them, and that’s why

6
I trust that the reader has already read the essay on which I am commenting and thus has no need for a
more complete description of the Pauline ontology. (That essay also contains answers to such “obvious”
objections to its author’s ontology as the following. According to Paul, parthood is not transitive. For
suppose that Amber the electron is a part of Hydra the hydrogen atom; if Paul is right, Amber is a fusion of
various properties, among them a part of the universal “negative charge.” But Hydra is not negatively
charged and thus does not have a part of negative charge as a part.)
7
I trust I need not explain this technical term or its foil ‘propertyPvI’; I promise to employ them only
when the sense of the word ‘property’ is not made clear by its context.
8
Or perhaps I should say ‘in a strict and mereological sense’, for a color is not a part of an apple in the
same sense of ‘part’ as that in which a seed is a part of an apple.
   

greenness is before you when the apple is before you. (“But Professor Paul, is it the
particular greenness of this apple that is before you when you look at the apple or is it
the universal greenness?” “Both. The universal greenness is nothing other than the
fusion of all particular greennesses. Or put the matter this way: a particular greenness is
a connected part, or perhaps a maximally connected part, of the universal greenness. If
we use the term ‘trope’ in the sense the word has in current analytical metaphysics, a
trope—like the greenness of this apple—is a part of the object of which it is a trope. For
an object like an apple to instantiate a universal is for it to overlap that universal, to
share a part, a trope, with that universal. The objects that are wrongly called particulars
by some metaphysicians—apples, all ‘moderately sized specimens of dry goods,’ stars—
are fusions, maximal fusions, of tropes. But when I was speaking of the property
greenness, I meant the universal.” “Ah, but then you never see the universal, you only
see parts of it.” “Yes, and, unless you’re an astronaut, you’ve never seen the Atlantic
Ocean—you’ve only seen parts of it. The analogy is exact: you can see the color
greenness, the universal, when you look at the apple in the same sense of ‘see the’ as
the sense in which you can see the Atlantic Ocean when you’re standing on the Outer
Banks facing east.”)
This was a revelation to me. Suddenly I realized that “platonism” concerning
universals (my sort of platonism, at any rate) and “Aristotelianism” concerning
universals are not sisters-in-arms, allies in the battle against nominalism who
disagree only on a few small, technical points. (Of course Paul is not an Aristotelian,
since she denies that there are “substances,” things that are not properties or
universals, things of which other things are predicated but which are not themselves
predicated of things. But her picture of the intrinsic nature of universals is very like
that of an Aristotelian who contends that universals exist by inhering in substances.
She, as it were, “subtracts” the substances from Aristotelianism, and identifies all the
objects the Aristotelian typically gives as examples of substances—the Granny Smith
apple, a bronze ball, Socrates—with fusions of parts of universals.) Rather, they differ
from each other quite as much as either differs from nominalism, even from austere
nominalism. It is, indeed, a defensible position that (my sort of) platonism is more
like austere nominalism than it is like Aristotelianism in that each gives the same
account of concrete particulars: particulars are what David Armstrong has called
“blobs”; the only kinds of “structure” a concrete particular has are spatial and
mereological structure—and the only proper parts a concrete particular has are
other, smaller concrete particulars.
For my part, I do not believe in propertiesPvI because I can see them. I believe in
them because I think that many sentences that express true propositions—‘Spiders
and insects share many important anatomical characteristics’, for example—would
not express true propositions if ‘There are propertiesPvI’ did not also express a true
proposition.9 Indeed I believe that no one can see them—just as no one can extract a
cube root with a forceps.10 PropertiesPvI are very much the same sort of thing as

9
For the whole story, see van Inwagen (2004).
10
I mean . . . I mean how could you see a property—on anybody’s understanding of ‘property’? You can
only see things if they absorb electromagnetic radiation and re-emit it. You can see a book or an apple
because they, or parts of their surfaces, do just that. They can do that because they are made of atoms and
   

propositions—but where propositions are true or false simpliciter, properties are true
or false of things. PropertiesPvI stand to declarative sentences in which one variable is
free much as propositions stand to closed declarative sentences: ‘The apple that
Newton saw fall was green’ expresses the proposition that the apple that Newton
saw fall was green; ‘x is green’ expresses the property greenness.
While Paul and I mean different things by ‘property’—at least when we are doing
metaphysics—, there is nevertheless a certain role that English words like ‘property’,
‘attribute’, ‘quality’, ‘feature’, and ‘characteristic’ play in our discourse, a role defined
by the tacit rules embodied in and governing that discourse. Paul thinks that proper-
tiesLP (assuming that there are such things) are better fitted to play this role than
propertiesPvI (assuming that there are such things). That is not why she believes in
their existence, but that is why she calls them ‘properties’. And I, of course, think just
the opposite. How could we have arrived at these two radically opposed positions?
I believe that a large part of the explanation—perhaps the whole of it—is meth-
odological. Paul believes that the proper method of metaphysics is to construct
explanatory theories: there are certain metaphysical data (for example: the book
and the apple are both green; we can establish that they are both green by visual
examination) and our job as metaphysicians is to order and explain those data.11 It
turns out that the best explanation of those data involves postulating properties—
propertiesLP. My method is to examine those truths that we bring to metaphysics
from everyday life or from the sciences and other non-philosophical disciplines (and
perhaps even from other parts of philosophy, such as ethics) and to attempt to
discover their metaphysical consequences—to discover which of their logical conse-
quences (if any) are metaphysical propositions. I have tried to explain (in “Relational
vs. Constituent Ontologies”12) why I think that Paul’s method (and she is hardly the
only analytical metaphysician to employ—and to endorse—this method) and my
method result in radically opposed ontologies.13 What is really at stake in our
methodological disagreement is this: is there such a thing as explaining what it is
for something to be, e.g., green (other than a physical explanation: it emits photons
with wavelengths predominantly in the 520–570 nm range). As I said in “Relational
vs. Constituent Ontologies,”

molecules that have electronic structures: a photon kicks an electron momentarily into a more energetic
atomic or molecular orbital, and another photon is emitted when the electron falls back into a lower-energy
orbital; some of these emitted photons interact with the visual apparatus of the observer. Properties (on
anybody’s account of properties) aren’t composed of atoms and molecules, are they? Granted, the fact that
a certain book or apple has the property greenness (the fact that it’s green) has its formal cause in the
electronic structure of the atoms and molecules that compose its surface layers, but that doesn’t mean that
one can see the property. One can see that the book or apple is green, of course, but that doesn’t mean
that one sees the property—any more than the fact that an appropriately placed observer can see that Laurie
Paul is a biped means that that observer can see the number 2. “But one can see the property blueness when
one looks up on a fine summer day, and when one is having that visual experience characteristic of that
condition, there’s nothing above one such that one sees that it is blue.” This footnote is getting out of hand.
For a reply to the “blue sky” objection, I refer the reader to van Inwagen (2004: 136).
11 12
See Paul (2012). Van Inwagen (2011).
13
See also my reply to Michael Loux in the present volume.
   

. . . no set of statements among all possible sets of statements counts as [a metaphysical or

ontological] explanation of what it is for a particular to have a property or for two distinct
particulars to have the same property.
In my view, Paul’s methodology is a methodology for solving a problem that does not
exist (a problem that cannot be coherently posed), and thus it is not surprising that it
leads her to postulate entities that make no sense. (I should perhaps mention that
I would say the same thing about On the Plurality of Worlds,14 which I regard as a
wholly admirable book, one of the greatest philosophical achievements of the twen-
tieth century. As I’ve said many times, meaninglessness is what we risk in doing
metaphysics; what one risks in metaphysics isn’t being wrong—except in the sense in
which someone who believes that a cube root can be extracted with a forceps is
“wrong.” What we risk is not even being wrong.)
I am going to say just two more things about the issues that separate Paul and me.
The first is a comment on the following definition.
Substance theory takes objects to be, fundamentally, primitively unanalyzable or irreducible
substances of different sorts, and holds that substances have properties by standing in some
sort of relation to universals or other entities. (34)

My ontology certainly includes substances. (There are at least two importantly

different definitions of substance. No matter: the things I call substances satisfy
them both. I believe, moreover, that the terms ‘substance’, ‘concrete object’, ‘particu-
lar’, ‘individual’, and ‘object that can be an agent or a patient’, although not equiva-
lent in meaning, all have the same extension.) And, as I have said, I agree with the
austere nominalist (who may or may not be happy with the term ‘substance’) about
the nature of such things. But I reject, I most emphatically reject, the thesis that
“substances have properties by standing in some sort of relation to universals or other
entities.” I give my reasons for this in my reply to Michael Loux’s paper in this
volume. (The reader of that reply will see how emphatic my rejection of this thesis is:
it would not be entirely unfair to describe what I say there as a rant.)
Secondly, it seems to me that the Pauline ontology must recognize the existence of
propertiesPvI. Consider our Granny Smith apple. Granny (so to name the apple) is
green. Granny has (exemplifies, instantiates) the universal greenness. That is to say
(according to the Pauline ontology) Granny overlaps the universal greenness: its
greenness is a common part of the apple and the universal. Well and good. But now
consider Great-granny, a slightly larger but very similar apple in the vicinity of
Granny. One of the things one can say (and say truly, according to the Pauline
ontology; in the sequel, I’ll suppose that the Pauline ontology gets the ontological
structure of apples right) about Granny is that it overlaps the universal greenness.
(I myself said that very thing about Granny a few sentences back.) And one can say
the same thing about Great-granny (also truly). Here, I’ll do it: Great-granny overlaps
the universal greenness. So: one of the things that is true of both Granny and Great-
granny is “that it is green” (or “the thing that one says of something when one says of

14
Lewis (1986).
   

it that it is green”). And, of course, that’s not the only thing that is true of them both.
Surely the following statement is true?
There are many things that are true of both Granny and Great-granny.
And that statement certainly appears to imply the existence of things that can be
true of things; that is, the existence of “unsaturated assertibles”; that is, the existence
of propertiesPvI. I do not see any way in which to paraphrase this statement in such a
way that the paraphrase implies the existence of nothing but propertiesLP. For—
surely?—Paul does not want to say that there is such a propertyLP as “overlapping the
universal greenness” or that overlapping the universal greenness is a part of both
Granny and Great-granny.

17.3 Reply to Louise Antony

Louise Antony’s paper is an admirable combination of clarity of argument and moral
passion. Nevertheless, either I have misunderstood its central line of argument or it turns
on a misunderstanding of my project in The Problem of Evil (van Inwagen 2006).15 I’m
going to quote a paragraph from the paper that illustrates the nature of the misunder-
standing on which I believe the paper rests.16 (The paragraph nicely illustrates the
misunderstanding, but it is not the only place at which that misunderstanding is
manifested.) At one point in the book I said, “It is at least very plausible to suppose
that it is morally permissible for God to allow human beings to suffer if the inevitable
result of suppressing the suffering would be to deprive them of a very great good, one that
far outweighs the suffering.” The following paragraph is a response to this statement.
But this is not good enough. It’s not good enough for a Defense to say that it is “at least very
plausible to suppose” that there’s something that makes it all right for God to behave in a way
that would be morally wrong if a human being did it. The whole point of the Defense is to show
us what that difference might be. Just as the fundamentalist wants an example of how a part of
an eye or a proto-wing might be adaptive, Atheist wants to know what kind of difference
between God’s situation and the situation of the human doctor would make it morally
permissible for God to do what it would not be permissible for the doctor to do. (183)
To explain why I say that this paragraph illustrates a misunderstanding of my
project—of what I was up to, if you like—, I will remind you of the “dialectical
context” in which the sentence ‘It is at least very plausible to suppose that it is morally
permissible for God to allow human beings to suffer . . . ’ occurs.
Theist and Atheist are conducting a debate before an audience of neutral agnos-
tics.17 At a certain point in this debate, Atheist is trying to convince the audience of
agnostics that, in light of the (global) argument from evil, they should cease to be

15
Perhaps she will reply to this statement as R. M. Chisholm once replied to a similar statement:
“I accept the disjunction.”
16
I’m going to suppose that the reader is familiar both with her essay in the present volume and The
Problem of Evil.
17
A neutral agnostic either assigns a subjective probability of 0.5 to the proposition that God exists
(and, of course, to its denial) or declines to assign it any subjective probability (or to place it within any
range of subjective probabilities—other than [0,1]), and, of course, does the same for its denial.
   

agnostics and become atheists like herself.18 Theist is not trying to convince the
agnostics to become theists (not in this part of the debate, anyway); he is simply
trying to convince the agnostics that their reaction to the argument Atheist has
presented should be this: “For all we know, at least one of the premises of that
argument is false.”19
In this dialectical context, Atheist bears a much heavier burden of proof than Theist.
She has to convince the agnostics that all the premises of the argument are true. (That is
what she has to do if her intention is to convince them that God does not exist. But if,
failing that, she would at least like to convince them that it’s significantly more
probable that God does not exist than that God does exist—see note 18—then she
must at least convince them that it’s significantly more probable that all the premises of
the argument are true than it is that at least one of them is false.) All he has to convince
the agnostics of is that they shouldn’t be convinced by her arguments. The relations
between Atheist and Theist and their audience are very like the relations between the
counsel for the prosecution, the counsel for the defense, and the jury in a criminal trial.
(Think of matters this way: the proposition that God exists is on trial; the charge is
falsity.) In a common-law criminal trial, the burden of proof (the burden of having to
prove things) falls on the prosecution. (Or—a picky philosopher’s qualification—a very
light burden of proof falls on the defense. The defense will sometimes tell the jury
things like this: that the prosecution has failed to show that the defendant was in
O’Malley’s Bar on the night of the 11th. The defense is then required to prove—that is,
is required to convince the jury—that the prosecution has indeed failed to show this.)
As I have described the debate, Theist attempts to convince the “jury” of agnostics
that, for all they know, at least one of the premises of the (global) argument from evil
is false by telling a story. The story is to have this feature:
(a) It logically implies that at least one of the premises of the argument from evil is
false. (If the story implies both that God exists and that the world contains evils of the
same sorts and in the same amounts and distributed in the same way as the evils of
the actual world, then the story will imply that at least one of the premises of any
logically valid version of the global argument from evil is false. And I represent Theist
as telling a story that has just those two properties.)
And Theist hopes it will have this feature:
(b) In the end, when all is said and done, when the debate has run its course, the
members of audience of neutral agnostics will react to it, the story, by saying—
imagine that they are sequestered and are discussing the plausibility of Theist’s story
among themselves—things along the lines of, “For all I know, that story is true” or
“The story he has told seems to be a real possibility, one we should take seriously” or
“Gee, if God exists, the rest of that story could be true, too. I mean for all we know.”

18
That’s what I said in The Problem of Evil. Conversation with several philosophers has convinced me
that it would have been better if I had said, ‘ . . . either cease to be agnostics and become atheists, or, failing
that, should at any rate cease to be neutral agnostics—should conclude that atheism is significantly more
probable than theism’.
19
We assume that the argument under consideration is formally valid.
   

I had thought that the ideas appealed to in my statement of (b) were reasonably
clear. But I may have been wrong. Antony sometimes uses language that suggests to
me that she thinks I use variants on the phrase ‘for all one knows’ with a meaning
weaker than the meaning I intended.20 I will, therefore, try to make my intended
meaning clearer than I perhaps have.21
Let us imagine a murder trial. Winifred has died of arsenic poisoning, the coroner’s
verdict was “unlawful homicide,” and Charles has been arrested for the murder and
brought to trial. The prosecution’s case against Charles is essentially that he was the only
person who both had a reason to desire Winifred’s death (he stood to inherit her vast
wealth upon her death) and had the means and opportunity to introduce arsenic into
her final meal. If the counsel for the defense suggests to the jury—by an adroit series of
questions put to a witness she has called—that a certain person other than Charles also
had “motive, means, and opportunity,” and if the prosecution is unable to rebut this
suggestion, it is reasonable to suppose that the members of the jury will conclude that
“for all they know” Charles is not the murderer, that that other person’s having
committed the murder is a real or serious possibility. If, in contrast, the defense counsel
calls the lead investigating officer as a witness, and says to him (out of clear blue sky, as it
were), “I put it to you, Detective, that the evidence you have presented to this court does
nothing to rule out the possibility that Winifred was a spy in the service of a hostile
power, and that she was poisoned by a C.I.A. assassination squad—the C.I.A. of course
having known that everyone would suppose that she had been poisoned by the
accused,” she will get a very different reaction from the members of the jury. None of
them will say, “Gee, for all I know that’s how things happened.” They, or the comedians
among them, will rather say things along the lines of, “Oh, sure—and for all I know, my
Aunt Harriet’s turnip pancakes are a cure for cancer” and “Yeah, that might be what
happened. And flaming monkeys might fly out of my . . . um, ears.”22 (The trial judge
and the Bar Association will no doubt have something to say about the matter as well.)
A story that has the feature (a)—and which is presented by someone who hopes it
will have feature (b)—I call a defense:23 Theist attempts to convince the agnostics
that, for all they know, at least one of the premises of the argument from evil is false
by presenting them with a defense. Antony has no objection to this strategy, but
insists that the two defenses I have presented24 are defective because they include

20
See, for example, the two paragraphs that precede her statement of the No Tolerance Principle. But
I confess to an imperfect understanding of the meaning of these two paragraphs.
21
The concept “for all one knows” obviously has some sort of important connection with the concept of
(subjective) probability, but drawing that connection would not be a trivial exercise. If you are about to be
dealt a card from a well-shuffled standard deck, then for all you know that card will be the four of
diamonds, despite the fact that there is only one chance in 52 of your being dealt that card.
22
A bowdlerized adaptation of a colorful demotic illustration of the distinction between “a possibility”
and “a real possibility.”
23
Feature (a) is very close to being an intrinsic feature of the stories that have it (one might contend that
it is technically a relational feature of those stories, since they might lack it if, say, the evils of the world were
significantly different). But feature (b) is blatantly relational; moreover, Theist’s hope may fail: the story
may turn out not even to have feature (b). It would therefore be more perspicuous to speak of a story’s
being on a certain occasion presented as a defense than of a story’s being a defense.
24
One is a story that entails the existence of both God and all the evils that have befallen human beings;
the other is a story that entails the existence of both God and all those evils that have befallen non-human
   

mysteries, things that—if they exist, and there is no reason to think that they do—are
beyond our understanding. It is as if, she says (well, she doesn’t say it; the example is
mine, not hers, but I think she will not object to it), I tried to show that a time
machine was (for all we know) possible by presenting a blueprint for the construction
of a time machine that included a neat box labeled “flux capacitor,” and explained
that a flux capacitor is a mechanism such that, if you constructed a device according
to my blueprint, and placed the mechanism in the device at the place indicated in the
blueprint, the result would be a working time machine. (After all, “for all we know”
it’s possible for there to be a mechanism that has exactly that property.)
I reply that it’s perfectly legitimate for Theist to include “mysteries” in a story that
is to serve as a defense—if it is reasonable for him to hope that the agnostics will
(despite Atheist’s best attempts to block this reaction) react to that story by saying,
“For all we know, that story is true.”25 Or, to put my point in terms of propositions
and alethic modality, it’s perfectly legitimate for Theist to include a proposition p in a
defense even if he has no argument for the metaphysical possibility of p. The
metaphysical possibility of the propositions his defense comprises is not something
Theist hopes to establish or to convince the agnostics of. He hopes, rather, that the
agnostics will—in the end, when Atheist has said everything about the story that she
has to say—react to each proposition contained in his defense by saying, “For all we
know, that proposition is true”—a statement that is, of course, consistent with, “For
all we know that proposition is true, and, for all we know, that proposition is
metaphysically impossible.”26 (For all I know, a sequence of thirty 7s occurs some-
where in the first 1030 digits of the decimal part of π; and, for all I know, it is
metaphysically impossible for there to be such a sequence. For all I know, signals are
sometimes transmitted faster than light—perhaps by advanced extraterrestrial beings
who would regard our physics as quaint; and, for all I know, it’s metaphysically
impossible for signals to be transmitted faster than light.)
I will give an example of an obviously legitimate defense that employs a mystery.
Many people who would like to believe that there are many technologically
advanced civilizations inhabiting the planets of other stars are troubled by the fact
that we have—as of this date—detected no radio signals whose origin was an extra-
solar civilization. Let us call this phenomenon “cosmic radio silence”—‘silence’ for
short. And let us use the term “alienist” to describe someone who believes that there
is at least one technologically advanced civilization other than our own within 1000

animals—throughout the hundreds of millions of years during which there have been organisms capable of
suffering—that cannot be ascribed to the actions of human beings. (The first defense was at a later point
supplemented with further elements intended to address the “local” argument from evil. In this reply, I will
not consider that aspect of the defense.)
25
“It’s not good enough for a Defense to say that it is ‘at least very plausible to suppose’ that there’s
something that makes it all right for God to behave in a way that would be morally wrong if a human being
did it. The whole point of the Defense is to show us what that difference might be.” Why? Who says that’s
the whole point of the defense? In her paper, Antony capitalizes ‘defense’ to remind her readers that the
word is my technical term. Well, yes, it’s my technical term, and I have to insist that nothing I said about its
meaning implies that the whole point of any defense is to show how something could be.
26
And, of course, he wants them so to react to the conjunction of all the propositions the defense
comprises—to the “whole defense.”
   

light-years of the earth. (There’s no other good use for the word.) If theists face “the
problem of evil,” alienists face “the problem of silence.”
It would be easy to construct an argument for the falsity of alienism that is similar to
the (global) argument from evil in its structure, and in which silence plays the role that
evil plays in the argument from evil—the “argument from silence.” It would be so easy
that I won’t bother to do it. Now imagine a debate before an audience of neutral
agnostics (sc. about whether there are other technologically advanced civilizations
within 1000 light-years of us). The debate is between Alienist (male) and Humanist
(female). (Humanists think that Homo sapiens is the only technologically advanced
species within 1000 light-years of the earth—“within the Sphere” I’ll say from now on.)
Humanist has presented the argument from silence, and Alienist must respond. His
response need not be in the form of a refutation of the argument from silence—that is,
he need not show that at least one of its premises is false.27 It would, of course, be nice
(from his point of view) if he could do that, but he need not do it. To respond effectively
to Humanist’s argument, he need only convince the agnostics that, for all they know, at
least one of its premises is false. What will he say? Well, there is a great deal of relevant
and easily available material he can draw upon, for many scientifically literate people
have proposed reasons why we might detect no radio transmissions from extra-solar
civilizations—reasons other than there being no extra-solar civilizations near enough to
us for their radio transmissions to be detectable. Imagine that Alienist offers the
following defense, a defense based on one strand of this voluminous material:
Many advanced civilizations inhabit planets of some of the approximately fifteen million stars
within the Sphere. Many of them have discovered that other such civilizations exist, and commu-
nicate with those neighboring civilizations across intersidereal distances—or with the colonies they
have themselves established on the planets of other stars. But only briefly in the “career” of a
technological civilization does that civilization use radio waves to communicate over large dis-
tances. Technological civilizations typically exist for many millions of years, and typically use radio
waves for communication for only a few hundred years. There is at least one other method—
perhaps there are many—of sending signals between the stars that is vastly more efficient than
radio waves. We have not yet discovered this method—or any of these methods—for it, or they,
depend on a level of understanding of the physical world that we have not yet achieved. When a
civilization achieves a certain level of understanding of the physical world, it abandons radio waves,
just as we abandoned smoke signals and hilltop beacons when we achieved a sufficient under-
standing of the physical world to invent telegraphy. We can no more detect the signals they send by
this method, or these methods, than the Elizabethans could have detected radio waves.
My story, of course, is pure speculation, for obviously we do not know anything about what
could be accomplished by physics we have not yet discovered. But it is plausible speculation, as
a glance at the history of human signaling shows. Advances in physics have frequently led to
the discovery of new and better ways of sending signals, signals it would have been impossible
for anyone to have detected before those advances were made. And we early-twenty-first-
century human beings still have much to learn about physics. We know this because we know
that our physics is radically incomplete. We have no inkling of an understanding of why the

27
Again, we assume that the argument we are considering is formally valid.
   

expansion of the universe is speeding up, for example. And our two finest physical theories, the
general theory of relativity (our theory of gravity) and the “standard theory” of elementary
particles (our theory of everything else) are logically inconsistent with each other.
Here, then, is a defense—a defense that Alienist has presented in an effort to counter
Humanist’s attempt to use the argument from silence to convert the audience of
neutral agnostics to humanists like herself.
But will the effort be successful? Will Alienist’s defense lead the agnostics to
conclude that—for all they know—at least one of Humanist’s premises is false?
(Presumably the defense would lead them to that conclusion with respect to at
most one of Humanist’s premises—‘All or almost all technologically advanced
species that send signals to other stars use radio waves for that purpose’, or some
premise very much like it.) I can say only that it would lead me to that conclusion. (In
the absence of some cogent argument from my creature Humanist that I should not
take the story seriously. And I can’t think of one to put into her mouth.) And I am no
neutral agnostic. I think alienism is almost certainly false—for reasons unrelated to
“cosmic radio silence.” But, of course, an argument whose conclusion is true can have
false premises, and a fortiori, premises that are false for all anyone knows.
Alienist’s defense convinces me—and I believe should convince anyone—that for
all we (all we early-twenty-first-century human beings) know, the proposition
All or almost all technologically advanced species that send signals to other stars use radio
waves for that purpose.

is false. (That is, it convinces me that the falsity of this proposition is a real possibility,
a possibility that is to be taken seriously.28) It doesn’t, of course, convince me (nor
should it convince anyone) that this proposition is false. For all I know, for all any
early-twenty-first-century human being knows, there are many technologically
advanced species (very distant ones, ones too far away for us to “hear” their signals)
who communicate with one another by radio—and none of them uses any other
method because radio is pretty much it: the regions of physics as yet unexplored by
human beings contain nothing that would permit intersidereal signaling that was in
any way preferable to good, old-fashioned electromagnetic radiation. But that’s
beside the point, for—for all we know—the “unexplored regions” do contain some-
thing of exactly that description, and many “nearby” extra-solar species do make use
of it to communicate.
Now Alienist’s defense, you will notice, contains a mystery: it entails the existence
of something that is beyond—far beyond—our present understanding. For all we
know, Alienist’s defense appeals to something that is simply not permitted by the
laws of physics. He certainly—by his own forthright admission—has no proof that,
no argument of any sort for the conclusion that, a means of intersidereal signaling
more efficient that radio waves is physically possible. But physical possibility is not
what Alienist’s defense appeals to: it appeals to epistemic possibility.

28
Unlike this possibility: An extraterrestrial civilization has surrounded the solar system with an
enormous force field that reflects radio waves from extra-solar artificial sources but has no effect on
radio waves from extra-solar natural sources. That might be true. And flaming monkeys might . . .
   

Are we to make anything of the fact that my example of a defense that contains a
mystery contains a mere “physical mystery,” contains something that is, for all we
know, physically impossible, while Theist’s defense (and likewise the defense that
I advance in propria persona to account for “the sufferings of beasts”) appeals to
things that for all we know are metaphysically impossible—and asserts that certain
actions are morally permissible for God, when, for all we know, those actions would
be morally impermissible for any rational being? No, for what counts is only this: that
it is epistemically possible (for us, for anyone in our present epistemic condition) that
these things are actual—or that it is epistemically possible that a morally perfect
being should act as God is represented as acting in the two defenses.
All Theist has to do, all I have to do, is lay these defenses before the agnostics and
contend that they are true for all they, the agnostics, know—and then wait for
Atheist’s rejoinder. For, of course, we must not forget that the defenses are being
presented to the agnostics in the presence of Atheist: that’s an essential part of my
“set-up.”29 Atheist is perfectly free to present arguments to the agnostics intended to
show that they should not regard Theist’s defenses as epistemically possible. That is,
Atheist is perfectly free to offer arguments for the falsity of those defenses. And she is
perfectly free to present arguments for the conclusion that, for one reason or another,
the defenses are simply unbelievable—and thus do not represent real possibilities.30
(Of course, she must do this in the presence of Theist, part of whose dialectical
function is to do everything—provided it is rational and intellectually honest—that
can be done to undermine the plausibility of such arguments as Atheist may present
to the agnostics in her attempts to convince them that they should not regard his
defenses as “true for all they know.”) She is perfectly free to offer arguments for their
metaphysical impossibility or their “moral impossibility,” or to contend that they are
inconsistent with some known facts—the known facts of human evolution, for
example—and thus not “true for all anyone knows.” That’s her dialectical function.
I do not, therefore, think I should have included more in my stories, my defenses,
than I did. But I could have. I included as much as I did and did not include more for
a simple, practical reason. The story, when it included everything I regarded as
essential, had got pretty long—to the point of bordering on unwieldiness. And
I did not want to push it over that border. But I will give an example of something
more I could have included. Toward the end of her paper, Antony asks,
Why think that, once they are removed from the world of horrors, restored to safe and blissful
union with God, that these human beings won’t just do the same damn thing again?

(That is, why think that they won’t rebel against God a second time?) I could have
included something in the story I put in Theist’s mouth to explain why they won’t. In

29
As I implied in the preceding paragraph, I did not put the “second” defense into the mouth of an
imaginary theist; I rather presented it in my own voice. This was a matter of literary convenience. Readers
were invited to imagine it as being presented in the same “debate” format as the first defense.
30
My imaginary defense counsel’s “C.I.A. defense” and the defense suggested in note 28 are examples of
defenses that are prima facie “simply unbelievable.” There are also stories that are not themselves prima
facie “simply unbelievable” but which have demonstrable entailments with that feature. The story of the
Spanish Barber is a trivial example. Antony at one point mentions a very non-trivial example indeed:
Frege’s unrestricted comprehension principle—a “story” about the existence of the extensions of concepts.
   

fact, in another presentation of more or less the same defense (there called a
‘theodicy’—but the word was being used in a carefully defined but non-standard
sense) I included just such a “something”.31 The core idea of that “something” is well
expressed by the proverb, “Once burned, twice shy.” My present point is not to
present that “something” but to point out that Theist is not required to include any
such thing in his defense. All he is required to do is tell stories, and if necessary, to
reply to Atheist’s critiques of his stories, her arguments for the conclusion that the
stories—if they are possible at all—do not represent possibilities that the agnostics
should take seriously. Now it happens that one of the stories he tells includes the
following statement (it is made in different words):
Once they are removed from the world of horrors, restored to safe and blissful union with God,
these human beings won’t just do the same damned thing again.

Of course, while he’s telling the story that includes this statement, Atheist is right
there listening to him tell it. And she’s free to try to turn the idea contained in
Antony’s pointed question into an argument for the falsity or “simple unbelievabil-
ity” of that statement. And, if she does try that, Theist will be right there listening to
that attempt, and will be ready to present the “Once burned, twice shy” rejoinder or
any other rejoinder that may occur to him. Antony writes as if the “debate” aspect of
the interaction between Theist, Atheist, and the neutral agnostics did not exist, as if
that all Theist had to do was first to present a defense to the agnostics—the Extended
Free Will Defense, let us say—and then, having presented it, to sit back in the hope
that they will react to it by saying, “Golly, that story is true for all we know.” But that
is not the point at which he hopes to hear those words, or not the only point. He
hopes to hear the agnostics say, “Golly, that story is true for all we know,” after
Atheist has done her worst—after she has brought all the dialectical pressure to bear
on the Extended Free Will Defense that an ideal representative of atheism can.
In the end, I have to say that I am puzzled by Antony’s demand that a proper
defense would contain more than my proposed defenses do—more details about how
various things that figure in the defenses work, arguments in support of the thesis
that this, that, or the other element in the defenses is metaphysically possible, reasons
for supposing that an action that would be morally impermissible for human beings
is, nevertheless, morally permissible for God . . . . I think the explanation has to be
this: she thinks I was doing something other than what I was doing. I can only
suppose that she thinks I was trying to do at least one of these two things:
To show that the vast amounts of ill-distributed suffering the world contains are consistent
(and not merely logically consistent) with the existence of God—that is, that there are
metaphysically possible worlds in God and “actual evil” co-exist.
To show that it is or at least could be reasonable for someone who was aware of the vast
amounts of ill-distributed suffering the world contains to believe in the existence of God.
Well, I was not trying to show that the existence of God was consistent with existence
of “actual evil.” I wouldn’t know how to do that—although of course I believe that

31
See van Inwagen (1988: 176–7).
   

they’re consistent. In fact, I wouldn’t know how to show that any two things were
consistent (other than by showing that they both existed or were both true, depend-
ing on the nature of the “things”). I know what it is to rebut arguments for the
conclusion that two things are inconsistent, but I’m not at all sure what would count
as showing that two things were consistent. (Formally or logically consistent, yes,
that’s certainly possible—for what it’s worth, and that’s not much. Consistent relative
to something else, yes: it’s certainly possible to show that Riemannian plane geometry
is consistent if Euclidean solid geometry is consistent. But to show that two things are
consistent simpliciter—what could that mean?)
And I was not trying to show that it is or could be reasonable to believe in the
existence of God given our knowledge of “actual suffering.” Again, I wouldn’t know
how to do that. I wouldn’t even know how to prove that belief in the existence of God
was or could be reasonable in a world that contained no suffering at all. (Of course,
I believe that belief in the existence of God is reasonable—and reasonable even in this
sorry world—, but I wouldn’t know how to show that it was reasonable or even that it
is reasonable for some beings in some metaphysically possible world containing
suffering comparable to that of the actual world.)
I had thought I had made it clear that I had only this goal: to defend the thesis that
the argument from evil has premises that are, for all anyone knows, false. That’s a
much weaker thesis than either ‘It is metaphysically possible for God and “actual evil”
to co-exist’ or ‘It could be reasonable for someone who was aware of the existence of
“actual evil” to believe in the existence of God’. Nevertheless, that conclusion is not
without interest. The argument from evil is an important philosophical argument—
one of the most important philosophical arguments. And if, for all anyone knows, at
least one of its premises is false, then no one should regard it as a cogent argument for
its conclusion.32
I would also remind Antony that my intention was not to convince someone like
her—an atheist; if I may so express myself, a hardened atheist—that at least one of the
premises of the argument from evil is false for all she knows. I’m not up to that task.
But it might be easier for one to convince neutral agnostics that a premise of the
argument was false for all they knew (and to convince them of this in the presence of
someone like her, someone whose dialectical function it was to call their attention to
every debatable point in one’s arguments) than it would be for one to convince
Antony that that premise was false for all she knew.33

32
And yet at one point at least Antony does seem to understand that this was the way I was arguing. See
her footnote 4. My project in The Problem of Evil was to attempt to convince my readers that the argument
from evil was a “failure” in the precise technical sense that I gave to that word.
33
At one point (p. 182) Antony quotes a passage from The Problem of Evil and says of it, “I quoted this
passage in full because I find it stunning.” (That is to say, she finds it a stunningly inept dialectical lacuna in
Theist’s presentation of his defense.) Now I don’t think she should have found it “stunning,” and I think
that the reasons she proceeds to give for regarding it as stunning are specious—in fact, stunningly so.
My point here, however, is that it’s a matter of indifference to me and to my mouthpiece Theist whether she
finds this passage stunning. What Theist and I care about is whether the audience of agnostics will find it
stunning—that is to say, we care about how the agnostics will regard the passage after listening to an
extended debate between Theist and Atheist about whether it is a stunningly inept (etc.). I am confident
that they will regard Theist as having “won” this debate—although, quite possibly, Atheist will be
convinced that she has won it.
   

I would remind her, finally, that in my view, whatever else it is, the argument from
evil is a philosophical argument—and, in one important respect, at least, a typical
philosophical argument. There is, in my view, no philosophical argument for any
substantive, positive34 conclusion that would fare well in a debate before an audience of
neutral agnostics (neutral agnostics with respect to its conclusion). I might, for
example, debate with an able compatibilist (let’s suppose it was David Lewis) about
the compatibility of free will and determinism before an audience of neutral agnostics
(agnostics about whether free will and determinism are compatible). At one point in
the debate, I might propose to convert the agnostics to incompatibilism by presenting
the Consequence Argument. But I am very close to being certain that the result of an
exchange in which Lewis and I argued about the premises of the Consequence
Argument till the cows came home would be this: in the end, he would be able to
convince the agnostics that the argument had at least one premise that was, for all they
knew, false. He wouldn’t be able to convince me—a hardened incompatibilist—of this
(even David Lewis wasn’t up to that task) but he would be able to convince them.35

17.4 Reply to Alex Rosenberg

There are several points in Alex Rosenberg’s description of the content of “Weak
Darwinism”36 that I’d like to dispute. But nothing a scholar can write is more boring
than a detailed, paragraph-by-paragraph documentation of the thesis that that
scholar has been misrepresented by a critic. I choose instead to discuss two substan-
tive arguments in Rosenberg’s paper, arguments that do not depend on the way
Rosenberg has represented my views—arguments that, if they are cogent, refute what
I would concede are the central theses of “Weak Darwinism.” I will first discuss
Rosenberg’s argument for the conclusion that “Allism” is supported by the second
law of thermodynamics. I will then turn to his argument for the conclusion that one
of my arguments—the argument for the compatibility of Darwinism and theism—is
flawed.

17.4.1 Allism and the Second Law of Thermodynamics

Allism, Rosenberg contends, is supported not so much by the data of biology and
paleontology (although data from those sciences certainly support a thesis I explicitly
affirmed, namely that natural selection must play an important and essential role in
any explanation of the features of life on the earth, either as it is at present or as it has
been at any point in its long history) as by physical theory: by the second law of
thermodynamics.
Rosenberg’s argument—if I understand it—proceeds from two premises: the thesis
of a “zero adaptation starting point,” and the thesis that every episode in which some

34
For a defense of this position and an account of ‘substantive’ and ‘positive’ (and a defense of the thesis
that atheism is a “positive” thesis), see van Inwagen (2009a), especially note 3, p. 17.
35
“Are you saying, then, that the Consequence Argument, is not a ‘cogent’ argument for incompati-
bilism? That it is a ‘failure’? If so, why are you an incompatibilist?” Well . . . it’s complicated. See the essay
referred to in the previous note and my replies to the papers in Part IV (“Method”) of the present volume.
36
Van Inwagen (2009b).
   

type of organism becomes better adapted to its environment must (at least to a very
high probability—a qualification I will ignore in the sequel) take place in accordance
with the second law.
I will discuss the second premise first. I would not dream of disputing it. I agree
that every aspect of the biological history of the earth—whether it involves adaptation
or not—has happened in accordance with the second law. The surface of the earth, its
atmosphere, and its oceans contain high local concentrations of order (Rosenberg
and I are two of them and a bacterium is another), but that order was not created ex
nihilo: it was “paid for” by a vast reduction in the order inherent in the physical
content of other places, such as the solar core. When the earth was lifeless—when
there were none of these organisms, these high local concentrations of order, on its
surface or in its atmosphere and seas—, the earth and sun taken together contained
vastly more order than they do now (most of it distributed pretty uniformly through-
out the solar core, among as-yet-unfused nuclei under immense pressure). Or
consider an automated factory, which builds, let us say, widgets out of materials
available in its immediate environment. Each individual widget may contain more
order that can be found in any widget-sized region of space within the factory or
within the local deposits of raw materials it draws upon; nonetheless, the total
amount of order in the factory, its sources of energy, the remaining deposits of raw
materials, and the stock of produced widgets, not only does not increase when a new
widget is produced but must decrease every time one is produced. (The “lost” order
will be dispersed over an eternally expanding volume of space in the form of thermal
radiation—heat produced, for example, when one of the factory’s automated lathes
breaks molecular bonds in a chunk of raw material, heat that the factory is unable to
utilize to power its operations.)
But why is it that any explanation of the high local concentrations of order in the
biological world (whether in terms of adaptation or not) must appeal to no order-
concentrating mechanism but natural selection? Natural selection can certainly
concentrate order, but there are other mechanisms by which order that is dispersed
over some region can be concentrated or “pumped into” a smaller region, and these
other mechanisms do not violate the second law. We human beings, for example,
often produce high local concentrations of order by designing and building things, by
applying foresight and knowledge and intelligence, and we are not miracle workers:
the high local concentration of order in a stack of firewood or in a twenty-dollar bill
or in an ampule of Botox has been paid for by the loss of a greater amount of order
somewhere in the world. I do not say this to suggest that “intelligent design” has been
at work in the history of life on the earth. I am rather pointing out that any natural
process that produces high local concentrations of order—even intelligent design—
will have to work within the constraints of the second law.
Let me tell you a story (the genre is science fiction). It is 100,000 years in the future,
and human beings have become like gods (albeit they are only what the sci-fi writer
Charles Stross has called “weakly godlike”: they cannot violate the laws of physics).
Our remote descendants of those days change the trajectories of stars at a whim and
design and implement entire biospheres for hitherto lifeless planets. And the bio-
spheres they design, while natural selection inevitably plays a significant role in
them, employ other mechanisms of—well, let us not say evolution but biological
   

development. Inherent in each biosphere there is a “biopilot,” a sort of computer.

(The physical substrate of this “computer” is entirely biological, the biosphere as a
whole—its operations are “distributed” throughout the totality of living things. The
biopilot exists as an aspect of the biosphere: it is a virtual and not a “real” entity.) Its
function is to search out ecological niches that are unoccupied or only marginally or
inefficiently occupied and to design taxa to occupy them. And, being a part of the
physical world—if only a virtual one—the biopilot is able to bring its designs to
reality. It “observes,” for example, that the niche “apex predator” in a certain
rainforest is occupied by an organism that is not exploiting the opportunities that
niche provides very efficiently, designs an optimum apex predator, works its magical
genetic engineering—“Any sufficiently advanced technology is indistinguishable
from magic”—, and within a few score generations, the old apex predator either
conforms to the biopilot’s design or has become extinct and has been replaced by a
new species that does. (Unaided natural selection does things like that, too, of
course—but much, much slower. That is precisely why our godlike descendants
build biopilots into their designer biospheres.) It seems to me that a species that
can control the behavior of stars might well be able to design and implement
biospheres with such virtual pilots. “Allism” would of course be false as a thesis
about such biospheres—and yet there would be nowhere within them any violation
of the second law: every local decrease of entropy brought about by the actions of the
biopilot would, must, be paid for, and more than paid for, by an increase in the
entropy of other parts of the universe (mainly in the core of the primary of the planet
“inhabited” by the biopilot and the organisms in its charge). The biopilot is a
designer, to be sure, but not an intelligent designer, for it is no more intelligent
than Watson (or, more exactly, than the DeepQA computer architecture).
“But, the biopilot is a product of intelligent design—as was the DeepQA architec-
ture. So its productions would be a result of intelligent design—at one remove.”
True—although that doesn’t touch the point that the biopilot operates within the
constraints of the second law—, but who’s to say that natural selection couldn’t at
some point produce something like the biopilot? (This would be something like the
“Gaia hypothesis,” although, I concede, considerably stronger.) After all—I know
that you, the interlocutor, will not dispute this statement—natural selection pro-
duced us, produced intelligence. In my sci-fi story, therefore, natural selection
produced the biopilot at, to use your phrase, one remove. Are you sure that natural
selection couldn’t, as it were, eliminate the middleman and produce something like
the biopilot without “going through” intelligence? If natural selection can indeed
produce intelligence, I’d be very cautious about saying what biological objects or
phenomena (consistent with the laws of physics) natural selection can’t produce. In
“Weak Darwinism,” I suggested that, for all anyone knows, other biological mech-
anisms than natural selection may have played a central role in the story of how life
on the earth got into its present state. In saying this I explicitly left open the
possibility that at one time natural selection was the only such mechanism and
that “additional” mechanisms as there may be came to be solely as a result of the
operations of natural selection. But if that has happened, Allism is false, for Allism is
not simply the thesis that natural selection is source of all other mechanisms that have
operated to produce such things as taxonomic diversification, biological complexity,
   

and apparent teleology in nature: it is the thesis that natural selection is the only such
mechanism there has ever been.
I suppose I should explicitly say that I am not putting forward the hypothesis that
something like a biopilot has actually played a role in the history of terrestrial life as a
real possibility, one we should assign a significant credence to. The biopilot-
produced-by-natural-selection is nothing more than a philosopher’s counterexample
to another philosopher’s contention that a certain proposition is a necessary truth: if
there were a natural-selection-produced biopilot, Allism would be false and both
Rosenberg’s premises would be true.) The intent of my story has been to demonstrate
that it is not evident that there is any inherent impossibility in the idea of other
mechanisms than natural selection that both figure in the explanation of taxonomic
diversification (etc.) and are consistent with the second law of thermodynamics. If
there were such mechanisms, I could hardly be expected to produce a plausible story
about what they might be—any more than Lord Kelvin’s contemporaries could have
been expected to produce a plausible story about what mechanisms might have
enabled the sun to shine for many hundreds of millions of years.
I contend, therefore, that Rosenberg’s conclusion does not follow from his
premises.
17.4.2 The Consistency of Darwinism and Theism
Rosenberg’s critique of my argument for the consistency of Darwinism and theism
makes extensive use of the concept of objective chance. I find his use of this concept
extremely puzzling.
Here is one of the questions that puzzles me: is the existence of “objective chance”
consistent with strict causal determinism? If it is not, then Darwinism entails causal
indeterminism, which is a very implausible idea. Adherence to Laplacian determin-
ism was widespread in the nineteenth century, and no one in those days saw
Laplacian determinism as an objection to Darwin’s theory. Computer programs
that mimic and illustrate the operations of natural selection are common enough,
and these programs are entirely deterministic: the only source of “objective chance”
in a particular application of one of these programs is the filling in of open param-
eters in the program by the human beings making the application—the fixing of
initial conditions, as it were. In any case, it is not the business of a biological theory to
pronounce on the question of determinism. Whether the world is deterministic may
be a question for physicists or it may be a question for metaphysicians. One could
argue about that. But it is certainly not a question to be answered by biologists.
Suppose then that the existence of objective chance (in whatever sense Darwinism
requires “objective chance”) is consistent with determinism. And suppose that Allism
is true and that the universe is deterministic. (Let us say that it is deterministic in the
sense imagined by Laplace, just to have a version of determinism to play with.) Then
an omnipotent God can easily bring it about that natural selection produces any
given organisms that it is possible for natural selection to produce. And, of course, the
way he can do this is by an exquisitely precise choice of initial conditions. Suppose,
for example, that God wanted natural selection eventually to produce human beings.
Well—so the Darwinists say—, natural selection did produce human beings. If they
are right, therefore, all God had to do to achieve his end was to create an array of
   

particles with certain precisely specified relative positions and momenta—these being
the positions and momenta those particles actually had at some point in time before
there were any living things. Being omnipotent, he would able to do that. Being
omniscient, he would know that if he created such an array, the physical universe
would after such and such an interval evolve (in the physicists’, not the biologists’
sense of ‘evolve’) into a state that included the existence of human beings (and, for
that matter, Darwinilus sedarisi and the Taj Mahal).
I have not been able to see the force of Rosenberg’s objection to my little argument
for the compatibility of theism and Darwinism.37 I will therefore present a slightly
more elaborate version of the argument and ask him what his objection to this
version of the argument would be:
God is often said to be achronic or timeless or “outside time.” (This was the position of Augustine
and Anselm and Thomas Aquinas and many other great Christian philosopher-theologians.) Let
us suppose that this doctrine is correct. Consider the timeless God contemplating (timelessly) the
creation of a cosmos. Since he is omniscient, all possible distributions of matter and radiation in
space-time are present to his mind. Consider the one among them that is displayed by the actual
cosmos. God chooses that one and says, “Let it be!” or “Let there be a cosmos that consists of
matter and radiation distributed in space-time in that way!” And, by that act, a complete four-
dimensional whole—from the Big Bang to, well, whatever—timelessly is.
Now if Darwinism is both a true theory and a scientific theory, its truth must have been
established by this timeless act of creation. The truth of a scientific theory must “supervene on”
(must be “settled by”) the distribution of matter and radiation in space-time. A theory whose
truth or falsity does not supervene on the distribution of matter and radiation in space-time is
not called a “scientific” theory. Such a theory is called a “metaphysical” theory.
I have told a story in which an omnipotent being has created a Darwinian world. At any rate,
those who accept the Darwinian theory should agree that I have told a story with that feature,
since they believe that Darwinism is both a true theory and a scientific theory, and every
(actually) true scientific theory must be true in any cosmos in which matter and radiation are
arranged as they are in the cosmos we inhabit.

(A few years ago, I presented this argument in a sort of symposium on science and
religion in Rome. One of my fellow symposiasts was the mathematical biologist
Martin A. Nowak, who is Professor of Biology and Mathematics at Harvard, and
Director of the Program for Evolutionary Dynamics. Professor Nowak and I had not
seen each other’s papers before they were presented, and we were amused—and I, at
least, was gratified—when it turned out that our two papers presented essentially the
same argument for the compatibility of Darwinism and theism.) I should say that
I do not think the above “four-dimensionalist” story is actually true. (For one thing, it
seems to me to be inconsistent with human free will.) But I do not put it forward as
true. I put it forward as a logically consistent story in which the physical universe has

37
It is something like this: according to Darwinism, the history of life has been a radically contingent
process; and no one, not even an omnipotent being, can control the outcome of a radically contingent
process. I have examined this kind of argument in great detail in van Inwagen (2003). See especially
pp. 356–61.
   

been created by God and which incorporates the truth of any scientific theory that is
true in reality. That the story is logically consistent does not prove that it
is metaphysically possible, of course, but the story is metaphysically possible if it is
metaphysically possible for there to be a timeless omnipotent being and metaphys-
ically possible for Allism to be both a scientific theory and true. And if the story is
possible, its possibility implies that God can create a world—it need not be governed
by deterministic laws—in which human beings are the product of natural selection.
And does “objective chance” exist in this world? Well, it does if the existence of
objective chance is necessitated by the way matter and radiation are distributed in the
space-time of the actual world. If the existence of objective chance is necessitated by
the way matter and radiation are actually distributed in space-time, then God can
create a world that contains all the objective chance any scientific theory that applies
to the actual world needs. If it is not so necessitated, then either (a) objective chance
does not exist and ‘Objective chance exists’ is not a component of any true theory, or
(b) ‘Objective chance exists’ is a metaphysical proposition—owing to the fact that its
truth-value does not supervene on the distribution of matter and radiation in space-
time—and should not be a component of any scientific theory.

17.5 Reply to David J. Chalmers

I am largely in agreement with David Chalmers’s fine paper. But I wish to say a few
things about some of his statements about my own views (or, in one case, Colin
McGinn’s views)—to set the record straight, as it were. I shall quote some passages
from the paper and comment on them. None of the things I say is of very
great relevance to any of the main theses of the paper. Each quoted passage from
Chalmers’s paper (after the first) marks a transition from a discussion of the
previously discussed passage to a discussion of the “new” passage.

17.5.1 Disagreement and Knowledge

I begin with:
A strong version of this view, suggested by van Inwagen’s discussion, is that where there is
sufficient disagreement among experts, no individuals can be said to know the truth. Even if
some individuals have hit on good arguments for true conclusions, how can they have justified
confidence that these are good arguments, when so many of their peers disagree? I am not so
sure: I think that at least in some cases, a good argument can ground an individual’s knowledge of
a conclusion even when peers reject it. For example, I think that the presence of any number of
peers who deny the existence of consciousness would not undermine my knowledge that I am
conscious. Likewise, it would not undermine arguments that take this claim as a premise. (p. 285)

Whether that view is suggested by my discussion or not, I will point out that I have
presented explicit arguments for a position that might be loosely expressed in these
words: ‘Where there is sufficient disagreement among philosophers, no individual
philosopher can be said to know the truth’.38 But that position applied only to

38
See van Inwagen (2009a).
   

disagreement about philosophical theses or beliefs that satisfied the following con-
dition: the theses must not be theses are accepted
. . . by almost all human beings. I shall not be concerned with philosophical theses that have
been accepted by all sane non-philosophers and have been denied only by a few philosophers—
generally practitioners of “revisionary metaphysics.” I assume that there are such philosophical
beliefs because I assume that the denial of a philosophical belief is itself a philosophical belief,
and many philosophers have believed things (in, as it were, their professional capacity) that
almost everyone—even most philosophers—would deny. Or so it seems at least plausible to
maintain. Plausible examples of things that fall into this category would be: “Change and
motion are not real features of the world”; “One has no reason to suppose that there are minds
other than one’s own”; “There are no material objects.” . . . Thus, philosophical beliefs like
“Change and motion are real features of the world,” “One does have reason to suppose that
there are minds other than one’s own,” and “There are material objects” do not satisfy my
second condition.39

Let us call philosophical propositions that are not “accepted by almost all human
beings” “controversial philosophical propositions.”40 The thesis I defended was (it
was not stated in these words; I have rewritten it to make it as verbally similar to
‘Where there is sufficient disagreement among experts, no individuals can be said to
know the truth’ as possible):
If there is extensive disagreement among philosophers41 about whether a certain controversial
philosophical proposition p is true (assuming that all philosophers who accept either p or its
denial are aware of and have a perfect understanding of all the arguments that any of them has
given that are relevant to the question the truth or falsity of p42), then it is irrational for any
philosopher to accept p.

Call this thesis the Philosophical Disagreement Thesis or PDT. (If PDT is true, then
I believe many, many things it is irrational for me to believe—and I expect you do,
too. My only defense is of the “ought implies can” variety: I can’t help it; I can’t help
believing things that seem to me to be obviously true.)
Now let us ask, is ‘Consciousness exists’ a counterexample to PDT? Before we can
address this question, we must answer a prior question. Suppose a philosopher
uttered the sentence ‘Consciousness does not exist’ and intended thereby to be stating
a philosophical thesis. What would that thesis be—what would it imply? The only
answer to this question that comes to my mind is that it would imply that all the
following statements (and many millions of statements of the same sort) were false:

39
Van Inwagen (2009a: 16–17).
40
This definition may need a little Chisholming. Some philosophical propositions, even some quite
trivial ones, might be said not to be “accepted by almost all human beings” simply because they involved
concepts that a significant proportion of human beings do not have—‘Eliminative materialism is incom-
patible with reductive materialism’, for example.
41
Generalizations about “philosophers” are generalizations about human philosophers unless other-
wise noted—and we shall be discussing some (fictional) non-human philosophers.
42
This is a broader class of arguments than the class of arguments (that have been given by philo-
sophers) whose conclusion is p or its denial. Suppose, for example, that A is an argument (given by
philosophers) whose conclusion is p, and that A0 is an argument (given . . . ) whose conclusion is the denial
of one of the premises of A (or whose conclusion is that A is invalid or that it assumes the point at issue . . . ).
Then A0 is an argument that is “relevant to the question of the truth or falsity of p.”
   

Lots of people suffer from migraine—and migraines are extremely painful.

Sometimes people have to make difficult and unpleasant decisions—but they do often some-
how manage to make them.
There are now [i.e., in the summer of 2015] people who believe that Hillary Clinton will almost
certainly be the next President of the United States.
Some couples who want to adopt a child are unable to.

At any rate, I am unable to see why a thesis would deserve to be called ‘the thesis that
consciousness does not exist’ if it didn’t imply that those statements (and perhaps
90 percent of the statements that human beings make in the course of conducting the
business of their lives) were false—and, of course, if it didn’t imply that their denials
were true.
Is ‘Consciousness exists’, so understood, a controversial philosophical propos-
ition? I’m willing to stipulate (as the lawyers say) that it’s a philosophical proposition.
But it’s certainly not “controversial”: it’s certainly held by almost all human beings—
at least tacitly or implicitly. That is to say, almost all human beings accept, if not
precisely the four statements given above, millions upon millions of statements of
much the same kind. So even if a given “man on the Clapham omnibus” has never
considered the proposition ‘Consciousness exists’ he can certainly be said to accept it
in the same sense as that in which he can be said to accept ‘Fish do not wear shoes’ or
‘Butter is not a high explosive’. And, I expect, what goes for the passengers on the
Clapham omnibus goes for most philosophers. I very much doubt whether very
many philosophers accept any proposition that implies that the above four state-
ments are false. And I am convinced of the truth of a rather stronger statement: while
there may well be possible worlds in which a sizable proportion of philosophers
accept such a thesis, such worlds are very distant from the actual world. (And I am
convinced that such worlds are not only distant but rare—that is, that the objective
probability of philosophy’s being is such a state, while greater than 0, is very low.)
Now let’s suppose that the stronger statement is true. Let w be such a distant (and
improbable43) world: although the medical and political (and so on) features of w are
much like those of the actual world, the condition of philosophy in w is very different:
w is a world in which a sizable proportion of philosophers accept a thesis that they
express as ‘Consciousness does not exist’—words that, in their mouths, have a meaning
that entails the falsity of our four statements. I am certainly willing to concede that if
David Chalmers exists in w (and if his “doxastic state” in w is as similar to his actual
doxastic state as the bizarre condition of philosophy in w allows44) he knows (in w) that
“consciousness exits.” (And his belief that “consciousness exists” is rational—in case
that doesn’t follow from his knowing that “consciousness exists.”) Nevertheless,
‘Consciousness exists’ is not a counterexample to PDT (in w), owing to the fact that,
in w as in the actual world, that proposition is accepted by almost all human beings.

43
This is loose talk; strictly speaking every world is of 0 (or perhaps infinitesimal) probability; what is
“improbable” is not w itself but, rather, that a world having the features ascribed to w (there are infinitely
many) should be the one that is actual.
44
I mean this to imply that in w he believes that consciousness exists; and, moreover, in w, his beliefs on
this matter are consistent: he does not also believe that consciousness does not exist.
   

Are there possible worlds in which ‘Consciousness exists’ is not accepted by almost
all human beings? I very much doubt whether there’s a world that has that feature and
contains “extensive disagreement among philosophers” about whether consciousness
exists. (A “Zombie” world would not satisfy both these requirements—for, although in
a Zombie world no proposition is accepted by any human being,45 by that very token
there will be no such thing as extensive disagreement among philosophers in a
Zombie world.) But even if I’m wrong about that and there is an “oddfolk-madphil-
saneDave” world—a world in which ‘Consciousness does not exist’ is consistent with
all the beliefs of the Clapham-omnibus passengers, in which a significant proportion
of philosophers accept that proposition, and in which Chalmers (nevertheless) knows
that consciousness exists—, an oddfolk world (whether or not it was also a “madphil”
and “saneDave” world) would be vastly, radically different from the actual world. And
it would be vastly more different from the actual world than a “mere madphil”
world—a world in which, although the folk, almost to a person, tacitly accept
‘Consciousness exists’, a goodly proportion of philosophers deny it. If an oddfolk-
madphil-saneDave world exists, its existence implies that PDT is not a necessary
truth. PDT is, nevertheless, false only in worlds that are vastly, radically different from
the actual world. In that respect it could perhaps be compared with ‘If human beings
exhibit the kinds of complex linguistic and non-linguistic behavior they exhibit in
actuality, they are conscious’: even if it is not a necessary truth, it is true and its truth is
firmly grounded in the Nature of Things as They Are.

17.5.2 Establishing Lack of Success

Does this mean that all philosophical arguments for positive theses are unsuccessful, as van
Inwagen . . . has suggested? . . . This depends on what one means by success. If one defines
success in sociological terms, so that success requires convincing almost everyone in a
community, then we have seen that at best very few philosophical arguments for positive
theses have been successful in our community. Van Inwagen defines success in idealized
epistemological terms: a successful argument for a proposition p is one that would convince
an audience of ideal reasoners who are initially agnostic concerning p, in the presence of an
ideal opponent of p. I do not think that the sociological observations above (or the sociological
observations that van Inwagen appeals to) come close to establishing that no philosophical
arguments are successful in that sense. (p. 289)
Well, of course my arguments don’t establish that conclusion. Any argument whose
conclusion is a philosophical thesis is a philosophical argument. (At any rate, I don’t see
any other way to understand ‘philosophical argument’.) Say that an argument for a
proposition p that “would convince an audience of ideal reasoners who are initially agnostic
concerning p, in the presence of an ideal opponent of p” is ‘successfulPvI’. The thesis
Neg No philosophical argument is successfulPvI
is a philosophical thesis. If some argument did establish the conclusion that no
philosophical arguments are successfulPvI—did establish the truth of Neg—, then
Neg would be false!

45
Let’s not quibble about words: I stipulate that Zombies would be human beings in the sense of
‘human being’ that these words have in the definition of ‘controversial philosophical proposition’.
   

And while I didn’t explicitly say that no philosophical argument “comes close” to
establishing its conclusion, I believe that, too. (And, anyway, suppose that some
philosophical argument “came close” to establishing its conclusion—that human
beings have free will, let’s say. It’s at least a defensible position that that argument
would establish the conclusion that human beings probably have free will—and isn’t
‘Human beings probably have free will’ a philosophical thesis?) I will say that I regard
my arguments for Neg as pretty good ones—as philosophical arguments go. They are,
I would say, no worse than the arguments of Kripkenstein or Frank Jackson’s “Mary
the color scientist” argument or one of the more sophisticated versions of the
cosmological argument. But, despite being pretty good as philosophical arguments
go, those three arguments don’t establish their conclusions—nor, sadly, do they come
close to establishing their conclusions. (“But Jackson’s argument not only comes
close to establishing its conclusion, it does establish its conclusion.” Really? If so, why
does Dan Dennett think it’s a tissue of fallacies? If it establishes its conclusion, what
are we going to say about poor Dan? What is he? Stupid? Philosophically inept?
Intellectually dishonest?)

17.5.3 Explaining Lack of Success

Evolutionary explanations: It is sometimes suggested (e.g. by McGinn 1993) that there is a
Darwinian explanation for the lack of progress in philosophy. The rough idea is that we did not
evolve to be good at philosophy, since in the evolutionary environment there were no selection
pressures that favored philosophical ability or anything that strongly correlates with it. Perhaps
there is something to this, though it would take some work to explain why the same does not
apply to the ability to do abstract mathematics or highly theoretical science. (p. 295)
McGinn’s position46 is that that the same does apply to our ability to do abstract
mathematics and highly theoretical science. That we (as a species, if not in every case
as individuals) happen to be good at these things and bad at philosophy has no
evolutionary explanation. It’s just one of those things that have happened in the
course of biological history—like the pentadactyl limb. (A species can’t be good at
everything. As Chomsky says somewhere, to be equally good at everything would be
to be very bad at everything.) We may one day discover among the stars a species
that, for no reason having to do with selection pressure or reproductive advantage, is
good at philosophy.47 And we may also imagine that this species is bad at physics. (At
any rate, the philosophers of this alien species all agree about the answers to
philosophical questions, or come as close to agreeing as human scientists do to
agreeing about the answers to scientific questions. The alien philosophers are—
they tell us—puzzled and confounded by the inability of human philosophers simply
to see that Zombies are metaphysically impossible and that free will is compatible

46
See McGinn (1993). See also van Inwagen (1996), a review of McGinn (1993).
47
The idea of such a species first turned up in some speculations of Noam Chomsky and was further
developed by McGinn. For a “real” science-fiction story (of course unrelated to the writings of Chomsky
and McGinn; it was first published in 1984) that features such a species, or at least a species that claims to
know the answers to all philosophical questions, see George Alec Effinger, “The Aliens Who Knew, I Mean,
Everything.” The story—originally published in The Magazine of Fantasy and Science Fiction—can easily be
found on line by Googling its title.
   

with determinism. And when the alien philosophers present what they regard as
decisive arguments for such conclusions to the human philosophers, the humans
can’t follow them: every time an alien philosopher presents human philosophers with
a careful, step-by-step presentation of one of the aliens’ supposedly decisive argu-
ments, at some crucial point in presentation of the argument, the human philo-
sophers confess that—to borrow a useful term from the Lexicon48—the reasoning
they are being asked to follow goes all wilfrid in their minds. And the aliens, for their
part, very soon become confused and unable to understand what we’re saying when
we try to explain rainbows or planetary orbits to them.)
[Van Inwagen has] advocated unknowability: humans are just not smart enough to answer the
big questions. The idea is that there is some level of intelligence or aptitude that would suffice
to answer these questions, but that humans fall below that level.
Van Inwagen argues for this conclusion as follows. He suggests that it is implausible that we are
much above that level, given the lack of progress to date, and that it is antecedently improbable
that we should be just barely at that level. So it is much more likely that the level lies above us.
I am not so sure about this argument. I think we already know that for a vast range of
questions, humans are just barely at the level for doing them well: scientific and mathematical
questions, for example. Because of this, it is arguable that we lie at a special intelligence
threshold at which an extraordinarily wide range of questions come to be within our grasp
over time. It is not obvious whether or not philosophical questions fall within that range, but it
is not obviously more likely that they do not than that they do. (p. 297)

This misrepresents the conclusion of my argument. The conclusion of the argument

is not that we are “not smart enough to answer the big questions.” (The argument
pertained only to metaphysics—it occurred in the “Concluding Meditation” of an
introductory textbook of metaphysics49—, but I am willing to say that a parallel
argument would apply to all questions of the form ‘Is the case that p?’, where p is one
of those “controversial philosophical propositions” I spoke of earlier.) The conclu-
sion of the argument was closer to this statement.
If it turns out that we human beings are not smart enough to answer the big questions (sc. of
metaphysics), we should not find that surprising.
It was in fact this statement:
. . . if metaphysics does indeed present us with mysteries we are incapable of penetrating, that
fact is not itself mysterious. (p. 289)
(Suppose that Jane is about to draw a card at random from a standard deck.
Obviously the two statements ‘The card Jane will draw is the three of clubs’ and ‘If
it turns out that the card Jane draws is the three of clubs, we should not find that
surprising’ are not equivalent.50) And I would say the same thing with respect to

48 49
Dennett and Steglich-Petersen (2008). Van Inwagen (2015: Chapter 13).
50
The following case is more closely parallel to the case we are considering: Jane must draw a card, and
has been informed that she will be shot if the card she draws is neither a red card nor a face card. The
following two statements are not equivalent: ‘The card Jane draws will be either a red card or a face card’: ‘If
the card Jane draws is either a red card or a face card, we should not find that surprising’. (We should not
find that outcome surprising because its probability is about 0.615.)
   

physics: if physics presents us with mysteries we are incapable of penetrating, we

should not find that mysterious—not, at any rate, if we know that we are just able to
get as far in physics as we have. Suppose, for example, that the nature of “dark
energy” is a mystery—in the sense that we human beings simply do not, as a species,
have within us cognitive resources that would enable us to discover its nature.
(If physicists and cosmologists actually manage to refer to something when they
use the phrase ‘dark energy’, it must be that “something” has a nature.) Suppose we
one day encounter another intelligent species, the Spicans, and discover that they are
demonstrably less good at physics than ourselves, although only a bit less good. The
Spicans had a pretty good analogue of Newtonian mechanics before we first visited
them, for example, and a sort of gesture in the direction of special relativity, but they
did not have not have anything resembling general relativity. When we teach general
relativity to their physicists, they can learn it, but we’re convinced, and they them-
selves sadly admit, that they’d never have figured it out for themselves: no member of
their species (we believe and they concede) would ever have been able to do what
Einstein did (and what Hilbert would almost certainly have done in 1915 or 1916 if
Einstein hadn’t done it first). Suppose we’re all agreed—on the basis of these facts
about the Spicans’ scientific abilities—that the Spicans (now that we’ve informed
them of the cosmological phenomenon that dark energy has been postulated to
account for) are unable, biologically unable, to discover the nature of dark energy.
Then, I contend, it would be reasonable for us to conclude that if the nature of dark
energy is a mystery (to us), there’s nothing mysterious about its being a mystery. If
the Spicans, who are only a bit less good at physics than ourselves, are unable to
discover its nature, then, if we are able to discover its nature, we’re only just barely
able to discover its nature. (As some of us are just barely able to run a mile in four
minutes—unlike horses and Houyhnhnms, who are able to run a mile in
four minutes and are not just barely able to run a mile in four minutes.) And that
seems rather improbable. Why should we be “just over the line”? So there’s a
reasonable probability that we’re not able to discover the nature of dark energy—a
probability high enough that we shouldn’t be surprised if we’re not able to discover
its nature.51 If moreover, we are able to penetrate all the mysteries—or what had at
some point in the history of science seemed to be mysteries—of physics, we should
say (on the basis of our knowledge of the Spicans)

51
An analogy. Suppose I’ve been set down at a random place on a roughly circular island about 500
meters in diameter. I know that a roughly circular region that occupies about a third of the island is an
unmarked minefield. Naturally, I’d like to know whether I’m inside the minefield—but I don’t know, and
have no way of finding out, whether I’m inside it. I do, however, somehow know that a boulder that is only
ten meters from me is inside the minefield. I reason as follows: if I’m outside the minefield, I’m just barely
outside it. And that’s rather improbable—it’s rather improbable, given the size of the island and the size of
the minefield, that my being randomly placed on the island should have resulted in my being outside the
minefield but within ten meters of its boundary. (If we assume that every point in the minefield is more
than ten meters from the water’s edge, a simple geometrical calculation gives a probability of 0.86 that I am
in the minefield, conditional on my being ten meters from an object in the minefield. Of course, in the
“dark energy” case, no calculation of the numerical value of the probability of our being unable to discover
its nature, conditional on the Spicans being unable to discover it, is possible.) An oracle then unexpectedly
resolves the matter for me: I am indeed inside the minefield. I ought not to find this revelation particularly
surprising, given what I already knew before the oracle spoke.
   

If physics had presented us with mysteries we were incapable of penetrating, that fact would
not have been mysterious.

As I said in Metaphysics, commenting on an argument similar to the argument I have

just presented,
What the argument . . . suggests is that if we were unable to explain [rainbows and planetary
orbits and such], this would not be mysterious. The success of physical science does not suggest
that the inability of metaphysics to achieve the same sort of success is a mystery. If anything in
this area is a mystery, it is the fact that science is a success. (p. 290, n 3)

I have said that I did not claim to have shown that we are unable to solve the problems
of metaphysics. I will add that I did not claim to have shown that we are probably
unable to solve the problems of metaphysics. My conclusion was rather this: that the
considerations I appealed to showed that the probability that we are unable to solve the
problems of metaphysics is high enough that, if we somehow come to learn that we are
unable to solve them, we should not find that particularly surprising.

17.6 Reply to John A. Keller

I do not have the space to discuss everything John Keller says in his very rich and
closely argued paper. I am going to do only two things: I will defend the criterion of
philosophical success and failure I have proposed in various places (but primarily in
Lecture 3 of The Problem of Evil 52), and I will say something about the alternative
criterion that Keller proposes. I will proceed mainly by quoting some passages from
his paper and commenting on them.

17.6.1 Analyzing Success

I begin with these two passages:
. . . van Inwagen’s is not the most liberal possible criterion. Philosophical individualism—
roughly, the view that an argument A for conclusion c is successful for individual i if and only if
A is convincing to i (regardless of i’s previous attitude towards c)—is more liberal, in that it
allows some arguments for substantive philosophical conclusions to be successes. (p. 305)
What makes philosophical individualism “individualistic” is that it relativizes success to
individuals, holding that ‘x is a successful argument’, like ‘x is in motion’, has implicitly
relativistic truth-conditions, such that an argument might be a success for you without being
a success for me. Philosophical individualism relativizes the success of an argument to the
person evaluating it. (p. 306)

I don’t think ‘philosophical success’ has enough pre-analytic content for it to be

possible to say that ascriptions of philosophical success do or do not have “truth-
conditions that relativize success to individuals.” I think that there is room for criteria
of success that relativize success to individuals and for criteria that relativize success
to very large subsets of the set of all human beings who have lived or will live. (My

52
Van Inwagen (2006).
   

own criterion, as we shall see, relativizes success to historical period and culture.)
Now it may be that it is impossible for a philosophical argument (for a substantive,
positive conclusion) to be a success relative to almost all persons of a certain time and
culture (or to any other such comprehensive class of people) and that it would
therefore be advisable for us to turn our attention to the possibility of criteria of
success that are relativized to individuals. I don’t dispute that. My thesis about
philosophical success, after all, is just exactly that there are no philosophical argu-
ments (for etc.) that are successes relative to almost all persons of our time and
culture (in the most liberal possible sense of ‘success relative to almost all persons of
our time and culture’). It seems to me that even if it is impossible for there to be an
argument that is a success according to my definition, that such an argument is
impossible is an interesting and important thesis. And, I believe, it would remain an
interesting and important thesis if it could be shown that there were other standards
of success, reasonably plausible and intuitive standards, according to which some
philosophical arguments (for etc.) were successes. In any case, Keller has not pre-
sented a definition or analysis of “success relative to persons of our time and culture”
that is more liberal than mine because that is not the concept he has presented a
definition or analysis of. His criterion and mine are not in competition. One could
consistently employ them both, although not, of course, for the same purposes. It
does not follow, however, that he has not shown that my time-and-culture-relative
criterion of philosophical success is, well, not a success.

17.6.2 Three Difficulties for SuccessPvI

Keller has found three principal difficulties with my criterion of success—SuccessPvI,
let us call it. The remainder of this reply will largely be an examination of these
difficulties. (In most of what follows, I will suppress reference to ‘persons of our time
and culture’.)
Difficulty 1: SuccessPvI cannot serve as a criterion of success for first-person
arguments—arguments such that each individual member of any audience to
whom the argument is addressed is expected to go through the argument “for
himself” or “for herself.”
‘Cogito, ergo sum’ is an example of a first-person argument—an example that is,
I think, sufficient to establish the philosophical importance of first-person arguments.
Here, I think, Keller is right: SuccessPvI cannot, as it stands, be applied to first-
person arguments.53 I am, however, fairly confident that it could be revised so as to
apply to them, and in such a way that the revised criterion would be very much in the
spirit of the original. If one presents a first-person argument to an audience, one
generally does so by going through the argument for one’s own case—that is, by using
one’s own case as a model or paradigm or template—and inviting the members of
one’s audience to follow one’s example. Here is an imaginary application of this
dialectical method. René is lecturing to Jean-Paul and Simone. He intones, “Je pense,

53
If there are first-person arguments, there are probably also second-person arguments. If so, Keller’s
point applies to them mutatis mutandis.
   

donc je suis,” and looks expectantly at the two of them. Jean-Paul, following René’s
lead, says, perhaps subvocally, “Je pense, donc je suis” and Simone also says, “Je pense,
donc je suis.” Each of them, or so we may imagine, is going through a certain
argument. Following David Kaplan, we may say that Jean-Paul’s argument and
Simone’s argument are the same in character but differ in content. (The proposition
Jean-Paul expresses by saying “Je suis” is true in exactly those possible worlds in
which he exists; the proposition Simone expresses by saying “Je suis” is true in exactly
those possible worlds in which she exists. Since those are distinct sets of worlds, the
conclusion of Jean-Paul’s argument and the conclusion of Simone’s argument are
distinct propositions.) We may say that our story involves one “character argument”
and three “content arguments.” (If a content argument can be identified with a
sequence of propositions, a character argument can be identified with a function
from sets of indices—of some agreed-upon kind: person, moment of time, and place,
it may be—to content arguments.) I propose to mend the flaw (the very real flaw)
that Keller has found in SuccessPvI by modifying SuccessPvI in such a way that it
applies to character arguments as well as content arguments. I will not undertake this
task here, however. In the sequel, I will defend only the thesis that SuccessPvI is, as it
stands, a useful and interesting criterion of success for “third personal” philosophical
arguments—that is, arguments such that every member of the audience to whom the
argument is addressed is considering the same content argument.
Difficulty 2: SuccessPvI involves the concept of an audience of whose members are
neutral agnostics with respect to the conclusion of an argument. But strict neutrality
with respect to a proposition is a condition that is much more demanding than I have
supposed it to be; it is in fact so difficult to achieve that a criterion of success that
appeals to it is bound to be unworkable.
Keller says:
There are, however, a number of problems with thinking that successful arguments must
convince neutral agnostics. First, neutral agnosticism is infectious. If neutral agnostics about p
only had to be agnostic about p itself, some such agnostics might firmly believe that almost all
arguments for or against p are sophistical. Such agnostics would be unlikely to be convinced by
any argument for or against p, successful or not.
But once we see that neutral agnostics about p must be agnostic about more than p itself, it is
hard to know where to stop. Would neutral agnostics about the existence of free will have to be
compatibilists? Moral realists? Consequentialists? Physicalists? Theists? Scientific realists? Even if
our agnostics were strictly neutral about the existence of free will, consider how differently
“primed” towards free will someone would be who was an incompatibilist, moral nihilist, atheist,
physicalist, and scientific realist compared with someone who was a compatibilist, moral realist,
theist, dualist, and scientific anti-realist. Of course, one can believe in free will as an incompatibi-
list, moral nihilist, atheist, physicalist, and scientific realist without manifest irrationality: Mark
Balaguer accepts all of those except moral anti-realism, and there is no reason to think that
adding moral anti-realism to the mix would make his position inconsistent. But convincing
Balaguer to reject free will seems much easier than convincing someone who is agnostic about
free will, but who is a compatibilist, moral realist, theist, dualist, and scientific anti-realist. Explicit
opposition to a claim doesn’t guarantee that the totality of one’s views doesn’t “mesh” better with
that claim than does another totality that is neutral about the claim. Of course, to be truly neutral
   

with regard to free will one would also have to be agnostic about determinism, consequentialism,
reductionism, and much else. If the members of a jury of “ideal neutral agnostics” were not
agnostic about these theses, they might come to different verdicts on the basis of their differing
opinions. But an audience that was agnostic about all of these would believe so little that it is hard
to imagine convincing them of anything at all. (p. 308–9)
I will first consider the case of those agnostics who ‘firmly believe that almost all
arguments for or against p are sophistical’. When laying out the features of the
members of an ideal “audience of agnostics” in The Problem of Evil, I said
The audience is composed of what we may call agnostics. That is, they are agnostic as
regards the subject-matter of the debate. If the debate is about nominalism and realism . . . ,
each member of the audience will have no initial opinion about whether there are universals,
and no predilection, emotional or otherwise, for nominalism or for realism.54 As regards a
tendency to accept one answer or the other, they will stand to the question whether there are
universals as you, no doubt, stand to the question whether the number of Douglas firs in
Canada is odd or even. But that is not the whole story; for you, no doubt, have no desire to
have the question whether that number is odd or even settled. My imaginary agnostics are
not like that in respect of the question of the existence of universals. They would very much
like to come to some sort of reasoned opinion about the existence of universals—in fact, to
achieve knowledge on that matter if it were possible. They don’t care which position,
nominalism or realism, they end up accepting, but they very much want to end up accepting
one or the other. (p. 44)
Perhaps ‘firmly believe that almost all arguments for or against nominalism are
sophistical’ is consistent with the letter of this description. It doesn’t exactly seem
to be consistent with the spirit of ‘They would very much like to come to some sort of
reasoned opinion about the existence of universals—in fact, to achieve knowledge on
that matter if it were possible’. I should have no objection to explicitly adding ‘regard
it as an open question whether there can be cogent arguments for p’ to my list of the
features of ‘neutral agnostics with respect to p’. I would regard that as a friendly
amendment. But suppose we don’t add that. That will no doubt make the task
assigned to the proponent of the argument more difficult—but it will not, by itself,
make that task hopeless. Suppose the debaters are Atheist and Theist, that the
proposition in question is ‘God does not exist’, and that the argument for that
proposition that is being tested the argument from evil. I had been supposing that
in the course of the debate before the audience of agnostics only Atheist and Theist
would speak. But we could allow questions and remarks from the audience as well.
(In fact, that’s probably a good idea quite independently of the question whether it is
possible for there to be an audience of “strictly neutral” agnostics.) Suppose a
member of the audience says, or says something that implies, that any argument

54
A potential predilection is not a predilection. A and B may both have no predilection to accept p and
yet it may also be that it would be much easier to cause A to have such a predilection than to cause B to
have one. It may be, for example, that A has a predilection to accept certain propositions that logically
imply p and that B has no predilection to accept any propositions that logically imply p. (Note added in the
present essay.)
   

either for the existence or the non-existence of God is almost certain to be sophistical.
It’s certainly open to Atheist to introduce a line of reasoning that opens with a
remark along the lines of, “All right, but do you think that this argument, the
argument from evil as I have presented it, is sophistical? This is a particular, concrete
argument. If it is sophistical, wherein does its sophistry lie?” In other words, it’s open
to Atheist to try to convince the members of the audience that, whatever may be the
case in general with arguments for the non-existence of God, this one, the one she has
presented, is not sophistical.
“Yes, but in such a case, at least some members of the audience are not strictly
neutral.” The Interlocutor’s question brings us to Keller’s more general argument as
regards the impossibility or near-impossibility of there being an audience of “strictly
neutral” agnostics. I can say only that I did not intend the audience to be “strictly
neutral with respect to p” in Keller’s very demanding sense but only in my sense. That
is, the sense I intended is the sense I intended: what I have written, I have written.
And that sense is: when the debate begins, each of them refuses to assign a higher
subjective probability to the conclusion of the argument than to its denial—and vice
versa. Suppose that the argument under consideration (call it ‘FW’) is an argument
for the conclusion ‘Human beings have free will’. And suppose the audience com-
prises people who are neutral agnostics in my sense (“neutralPvI agnostics”) on the
question whether human beings have free will. Suppose, finally, that some of them
are dualists and some are materialists. Ophelia, the opponent of the argument is able
to convince the members of the audience that premises (2) and (4) of FW jointly
imply dualism. Then, unless Protagoras, the proponent of the argument, is able
somehow to convert the materialists in the audience to dualism or is able to convince
them that (2) and (4) do not, after all imply dualism, he will find that Ophelia has
effectively blocked his attempt to convince the audience (all the members of the
audience) of the existence of free will by presenting the argument FW. But I don’t see
why this case should be supposed to imply that the members of the audience who are
materialists should not be described as (having initially been) neutral agnostics with
respect to the existence of free will. What the case does do is provide an additional
reason to suppose that it is no easy task to find an argument for a positive, substantive
philosophical conclusion p that will convince an audience whose membership com-
prises neutralPvI agnostics with respect to p to accept p. And that reason is: any given
argument A whose conclusion is p may well be such that various members of that
audience are convinced adherents of various philosophical theses that are inconsist-
ent with various premises of A. I am happy to allow the following addition to my
description of the audience: it is large and its members hold all possible combinations
of philosophical positions that are consistent with their being neutral agnostics—
neutralPvI agnostics—with respect to the conclusion of the argument under consid-
eration. (By ‘x holds a combination of philosophical positions consistent with x’s
being a neutralPvI agnostic with respect to p’ I mean this: Let ϕ be the combination of
philosophical positions that x holds; the statement that x holds the combination of
philosophical positions ϕ and is a neutralPvI agnostic with respect to p is not a self-
contradictory statement. I am quite willing to allow that many of the combinations of
philosophical positions held by individual members of the audience are inconsistent
with neutral agnosticism with respect to p—although, of course, the contradictions
   

can’t be very obvious ones, owing to the agnostics’ enjoying “the highest degree of
logical acumen.”55)
Note that it is not clear to what extent bias accounts for disagreement in philosophy, since
there is persistent apparently rational disagreement about philosophical questions about which
no one (or hardly anyone) has a preconceived opinion: Newcomb’s Paradox, the nature of
mental representation, quantifier variance, whether properties are parts of their instances, and
so on. (p. 309, n 24)

I can’t agree. Even in the most abstract areas of philosophy, we tend to accept the
propositions (ideas, points of view . . . ) that were advanced and defended by the most
persuasive (charismatic, forceful, overbearing, dialectically able . . . ) of those who
taught us in graduate school. If one studied metaphysics at Notre Dame when John
Keller did, one would, ten years on, be much more likely to take a Quinean meta-
ontology for granted than one would be if one had been an MIT or a Pittsburgh
“product”; and one would be much more likely to reject a constituent ontology than
one would be if one’s doctoral studies had been at UNC or Durham University.
Consider this passage from Mark Lilla’s The Reckless Mind: Intellectuals in
Politics:56
[At Marburg] Heidegger attracted students who traveled from the four corners of Europe to
study with him. Although he had not yet published any works, he quickly became known
throughout Germany as a genius and radical thinker simply by the force of his teaching.

It would be hard to deny that almost all the adoring students who hung on every
word of “the little magician from Messkirch” would have been strongly inclined to
dispute Hegel’s characterization of Sein as “the most barren and abstract of all
categories”—and would have been so inclined for the rest of their lives. This is
simply an extreme example of a very common phenomenon. (It is true that in
graduate school we academics tend to be stripped of the biases we acquired as
undergraduates. Nevertheless, even our undergraduate education has almost cer-
tainly had a lasting influence on our philosophical views, for the opinions of our
undergraduate teachers influenced our choice of where to pursue graduate study—
and the philosophical “schools” to which our letter-of-recommendation writers
belonged had a lot to do with where our candidacy for admission had even a chance
of success. A Marquette undergraduate is much more likely to get a PhD in
philosophy from Penn State than from Princeton.)
An additional problem is that agnostics are not neutral in debates about agnosticism itself,
religious or otherwise. Pyrrhonian skeptics aim only to get us to suspend belief about whether
there is an external world, so someone who is agnostic about whether there is an external world
already accepts the skeptic’s conclusion. When it comes to arguments that are skeptical in this
sense, there is no neutral vantage point. But if anti-skeptical arguments do not need to appeal

55
‘The highest degree of logical acumen’ was not meant to imply “logical omniscience” on their part.
See the reply to Thomas Kelly and Sarah McGrath, p. 383.
56
Lilla (2001). Available on line at <http://www.worldcat.org/wcpa/servlet/DCARead?standardNo=
0940322765&standardNoType=1&excerpt=true>.
   

only to premises acceptable to the skeptic, and they do not need to appeal only to premises
acceptable to someone agnostic about skepticism, what is left? The only plausible answer is
that they can appeal to premises that non-skeptics themselves accept. But that supports
philosophical individualism over views that claim that successful arguments must utilize
premises acceptable to everyone, a neutral audience, etc. (p. 309)
Let us consider the case of the Skeptics. (I will not call them Pyrrhonian skeptics for
historical reasons that I will not go into.) Let us in fact consider two particular
skeptics, Pyrrho and Carneades. Each is standing before an audience comprising
persons of our time and culture57 and each wishes to bring it about that the members
of his audience accept neither the proposition that there is an external world nor the
proposition that there is no external world. But how do they attempt to do this?
There would seem to be two dialectical procedures such a skeptic might employ.
Pyrrho choses one and Carneades the other.
Dialectical procedure 1: Pyrrho will proceed by presenting the members of his
audience (AP) with irrefutable arguments for the existence of an external world and
irrefutable arguments for the non-existence of an external world.58 (Let us call these two
propositions the Positions.) And he will, as it were, allow these arguments speak for
themselves. He will present no final, comprehensive “the lesson of these arguments is”
argument that refers to them (as Kant did with respect to his Antinomies of Pure Reason)
and whose conclusion is a proposition along the lines of ‘Both Positions are without
epistemic warrant’: He will not go on to say, “Since, as we have seen, there are irrefutable
arguments for both Positions, both Positions are therefore without epistemic warrant.”
A philosopher who, like Pyrrho, employs dialectical procedure 1 presents no
problems for SuccessPvI, owing to the fact that that procedure involves no argument
that is intended to convince those to whom it is addressed of the truth of its conclusion.
It is, in fact, essential to Pyrrho’s strategy that none of the arguments the skeptic
employs should, in the end, convince any member of AP that its conclusion is true—
owing to the fact that, for each argument he presents, he will have presented at least one
equally convincing argument for the denial of its conclusion.
Dialectical procedure 2: Carneades will proceed by presenting the members of his
audience (AC) with an argument whose conclusion is a certain epistemological
proposition. (Or more than one argument. But let us make the case as simple as
possible and suppose he presents only one argument.) Let us call that argument SKA
and let us suppose that the conclusion of SKA is
Skepticism Both Positions are without epistemic warrant.

57
It can therefore hardly be the case that “Pyrrho” and “Carneades” are the historical Pyrrho of Elis and
the historical Carneades of Cyrene. In any case, their dialectical procedures do not correspond to those of
their respective namesakes.
58
I.e., arguments each of which, when considered individually, seems to the members of AP and to
Pyrrho himself obviously to be a proof or demonstration of its conclusion. Pyrrho, of course, concedes that
some of or all his arguments are unsound: he does not reject the law of the excluded middle. But he is
convinced that he hasn’t the faintest clue as to what the flaw(s) in any of his arguments might be; and he is
attempting to induce (simply by presenting those arguments and defending them against all known
objections) a like conviction in the minds of the members of AP.
   

Note that, unlike the arguments Pyrrho presented, which were not intended to
convince the members of AP that their conclusions were true, this argument is
intended to convince the members of AC that its conclusion is true. But should
SKA convince them that its conclusion is true? Suppose I say that that SKA should
convince the members of AC that everyone should suspend judgment about whether
there is an external world just in the case that SKA is a SuccessPvI. Does that mean
that AC must be entirely composed of persons who neutralPvI agnostics about the
truth or falsity of Skepticism? No, not at all. What it implies is this:
SKA should convince the members of AC that Skepticism is true just in the case
that:
If presented in IDEAL CIRCUMSTANCES59 SKA would convince the members of an audience
of initially neutralPvI agnostics (about the truth or falsity of Skepticism) that Skepticism is true.

Here is a strictly parallel statement (let ‘MAT ’ be any argument for materialism,
understood as the thesis that human beings are wholly material):
MAT should convince Robert M. Adams and Richard Swinburne and Alvin
Plantinga that human beings are wholly material just in the case that:
If presented in IDEAL CIRCUMSTANCES MAT would convince the members of an audience
of initially neutralPvI agnostics about the truth or falsity of materialism that human beings are
wholly material.

Each of these two biconditionals makes reference to two audiences: the former to AC
and an audience of agnostics about Skepticsm, and the latter to Adams, Swinburne,
and Plantinga (on the one hand) and an audience of agnostics about materialism (on
the other).
Let us suppose that INA is a representative or typical audience of “initially
neutralPvI agnostics about the truth or falsity of Skepticism.” It may well be that
each of the members of INA must—in order to exhibit the appropriate species of
agnosticism about Skepticism—accept neither Position. (For it seems plausible to
suppose that there is some sort of pragmatic contradiction involved in saying
anything of the form, ‘p, and my belief that p is without epistemic warrant’.) Let us
suppose that that is so. Has it any consequences that are relevant to the question
whether SuccessPvI is a useful or interesting criterion of philosophical success? It
certainly does not follow that, since AC is presumably composed mostly of people
who believe that there is an external world, Carneades should not present SKA to the
members of AC with the intent of demonstrating to them that both their Position and
its denial are without epistemic warrant. And it would not follow that ‘SKA is a
SuccessPvI’ is not sufficient for ‘SKA should convince the members of AC that both
Positions are without epistemic warrant’. There would, of course, be no point in
Carneades’ presenting SKA to an audience whose members were neutralPvI agnostics
about the existence of an external world in order simply to convince them neither
to accept nor to reject the proposition that there is an external world, since they
“already” neither accept nor reject that proposition. That fact, however, does not

59
I trust that everyone who has got this far in the discussion will know what ‘in IDEAL CIRCUM-
STANCES’ abbreviates.
   

prevent us from “testing” SKA by asking whether it could be used to cause an

audience of initially neutralPvI agnostics about the truth of the proposition Skepticism
to accept that proposition. As I have said, it is plausible to suppose that an audience
that was agnostic about Skepticism would also have to be agnostic about whether
there was an external world. But an audience that was agnostic about whether
there was an external world could certainly also be agnostic about whether both
‘There is an external world’ and ‘There is no external world’ are without epistemic
warrant. There is no contradiction, pragmatic or otherwise, in my saying something
of the form:
I accept neither p nor its denial; and I accept neither the proposition that p is without epistemic
warrant nor the proposition that the denial of p is without epistemic warrant.

Consider, for example, the proposition that the power of the continuum is either
aleph-one or aleph-two. I accept neither that proposition nor its denial. It has been
reported that Gödel believed that the power of the continuum was either aleph-one
or aleph-two. Let us suppose that that report is correct. I do not believe (do not have
the belief that) Gödel’s belief was without epistemic warrant. (For all I know, his
belief was based on some profound mathematical insight.) Nor do I have the belief
that the proposition that the power of the continuum is neither aleph-one nor aleph-
two is without epistemic warrant. (For all I know, Gödel’s “profound insight” was
non-veridical—even Homer nods—, and the proposition that the power of the
continuum is neither aleph-one nor aleph-two is true. And, for all I know, some
other mathematician has had a veridical mathematical insight that revealed its truth
to him or her.) Therefore:

I accept neither the proposition that the power of the continuum is either aleph-one or aleph-
two nor its denial; and I accept neither the proposition that (the proposition that the power of
the continuum is either aleph-one or aleph-two is without epistemic warrant) nor the
proposition that (the proposition that the power of the continuum is neither aleph-one nor
aleph-two is without epistemic warrant).

Difficulty 3: SuccessPvI is incompatible with (subjective) Bayesianism.

Keller says,
Bayesians hold that what is rational to conclude on the basis of new arguments or evidence
depends on one’s priors. But people have different priors, and according to (subjective)
Bayesian orthodoxy no set of coherent priors is more rational than another. Hence, two people
exposed to the same evidence and arguments may reach different conclusions. Indeed, if they
are rational, they must reach different conclusions, each in accordance with her priors. But
then there can be no objective fact about whether an argument makes acceptance of its
conclusion rational, and hence whether the argument is successful. For this will depend on
one’s priors, and priors are individualistic. Hence, so is argumentative success. [The following
footnote is appended to this passage: ‘Yes, priors can often be “swamped” by the evidence. . .
. But they are not always swamped, and so as long as we hold that any consistent set of priors is
rational—or even a wide range of consistent priors—exposure to the same arguments will not
necessarily lead to agreement, even between people who respond to evidence perfectly.’] (pp.
311–12)
   

If this argument is cogent, it tells against the possibility of there being any argument
in any field of enquiry that is a non-individual-relative success. This one, for example:
‘Only a ball casts a circular shadow from all angles; hence, if the boundary of the
shadow the earth casts on the moon during a partial lunar eclipse is always an arc of a
circle, then the earth must be a ball.’ Or consider currently available arguments for
the conclusion that smoking cigarettes causes lung cancer or for the conclusion that
the continents are in motion. (I don’t mean the arguments that have convinced
epidemiologists and geologists that these propositions are true; I am thinking of
arguments addressed to the general public that rest, at least in part, on appeals to the
testimony of such experts.) These arguments are successes for (almost) all of us, for
all us twenty-first-century adult human beings who are members of a European or
“European-descended” culture—or almost all. (Remember that in my account of the
audience of agnostics, I said, “I mean the agnostics to be drawn from our time and
our culture.”60) Almost all adult persons of our time and our culture share certain
priors—and a very large and rich class of them. Almost everyone “of our sort” assigns
a subjective probability of 1 to ‘Some people live in houses’ and ‘Lions are carnivores’
and ‘It is colder at the poles than at the equator’.61 Almost everyone assigns a
subjective probability of 0 to ‘Butter is a high explosive’ and ‘Tomatoes are imported
from the moon’ and ‘Bacteria are capable of learning calculus’. Almost everyone
assigns a subjective probability of 0.5 to ‘When Vladimir Putin dies, he will have had
an odd number of birthdays’. If these rather fanciful examples are not examples of
true priors—owing to the fact that most people who accept them accept them on the
basis of the probabilities they assign to other propositions—, their near-universality
implies that there must be some probability assignments that constitute the priors of
almost every adult of our time and culture.62 I think we may speak of the “core
priors” of the (adult) members of that class of people, the priors that they (almost) all
share. (Naturally, there are bound to be borderline or disputable cases of membership
in this class: perhaps there is no determinate answer to the question whether ‘If a fair
coin has fallen heads ten times in a row, the probability that it will fall heads on the
next toss is 0.5’ is a core prior.) I am happy to relativize my account of success to
“success with respect to those whose set of priors (whose consistent set of priors)
includes the core priors of our time and culture.” My point, after all, was that no
philosophical argument for a positive, substantive conclusion is an “impersonal”
success in the way ‘Only a ball casts a circular shadow from all angles; hence, if the
boundary of the shadow the earth casts on the moon during a partial lunar eclipse is
always an arc of a circle, then the earth must be a ball’ is an impersonal success. It is
therefore essential to my purposes that I should appeal to a criterion of success that
allows some arguments to be impersonal successes in that way.

60
Van Inwagen (2006: 47). My account is, therefore, as I said earlier, audience-relative. Following the
quoted sentence, I went on to say, “ . . . so limiting the jury pool, of course, relativizes our criterion of
success to our time and our culture.” I direct the interested reader to the discussion of this “relativization”
on pp. 47 and 48 of The Problem of Evil.
61
Or at least a probability that is essentially 1—a probability so close to 1 that we unreflectively treat it as
1 when we are engaged in practical reasoning.
62
Assuming that everyone really can be said to “have” a set of priors. If that assumption is wrong, of
course, then Bayesianism is wrong and thus does not present SuccessPvI with any sort of difficulty.
   

17.6.3 A Difficulty for SuccessJK

This concludes my discussion of the three difficulties that, in Keller’s view, confront
my position on dialectical success and failure in philosophy. I will close this reply by
mentioning a difficulty I see in Keller’s own individual-relative account of success,
the “Knowledge Account”:
An argument is successful for an individual just if she knows it is sound (and non-fallacious).
I do not think this will do. I know that Genuine Modal Realism is not an acceptable
modal ontology, and I know that Alvin Plantinga has said that Genuine Modal
Realism is not an acceptable modal ontology. I know that modus ponens is a valid
argument-form. I therefore know that the following argument is sound:
Alvin Plantinga has said that Genuine Modal Realism is not an acceptable modal ontology
If Alvin Plantinga has said that Genuine Modal Realism is not an acceptable modal ontology,
then Genuine Modal Realism is not an acceptable modal ontology63
hence,
Genuine Modal Realism is not an acceptable modal ontology.
But I can see no sense in saying that this argument is a success “for me.”64 If our
definiendum is
A is a success for i,
I propose that the following clause, or something very much like it, appear some-
where in the definiens:
i knows the premises of A to be true, and would know those premises to be true
whether she knew its conclusion to be true or not.65

63
If you think that knowing that the consequent of an indicative conditional is true is not sufficient for
knowing that the conditional is true, substitute this disjunction for the conditional: Either (it’s not the case
that Alvin Plantinga has said that Genuine Modal Realism is not an acceptable modal ontology) or
Genuine Modal Realism is not an acceptable modal ontology.
64
If you doubt whether I know that Genuine Modal Realism is false, I could state my conclusion
hypothetically: If I did know that Genuine Modal Realism was false, this argument would not be a success
for me.
65
Might it be that the argument I have offered as a counterexample to Keller’s definition “begs the
question,” and is thus fallacious? That is hard to say, because it is hard to say what it is for an argument to
beg the question. (It would certainly be possible for someone to know that the conditional second premise
of the argument was true without knowing whether its conclusion was true.) But if Keller’s definition is
modified in the way I have suggested, the proponent of the revised definition who undertakes to determine
whether (e.g.) the argument from evil is a success for Eleonore Stump need not consider whether the
argument from evil is guilty of the fallacy of begging the question—whatever, precisely, that fallacy may be.
Note also that if God has revealed to Meghan Sullivan that Alvin Plantinga is always right about modal
ontology, the argument in the text will be a success for her even if, in some impersonal sense, it begs the
question. (One who accepts my revision of Keller’s definition and who believes that there is such a fallacy as
begging the question should probably replace Keller’s ‘and non-fallacious’ with something like ‘and is non-
fallacious except, possibly, in the matter of begging the question’.)
   

17.7 Reply to Thomas Kelly and Sarah McGrath

Thomas Kelly and Sarah McGrath’s paper is stimulating, thoughtful, clear, and closely
reasoned. Because it is stimulating and thoughtful, I am moved to reply to it at some
length. Because it is clear and closely reasoned, it is easy for me to see on what points
I disagree with them about success and failure in philosophical argument.
In The Problem of Evil, I gave a definition of what it is for a philosophical argument
to be a failure. And I presented an argument for the conclusion that all the members
of a very broad class of philosophical arguments were, in the sense provided by that
definition, failures. (If that argument is sound, its soundness implies that it is itself a
failure: its conclusion is a substantive metaphilosophical thesis, and substantive
metaphilosophical theses are substantive philosophical theses. I don’t mind that.
There are many good philosophical arguments in the “broad class” to which I have
alluded.66 On pain of pragmatic inconsistency, I can’t hope to produce a compelling
philosophical argument—so to call a philosophical argument that is not a failure67—
for that conclusion, but I think my argument is a good argument.)
The definition was framed in terms of an ideal debate between a defender of the
conclusion of an argument and an opponent of its conclusion (Desmond and
Ophelia, let us call them) before an audience of ideal, neutral agnostics (agnostics
with respect to the conclusion of the argument)—an audience whose members are, in
Kelly and McGrath’s happy phrase, disinterested but not uninterested with respect to
the conclusion of the argument. Very roughly speaking, the argument in question will
be a failure just in the case that Desmond will ultimately be unable to use that argument
to turn the agnostics into believers like himself—will be unable to use the argument to
convince them to accept its conclusion. I say ‘ultimately’ because he must do this in the
presence of Ophelia, his opponent in the debate, whose dialectical function is to do
everything possible to block his attempt at “conversion.” (This statement of the
definition of failure is intended only to serve as a reminder of its content. My “official”
statement of the definition remains the one presented in the book.)
Before I proceed to a discussion of Kelly and McGrath’s essay, I wish to make two
comments on the definition.
(1) The definition was intended to apply only to arguments for “controversial”
conclusions—an uncontroversial conclusion being an argument that practically
everyone would accept. (Not an entirely straightforward concept, I concede: Berkeley
thought that practically everyone accepted his conclusions—whether they knew it or
not; most philosophers reject this judgment.) Consider, for example, arguments
against solipsism. Few philosophers, if any, who construct such arguments construct
them with the intention of convincing the generality of humanity that solipsism is
false. The purpose of those arguments must be something other than that (perhaps it
is to display the basis of our “knowledge of other minds”; perhaps each of them is an
attempt to provide an explicit statement of an argument that—its author supposes—
almost everyone accepts tacitly). The criterion of success and failure for arguments

66
At least some versions of the argument from evil are very good philosophical arguments indeed.
67
Think of the relation between these two terms this way: to say that an argument is a failure is to say
that it fails to be a compelling argument.
   

against solipsism or for the reality of change and motion or for the existence of truths
about the past must, in my view, be of a different sort from the criterion of success
and failure for arguments for (and arguments against) consequentialism or materi-
alism or mathematical platonism. My definition is intended to provide a criterion of
success and failure only for arguments of the latter sort.
In connection with this point, I will distinguish two senses in which a philosophical
argument may be said to be an argument for a certain philosophical proposition ϕ. This
may mean no more than that ϕ is the conclusion of the argument. But it may mean that
and more. It may also mean that the argument has been constructed with an intended
audience in mind, an audience of philosophers (and perhaps non-philosophers as well)
who do not accept ϕ (who either accept the denial of ϕ or who accept neither ϕ nor its
denial) and which was constructed for the purpose of convincing that audience to
accept ϕ. I will use “argument for” in the latter, stronger sense.
(2) If I were given the opportunity to revise the account of philosophical success
and failure that I presented in The Problem of Evil, I would no longer say that an
argument was a failure just in the case that Desmond (the defender of its conclusion)
“will ultimately be unable to use that argument to turn the agnostics into believers
like himself”; I would say, rather, that an argument is a failure just in the case that he
will ultimately be unable to do either of the following two things:
• turn the agnostics into believers like himself
• turn the agnostics into positively weighted agnostics.
And this is what italicized phrase means:
At the beginning of the debate, the agnostics were neutral agnostics: either they assigned a
subjective probability of 0.5 to the conclusion of the argument (and of course to its denial) or
they refused to assign any subjective probability to either the conclusion or its denial (nor were
they willing to ascribe to those propositions membership in any range of subjective
probabilities—other than [0,1], of course). A positively weighted agnostic (with respect to
a proposition p) is a person who accepts neither p nor its denial and who either assigns to p a
particular probability significantly higher than 0.5 or who (more likely) is willing to say that the
probability of p, whatever exactly it may be, is significantly higher than 0.5.68
Suppose, for example, that the outcome of the debate imagined in the discussion of
the global argument from evil in The Problem of Evil was this: the formerly neutral
agnostics all say something along the lines of, “We still don’t accept either atheism or
theism, but now—now that the debate is over, now that both Theist and Atheist have
said everything they have to say about the global argument from evil—we have been
convinced by that argument to assign a significantly higher probability to atheism
than to theism.” If that is what the outcome of the debate would be, then the
argument would be a failure according to the definition in The Problem of Evil but
a success according to the revised definition.69

68
If one uses the revised definition, one should not say that an argument that is not a failure is
“compelling.”
69
I touched on this point in The Problem of Evil. See pp. 50–1.
   

In the sequel, however, I’ll keep to the original definition, if only because that is the
definition that Kelly and McGrath were working with (and they have not criticized
the definition on the ground that it entails that an argument with the power to turn
neutral agnostics into positively weighted agnostics—but not into “believers”—would
count as a failure).
I turn now to various matters that pertain to Kelly and McGrath’s critique of my
argument for the Pessimistic Verdict.

17.7.1 Uniformity of Response?

In The Problem of Evil, I assumed “uniformity of response” from the audience of
agnostics. That is, I assumed that, when the debate was finished, either they would all
be convinced by the argument that had been advanced and would all, therefore,
accept its conclusion, or they would all say, “For all I know, at least one of the
premises of that argument is false.”70 (I conceded that this was an assumption I was
making, and said that it would be interesting to look into the consequences of
proceeding without it.) It still seems pretty reasonable to me, given the high degree
of idealization involved in the debate. But, rather than defend it, I will look into the
consequences of proceeding without it.
On consideration, I’m willing to accept the definition without the assumption. My
reasoning is this. Suppose you did present a certain argument α for, let’s say, the
existence of abstract objects or platonism, to an audience of ideal agnostics in the
context of an ideal debate between a platonist and a nominalist (a denier of platon-
ism). Suppose the response from the audience of agnostics is, as they say, all over the
shop. Some are convinced by α and become platonists. Others, remain agnostic on
the question of the existence of abstract objects on the ground that that for all they
know, some of the premises of α are false (some say premises (3) and (6) are, for all
they know, false; others find those two premises luminously evident but are doubtful
about premise (4) . . . ). Perhaps a few of them remain agnostic owing to their
conviction that one or two of the premises of α are simply false, false without
qualification, and self-evidently so. Suppose further, that when α is presented (in
the context of an ideal debate) to different (sizable) groups of ideal neutral agnostics,
the non-uniform outcomes vary widely: in Group A, 32 percent remain agnostics and
the remainder are convinced by α and become platonists; in Group B, it’s 48 percent
and 52 percent. . . .
That certainly seems to me to be an excellent reason to say that α is a failure. Or
put the matter this way. It seems to me that anyone who accepts the proposition
If, when α is presented to an audience of ideal neutral agnostics in the context of an ideal
debate, they all, to a man and to a woman, respond by saying, “For all I know, some of the
premises of α are false,” then α is a failure.

70
We assume throughout, for the sake of simplicity, that all the arguments we are considering are
uncontroversially logically valid. (If there are questions about what that means, I should willing to discuss
them—but to attempt within the scope of this brief reply to anticipate and address such questions would be
what one of my teachers called philosophical shadow-boxing.) In consequence, the only questions relevant
to our purposes that can be raised about the arguments pertain to the truth-values of their premises.
   

should also be willing to accept the proposition

If, when α is presented to an audience of ideal neutral agnostics in the context of an ideal
debate, their responses vary widely—are, in fact, all over the shop—then α is a failure.
That is to say, an argument (for a controversial conclusion) is a success only if, if it
were presented to an audience of ideal neutral agnostics in the context of an ideal
debate, the members of that audience would uniformly respond by accepting its
conclusion. (We could weaken that principle, perhaps—say to ‘almost uniformly
respond’. But we should, I believe, certainly count an argument as a failure if we
somehow knew that the “ideal response” would be very far from uniform.)

17.7.2 Logical Omniscience?

Kelly and McGrath correctly identify a problem that would confront my definition of
philosophical success and failure if my ideal agnostics were “logically omniscient”—
if, for any proposition they accepted, they immediately saw, and accepted, all its
logical consequences. Fortunately, I do not have to deal with that problem, since I did
not say anything that implied that they were logically omniscient. I did want my
agnostics to be human beings, and—obviously I would say—human beings can
consider or grasp or hold before their minds only an infinitesimal proportion of
the logical consequences of any proposition. When I said that the agnostics possessed
“the highest degree of logical acumen,” I meant only that, for any argument Des-
mond or Ophelia might present them with (it would have to be simple enough for a
human being actually to understand—but then Desmond and Ophelia are human
beings, too, and they are unlikely to present the agnostics with arguments so complex
that they themselves are unable to understand them), they would, given a reasonable
amount of time to consider the matter, be able to determine, infallibly, whether that
argument was logically valid. Few would deny “the highest degree of logical acumen”
to Gottlob Frege, and yet for many years he was unaware of the fact that the three
propositions (the language of this example is anachronistic)
For any one-place open sentence x such that, for every object y, either y determinately satisfies
x or y determinately fails to satisfy x, there exists exactly one set whose membership comprises
exactly those objects that satisfy x
For every object y, either y determinately satisfies ‘x is a member of x’ or y determinately fails to
satisfy ‘x is a member of x’
For every object y, either y determinately satisfies ‘it is not the case that x is a member of x’ or y
determinately fails to satisfy ‘it is not the case that x is a member of x’
logically implied the proposition
There exists a set x such that x is a member of x and it is not the case that x is a member of x.
And yet, when Russell presented him with the argument whose premises were
(essentially) those three propositions, and whose conclusion was (essentially) the
fourth proposition, he immediately saw—owing to his possessing logical acumen in
the highest degree—that that argument was logically valid.
I meant only to imply that my agnostics possessed Frege’s sort of logical acumen.
   

17.7.3 What is a Positive Thesis?

Kelly and McGrath raise questions about what a positive thesis is, and whether, by
any reasonable account, atheism is a positive thesis. (The latter point is relevant to the
use I made of the Pessimistic Verdict in The Problem of Evil, where I suggested that it
would not be surprising to one who accepted the Verdict if the argument from evil
turned out to be a failure.) On this point, I will quote a scattering of passages from a
paper Kelly and McGrath do not consider in their essay.71 What is said in these
passages will not perhaps satisfy those who think that the concept of a positive thesis
is insufficiently well defined to be a load-bearing member in a serious philosophical
argument (or satisfy those who insist that an account of ‘positive thesis’ that had the
consequence that atheism was a positive thesis would be ipso facto objectionable),
but it is the best I can do. The first passage is,
What it means to say that a belief (proposition, thesis, conjecture, theory, hypothesis . . . ) is
positive or negative is hard to explain in any philosophically satisfactory way, and I will not
attempt to do so. I shall have to be content to give a few examples of philosophical beliefs or
propositions that are paradigmatically not positive: ‘Formalism is not the correct philosophy of
mathematics’; ‘Utilitarianism is not an acceptable ethical theory’; ‘Knowledge is not simply
justified true belief.’ And, by the same token, ‘Knowledge is justified true belief,’ although it is
no doubt a false thesis, is a positive thesis, and to assent to it is to have a positive philosophical
belief. Formalism and utilitarianism—assuming that these terms have been sufficiently well
defined that they denote particular propositions—are positive theses, and anyone who accepts
formalism or accepts utilitarianism thereby has a positive belief. (p. 16)

A few sentences later, speaking of the theses ‘Change and motion are not real features
of the world’, ‘One has no reason to suppose that there are minds other than one’s
own’, and ‘There are no material objects’, I said,
All these theses, or my statements of them, contain some sort of negative construction.
Nonetheless, all of them are what I would call ‘positive’ theses. As I said, ‘positive’ is a very
hard term to explain. (pp. 16–17)
And to these two sentences I added a footnote (footnote 3, p. 17):
I would say that the negation of a negative belief must be a positive belief, but that the negation
of a positive belief will in some cases also be a positive belief. An analogy is perhaps provided by
the concept of positive and negative geographical information. That the spy whose where-
abouts we should like to know is not in London is a negative piece of geographical information,
and that he is in London is a positive piece of geographical information. That he is in the
Western Hemisphere is a positive piece of geographical information, but so is the information
that he is not in the Western Hemisphere—at least given that he must be either in the Eastern
or the Western Hemisphere—, for the latter piece of information narrows down our range of
possible specific hypotheses as to his location precisely as effectively as its negation does.
I might put my point this way: ‘Theism is false’ is a positive philosophical belief because both
theism and its negation, atheism, are philosophical theories or at any rate philosophical

71
Van Inwagen (2009b).
   

positions. ‘Utilitarianism is false’ is not a positive philosophical belief because [the negation of
utilitarianism] is not a philosophical theory or position. There are many philosophical
theories—many ethical theories—that are incompatible with utilitarianism, but [the negation
of utilitarianism] or the disjunction of all ethical theories (indeed, of all propositions) incom-
patible with utilitarianism, is not one of them: it’s incompatible with utilitarianism all right, but
it’s not an ethical theory—and not a theory of any sort.

17.7.4 The Pessimistic Verdict

I turn now to the Pessimistic Verdict. I prefer to state it this way:
Every known72 argument for any substantive, positive philosophical thesis is a failure.
(Here, of course, the sense of ‘failure’ is to be provided by my definition. I do not deny
that there may be other senses of ‘failure’, eminently defensible senses, in which some
arguments for substantive philosophical theses are not failures. Anyone who thinks
that this is so and who thinks the point an important one may feel free to substitute
some other word or phrase for ‘failure’ in my statement of the Pessimistic Verdict;
‘failurePvI’ perhaps. In philosophy, we must at all costs avoid verbal essentialism.)
But I am, just for the purposes of the present reply, going to make one modification
of the Verdict.
Kelly and McGrath rightly raise the question of how the fact that some arguments
for philosophical conclusions have empirical premises (propositions whose truth-
values can be determined only by observation and experiment) factors into the
question of what it is for an argument to be a success or a failure. This is a very
good question, and no doubt it is one I should have paid more attention to than I did.
(I did remark that it was conceivable that an argument that was a success in one
century might be a failure in another, owing to advances in empirical knowledge.)
But for present purposes, I am going to do an end run around this very good
question. I will accomplish this end run simply by replacing the Pessimistic Verdict
with a pessimistic verdict whose scope is restricted to arguments a priori. (An
‘argument a priori’, for present purposes, is an argument none of whose premises
is an empirical proposition.) And here is
The Restricted Verdict: For any empirical proposition p, and any substantive, positive
philosophical thesis ϕ, every known argument a priori for the conditional proposition whose
antecedent is p and whose consequent is ϕ is a failure.73
I will use an example to explain the way in which I mean to apply the Restricted
Verdict. Suppose that a philosopher offers an argument for the existence of a first
mover, and that one of the premises of this argument is ‘The age of the cosmos is
finite’—which I hope everyone will agree is an empirical proposition. (If you don’t

72
‘Known’ is short for ‘well known to philosophers’. An argument that exists only in the mind or the
unpublished ms. of some reclusive philosophical genius is of course “known” to her, but it is not known in
the sense I intend.
73
Note that the following sound argument, ‘The Earth is round; hence, if the Earth is not round, the will
is free’ is not an argument a priori. The valid argument, ‘The Earth is both round and not round; hence, If
Descartes died in Sweden, persons are immaterial substances’ is an argument a priori, but pretty clearly a
failure.
   

think it is, change the example.) Suppose further that this is the only premise of that
argument that is an empirical proposition. Suppose, finally, that the truth of that
premise has been established by empirical investigations that only people who have
mastered some very difficult mathematical physics can understand. If we imagine
Desmond presenting this argument to an audience of ideal agnostics in an attempt to
convince them of the existence of a first mover, must we, in attempting to anticipate
their reaction to the argument, suppose that they are all trained mathematical
physicists—and Desmond and Ophelia as well? If not, how are we to apply our
definitions of success and failure to this argument? I have no good answer to these
questions. But if the Restricted Verdict is correct, I don’t need to apply these
definitions to his argument—not to that argument. Let me explain.
Although Desmond’s argument does not fall within the scope of the Restricted
Verdict, there is an argument that bears an important and intimate relation to his
argument that does fall within its scope, an argument whose conclusion is the
conditional proposition ‘If the age of the cosmos is finite, then there is a first
mover’. If the premises of Desmond’s argument for the existence of a first mover
were A, B, ‘The age of the cosmos is finite’, and D, the “new” argument will have
premises A, B, and D (none of them empirical propositions) and the conclusion ‘If
the age of the cosmos is finite, there is a first mover’. We may call an argument
obtained by this method from a philosophical argument with empirical premises the
“a priori conditional transform” of the original argument—or, for short, its Trans-
form. The Restricted Verdict implies that if an argument for a substantive, positive
philosophical thesis has empirical premises, its Transform is a failure. And if the
Transform of such an argument is a failure, so must the original argument be.
Intuitively, an argument some of whose premises are empirical propositions may
be said to have an empirical part and an a priori part, and it will be a success only if
the agnostics find both parts convincing. The Restricted Verdict implies that what-
ever may be the case with the empirical part of the argument, the agnostics will find
the a priori part unconvincing: no philosophical argument for a substantive, positive
conclusion will be a failure only because it has empirical premises that the agnostics
find unconvincing. If, therefore, the Restricted Verdict is correct, the original Pes-
simistic Verdict is also correct. Thus the “end run”: if the Restricted Verdict is
correct, the problem of specifying the methods Desmond is to use to convince the
agnostics of his empirical premises is irrelevant to the question of the truth or falsity
of the Pessimistic Verdict—for, however that problem is solved, the Pessimistic
Verdict is correct if the Restricted Verdict is correct.
I hereby endorse the Restricted Verdict. In the sequel, however, I shall defend my
argument for the original Pessimistic Verdict. That, after all, is the argument that
Kelly and McGrath have criticized. And my argument for the original Pessimistic
Verdict could easily be modified to produce a parallel and equally good (or equally
bad, as the case may be) argument for the Restricted Verdict.74

74
The modification would be not only easy but not needed at all if it were true that any conditional
proposition whose antecedent was an empirical proposition and whose consequent was a significant
philosophical thesis was itself a significant philosophical thesis. But that is obviously not the case: the
true proposition ‘If the earth is not round, the will is free’ is not a significant philosophical thesis. Cf. n 73.
   

And here is the argument for the original Pessimistic Verdict:

Suppose that there were an argument α for a substantive, positive philosophical thesis ç, an
argument that was well known to philosophers, such that:
If α were presented to an audience of ideal neutral agnostics (sc. about the truth or falsity of ϕ)
in the context of an ideal debate about whether ϕ was true or false, then, when the debate had
run its course, they, the agnostics would uniformly agree that α (considered in the light of
everything said for and against it during the debate) had led them to assent to ϕ.
If there were indeed a known argument that had that counterfactual property, then a
significantly higher proportion of philosophers would accept ϕ than actually do accept ϕ.
(This conditional statement, of course, is the premise on which the argument turns.) Therefore,
no known argument has this feature. There may be arguments subsisting in the Platonic
heaven (or even arguments actually written down somewhere—perhaps in Berkeley’s lost
manuscript on ethics) that have it, but no argument with a substantive, positive conclusion
that is known to philosophers has the power to produce uniform acceptance of its conclusion
among an audience of ideal agnostics when it is presented to them in the context of an ideal
debate. That is to say: Every known argument for any substantive, positive philosophical thesis
is a failure (a failurePvI, if you like).
Now what about that premise? Let us consider an example. According to the study
David Chalmers cites in his essay in the present volume (the 2009 PhilPapers
Survey), 39 percent of us analytical philosophers accept platonism, and 38 percent
accept nominalism. (Or at any rate, this is how matters stood in 2009. I will assume
that these figures are still accurate.) Now suppose that there exists, Platonically
speaking, an argument α for platonism that has never been discovered in the actual
world. α, moreover, is not an argument that only God or an archangel or the Wise
Old Beings from the Galactic Core could comprehend: α is an argument that might
have been discovered by some human philosopher (and it could have been dis-
covered without philosophy’s having gone in some entirely different direction; α
employs only concepts and logical methods that have actually been available to
analytical philosophers since, let us say, the middle 1960s). Let us suppose, in fact,
that it is simple enough that it could successfully be taught to serious undergraduate
students of philosophy. And suppose that α has the conditional property with respect
to determining the outcome of an ideal debate about the truth-value of its conclusion
described above—that it is a philosophical success.
If all that is true, the premise endorses the following counterfactual thesis:
Consider the closest75 possible worlds that satisfy the following conditions:
In those worlds, α was discovered and published in 1970; in them, the state of analytical
philosophy in 1970 was as much like the way it actually was in 1970 as is consistent with the
discovery and publication of α in that year; by 1975, α was well known to all analytical
philosophers interested in ontological questions.

75
The closest such worlds do not include worlds in which all life on earth ended in 1979 or in which all
universities eliminated their philosophy departments in the 1980s or worlds in which in that decade all
philosophers happened to suffer head injuries that caused them to lose their memories or in which
philosophers perversely conspire to ensure that certain arguments they know of “die with them”—conspire
to keep them secret from their students. . . .
   

In those worlds, the following proposition is true:

Now, in the teens of the twenty-first century, the proportion of analytical philosophers who
assent to platonism is significantly higher than 39%.
And isn’t that counterfactual thesis at least highly plausible? Certainly the discovery
of an argument with such marvelous powers of rational persuasion would not have
the consequence that the proportion of philosophers who accept platonism is lower
than it actually is, would it? And if α indeed has what we might call the intrinsic
rational power to produce uniform assent to platonism among a population of ideal
and initially neutral agnostics in the context of an ideal debate, can we really suppose
that its having become known to philosophers-more-or-less-as-they-actually-were-
in-1970 would not have significantly raised the proportion of those philosophers and
their “professional descendants” that were platonists? Were philosophers active in
the 1970s immune to reason to that degree? Were they then—and did they continue
to be—so wedded to the convictions that were theirs before they encountered α that
careful consideration of α would have no power to move them? And, if they were so
wedded to their prior convictions, what about those “descendants” of theirs? What
about the people whose philosophical formation included exposure to α, who first
encountered the argument as “innocents”—when they were undergraduates—and
who later became professional philosophers? (In the closest worlds that satisfy the
above conditions, the majority of philosophers professionally active “in the teens of
the twenty-first century” first encountered α when they were undergraduates.)76
I have presented an argument for the Pessimistic Verdict. The argument, of course,
is a failure. It is a failure by my standards, not to mention the more stringent standard
devised, if not endorsed, by Alvin Plantinga, and that perhaps “maximally stringent”
standard that, according Robert Nozick, he accepted as a young man.77 It is not a
“compelling” argument.78 I think, however, that it is a good argument, an argument
that should lead philosophers to take its conclusion seriously. And, in my view, that is
the best that can be hoped for from a philosophical argument for a substantive,
positive conclusion.

76
I cannot, alas, claim that the “Consequence Argument” has been a philosophical success. But consider
the following two “populations”: professional philosophers with an interest in the free-will problem who
first encountered the Consequence Argument when they were undergraduates or graduate students;
professional philosophers with an interest in the free-will problem who first encountered the Consequence
Argument more than ten years after receiving their doctorates. At every time at which both populations
were large enough for this judgment to be meaningful, the proportion of the former who were then
incompatibilists has been significantly higher than the proportion of the latter population who were then
incompatibilists.
77
See The Problem of Evil, pp. 37–9.
78
I must protest a statement in Kelly and McGrath’s essay, which I will quote. Speaking of some
passages they have quoted from the version of the argument presented in The Problem of Evil, they say, “In
these passages, the fact that the conclusion of a philosophical argument is typically denied by a significant
number of philosophers is treated as compelling evidence that the argument would not convince the ideal
agnostics” (p. 335). No, not compelling evidence. Good evidence, however—evidence that should lead one
to take the Pessimistic Verdict seriously.
   

References
Dennett D., and Steglich-Petersen, A. (2008), The Philosophical Lexicon, <http://www.
philosophicallexicon.com>.
Lewis, D. (1986), On the Plurality of Worlds (Oxford University Press).
Lilla, M. (2001), The Reckless Mind: Intellectuals in Politics (New York Review of Books).
McGinn, C. (1993), Problems in Philosophy: The Limits of Inquiry (Blackwell Publishers).
Paul, L. A. (2012), ‘Metaphysics as Modeling: The Handmaiden’s Tale’, in Philosophical Studies
160: 1–29.
Van Inwagen, P. (1988), ‘The Magnitude, Duration, and Distribution of Evil: A Theodicy’, in
Philosophical Topics XVI, no. 2: 161–87.
Van Inwagen, P. (1996), ‘Review of Problems in Philosophy by Colin McGinn’, in The
Philosophical Review CV: 253–6.
Van Inwagen, P. (2003), ‘The Compatibility of Darwinism and Design’, in N. A. Manson (ed.),
God and Design: The Teleological Argument and Modern Science (Routledge), 348–63.
Van Inwagen, P. (2004), ‘A Theory of Properties’, in D. Zimmerman (ed.), Oxford Studies in
Metaphysics, Vol. 1: 107–38.
Van Inwagen, P. (2006), The Problem of Evil (Oxford University Press).
Van Inwagen, P. (2009a), ‘Listening to Clifford’s Ghost’, in A. O’Hear (ed.), Conceptions of
Philosophy: Royal Institute of Philosophy Supplement 65 (Cambridge University Press),
15–35.
Van Inwagen, P. (2009b), ‘Weak Darwinism’, in L. Caruana (ed.), Darwin and Catholicism
(T&T Clark), 107–20.
Van Inwagen, P. (2011), ‘Relational vs. Constituent Ontologies’, in Philosophical Perspectives
25: 389–405.
Van Inwagen, P. (2014), ‘What is an Ontological Category?’, in Existence: Essays in Ontology
(Cambridge University Press).
Van Inwagen, P. (2015), Metaphysics, 4th edn. (Westview Press).
The Complete Works of Peter van
Inwagen (in English)

(1972) ‘Lehrer on Determinism, Free Will, and Evidence’, in Philosophical Studies 23: 351–7.
(1973) ‘Review of Determinism by Bernard Berofsky’, in The Philosophical Review 82: 399–404.
(1974) ‘A Formal Approach to the Problem of Free Will and Determinism’, in Theoria 40: 9–22.
(1975) ‘The Incompatibility of Free Will and Determinism’, in Philosophical Studies 27: 185–99.
(1977) ‘Counterfactuals with Disjunctive Antecedents’ (with T. McKay), in Philosophical
Studies 31: 353–6.
(1977) ‘Creatures of Fiction’, in American Philosophical Quarterly 14: 299–308.
(1977) ‘Ontological Arguments’, in Noûs 11: 375–95.
(1977) ‘Reply to Gallois’, in Philosophical Studies 32: 107–11.
(1977) ‘Reply to Narveson’, in Philosophical Studies 32: 89–98.
(1978) ‘Ability and Responsibility’, in The Philosophical Review 87: 201–24.
(1978) ‘A Definition of Chisholm’s Notion of Immanent Causation’, in Philosophia 7: 567–81.
(1978) ‘The Possibility of Resurrection’, in International Journal for Philosophy of Religion 9:
114–21.
(1978) ‘Review of Naming, Necessity, and Natural Kinds by Stephen P. Schwartz’, in Inter-
national Studies in Philosophy 10: 197–9.
(1978) ‘Review of Will, Freedom, and Power by Anthony Kenny’, in The Philosophical Review
87: 99–101.
(1979) ‘Laws and Counterfactuals’, in Noûs 13: 439–53.
(1979) ‘Review of The Coherence of Theism by Richard Swinburne’, in The Philosophical
Review 88: 668–72.
(1980) ‘Compatibilism and the Burden of Proof ’, in Analysis 40: 98–100.
(1980) ‘Critical Study of Worlds, Times and Selves by A. N. Prior and Kit Fine’, in Noûs 14: 251–9.
(1980) ‘The Incompatibility of Responsibility and Determinism’, in M. Brand and M. Bradie
(eds.), Action and Responsibility, Bowling Green State University Press.
(1980) ‘Indexicality and Actuality’, in Philosophical Review 89: 403–26.
(1980) ‘Philosophers and the Words “Human Body” ’, in P. van Inwagen (ed.), Time and
Cause: Essays Presented to Richard Taylor, Springer.
(1980) Time and Cause: Essays Presented to Richard Taylor, (ed.), Springer.
(1981) ‘The Doctrine of Arbitrary Undetached Parts’, in Pacific Philosophical Quarterly 62:
123–37.
(1981) ‘Why I Do Not Understand Substitutional Quantification’, in Philosophical Studies 39:
281–5.
(1982) ‘Abstract of Comments: Abnormal Experience and Abnormal Belief ’, in Noûs 16:
13–14.
(1983) An Essay on Free Will, Oxford University Press.
(1983) ‘Fiction and Metaphysics’, in Philosophy and Literature 7: 67–77.
(1984) ‘Dennett on “Could Have Done Otherwise” ’, in Journal of Philosophy 81: 565–7.
(1984) ‘On Computer Simulation of the Presence of a Thinking Being’, in The Syracuse Scholar
Fall: 15–21.
(1984) ‘Review of The Logical Structure of Spinoza’s Ethics, Part I by Charles Jarrett’ (with
J. Bennett), in The Journal of Symbolic Logic 49: 996–7.
        ( )

(1984) ‘Review of Was Spinoza Fooled by the Ontological Argument? by Joel I. Friedman’, in
The Journal of Symbolic Logic 49: 997–8.
(1985) Alvin Plantinga, H. Tomberlin and P. van Inwagen (ed.), Springer.
(1985) ‘Compatibilistic Reflections’, in Australasian Journal of Philosophy 63: 349–53.
(1985) ‘On Two Arguments for Compatibilism’, in Analysis 45: 161–3.
(1985) ‘Plantinga on Trans-World Identity’, in H. Tomberlin and P. van Inwagen (eds.), Alvin
Plantinga, Springer.
(1985) ‘Pretense and Paraphrase’, in P. McCormick (ed.), The Reasons of Art, University of
Ottawa Press.
(1985) ‘Review of The First Person by Roderick M. Chisholm’, in Noûs 19: 122–9.
(1986) ‘How to Reason About Vague Objects’, in Noûs 20: 72–3.
(1986) ‘Review of Philosophical Papers, Vol. I by David Lewis’, in Mind 95: 246–57.
(1986) ‘Two Concepts of Possible Worlds’, in Midwest Studies in Philosophy 11: 185–213.
(1987) ‘When Are Objects Parts?’, in Philosophical Perspectives 1: 21–47.
(1987) ‘Review of Without Proof or Evidence: Essays of O. K. Bouwsma’, in Faith and
Philosophy 4: 103–8.
(1988) ‘And Yet They Are Not Three Gods But One God’, in T. Morris (ed.), Philosophy and
the Christian Faith, University of Notre Dame Press.
(1988) ‘Critical Study of Elbow Room by Daniel Dennett’, in Noûs 22: 609–18.
(1988) ‘How to Reason About Vague Objects’, in Philosophical Topics 16: 255–84.
(1988) ‘The Magnitude, Duration, and Distribution of Evil: A Theodicy’, in Philosophical
Topics 16: 161–87.
(1988) ‘On Always Being Wrong’, in Midwest Studies in Philosophy 12: 95–111.
(1988) ‘The Place of Chance in a World Sustained by God’, in T. Morris (ed.), Divine and
Human Action, Cornell University Press.
(1989) ‘When is the Will Free?’, in Philosophical Perspectives 3: 399–422.
(1990) ‘Four-Dimensional Objects’, in Noûs 24: 245–55.
(1990) ‘Logic and the Free Will Problem’, in Social Theory and Practice 16: 279–90.
(1990) Material Beings, Cornell University Press.
(1990) ‘Response to Slote’, in Social Theory and Practice 16: 385–95.
(1991) ‘Entia Successiva’, ‘Determinism’, in H. Burkhardt and B. Smith (eds.), Handbook of
Metaphysics and Ontology, Philosophia Verlag.
(1991) ‘Modal Inference and the Free Will Problem’, in Daimon Revista Internacional de
Filosofía 3: 57–63.
(1991) ‘The Problem of Evil, the Problem of Air, and the Problem of Silence’, in Philosophical
Perspectives 5: 135–65.
(1991) ‘Searle on Ontological Commitment’, in E. Lepore and R. van Gulick (eds.), John Searle
and His Critics, Wiley-Blackwell.
(1992) ‘Reply to Christopher Hill’, in Analysis 52: 56–61.
(1992) ‘There Is No Such Thing As Addition’, in Midwest Studies in Philosophy 17: 138–59.
(1993) ‘Critical Studies of the New Testament and the User of the New Testament’, in T. Flint
and E. Stump (eds.), Hermes and Athena: Biblical Exegesis and Philosophical Theology,
University of Notre Dame Press. Reprinted as ‘Do You Want Us to Listen to You?’ in
C. Bartholomew, C. S. Evans, M. Healy, and M. Rae (eds.), ‘Behind’ the Text: History and
Biblical Interpretation, Zondervan.
(1993) ‘Critical Study of Identity, Consciousness, and Value by Peter Unger’, in Noûs 27: 373–9.
(1993) ‘Genesis and Evolution’, in E. Stump (ed.), Reasoned Faith, Cornell University Press.
(1993) ‘Naive Mereology, Admissible Valuations, and Other Matters’, in Noûs 27: 229–34.
(1993) ‘Precis of Material Beings’, in Philosophy and Phenomenological Research 53: 683–6.
(1993) ‘Reply to Reviewers’, in Philosophy and Phenomenological Research 53: 709–19.
        ( ) 

(1993) ‘Review of Universes by John Leslie’, in Faith and Philosophy 10: 439–43.
(1994) ‘Composition as Identity’, in Philosophical Perspectives 8: 207–20.
(1994) ‘Doubts About Darwinism’, in J. Buell and V. Hearn (eds.), Darwinism: Science or
Philosophy?, Foundation for Thought & Ethics.
(1994) ‘Not by Confusion of Substance but by Unity of Person’, in A. Padgett (ed.), Reason and
the Christian Religion: Essays in Honor of Richard Swinburne, Clarendon Press.
(1994) ‘Response to Frederick Grinnell’, in J. Buell and V. Hearn (eds.), Darwinism: Science or
Philosophy?, Foundation for Thought & Ethics.
(1994) ‘When the Will is Not Free’, in Philosophical Studies 75: 95–113.
(1995) ‘Dualism and Materialism: Athens and Jerusalem?’, in Faith and Philosophy 12: 475–88.
(1995) God, Knowledge, and Mystery: Essays in Philosophical Theology, Cornell University Press.
(1995) ‘Non Est Hick’, in T. Senor (ed.), The Rationality of Belief and the Plurality of Faith,
Cornell University Press.
(1995) ‘Subsistence’, The Cambridge Dictionary of Philosophy.
(1996) ‘Foreword’, in M. Rea (ed.), Material Constitution, Rowman & Littlefield Publishers.
(1996) ‘ “It is Wrong, Everywhere, Always, and for Anyone, to Believe Anything Upon
Insufficient Evidence” ’, in J. Jordan and D. Howard-Snyder (ed.), Faith, Freedom, and
Rationality, Rowman & Littlefield. Reprinted in E. Stump and M. Murray (eds.), Philosophy
of Religion: The Big Questions, Blackwell, 1999.
(1996) ‘Persistence’, The Encyclopedia of Philosophy (Supplement).
(1996) ‘Quam Dilecta’, in T. Morris (ed.), God and the Philosophers, Oxford University Press.
(1996) ‘Reflections on the Essays of Draper, Gale, and Russell’, in D. Howard-Snyder (ed.), The
Evidential Argument from Evil, Indiana University Press.
(1996) ‘Review of Problems in Philosophy by Colin McGinn’, in The Philosophical Review 105:
253–6.
(1996) ‘Why Is There Anything at All?’, in Proceedings of the Aristotelian Society 70: 95–110.
(1997) ‘Against Middle Knowledge’, in Midwest Studies in Philosophy 21: 225–36.
(1997) ‘Fischer on Moral Responsibility’, in The Philosophical Quarterly 47: 373–81.
(1997) ‘Materialism and the Psychological-Continuity Account of Personal Identity’, in Philo-
sophical Perspectives 11: 305–19.
(1997) ‘Of “Of Miracles” ’, in The Possibility of Resurrection and Other Essays in Christian
Apologetics, Westview Press.
(1997) The Possibility of Resurrection and other Essays in Christian Apologetics, Westview Press.
(1997) ‘Probability and Evil’, in The Possibility of Resurrection and Other Essays in Christian
Apologetics, Westview Press.
(1997) ‘A Reply to Professor Hick’, in Faith and Philosophy 14: 299–302.
(1997) ‘Skeptical of the Skeptics’ (review of The Historical Christ and the Jesus of Faith: The
Incarnational Narrative as History by C. Stephen Evans), in Books and Culture May/June
1997.
(1998) ‘Incarnation and Christology’ ‘Lewis, David’, ‘Trinity’, ‘Resurrection’, Routledge
Encyclopedia of Philosophy.
(1998) ‘Meta-ontology’, in Erkenntnis 48: 233–50.
(1998) Metaphysics: The Big Questions, P. van Inwagen and D. Zimmerman (eds.), Wiley-Blackwell.
(1998) ‘Modal Epistemology’, in Philosophical Studies 92: 67–84.
(1998) ‘The Mystery of Metaphysical Freedom’, in P. van Inwagen and D. Zimmerman (eds.),
Metaphysics: The Big Questions, Wiley-Blackwell.
(1998) ‘The Nature of Metaphysics’, in S. Laurence and C. Macdonald (eds.), Contemporary
Readings in the Foundations of Metaphysics, Wiley-Blackwell.
(1998) ‘Ontological Arguments’, in B. Davies (ed.), Philosophy of Religion: A Guide to the
Subject, Oxford University Press.
        ( )

(1999) ‘Christian Belief and the Platonic Conception of Rationality’, in G. Brüntrup and
R. K. Tacelli (eds.), The Rationality of Theism, Springer.
(1999) ‘Moral Responsibility, Determinism, and the Ability to Do Otherwise’, in Journal of
Ethics 3: 341–3.
(1999) ‘Some Remarks on Plantinga's Advice’, in Faith and Philosophy 16: 164–72.
(2000) ‘Double Dactyls’, in Mind 109: 23–4.
(2000) ‘Free Will Remains a Mystery’, in Philosophical Perspectives 14: 1–19.
(2000) ‘Generalizations of Homophonic Truth-Sentences’, in R. Schantz (ed.), What is Truth?,
De Gruyter.
(2000) ‘Quantification and Fictional Discourse’, in A. Everett and T. Hofweber (eds.), Empty
Names, Fiction and the Puzzles of Non-existence, Center for the Study of Language and
Information.
(2000) ‘Temporal Parts and Identity Across Time’, in The Monist 83: 437–59.
(2001) ‘The Argument from Particular Horrendous Evils’, in Proceedings of the American
Catholic Philosophical Association 74: 65–80.
(2001) ‘Genes, Statistics, and Desert’, in D. Wasserman and R. Wachbroit (eds.), Genetics and
Criminal Behavior, Cambridge University Press.
(2001) ‘ “I am Elizabeth Anscombe” is Not an Identity Proposition’, in Metaphysica 2: 5–8.
(2001) ‘ “It is Nonsensical to Speak of the Total Number of Objects” ’, in U. Meixner (ed.),
Metaphysik im postmetaphysichen Zeitalter, öbvahpt Verlagsgesellschaft.
(2001) Ontology, Identity, and Modality, Cambridge University Press.
(2002) ‘Can Variables Be Explained Away?’, in Facta Philosophica 4: 3–9.
(2002) ‘The Number of Things’, in Philosophical Issues 12: 176–96.
(2002) ‘Review of Persons and Bodies by Lynne Rudder Baker’, in The Philosophical Review
111: 138–41.
(2002) ‘What Do We Refer to When We Say “I”?’, in R. Gale (ed.), The Blackwell Guide to
Metaphysics, Wiley-Blackwell.
(2002) ‘What is the Problem of the Hiddenness of God?’, in D. Howard-Snyder and P. Moser
(eds.), Divine Hiddenness: New Essays, Cambridge University Press.
(2002) ‘Why Vagueness is a Mystery’, in Acta Analytica 17: 11–17.
(2003) ‘ “Carnap” and “The Polish Logician” ’, in Acta Analytica 17: 7–17.
(2003) ‘The Compatibility of Darwinism and Design’, in N. Manson (ed.), God and Design: The
Teleological Argument and Modern Science, Routledge.
(2003) ‘Existence, Ontological Commitment, and Fictional Entities’, in M. Loux and
D. Zimmerman (eds.), The Oxford Handbook of Metaphysics, Oxford University Press.
(2003) ‘Three Persons in One Being: On Attempts to Show that the Doctrine of the Trinity is
Self-Contradictory’, in M. Stewart (ed.), The Trinity, Springer.
(2004) ‘The Argument from Evil’, in P. van Inwagen (ed.), Christian Faith and the Problem of
Evil, Wm. B. Eerdmans Publishing Company.
(2004) ‘Can Science Disprove the Existence of God?’, in Philosophic Exchange: Annual
Proceedings 34: 38–53.
(2004) Christian Faith and the Problem of Evil (ed.), Wm. B. Eerdmans Publishing Company.
(2004) ‘Freedom to Break the Laws’, in Midwest Studies in Philosophy 28/1: 334–50.
(2004) ‘Human Destiny’, in W. Mann (ed.), The Blackwell Guide to the Philosophy of Religion,
Wiley-Blackwell.
(2004) ‘Properties’, in T. M. Crisp, D. Davidson, and D. Vander Laan (eds.) Knowledge and
Reality, Springer. (This is a shorter version of ‘A Theory of Properties’.)
(2004) ‘The Self: The Incredulous Stare Articulated’, in Ratio 57: 478–91.
(2004) ‘A Theory of Properties’, in Oxford Studies in Metaphysics 1: 107–38.
        ( ) 

(2004) ‘Van Inwagen on Free Will’, in J. K. Campbell, M. O’Rourke, and D. Shier (eds.),
Freedom and Determinism, A Bradford Book.
(2005) ‘Is God an Unnecessary Hypothesis?’, in A. Dole and A. Chignell (eds.), God and the
Ethics of Belief, Cambridge University Press.
(2005) ‘Reply to Sean Carroll’, in Faith and Philosophy 22: 636–40.
(2006) ‘Can Mereological Sums Change Their Parts?’, in Journal of Philosophy 103: 614–30.
(2006) ‘Names for Relations’, in Philosophical Perspectives 20: 535–477.
(2006) The Problem of Evil, Oxford University Press.
(2006) ‘What is Analytic Philosophy? What is Naturalism?’, in A. Corradini, S. Galvan, and
E. J. Lowe (eds.), Analytic Philosophy without Naturalism, Routledge.
(2007) ‘Impotence and Collateral Damage: One Charge in Van Fraassen’s Indictment of
Analytical Metaphysics’, in Philosophical Topics 35: 67–82 (appeared in 2010).
(2007) ‘A Materialist Ontology of the Human Person’, in P. van Inwagen and D. Zimmerman
(eds.) Persons: Human and Divine, Oxford University Press.
(2007) ‘Metaphysics’, Stanford Encyclopedia of Philosophy.
(2007) Persons: Human and Divine, P. van Inwagen and D. Zimmerman (eds.), Oxford
University Press.
(2007) ‘Plantinga’s Replacement Argument’, in D. Baker (ed.), Alvin Plantinga, Cambridge
University Press.
(2008) ‘The Consequence Argument’, in P. van Inwagen and D. Zimmerman (eds.), Meta-
physics: The Big Questions (2nd edn.), Wiley-Blackwell.
(2008) ‘How to Think About the Problem of Free Will’, in Journal of Ethics 12: 327–41.
(2008) Metaphysics: The Big Questions (2nd edn.), P. van Inwagen and D. Zimmerman (eds.),
Wiley-Blackwell.
(2008) ‘McGinn on Existence’, in The Philosophical Quarterly 58: 36–58.
(2008) ‘Quine’s 1946 Lecture on Nominalism’, in Oxford Studies in Metaphysics 4: 125–42.
(2008) ‘Was George Orwell a Metaphysical Realist?’, in Philosophia Scientiae 1: 161–85.
(2008) ‘What Does an Omniscient Being Know About the Future?’, in Oxford Studies in
Philosophy of Religion 1: 216–30.
(2009) ‘Being, Existence, and Ontological Commitment’, in D. Chalmers, D. Manley, and
R. Wasserman (eds.), Metametaphysics, Oxford University Press.
(2009) ‘Explaining Belief in the Supernatural: Some Thoughts on Paul Bloom’s “Religious Belief as
an Evolutionary Accident”’, in J. Schloss and M. Murray (eds.), The Believing Primate: Scientific,
Philosophical, and Theological Reflections on the Origin of Religion, Oxford University Press.
(2009) ‘God and Other Uncreated Things’, in K. Timpe (ed.) Metaphysics and God: Essays in
Honor of Eleonore Stump, Routledge.
(2009) ‘The Hiddenness of God’, in K. Timpe (ed.) Arguing About Religion, Routledge.
(2009) ‘Indeterminacy and Vagueness: Logic and Metaphysics’, in European Journal for
Philosophy of Religion 1: 1–19.
(2009) ‘Listening to Clifford’s Ghost’, in Royal Institute of Philosophy Supplement 85/65.
(2009) Metaphysics (3rd edn.), Westview.
(2009) ‘The New Anti-Metaphysicians’, in Proceeding and Addresses of the American Philo-
sophical Association 83/2: 45–61.
(2009) ‘Some Remarks on the Modal Ontological Argument’, in Philo 12: 217–27.
(2009) ‘Weak Darwinism’, in L. Caruana (ed.), Darwin and Catholicism: The Past and Present
Dynamics of a Cultural Encounter, Bloomsbury T&T Clark.
(2010) ‘Changing the Past’, in Oxford Studies in Metaphysics 5: 3–28.
(2010) ‘A Kind of Darwinism’, ‘Darwinism and Design’, ‘Science and Scripture’, in M. Stewart
and (eds.), Science and Religion in Dialogue, Vol. 2, Wiley-Blackwell.
        ( )

(2010) ‘The Inspiration of the Hebrew Bible and the Morality of God’s Commands’, in The
Harvard Ichthu Spring: 12–18.
(2010) ‘Peter van Inwagen’, in A. Steglich-Peterson (ed.) Metaphysics: Five Questions, Auto-
matic Press/VIP.
(2010) ‘We’re Right, They’re Wrong’, in R. Feldman and T. Warfield (eds.), Disagreement,
Oxford University Press.
(2011) ‘Comments on Professor Curley’s “The God of Abraham, Isaac, and Jacob” ’, in
M. Murray and M. Rea (eds.), Divine Evil? The Moral Character of the God of Abraham,
Oxford University Press.
(2011) ‘A Promising Argument’, in R. Kane (ed.), The Oxford Handbook of Free Will, 2nd edn.,
Oxford University Press.
(2011) ‘Relational vs. Constituent Ontologies’, in Philosophical Perspectives 25: 389–405.
Revised and reprinted as ‘Against Ontological Structure’, in G. Galluzzo and M. Loux
(eds.), The Problem of Universals in Contemporary Thought, Cambridge University Press.
(2012) ‘Causation and the Mental’, in K. J. Clark and M. Rea (eds.), Reason, Metaphysics, and
Mind: New Essays on the Philosophy of Alvin Plantinga, Oxford University Press.
(2012) ‘Russell’s China Teapot’, in D. Łukasiewicz and R. Pouivet (eds.), The Right to Believe:
Perspectives in Religious Epistemology, Ontos Verlag.
(2012) ‘ “Who Sees Not that All the Dispute is About a Word?”: Some Thoughts on Bennett’s
“Proxy ‘Actualism’ ” ’, in Hungarian Philosophical Review 3: 69–81.
(2013) ‘CS Lewis’ Argument Against Naturalism’, in Res Philosophica 90: 113–24.
(2013) ‘A Dialogue on Free Will’, in Methode 3: 212–21.
(2013) ‘A Modal Argument for Incompatibilism’, in P. Russell and O. Deery (eds.), The
Philosophy of Free Will, Oxford University Press.
(2013) ‘Three Versions of the Ontological Argument’, in M. Szatkowski (ed.), Ontological
Proofs Today, De Gruyter.
(2014) ‘Alston on Ontological Commitment’, in Existence: Essays in Ontology, Cambridge
University Press.
(2014) ‘Dispensing with Ontological Levels: An Illustration’, in Disputatio VI, 38: 25–43.
(2014) Existence: Essays in Ontology, Cambridge University Press.
(2014) ‘Fictional Nominalism and Applied Mathematics’, in The Monist 97: 479–502.
(2014) ‘Inside and Outside the Ontology Room’, Introduction to Existence: Essays in Ontology,
Cambridge University Press.
(2014) ‘Modes of Being and Quantification’, in Disputatio VI, 38: 1–24.
(2014) ‘What is an Ontological Category?’, in Existence: Essays in Ontology, Cambridge
University Press.
(2015) ‘Did God Create Shapes?’, in Philosophia Christi 17: 285–90.
(2015) ‘A Reply to Craig’, in Philosophia Christi 17: 299–305.
(2015) ‘Some Thoughts on An Essay on Free Will’, in The Harvard Review of Philosophy 22: 16–30.
(2016) ‘The Rev’d Mr Bayes and the Life Everlasting’, in M. Bergmann and J. Brower (eds.),
Reason and Faith: Themes from Richard Swinburne, Oxford University Press.

Acknowledgments
Compiled using the PhilPapers and http://andrewmbailey.com/pvi/ as primary sources. Thanks to
Kelli Asklar and Jacqueline DiFonzo for their assistance.
Index
abstracta 3, 5, 11–31, 32–61, 62–79, 112–16,
120, 121, 132–3, 244, 258, 266–81, 307,
345–52, 386
adaptation 163–4, 227–40, 241–62, 361–2
agnosticism 154–5, 173, 289–91, 299–323,
324–39, 352–66, 369, 373–83
An Essay on Free Will 6, 95–117, 118–39,
140–56, 395
Anselm 6, 263–73, 365
Antony, Louise 6, 344, 352–61
Aquinas 11, 186–208, 245 n.15, 310, 365
argumentative success/failure see Method
argument from evil see Problem of Evil
atheism 159–85, 203 n.60, 227–40, 241–62, 272,
281–2, 284, 286, 288 n.4, 309, 352–66,
375–6, 388
atonement 6, 186–208
being/existence 5, 38, 68, 82–4, 86–7, 89–90,
263–73
Chalmers, David 7, 149–50, 306 n.16, 344,
366–73
chance 97, 100 n.18, 164, 209–26, 227–40, 278
n.1, 347, 354 n.21, 361–6
Chisholm, Roderick 2 n.4, 14 n.12, 19, 26 n.57,
27, 28 n.61, 29, 103 n.22, 352 n.15
cognitive-ability assumption 263–73
compatibilism 5–6, 95–117, 118–39, 140–56,
181 n.24, 281, 286, 288, 309, 336, 361,
392 n.76
consequence argument 6, 118–39, 140–56, 301,
361, 392 n.76
constituent ontology 5, 11–31, 32–61, 62–79,
348–52, 378
contextualism 6, 96, 112–16, 281
convergence 277–98, 305, 315, 336
defense 159–85, 352–61
design argument 162–6, 173–4, 177, 227–40
disagreement 7, 181, 277–98, 309 n.24, 313,
317, 335–8, 344, 348–50, 366–73, 377–8
Draper, Paul 6, 161 n.5, 241–62
drift (evolutionary) 228 n.4, 230, 237
Eckhart, Meister 192–200
evidential argument from evil 159–85, 352–61
evolution 6, 23, 162–6, 173–4, 177, 227–40,
241–62, 292, 295–7, 361–6
existence in intellectu 263–73
existence in re 263–73
failure see method
fine-tuning argument 177 n.18
fixity of the past 140–56
free will 5–6, 95–117, 118–39, 140–56, 171–84,
200–2, 227, 237, 239–40, 278, 281–2, 292,
294, 309, 359–61, 365, 370, 375, 377
Goldman, Alvin 210–13, 224
grace 200–7
great making principle 266–7
grounding 6, 64, 119, 128–38
growing block 80–91
hypertime 80–91
identity of indiscernibles 47–59
incompatibilism 5–6, 95–117, 118–39, 140–56,
181 n.24, 281, 286, 288, 309, 336, 361,
392 n.76
Keller, John A. 7, 343, 344, 373–83
Kelly, Thomas 7, 289, 300 n.4, 304 n.12, 305
n.15, 310 n.26, 313 n.34, 344, 378 n.55,
384–92
Leibniz 95, 103 n.22, 252–7, 261
Lewis, David v, 3 n.6, 4, 15–17, 24–5, 27, 29, 32,
46, 84, 101 n.19, 122 n.19, 141 n.3, 142 n.6,
159 n.1, 216 n.20, 219, 239, 284, 287, 304
n.13, 312, 337, 351 n.14, 361
Loux, Michael 5, 36, 37, 46 n.24, 63, 64, 70, 343,
344–8, 350 n.13, 351
Material Beings 5, 63 n.2, 95, 270 n.16, 325,
326, 396
McGinn, Colin 277, 295, 297, 366, 370
McGrath, Sarah 7, 289, 300 n.4, 305 n.15,
310 n.26, 313 n.34, 326, 344, 378 n.55,
384–92
Meinong, Alexius 6, 263–73
mereological bundle theory 32–61, 348–52
mereology 17–19, 32–61, 62–79, 348–52
metaphilosophy see method
method 6, 7, 144–7, 153–5, 159–85, 235–6,
254–7, 277–98, 299–323, 324–39, 366–92
Mind Argument 6, 118–39
modality 118–39, 140–56, 168–84,
263–73, 383
morality 6, 99–101, 115, 159–62, 166–84, 188,
192–4, 196–8, 201, 209–26, 278, 281, 283,
294, 309, 326, 352–61
 
moral responsibility 6, 95–117, 167, 209–26,
288 n.4
movable present 80–91
mystery 25, 57, 72, 74, 159–85, 352–61, 373
natural selection 163–5, 173, 177, 227–40,
241–2, 283, 295, 361–6, 370
omniscience 227, 237, 239–40, 332, 387
ontological argument 6, 256, 263–73
Parfit, Derek 210, 212, 217–23
Paul, L. A. 5, 17, 348–52
Plantinga, Alvin 19, 27, 29, 120 n.10, 159 n.2,
186 n.6, 249–50, 343, 380, 383, 392
presentism 80–91, 288 n.4, 292
probability 52, 160–2, 171, 173, 209–26,
227–40, 241–62, 308–9, 311–12, 354 n.21,
361–6, 372 n.51, 382, 385
problem of evil 6, 159–85, 186–208, 223–4, 227,
238, 288 n.4, 325, 329–30, 336–8, 345 n.1,
352–61, 383 n.65
progress 7, 277–98, 315–17, 366–73
properties see abstracta
Proslogion II 263–73
Quine, W. V. O 1, 4, 13, 15, 27, 29, 236, 288,
290, 294, 295, 344, 378
rationality/reasons 95–117, 160 n.4, 161, 165–6,
172, 174–5, 209–26, 246, 251–7, 290, 295–6,
299–305, 306 n.16, 307–13, 330–7, 367–8,
387, 392
relational ontology 5, 11–31, 34–6, 63–4, 66–7,
69, 78, 344–8, 348–52
Rosenberg, Alex 6, 361–6
salvation 186–208
Sartorio, Carolina 210, 213–17, 224–5
success see method
suffering 159–85, 186–208, 209, 211, 214,
223–4, 238 n.10, 352–61
theism 6, 159–85, 186–208, 227–40, 241–62,
263–73, 352–61, 361–6, 385, 388
theism and evolution 227–40, 241–62, 361–6
The Problem of Evil 6, 7, 147 n.12, 154, 159–85,
187 n.1, 210 n.3, 223 n.31, 228 n.3, 277–98,
299–323, 324–39, 399
time travel 80–91
universals see properties
unmediated causal powers 263–73
voting 209–26
Woltersdorff, Nicholas 17–20, 35 n.6, 63 n.1, 64