Miguel Ángel Fernández Vargas

Performance Epistemology
Performance
Epistemology
Foundations and Applications
edited by
Miguel Ángel Fernández Vargas
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 2016
The moral rights of the authors have been asserted
First Edition published in 2016
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: 2015942037
ISBN 978–0–19–874694–2
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.
Contents
Acknowledgmentsvii
Contributorsix
Introduction—Performance Epistemology: Foundations and Applications 1

Part I. Foundational Issues

1. Epistemic Competence and Judgment 19
Ernest Sosa
2. Dispositional Robust Virtue Epistemology versus Anti-luck Virtue
Epistemology31
Jesper Kallestrup and Duncan Pritchard
3. Knowledge, Virtue, and Safety 51
John Greco
4. Abilities, Competences, and Fallibility 62
Alan Millar
5. Against Actual-world Reliabilism: Epistemically Correct
Procedures and Reliably True Outcomes 83
Peter J. Graham
6. Who Knows? 106
Baron Reed
7. Knowledge as Achievement, More or Less 124
John Turri
Part II. Applications and Problems

8. Intuitions and the Understanding 137
Paul A. Boghossian
9. Rising above the Animals: The Search for Intellectual Assurance 151
Richard Fumerton
10. Epistemic Agency 167
Hilary Kornblith
11. The Value of Reflection 183
Stephen R. Grimm
vi Contents
12. The Epistemology of Stupidity 196

Pascal Engel
13. Epistemic Circularity and Virtuous Coherence 224
Ram Neta
Index249
Acknowledgments
The idea of producing this book started to grow in 2011, during the Conference
“The Present and Future of Virtue Epistemology” that took place at the Instituto de
Investigaciones Filosóficas of UNAM, in Mexico City, to celebrate the 70th birthday
of Ernest Sosa, the father of contemporary Virtue Epistemology. Sosa himself was
among the speakers in the Conference and versions of most of the essays presented on
that ocassion are included in this volume. Like that Conference, this book owes its
existence to Sosa’s epistemology, for without it the book will lose most of its subject
matter; to that extent Sosa’s work is fundamental to understanding and discussing
Performance Epistemology. I want to express my gratitude to Ernie for his kind advice
throughout the process that led to this book.
I also want to thank Peter Momtchiloff at Oxford University Press for believing
in this project, and to all those at Oxford University Press who where involved in
putting this volume together. Thanks also go, of course, to the contributors for their
patience throughout this project.
Ciudad Universitaria, UNAM
Mexico City
October 2015
Contributors
Paul A. Boghossian, New York University

Pascal Engel, EHESS, Paris
Miguel Ángel Fernández Vargas, Universidad Nacional Autónoma de México
Richard Fumerton, University of Iowa
Peter J. Graham, University of California, Riverside
John Greco, St. Louis University
Stephen R. Grimm, Fordham University
Jesper Kallestrup, University of Edinburgh
Hilary Kornblith, University of Massachusetts, Amherst
Alan Millar, Stirling University
Ram Neta, University of North Carolina
Duncan Pritchard, University of Edinburgh
Baron Reed, Northwestern University

Ernest Sosa, Rutgers University
John Turri, University of Waterloo
Introduction
Performance Epistemology: Foundations
and Applications
I
This collection of essays brings together previously unpublished material by leading
epistemologists who discuss key issues concerning the foundations and applications
of a prominent branch of virtue epistemology that we will call “performance-based
epistemology.”
Performance-based epistemology (“PBE,” hereafter) conceives the normativity
involved in epistemic evaluation as a special case of a pattern of evaluation that can be
applied to any domain where there are agents that carry out performances with an aim.
For example, it conceives believing and judging as types of performances with an epis-
temic aim that are carried out by persons. Evaluating beliefs epistemically becomes
then a task with essentially the same structure as that evaluating athletic, culinary, or
any other sort of performance; in all cases the performance in question is evaluated in
terms of how it relates to certain relevant competences and abilities of the subject that
carries it out. In this way, PBE locates epistemic evaluation within a general normative
pattern that spreads across many different human activities and disciplines.
PBE is the present-day result of the evolution of what is normally called “virtue
reliabilism.” Virtue reliabilism was the original virtue-theoretic approach introduced
in contemporary epistemology through the writings of Ernest Sosa during the
1980s;1 later, alternative aretaic approaches were introduced in epistemology, the most
important of them are usually grouped together under the label “character-based virtue
epistemology.” But those alternative approaches developed partly due to alleged limita-
tions of virtue reliabilism. In view of the growth of those alternative aretaic approaches
it now seems to be the appropriate time to take a close look at what has happened to the
original virtue-theoretic insights in contemporary epistemology, which is to say that it
1
Sosa’s collection of such writings is his 1991.
2 Miguel Ángel Fernández Vargas
is the appropriate time to take a close look at PBE, for it embodies the present-day form
of virtue reliabilism. Through a systematic discussion of the foundations of PBE and of
its applications to different problems, we will gain a clearer idea of how much the origi-
nal virtue-theoretic insights in epistemology can offer to epistemologists today. This
will contribute to a better delineation of the proper thematic scope of PBE, that is, the
proper range of epistemological topics that PBE is well equipped to handle. A similar
contribution would be desirable concerning character-based approaches in virtue epis-
temology, a better delineation of their proper thematic scope.
Standard introductions to virtue epistemology contrast PBE and character-based
virtue epistemology in terms of how the approaches differ in their conception of
epistemic virtue and in terms of certain meta-philosophical ambitions. While PBE
conceives intellectual virtues as reliable competences that include even basic faculties,
like perception and memory, character-based approaches conceive them as refined
traits of character, like open-mindedness and intellectual fairness, and emphasize their
contribution to making the subject a responsible epistemic agent, marginalizing the
issue of their reliability. While PBE aims to address some of the perennial problems
of epistemology, like the definition of knowledge and justification, skepticism and
epistemic circularity, character-based approaches aim to steer clear of such tradi-
tional problems and instead seek to broaden epistemology’s scope by addressing
non-traditional issues like the value of understanding and wisdom and the nature of
individual virtues and vices. Instead of adopting this or another way of drawing the
contrast between PBE and character-based approaches,2 I want to emphasize that
the questions concerning the extent to which the proper domains of application of
those approaches are disjointed and the extent to which work within one approach can
benefit from work done in the other, remain open questions. But in order to answer
these questions properly it is necessary first to critically examine the conceptual issues
that lie at the foundation of the theoretical frameworks of each approach and to have a
clear understanding of the problems that each approach faces when applied to the
range of problems that it regards as its proper domain. The present volume is an
attempt to carry out these critical tasks with respect to performance-based epistemol-
ogy. Accordingly, the volume is divided into two parts; the essays in Part I examine
some foundational issues in the conceptual framework of PBE and the essays in Part II
discuss the application of PBE to some outstanding problems in contemporary episte-
mology. Overall, the volume can therefore be seen as an examination of the explana-
tory power of the virtue-theoretic insights contained in PBE with respect to a wide
range of epistemological issues.
All the essays in this volume make substantial reference to Ernest Sosa’s work to
criticize it, to propose alternatives to his views, or to use and improve on some of his
proposals. This is unavoidable and for very good reason. Given that PBE originates in
2
See Greco and Turri 2011, Baehr 2004, Sosa 2015, Pritchard 2009, and Battaly 2008 for discussion rel-
evant for the distinction between PBE and character-based approaches.
introduction 3
Sosa’s work and that he has been its main architect over the years, a volume dedicated
to discussing the present state of PBE that didn’t take Sosa’s contributions heavily into
account would be inconceivable.
II
As I pointed out above, a hallmark of PBE is that it conceives beliefs as a type of perfor-
mance and then applies to them a form of evaluation that applies to performances
generally. In evaluating performances of an agent we investigate the ways in which
they relate to the agent’s competences, for those relations between competences and
performances are thought to determine various normative statuses of the perfor-
mances. When we apply to belief this conception of how performances are evaluated
generally two sets of conceptual issues become of fundamental importance: one is a
group of issues concerning the nature of the cognitive competences an agent may pos-
sess, the other is a group of issues concerning the nature of the normative-inducing
relations between those competences and the performances that they help to produce.
The essays in Part I of the present volume are dedicated to discussing those sets of
conceptual issues lying at the foundation of PBE.
Ernest Sosa has proposed what has proven to be a very fruitful framework to study
the conceptual issues mentioned in the previous paragraph. We can term Sosa’s
framework the SSS + AAA approach (pronounced “Triple S plus Triple A”). The SSS
part in Sosa’s approach deals with the nature of competences, and the AAA part
with the nature of the normative-inducing relations between competences and
performances.3
The three S’s make reference to the structure Sosa thinks that competences in
general, and cognitive competences in particular, have. In his view, a competence is
constituted by an innermost Skill (the first ‘S’), that one retains even when placed
in very unfavorable circumstances for its exercise and when one is in a state that is
also unfavorable for its manifestation. For example, a competent driver retains an
innermost skill to drive even when he is driving in complete darkness or when he is
completely drunk. But an innermost skill can be combined with a favorable inner
State of the agent (the second ‘S’), which results in possessing an inner competence,
this would be, for example, the competence that a driver has when he not only possesses
his driving skill but is also not drunk and is otherwise in a perfectly good inner condi-
tion to drive. An inner competence, however, is not yet what Sosa calls a “complete com-
petence.” To be this a combination of Skill and State has also to combine with a good
Situation (the third ‘S’): in addition to being in a good state and having a good skill,
the driver also needs to be in a good situation for the exercise of her inner competence,
for example, with good roads and good lighting conditions. In order to possess the
3
Sosa’s development of the SSS + AAA approach spreads across all his work on epistemology. His most
up-to-date and systematic presentations of the approach can be found in his 2011 and 2015.
complete competence to drive well and safely, a driver needs the relevant triple S of that
competence to obtain.
The three A’s in Sosa’s framework make reference to three normative statuses that
performances in general, and cognitive performances in particular, may have in virtue
of being related in certain ways with the agent’s competences. A performance can be
Accurate (the first ‘A’) when it succeeds in its aim, whether such success derives or not in
an appropriate way from the agent’s competences. So, mere accuracy of a performance
tells us very little about the standing of the agent’s competences. But a performance can
also be Adroit (the second ‘A’), and adroitness is a normative status of a performance
that does tell more about the agent’s competences, for a performance is adroit when it
manifests some relevant competences of the agent. For example, in Sosa’s favorite
example, when an archer hits the bulls-eye by virtue of her long-trained competence,
her shot was not just accurate but also adroit. An adroit performance, however, can still
suffer from a normative failing if it is not also Apt (the third ‘A’), for a performance can
manifest the agent’s competences but its success still not manifest the agent’s compe-
tences; for the success of a performance to manifest the agent’s competences it needs to
be accurate because it is adroit, it has to succeed because it manifests the agent’s com-
petences. The archer’s shot can be adroit because its execution manifests her compe-
tences, but if compensating gusts of wind are causally responsible of the arrow hitting
the bulls-eye, then the shot is not apt, for its success does not manifest her competence;
for the success of her shot to manifest her competences it needs to hit the bulls-eye
because it was adroit, it has to succeed because it manifests her shooting competence.
Sosa’s SSS+AAA approach provides an elegant framework to begin the systematic
study of the foundational conceptual issues discussed in Part I of this volume and, as
will become evident, the contributors to this Part take Sosa’s views heavily into account.
It is therefore natural that the Part begins with a contribution by Sosa himself.4
In his essay “Epistemic Competence and Judgment,” Sosa shows how his SSS con-
ception of the structure of competences helps to appreciate how apt success, whether
apt action or apt belief, is consistent with certain kinds of luck. The compatibility of
such achievements with luck is of great importance especially in the case of apt belief,
given that apt belief is regarded by Sosa as a sort of knowledge that he calls “animal
knowledge,” and knowledge in general is regarded as incompatible with epistemic
luck. The types of luck that in Sosa’s view are compatible with animal knowledge are
those that would affect possession of each of the three S’s constitutive of a complete
competence. Animal knowledge requires that one’s true belief manifests one’s relevant
complete competence, and as we have seen a complete competence is constituted by a
relevant skill, state, and situation; he holds that one could be lucky in having (or being
4
In what follows I do not intend to give a full summary of the issues addressed in each of the essays
included in this volume, what I do instead is to mention some of the themes and theses that are discussed
in the essays that help to see either some unifying concerns of the authors working in the field or some
outstanding challenges posed to performance-based epistemology.
introduction 5
in) any of the appropriate S’s, and therefore lucky in having a complete competence,
and yet gain animal knowledge through the exercise of such complete competence. So
animal knowledge is compatible at least with those types of luck. One way of illustrat-
ing this compatibility is by imagining close counterfactual scenarios where one lacks
any of the S’s constitutive of one’s complete competence and one avoids being in any of
them by sheer luck. Avoiding such close threats by luck is compatible with animal
knowledge because so avoiding them leaves intact the actual SSS structure of the rele-
vant competence, the exercise of which is what matters for animal knowledge. Sosa
approaches the compatibility of apt action and lucky possession of the relevant compe-
tence through the question whether “apt means–end action requires animal knowl-
edge that the means will yield the end,” and his answer is “no.” His answer is based on
cases where the agent’s means–end belief is Gettiered, and therefore fails to be knowl-
edge, but the action based on that belief is nevertheless apt. Imagine, for example, a
prisoner who has a Gettiered belief that his cell will be opened at midnight (his guard
lied to him that this will be so but by accident it happens to be so); the prisoner acts on
that belief and succeeds. Sosa holds that the prisoner’s action is apt despite being based
on an inapt belief because the inaptness of his belief amounts to the fact that the pris-
oner finds himself by sheer luck in the appropriate Situation constitutive of his com-
plete competence to escape. But we’ve seen that one can be lucky in possessing any of
the S’s constitutive of one’s competence and still achieve apt success through its exer-
cise, and this is precisely what the prisoner achieves: apt action through a complete
competence he possesses by mere luck. Despite that apt belief and apt action can be
compatible with one being lucky in possessing the relevant complete competences, it is
very important for Sosa to insist that apt success is “enhanced,” it becomes something
even better, if it is free from reliance on such luck. To achieve this one has to ascend to
the epistemic second-order and in the case of belief engage in judging, which he con-
ceives as a performance that aims not only to believe the truth but also to believe it
reliably and competently. Such “endeavor”, as he calls it, is in the second-order because
when one engages in it “one must consider one’s relevant, first-order, complete compe-
tences, and the first-order options of affirming, denying and double-omitting.” It is a
hallmark of Sosa’s normative theory that any apt success will not be “fully apt” until it is
the result of a second-order endeavor of that sort.
In their essay “Dispositional Robust Virtue Epistemology versus Anti-luck Virtue
Epistemology,” Jesper Kallestrup and Duncan Pritchard (“K & P”, hereafter) criticize
Sosa’s analysis of knowledge for its alleged inability to deal with a couple of phenomena
they call “negative” and “positive epistemic dependence.” What is at stake in this debate
between K & P and Sosa is whether a virtue-theoretic condition, like aptness, is neces-
sary and sufficient for knowledge. We’ve seen that Sosa thinks that it is, since animal
knowledge is just apt belief; they think it isn’t, for they argue that “negative epistemic
dependence” shows that aptness is insufficient for knowledge and “positive epistemic
dependence” shows that it is unnecessary. In negative epistemic dependence “an
agent . . . fails to know because of factors outwith her cognitive agency.” Their thought is
that the environment can be unfavorable to the agent in presenting her risks of error
that undermine knowledge by making her belief unsafe (i.e. one that could easily have
been false), without undermining aptness; hence aptness is insufficient for knowledge,
knowledge may require more than apt belief. We saw that for Sosa the modal proximity
of certain threats that endanger possession of a complete competence is compatible
with animal knowledge. K & P agree that the modal proximity of such threats leaves
the aptness of a belief intact, but think that it may nevertheless undermine knowledge
because knowledge requires more than apt belief, it requires safety, which is itself a
modal condition that can be undermined by the modal proximity of threats of the sort
considered by Sosa. In contrast, in positive epistemic dependence “an agent . . . possesses
knowledge on account of factors outwith her cognitive agency” and when this phe-
nomenon occurs the environment is favorable to the agent in such a way that despite
that her belief doesn’t qualify as apt, she nonetheless has knowledge; hence aptness is
unnecessary for knowledge, knowledge may require less than apt belief. They focus on
what is generally regarded as a paradigm of this phenomenon: casual testimonial
knowledge. They argue that when an agent gains knowledge by simply relying on what
someone else tells her, the cognitive success of the belief she thereby forms does not
manifest her cognitive competences (since she is merely listening), and therefore her
belief should not qualify as apt, but undoubtedly it is a case of knowledge. K & P think
that together, positive and negative epistemic dependence show that knowledge some-
times is more and sometimes is less than apt belief, and they propose an analysis that
preserves some fundamental insight from performance-based epistemology but also
incorporates a modal condition for knowledge, they call their proposal “anti-luck virtue
epistemology.” In their analysis knowledge is safe cognitive success that is adroit, it is
not required that it is also apt. They explain how this conception of knowledge deals
successfully with the cases of positive and negative epistemic dependences. Despite that
the agent has an apt belief in a case of negative epistemic dependence, their analysis
rules it out as a failure to be knowledge because the belief fails to be safe; and despite that
the agent doesn’t have an apt belief in a case of positive epistemic dependence, their
analysis rules it in as a case of knowledge because the belief is at least adroit and also safe.
K & P’s critique of Sosa is meant to be applicable to other performance epistemology
theorists who also have the “aspiration to exclusively analyze knowledge in terms of a
virtue-theoretic condition, thereby avoiding a separate anti-luck requirement.” But
they are aware that sometimes Sosa seems to recognize an anti-luck or modal condi-
tion for knowledge, like safety. In that case their critique of him seems instead to
be that the addition of such a condition is merely ad hoc. In his essay “Knowledge,
Virtue, and Safety” John Greco argues that the correct analysis of aptness (which is one
inspired by some of Sosa’s writings) actually guarantees that a safety condition is ful-
filled; so the modal condition for knowledge, far from being a mere ad hoc addition
to the virtue-theoretic condition, is actually motivated and explained by it. The key in
Greco’s analysis is the way he understands the reliability that cognitive competences
must possess. Such competences are “dispositions to believe truths in a relevant range,
introduction 7
when in relevant circumstances, with a sufficient degree of reliability”; many authors

believe that the sufficient reliability in question is to be judged relative to normal
conditions. Greco’s novel proposal is that the reliability needed for knowledge arising
from cognitive competences is not always reliability in normal circumstances, but
rather the reliability needed in conditions set by some informational needs. For exam-
ple, a miner may need to be very good at distinguishing lead from gold in abnormally
low lighting conditions, a reliable visual competence to distinguish lead from gold in
normal conditions will be useless to her, it will give her no knowledge. To gain knowl-
edge she needs a competence that is reliable in the conditions where she needs to
obtain the information she wants, regardless of how reliable that competence is in
normal conditions. In Greco’s example we can suppose that the miner’s competence
that is reliable in abnormally low lighting conditions becomes unreliable in normal
conditions because normal light is too much light for her visual system. Greco argues
that if an agent has a true belief that is produced by one of her cognitive competences
that is reliable in a set of circumstances defined by certain informational needs, that
belief automatically is safe (i.e. one that could not easily have been false), but relative
to the exercise of that competence in those circumstances. In this way, fulfilling the
virtue-theoretic condition guarantees that a restricted or relativized safety condition
is also fulfilled.
Clarifying the kind of reliability that a competence must have to generate knowl-
edge is key in Greco’s argument to link aptness and safety; the proper elucidation
of reliability is also a key issue in Alan Millar’s essay “Abilities, Competences, and
Fallibility.” Millar defends what he calls the “generalized success thesis” concerning
abilities, which says that one exercises an ability only if one succeeds in doing that
which the ability is an ability to do. He specifically discusses the extension of such the-
sis to perceptual-recognitional abilities, which he then conceives as abilities that one
exercises only if a relevant recognition is indeed effected on the basis of perception.
Like all the contributors we have discussed so far, and many other epistemologists,
Millar thinks that the reliability of an ability or competence is to be evaluated rela-
tive to a specific set of circumstances; he says that the ability has to be considered in
“favorable environments,” which are environments that, at least, are not populated by
deceptions that would make the recognitional ability go astray. But Millar discusses
another aspect of the exercise of abilities that needs to be taken into account to deter-
mine how reliable one has to be at Φ if one is to count as possessing the ability to Φ:
that aspect is how hard or difficult is the relevant performance. The idea is that the
degree of reliability required for possession of an ability is relative to how hard the type
of performance in question is, the harder the performance the lesser the degree of reli-
ability required to count as possessing the relevant ability. In light of the generalized
success thesis, when one doesn’t succeed in a performance one is in fact failing to exer-
cise the relevant ability at all; this is a manifestation of the fallibility of the ability in
question, which amounts to the fact that one is not bound to exercise the ability
(i.e. one is not bound to Φ) every time that one acts with the intention of exercising
it (i.e. every time one acts with the intention of Φing). Given this conception of the
fallibility of abilities, the reliability required for possession of an ability consists in suc-
ceeding at exercising the ability, and hence at achieving what the ability is an ability to
do, a high enough proportion of times given how hard the performance is.
As is clear from the issues discussed in the essays mentioned so far, a central com-
ponent of PBE is the tenet that an epistemically good competence, one capable of
delivering knowledge and justification, has to satisfy some or another reliability con-
straint. By adopting reliability constraints PBE faces a classic problem for reliabilism
known as the New Evil Demon Problem. This problem arises from considering certain
counterfactual scenarios where a subject’s competences systematically fail to fulfill a
relevant reliability constraint, due to the undetectable intervention of an extraneous
force (like an evil demon or a super-computer manipulating her brain) and yet such
scenarios elicit the intuition that she is justified in believing what she does. In his essays
“Against Actual-world Reliabilism” Peter J. Graham argues against the so-called “actual-
world response” to the New Evil Demon Problem and proposes a different solution to
it. Instead of talking of competences Graham talks of procedures, I will follow him in
this regard. According to actual-world reliabilism the epistemically correct proce-
dures, that is, those that confer justification, are those that are reliable in the actual
world; if a procedure is reliable in the actual world then it is correct, that is, it confers
justification in all possible worlds. This theory explains the intuition that seemed
incompatible with reliability constraints saying that the beliefs of an agent in the
demon world are indeed justified because they are produced by procedures, that even
though they are not reliable in the demon world, they are reliable in the actual world
and therefore confer justification in all possible worlds, including the demon world.
Graham examines two variants of actual-world reliabilism, which he calls “absolutist”
and “indexicalist,” and rejects both. According to the absolutist version of the view
there is one and only one actual world, as he puts it “being actual is an absolute,
non-perspectival feature of just one world.” He thinks this absolutist version has a
clear counterexample, since there are possible but non-actual epistemically correct
procedures that would fail to be reliable in the actual world. Imagine, for example,
a possible world inhabited by some aliens whose epistemic procedures are highly
reliable in that possible world, but would be unreliable if they existed in the actual
world. Clearly, the beliefs those aliens would form on the basis of their procedures
would be justified, however the absolutist view mistakenly rules them as non-justified.
According to the indexicalist version of the view “actual” refers to the world of utter-
ance or thought, so which world is the actual world is determined by the world in
which the speaker or thinker is. This version of the view seems to account for one’s
intuition concerning the demon world and seems to avoid the alien counterexample.
When we think that the demon’s victim uses procedures that are reliable in the actual
world, we are thinking something true, for those procedures are indeed reliable in our
world, and therefore they confer justification in all worlds. Similarly, when an alien
thinks that her procedures are reliable in the actual world, she is thinking something
introduction 9
true, for those procedures are indeed reliable in their world, and therefore they con-
fer justification in all worlds. However, it is obvious that this indexicalist approach
generates some implausible results. We’ve just seen that from an alien’s perspective her
procedures confer justification in all possible worlds, but if we consider their proce-
dures from our perspective we would have to say that they are unreliable in the actual
world, which is true, for those procedures are unreliable in our world, and therefore
they do not confer justification in any world, which is the negation of the alien’s view.
Graham rejects the indexicalist variant of actual-world reliabilism for making the
power of a procedure to confer justification too perspectival, in the way just described.
He then moves on to propose a form of reliabilism that relativizes the power of a pro-
cedure to confer justification not to its reliability in the actual world (however it is
conceived) but to its reliability in “normal” or “natural” circumstances or conditions.
We saw that Greco argues that the reliability that makes a competence epistemically
good is not always its reliability in normal conditions; confronting Graham’s and
Greco’s views on this issue can be a very fruitful task.
Amidst their critique of Sosa’s analysis of knowledge, Kallestrup and Pritchard
acknowledge that performance-based epistemology honors the fundamentally true
idea that when one knows “it is down to the proper exercise of one’s cognitive agency.”
But how exactly is one’s agency involved in knowledge? The essays we’ve addressed so
far discuss various aspects of the competences one exercises in obtaining knowledge
and justification, like their structure, the kind of reliability they must have and the way
they relate to modal conditions for knowledge, but the issue of what it means for an
exercise of a competence to be the manifestation of one’s own agency has not been dis-
cussed. In his essay “Who Knows?” Baron Reed addresses this question. He argues that
the mere reliability (of whatever stripe) of a competence will always be insufficient for
knowledge if the subject is not properly active with respect to her competence and the
true belief it produces, as he says: “the person cannot be merely the passive location in
which reliable cognitive faculties are at work.” He suggests that in order for a subject to
be properly active with respect to her beliefs they must constitute a kind of rational
synthesis of the inputs delivered by the agent’s various faculties; that kind of synthesis
is something that the agent herself must do, something of which she, and not merely
any of her discrete competences, must be responsible for. When this sort of synthetic
response takes place, the agent herself, and not just her cognitive competences taken
individually, is properly related to the environment; only then the cognitive success of
her beliefs is an achievement that genuinely belongs to her. Now, when the agent
responds rationally to the inputs from her environment, she does so inevitably from
her subjective perspective. For this reason, the elucidation of the sense in which cog-
nitive success and cognitive competences belong to an agent “is implicitly an attempt
to locate an internal nature in the cognitive sense—a deeper self that doesn’t merely
react to impulses from the outside but rather shapes its own responses.” Many theo-
rists working in performance-based epistemology believe that an attributability or
credit condition is necessary for knowledge, but Reed gives reasons to think that the
notion of attributability in play is in need of a deeper explanation than it has received

so far. He believes the explanation he sketches, that gives primacy to how the agent
as a whole responds to the environment from her subjective perspective, is in the
right track to deliver the kind of elucidation needed.
We’ve pointed out that one set of fundamental conceptual issues in performance-
based epistemology is that concerning the nature of the normative-inducing relations
between competences and the performances they help to produce. We’ve come across
two of those relations that have a place in Sosa’s SSS + AAA framework, one is
Adroitness, when one’s performance manifests one’s competences, and the other is
Aptness, when the success of one’s performance manifests one’s competences. In his
essay “Knowledge as Achievement, More or Less” John Turri suggests ways to enrich
the AAA side of Sosa’s framework by defining other normative-inducing relations
between competences and performances. In particular, Turri defines two more of such
relations whose names also happen to begin with the letter “A”, he calls them Amplitude
and Adequateness; in this way Sosa’s original triple A structure becomes a quintuple A
structure. A performance is ample when its safety (not just its accuracy) manifests the
agent’s competences; a belief is then ample when its safety (not just its truth) manifests
the agent’s competences. A performance is adequate when its success manifests reliable
or unreliable competences. When a performance’s success manifests reliable compe-
tences, the performance is apt, and the outcome of the performance is an “achieve-
ment.” Turri calls the successful outcome of an adequate performance “attainment,”
therefore all achievements are attainments, but not all attainments are achievements;
he reserves the name “mere attainment” for the outcome of a performance that mani-
fests unreliable competences, we may then call such performances “merely adequate,”
for they fail to be apt.5 Given what we’ve seen about the previous essays in the present
volume, some of the explanatory power of Turri’s suggested extensions to the AAA
framework must be clear. Consider, for example, Kallestrup and Pritchard’s objections
to Sosa. According to their first objection aptness is insufficient for knowledge because
knowledge is more than an achievement, it also requires safety. Turri’s notion of ampli-
tude can be used to account for that objection: an ample belief is one such that its safety
manifests the agent’s competences, we can then capture Kallestrup and Pritchard’s
intuition that in addition to a virtue-theoretic condition knowledge requires also a
safety condition if we define knowledge as ample belief. One would expect (whether
or not this is his intention) that Turri’s notion of mere attainment could be used to
capture the intuition in Kallestrup and Pritchard’s second objection to the effect that
aptness is unnecessary for knowledge because knowledge is sometimes less than an
achievement, like when one learns the location of a building from casual testimony. But
this doesn’t seem to be so. In the testimony cases discussed by Kallestrup and Pritchard
5
Turri calls “adeptness” what I’ve been calling “aptness”, and reserves “aptness” for a slightly weaker
notion. In order to avoid unnecessary confusion within this Introduction I use “aptness” uniformily
throughout and refer the reader to Turri’s essay for his reasons to opt for a different usage.
introduction 11
the intuition is that knowledge doesn’t require achievement because the agent doesn’t
deserve enough credit for her true belief and she nevertheless knows. In contrast, in the
cases discussed by Turri the relevant agents do deserve enough credit for their success,
but their success doesn’t derive from a reliable enough competence and therefore fails to
count as an achievement, it is a mere attainment. However, it is doubtful that a true
belief that is a mere attainment, in that sense, would be considered by most epistemolo-
gists as a case of knowledge. Turri points out that in other domains of evaluation mere
attainments may be highly praised and wonders why not consider some true beliefs
that are mere attainments as constituting a special category of knowledge. In any case, it
seems clear that the difference between a mere attainment and an achievement (which
is a difference in reliability) is not of the right kind to capture the difference between
knowledge obtained through casual testimony and an arduous cognitive achievement
of the agent (which is not a difference in reliability but in how creditworthy the agent is).
This critical remark, however, should not obscure the central message in Turri’s essay:
the basic conceptual framework of performance-based epistemology can be fruitfully
extended in simple ways that increase its explanatory power in valuable ways.
III
The essays in Part II of this volume discuss the application of performance-based epis-
temology to some outstanding problems in contemporary epistemology. Part II opens
with Paul A. Boghossian’s essay “Intuitions and the Understanding,” its theme is the a
priori justification we may have for believing philosophical theses or theories and his
critical focus is Sosa’s epistemology of intuitions that conceives them as attractions to
believe that derive from a competence to tell reliably, on the basis of one’s understand-
ing of p alone, whether p is true or false. Boghossian presents several objections to
Sosa’s view, a couple of them are these. Sosa thinks that intuitions play a foundational
role in justification, they work like regress stoppers, Boghossian objects that if intui-
tions are attractions to assent then they cannot play a foundational role in justification
because attractions to assent can themselves be epistemically evaluated. Another
objection of Boghossian concerns the scope of application of Sosa’s theory, Boghossian
argues that many cases of a priori knowledge and justification cannot be explained as
deriving from a competence to tell on the basis of understanding of p alone, whether p
is true or false. Without going into the details of the cases Boghossian thinks are
problematic for Sosa’s views, I think that the challenge he is issuing for performance-
based epistemology is to identify other, perhaps more complex competences, whose
deliverances can help explain the whole range of a priori knowledge and justification.
Boghossian closes his essay by sketching an alternative picture on which intuitions are
impressions of obviousness, which are pre-doxastic states. Although he rejects reliabi-
lism as a theory of justification, he thinks that the reliability of our judgments based on
such impressions needs to be explained by a scientifically respectable account of why
we have certain concepts and make judgments involving them and why we would be
reliable in making them.
The second essay in Part II is Richard Fumerton’s “Rising above the Animals: The
Search for Intellectual Assurance.” Fumerton expresses several objections that inter-
nalists would make to Sosa’s views on second-order knowledge. Perhaps the two most
pressing of these objections are those that concern whether Sosa’s views can escape
the so-called “problem of easy knowledge” and whether they can accommodate the
intuitions behind the internalist notion of assurance. The problem of easy knowledge
arises more clearly for externalist theories of knowledge because they recognize a
category of “basic knowledge,” which is a type of knowledge that one obtains through
a source without having to know that the source is reliable. The problem is that if
basic knowledge is allowed it seems that we can use the readings of a particular source
to confirm their own accuracy; for example, if a thermometer gives us basic knowl-
edge that p, it seems that we can use that knowledge to confirm the accuracy of that
very same reading. If we can do that, then we can confirm the accuracy of many read-
ings of the thermometer in the same way and then inductively conclude that the ther-
mometer is reliable. But that is an unacceptably easy way to come to establish the
reliability of a source. Now, since the notion of apt belief is a category of basic knowl-
edge, performance-based epistemology faces the problem of easy knowledge.
Fumerton argues that Sosa’s views actually exacerbate the problem because they
make it even easier to get justification for the reliability of our belief sources. The crux
of Fumerton’s criticism is that he thinks that a consequence of Sosa’s views is that if
one fulfills the conditions for apt belief one automatically, without having to engage
in any reasoning, has justification for an underlying “commitment” concerning the
reliability of the competence exercised in obtaining the relevant apt belief. Fumerton
also objects to Sosa that his views are unable to properly handle the intuitions that
sustain the internalist notion of assurance. The issues involved here are quite subtle
but the essence of the problem is whether Sosa can give a satisfactory answer to the
question of how we can discover that, or assure ourselves that, we have animal,
first-order knowledge. Sosa thinks he can, as he explains at different places, including
his essay in the present volume: we can ascend to the epistemic second-order through
the use of second-order competences that enable us to know whether our first-order
competences are properly exercised in favorable conditions and if their exercise
results in apt belief. But Fumerton thinks that answer is not satisfactory, because in
the same way that the first-order competences need to be exercised in favorable con-
ditions to give us first-order knowledge, the second-order competences also need to
be exercised in favorable conditions to accomplish their feat. But when we exercise
our second-order competences it seems that we merely assume that we are in the rele-
vant favorable conditions for their exercise. It is this residue of assumption that in
Fumerton’s view seems to prevent achieving the kind of philosophical assurance that
he seeks concerning our possession of knowledge. Perhaps what is ultimately at stake
in the debate between Fumerton and the internalists on the one hand and Sosa and
introduction 13
performance-based epistemology on the other, is what is to count, and why, as a satis-

factory answer to the question of how we know that we have first-order knowledge.
The importance that Sosa gives to second-order or reflective knowledge is also a cen-
tral theme in Hilary Kornblith’s essay “Epistemic Agency.” He develops a critique of the
role that epistemic agency plays in Sosa’s version of performance-based epistemology,
focusing on one manifestation of epistemic agency that has a particularly prominent
role in Sosa’s thinking: the role of the agent in reflectively assessing her first-order
beliefs. Contrary to Sosa’s implicit optimism that reflection can increase the chances of
success at the first-order, Kornblith draws on empirical literature to argue that not only
the real sources of our first-order beliefs are hidden to ordinary reflection, but the
reflective agent very often actually confabulates to depict such sources as reliable,
when in fact they are not, while creating the illusion that she is giving an accurate
account of them. In this way, the ordinary reflective agent misrepresents both the way
in which she forms first-order beliefs and the way in which she examines them. From
this perspective it doesn’t look as if the epistemic agency manifested in ordinary reflec-
tion is in good shape to carry out the ameliorative project that Sosa assigns to it. But
Kornblith seems to have an even deeper critique of the role that Sosa and other perfor-
mance-based epistemologists give to epistemic agency, he argues that in fact agency is
not involved in the formation of belief itself, as they suppose, but merely in actions
which provide inputs to our belief forming processes. He thinks this is true not only of
unreflective belief acquisition but of reflectively formed beliefs too. In particular, he
argues that the features of reflectively formed beliefs on which performance-based
epistemologists focus their attention as paradigms of epistemic agency, in fact do not
exhibit any distinctive epistemic agency at all, because sub-personal processes take
over in the case of reflectively formed beliefs as much as they do in the case of unreflec-
tively formed beliefs. If the phenomenology of ordinary reflection is misleading in the
ways and to the extent that Kornblith thinks empirical work shows, then perfor-
mance-based epistemologists have the outstanding challenge to reshape their notion
of epistemic agency and rethink the roles it can genuinely play in our cognitive lives.
One question that emerges in Kornblith’s discussion is this: if ordinary reflection is
not valuable for enhancing reliability at the first-order of cognition, then why is it valu-
able? He criticizes the hypothesis that reflection is valuable because of the autonomy
that it endows the agent with; given that autonomous reflection can in fact undermine
reliability, he finds no reason to value it epistemically. The hypothesis that tries to link
the value of reflection with autonomy, takes center stage in Stephen Grimm’s essay
“The Value of Reflection.” Grimm begins by considering Sosa’s view that reflective
knowledge is specially valuable because it enables one to avoid some sort of cognitive
disharmony exhibited by a subject who is not in a position to reflectively endorse her
first-order beliefs because she is in doubt about their epistemic status. But Grimm notes
that there is no such disharmony in those who fail to reflect at all, simply because they
don’t have any second-order beliefs at all, and asks why should those unreflective
subjects value reflective knowledge? He then considers the idea that “something is
normatively lacking in a belief unless one is able to respond to doubt”; the thought is
that human beings should doubt their first-order beliefs and then assure themselves that
they are true, only with this second-order certification will beliefs be fully correct from
the epistemic point of view. Grimm objects that in order to resolve doubts one need not
go up to the second-order, and besides, if responding to doubts is a desideratum for full
normative adequacy at the first-order, why should it not also be so at the second-order
and beyond? A regress looms. Instead of linking the value of reflection to its potential
power to bring some sort of harmony to the cognitive states of the agent, Grimm pro-
poses that reflection is valuable because in it one manifests intellectual autonomy. But he
points out that the reflection that manifests such autonomy is not the reflection about
the reliability of one’s competences, but the reflection through which we endorse our
“given” epistemic ends as believers as good or worthwhile. The second-order endorse-
ment that manifests our autonomy is not (or not primarily) the second-order
endorsement of the reliability of our first-order competences, but of our first-order
epistemic ends. However, he underlines that this second-order endorsement doesn’t
make our first-order beliefs epistemically better but, as he puts it, “will make them
more fully our own.” The fact that the intellectual autonomy that makes reflection
valuable doesn’t make our beliefs epistemically better, raises the question whether the
value that reflection has in virtue of manifesting that autonomy is an epistemic value;
Grimm himself suggests that it is not and that it is better thought of as a moral or eth-
ical good. So, Grimm agrees with Kornblith on this much: if the autonomy manifested
in certain forms of reflection about our beliefs is valuable, that value is not epistemic.
An evident tendency in mainstream performance-based epistemology is its almost
exclusive focus on the analysis of the nature of competences and their role in our cogni-
tive lives; performance-based epistemologists are almost completely silent about the
opposite of competences: incompetences, which pre-theoretically one might think
play an equally important role in our lives. In his essay “The Epistemology of Stupidity”
Pascal Engel undertakes the task of analyzing a generic kind of epistemic vice he
calls “stupidity,” attempting to determine to what extent the framework of perfor-
mance-based epistemology can be used in such a task. Engel begins by stressing the
diversity of epistemic vices that fall under the adjective “stupid” and remarks that
those that consist of basic inabilities or forms of incompetence fall under the reach of
performance-based epistemology. However, he rejects what he calls the “intellectual-
ist” conception of stupidity, which equates it with a lack of judgmental and inferential
ability. He develops several reasons to reject such a conception but the most important
for his purposes is that it doesn’t capture the kind of stupidity he is most concerned
with, which consists in a failure to appreciate and respect the relevance and importance
of fundamental facts about our epistemic goals; it is a failure to ponder them correctly
and give them their proper weight across changing circumstances. The fool, he says, is
a person who “doesn’t have the proper conception of the right attitude that one ought
to have with respect to epistemic objects.” On the face of it, this form of foolishness
or stupidity would seem to be properly conceptualized within the framework of
introduction 15
performance-based epistemology as a form of second-order incompetence to properly

handle epistemic goals and objects. But Engel insists that the form of stupidity he is
characterizing “does not bear on competence or basic intelligence,” since even “learned
and competent individuals” can suffer from it, and suggests that it is more amenable to
treatment within a framework of responsibilist-virtue epistemology. Engel’s argu-
ments do not only pose the challenge to performance-based epistemologists to work
out a proper account of the form of stupidity that he discusses, they also raise issues
concerning relations between performance-based epistemology and responsibilist-
virtue epistemology.
The present volume closes with Ram Neta’s essay “Epistemic Circularity and
Virtuous Coherence,” where he provides a solution to the problem of explaining what’s
wrong with epistemically circular arguments, and criticizes an important aspect of
Sosa’s solution to that problem. The problem of epistemic circularity, as Neta and oth-
ers conceive it, is the problem of explaining in a principled way why one can’t use a
belief-source to establish its own reliability. He begins by focusing on a more circum-
scribed problem that he calls the problem of “single-case circularity,” which consists
in explaining why we can’t rely on a particular reading of a belief-source to establish
the accuracy of that very reading. He proposes a principle he calls “Generalized
Transmission Failure” (GTF) that he believes correctly explains which inferences suf-
fer from “single-case circularity” and which ones don’t. He then moves to consider
“bootstrapping inferences,” which try to establish not the accuracy of one particular
reading of a belief-source on the basis of that very reading, but rather the reliability of a
belief-source relying on various readings of that very source. Neta argues that his prin-
ciple GTF also gives the correct diagnosis of bootstrapping inferences. He then points
out that the background needed to apply GTF generates a regress of independent justi-
fications, criticizes Sosa’s account of how to deal with such a regress, and proposes
what he argues is a better way to handle it. His criticism of Sosa is that the application
of GTF presupposes that there are certain asymmetric relations of epistemic depend-
ence, but Sosa responds to the regress of independent justifications introducing a
coherentist element in his epistemology according to which all epistemic relations
of dependence are symmetric to some degree, and so Neta thinks that such an episte-
mological outlook prevents the best explanation of the wrongness of epistemically
circular inferences, that is, precisely the one based on his principle GTF. Surprisingly,
Neta’s response to the regress of independent justifications also includes a coherentist
element, but one which is subtly different from the one present in Sosa’s epistemology.
Neta’s essay raises the general question of how performance-based epistemology
should best address the problem of epistemic circularity, and the particular question
of whether in providing such explanation a commitment to one form or another of
coherentism is inevitable.*
* I want to acknowledge that the work I did in putting together this volume was made possible thanks
to the financial support of research project PAPIIT–IN400513, “Conocimiento y escepticismo”, granted by
DGAPA-UNAM.
References
Baehr, J., 2004. “Virtue Epistemology.” In J. Fieser and B. Dowden (eds) The Internet Encyclopedia
of Philosophy. <http://www.iep.utm.edu/> accessed July 7, 2015.
Battaly, H., 2008. “Virtue Epistemology.” Philosophy Compass 3(4): 639–63.
Greco, J., and J. Turri 2011. “Virtue Epistemology.” In E. N. Zalta (ed.) The Stanford Encyclopedia
of Philosophy (Summer 2015 Edition). <http://plato.stanford.edu/archives/sum2015/entries/
epistemology-virtue/> accessed July 7, 2015.
Pritchard, D., 2009. Knowledge. New York: Palgrave Macmillan.
Sosa, E., 1991. Knowledge in Perspective: Selected Essays in Epistemology. Cambridge, UK:
Cambridge University Press.
Sosa, E., 2011. Knowing Full Well. Soochow Lectures in Philosophy. Princeton University Press.
Sosa, E., 2015. Judgment and Agency, Oxford: Oxford University Press.
PA RT I
Foundational Issues
1
Epistemic Competence
and Judgment
Ernest Sosa
Any performance with an aim can be assessed on three dimensions. First, does it attain
its aim? Second, is it competent? Third, does its success manifest its competence? That
introduces a kind of normativity, performance normativity.
A gust of wind, for example, can make an archer’s shot accurate while maladroit, or
adroit while inaccurate. Compensating gusts can make the shot both accurate and
adroit, moreover, without being apt, without the accuracy manifesting the adroitness.
An initial gust can divert the skillful shot, so that it would miss, although a second gust
puts it back on course to the bull’s-eye. Only if no spoiler gust intervenes can the shot
be apt: only thus would its accuracy manifest its adroitness (and not just derive from
the luck of that second gust).
The account of epistemic normativity as involving such performance normativity
helps explain the nature of knowledge, which amounts to belief that is apt, belief that is an
apt epistemic performance by manifesting the relevant competence of the believer in
attaining the truth reliably enough. And it explains also an extra value of knowledge
beyond that of merely true belief, the value that success through competence has beyond
that of mere success. We might thus plausibly understand a certain sort of knowledge,
animal knowledge, as simply being aptly right.
A. Animal Knowledge
The importance of knowledge derives from how it relates to human achievement gen-
erally, as emerges in examples like the following.
1. Superstitious Hunter believes that his shot in the dark would succeed because
his god of the hunt would ensure it regardless of the direction or speed or tim-
ing of his shot. Alternatively, he might believe that if he next shot in this specific
way (with a certain imaged orientation and speed), his shot would succeed, and
that this is so because this belief is guided by the god, although in fact it’s just
20 Ernest Sosa
what first came to mind, with no reliable source at all. In neither case does it
seem plausible that Superstitious Hunter now has a competence to hit some
prey. He does have a disposition to do so, based essentially on his belief, but can
that sorry belief really constitute a true competence?
2. What makes it so implausible to think that Superstitious Hunter has any rele-
vant competence to shoot prey with shots in the dark? Perhaps a relevant com-
petence requires a broader field of accomplishment. The supposed competence
of Superstitious Hunter is stipulatively restricted to the particular occasion
involved. If he tried to succeed similarly with other shots in the dark, while
supposedly aided by his god of the hunt, his success ratio would be near zero.
He seems to have no competence of relevant interest.
3. A comparison may help. Suppose a pewter mug would shatter upon hitting a
certain surface but only because a hovering fiend is determined to zap it if and
only if it hits the surface, at the moment of impact. The mug would thus satisfy
the relevant conditional—it would shatter on that occasion—without it being
thereby fragile. For the latter to be the case, one would need at a minimum that
the mug would shatter whenever relevantly impacted in a relevantly interesting
combination of condition and situation.
4. Plausibly, what a hunter needs for competence is know-how. Sometimes this know-
how involves knowledge of what means can be used to kill a prey. Competence
is sometimes constituted by a belief that means M will bring about end E, but
only if this belief is true, competent or justified, and, indeed, a case of knowledge.
In a slogan: if knowledge-how is constituted essentially by a means–end belief, then
this belief must be a case of knowledge-that.
B. How Does Near-actual Simulation Affect

Competence and Aptness?
Consider the case of Simone—a pilot in training who could easily be, not in a real
cockpit, but in a simulation, with no telltale signs. In this thought experiment, trainees
are strapped down asleep in their cockpits, and only then awakened. Let us suppose
Simone to be in a real cockpit, flying a plane and shooting targets accurately. She now
justifiably believes that her training is long over, that now every morning she goes up in
a real plane. But most days she is still in the simulation cockpit. Nevertheless, on the
occasions when she is up aloft shooting missiles, her shots can be not only accurate,
surely, but also competent, and even apt.
So much for Simone’s physical shots and for how their aptness is affected by the
danger that the modal proximity of the simulation poses to her competence. What of
her intellectual shots, her judgments and beliefs? Suppose again she happens to be
really up piloting and believes she shoots a certain target. And suppose this belief to
be accurate and even competent. Can it also be apt? That is to say, can it be a belief
whose accuracy manifests Simone’s epistemic competence? Can it be apt despite the
epistemic competence and judgment 21
great threat to her epistemic competence, since she might so easily have been in
the simulation cockpit?
It is hard to deny that Simone’s belief can be apt, given how clearly apt her shot can
be when it hits its target. Despite its fragility, her shooting competence can be manifest
in the accuracy of her shot, which would thereby be apt. What then should we say of
her belief, the intellectual shot? Why can it not be apt, like the physical shot? Even if
this analogy makes it plausible that the belief is apt, however, maybe it is apt without
being knowledge. After all, it is correct only through a kind of luck seemingly inimical
to real knowledge.
Simone retains her relevant inner competence. She has the constitutional, structural
skill required. And she is in good shape: awake, sober, alert, and so on. However, she
lacks or is in danger of lacking the situation required for the relevant complete compe-
tence. She might easily have been in the simulation cockpit.
Three options open in the case of Simone: 1. Do we deny that she so much as aptly
believes? 2. Do we say that she aptly believes and knows accordingly? 3. Do we say she
aptly believes but does not know?
1. Suppose we deny that Simone so much as aptly believes, alleging that the threat
of the simulation cockpit deprives her of complete competence. It is thus untrue that if
she were to believe she sees a target, she would likely enough be right. She might too
easily be in the simulation. For this reason we might deny that her belief is apt: “She
lacks the complete competence required to tell when she sees a real target and not just
a simulation.”1
Note how awkward it would be to hold beliefs to that standard, however, without
doing so for performances generally. Yet we impose no such requirement on archers,
pilot trainees, or athletes. A basketball player, for example, might be in an indoor venue
where her shots are calmly apt, even though high winds would impair them in all
nearby venues. And this is so even if the player might then easily have performed
in a nearby venue, so that the impairing situation is proximate not only spatially but
also modally.
Most plausibly, Simone does perform aptly on the days when she is really up in her
plane, with success that manifests competence. Her condition and situation then seem
appropriate for the relevant trigger-manifestation conditionals to help determine posses-
sion of the relevant competence. What is manifest in her success is not only her structural
skill, but also her being in appropriate shape while appropriately situated.
2. Still open, however, is this question: Does aptness of belief amount to any kind of
knowledge? Simone, we say, does believe aptly. Nevertheless, she surely cannot know
she is aloft shooting targets when so easily she might have been indistinguishably in
the simulation cockpit instead.
Nevertheless, Simone’s competent, apt shooting cannot be divorced from her apt
choice of effective means, and this latter seems a kind of knowledge how. Apt action
1
For defense of this approach, compare Greco 2009 and Millar 2010.
22 Ernest Sosa
often requires apt means–end belief, which cannot constitute relevant know-how
unless it amounts to animal knowledge (whether or not this animal knowledge also
constitutes reflective knowledge).
3. Our question is whether there is a sort of knowledge that one necessarily has in
believing aptly. Unfortunately, it is implausible that Simone can then know even that
she is aloft and not in a simulation. Anyone with this strong intuition—that Simone
does not know—might thus insist that she has no sort of knowledge whatsoever that her
shot hits its target.
With recourse to reflective knowledge, we could share that reaction. We could
hold that the only literal sort of “knowledge” that we ever need recognize is reflective
knowledge. So-called “animal knowledge” might then be thought to involve only a
metaphorical use of epistemic vocabulary. It could of course be well-motivated use of
that vocabulary, as when we say that the electric-eye door “knows” when someone is
approaching. People fall in naturally with that use. We all understand well enough
what is being said. Something significant is being said, moreover, even if the speech is
metaphorical.
C. Why It Is More Complicated

So we have reached the conclusion that apt means–end action requires knowledge of
the relevant means–end proposition. If her hit is to be apt, Simone must know when
aloft that by pulling her trigger she will hit her target. It is objected that Simone cannot
know that she is aloft, given how easily she might have been acting under simulation,
unbeknownst to her. But this objection may pertain only to reflective knowledge. That
is what Simone intuitively lacks, which enables us to insist that she does have knowl-
edge of the lower, animal grade. And indeed we need not insist that this is in line with
literal ordinary English. Our concern may be to understand the metaphysics of knowl-
edge and not so much the semantics of ‘knowledge’. Through metaphor we may still be
able to formulate and communicate an account of the objective phenomena of interest
to us, the phenomena of human knowledge and belief.
Again, we have been exploring the thesis that apt means–end action requires animal
knowledge that the means will yield the end.2 Although that thesis has much to be said
in its favor, we next turn to some serious problems that it must face:
1. There are many instances where we act on probabilities, where we are far from sure that our
means will yield success. Shipwrecked, you may follow your best guess by swimming west. If
you do reach landfall in some hours, this may still be an apt accomplishment highly creditable
to you. The probability was low, let us say, but not much lower than that of a soccer shot-on-goal,
or a baseball batter’s swing, each of which aspires to a top success rate of about 15%.
2
My forthcoming “Knowledge in Action” considers this thesis more fully while relating it to Aristotelian
virtue theory.
Performances can thus clearly be apt even while manifesting a competence with low success
rate. They must sufficiently manifest competence, of course, in order to be admirably apt, but
in general the threshold for competent performance falls well below 50%. And it seems in line
with this, and plausible anyhow, that our strong swimmer’s success in reaching shore can
manifest competence so as to count as creditably apt, as a success that manifests competence
despite being based essentially on a good guess.3 The key point is that much luck is compati-
ble with possession of competence. One might be lucky in possessing each of the three S’s
constitutive of a complete competence (skill,4 shape, situation). The present point is that a
means–end belief can itself be correct by luck while still constitutive of a competence manifest
in apt performance.
It may be replied that the swimmer may still know quite reliably that it is somewhat probable
that reachable land lies in the direction that he takes. And this might be worth exploring,
except that even a swimmer who takes a wild and arbitrary guess might attain apt success any-
how in reaching his destination, so long as the guess is right. Such action under a supposition
may still count as apt provided the supposition turns out to be true. This gives a further twist
to Plato’s Meno question: Why is even true belief better than true supposition on which one is
poised to act?
2. What is required in order to possess a skill, an ability, a competence, and indeed a
general competence? Take a prisoner who believes his guard’s barefaced lie that the cell
door will be open that night from nightfall to dawn, when the door just happens indeed to
be unlocked during that span, accidentally so. As he lies in bed trying to decide whether to
make a break, does our prisoner have the ability to escape? The capacity? The competence?
He certainly can get up from his bed, turn the handle, and open the door. Indeed, through-
out that period, from dusk to dawn, he has a set of motor skills, while in the appropriate
shape (awake and alert) and situation (not tied down, no obstacles, unlocked door) that
give him the specific competence and skill required for escape. That seems plausible enough.
And yet an essential constituent of that complexly constituted competence is a means–end
belief that is far from constituting knowledge: namely, his belief that by turning the handle
he would open the door.
Competence comes of course in degrees, as does, correspondingly, aptness. Consider

again our prisoner who does plausibly acquire relevant competence based on his
Gettiered belief, and who does make his jailbreak. Exiting from his cell that night is a
successful, competent performance of his, and even apt. In these respects it is like
one’s reaching Larissa based on a merely true belief about the right road. But such apt
performance still falls short. It falls short by comparison with its status had the perti-
nent belief been not merely true but also knowledge. Despite its success as an escape,
the prisoner’s performance suffers from an important element of luck owed to the
3
A distinction is required here between a performance’s manifesting competence and its success doing
so. In a Gettier case, the belief manifests competence but its success does not.
4
‘Skill’ in a broad sense that allows natural skills, which need not have been learned, but may have come
naturally with normal maturation.
24 Ernest Sosa
epistemic failure involved. And so there is a relevant way in which its success is owed
to sheer luck. In that respect it falls short as a performance.
Such luck does not preclude aptness, in which respect it is like many other kinds of
luck that can attend apt performance. Again, it can be lucky that one is in appropriate
shape, in an appropriate situation, and even that one retains the appropriate innermost
skill. But these elements of luck need not deprive one of success that is apt, success that
manifests sufficient competence, even if reliant on luck. Compare then a means–end
belief whose truth is constitutive of one’s competence (as with our prisoner). The pris-
oner’s belief need not constitute knowledge in order to constitute his competence to
escape. Compatibly with that, his competence to escape would have been enhanced by
his belief ’s being not only true but also knowledge. That way an element of luck would
have been removed, in such a way that his escape would have been more fully creditable
to relevant competence.5
D. Judgment and Judgmental Belief

A further important value imported by human knowledge is not visible until we look
above the animal, functional, implicit level, to the reflective, judgmental, consciously
explicit level. Let us next turn to that.
Elsewhere I have argued for a view of judgment as an act of affirmation, public or
private. Judgment is to be distinguished from mere thought, as when one thinks that
the right answer in a quiz show is such and such, even if one is only guessing. So,
psychological guessing can take the form of a mere thinking that something is so. And
psychological guessing can take an even weaker form, as when one merely supposes
that the answer is such and such, for the sake of giving an answer that has as good a
chance to win as any, but no better than any other. Here again one is guessing, but here
one does not so much as think that one’s answer is right.
Mere supposition cannot constitute knowledge, of course, but apt thinking that
something is so can plausibly constitute a low grade of knowledge, a possibly minimal
sort of animal knowledge, if it manifests some degree of truth-directed competence.
Attribution of “knowledge” in such an instance might, again, be just metaphorical.
However, we do reserve more serious and literal attributions of knowledge for some-
thing more substantial epistemically.
Accordingly, in assessing our acts of affirmation, let us reserve the terminology of
‘judgment’ for this more substantial usage, where the affirmation must be not just in
the endeavor to get it right, which might be just a guess. Judgment requires more
than guessing; it requires rather that the affirmation be in the endeavor to get it right
5
And indeed Descartes’s exalted status of full certainty, fully apt knowledge, precludes also other
e lements of luck: luck in one’s situation, for example, and luck in one’s constitution. Thus, recall the skepti-
cal scenario in which one is created by a lesser being, one who falls short of divine perfection. This imports
luck in one’s proper constitution and situation, and such luck deprives us of certainty, or so Descartes
advises.
competently or reliably enough. But there is a contextual element here, depending on

whether the judgment is just a quotidian judgment, or whether it is rendered in a
context of expertise, where it represents one’s professional judgment, or one’s expert
opinion. What counts as competent enough may thus rise considerably from the
everyday. So, we need to distinguish ordinary knowledge from expert knowledge, as
we normally do in any case.
Thus, the more fully adequate way to connect apt action with epistemology is by rec-
ognizing varieties of epistemic performance. At the lowest end lies mere supposition
for the sake of giving a try, as when one swims in a certain direction based on an arbi-
trary guess among 360 options. Next would come the varieties of more substantial
guessing, where one does think a certain option to be more likely right, however
slim may be one’s basis. Exclaiming “I knew it” would now seem a more appropriate
reaction to one’s success, even if one speaks metaphorically. Eventually, we reach the
perhaps more literal crediting of the subject as having known, which requires belief
beyond guessing.
Let us restrict ourselves to the domain of free, conscious judgment. Let’s abstract
from functionings, from functional beliefs that may guide our conduct implicitly. Even
before the advent of advanced technology, much of our most important interpersonal
concerns play out on that level of conscious judgment. Othello is deceived by Iago into
doubting Desdemona. He exclaims, in effect, “I think that she is faithful and I think
that she is not faithful,” and demands release from his doubt. Iago is only too willing to
comply and reports an overheard soliloquy that implicates Cassio, appealing also to
some incriminating (planted) evidence. When he has enough evidence consciously in
view, Othello draws his conclusion, his conscious judgment, which leads to his wrath-
ful revenge.
Whatever deeper psychological forces may also be at work, such dialogue and such
conscious reasoning seem often to have fateful consequences (if less dramatically).
And this is multiplied vastly, once we reach a technological, information-dependent
civilization. Such judgment is important well beyond its place in Pyrrhonian or
Cartesian epistemology.
Elsewhere I have argued that such acts of affirmation (in the endeavor to answer
a question correctly and even aptly) are about as plausibly free and relevantly
will-dependent as are ordinary choices and actions. Both are very largely and
importantly reason-based with no loss of freedom or choice. Moreover, both can
manifest competence, rational competence. And both can be competent and even
apt, if all goes well.
In any case, a fuller value of human knowledge is realized only with a more humanly
distinctive level of knowledge, the reflective level. Fully reflective knowledge requires
a free act of affirmation in the endeavor to get it right on some question, and to do
so reliably enough. Here again, as with animal knowledge, epistemic competence is
required. Only judgment that manifests such competence can qualify as true knowl-
edge. Such reflective knowledge requires us to perform on a higher order, with access
26 Ernest Sosa
to our lower-level competences. What are these epistemic competences, and how do
they accomplish their good work?
Next I would like to sketch a view of such competence in general, and of epistemic
judgmental competence in particular.
E. The Structure of Competences (and Abilities)

1. A competence is a disposition (ability) to succeed when one tries. How are such
ispositions in general constituted? According to my suggestion, they have a triple-S con-
d
stitution when complete: innermost (skill), inner (skill + shape), complete (skill + shape +
situation). With regard to driving competence (or ability), for example, we can distinguish
between (a) the innermost driving competence, that is, the seat, the structural skill that
one retains even while asleep or drunk, (b) the fuller inner competence, which requires
also that one be in proper shape, that is, awake, sober, alert, and so on, and (c) the complete
competence or ability to drive well and safely, which requires also that one be situated with
control of a vehicle, along with appropriate road conditions, which pertain to the surface,
the lighting, and so on. The complete competence is thus an SSS competence.
2. Finks vs masks.
Once we understand competences and dispositions in terms of such threefold
components, we can distinguish finks from masks for each. In each case a mask pre-
vents a disposition from manifesting, without removing it. Thus a mask might be an
agent bent on stopping the manifestations by intervening in the process that normally
leads to the manifestation, but without removing any of the relevant S’s (the skill,
shape, or situation). By contrast, a fink would block the manifestations by removing
one or another of the S’s involved. This might be the structural skill (innermost compe-
tence), or the shape, or the situation.
A situation-fink masks both the inner competence and the innermost competence,
finally, whereas a shape-fink masks the innermost competence.
3. Competences come in degrees of reliability, along with a threshold.
a. A competence is a disposition to succeed when you try. So, exercise of a com-
petence involves aiming at a certain outcome. It is a competence because it is
a disposition to succeed reliably enough when one makes such attempts. So, a
competence is necessarily a competence to ø successfully, for some ø. And it is
thus tied to a conditional of the form: if one tried to ø, one would succeed.
b. However, in order to possess a competence to ø it is not enough that the following
conditional be true: that one would ø reliably enough if one tried to ø, that is, that
one would not too easily try to ø without actually ø-ing. For, this latter could be
true simply because, knowing one’s limits, one ø’s extremely rarely and only when
one is in the extremely narrow range where one would indeed succeed. Thus, one
might restrict one’s shots as an archer to situations where one is two feet away from
the target. This might show good shot selection, assuming that one is not reliable
enough at any longer distance, but it would not show archery competence.
c. What then is required for possession of a competence? What is required for

archery competence, for example, is rather a “sufficient spread” of possible shots
(covering enough of the range of relevant possible shapes and situations one
might be in as an agent) where one would succeed if one tried, an extensive
enough range. What constitutes this range is the following: There must be a
close enough sphere of possible worlds where one takes shots, varied enough across
the relevant range(s), and these shots must easily enough succeed, extensively
enough across the relevant range(s).
It might be objected that even if one is tied down, so that, knowing one’s condition,
one would not try to shoot, this does not take away one’s competence. But that would
surely take away one’s complete SSS competence, even if it would not take away the
inner, SS competence, nor the innermost, S competence. Moreover, if indeed one is so
disabled psychologically by a phobia that one cannot so much as try to shoot, then this
does take away even one’s innermost skill S. One is now no longer so structurally con-
stituted psychologically as to be competent to ø.
F. Competences, Epistemic and Otherwise

1. We distinguished judging that p from merely thinking that p. The latter requires
that one affirm in the endeavor to get it right on the question whether p. But that is
compatible with just psychologically guessing. Full judgment requires more, as one
must endeavor not only to get it right on that question, but also to do so competently,
reliably enough, and even aptly.
2. And here we find a reason why judgment is automatically on a reflective higher
order. For, what does it mean that one endeavors to get it right “competently, reliably
enough”? Succeeding in this endeavor requires that one would affirm in such a way, in
such circumstances, that one would be likely enough to affirm correctly. That is to say,
one’s success rate for such attempts must be high enough. And what would go into
such an endeavor? What must one consider as one aims for that objective: the objec-
tive of affirming reliably enough? Must one not consider one’s relevant SSS situation
and how likely it is that a relevant attempt manifesting such a complete competence
would succeed? In that case one takes a higher order attitude towards one’s three
options: affirming, denying, suspending. One considers which of these would be best
from the point of view of taking a risk if but only if it is proper. What one considers
then is the reliability of affirmation or denial on the question at hand, and whether
double-omission would not be preferable given the risks involved in either affirming
or denying. And this is all on the second order, since one must consider one’s relevant,
first-order, complete competences, and the first-order options of affirming, denying,
and double-omitting.
3. For performances more generally something similar is true across the board.
Thus consider three-point shots. Even a basketball player who overconfidently takes
low-percentage shots too frequently may retain an excellent ability to sink three-point
28 Ernest Sosa
shots close enough to the three-point line. So, success in such an attempt does seem
properly creditable to her competence. This is not affected by the fact that she tries
too often when she should not do so, when she is so far from the basket that her
reliability plummets.
4. But what is the aim of a basketball shooter? Does she aim merely to get the ball
in the hoop? If that is her sole relevant aim, then it seems she does succeed in her aim,
creditably so, even when she is far out, especially if her success even that far out is due
to a level of competence well above the average. Important considerations about her
performance as a team player are relevant to this assessment. For example, should she
pass rather than shoot? However, normally performers do not aim just to succeed in
the sense of attaining their basic aim no matter how unreliably. Normally they aim to
succeed reliably enough, while avoiding too much relevant risk.
5. Consider now the three-point shooter as she approaches the distance to the
basket where her reliability is indiscernibly near to the relevant threshold. And sup-
pose that in fact her reliability is above the relevant threshold, but indiscernibly so to
her. A statistician-coach-observer might know perfectly well that the player’s reliabil-
ity is still barely above the threshold. Suppose he has studied her success rate exten-
sively, aided by a device that measures with exactitude her distance from the basket.
Given all of this, he can tell that she is still reliable enough at that distance. But the
player is very far from knowing any such thing. Consider now her aims as she shoots
from that distance. She may attain her basic aim: namely, to sink that basket. And in
that respect her shot may be apt. It may succeed in a way that manifests the compe-
tence that the statistician knows her to possess even at that distance. So, she has ani-
mal competence and her shot is not just successful, not just competent, but also apt.
The success of the shot and not just its existence manifests the competence resident in
the player even at that distance.
6. What then is she missing? Anything? Well, although she attains her first order
objective, and does so aptly, she does not aptly attain her reflective aim of succeeding
competently, reliably enough. She does succeed reliably enough, but her attaining this
objective is a matter of luck. Unlike the statistician, she is unable to tell that her shot
is still reliable enough even at that distance. If she shoots anyhow from that distance,
and it turns out that her shot is reliable enough, then she may attain her objective of
succeeding reliably enough. However, her attainment of this objective will not be apt.
It will be relevantly by luck, so that the shot is not creditable to her, not fully anyhow.
So, her first-order success will be apt but it will not be meta-competent and hence not
meta-apt, and so it will not be fully apt.
G. The Nature and Epistemology of Competences

Competences are a special case of dispositions. Suppose a solid iron dumbbell would
shatter upon hitting a certain surface only because a hovering fiend is determined to
zap it if and only if it hits the surface, and just as it does so. This would make the
relevant conditional true: the dumbbell would shatter on that occasion, but this would
not make it fragile. For the latter to be the case, one would need at a minimum that the
dumbbell would shatter whenever relevantly impacted in a relevantly “interesting”
combination of condition and situation.
Just what makes a triple-S combination of structure, shape, and situation “interest-
ing” is an interesting, and neglected, question. Such combinations constitute the innu-
merable competences of interest in the many performance domains recognized by
human communities. Such domains—whether athletic, artistic, intellectual, medical,
scientific, legal—contain performances aimed at certain distinctive aims, along with
the competences whereby performers might succeed more or less competently. When
the success manifests such competence, then it is apt, and only then. The SSS profile
that constitutes any such competence is not always subject to full and detailed linguis-
tic formulation. But in that respect it joins much else of substantial human interest and
importance, such as when conduct counts as polite.
What conduct counts as polite? This is much easier to discern by perception in a
particular situation, than it is to formulate linguistically in full detail. This is so for
politeness in general, as well as for particular matters of politeness, such as the distance
that is politely appropriate for normal conversation in person, and such as the tone and
volume of voice that counts as rude, to take just two out of many instances.
Moreover, that is not to say that rules of polite conduct are just a myth. Not all rules
need to be formulated linguistically. But if what is polite is defined by human conven-
tion, then there must be rules in some broad sense that communities agree upon ahead
of time. There need not be a constitutional convention to settle what those rules are.
Their institution may be more organic, less artificial than that. Moreover, the agree-
ment within the community will not require so much as linguistic communication. It
may be instituted more implicitly than that, through persistent profiles of approval/
disapproval, praise/blame, and systematic and implicit manifestation in individual
and social conduct.
Returning to our own issues, this pattern of normativity seems applicable also to
competences. Thus, the SSS profiles of the competences that acquire salience for com-
munities may be determined not by antecedent, linguistically formulable convention,
but rather by persistent implicit profiles of the sort that determine the content of
politeness for that community.
What is more, the relation is not just a matter of coincidence or mere analogy. There
is after all such a thing as competence in etiquette, and even in morality. It seems possi-
ble to conceive of these domains either in terms of rules implicitly operative, or in
terms of competences determined by community interest.6
6
Of course, that is not to say that all such domains have their normative source or basis in human con-
vention. Nor need we suppose convention-based domains to lie beyond objective assessment in global or
specific respects. They may still be subject to such assessment on how well they further any value they may
be designed to secure.
30 Ernest Sosa
References
Greco, J., 2009. “Knowledge as Success from Ability.” Philosophical Studies 142: 17–26.
Millar, A., 2010. “Knowledge and Recognition.” In D. Pritchard, A. Millar, and A. Haddock
(eds) The Nature and Value of Knowledge: Three Investigations. Oxford: Oxford University
Press.
Sosa, E., (forthcoming). “Knowledge in Action.” In A. Bahr and M. Seidel (eds) Ernest Sosa:
Targeting His Philosophy, Springer, 18th Münster Lectures.
2
Dispositional Robust Virtue
Epistemology versus Anti-luck
Virtue Epistemology
Jesper Kallestrup and Duncan Pritchard
1. Introductory Remarks
It is difficult to express the tremendous importance of Ernest Sosa’s work for contem-
porary epistemology. One litmus test in this regard is the fact that Sosa’s work is so
much a part of the furniture of mainstream epistemology that the student of this disci-
pline simply “swallows down” a grip on his ideas in virtue of studying the area.1 Our
interest in this essay is one particular aspect of Sosa’s thinking in epistemology, which
is his distinctive theory of knowledge.
In §2, we set out Sosa’s theory of knowledge, which we refer to as dispositional robust
virtue epistemology. In §3–4, we contend that Sosa’s view is nonetheless ultimately
untenable. In particular, we argue that Sosa’s account of knowledge is unable to accom-
modate what we call the epistemic dependence of knowledge, where this epistemic
dependence is shown to have both a negative aspect (§3) and a positive aspect (§4).
Finally, in §4, we demonstrate that there is an alternative proposal available, anti-luck
virtue epistemology, which can accommodate the phenomenon of epistemic depend-
ence, and which should thus be preferred to Sosa’s proposal.
2. Sosa’s Dispositional Robust Virtue Epistemology

Sosa is the father of contemporary virtue epistemology (indeed, arguably he is the
father of contemporary epistemology simpliciter). While there are many different vari-
ants of virtue epistemology, the core idea is that the notion of an epistemic virtue
should play a fundamental role in one’s epistemology. It is through Sosa’s pioneering
1
We here adopt an analogy offered by Wittgenstein (1969: §143).
32 Jesper Kallestrup and Duncan Pritchard
work in this area that virtue epistemology has gone from being largely an ignored
option to one of the central movements in contemporary philosophy.2
Our interest here is in the specific virtue-theoretic theory of knowledge which
Sosa advances. In particular, we are interested in Sosa’s claim to have offered a the-
ory of knowledge along exclusively virtue-theoretic lines. This is controversial, for
while there is a strong prima facie case for the weak claim that there should be some
sort of virtue-theoretic condition on knowledge, the case for the strong claim that
one could offer an exclusively virtue-theoretic theory of knowledge is prima facie
problematic.
The motivation for the weak claim, in broad outline, is that a virtue-theoretic condi-
tion on knowledge is required in order to capture the sense in which when one knows
it is down to the proper exercise of one’s cognitive agency. The point is that virtue the-
ory offers us the best way of understanding what the proper exercise of one’s cognitive
agency involves. That is, it involves the exercise of cognitive abilities—epistemic virtues,
broadly conceived—where these are stable and integrated features of one’s overall
cognitive character.
On the face of it, however, the strong claim that knowledge might be nothing more
than cognitive success (i.e., true belief) which is due to one’s epistemic virtue looks
dubious. After all, doesn’t the agent in a typical Gettier-style case exhibit epistemic
virtue in their formation of a true belief and yet nonetheless lack knowledge?3 In a
number of works, however, Sosa (1988, 1991, 2007, 2009, 2010, 2011, 2013, forthcom-
ing) has developed a particular rendering of virtue epistemology which can offer a
compelling account of knowledge.
In terms of Sosa’s terminology, knowledge has a “triple A” structure. A belief which
is true is accurate, and a belief which is formed via an epistemic virtue is adroit. In
addition, Sosa delineates a further epistemic category whereby the subject’s belief
is not only adroitly formed and accurate, but also accurate because adroitly formed.
Sosa refers to this as apt belief. Apt belief, for Sosa, is knowledge. It is held to be pre-
cisely aptness which is lacking in the Gettier-style cases, in that while there is accuracy
(the agent believes truly) and while there is adroitness (the agent forms her belief via a
reliable cognitive disposition), the belief is not accurate because adroit, and that’s why
it doesn’t amount to knowledge.
This requires some unpacking. First, another piece of terminology, this time of our
own. We will refer to a theory of knowledge which is purely virtue-theoretic—that is,
which requires of a knowing agent who has a true belief only that, in addition, they
2
For some helpful surveys of the literature on virtue epistemology, see Axtell (1997), Greco (2002),
Baehr (2004), Greco and Turri (2009), Kvanvig (2010), and Turri and Sosa (2010). See also Baehr (2008).
3
Indeed, many virtue epistemologists have gone down the route of arguing that virtue epistemology
should not be in the business of offering a theory of knowledge in the first place; that virtue epistemology
should instead be seen as “re-orienting” the concerns of traditional epistemology rather than responding
to them. See, for example, Code (1987), Kvanvig (1992), Montmarquet (1993), Hookway (2003), and
Roberts and Wood (2007).
Dispositional Robust Virtue Epistemology 33
meet a virtue-theoretic condition—as robust virtue epistemology. Sosa’s proposal is of

this sort, since there is nothing in his theory of knowledge beyond epistemic success
(true belief, or accuracy as he calls it) and the appropriate relationship between this
epistemic success and the manifestation of epistemic virtue.
Second, we need to note that Sosa explicitly understands epistemic virtue (or
adroitness, in his terminology) in terms of the manifestation of a cognitive disposi-
tion, where these cognitive dispositions have a physical basis resident in the cognitive
subject (see, e.g., Sosa 2007: 29, 2009: 135). Relatedly, when Sosa talks of knowledge
(i.e., apt belief) being the result of accuracy that is because of adroitness, the “because of ”
at issue here should be understood in terms of a particular kind of manifestation of a
disposition. This dispositional aspect of Sosa’s view is distinctive, and this is why we
will henceforth refer to his position as dispositional robust virtue epistemology, or
“DRVE” for short.4
To understand the distinctiveness of Sosa’s position on this score, it is worthwhile
comparing it with a leading alternative robust virtue-theoretic proposal, due to John
Greco (2003, 2007, 2008, 2009a, 2009b, 2009c). Although superficially similar to Sosa’s
proposal (and for good reason, since it was largely inspired by Sosa’s pioneering work
on virtue epistemology), in that it also regards knowledge as being true belief that is
because of epistemic virtue, Greco’s account of knowledge differs on a fundamental
level by offering a very different account of what this “because of ” relation amounts to.
In particular, according to Greco we are to read the “because of ” not in terms of the
manifestation of a cognitive disposition, as Sosa suggests, but rather along causal
explanatory lines. That is, to say that an agent’s cognitive success (i.e., true belief) is
because of epistemic virtue is on this view to say that the agent’s epistemic virtue is the
most salient part of a causal explanation of that cognitive success.5
To see how these two accounts of the “because of ” relation can come apart, consider
a glass that was broken as a result of someone dropping it onto a wooden floor.
Ordinarily, the most salient part of the causal explanation of why the glass broke will be
that someone dropped it on the floor, and in this sense it will be true to say that the
4
In more recent work, Sosa (2013; forthcoming) proposes a triple-S analysis of a complete compe-
tence comprising an innermost S-competence, which is the seat (or skill), an inner SS-competence,
which is the combination of seat and shape, and a complete SSS-competence, which is the conjunction
of seat, shape, and situation. The connection between the triple-A analysis of a performance and the
triple-S analysis of a competence is the following: a performance is apt when its success manifests com-
petence, which happens just in case the innermost skill causally produces the success in combination
with the appropriate shape and situation. The seat of the competence is determined as the causal basis
for a success-response of an object when subjected to a stimulus in certain shape and situation combi-
nations. Since a complete competence is necessarily a competence to succeed when trying for some
outcome such that if one tried, one would very likely succeed, no such competence can bring about
that outcome. Only the innermost seat can do that when the shape and situation are conducive to
the outcome.
5
For example, Greco (2009a: 12) writes that “the term ‘because’ . . . marks a causal explanation.” Later on
(Greco 2009a: 12) he makes clear that the agent’s abilities must be the overarching element in the causal
explanation in question when he states that “in cases of knowledge, S’s believing the truth is explained by
S’s abilities, as opposed to dumb luck, or blind chance, or something else.”
glass broke because it was dropped in this way. Note, however, that this is consistent
with the claim that it was because of the glass’s fragility that it broke, since here we are
talking about the manifestation of a disposition and not offering a causal explanation.
If fragility is the (second-order) property of having a property that causes breaking if
dropped, then we cannot causally explain why the glass broke when dropped in terms
of its fragility. We can say the glass broke when dropped because the glass has a
(first-order) molecular bonding property. What is causally responsible for the shatter-
ing is this micro-structural property together with the dropping. The dispositional
property itself, thus understood, is causally inefficacious of the effects in terms of
which it is defined.6
In the same way, according to Sosa we are to think of apt belief in terms of a belief
which is accurate because adroit in the specifically dispositional, rather than causal
explanatory, sense that the accuracy manifests an epistemic ability on the part of the
subject. Unlike Greco’s alternative proposal, then, Sosa’s virtue epistemology thus trades
on a broader metaphysical picture of dispositions and powers, where the manifestation
of a cognitive disposition mirrors the manifestation of dispositions and powers more
generally. We suggest that this is an attractive feature of the view.7
We will not be delving into the relative merits of DRVE over its causal explanatory
counterpart here.8 Our interest is rather in whether DRVE is a tenable theory of knowl-
edge. We will be arguing that it isn’t.
6
Even so, dispositional properties frequently enter into causal explanations in science. For instance,
Richard Feynman explained why the space shuttle Challenger blew up in terms of the failure of an O-ring
in one of the solid rocket boosters to expand at lift-off. Jackson (1996: 397) suggests that causal expla-
nations by dispositional properties provide two kinds of information: (i) the effect was caused by the
categorical basis of the disposition, and (ii) the effect is one of the outputs in terms of which the disposi-
tion is defined. (ii) is required because some base properties can ground more than one disposition—e.g.,
electrical and thermal conductivity in metals share the same categorical basis. What Feynman discovered
was that the categorical basis of the rigidity caused the disaster, and the disaster resulted from the kind
of output distinctive of rigidity: the O-ring failed to expand after compression and its failure led to the
disaster.
7
We explore the metaphysical foundations of Sosa’s proposal in more detail in Kallestrup and Pritchard
(2013b). Is the causal explanatory reading of “because of ” the only viable alternative to DRVE? Arguably,
it is the only alternative which has been properly developed in the literature. Zagzebski (e.g., 1999) favors
a virtue-theoretic account of knowledge which treats the “because of ” relation as an undefined primitive.
While this is a dialectical option, it is obviously preferable to have a virtue-theoretic account of knowledge
which can spell out what this relation amounts to. In his most recent statement of his view, Greco (2012)
explores some possible ways of cashing out the virtue-theoretic account of knowledge, and sketches a
variant on the “causal-explanatory” proposal we have noted above, one that puts more of an emphasis
on pragmatic factors.
8
One potential advantage that Sosa’s DRVE has over Greco’s competing view is that the latter is clearly
wedded to a form of pragmatic encroachment in epistemology, a consequence of the view which Greco has
increasingly embraced. See especially Greco (2012), but also the exchange between Greco (2008) and
Pritchard (2008a). While pragmatic encroachment is not without its defenders—see, for example,
Hawthorne (2004), Stanley (2005), and Fantl and McGrath (2009)—it is nonetheless a highly controversial
thesis. Accordingly, that Sosa’s rendering of robust virtue epistemology can avoid pragmatic encroachment
potentially puts it at an advantage.
3. DRVE and Negative Epistemic Dependence

The problem that we will be posing for Sosa’s DRVR is that it is unable to accommodate
a phenomenon which we have elsewhere termed epistemic dependence.9 This phenom-
enon concerns the fact that the conditions under which knowledge is possessed, and
fails to be possessed, can be significantly dependent upon factors outwith the cognitive
agency of the subject. In particular, knowledge possession can be dependent upon fac-
tors outwith the cognitive agency of the subject to an extent which is inconsistent with
a robust virtue-theoretic account of knowledge like DRVE.
There are two sides to epistemic dependence, negative and positive. Negative epis-
temic dependence is when an agent who would ordinarily count as knowing fails to
know because of factors outwith her cognitive agency. Positive epistemic dependence
is when an agent who would ordinarily not count as knowing nonetheless possesses
knowledge on account of factors outwith her cognitive agency. We will take these two
types of epistemic dependence in turn.
The kind of negative epistemic dependence that interests us is one which is extensive
enough to be incompatible with a purely virtue-theoretic account of knowledge like
DRVE. In particular, and contrary to DRVE, negative epistemic dependence occurs
when an agent’s belief meets the conditions that Sosa lays down for aptness and yet
does not constitute knowledge because of factors external to the cognitive agency of
the subject. That knowledge is subject to negative epistemic dependence of this sort
can be neatly brought out by appealing to what we call epistemic Twin Earth cases.10
Standard Twin Earth arguments run as follows.11 Despite appearances there is no
water on twin earth. Water is essentially H2O, and all the watery stuff on twin earth has
the different microstructure XYZ—earthlings call that “twin-water.” When S on earth
utters “water is wet,” she expresses the proposition that water is wet, but when S’s
intrinsic physical duplicate on twin earth utters the same sentence, twin-S expresses
the proposition that twin-water is wet. Since S and twin-S refer to different kinds of
stuff when they token “water” the truth-conditions of their respective utterances differ.
Assuming the contents of their beliefs are fixed by the truth-conditional contents of
the sentences that they use to express those beliefs, then these belief contents also
fail to supervene on their intrinsic physical properties. Indeed if belief states are indi-
viduated in part by their contents, then what belief states S and twin-S are in fail to
supervene on their intrinsic physical properties. Instead these states depend partially
9
See, especially, Kallestrup and Pritchard (2013a).
10
We first proposed epistemic Twin Earth cases in Kallestrup and Pritchard (2011). See also Kallestrup
and Pritchard (2012, 2013a, 2013b). It is worth noting that the epistemic Twin Earth argument that we
propose is significantly different to the moral Twin Earth argument that has been proposed by Horgan and
Timmons (1991, 1992) and Timmons (1999: chapter 2), and which calls into question a certain form of
moral naturalism.
11
See, for example, Putnam (1975). For a recent discussion of standard Twin Earth arguments, see
Kallestrup (2011: chapter 3).
for their individuation on which patterns of causal relations S and twin-S bear to their
respective physical environments.
Consider now epistemic twin earth on which most watery stuff is H2O. In between
there is some scattered twin-water the exact location of which varies from case to case.
Our contention is that an epistemic Twin Earth argument shows that whether a subject
is in a perceptual knowledge state cannot merely be a question of getting things right
through exercising her cognitive abilities in the way that robust virtue epistemology
suggests.
Let’s divide epistemic twin earth into three regions. The subject’s local environment
is where the subject is currently located. It contains the objects and properties that are
the proximate causes of her current perceptual experiences. Take facts to be objects
instantiating a property at a time. If the subject now perceives that p, then the fact that
p (the “p-fact”) is one that concerns her local environment—that is, it is a local fact.
Other local features have to do with aspects of the perceptual process and various
background conditions on perception—for example, distorting noise, brightness.
The subject’s regional environment is neither where the subject is currently located,
nor where she typically forms any beliefs. Still, it contains the objects and properties
with which she might easily have been causally connected. If the q-fact is such that if
the subject had not now perceived that p then she would have perceived that q, then the
q-fact is one that pertains to her regional environment—that is, it is a regional fact.
Regional facts, thus understood, are nearby perceptual possibilities, but they play no
causal role in producing the subject’s current perceptual experience on which she
bases her belief that p.
Finally, the subject’s global environment is where she is normally located although
not at present. It contains the objects and properties with which she ordinarily caus-
ally interacts. The global facts thus comprise all the facts that extend in space-time
beyond the regional facts. Assuming the subject now perceives the local fact that p,
the fact that r is a global fact only if she would not have perceived that r had she not
perceived that p. Given the subject’s current location, global facts are not only dis-
tant perceptual possibilities, they are also causally inefficacious in producing her
current perceptual experiences.
We can now mount an epistemic Twin Earth argument to the effect that robust vir-
tue epistemology is an inadequate account of knowledge. The subject, S, is on earth
where all watery stuff is H2O. S’s perceptual apparatus is highly reliable in that a high
frequency of S’s perceptual beliefs is both actually true and true across relevantly close
worlds. Based on a perceptual experience as of water, S forms the demonstrative belief
that that’s water. There is no question that S thereby comes to know just that.
On epistemic twin earth S has an intrinsic physical duplicate called “twin-S”. S and
twin-S are conceptually competent but chemically ignorant. On epistemic twin earth
all watery stuff in twin-S’s global environment is H2O. Not only is twin-S therefore able
to entertain water-thoughts, a high frequency of twin-S’s water-beliefs as formed in
her global environment is true both in actual fact and across relevantly close worlds.
Twin-S’s perceptual apparatus, as exercised in that environment, is thus equally relia-

ble. Moreover, all watery stuff in twin-S’s local environment is H2O. When twin-S
forms the demonstrative belief that that’s water on the basis of a perceptual experience
as of water, her belief is true. Yet, unbeknownst to twin-S, twin-water is abundant in
her regional environment. The basis on which twin-S holds that belief is thus such that
her belief is only luckily true—it is unsafe, to use contemporary terminology—in that
given the basis for her belief it could very easily have been the case that she would have
formed a false belief (e.g., had she been interacting, unbeknownst to her, with twin-
water). That is to say, very easily could twin-S have believed that that’s water on the same
basis—a perceptual experience as of water—without that being so.12 On the plausible
assumption that knowledge excludes such environmental luck—in virtue of the fact
that knowledge requires safe belief; that is, true belief formed on a basis such that it
could not have very easily been formed on that basis and be false—it follows that
twin-S lacks knowledge.13
We can illustrate what is going on here with the following diagram:
Earth Epistemic Twin Earth
Global: H2O Global: H2O
Regional : H2O Regional : XYZ
Local : H2O Local : H2O

S Twin-S
The explanation a robust virtue epistemic account of knowledge like DRVE offers of
why S has knowledge on earth is that her cognitive success is because of her cognitive
12
If twin-S uttered “that’s water” while demonstrating twin-water she would express the false proposi-
tion that that’s water. We assume that the concept of water as deployed on both earth and epistemic twin
earth is a natural kind concept that applies to all and only H2O. One might envisage a loophole here for the
robust virtue epistemologist if twin-S’s utterance has the purely descriptive truth-condition: “that’s water”
is true iff that’s watery stuff, or the disjunctive truth-condition: ‘that’s water’ is true iff that’s water or that’s
twin-water. We find both views implausible. The presence of small amounts of twin-water on epistemic
twin earth implies neither that water is a functional kind in the way that, say, vitamin is, nor that water is a
disjunctive kind in the way that, say, jade is. Even those with descriptivist or semantic internalist leanings
insist that, to a first approximation, “water” in someone’s mouth picks out the dominant watery stuff of
their acquaintance. XYZ is neither dominant nor stuff with which adequate causal connections are
sustained. In fact, Chalmers (1996: 58) is explicit that if the watery stuff in our world turned out to be a
mixture of 95% H2O and 5% XYZ, the primary intension of “water” would pick out only H2O. For more
details, see Kallestrup (2011: chapters 3 and 4).
13
For some key elaborations of the safety condition, see Sainsbury (1997), Sosa (1999), Williamson
(2000: chapter 5), Pritchard (2002, 2005, 2007, 2008b, 2012a, 2012b, 2015, forthcomingb), and Luper
(2003; cf. Luper 1984). For a recent debate regarding the necessary of safety for knowledge, see the exchange
between Hetherington (2013) and Pritchard (2013).
ability. That is, her belief is apt. The challenge, however, is to explain why twin-S lacks
knowledge on epistemic twin earth, since her belief seems to be no less apt. Indeed, the
fact that S and twin-S are intrinsic physical duplicates embedded in physically identi-
cal global environments means that the one subject cannot possess a cognitive ability
that the other lacks. And the fact that S and twin-S are currently located in physically
identical local environments rules out the possibility that only one of them manifests
that ability. So if S’s belief is apt, then so too must be twin-S’s belief.
To use an analogy that Sosa is himself fond of, suppose S is an expert archer. S
possesses that ability in virtue of relevant bodily/psychological features and mostly
occupying an environment that is conducive for her to frequently hit the innermost
rings when dispatching arrows. Given that the latter are equally true of twin-S, she will
be an expert archer too. And the fact that both S and twin-S currently occupy physically
identical local environments means that their cognitive successes must arise in the very
same way. To use the analogy, the ways in which S and twin-S propel their respective
arrows into the yellow ring are identical. After all, fletching, bow strings, body posi-
tions, prevailing winds, distances to target, energy imparted to arrows, and so on, are
identical in the two cases. Combining these two facts spells trouble for DRVE, for it
deprives Sosa of a principled basis on which he can treat the two cases differently—on
Sosa’s view, both agents exhibit apt belief. And yet there clearly is an epistemic differ-
ence between them, in that twin-S, unlike her counterpart S, lacks knowledge.
It’s not clear how Sosa could respond to this argument. Recall that he explicitly con-
ceives of cognitive abilities in terms of cognitive dispositions that have a physical basis
resident in whoever has those dispositions (see Sosa 2007: 29, 2009: 135). If that’s right,
however, then it is difficult to see why twin-S should lack a cognitive ability that S pos-
sesses given that they are physically identical. For whatever physical basis is sufficient
for S to possess her cognitive ability is a basis shared by twin-S. Of course, which cogni-
tive abilities S possesses depend on environmental features such as operative laws or
law-like regularities and physical background conditions. But the relevant environ-
ment here is the one in which S is typically embedded. There are neither nomological
differences between earth and epistemic twin earth, nor any physical differences
between S and twin-S’s global environments.14
The analogy with other physical dispositions, such as solubility, is instructive here.
After all, these dispositions are such that the instantiation by an object of the physical
14
Sosa’s more recent account of competences, as sketched in note 4, offers no solution to the current
problem. Sosa takes cases of environmental and situational luck to be cases where the agents are completely
competent even though they are in danger of lacking the situation required for possession of such compe-
tences. Barney in fake-barn country and Simone in the cockpit are cases in point. What the threat of the
barn façades/simulation cockpit takes away is not their complete competence, but rather the safety of their
belief. Barney and Simone’s beliefs are apt yet unsafe. It should be clear that by Sosa’s own lights, twin-S is
completely competent if S is. Their seat and shape are certainly identical, but so is also their situation. Keep
in mind that Sosa understands a situation as pertaining to local features: Barney is not too far away, and
Simone is not too far up. Both are in good light, with a clear line of vision, and so on. It follows that the
situations (required for complete competence) that S and twin-S are in are identical since their local envi-
ronments are identical.
base property for the disposition physically necessitates the instantiation of the dispo-
sitional property. So, taking the case of solubility as an example, as long as the laws of
physics are fixed, any intrinsic physical duplicate of a solute is also soluble. To find such
a duplicate that is not soluble you must go to a world with deviant laws of physics. It is
hard to see why cognitive dispositions should be so different as to come and go with
hidden variations in particular physical facts in the regional environment.
Following Greco (2009c: 21–2), one possible response to the epistemic Twin Earth
argument is to maintain that twin-S lacks the cognitive disposition to tell water from
non-water when occupying the particular local environment we have sketched due
to the fact that manifesting that discriminatory disposition in that environment is
unreliable, where reliability requires a high frequency of true beliefs, actually and
across relevantly close possible worlds. In contrast, twin-S clearly does possess that
disposition relative to her global environment, because manifestations in that environ-
ment do issue in a high frequency of true beliefs as a matter of fact and in nearby possi-
ble worlds. In reply, we agree that disposition possession is both environment-relative
and requires reliability. But we maintain that whether a cognitive disposition is pos-
sessed in a given environment depends on whether the agent reliably manifests that
disposition in her global environment. On the assumption that twin-S possesses the
disposition to discriminate between water and non-water relative to her global envi-
ronment she also possesses that disposition in her local environment. True, manifest-
ing that disposition in the latter environment could very easily lead S astray, but that
does not rob her of the disposition. Compare with archery. Suppose an expert archer
dispatches an arrow which then propels through the innermost ring. Nothing in the
local environment prevents her from manifesting that ability. Suppose a sudden and
unexpected gust could very easily have diverted the arrow off course. That regional fact
does not imply that she fails to possess her archery ability. All it shows is that in her
local environment manifestations of that ability could very easily have resulted in her
missing the target. Contrast with a novice archer who surprisingly also hits the target
in identical environmental conditions, despite not having previously dispatched a sin-
gle arrow. Surely, the expert deserves more praise, because only her achievement stems
from a praiseworthy ability.
On our view, dispositions—cognitive or not—are characterized by a kind of modal
sturdiness that knowledge lacks in the sense that nearby unactualized possibilities may
well undermine reliability and hence knowledge, while leaving dispositions and their
manifestations intact. Once cognitive dispositions are acquired through nurture or
nature, they are retained even when the regional facts prevent their manifestations
from being reliable. Indeed, to appreciate that possession of such dispositions are rela-
tive only to global environments they need only be specified in sufficient detail. For
instance, on the assumption that one has the disposition to tell liquids that resemble
water from those that do not, having the disposition to tell water from non-water con-
sists in having the dispositions to tell water from distinct resembling liquids and to tell
water from distinct non-resembling liquids. Consequently, on the one hand, as S and
twin-S do not possess the former disposition to tell water from twin-water relative to
their shared global environment, neither possesses that disposition in their identical
local environment. That is not a disposition they possess anywhere. On the other hand,
as S and twin-S do possess the latter disposition to tell water from, say, beer or petrol
relative to their shared global environment, both possess that disposition in their iden-
tical local environment. That is a disposition they possess everywhere. The upshot is
that the difference in regional environments between S and twin-S makes for no differ-
ence in which dispositions are possessed.
Although Sosa has not responded to the problem of negative epistemic dependence
as presented in the way that we have just described it (i.e. via an epistemic Twin Earth
case), he does offer remarks in recent work which suggest how he might be inclined to
deal with this problem.
Sosa draws a distinction between first-order, animal knowledge, and second-order
reflective knowledge. The former is just apt belief, as described above. The latter, how-
ever, is more demanding, in that it is, as he puts it, “apt belief, aptly noted,” where this
means that it is an apt second-order belief that one has the corresponding first-order
apt belief. Elsewhere in his work Sosa exploits this distinction between animal and
reflective knowledge in order to explain why on his view subjects have knowledge in
cases where epistemologists often deny that the subject concerned is a knowing sub-
ject. That is, in such cases he contends that what the subject is lacking is not knowl-
edge simpliciter (i.e., animal knowledge), but specifically reflective knowledge.15
Perhaps, then, Sosa could extend this strategy to the epistemic Twin Earth argument
and hence contend that while S on earth has both animal and reflective knowledge,
twin-S on epistemic twin earth qualifies only for the former. In this way, he could
argue that while it is true that if S has knowledge of the target proposition then so does
twin-S, there is nonetheless an epistemic difference between the two subjects just as
intuition dictates, albeit an epistemic difference which is at the level of reflective
knowledge rather than knowledge simpliciter. In effect, the idea would be to bite the
bullet of our objection, while nonetheless mitigating the overall force of the objection
for the view.
One immediate problem with this line of argument is that it does not seem at all
essential to the epistemic Twin Earth argument that either of these subjects should
even have the relevant second-order belief. But insofar as they lack this belief then they
are thereby both excluded from having reflective knowledge, and hence one cannot
explain away the intuition that there is an epistemic difference between these two
virtue-theoretic duplicates by appealing to the thought that one of them has reflective
15
For example, Sosa (e.g., 2007: chapter 5; cf. Sosa 2011) grants that on his view the subject in the “barn
façade” case can have knowledge, even though most epistemologists deny this (on account of the fact that
the belief in question is only luckily true, and hence unsafe). His explanation of what is going on here is
that the subject concerned merely has animal knowledge of the target proposition, and not also reflective
knowledge. For a specific discussion of Sosa’s view on this score, see Pritchard (2009).
knowledge that the other lacks. Furthermore, note that this problem does not trade on
the particular rendering of robust virtue epistemology that is in play.16
Even if we ignore this problem and allow that the subjects concerned have the rele-
vant second-order beliefs, however, this line of response to the epistemic Twin Earth
argument still flounders. The reason for this is that we appear to be perfectly able to run
a corresponding epistemic Twin Earth argument which is specifically focused on
reflective knowledge.
Recall that animal knowledge and reflective knowledge differ on Sosa’s view in
terms of the different cognitive dispositions that are being manifested in each case.
Whereas animal knowledge involves the manifestation of a first-order cognitive dis-
position, reflective knowledge involves the manifestation of a second-order cognitive
disposition that is tracking the cognitive success of the relevant first-order cognitive
disposition. Presumably, however, all dispositions, whether first- or second-order, have
physical bases that are resident in whoever has the disposition. Certainly if, as Sosa
acknowledges, first-order cognitive dispositions have physical bases, then the claim
that second-order cognitive dispositions lack physical bases would seem ad hoc. But
then how is the proponent of this hypothetical line of response to the epistemic Twin
Earth argument to explain how S on earth and twin-S on epistemic twin earth could
enjoy different second-order epistemic statuses when they and their local and global
environments are physically identical?
The foregoing may be thought slightly unfair to Sosa. In response to his “kaleido-
scope” case, Sosa (2007: 33; cf. Sosa 2009: 238–9) proposes that S’s meta-belief is true
because of a competence only if “it derives from the exercise of that competence in
appropriate conditions for its exercise, and that exercise in those conditions would not
then too easily have issued a false belief.” The latter supplementation is certainly fit for
purpose. Since the identical, local environments in which S and twin-S are embedded
are identical to their global environments, both form true beliefs in conditions appro-
priate for exercise of their respective cognitive abilities. But while S’s exercise of her
cognitive ability results in true beliefs across nearby worlds, twin-S’s exercise would
indeed all too easily have issued a false belief. For that reason S has animal knowledge
that she has animal knowledge, but twin-S lacks such reflective knowledge. Assuming
with Sosa (2007: 31–40) that first-order animal knowledge does not also require that
the exercise of the pertinent cognitive ability in normal conditions would not very eas-
ily have issued a false belief, twin-S can retain such knowledge.
It should be pretty obvious by now why this proposal offers no comfort for DRVE.
The explicit recommendation is that reflective knowledge requires the exercise of a
cognitive ability in conditions that would not lead to such knowledge being under-
mined by epistemic luck. For what prevents twin-S from having reflective knowledge is
that the conditions in which she exercises her meta-cognitive ability are such that her
cognitive success could very easily have been cognitive failure. A non-virtue-theoretic
16
This point is also discussed in Kelp (2012).
condition designed to deal with knowledge-undermining epistemic luck is thus explic-

itly built into reflective knowledge. Consequently, the proposal cannot explain away
the intuition that S and twin-S differ epistemically in a way that is consistent with the
fact that DRVE is meant to be a type of robust virtue epistemology.17
Finally, note that if Sosa does respond to the epistemic Twin Earth cases by appeal-
ing to the distinction between animal and reflective knowledge, then he is committed
to allowing that knowledge—animal knowledge at any rate—is compatible with lucky
(i.e., unsafe) cognitive success. In particular, he is committed to allowing that one can
have animal knowledge even while forming one’s true belief on an unsafe basis—that
is, such that one could very easily have formed a belief on that basis and believed falsely.
Given that it is widely accepted that knowledge demands safety, or a similar modal
condition, this is far from being a negligible concession for Sosa to make.18
If the epistemic Twin Earth argument goes through, however, then it will not be
possible to offer a pure virtue-theoretic account of knowledge of the sort that Sosa
offers with DRVE. The problem is that such a proposal cannot accommodate the extent
to which knowledge can be negatively epistemic dependent. In the epistemic Twin
Earth case we have two subjects who are internal duplicates and who manifest their
cognitive abilities to the same extent within identical local and global environments
(the only environments relevant to the manifestation and possession, respectively, of a
cognitive disposition, as we have seen), but where factors outwith one of the subject’s
cognitive agency ensures that she fails to have knowledge, unlike her counterpart. In
particular, in terms of the specifics of DRVE, both the beliefs in play in these two cases
meet the conditions for aptness, but only one of them, because of the negative epis-
temic dependence in play, meets the conditions for knowledge. The moral to be drawn
is that knowledge should not be identified with aptness as DRVE suggests.19
4. DRVE and Positive Epistemic Dependence

We come now to positive epistemic dependence. The kind of positive epistemic
dependence that interests us is one that is extensive enough to be incompatible with a
pure virtue-theoretic account of knowledge like DRVE. In particular, and contrary to
17
In Kallestrup and Pritchard (2011: §5) we raise a similar complaint against a modification of DRVE
offered by Turri (2011). Roughly, Turri argues that we should require for knowledge not just apt belief, but
also ample belief, where this means that the belief is safe and where its safety manifests adroitness. To take
such a line is, however, to move away from the project of offering a robust virtue-theoretic account of
knowledge of a kind exemplified by DRVE. Moreover, this proposal will not have the advantages that we
show anti-luck virtue epistemology (“ALVE”) to have below. For while ALVE also incorporates a safety
condition, it is crucial to the success of this proposal that it does not merely add a safety condition to
aptness. Instead, ALVE combines a safety condition with a weaker virtue-theoretic condition than aptness.
18
Though widely accepted, the idea that knowledge demands safety has been contested in the recent
literature. For a good overview of the issues in this regard, see the exchange between Hetherington (2013)
and Pritchard (2013). See also Pritchard (2012b, 2015, forthcomingb).
19
The problem that epistemic Twin Earth cases pose for robust virtue-theoretic accounts of knowledge
is explored in more detail in Kallestrup and Pritchard (2011).
DRVE, positive epistemic dependence occurs when an agent’s belief fails to meet the
conditions that Sosa lays down for aptness and yet nonetheless constitutes knowl-
edge because of factors external to the cognitive agency of the subject. We can illustrate
the phenomenon of positive epistemic dependence by appeal to a certain kind of testi-
monial knowledge.
On standard views of the epistemology of testimony, in epistemically favorable con-
ditions it is possible to gain testimonial knowledge by, for the most part, simply trusting
the word of one’s informant. That is, while one will be expected to exercise some signif-
icant degree of epistemic skill in one’s acquisition of this testimonial knowledge—for
example, it had better not be that one would believe anything that one is told, no matter
how outlandish—it is nonetheless the case that to a large extent one’s cognitive success
is down to factors which are outwith one’s cognitive agency.20
To take a standard kind of example to illustrate this point, imagine an agent who is
newly arrived in an unfamiliar city and who asks for directions.21 Let us stipulate that
the conditions are epistemically favorable, in the sense that all the informants in the
vicinity would be inclined to offer truthful and informative answers, and that there is
nothing else epistemically amiss occurring which might lead our hero to form false
beliefs on the basis of the testimony provided. Moreover, let us stipulate that our hero is
exercising some degree of epistemic skill in acquiring her testimonial belief. She would
not just ask anyone, but only plausible informants; she would not just believe anything
she is told, even something outlandish; and so on. This is thus not a case where our
agent is merely trusting an informant, since the intuition that someone can gain knowl-
edge from mere trust is far from secure. Nonetheless, the intuition is that a testimonial
belief formed via the exercise of relatively minimal levels of epistemic ability can in
epistemically favorable circumstances amount to knowledge. Indeed, if one does not
gain testimonial knowledge in epistemically favorable circumstances like these, then
testimonial knowledge is far less often possessed that we ordinarily suppose.
Here is the crux. While our hero is manifesting her cognitive abilities to some degree
in this case, it does not seem at all plausible to suppose that her cognitive success is
because of her cognitive abilities. That is, it does not appear that our hero’s belief quali-
fies as apt. Indeed, it seems that her cognitive success is to a large degree due to factors
outwith her cognitive agency, such as the cognitive abilities of her informant and the
20
That is, most epistemologists of testimony are inclined towards some version of anti-reductionism,
which is epistemically more liberal than its reductionist counterpart. It is precisely because of their anti-
reductionism that most epistemologists would tend to treat the subject in this case as having testimonial
knowledge. Reductionists, in contrast, would tend to regard this subject as lacking knowledge on account
of the degree of trust in play. While reductionism is not a popular view in the epistemology of testimony, it
does have some adherents. See, for example, Fricker (1995). For a very useful survey of contemporary work
on the epistemology of testimony, with special focus on the reductionism/anti-reductionism distinction,
see Lackey (2010). See also Carter and Pritchard (2010).
21
The basic kind of case in play here is attributable to Lackey (2007). Note, however, that we have made
certain changes to the case in order to ensure that it demonstrates the point that we have in mind (which
is importantly different from that which is the focus of Lackey’s paper). For more details, see Kallestrup and
Pritchard (2012: §2).
epistemically favorable nature of the environment. And yet she does seem to have
knowledge nonetheless.
We can bring this point into sharp relief by imagining, as before, two agents who are
internal physical duplicates who occupy essentially the same local and global environ-
ments, and who thus manifest the very same cognitive abilities. As before, the only
difference concerns the regional environment that the agents are in. For one of the
agents, the regional environment is epistemically favorable, just like the conditions
faced by our hero in the example just considered. For the other agent, in contrast, the
regional environment is not epistemically favorable, but in fact one which is highly
unfavorable. Perhaps, for example, the agent’s environment could so very easily have
been populated by dishonest informants rather than the honest informants she hap-
pens to interact with.
The problem should now be manifest. For while the agent who has an epistemically
favorable regional environment gains testimonial knowledge, her internal duplicate
who has an epistemically unfavorable environment does not gain testimonial knowl-
edge. And yet, as we have seen, there can be no difference in these agents’ possession or
manifestation of cognitive abilities. Whatever the reason for why they differ in terms of
what they know, then, it is not a difference that is a function purely of their manifesta-
tion of their cognitive abilities. In short, the epistemic difference between these agents
cannot be explained by DRVE, since neither of them are forming apt beliefs, and hence
neither of them should on this view count as having knowledge.22
Now Sosa does have something to say about such a case. He remarks that in such
cases the agent’s cognitive success is “attributable to a complex social competence only
partially seated in the individual believer” (2007: 97). It is unclear how to understand
this suggestion, however, and Sosa doesn’t offer much by way of explanation. We take it
that his idea is that whereas in standard cases of knowledge the epistemic competences
on display are solely that of the individual knowing agent, in testimonial cases like the
one under consideration there is instead a shared “social competence” that is displayed
by the cognitive whole of a “testifier-and-testifiee.”
The problem with this proposal is that it is entirely antithetical to the spirit of robust
virtue epistemology. This, after all, is the view that an agent has knowledge when her
cognitive success is because of her cognitive ability (only then is her belief apt, as Sosa
himself puts it). Here, for example, is Sosa describing what constitutes a competence:
[A] competence is a disposition, one with its basis resident in the competent agent, one that
would in appropriately normal conditions ensure (or make highly likely) the success of any
relevant performance issued by it. (Sosa 2007: 29, our italics)
But what Sosa is now claiming is directly in tension with this claim, since he seems to
be conceding that at least in some cases an agent can gain knowledge not in virtue of
22
For a more detailed discussion of positive epistemic dependence, and its relevance to virtue episte-
mology, see Kallestrup and Pritchard (2012, 2013a).
the exercise of their own epistemic competences but instead as a result of the part they
play in some wider social epistemic competence. This is not a minor “tweak” to DRVE,
but rather a radical departure, one that requires explanation and motivation.
Whereas the phenomenon of negative epistemic dependence demonstrates that
there is sometimes much more to knowledge than cognitive success that is because of
cognitive ability, the phenomenon of positive epistemic dependence demonstrates
that there is also sometimes much less to knowledge than cognitive success that is
because of cognitive ability. Either way, the upshot is that knowledge cannot be ana-
lyzed along robust virtue-theoretic lines purely in terms of the manifestation of the
subject’s epistemic virtues, and hence DRVE is untenable.
Finally, note that this problem of epistemic dependence, in virtue of having both a
negative and a positive aspect, effectively pulls the proponent of a robust virtue episte-
mology like DRVE in two directions. Negative epistemic dependency puts pressure on
the proponent of DRVE to strengthen the proposal, to make it less permissible; whereas
positive epistemic dependency puts pressure on proponents of DRVE to weaken the
proposal, to make it more permissible. In this way, the two aspects of epistemic depend-
ency pull a robust virtue-theoretic account of knowledge like DRVE asunder.
5. DRVE versus Anti-luck Virtue Epistemology

What makes matters worse for DRVE is that there is a different proposal available—
anti-luck virtue epistemology, or “ALVE” for short—which can accommodate the phe-
nomenon of epistemic dependence, in both its negative and positive aspect. In its most
general form, ALVE maintains that knowledge is safe cognitive success that is due to
the exercise of the agent’s relevant epistemic virtue, where a safe cognitive success is a
cognitive success that could not very easily have been a cognitive failure.
We saw above the plausibility of having a virtue-theoretic condition on knowledge.
By incorporating a safety condition as well in this way, ALVE can effectively eliminate
the knowledge-undermining epistemic luck that is in play in such scenarios as Gettier-
style cases. But with this epistemic luck eliminated by the safety condition, there is no
need to “beef up” the virtue-theoretic element of the proposal in order to deal with the
problem posed by knowledge-undermining epistemic luck in the first place. In par-
ticular, one does not need to demand that one’s cognitive success be because of one’s
cognitive ability, and thus one can allow that apt belief is not necessary for knowledge.
One can instead rest content with safe cognitive success which is due to the exercise of
one’s relevant epistemic virtue. In Sosa’s terminology, knowledge entails accuracy and
adroitness, but it does not entail aptness.23
In cashing out what it means for one’s safe cognitive success to be “due to” the exercise
of one’s relevant epistemic virtue, one can go down either the dispositional (Sosa-style)
23
For the key defences of anti-luck virtue epistemology, see Pritchard, Millar, and Haddock (2010:
chapter 3) and Pritchard (2012a). See also Pritchard (forthcoming).
or causal-explanatory (Greco-style) route. On the latter rendering, knowledge is safe

cognitive success that is to a significant extent creditable to one’s cognitive agency, where
this means that one’s cognitive agency (i.e., the exercise of relevant epistemic virtue) is a
significant factor, but not necessarily the primary or overarching factor, which causally
explains one’s safe cognitive success. On the former reading, knowledge is safe cognitive
success that is the product of the exercise of the relevant epistemic competence, but
where it is not demanded that the agent’s cognitive success, much less the agent’s safe
cognitive success, be because of the manifestation of this epistemic competence.
In order to keep the debate between DRVE and ALVE as clean as possible, let us opt
for the dispositional reading of ALVE. A good way of fleshing out what this account
demands is to see how it deals with Gettier-style cases. Recall that DRVE responds
to these cases by arguing that while there is accuracy and adroitness there is not accu-
racy because adroitness (i.e., aptness), and hence there is not knowledge. ALVE agrees.
In contrast to DRVE, however, ALVE explains the lack of knowledge in play here not in
terms of the failure of the belief to be apt, but rather in terms of the fact that the belief so
formed is unsafe. In this way, it avoids making aptness necessary for knowledge.
Note how the problems of both negative and positive epistemic dependence do not
arise by the lights of such a rendering of ALVE. Consider first negative epistemic
dependence. What is the epistemic difference between S and twin-S’s beliefs which
explains why S has the knowledge which twin-S lacks? Well, the difference is clearly
not down to their manifestation of a relevant epistemic competence, since as we’ve
noted above both subjects are identical on this score. According to ALVE, however,
there is more to having knowledge than merely forming a true belief via the manifesta-
tion of a relevant epistemic competence, since one’s true belief should also be safe.
Clearly, however, this is lacking in twin-S’s case, since given the nature of her regional
environment it could very easily have been the case that she forms a belief on the same
basis and yet believes falsely. ALVE thus delivers the right result. In particular, it can
account for the phenomenon of negative epistemic dependence because it can accom-
modate the idea that a display of cognitive agency that would normally suffice for
knowledge will not suffice in those conditions where features of the environment
ensure that the belief so formed is nonetheless unsafe.
Now we turn to positive epistemic dependency. Recall that ALVE does not treat apt-
ness as necessary for knowledge. As a result, it does not follow on this view that because
the two counterpart agents in the testimonial cases described earlier do not form apt
beliefs that they thereby lack knowledge. Both agents form accurate beliefs by exercis-
ing relevant epistemic competences, however, even if the accuracy is not because of the
manifestation of these competences. What differentiates the agent in the epistemically
favorable environment from the agent in the epistemically unfavorable environment is
thus not the degree of cognitive agency on display (i.e., the adroitness of her belief), but
rather the fact that unlike her counterpart her belief so formed is safe. This is why in
incorporating a safety condition ALVE can deal with such cases of positive epistemic
dependency. In particular, it can account for how the manifestation of a very limited
degree of cognitive agency can sometimes suffice for knowledge when the conditions
in which this cognitive agency is displayed are sufficiently epistemically friendly to
ensure that the belief so formed is safe.
We can diagnose where robust virtue epistemologies like DRVE go awry when it
comes to the phenomenon of epistemic dependence in terms of their aspiration to
exclusively analyze knowledge in terms of a virtue-theoretic condition, thereby avoid-
ing a separate anti-luck requirement. In doing so, they make it impossible to account
for how knowing can sometimes involve a lot more, and sometimes a lot less, than
cognitive success that is because of cognitive ability, where in both cases what is mak-
ing the difference is the epistemic favorability/unfavorability of the environment. In
contrast, by incorporating a safety condition into its theory of knowledge, ALVE
doesn’t face these problems. So not only does DRVE face formidable difficulties when
it comes to dealing with the phenomenon of epistemic dependence, but, worse, there is
also an alternative proposal available which doesn’t face these difficulties. We conclude
that the prospects for DRVE do not look bright.24
References
Axtell, G., 1997. “Recent Work in Virtue Epistemology.” American Philosophical Quarterly 34:
410–30.
Baehr, J., 2004. “Virtue Epistemology.” In B. Doden and J Fieser (eds) Internet Encyclopaedia of
Philosophy. <http://www.iep.utm.edu/virtueep/> accessed June 19, 2015.
Baehr, J., 2008. “Four Varieties of Character-Based Virtue Epistemology.” Southern Journal of
Philosophy 46: 469–502.
Carter, J. A., and D. H. Pritchard, 2010. “Epistemology of Testimony.” Oxford Bibliographies:
Philosophy, doi: 10.1093/OBO/9780195396577-0119.
Chalmers, D., 1996. The Conscious Mind: In Search of a Fundamental Theory. Oxford: Oxford
University Press.
Code, L., 1987. Epistemic Responsibility. Hanover, NH: University Press of New England.
Fantl, J., and M. McGrath, 2009. Knowledge in an Uncertain World, Oxford: Oxford University
Press.
Fricker, E., 1995. “Telling and Trusting: Reductionism and Anti-Reductionism in the
Epistemology of Testimony.” Mind 104: 393–411.
Greco, J., 2002. “Virtues in Epistemology.” In P. Moser (ed.) Oxford Handbook of Epistemology,
287–315. Oxford: Oxford University Press.
Greco, J., 2003. “Knowledge as Credit for True Belief.” In M. DePaul and L. Zagzebski (eds)
Intellectual Virtue: Perspectives from Ethics and Epistemology, 111–34. Oxford: Oxford
University Press.
Greco, J., 2007. “The Nature of Ability and the Purpose of Knowledge.” Philosophical Issues 17:
57–69.
24
Thanks to J. Adam Carter, Miguel Fernández, and to two anonymous referees who supplied com-
ments on a previous version of this essay. We are also very grateful to Ernie Sosa, for many long and
enlightening (for us at least!) conversations about epistemology over the years.
Greco, J., 2008. “What’s Wrong With Contextualism?” Philosophical Quarterly 58: 416–36.
Greco, J., 2009a. Achieving Knowledge. Cambridge: Cambridge University Press.
Greco, J., 2009b. “The Value Problem.” In A. Haddock, A. Millar, and D. H. Pritchard (eds)
Epistemic Value, 313–21. Oxford: Oxford University Press.
Greco, J., 2009c. “Knowledge and Success From Ability.” Philosophical Studies 142: 17–26.
Greco, J., 2012. “A (Different) Virtue Epistemology.” Philosophy and Phenomenological Research
85: 1–26.
Greco, J., and J. Turri, 2009. “Virtue Epistemology.” In E. Zalta (ed.) Stanford Encyclopaedia
of Philosophy. <http://plato.stanford.edu/entries/epistemology-virtue/> accessed 19 June
2015.
Hawthorne, J., 2004. Knowledge and Lotteries. Oxford: Oxford University Press.
Hetherington, S., 2013. “There Can Be Lucky Knowledge.” In M. Steup and J. Turri (eds)
Contemporary Debates in Epistemology, 2nd edn., 164–76. Oxford: Blackwell.
Hookway, C., 2003. “How to Be a Virtue Epistemologist.” In M. DePaul & L. Zagzebski (eds)
University Press.
Horgan, T., and M. Timmons, 1991. “New Wave Moral Realism Meets Moral Twin Earth.”
Journal of Philosophical Research 16: 447–65.
Horgan, T., and M. Timmons, 1992. “Troubles on Moral Twin Earth: Moral Queerness Revived.”
Synthese 92: 221–60.
Jackson, F., 1996. “Mental Causation.” Mind 105: 377–413.
Kallestrup, J., 2011. Semantic Externalism. London: Routledge.
Kallestrup, J., and D. H. Pritchard, 2011. “Virtue Epistemology and Epistemic Twin Earth.”
European Journal of Philosophy (Online First, doi: 10.1111/j.1468-0378.2011.00495.x).
Kallestrup, J., and D. H. Pritchard, 2012. “Robust Virtue Epistemology and Epistemic Anti-
Individualism.” Pacific Philosophical Quarterly 93: 84–103.
Kallestrup, J., and D. H. Pritchard, 2013a. “Robust Virtue Epistemology and Epistemic
Dependence.” In T. Henning and D. Schweikard (eds) Knowledge, Virtue and Action: Putting
Epistemic Virtues to Work, ch. 11. London: Routledge.
Kallestrup, J., and D. H. Pritchard, 2013b. “The Power, and Limitations, of Virtue Epistemology.”
In J. Greco and R. Groff (eds) Powers and Capacities in Philosophy: The New Aristotelianism,
ch. 13. London: Routledge.
Kelp, C., 2012. “Knowledge: The Safe-Apt View.” Australasian Journal of Philosophy (Online
First, doi: 10.1080/00048402.2012.673726).
Kvanvig, J., 1992. The Intellectual Virtues and the Life of the Mind. Savage, MD: Rowman &
Littlefield.
Kvanvig, J., 2010. “Virtue Epistemology.” In S. Bernecker and D. H. Pritchard (eds) Routledge
Companion to Epistemology, 199–207. London: Routledge.
Lackey, J., 2007. “Why We Don’t Deserve Credit for Everything We Know.” Synthese 158:
345–61.
Lackey, J., 2010. “Testimonial Knowledge.” In S. Bernecker and D. H. Pritchard (eds) Routledge
Companion to Epistemology, 316–25. New York: Routledge.
Luper, S., 1984. “The Epistemic Predicament.” Australasian Journal of Philosophy 62: 26–50.
Luper, S., 2003. “Indiscernability Skepticism.” In S. Luper (ed.) The Skeptics: Contemporary
Essays, 183–202. Aldershot: Ashgate.
Montmarquet, J., 1993. Epistemic Virtue and Doxastic Responsibility. Lanham, MD: Rowman &
Littlefield.
Pritchard, D. H., 2002. “Resurrecting the Moorean Response to the Sceptic.” International
Journal of Philosophical Studies 10: 283–307.
Pritchard, D. H., 2005. Epistemic Luck. Oxford: Oxford University Press.
Pritchard, D. H., 2007. “Anti-Luck Epistemology.” Synthese 158: 277–97.
Pritchard, D. H., 2008a. “Greco on Knowledge: Virtues, Contexts, Achievements.” Philosophical
Quarterly 58: 437–47.
Pritchard, D. H., 2008b. “Sensitivity, Safety, and Anti-Luck Epistemology.” In J. Greco (ed.) The
Oxford Handbook of Scepticism, 437–55. Oxford: Oxford University Press.
Pritchard, D. H., 2009. “Apt Performance and Epistemic Value.” Philosophical Studies 143:
407–16.
Pritchard, D. H., 2012a. “Anti-Luck Virtue Epistemology.” Journal of Philosophy 109: 247–79.
Pritchard, D. H., 2012b. “In Defence of Modest Anti-Luck Epistemology.” In T. Black and
K. Becker (eds) Modal Epistemology: New Essays on Sensitivity and its Descendants, 173–92.
Cambridge: Cambridge University Press.
Pritchard, D. H., 2013. “There Cannot be Lucky Knowledge.” In M. Steup and J. Turri (eds)
Contemporary Debates in Epistemology, 2nd edn., 152–64. Oxford: Blackwell.
Pritchard, D. H., 2015. “Anti-Luck Epistemology and the Gettier Problem.” Philosophical
Studies 172, 1: 93–111.
Pritchard, D. H., Forthcoming. “Knowledge, Luck, and Virtue: Resolving the Gettier Problem.”
In C. Almeida, P. Klein, and R. Borges (eds) The Gettier Problem. Oxford: Oxford University
Press.
Pritchard, D. H., A. Millar, and A. Haddock, 2010. The Nature and Value of Knowledge: Three
Investigations. Oxford: Oxford University Press.
Pritchard, D. H., and J. Turri, 2011. “Knowledge, the Value of.” In E. Zalta (ed.) Stanford
Encyclopaedia of Philosophy. <http://plato.stanford.edu/entries/knowledge-value/> accessed
June 19, 2015.
Putnam, H., 1975. “The Meaning of ‘Meaning’.” Minnesota Studies in the Philosophy of Science
7: 131–93.
Roberts, R., and W. J. Wood, 2007. Intellectual Virtues: An Essay in Regulative Epistemology.
Oxford: Oxford University Press.
Sainsbury, R. M., 1997. “Easy Possibilities.” Philosophy and Phenomenological Research 57:
907–19.
Sosa, E., 1988. “Beyond Skepticism, to the Best of Our Knowledge.” Mind 97: 153–89.
Sosa, E., 1991. Knowledge in Perspective: Selected Essays in Epistemology. Cambridge: Cambridge
University Press.
Sosa, E., 1999. “How to Defeat Opposition to Moore.” Philosophical Perspectives 13: 141–54.
Sosa, E., 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge. Oxford: Clarendon
Press.
Sosa, E., 2009. Reflective Knowledge: Apt Belief and Reflective Knowledge. Oxford: Clarendon
Press.
Sosa, E., 2010. Knowing Full Well. Princeton, NJ: Princeton University Press.
Sosa, E., 2011. “How Competence Matters in Epistemology.” Philosophical Perspectives
24: 465–75.
Sosa, E., 2013. “Animal Versus Reflective Orders of Epistemic Competence.” In T. Henning and
D. Schweikard (eds) Knowledge, Virtue and Action: Putting Epistemic Virtues to Work, ch. 7.
London: Routledge.
Sosa, E., Forthcoming. Epistemic Agency: Responsible Character and Reliable Competence: The
Unity of Virtue Epistemology. Oxford: Oxford University Press.
Stanley, J., 2005. Knowledge and Practical Interests. Oxford: Oxford University Press.
Timmons, M., 1999. Morality Without Foundations: A Defence of Ethical Contextualism.
Turri, J., 2011. “Manifest Failure: The Gettier Problem Solved.” Philosophers Imprint 11: 1–11.
Turri, J., and E. Sosa, 2010. ‘Virtue Epistemology’, Oxford Bibliographies: Philosophy, doi:
10.1093/OBO/9780195396577-0123.
Williamson, T., 2000. Knowledge and its Limits. Oxford: Oxford University Press.
Wittgenstein, L., 1969. On Certainty, ed. G. E. M. Anscombe and G. H. von Wright, tr. D. Paul
and G. E. M. Anscombe. Oxford: Blackwell.
Zagzebski, L., 1999. “What is Knowledge?” In J. Greco and E. Sosa (eds) The Blackwell Guide to
Epistemology, 92–116. Oxford: Blackwell.
3
Knowledge, Virtue, and Safety
John Greco
In a series of papers beginning in the late 1990s, Ernest Sosa argued in favor of a
“safety” condition on knowledge: viz., S knows that p only if not easily would S falsely
believe that p (Sosa 1999a, 1999b, 2000, 2004). Moreover, Sosa argued, a safety condi-
tion on knowledge is properly grounded in a virtue condition: viz., S knows that p only
if S’s belief that p is safe, and only if such safety has its residence in S’s cognitive
competence.1
More recently, however, Sosa at least seems to retract his position that knowledge
requires safety, arguing that safety and virtue can come apart, and that knowledge
requires only the latter (Sosa 2007). This possibility is demonstrated by the phenome-
non of normal dreaming. It is plausible, Sosa says, that our beliefs about the world are
unsafe, due to the close possibility where we are fooled by a dream. That possibility is
indeed close, Sosa reasons, because we are fooled by dreams often, and so might easily
be fooled at some given moment, even when in fact awake. This possibility under-
mines safety, but not aptness, Sosa argues, since, so long as we are actually awake, our
true belief might still manifest competence in normal conditions:
The proximate possibility that one is now asleep and dreaming might thus render fragile both
one’s competence and also, jointly or alternatively, the conditions appropriate for its exercise.
That is how the possibility that one is asleep and dreaming might endanger our ordinary per-
ceptual beliefs. But this is just one more case where safety is compromised while aptness
remains intact. (Sosa 2007: 30)
If our true belief does manifest competence, then this is enough for knowledge, despite
the close possibility of dreaming, and so despite lack of safety: “The knower’s belief can
thus remain apt even if unsafe through the proximity of the dream possibility”
(Sosa 2007: 31).
1
In these papers Sosa argued that knowledge requires safety over “sensitivity,” as understood by such
writers as Nozick, Dretske, or DeRose, thereby preserving plausible closure principles, and yielding a more
satisfying response to skepticism.
52 John Greco
Others have argued for yet different relations between a safety condition and a vir-
tue condition on knowledge. Responding to Sosa’s earlier papers, for example, Duncan
Pritchard argued that knowledge requires a safety condition but not a virtue condition
(Pritchard 2005). More recently, however, Pritchard argues that knowledge requires
both a virtue condition and a safety condition, and that the two are independent of
each other (Pritchard 2010). Notice that Sosa (early and late) and Pritchard (early and
late) agree on at least this much: safety and virtue can come apart. But whereas Sosa (in
his latest work) draws the conclusion that knowledge requires virtue but not safety,
Pritchard (in his latest work) concludes that knowledge requires both.
Finally, at least one author has maintained a consistent position throughout, thereby
manifesting either the vice of intellectual stubbornness or the virtue of intellectual
steadfastness, depending on one’s point of view. Thus I have argued that knowledge
requires both virtue and safety, and that the virtue condition (properly understood)
entails the safety condition (properly understood). Moreover, I have argued, a signifi-
cant advantage of a virtue-theoretic account of knowledge is that it explains why, and
in what sense, knowledge requires safely true belief. Put differently, a virtue-theoretic
account motivates a safety condition on knowledge. This is an advantage over an
account that requires safety alone, without an explanation as to why knowledge
requires safety rather than some other modally strong relation to the world. For simi-
lar reasons, it is an advantage over an account that requires both virtue and safety for
knowledge, but as conditions that are independent from each other.
In Part One of the paper I explain the nature of safety in more detail, and I draw
some important distinctions regarding different kinds of safety. I also show that Sosa’s
latest position regarding the relation between virtue and safety is what I take to be the
correct one. In other words, Sosa now holds that knowledge requires both virtue (of a
sort) and safety (of a sort), and that the safety condition is entailed by the virtue condi-
tion. I also say a bit more about why I think this constitutes a theoretical advantage
over alternative positions. In Part Two I argue for a specific version of the position that
Sosa and I both endorse. In particular, I argue for a specific account of the virtue condi-
tion on knowledge, and hence the safety condition that falls out of it. In Part Three I
argue that the position that results explains why there is no knowledge in Sosa’s
Kaleidoscope Jokester case, and without having to invoke a distinction between ani-
mal and reflective knowledge.
Part One: Varieties of Safety

In the early papers mentioned above, Sosa introduces his notion of safety by compar-
ing it with the Dretske-Nozick notion of sensitivity. A belief is sensitive just in case: if p
were false then S would not believe p. Where ‘⇒’ represents the subjunctive
conditional,
∼ p ⇒ ∼ B( p).
knowledge, virtue, and safety 53
A belief is safe, Sosa tells us, just in case: S would believe p only if p were true. More
formally,
B ( p ) ⇒ p.2
Sosa resists an analysis of safety in terms of possible worlds. Nevertheless, we can use
that language as a heuristic. Hence,
SENSITIVITY. A belief is sensitive just in case: In the closest possible world where p
is false, S does not believe p.
SAFETY. A belief is safe just in case: In close possible worlds where S believes p, p is
true.
With the possible worlds heuristic in place, we can note an ambiguity in Sosa’s notion
of the safety condition as expressed above. Namely, safety is ambiguous between:
Strong Safety. In all close worlds where S believes p, p is true. Alternatively: In close
worlds, never does S believe p and p is false.
Weak Safety. In most (almost all) close worlds where S believes p, p is true.
Alternatively: In close worlds, almost never does S believe p and p is false.
Given the way that Sosa treats the lottery problem and some other relevant cases, it is
clear that he intends to place a weak safety condition on knowledge in the papers cited
above. Thus Sosa holds that one can know that she will lose the lottery, and in general
can have knowledge through inductive reasoning, even if in some close worlds S
believes p and p is false (Sosa 2007: 25).3
Early and late, Sosa employs a second distinction between kinds of safety: that
between “absolute” or “outright” safety and “restricted” safety. S’s belief is absolutely
safe just in case it satisfies SAFETY above. S’s belief is safe in a restricted way just in case
S would not easily go wrong, relative to some restricted set of circumstances. One
important kind of restricted safety is basis-relative safety. A belief is basis-relative safe
just in case S would not easily go wrong by believing on the same basis that she does.
For example, suppose that S believes that there is a red ball, basing her belief on a visual
experience having a particular phenomenal character. S’s belief is safe relative to that
basis, just in case S would not easily believe that there is a red ball on that same basis,
and that belief turn out false. Alternatively:
Restricted Safety. A belief is safe restricted to circumstances C just in case: In close
worlds where S believes p in C, p is true.
Basis-relative Safety. A belief is basis-relative safe just in case: In close worlds where
S believes p on basis B, p is true.
2
As Sosa explains, subjunctive conditionals do not contrapose.
3
See also (Sosa, 2007: 25): “A belief that p is safe provided it would have been held only if (most likely) p.”
54 John Greco
With these notions in place, we can easily define other kinds of relative or restricted
safety. Important for presents purposes:
Ability-relative Safety. A belief is ability-relative safe just in case: In close worlds
where S believes p from ability A, p is true.
(Ability + Proper Conditions)-relative Safety. A belief is (Ability + Proper Conditions)-
relative safe just in case: In close worlds where S believes p from ability A, and in
conditions proper for the exercise of A, p is true.
Returning to Sosa’s early and late positions, we may now see that early on Sosa endorsed
a weak, basis-relative safety condition on knowledge:
What is required for a belief to be safe is not just that it would be held only if true, but rather
that it be based on a reliable indication. What counts as such an indication? Indications are
deliverances, as when you ostensibly perceive, or remember, or deduce something or
other . . . Such a deliverance is an indication if and only if it would occur only if the delivered
proposition were true. (Sosa 1999b: 149)
We have seen that, in his most recent work, Sosa now rejects any basis-relative safety
condition on knowledge. However, he does endorse a different kind of restricted safety
condition. Specifically, he endorses (Ability + Proper Conditions)-relative Safety. Thus
he writes:
In conclusion, neither aptness nor safety entails the other. The connection that perhaps
remains is only this. Aptness requires the manifestation of a competence, and a competence is
a disposition, one with a basis resident in the competent agent, one that would in appropriately
normal conditions ensure (or make highly likely) the success of any relevant performance
issued by it. Compatibly with such restricted safety, the competence manifest might then be
fragile, as might also be the appropriate normalcy of the conditions in which it is manifest.
(Sosa 2007: 29)
The way to read this passage, I suggest, is as follows:

In conclusion, neither aptness nor basis-relative safety entails the other. The connection that
perhaps remains is only this. Aptness requires the manifestation of a competence . . . one that
would in appropriately normal conditions ensure (or make highly likely) the success of any
relevant performance issued by it; in other words, aptness insures ability-relative safety, at least
relative to normal conditions. Compatibly with that kind of restricted safety, the competence
manifest might then be fragile, as might also be the appropriate normalcy of the conditions in
which it is manifest; in other words, aptness does not insure outright safety, or even basis-relative
safety, but rather (Ability + Proper Conditions)-relative Safety.
Putting aside Sosa’s intentions in the above passage, it is fairly easy to see that a Sosa-
type virtue condition on knowledge does entail a kind of ability-relative safety condi-
tion. Thus an intellectual virtue, according to Sosa, is a reliable disposition to believe
truths in some relevant range, when in appropriately normal conditions or circum-
stances. Moreover, Sosa’s notion of reliability is modally strong: a reliable disposition
would produce true belief, in appropriate normal conditions, with a high degree of
regularity. Accordingly:
S’s belief p is produced by intellectual virtue V only if, for some disposition D, in most (almost
all) close worlds where S believes p from D in appropriately normal conditions, p is true.
But, of course, that just is a weak, ability-relative safety condition, restricted to

appropriately normal conditions. Put differently, it is the weak, (Ability + Proper
Conditions)-relative Safety condition.
To my mind, the fact that Sosa’s virtue condition entails a safety condition is a signif-
icant theoretical advantage of that account. After all, the debate among safety and sen-
sitivity theorists is a debate over what modally strong relation between mind and world
is required by knowledge. Likewise, the debate over different kinds of safety condition
is a debate about that issue. What all parties to these debates agree on, though, is that
knowledge does require some such modal relation or another. What the foregoing dis-
cussion shows is that a virtue-theoretic account motivates just such a condition on
knowledge. In that sense, a virtue-theoretic approach explains why knowledge requires
a modally strong relation between mind and world, and explains in some detail just
how that relation should be characterized.
Another way to make the same point is to say that a virtue-theoretic account
explains why knowledge is inconsistent with certain kinds of luck, and explains in
some detail just what kind of luck knowledge cannot tolerate. Specifically, knowledge
excludes luck in the way that success from ability in general excludes luck. Put differ-
ently, in cases of knowledge it is “no accident” that S believes the truth. A virtue-theoretic
approach gives a fairly precise explication of that intuitive idea: In cases of knowledge,
S believes from ability, and in most (almost all) close worlds where S believes from
ability (and in relevant conditions), S believes the truth.4
Part Two: How to Think about Virtue and Safety

An intellectual virtue, in the sense required for knowledge, is a disposition to believe
truths in a relevant range, when in relevant circumstances, with a sufficiently high
degree of reliability. Accordingly, knowledge-relevant abilities have the following
structure:
KRA. S has a knowledge-relevant ability A(R/C/D) = S has a disposition to believe
truths in range R when in circumstances C, with degree of reliability D.
4
In fact, the position that Sosa and I defend entails an even stronger modal relation between mind and
world. That is because we characterize knowledge as an achievement, and achievements in general require
not only that one’s successful performance is produced by ability, but also that one’s performance is suc-
cessful because produced by ability. In cases of knowledge, then, one’s belief is true because one’s belief is
produced by ability. For extended discussion, see my (Greco 2003, 2010) and (Sosa 2007).
56 John Greco
How do we fill in the relevant parameters in KRA? For example, what counts as the
relevant sort of circumstances? As we have seen, Sosa suggests that what is relevant is
what is “normal” or “appropriately normal”; that is, appropriately normal for the exer-
cise of the ability in question:
Failed attempts in abnormal circumstances do not show lack of ability. Despite such failures we
might still depend on you in normal circumstances. What is required is only that your attempts
tend to succeed when circumstances are normal. (Sosa 2007: 83–4)
I want to suggest, however, that intellectual abilities should not be defined in terms of
normal, or even “appropriately normal”, conditions. Not if we are trying to define
“intellectual ability” in a way that ties ability to knowledge. On the contrary, some-
times what knowledge requires is ability in abnormal circumstances.
For example, suppose that we need to distinguish lead from gold, and we need to do
this while standing in a cave, where lighting conditions are abnormally low for human
vision. If I form the true belief that x is gold, based on vision that is unreliable in pres-
ent lighting conditions, then I hardly come to know that x is gold, even if my vision is
highly reliable in normal lighting conditions. (Sosa’s view, stated in terms of appropri-
ately normal conditions, rules correctly here.) On the other hand, suppose that I can
reliably distinguish lead from gold in the abnormal light we have. Plausibly, that sort of
ability does ground knowledge in present circumstances, and whether or not I can dis-
tinguish lead from gold in normal conditions. Suppose, for example, that my eyes are
damaged, and in such a way that “normal” light is too much light for me.
The alternative I want to propose is that we look to relevant informational needs to
set the parameters in KRA. That is, those parameters should be set by the needs for
information attaching to some relevant practical task. This will often set parameters at
normal values, but sometimes not.
Following this line of reasoning, we should fill in the other parameters in KRA the
same way. Suppose, for example, that what is normal for human beings is that they can
visually distinguish lead from gold, but not gold from iron pyrite. If our practical task
requires the latter, then we want someone who is reliably successful in the abnormal
range, not the normal one. Put differently, we want someone with an ability that will
serve relevant needs for information. Relevant needs for information will often be
equivalent to “normal” needs for information, but sometimes not.
The present proposal is that the concept of knowledge tracks relevant needs for
information, and that the concept of an intellectual virtue should do the same. This
proposal is motivated by remarks that Sosa makes himself:
We care about justification because it tends to indicate a state of the subject that is impor-
tant and of interest to his community, a state of great interest and importance to an infor-
mation-sharing social species. What sort of state? Presumably, the state of being a
dependable source of information over a certain field in certain circumstances. In order
for this information to be obtainable and to be of later use, however, the sort of field F and
the sort of circumstances C must be projectible, and must have some minimal objective
likelihood of being repeated in the careers of normal members of the epistemic community.
(Sosa 1991: 281–2)
The present idea is that relevant Fs and Cs will not always correspond to what is “nor-
mal,” since sometimes our informational needs will not be normal. I take it that this is
an idea that Sosa can accept, and even makes room for when he talks in terms of
“appropriate,” “proper,” and “relevant” conditions, as opposed to merely “normal”
conditions.5
Accordingly, the proposal is this:
S knows that p only if S’s believing that p is produced by an intellectual ability of the relevant
sort, i.e. of a sort defined by parameters that would serve relevant informational needs.6
Adding the other conditions that Sosa requires for knowledge:

AK. S knows that p if and only if:
1. S’s believing that p is produced by an intellectual ability of the relevant sort, i.e.
of a sort defined by parameters that would serve relevant informational needs;
2. S is in conditions relevant for the exercise of that ability; and
3. S has a true belief because S’s belief is produced by an ability of the relevant sort,
while in relevant conditions.7
Part Three: Dreamers and Jokesters

As we saw above, Sosa’s most recent work emphasizes that basis-relative safety and
virtue can come apart. This possibility is demonstrated by the phenomenon of normal
dreaming, where S believes some p on the usual basis (for example the usual appear-
ances), but where p is false because S is merely dreaming. This possibility undermines
basis-relative safety, Sosa argues, because S could easily be dreaming and thereby have
a false belief on the usual basis. The possibility of normal dreaming does not under-
mine aptness, however, so long as we are actually awake. This is because, while awake,
our true belief might still manifest competence in normal conditions. Hence, Sosa
concludes, “The knower’s belief can thus remain apt even if unsafe through the prox-
imity of the dream possibility” (Sosa 2007: 31).
5
For example, in (Sosa 2007), where Sosa alternates among these formulations.
6
Moving up a level of assent:
An assertion “S knows that p” is true relative to a conversational context only if S’s believing that p is
produced by an exercise of intellectual ability of the relevant sort, i.e. of a sort defined by parameters that
would regularly serve relevant informational needs, as determined by the conversational context.
Notice that these theses (meta-linguistic and object-level) are consistent with various options regarding
what counts as “relevant sort” and “relevant informational need.” For example, the meta-linguistic thesis
does not entail attributor contextualism. On the contrary, that depends on whether what counts as “rele-
vant” varies across conversational contexts. The present position, then, is consistent with attributor contex-
tualism but does not entail it. For an extended defense of the present approach, see Greco (2012).
7
I defend AK at length in Greco (2012).
58 John Greco
This approach to perceptual knowledge is threatened, however, by Sosa’s case of the

Kaleidoscope Jokester:
You see a surface that looks red in ostensibly normal conditions. But it is a kaleidoscope surface
controlled by a jokester who also controls the ambient light, and might as easily have presented
you with a red-light+white-surface combination as with the actual white-light+red-surface
combination. Do you then know the surface you see to be red when he presents you with that
good combination, despite the fact that, even more easily, he might have presented you with the
bad combination? (Sosa 2007: 31)
Sosa’s account of perceptual knowledge is threatened, he tells us, because the subject in
the Jokester case has apt belief. But then, implausibly, he has knowledge on Sosa’s account:
Arguably, your belief that the surface is red is an apt belief, in which case it amounts to knowl-
edge, or so it does according to our account. For you then exercise your faculty of color vision
in normal conditions of lighting, distance, size of surface, etc., in conditions generally appro-
priate for the exercise of color vision. Yet it is not easy to insist that you therefore know that
surface to be red. (Sosa 2007: 31)
Sosa’s solution is to employ his now familiar distinction between animal and reflective
knowledge. Whereas animal knowledge requires apt belief, reflective knowledge
requires apt belief aptly noted. That is, reflective knowledge requires an apt perspective
on the fact that one’s first-level belief is aptly formed. Sosa’s diagnosis is then as follows:
whereas in typical cases of perception one enjoys both animal and reflective knowl-
edge, the believer in the Jokester case enjoys only animal knowledge. That is, the
Jokester’s victim does have apt belief, but he fails to have an apt perspective on his apt
belief. That is because, too easily, the Jokester could have switched from a good combi-
nation to a bad combination, and were he to have done so, this would have gone unno-
ticed by our believer.
According to Sosa, then, there is a disanalogy between the Kaleidoscope Jokester
Case and typical cases of perceptual knowledge:
In the typical case, where no jokester is lurking, one competently believes in normal conditions,
and one’s belief that conditions are normal is itself aptly formed; that is, one’s belief that condi-
tions are normal itself derives from a competence exercised in normal conditions, and one that
would not easily go wrong when exercised in those conditions.
In the Kaleidoscope Jokester Case, where a jokester is lurking, one competently believes in nor-
mal conditions, but one’s belief that conditions are normal is not itself aptly formed; that is, one’s
belief that conditions are normal does not itself derive from a competence exercised in normal
conditions, and one that would not easily go wrong when exercised in those conditions.8
8
According to Sosa, either the presence of the jokester makes conditions for the exercise of one’s
meta-competence inappropriate, or conditions for the exercise of one’s meta-competence are appropriate,
but too easily might one have gone wrong in believing that conditions for one’s perceptual competence are
appropriate. Either way, though, one fails to aptly believe that one’s perceptual belief is aptly formed. That
is, either way one does not reliably form one’s meta-belief in conditions appropriate for its exercise, and in
a way that one could not easily have gone wrong by so believing. See Greco (2012: 36).
Sosa’s treatment of the Jokester case is ingenious, but to my mind it is not optimal.
First, doesn’t the attribution of knowledge in the Jokester case concede too much? Put
differently, isn’t it too generous? Of course, Sosa takes with one hand what he has given
with the other, because he also insists that S does not have reflective knowledge in the
Jokester case. But the present point is that, according to Sosa, S does have animal
knowledge in the Jokester case, and that is what seems too generous. This point is
related to a second: that on Sosa’s account, animal knowledge is supposed to be valua-
ble. That is, Sosa’s account of the value of knowledge in terms of apt success is an
account of the value of animal knowledge. It is animal knowledge that counts as a kind
of apt success, and thus shares in the value of apt success generally. But how valuable
can animal knowledge be, and how valuable can apt success be more generally, if S has
these in the Jokester case?
There is an alternative diagnosis of the Jokester case available, however. Specifically,
why think that conditions in that case are “relevant” or “proper” for the exercise of
ability? After all, there is a jokester present who might easily fool us! Lighting conditions
are normal—that’s true. But why should that be enough to say that conditions are in
general relevant or proper? If they are not, then there is a straightforward diagnosis for
why S lacks knowledge in the case: namely, S does not satisfy condition 2 of AK (and,
for that reason, does not satisfy condition 3, either).9
Sosa does not want to ruin his account of perceptual knowledge—his account of
why we can often have perceptual knowledge in normal conditions, despite the prox-
imity of the dream possibility. That is why he needed a relevant difference between the
“good” cases, where there is perceptual knowledge in normal conditions, and the
Jokester case, where there is not. But our alternative diagnosis gives us a relevant differ-
ence: In the good cases, but not in the Jokester case, S believes from a relevant ability
and in relevant conditions. In other words, in the good cases S does satisfy conditions 2
and 3 of AK, as well as condition 1. And this is so despite the proximity of dreaming in
one’s sleep.
The present diagnosis depends on a plausible assumption: that relevant informa-
tional needs will determine the “circumstances” parameter in KRA and AK so as to
9
One might argue that conditions in the Jokester case are “relevant” or “proper” by stressing the analogy
with non-epistemic cases. For example, we would not judge that a baker is not working in conditions
proper for the exercise of his baking abilities, because there is a jokester lurking who might easily (but does
not) switch salt for sugar. But here again we need to consider the practical interests in play. For good prac-
tical reasons, baking abilities and their relevant conditions are not to be understood in terms of how well
one does with jokesters out to ruin one’s cake. But cognitive abilities and their relevant conditions are to be
understood in terms of how well one does with ambient dangers of deception. Accordingly, we may say
that S has a perceptual ability to discriminate red surfaces from white in normal conditions (her perceptual
dispositions are reliable in those conditions), but lacks the ability to discriminate red surfaces from white
in the Kaleidoscope Jokester case (because her perceptual dispositions are not reliable in those conditions).
NB: The present position is not that perceptual abilities “come and go” as S moves from one set of environ-
mental conditions to another. Rather, S has one internal disposition, which is reliable relative to conditions
C (and hence is an ability relative to conditions C) and which is unreliable relative to conditions C' (and
hence is not an ability relative to conditions C'). Put differently, S retains the ability to discriminate red
surfaces in conditions C, whether or not she is in conditions C.
60 John Greco
specify that S is awake. That is, we are assuming, with Sosa, that knowledge requires
reliability not just in any circumstances whatsoever, but in circumstances where,
among other relevant considerations, S is awake. Sosa made that come out true by
specifying that knowledge-producing circumstances must be normal for the exercise
of S’s abilities. And that does do the trick. But we can also do the trick by specifying that
knowledge-producing circumstances answer to relevant informational needs.
When we opt for this understanding of intellectual ability, and for the conditions on
knowledge that this entails, we get the right result for all of our cases, including the
Jokester case. Moreover, we do not have to concede that the Jokester’s victim has even
animal knowledge. In fact, we do not need a distinction between animal knowledge
and reflective knowledge at all, at least not for understanding these cases. All this is
consistent with invoking the animal/reflective knowledge distinction elsewhere, as
Sosa does in his treatment of the Pyrrhonian problematic (Sosa 1997, 2011). It is also
consistent with arguing for the added value of reflective knowledge over animal
knowledge, as Sosa does in various places (Sosa 2007, 2010, 2011). But the present
proposal, I would argue, better explains the value of animal knowledge alone, and
does so by means of an understanding of intellectual ability that is independently
motivated.10
References
Greco, J., 2003. “Knowledge as Credit for True Belief.” In M. Depaul and L. Zagzebski (eds)
Intellectual Virtue: Perspectives from Ethics and Epistemology. Oxford: Oxford University
Press.
Greco, J., 2010. Achieving Knowledge: A Virtue-Theoretic Account of Epistemic Normativity.
Greco, J., 2012. “A (Different) Virtue Epistemology.” Philosophy and Phenomenological Research,
85: 1–26.
Pritchard, D., 2005. Epistemic Luck. Oxford: Oxford University Press.
Pritchard, D., 2010. “Knowledge and Understanding.” In A. Haddock, A. Millar, and D.
Pritchard (eds) The Nature and Value of Knowledge: Three Investigations. Oxford: Oxford
University Press.
Sosa, E., 1991. Knowledge in Perspective: Selected Essays in Epistemology. Cambridge: Cambridge
University Press.
Sosa, E., 1997. “How to Resolve the Pyrrhonian Problematic: A Lesson from Descartes.”
Philosophical Studies, 85: 229–49.
Sosa, E., 1999a. “How Must Knowledge Be Modally Related to What Is Known?” Philosophical
Topics, 26: 373–84.
Sosa, E., 1999b. “How to Defeat Opposition to Moore.” Noûs, 33: 141–53.
Sosa, E., 2000. “Skepticism and Contextualism.” Philosophical Issues, 10: 1–18.
10
Thanks to Miguel Ángel Fernández Vargas, Jesper Kallestrup, Christoph Kelp, Joshue Orozco, Duncan
Pritchard, and Ernie Sosa for helpful discussions. Thanks also to two anonymous referees for helpful
comments.
Sosa, E., 2004. “Response”. In J. Greco (ed.) Ernest Sosa and His Critics. Oxford: Blackwell.
Sosa, E., 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, i. Oxford: Oxford
University Press.
Sosa, E., 2011. Reflective Knowledge: Apt Belief and Reflective Knowledge, ii. Oxford: Oxford
University Press.
4
Abilities, Competences, and
Fallibility
Alan Millar
1. Introduction
Many of us have an ability to recognize lemons as lemons visually, that is, from the way
they look. This is an example of what I call a perceptual-recognitional ability. To pos-
sess this ability is to have command of a way of telling, thus coming to know, of lemons
that they are lemons from the way they look. So if seeing lemons on the supermarket
shelf I recognize them to be lemons, I thereby come to know that they are in virtue of
having exercised an ability to tell this sort of thing—that something is a lemon—in this
sort of way—from the visual appearance of the thing.
My aim in this discussion is to show that a conception of perceptual-recognitional
abilities that I have been developing is in keeping with a plausible wider conception of
abilities in general and of our fallibility with respect to our abilities.1 Henceforth when
I speak simply of recognitional abilities I shall be referring to perceptual-recognitional
abilities.
As I conceive them, recognitional abilities are exercised only in acts of recognition,
that is, acts in which a perceived object is recognized, and thus known, to be of some
kind, or to have some property, or to be some specified individual. If I were to judge
falsely of something I see that it is a lemon from the way it looks I would have done
something that is in important respects like what I would have done had I recognized
the thing to be a lemon. Nonetheless, in such a case I would not have exercised the
ability to recognize lemons as lemons from the way they look.
Another feature of recognitional abilities is that they are abilities with respect to
favorable environments.2 I would not count as having the ability to recognize lemons in
the environment I inhabit to be lemons, from the way they look, unless that environ-
ment was favorable to the possession of that ability. A crucial requirement is that lemons
1
The conception of perceptual-recognitional abilities is worked out in Millar 2007, 2008a, 2008b, 2009b,
2011a, 2011b, 2012, 2014, and in my contribution (Part II) to Pritchard, Millar, and Haddock 2010.
2
I take this to be the chief lesson to be drawn from examples of the fake-barn type (Goldman 1976).
abilities, competences, and fallibility 63
have a (nearly enough) distinctive visual appearance in that possession of that appear-
ance by things is a very highly reliable indicator that they are lemons. It follows that
I would not be able to tell of lemons that they are lemons from the way they look
if I were in an environment in which ringers for lemons were legion and accessible. Of
course, even in the environments that we inhabit—ones in which many have the ability
to recognize lemons to be lemons visually—there could be, and perhaps are, some ring-
ers for lemons. But let it not be inferred from this that when judgments are made by
way of exercising the relevant recognitional ability they are luckily true in any way that
is at odds with the case being one of knowing in that way that the objects in question are
lemons. In a sophisticated environment, of course, a robust visual lemon-recognizing
ability incorporates sensitivity to contexts (soap shops, bathrooms?) in which ringers
might appear. This sensitivity helps us to avoid false judgments and so to maintain high
levels of reliability at making true judgments. I shall take that to be understood.
That perceptual-recognitional abilities have these and the other features to which
I shall draw attention strikes me as being in keeping with common sense and a fruitful
starting point for a philosophical account of perceptual knowledge. Not everyone
agrees so I shall defend the picture by, among other things, presenting it in the light of a
treatment of abilities in general. En route I shall engage with some recent remarks of
John McDowell about fallible capacities. Then I shall explore some contrasts between
the perspective outlined here and Ernest Sosa’s virtue epistemology.
2. The Ability to Ride a Bicycle and Kindred Abilities

It will be helpful to begin by considering common abilities like the ability to ride a
bicycle, or to read English or to walk, or to swim, or to play football. Plausibly, the fol-
lowing are true of abilities such as those.
(1) People exercise the ability to do something only if they do, or are doing, that
thing. (The point of adding ‘or are doing’ will become clear towards the end of
section 4.) Call this the success thesis. It would be odd to suppose that someone
is exercising the ability to ride a bicycle and yet is not riding a bicycle, or is
exercising the ability to calculate the sum of a column of figures and yet is not
calculating the sum of that column of figures.3
3
Someone who can jump over a 50 centimetre high bar can jump over a 30 centimetre high bar. Adam
Carter and Michael Wheeler have independently raised the question whether one would manifest the
lesser ability in exercising the more onerous one. If so, then it looks as if an ability can be manifested by
doing something other than the thing that the ability is an ability to do. I think it best to distinguish
between manifesting that one has an ability and exercising that ability. Possession of the lesser ability could
be manifested by an exercise of the more onerous ability in the sense that this exercise could indicate that
the agent has the lesser ability, but it does not follow that the lesser ability would in that case be exercised.
Another worry might arise in connection with tasks that can be interrupted, like performing a sonata and
making an omelette. One might think that on my view the ability to carry out such tasks is not exercised in
cases in which the agent was in the course of carrying them out, but did not complete the task. I suggest in
64 Alan Millar
(2) There are many factors that can prevent an agent who has an ability from exer-
cising it when acting in order to do what the ability is an ability to do. An
ability is exercised only if nothing impedes its exercise. With regard to any
ability of which we have an understanding we have some grasp of the sorts of
factor that could impede its exercise. Among these is lack of means. We cannot
read an English text unless such a text is available to be read. We cannot swim
unless water is available in which we can swim. Another impeding factor is
lack of fitness on the part of the agent. We cannot ski for long if very drunk.
A further impeding factor is unfavorable environmental circumstances. We
cannot ride a bicycle in the face of a hurricane or visually check the contents of
a room in total darkness.4
(3) Notwithstanding the points made in (2), a statement to the effect that a person
has an ability to do something is not shorthand for a longer statement to the
effect that the person has the ability to do that thing provided that . . . , where
the gap is filled by specific conditions that together amount to enabling condi-
tions for the exercise of the ability. It seems plausible that such grasp as we
have of enabling conditions for exercising abilities does not in general take the
form of a conception of some totality of conditions that could be specified in
statements of that sort. Our general conception of the conditions enabling the
exercise of an ability depends on our having an understanding of what that
ability involves that enables us not to be surprised that certain conditions we
come across are liable to hinder or prevent one from exercising it. In the case
of skiing it depends on our understanding that skiing requires one to stay
upright and in control while moving over snow equipped with skis. With that
understanding it is no surprise that skiing on the ice is near impossible.5
(4) Abilities are abilities with respect to favorable environments, that is, environ-
ments that enable the ability to be acquired and in suitable circumstances to be
exercised. My ability to jump over a bar 50cm high depends on the force of
gravity not being significantly stronger than it is near the surface of the Earth.
So it is not just that in an environment with a significantly stronger force of
gravity I would not exercise the ability, but that I would lack the ability with
section 4 that just as performances can be complete or incomplete so the relevant abilities can be com-
pletely or incompletely exercised.
4
There is a distinction between an environment considered as a geographical region, or an analogue of
such a region beyond the Earth, and varying circumstances that obtain in an environment. The occasional
occurrence of a hurricane does not make it the case that one’s environment is unfavorable to the possession
of an ability to ride a bicycle.
5
Peter Sullivan drew my attention to a passage in Philosophical Investigations in which Wittgenstein
comments on claims we make about what people can do. He writes, “We do say, ‘Now I can go on, I mean
I know the formula,’ as we say ‘I can walk, I mean I have time’; but also ‘I can walk, I mean I am already
strong enough’; or ‘I can walk, as far as the state of my leg is concerned’, that is, when we are contrasting this
condition for walking with others. But here we must be on our guard against thinking that there is some
totality of conditions corresponding to the nature of each case (for example, for a person’s walking) so that,
as it were, he could not but walk if they were all fulfilled” (Wittgenstein 2009: §183).
respect to such a place—I would lack the ability to make the jump in such a
place. That does not mean that if I were to be transported to such an environ-
ment I would cease to have the ability we are supposing that I actually have, for
that ability is with respect to the kind of environment that I actually inhabit.
(This turns out to be important in section 6.) Even so, we should not suppose
that the ascription of an ability to jump over a bar 50cm high is shorthand for
a more specific statement that ties the ability to a detailed and specific charac-
terization of favorable environments.
(5) Our ascriptions of abilities tend to underdetermine the character of those abil-
ities. A novice skier and a skier of Olympic medal standard both count as hav-
ing the ability to ski though their abilities differ enormously. Similarly, both a
child who has just learned to read and the highly literate adult count as being
able to read, though the adult is far better able to read quickly and to compre-
hend a wide range of texts. Just as it would be wrong to take ascriptions of an
ability to ride a bicycle, for instance, as shorthand for a longer statement char-
acterizing the ability as an ability to ride a bicycle in such-and-such specified
conditions, so it would be wrong to suppose that an ascription of the ability to
ski is shorthand for a longer statement that specifies the level of the ability. Yet
we can more fully characterize abilities when there is a need to differentiate
between levels of ability.
(6) We are fallible with respect to our abilities. Our fallibility with respect to an
ability to Φ of the sort under consideration consists in its being the case that
we are not bound to Φ every time we act with the intention of Φing. There
could be cases when I act with a view to Φing but circumstances impede the
exercise of the ability and I fail to Φ. I do not think we should rule out the
possibility that even if there is nothing evident that impedes a person’s perfor-
mance—to all intents and purposes the conditions are such as to enable the
ability to be exercised—the agent should nonetheless fail. In the nature of the
case such occasions must be very rare if they occur at all.6 If they were frequent
one would not be reliable enough at carrying out intentions to do the thing in
question to count as being able to do that thing. This conception of fallibility
does not invoke the notion of defective exercises of abilities in the sense of
cases in which a subject counts as exercising an ability while failing to do that
which the ability is an ability to do. The very idea of a defective exercise in that
sense is at odds with the success thesis expressed in (1) above. It looks suspect
at least with respect to the kind of abilities that have been under consideration
so far. In section 4 I consider abilities that might seem to call for the introduc-
tion of the notion of a defective exercise.
6
One might so define “enabling circumstances” that failure on the part of someone with the ability
would entail that the circumstances are not enabling. I am not sure that this would reflect our actual judg-
mental practice. People can have random lapses of concentration that lead to failure even though to all
intents and purposes conditions are favorable.
66 Alan Millar
I have been focusing on abilities such as skiing, reading, and riding a bicycle. I con-
tinue this focus in the next section.
3. Having an Ability and Being Reliably Successful

J. L. Austin pointed out that “there is some plausibility . . . in the suggestion that ‘I can do
X’ means ‘I shall succeed in doing X, if I try’” (Austin 1956/79: 218), though he quickly
drew attention to objections to the view. The analogue for abilities—the claim that
“I am able to Φ” means that I shall Φ if I try—will not do since it fails to connect having
the ability with being reliably successful. It is possible, if highly unlikely, that I lack the
ability to Φ yet Φ by way of a fluke on the only occasion on which I try, and would fail
on nearly every other occasion on which I tried to bring off this feat.7 The proposed
condition is not sufficient for having the ability. We might try to repair by making the
analysans a subjunctive. Allowing at the same time for fallibility, it might be thought
plausible that “I am able to Φ” means that I would very likely Φ if I were to try to, but
then what if someone able to swim tries to swim in a quite unsuitable situation? Ruth
Millikan imagines a case in which a person tries to swim while sitting indoors at a desk.
This person would surely fail if he tried to swim. He would not even be likely to succeed
if he tried (Millikan 2000: 53–4). The example might seem too unrealistic. It is constitu-
tive of having an ability that one is good at selecting occasions to exercise it. So one is
hardly going to try to swim while sitting at one’s desk. One might also wonder what
would count as trying to swim in the envisaged situation. Still, Millikan is right to find
the condition suspect. There can be moments of madness and occasions on which one
is not in a position to determine that conditions for the exercise of the ability are not
favorable or in which one is misled into thinking that they are favorable.
Any proposal that accounts for possession of an ability in terms of success on trying
along the lines considered faces a deeper challenge. Why should trying be thought
central to the explanation of what an ability is? Suppose that the response is that trying
is common to the occasions on which one is successful in doing something and those
on which one fails to do it. Many find this view unexceptionable, but some take issue.
Where, one might ask, is the trying when nothing stands in the way of doing what
one is doing?8 I shall not attempt to adjudicate. Irrespective of whether the worry is
justified I take it that the plausibility of linking ability to success on trying depends on
7
In Millar 2009b: footnote 8, I advanced a manifestation principle to the effect that for all ϕ, one mani-
fests an ability to ϕ if and only if one ϕs. There I used “manifest” where I would now use the term “exercise.”
(See further footnote 3 above.) A similar slip occurs in footnote 10 of Millar 2012. In neither case is the
main line of argument in those places affected. The manifestation principle as stated will not do since it is
not true that, for every ϕ, ϕing is a sufficient condition of manifesting, in the sense of exercising, an ability
to ϕ. Scoring a hole in one at golf by a fluke is not a sufficient condition of exercising an ability to score a
hole in one. (See section 4 here on abilities underlying hard performances.)
8
This concern is swiftly dismissed in Thompson (2008: 102). For sustained discussion of the issue
expressing skepticism that all action implicates trying, see R. A. Duff (1996: chapter 10). Duff ’s targets
include Hornsby (1980 and 1993) and O’ Shaughnessy (1980).
it being the case that exercises of the abilities we have been considering thus far are
intentional actions. I walked to work. Had I fallen seriously ill en route I would not
have succeeded in getting there. At least part of the motivation for assuming that all
intentional actions implicate tryings is the plausible thought that there would be some-
thing common between such a case and the success case in which I walked all the way
to work. Yet there is a perfectly natural way to articulate what is in common without
commitment to the doctrine of trying. It is true of both the success case and the failure
case that I act with the intention of—with a view to—getting to work. It is this notion of
acting intentionally that figures in the account of fallibility in (6) above: the fallibility of
my ability to walk to work consists in the fact that I am not bound to walk all the way to
work whenever I act with the intention of doing so.
There is a connection between having an ability to Φ, where that is an ability of the
sort under consideration, and being reliably successful at carrying out intentions to Φ.
To count as having the ability to Φ one must be reliably successful at Φing in that in the
absence of impediments one would nearly always Φ if one were to act with the intention
of Φing. To count as exercising an ability, then, one must not only do what the ability is
an ability to do but also be reliably successful at doing such things in the sense explained.
There can be cases in which one has an ability, and does the thing that the ability is
an ability to do, but not by way of exercising the ability. I may have the ability to make a
certain sort of arithmetical calculation but on some occasion correctly guess the result
of such a calculation, without having applied the relevant algorithm.9 That I get the
right result has nothing, or at least not enough, to do with having the ability. One might
try to find some principled way to specify the right kind of connection between the
ability and the outcome of interest. I doubt that the notion of the exercise of an ability
will yield to such analysis. For each ability the character of which we have a grasp we
have a working understanding of what it is to exercise it. For some cases, like the exam-
ple just given, that understanding accommodates the possibility of there being factors
entering into the aetiology of an action that prevent that action from being an exercise
of an ability to do that thing, despite its consisting in doing something that the ability is
an ability to do. We could recognize these factors if they occurred but this does not
mean that we have some general specification of what it takes for an action to be an
exercise of the ability that will filter out such cases. That is just the way it is. We are
unlikely to be hampered in our judgments in such matters since bringing off a perfor-
mance of the right kind on the part of an agent who has an ability to do such things is a
highly reliable indicator that the agent’s action is by way of exercising the ability.
4. Hard Performances and Completable Performances

In the article previously cited Austin drew attention to a problem for the view that
“I can Φ” means I shall Φ if I try (1956/79: 218, footnote 1). He remarked that “a human
9
Thanks to Peter Sullivan for suggesting the example.
68 Alan Millar
ability or power or capacity is inherently liable not to produce success, on occasion,

and that for no reason (or are bad luck and bad form sometimes reasons?).” This talk of
abilities not producing success could be taken to suggest that abilities sometimes pro-
duce success and sometimes do not in that they are sometimes exercised in actions that
are doings of the thing that the ability is an ability to do, and sometimes exercised in
actions that are not doings of the thing that the ability is an ability to do.10 On this way
of thinking exercises that fail would be defective exercises. This notion is invoked by
John McDowell in the following passage in which the key point is expressed in terms of
capacities:
A perceptual capacity . . . is a capacity—of course fallible—to get into positions in which one has
indefeasible warrant for certain beliefs. That is what the capacity is a capacity to do, and that is
what one does in non-defective exercises of it, exercises in which its acknowledged fallibility
does not kick in. (2010: 245)
At this point it is not McDowell’s view of perceptual warrant that matters but his con-
ception of fallibility, which is meant to apply to capacities other than cognitive ones.
He gives this example:
Some people have a capacity to throw a basketball through the hoop from the free-throw line.
Any instantiation of such a capacity is imperfect; even the best players do not make all their
free throws. (2010: 245)
It is clear from the context, which includes the first of the passages quoted above, that
McDowell thinks that cases in which the player does not get the ball into the basket
from the free-throw line can be defective exercises of a capacity to throw the ball into
the basket from the free-throw line. In the following passage from another recent work
he is more explicit:
If a capacity is fallible, or if . . . anyone who has it is fallible in respect of it, that means that there
can be exercises of the capacity in which its possessor does not do what the capacity is specified
as a capacity to do. (2011: 37)
Though this is addressed to the topic of cognitive capacities, I take it that it is meant to
apply to all capacities. In the ensuing discussion I shall revert to speaking in terms of
abilities.
Examples like that of the basketball player are liable to mislead us. In discussion
I have encountered the following:
Consider the basketball player making his throws. Plausibly, (i) something that is central to
what is done in the cases of failure is also central to what is done in success cases that indisput-
ably count as exercises of the ability to get the ball into the basket, and (ii) the similarity in the
cases warrants thinking that what is done in the cases of failure is an exercise of the ability. The
second of these claims receives support from cases in which the player throws in such a way
that the ball would have gone into the basket but for interference (for instance, a member of the
10
I don’t suggest that this is the only way to interpret Austin’s comment.
audience leaping forward and deflecting the ball). In such a case the player has exercised an
ability to throw the ball into the basket but has not thrown the ball into the basket. That is to
say, the player’s exercise of the ability is defective.
I do not attribute this line of thought to McDowell but merely consider it as a possible
motivation for introducing the notion of a defective exercise of an ability. It is prob-
lematic because it assumes that the player has, without qualification, an ability to
throw the ball into the basket. Perhaps ordinary language does not proscribe talking
in this way. I aim to show that, nonetheless, it is misleading as to the character of
the ability.
When the player throws the ball into the basket there is surely some sense in which
he is able to do so. Moreover, this comes to more than that it is merely possible that he
should get the ball into the basket. For this player is rather good at getting the ball into
the basket and when he succeeds it will usually be because he has done something that
raises the chance of his succeeding. Nonetheless, saying of him that he is able to throw
the ball into the basket runs the risk of assimilating his ability to that of someone who
can read English, when it is more like that of the darts player with respect to throwing a
dart into the bullseye.
There is an important distinction to be made here. Performances like throwing the
dart into the bullseye are hard performances in that they are so difficult to bring off that
it is not to be expected that those with expertise, acting to bring them off, will always,
or even most of the time, succeed. Easy performances, by contrast, are ones for which
such an expectation is appropriate. To understand why someone does not bring off a
hard performance, no special explanation in terms impairment, unfavorable condi-
tions, or the like, is required. By contrast, it is reasonable to seek a special explanation
when someone able to read cannot do so on some occasion. It is overwhelmingly likely
that some factor has impeded the agent’s performance. Easy performances in the some-
what specialized sense in play here need not be easy in the ordinary sense. Landing a
passenger aircraft safely is difficult in that it takes a lot of skill born of rigorous training.
Nonetheless, in the specialized sense, for those with the requisite expertise, it is easy:
commercial airline pilots land safely except in the most exceptional circumstances.
Throwing a basketball into the basket from the free-throw line, throwing a dart into
the bullseye, potting a snooker ball with a long shot, and putting a golf ball into the
hole from ten metres are all hard performances. It is not to be expected that even
experts will succeed on every, or even nearly every, occasion when they act with a view
to doing so.11
11
The difference between hard performances and easy performances is of interest to us. Someone
with an impressive success rate at a hard performance, such as throwing the basketball into the basket
from the free-throw line, is not open to reasonable censure or criticism for not being successful at a single
attempt. Even if a player’s failed attempt may disappoint, it would be inappropriate to be disappointed by
the player unless his routine for bringing off the performance had been carelessly executed. By contrast,
we expect a person who can drive safely to drive safely whenever he aims to do so, barring exceptional
circumstances.
70 Alan Millar
Two issues arise at this point. Towards the end of section 3, in connection with the
ability to ride a bicycle and others like it, I claimed that “to count as having the ability to
Φ one must be reliably successful at Φing in that in the absence of impediments one
would nearly always Φ if one were to act with the intention of Φing.” Call this the strong
reliability requirement. The first issue is whether this requirement, or something very
similar, applies to the abilities in play in hard performances. The second issue is
whether hard performances pose a problem for a generalized version of the success
thesis. Is it true of abilities in general, thus of the abilities in play in hard performances,
that one exercises them only if one does, or is doing, the thing that the ability is an
ability to do? If the right answer to this were, “No,” we would need to invoke the notion
of the defective exercise of an ability.
It might seem straight off that nothing like the strong reliability requirement can
apply to abilities relevant to being good at hard performances. These are by definition
performances such that it is not to be expected that even those with expertise
can nearly always bring them off when they attempt to in favorable circumstances.
Even so, we need to think more carefully about the nature of the abilities in play in
hard performances.
If we allow ourselves to regard the basketball player as having, without qualification,
the ability to throw the ball into the basket, we mask a significant difference between
his actual ability and the kind of ability we attribute to people who can ride a bicycle.
Saying that the basketball player has the ability to get the ball into the basket can
amount to no more than saying that the player has an ability to bring off this particular
performance a good number of the times on which he aims to do so. The kind of ability
we attribute to the person who has the ability to ride a bicycle is not like that, except
perhaps in the very early stages of learning. What this suggests is that we need the
notion of a success-rate ability. Success-rate abilities are abilities to succeed at a hard
performance a good proportion of the times on which one acts to do so, where that
proportion falls well short of success every time but is not so low as to make successes
no better than flukes. The basketball player has a success-rate ability. We can best bring
out its character if we say that it is an ability to achieve an impressive rate of success in a
sequence of attempts. Such an ability is not to count, without qualification, as an ability
to throw the ball into the basket, on pain of obscuring its character.
Return now to the line of thought displayed above that was in support of the idea
that there are defective exercises of abilities. Barring unfavorable conditions, in many
cases of failure and many cases of success, the expert player goes through a routine that
involves throwing the ball in such a way that there is a good chance that it will go into
the basket. There is an ability exercised in both successful performances and unsuc-
cessful performances in which the player’s routine is implemented. It is the ability to go
through the routine, that is, to do what the player generally does to get the ball into the
basket—the performance that raises the chance that the ball will go in. Notice though
that this ability can be non-defectively exercised in both sorts of case. Even when
the player does not get the ball into the basket it can be that he has gone through the
routine just as he has in cases in which the ball goes into the basket. If he makes a mess
of performing the routine, then, of course, there is no question of his having exercised
the ability to go through the routine.
We are now in a position to address the first of the issues raised above. This was
whether anything like the strong reliability requirement applies to the abilities brought
into play in hard performances. As it stands this requirement was designed to accom-
modate abilities exercised by intentional actions. That is fine so far as the ability to
ride a bicycle is concerned. But a success-rate ability can be exercised by going through
a sequence of performances even if the agent has no intention to go through the
sequence. A basketball player can be exercising a success-rate ability in the course of
a sequence of attempts to throw the ball into the basket from the free-throw line.
While each of the attempts would be an intentional action, the player need not have
any intention regarding any sequence of attempts, thus no intention to achieve an
impressive level of success in any such sequence. He just does his best each time he
throws. So it is best not to think of his reliability as a matter of his being such that
he would nearly always achieve an impressive success rate if he intended to achieve an
impressive success rate.12 In the light of the previous considerations we can see that
there is a strong connection between abilities in general, including those exercised in
achieving impressive success rates, and high levels of reliability—one that is closely
analogous to that captured by the strong reliability requirement. Suppose that Φing is
achieving an impressive success rate in a sequence of attempts at Ψing, where Ψing
is hard. Agents count as having a success-rate ability to Φ only if, absent impediments,
they nearly always achieve an impressive success rate in sequences of attempts at
Ψing. The high degree of reliability that is associated with Φing is compatible with
there being a significantly lesser degree of reliability at Ψing, albeit one that is still
impressive given that Ψing is hard.
The ability of our expert basketball player to throw the ball into the basket from the
free-throw line is not to count, without qualification, as an ability to bring off this feat.
This player has (a) an ability to go through his routine, and (b) an ability to achieve an
impressive success rate at getting the ball into the basket from the free-throw line. Both
of those abilities implicate a high degree of reliability. Ability (a) is such that the player
is reliably successful to a very high level at doing what the ability is an ability to do, that
is, go through the routine. Ability (b) is such that the player is reliably successful to a
very high level at doing what the ability is an ability to do, that is, achieve an impressive
success rate over a sequence of attempts.
The second issue was whether we can make sense of the character of the basketball
player’s ability, and his successes and failures, while respecting a generalized success
thesis along the following lines: for any ϕ, one exercises an ability to ϕ only if one ϕs (or
is ϕing). It is clear now that abilities brought into play in hard performances give us no
12
Such a characterization might be true of him, but it might not be. Setting himself to undertake a
sequence of attempts to achieve an impressive success rate might inhibit him.
72 Alan Millar
reason to invoke the notion of a defective exercise of an ability. Abilities like (a) and
abilities like (b) are exercised only in doing what the ability is an ability to do.
There is another class of performances that might be thought to put pressure on the
generalized success thesis. This comprises performances like making an omelette,
planting a row of seeds, walking to work, and building a hut. It can be true of me that
I am making an omelette but have not yet made it. By contrast, it is never true of me
that I am skiing but have not yet skied. Accordingly, by contrast with skiing, perfor-
mances like making an omelette might be called completable performances.13 The
ability to make an omelette is presumably not just the ability to engage in the process of
making an omelette. Rather, it is the ability to bring the procedures for making an ome-
lette to (successful) completion. Yet while in the process of making an omelette I can be
interrupted and never complete the job. It might seem that by my account of what it is
to exercise an ability I will not have exercised the ability to make an omelette if I have
not completed the job. The thought here is that the ability in question is nothing less
than the ability to make, and complete the making of, an omelette and, it seems, by the
account, will be exercised only if one makes, and so completes the making of, an ome-
lette. Yet it is odd to suppose that while in the process of making an omelette I am not
exercising the ability to do so. We are not compelled, though, to make this odd suppo-
sition. Just as one can be making an omelette without having completed making it, so
one can be exercising the ability to make an omelette without having completed the
exercise of the ability. We may call such incomplete exercises imperfect exercises, but
the important point for present purposes is that they are not defective exercises, con-
ceived as cases in which an ability is exercised though the agent is not doing what the
ability is an ability to do. In the midst of making an omelette one is by this account
exercising the ability to make an omelette. (This accounts for the qualification to which
I alluded in (1) of section 2.)
5. Mental Acts
Thus far we have been dealing with abilities the exercises of which are intentional
actions or sequences of intentional actions. I turn now to consider a kind of cognitive
ability the exercises of which are mental acts, not intentional actions. By “mental acts”
I mean dateable episodes like deciding to do something and judging that something
is so.14 A central case of decision is settling on a certain course of action from among
others one might have settled on. Central cases of judging include determining that
something is so on seeing that it is so and it striking one that something is so in the face
of compelling evidence.
13
I thank Adrian Haddock for pressing me to think about such cases and for helping me to correct my
initial reaction to them. The philosophical import of “imperfective” or “incomplete” characterizations of
actions is interestingly explored in Thompson (2008: chapter 8).
14
The term and the concept figure in Peter Geach’s little book Mental Acts (Geach 1957).
Deciding and judging are not intentional actions. Whereas I may intentionally go
for a walk my decision to go for a walk, rather than watch sport on television, is not an
intentional action. The decision might have been preceded by some consideration of
possibilities for action. If so that consideration would have been intentionally pursued,
but the decision itself would not have been intentional in that it would not have been
an action I engaged in to bring about or realize an end that I had set myself. Similarly,
although a judgment might have been preceded by active observation, or considera-
tion of evidence—actions that are intentionally pursued—the judgment itself is not an
intentional action.
We can make sense of fallibility in relation to abilities exercised by non-intentional
mental acts in a manner that is closely analogous to the treatment of the fallibility of
the abilities that were discussed earlier.
It is crucial that deciding and judging are in a sense directed acts, though not in the
sense that we direct them at some end. Decision raises problems that will take us too far
off topic. For present purposes it is judgment that matters and, specifically, judgments
made from the appearance of something currently perceived. All such judgments are a
species of recognitional judgment in that they are directed at recognition. Their being
so directed does not entail that they are all acts of recognition.15
There are two aspects to directedness in such cases: (i) a recognitional judgment
admits of evaluation in terms of whether it is successful or not and (ii) the subject
making such a judgment is suitably sensitive to considerations bearing on whether the
judgment is successful. The success condition for a perceptual-recognitional judgment
that P is not just that it should be that P but that the subject should have grasped—
come to know—that P in an act of perceptual recognition. Illustrations of the right
kind of sensitivity include one’s preparedness to amend a recognitional judgment that
a bird is a buzzard on seeing a feature of the bird’s tail that shows it to be a kite, or to
withdraw such a judgment if told by a source one takes to be reliable that no buzzards
are in the area. It also seems plausible that an aspect of sensitivity in relation to recogni-
tional judgments concerning buzzards would be preparedness to withhold a judgment
that a bird is a buzzard if it emerges that one has not yet mastered the ability to discrim-
inate buzzards from other raptors with which they might be confused.
As I emphasized previously, to recognize something as belonging to some kind, or
as having some property, from the way it appears to some sense modality, is to exercise
an ability, specifically, an ability to tell, that is, come to know, of certain things that they
are of that kind or have that property from their appearance to that sense modality.
Similarly, to recognize an individual as your friend Sally from the sound of her voice is
to exercise an ability to tell, that is, come to know, of that individual that she is your
friend Sally from the way her voice sounds.
15
Decision is complicated because it is implausible that decision is directed at the good even when
allowance is made for what Raz (1999: 30) calls “small” goods. I am inclined to think that decision is
directed at an action’s having some point. For related discussion, see Millar 2009a.
74 Alan Millar
It is time now to review the extent to which the features I described in connection
with the abilities discussed in sections 2 and 3 are possessed by perceptual-recognitional
abilities conceived along these lines. Each of the following, which correspond to the
list (1)–(6) in section 2, is plausible:
(1´) The success thesis extends to perceptual-recognitional abilities: one exercises

such an ability only if the relevant recognition is effected, since what the ability is an
ability to do is nothing short of recognition. We have failures of recognition when we
make a judgment directed at recognition, or aspire to make such a judgment, but in
either case fail to effect recognition. There can, of course, be failure because we lack the
requisite ability. More interesting are cases in which we have the requisite ability but
for one reason or another fail to exercise it. Within this class are two broad types of
failure. The most common, type A, are cases of failure to recognize a perceived object as
being of some kind or as having some property or as being some specified individual.
Among those are cases in which one judges a perceived object from its appearance to
be such-and-such when it is not (type A1). When this happens we have a judgment
that is directed at recognition but which fails to be recognition because the judgment is
false. Also of type A are cases (type A2) in which judgment is suspended. These include
cases like that in which we have a recognitional ability appropriate for recognizing a
flower as belonging to some kind, but fail to make out whether a flower at which we are
looking is of that kind and have to suspend judgment. The less common broad class of
failures, type B, comprises cases in which it is to the subject as if he or she is recognizing
an object as being some way but no such object is perceived. Examples include percep-
tual Gettier cases (type B1). Suppose you are looking at a depiction of your friend Bill
that makes it look to you as if Bill is at a certain location, and you judge correctly that he
is at this location—he’s behind the depiction. You have not visually recognized Bill
because you have not seen him. Accordingly you have not recognized him to be at the
location at which you correctly take him to be. Also of type B are cases of hallucination
(type B2). Suppose it looks to you just as if a rat is crawling up the wall, but you are
hallucinating, and no rat is present. You are so disposed that if you were to try to
express what you think you would say, “That is a rat,” as if picking out a rat in a demon-
strative judgment. But no such judgment is made and no recognition of anything as a
rat is effected. (Since no rat has been picked out there is no object to which anything
would be attributed if one made such an utterance. Other judgments are likely to be in
the offing, for instance, that a rat is crawling up the wall.)
(2´) Exercise of a perceptual-recognitional ability might be prevented by a number
of factors. There seems to be nothing comparable to the lack of means noted in relation
to the abilities considered in (2) of section 1. There the point was that someone who
has ability might lack the means to exercise it, as when someone with the ability to ski
cannot exercise the ability for want of equipment. Of course, if we can recognize light
switches as light switches from the way they look we must possess the concept of a light
switch, but if we lack the concept we lack the ability as well, not just the means to
e xercise it. Clearly, it is possible to have a perceptual-recognitional ability and, due to

lack of fitness, exhibit a failure of recognition. Cases of hallucination (type B2 failure)
evidently exhibit lack of fitness. And if you have just eaten anchovies you might judge
other cooked food to be under salted when it is adequately salted (type A1 failure). You
might try to make out who is on the other side of the road, but fail to do so and suspend
judgment due to temporarily blurred vision (type A2 failure). Finally, as with the other
abilities discussed, there can be failures due to temporary environmental circumstances
that are unfavorable to the exercise of an ability. In dim light you might suspend judg-
ment on whether a bird at which you are looking is a willow warbler (type A2 failure).
The Gettier case described under (1´) (type B1 failure) illustrates this sort of impedi-
ment. In such cases the environment misleads a subject into making a recognitional
judgment yet prevents the relevant recognition.
(3´) A statement to the effect that a person has an ability to recognize Fs as Fs (G
things as G things/or X as X) from the way they look (or it looks) is not shorthand for a
statement that the person has the ability to effect such recognition, provided that this
and that specified condition is satisfied. As with the other abilities considered it seems
plausible that such grasp as we have of conditions enabling recognition flow from our
understanding of the character of that type of recognition. We do not expect people
with an anesthetized tongue to be able to recognize tastes while in that condition, and
nor do we expect people with blocked-up ears to be able to recognize sounds. Even so,
we are not set up in advance with a command of all of the specific conditions that pre-
vent recognition.
(4´) Perceptual-recognitional abilities are abilities with respect to favorable envi-
ronments. In a fake-barn scenario we could not tell of structures there that they are
barns from their visual appearance. It is not just that we could not exercise the ability;
we would not have the ability with respect to structures in that place because we cannot
discriminate barns there from fake barns. (I return to this theme at greater length in
section 6.)
(5´) Our ascriptions of perceptual-recognitional abilities routinely underdetermine
the character of those abilities. An expert at recognizing orchids as orchids from their
visual appearance and a person able to recognize only supermarket orchids, and others
sufficiently like them, might both count as having the ability to recognize orchids as
orchids from their visual appearance. Both might be reliably successful in the circum-
stances in which they are prepared to make the relevant judgments, and normally able
to make true judgments when instances that their ability comprehends are evidently
present. Yet their respective abilities would be very different. We should not infer from
this that our ascriptions of abilities are shorthand for some more specific ascription.
Yet, where circumstances call for greater precision a more specific characterization of
the ability can be given.
(6´) In the light of our discussion of directedness of judgments we can now see that
there is a close analogue of the account of fallibility set out in (6) of section 2. Our falli-
bility with respect to an ability we have to recognize things as being some way amounts
76 Alan Millar
to it being the case that we are not bound to exercise it every time we make a judgment
directed at recognition of the relevant sort or merely aspire to do so. The latter alterna-
tive accommodates both cases of failure to recognize in which judgment is suspended
and type B failures in which it is to the subject as if he or she is making a demonstrative
judgment, thereby picking out an object and recognizing it to be some way, but no such
judgment is made, because no object is picked out.
Against the background of the discussion in section 4 it should be emphasized that
perceptual-recognitional abilities are not success-rate abilities. Barring freak circum-
stances those who have the concept of a crow, and have learned to apply it in visual-rec-
ognitional judgment, would nearly always apply it correctly so long as the
circumstances do not impede recognition. Adapting the terminology introduced in
section 4, we may think of recognitional judgments as easy performances, rather than
hard performances.
There is kind of case that might be thought to put pressure on the success thesis for
recognitional abilities. An expert physician might make a false diagnosis of a patient.
The falsity need not be due to any incompetence on the part of this person since in the
light of received theory and clinical practice the diagnosis might have been entirely
reasonable and accepted by other experts. The worry here might be that this doctor has
exercised suitable abilities even though his judgment is false. But while the doctor has
surely exercised suitable abilities in such a situation, the case is not a counterexample
to the success thesis for recognitional abilities or for abilities in general. What counts as
successful exercise of an ability depends on what the ability is an ability to do. The kind
of ability in play in this case is an ability to select a diagnosis that is reasonable in the
light of the evidence and background information available. Since ex hypothesi the
doctor has done that, it is not the case that he has exercised the relevant ability by doing
something other than what the ability is an ability to do.
The resulting perspective enables us to capture a point on which McDowell has
insisted without invoking his notion of a defective exercise of a cognitive ability. From
it being possible that one should mistakenly judge animals one sees to be zebras from
the way they look it does not follow that one never has warrant for believing an animal
to be a zebra constituted by the fact that one sees that it is. This strikes me as being
exactly right, but we may quite naturally think of the case of incorrect judgment as one
in which the ability is not exercised at all, while preserving the asymmetry with respect
to warrant between the good case in which recognition is effected and the bad case in
which it is not. What matters is that in the good case one has a warrant that one lacks in
the bad case. In the good case the warrant to believe that the animal is a zebra is given
by the fact that one sees that the animal is a zebra. How such a fact is available to one
needs explanation, but that it can be available is presumably not in dispute. In the bad
case the ability to recognize zebras as zebras is not exercised. Accordingly, in that case
there is no fact to the effect that one sees that the animal is a zebra that can serve as a
warrant to believe that it is. So the cases are asymmetrical with respect to warrant. They
provide no decisive reason to invoke the notion of a defective exercise as McDowell

does in the works cited earlier in section 4.16
It might be thought that the similarity between what goes on in a case of perceptual
recognition, and what goes on in a counterpart case in which a judgment directed at
recognition fails to be recognition, gives us reason to suppose that the same ability is
exercised in the two cases. There is no question that there is a similarity. Reverting to
the zebra example, in both the good and bad cases there is an act directed at recogni-
tion—a judgment concerning the object seen that is made from its visual appearance
and to the effect that it is a zebra. Since recognition is not effected in the bad case the
ability is not exercised in that case. We are not at a loss to characterize a way in which
the cases are the same. I have just done so.
6. Sosa and Competence

The account of perceptual knowledge in terms of recognitional abilities falls within a
spectrum of philosophical accounts of perceptual knowledge in which the acquisition of
such knowledge is conceived as the exercise of a competence, capacity, or ability. Ernest
Sosa has over many years developed and refined a general virtue theory of knowledge
that makes a certain conception of competence central. He conceives of knowledge as
apt belief, and aptness of a belief as true belief the formation or maintenance of which
is a manifestation of relevant competence on the part of the believer in attaining the
truth (Sosa 2007: Lecture 2, 2010, 2011: 4). Competences are taken to be abilities (2011:
chapter 4, section 7).
It is instructive to reflect on respects in which Sosa’s general approach compares
with that set out above. A salient difference has to do with his treatment of examples of
the fake-barn type, in which objects of a sort that in familiar environments have a dis-
tinctive appearance are imagined to be in an unusual environment in which they are
16
McDowell himself says what needs to be said in the following passage from an earlier article without
invoking the notion of a defective exercise of an ability:
I can tell a zebra when I see one. . . . If what I believe to be a zebra is actually a cunningly
painted mule, then of course I do not recognize it as a zebra, as I suppose, and I do not have
the warrant I think I have for believing it is a zebra, namely that I see it to be a zebra. My
ability to recognize zebras is fallible. It does not follow—this is the crucial point—that I
cannot ever have the warrant for believing that an animal in front of me is a zebra consti-
tuted by seeing that it is a zebra. If the animal in front of me is a zebra, and conditions are
suitable for exercising my ability to recognize zebras when I see them . . . then that ability,
fallible though it is, enables me to see that it is a zebra. . . . (McDowell 2008: 387)
These claims are compatible with the conception of fallibility that I have been outlining. The remainder
of the paragraph from which the quotation comes makes it clear that McDowell thinks that in exercising
the ability whereby one recognizes the zebra to be a zebra, one at the same time comes to know that one
sees the animal to be a zebra. One and the same ability accounts both for recognition and for being apprised
of the reason that provides one with a warrant to believe the animal to be a zebra. I provide a different
account in Millar 2011a and 2011b—one that does not assume that being apprised of the reason is inextri-
cable from recognition, but has it that in reflective agents functioning normally the two go together.
78 Alan Millar
interspersed with, or accessibly nearby, objects that have this appearance but are not of
the sort in question. As the following passage makes clear, the issue for Sosa is whether
subjects placed in those environments lose a competence that they would normally be
taken to have.
Someone views a red surface in good light and believes it to be red. He would seem thereby to
manifest his competent color sight. What if all nearby surfaces that also look red are actually
white surfaces bathed in red light? Does the fake color environment take away the subject’s
color sight competence? I cannot see that it does. (2011: 82. Cf. 2007: 35–6.)
Sosa thinks of the relevant color sight competence as an ability, and he treats abilities
as dispositions that, like all dispositions, are associated with trigger-manifestation
conditionals (2010: 466, 2011: 80–2). The relevant trigger-manifestation conditional
for color-sight competence would, I take it, be something like, “If you were to make a
judgment as to the color of a perceived surface from the way it looks, you would likely
judge correctly.”17 Sosa emphasizes that conditionals such as this hold true only when
appropriate conditions obtain. He takes this to reflect our conceptions of dispositions.
We do not think it likely that a person who is competent at judging the colors of things
from their visual appearance will be reliable at judging correctly in circumstances in
which, as a result of unusual lighting conditions, the apparent colors of many surfaces
are not reliably indicative of their true colors.
Against this background Sosa considers a reaction to the attitude he expresses in the
passage displayed above. He imagines it being suggested that the subject looking at
surfaces under the misleading light does not have the relevant competence and spells
out the thinking underpinning this reaction as follows:
the competence to spot barns, or red surfaces, cannot be restricted to the object one happens to
be viewing at that moment. A relevant perceptual competence must extend also across a broader
neighborhood. An environment containing enough fakes would deny one such a spread com-
petence, even when one forms a belief about a barn or surface to which one is appropriately
related. One is denied such a fuller competence even if one is appropriately related to that
particular object. (Sosa 2011: 82–3)
Before we turn to Sosa’s response to the line of thought he articulates in the above pas-
sage it will be useful to take stock of what should be said about the scenario with the red
light conditions from the perspective of the recognitional-abilities account of percep-
tual knowledge. Bear in mind that from this perspective an ability to recognize red
surfaces to be red from their visual appearance is an ability to tell, thus come to know,
17
In Knowing Full Well (2011, especially, pp. 75–8) Sosa has a conception of perceptual competence the
outputs of which are experiences with propositional contents. Having an experience representing there to be
a red surface before one would on this view fall short of believing or judging that the surface is red. So one
might wonder whether Sosa takes color-sight competence to be tied to trigger-manifestation conditionals,
along the lines of, “If you were to look at a surface, you would likely have an experience that represents it to
be the color that it is” rather than the sort I have illustrated in the main text. However, in the quoted passage
on which I am commenting it seems clear that the competence of which Sosa is speaking has beliefs as out-
puts. Sosa calls this kind of competence a perceptual epistemic competence (2011: 81 (bottom of the page)).
of red surfaces that they are red from the way they look. It is entirely natural to think of
people as having this ability, even if in an unusual environment they would incorrectly
judge some surfaces to be red. (This corresponds to Sosa’s claim about how to under-
stand trigger-manifestation conditionals.) Even so, a person could count as having the
ability simpliciter while lacking the ability to recognize red surfaces to be red in circum-
stances in which the unusual light conditions prevail. (We need to recall here that speci-
fications of abilities routinely underdetermine their character. See (5) of section 2 and
(5´) of section 5.) On the face of it there is reason to deny a subject this more finely
specified ability. The thought here is simply that unless the visual appearance of being
red is a very highly reliable indicator of being red you cannot tell that a thing is red
from the way it visually appears. It is not true of the unusual environment that having
the visual appearance of something red is a highly reliable indicator of being red.
This way of thinking is in keeping with the way we think of abilities more generally.
One may count as having the ability to ski simpliciter even if one is unable to ski in a bliz-
zard or to ski on a very steep slope. Lacking the more finely specified ability is compatible
with having the ability to ski simpliciter. From this perspective it makes sense to deny
people the ability to recognize red surfaces to be red with respect to an environment in
which having a red appearance is not a highly reliable indicator of being red. That people
in those circumstances lack the more finely specified ability does not show that they lack
an ability to recognize red surfaces as red simpliciter. Sosa sometimes writes as if he
thought that those who deny the subject the relevant competence with respect to the
unusual scenario are committed to supposing that the subject has lost a competence that
he or she so plainly retains. In one place he represents his opponents as supposing that
the subject entering fake-barn territory “loses his full competence [at spotting barns]”
(Sosa 2010: 469, first paragraph). But Sosa’s opponents need not suppose that anything is
lost on entering fake-barn territory. On my account of recognitional abilities the prob-
lem in fake-barn territory is not due to losing anything one had but simply to lacking an
ability to recognize barns around there as barns from the way they look.
Sosa resists the view that we should deny the observer, in the unusual scenario
in which some white surfaces look red because bathed in red light, a competence at
spotting red surfaces. Taking abilities to be dispositions, he sets out his reasons in the
following passage:
Rarely if ever does a recognized, commonsense disposition require that its triggering condi-
tions would trigger its resultant manifestations not only at the very place and time where the
host of the disposition is located, but also throughout some wider neighborhood of places and
times. . . . If a wine glass is dropped on a hard counter, its shattering can manifest its fragility
even if all nearby horizontal surfaces are outfitted with cushions. (Sosa 2011: 81)
Applying these considerations to the case of color judgment, the conclusion drawn is
that it would be wrong to deny that a correct judgment that a surface is red, made in the
unusual scenario, is a manifestation of the relevant competence simply on the grounds
that one would have incorrectly judged surrounding surfaces to be red.
80 Alan Millar
Suppose, however, that we take into account the idea that an epistemic competence
is, as Sosa says, “competence to judge well” (Sosa 2010: 467) and that a perceptual epis-
temic competence is “an ability to discern the true from the false in a domain corre-
sponding to that perceptual competence” (Sosa 2011: 82–3). That suggests that having
a perceptual epistemic competence with respect to judgments in some class (a domain)
entails being reliably successful at making correct judgments in that class. Within this
perspective there are natural analogues of the points about Sosa’s example that I made
earlier in this section from within the recognitional-abilities perspective. An implica-
tion of Sosa’s remarks is that one does not fail to be good at judging surfaces to be red
from their visual appearance just because such judgments might easily go awry in the
unusual scenario. That implication looks right but accepting it is compatible with
denying to a person a more finely specified competence, in particular, the competence
consisting in being good at correctly judging surfaces in the unusual conditions to be
red from their visual appearance. There is no tension here, nor is there any suggestion
that our abilities are generally specified in fine-grained terms. (Recall (3) in section 2
and (3´) in section 5.) Furthermore, we have an explanation of why it would be correct
to deny the subject the more finely specified ability. To be good at judging surfaces in
the unusual scenario to be red from their appearance one has to be reliably successful
in making such judgments there—good at making correct judgments there—but one
would not be reliable enough in this respect on account of the fact that too often, there,
having the visual appearance of being red is not indicative of being red. We also have
within Sosa’s perspective an explanation of why a subject who is unreliable at making
correct judgments to the effect that a perceived surface is red in the unusual scenario
could still count as being good at judging surfaces to be red from their appearance: the
general competence—being good at judging surfaces to be red from their appear-
ance—does not require that one be good at making such judgments in environments
in which visual appearance is not a reliable indicator of color.
There are clearly significant differences between Sosa’s perspective and that pro-
vided by the recognitional-abilities account. The most obvious is that, whereas on
Sosa’s account the epistemic competences he invokes in connection with perception
are manifested by formations of beliefs, on my account recognitional abilities are exer-
cised only in episodes of recognition, conceived as acquisitions of knowledge. There is
no recognition in the scenario with the unusual light conditions because a pre-requi-
site of recognition—(near) distinctiveness of the relevant appearance—does not
obtain.18, 19
18
I am grateful to Ernest Sosa for the challenges posed by his calmly tenacious defense of his views. The
challenges have in recent years provided a helpful stimulus to the development of my own thinking.
19
Colleagues at Stirling have been especially helpful. I’d like to thank, in particular, Adrian Haddock,
Simon Hope, Colin Johnston, Sonia Roca Royes, Ben Saunders, Peter Sullivan, and Michael Wheeler for
useful comments. I am also grateful to Santiago Echeverri for written comments on an earlier draft and to
Sara Neva for both her comments on a related presentation I gave at a conference on skepticism in Bologna
in December 2010 and for subsequently supplying further written comments. Finally, I thank two anony-
mous readers for comments that prompted improvements.
References
Austin, J. L., 1956/79. “Ifs and Cans.” Proceedings of the British Academy, 1956; repr. in Austin’s
Philosophical Papers, 3rd edn., 205–32. Oxford: Clarendon Press, 1979. Page references are
to this reprint.
Duff, R. A., 1996. Criminal Attempts. Oxford: Clarendon Press.
Geach, P., 1957. Mental Acts. London: Routledge and Kegan Paul.
Goldman, A. I., 1976. “Discrimination and Perceptual Knowledge.” The Journal of Philosophy
73: 771–91.
Hornsby, J., 1980. Actions. London: Routledge.
Hornsby, J., 1993. “On What’s Intentionally Done.” In S. Shute, J. Gardner, and J. Horder (eds)
Action and Value in Criminal Law, 55–74. Oxford: Oxford University Press.
McDowell, J., 2008. “The Disjunctive Conception of Experience as Material for a Transcendental
Argument.” In A. Haddock and F. Macpherson (eds) Disjunctivism: Perception, Action,
Knowledge. 376-89. Oxford: Oxford University Press.
McDowell, J., 2010. “Tyler Burge on Disjunctivism.” Philosophical Explorations 13: 243–55.
McDowell, J., 2011. Perception as a Capacity for Knowledge. Milwaukee, WI: Marquette
University Press.
Millar, A., 2007. “What the Disjunctivist Is Right About.” Philosophy and Phenomenological
Research 74: 176–98.
Millar, A., 2008a. “Perceptual-Recognitional Abilities and Perceptual Knowledge” In A. Haddock
and F. Macpherson (eds) Disjunctivism: Perception, Action, Knowledge. 330–47. Oxford:
Oxford University Press.
Millar, A., 2008b. “Disjunctivism and Skepticism.” In J. Greco (ed.) The Oxford Handbook of
Skepticism, 581–604. New York: Oxford University Press.
Millar, A., 2009a. “How Reasons for Action Differ from Reasons for Belief.” In S. Robertson
(ed.) Spheres of Reason: New Essays in the Philosophy of Normativity, 140–63. Oxford: Oxford
University Press.
Millar, A., 2009b. “What Is It that Cognitive Abilities Are Abilities to Do?” Acta Analytica 24:
223–36.
Millar, A., 2011a. “Knowledge and Reasons for Belief.” In A. Reisner and A. Steglich-Peterson
(eds) Reasons for Belief, 223–43. Cambridge: Cambridge University Press.
Millar A., 2011b. “How Visual Perception Yields Reasons for Belief.” Philosophical Issues 21:
The Epistemology of Perception, 332–51.
Millar, A., 2012. “Scepticism, Perceptual Knowledge, and Doxastic Responsibility.” Synthese
189: 353–72.
Millar, A., 2014. “Perceptual Knowledge and Background Beliefs.” In D. Dodd and E. Zardini
(eds) Scepticism and Perceptual Justification, 128–48. Oxford: Oxford University Press.
Millikan, R., 2000. On Clear and Confused Ideas: An Essay about Substance Concepts. Cambridge:
O’ Shaughnessy, B., 1980. The Will: A Dual Aspect Theory. Cambridge: Cambridge University
Press.
Pritchard, D., A. Millar, and A. Haddock, 2010. The Nature and Value of Knowledge: Three
Investigations. Oxford: Oxford University Press.
Raz, J., 1999. Engaging Reason: On the Theory of Value and Action. Oxford: Oxford University
Press.
82 Alan Millar
Sosa, E., 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, i. Oxford: Clarendon
Press.
Sosa, E., 2010. “How Competence Matters in Epistemology.” Philosophical Perspectives 24:
465–75.
Thompson, M., 2008. Life and Action: Elementary Structures of Practice and Practical Thought.
Cambridge, MA: Harvard University Press.
Wittgenstein, L., 2009. Philosophical Investigations, revised 4th edn. Oxford: Wiley-Blackwell.
5
Against Actual-world Reliabilism
Epistemically Correct Procedures and Reliably
True Outcomes
Peter J. Graham
I. Being in the Right

Being justified in general means being in the right. This involves meeting some standard
or norm for correctness. Justified in theology means being made righteous in the eyes
of God. Justified in printing means the text lines up along the left, right, or both mar-
gins; the edges set the standard. Justified in the law means having or showing reasons
for having committed the act to be answered for. What does being justified in episte-
mology mean?1
In general, being justified in epistemology means being in the right vis-à-vis the
goal of believing truth and avoiding error. That tells us something. But it leaves quite
a lot open. Let us narrow the scope to beliefs. A justified belief is then a belief that
is in the right vis-à-vis the goal of believing truth and avoiding error. A justified
belief then meets a standard or norm understood in terms of promoting truth and
avoiding error.
One way for a belief to be in the right in this sense is to be true; truth would set the
standard. Are all true beliefs then justified? No. This is not the sense intended in episte-
mology. A belief may be false yet justified and a belief may be true but unjustified.
Being justified in epistemology, though understood in terms of promoting truth
1
I have presented previous versions under different titles to audiences at Pomona College, Universidad
Nacional Autonoma de Mexico, National University of Singapore, and the University of Aberdeen. I recall
stimulating and rewarding discussions. I recall comments that led to improvements from Juan Comesana,
Richard Fumerton, Claudia Lorena Garcia, James Jhun, Jinho Kang, Hilary Kornblith, Peter Kung, David
Sosa, Ernest Sosa, and Crispin Wright. I am sure I have forgotten just as many, if not more. To those I have
forgotten, I apologize. For comments on the penultimate draft that led to substantial improvements, I am
grateful to the referees and to Zachary Bachman, Meredith McFadden, and Megan Stotts. I am intellectually
and professionally indebted to Alvin Goldman and Ernest Sosa for their landmark work in epistemology
and for their continual encouragement and advice.
84 Peter J. Graham
and avoiding error, is not the same as being true. Truth is one standard; justification
is another.
Many traditional epistemologists connect justified belief to the individual’s ability to
justify her belief. For an individual’s belief to be justified, the individual must be able
to refer to her evidence, reasons or grounds in favor of the belief. She must be able to
apply standards or norms to her beliefs; she must be able to use the standard or norm as
a guide in deliberation. According to these epistemologists, only individuals who have
the capacity to engage in critical reasoning have justified beliefs.2
This view has fallen on hard times. Not because being able to justify a belief isn’t a
way of being in the right vis-à-vis promoting truth and avoiding error. It has fallen
on hard times because it overly narrows the scope of justified beliefs. Small chil-
dren and many non-human animals have justified beliefs. But they lack the capacity
to critically reason in support of their beliefs. They cannot justify their beliefs. Even
mature, reflective humans have a hard time justifying most of what they believe. Most
epistemologists recognize that the property of a belief ’s being justified is much broader
in scope than the traditional view.3
I shall speak to this large audience of epistemologists who allow that non-human
animals, small children, and ordinary adults have justified beliefs, even though they
may be unable to justify them, or lack justifications (as reflectively accessible argu-
ments) in favor of their beliefs.4
II. The Correct Procedures Conception

Nearly everyone in my audience—“internalists” and “externalists” alike—believes the
following fact:
FACT ONE: On the one hand, beliefs based on perception, introspection, good reasoning and
memory are, for the most part, prima facie justified beliefs. These psychological capacities are
paradigm ways of coming to form justified beliefs. Relying on these capacities are all paradigm
ways of being in the right vis-à-vis promoting truth and avoiding error. Beliefs so formed and
maintained are correctly formed beliefs.
On the other hand, beliefs based on wishful thinking, emotional attachment, hasty general-
ization and random blows to the head are not justified beliefs. These psychological influences
on belief are paradigm ways of forming unjustified beliefs. Relying on these influences are all
2
For views that tie justifiedness to justifying, see Audi (2001), BonJour (2009), Foley (1993), Leite
(2004), and Smythies (2015). For further discussion see Graham (2012b: 138–41). For more on critical
reason, see Burge (1986).
3
For this reason it can prove useful to reserve ‘being justified’ for the narrower positive epistemic status
associated with critical reasoning, and use other words such as ‘warrant’ and ‘entitlement’ for the broader
positive epistemic status. I prefer this practice (see my 2011, 2012, partly following Burge 1993, 2003,
2013). However, since the authors I am engaging in this essay prefer to use ‘justification’ and ‘justifiedness’
for this broader status, I shall follow their use here.
4
I thereby also exclude social constructivists about justification such as Robert Brandom, Richard
Rorty, and Michael Williams.
against actual-world reliabilism 85
paradigm ways of being in the wrong vis-à-vis promoting true belief and avoiding error. Beliefs
so formed and maintained are not correctly formed.5
Generalizing, the idea is that justified beliefs are well-formed beliefs, beliefs that are
correct or proper responses to inputs to our cognitive systems, where perception, good
reasoning, and so on, are paradigm cases of correct or proper responses to inputs to
our cognitive systems, epistemically correct belief-forming procedures.
Here’s how Alvin Goldman put the idea:
Justifiedness seems to be a function of how a cognizer deals with his environmental input,
i.e. with the goodness or badness of the operations that register and transform the stimulation
that reaches him. (‘Deal with’, of course, does not mean purposeful action; nor is it restricted to
conscious activity.) A justified belief is, roughly speaking, one that results from cognitive oper-
ations that are, generally speaking, good or successful . . . A belief is justified iff it is ‘well-formed’
(1979: 12–14).
Here’s how John Pollock put the idea:

In arriving at our beliefs, epistemic agents follow various procedures. Some of these procedures
are epistemically praiseworthy . . . There is a procedural sense of epistemic justification according
to which a belief is epistemically justified iff it was arrived at or held on the basis of procedures
that are epistemically praiseworthy. (1999: 385)
[T]he justifiedness of a belief is determined by whether it was arrived at or is currently sus-
tained by “correct cognitive processes”. The view is that being justified in holding a belief con-
sists in conforming to epistemic norms, where the latter tell you “how to” acquire new beliefs
and reject old ones . . . being justified consists of “making the right moves.” (Pollock and Cruz
1999: 25)
Here’s how Richard Feldman put the idea:

Reliabilists, proper functionalists, evidentialists of various stripes, and others, all agree that
there is some notion of a proper response to information (or evidence or stimuli), and that the
paradigmatic epistemic evaluations [being rational, being justified, being warranted] are about
this. A belief is favorably evaluated [justified, in the right] when it is a proper response and
unfavorably evaluated when it is an improper response. (2008: 347)
Let us call this first fact the correct procedure conception of justified belief. This con-
ception is a “horizontal” or mind-to-mind conception of justified belief, for it locates
justifiedness in relations between inputs to cognition (stimuli, representations, experi-
ences, beliefs, understanding) and the outputs of cognition (further beliefs). It locates
5
The “externalist” Alvin Goldman writes: “[S]tandard perceptual processes, remembering, good rea-
soning and introspection [are] intuitively justification-conferring . . . [C]onfused reasoning, wishful think-
ing, . . . emotional attachment, mere hunch or guesswork, and hasty generalization [are not]” (1979: 9–10,
1986: 103–4, 2011). The “internalists” Pollock and Cruz write: “[T]here are a number of natural processes
that lead to belief formation. Among these are such ‘approved’ processes as vision, inductive reasoning,
deductive reasoning, memory, and also some ‘unapproved’ but equally natural processes such as wishful
thinking . . . [W]e do not accord it the same status as some other belief-forming processes like vision”
(1999: 126).
86 Peter J. Graham
justifiedness in relations between mental states (inputs to cognition) and other mental
states (outputs of cognition).
This point is often put in terms of epistemic norms. Epistemic norms are standards
of correct cognition. Though some epistemic norms guide cognition, in general they
do not. Epistemic norms as standards for correct cognition need not guide cognition
or be accessible to the individual. They are standards that apply to an individual’s cog-
nition even if the individual cannot recognize or represent the standard. They are more
like standards for a healthy heart than recipes for cooking chicken livers.6
Epistemic norms of correct cognition take an input–output form, where the input
consists in mental or other internal states of the individual and the output consists in
what the individual is permitted or obligated to believe. If the individual conforms to
the norm in her cognition (i.e., believes what she is permitted or obligated to believe on
the input) then her belief is prima facie justified.7
Psychologists investigate the actual input–output relations in cognition. On the
assumption that human perception, reasoning, and so on produced justified beliefs,
then in studying how we actually cognize, psychologists are discovering, inter alia,
how we ought to cognize.
III. The Reliabilist Programme

Nearly everyone in my audience believes the following as well:
FACT TWO: As a matter of fact, perception, introspection, good reasoning, and memory are,
for the most part, reliable belief-forming capacities. They have a tendency to produce more
true beliefs than false beliefs. These capacities are good routes to truth.
6
Goldman: “[The justification-rules] I shall be discussing should not be understood as rules for guiding a
cognizer’s intellect. A person need not even understand the rules, and if he does, he need not be able to apply
them in the process of belief formation” (1986: 59). Pollock and Cruz: “If [belief-formation] is governed by
epistemic norms, just how is it governed? There is a model of this regulative process that is often implicit in
epistemological thinking, but when we make it explicit it is obviously wrong. The model assimilates the
functioning of epistemic norms to the functioning of explicitly articulated norms. . . . This “intellectualist”
model is almost always wrong. [Knowing how to form beliefs in approved ways] is a form of procedural
knowledge. . . . Our epistemic norms are just the norms that describe this procedural knowledge, and [justi-
fied] cognition is cognition in compliance with the norms” (1999: 125–9). Burge: “Natural” epistemic norms
“for perception, deductive reasoning, perceptual belief, primitive agency, and agency guided by perception or
perceptual belief do not depend on any individual’s setting, appreciating, or acceding to the norms. Such
norms do not depend on intention, convention, or rational agency. The norms apply whether or not anyone
recognizes them” (2010: 315).
7
Pollock and Cruz (1999: 122–43). Reliabilists agree. This is the whole idea behind Goldman’s “rule
framework” from Epistemology and Cognition (1986: chapters 4 and 5). Witness also Burge: “I believe that
entitlements always make reference to the way an individual’s cognitive (or practical) psychological com-
petencies operate. So they are partly internal. They are never merely statements that the individual’s beliefs
are produced by just any old process that is reliable in producing true beliefs. They are certain norms gov-
erning operation of the individual’s internal states. For example, the norms may make reference to the
belief ’s dependence on a reliably veridical type of perceptual representation” (Burge 2013: 367; cf. Burge
2010: 49ff). See also Graham (2010a and 2012b: 142–5).
Wishful thinking, emotional attachment, hasty generalization, and random blows to the
head, on the other hand, are not reliable for the most part. They do not have a tendency to pro-
duce more true beliefs than false beliefs. These influences on belief make a poor route to truth.
My audience disagrees over the significance of this fact.

Reliabilists believe this second fact explains the first. Reliabilists hold that the “good”
psychological processes—the good mind-to-mind transitions—are epistemically cor-
rect because they are good routes to truth. Being based on a good route to truth is the
standard—the criterion of correctness—that a justified belief meets; to be in the right
in the way of belief is to be based on a reliable belief-forming psychological capacity.
Justification is understood in terms of promoting true belief and avoiding error.8 In
canonical form:
SIMPLE RELIABILISM: In all possible worlds W, a belief is prima facie justified in W if and
only if (to the extent that) the psychological process that caused or sustained the belief reliably
produces true beliefs in W.
And by “explain” the reliabilist sets out to really, genuinely explain. In Goldman’s words:
What I have in mind is an explanatory theory, one that explains in a general way why certain
beliefs are . . . justified and others . . . unjustified . . . I want a set of substantive conditions . . . I seek
an explanatory theory, i.e. one that clarifies the underlying source of justificational status. [I]t is
not enough for a theory to state ‘correct’ necessary and sufficient conditions. Its conditions must
be appropriately deep or revelatory. (1979: 1–2, emphasis added)
Simple reliabilism is a “vertical” mind-to-world account of justification.9 Beliefs are

justified because the psychological capacity (the mind) stands in the relation of reliably
representing its subject matter (which is, for the most part, the world). Simple reliabil-
ists believe that perception, reasoning and so on are good ways of forming beliefs
because they reliably form true beliefs. Epistemic norms are true because conforming
to the norms results in reliably true beliefs. The correct mind-to-mind procedures are
those that result in reliably true outcomes. Justifiedness consists in both mind-to-mind
and mind-to-world relations, where correct mind–mind relations are made correct
8
See Goldman (1979, 1986: 103–4, 2011). Here’s how John Greco puts the idea: “Generic reliabilism is
a powerful view. For one, it accounts for a wide range of our pre-theoretical intuitions regarding which
beliefs have epistemic justification. Thus reliabilism explains why beliefs caused by perception, memory,
introspection, logical intuition, and sound reasoning are epistemically justified, and it explains why beliefs
caused by hallucination, wishful thinking, hasty generalization, and other unreliable processes are not”
(2005: 290, emphasis added).
9
For the use of the phrases ‘horizontal’ and ‘vertical,’ see Burge (2010: 50–1). I have discussed this con-
trast between mind–mind and mind–world conceptions in earlier papers (Graham 2009, 2010a, 2012a,
2012b). I earlier contrasted “actual-result” (mind–world) conceptions with “proper-aim” (mind–mind)
conceptions, partly inspired by Robert Audi’s (1988) contrast between “ontological” (mind–world) and
“teleological” (mind–mind) conceptions. Moral philosophers may find Audi’s labels misleading, as ‘teleo-
logical’ connotes consequentialist notions in ethics, and so might be misread as connoting reliabilism in
epistemology. That would not be Audi’s intent. Goldberg (2015) identifies “the” internalism–externalism
debate as asking whether a mind–world relation is required as well as a mind–mind relation for a complete
account of justifiedness. “Externalists” say yes but “internalists” say no.
88 Peter J. Graham
because of mind–world relations; the mind’s relations to the world constitutes the
correctness of its psychological procedures.10
IV. Cognitive Essentialism

Most non-reliabilists allow that the reliability of the process in the world of use matters
to knowledge; our second fact surely matters to knowledge. But many epistemologists
deny that reliability matters to justifiedness. Many readers do not agree that correct
procedures are correct because they produce reliably true outcomes. There are two
standard counterexamples that purport to sever the connection between correct pro-
cedures and reliably true outcomes: the clairvoyance case and the brain-in-a-vat (BIV)
case.11 I shall focus in what follows on a metaphysically possible version of the BIV or
“demon world” case.12
BRAIN-IN-A-VAT. Imagine Dennis1. Dennis1 is an ordinary human adult, with normally
functioning perceptual and cognitive capacities. Dennis1 forms reliably true perceptual beliefs
and other beliefs about his environment.
Then imagine a twin duplicate of Dennis1, Dennis2 in another possible world. Imagine that
Dennis2’s sensory systems are hooked up to a massive super-computer that continually induces
perceptual representations type identical to Dennis1’s. However, Dennis2’s perceptual rep-
resentations are massively in error; the super-computer causes Dennis2 to be massively deceived
about his external surrounds. In Dennis2’s possible world, human perception is massively unre-
liable. Dennis2 is completely unaware of the deception. He continues to form exactly the same
type of perceptual beliefs Dennis1 forms.
Dennis1 forms reliably true beliefs about his environment on the basis of his perceptual
capacities; human perception is reliable in his world. Dennis2 forms massively false beliefs
about his environment on the basis of the very same perceptual capacities; those capacities are
massively unreliable in his world.
Dennis1 and Dennis2 use the same capacities. Intuitively they both arrive at justified
beliefs. Intuitively Dennis2’s perceptual beliefs and beliefs derived from reasoning are
just as justified as Dennis1’s—he is conforming to the correct epistemic procedures in
cognition—even if his beliefs are no more reliably true than wishful thinking or sheer
guesswork. Reliability in the world of use does not seem necessary for justifiedness.
10
Pollock calls this view “norm externalism.” However, he sees norm externalism as committed to the
view that epistemic norms are revisable because contingent. He then purports to “refute” externalism by
arguing that epistemic norms are necessary (Pollock and Cruz 1999: 137–43). But this is not a necessary
piece of the norm externalist package. As we will see, the norm externalist (even Goldman) can (and does)
agree that (basic, primary) epistemic norms are necessary.
11
BonJour (1980), Cohen (1984), Feldman (1985), Foley (1985), Lehrer (2000).
12
Call any case where a belief-forming capacity that is intuitively reliable and justification conferring in
one world but massively in error but still intuitively justification conferring in a “demon-world” case. I do
not believe the classic Cartesian case of a lone disembodied spirit and a powerful malevolent demon is
metaphysically possible, and I do not believe Descartes thought so either. I am enough of a materialist to
require a functioning brain or equivalent for cognition.
This case and the clairvoyance case have led many epistemologists to reject reliabi-
lism as a theory of justified belief. Beginning in the 1980s, reliabilism has been
rejected by Robert Audi, Laurence BonJour, Stewart Cohen, Earl Conee, Joseph
Cruz, Catherine Elgin, Richard Feldman, Richard Foley, Susan Haack, Michael
Huemer, Keith Lehrer, John Pollock, Michael Slote, Matthias Steup, Jonathan Vogel,
Ralph Wedgwood, and many, many others on the basis of either or both of these two
cases. Nearly all of these philosophers believe that justification consists entirely in
mind-to-mind relations. Reliably true outcomes do not make epistemically correct
procedures correct.13
A number of epistemologists even hold that the BIV case establishes a stronger
point. They believe that it is not just contingently true that human perception and rea-
soning produce justified beliefs, but it is necessarily true that they do, that it is part of
the nature or essence of these psychological capacities to produce justified beliefs.
Witness John Pollock and Joseph Cruz:
COGNITIVE ESSENTIALISM. Internalist theories are committed to the principle that the
correctness of an epistemic move (a cognitive process) is an inherent feature of it. For example,
it may be claimed that reasoning in accordance with modus ponens is always correct, whereas
arriving at beliefs based on wishful thinking is always incorrect. This is implied by the claim
that the justifiability of a belief is a function of one’s internal states, because what that means is
that we can vary everything about the situation other than the internal states without affecting
which beliefs are justifiable. In particular, varying contingent properties of the cognitive pro-
cesses themselves will not affect whether a belief is justified. This is called cognitive essentialism.
According to cognitive essentialism, the epistemic correctness of a cognitive process is an
essential feature of that process and is not affected by contingent facts such as the reliability of
the process in the actual world. (1999: 25)
Goldman agrees:
[T]here is a right system of epistemic norms . . . The system that is right in the actual world is
right in all possible worlds . . . Positive judgments of justification about demon-world cases sup-
port the idea that norm-rightness may be rigidified rather than be allowed to vary across
worlds. (2011)
Epistemic norms are then necessary truths about the nature of the psychological
capacity. Justifiedness is constitutively associated with the nature of the psychology.
13
Pollock and Cruz, citing the “demon” world and clairvoyance cases, conclude: “[R]eliability has noth-
ing to do with epistemic justification. . . . [B]eliefs are justified because the believer is “reasoning correctly”
(in a broad sense of “reasoning”). If one makes all the right epistemic moves, then one is justified regardless
of whether her belief is false or nature conspires to make such reasoning unreliable” (1999: 113–14).
Correct procedures are not made correct by reliably true outcomes. Sosa comments on this idea: “A further
issue is now pressed by internalist epistemologists. Is there not some . . . notion of epistemic justifica-
tion. . . . However it may relate to knowledge . . . According to which the victim of the demon would still be
justified. . . . Sometimes internalist epistemologists even claim that traditional issues of epistemology con-
cern only such rational justification. . . . It is this kind of justification that, according to internalists, you
might still have even if unlucky circumstances make your animal mechanisms unreliable” (2003: 160–1).
Sosa sets out to capture this kind of justification in reliabilist terms.
90 Peter J. Graham
But if the reliability of the capacity is a contingent feature, how could reliably getting
things right explain why the psychology necessarily confers justified beliefs? The BIV
case seems to conclusively show that what makes correct procedures correct cannot be
the reliability of the procedure.
V. Special-worlds Reliabilism
Leading reliabilists like Alvin Goldman and Ernest Sosa are well aware of these cases
and remain undeterred. Goldman and Sosa believe they can accommodate these cases.
So how can a reliabilist accommodate the “rigidity” of epistemic norms, given the
“flaccidity” of reliability in the circumstances of use? Confronted with the scenario, the
idea is to modify the theory so that the victim’s perceptual beliefs turned out justified
after all. How does one do that?
Here is the idea. Instead of explaining the correctness of procedures in terms of
reliability in the world of use, explain the correctness of procedures in terms of reliably
true outcomes in a special set of worlds. If the capacity is reliable in the special set, then
the belief-forming process is correct in all possible worlds; if the capacity is not reliable
in the special set, then the process is always incorrect. Necessary epistemic norms thus
turn on reliability in a special set of worlds.
SPECIAL-WORLDS RELIABILISM: In all possible worlds W, a belief is prima facie justified in
W if and only if (to the extent that) the psychological process that caused or sustained the belief
reliably produces true beliefs in special worlds.
If Dennis2’s perceptual capacities are reliable in special worlds SW, then his beliefs even
while envatted and massively deceived are still justified, for they are justified in all
worlds because they are reliable in special worlds SW.
Reliability theories of justifiedness will supposedly not run afoul of cognitive essen-
tialism provided they identify a kind of reliability that is a necessary property of the
process in question. Find the right kind and the reliabilist is free to explain what makes
epistemically correct procedures correct in terms of reliably true outcomes. The chal-
lenge, however, is to find the right kind.14
VI. Normal-worlds Reliabilism

Goldman’s first major attempt was the normal-worlds theory (1986, 1988).15 It shall
prove illustrative to see why it failed.
14
Tyler Burge (2003), Sandy Goldberg (2012), David Henderson and Terry Horgan (2011), and Jack
Lyons (2013) provide alternative reliabilist solutions to the demon-world problem. I provide another
(Graham 2012a). I discuss Burge in Graham (forthcoming), Henderson and Horgan in Graham (2014a) and
Lyons in Graham (2011b, 2014c). I intend to further discuss these alternative approaches in future work.
15
In “What is Justified Belief?” he tentatively suggested relativizing justifiedness to reliability in natural
environments or to the actual-world (1979: 16–17).
Here is how Goldman defines normal worlds:

We have a large set of common beliefs about the actual world: general beliefs about the sorts of
objects, events, and changes that occur in it. We have beliefs about the kinds of things that,
realistically, do and can happen. Our beliefs on this score generate what I shall call the set of
normal worlds. These are worlds consistent with our general beliefs about the actual world.
(1986: 107)
We might as well call them doxastic-general worlds, for they are generated by our beliefs
about the general features actual world. These worlds are relative (relative to what we
believe) and subjective (they are determined by what we believe). We can now state the
theory:
NORMAL-WORLDS RELIABILISM. In all possible worlds W, a belief is prima facie justified
in W if and only if (to the extent that) the psychological process that caused or sustained the
belief reliably produces true beliefs in normal worlds.
Normal-worlds reliabilism apparently avoids the counterexample:

Now an evil demon world is a paradigm case of a non-normal world. So it does not matter that
the processes in question are highly unreliable in that world. It only matters whether they are
reliable in normal worlds, and that is apparently the case. (1986: 113)16
Despite that desirable outcome, Goldman soon gave up on the theory entirely. In
“Strong and Weak Justification” (1988) he listed the following three problems. The
fourth is from Pollock and Cruz (1999: 115). I claim credit for the fifth. Many, I am
sure, have asserted the sixth.
1. Which general beliefs count for determining normal worlds? There seem to be
too many choices.
2. Whichever ones we select, it looks like dramatically different worlds might fall
in the class of normal worlds. Does justification turn on reliability in all of these
worlds? Is any process even a candidate for reliability in all of these worlds?
3. Who is the “we”? All humans ever? Does the referent change over time? Does it
mean a special subset?
4. The theory puts no constraints on how we get our general beliefs. What if the
beliefs are unjustified? Should justification turn on crazy or wild beliefs? Do
normal worlds involve wizards and witchcraft?
5. What if the beliefs include (hidden, unnoticed) contradictions? Surely
everything we’ve ever believed about the general features of the actual world
16
Megan Stotts pointed out to me that Goldman’s claim here can be difficult to evaluate. It is true that a
world where only a malevolent demon and a massively deceived disembodied spirit exist is a paradigm case
of a non-normal world. But it is also a paradigm case of an imaginable world that is not a metaphysically
possible world, so it is hardly a counterexample to anything. The metaphysically possible “demon” world case
with a massive super-computer causing misleading stimulations on a normally functioning human percep-
tual system (e.g. Dennis2’s case), though possible, does not clearly fall outside of the scope of Goldman’s
“normal-worlds.” There could be a world like that within the range of Goldman’s “normal-worlds.”
92 Peter J. Graham
can’t be consistent. Does that mean there are no normal-worlds? Does that
mean no belief is ever justified?
6. Should justification turn on beliefs at all? Why relativize justification to what
“we” believe? Isn’t that too subjective, too non-realist, to fall within the spirit of
reliabilism? Why should what we believe determine what beliefs are really
justified?
The “normal-worlds” maneuver fails to provide a good explanation of correctness of
epistemic procedures. We’re at a loss to understand why correctness in all worlds
should turn on reliability in “doxastic-general” worlds. Even if the theory provides
“correct” necessary and sufficient conditions (which seems impossible to evaluate), it
clearly does not “clarify” the “underlying source” of justificational status.
Even worse, the theory fails to provide “correct” necessary conditions. There’s a clear
counterexample imagined by Stewart Cohen that Goldman reported as follows:
ALIEN. Finally, even if all of these problems could be resolved, it isn’t clear that the nor-
mal-worlds approach gets things right. Consider a possible non-normal world W, significantly
different from ours. In W people commonly form beliefs by a process that has a very high
truth-ratio in W, but would not have a high truth-ratio in normal worlds. Couldn’t the beliefs
formed by the process in W count as justified?
To be concrete, let the process be that of forming beliefs in accord with feelings of clairvoy-
ance. Such a process presumably does not have a high truth-ratio in the actual world; nor
would it have a high truth-ratio in normal worlds. But suppose W contains clairvoyance waves
analogous to sound or light waves. By means of clairvoyance waves people in W accurately
detect features of their environments just as we detect features of our environment by light and
sound. Surely, the clairvoyance belief-forming processes of people in W can yield justified
beliefs (1988: 62).17
Couldn’t there be another kind of psychological capacity in another kind of cognitive

being that is among the right ways of forming beliefs? Couldn’t relying on that capac-
ity be among the epistemically correct procedures—correct in all possible worlds—
even if not reliable in “doxastic” normal worlds? Why restrict correct procedures to
procedures reliable in normal worlds? Goldman concluded: “For all the foregoing
reasons, it seems wise to abandon the normal-worlds version of reliabilism” (1988: 62).
Simply relativizing to worlds won’t do the trick; you have to relativize to the right set of
worlds.
VII. Actual-world Reliabilism

If the normal-worlds version does not work, does another? Alvin Goldman, Ernest
Sosa, and Juan Comesana have each advanced versions of actual-world reliabilism.
17
Compare Sosa’s “extraterrestrial mechanisms of belief-formation” (1980: 188, 2001: 390). See also
Graham (2012: section 5.4 “Aliens”).
ACTUAL-WORLD RELIABILISM: In all possible worlds W, a belief is prima facie justified in

W if and only if (to the extent that) the psychological process that caused or sustained the belief
reliably produces true beliefs in the actual world.
Epistemically correct procedures are then correct in all possible worlds because they
are correct in the actual world; reliable outcomes in the actual world makes correct
procedures correct in all worlds. Here is Goldman:
First, there is a right system of epistemic norms or principles, norms that govern which
belief-forming processes [are correct]. These norms are grounded in considerations of reliabil-
ity . . . The right set of norms is “made” by the true facts of reliability pertaining to our cognitive
processes and the actual world . . . a belief is really justified iff it is arrived (or maintained) in
conformity with the right set of norms or principles . . . The system that is right in the actual
world is right in all possible worlds. (Goldman 2011)
Goldman, Sosa, Comesana, and many others think this is enough to avoid the demon-
world case. Witness Goldman:
A belief is “really” justified if (and only if) it meets a correct standard, where a correct standard
specifies a process that is genuinely reliable in the actual world. Rigid use of a correct standard
would render perceptual beliefs in demon worlds “really” justified. (2002: 49)
Witness Sosa:
If one were a demon’s victim, one’s intellectual performance might still be adroit (adroit-justified).
That is to say, in a world W in which one was a victim, one’s beliefs acquired through our nor-
mal perceptual faculties would come out of processes that in our actual world are reliable, and
hence those beliefs of ours in that world would be adroit (adroit-justified). (2001: 391–2)
Will the actual-world theory save the day where the normal-worlds theory failed? I
have grave doubts.
VIII. Absolutist Actual-world Reliabilism

There are two views of what counts as the “actual” world, the “absolutist” and the
“indexicalist” conception. And so there are at least two versions of actual-world relia-
bilism, “absolutist” actual-world reliablism and “indexicalist” actual-world reliabilism.
I shall discuss them in turn.
On the “absolutist” view, among all the possible worlds there is but one and only one
actual world; being actual is an absolute, non-perspectival feature of just one world.
Draw a number of circles in the margin and label one the actual world. Once you
have done that, all of the other circles represent non-actual, possible worlds, where in
no sense will any of those other possible worlds ever be the actual world. The actual
world is, as it were, “intrinsically” actual, and all other possible worlds are, as it were,
“intrinsically” non-actual. On the “absolutist” view, the word ‘actual’ is like a name or a
predicate that applies to one and just one world, the actual world. Being actual is never
94 Peter J. Graham
a “perspectival” or “varying” feature of worlds. Being actual is a “real” and “non-varying”

feature of just one world.
Goldman advocates the absolutist variant of actual-world reliabilism (2001: 467,
2002: 49, 2011). According to Goldman, in all possible worlds, a belief is prima facie
justified iff based on a psychological process that is reliable in the actual world. Reliably
true outcomes in the actual world then set the standards for correctness in all possible
worlds.
This view faces a clear counterexample, the aliens counterexample just raised to
Goldman’s normal-worlds theory. It is just as effective against the “absolutist” actual-
world reliability theory, for the process we imagined was a possible, non-actual process.
And so it is not reliable in the actual world, for it does not exist in the actual world. And
were it to exist in the actual world, so to speak, it would not be reliable. Hence their beliefs
are ruled non-justified by the absolutist actual-world reliability theory; their belief-form-
ing procedures are ruled epistemically incorrect in all possible worlds, and so in their
world. This is counterintuitive. Surely there are possible, non-actual belief-forming pro-
cesses (that are as reliable as you please) in possible, non-actual worlds that confer
justification on the beliefs they cause and sustain. Surely there are possible but non-
actual epistemically correct belief-forming procedures. Surely the scope of epistemi-
cally correct processes transcends the actual processes. Our actual procedures may be
correct, but not all correct procedures are actual. This view is clearly mistaken. It is just
as “wise to abandon” the absolutist actual-world reliability theory as it was to abandon
the normal-worlds theory.
IX. Indexicalist Actual-world Reliabilism

I now turn to the indexicalist variant. Perhaps it will do better.
On the indexicalist view of “actuality,” the word ‘actual’ is an indexical. The words
‘now’ and ‘here’ are paradigm indexicals. When I utter ‘I am here’ or think I am here
I am referring to the location where I am. When you utter ‘I am here’ or think I am here
you are referring to the location where you are. That’s how indexicals work.
On the indexicalist view of “actuality,” being the actual world is like being here. On
the indexical view, ‘actual’ refers to the world of utterance or thought. When Dennis1 in
his world says ‘I am in the actual world’ or thinks I am in the actual world he refers to
the possible world that he is in, and what he says and thinks is true. It’s like saying or
thinking I am here. And when Dennis2 in his distinct possible world says ‘I am in the
actual world’ or thinks I am in the actual world he refers to the possible world that he is
in—a world different from Dennis1’s world—and what he says and thinks is true too.
Again, it’s like saying or thinking I am here.
And so when Dennis1 in his world says ‘Dennis2 is in a possible world, not the actual
world’ what Dennis1 says or thinks is true. It’s as if Dennis1 said ‘I am here and Dennis2
is there’ which is clearly true. So when Dennis2 says ‘Dennis1 is in a possible world, not
the actual world’ what Dennis2 says is also true. He might as well have said ‘He is there
and I am here.’
If you drew those circles in the margins when reading the previous section, then to
get your mind around the indexicalist view, draw a stick figure in one circle uttering
‘actual.’ That circle is then the “actual” world (that circle is now “here”) from the point
of view (the world) of the utterance. All the other circles are then possible, non-actual
worlds. But then draw a stick figure in a different circle uttering ‘actual.’ That circle is
then the “actual” world (that circle is now “here”) from the point of view (the world) of
the utterance. All the other circles are then possible, non-actual worlds. No world is
then, as it were, “intrinsically” actual, just as no location is, as it were, “instrinsically”
here. Being actual, like being here, is then a varying, “perspectival” feature of worlds; it
all depends upon where the utterer stands.
On the indexicalist variant of actual-world reliabilism, in all possible worlds a belief
is prima facie justified iff based on a psychological process that is reliable in the actual
world, where ‘actual’ is an indexical, so “the” actual world depends on the world of utter-
ance. As we will see, this variant avoids the counterexample to the absolutist variant,
and for that reason is initially attractive. Maybe it is exactly what we’re looking for.
Ernest Sosa (2001, 2003) and Juan Comesana (2002) have both advanced this variant.
I shall focus on Sosa’s discussion. For a period leading up to Sosa’s AAA “perfor-
mance-theoretic” analysis of knowledge, Sosa used the phrase ‘apt-justification’ to refer
to the reliability of the belief-forming process in the world of use and ‘adroit-justification’
to refer to the reliability of the belief-forming process in the actual world. ‘Adroit-
justification’ is intended to follow our use of ‘justifiedness’ in the present paper.
To get your mind around Sosa’s proposal, consider first the “demon” (BIV) world
where human perception is not reliable. Since human perception is not reliable there,
perceptual beliefs there are not apt-justified there. However, perceptual beliefs there
are still adroit-justified there, because human perception is reliable here, in our actual
world (2001: 391–2, 2003: 160–1). Sosa thus claims to “solve” the “demon” world case.
So far, so good. But what about the counterexample to Goldman’s variant? How does
the indexicalist variant avoid that counterexample? Easy. Consider the aliens in their
possible world. When one of the aliens utters or thinks ‘I am in the actual world’ he or
she refers to their world. So when one of the aliens utters or thinks ‘our psychological
capacities are reliable in the actual world’ he or she is referring to their “actual” world.
Since by hypothesis their psychological capacities are reliable in their world, their
capacities reliably produce true beliefs in the actual world (their “actual” world). And
so according to indexicalist actual-world reliabilism, their beliefs are adroit-justified.
Counterexample diffused (Sosa 2001: 390, 400).
We can even imagine a “demon” case for the aliens. Imagine a world where the aliens
are hooked up to a massive super-computer that induces massive error, unbeknownst
to them. Are their beliefs adroit-justified on the indexicalist theory? Indeed they
are, for those psychological capacities are reliable in their “actual” world, and so
96 Peter J. Graham
adroit-justified in all possible worlds, and so adroit-justified in this “demon” BIV world.
Pretty clever.
All, however, is not well. I shall argue that the indexicalist variant is implausibly
“perspectival” and “promiscuous.” It causes more problems than it solves.
Here’s why it is implausibly “perspectival.” Ask yourself this: is the alien psychology
reliable in the actual world? Well, if you are thinking that thought then, on the view,
you are referring to your “actual” world. In which case the answer is “no” for the psy-
chology does not exist in your “actual” world, and were it to exist, so to speak, it would
not be reliable in your “actual” world. Hence, on the indexicalist variant, the alien’s
psychology is an epistemically incorrect procedure in all possible worlds; the beliefs of
the aliens are not adroit-justified in any possible world. (It’s like saying “because the
process is not reliable here, it does not confer justification anywhere.”) But, as we’ve just
seen, the exact opposite is true if we start from the alien’s utterances or thoughts in their
“actual” world. (It’s as if they’ve said, “because the process is reliable here, it confers
justification everywhere.”) So from our perspective their beliefs are not adroit-justified
in any possible world (and so not in their world), but from their perspective their
beliefs are adroit-justified in their world (and so in all possible worlds). So their belief-
forming procedures confer justification in no worlds and in all worlds at the same time.
We don’t even need the alien psychology to reveal the implausibly perspectival
nature of the view. Take our two Dennises once more. Dennis1’s belief-forming capaci-
ties are reliable in our (and his) “actual” world. According to indexicalist actual-world
reliabilism, beliefs based on those types of belief-forming capacities are thus adroit-
justified in all possible worlds, and so in Dennis1’s world and Dennis2’s world. (This is
how Sosa “solves” the BIV case.)
Dennis2’s type-identical belief-forming capacities, on the other hand, are not relia-
ble in his (not our) “actual” world (he’s being massively fooled in his “actual” world by
the super-computer). According to indexicalist actual-world reliabilism, beliefs based
on those types of belief-forming capacities are thus not adroit-justified in all possible
worlds, and so not in Dennis2’s or in Dennis1’s world. The cases are symmetrical; going
“indexical” guarantees that nothing privileges one world over another; each have equal
right to determine what is true for all possible worlds.
So from Dennis1’s perspective his belief-forming procedures are correct in all possible
worlds (and so from his perspective his beliefs and Dennis2’s beliefs are adroit-
justified), but from Dennis2’s perspective his belief-forming procedures are incorrect
in all possible worlds (and so from his perspective his beliefs and Dennis1’s are not
adroit-justified). But their psychologies are type-identical. So their belief-forming
procedures confer justification in no worlds and in all worlds at the same time.
The source of the trouble arises from the built-in “symmetry” in the indexicalist
account. Since ‘actual’ is an indexical, all possible worlds are on a par; every world can,
as it were, be “actual” (just as every location can be, as it were, “here”). Hence what
happens in each possible world has just as much right to set the standard for epistemic
correctness in every possible world as any other world. The indexicalist variant has a
built in symmetry while the absolutist variant has a built-in asymmetry. The asymme-
try invites the alien counterexample while the symmetry avoids it. But the symmetry
invites ever more troubles.18
The indexicalist actual-world reliability theory clearly generates a bizarre, implausi-
ble result. It generates the counterintuitive consequence that the aliens’ psychological
procedures are epistemically correct in all worlds and no worlds at the same time. And
what applies to the aliens applies equally well to us. Going “indexical” may seem attrac-
tive for it allows one to initially dodge the aliens counterexample, but going “indexical”
and then rigidifying across all possible worlds makes matters even worse.19
But that’s not all. The indexicalist variant is not just implausibly “perspectival,”
it generates an implausible promiscuity of degrees of adroit-justification. Adroit-
justification arises from reliability in “your” “actual” world. And it is rigid; the process
in question has it in all possible worlds. But if there is a plurality of worlds, then there is
a plurality of degrees or kinds of adroit-justification. For adroit-justification is reliabil-
ity in the “actual” world. Since there is a plurality of worlds, where each is “actual”
(from its own world), there is a plurality of degrees or kinds of adroit-justification, one
for each world, for a psychological process P will be reliable to various degrees or kinds
in different worlds. So a structure may be extremely reliable in W1, very reliable in W2,
somewhat reliable in W3, barely reliable in W4, unreliable in W5, very unreliable in
W6, and so on. In each world we will end up constructing a different degree of
adroit-justification. High degrees, low degrees, zero degrees, and so on. So if there is a
plurality of possible worlds and so a plurality of degrees or kinds of adroit-justification,
then beliefs produced by P will have all of these different degrees of adroit-justification
at one and the same time. Going “indexical” gets it all wrong.
18
Sosa raised a problem for his view that, once solved, walks right into the problem I am raising in the
text. If justification amounts to reliability in “our” world, then when the victim of the brain-in-a-vat sce-
nario thinks to herself that her perceptual beliefs are justified, she must be thinking that her beliefs are
reliably formed in our world, but not in her world (Sosa 2001: 397–8). But why should she be thinking of
our world and not hers? After all, she may have no idea about us at all!
Sosa’s reply is that, on his theory, she isn’t thinking of our world at all. Rather she is really thinking of her
world. The objection failed to track the shift in the use of the indexical “our.” She is thinking that her beliefs
are justified because they are reliable in “our” world, where she utters “our” and so refers to herself and her
conspecifics, and not to us. ‘Actual’, Sosa says, is an indexical; it shifts its reference from world to world.
That may avoid that problem, but it walks right into the problem raised in the text: when she thinks to
herself that her belief is adroit-justified, she would be mistaken, for her belief is not reliably formed in her
world. But if not reliably formed in her world, then the process is not adroit-justified in all possible worlds,
even hers.
19
Have I been unfair to the indexicalist? In attributing an apparent contradiction, have I unfairly attrib-
uted to the indexicalist the realist view that reliability in one world generates correctness in all worlds,
where the indexicalist holds the perspectivalist view that reliability in one world W1 generates adroitness-
vis-à-vis-W1 in all worlds, reliability in W2 generates adroitness-vis-à-vis-W2 in all worlds, and so on,
where each kind of adroitness is compatible with every other kind? That view might seem better to the
indexicalist, but not to me. For it creates a possibly infinite number of perspectivalist “kinds” of justifica-
tion a belief-forming process will have, for once we rigidify it will have all of these “kinds” in all possible
worlds where it exists.
98 Peter J. Graham
X. The Explanatory Burden

I’m told the actual-world reliability theory is intuitive; that probably explains its popu-
larity. If you still find it intuitive, despite its counterintuitive consequences just discussed,
then I still have some work cut out for me. In this section and the next I want to remove
its intuitive plausibility. First I’ll argue that, even if we ignore the counterintuitive conse-
quences just raised, it doesn’t really explain what makes epistemically correct procedures
correct. Then in the next section I’ll argue that the theory derives all of its appeal from a
related (though rival) theory.
Remember first that our goal is to provide an explanatory, substantive theory that is
“appropriately deep or revelatory” that “clarifies the underlying source of justifica-
tional status” that does not simply provide “necessary and sufficient conditions” for
justifiedness. So even if the actual-world reliability theory avoids counterexamples
that may not be enough to meet the explanatory burden.
Remember second that our goal is to explain an “inherent,” essential property of
belief-forming procedures. Remember that we have accepted cognitive essentialism.
That means if we are to explain what makes certain mind–mind transitions epistemi-
cally correct—an “inherent” or essential property of those procedures—in terms of
certain mind–world relations then those mind–world relations must be “inherent” or
essential properties of the belief-forming capacities, or constitutively associated with
the nature or essence of the capacity.
Special-worlds reliabilism tries to meet these explanatory goals by “rigidifying”
epistemic correctness to mind–world relations in a special world or worlds. If the pro-
cess is reliable in that world or worlds, then it is epistemically correct in all possible
worlds. I shall argue that rigidity isn’t enough. The actual-world reliability theory
needs supplementation to discharge its explanatory burden.
To warm you up, let me remind you of a point that is gaining ever-wider acceptance:
rigidity is not enough for essence. Rigid properties are not always essential properties
(or properties constitutively associated with a nature or essence), even if essential
properties are rigid properties. Take any obviously contingent, non-essential or non-
constitutive property of anything. Take the color of this cell phone. Surely its color is a
contingent, non-essential property. I bought a black one, but they also came in red.
I think you can probably buy a transparent one in Japan if you really wanted one. Now
suppose someone said that its color is essential to it, part of its nature. No, we would
respond, for we can imagine a world where it is red, blue, or even transparent. Well, our
interlocutor responds, if we “actualize” being black, then it turns out that in all possible
worlds where the cell phone exists it has the property of being-black-in-the-actual-
world; it has the “actualized” property necessarily. You can take any contingent prop-
erty of a thing, or contingent relation between two properties (no matter how irrelevant
to the nature of the thing or the nature of the two properties), and rigidify or “actualize”
the property and thereby create a necessity. Given the trick, necessities are cheap;
you get at least one for every contingency. But clearly these cheap necessities aren’t
constitutively associated with anything. They are just “actualized” contingent properties.
“Actualizing” non-essential, non-constitutively associated properties doesn’t create an
essential, constitutive, or constitutively associated property.
So if our goal is to identify substantive explanatory conditions that are essential
to the natures of belief-forming processes, or constitutively associated with those
processes, identifying a contingent property and then rigidifying it is not enough to
identify such a property. Goldman and Sosa might have thought they had discharged
their explanatory burden when they (thought) they found “correct” necessary and
sufficient conditions for correctness in all possible worlds, but in fact they were just
getting started. Going “actual” to rigidify a property is a trick that will avoid counter-
examples, but it won’t, as such, provide an “appropriately deep or revelatory” theory of
anything. A trick of modal logic can generate strong supervenience without even being
in the ballpark for providing constitutive conditions for the nature or essence of a
property or thing.20
All of that was just to warm you up. You should now see why, technically speaking,
actual-world reliability theories still have some work to do. Since rigid properties are
not sufficient for explaining essences or natures, simply rigidifying isn’t yet to explain
essences or natures; you still need to explain why those rigid properties do what they
are supposed to do.
In the remainder of this section I will say why I find it intuitive that they don’t. Then
in the next section I’ll say why the appeal of the actual-world reliability theory actually
derives from a rival theory, a rival with a better shot at discharging its explanatory
burden.
Here’s why I don’t think the actual-world reliability theory explains what it is sup-
posed to explain. I’ll put my point rhetorically: why should epistemic correctness in all
worlds be constituted by, or constitutively associated with, reliably true outcomes in
the actual world (either the actual world or my “actual” world)? Why should reliability
in that world constitute correctness in all worlds? Why should the fact that the process
is a good route to truth in that world make it the correct procedure to follow in all
worlds, even in worlds where it is bound to lead one astray? Why should reliability
there constitute epistemic correctness everywhere? Has the actual-world reliabilist
even asked this question?
Remember that the original motivation behind simple reliabilism was that correct-
ness was tied to reliably promoting truth in the world of use; justifiedness is a good
route to truth: conditional on using the process, the belief produced has a good shot at
being true (justifiedness is not the same thing as truth, even so it is pretty close). Now
that idea can be initially very appealing. “Oh, I see! It’s because these procedures relia-
bly get us to the truth that using them is correct!” This idea can feel “deep or revelatory”
about the very nature of justifiedness. But once we confront the BIV and clairvoyance
20
This is a theme in Kit Fine’s work (e.g. Fine 1994).
100 Peter J. Graham
cases, we realize that justifiedness isn’t the same as being a good route to truth in the
world of use. “Shoot, I guess that idea didn’t work.”
The “actual world” gambit says to try again, but to relativize to the actual world. But
the initial idea that was so attractive doesn’t apply in this case. The maneuver won’t
seem initially as attractive. “Hmm, reliability in the actual world? Well, maybe, but
I don’t see it. Why should the ‘demon’ victim have justified beliefs because his proce-
dures produce true beliefs somewhere else? After all, in the vat they are all false.” Why
does the special world maneuver in general, and the actual-world variant in particular,
explain why the victim’s beliefs are justified? It is completely natural to find oneself
puzzled or unsure.
Since the obvious motivation of the actual-world reliability theory was to dodge
the counterexamples, its no surprise that it hasn’t won many (if any) converts from the
other side. For it doesn’t really explain to the philosophers on the other side why the
correctness of epistemic procedures should be partly constituted by reliably getting
things right in the actual world. For if the actual-world maneuver really explained
what makes a procedure correct, philosophers on the other side should have an “Oh
I see! I get it!” experience, for explanations are supposed to explain, especially deep
and revelatory explanations. But as far as I know, no one has had such an experience.
I know I haven’t. I don’t see why epistemic correctness in all worlds is constituted by
reliably true outcomes in the actual world.21
To really explain the actual-world theory needs considerable supplementation. But
given the counterexamples and counterintuitive consequences we discussed before, it
is not even a candidate explanation. Supplementation won’t be enough. We need to
supplant the theory, not supplement the theory.
XI. Moving Forward: Normal-conditions Reliabilism

Should the reliabilist be forced to throw in the towel? I know many so-called internalists,
mentalists, and evidentialists would like to think so. But I don’t think the fight is over yet.
In what little time we have remaining together, I shall make three suggestions.
First, I think the entire analytical tool of “worlds” should be replaced with “circum-
stances” or “conditions” instead. If the reliabilist is going to relativize correctness always
and everywhere, the reliabilist should relativize to special circumstances or conditions
instead of special worlds. Second, I think the reliabilist should switch from relativizing
correctness to reliability in the actual world to reliability in “normal” or “natural” cir-
cumstances or conditions; the right “special” circumstances are normal circumstances,
not necessarily actual circumstances. Third, to meet the explanatory burden, the
21
I have a hunch (urged on me by Peter Kung) that “being actual” never explains. It is always some other
property of the thing we want to explain that does the explaining. We explain why a glass fell by citing the
ball hitting the glass. We do not explain why a glass fell by citing the actuality of these properties; we simply
presuppose, in explaining them, that they are actual.
reliabilist should explain why epistemically correct procedures are reliable in normal
conditions as a part of their nature or essence, that what makes epistemically correct
procedures correct arises from the fact that they are reliable in normal conditions as a
part of their nature or essence, or that their nature or essence is constitutively associated
with reliability in normal conditions.
Since this third suggestion is clearly the topic for another paper, I won’t say anything
further about it here, except for a brief remark when closing. Since the first two sugges-
tions, however, are relevant to my overall case against actual-world reliabilism, a few
more words are clearly in order. And since the first two are really just two aspects of one
idea (the reliabilist should switch from reliability in the actual world to reliability in
normal circumstances) I will talk about both aspects at once.
What do we all already know about normal conditions? First, normal conditions or
circumstances are relative to kinds. What are normal circumstances for fish? Being
submerged in water. Take them out of the water, and they won’t last for long. What are
normal circumstances for humans? Certainly not being fully submerged in water, at
least not for very long. Deep inside the center of the Sun certainly does not count as
normal either.
Second, normal circumstances, so understood, are repeatable types of circum-
stances. A fish that has always lived in a lake would still be in normal conditions if we
moved it to another lake. A human that has always lived in Alaska would still be in
normal conditions if he or she moved to Nebraska. The same type of circumstance can
be instantiated in a variety of locales, even locales where one in never likely to go.
Third, “normal” here does not mean typical or average. Normal conditions are not
conditions typical in the Universe. There’s precious little water throughout the entirety
of the Universe. Of all the conditions in the Universe, very few may be normal. Nor are
the conditions that an animal typically occupies ipso facto normal. The majority of
rhesus monkeys may, unfortunately, find themselves in cages for laboratory experi-
ments. It happens.
Fourth, normal and abnormal environments transcend worlds. Fish in the actual
world are submerged in water. And in most possible worlds I can imagine, fish that
aren’t dead are submerged in water too. Normal conditions for fish exist in actual
and possible worlds. Normal conditions are not “world-bound.” So too abnormal
conditions.
Fifth, normal circumstances or conditions comprise a subset of all of the conditions
in a possible world. If being submerged in water comprise normal circumstances for
fish, then the actual world is awash with non-normal circumstances for fish, for the
actual world is a vast Universe, possibly infinite in time-and-space, where water is, as
I just said, a rare occurrence.
In sum, here is what we all already know: normal conditions are (1) relative to kinds,
(2) repeatable types of circumstances (3) not (necessarily) typical, (4) transcend
worlds, and (5) (typically) comprise but a subset of all of the different kinds of circum-
stances in a world.
102 Peter J. Graham
With this admittedly brief discussion of normal conditions, I think the way forward
for the reliabilist is to relativize correctness in all conditions to reliability in normal
conditions:
NORMAL-CONDITIONS RELIABILISM: In all possible circumstances C, a belief is prima
facie justified in C if and only if (to the extent that) the psychological process that caused or
sustained the belief reliably produces true beliefs in normal conditions.
The theory already has weighty advocates. Ruth Millikan (1984) defends such a view
for knowledge, appealing to Mother Nature. Alvin Plantinga (1993) defends such a
view for knowledge and justified belief, appealing to God. Tyler Burge (2003) defends
such a view for knowledge and justified belief, appealing to anti-individualism about
psychological states. Michael Bergmann (2006) defends such a view for justification,
but without an appeal to Mother Nature, God or anti-individualism.22
This theory differs from Goldman’s “normal-worlds” theory for it refers to conditions
or circumstances (aspects of worlds, habitats, or environments) instead of complete
possible worlds, and it differs to the extent that it explicates what counts as “normal”
differently. Burge, Millikan, and Plantinga have all done just that.23 Let us assume a
realist development, letting our intuitive grasp of normal conditions from the few
examples provided so far be our guide.
We can see why this is the path for every reliabilist to take for it avoids all of the
problems raised for the two variants of the actual-world reliability theory. First, it
avoids the alien counterexample to the absolutist variant, for even though the alien
psychological capacity is not reliable in the actual world, it is reliable in normal con-
ditions (conditions normal for the aliens). What happens or does not happen in the
actual world is then irrelevant to whether their psychologies are epistemically cor-
rect procedures of belief-formation. And here is another difference with Goldman’s
“normal-worlds” theory, for what happens in worlds that share the general features we
believe to hold true of the actual world has nothing to do with what makes conditions
normal for the aliens.
Second, taking this path avoids the objections we have raised to the indexicalist var-
iant. Human perceptual capacities reliably induce true beliefs in normal conditions for
humans. Thus on the view they confer prima facie justifiedness in all possible circum-
stances when functioning normally. Both Dennis1 and Dennis2 are relying on normally
functioning human belief-forming capacities that produce reliably true outcomes in
normal conditions for humans. Dennis1 is in normal conditions though Dennis2 is not.
Even so, both of their perceptual beliefs are prima facie justified. And that is the end of
22
I have also developed such a view, appealing to etiological functions. See my 2010b, 2011a, 2011b,
2012a, 2014b.
23
For example, see Millikan (1984) or Burge (2003). The general idea behind Millikan and Burge is that
normal conditions are those that are relevant to individuating and/or constituting the nature of the psycho-
logical process. For Plantinga’s divinely inspired approach, the idea is that normal conditions are those that
God intended for the use of the psychological process.
the story. There is no room in the theory, unlike the indexicalist theory, for generat-
ing the judgment that Dennis1’s perceptual beliefs are not adroit-justified because
Dennis2’s perceptual capacities are not reliable in his “actual” world. And there is no
room in the theory, unlike the indexicalist theory, for generating a plurality of degrees
of justifiedness, for what matters is reliability in normal conditions for the capacity, not
any possible conditions or circumstances where the capacity might exist.
The normal conditions path also captures what is plausible—what is initially intui-
tively attractive—about the actual-world reliability theory. The plausibility of that the-
ory derives from our two initial facts: that perception (etc.) is a correct procedure, and
that perception (etc.) is a reliable procedure in normal conditions. Since the conditions
we are actually in are normal conditions for our psychological capacities, it is natural
to confound normal conditions for actual conditions, and reliability in actual condi-
tions for reliability in the actual world. But since the metaphysical profile of the actual
world is so very different (as we have seen) from the metaphysical profile of normal
conditions, these are very different theories. But since prior to such metaphysical
reflections they can seem so similar, it is natural for one to fail to notice the difference,
and so natural to find the actual-world theory plausible when really it’s the normal
conditions theory that’s so appealing.
Indeed, I conjecture this is what drove Goldman and Sosa. I have little evidence that
this is so for Goldman, so it shall remain sheer conjecture. But there is plenty of evi-
dence that something like this is going on in Sosa’s mind. For even though the letter
of Sosa’s view is an actual-world view, the spirit of the view seems to be a normal-
conditions view, for his glosses frequently advert to normal conditions. For example:
What powers or abilities . . . enable a subject to achieve . . . justification? They are presumably
powers or abilities to distinguish the true from the false . . . One’s power or ability must . . . make
one such that, normally at least, in one’s ordinary habitat, or least in one’s ordinary circum-
stances when making such judgments, one would believe what is true and not believe what is
false. (1988: 151, emphasis added)
We are . . . a certain way by nature, a way that, given our normal environment, enables us to
attain truth and understanding on questions of interest . . . Our inbuilt mechanisms may still
operate correctly even if, unfortunately, we are in an abnormal environment relative to which
those very mechanisms distance us from both truth and understanding . . . This is a way we can
become justified by using our basic faculties . . . Even victims of a Cartesian-Evil Demon would
retain such justification. (2004: 308, emphasis added)
Failed attempts in abnormal circumstances do not show lack of ability . . . what is required is
only that your attempts tend to succeed when circumstances are normal. (2007: 84)
Talk of normal conditions and normal functioning fills Sosa’s writings on justification;
they just don’t enter into explicit statements of the theory. If the actual-world is the
letter of his theory, normal-conditions are the spirit.
This concludes my case against actual-world reliabilism and my gestures in favor of
normal-circumstances reliabilism. Reliabilists want to capture what makes epistemi-
cally correct procedures correct in all possible circumstances in terms of reliability in a
104 Peter J. Graham
special set of circumstances. I’ve argued against normal-worlds reliabilism and two
variants of a ctual-world reliabilism, my main target. I have suggested that normal-
circumstances reliabilism is the right way forward for the reliabilist. But for it to
succeed, it too has to discharge its explanatory burden; it must explain why correctness
in all circumstances should turn on reliably true outcomes in normal circumstances.
Can it do that? It can, I believe, if it constitutively interrelates the nature of the belief-
forming psychology in a substantive and explanatory way with reliably getting things
right in normal conditions, so that producing reliably true outcomes in normal condi-
tions is constitutively associated with the nature of the psychology. But discussing this
further is clearly a topic for another occasion.
References
Audi, R., 1988. “Justification, Truth and Reliability.” Philosophy and Phenomenological Research
49: 1–29.
Audi, R., 2001. “An Internalist Theory of Normative Grounds.” Philosophical Topics 29: 19–46.
Bergmann, M., 2006. Justification Without Awareness. Oxford: Oxford University Press.
BonJour, L., 1980. “Externalist Theories of Knowledge.” Midwest Studies in Philosophy V: 53–73.
BonJour, L., 2009. Epistemology, 2nd edn. Washington, DC: Rowman & Littlefield.
Burge, T., 1986. “Our Entitlement to Self-Knowledge.” Proceedings of the Aristotelian Society.
New Series 96: 91–116.
Burge, T., 1993. “Content Preservation” Philosophical Review 102: 457–88.
Burge, T., 2003. “Perceptual Entitlement.” Philosophy and Phenomenological Research 67:
503–48.
Burge, T., 2010. Origins of Objectivity. Oxford: Oxford University Press.
Burge, T., 2013. Cognition Through Understanding: Philosophical Essays, iii. Oxford: Oxford
University Press.
Cohen, S., 1984. “Justification and Truth” Philosophical Studies 46: 279–95.
Comesana, J., 2002. “The Diagonal and the Demon.” Philosophical Studies 110: 249–66.
Feldman, R., 1985. “Reliability and Justification.” The Monist 68: 159–74.
Feldman, R., 2008. “Moderate Deontologism in Epistemology.” Synthese 161: 339–55.
Fine, K., 1994. “Essence and Modality.” Philosophical Perspectives 8: 1–16.
Foley, R., 1985. “What’s Wrong with Reliabilism?” The Monist 68: 188–202.
Foley, R., 1993. Working Without a Net. Oxford: Oxford University Press.
Goldberg, S., 2012. “A Novel (and Surprising) Argument against Justification Internalism.”
Analysis 27: 239–43.
Goldberg, S., 2015. “What Is the Subject-Matter of the Theory of Epistemic Justification?” In
J. Greco and D. Henderson (eds) Epistemic Evaluations: Purposeful Epistemology. Oxford:
Goldman, A., 1979. “What is Justified Belief?” In G. Pappas (ed.) Justification and Knowledge,
1–23. Dordrecht: D. Reidel.
Goldman, A., 1986. Epistemology and Cognition. Cambridge, MA: Harvard University Press.
Goldman, A., 1988. “Strong and Weak Justification.” Philosophical Perspectives 2: 51–69.
Goldman, A., 2001. “Reply to Sosa.” Philosophical Topics 29: 465–8.
Goldman, A., 2002. Pathways to Knowledge. Oxford: Oxford University Press.
Goldman, A., 2011. In E. Zalta (ed.) The Stanford Encyclopedia of Philosophy, <http://plato.
stanford.edu/archives/spr2011/entries/reliabilism/> accessed July 7, 2015.
Graham, P. J., 2010a. “Theorizing Justification.” In J. Campbell, M. O’Rourke, and H. Silverstein
(eds) Knowledge and Skepticism, 45–71. Cambridge, MA: MIT Press.
Graham, P. J., 2010b. “Testimonial Entitlement and the Function of Comprehension.” In
D. H. Pritchard and A. Millar (eds) Social Epistemology, 148–93. Oxford: Oxford University
Press.
Graham, P. J., 2011a. “Does Justification Aim at Truth?” Canadian Journal of Philosophy 41:
51–71.
Graham, P. J., 2011b. “Perceptual Entitlement and Basic Beliefs.” Philosophical Studies 153:
467–75.
Graham, P. J., 2012a. “Epistemic Entitlement.” Noûs 46: 449–82.
Graham, P. J., 2012b. “Psychological Capacity and Positive Epistemic Status.” In J. Graper
Hernandez (ed.) The New Intuitionism, 128–50. London: Continuum.
Graham, P. J., 2014a. “Against Transglobal Reliabilism.” Philosophical Studies 169: 525–35.
Graham, P. J., 2014b. “Warrant, Functions, History.” In A. Fairweather and O. Flanagan (eds)
Naturalizing Epistemic Virtue, 15–35. Cambridge: Cambridge University Press.
Graham, P. J., 2014c. “Against Inferential Reliabilism: Making Origins Matter More.” Korean
Journal for Analytic Philosophy 31: 87–121.
Graham, P. J., (forthcoming). “Perceptual Entitlement and Natural Norms.” In P. Graham and
N. J. L. L. Pedersen (eds) Epistemic Entitlement. Oxford: Oxford University Press.
Greco, J., 2005. “Virtues in Epistemology.” In P. Moser (ed.) The Oxford Handbook of
Epistemology, 287–315. Oxford: Oxford University Press.
Henderson, D., and T. Horgan, 2011. The Epistemological Spectrum. Oxford: Oxford University
Press.
Lehrer, K., 2000. Theories of Knowledge, 2nd edn. Boulder, CO: Westview.
Leite, A., 2004. “On Justifying and Being Justified.” Philosophical Issues 14: 219–53.
Lyons, J., 2009. Perception and Basic Beliefs, Oxford: Oxford University Press.
Lyons, J., 2013. “Should Reliabilists Worry about Demon Worlds?” Philosophy and
Phenomenological Research 86: 1–40.
Millikan, R. G., 1984. Language and Other Biological Categories. Cambridge, MA: MIT Press.
Plantinga, A., 1993. Warrant and Proper Function. Oxford: Oxford University Press.
Pollock, J., 1999. “Procedural Epistemology.” In J. Greco and E. Sosa (eds) The Blackwell Guide
to Epistemology, 383–414. Oxford: Blackwell.
Pollock, J., and J. Cruz, 1999. Contemporary Theories of Knowledge, 2nd edn. Washington, DC:
Rowman & Littlefield.
Smythies, D., 2015. “Why Justification Matters.” In J. Greco and D. Hendersen (eds) Epistemic
Evaluations: Purposeful Epistemology. Oxford: Oxford University Press.
Sosa, E., 1980. “The Raft and the Pyramid.” Midwest Studies in Philosophy V.
Sosa, E., 1988. “Methodology and Apt Belief.” Synthese 74: 415–26.
Sosa, E., 2001. “Goldman’s Reliabilism and Virtue Epistemology.” Philosophical Topics 29:
383–400.
Sosa, E., 2003. Epistemic Justification. Oxford: Blackwell.
Sosa, E., 2004. “Replies to Critics.” In J. Greco (ed.) Ernest Sosa and His Critics, 275–326.
Oxford: Blackwell.
Sosa, E., 2007. A Virtue Epistemology. Oxford: Oxford University Press.
6
Who Knows?
Baron Reed
Ernest Sosa says, “Virtue epistemology is distinguished by its emphasis on the subject
as seat of justification.”1 In what follows, I shall modify this claim in three ways and
argue that it amounts to a fundamental truth in epistemology. The first two modifica-
tions can be made fairly quickly; the third will require rather more argument.
First, where Sosa talks about justification, my focus will be knowledge. Some philos-
ophers think that we can distinguish between a belief ’s being justified and a person’s
being justified in holding a belief. Presumably, this would imply that there is a notion of
justification that is divorced to some extent from how we evaluate the person who has
the belief. I myself do not think that there is a genuine distinction here, but I am content
simply to point out that no one thinks such a distinction holds in the case of knowledge.
That is to say, there is no sense in which a belief could count as knowledge without it
being the case that someone knows something, and we cannot say that someone knows
something without it being the case that the belief in question is knowledge.
Second, although it may be distinctive of virtue epistemology that it is explicitly
focused on the subject as the seat of justification, the recognition of the subject’s
importance is by no means limited to virtue epistemologists. Indeed, as I shall argue,
the intuitions underlying Sosa’s claim are widely—perhaps universally—shared.
The third modification involves replacing Sosa’s mention of “subjects” with a focus
instead on persons. The resulting claim, then, is that in epistemology the focus is on the
person as the seat of knowledge. In what follows, I shall explore what this means and
trace some of its consequences. In particular, making explicit the link between persons
and knowledge will begin to make clear both the nature and the value of human
knowledge.
1. What Happens When Someone Acts?

Before turning to a consideration of what is involved in knowing, it will prove useful
to begin with a somewhat different question: what happens when someone acts? In
1
Sosa (1997: 419).
who knows? 107
asking that question, David Velleman raises a significant objection to one of the tradi-
tionally dominant accounts of action.2 According to the sort of view in question, what
happens when someone acts is roughly this: the agent has a desire to bring about some
end, and she has some belief about how to satisfy that desire; the belief and the desire
together cause the agent’s action.3 Further complications may be added to the view. For
example, the belief and desire may directly cause an intention, which then itself directly
causes the action; if so, then the belief and desire are only mediate causes of the action
in question. It may also be the case that the belief and desire together rationalize the
resulting action (and intention, if there is one).
The objection Velleman then raises is something like the following: although this
model of action may be right insofar as belief and desire genuinely are causally related
to action, it nevertheless cannot be the whole story. What is left out is what should be
the most important part of any account of action—the agent. As Velleman says, on
this model “psychological and physiological events take place inside a person, but the
person serves merely as the arena for these events: he takes no active part” (189). There
are two places in which we naturally take the agent to play an active role in acting. First,
although the relevant desire and belief may influence the agent, it is still the agent her-
self who forms an intention or decides how to act. And, second, it is not the intention
that directly causes the action; rather, it is the agent who implements the intention
in acting as she does.4
Velleman is quite clear that this objection does not preclude a reductive account of
action. There is no need to take the agent to be a primitive, irreducible element in
agency, separate from her beliefs and desires. However, he says, the agent cannot be
reduced to the belief and desire in question, which purportedly cause the action (or
intention). After all, if the agent is supposed to be the element in agency that acts upon
the basis of that desire and belief (perhaps taking them as reasons), then she must be
something other than them.
Here, then, is the upshot of Velleman’s objection: “Just as the mind-body problem is
that of finding a mind at work amid the workings of the body, so the problem of agency
is that of finding an agent at work amid the workings of the mind” (196). What I shall
now propose is that this problem of agency is in fact two problems. The one with which
Velleman has been concerned we may call the problem of practical agency. But there is
also a problem of epistemic agency—that is to say, we must be able to find an epistemic
agent at work amid the cognitive workings of the mind. Knowing is not merely some-
thing that happens within a person’s mind—it is something that the person does. To
put the point a different way, someone who knows is not merely a subject who has
knowledge, she is a person to whom knowledge belongs.
2
Velleman (1992).
3
Velleman mentions Davidson as one of the major defenders of this sort of view but says that it can also
be traced back as far as Hobbes.
4
See Velleman (1992: 189–90) for both points.
108 Baron Reed
Before presenting the case for this claim, let me first address an objection that is
likely to occur immediately to many philosophers: in saying that knowledge does not
merely happen in a person’s mind but is rather something that the person does, I am
taking knowledge to be an action, but there are at least two reasons why this is implau-
sible. First, the thesis of doxastic voluntarism—that we are able to choose what we
believe—is probably false. Although it may be difficult to provide a decisive argument
against it, the phenomenology of belief seems to show that this is not an ability we
have. However, we do apparently have the power to choose how we act. Second, much
of the knowledge we have is gained in what is apparently an automatic sort of way.
Merely opening my eyes, I learn a tremendous amount about my environment—even,
at times, when I do not want to. Much of our knowledge seems to come to us without
any prior deliberation. This is true not only of much of our perceptual knowledge, but
also of many of the things we know by memory, by testimony, by introspection, and
by inference.
I am willing to grant both of these points, that doxastic voluntarism is false and that
much of our knowledge is automatic. I am even willing to grant the broader point they
are meant to establish—knowledge is not an action. But this concession is not as large
as it might look. First, even if we do not have direct control over what we believe (and,
for that reason, do not have direct control over what we know), we still have an indi-
rect control over our beliefs (and, thereby, over our knowledge). Once in a cognitive
situation, I have no control over how I respond doxastically to it. But I can affect my
beliefs by acting so as to put myself in a different cognitive situation.5 In short, knowl-
edge is not something that we produce directly, but it is a goal toward which we direct
our actions. Second, many of our actions are as automatic as the perceptual beliefs I
form simply by looking out of the window. When a squirrel darts in front of my car, I
brake without giving it a thought. This is still an action of mine, shaped by my charac-
ter. In the same way, the automatic beliefs I form through perception, inference, and
so on, are shaped by my intellectual character. Though this is often hidden from sight,
the automatic responses I make in a given situation may well differ from those of
another person (e.g., we may bring different conceptual resources to bear on the
circumstances).
So, even if knowledge is not itself an action, this is no bar to its being action-like in
important ways. Although the distinction between actions and events that merely hap-
pen to us is a fundamental and important one, it falls under a distinction that is broader
still. What I have in mind here is the difference between being active and being passive.6
5
For more on this sort of indirect control over belief, see my 2001. There may be limits to how much
control over belief is possible—e.g., there may not be any action I could take that could put me in a position
to believe that I am currently living in the fourteenth century—but there are also limits to control over
action (no amount of training, at this point in my life, will ever allow me to run a marathon in under three
hours). So, there is no point of dissimilarity here.
6
See Frankfurt (1976). The distinction is also important in Spinoza’s account of freedom and moral
psychology.
who knows? 109
To illustrate what the distinction is, suppose for a moment that determinism is true:
everything that happens is the necessary consequence of what came before. Although
some philosophers have taken determinism to be incompatible with the existence of
free and responsible action, there is a long tradition that resists this way of thinking.
The ancient Stoics explained the point by way of analogy: a cylinder standing on the
side of a slope cannot move on its own but, once it has been given a push by something
outside, it continues to roll down the slope through its own nature.7 In the same way,
humans cannot initiate action on their own. Once pushed from outside, however, their
own natures determine how they will behave. Given the internal complexity of
humans, moreover, it is not difficult to see that the greatest part of the explanation for
their behavior will rest in their internal natures, rather than in the external stimuli that
prompted them into action. This behavior belongs to the agent in a fundamental and
important sense.
The major challenge for philosophers in this tradition, which of course often goes by
the name of “compatibilism,” is to identify what the “internal nature” of an agent might
be.8 Many of the early compatibilists identified the agent’s will as constituting the inter-
nal nature from which free and responsible action flowed.9 The standard objection
from their opponents has been that the causal link between the agent’s will and her
action is insignificant when the agent’s will has been entirely determined by factors
that are ultimately out of her control.10 According to a second, and perhaps more com-
pelling objection, the above, simple form of compatibilism is unable to explain why
agents are not responsible for actions that are caused by compulsive desires. In such
cases, the agent’s action is still caused by the agent’s will. Because the simple form of
compatibilism takes the causal antecedents to the will to be irrelevant, it is unable to
rule out especially problematic causes like addiction or compulsion. Indeed, they
count as free and responsible actions for this simple sort of compatibilist view.
Recent, sophisticated forms of compatibilism have attempted to respond to both
objections by offering more robust accounts of what the internal nature of the agent is
supposed to be. Harry Frankfurt has proposed that a free will is one with the proper
hierarchical structure: the agent has a second-order volition through which she
chooses (“identifies with”) one of her first-order desires as the one that she wants to
be productive of action.11 Other philosophers, such as Gary Watson and Susan Wolf,
7
See the passages from Cicero (On fate) and Gellius in Long and Sedley (1987: 62C–D).
8
For philosophers on the other side of this debate—the libertarians—the question is easily
answered: the internal nature of an agent is the free will, which operates free from all causal constraint.
Other, more difficult questions will have to be answered by the libertarians, but it should be clear that
they would not reject the basic distinction between being active and being passive. I focus here on the
compatibilist tradition merely because it has the harder time distinguishing agents from the rest of
the universe and thus has tended to devote more attention to the distinction between being active and
being passive.
9
See, for example, Hume’s definition of ‘liberty’ as “a power of acting or not acting, according to the
determinations of the will” in the Enquiry, section viii, no. 73.
10
See van Inwagen (1983). 11 Frankfurt (1971).
110 Baron Reed
have suggested instead that reason constitutes the internal nature of the agent.12 There
are powerful objections to each of these views in turn, but what I want to focus on is the
broader, shared perspective: an agent is active to the extent that her behavior flows
from her internal nature (whatever that might be). When that happens, the action
belongs in an important way to the agent.
From the practical point of view, then, action is the result of an agent’s being active.
The analogous claim in epistemology is this: knowledge is the result of an epistemic
agent’s being active. Knowledge belongs in an important way to the epistemic agent
who has it because it flows from her internal cognitive nature. In what follows, I shall
argue, first, that this perspective is not a foreign imposition on epistemology but rather
a way of bringing to light intuitions that have already guided the debate about the
nature of knowledge and, second, that becoming more explicit about the nature of the
epistemic agent will help us better understand the nature of knowledge.
2. Why Does Knowledge Require an Epistemic Agent?

Let us approach this question by considering a couple of epistemological views that
attempt to account for knowledge without any substantive reference to the epistemic
agent who has it. Perhaps the clearest such example is Robert Nozick’s tracking account
of knowledge.13 In its simplest form, the tracking account holds knowledge to obtain
when the following four conditions are satisfied:
(1) It is true that p,
(2) S believes that p,
(3) if it were not the case that p, S would not believe that p, and
(4) if it were the case that p, S would believe that p.
Although the last three conditions do refer to the person, her role is rather minimal. In
particular, there is nothing in the tracking view that requires the person to be active in
any way in having the belief that p. To see this, notice that we could construe belief as
something much like an allergic reaction and yet all of Nozick’s conditions could be
met. As long as the person’s allergic reaction could be said to have the property of “rep-
resenting” the allergen that causes it, it could easily track the presence of that allergen:
if the allergen were not present, S would not have a reaction, and, if the allergen were
present, S would have a reaction. Although we do sometimes talk about a person hav-
ing an allergic reaction, it is fundamentally not a way in which someone is active.
Rather, allergic reactions merely happen to one. They are ways in which one’s immune
system responds to stimuli. We do not typically see any role for an agent to play in this
sort of transaction with the environment.
12
See Watson (1980) and Wolf (1990). Velleman’s proposal is somewhat different; he takes rational
agents to be constituted essentially by the desire to “do what makes sense, or what’s intelligible to them, in
the sense that they could explain it” (207).
13
Nozick (1981).
who knows? 111
It should be noted that Nozick does modify his view so as to take into account the
method by which S’s belief is formed. He does so to handle cases where the person
tracks via one method though, if it were false that p, she would use another method
that fails to track whether p. For example, a grandmother may see that her grandson is
well and track by means of vision that this is so, though if he were ill his family would
lie about his health to spare her feelings (in which case her belief acquired through
testimony would not track the truth about her grandson’s well-being).14 For this rea-
son, conditions (3) and (4) above need to be modified:
(3*) If it were not the case that p, and S were to use method M to arrive at a belief
whether or not p, then S would not believe via M that p, and
(4*) if it were the case that p, and S were to use M to arrive at a belief whether or not
p, then S would believe via M that p.15
Still, this change does not bring the person, as epistemic agent, into any more of an
active role than she had in the initial account. At most, to continue the prior analogy
between beliefs and allergic reactions, the change in Nozick’s account is merely a rec-
ognition that the immune system can function with multiple subsystems, each of
which may have different counterfactual profiles. An allergic reaction produced by one
such subsystem may track the presence of an allergen even though, if the allergen were
absent, a different subsystem that fails to track its presence would be dominant. Again,
there is no need to see the person involved as an agent in any of these transactions with
the environment.
Perhaps the most compelling objection to Nozick’s system has been that it leads to
failures in closure—that is, cases in which one knows the premises of an inference and
knows that they license a particular conclusion but is then unable to know that the
conclusion is true. To be sure, Nozick regards it as an advantage of his view that, if it is
correct, knowledge turns out not to be closed under deductive inference. This is par-
ticularly supposed to be helpful in response to skepticism—Nozick thinks that we can
concede to the skeptic that we don’t know that we are not brains in vats, but he also
thinks that this has no direct bearing on our ordinary instances of knowledge. Perhaps
this is a viable response to skepticism.16 In epistemology, it often seems that when skep-
ticism is at issue all bets are off; perhaps the deductive inference linking my inability to
rule out the brain-in-a-vat scenario to my knowing that I have hands really does not
hold. Nevertheless, Nozick’s view leads to other failures of closure—ones that are much
harder to accept. Consider the following case, adapted with minor changes from one
given by Richard Fumerton:
Assassination.17 Maria forms the true belief that Jones has assassinated Smith by reading an
article to that effect in the local newspaper. The report is accurate and permits Maria’s belief to
14
Nozick (1981: 179). 15 Nozick (1981: 179).
16
But see DeRose (1995) on what he calls Nozick’s “abominable conjunction.”
17
This case comes, with minor modification, from Fumerton (1987: 172). See Kripke (2011: chapter 7)
for a similar case.
112 Baron Reed
track the truth that Jones has assassinated Smith. In particular, if it were not true, Maria would
not form the belief by reading in the newspaper that Jones had assassinated Smith and, if it
were true, she would form that belief by reading about it in the newspaper. Now, it is also the
case that the group behind Smith’s assassination has a back-up plan to sow discord in case Jones
fails to kill Smith: if that were to happen, they would plant a false story in the newspaper saying
that someone had killed Smith (though the story would name Jones as one of the conspirators,
it would not say that Jones was the assassin). Consequently, Maria’s belief that someone has
killed Smith does not track the truth. If it were false that Smith had been assassinated, she
would still believe that he had been.
Fumerton thinks it is absurd that Maria could know that Jones killed Smith but not that
someone killed Smith. Nozick could respond to this case by insisting that we pay atten-
tion to the method Maria uses in forming the belief that someone has assassinated
Smith. It seems to be a belief that she has actually inferred from her belief that Jones
assassinated Smith. In the counterfactual situation, however, she does not acquire the
belief by inference but rather by reading the newspaper. So, this is not a successful
counterexample to the tracking view.18
This reply on Nozick’s behalf is correct, and yet it does not remove the problematic
nature of the case. There are two separate ways in which it can be modified so that the
method Maria uses is taken into account. First, let us specify that Maria actually forms
her belief that someone assassinated Smith, not by inference, but directly from reading
the newspaper article (perhaps the article she actually reads begins by stating that
someone has assassinated Smith and only later identifies the assassin as Jones). In that
case, the relevant method, both actually and counterfactually, is reading the newspa-
per article. That method yields the problematic combination of beliefs: that Jones
assassinated Smith (which does track the truth) and that someone assassinated Smith
(which does not). Even though Maria did not actually infer the latter belief from her
belief that Jones assassinated Smith, it is remarkably odd that she could know that
Jones assassinated Smith but not that someone assassinated Smith. The two beliefs
form an obvious rational pattern in her mind; it is incredible that they could have such
different epistemic statuses.
The second way in which the case may be modified takes Maria’s belief that someone
has assassinated Smith to be inferential both actually and counterfactually. She actu-
ally infers it from her belief that Jones assassinated Smith. Nozick suggests that, when
the belief that p is inferred from the belief that q, we take the method used to be “infers
it from q.”19 So, we must look at the counterfactual situation in which it is false that
18
I am grateful to Sandy Goldberg for pressing this reply to the Assassination case.
19
Nozick (1981: 189). It should be noted, however, that this will cause problems for many ordinary cases
of inferred belief. Suppose I believe via visual perception that my brother is standing in front of me and
infer from this belief that a relative of mine is standing in front of me. Does this inferred belief track the
truth? One of the subjunctive situations to consider is that in which none of my relatives is standing
in front of me. The closest possible worlds in which none of my relatives is standing in front of me and yet
I try to infer something from the proposition that my brother is standing in front of me will be the (pre-
sumably distant) worlds in which I hallucinate my brother. In that case, I will believe one of my relatives is
who knows? 113
someone assassinated Smith and yet Maria forms a belief by inference from the propo-
sition that Jones assassinated Smith. Most plausibly, this happens in virtue of Maria
misreading the conspirators’ false article as stating that Jones was the assassin (though
it doesn’t really say this). Maria would thus believe that someone assassinated Smith
via this inferential method even when it wasn’t true. So, her belief that someone assas-
sinated Smith would fail to track the truth, even though the belief from which she
actually inferred it—that Jones assassinated Smith—does track the truth.20
The inability to draw some obvious inferences from what one knows—and thereby
know the conclusions—is widely considered to be a serious problem for Nozick’s view.
What has not been generally recognized is that his tracking account licenses another
sort of rational failing—one that is even more disturbing. It is not merely that we are
sometimes unable to extend our knowledge via rational inference, but also the case
that we are sometimes able to acquire (what is purportedly) knowledge via inferences
that are clearly defective. To see this, consider the following:
Vegetable. Samantha tends to get confused about whether the tomato is a fruit or a vegetable. In
particular, when tomatoes are presented to her as garnish for a sandwich, she thinks of them as
vegetables. But, when presented with tomatoes alone, she thinks of them as fruit (as, in fact,
they are). Samantha’s mother makes a tomato sandwich with a side of fried potatoes for her
lunch. After eating, Samantha believes that she has eaten a tomato and infers from this the belief
that she has eaten a vegetable. (Let us assume that she does not recognize the potato as a vege-
table, and she generally does not eat other vegetables.) Because Samantha’s mother always
combines tomato sandwiches with fried potatoes, Samantha’s inferred belief meets condition
(4*): if she were to eat a vegetable and form a belief as to whether she had done so by inferring
this from the proposition that she had eaten a tomato, she would believe that she had eaten a
vegetable. What would be the case if Samantha had not eaten the sandwich? Let us suppose that
her mother would have given her a cherry tomato to eat, in which case Samantha would have
inferred, from her belief that she has eaten a tomato, that she had eaten a fruit. Thus, her inferred
belief would meet condition (3*): if it were not the case that she had eaten a vegetable and she
were to form a belief as to whether she had eaten a vegetable by inference from the proposition
that she had eaten a tomato, she would not have believed that she had eaten a vegetable.
A point of clarification is in order: when Samantha is presented with a tomato in the

context of a sandwich, she simply sees it as a vegetable. Similarly, when she is presented
standing in front of me—which means that this belief will not track the truth. Nevertheless, my inferred
belief seems to be perfectly respectable in the actual world.
20
Wouldn’t Maria’s misreading the conspirators’ false story mean that she would believe Jones was the
assassin even when he wasn’t? And, therefore, wouldn’t her belief that Jones assassinated Smith also fail to
track the truth? No; although the possible world in which she misreads the false story is relevant to the
counterfactual regarding what she would do if it were false that someone assassinated Smith and she
formed a belief by inference from the proposition that Jones was the assassin, that world is sufficiently
distant that it is not relevant to the counterfactual regarding what Maria would believe via reading the
newspaper if it were false that Jones was the assassin. In the nearest possible worlds in which Jones is not
the assassin, Maria would continue to believe that someone assassinated Smith, but she would not continue
to believe that Jones did so. It is the latter fact, of course, that is relevant to determining whether Maria’s
belief that Jones assassinated Smith tracks the truth.
114 Baron Reed
with a cherry tomato alone, she simply sees it as a fruit. In neither case is she drawing an
inference from any background beliefs regarding whether the tomato is a fruit or a veg-
etable. Hence, both actually and counterfactually, she would be drawing an inference
directly from her belief that she has just eaten a tomato. In both cases, then, she would be
using the same method to form a belief as to whether or not she has eaten a vegetable.
I think it will strike most philosophers as fairly obvious that Samantha does not
know that she has eaten a vegetable in this case. Her way of arriving at that belief is
rationally defective in a way that is generally regarded as incompatible with the belief
being knowledge.21 But the belief, though acquired in this defective way, does track the
relevant truth. Why does it not count as knowledge, then? The answer must be that,
although the belief is successful in itself and in its own relation to the environment,
that success does not belong to the person. Although the belief is well connected to the
environment, the person is not. So, even though the belief does belong to the person,
it does not permit the person to know how things are in her environment.
The upshot of these considerations is that knowledge requires something more than
merely a belief that successfully fits with the environment.22 That does not yet show,
however, that we must find an active role for the person in the possession of knowl-
edge. Compatibly with the conclusion just drawn, it may turn out that knowledge has
to be grounded simply in a proper fit between the environment and some cognitive
faculty that belongs to the person. To put the point another way, when we understand
the person to be merely a collection of cognitive faculties, perhaps all that needs to
be the case is that at least one of these faculties bears the right sort of relation to the
environment.23 If so, there is no need to see the person herself, as distinct from these
individual faculties, as involved in the possession of knowledge.
Here is one way of developing this suggestion: take a faculty to be connected in
the right sort of way with the environment when it is reliable, in the sense of being
truth-conducive. This, of course, is a rough characterization of the way that Alvin
Goldman begins the presentation of reliabilism.24 But, as he recognizes, revision is
necessary. Consider the following:
Birding Book. Molly sees (and thereby believes) that a downy woodpecker is in the tree in her
front yard. Her visual perception is generally reliable, and she also is reliable in identifying
21
Notice that a defender of the tracking view could not fix the problem by stipulating that beliefs formed
on the basis of inference must be inferred from beliefs that count as knowledge. In Vegetable, the belief
from which Samantha is drawing inferences (i.e., that she has eaten a tomato) is something she knows.
22
Condition (3*) of Nozick’s account expresses sensitivity, a modal relation that a belief has to the fact it
represents. The arguments I have given here target sensitivity, but similar arguments could be given for
alternative modal relations. For example, Sosa (1996) and (1999) and Williamson (2000) propose a safety
condition for knowledge in place of sensitivity; a belief is safe when it couldn’t easily be false (or, in other
words, if one were to believe that p, it would be true that p). For arguments of this sort see Lackey (2008a).
23
This would nevertheless be a stronger requirement than what we find in the modified tracking
account, where the inclusion of methods means only that we must individuate beliefs by their causal origin.
The modified tracking account does not further require that the method that gives rise to the belief be
reliable in any broader sense.
24
Goldman (1979).
who knows? 115
birds. Molly then reads in a reputable birding book that downy woodpeckers are not to be
found in her part of the country at that time of year, but she chooses to ignore the book. As it
happens, the birding book, which is generally reliable, has made a mistake—downy woodpeck-
ers are in fact year-round residents, and it really is a downy woodpecker in her yard.
Molly’s belief, acquired via perception, is thus true and reliable, but it does not seem
to be knowledge. The conclusion that Goldman draws from considering a case like
Birding Book is that the epistemic status of a belief “is not only a function of the cogni-
tive processes actually employed in producing it; it is also a function of processes that
could and should be employed.”25
In the case considered, the product of one reliable process is in conflict with (what
would be) the deliverance of another reliable process. When this happens, the belief
produced by the process one actually uses cannot count as justified or as knowledge:
reliability can be trumped by reliability, it seems. It is worth noting, however, that relia-
bility can also be overridden by the mere appearance of reliability:
Lying Neighbor. Instead of looking at the birding book, Molly encounters a neighbor who tells
her that the bird was not really a downy woodpecker. Moreover, the neighbor, who is a patho-
logical liar, spins a very plausible fabrication about why it has to be a hairy woodpecker, which
is very easily confused with the downy woodpecker. Although the testimony is very plausible,
and Molly has no reason to be suspicious of the testifier, she simply disregards it, thereby
retaining her true belief.
As in Birding Book, Molly has a reliable, true belief, but the belief does not seem to be
knowledge. She should have given some weight to the contrary testimony, even though
it was not in fact reliable.26 For that reason, the reliability of her perceptual faculties is
not enough to ground the belief as knowledge.
As we saw before when considering the objection to Nozick’s tracking view, the per-
son in these cases has a belief that fits well with the environment. Even more strongly,
the belief is the product of a reliable faculty that itself fits well with the person’s envi-
ronment. Nevertheless, the success of the belief and the success of the faculty do not
belong to the person in the right sort of way. Although her belief and her faculty are
well connected to the environment, she is not. This is so even though the belief and the
faculty are both hers. Because the person herself is not in a position to know, in these
cases, her belief does not count as knowledge.
Any plausible account of knowledge, then, must be able to show how the person—
and not merely her belief or her cognitive faculty—is cognitively related in the right
sort of way to what is known. But why think this means that the person must be
active—that is, must be an epistemic agent? To see why the person cannot be merely
the passive location in which reliable cognitive faculties are at work, return to
25
Goldman (1979: 123).
26
See BonJour (1985: chapter 3), for a variety of cases of this general kind; compare Gilbert Harman’s
newspaper case (1973: 143–4). It is interesting to note that both Goldman and Nozick think it is necessary
to respond to BonJour’s cases with a no-defeaters condition.
116 Baron Reed
Goldman’s conclusion above: the belief ’s epistemic status is determined not only by
the reliable processes the person actually used, but also by those that “could and should
be employed.” Now, presumably, the facts about which faculties or processes could be
used in a given situation may turn out to be independent of anything the person does.
So, for example, the availability of her neighbor’s testimony does not depend on Molly
doing much of anything. He has spoken in her presence, using words that she under-
stands; this is all it takes to make the use of his testimony a belief-forming process that
Molly could use. But is that enough to make it the case that she should rely on his testi-
mony when it comes into conflict with her own belief?
There are different senses of “should”; which is relevant here? In one sense, the
person should believe only what is true. That this is so is obvious, given that the dis-
covery that one’s belief is false provides one with an immediate motivation to change
it (and immediately serves as the basis for a kind of criticism of the belief). However,
this is not the sense of “should” that is at issue. In Birding Book, the upshot was that
Molly should have given some weight to what the book said—even though it was
explicitly part of the case that the book was mistaken. The same moral was drawn
from Lying Neighbor: Molly should have taken into account her neighbor’s testimony,
even though it was nothing but a complex of lies and even though the neighbor is
generally (though unbeknownst to her) a pathological liar. Molly should have taken
it into account, despite the fact that it was both false and unreliable, because it is
rational to do so from her perspective. As far as she can tell, when she is discussing
the matter with her neighbor, the testimony is credible. People usually do not lie in a
situation like this where nothing is at stake for the speaker, and complex chunks of
testimony (where, e.g., some explanatory context is provided, rather than merely a
bald assertion) are usually grounded in some competence governing what is asserted.
Although from a purely consequentialist point of view, Molly turns out to have done
the best thing by ignoring her neighbor, this is not something that is warranted when
our focus is restricted to how things appear to her. From that point of view, she should
listen to what the neighbor has to say—even though it really is nothing but a pack
of lies.
The lesson often drawn from considering cases like those above is that, as BonJour
says, “external or objective reliability is not enough to offset subjective irrationality.”27
But what is it about subjective irrationality that makes it so important in epistemology?
Why, that is, do we focus on how things look from the person’s perspective? There are
two things that we expect the person to be able to do—and it is because these are things
the person must be able to do that we have at last our reason why the person must be an
epistemic agent. First, the epistemic agent must be able, not only to receive the inputs
from her various faculties, but also to synthesize them into a single cognitive response
to the environment. Even when she fails to do so, as Molly did in the above two cases,
she is evaluated with this in mind. Ultimately, it is not discrete cognitive faculties that
27
BonJour (1985: 41).
who knows? 117
must be appropriately related to the environment; it is the epistemic agent’s response

that matters.
The second thing the epistemic agent must be able to do is to take into account nor-
mative constraints on her cognitive response. The information from the birding book
is not relevant to Molly’s belief simply in virtue of being physically in front of her—if
the book had been written in a language she couldn’t read, it would have been physi-
cally present yet still evidentially unavailable to her. Rather, the birding book makes
available the information to her in such a way that she is reasonably expected to include
it in her synthesis of her total cognitive state. The epistemic agent is thus normatively
required to recognize the epistemic significance of how things are in her environment.
Moreover, the synthesis she is expected to make, drawing together all of the inputs
from her various cognitive faculties, must be a rational one. In other words, the deliv-
erances of her faculties do not simply enter into a merely causal competition with one
another, in the way that, for example, one’s desire to have a piece of candy may over-
whelm one’s conflicting desire to stick to a diet. If the epistemic agent is to be successful
in matching her cognitive response to how things are, she must weigh the deliverances
of her faculties so that it is their epistemic properties that matter.
To summarize the results of this section, then, it is the person—and not merely her
belief or her cognitive faculties taken individually—that must be properly related to
the environment. Our intuitions about what knowledge is make clear that, when we
attribute knowledge to someone, we are assessing her cognitive performance with the
expectation that she has made a rational integrated or synthesized response to her
environment.
This is what our intuitions about knowledge tell us. But we are now in a position to
see them in a broader perspective, the one sketched in the previous section. What is
driving the objections considered above, first to Nozick’s tracking view and then to the
simple form of reliabilism, is the underlying sense that those views have not succeeded
in capturing something we think essential to knowledge: they have not shown how
knowledge belongs to the person in the right sort of way. The belief in question and the
individual faculties in question are undeniably the person’s, but this is not enough. In
much the same way, a person’s allergic reaction and her immune system are undenia-
bly hers, but they do not belong to her qua practical agent—the allergic reaction is not
an action of hers. More fundamentally, it is not a way of being active.
As we have seen, honoring this intuition pushes us in the direction of acknowledg-
ing the importance of the subjective perspective of the epistemic agent. Given the
framework established in the first part of this essay, we can now explain why this is
necessary: it is implicitly an attempt to locate an internal nature in the cognitive
sense—a deeper self that doesn’t merely react to impulses from outside but rather
shapes its own responses. This is not to say that one’s full cognitive nature is wholly
implicated in every instance of knowledge. Rather, it is enough that it could be. All that
is required is that one is monitoring, from the rational point of view, how one’s beliefs
are adapting to changes in one’s environment. This means, at a minimum, that one is at
118 Baron Reed
least tacitly aware of the epistemic profiles of those beliefs, so that one can keep track
not only of the beliefs but of what makes them the proper beliefs to have.28
3. Clarifications
Before considering some implications of the framework sketched thus far, let me
offer several clarifications. First, although I have been approaching the claim that
knowledge belongs in a special way to the person who has it by considering how
various externalist theories have developed in the face of criticism, it should be clear
that the point generalizes beyond externalism. It is not only the deliverances of one’s
reliable faculties that must be rationally weighed and synthesized—this holds true
for evidence as well. Justification and knowledge, as conceived by the evidentialist,
cannot be the result of a belief ’s being supported by a single bit of evidence in isolation
from everything else. Someone who ignores counterevidence to her belief is
irrational—and therefore not in a position to have knowledge—in just the same way
as someone who ignores some of the inputs she receives from her reliable faculties.
So, wherever it is possible for the person to react cognitively to the environment in
complex and fallible ways—however we conceive of those reactions—the person
will be required to play the role of an active rational cognizer. To put the point a
different way, it is not only epistemologists coming from an externalist perspective
who must incorporate into their accounts some recognition of the central
importance of reflective rationality. It is a constraint on just about every view in
epistemology, including all fallibilist forms of evidentialism.
Second, one of the common motivations for externalist views is that they are
uniquely able to account for the ability of animals and young children to have knowl-
edge. But, if we take knowledge to be possible only for persons who are epistemic
agents and thus have rational capacities, it seems we will be unable to allow that ani-
mals and children do in fact have knowledge. There is much that can be said in response
to this kind of worry.29 Here, I shall only sketch a few replies. In the first place, it is by no
means clear that rationality is characteristic of adult humans alone. The Stoics, for
example, mention a dog that, in tracking its prey, comes to a three-way fork in the path.
The dog sniffs down the first two branches without picking up the scent and then heads
down the third path without bothering to sniff. The dog has performed a disjunctive
syllogism, though of course it would never be able to think of itself as having done so.
More generally, animals and young children seem to be able to modify their beliefs
in accordance with the evidence they take in, and they are often pretty good at discern-
ing when the evidence available is inadequate or inconsistent. This may not amount
to full-blown, reflective rationality, but the claim on behalf of which I am arguing
holds only that knowledge requires the person to be properly connected to how
28
For more on the relation between justification and how we respond to it, see Reed (2001).
29
See Lackey (2008b).
who knows? 119
things are. This will involve an evaluation of the person’s overall cognitive response to
her environment. In fully developed adult humans, that overall cognitive response
may be provided by full-blown, reflective rationality, but it is also possible that in other
creatures it is the product of a more primitive faculty or a more primitive form of
rationality.
But suppose, finally, that in a particular instance an animal or a child acquires a
belief through some reliable method but is then incapable of adjusting that belief in
accordance with new evidence. Should we really count that belief as knowledge? It is
hard to see why we should feel compelled to do so. Surely it is enough for such a subject
that her belief is reliably true—as Plato says in the Meno, true belief can be just as prac-
tically useful as knowledge.
Third, the framework for epistemology that I have attempted to articulate above
should be distinguished from a view defended recently by several virtue epistemolo-
gists.30 According to these philosophers, knowledge is true belief where the truth of the
belief is due to the abilities of the person; in that sense, the person deserves credit for
the success of her belief. Although this sort of virtue epistemology sounds quite similar
to the thesis that knowledge belongs to the person, there is at least the following signif-
icant difference. Where this version of virtue epistemology takes the credit thesis to be
explanatory of what knowledge is, the arguments I have presented here are in service
of the idea that knowledge belongs to the person who has it and that this feature of
knowledge—its attributability—is in need of further explanation.
This can be seen in two ways, one specific and one general. The specific need for a
deeper explanation can be seen in a dilemma that Jennifer Lackey raises for the credit
thesis. Lackey points out that, in many ordinary cases of knowledge acquired via testi-
mony, the vast majority of the epistemic work is done by the speaker, not by the hearer.
It is implausible to credit the hearer with the success of her belief in such a case. On this
horn of the dilemma, then, there are cases where the knower does not deserve credit
for the knowledge.
A natural response to this problem is to weaken the sort of credit a knower needs
in order to have knowledge. Thus, in response to this problem, Sosa argues that the
relevant notion of credit is a merely causal one and that it is consistent with the credited
person making only a quite small contribution to the success of her belief. But this
response leads to the second horn of Lackey’s dilemma: someone in a Gettier case
makes a contribution toward the success of her belief. In one of Gettier’s original cases,
the person ends up with a justified, true belief by correctly using the disjunction-
introduction rule in first-order logic.31 The belief is accidentally true, but its success is
at least partially grounded in the person’s logical competence. On this horn of the
dilemma, then, deserving credit for the success of one’s beliefs is not unique to—and
therefore not distinctive of—knowledge.
30
See especially Sosa (2011: 86, 2007, and 2003), Greco (2003), and Riggs (2002a and b).
31
Gettier (1963).
120 Baron Reed
The more general reason for thinking that the attributability of knowledge needs a
deeper account derives from the skeptical problem that arises from combining this
claim with fallibilism, the view that one can know something (on the basis of a particu-
lar justification) even though one might have failed to know it (while retaining that
same justification).32 If fallibilism is correct, there is always a gap between one’s cogni-
tive performance and the truth of one’s belief.33 Because the basis for one’s belief does
not guarantee the success of the belief, the person depends on the world to cooperate.
In purported cases of knowledge, the gap between the person’s epistemic performance
and the relevant fact is closed by the world being as she thinks it is; in straightforward
Gettier cases, on the other hand, the gap is closed in an unexpected way. But in both
sorts of cases, the gap is closed by something other than what she does. Knowledge is
supposed to be attributable to the person who has it, and yet—if fallibilism is correct—
it never is.
If this skeptical problem can be resolved, it will happen only through a fuller expla-
nation of what is involved in attributing knowledge to a person. Pointing out that
relationship is the beginning of epistemological inquiry—not the end of it.34
4. The Value of Knowledge

One of the usual objections to the claim that knowledge has value points to the appar-
ent worthlessness of some instances of knowledge. For example, I know that there are
now forty-three books on the top shelf of the bookcase in my office. Surely, that knowl-
edge serves no purpose (beyond its role as an example in this essay) and has no value.
My life would not be one whit the worse if I did not know it. But this is not supposed to
be the case for things that have value, and have it non-instrumentally—their absence
does leave one worse off, even if only by a little bit.35
It is interesting to note that the very same argument can be given in the case of free
action, which is also supposed to have non-instrumental value. Counting the books on
the top shelf of the bookcase in my office was something that I freely did. As far as I can
see, having done so added nothing of value either to my life or to the universe as a
whole. However, it is interesting to further note that philosophers do not appear to be
inclined to draw the conclusion that there is nothing of non-instrumental value in the
32
See Reed (2002) and (2012) for more on fallibilism.
33
See Reed (2007) and (2009) for arguments in support of this claim.
34
Sosa (2011) does offer a much more substantive account of knowledge in terms of aptness. I suggest
that this can be read as a way of fleshing out what it means to deserve credit for the success of a belief.
Insofar as a belief is apt when it is the manifestation of a competence in a particular environment, I think
(but will not here argue) that Sosa’s view will still face Lackey’s dilemma for testimony and my general
argument for skepticism.
35
Notice that this argument applies equally well to the view that the fundamental item of epistemic
value is true belief. If it is plausible to say that my knowledge about how many books are on the shelf in my
office is lacking in value, it is equally plausible to say that this true belief is also lacking in value.
who knows? 121
case of freedom. I think there are two reasons why this is so; I’ll conclude by explaining
why I think both reasons apply to the case of knowledge as well.
First, many would agree that morally right actions have non-instrumental value.
But it is possible to perform a morally right action only if one has acted freely. In that
sense, freedom is the precondition for morally right action. This does not mean that
freedom has only instrumental value—it is not a means by which someone performs a
morally right action. Rather, it is what puts one in the game of morality, so to speak. Of
course, not every exercise of one’s freedom has to contribute to one’s standing in that
game, but that does not diminish the value of freedom in general.
Something similar can be said about knowledge. Whether or not every instance of
knowledge has value, everyone would agree that at least some instances of knowledge
clearly do. Galileo’s discovery of the moons orbiting Jupiter, the invention of calculus
by Newton and Leibniz, Mendel’s working out the mechanism for the inheritance of
biological traits—these were all monumental achievements. Other uses of human cog-
nitive faculties—for example, the attempt to argue on behalf of the “intelligent design”
hypothesis—are far less admirable. Still, what makes possible the evaluation of one’s
broad intellectual efforts as good or bad is the capacity for knowledge (where that
capacity includes both the normative requirements articulated above—that one’s belief
be an overall response to one’s epistemic position—and the having of at least some
knowledge).36 Knowledge has non-instrumental value in virtue of suiting us for this
sort of intellectual endeavor. And, again, knowledge in general has this value even if
there are specific instances of knowledge that contribute little or nothing to intellectual
success in the broader sense.
Second, as Frankfurt has argued, the concept of a person can be “construed as the
concept of a type of entity for whom the freedom of its will may be a problem.”37
Persons and less sophisticated creatures alike have desires that find expression in
behavior. But persons differ in that they also have desires about those desires. That is to
say, a person is someone who reflects on the causes of her actions and to whom it mat-
ters how they have come to be. She may choose to continue acting as she has, or she
may decide instead to change her practical character. What really matters is that this
matters to her.
Much the same is true in the case of knowledge. Both persons and less sophisticated
creatures have beliefs, and these beliefs can equally well be true or false. But persons
differ in that they take an interest in how those beliefs have been formed. Where there
appears to be a mismatch between the belief and what the belief is about, there is a
problem for the person. She may perform better or worse as an epistemic agent, in
36
The capacity for knowledge includes the capacity for justification (where the latter is understood as
one of the necessary conditions for knowledge). A full examination of the relation between knowledge and
justification is beyond the scope of this essay, but see Reed (2010) and (2013) for more on how they interact
in the context of practical deliberation.
37
Frankfurt (1971: 19).
122 Baron Reed
acting to correct the mismatch, but (to echo Frankfurt) a person is a type of entity for
whom knowledge is a problem.38
Freedom and knowledge both have value insofar as they are both partially constitu-
tive of being a person. Their value is neither merely instrumental nor extrinsic because
their value is also partially constitutive of the value of being a person. Having carried
the matter to this point, I find it difficult to go much further. The value of freedom and
knowledge can be illuminated, to some degree, by bringing to light their relationship
with personhood. The converse is also true: some light can be shed on the value of
personhood by revealing its connections with freedom and knowledge. But I am afraid
that, in the end, no further explication is possible. If we are to acknowledge the value
of all three, we shall have to do so, not out of further philosophical argument, but out of
our first-hand experience as persons who act and know.39
References
BonJour, L., 1985. The Structure of Empirical Knowledge. Cambridge, MA: Harvard University
Press.
DeRose, K., 1995. “Solving the Skeptical Problem.” Philosophical Review 104: 1–52.
Frankfurt, H., 1971. “Freedom of the Will and the Concept of a Person.” Journal of Philosophy
68(1): 5–20. Reprinted in Frankfurt (1988): 11–25.
Frankfurt, H., 1976. “Identification and Externality.” In A. O. Rorty (ed.) The Identities
of Persons. Berkeley, CA: University of California Press. Reprinted in Frankfurt (1988),
pp. 58–68.
Frankfurt, H., 1988. The Importance of What We Care About. Cambridge: Cambridge University
Press.
Fumerton, R., 1987. “Nozick’s Epistemology.” In S. Luper-Foy (ed.) The Possibility of Knowledge:
Nozick and His Critics, 163–81. Totowa, NJ: Rowman & Littlefield.
Gettier, E., 1963. “Is Justified True Belief Knowledge?” Analysis 23: 121–3.
Goldman, A., 1979. “What Is Justified Belief?” In G. Pappas (ed.) Justification and Knowledge,
1–23. Dordrecht: Reidel.
Greco, J., 2003. “Knowledge as Credit for True Belief.” In M. DePaul and L. Zagzebski (eds) 2003.
Intellectual Virtue: Perspectives from Ethics and Epistemology, 111–34. Oxford: Clarendon
Press.
Harman, G., 1973. Thought. Princeton, NJ: Princeton University Press.
Hume, D., 1777. An Enquiry Concerning Human Understanding, 3rd ed. by L. A. Selby-Bigge
and P. H. Nidditch. Oxford: Clarendon Press.
Kripke, S., 2011. Philosophical Troubles. Oxford: Oxford University Press.
38
Here, I leave open whether the way in which knowledge is a problem for a person is best understood
as a first-order or second-order phenomenon.
39
For helpful discussion of an earlier draft of this essay, I am grateful to audiences at a conference on
epistemic agency at University Carlos III in Madrid and at the University of Wisconsin-Milwaukee, as well
as to my colleagues Kyla Ebels-Duggan, Rachel Zuckert, and, especially, Jennifer Lackey. I am also grateful
to two anonymous referees for their helpful comments.
who knows? 123
345–61.
Lackey, J., 2008a. “What Luck Is Not,” Australasian Journal of Philosophy 86: 255–67.
Lackey, J., 2008b. Learning from Words. Oxford: Oxford University Press.
Long, A. A., and D. N. Sedley, 1987. The Hellenistic Philosophers, i. Cambridge: Cambridge
University Press.
Nozick, R., 1981. Philosophical Explanations. Cambridge, MA: Harvard University Press.
Reed, B., 2001. “Epistemic Agency and the Intellectual Virtues.” The Southern Journal of
Reed, B., 2002. “How to Think about Fallibilism.” Philosophical Studies 107: 143–57.
Reed, B., 2007. “The Long Road to Skepticism.” Journal of Philosophy 104: 236–62.
Reed, B., 2009. “A New Argument for Skepticism.” Philosophical Studies 142: 91–104.
Reed, B., 2010. “A Defense of Stable Invariantism.” Noûs 44: 224–44.
Reed, B., 2012. “Fallibilism.” Philosophy Compass 7(9): 585–96.
Reed, B., 2013. “Fallibilism, Epistemic Possibility, and Epistemic Agency.” Philosophical Issues
23: 40–69.
Riggs, W., 2002a. “Beyond Truth and Falsehood: The Real Value of Knowing that P,”
Philosophical Studies 107: 87–108.
Riggs, W., 2002b. “Reliability and the Value of Knowledge.” Philosophy and Phenomenological
Research 64: 79–96.
Sosa, E., 1996. “Postscript to ‘Proper Functionalism and Virtue Epistemology’.” In J. Kvanvig
(ed.) Warrant in Contemporary Epistemology, 271–80. Lanham, MD: Rowman & Littlefield.
Sosa, E., 1997. “Reflective Knowledge in the Best Circles.” Journal of Philosophy 94: 410–30.
Sosa, E., 1999. “How to Defeat Opposition to Moore.” Philosophical Perspectives 13: 141–53.
Sosa, E., 2003. “The Place of Truth in Epistemology.” In M. DePaul and L. Zagzebski (eds) 2003.
Intellectual Virtue: Perspectives from Ethics and Epistemology, 155–79. Oxford: Clarendon
Press.
Sosa, E., 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, i. Oxford: Clarendon
Press.
van Inwagen, P., 1983. An Essay on Free Will. Oxford: Clarendon Press.
Velleman, J. D., 1992. “What Happens When Someone Acts?” Mind 101: 461–81. Reprinted in
J. M. Fischer and M. Ravizza (eds.) Perspectives on Moral Responsibility, 188–210. Cornell
University Press.
Watson, G., 1980. “Free Agency.” Journal of Philosophy 72(8): 205–20.
Williamson, T., 2000. Knowledge and Its Limits. Oxford: Oxford University Press.
Wolf, S., 1990. Freedom Within Reason. New York: Oxford University Press.
7
Knowledge as Achievement,
More or Less
John Turri
This essay enhances and extends a powerful and promising research program,
performance-based epistemology, which stands at the crossroads of many important
currents in contemporary epistemology, including the value problem, epistemic
normativity, virtue epistemology, and the nature of knowledge. Performance-based
epistemology offers at least three outstanding benefits: it explains knowledge’s dis-
tinctive value, it places epistemic evaluation into a familiar and ubiquitous pattern of
evaluation, and it solves the Gettier problem. But extant versions of performance-based
epistemology have attracted serious criticism. This essay shows how to meet the objec-
tions without sacrificing the aforementioned benefits.
The essay proceeds as follows. Section 1 introduces the basic model of performance
assessment. Section 2 introduces the leading version of performance-based epistemol-
ogy, the achievement account of knowledge. Section 3 considers the objection that the
achievement account is too weak, and shows how to strengthen it. Section 4 considers
the objection that the achievement account is too strong, and shows how to weaken it.
Section 5 shows how to adjust performance-based epistemology if it turns out that
knowledge doesn’t require truth. Section 6 briefly concludes.
1. A Basic Model of Performance Assessment

A raging Achilles leads the Greek charge, driving the Trojans all the way to the Scaean
gate. The siege of Troy might have ended this day, but for one fateful shot. The Trojan
prince Paris, perched in the tower high above the gate, places an arrow on his bow,
surveys the bloody chaos below, identifies the indomitable Greek hero, takes aim and
lets the shot fly. Apollo guides the arrow through the mass of bodies and trampling
feet to bury deep into Achilles’s ankle, just below his leg armor. Only in that very spot
could Achilles receive his death blow. Blood gushing, strength failing, eyes dimming,
Achilles staggers forward in one last feeble defiant gesture before toppling dead to the
ground.
knowledge as achievement, more or less 125
We might assess Paris’s shot along any number of dimensions. We might think it
cowardly that he assailed Achilles from afar. Or we might think it foolish to shoot from
the tower rather than help rally flagging Trojan spirits down at the gate, or unwise to
shoot at the nearly invincible Achilles rather a more vulnerable Greek soldier. But set-
ting aside such evaluation that places Paris’s shot in a broader critical context, be it
moral, political, or strategic, we can assess it more narrowly as an archery shot in battle,
as a martial shot qua shot.
First, we might assess whether it attained its aim, that is, hit its target, Achilles’s vul-
nerable heel. Here it scores perfectly. In virtue of this, call it accurate.
Second, we might assess it for skillfulness, that is, whether it manifests relevant
archery skill or competence. Here our myth-makers omit the pertinent detail, so we
may feel free to add details as we like. The son of a wealthy and powerful king, Paris no
doubt received expert training. He also had a decade of experience with the bow in
battle, warding off Greek invaders. So let’s say his shot manifested competence. In vir-
tue of this, call it adroit.
Third, we might assess whether it was accurate because it was competent. Here again
our myth-makers are silent, so let’s add some detail. Apollo had orders from Zeus to:
guide the arrow into Achilles’s ankle just in case, and precisely because, Paris shoots
competently. Thus Paris’s display of competence figures crucially in explaining why the
shot is accurate. His shot is accurate because competent. In virtue of this, call it apt.
Fourth, we might assess whether the shot’s outcome—the arrow’s striking Achilles’s
heel—manifests Paris’s competence. A competence is a special type of power or dispo-
sition, and we almost certainly cannot specify in general when an outcome manifests a
disposition. But case by case we find it relatively easy to tell, and in this case the out-
come clearly doesn’t manifest Paris’s competence. It manifests Apollo’s competence at
guiding the arrow’s flight through the tangled and shifting mass of humanity. (Should
that verdict not strike you as obviously correct, we can add that as the arrow flew, a
Trojan chariot unexpectedly and improbably tumbled, interposing itself between the
arrow and Achilles’s heel, whereupon Apollo immediately altered the arrow’s flight to
compensate for the unforeseen obstacle.) As one modern teller of the tale puts it,
“Apollo,” not Paris, “killed Achilles.”1 In virtue of this, Paris’s shot fails to be adept. Had
the outcome manifested his competence, the shot would have been adept.
Ernest Sosa (2007) introduced the triple-A performance-assessment structure of
accuracy, adroitness, and aptness. Accurate performances achieve their aim, adroit
performances manifest competence, and apt performances are accurate because
adroit. To that I add the fourth category, adeptness.2 An adept performance is one
1
Davidson (2008). Davidson recounts a different version of Achilles’s death, which the medievals attrib-
uted to the ancient Trojan cleric Dares Phrygius, wherein Achilles dies in an Apollonian temple after killing
Troilus and Hector. But the central point remains the same in that version, namely: Apollo guides Paris’s
shot into Achilles’s heel. My version of Achilles’s death hews closer to Rosemary Sutcliff ’s (1993) retelling.
2
All of these statuses are a matter of degree, but the details needn’t detain us here. Sosa might intend
‘aptness’ to name what I call ‘adeptness’, though the official characterization falls short of it. In any event,
we might want a name for aptness as described in the text.
126 John Turri
whose accuracy manifests competence (as opposed to being accurate merely because
competent). Call an adept performance’s outcome an achievement.
In a narrow sense, any accurate performance succeeds, be the outcome good, bad, or
indifferent. The assassin succeeds if he fatally wounds the victim, though he deserves
no plaudits, only condemnation, for this deplorable outcome. The despotic cabal suc-
ceeds if its activities establish a despotism, another deplorable outcome. Labeling some
performance a ‘success’ here means only that it is accurate.3 It implies no approval of
either the aim, outcome, or activity. Similarly, labeling some performance or outcome
an ‘achievement’ doesn’t imply approval. Many achievements ought never have come
to pass. (If such positive connotations unfailingly accompany ‘achievement’ in your
idiolect, please substitute whichever works best of ‘feat’ or ‘deed’.) Consistent with
those general cautionary points about the relation between success, achievement, and
approval, some types of success or achievement might be inherently good (or bad),
and recognizing them as such would imply that they merit approval (or disapproval).
2. The Achievement Account of Knowledge

On one influential contemporary view, we best understand knowledge as a cognitive
achievement (Sosa 2007, Riggs 2009, Zagzebski 2009, Greco 2010). This view treats
belief-formation as a cognitive performance whose aim is truth. Accurate beliefs are true.
Adroit beliefs manifest cognitive competence, including discriminating perception,
cogent reasoning, and good memory. Apt beliefs are true because competent. Finally, your
belief that Q is adept just in case your truly believing Q manifests your cognitive compe-
tence. Adept beliefs are cognitive achievements and are identified with knowledge. Call
this the achievement account of knowledge, or AA for short (pronounced “double A”).
AA promises at least three outstanding benefits. First, it helps explain the added
value of knowledge over mere true belief, a question which has been central to episte-
mology ever since Plato’s Meno (Riggs 2002, Kvanvig 2003, Zagzebski 2003, Pritchard
and Turri 2011). We value achievement over mere success, at least when the relevant
outcome is unobjectionable or better. For instance, compare a novel whose enthralling
plot manifests the author’s literary genius to a novel whose enthralling plot is due to a
monkey randomly striking a keyboard repeatedly. We value the former more than the
latter. Or compare a dance whose visual beauty manifests the dancer’s skill to a dance
whose visual beauty is due to a series of fortuitous stumbles. The former is better than
the latter. Likewise, compare a stretch of inquiry whose eventual true conclusion man-
ifests the inquirer’s intellectual skill to one whose true conclusion is due to a series of
lucky guesses. The former is better than the latter.
3
Distinguish this from the performance’s succeeding in any of the broader senses mentioned earlier, or
satisfying any ulterior aims the agent has in performing. Failing in the narrow sense might be best overall
or best serve the agent. For instance, she might engage her enemy in a game of chess to distract him, which
goal is best served by playing poorly and setting herself up to be easily checkmated. The outcome is a
chess-failure yet, relative to the agent’s own goals, a smashing success.
Second, AA places epistemic evaluation in a familiar pattern (Sosa 2007, Greco

2007a: 57–8; see also Turri 2006, 2010, 2011c, 2011d). Whether it’s art, athletics,
oratory, or inquiry, we’re keen to assess how outcomes relate to the relevant skills
and abilities. The basic model of performance assessment applies across the entire
range of evaluable rational activity. Knowledge takes its place as an intellectual
achievement, fundamentally similar to artistic, athletic, political, and other sorts of
achievement.
Third, AA solves the Gettier problem.4 Consider this standard Gettier case.
(HUSBAND) Mary enters the house and looks into the living room. A familiar appearance
greets her from her husband’s chair. She thinks, “My husband is sitting in the living room,” and
then walks into the den. But Mary misidentified the man in the chair. It’s not her husband, but
his brother, whom she had no reason to think was even in the country. However, her husband
was seated along the opposite wall of the living room, out of Mary’s sight, dozing in a different
chair. (Adapted from Zagzebski 1996: 285–6)
HUSBAND follows the standard recipe for generating Gettier cases. Start with a belief
sufficiently justified to meet the justification requirement for knowledge. Then add an
element of bad luck that would normally prevent the justified belief from being true.
Lastly add a dose of good luck that “cancels out the bad,” so the belief ends up true
anyhow. It has proven difficult to explain why this “double luck” prevents knowledge
(Zagzebski 1994). AA diagnoses the problem as follows. The outcome—in this case,
Mary’s true belief that her husband is home—doesn’t manifest Mary’s intellectual
skills, so Mary doesn’t know that her husband is home. This verdict fits seamlessly into
a more general pattern of manifestation failure, wherein the relevant outcome fails to
manifest the relevant power or ability (Turri 2011a, 2012a, 2013).
I emphasize that this is not the place to argue at length that AA delivers on these bene-
fits, a task accomplished in numerous venues by many theorists over the last decade and
more. For present purposes, I assume that AA does deliver on those benefits. I’ve explained
the basic reasons for thinking that it does, so we can later appreciate how the alternative
versions of performance-based epistemology, introduced over the next couple sections,
can likewise deliver on those benefits, while avoiding objections that plague AA.
3. The Ample Achievement Account of Knowledge

Duncan Pritchard (2008: 445–6) rejects AA, arguing that cognitive achievement does
not suffice for knowledge. Achievements can be unsafe. But knowledge cannot be
unsafe. So knowledge is not (essentially) an achievement.5
4
Recent empirical work on Gettier cases should inform our view of what exactly the Gettier problem
amounts to and what the best solution to it is. For example, see Starmans and Friedman (2012);
Turri (2013); Nagel et al. (forthcoming); Turri, Buckwalter, and Blouw (2015); Blouw, Buckwalter, and
Turri (forthcoming).
5
Says Pritchard (2008: 445), “knowledge seems to be intolerant to luck in a way [that] achievements are
not … .” Pritchard pairs this objection with another, arguing, following Lackey (2007), that achievement is
128 John Turri
Suppose a performance succeeds. We can ask how easily it might have failed. If the
correct answer is ‘not easily’, then the performance is safe. More generally, a perfor-
mance is safe just in case it (i) succeeds and (ii) not easily would have failed. A belief is
safe just in case it (i) is true and (ii) not easily would have been false.6
To appreciate Pritchard’s critique, let’s start by considering this case:
(FORCE FIELD) Fielding is participating in an archery competition. He selects one target out
of hundreds to shoot at. Fielding is focused, the wind perfectly still. He nocks the arrow,
assumes his stance, takes aim and skillfully releases the shot, which glides unerringly into the
heart of the stationary target. Bulls-eye. (Adapted from Pritchard 2008: 445)
This counts as an achievement. Adding the following details does not disrupt the
intuition:
(FORCE FIELD cont’d) Fielding had randomly selected the only target on the range lacking a
hidden force-field apt to repel any incoming arrow.
Pritchard concludes that Fielding’s bullseye is an achievement despite being “clearly

lucky, in the sense that [he] could very easily have” missed (Pritchard 2008: 445).
Pritchard then compares FORCE FIELD to,
(BARN) Henry and his son are driving through the country. Henry pulls over to stretch his
legs, and while doing so regales his son with a list of currently visible roadside items. “That’s a
tractor. That’s a combine. That’s a horse. That’s a silo. And that’s a fine barn,” Henry added,
pointing to the nearby roadside barn. It was indeed a fine barn Henry saw. But unbeknownst
to them the locals recently secretly replaced nearly every barn in the county with fake barns.
Henry happens to see the one real barn in the whole county. But had he instead set eyes on any
of the numerous nearby fakes, he would have falsely believed it was a barn. (Adapted from
Goldman 1976: 172–3)
Pritchard notes that BARN exhibits the same structure as FORCE FIELD, so consist-
ency requires us to count Henry’s true belief as an achievement too. But Henry’s belief
doesn’t count as knowledge. More generally, adept beliefs can be unsafe whereas
knowledge cannot, ruining the identification of knowledge with adept belief.
AA’s proponents can respond in at least three ways. First, they might argue that
Henry does indeed know, so the disanalogy dissipates (Turri 2011a, 2012a; Turri,
Buckwalter, and Blouw 2015; Blouw, Buckwalter, and Turri, forthcoming). I think this
response reflects the way we ordinarily view knowledge and is, moreover, correct upon
reflection, but I will grant for the sake of argument that Henry doesn’t know. Second,
they could argue that whereas other achievements can be unsafe, knowledge cannot.
While not obviously implausible—knowledge might just be special in this way—this
not necessary for knowledge either. But this latter worry has been repeatedly adequately addressed else-
where (see Greco 2007a, Riggs 2009, and Turri 2011a).
6
“Possible worlds” terminology provides a convenient intuitive gloss on ‘not easily’. In this idiom, it’s
true that some event would not easily occur (in certain conditions) just in case, at all or most nearby worlds
(where those conditions obtain), it does not occur. Here I don’t commit myself to the adequacy of this
idiom.
response runs the risk of appearing ad hoc. Better would be, third, to identify a class of
achievements that must be safe and contend that knowledge patterns with these. This
deviates from AA but still identifies knowledge with a type of achievement. I will now
develop this third response.
Begin with a natural extension to our theory of performance-assessment.
Performances have a quintuple-A structure. Add amplitude to the four previously
mentioned. A performance is ample just in case its safety (not just its accuracy) mani-
fests the agent’s competence.7 Your belief is ample just in case its safety (not just its
truth) manifests your competence. Ample performances form a proper subset of adept
performances. Call the outcome of an ample performance an ample achievement, and
call the view that knowledge is ample belief the ample achievement account of knowl-
edge or AA+ for short (pronounced “double A plus”).
(Let’s define a further property—call it supersafety—that a performance has just in
case it (i) succeeds and (ii) couldn’t have failed. In line with the current suggestion, we
might add that a performance is adamantine just in case its supersafety (not just its
safety) manifests the agent’s skill. It’s noteworthy that the underlying performance-based
model extends to accommodate the ultra-Cartesian view that knowledge is adamantine
belief. In unpublished work, Ernest Sosa proposes a fascinating reading of Descartes
along these very lines. Many statuses betwixt safety and supersafety are possible.)
AA+ shares the benefits of AA. First, AA+ can explain the added value of knowledge
over mere true belief. We value ample achievement over mere success, at least when
the relevant outcome is unobjectionable or better. Second, it places epistemic evalua-
tion in a familiar pattern, that of ample achievement. Compare adversarial competi-
tion. In line with our theory of performance-assessment, when victory is the outcome,
we may ask how it relates to the agent’s performance. Victory manifesting your skill
counts as an achievement. But not every victory is an overwhelming victory, which can
also be an outcome of competition. To overwhelm an opponent, you must not only
succeed, but do so by a wide margin, as was the case with Usain Bolt’s crushing victo-
ries in the men’s 100- and 200-meter sprints at the 2008 Olympics. (Compare that to
when Michael Phelps won the men’s 100-meter butterfly by one-hundredth of one sec-
ond over Milorad Cavic at the 2008 Olympics.) An advocate of this third response,
then, might liken knowledge to overwhelming a fact. Third, AA+ solves the Gettier
problem exactly the same way as AA, namely, by pointing out that the relevant out-
come fails to manifest the agent’s relevant ability.
4. The Attainment Account of Knowledge

The last section entertained the objection that cognitive achievement isn’t sufficient for
knowledge. This section entertains the objection that it isn’t necessary for knowledge.
7
Greco (2007b) might advocate this type of response when he suggests that knowledge requires “agent-
based safety.” He doesn’t explain what agent-based safety is.
130 John Turri
Start by noticing something I didn’t emphasize earlier in section 3. AA’s proponents

claim that knowledge is an achievement that proceeds from select intellectual disposi-
tions. Sosa (2007: 29) says knowledge must be produced by a competence, which is a
disposition “that would in appropriately normal conditions ensure (or make highly
likely) the success of any relevant performance issued by it.” Greco (2002: 308) identi-
fies knowledge as true belief manifesting “reliable cognitive abilities or powers.” Linda
Zagzebski (1996: II.4.1.2 and III.2) identifies knowledge as true belief manifesting
intellectual virtue, where a virtue is an entrenched character trait enabling “reliable
success.” To settle on convenient terminology, I’ll say these theorists claim that knowl-
edge-producing dispositions must be normally successful.
(As I use the term here, ‘disposition’ denotes the genus that includes as species abili-
ties, from the maximally reliable down to the extremely weak. Weak abilities are unre-
liable but not completely useless. I understand competences, skills, powers, excellences,
etc., as types of ability.)
Here we encounter a potential problem for AA. Achievement does not generally
require normally successful dispositions. Ted Williams is the best baseball hitter ever.
At least often when he got a hit, he performed adeptly. But he normally failed to get a
hit: the relevant ability could at best be counted on to produce a hit about four in ten
times (his best yearly average ever was .407). And many lesser hitters sometimes per-
form adeptly too, despite normally failing seventy-five percent of the time or more. No
disposition that normally fails is normally successful.
Consider also this case.
(BABY STEPS) Geno is eleven months old. He’s daily been gaining confidence in his ability to
balance on two feet. Today he stood holding onto the couch when an object in Dad’s hand
commanded his attention. Inspired, Geno let go and, to his parents’ delight, took his first two
steps in Dad’s direction.
Those first steps were an achievement. They manifested Geno’s blossoming bipedal-
ism. But of course he’s still a highly unreliable walker. (At least, he was unreliable when
those first steps occurred.) So achievement can issue from even highly unreliable
ability.
Baby steps are not unique. Achievements populate the road to proficiency in many
spheres. A child’s first grammatical sentence manifests her incipient linguistic ability.
A rookie golfer’s first par manifests his incipient putting skill. A pleasing chord mani-
fests a novice musician’s incipient musical ability. An undergraduate’s essay might even
manifest incipient compositional skill. Even though their authors could not reliably
produce such results, that first sentence, first par, first pleasing chord, and first thor-
oughly grammatical essay are achievements (the last of which is unfortunately far too
rare).
Reflection shows that this result is expected. Outcomes often manifest unreliable
dispositions in mundane cases. A car’s starting might manifest an unreliable starter’s
power. A room’s being illuminated might manifest an unreliable flashlight’s power.
A candle’s burning might manifest an unreliable lighter’s power. A door’s opening

might manifest an unreliable sensor’s power. Such outcomes are possible even if unex-
pected from unreliable items.
This raises a question for AA (Turri 2012b). If achievements don’t generally require
normally successful dispositions, then why insist that knowledge must proceed from
normally successful dispositions? Does not that restriction threaten to rule out too
much?
But wait. We’ve seen that success can manifest unreliable ability. I called those out-
comes “achievements,” said that “achievements” populate the road to proficiency in
many spheres, and posed the challenge to AA in terms of “achievements.” Yet given the
way we earlier characterized ‘achievement’, AA’s proponents might object. You per-
form adeptly just in case the successful outcome manifests your competence.
Competence requires more than weak, unreliable ability. (The same goes for other cat-
egories AA theorists tend to favor, such as excellence and virtue.) This fair point forces
me to reformulate the question, which I’ll now do.
We first need a term for performances whose success manifests reliable or unreliable
abilities. Call such a performance adequate. All adept performances are adequate, but
not vice versa. We also associated ‘achievement’ with the outcome of adept perfor-
mance. We need a term for the outcome of adequate performance. Call them attain-
ments. All achievements are attainments, but not vice versa. Call attainment that is not
also an achievement a mere attainment. Finally, call the view that identifies knowledge
with intellectual attainment the attainment account of knowledge or AA– for short
(pronounced “double A minus”).
Now let’s reformulate our question. Why identify knowledge with intellectual
achievement rather than attainment? Why prefer AA over AA–? Why exclude mere
intellectual attainments? What’s wrong with an intellectual performance as adequate
as a typical Ted-Williams double?
These are important questions, which AA’s proponents ought to have answers to.
But for present purposes, I want to concentrate on something else: even if it turns out
that AA– is preferable, we’re still left with a version of performance-based epistemol-
ogy that delivers on all three benefits mentioned earlier. First, AA– can explain the
added value of knowledge over true belief. We value attainment over mere success, at
least when the relevant outcome is unobjectionable or better. Second, AA– places epis-
temic evaluation in a familiar pattern, namely, that of attainment. Third, AA– can solve
the Gettier problem the same way AA does, by pointing out that the relevant outcome
fails to manifest the relevant disposition.
5. Knowledge and Truth

Suppose you think it is possible to know false things. You think that even if Q is false,
it’s still possible to know Q, so long as it is approximately true. So you’ll think that AA is
too strong, but not for the same reason as proponents of AA–. To accommodate this
132 John Turri
intuition, we need only adjust the admissible outcomes. Call a performance that fails
but nearly succeeds approximate. Call an approximate performance’s outcome an
approximation. Now we can simply append a disjunct to one of the earlier definitions
of knowledge. You know Q just in case your believing Q is either an achievement
(attainment) or an approximation.8 Call this the approximation (or better) account of
knowledge.
Or suppose you think that knowledge doesn’t even require approximate truth.
Rather, you think completely false beliefs can be knowledge, so long as they’re reliably
produced.9 Performance-based epistemology can accommodate this intuition as well.
Adroit beliefs manifest competence, and competence must be reliable. But adroit
beliefs can be false. So we can understand knowledge as adroit belief. Call this the
adroit-belief account of knowledge.
6. Conclusion
Further variations on the basic theme are possible. But I’ve said enough to clarify how
to manipulate the model. You can strengthen or weaken the admissible outcomes.10 Or
you can strengthen or weaken the admissible dispositions. Or you can even adjust
both. The relation of manifestation is common to all variants. Performance-based
epistemology is a substantive, powerful, and versatile approach. The objections to AA
canvassed here shouldn’t convince us to abandon performance-based epistemology
and its considerable benefits just yet. To do so would be to throw the proverbial baby
out with the bath water.11
8
More fully spelled out: you know Q just in case either (a) your truly believing Q manifests your
cognitive competence (ability) or (b) your approximating Q manifests your cognitive competence
(ability).
9
Again, recent empirical work should inform theoretical work on this question, at least to the extent
that the theoretical work purports to describe knowledge ordinarily understood. See Buckwalter forth
coming; Buckwalter and Turri under review. See also Hazlett 2010, Turri 2011b, and Hazlett 2012.
10
One objection not considered here is that knowledge doesn’t require belief, in which case the domain
of admissible outcomes must include some other sort of representational state or cognitive pro-attitude.
Going back to Radford (1966), some contemporary philosophers have questioned whether knowledge
entails belief. Recent experimental work has re-ignited that old theoretical debate (Myers-Schulz and
Schwitzgebel 2013; Murray et al. 2013), suggesting that knowledge ordinarily understood does not require
belief. However, further empirical work has uncovered important nuances in the folk psychology of belief,
and the balance of evidence seems to support the view that knowledge ordinarily understood does require
belief in one form or another (see Rose and Schaffer forthcoming; Buckwalter, Rose and Turri 2013; Turri
and Buckwalter, forthcoming; Rose, Buckwalter, and Turri 2014). Thus it seems that the envisioned objec-
tion to the performance-based approach can be overcome.
11
This research was kindly supported by the Social Sciences and Humanities Research Council of
Canada, The Character Project at Wake Forest University and the John Templeton Foundation (neither of
which necessarily endorses any opinion expressed here), and an Ontario Early Researcher Award.
References
Blouw, P., W. Buckwalter, and J. Turri, forthcoming. “Gettier Cases: A Taxonomy.” In R. Borges,
C. de Almeida, and P. Klein (eds) Explaining Knowledge: New Essays on the Gettier Problem.
Buckwalter, W., forthcoming. “Factive Verbs and Protagonist Projection.” Episteme.
Buckwalter, W., D. Rose, and J. Turri, 2013. “Belief through Thick and Thin.” Noûs: 1–28.
doi:10.1111/nous.12048.
Buckwalter, W., and J. Turri, under review. “In the Thick of Moral Motivation.” University of
Waterloo.
Davidson, J., 2008. “Troilus and the Death of Achilles.” Guardian (London), 17 January. Part of
“The Greek Myths” series.
Greco, J., 2002. “Virtues in Epistemology.” In Paul Moser (ed.) The Oxford Handbook of
Greco, J., 2007a. “The Nature of Ability and the Purpose of Knowledge.” Philosophical Issues 17:
57–69.
Greco, J., 2007b. “Worries about Pritchard’s Safety.” Synthese 158: 299–302.
Greco, J., 2010. Achieving Knowledge. Cambridge: Cambridge University Press.
Hazlett, A., 2010. “The Myth of Factive Verbs.” Philosophy and Phenomenological Research
80(3): 497–522.
Hazlett, A., 2012. “Factive Presupposition and the Truth Condition on Knowledge.” Acta
Analytica 27(4): 461–78.
Kvanvig, J. L., 2003. The Value of Truth and the Pursuit of Understanding. Cambridge: Cambridge
University Press.
345–61.
Murray, D., J. Sytsma, and J. Livengood, 2013. “God Knows (But Does God Believe?).”
Philosophical Studies 166(1): 83–107.
Myers-Schulz, B., and E. Schwitzgebel, 2013. “Knowing that P Without Believing that P.” Noûs.
47(2): 371–84.
Nagel, J., V. San Juan, and R. A. Mar, forthcoming. “Lay Denial of Knowledge for Justified True
Beliefs.” Cognition.
Pritchard, D., 2008. “Greco on Knowledge: Virtues, Contexts, and Achievements.” Philosophical
Quarterly 58 (232): 437–47.
Pritchard, D. and J. Turri, 2011. “The Value of Knowledge.” Stanford Encyclopedia of Philosophy
(Nov. 2011) <http://plato.stanford.edu/entries/knowledge-value> accessed July 7, 2015.
Radford, C., 1966. “Knowledge—By Examples.” Analysis 27: 1–11.
Riggs, W., 2002. “Reliability and the Value of Knowledge.” Philosophy and Phenomenological
Research 64(1): 79–96.
Riggs, W., 2009. “Two Problems of Easy Credit.” Synthese 169(1): 201–16.
Rose, D., and J. Schaffer, forthcoming. “Knowledge Entails Dispositional Belief.” Philosophical
Studies.
Rose, D., W. Buckwalter, and J. Turri, 2014. “When Words Speak Louder than Actions: Delusion,
Belief and the Power of Assertion.” Australasian Journal of Philosophy 92(4): 683–700.
University Press.
134 John Turri
Starmans, C., and O. Friedman, 2012. “The Folk Conception of Knowledge.” Cognition 124(3):
272–83.
Sutcliff, R., 1993. Black Ships before Troy: The Story of The Iliad. New York: Laurel-Leaf.
Turri, J., 2006. “Epistemic Reasons.” PhD dissertation, Brown University.
Turri, J., 2010. “On the Relationship Between Propositional and Doxastic Justification.”
Philosophy and Phenomenological Research 80(2): 312–26.
Turri, J., 2011a. “Manifest Failure: The Gettier Problem Solved.” Philosophers’ Imprint 11(8):
1–11.
Turri, J., 2011b. “Mythology of the Factive.” Logos & Episteme 2(1): 143–52.
Turri, J., 2011c. “Believing for a Reason.” Erkenntnis 74(3): 383–97.
Turri, J., 2011d. “The Express Knowledge Account of Assertion.” Australasian Journal of
Philosophy 89(1): 37–45.
Turri, J., 2012a. “Is Knowledge Justified True Belief?” Synthese 184(3): 247–59.
Turri, J., 2012b. Review of John Greco, Achieving Knowledge: A Virtue-Theoretic Account of
Epistemic Normativity. Mind 121(481): 183–7.
Turri, J., 2013. “A Conspicuous Art: Putting Gettier to the Test.” Philosophers’ Imprint 13(10):
1–16.
Turri, J., forthcoming. “Sustaining Rules: A Model and Application.” In J. A. Carter, E. C.
Gordon, and B. Jarvis (eds) Knowledge First. Oxford: Oxford University Press.
Turri, J., and W. Buckwalter, forthcoming. “Descartes’s Schism, Locke’s Reunion: Completing
the Pragmatic Turn in Epistemology.” American Philosophical Quarterly.
Turri, J., W. Buckwalter, and P. Blouw, 2015. “Knowledge and Luck.” Psychonomic Bulletin &
Review 22(2): 378–90.
Zagzebski, L., 1994. “The Inescapability of Gettier Problems.” Philosophical Quarterly 44(174):
65–73.
Zagzebski, L., 1996. Virtues of the Mind: An Inquiry into the Nature of Virtue and the Ethical
Foundations of Knowledge. Cambridge: Cambridge University Press.
Zagzebski, L., 2003. “The Search for the Source of Epistemic Good.” Metaphilosophy 34: 12–28.
Zagzebski, L., 2009. On Epistemology. Belmont, CA: Wadsworth.
PA RT I I
Applications and Problems
8
Intuitions and the Understanding
Paul A. Boghossian
Introduction
In this essay, I take for granted that intuitions play a significant role in providing a pri-
ori justification within various domains, including philosophy itself.1 I focus on the
question of what intuitions are, and how they might be able to supply the justification
at issue. I will develop my account partly by contrasting it with Ernest Sosa’s well-
known views on these matters.
For concreteness, let us begin by looking at the role of intuitions in thought experi-
ments and, in particular, in the famous case of Twin Earth, as described by Hilary Putnam.2
In putting forward his Twin Earth thought experiment, Putnam was concerned to
refute a view that he (appropriately) called the “Received View of Meaning.” He took it
to consist in the conjunction of the following two claims:
(1) The meaning that a speaker associates with a word is determined by individual-
istic facts about that speaker,
and,
(2) The meaning that a speaker associates with a word determines that word’s
extension.
In order to refute the Received View, Putnam described a possible world in which, in
another part of our galaxy, there is a Twin Earth, a planet as much like Earth as possi-
ble, except that the liquid that flows through its lakes, rivers, and faucets, although
superficially quite similar to water, is not H2O but XYZ.
On Earth, we suppose there to be a user of the word ‘water,’ Oscar, who, while not
knowing anything about water’s chemical composition, is nevertheless perfectly com-
petent at using it. While on Twin Earth, there is a molecule for molecule duplicate of
Oscar’s, Toscar, who also counts as a competent user of ‘water’ by the relevant standards,
1
For arguments in favor of the view that I here take for granted, see Boghossian (forthcoming).
2
See Putnam (1975). The reasons for choosing this example are largely historical. Nothing I say should
depend on the specifics of this particular case.
138 Paul a. Boghossian
although he, too, does not know anything about the chemical composition of the
local liquid.3
Putnam asks whether the extension of Oscar’s word ‘water’ is the same as Toscar’s.
He believes that, when we reflect on this question, we will see clearly that, in the sce-
nario he described, Oscar’s word ‘water’ will have H2O in its extension but not XYZ;
while Toscar’s word ‘water’ will have XYZ in its extension but not H2O.4
If Putnam were right about this judgment, the Received View, which predicts that
the extensions would be the same, would stand refuted. And, since the Received View
is enmeshed with a large number of other important and well-entrenched claims—
claims involving privileged access, psychological explanation, and the relation
between meaning and reference—the implications of that refutation would be
dramatic.
Accepting Putnam’s judgment, then, threatens to upend many central and seem-
ingly well-supported philosophical theories. Given this fact, his judgment had better
be pretty strongly justified if it is to be accepted: we don’t want to rewrite large
swathes of the philosophy of mind, the philosophy of language, and epistemology, if
Putnam’s judgment is just something that he is inclined to believe, without adequate
justification. Clearly, Putnam believes his judgment to be strongly justified; and so
does each of us when we arrive at the same intuitive judgment about this particular
case.
The question is: What could be the justification for this judgment about this highly
contrived and specialized case, far removed from any actual scenario that we have pre-
viously encountered or thought about? How do we come to know what would be true
in such remote and unusual scenarios?5
In the earlier work alluded to above, I argued that we could not hope to explain the
source of the justification for this and other thought experimental judgments simply
by appeal either to perception (broadly understood to include sensory experience,
memory, introspection, and testimony) or to the understanding—our conceptual
competence with the relevant concepts. Rather, I argued, we must make appeal to the
traditional notion of intuition, understood as a species of intellectual seeming or
insight.
According to this traditional idea, in reflecting on Putnam’s scenario, and on his
question, we have an intuition to the effect that Oscar’s and Toscar’s tokens of ‘water’
have non-overlapping extensions, and it is that intuition that leads us to, and that justi-
fies, the corresponding Twin Earth judgment, and thereby to a rejection of the Received
View.
3
As usual, we set aside the complication that we ourselves are largely made up out of water.
4
The important claim here is not that XYZ is in the extension of Toscar’s ‘water.’ That much we may take
to have been stipulated by Putnam (though it needn’t have been). The important claim is that H2O isn’t in
that extension (and vice-versa for Oscar’s word ‘water.’)
5
There is also a question about how we know that the scenario that Putnam describes is metaphysically
possible. For the sake of simplicity, I will ignore this question for now.
Intuitions and the Understanding 139
A crucial feature of this justification, as I have been emphasizing, is that it must be

strong enough to trump the justification we previously had for (what is often) a large
number of central beliefs that conflict with it. It must be able to justify our tossing out
those well-entrenched conflicting beliefs, or, at the very least, to justify our reconsider-
ing them.
Any intuition-friendly view of a priori justification confronts three urgent questions:
(I) What are intuitions?
(II) Precisely what kind of justification do they supply?
(III) How exactly do they supply the justification that is claimed for them?
Sosa on Intuitions
In a series of influential and important essays, Sosa has developed non-skeptical
answers to these central questions, answers that may be summarized in the following
theses:6
( S-I) An intuition that p is an intellectual seeming that p, which is itself to be under-
stood as consisting in an attraction to assent to p that is based merely on an under-
standing of p.
(S-II) Intuitions supply a foundational and a priori justification for propositions.
(S-III) An intuition that p is able to justify the belief that p if and only if it is explained by
a competence to tell, on the basis of one’s understanding of p alone, whether p is true.
Since I agree with Sosa on (S-II), I will simply assume it in what follows. My focus will
be on assessing his picture of what intuitions are (S-I), and of how they justify (S-III).
In particular, I will be interested in whether the understanding deserves the role that it
is accorded by Sosa’s account.7
The Nature of Intuitions and the Understanding

We may break up Sosa’s account of the nature of intuitions, (S-I), into two theses:
(S-Ia) Intuitions are intellectual seemings.
(S-Ib) Intellectual seemings that p are attractions to assent to p that are based on
one’s understanding of p alone.
I agree that intuitions are intellectual seemings. But why should we go along with Sosa’s
other claim—that intellectual seemings are attractions to assent that are based on one’s
understanding of the relevant proposition alone? Why attractions to assent? And:
Why based on the understanding alone?
6
See Sosa (2007: 45) and other essays (1998, 2006, 2009).
7
The discussion in this essay complements and, in some respects, supersedes my earlier discussion of
Sosa’s views on these topics—see Boghossian (2009).
The view that intellectual seemings are to be understood as a species of attraction to

assent is not without motivation. In its absence, intellectual seemings can seem myste-
rious. For what are they to be?
We can agree, of course, that an intellectual seeming is a kind of seeming. And a
seeming, we can say, is a conscious mental state that presents a proposition as true. It’s
a conscious mental state that presents the world as being so, or perhaps even as being
necessarily so. But what makes such a seeming an intellectual seeming?8
The intended contrast is with sensory seemings, seemings that some sense organ or
other delivers, and that are individuated by their characteristic phenomenologies. But
how are intellectual seemings individuated? Do they, too, have a characteristic
phenomenology?
This is an area of some controversy. Many philosophers balk at talk of states with a
special intellectual phenomenology. Many philosophers report that they can detect no
such states in themselves when they contemplate thought experiments and respond
with intuitive judgments. But what then is an intellectual seeming to be, if it is not to be
picked out by its distinctive phenomenology?
In response to this problem, Sosa’s answer can seem quite compelling: an intellectual
seeming that p is an attraction to assent to p. This proposal identifies an intellec-
tual seeming with something we appear to have an independent purchase on—namely,
assent. And it approximates the pre-doxastic character of a seeming by equating it not
with actual assent, and not even with a disposition to assent, but merely with an attrac-
tion to assent. I can be attracted to assent to the proposition that the Müller-Lyer lines
are unequal in length, even while not being disposed to assent to that proposition (as a
result of knowing that they are in fact equal in length).
But once we remove the mystery of what intellectual seemings are by equating them
with attractions to assent, we face another problem. Intellectual seemings can’t just be
any old attractions to assent, as the example used above shows, since such an attraction
to assent to p might be based on one’s sensory experience of p; and such an attraction
would obviously not be an intellectual seeming.
This observation motivates Sosa’s (S-Ib): an intellectual seeming is not any old
attraction to assent to p, but one that is based on one’s understanding of p alone. That
restriction seems to give us a sufficiently distinctive account of what an intellectual
seeming is, one that distinguishes it from a mere sensory seeming, since sensory seem-
ings are attractions to assent that are not based on the understanding alone but are
partly based on sensory experience.
So, a compelling consideration in favor of an understanding-based view of the
nature of intuitions is that it appears to give us a prima facie plausible account of what
they are, one that distinguishes them in what appears to be the right way from mere
sensory seemings.
8
I follow John Bengson (2015) in emphasizing the importance of presentations (as opposed to rep-
resentations) as sources of epistemic justification. Unlike Bengson, though, I see no daylight between pres-
entations and seemings and use them pretty much interchangeably.
Understanding and A Priori Justification

The understanding also figures in Sosa’s account of the justificatory power of intuitions,
(S-III).
According to Sosa, what explains why an intuition is able to a priori justify a belief
that’s based upon it is that the attraction to assent to the relevant proposition derives
from a competence to tell, on the basis of the understanding alone, the true from the
false, in the domain to which the proposition belongs. There are two questions here to
which Sosa may be seen as giving answers:
(A) How could intuitions be the source of any sort of justification?
(B) How could they be the source specifically of a priori justification?
In some sense, there ought not to be too much of a puzzle about why intuitions could
be a source of some justification for belief. For intuitions are a kind of seeming, and,
surely, seemings must be the source, at the very least, of prima facie justification. Where
could rational belief begin if not with how things seem to be? Such justification may be
defeasible by other beliefs that are more strongly grounded, but that concession cannot
interfere with a seeming’s ability to provide prima facie justification.
What I have just said about prima facie justification will be found plausible by those
with a generally internalist orientation in the theory of justification. But those philoso-
phers, like Sosa, for whom justification involves the reliability of the mechanism by
which belief is fixed, will not be satisfied. They will insist that it is a condition on intui-
tion’s being able to supply justification that it be a reliable source of truth about the
relevant domain. And they will want to know how our having an intuitive response to a
remote possible world scenario could possibly be a reliable source of truth about the
modal facts? How, to return to the Twin Earth example, they might ask, could we relia-
bly know, just by reflecting on the scenario in question, what the extensions of words
would have been on Twin Earth?
The demand for an explanation of intuition’s reliability about modal facts may be felt
even by those who accept a broadly internalist view of justification. Some philosophers
may feel that, regardless of whether reliability is a condition on justification, there had
better be a scientifically respectable explanation of how a given epistemic faculty could
be a reliable source of truths about a particular domain.
Either way, we seem to be faced here with an embarrassing question: what’s the
informational route whereby our intellectual seemings can be a source of reliable views
about abstract or modal truths?
The answer can’t be that there’s a causal channel open between abstract reality and
our cognitive faculties that explains why we’re reliably attracted to assent to truths
about abstracta, on analogy with the way in which sensory perception serves as such a
causal channel to concrete reality. But what’s the alternative?
Here, too, it looks as though play with the understanding provides a potentially sat-
isfying resolution to these worries. If we have a competence to reliably tell, on the basis
of our understanding alone, which modal propositions are true and which false, we
would have a demystifying answer to the potentially embarrassing question about reli-
ability. The ‘causal channel’ in question would be open only between our cognitive
faculties and the information that’s encoded in our concepts; and, presumably, every-
one should be comfortable with the existence of that sort of mechanism.
Finally, (S-III) promises to give a satisfying answer to question (B) as well. If the
competence for reliable modal belief were based upon the understanding alone, that
would explain why the justification involved is a priori, since, by relying on the under-
standing alone, one wouldn’t be relying on sense experience.
Rejecting the Link between the Nature of Intuitions

and the Understanding
I believe that these considerations build a powerful case for deploying the notion of the
understanding in an account of intuition in the way that Sosa does. Nevertheless, I
believe that such accounts are ultimately mistaken. In what follows I will first try to
explain why they are mistaken. I will then turn to defusing the considerations I have
adduced in their favor and suggest an alternative picture.
Let me start with a discussion of (S-I). According to Sosa’s account, a person’s intui-
tion that p is an attraction to assent to p that is epistemically based on that person’s
understanding of p alone.
It is a distinctive feature of views that identify intuition with some species or other of
assent that an intuition can have an epistemic basis, rather than merely serving as one.
A presentational state, such as a visual percept, could serve as an epistemic basis for
assent, but could not itself be epistemically evaluated. An attraction to assent, by con-
trast, in being a species of assent, may be epistemically evaluated and so could itself
have an epistemic basis, as opposed simply to being able to serve as one.
Sosa tells us very little about basing and even less about what it is to grasp or under-
stand a proposition. This is not surprising: these notions have proven to be very diffi-
cult to explicate. But they are, of course, notions that are needed quite generally; and
we are entitled to some confidence that each is in good standing, even in the absence of
a satisfying account.
However, what we are not entitled to be confident about is that when we finally get a
satisfyingly general account of grasp, that it will be clear how mere grasp of p could
serve as an epistemic basis for assenting (or being attracted to assent) to p.
There is one model for this that we (think we sort of) understand reasonably well.
And that is when grasp of p consists in grasp of some sort of explicit definition for p, and
our basis for assenting to p consists in inferring p from its definition.
But this is clearly a very special case—special both in that grasp rarely consists in
grasping an explicit definition, and in that ‘basis’ rarely means inferential basis.
But if the relation between grasp and assent is not inferential, what else could it be?
Do we understand how the relation between grasp and assent could be like the relation
between a perceptual state and the belief that is based upon it? What notion of grasp
would work like that?
Rejecting the Link between Justificatory Power of

Intuitions and the Understanding
Let me turn to the role the understanding is said to play in explaining intuition’s justifi-
catory powers.
As we saw, Sosa’s account of this (S-III) claims that an intuition that p is a potential
provider of foundational a priori justification for the belief that p if and only if it is an
attraction to assent to p that is explained by a competence to tell, on the basis of one’s
understanding of p alone, whether p is true.
Any account of intuitions that gives a central place to the notion of understanding is
likely to seem surprising against the background of the history of analytic philosophy.
For, historically, explanations of a priori justification on the basis of the understanding
alone—explanations in terms of what I have called “epistemic analyticity”—were
thought to be rivals to accounts in terms of intuition.9 Much of the early interest in the
notion of analyticity was fueled precisely by a desire to do the epistemology of a priori
justification without appeal to intuition.10
One immediate puzzle about Sosa’s view concerns the foundational aspect. How
could the justification provided by intuitions be foundational, on this view, when what
is providing the justification is itself epistemically evaluable? That flies in the face of
what foundational justification is.
Since I believe it is correct that intuitional justification is foundational justification, I
am committed to finding a way of thinking about what intuitions are that doesn’t
equate them with a species of assent, but rather construes them on analogy with the
sorts of pre-doxastic presentational states that visual perceptions are.11
A second problem for the account, from my point of view, derives from its reliance
on a reliabilist conception of epistemic justification. I do not pretend to have anything
new to say about this extensively discussed topic. As far as I’m concerned, the standard
objections to such a conception are fatal. A reliabilist conception of justification, in my
view, loses the topic.12
A final objection to (S-III) concerns the reach of such a theory of intuitional justifi-
cation. Sosa makes it clear that he thinks intuitions supply all (or most) of our a priori
knowledge. But if intuitions are constrained to justify only what can be reliably settled
9
For the notion of epistemic analyticity see Boghossian (1996, 1997, 2003).
10
For an account of the rivalry between intuition and analyticity-based accounts of the a priori, see
Creath (1990).
11
Sosa finds important a distinction between ‘‘basis-dependent foundational justification’’ and ‘‘virtue
foundational justification.’’ And he holds that his account is able to supply the latter but not the former. For
my part, I don’t really understand this distinction and recognize only the former variety of foundational
justification.
12
See Cohen (1984), Foley (1985), and BonJour (1985).
on the basis of one’s understanding, then there will be lots of a priori knowledge that
intuitions won’t be able to explain. As a result, the story in terms of intuitions will be, at
best, radically incomplete.
What are some of the propositions that look resistant to explanation by the
understanding alone? I will mention two examples, although there are likely many
others.
The first is the famous color exclusion case. We can establish on the basis of a thought
experiment that nothing can be red and green all over at the same time. But how could
this be accounted for purely conceptually?
Knowledge of the exclusion could not be built into one’s grasp of the concept red,
because one could have red without having green. Mutatis mutandis, it could not reside
in one’s grasp of the concept green.
Hence, at best it could reside only in the joint possession of red and green. But what
could the joint possession of red and green be if not the simultaneous possession of
each?
So it looks as though, in addition to grasping the ingredient concepts, some further
act of insight is required if someone is to be able to get justification for the color exclu-
sion claim from whatever information is encoded in our color concepts.
A second, and arguably more important, case is provided by normative judgments,
whether these are the judgments of morality or those of rationality.
In a Trolley Problem, for example, one might conclude that it is not morally permis-
sible to throw the fat man off the bridge in order to save the lives of the five innocent
victims who are trapped below. But it is not plausible to say that this judgment derives
solely from our understanding of the ingredient concepts.
Hume (1978: book 2, part 3, section 3) famously said:
’Tis not contrary to reason to prefer the destruction of the whole world to the scratching of my
finger.
Perhaps one wouldn’t want to go as far as Hume. However, the kernel of truth in Hume’s
memorable remark is that it is not contrary to our understanding of the ingredient con-
cepts to prefer the destruction of the whole world to the scratching of my finger.
Whatever deficit one would be manifesting in having such a perverse preference, it
wouldn’t be a lack of understanding of what one’s preferences actually are.
This argument is, in a sense, an application of Moore’s Open Question Argument, an
argument that is widely accepted as showing that substantive moral conclusions are
simply not built into our concepts of right and wrong. If someone were to insist that,
according to their concept of ‘good’, the good always involves maximizing happiness, it
seems that it is always coherent to ask: But is that the correct concept of good, the one
that delivers genuinely normative results?
Notice the contrast here with the concept square. If someone said: According to my
concept square, a square always has four sides, it wouldn’t make any sense to ask: But is
that the correct concept square?
By contrast, it seems always to make sense to ask about the correctness of a given
normative concept.13
If this is right, and yet moral judgments are knowable a priori, this shows that the
story in terms of understanding can’t be the whole story. There must be more to a priori
justification of normative claims than is provided by the understanding account.
Intuitions without the Understanding

So we appear to have many reasons for doubting that the understanding can play the
roles that Sosa has assigned it, either in our theory of the nature of intuitions or in our
account of their justificatory powers. Can we do better? I will try here to sketch the
outlines of an alternative view.
We are agreed that an intuition is an intellectual seeming: a state that non-sensorily
presents a proposition as true.
The phenomenon of a proposition’s intellectually seeming true to you is undoubt-
edly real. To use an example of George Bealer’s:
When you first consider one of de Morgan’s laws, [for example,
(3) Not (p and q) is equivalent to (Not p) or (Not q)]
often you draw a blank; after a moment’s reflection, however, something happens: it
now really seems obvious to you.14
Elijah Chudnoff ’s (2013: 50) example can elicit a similarly compelling impression.
(4) Two circles can have at most two common points.
But it is very implausible that its so seeming simply consists in some species or other of
assent. Rather, the phenomenon seems pre-doxastic: it compels assent from you, and
explains why you are attracted to assent to it. It’s not the assent or the attraction to
assent itself.
And it may well be that there is no good reductive account of this phenomenon.
Why should there be, when there are so few good reductive accounts of other
phenomena?
13
An anonymous referee for the publisher raised the question whether a similar question might not arise
for a non-normative concept such as that of species. Someone might insist on a particular concept of species,
but it might still be coherent to ask whether it is the correct concept of species, the one that genuinely carves
nature at its joints. The sense of the question in the two cases seems to me to be distinct. In the case of the
normative concept, we can ask whether any of the things yielded by the concept are actually true. In the case
of a concept about which we can ask whether it ‘‘carves at the joints’’ we are asking not whether the proposi-
tions it embeds are true, but whether they are nomological or counterfactual-supporting.
14
Bealer (1992: 101–2). Some philosophers deny that intuitions are best understood as a species of
seeming. John Bengson, for example, thinks that intuitions should be thought of as presentations, rather
than seemings. The difference between them is supposed to be that while the content of a seeming would
be explicitly available to a subject, the content of a presentation need not be. I believe that, in this sense,
presentations won’t be able to epistemically justify a belief that is based upon them, but I won’t go into this
issue here.
As these examples bring out, in addition to there being conscious episodes in which
a proposition seems true to you, there are conscious episodes in which a proposition
seems obviously true to you, conscious episodes in which a proposition seems neces-
sarily true to you, and, indeed, conscious episodes in which a proposition seems both
things to you at once.
Often, though not always, when we report on having the intuition that p we mean
not merely that it seems true to us that p, but that it seems obvious to us that p is (neces-
sarily) true.
I think this is probably what Gödel had in mind in talking about certain proposi-
tions (he had in mind some of the axioms of set theory) that ‘‘force themselves upon
you’’ when you contemplate them. They are not forced upon you by sensory evidence,
or by following easily from other things that you already recognize yourself to have
reason to believe. Rather, when you contemplate one of these propositions, you cannot
help but have the impression that it is obviously true.15
A Priori Justification without the Understanding

This, then, is the somewhat minimalist account of intuitions that I favor. An intuition
is an intellectual seeming: there need be no further reductive characterization. Such an
intellectual seeming could be one of obvious truth; or one of necessary truth; or just a
seeming of plain truth; or of some combination of these different types. And in all
these flavors, the intuition can come in gradable quality: it can be more or less strong or
more or less vivid.
The question now before us is whether we can account for the justificatory role of
such states without appealing to the understanding.
We have noted at least four features of that justificatory role that need to be
accounted for:
(5) Intuitions are treated as data—that is, they are treated as providing justifiers
that are themselves beyond justification.
(6) Intuitions are regarded as providing a priori justification.
(7) Intuitions are regarded as providing such strong justification that they are capa-
ble of overturning entrenched and highly justified theories.
(8) Intuitions are regarded as a source of reliable truth.
By contrast:
(9) Intuitions are not regarded as infallible or indefeasible.
(10) If p is obvious to someone, it doesn’t follow that everyone else will find p
obvious.
15
Finding p to be obvious seems not only to be an intentional state—a state with propositional c ontent—
it seems to be a relatively sophisticated propositional state. It has epistemic subject matter: it presents a
proposition as worthy of assent. And yet it is not a judgment, but more like a presentation. There are many
difficult questions about how it is possible for us to enjoy such states, questions that I cannot go into here.
(11) If p is obvious, it doesn’t follow that p cannot also be supported by argument.

Intuitions can mislead; and they can be either defeated or supported by further consid-
erations. All that is perfectly compatible with their having features (5)–(8).16
What we need to explain is not why intuitions are infallible, but only why they can be
the source of enormously strong, but defeasible, evidence.
Along with many others, I have emphasized that there is no mystery about why a
seeming, sensory or otherwise, can be the source of some prima facie justification. If
we cannot start with seemings, we cannot start anywhere. If it seems to me that p, I am
prima facie justified in believing that p. That’s how it is with vision. And that’s how it is
with intellectual seemings, too.17
Why should the justification provided by intellectual seemings be thought to be a
priori? The short answer is that when we say that a justification is a priori we mean that
its source does not depend on perception (broadly understood) but may depend on
intuition or conceptual competence.
The trickier question is what epistemologically interesting principle underlies the
distinction illustrated by these lists. That is a much harder question to which I won’t try
to give an answer on this occasion.18
The crucial question right now is whether this view is able to explain why intuitions
are accorded as much weight as they are. How could a deeply entrenched view like the
JTB view of knowledge be overturned on the basis of its simply intellectually seeming to
us that Mr. Smith, in a hypothetical scenario, has a justified true belief but does not know?
Part of the explanation is already contained in the fact that the justification that an
intuition provides is foundational, so not dependent on any other beliefs for its plausi-
bility. But that’s not a full explanation by itself, since justification can both be founda-
tional and, in a given context, quite weak. (For example, with respect to the proposition
that there are some tomatoes in front of me, vision might be thought to supply quite
strong justification, touch somewhat weaker justification, smell perhaps weaker still,
and hearing very little.)
When we get strong justification from our intuitions, I believe that the reason is nearly
always that those intuitions are impressions not merely of truth but of obvious truth.
If I take it that p is obvious, then I’m taking it that any competent, rational person
would be willing to consent to p on no evidence at all.
I may be wrong to take p to be obvious; I may even be wrong to take it to be true. But
in the absence of defeaters, I am prima facie entitled to believe p and to insist that it
should take a great deal to defeat p. That just is what it is to find p obvious.
So, unless we are to question the probative value of finding certain propositions to
be obvious, we seem to have all the explanation we need.
16
Skeptics are fond of pointing to the cases where impressions of obvious truth turned out to be spec-
tacularly wrong—for example, Kant’s (alleged) intuition that physical space had to be necessarily Euclidean.
No one should claim that intuitions can’t mislead.
17
This is true both on a liberal and a conservative view of perceptual justification.
18
See the interesting discussion in Casullo (2003).
This brings us, finally, to the question about reliability. According to the view I am
defending, we have impressions of obviousness that help us reasonably answer such
questions as:
(a) Is the Twin Earth scenario metaphysically possible?
(b) What would the extension of ‘water’ have been on Twin Earth?
And we take it that these impressions of obviousness are reliable. How could we be
justified in taking these intuitions to be reliable?
Even though I am not a reliabilist about justification, I take this question seriously.
Even if reliability is not what makes our judgments justified, evidence of the unreliabil-
ity of a given putative source of justification can undermine the justification that the
source is presumed to provide.
Do we have evidence of the unreliability of our intuitions? Philosophers who have
engaged in empirical work on intuitions say yes. I think this work is flawed, but will not
engage it here.
I am more concerned about an a priori argument that purports to show that our
intuitions could not be reliable because there could be no plausible mechanism in vir-
tue of which they could be reliable.
The worry here is just an application of Hartry Field’s version of Paul Benacerraf ’s
problem about mathematical knowledge, on a Platonistic view of mathematics. We
assume that the subject matter of mathematics is abstract. We also assume that we have
a reliable capacity for forming true mathematical beliefs. But how could we have such a
capacity on a Platonistic view? Wouldn’t such a capacity necessarily involve our being
able to track the mathematical facts? But how could we possibly be doing that, if there
could not, in the nature of things, be a causal channel open between us and the abstract
mathematical facts?
This worry applies equally to facts about logic, modality, epistemic rationality, and
morality. And it raises a quite general concern, which applies equally to intuitions: how
could any capacity we have, including that of intuitions, be a reliable guide to facts in
these domains?
The dialectic here is that the skeptic thinks he has a proof that would show, on com-
pletely a priori grounds, that states of intuition could not reliably track the modal and
abstract facts.
Sosa, as we have seen, deals with this challenge by postulating that the relevant
abstract facts are somehow or other encoded in our understanding of concepts, so all
we have to do is exercise our capacity to say what is in our concepts.
But he doesn’t explain why we should be confident that the relevant facts are encoded
in our concepts; nor how it is that we access what is in them.
In any case, we have seen that even if those problems could be overcome, explana-
tions that lean entirely on our understanding of concepts cannot provide a sufficiently
general explanation of a priori knowledge.
If our reliability cannot be explained entirely in terms of understanding, how could
it be explained? Recall, what’s needed is a ‘‘proof of concept,’’ not an actual detailing of
the mechanisms that underlie our putative reliability. We need to show that there is a
possible story that could explain our reliability, consistent both with the absence of a
causal channel between the subject matter and us and with scientific scruple.
I think that there clearly is such a story. On the kind of explanation I have in mind,
the problem of reliability is split into two parts: on the one hand, into a scientifically
respectable account of why we have certain concepts, and judgments involving them;
and, on the other, into a scientifically respectable account of why, given that we have
such concepts and make such judgments, we would be reliable about them.
Once the problem is split up in this way, we can easily imagine a scientifically
respectable answer to each of its two parts.
Thus, it seems plausible to suppose that having the capacity to think about what fol-
lows from what would be evolutionarily advantageous. And it also seems plausible to
suppose that once we develop the capacity to make such judgments and deploy them
in our thinking, that it would be evolutionarily advantageous for us to be at least fairly
reliable in how we arrive at them. If these judgments were arrived at on the basis of
intuition, it would be evolutionarily advantageous for our intuitions to be at least fairly
reliable.
Obviously, there are many outstanding issues for such a style of explanation. But its
availability is sufficient to refute the suggestion that we have an a priori proof at hand
that shows that there can be no scientifically respectable explanation of the reliability
of our intuitions.19
Conclusion
I am increasingly inclined to the view that we cannot adequately explain a priori justi-
fication without appeal to intuitions.
If such an appeal to intuitions is to help, it must provide epistemological resources
that go beyond those provided by explanations in terms of epistemological analyticity
(appeals to our understanding of concepts). Accounts, like Sosa’s, which reduce intui-
tions to attractions to assent, and which give the understanding an indispensable role
in explaining the justificatory powers of such attractions, cannot provide such a
resource. As a result, such accounts must be rejected.
I have briefly presented an alternative account of these issues, one that seems to me
to hold greater promise.
References
Bealer, G., 1992. “The Incoherence of Empiricism.” Proceedings of the Aristotelian Society,
Supplementary Volumes 66: 99–143.
19
The possibility of an explanation along these lines first occurred to me in the 1990s. Subsequently,
I benefited from a number of conversations about this idea with Joshua Schechter, who explored it in con-
nection with explanations of logical reliability, both in his dissertation and in his (2010).
Bengson, J., 2015. “The Intellectual Given.” Mind. doi:10.1093/mind/fzv029

Boghossian, P., 1996. “Analyticity Reconsidered.” Noûs 30(3): 360–91.
Boghossian, P., 1997. “Analyticity.” In C. Wright and B. Hale (eds) A Companion to the
Philosophy of Language, 331–68. Oxford: Blackwell.
Boghossian, P., 2003. “Epistemic Analyticity: A Defense.” Grazer Philosophische Studien 66(1):
15–35.
Boghossian, P., 2009. “Virtuous Intuitions: Comments on Lecture 3 of Ernest Sosa’s A Virtue
Epistemology.” Philosophical Studies 144(1): 111–19.
Boghossian, P., forthcoming. “The Indispensability of Intuitions.” In P. Boghossian and
T. Williamson, Debating the A Priori and the Analytic. Oxford: Oxford University Press.
BonJour, L., 1985. The Structure of Empirical Knowledge. Cambridge, MA: Harvard University
Press.
Casullo, A., 2003. A Priori Justification. Oxford: Oxford University Press.
Chudnoff, E., 2013. Intuition. Oxford: Oxford University Press.
Cohen, S., 1984. “Justification and Truth.” Philosophical Studies 46: 279–95.
Creath, R., (ed.) 1990. Dear Carnap, Dear Van: The Quine–Carnap Correspondence and Related
Work: Edited and with an Introduction by Richard Creath. Oakland, CA: University of
California Press.
Foley, R., 1985. “What Is Wrong with Reliabilism?” Monist 68: 188–202.
Hume, D., 1978. A Treatise of Human Nature. Edition by L. A. Selby-Bigge and P. H. Nidditch.
Putnam, H., 1975. “The Meaning of ‘Meaning’.” Language, Mind and Knowledge: Minnesota
Studies in the Philosophy of Science. Minneapolis: University of Minnesota Press.
Schechter, J., 2010. “The Reliability Challenge and the Epistemology of Logic.” Philosophical
Perspectives 24(1): 437–64.
Sosa, E., 1998. “Minimal Intuition.” In M. DePaul and W. Ramsey (eds.) Rethinking Intuition,
257–69. Washington, DC: Rowman & Littlefield.
Sosa, E., 2006. “Experimental Philosophy and Philosophical Intuition.” Philosophical Studies
132(1): 99–107.
University Press.
Sosa, E., 2009. “A Defense of the Use of Intuitions in Philosophy.” In M. Bishop and D. Murphy
(eds) Stich and His Critics, 101–12. Oxford: Blackwell.
9
Rising above the Animals
The Search for Intellectual Assurance
Richard Fumerton
In this essay1 I will examine Sosa’s epistemological views primarily as they have been
set forth in his two volumes on Reflective Knowledge. As the title of this essay might
suggest, I will ultimately concentrate on the way in which Sosa tries to develop the
concept of a more intellectually satisfying concept of justification and knowledge
(scientia).
Points of Agreement
It is no secret that on the controversy between internalism and externalism I am gener-
ally more sympathetic to certain versions of internalism. I would readily concede,
however, that when fully entrenched in the internalist camp one has an uphill battle
avoiding a rather radical skepticism. Most epistemologists take the fact that an episte-
mological view leads to skepticism to be a conclusive reason to reject that position.
Once we emphasize, however, that the epistemological assurance the philosopher seeks
is something rather unusual, something, perhaps, that most people don’t seek, that
seems to me to take a bit off the edge of the skeptic’s position. It has always seemed to
me a touch of philosophical hubris to suppose that just because the philosopher wants
something, the world will cooperate so as to allow the philosopher to get it. Sosa
(2011a) suspects that the internalists want something that they can’t possibly get and
that they would be better off changing their goal. I’m more willing to stay frustrated.
But on the general issue of whether the internalist will be able to avoid serious skeptical
conclusions Sosa and I may well agree.
1
I’d like to thank Sam Taylor at the University of Iowa for very useful discussions about the issues dis-
cussed in this essay. I would also like to express special thanks to Miguel Ángel Fernández Vargas, Ernie
Sosa, Matt McGrath, and the other participants in the international conference on “The Present and Future
of Virtue Epistemology” sponsored by the Instituto de Investigaciones Filosóficas at the Universidad
Nacional Autononoma de México.
152 Richard Fumerton
I do want to underscore some other respects in which I agree with Sosa. First, and
most importantly, I agree with Sosa that there must be genuinely basic knowledge or
justification—knowledge or justification that does not require one’s moving up a level
to reflect philosophically on the source of that knowledge or justification. I reject
access requirements as general requirements on knowledge or justification.
Controversially, however, I do think that inferential justification, at least ideal philo-
sophically satisfying inferential justification, does require access of some sort to the
relevant connection between one’s premises and one’s conclusion.2
When I am careful, as I just tried to be in the last sentence, I stress that the kind of
justification I am usually talking about, is an ideal, philosophically satisfying, justifica-
tion. And that contextually implies another area on which I have come to agree with
Sosa. Talk of ideal justification suggests that we might be able to recognize justification
that is less than ideal. Sosa famously distinguishes animal-level knowledge from a
more intellectually satisfying sort of knowledge that builds upon it. For a number of
reasons, I think it would be better to start from an ideal concept of knowledge and jus-
tification and recognize kinds of knowledge and justification that are, in a sense,
degenerate and less demanding. In particular, as I will explain shortly, starting with the
ideal seems to me to get the right answers about justified belief in hypothetical situa-
tions involving massive and successful deception, both in the ideal case and in degen-
erate cases.
Animal-level Knowledge
In talking about the animal level, there is a reason, I think, that Sosa is much more
comfortable talking about knowledge than justification. It is an almost uncontrover-
sial linguistic datum that we do use the word “know” in characterizing the states of not
only young children but also non-human animals. In many of the contexts in which we
use the term “know,” it would seem much more strained to talk about justified belief.3
We say that the nervous dog in the car “knows” that it is going to the vet. That same dog
when it sees its master getting the leash knows that it is about to go for a walk. Rats (and
almost all other living creatures) “learn” from experience. The baby recognizes (a form
of knowledge?) that his mother is holding him, and two-year-olds seem to have exten-
sive knowledge of all sorts of truths. If the philosopher loads up the concept of knowl-
edge with conditions that are too intellectually demanding, it’s hard to see how we will
ever be able to make sense of much of this talk.
Now there is a reason I put “know” and “learn” above in scare quotes. It is not crystal
clear that we are using the expressions “know” and “learn” with their literal meaning in
2
But see Fumerton’s “What the Internalist Should Say to the Tortoise” (forthcoming) for an evaluation
of quite different ways in which to think about the relevant access.
3
Some would agree with Plantinga that the concept of justified belief is a relatively recent player on the
epistemological scene. And it is striking how seldom philosophers like Descartes ever talk about evidence
making probable a given conclusion.
rising above the animals 153
some of examples described above.4 We do have a penchant for anthropomorphizing.

We say that our computer is trying to retrieve a document, that our GPS seems to be
confused, that our white blood cells know when our body has been invaded by a virus
and that they try to kill it off. I suppose we might even describe the water-deprived
plant as realizing that it needs to extend its roots in order to seek out more water. But
I’m not sure that the above descriptions are intended to be anything other than “as if ”
claims. The behavior of the computers, the guidance system, the white blood cells, and
the plant remind us of our behavior when it is guided by knowledge and desire, and
perhaps we are simply taking an “intentional stance” that we know is a kind of fiction
towards all sorts of entities that don’t really have intentional states. But the burden is
probably on those who would deny that animals have knowledge to convince us that
the terms are not being used in a literal sense.5
Sosa and the New Evil Demon Objection to Externalism

Sosa’s well-known account of basic, animal-level knowledge is paradigmatically exter-
nalist. Roughly, one knows in this sense when one believes truly, and when one’s true
belief is caused by a relevant cognitive skill. The cognitive skill in question is ultimately
going to be cashed out in terms of an ability or a capacity to get at the truth in a reliable
way. One doesn’t need to know that one has the ability in order to use it to arrive at the
truth. In this sense there are many similarities between Sosa’s account of the skill that is
an element of animal-level knowledge and the reliabilist’s understanding of the relia-
bility condition for justified belief. In developing his view, Sosa often relies on the way
we talk about athletes and their achievements. Almost anyone can hit a hole-in-one.
But when I do it (and I actually have twice) it is surely a matter of luck. When Tiger
Woods does it, we attribute the success to his skill. To be sure, this raises all sorts of
difficult questions about causation and explanation. The “causation” in play when
Woods hits the ace sure doesn’t seem to have much to do with regularity. Indeed, while
he has had many more holes-in-one that I have, he holes out from that sort of distance
only a tiny fraction of the time. But Sosa is certainly right that we do talk as if the skill
explains in some sense the relevant success.6 In the same way Sosa thinks we can attrib-
ute true belief to the successful exercise of a cognitive skill.
Because of the close connection between Sosa’s view and some classic versions of
reliabilism, it is not surprising that critics think that both views encounter similar
4
Indeed, Sosa himself seems willing to entertain the possibility that he is using the term “knowledge” in
describing animal-level knowledge in a less than literal sense (see Sosa 2011a and 2011b).
5
See Kornblith (2002) for an enthusiastic defense of the view that “know” is used correctly and literally
to describe the states of all sorts of creatures.
6
An anonymous referee suggests that the example of a hole-in-one is bad precisely because the odds are
so low of anyone achieving that feat (at least on any given shot). That same referee suggested using instead
the example of a home run by a skilled baseball player. But the probably of a given at-bat resulting in the
achievement of a home run is also very, very small. All this simply underscores that it isn’t going to be easy
to define the critical notion of a success explained by possession of a skill.
objections. Certainly, both face the “in-house” problem of dealing with the so-called
generality problem—the problem of how to specify the circumstances relative to
which we identify the reliability that defines the relevant ability. But these are technical
difficulties. The heart of the internalist’s objection to all paradigmatic externalist
accounts of justification is the New Evil Demon Problem. Now as we have already
noted, Sosa seems primarily interested in giving a foundationalist account of ani-
mal-level knowledge. But the ability or intellectual skill that figures in that account
seems to be the analogue of what other philosophers are talking about when they talk
of justified belief.
So what is the New Evil Demon Problem that is supposed to afflict externalism?
Well it’s easiest to understand in terms of the externalists’ rejection of the internalist’s7
cryptic slogan:
(Int) Same internal states of a believer; same epistemic justificatory8 properties of
that person’s beliefs.
(Int) is advanced by the internalist not as a contingent, empirical truth, but as a strongly
necessary truth. Externalists, it is argued, are typically committed to allowing for the
conceivability of my having an internal twin who is justified in believing propositions
for which I possess no justification (or vice versa). The demon thought experiment is
familiar. Suppose that someone S is consistently and vividly deceived through
demon-induced sensation to believe the same sorts of propositions we actually believe
when having the same sorts of sensations (same in terms of their intrinsic character).
The victim of demonic machination has mostly false beliefs (of this sort). We, let us
suppose, have mostly true beliefs (of this sort). The victim of demonic machination
has belief-forming dispositions that are rather unfortunate with respect to the goal of
believing what is true, while those same dispositions in us (though not exercised in the
same environment) work out rather well when it comes to achieving our cognitive
goals. There is no question that with respect to some goals a rational agent has, those
goals are, in fact, better achieved by the “normal” perceiver than by the “victim.” But
surely, the internalist pleads, whatever epistemic assessment we make of the demon
victim’s rationality should be made of the person whose experiences happen to be
veridical. Any view that can’t secure that result should be abandoned, and it should be
abandoned even if the alternative metaepistemologies leave us exposed to the skeptic’s
relentless attack.9
7
There are actually a number of different forms of internalism. The kind I’ll focus on here is what I often
call internal state internalism.
8
I used this awkward locution to stress that the epistemic concept that is the target of an internalist’s
view is the concept of a belief for which there is epistemic justification. On most accounts of knowledge,
any that allow one can know despite having fallible justification, the truth condition for knowledge will
usually preclude an internalist account of knowledge.
9
A similar sort of argument is often also used (successfully, I think) to attack actual consequence
accounts of rational action. It seems almost obvious that a perfectly rational action can through bad luck
frustrate one’s important goals. Conversely, a painfully irrational action can end up (through sheer luck)
These days many philosophers will reject the intelligibility of the thought experi-
ment. Extreme content externalists might argue that a person whose internal states are
consistently caused (directly) by a conscious being like the demon simply can’t believe
what we believe (on the supposition that our internal states have radically different
sorts of causes). Indeed, such externalists are likely to argue that the subject in question
can’t even have the same experiences as we have (on the supposition that experiences
are intentional states whose content is somehow fixed by external conditions). But it is
only an extreme (and I believe implausible) semantic externalism that can achieve that
result. Even if (contrary to fact) simple thought were to have its content fixed by its
causal history, almost all content externalists will concede that complex thought can
have an intentional content that does not now and never did correspond to anything
real. I will argue later that even thought of apparently simple properties of physical
objects is actually quite complex, for example, the thought of color properties. But the
more general point is that content externalists must restrict their thesis to a sub-class of
ideas. Everyone concedes that we can form the ideas of a unicorn, a golden mountain,
a God, a fountain of youth, and so on regardless of whether we have causally interacted
with such beings. The natural response, learned from the radical empiricist, is to con-
cede that we can combine through our imagination simple ideas (whose content gets
an externalist analysis) to form complex ideas whose content is a function of the sim-
ple ideas. But that’s all the skeptic needs to form beliefs about a physical world. As long
as we have the idea of sensations (however those sensations are understood), we can
form the idea of whatever it is that is both non-mental and causally responsible for
those sensations. And now one can wonder whether or not that idea actually corre-
sponds to anything in reality.10
Sosa seems to agree that the victim of the demon has beliefs that enjoy the same jus-
tificatory epistemic status as our beliefs have (again on the assumption that we are not
so misled). He seems to think that he can reach this conclusion by an actual-world
indexed account of the reliability (exercise of virtue) that is relevant to the epistemic
assessment of beliefs in this sort of thought experiment involving other possible
worlds. So the basic idea is that if we are thinking about the question of whether the
demon victim has justified beliefs, we should answer “Yes” just insofar as we are
convinced that the belief-forming dispositions of the demon victim would yield
mostly true beliefs in the actual world. As Sosa knows full well, his frustrated internal-
ist critic is likely to want to know what the epistemic verdict would be should the actual
world turn out to be a demon world (2009: 40). The critic is convinced that Sosa (like
the classic reliabilist) will be forced to concede (implausibly) that if the actual world is a
demon world then our beliefs are unjustified. But Sosa responds to that critic by
benefiting a person. As Hobbes (1651: Part I, Chapter IV) said in replying to “the fool,” there is a huge
difference between being lucky/unlucky and being rational/irrational.
10
The argument here is, of necessity, highly compressed. In Fumerton (1985, 2007, and 2013) I defend
in much more detail the framework I sketch here.
a rguing that the following conditional is true: if the actual world is a demon world then
our beliefs are justified. The relevant conditional is either a material condition or
something stronger (2009: 40–1). If it is a material conditional, Sosa says, it has a false
antecedent and is true. If it is a disguised subjunctive, it is a roundabout way of asking
us to perform the same thought experiment we have already considered, a thought
experiment that issues the verdict that in this possible (but not actual) world—the
world in which contrary to fact the actual world is a demon world—we are to assess the
epistemic status of a subject’s beliefs by reference to the way in which that subject fares
with respect to getting at the truth in the actual world. One might think of the move
this way: Let’s name the actual world W* and the demon world Wd. In W* people are
getting at the truth by taking sensations at face value, and the resulting beliefs enjoy at
least animal-level justification. The internalist is trying to ask about the epistemic sta-
tus of these beliefs we form in W* if W* were Wd (a demon world). To answer that
question we travel in imagination to a demon world, Wd, but Sosa’s view still requires
us to evaluate epistemic truths in Wd by looking back to W*, the actual (non-demon)
world, a world in which we are supposing that the outcome of having the relevant
belief-forming dispositions is gaining true beliefs.
Now internalists worry about why most externalists are so sanguine when it comes
to making assertions about what is and isn’t truth-conducive in the actual world. I’m
not sure I know what the norm of an assertion is, nor even that I understand claims
about norms of assertion, but I do know that the skeptic is eager to ask externalists how
they know all these truths that seem often not discussed, but simply assumed, about
the nature of the world in which they live. For now, though, let’s leave that alone and
ask ourselves if Sosa’s maneuver can give the internalists what they want.
I worry that the above discussion is just another example of confusion caused by
taking talk of possible worlds too seriously. There is (of course) only one world, and all
truths (including modal truths of various sorts) are made true by some feature of that
one and only world. When the internalist asks the reliabilist or the virtue epistemolo-
gist that subjunctive question about the actual world, the internalist is simply asking a
question about the conceptual connections to which the view is committed. Consider
the following propositions:
(1) We have various sensations and they are caused directly by a very powerful
mind.
(2) We have a stable disposition to form beliefs about the existence of physical
objects and their properties as a result of having those sensations, beliefs that
are mostly false.
According to Sosa’s view, do (1) and (2) conceptually entail:
(3) Our perceptually formed beliefs are unjustified?
The internalist thinks that Sosa and the classic reliabilist must say “Yes.” But most inter-
nalists think that the answer is clearly “No.” When the dust settles, it seems to me that
the demon described above still plays havoc not only with true belief but with finding a
plausible metaepistemology that builds on an externalist account of foundational
justification.
I noted earlier that I am sympathetic to Sosa’s idea that we should distinguish differ-
ent levels of knowledge and justification. I’m inclined to think that ideal justification
for even commonplace beliefs about the external world is actually inferential in nature
and that as philosophers we seek access to the (rather complex) evidential connections
that would assure us that what we believe is probably true. But it also seems problem-
atic to suppose that children, the philosophically unsophisticated, or even the philo-
sophically misguided, are aware of the relevant evidential connections.11 As Hume
(1888: 187) so poetically suggested, when it comes to reaching critical conclusions
about the world around us, we may have evolved in such a way that all is well as far as
finding the truth: “[Nature] has doubtless esteem’d it an affair of too great importance
to be trusted to our uncertain reasonings and speculation.” If we are caused to believe
propositions about the past and the world around us by sensations and apparent mem-
ories, we might still allow that the resulting beliefs are justified if the sensations and
apparent memories are truth makers for propositions that are by themselves, or are
parts of complex conjunctions of propositions that are evidentially connected to what
we believe. If the internalist is right about evidential connections, this “degenerate”
justification will exist whether we are massively deceived or not.
Reflective Knowledge and the Search for Assurance

We have been talking mainly about Sosa’s conception of animal-level knowledge. But
as we have already noted, Sosa is sensitive to the philosopher’s concern that we want to
rise above this sort of animal-level knowledge. We want something that is intellectu-
ally more satisfying. We want what he calls reflective knowledge. Most of the time, Sosa
argues that we gain this reflective knowledge through finding coherence among our
beliefs. But this is only a contingent fact about human beings and our epistemic predic-
ament. Sosa acknowledges that all that is necessary for reflective knowledge is apt
belief that a belief is aptly formed (where apt belief is shorthand for belief that satisfies
the conditions for animal-level knowledge). So, for example, if we were so constituted
that we always instinctively believed that a belief was aptly formed when and only
11
Though I would also caution against underestimating the conceptual sophistication of even very
young children. Consider color judgments, for example. Arguably, one does not really possess the concept
of red as it applies to physical objects until one understands that appearances can be deceptive. So a child
who insists that the color of the ball has actually changed as the child puts on tinted sunglasses doesn’t
know how to use color expressions. Mastery of the language requires one to learn the distinction between
how things look and how they are. Minimally, one needs to understand that something can be red even if
it doesn’t look red (as when the lights are off). If it is true that one needs the concept of distorting condi-
tions (whether one has a vocabulary to describe them or not) to possess the concept of color, then even
very young children are deploying very complex concepts, that may require having in place rather sophis-
ticated beliefs about background conditions.
when it was, that would give us reflective knowledge (second-level knowledge). And if
we were so constituted that we always instinctively believed that a belief was aptly
formed only when it was, that would give us yet another level of reflective knowledge
(and of course this could continue indefinitely). But this, of course, frustrates internal-
ists. They want to rise above animal-level knowledge. But animal-level knowledge that
one has animal-level knowledge seems to be just more animal-level knowledge (albeit
with a different propositional content).
Sosa is, I believe, prepared to bite the bullet with respect to the above complaint.
Indeed, he seems to suggest that we do have abilities like those described above
with respect to certain sorts of propositions, and he appeals to such abilities in his
account of a priori knowledge. But Sosa doesn’t want reflective knowledge to come
too easily and, in particular, he seems suspicious of what is now often called “easy
knowledge.”
I sketched the problem of easy knowledge facing externalists in Metaepistemology
and Skepticism (1995) and the issue was crystallized nicely by Stew Cohen’s (2002) now
classic paper. There are actually different kinds of problematic easy knowledge, but all
trade on the idea that once we have a foundational source of knowledge or justified
belief, there seems to be no principled reason why we shouldn’t be able to use that
source to certify its own legitimacy. So Cohen imagines his son worried that an object
that looks red might not really be red—the lighting conditions, for example, might be
causing the object to appear other than it is. Not to worry. If we can have foundational
knowledge of the colors of objects, and introspective knowledge of appearances, we
have available to us the following argument.
1) It looks as if there is a red object there (introspectively justified)
2) There is are object there (justified by the experience)
Therefore,
3) This experience isn’t deceptive.
Of course, if this reasoning is unproblematic we can easily use track-record arguments
to certify the reliability of any source of belief that is, in fact, reliable (at least when
beliefs that rely on memory and induction are also reliably informed). You want an
argument that we can rely on apparent memory. Here it is:
1) I seem to remember X (introspectively justified)
2) X (justified by apparent memory)
3) I seem to remember Y (introspectively justified)
4) Y
5) And so on
6) I don’t now seem to remember (very often) seeming to remember things that
didn’t happen (justified introspectively and by apparent memory).
Therefore,
7) Apparent memory is reliable.
Similar arguments can easily be deployed to certify the reliability of perception. Of

course, arguments with that structure could also be deployed to certify the reliability of
astrological inference, or palm-reading inferences, but they wouldn’t succeed on the
assumption that astrological inferences and palm-reading aren’t really reliable ways of
discovering truth. If perchance such problematic inferences turned out to be reliable
then the astrologer and the palm reader would also be in business.
At least some epistemologists worry that track-record arguments of the sort just
discussed are obviously problematic, though there is no real consensus on what makes
the arguments problematic. Some (Cohen, for example) think that the real source of
the problem is foundationalism—he thinks that the problem will inevitably arise for
any philosopher who allows that a belief can be noninferentially justified. I have argued
(1996) that the problem really arises only for the foundationalist who has an incorrect
account of noninferential justification. Cohen’s son is unsatisfied because at some level
Cohen’s son knows that one doesn’t have noninferential knowledge of the color of
objects—such knowledge already requires that one know something about the condi-
tions of perception. And that suggests one obvious diagnosis of the problematic “easy
knowledge” arguments—the arguments are question-begging. For example, one
might argue, the conclusion of the problematic argument certifying memory must be
justifiably believed prior to one’s making the move from apparent memory to beliefs
about the past. In the case of perception, one needs to be justified in believing that
appearances aren’t deceptive before one can rely on appearance to reach a conclusion
about one’s environment.
Sosa himself seems to diagnose the problematic nature of track-record arguments
in a way that is at least superficially similar. He doesn’t want one to be able to get apt
belief that one has apt belief in way that is “too” easy. But he also doesn’t want ani-
mal-level knowledge (the kind of knowledge he thinks we have of our immediate
bread-boxed-size physical environment) to rely on explicit premises describing reli-
able correlations between appearances and reality. Nevertheless, he seems to admit
(2009: chapter 10) that our sensation-based beliefs, even at the animal level, carry
with them (or are accompanied by) presuppositions (he sometimes calls them com-
mitments) about the conditions of perception. He seems to argue that in some sense
when we form such beliefs on the basis of sense experience, we commit ourselves to
the position that the conditions of perception are propitious with respect to trusting
appearance as an indicator of reality. These presuppositions or commitments may
not make it to the level of full-fledged conscious belief, but they are lurking at some
level of consciousness in the background and they had better be justified if our seem-
ingly spontaneous doxastic reactions to our sense experiences are to be justified.
But how do we get non-circular justification, for example, for a presupposition with
something like the content Sosa discusses?
(R) Reliably, if I have a visual experience as of something red, I tend to see some-
thing red (2009: 241) [where “see” is used as a factive].
As I understand Sosa’s position, justified commitments with the content of (R) must be
in place in order for us to justifiably believe on the basis of our experience that a given
object is red. But in the final analysis I think Sosa takes the relevant commitment to be
nothing other than the relevant stable disposition to form the belief in question (2009:
223, footnote 14). The commitment/disposition, Sosa agrees, can be assessed epistem-
ically. But for Sosa this is ultimately only to say that the relevant disposition can be apt
or not (again, 2009: 223, footnote 14). But now the account seems a bit odd to me. My
animal-level justification for believing that the object is red derives from the fact that
the belief is apt. It is apt only because it results from a cognitive ability or skill that gets
me to the truth. That cognitive ability/skill/commitment needs to be epistemically jus-
tified (non-derivatively) in order to generate justified belief about the color of objects.
Yet its epistemic justification is nothing other than the fact that it results in apt belief. In
one sense, then, the problem of easy knowledge seems even worse to me on Sosa’s view
that on the views that Cohen discusses. We can’t use the question-begging argument
Cohen describes to justify our belief in the commitment that must be in place, accord-
ing to Sosa, because the commitment’s justification must be prior to (or at least simul-
taneous with) the resulting justification enjoyed by the belief about color. But for Sosa,
it seems to me that the conditions that constitute the belief about color being justified
(at the animal level) contain as constituents the conditions that constitute the relevant
commitment/disposition being justified. At the animal level, there is no separate prob-
lem of justifying underlying commitments. On Sosa’s view, I would suggest, it is even
easier to get justification for the relevant underlying commitment than it is through
the reasoning Cohen imagines. There isn’t any difference between having animal-level
justification for belief about the color of an object and having justification for the rele-
vant underlying commitment. But, of course, if we were worried about getting assur-
ance of the sort the skeptic seeks, the internalist is convinced, we are getting nowhere.
Moreover, since the epistemic status of the commitments are already taken care of at
the animal level, I don’t see why we can’t get explicit, fully conscious, apt metabelief
about the aptness of first-level belief about color in the easy way that Cohen (and
sometimes Sosa) criticize.
Coherence as the Source of Reflective Knowledge

If all of the above were true, then it is not clear to me why Sosa talks so much about
coherence as the most common way to get apt belief that one has apt belief. Why isn’t
the simple track-record argument fine?12 To get apt belief that I have apt belief that P,
12
And perhaps in the end he is fine with a bit of bootstrapping as long as the reasoning is appropriately
fundamental. He argues (2011a) that internalists and externalists alike trivially have no choice but use the
sources of belief they have to certify their legitimacy. One simply can’t respond to the skeptic who insists
that you use none of the ways you have of reasoning to certify that your reasoning is legitimate. And that
does seem right to me. The only issue is whether you can justify a fundamental inference rule empirically.
The real issue, I think, is whether evidential connections hold necessarily and are knowable a priori.
I just need to have an apt belief that P (trivially satisfied when I satisfy the conditions
for animal-level knowledge that P), an apt belief that I have a skill when it comes to
getting at truths of this sort (satisfied by the inductive argument correlating appear-
ance with reality), and an apt belief that the skill causes (in the right way under the right
conditions—more about this shortly) the relevant true belief. It is not that obvious how
one goes about justifying a causal claim, but correlations of the sort that come with
establishing the skill seem also a critical part of establishing the causal connection—
the possession of the skill will be easy to correlate with the success.
Perhaps in the end Sosa will be on board with the right sort of bootstrapping. He
does argue (2011) that internalists and externalists alike (including Descartes, the
internalist’s internalist) trivially have no choice but to use the sources of belief they
have to certify their own legitimacy. There is no view from nowhere when it comes to
discovering truth. You simply can’t respond to the skeptic who insists that you use
none of the ways you have of justifying belief to certify that your beliefs are justified.
And that does seem right. The question, though, is whether you can justify a funda-
mental inference rule empirically. The real issue, I think, is whether evidential connec-
tions hold necessarily and are knowable a priori. Internalists will be in the end
committed to this principle. Sosa and most externalists reject it (at least for evidential
connections other than forms of entailment).
Part of the initial worry about some track-record arguments is that they seem point-
less. When we use a single source of belief to certify inductively its reliability, it seems
that we would be able to know what the result of the inductive reasoning will be before
we perform it. That’s because, as philosophers, we know that once we start trusting a
given sort of appearance as a guide to reality in formulating our premises, we know we
were going to be able to generate the result we want. Our “investigation” would be as
pointless as trying to certify that a yardstick was really a yard long by using it repeat-
edly to measure itself. As we’ll see in a moment, however, this argument might be a bit
too quick.
Other sorts of checks and balances on sources of belief seem at least slightly more
promising. If one has what Sosa calls a commitment to the reliability of instinctive
beliefs based on visual experience and an equal commitment to the reliability of
instinctive beliefs based on tactile experience, at least we know that we can encounter
problems. If when something looks round to me, it always also feels square, one of my
commitments is in trouble. So when I don’t encounter such trouble, things are going,
at the very least, better than they could have gone. And even some single sources of
belief are like that. They can undermine themselves. I often give the example of the
magic eight ball that some of us used to play with as children. One asks the eight ball a
question, shakes it, and an answer floats to a transparent opening on the ball. When I
first used the example, I was trying to point out that we would hardly expect the skep-
tic about eight-ball reasoning to be satisfied if we ask the eight-ball to respond to the
question “Are you reliable?” and we get the answer “Yes.” But it may not be true that
the exercise was completely worthless. At the very least, the eight-ball reasoner is
etter off than he would have been had he received the answer “No.” And if we kept
b
asking the same question (perhaps in different ways) and we kept getting answers like
“I already told you, Yes,” “How many times do I have to tell you the same thing?” when
we know that there are other answers that the eight ball could have given, we might
start wondering if there was something to this whole idea of relying on eight-ball
answers. Our next step might be to repeat all sorts of other questions to see if we get a
consistent answer to those questions as well. Let’s suppose we do. Have we made a
strong prima facie case for eight-ball reasoning. Not really. There are still alternative
eight-ball skeptical hypotheses that would be consistent with the data. These days we
could easily imagine an eight ball that is sophisticated enough that it can “recognize”
various forms of the “Are you reliable?” question and that is programmed by its man-
ufacturer to respond to those questions in just the way described above. It might also
be programmed in such way that it “remembers” questions (and indeed can translate
questions of equivalent meaning both within and between languages) and is pro-
grammed to respond the same way to any question with the same meaning no matter
how many times the question is asked. Of course, if we had what we took to be inde-
pendent ways of confirming a wide range of predictions made by the eight-ball (the
way we think that tactile experience can lead us to the same conclusion as visual expe-
rience), then we might develop an even stronger belief about the reliability of eight-
balls, but we still might not think that eight-balls are non-derivative sources of
justification. The eight-ball might, after all, be a disguised computer loaded with
Wikipedia software.
Consider another sad but more realistic example. The track-record argument for the
reliability of memory discussed above contained the critical premise that I don’t seem
to remember often seeming to remember some p when not-p. That premise is only
contingently true and I realize that there might come a time when I discover that is no
longer true of me. Those in the agonizing early stages of Alzheimer’s are often in the
terrifying position of seeming to remember lots of occasions on which they seem to
remember having done things that they also now seem to remember not having done
(or vice versa). Memory can turn on itself, just the way an eight-ball could tell you that
it isn’t reliable. When memory doesn’t turn on itself, that’s at the very least epistemi-
cally better than when it does. Should the fact that the track record for memory doesn’t
get in trouble satisfy the skeptic? Probably not. There are, of course, skeptical hypothe-
ses that would also explain memory’s reinforcement of itself, and we would need to
give the skeptic reasons for rejecting these in favor of “commonsense.” Similarly, there
certainly seem to be skeptical hypotheses that would explain why visual and tactile
sensations typically lead us to the same conclusion about shape even should such
beliefs be almost always false. In particular, grand powerful deceiver and massive hal-
lucination hypotheses would always seem to be consistent with all of the available sen-
sory evidence. But even deceiver hypotheses typically posit a common cause of various
sensory states and the connections between them. If part of the concept of a physical
object just is the concept of that which has the power to affect our sensory states in
various ways,13 we might be able to narrow the task of refuting skepticism to that of
giving reasons why we should reject the hypothesis that the relevant causal powers
belong to a conscious being.
I’m not sure I would seek the relevant justification in coherence. Clever skeptical
hypotheses will cohere just as well as the hypotheses of commonsense. Moreover,
global coherence of beliefs seems neither necessary nor sufficient for justified belief
(see Foley 1979 and Fumerton 1994). The best hope still seems to me to lie in coming
up with some satisfying criteria for choosing among competing explanations (as well
as the hypothesis that there is no explanation).14 But I have my doubts that one can
defend reasoning to the best explanation against the charge that it simply collapses
back into inductive reasoning (see Fumerton 1980).15 As I admitted at the outset, it is
not obvious that the skeptic won’t prevail.
Attributing Success to Skill—Another Issue Involving

Reflective Knowledge
In the preceding discussion, we haven’t worried much about what is involved in suc-
cess being attributable to a capacity or skill. I have talked mainly about the abstract
issues that will bother internalists about really any attempt to “rise above the animals”
by employing methods whose success is still measured by externalist criteria. But there
is a sense, I think, in which one can appreciate even more clearly these internalist con-
cerns by looking carefully at some of the problems that surround an attempt to capture
the relevant conditions that contribute to the sort of success necessary (according to
Sosa) for both animal and reflective knowledge.
Sosa (2012) returns to the question of what is required for reflective knowledge. He
reminds us that for both first-level animal and reflective knowledge one must distin-
guish the question of whether a given subject has the relevant inner competence/skill
(faculty) from the question of whether the inner competence is relevantly responsible
in a given situation for a true belief in the target proposition. He points out that it is
critical in answering this question to take into account the relevant conditions in which
the faculty is exercised. The fact that I can’t use a skill to achieve a result on a given
occasion because conditions are unfavorable doesn’t mean that I don’t have the skill.
The best archer in the world can’t be expected to hit a target in gale force winds, and
even if the target is, by luck, hit, we won’t attribute the success to the skill possessed by
the archer.16 But what about epistemic skill and success at getting true belief?
13
A view for which I have argued (Fumerton 1985, 2006).
14
It is interesting that in 2011b, Sosa seems to agree that the coherence he thinks is important involves
a kind of abductive inference, p. 56.
15
But see Vogel (2004) for one of the subtle attempts at defending a version of reasoning to the best
explanation as a solution to skepticism about the external world.
16
Though as our discussion of the hole-in-one made vivid, the issue is difficult. The archer whose arrow
was buffeted about by the wind still wouldn’t have hit the target but for his careful aim. The exercise of the
skill was a causally critical “but for” condition for the success.
Sosa famously discusses the person who judges, based on appearance, that the sur-
face of an object is red, but where the environment includes a “jokester” who is con-
stantly altering the light so as to produce misleading appearances of color. Sosa
eventually seems to suggest that as long as one was lucky enough to be relying on
appearance at the one moment the jokester wasn’t altering the lighting conditions one
could still have knowledge relying on that appearance.17 Why? Because the success
(getting a true belief about color) was due to an animal-level skill that was operating in
favorable conditions (for however briefly those conditions might have lasted). Of
course, it would be nice to have a general account of how to distinguish conceptually
favorable from unfavorable conditions—it presumably should have something to do
with the question of whether the conditions are compatible with a claim about the true
belief being caused by the relevant skill. In a related discussion in the same article, Sosa
suggests that when it comes to distinguishing being awake from dreaming, the favora-
ble conditions under which our skill can be said to be responsible for our true belief
include being awake. The idea, I guess, is that I can be pretty good at telling whether I
am asleep or awake when I’m awake, and if I am I can know now (assuming I am awake)
that I’m awake. There may be other conditions under which I can’t tell the difference,
but that doesn’t mean I don’t have the relevant cognitive skill—those conditions will be
like the gale force wind that prevents the archer from using skill to hit a target.
The internalist will again be frustrated, of course. We are told that if we are awake,
we can tell that we are awake. We are also told that if we are asleep and dreaming we
wouldn’t be able to tell whether or not we are awake. But Descartes got some of us
worried about which of the two states we are in and it doesn’t seem to help much to be
told that we don’t have to worry about being asleep and dreaming as long as we aren’t.
If we internalists can’t ask the question we want at the animal level, surely we can
express our philosophical worry at the reflective level. Surely, we lack a satisfactory
answer to the question of how we can assure ourselves that we are in favorable as
opposed to unfavorable situations with respect to being able to tell whether our
first-order belief that we are awake was formed as a result of an epistemic skill exer-
cised under the “right” circumstances. Not so fast, suggests Sosa. After all, we need
only to be able to have an ability that explains our true belief that we have animal-level
knowledge. To be sure, the ability will provide the relevant explanation only when
formed under favorable circumstances. What are the favorable circumstances? Those
in which we would typically be able to tell whether or not first-level beliefs were aptly
formed. We could find ourselves in a situation in which we wouldn’t be able to pick up
on any clues that they were not aptly formed because such clues might not exist.
Presumably, demons could insure that there were no differences between veridical and
non-veridical experiences, differences that a cognizer with the right skills could pick
up on. But as long as that is not the situation, we are pretty good (the assumption is) at
discovering that we have the relevant animal-level knowledge.
17
For similar reasons Sosa is inclined to think that the person travelling through fake-barn country can
know that there is a barn before him if he is lucky enough to be looking at a real barn.
To underscore the kind of frustration the internalist feels at this point, think about
an analogue of the preceding discussion. Imagine a more mundane skeptical worry.
Someone tells me something and though I am initially inclined to believe him, I begin
to worry that my trust might be misplaced. First, do I have animal-level knowledge if I
simply believe what the person tells me and the belief is true? Well, I suppose I might
think that the answer depends at least in part on whether I am pretty good at distin-
guishing truth-tellers from liars. Suppose I conclude that I am good at distinguishing a
truth-teller from a liar at least when conditions are favorable, where I then go on to
identify favorable conditions as restricted to those in which the person talking to me
generally tells the truth. I suppose that might be plausible enough. People who usually
tell the truth might have a much harder time keeping the proverbial “straight face”
when lying. The skeptic will not be terribly impressed even if the skeptic grants that I
have that skill—I can tell when people who don’t usually lie are lying. The skeptic will
obviously start worrying whether I can satisfy myself that I am talking to someone who
doesn’t usually lie. We are, of course, now seeking what Sosa would call reflective
knowledge. Suppose Sosa tells us that I can also tell when I’m dealing with someone
who isn’t usually lying. How? Because I am the kind of person who has the ability to tell
when I’m dealing with someone who usually tells the truth—at least I have that ability
when that fact about the person isn’t hidden (as it might be if someone is a really good
liar).18 But now the search for philosophical assurance at the level of reflective knowl-
edge is surely frustrated. The skeptic worries that I might be deceived, and I am told
only that I’ll be able to tell that I’m not deceived as long as I’m not being deceived by a
really good deceiver.
References
Cohen, S., 2002. “Basic Knowledge and the Problem of Easy Knowledge.” Philosophy and
Phenomenological Research 65: 309–29.
Foley, R., 1979. “Justified Inconsistent Beliefs.” American Philosophical Quarterly 16: 247–58.
Fumerton, R., 1980. “Induction and Reasoning to the Best Explanation” Philosophy of Science
47: 589–600.
Fumerton, R., 1985. Metaphysical and Epistemological Problems of Perception. Lincoln, NE:
University of Nebraska Press.
Fumerton, R., 1994. “The Incoherence of Coherence Theories.” Journal of Philosophical Research
XIX: 89–102.
Fumerton, R., 1995. Metaepistemology and Skepticism. Boston: Rowman & Littlefield.
Fumerton, R., 2006. “Epistemic Internalism, Philosophical Assurance, and the Skeptical
Predicament.” In T. M. Crisp, M. Davidson, and D. Vander Laan (eds) Knowledge and Reality:
Essays in Honor of Alvin Plantinga, 179–92. Dordrecht: Kluwer.
18
See Sosa (2012) on how I can tell that I have veridical color experience at least when the presence or
absence of that fact is not hidden (149).
Fumerton, R., 2007. “Price on Hume on Our Idea of the External World.” Soochow Journal of
Fumerton, R., 2013. Knowledge, Thought and the Case for Dualism. Cambridge: Cambridge
University Press.
Hobbes, T. [1651] 1981. Leviathan. London: Penguin Classics.
Hume, D., 1888. A Treatise of Human Nature, ed. L. A. Selby-Bigge. Oxford: Oxford University
Press.
Kornblith, H., 2002. Knowledge and its Place in Nature. Oxford: Oxford University Press.
Sosa, E., 2009. Reflective Knowledge, ii. Oxford: Oxford University Press.
Sosa, E., 2011a. “Book Symposium on Reflective Knowledge: Replies to Richard Fumerton, John
Greco, and Michael Williams.” International Journal for the Study of Skepticism 1(2):
138–49.
Sosa, E., 2011b. “Book Symposium on Reflective Knowledge: Replies to Ram Neta, James Van
Cleve, and Crispin Wright.” Philosophical Studies 53(1): 43–59.
Sosa, E., 2012. “Animal versus Reflective Orders of Epistemic Competence.” In T. Henning and
D. Schweikard (eds) Knowledge, Virtue, and Action, 140–50. London: Routledge.
Vogel, J., 2004. “Skeptical Arguments.” Philosophical Issues 14(1): 426–55.
10
Epistemic Agency
Hilary Kornblith
Over the years, epistemic agency has played a larger and larger role in Ernest Sosa’s
epistemology. In this essay, I examine the ways in which Sosa’s views on epistemic
agency have changed over time. I argue that there are substantial problems with Sosa’s
accounts, both early and late. Sosa puts epistemic agency to work in his epistemologi-
cal theory, but I argue that that work is better carried out in ways which make no appeal
at all to the problematic notion of epistemic agency.
I
Let us begin with the distinction between animal knowledge and reflective knowledge.
Here is how Sosa drew that distinction in 1985:
One has animal knowledge about one’s environment, one’s past, and one’s own experience if
one’s judgments and beliefs about these are direct responses to their impact—e.g., through
perception or memory—with little or no benefit of reflection or understanding.
One has reflective knowledge if one’s judgment or belief manifests not only such direct
response to the fact known but also understanding of its place in a wider whole that includes
one’s belief and knowledge of it and how these come about. (Sosa 1991: 240; see also Sosa 1997)
This distinction, famously, plays a central role in Sosa’s epistemology. Reflective

knowledge, on Sosa’s view, is especially highly valued. What is so valuable about it?
The question may seem silly. When we have animal knowledge, that is, when we
form our beliefs unreflectively, our beliefs are arrived at passively. Unreflective beliefs
are merely produced in us; we play no active role in their formation. Sometimes, how-
ever, we are not merely passive vessels in which a variety of belief-producing processes
are prompted to go to work. Rather, we stop and reflect on our beliefs, and their
sources, and the logical relations among them, and we ask ourselves whether these are
beliefs which we really ought to have. When we do this, it seems, we take an active role
in belief acquisition. We become epistemic agents.
Of course, this only postpones the question. After all, what is so special about taking
on this role as an epistemic agent? And it is here, I think, that the answer may seem
7.5.3 Fraudulent Trade Assignment
Fraudulent trade assignment can occur when a broker executes orders on the
same side of the market for more than one client. Each client should get
the price at which his or her order filled. A dishonest broker, however, may
assign the best prices to his favorite clients.
Fraudulent trade assignment may be especially problematic when bro-
kers also act as dealers. Without appropriate safeguards, broker-dealers may
be tempted to take the best trades for themselves and leave the worst trades
for their clients.
7.5.4 Prearranged Trading and Kickback Schemes
Prearranged trading occurs when a broker arranges a trade without properly
exposing her client's order to other traders who might be willing to offer
better prices. Under such circumstances, the client often receives a worse
price than he might have received if his broker had properly exposed
the order.
Prearranged trading is illegal in floor-based futures markets. In such mar-
kets, traders must shout out their bids and offers so that all traders have an
opportunity to trade. It is also illegal in most electronic futures exchanges.
Many equity markets and some futures markets allow block traders to
prearrange trades that they want to print on the floor of the exchange. The
matched trades must be brought to the floor to give floor traders an op-
portunity to offer better prices if they choose to. These special procedures
allow brokers to profit when they have arranged both sides of a difficult
transaction and at the same time protect both sides of the trade from po-
tential abuse.
In a kickback scheme, a broker sends an order to a dealer with the under-
standing that the dealer will fill it at a poor price. The dealer gives the bro-
ker some consideration—the kickback—in exchange for the opportunity to
cheat the client. The dealer may pay the kickback in cash or with non-
monetary considerations.
Brokers often arrange to send dealers order flow in exchange for mone-
tary or nonmonetary payments. Although these payments for order flow
arrangements appear a lot like kickback schemes, they generally are not. We
discuss the economics of payments for order flow in chapter 25.
7.5.5 Unauthorized Trading and Churning

Brokers engage in unauthorized trading when they make trades for their
clients that their clients have not authorized. Brokers generally make these
trades to generate commissions or to manipulate prices. Not surprisingly,
the problem is most serious among unsophisticated retail investors.
Clients must pay close attention to their accounts to ensure that their
brokers are not making trades of which they do not approve. They must pay
particular attention to the trade confirmations that they receive.
Unauthorized trading is especially difficult to detect if the broker has
changed the mailing address on the defrauded account. In that case, the vic-
tim may not quickly detect the unauthorized trading. To prevent this prob-
lem, most brokerage firms do not allow their brokers access to the systems
that maintain client mailing addresses. They also require signed instructions
to change client mailing addresses, and they compare the signatures with
those on file. Finally, they send letters to their clients to advise them when-
epistemic agency 169
may thereby fail to increase our overall reliability. But the mere possibility of unrelia-
bility is not enough to undermine the suggestion that reflective scrutiny is actually a
good thing. We need to know how our reflective processes actually work and to what
extent they actually serve to increase or decrease our overall reliability. This is, obvi-
ously, an empirical question, and it is an empirical question on which a great deal of
work has been done.
When we stop to reflect on the source of our beliefs, we do often come to form
(second-order) beliefs about the source of our first-order judgments. If you ask people
why they hold the beliefs they do, then, in a very wide variety of cases, they will give
quite confident answers about how they arrived at their beliefs. It is, however, well
known that a very large part of the cognitive processes by which beliefs are produced is
unavailable to introspection. For example, the vast majority of the information pro-
cessing that goes on in visual perception simply eludes introspection; one can’t even
begin to understand the complexity of the psychological processes involved in vision
by introspective means.1 The same is true, of course, of the processes involved in
language acquisition and use.2 Indeed, this is true of belief acquisition generally.
The parts of these processes that even seem to be available to introspection are just the
tip of the iceberg. One can’t do cognitive psychology by simply reflecting on one’s
mental processes.
All of this is to say that the view we have of the processes by which our beliefs are
produced, when we reflect on our beliefs, is at best partial. But this might not be a
problem if the part of the process which we seem to have access to allows us to accu-
rately evaluate the reliability of the entire process. Thus, for example, if any errors we
might be making are most likely to show up, or leave some trace, in the parts of the
process which are available to introspection, then reflective evaluation of the source of
our beliefs might be an extremely useful thing. So we need to know to what extent the
view we have of belief acquisition, when we reflect, can contribute to an accurate eval-
uation of the reliability of that process. And since the point of reflective evaluation of
these processes is amelioration, we will be especially interested in whether errors we
might otherwise make, were we unreflective, are likely to be identified when we stop to
reflect.
The beliefs we form are often influenced in powerful ways by factors which do not in
any way make it likely that the resulting beliefs are true. Our beliefs about the quality of
various consumer goods may be influenced by their relative position: we have a strong
tendency to believe that objects further to the right are superior to objects further to
the left (Nisbett and Wilson 1977). We are influenced in the numerical judgments we
make when, in the course of arriving at our judgments, we are exposed to obviously
irrelevant numerical information. Thus, for example, subjects who were asked to estimate
how many African countries are members of the United Nations were dramatically
1
For one particularly useful account of these processes, see Marr (1982).
2
For an early and important account, see Chomsky (1965).
170 Hilary Kornblith
influenced in their judgments by the number which turned up on the spin of a roulette
wheel (Tversky and Kahneman 1974). We are influenced in our interactions with
others, and the beliefs we form about them, by simple exposure to words such as ‘‘rude’’
or ‘‘polite’’ (Bargh, Chen, and Burrows 1996). The colors of objects influence our judg-
ments about a variety of matters in ways which are completely unreliable.3 We are
influenced by stimuli presented below the threshold of awareness, and these influences
are often only randomly correlated, or negatively correlated, with the truth of the
resulting beliefs (Merikle 1992). Our judgments are often influenced by racial stereo-
types, even when we do not believe the very features of those stereotypes which play a
crucial role in affecting our judgments (Devine 1989; Hauser’s Moral Sense Test). In all
of these cases, subjects are unaware that their beliefs are influenced by these extrane-
ous factors. When directly asked whether their judgments are influenced in these
ways, subjects deny, often quite vehemently, that their beliefs were affected in any way
by the factors which played the most direct causal role in bringing about their beliefs
(Nisbett and Wilson 1977). Ziva Kunda nicely summarizes the upshot of this large
body of literature.
Our judgments, feelings, and behaviors can be influenced by factors that we have never been
aware of and have only been exposed to subliminally, by factors that we were aware of at one
time but can no longer recall, and by factors that we can still recall but whose influence we are
unaware of. (Kunda 1999: 308)
Asking subjects to introspect more carefully, or think longer and harder about the
sources of their beliefs, is entirely useless in these cases.4 Subjects are ignorant of the
actual source of their beliefs, and reflection is incapable of revealing it to them.
This does not mean, however, that when subjects in these cases reflect on the source
of their beliefs, they find themselves at a loss or recognize that they are ignorant on the
issue of how their beliefs came about. Nothing like this is true. Instead, subjects in
these examples offer quite confident accounts of the source of their beliefs, accounts
which are not even close to correct. By and large, the accounts they offer serve to
rationalize their beliefs. Were these accounts correct, they would serve to show how it
is that the subjects arrived at their beliefs on the basis of good reasons (Nisbett and
Wilson 1977). Finally, the phenomenology of these subjects is quite interesting. These
subjects do not seem to themselves to be offering some sort of retrospective hypothesis
3
Thus, for example, judgments about politicians were found to be influenced by the colors of their
campaign posters (in ways having nothing to do, for example, with the ways in which certain colors may
reliably indicate the party of which the politician is a member) in Rubinoff and March (1980). The color of
ballots was found to influence voting behavior in Garret and Brooks (1987). The importance of the color of
packaging has long been known to those in advertising and sales, and often used to great advantage. See,
e.g., Fehrman and Fehrman (2003).
4
Laurence BonJour, in an attempt to defend reflection against the kind of concerns developed here,
suggests otherwise, although he offers no experimental evidence for his claim. I have not been able to find
any experimental evidence at all which suggests that BonJour is right here. See BonJour (2006: esp. 324–6).
Richard Feldman also suggests, in a case of this sort, that one might be able to discover the source of one’s
belief by way of reflection, but he too offers no experimental evidence. See Feldman (2003: 607).
about how it is that they must have arrived at their beliefs. Rather, it seems to them that
they are directly aware of the manner in which they arrived at their beliefs. Their
self-knowledge, as they see it, is as direct as self-knowledge ever is (Gopnik 1993,
Nisbett and Wilson 1977).
Thus, in an extremely wide variety of cases, subjects form beliefs in ways which are
quite unreliable. They are moved by factors of which they are unaware, and reflection
on the source of their beliefs cannot make them aware of these factors. When they do
stop to reflect, however, they come to form confident beliefs about how it is that they
arrived at their first-order beliefs. These judgments, which are entirely inaccurate, por-
tray them as properly moved by good reasons. These judgments about the source of
their first-order beliefs seem to be direct and unmediated by inference or theorizing.
What these subjects are doing, however, is confabulating (Wason and Evans 1975,
Nisbett and Wilson 1977, Wilson 2002, Lucas and Ball 2005, Halberstadt and
Wilson 2008, Hauser et al. 2007). These subjects are sincere in their claims about the
sources of their beliefs, and they are entirely taken in by their own confabulation.
Moreover, these subjects are not all facing situations which are contrived or unrealis-
tic. The factors which influence them are regularly present in our everyday environ-
ment. We are all subject to these influences, and we are all bound to form an immense
number of beliefs in these unreliable ways, despite the fact that, when we reflect on
these very beliefs, we each seem to ourselves to be a paradigm of rationality.
The idea, then, that by reflecting on the source of our beliefs, we may thereby subject
them to some sort of proper screening, and thereby improve on the accuracy of the
resulting beliefs, is simply misguided. When we reflect in this way, we get the impres-
sion that we are actually providing some sort of extra screening of our beliefs, and we
thus have the very strong impression that we are actually doing something to assure
that our beliefs are, indeed, reliably arrived at. But this is not what we are doing at all.
Instead, we are engaged in a process which, in a very wide range of cases, makes us
more confident that we are right than before we began, but a process, nevertheless,
which is almost useless for improving our reliability. In a large class of cases, the pro-
cess of reflection is an exercise in self-congratulation. It does nothing, however, in
these important cases, to improve on the accuracy of our first-order beliefs.
We have a number of general tendencies which aid in this process of self-
misunderstanding. We have a very strong tendency to persevere in the beliefs we form.
Once we do form a belief, we are strongly disposed to scrutinize evidence against it,
and we are usually able to come up with commonsensical explanations for why the
evidence should not be trusted in this particular case. Evidence in favor of beliefs we
already hold, however, is not so carefully scrutinized, and it is typically taken at face
value. In addition, our memory is not even-handed in the way it treats evidence for and
against the beliefs we hold. Evidence in favor of existing beliefs is better remembered
than evidence against them. When we do stop to reflect on our beliefs and scrutinize
the evidence we can remember, it is no surprise that we find our beliefs to be well sup-
ported by the available evidence. The ways in which memory and reflection work
Sunpoint Securities
Sunpoint Securities was a full-service self-clearing broker-dealer based in
Longview, Texas. It started its business in 1989. It ceased operations on
November 18, 1999, when it became apparent that it did not have enough
assets to cover liabilities owed to its clients.
In a civil lawsuit, the U.S. Securities and Exchange Commission charged
the CEO and the CFO of Sunpoint with systematically stealing 25 million
dollars from a money market account that the firm maintained for its clients.
The SEC's complaint alleged that from December 1997 through November
18, 1999, Sunpoint illegally transferred money market funds, belonging to
its clients, to the firm's clearing account. The firm then improperly transferred
the funds to satisfy the firm's net capital requirements. The SEC further
alleged that the firm's president and CEO also used the funds for their
personal benefit. The diversion of client funds resulted in the firm having only
12 million dollars in its client money market account to cover 37 million
dollars in money market obligations to its clients. Consequently. Sunpoint was
grossly below its net capital and client reserve requirements.
For more information, see SEC Litigation Release no. 16366, dated November 19, 1999
at www.sec.gov/enforce/litigrel/lrl6366.htm.
settle their short sales, the lending brokers are no longer owners of record.
The new purchasers become the holders of record on the issuer's registry.
7.5.6.2 Brokerage Bankruptcies
When a broker goes bankrupt, traders who deposited assets with the bro-
ker risk losing those assets. Bankruptcies often occur when the broker in-
curs significant trading losses on its own account, when one or more of its
clients default on their obligations to the broker, or, most commonly, when
someone steals assets from the broker. Traders therefore should carefully
consider whether their brokers are creditworthy before they entrust their as-
sets to them.
Brokers naturally want to assure their clients that they are trustworthy
and creditworthy. To increase investor confidence, brokers publish their fi-
nancial accounts for all to see. They may also take out excess insurance poli-
cies to protect their clients. The insurance companies help regulate brokers
to ensure that they do not get into trouble.
Many other organizations subject brokers to regulatory oversight. Ex-
changes and clearinghouses regulate their members to ensure that they are
financially viable, to minimize the costs that insolvent members can impose
upon others, and to increase public confidence in their membership. Clear-
ing members that clear for other broker-dealers regulate them to avoid losses
that they may inherit if their broker-dealer clients go bankrupt. Broker-
dealer associations and governments regulate brokers for similar reasons.
The various regulators ensure that brokers have adequate capital
reserves to meet their obligations. They also ensure that brokers have well-
functioning managerial controls in place to prevent unexpected losses due
to negligence, stupidity, poor luck, or fraud. Finally, they require that bro-
kers have accounting systems which can quickly detect such problems. Bro-
kers accept these regulatory relationships in order to increase investor con-
fidence in their financial integrity.
The SIPC
The U.S. Congress created the Securities Investor Protection Corporation
(SIPC) in 1970 to increase confidence in U.S. brokers. Almost all broker-
dealers that register with the Securities and Exchange Commission are
automatically members of the SIPC. (The only exempt brokers are those who
exclusively distribute mutual fund shares, sell variable annuities or insurance,
or conduct their business outside the United States.)
If a brokerage fails, the SIPC distributes all securities registered in clients'
names and held by the firm to those clients. The remaining securities—those
held in street name—and cash are then distributed on a pro rata basis to
the clients. The SIPC will satisfy any remaining investor claims up to a
maximum of 100,000 dollars for cash and a combined maximum of
500,000 dollars for securities and cash. The SIPC makes such distributions
from a special fund it maintains for this purpose. The money comes from
assessments that the SIPC levies on its members and from interest that the
fund earns on its investments in U.S. Treasury securities. Should the SIPC
need more money to satisfy claims, it can borrow up to a billion dollars
from the U.S. Treasury.
The largest payout the SIPC has made was in the Sunpoint Securities
liquidation. The SIPC paid 31 million dollars to restore stocks and cash that
9,738 investors apparently lost to theft at Sunpoint.
When a brokerage goes bankrupt, clients must quickly file their claims
with the bankruptcy court. Most courts accept only claims filed within 30 or
60 days of the publication date of the bankruptcy. In any event, the law
prohibits the SIPC from satisfying any claims that it receives more than six
months after the bankruptcy is published.
Clients should be notified by mail when a brokerage with which they do
business goes bankrupt. In practice, clients may not receive notice because
their broker's records are poor or because clients failed to notify their
broker of a change of address.
If you maintain an account with a brokerage whose bankruptcy might
not immediately come to your attention, you should regularly open your
mail to make sure that your brokerage has not gone bankrupt. You should
also inquire quickly into your brokerage's financial health if you fail to
receive your monthly statement.
For more information, click on www.SIPC.org.
7.5.7 Summary
What Works Best When
Most brokers are honest, trustworthy, and creditworthy. They behave well You Never Use It?
because most brokers are good and honorable people; because they know
All types of security systems—
that a good name is good for their business; and because regulators, mar-
burglar alarms, guards,
kets, competitors, clients, and broker associations have established systems armies, criminal justice
to deter bad behavior. systems, and regulatory
Unfortunately, not all brokers behave well all the time. To weed out rogue systems—work best when they
brokers and to help traders recover losses, regulators maintain grievance deter bad behavior.
programs. Consequently, the better they
Traders who suspect that they have been defrauded should complain to work, the less necessary they
seem. Unfortunately, many
the appropriate regulator. Although many complaints are due to misunder-
people do not fully appreciate
standings, some are due to dishonest or irresponsible behaviors that typi- the value of deterrence when
cally stop only when regulators take disciplinary actions. If you have been confronted with its cost. The
defrauded, your complaint may eventually lead to reparations. At a mini- former is intangible, whereas
mum, your complaint may prevent someone else from losing. the latter is concrete.
than our perceptual faculties; imagine that we could have belief-producing mecha-
nisms which were perfectly reliable. More than this, imagine that these mechanisms
might operate in us unreflectively. We simply pick up information about the world in
a perfectly reliable manner without having to reflectively monitor our belief acquisi-
tion at all. Would we be missing out on something valuable here insofar as we fail to
reflect and thereby assert our autonomy? Would reflection be valuable, in such a situ-
ation, even if it compromised our reliability? From an epistemological point of view, I
cannot see why it would. Perhaps autonomy is a valuable thing that we should care
about in its own right, but if it is, I cannot see why we should think of it as epistemi-
cally valuable.9
From an epistemological point of view, we should value reflection to the extent that,
and only to the extent that, it contributes to our reliability. The appeal to the value of
autonomy either presupposes, what is not true, that reflective belief acquisition is likely
to be more reliable than unreflective belief acquisition, or it is simply an appeal to some
sort of extra-epistemic value. Epistemologically speaking, there is no reason to value
reflectively arrived at belief in general over unreflective belief.
III
We may, perhaps, however, better see what it is that Sosa has in mind by returning to
the issue about the connection between reflection and agency. What is it about reflec-
tion, and beliefs formed under the guidance of reflection, that makes it so important,
according to Sosa? Let us return to the passage I quoted earlier, this time pursuing it
just a bit further:
Reflection aids agency, control of conduct by the whole person, not just by peripheral modules.
When reasons are in conflict, as they so often are, not only in deliberation but in theorizing, and
not only in the higher reaches of theoretical science but in the most ordinary reasoning about
matters of fact, we need a way holistically to strike a balance, which would seem to import an
assessment of the respective weights of pros and cons, all of which evidently is played out
through perspective on one’s attitudes and the bearing of those various reasons. (Sosa 2004: 292)
A connection is being made here between reflective belief formation and agency. My
unreflective belief which simply registers the presence of the table, like my dog’s unre-
flective belief which registers the presence of his food, is merely passive. But when I
stop to reflect—something my dog cannot do—I become an agent with respect to my
beliefs. Sosa believes that human beings are epistemic agents, and our agency comes
into play when we form beliefs reflectively.10
9
One may hold, of course, that reflection has value even if it does not have epistemic value. On such a
view, however, the value of reflection would not be something that needs to be taken account of within an
epistemological theory.
10
It is for this reason, no doubt, that Sosa thinks that belief formation may, at least at times, when it is
apt, be something a person deserves credit for. (See Sosa 2007: chapter 5.) The notion of credit at work here
What is the view of cognition which is implicit in this suggestion? Consider, first,
the case of unreflective belief acquisition. The mechanisms at work in a person which
produce beliefs of any sort, including unreflective belief, are extraordinarily complex.
They are not, for the most part, available to introspection. When I form perceptual
beliefs, for example, my perceptual apparatus engages in a process of edge-detection,
which is made possible by way of mechanisms which are responsive to sudden changes
in illumination across my visual field.11 These mechanisms operate sub-personally.
They are, as Sosa puts it, “peripheral modules.” Edge-detection is not something which
I engage in, at least in standard cases of perception; rather, it is something done by
sub-personal mechanisms within me. Here, at least, I do not act. Mechanisms within
me are at work which simply produce perceptual beliefs.
How then are things supposed to be different when I engage in reflection? It will be
best to have a simple example before us. Suppose that I am serving on a jury in which
someone is charged with murder. Imagine as well that I don’t simply react to the evi-
dence presented. Instead, I stop to reflect. I self-consciously consider whether the evi-
dence presented supports a guilty verdict. Here, when I stop to reflect, is where
epistemic agency is supposed to be found. But where, precisely, does my agency come
into play?
There certainly are things that I do in the course of reflecting on the evidence pre-
sented at trial. I may focus my attention on various pieces of evidence and question
their relevance as well as their probity. The focusing of my attention is arguably
something that I do, as is the activity of questioning both the relevance and the pro-
bity of the evidence. So there is genuine agency at work here, at least if we accept
these commonsense accounts of what is going on.12 But activities of this sort, while
they are certainly present when a person reflects on his or her beliefs, are no different
in kind from various activities we all engage in when forming unreflective beliefs.
Thus, for example, just as I focus my attention on various bits of evidence when I
carry out my jury duties, I turn my head in the direction I wish to look when I form
various perceptual beliefs. Turning my head is certainly a voluntary activity; it is a
manifestation of my agency. But the fact that I turn my head voluntarily does not
show that my perceptual belief itself is a manifestation of epistemic agency. Whether
I turn my head is determined by my choice, but once my head is turned in a certain
is not merely the notion of causal responsibility, as when the proper spelling of a certain word in some text
is “credited” to the automatic operation of a spell-checking program. Talk of credit in this context is far
more substantive, as befits a virtue-theoretic approach.
11
See, for example, David Marr (1982).
12
I don’t believe that we should accept these commonsense accounts of our mental lives. Indeed, the
history of the cognitive sciences over the last fifty years seems to me to show very clearly that the phenom-
enology of mental processes is not even roughly reliable in producing an understanding of the mechanisms
which actually operate. It is not just that the phenomenology leaves out important features of those mech-
anisms. Rather, even when it comes to those features of the mechanisms which the phenomenology repre-
sents, it very often misrepresents their role. For a recent defense of this view, see Wilson 2002. I take the
commonsensical view at face value in the text here, however, not because I believe it to be correct, but
rather because I believe that it presents the best possible case in favor of epistemic agency.
direction, with my eyes open, and the lighting just so, my perceptual mechanisms
will simply operate in me in ways which have nothing at all to do with the fact that I
am an agent. The fact that I focus my attention, and question the relevance and pro-
bity of the evidence, thus show no more agency when I reflect than goes on in unre-
flective cases. Indeed, these activities not only show no more epistemic agency than
goes on in unreflective cases in human beings; they show no more epistemic agency
than goes on in lower animals when they form perceptual beliefs. But this is just
to say that these features of reflectively formed belief do not exhibit epistemic agency
at all.13
So, once again, we need to ask, just where are we supposed to find the workings of
our epistemic agency? As I’ve mentioned, there are a great many sub-personal pro-
cesses at work whenever we form unreflective beliefs. But in this respect too, reflec-
tion is no different. We certainly shouldn’t think that what goes on in reflection is
fully and accurately represented in its phenomenology, any more than it is in unre-
flective belief acquisition. So when we get done focusing on various bits of evidence,
and considering their relevance and probity, a host of sub-personal processes go to
work eventuating in the production of a belief. How, indeed, could things possibly
be any different? There is, after all, a causal explanation to be had of how it is that
beliefs are formed, whether belief acquisition is reflective or unreflective. We should
certainly not think that while unreflective belief acquisition takes place within a
causally structured series of events, leaving no room for epistemic agency (just the
workings of “peripheral modules”), reflective belief acquisition somehow takes place
somewhere outside the causally structured network of events. There is, of course, no
such location. But now the appeal to epistemic agency seems to be nothing more
than a bit of mythology. A demystified view of belief acquisition leaves no room for
its operation.
Epistemic agency does not seem to be a feature of belief acquisition generally. But
any attempt to tie agency to reflection seems doomed, since our agency is involved
only in actions which provide input to our belief-forming processes, not in the forma-
tion of belief itself. And this therefore provides us with no more reason to speak of
agency in the case of belief acquisition than the fact that eating is undertaken voluntar-
ily gives us reason to speak of digestion as a manifestation of our agency.
IV
In Sosa’s more recent work, epistemic agency has come to play a more central role. In
Sosa 2007, there is an analogy that runs throughout. Sosa begins the book with the
13
One might argue that one should attribute epistemic agency in the case where one turns one’s head in
order to see something, even though the processes by which perceptual beliefs are formed, once one’s head
is turned, are entirely automatic. On such a view, there is not only epistemic agency without reflection in
the human case; there is epistemic agency in creatures who are incapable of reflection. This is not, however,
Sosa’s view.
example of an archer shooting at a target. Sosa remarks that the archer’s performance,
like all performances, may be assessed along three different dimensions. We may ask
whether it was accurate (i.e., whether it succeeded in its aim); whether it was adroit
(i.e., whether it manifested a skill); and whether it was apt (i.e., assuming it was both
accurate and adroit, whether it was accurate as a result of being adroit). Sosa refers to
this as the “AAA structure.” He then goes on to say, “Beliefs fall under the AAA struc-
ture, as do performances generally” (Sosa 2007: 23). But there is certainly something
puzzling about this. Beliefs don’t seem to be performances. They are not actions. They
do not seem to be something that we do.
Sosa gives a very brief response to this worry. He comments:
Some acts are performances, of course, but so are some sustained states. Think of those live
motionless statues that one sees at tourist sites. Such performances can linger, and need not be
constantly sustained through renewed conscious intentions. The performer’s mind could wan-
der, with little effect on the continuation or quality of the performance. (Sosa 2007: 23)
But this does not really respond to the concern. The worry is not about whether there is
a conscious intention at work here. The worry is about whether there is any intention at
all. The performer who remains motionless is clearly doing so as a result of an inten-
tion. No one can remain as motionless as these performers do without intending to do
so. But believers, at least typically, do not form beliefs as a result of an intention. At least
typically, when I look at a table in front of me and come to form the belief that there is a
table, I am not moved by any intention, conscious or otherwise, any more than my dog
is moved by an intention to form a belief when he comes to believe that there is food in
his dish. So the suggestion that we may see belief formation, like the shooting of an
arrow, as a kind of performance seems to be just a mistake.14
Sosa’s most recent work, Sosa 2011, continues this theme. The chapter on epistemic
agency begins as follows:
Some performances are consciously aimed at a certain outcome, as when an archer aims his
shot at a target. Some have an aim in a broader sense, as when a heartbeat aims to help the
blood circulate. Performances may be said to be “endeavors” when they have a certain aim,
even if the aim is not conscious. Such a performance is assessable as correct or incorrect, in
terms of whether it attains its “aim.” (Sosa 2011: 14)
Here we see the same problem that emerged in the earlier work. Sosa begins by talking
about performances, such as an archer shooting at a target. The archer has an aim,
whether conscious or not, and his performance of shooting the arrow is an intentional
14
The case of perceptual belief is a particularly pure case because in many such cases, no intention plays
any role in the formation of the belief. One might think, however, that there are many beliefs in which
intentions do play some significant role. We do, after all, undertake lines of inquiry with the intention of
addressing various problems. This will still not help Sosa, however, since he wishes to offer an account on
which believing generally is a performance. Moreover, even in the cases where an inquiry is the product of
some intention, it seems that it is the act of investigating which is a performance, rather than the believings
which result from the investigation.
action. It involves genuine agency. But in the second sentence, Sosa speaks of an aim
“in a broader sense,” “when a heartbeat aims to help the blood circulate.” But heartbeats
do not literally have aims at all. Heartbeats are not agents; they do not have intentions;
they do not have goals; they do not have aims. They do certainly have functions, but
that is another matter entirely. Now in the third sentence, Sosa says that performances
may count as “endeavors” “when they have a certain aim, even if the aim is not con-
scious.” Now, of course, the archer has an aim or goal even when that aim or goal is not
consciously entertained. The shooting of the arrow is a genuine action; it is a perfor-
mance; it is an endeavor. But the heartbeat is entirely unlike the archer, and the differ-
ence does not lie in whether the aim of the heartbeat is conscious, for the heartbeat has
no aim, and, having no mental states, it cannot have states which are either conscious
or unconscious. Talk of an aim, in the case of the heartbeat, is at best metaphorical. It
would be better, I think, to talk of its function. Sosa seems to recognize this when he
says that the heartbeat has an aim “in a broader sense” and when, in the last sentence
quoted, he puts the word ‘aim’ in scare quotes.
Once all of this is made clear, it is difficult to see why we should talk of epistemic
agency. Just as heartbeats are not agents, and only have aims in some broader sense,
namely, they have functions, beliefs too are not agents, and do not have aims, but only
have certain functions. Beliefs are not performances, not because the aim of belief is
not consciously entertained, but rather because beliefs do not literally have aims at all.
Now perhaps this is being unfair to Sosa. Although Sosa does speak of belief as a
performance, and he does speak of beliefs as having aims, perhaps the suggestion,
charitably interpreted, should be that believing is a performance, and that believers
have aims in forming their beliefs. Thus, when discussing the suggestion that the sole
aim of belief might be truth, Sosa remarks, “There is such a thing as wishful thinking of
a sort that aims at the intellectual comfort of the believer.” Now in this sort of case, one
might reasonably conjecture that belief acquisition is an intentional activity. The
believer has a certain aim—to form beliefs on a particular subject in such a way as to
receive comfort—and this aim, though not consciously accessible to the believer, plays
a role in directing the course of belief acquisition. Believing, in this kind of case, would
then be an intentional activity; a performance; an endeavor. The aim here is the aim of
the believer, and only indirectly, and metaphorically, the aim of the particular belief.
Indeed, Sosa says something very much like this:
In believing one might or might not be endeavoring to attain truth. Believings whereby one is
not so endeavoring cannot constitute knowledge . . . Believings constitute knowledge only when
the believer thereby endeavors to attain truth, which is not always the case. (Sosa 2011: 15–16)
Here the aim is rightly attributed to the believer, not to the belief itself. But now, the
suggestion is that, just as the believer who is self-deceived forms a belief intentionally,
since his belief acquisition is directed by his desire to avoid discomfort, the ordinary
believer forms his or her belief intentionally as well, and belief acquisition is directed
by his or her desire for truth. But although it may be quite plausible to suggest that in
cases of self-deception we should see belief acquisition as an intentional activity, it is

far more controversial to suggest that ordinary belief acquisition is properly viewed as
some sort of intentional activity or performance.
Let us begin with the case of perceptual beliefs. I open my eyes and see a table in
front of me and thereby come to acquire the belief that I am standing in front of a table.
The perceptual mechanisms which come into play here work automatically; they do
not require any mediating aim, or goal, or intention on the part of the believer. I may
sincerely care about having true beliefs about what is in front of me, or I may be indif-
ferent to the matter entirely. My perceptual mechanisms, and the cognitive mecha-
nisms to which they are linked, will simply go to work whatever my concerns. If I do
have an earnest desire to believe the truth about what is in front of me, or to believe
truths in general, it seems that these desires are epiphenomenal with respect to the
fixation of perceptual belief, at least in the ordinary case. After all, infants and non-
human animals form perceptual beliefs, and they lack the conceptual resources to
form a desire for truth in general, or truth on some particular topic. It cannot be a
coincidence that the manner in which my perceptual belief is produced seems to mir-
ror theirs exactly. Any desires I may have with respect to true belief are simply failing to
play a role in how my perceptual beliefs are processed. And if Sosa then insists that
“believings constitute knowledge only when the believer . . . endeavors to attain truth,”
then perceptual knowledge thereby becomes an impossibility, not only for infants
and non-human animals, but for fully functioning adults as well. And this, I think, is
a problem.
Nor is more sophisticated belief obviously better off. A sophisticated agent, engaged
in inquiry, may undertake a variety of activities which are motivated by a desire to
determine the truth about some particular matter. In this, we see the goal of believing
truths actually playing some causal role in directing the agent’s behavior. But as we saw
in the discussion of the juror, this kind of intentional action plays a role in providing
the mechanisms of belief acquisition with input, but the workings of these mechanisms
themselves is not the product of any sort of action, intention, or aim. The believing
itself does not involve an exercise of agency, for it is produced by mechanisms which
operate automatically.15 If believing itself must be an intentional activity guided by the
aim to have true beliefs in order for knowledge to result, it seems that knowledge
becomes unattainable. The mechanisms of belief acquisition, retention, and modifica-
tion do not work in ways which involve the mediation of our desires, at least in the
15
An anonymous referee raised the following very interesting objection: “I am not sure that production
of a result by automatic mechanisms precludes exercises of agency. Suppose I am standing in the rain and
intend to get wet. I know I can achieve this result by not getting out of the rain. As a result, I don’t move
and get wet. Here my getting wet involves an exercise of agency even though my getting wet is produced by
(meteorological) mechanisms which operate automatically.” While I certainly intend to get wet in the case
described, my getting wet is something that happens to me rather than something that I do. Standing in the
rain is a performance; getting wet is not. Similarly, I would want to say that, in the jury case, focusing one’s
attention on the evidence is a performance; forming a belief about the guilt or innocence of the accused
is not.
ordinary cases which do not involve wishful thinking and other forms of self-deception.
So Sosa’s account seems to leave us not only without any perceptual knowledge, but
without any knowledge at all.
V
Why is Sosa so concerned to see beliefs as performances, or believings, as perfor-
mances, guided by the aim to attain the truth? On Sosa’s view, the normative con-
straints on belief derive from the aims with which they are formed. Knowledge is
attained only when a particular act of believing achieves its aim. So epistemic agency
plays a central role, setting the normative standards which belief must meet. And, as
we have seen, such a view has the consequence that if there is no epistemic agency—if
belief acquisition is not a voluntary activity, or performance, or endeavor—then there
is no knowledge, and there are no normative standards which belief must meet. The
result is a kind of epistemic nihilism.
Why should we hold knowledge and epistemic normativity hostage to such a con-
troversial view about epistemic agency?16 Sosa’s idea seems to be that it is only by way of
an aim which plays a role in a performance of believing that normative standards may
be set. When there is no aim, and no performance, we are left without standards for
belief to meet. But we have already seen one clear alternative to this view. Just as a
heartbeat may have a function even apart from anyone having an aim involving that
heartbeat, believing may have a function which is quite independent of any aims that
the believer may have. And we may now assess particular cases of belief acquisition
by the extent to which they carry out that function. Particular heartbeats may carry out
the function of a beating heart either well or badly, and the same is true of particular
cases of belief acquisition. No one’s aim is needed to make sense of this sort of norma-
tive standard.
And much as the appeal to functions as a way of setting standards provides one way
to avoid epistemic nihilism, it is certainly not the only way. We may assess the workings
of our cognitive mechanisms against any number of different standards, where the
standards are set by something other than the believer’s particular aims. One might
think, for example, that there is something special about true belief, a reason why hav-
ing such beliefs is in the interest of every agent, whatever their particular aims and
goals, whether they actually have the desire or aim for true belief or not.17 On such a
view, the fact that true belief has this special feature gives rise to certain normative
demands. But these demands do not arise from any aim or goal or particular desire
which any believer must have.
16
The objection here is an attempt to develop the objections which Sosa considers on pages 19 and 20
of Sosa (2011).
17
I have defended such a view in chapter 5 of Kornblith (2002).
Much of what Sosa says about apt belief remains entirely unaffected by a move away
from the account of normativity which ties it to epistemic agency. Apt belief remains
important even if we give up the view of beliefs or believings as performances. And this
is a good thing, for there is a great deal of insight to be found in Sosa’s discussion of the
importance of apt belief. If we can retain all of this even without a commitment to
epistemic agency, then I believe that our epistemology, and epistemic normativity in
particular, will have a firmer foundation.18
References
Adler, J., and L. Rips (eds) 2008. Reasoning: Studies of Human Inference and its Foundations
Bargh, J. A., M. Chen, and L. Burrows, 1996. “Automaticity of Social Behavior: Direct Effects of
Trait Construct and Stereotype Activation on Action.” Journal of Personality and Social
Psychology 71: 230–44.
BonJour, L., 2006. “Kornblith on Knowledge and Epistemology.” Philosophical Studies 127:
317–35.
Chomsky, N., 1965. Aspects of the Theory of Syntax. Cambridge, MA: MIT Press.
Devine, P. G., 1989. “Stereotypes and Prejudice: Their Automatic and Controlled Components.”
Journal of Personality and Social Psychology 56: 5–18.
Fehrman, K., and C. Fehrman, 2003. Color: The Secret Influence, 2nd edn. Englewood Cliffs, NJ:
Prentice-Hall.
Feldman, R., 2003. “Chisholm’s Internalism and its Consequences.” Metaphilosophy 34:
603–20.
Garnham, A., and J. Oakhill, 1994. Thinking and Reasoning. Oxford: Blackwell.
Garret, J., and C. Brooks, 1987. “Effect of Ballot Color, Sex of Candidate, and Sex of College
Students of Voting Age on their Voting Behavior.” Psychological Reports 60: 39–44.
Gilovich, T., D. Griffin, and D. Kahneman (eds) 2002. Heuristics and Biases: The Psychology of
Intuitive Judgment. Cambridge: Cambridge University Press.
Gopnik, A., 1993. “How We Know our Minds: The Illusion of First‑Person Knowledge of
Intentionality.” Behavioral and Brain Sciences 16: 1–15, and 90–101.
Halberstadt, J., and T. Wilson, 2008. “Reflections on Conscious Reflection: Mechanisms of
Impairment by Reasons Analysis.” In J. Adler and L. Rips (eds) Reasoning: Studies of Human
Inference and its Foundations , 548–65. Cambridge: Cambridge University Press.
Hauser, M., “Moral Sense Test”, <http://mprlab327.webfactional.com> accessed July 7, 2015.
Hauser, M., et al. 2007. “A Dissociation between Moral Judgments and Justifications.” Mind and
Language 22: 1–21.
Kahneman, D., P. Slovic, and A. Tversky (eds) 1982. Judgment under Uncertainty: Heuristics
and Biases. Cambridge: Cambridge University Press.
Kornblith, H., 2002. Knowledge and its Place in Nature. Oxford: Oxford University Press.
Kunda, Z., 1999. Social Cognition: Making Sense of People. Cambridge, MA: MIT Press.
18
I am indebted to the audience at UNAM for helpful comments, and especially to Ángeles Eraña, Ernie
Sosa, and two anonymous referees.
Lucas, E. J., and L. J. Ball, 2005. “Think-Aloud Protocols and the Selection Task: Evidence for
Relevance Effects and Rationalisation Processes.” Thinking and Reasoning 11: 35–66.
Marr, D., 1982. Vision. San Francisco:W. H. Freeman.
Merikle, P. M., 1992. “Perception without Awareness: Critical Issues.” American Psychologist 47:
792–5.
Nisbett, R., and L. Ross, 1980. Human Inference: Strategies and Shortcomings of Social Judgment.
Englewood Cliffs, NJ: Prentice-Hall.
Nisbett, R., and T. Wilson, 1997. “Telling More than We Can Know: Verbal Reports on Mental
Processes.” Psychological Review 84: 231–59.
Roush, S., 2009. “Second Guessing—A Self-Help Manual.” Episteme 6: 251–68.
Rubinoff, M. R., and D. T. March, 1980. “Candidates and Color: An Investigation.” Perceptual
and Motor Skills 50: 868–70.
Sosa, E., 1991. “Knowledge and Intellectual Virtue.” Repr. in Knowledge in Perspective: Selected
Essays in Epistemology, 225–44. Cambridge: Cambridge University Press.
Sosa, E., 1997. “Reflective Knowledge in the Best Circles.” Journal of Philosophy XCIV:
410–30.
Sosa, E., 2004. “Replies.” In Greco (ed.), 275–325.
Stanovich, K., 1999. Who Is Rational? Studies of Individual Differences in Reasoning. Mahwah,
NJ: Laurence Erlbaum.
Tversky, A., and D. Kahneman, 1974. “Judgment under Uncertainty: Heuristics and Biases.”
Science 185: 1124–31.
Wason, P., and J. St. B. T. Evans, 1975. “Dual Processing in Reasoning.” Cognition 3: 141–54.
Wilson, T., Strangers to Ourselves: Discovering the Adaptive Unconscious. Cambridge, MA:
Harvard University Press.
11
The Value of Reflection
Stephen R. Grimm
Unlike other animals, human beings do not simply have beliefs and desires. We are
also the kind of beings who can take a step back (or a step up?) and reflect on our beliefs
and desires. We can do that, but why exactly would we want to? What is the good of it?
This, it turns out, is not so easy to explain. It is clear that the ancient Greeks held
reflection in high esteem, for instance, but their reasons are unclear. Thus the Delphic
Oracle tells us to “Know Thyself,” but not why this reflective knowledge is a good thing
for humans to have, or why a life without it would be diminished.
A similar preference for reflection lives on in the epistemology of Ernest Sosa, and
in his influential distinction between animal and reflective knowledge. For Sosa,
animal knowledge comes about when one believes the truth as a result of one’s intel-
lectual abilities—for example, when you form a true belief that the cup is on the table
as a result of your visual ability to pick out and identify cups. Beyond the animal,
however, Sosa claims that human beings are also capable of knowing something in a
reflective way. Thus we are capable not just of knowing that the cup is on the table (as
an animal might), but of knowing this in light of our further knowledge that the
source of our belief, in this case our vision, is trustworthy or reliable. Reflective
knowledge is thus “enlightened knowledge” (Sosa 2009: 153). It is knowledge that
does not just result in a mechanical way from our faculties but is instead guided by,
and gains some sort of extra status from, our further knowledge that the source of
our belief is reliable.
Although Sosa is an heir of Socrates in valuing reflection, his focus is also clearly
different. For Socrates, the emphasis is on reflecting on one’s own views regarding what
is good or valuable or just. For Sosa, the emphasis is on reflecting on all of one’s beliefs,
and evaluating whether their sources are (or were) trustworthy. Further, while
Socrates’ judgment about the unexamined life is quite harsh—that it is not worth liv-
ing—Sosa’s view of “unexamined” or animal knowledge is considerably more gener-
ous. None of us, he says, would disdain acquiring the truth as a result of our animal
abilities, just as none of us would disdain acquiring gold in the dark. But it is much
better, Sosa claims, to acquire the truth while guided by a reflective perspective on our
184 Stephen R. Grimm
own abilities; this would be like acquiring the truth in the light of day, where one can
properly savor or appreciate it, or at least, where we can put to rest one’s doubts.
Animal knowledge therefore is not epistemically bad, it is just that reflective knowl-
edge is epistemically “better” (2007: 129, 2009: 193, 2011: 13); it deserves a certain
“pride of place” (2004: 291) in our intellectual lives.1 What I would like to do in this
essay, however, is argue that the reasons that Sosa offers in defense of this view are
unsatisfying. More exactly, I will argue that several of the reasons he provides do not in
fact favor reflective knowledge over animal knowledge while others favor not just
attaining a reflective perspective on one’s animal knowledge, but towards a perspective
on that perspective, and beyond. The question then becomes whether there is any
defensible sense in which it is better to know reflectively as opposed to animally, and I
will claim that there is such a sense, but it is not the one identified by Sosa: namely, that
those who know reflectively exercise a kind of autonomy over their beliefs that those
who know animally do not. Reflection therefore turns out to be valuable on this view
not for what we would normally think of as “epistemic” reasons, but rather for some-
thing like moral or ethical reasons. In order to appreciate these points, however, we
first need to get a better sense of what the distinction between animal and reflective
knowledge amounts to in Sosa’s work.
I. Animal vs Reflective
According to Sosa, a belief amounts to animal knowledge just in case it is apt: that is,
just in case the accuracy of the belief manifests the believer’s competence at attaining
the truth in this area (2011: 24). A bit of animal knowledge amounts to reflective
knowledge, moreover, just in case that apt belief is aptly noted at the second level; that
is, just in case the aptness of the object belief is guided by one’s (apt) meta-belief that
the source of one’s object belief is reliable. To have reflective knowledge that my neigh-
bor’s dog is in the yard therefore requires not just that I aptly believe, as a result of my
reliable eyesight, that the dog is in the yard, but also that this belief is guided by my apt
second-order belief that my eyesight is reliable in these circumstances.
A few points of clarification. First, Sosa sometimes insists that one can have a
second-order perspective on one’s beliefs—and in particular, on the reliability of one’s
faculties—that is only implicit, or not brought to full awareness.2 This complicates
things, as Hilary Kornblith (2004) notes, because it is likely not just that humans but
also other animals have some sort of implicit perspective on the reliability of their
faculties—in which case my neighbor’s dog might not just have animal knowledge
that, say, a squirrel is in front of her, but reflective knowledge of this same proposition.
1
Alternatively, the claim is sometimes that reflective knowledge is “higher” (2009: 151), or is “the best
human knowledge” (2009: 153).
2
E.g., “Conscious reflection on the spot is not required, however, since a second-order perspective can
work beneath the surface of consciousness” (Sosa 2004: 291–2).
the value of reflection 185
Here I set this complication to one side: as I am understanding Sosa’s thesis, in order to
know reflectively one must actually reflect.3 This might seem obvious, but some of the
subtleties of Sosa’s view make it worth saying.
Second, there is one way in which the question of why reflective knowledge is better
than mere animal knowledge has an easy answer—by my lights, too easy of an answer,
and hence one which we should again set to one side.4 Thus one might claim that
reflective knowledge that p is epistemically better than mere animal knowledge that p
simply because there is more “epistemic stuff ” on the scene when one reflectively
knows—not just the first-order animal knowledge but the second-order reflective
knowledge as well. Call this the “trivial answer” to our original question, one that
claims that reflective knowledge is better simply because it constitutes more knowl-
edge (cf. Sosa 2004: 285). I take it that this is not the right way to think about Sosa’s
claim because on his view it is the animal knowledge itself that is supposed to be
improved by the addition of the perspective (see, e.g., Sosa 2011: 13). It is this claim,
not the trivial one, which we are contesting.
II. The Main Argument

With these points in mind, Sosa’s most common argument in favor of the betterness of
reflective knowledge is that in the absence of reflective knowledge one’s beliefs run the
risk of incoherence.5 Suppose you are living contentedly with your animal beliefs but
then begin to wonder whether your beliefs are justified or amount to knowledge.
Unless you can answer in the affirmative, Sosa claims, your animal beliefs will then be
seriously undermined. As he writes:
Nevertheless, a mere thermometer reaction to one’s environment cannot constitute the best
human knowledge, regardless of whether that reaction is causally mediated by experience. It is
not enough that one respond to seeing white and round objects in good light with a ‘belief ’ or
‘proto-belief ’ that there is something white and round. Suppose one asks oneself ‘Do I know
that this is right and round’ or ‘Am I justified in taking this to be white and round?’ and one has
to answer ‘Definitely not’ or even ‘Who knows? Maybe I do know, maybe I don’t; maybe I’m
justified, maybe I’m not.’ In that case one automatically falls short, one had attained only some
3
As when Sosa talks of reflective knowledge where the source of one’s belief is explicitly, reflectively
“seen” or judged to be reliable (cf. Sosa 2004: 300). Two anonymous reviewers have pressed me on this
point, arguing that the emphasis on belief over-intellectualizes Sosa’s account, so let me add a bit more: If
one supposes that the reflective monitoring can be highly implicit and “beneath the hood,” then the inter-
est, and perhaps even the legitimacy, of Sosa’s distinction between animal and reflective knowledge would
be in danger. For it would then be hard to think of a case in which animal knowledge exists in the absence
of reflective knowledge, because any animal that has this sort of (reliable) monitoring ability will count as
having reflective knowledge—and that of course includes young children too. Whereas to begin with
reflective knowledge seemed like a great accomplishment, because it allowed for properly human knowl-
edge, rational defensibility, and the like, reflective knowledge would then almost come “for free,” once these
monitoring abilities are on the scene.
4
Though towards the end of the essay I will again return to it.
5
See, for example, Sosa (2004: 302, 310; 2007: 115, 132; 2009: 146, 153).
lesser epistemic status, and not any ‘real, or enlightened, or reflective’ knowledge. The latter
requires some awareness of the status of one’s belief, some ability to answer that one does know
or that one is epistemically justified, and some ability to defend this through the reliability of
one’s relevant faculties when used in the relevant circumstances. (Sosa 2009: 153)
In short, for those who consciously wonder about the status of their beliefs, Sosa claims
that three options present themselves. (To narrow things down, suppose one is only
wondering about some particular belief that p.)
Option A: One reflects on the status of one’s belief that p and comes to think p is
“definitely not” something one knows or is justified in believing.
Option B: One reflects on the status of one’s belief that p and comes to think, “maybe
I do know that p and maybe I don’t; maybe I’m justified, maybe I’m not.”
Option C: One reflects on the status of one’s belief that p and comes to think p is
something one knows or is justified in believing.
Sosa unsurprisingly takes Option C to represent the person who has reflective knowl-
edge. And his claim, again, is that Option C is definitely better than Option A and very
plausibly better than Option B. Why? Elsewhere he makes it clear that:
Answer (a), and even answer (b), would reveal a certain lack of integration in that stretch of
consciousness; only answer (c), of the three entirely avoids disharmony within that conscious-
ness at that time. If one has to give answer (a), or even answer (b), one thereby falls short, and
one’s belief that p itself falls short. That belief is then not all it could be. (Sosa 2007: 115)
In other words, the idea is that Option C is better than Options A or B because only C
displays intellectual coherence—only C “entirely avoids disharmony within that con-
sciousness at that time.”
What should we make of this argument? Notice, for one thing, that it is a much more
limited version of the claim we began with, which seemed to hold, unrestrictedly, that
reflective knowledge is better than animal knowledge. The argument here, by contrast,
at best shows that for those who reflect (or wonder, or express doubts about the status of
their beliefs) reflective knowledge is better. The only way to return to the more unre-
stricted version of the claim, as far as I can see, would be not just to say, “For those who
reflect, reflective knowledge is better,” but rather that human beings should so reflect
(or wonder or doubt). Or maybe that human beings are only properly satisfied, or can
only bring inquiry to a proper close, when they have attained this sort of reflective
endorsement. This would add an additional normative dimension to Sosa’s claim that
it seems he needs, if he is to defend its unrestricted form.
In sections 3 and 4 I will return to further thoughts Sosa offers along these lines, but
it is worth pointing out now that even this more limited defense of the value of reflec-
tive knowledge does not obviously get him everything he wants. For even if we agree
with what we might call the meta-incoherence norm—that for those who doubt, one
has cognitive disharmony (or meta-incoherence) in the absence of reflective knowl-
edge—it is nonetheless true that this norm can be satisfied in two different ways, both
recognizably Sosian. Thus on the one hand you could reflectively endorse your first-
order belief in an animally apt way: perhaps simply by affirming, at the meta-level,
your Reidian trust in the reliability of your first-order faculties. On the other hand, you
could reflectively endorse your first-order beliefs in a way that is backed by what Sosa
calls a “structure of reasons” (2007: 121)—perhaps, as he suggests, reasons you acquire
via cognitive science that support the reliability of your perceptual faculties.
It is clear however that Sosa thinks only the second sort of reflective knowledge, the
one backed by the structure of reasons, is the one we really want. Why is that? Because,
remember, the move to the meta-level in the first place was prompted by doubt that
one’s first-order belief was reliably formed, and Sosa seems to think that only reasons
can then dispel this doubt, not some sort of animal trust or faith.
Notice though how the argument has now shifted. What began with an appeal to the
uncontroversial norm that meta-incoherence is to be avoided has now evolved in
response to a more controversial norm—call it the doubt norm—along the following
lines: that doubt can only be properly resolved by appeal to reasons. Once the doubt
norm is on the scene, however, the main question is how it might be contained. If one
accepts on the meta-level not just that sense perception is reliable, but that sense per-
ception is reliable because of x, y, and z (where x, y, and z might be certain claims from
cognitive science), then it is natural to wonder how claims x, y, and z might escape
from doubt. But if they are subject to doubt, don’t we need reasons to support them as
well? And so on.
In the following sections we will see that Sosa tries to evade the threat of regress by
denying that doubt is cogent at the meta-level. Before we turn to this claim, however, it
will help to ask why Sosa thinks there is normative pressure to doubt or reflect at the
first-order level to begin with.
III. Other Norms of Belief

We have seen that for those who doubt the status of their beliefs, reflective endorse-
ment of some sort seems necessary to stave off cognitive disharmony. But to return to
our earlier question: What about those who fail to doubt or reflect? One can hardly say
that their beliefs are meta-incoherent for the simple reason that they have no meta-
level beliefs that might fail to cohere. Or again, what about the person who fails to
reflect not out of thoughtless neglect, but (as it were) willfully? The sort of person, that
is, who declines to reflect because he thinks the activity is useless or idle or maybe even
decadent.6 Suppose this person has animal knowledge that, say, his cup is on the desk,
but when asked whether he ever wonders if his vision is accurate, simply replies, “No
thanks, I’m fine.” What if anything is wrong with this response?
6
Here I leave out of consideration the still more complicated case where one might think, based on
empirical work, that reflection is an unreliable process, one that is apt to lead us farther away from the
truth. For more on this, see Tiberius (2010).
One intriguing reply is that resting content at the animal level is inadequate because
it somehow falls short of our natural goals as believers, and hence that there is some-
thing lacking or deficient about a belief that is not subjected to doubt or reflection. In
this section I will consider two ways Sosa tries to support this claim.
The first is that as human beings, or perhaps as rational or intellectual beings, we
have a natural desire to understand the world that leaves us unsatisfied with mere ani-
mal knowledge that p. Thus Sosa claims that when we consider what we care about
from an epistemic point of view, “several epistemic values stand out” (2009: 136). To
begin with, we want beliefs that are both (a) true and (b) safe. More than that, however,
we would like:
(c) Understanding/explanation: often we would like not only to know a given thing, but also
to understand it, to have an explanation. (And this leads to the next item.)
(d) Coherence: we would prefer that our minds not house a clutter of mere facts sitting there
loose from one another. (2009: 136–7)
In another passage he likewise stresses the tight connection between understanding

and coherence, along with their interlocking value:
Note how the explanatory coherence of a body of beliefs comports with its providing the sub-
ject a corresponding yield in understanding. The cement that binds Bp and Bq in that body
coherently would be a further belief of the form B(p because q) or B(q because p). And such
a belief would embody some measure of understanding as to why p or as to why q. Therefore,
the value of such coherence, explanatory coherence, is allied to the value of understanding.
(2004: 312)
All told, the idea therefore seems to be that when we have mere animal knowledge that
p, we have knowledge that satisfies our preference for the true and the safe, but not
knowledge that satisfies all of our epistemic preferences. In particular, it fails to satisfy
our preference to understand why p, a kind of understanding that brings with it a cor-
relative coherence among our beliefs (as opposed to a mere clutter).7
For the sake of argument let us grant there is something unsatisfying or incomplete
about merely knowing that p, and that we naturally desire to understand why p as well
(at least often, or characteristically, or insofar as we are human). What remains unclear,
however, is how exactly this claim is supposed to support the value of reflective knowl-
edge. Suppose I animally know that my coffee just spilled, and that as a human being I
naturally desire to push beyond this knowledge in order to achieve an understanding
of why it spilled (along with the correlative coherence this would bring). Surely what I
would need in order to achieve an understanding of this sort is not some sort of per-
spective on the reliability of my perceptual faculties, but rather some knowledge of the
“why” of p, or the cause of p: in this case, let us suppose, the knowledge that my coffee
7
As he reiterates elsewhere: “Understanding is intimately connected with coherence, since the explan-
atory interrelationship among our beliefs is bound to function as a, or even the, main component of epis-
temically efficacious coherence” (Sosa 2004: 298).
spilled because I bumped the desk with my knee. Notice, however, that this extra bit of
knowledge is quite different than the sort of reflective knowledge that Sosa has in
mind. For one thing, it is still an item of first-order knowledge, one that says nothing
about the reliability of my faculties; for another, it is the sort of knowledge that is fully
available to non-human animals and unreflective human beings. If the psychologists
are right, animals at least from mice on up seem capable of grasping the causal depend-
encies that obtain in the world.8
In addition to the desire for understanding, a second reason why Sosa thinks there is
normative pressure to move beyond the animal level is that as a rational human being I
should doubt my first-order beliefs. Or at least, I should be concerned about the possi-
bility of error among those beliefs. Thus Sosa approvingly quotes the ancient
Pyrrhonian Sextus:
Let us imagine that some people are looking for gold in a dark room full of treasures . . . .[N]one
of them will be persuaded that he has hit upon the gold even if he has in fact hit upon it. In the
same way, the crowd of philosophers has come into the world, as into a vast house, in search of
truth. But it is reasonable that the man who grasps the truth should doubt whether he has been
successful. (Quoted in Sosa 2007: 129, italics my emphasis)
What I should want then, is not just to believe that p in a way that I hope amounts to
knowledge—I should want to believe that p in a rationally defensible way, or in a way
that protects against doubt. And the move to the reflective level is apparently what it
takes to resolve these doubts, or to make my belief defensible in the arena of reflection.
Anything less would (or at any rate should) fail to satisfy me as a rational being.
What should we say about this? To begin with, we should resist the idea that doubt
can only be removed, or rational defensibility can only be secured, by moving to the
meta-level. Suppose that based on past experience I believe that my local bank will be
open this Saturday. You then tell me that several local banks have been eliminating
their Saturday hours in order to save money, leading me to doubt my belief. How might
I resolve this doubt? Why not: by calling up my bank and asking whether they will be
open on Saturday? Surely this strategy would be more natural and effective than taking
some sort of meta-perspective on my faculties. But then it seems that doubts can legit-
imately be resolved without moving to the meta-level at all.
Or again, suppose my wife tells me that the car keys are near the toaster, that I believe
her about this, but then I come to wonder whether the keys are really by the toaster,
knowing that the kids sometimes move things around. How could I resolve my doubt?
Here again, why not by going and taking a look? Then I will know for sure, and in a way
that legitimately dispels my previous doubt—meta-level not required.9
8
For more on this see Blaisdell et al. (2006) and Taylor et al. (2009).
9
Perhaps what Sosa will say here is that I have now achieved a kind of (implicit?) perspectival endorse-
ment of the previous source of my belief, but is that really plausible? In the bank case, for example, would
it have amounted to a perspectival endorsement that induction is reliable? It seems implausible, and
unnecessary, to attribute any such perspective to me, even an implicit one.
Perhaps it will be said here that my doubts should not be so easily removed, or that
rational defensibility is not so easily acquired. There is, after all, still the possibility that
someone was playing a trick on me when I phoned the bank, or that I was hallucinating
when I saw the keys near the toaster. But the more exotic the doubts become, the less
obvious it likewise becomes that I “should” resolve these doubts. Arguably a rational
human being could responsibly dismiss these doubts as idle, or exotic, or decadent, in a
way that the previous doubts could not responsibly be dismissed.
There is also a more profound worry for Sosa, one that we gestured at in the previous
section. Suppose the Pyrrhonians are right not just that we can doubt our first-order
beliefs but that we should doubt them—and moreover that only second-order reflec-
tive knowledge will adequately resolve our doubts. The problem then is that this
normative pressure will presumably apply not just to the object level but to the
second-order level as well. For instance, suppose I believe that my neighbor’s dog
is in the yard, that I begin to wonder whether the source of my belief (perception) is
reliable in these circumstances, and then as a result of studying the relevant cognitive
science I come to believe that the source is thus reliable. My first-order belief that the
dog is in the yard is now, on Sosa’s view, appropriately enlightened—it is now ration-
ally defensible, or so it seems—but what about now my meta-belief that perception is
reliable, based on the results of the lab? Can’t I—indeed, if the earlier reasoning was
on target, shouldn’t I—doubt that too? Isn’t it possible that all of the experiences in
the cognitive science lab were hallucinated? Or that (to take one of Sosa’s own recent
scenarios) all of my reflective reasoning was performed under the influence of the
Disablex, a pill that “induces a persistent illusion of coherent empirical reality”
(2011: 153)? Based on the Pyrrhonian reasoning, one would think that the same nor-
mative pressure to acquire a rational defense against doubt applies just as much on
the second-order level as it did on the object level. But then how in the world could
I do that?10
Sosa’s response to this concern is apparently to reject this last step and deny that the
same normative pressure is, in fact, felt at the meta-level. As he writes in response to a
similar objection by Barry Stroud:
However, it is not easy to understand this [Stroud’s] position. If our perceivers believe (a) that
their perception, if reliable, yields them knowledge, and (b) that their perception is reliable,
then why are they restricted to affirming only the conditional, a, and not its antecedent, b? Why
must they wonder whether they understand their relevant knowledge? Indeed, to the extent
that they are really convinced of both a and b, it would seem that, far from being logically con-
strained to wondering whether they know, they are, on the contrary, logically constrained from
so wondering. After all, first, if you are really certain that p, then you cannot well consider
10
There are of course strategies. With Descartes one could argue a priori to the claim that God is no
deceiver, hence would not allow it. Or, to take Sosa’s interesting recent strategy, one could argue transcen-
dentally that the Disablex (at least) is impossible. I do not think those strategies work, but for present
purposes it is enough to note that even on Sosa’s own view I do not think they are necessary, for reasons I
will note in the next paragraph.
whether you know it without thinking that you do. Moreover, second, isn’t it incoherent to be
convinced that p and yet wonder whether p? (2009: 201–2).
The general strategy, then, seems to be to deny that doubt is cogent at the meta-level
because of our conviction that sense perception is reliable.
Now we are in a puzzling situation, however, because if this move works at the meta-
level then it is unclear why it does not work at the object level as well. That is, why not
deny there too that doubt about whether the cup is on the table, or the bank is open, or
the keys are by the toaster is cogent because we are convinced (let’s stipulate) about all
of these things? But if this move is unsatisfying at the object level, perhaps because it
smacks too much of dogmatism, then it is unclear why it is suddenly satisfying on the
meta-level.
In short, if one thinks that something is normatively lacking in a belief unless it is
able to respond to doubt, then (as argued earlier) this norm at least in many cases
seems satisfiable at the object level, without any need to appeal to a meta-perspective.
Alternatively, if we need to turn to the meta-level to resolve doubt, then it is unclear
why one does not need to turn to the meta-meta-level to resolves further doubts as
well.11 But there seems to be no limit here, so in order to uncover the value of reflection
it seems best to turn elsewhere.
IV. Intellectual Autonomy

We have discussed a few different reasons why Sosa claims that reflective knowledge12
that p is epistemically better than stand-alone animal knowledge that p—that only
reflective knowledge avoids meta-incoherence, or allows for understanding, or
rational defensibility, or the resolution of doubt—and found all of these wanting. Let
us now return to the interesting case of the person who neglects to reflect or perhaps
even declines to reflect—in the later case, the person who says, “No thanks, I’m fine.”
Surely something seems wrong or lacking about such a response. But what is it, exactly?
In this last section I want to consider what is by my lights Sosa’s most intriguing
claim about the value of reflective knowledge, namely, that only by reflecting can we
ensure that our beliefs are properly creditable to us, or that we are believing in a way
that respects our intellectual autonomy. And previewing a bit, I want to defend two
things here. First, that while taking a reflective perspective on our beliefs is indeed
importantly connected to autonomy, Sosa misidentifies the point at which autonomy
enters the story. It is not by reflecting on the reliability of our faculties that we acquire
11
From discussions with Sosa, I now think that his preferred way to solve this problem is by claiming
because “ought implies can,” our obligations on this score will quickly peter out. That is, we might be too
cognitively limited, after a certain point, to be able to reflect further. It seems to follow for Sosa, however,
that a cognitively perfect being such as God would still be required to reflect further, and indefinitely,
which (after a few levels of iterations) does not seem like a great result.
12
In Sosa’s favored sense of reflective knowledge “supported by a structure of reasons.” That qualifica-
tion should be understood in what follows.
intellectual autonomy, or that our beliefs become properly creditable to us. Rather, it is
by taking our “given” ends as believers to be good or worthwhile that we acquire this
autonomy. The second point is that once the connection between reflection and auton-
omy is properly understood it becomes much less clear whether reflection is properly
thought of as an epistemic good or whether it is instead best thought of as a moral or
ethical good. Although this classificatory question is in some ways of secondary inter-
est, it is worth making because it helps us to better understand the significance of
reflection in our lives. In order to appreciate the difference between our proposal and
Sosa’s, however, we should first look more closely at the way Sosa connects reflection
and autonomy.
We know that for Sosa animal knowledge arises when one’s beliefs reliably track
how things stand in the world (or, more elaborately, when the accuracy of these beliefs
manifests one’s ability to track how things stand in the world). But notice that taken in
this way one’s animal knowledge might seem less than fully human, because not terri-
bly different from the way a good thermometer tracks the ambient temperature, or the
way a good supermarket door tracks the presence of nearby shoppers.13 Sosa expands
on this concern, as well as the way in which the appeal to reflective knowledge is sup-
posed to address the concern, in the following passages:
Why the pride of place for reflective knowledge? One answer is to be found in the special bear-
ing of reflective knowledge on the understanding and coherence dear to intellectuals,14 and on
the intellectual agency that we honor . . . . No matter how much we value consultation, we are
unwilling to yield our intellectual autonomy, which requires us to assess the place of consulta-
tion in the light of all our other relevant information and recognized desiderata . . . . In the end
reflection has a closer, more finally determinative influence on the beliefs we form, and the
deliverances of consultation bear properly only through reflection’s sifting and balancing.
(2004: 291, italics added)
One question remains pertinent: What could possibly give to reflective knowledge a higher
epistemic status than the corresponding unreflective cognitio(n)? . . . What favors reflective
over unreflective knowledge? Reflective acquisition of knowledge is, again, like attaining a
prized objective guided by one’s own intelligence, information, and deliberation; unreflective
acquisition of knowledge is like lucking into some benefit in the dark. The first member of each
pair is more admirable, something that might be ascribed admiringly to the protagonist, as his
doing. And we can after all shape our cognitive practices, individually and collectively, enhanc-
ing their epistemic virtue, their enabling us to grasp how matters stand. (2009: 142, italics
added)
The suggestion is therefore that while in one sense my animal knowledge is ascribable
to me because it manifests my own abilities (good eyesight, good memory, etc.), in a
13
As Sosa claims: “Nevertheless, a mere thermometer reaction to one’s environment cannot constitute
the best human knowledge, regardless of whether that reaction is causally mediated by experience”
(Sosa 2009: 153).
14
Here we see another appeal to understanding and coherence, but since I have addressed these earlier
I will not return to them again.
fuller or more accurate or perhaps more important sense Sosa seems to think it is not
my own doing. Why? Because again the beliefs that constitute this knowledge seem to
just brutely arise in me, much like the readings of a good thermometer just brutely
arise in response to the temperature. My beliefs can only be significantly or impor-
tantly ascribable to me—ascribable to me “as my doing”—when I have reflectively
endorsed them, which in this context means endorsed them as reliably oriented to
the truth.
On this approach reflective knowledge seems to acquire its special value due to
its connection to agency and autonomy, and the value of “intellectual autonomy”
in particular. It is not entirely clear how having a second-order perspective on one’s
first-order reliability is supposed to yield these goods, however. Consider a large
supermarket with a variety of doors, all of which are connected to a central system
monitoring their performance. When the system determines that a given door is relia-
ble, it allows it to continue functioning; when it determines that a door is unreliable, it
shuts it down and calls for repair. If the first-order door reliability was not autonomous
to begin with, because a simple brute response, how might it suddenly become autono-
mous with the addition of second-order monitoring—which, we can suppose, is itself
merely a brute response to stimulus, albeit now stimulus regarding how the first-order
door systems are performing?
Of course this example is in some ways misleading because with the door there are
no beliefs on the scene to begin with—beliefs which might be elevated beyond the
animal. But still, the first-order endeavorings of the door are nonetheless perfor-
mances on Sosa’s view, and the question remains as to how brute reactions on the
first-order level become performances of a significantly different kind with the addi-
tion of a second level brutely responding to stimuli from below. But if this way of con-
necting reflection and autonomy is unsatisfying, where should we look instead?
The alternative that I would like to sketch instead is I think closer to the earlier
Socratic claim that the unexamined life is not worth living, and to the Delphic claim to
“Know Thyself.” To see how, suppose for the sake of argument that truth is the natural
or default aim of belief and happiness is the natural or default aim of desire. Even
though our beliefs and desires might be reliably oriented towards these ends—even
though they might track these ends—one might nonetheless worry, along with Sosa,
that there is something less than fully human about simply believing and desiring in
this way because on this scheme our beliefs would be responding to indications of
truth, and our desires to indications of happiness, in much the same way that a super-
market door would be responding to indications of shoppers.
So what would it take to move us beyond this animal or servo-mechanical level?
Plausibly, what is needed is a second-order endorsement of our first-order ends. Even
if it turns out that we cannot choose for our beliefs to be oriented towards something
other than truth, or for our desires to be oriented towards something other than happi-
ness, we can at least on the reflective level take these ends to be worthy or deserving
ends, and we can approve of and try to beneficially shape our various ways of pursuing
these ends. As I am imagining it, this sort of reflective approval will not necessarily
improve upon the reliability of our beliefs (to say nothing of desires), and hence will
not obviously qualify as a kind of “epistemic” improvement, but they will make them
more fully our own. Or indeed, they will make them genuinely our own for the first
time. It is for this reason, as I suggested earlier, that the sort of reflective perspective
just sketched is better thought of as a moral or ethical gain, rather than an epistemic
one.15
Although I am not qualified to say that this is what Socrates had in mind by his claim
that the unexamined life is not worth living, it is at least an appealing way of making
sense of this provocative idea. On this view, the reason why the unexamined life is not
worth living is because it is unfree. It is a life that would be worth living, perhaps, if we
were simply confined to our first-order animal perspective, unable to reflect on and
endorse (or criticize) our first-order ends. But human beings are not like that, and our
dignity consists, arguably, in the fact that we do not merely have beliefs that are natu-
rally responsive to the truth, but we can look down on the reflective level and take this
natural responsiveness to be a good thing. We can endorse it or say yes to it.
Of course all of this gives rise to some puzzles: for example, if the belief at the object
level was not really my own doing, because just a kind of brute response, then why
think that the belief at the meta-level—wherein I evaluate and perhaps endorse my
first-order ends—is somehow more my own doing, or maybe even for the first time my
own doing? If this belief too is a kind of brute response to ends that I take to be worthy,
how can this be the source of my freedom or autonomy? As far as I can see, however,
this account is the most promising way to defend Sosa’s insistence on the importance
of reflection, and for the fully human way of living it makes available.16
References
Blaisdell, A., et al. 2006. “Causal Reasoning in Rats.” Science 311: 1020–2.
Kornblith, H., 2004. “Sosa on Human and Animal Knowledge.” In J. Greco (ed.) Ernest Sosa
and His Critics. Malden, MA: Blackwell.
Sosa, E., 2004. “Replies.” In J. Greco (ed.) Ernest Sosa and His Critics. Malden, MA: Blackwell.
15
A side note on the meaning of “epistemic”: what might it mean for one belief, one believing, to be
epistemically better than another? I don’t know what else this could mean except that the one belief is
better oriented to the truth than the other—in the sense that it is more reliably formed, or more likely to be
correct. But is it really true that beliefs guided by a reflective perspective—in either Sosa’s sense or the sense
suggested here—are more likely to be correct than beliefs that are not so guided? This does not seem to be
the case, at least not across the board. Consider again my belief, via my excellent vision, that my neighbor’s
dog is in the yard. Does this belief somehow become more reliable after I come back from the cognitive
science lab and learn that my vision is excellent? It does not seem so. But then we should reject the idea that
acquiring reflective knowledge necessarily brings about some epistemic benefit. Better instead to explain
the value of reflection along other lines, such as attempted here.
16
Thanks to Allan Hazlett, Ernest Sosa, Miguel Fernández, and two anonymous referees for helpful
comments on earlier versions of this essay.
Sosa, E., 2007. A Virtue Epistemology: Apt Belief and Reflective Knowledge, i. New York: Oxford
University Press.
Sosa, E., 2009. Reflective Knowledge: Apt Belief and Reflective Knowledge, ii. New York: Oxford
University Press.
Taylor, A. H. et al. 2009. “Do New Caledonian Crows Solve Physical Problems through Causal
Reasoning?” Proc. Biol. Sci. 276: 247–54.
Tiberius, V., 2010. The Reflective Life: Living Wisely with Our Limits. New York: Oxford
University Press.
12
The Epistemology of Stupidity
Pascal Engel
1. Negative Epistemology and the Nature of Stupidity

It is strange, although not completely surprising, that epistemologists—unlike sati-
rists, novelists, moralists, and essayists—have devoted so little thinking to the phe-
nomenon of stupidity. Strange, because stupidity being, prima facie, a certain lack of
knowledge or a pathological condition of knowing, it would seem that epistemology,
whose main task consists in the definition of the constitutive conditions of knowledge,
ought to discuss seriously this phenomenon. Not completely surprising, because it
seems natural to think that once one has defined knowledge and the conditions of its
exercise, one has automatically circumscribed, negatively, the domain of its absence or
of its failures. Indeed negative epistemology, as we may call the study of defective belief
formation or of the lack of knowledge, is but the mirror image of positive epistemol-
ogy. But not quite: for truth and knowledge, if they are to be defined and to be subject
to positive conditions, are one, whereas error, mistakes, and failures to know are
diverse and multifarious. As all the literature on Gettier counterexamples shows, there
are endless ways of going wrong or not satisfying the conditions of knowledge,
although these conditions are supposed to be unique. Negative epistemology seems to
cover many diverse phenomena: error, ignorance, irrationality, illusions and biases,
and so on. These have been widely studied in cognitive psychology, social psychology,
and psychopathology as well as in the history of science.1 Stupidity, however, is even
more elusive. It refers to so many diverse features that it is very hard to pin down. To be
stupid is not simply to be ignorant or to be prone to mistakes. The vocabulary of
1
To negative epistemology belongs a part of social epistemology which deals with situations where false
beliefs and absence of knowledge can be widespread in a society and in a culture. While a lot of the history
of science deals with ideologies, scientific errors, and failures of rationality, these situations have been
recently studied under the name of “agnotology”, a discipline devoted to the study of ignorance in all its
forms (Proctor and Schiebinger 2008). One can also consider as belonging to negative epistemology the
branch of formal epistemology which deals with the modeling of situations of ignorance as “logics of igno-
rance” (Hendricks 2010). One can also consider as part of negative epistemology, or at least tools for such
an epistemology, the numerous studies in cognitive social psychology which show how subjects can be
affected by all kinds of errors, cognitive illusions, or biases, and many studies in psychopathology about
delusions.
the epistemology of stupidity 197
s tupidity is definitely “thick”, referring to a set of properties which have a lot of family
resemblances, but which are so diverse that they seem to resist any attempt at a system-
atic inventory. To quote only a few expressions in various languages, people are called
stupid, dunce, dull, or dumb (stupidus, plumbeus, stupide, bête, mudo, dum), brutish,
(bestialis, brute), rude, rustic, or rough (grossus, rusticus), idiot, imbecile (idiota, imbe-
cilis), inept, stooge (ineptus), naïve, credulous, gullible, garrulous (credulous, inexper-
tus), clumsy, goofy (stolidus, incrassatus, balourd), fatuitous, conceited, vane (vacuus,
fat, vain), silly, fool, (stultus, insipiens, tonto, sot, tor). These words have as many con-
traries (“intelligent,” “smart,” “subtle,” “bright,” “clever,” “quick-witted,” etc.). The
adjectives “stupid” or “foolish,” as their contraries, apply both to certain acts or perfor-
mances (“that was a stupid thing to do”), to states (to be stupidus designates in Latin the
person who is in a state of stupor, of astonishment) and to individuals or character
traits (“you damn fool”). Another interesting feature is that “stupid” or “fool” and their
contraries are gradable adjectives: one can be more or less stupid. But how to put order
in such a mess? An important obstacle to this task is that adjectives like “stupid” and its
family most of time belong to the class of words which semanticists call “pejoratives”.2
Some people are called “stupid,” “morons,” or “nuts”. Does it follow that they are so?
This raises immediately the suspicion that the proper semantics for such words is
expressivist, and that they are mere projections of our moods, or of the social settings
in which they are used. It is true that in many cases such adjectives have an expressive
or social meaning (the French con, the Spanish gilipoyas, or the U.S. jerk are notori-
ously hard to translate). But it does not follow that what they denote is only in the eyes
of the beholder. As diverse and context-relative can be the range of properties desig-
nated by this multifarious vocabulary, we should not exaggerate their lack of unity. If
we presupposed that the notions stupidity or of foolishness have only an expressive or
social meaning the present investigation would have no point.
The most natural home for an investigation into the nature of stupidity is the
Aristotelian tradition of the description of characters, from Theophrastus to the
French and British moralists, taken up in literature by novelists like Fielding, Austen,
Eliot, and by many contemporary works.3 But here too the diversity is striking. It is not
difficult to see that Don Quixote and Sancho Panza, Dostoïevski’s Mychkin, Faulkner’s
Benjy, or Flaubert’s Bouvard and Pécuchet are not idiots or fools belonging to the same
kind. The first is a lunatic, the second a rough man, the third is an idiot in the clinical
sense, the fourth is a mentally retardate, and the last two characters are stupid although
intelligent in many ways. The most obvious candidate for unifying these properties is
to take them as features of a personal character, hence to take stupidity and its kin
properties as species of epistemic or cognitive vices. The negative epistemology of
2
On pejoratives see e.g., Hom (2010).
3
Philosophers, it has to be said, have done very little to illuminate the nature of stupidity. Poets, moral-
ists, writers, and satirists like Juvenal, Erasmus, Pope, Swift, Flaubert, Musil, and Kraus, have done a lot,
without being able to capture the very essence of the phenomenon (Musil’s über die Dummheit (Musil 1937)
is an exception).
198 Pascal Engel
stupidity is the counterpart of the positive epistemology of epistemic virtues. So it

seems natural to treat the epistemology of stupidity as a chapter of virtue epistemology.
Just as epistemic virtues, like moral virtues, are specific kinds of excellences based on
various dispositions and abilities, epistemic vices are specific kinds of bad dispositions
or of inabilities.
Virtue epistemology, however, comes into two varieties. On the one hand, accord-
ing to what is often called “responsibilist” or “character based” virtue epistemology,
virtues and vices are conceived as personal character traits, which are in general volun-
tary or under the agent’s control. On the other hand, according to what is often called
“reliabilist” virtue epistemology, virtues and vices are conceived as basic dispositions
and competences, which are not voluntary and not directly under the agent’s control.4
A number of writers (Hookway 2003, Lepock 2011, Baehr 2011) distinguish two kinds
of intellectual virtues: “low-level” knowledge-generating faculties, like perception,
memory, inference, or understanding language, and “high-level” cognitive character
traits that regulate inquiry and deliberation, such as conscientiousness, perseverance,
open-mindedness, intellectual humility, or intellectual courage. It seems natural to
consider that reliabilist virtue epistemology concerns virtues of the first kind, and that
responsibilist virtue epistemology concerns virtues of the second kind. Although these
are often presented as rivals, one might ask whether they actually compete.5 When we
consider epistemic vices such as stupidity, however, they seem to compete. For, on the
one hand, the property of being stupid seems to designate, in many cases, a certain lack
of competence or a lack of the innate dispositions and abilities which are necessary for
knowledge, and a defect for which the agent is not responsible. On the other hand,
stupidity is a defect in a certain sort of performance, an incapacity to exercise one’s
cognitive competence. In a number of cases, this incapacity is, at least partially, under
the control of the agent, not in the sense that he is willingly so, but in the sense that he
illustrates a certain kind of vanity or fatuity, for which he is accountable. In an even
wider sense, stupidity is more an instance of what is called in English foolishness, the
absence of wisdom, which is, in many ways, something for which we can be accounta-
ble. This feature raises the old problem of the relationship between intellectual and
moral virtues: to what extent is stupidity a matter of intellectual deficiency or a matter
of ethical deficiency? So where should we place the study of stupidity? At the level of
the basic dispositions or competences or at the level of regulative character traits? It is
not clear that these two forms of stupidity, a defect in competence, and a form of fool-
ishness are species of a common kind. It is, however, important to understand their
relations, if we want to understand the nature of the epistemic vice of stupidity, and
more broadly, the relationship between the two kinds of virtue epistemology.
4
Sosa (2007), Greco (2010), are usually considered as representatives of reliabilist virtue epistemology,
Montmarquet (1993), Zagzebski (1996), and Baehr (2011) as representatives of the responsibilist version.
5
Baehr (2011), Sosa (2015). Interestingly the division between the two kinds of virtues coincides in a
number of respects with the “dual mind” view of human intelligence defended, for instance, by Kahneman
(2011).
The problem of the nature of stupidity raises another familiar issue, about the value
of knowledge. Is it bad to be stupid or foolish, and why is so bad? It is bad to lack certain
intellectual capacities or to be unable to exercise them, but is it always bad? Blessed are
the poor in spirit, says the Gospel, and literature and movies often portrait dumb peo-
ple who turn out to be much wiser and morally good than those reputed to be intelli-
gent. Just as there is a problem of the relation between intellectual and moral virtue,
there is a problem of the relation between intellectual and moral vice. What are the
points of contact between the moral and epistemic vice here? So in many ways, stupid-
ity is a test case for virtue epistemology.
Although, to my knowledge, Ernest Sosa has never dealt directly with the problem
of stupidity as such, his general framework in epistemology is very relevant to the anal-
ysis of this phenomenon. According to Sosa’s (2007, 2011) metaphor, an archer can be
adroit (in having certain abilities and skills), and can make accurate shots (hit the tar-
get), but he is a good archer only when his successes are the result of his abilities and
skills. Similarly belief is a kind of performance which has to pass three dimensions of
evaluation to become knowledge: when it reaches its aim (when it is accurate, hence
true), when it is competent (adroit) and when it is accurate because of this competence
(apt). When knowledge of one’s apt performances becomes reflective we know “full
well.” Let us try to apply this scheme to stupidity. Is stupidity a failure at the level of
accuracy? No, since having a false belief does not make you stupid. Does having many
false beliefs make you stupid? It makes you ignorant, but not stupid. Ignorance can be
stupidity if the deficit comes at the level of the basic competence, when the basic low-
level dispositions are missing or are unreliable. Stupidity is at least in part a failure in
aptness. It occurs also at the level of the performance. An archer can have the abilities
to make good shots, and as a result of being clumsy, fail to hit the target. But some
archers are so bad that they very rarely hit the target, or if they do so, we are in doubt
whether it did not happen by sheer luck. We then tend to say that there is something
really wrong with them: perhaps they’d better try another sport, and go for football
instead. Stupidity may be located also as the failure of the higher-level performance.
But what if the archer goes even more seriously wrong, and, for instance attempts at
making the moves and gestures of archery when he actually is doing the moves associ-
ated to another game, say tennis? We have the intuition that it is not simply a matter of
competence, let alone of cognitive competence, but also of what is actually at stake in
the game: the person is not playing the game of archery, perhaps actually refuses to
play it. An agent who understands the rules of the game and has the relevant abilities,
but systemically fails to take the game seriously is not someone whose competence is
defective and he is not ipso facto stupid. Similarly someone who understands that the
aim of belief is knowledge and truth, has the proper competence and skills, but does
not try to reach these goals is not stupid. Or imagine someone whose whole life is
devoted only to archery, who does nothing else, at the expense of all of his other activi-
ties and duties in life. Something in them went wrong, but not with the achievement of
their aims; rather with the very meaning of the aims. Such agents are rather what we
200 Pascal Engel
normally call fools. The kind of vice that they exemplify is prima facie better described
within the framework of responsibilist virtue epistemology.
In what follows I shall first examine the intellectualist conception according to
which stupidity is a kind of cognitive defect and its difficulties. But my problem will be
here orthogonal to the one which occupies most writers on virtue epistemology, which
is the problem of whether virtues—either in the sense of reliable dispositions or in the
sense of acquired character traits—can define knowledge or be constitutive of it. I shall
assume that we can define knowledge as a form of safe belief based on certain kinds of
competences and which can be apt, accurate, and adroit. I shall reject the idea that stu-
pidity is only the absence of knowledge so defined. I shall envisage the alternative
hypothesis that stupidity—or at least a distinct species of stupidity lies elsewhere: in a
failure to appreciate our epistemic goals. This will lead us towards the view that there is
a kind of epistemic vice which consists in a failure to respect intellectual values, which
is more appropriately called folly, foolishness, or the lack of wisdom. Finally I shall sug-
gest that there is more continuity between the two kinds—stupidity and folly—than
first appears.
2. The Intellectualist Conception of Stupidity

The most common conception of stupidity is that according to which stupidity is a
certain lack of intellectual competence. This conception is well entrenched both in our
common-sense notion of stupidity as a cognitive deficit which is in many ways innate and
in the widespread view among psychologists that it consists in intellectual capacities—
of judgment, of reasoning, and of inference. Philosophers like Descartes who insist on
the idea that reason is an innate endowment, psychologists like Piaget, who hold that
the capacity in question comes through cognitive development and is in a large part
acquired, and cognitive psychologists who test intelligence through specific judgments
which display rationality or irrationality, all share the idea that intelligence is a special
kind of intellectual capacity, and stupidity a lack of it. Smart people have this capac-
ity—they are quick, have a good memory, reason well, and often know a lot. People
deemed stupid seem not to have the required competence—they are slow, forgetful,
reason badly, and in general know little. But there are different ways of understanding
this competence: is it a matter of knowledge, of reasoning, of intuition? Indeed idiocy
or cretinism are names of certain kinds of intellectual deficiencies, and often to be stu-
pid is simply to be ignorant. Stupidity may be associated to the lack of knowledge, but
the lack of knowledge does not entail stupidity—lots of people know little but are far
from being stupid—and being knowledgeable does not entail that one is intelligent or
clever, for there are very knowledgeable people who make stupid things. There are
“intelligent” idiots, like the “autistic savants” studied by Oliver Sacks (Sacks 1985). So it
would be wrong to assimilate intelligence to the possession of knowledge, and stupid-
ity to its absence. The most common cognitive defect associated to stupidity is d ifferent.
It is the lack of judgment. Let us call this the judgmental, intellectualist, or rationalist
conception of stupidity. Indeed knowledge consists in a number of dispositions, and
the capacity to judge is the capacity to manifest these dispositions, and in this sense
being intelligent and being stupid are dispositions associated to the presence of absence
of knowledge. Judgment, however, is a distinctive capacity. Kant’s famous definition of
stupidity locates it in a defect of the capacity to judge:
The right word for a lack in one’s power of judgment is “stupidity” (Dummheit) and there is no
help for it. Someone who is dull or limited in his thinking, having nothing wrong with him
except a low-grade understanding and a shortage of concepts, can be instructed—even to the
point of becoming learned. But people of that sort usually lack judgment (Mangel an
Urteilskraft) as well, so that it isn’t unusual to encounter learned men whose applications of
their science frequently show signs that lack, for which there is no cure. (Kant, Critique of Pure
Reason, KRV, I, 174).
Kant actually says here two things. The first is that stupidity is a lack of a certain kind of
skill, the skill of judging. The second is that it is the property of a type of character, the
one who lacks the skill in question. Let us for the moment leave aside this second fea-
ture, which ties Kant’s view to the traditional Aristotelian tradition of intellectual vir-
tues and vices. To say that stupidity is a defect of our judgments means that it is not
simply a matter of failing to know certain things—in Kantian terms to lack certain
concepts—but a matter of being unable to apply what one knows to a particular
circumstance—in Kantian terms, to fail to bring together the categories of the under-
standing and one’s intuitions about a particular case. Thus in La Fontaine’s fable, The
Bear and the Amateur Gardener, the bear, intending to remove a fly from the nose of his
sleeping master, throws a large stone on his face and kills him. The moral of the story is
that “A foolish friend may cause more woe/ Than could, indeed, the wisest foe.” The
stupidity lies in his being unable to adapt the proper tool for the circumstance as when
one “cracks a nut with a sledgehammer.” Flaubert’s novel Bouvard et Pécuchet is another
illustration of the judgmental conception. The two “idiots” are remarkably knowledge-
able in all sorts of things which they have learnt in books, from agriculture to astron-
omy and jurisprudence, but all their efforts at applying their knowledge fail: they lack
the power to judge correctly, that is, to apply the appropriate concepts in the relevant
situations. On this view, stupidity is a lack in a certain skill of judgment. This feature
can indeed affect a person as a matter of basic character, thus sorting out certain indi-
viduals as “idiotic” in the etymological sense. But it can be shared also by a lot of intel-
ligent people, at least at some moments. This is why it is not necessarily an enduring
trait of character.
The capacity to judge is indeed related to a certain competence, but the competence
in question is not usually understood as a form of knowledge. It is, rather, the posses-
sion of a power not only to issue judgments, but also to infer certain things from oth-
ers, hence to reason. Judging is supposed to be the main manifestation of our capacity
to be rational. So a lack of this capacity is a lack of rationality. Man being a rational
202 Pascal Engel
animal, it would thus seem that stupidity is the exception rather than the rule.
Contemporary cognitive and social psychology, however, seem to show exactly the
contrary. They are replete with studies showing how cognitive illusions and biases
affect ordinary human performance in reasoning and judgment. These mistakes are
not occasional, but systematic. They can affect deductive reasoning (Inhelder and
Piaget 1955, Wason 1968, Nisbett and Ross 1980), probabilistic and economic reason-
ing (Kahneman and Tversky 1982), and a number of our decisions and behaviors, in
particular in economics (Leavitt and Dubner 2005). The mistakes also affect a vast
variety of behaviors: people make bad estimates of risk, based only on success in the
past, they indulge in wishful thinking, they under-evaluate public opinion and techni-
cal change, or are overconfident. These persistent errors are generally considered as
failures of rationality in the human mind. So on this conception, stupidity is the same
thing as irrationality. We may not be stupid, since we have a modicum of rationality,
but we often think and do stupid things by reasoning badly, even on very simple tasks.
The intellectualist-rationalist conception of stupidity is both too coarse and too nar-
row. Too coarse because, in so far as it conflates intelligence with rationality and stu-
pidity with irrationality, it is too broad to capture what is specific to stupidity. If we
interpret it, on the one hand, as do many psychologists and philosophers (Nisbett and
Ross 1980, Stich 1985) as showing that human beings are, as a matter of empirical fact,
deeply irrational and bad in many of their ordinary judgments and reasoning, and in
this sense very stupid, the view is too broad, since it entails that everyone, more or less,
is stupid, most of the time. As many titles of publications say, “It’s not necessary to be
stupid to act stupidly” (Legrenzi 2010) or Why Are Smart People so Stupid? (Sternberg
2002). So on the intellectualist conception, everyone has his share of stupidity. On the
other hand, it has been argued that these pessimistic results can be obtained only
against the background of a general presupposition of general competence or rational-
ity (Davidson 1980, Cohen 1981): a kind of behavior or judgment can be predicated as
irrational only if one assumes, from the outset, that human beings are in general
rational, hence intelligent, and that mistakes and errors are products of performance,
not of competence. If one takes this line, human beings are in general intelligent, and
only stupid on that general basis. One may further argue that the kinds of mistakes that
people make are often artefacts of the formalisms and of the normative principles used
by the experimenters, and that when one considers intelligence as the capacity to use
heuristics and adaptable behavior—and not as competence judged according to a
principled normative theory—the so-called mistakes seem far less salient (Gigerenzer
1996). When the proper logical norms are employed, the overall irrationality disap-
pears, and people appear much smarter, for they use simple heuristics and cognitive
tinkerings, which show that human intelligence has much more adaptive resources
than those predicated on traditional models. “Ecological rationality” which uses
“quick and dirty” heuristics fares in general much better in reaching certain goals than
a rigid logical conception of reason. In the end, when intelligence is not understood as
logical rationality but as a form of adaptive behavior in evolutionist terms, humans
turn out to be much more intelligent and less stupid. This confirms the point noticed
above, that one does not need complex cognitive equipment, hence much knowledge,
to be able to make good judgments (Todd and Gigerenzer 2007). So everyone has his
own share of stupidity—as the movie character Forest Gump famously said, “stupid is
as stupid does” applies to all of us—but also everyone has his or her share of intelli-
gence. We may well be rational or very clever in one respect or another. Intelligence
and stupidity are very contextual: for instance people who are able to solve a problem
quite well in a certain kind of practical context are often unable to solve it in another
context or when it is posed in abstract terms. It is not even clear, on the rationality view
of intelligence, that we can set a standard of intelligence so that individuals of a certain
class can be definitely considered as “stupid” and others as “intelligent.”6 So the claims
of universal intelligence or stupidity are underdetermined by empirical evidence and
by the choice of our normative models of rationality. The intellectualist assimilation of
intelligence to rationality is too general to capture stable properties such as disposi-
tions. This generality is but a consequence of the psychological conception upon which
the rationality view rests: rationality is displayed in particular judgments that people
make, which are understood as events falling under laws which cognitive psychology
has to investigate. They are never understood as enduring properties of kinds of indi-
viduals or as character traits. The idea that intelligence is an enduring power of judg-
ment was indeed Kant’s. But contemporary psychologists have implicitly rejected this
conception by concentrating on particular manifestations of a general competence
and by rejecting the notion of an enduring power.7
The intellectualist-rationalist view is also too narrow. Ryle (1946) notoriously
objects to the view that intelligence consists in the contemplation of propositions and
in a form of propositional knowledge or of truths, and argues that it is more a matter of
knowing how or practical knowledge. He defends the thesis that it consists in disposi-
tions and skills. On Ryle’s view if intelligence is a kind of knowledge, it is a certain kind
of know-how. It is the style of performance which is intelligent or stupid, not the grasp
of propositional rules which govern it. The intellectualist is not necessarily committed
to the view that to be intelligent is to be able to grasp certain propositions, but it is an
important line of objection to at least some versions of it .The Rylean view is not neces-
sarily in opposition here with reliabilist virtue epistemology, since one can count a
number of reliable dispositions as forms of know-how. 8
Another common objection to the intellectualist view is that intelligence is not only,
or not mainly, the exercise of rationality. It has to do with a variety of factors, which
involve not only the faculties of reasoning and intellectual judgment, but also emotions
and feelings, sensitivity to context, the capacity to decide in various circumstances,
6
Ceci (1993), Lepock (2011), Olin and Doris (2014).
7
This rejection is clear in Gilbert Harman (2000). But perhaps the notion of a power to judge has been
abandoned too quickly.
8
Indeed this objection rests upon whether one accepts Ryle’s distinction between knowing how and
knowing that. But I won’t go into this here.
204 Pascal Engel
the disposition to learn from one’s errors rather than the rigid application of the rules
of logic. The use of reason, far from being the criterion of intelligence, is often the mark
of stupidity. This was actually the objection that the Romantics addressed to the
Enlightenment conception of reason. Schopenhauer accepted Kant’s definition of stu-
pidity (Dummheit) as a lack of understanding (Verstand), but he also said that stupidity
can affect reason (Vernunft),9 which can make mistakes of its own. For Schopenhauer
(who here follows Hegel), the substance and proper incarnation of reason is contained
in the principle of identity “A = A,” which contains all logical truths, but which is empty
and tautological. Stupidity for the Romantics is not, as it is for the Enlightenment phi-
losophers, a misuse of a faculty—reason—which is healthy by nature, but it is the mark
of an excessive respect for reason, which thus becomes a kind of illness. Nietzsche
denounced in the same way what he called the logic of identity.10 The theme of stupid-
ity as an excess of reason is most present in literature, where stupidity is represented by
characters, such as Flaubert’s Homais, Dickens’s Pickwick, or Lewis’s Babbitt, who
issue only tautologies and commonplaces, and who are paradigms of stupidity.
Flaubert wrote a Dictionary of Accepted Ideas, which is a repertoire of such truisms.
Like the Romantics, he understood stupidity as the trap in which one falls when one
has too much reverence for reason, when one tries to be too rational.
On the Romantic view, reason itself and the very search for truth are stupid. A famil-
iar Romantic theme is that we ought not only to praise feeling and sentiment, but also
folly and madness rather than cold reason. It’s not stupidity which is bad, it’s reason
which is stupid. The Nietzschean conception of values is, in many ways, a kind of over-
statement of the Romantic conception. The values of life are held to be superior to
those of knowledge and truth. Stupidity and silliness are, in this respect, to be praised
as displaying the dark side of reason.11
3. Stupidity and Understanding

One need not go to such extremes as the Nietzschean inversion of values in order to
see what is wrong with the intellectualist–rationalist view. Stupidity may be an excess
of reason, as the romantic conception has it, but it does not follow the very use of rea-
son is stupid. A less radical hypothesis is that stupidity does not involve the absence or
9
Schopenhauer (1909: 50). The main contemporary representative of this conception of stupidity is
Deleuze (1968). See the comments on these conceptions by Roger (2008) and Engel (2014).
10
Nietzsche: “Logic is bound to the condition: assume there are identical cases. In fact, to make possible
logical thinking and inferences, this condition must be treated fictitiously as fulfilled. That is: the will to
logical truth can be carried through only after a fundamental falsification of all events is assumed. From
which it follows that a drive rules here that is capable of employing both means. Firstly falsification, then
the implementation of its own point of view: logic does not spring from will to truth” (Nietzsche 1901:
512).
11
See in particular Deleuze (1968), who builds a whole theory of bêtise on the basis of his Nietzschean
critique of reason, and the comments by Roger (2008: 19–38). The fascination for stupidity as the dark side
of reason is indeed a common postmodern theme, which inspires a lot of contemporary literary essays (see
Engel 2014).
the excessive use of reason, but a misuse of reason. The rationalist view implicitly
identifies the epistemic goal as the attainment of true belief. In identifying implicitly
competence to the reliable capacity to maximize true beliefs, the rationalist equates
the possession of knowledge and the exercise of intelligence with the reliable manifes-
tation of the disposition to form true beliefs. Now, as we have already noticed, a num-
ber of people who possess knowledge in this sense can be stupid. The traditional
conception of knowledge presupposes that truth is our ultimate epistemic goal. If all
that it takes to achieve knowledge is reaching this goal in maximizing our true beliefs,
then someone could get a lot of true beliefs and thus reach knowledge without appre-
ciating the value of these beliefs. According to Catherine Elgin (1988), “currently
popular theories of knowledge have the surprising consequence that stupidity can
enhance, and intelligence diminish, one’s prospects for knowledge” and so promote
“the epistemic efficacy of stupidity.” Most contemporary theories of knowledge, she
argues, rest upon:
An unwitting bias in favour of stupidity is characteristic of both internalist and externalist
theories of knowledge. It derives from the shared convictions that (a) our epistemic goal is to
accept (or believe) a sentence if it is true and reject (or disbelieve) it if it is false, and (b) the
standard for acceptability cannot be set too high, else scepticism will prevail. The epistemic
inutility of intelligence that follows is not the sceptic’s fatalistic conclusion that since no one
knows anything, dullards are no worse off than the rest. (Elgin 1988: 297)
Stupidity, for Elgin, consists in the search for as many true beliefs as possible, without
caring for their meanings. A man who knows a lot of truths, without knowing why they
are truths, is in this sense stupid. To know that a proposition is true, and to attempt to
know as many true propositions as possible is one thing, but to know for what purpose
these propositions are relevant to our inquiries, and what role they play in our cogni-
tion, is another thing. The dull and repetitive registration of information, as it can be
accomplished by a mindless bureaucrat or by a computer, satisfies the criteria of reliabi-
list theories of knowledge. This argument suggests a diagnosis about the nature of stu-
pidity which is quite different from the intellectualist one: stupidity is not a failure in
cognitive competence, it is rather a failure to appreciate the real nature of our epistemic
goal. A number of writers have argued that the ultimate epistemic goal is not truth or
knowledge, but understanding. Stupidity is basically a failure in understanding:
Knowledge, as contemporary theories conceive it, is not and ought not be our overriding cog-
nitive objective. For to treat it as such is to devalue cognitive excellences like conceptual and
perceptual sensitivity, logical acumen, breadth and depth of understanding, and the capacity to
distinguish important from trivial truths. What is wanted is a wide-ranging study of cognitive
excellences of all sorts, and of the ways they contribute to or interfere with one another’s reali-
zation. (Elgin 1988: 310, 2006, see also Kvanvig 2003)
There are two sides to this argument. The first is that since according to reliabilism the
proper goal of cognition is to maximize true beliefs and to minimize errors, to be intel-
ligent cannot consist in our acquiring more true beliefs through our reliable capacities,
206 Pascal Engel
and to be stupid cannot consist in having fewer true beliefs and in being less reliable.
Understanding, on this view, is another sort of excellence than knowledge, and another
sort of epistemic value than truth or knowledge. The narrowness of our conception of
epistemic goals distorts our conception of the epistemic vice of stupidity: a man who
would aspire only to know truths would be a kind of idiot. Stupidity is the lack of some-
thing, which is over and above knowledge and truth, and the value of knowledge and
truth, a form of excellence, which is only in part cognitive. It involves other cognitive
capacities than those which are associated with knowledge and truth, such as the
capacity to grasp coherent sets of beliefs, to explain by relating kinds of facts, and the
capacity to grasp meanings. Stupidity is a failure to grasp the relevance of certain facts
rather than a failure to know truths. This is familiar from many jokes. A man has a hard
time peeling hot potatoes with a knife. He is advised to use a fork. He asks: “Why use a
fork to peel potatoes?” His question is stupid because he fails to understand that the
fork is not intended as a peeling instrument, but a means to avoid seizing the hot pota-
toes by hand. A man wants to enroll as a marine for his country. He meets a naval
officer, who asks him: “Can you swim?” He answers: “Why? Don’t you have any
boats?”12 An answer or a behavior is stupid because the agent fails to grasp what is rele-
vant in a given situation, or misapplies a category to a particular case. This view of stu-
pidity is actually very close to the Kantian conception of stupidity as a defect in the
capacity to judge. When one judges, according to Kant, one applies a given category to
an intuition in experience. Bad judging is a matter of failing to subsume the proper
intuition under the relevant concept. Judging is not a mechanical procedure; it is an art
of grasping what is peculiar to a given situation, and failure to exercise this art is a
manifestation of stupidity. The comic effect of stupidity comes from the misapplica-
tion of the categories. Stupidity in this sense turns out to be very close to the comic,
which is often the product of this kind of mismatch.13
The idea that understanding involves capacities other than reliable dispositions
towards truth also suggests that it may be located at the level of second-order reflective
knowledge rather than at the level of first-order knowledge. What the stories illustrate
are not cases of failure to use their basic faculties, but failure to reflect upon their own
immediate judgments. But either way, it is hard to see to what extent this conception is
distinct from the intellectualist one, since both views rest upon a certain conception of
judgment. Moreover, it is not clear that understanding is independent from knowl-
edge, and that it is not based on it (Elgin 2009). If understanding involves the grasp of
explanations and of causes, it is hard to achieve without some knowledge of these
causes and of facts and truths relevant to the explanations. So Elgin cannot be right
that stupidity is the lack of a certain capacity, over and above knowledge, such as
understanding. But she is right when she suggests that it can consist in a failure to
12
The first joke is from Adam (1975).
13
This view of the comic as a mismatch between categories (category mistake) or as a mismatch between
category and intuition, is Schopenhauer’s (who was actually inspired by Kant) is a form of incongruity or
of discordance of the faculties (Schopenhauer 1909: §13).
appreciate the proper goal and value of cognition itself. To be stupid does not involve
only an absence of reflective knowledge, but also an absence of proper evaluation of
the goal of cognition. In this respect, Elgin’s argument is close to two familiar argu-
ments in epistemology against “epistemic value monism,” to the effect that the epis-
temic goal cannot be truth, or at least not truth alone.
The first is Sosa’s objection to taking truth simpliciter as the prime epistemic goal
(Sosa 2001). If the aim of getting true belief were the only goal of our epistemic endeav-
ors, this aim could be satisfied even if the true beliefs failed to be apt. It would be satis-
fied if we collected a great quantity of trivial truths, such as how many numbers of
grain of sands are presently on this spot on the beach, or how many people whose first
name is “John” there are in the phone book for Chicago. To consider that our only
epistemic goal is truth, period, is a category mistake. We desire the truth, but only in so
far as certain truths answer the questions that are of interest to us in the context of an
inquiry. In some cases we are just uninterested in the truth of our beliefs, for instance
when we lazily read magazines while waiting at the dentist’s. In other cases, knowing
the truth is very important for us, for instance if we suspect that we have an illness or
that our bank account is empty. But we have no desire for truth in itself, nor desire of
truth for truth’s sake. In Sartre’s novel Nausea there is a character, named “The autodi-
dact” who endeavors to read all the books in the library, by alphabetical order of their
titles (at the moment when Roquentin, Sartre’s hero, meets him, he has reached the
letter L). The autodidact is a perfect example of this undifferentiated desire for truth
which amounts to stupidity. Why is it stupid to have the goal of believing every truth?
Because such a believer is indifferent to the significance of the true beliefs thus attained.
Sosa’s argument, however, does not show that truth is not the goal of inquiry. What it
shows is that undifferentiated search for truth is not the goal of inquiry and that “signif-
icant true belief ” is a more reasonable conception of the goal.14
This line of thought is close to the “swamping argument” about the value of knowl-
edge: if one assumes, as reliabilists do, that knowledge is true belief produced by relia-
ble process and that truth is the only value which is worth pursuing, then the value of a
belief is not enhanced by the fact that it was produced by a reliable process, and the
value of knowledge is not enhanced by the fact that it was so produced. Unreliably
produced true belief is no better than routinely dull true belief (Zagzebski 1996,
Kvanvig 2003).15
Reliabilists, however, have the means to answer such objections: in so far as they
agree that something has to be added to the mere accumulation of true belief in order to
14
The phrase “significant true belief ”, is Kitcher’s (1992). Sosa has, here and elsewhere, a tendency to
think in terms of epistemic goals rather than in terms of epistemic norms. For objections to this teleologi-
cal perspective, see Grimm (2009), Engel (2013). But the difference here is irrelevant for my purposes,
because the kind of perversion of the value of truth which is described as folly in the next section can be
described both in terms of values and in terms of norms.
15
The swamping argument is not only a problem for the reliabilists. But here I take it to be an objection
to the view that the goal of cognition is to maximize true beliefs, which Goldman calls “veritism.”
208 Pascal Engel
produce valuable true beliefs the swamping argument loses its grip (Goldman and
Olsson 2009). Indeed they still have to say what one needs to add to true beliefs to
make them valuable. But our problem here is not to say what the value of knowledge is,
or whether it is distinct or not from the value of true belief. Neither is it the problem of
identifying precisely the goal of cognition. We can assume here that it is truth and
knowledge. These questions are orthogonal to the one that we are examining. What we
are trying to identify is the kind of significance which is lacking when our cognition
becomes “stupid” or “silly.” It is a species of stupidity which is distinct from a lack of
basic rational competence. What the understanding view suggests is the idea that to be
stupid consists in the lack of a certain form of evaluation of the proper goal of cogni-
tion. It is a second-order kind of aptitude which is missing, bearing on the value of
knowledge itself. On this view, the stupid person is basically someone who adopts the
wrong normative attitude about the nature of inquiry. Arguments such as Elgin’s and
others show that something is missing in our theories of knowledge if we take the goal
of cognition to be the mere maximization of true beliefs. What is missing, however,
may not be a goal which would be distinct from the goal of truth and knowledge, such
as understanding. We need not reject epistemic value monism to characterize what
makes the mere collection of truths “stupid.” What may be missing is a proper appreci-
ation of the goal of truth and knowledge and of its significance.
4. Folly, Foolishness, and Epistemic Indifference

There is a form of epistemic vice which, far from being a cognitive failure, is a failure of
sensitivity to the value of knowledge and to the value of truth. People who are so insen-
sitive are often called fools.
There is no undifferentiated value of truth, in the sense that our epistemic good
would be to maximize our true beliefs, whichever they are. The problem is not that this
maximization ideal would lead us to accept any belief whatsoever, whether trivial or
not, interesting or not. The problem is that the attitude of the inquirer, if it is described
as directed towards believing truths just because they are truths, is not the attitude of a
genuine inquirer. A genuine inquirer does not aim at believing every truth whatsoever.
He aims at believing truths and only truths. He aims at believing in full consciousness of
the aim of belief or of the correctness condition for belief.16 If he does not, he runs the
risk of doing faked inquiry, and in this sense to be a fool. Indeed one can aim at acquir-
ing true beliefs without having a conception of what inquiry is: there are non-reflexive
inquirers. My claim, however, is that if one becomes aware of the aim of inquiry but
fails to respect this aim, one is instantiating a form of foolishness. I can only give some
examples of this kind of behavior.
All men, says Aristotle famously, have a natural desire to know. Curiosity is a dispo-
sition which all humans, and many animals, have. It is a disposition to aim at knowing
16
I have examined this elsewhere, alongside many others. See Engel (2013 a, 2013b).
truths, which, in many circumstances, has a strong survival value. Indeed curiosity,
when it characterizes an inquiring and attentive mind has an epistemic virtue. But the
mere collecting of truths without caring why they are truths and why they can be rele-
vant or not to our cognitive concerns, is a distortion of this natural bias. Leafing
through magazines, watching any program whatsoever on the TV, browsing leisurely
on the Internet, or simply looking at the passing show in a busy street are cases of care-
less curiosity or idle inquiry. These activities are not without purpose—most of the time
we aim at being entertained to pass the time, to get divertissement, as Pascal said—and
they are behaviors which result in our acquiring truths of various sorts. But these
behaviors are not aimed at truth, in the sense that our activities which have the effect of
making us believe truths are not accomplished for the sake of truth, and with a care for
truth as the goal of inquiry. They are not done with an inquiring mind, with the objec-
tive of acquiring something which would be cognitively significant. Idle curiosity is
the passive reception of undifferentiated truths. Many other activities involve the pro-
duction of such idle truths: gossiping, chattering, clicking a “like” on a so-called
“friend’s” page on social media. In many of these cases, we aim at getting truths, often
“interesting” truths—the gossiper undoubtedly exchanges information—but which
are neither acquired nor entertained seriously. The gossiper is “just talking”: by defini-
tion, he does not aim at saying something true, and he is not prepared to give reasons
for his claims.
Gossip or idle curiosity are what we may call benign pathologies of truth or of inquiry.
There are more severe forms. Thus gullibility consists, like idle curiosity, in welcoming
as true propositions without being prepared to ask whether they are justified. Just as we
are naturally curious, we have a tendency to believe what we are told. In so far as we can’t
help it, we cannot be blamed for that. But some people indulge in this tendency, and do
nothing to correct their instinctive acceptance of testimony. Independently of the epis-
temological debates about testimony, it is clear that being gullible is a form of epistemic
vice. Dogmatism, the disposition not to question one’s reasons is another.
None of these epistemic pathologies, however, are based on the rejection or on the
explicit disregard of the value of truth or of knowledge. They are, most of the time, non
voluntary, and better understood as forms of laziness or negligence in the epistemic
domain. There is, however, a kind of epistemic vice that is voluntary and based on an
explicit disregard for the value of truth. It is the one which Harry Frankfurt has suc-
cessfully described as the production of “bullshit” (Frankfurt 1988/2005).17 The
bullshitter, says Frankfurt, is not someone who actually believes what he says or writes.
He is “faking things,” what he says is “phony.” His assertions are not lies but pseudo-
assertions (a liar intends to make you believe what he says, whereas the bullshitter
does not). His game is a game of pretense, he is pretending to say things, but he actually
says nothing:
17
See also Black (1983), who labels “humbug” the phenomenon, which has engendered a large litera-
ture. See Cohen (2002), Schaubroeck and Maes (2006), Olsson (2008), Gjelsvik (2006), Engel (2014).
210 Pascal Engel
A bullshit statement is grounded neither in a belief that it is true nor, as a lie must be, in a belief
that it is not true. It is just this lack of connection to a concern with truth—this indifference to
how things really are—that I regard as of the essence of bullshit. . . . He does not reject the
authority of truth, as the liar does, and oppose himself to it. He pays no attention to it at all.
(Frankfurt 2005: 60–1)
Bullshitting constitutes a more insidious threat than lying does to the conduct of civilized
society. (Frankfurt 2006: 4–5)
Snobbery is another disease of the same family, which we may call mock-acceptance.
The name “snob” was invented by Thackeray for a certain kind of social behavior in
Victorian society, but the kind of character that it designates has been described much
earlier, for instance as Theophrastus’s microphilotimion (literally: the man who looks
for small honors), and as Molière’s Bourgeois gentilhomme.18 Snobbery is, like gullibil-
ity, a disease of testimony, most often about matters of taste or aesthetic subjects, but
virtually about any topic. Like the bullshitter, the snob accepts or mimicks the accept-
ance of propositions of which he does not care whether they are true or not, but which
he praises because they are uttered by people whom he wants to emulate or because
they are fashionable. Other cases of mock-believing are those of the followers of a
guru—or of other leaders of opinion—who “accept” propositions which they do not
understand, just because they have been uttered by the guru or the leader.19 C. S. Peirce
identified another type of perversion of inquiry as “sham reasoning”:
Men, then, continue to tell themselves they regulate their conduct by reason; but they learn to
look forward and see what conclusions a given method will lead to before they give their adhe-
sion to it. In short, it is no longer the reasoning which determines what the conclusion shall be,
but it is the conclusion which determines what the reasoning shall be. This is sham reasoning.
(Peirce 1937: 1. 56; see Haack 1998: 31–2)
Sham reasoning has affinities with willful belief and self-deception: one reasons in
order to reach the conclusion that one wants to reach, not because it is true, but because
it is pleasant, or comforting or desirable in some sense.
Many other forms of such epistemic indifference exist. There are indeed degrees.
Sometimes they are mere cases of negligence, and are involuntary. In other cases they
are voluntary, as kinds of epistemic counterparts of akrasia (absence of control) or
acedia (absence of interest), in the moral domain. Depending on how we describe the
case, a gossiper can indulge in his behavior without noticing that he does. It is harder to
say about the bullshitter or the snob, who most of the time are conscious of what they
are doing, even though it is not fully clear whether their behavior is voluntary or not.20
18
Thackeray (1848), Molière (1670), Theophrastus, Characters, 20. For a virtue theoretic account, see
Kerian (2010).
19
See the interesting analysis of such phenomena by Sperber (2010).
20
I owe this remark to Jon Elster. Although I cannot deal with this here, it would be interesting to com-
pare bullshitting and what is called “epistemic akrasia,” although the one has to do with assertion whereas
the other has to do with belief. The bullshitter says things which he does not care to believe to be true or
not, whereas the epistemic akratic believes things that he believes he ought not to believe. Not everyone
Such phenomena of distortions of inquiry are ubiquitous. Although we often call

these behaviors silly, the interesting fact about them is that they do not belong to the
same category of epistemic malfunctioning as stupidity. They do not affect the compe-
tence of epistemic agents, but the way they regulate their inquiry. Indeed gossipers,
snobs, and bullshitters are neither stupid nor idiotic. They may actually be very intelli-
gent and clever. It’s only that they don’t care for truth or knowledge (Frankfurt 2006,
Haack 1998). Many behaviors, such as vanity, intellectual laziness, absence of atten-
tion, self-indulgence are based on disregard for truth. A vain man is someone who
praises himself more than the truth. Someone who is intellectually lazy does not care
for finding reasons for his beliefs. Someone who is self-indulgent is content with the
reasons that he has for his beliefs, and does not try to find other or more solid ones.
Those who are affected by these vices can be highly intelligent. But they manifest a dif-
ferent sort of defect than stupidity. They are cases of an epistemic vice which has a
name stultitia in Latin, moria in Greek, folly in English, sottise in French, tonteria in
Spanish, Torheit in German. It has been widely described from Antiquity to modern
times by moralists and philosophers: foolishness or folly is a kind of absence of wis-
dom. The fool has many diverse incarnations. Seneca called stultus the man who is
busy at useless things and whose mind is in constant agitation, always after something
new (a form of misplaced curiosity, close to snobbery). Erasmus, in his famous Morias
Enkomion (In Praise of Folly, 1511) called folly (stultitia, moria) the absence of wisdom,
which is its proper name. Cervantes’s Quixote is a fool who lives in his own fancies. So
the contrary of fool is not intelligent or clever; it is wise.
There are, however, diverse forms of folly. But the one which is associated with a
perversion of inquiry, such as bullshitting, may be called epistemic folly or perhaps
epistemic indifference. Unlike stupidity, which can be temporary, folly is an enduring
trait. It is not, as we already noted, based on a lack of competence or of understanding.
Unlike stupidity, it deserves the name of epistemic vice, in the sense promoted by
responsibilist virtue epistemology, since it is most of the time voluntary. This kind of
fool is neither unintelligent nor stupid, for he can be cognitively very smart. But he
displays a lack of interest for truth and knowledge, which can be based on a certain
kind of insensitivity. He has no love for truth. He is indifferent to whether a proposition
is true or not, whether it is justified or not.21
agrees that this is even possible (Owens 2002). Zagzebski (1996: 154) compares epistemic vice with epis-
temic akrasia, and says of the latter that it is such that “the epistemically akratic person has ‘the desire to be
intellectually virtuous,’ though she fails to act on that desire.” But on the present conception, the bullshitter,
the snob, or the sham reasoner do not care for truth and have no desire to be intellectually virtuous. So they
would be rather, on Zagzebski’s view, epistemically vicious. Battaly (2014: 68) has a different description of
epistemic akrasia, which brings it closer to the kinds of attitude that I describe as epistemic indifference
here: “The epistemic akratic has a correct conception of which intellectual actions and processes are epis-
temically good or bad. He correctly believes that (say) it is bad to ignore evidence and employ wishful
thinking, and is thus motivated to avoid such actions and processes. But he can’t get himself to act in
accordance with his rational desires.”
21
I am here very indebted to Mulligan (2014) who comments on Musil, and to his many other contri-
butions on epistemic emotions and values. My diagnosis of the insensitivity, however, is slightly different.
212 Pascal Engel
Epistemic indifference or indolence is a kind of vice which has been most clearly
identified by French and British moralists and philosophers of the seventeenth cen-
tury. They often identified it with what they called in French bel esprit or in English wit.
These are the kinds of characters which La Bruyère aptly describes:
If I mention Eurypilus, you say he is a wit. . . . A workman is proud of his trade. Is Eurypilus
proud of being a wit? If he is proud of it, he is a coxcomb, who debases the natural dignity of
his intellect, and has a low and mechanical mind, which never seriously applies itself to what is
either lofty or intellectual; and if he is not proud of anything, and this I understand to be his
real character, then he is a sensible and intelligent man. (La Bruyère 1688, Of opinions, § 20)
Malebranche characterizes perfectly wit as opposed to stupidity when he says: “Both

the stupid and the wit (bel esprit) are closed to the truth. The difference is that the stu-
pid respects it,” whereas the wit despises it (Malebranche 1711).22 Locke opposes wit, as
an intellectual skill, to judgment:
And hence, perhaps, may be given some Reason of that common observation, That Men who
have a great deal of Wit and prompt Memories, have not always the clearest Judgment, or deep-
est Reason. For Wit lying most in the Assemblage of Ideas, and putting those together with
Quickness and Variety, wherein can be found any Resemblance or Congruity thereby to make
up pleasant Pictures and agreeable Visions in the Fancy; Judgment, on the contrary, lies quite
on the other Side. In separating carefully one from another, Ideas wherein can be found the
least Difference, thereby to avoid being misled by Similitude and by Affinity to take one thing
for another. This is a Way of proceeding quite contrary to Metaphor and Allusion; wherein, for
the most Part, lies that Entertainment and Pleasantry of Wit which strikes so lively on the
Fancy, and is therefore so acceptable to all People.23
Pope famously defined “true wit” as:

Nature to advantage dress’d
What oft was thought, but ne’er so well express’d;
Something whose truth convinced at sight we find,
That gives us back the image of our mind.24
and contrasted it with false wit as an incapacity to take truth seriously, which he often
calls foolishness. The fool fails to display the appropriate emotional response to truth
and knowledge. The “wit”—the bullshitter, the snob, the sham reasoner—is not stupid,
but he has no respect for truth. This is a superior form of epistemic vice. The fool is
someone who just happens to have no interest for truth and knowledge, and who does
not care for these, either willingly or not. In any case, he has no desire and no interest for
Where he emphasizes an affective deficit and insensitivity to values, I emphasize the kind of reasons which
the believer has.
22
The English translation (Cambridge University Press, 1997: 5) here says: “The dull and the sharp
mind”. But “dull” and “sharp mind” are very bad renditions of “stupide” and “bel esprit.”
23
Locke, Essay, 1710, II, 11, quoted by Addison, “True and False Wit,” The Spectator No. 62 (11 May
1711).
24
Pope, An Essay on Criticism, 1711, 297–300.
knowledge. No writer probably has given a better definition of epistemic folly than
Robert Musil in his lecture Uber die Dummheit. Musil here calls it “intelligent stupid-
ity” or “a higher pretentious form of stupidity” which “is not so much lack of intelli-
gence than a failure of intelligence for the reason that it presumes accomplishments for
which he has no rights”.25
Wits, bullshitters, and epistemic fools are everywhere.26 The utter diversity of these
characters seems to defy generalization. To what kind or genus should we assign them?
Indifference, disregard, or insensitivity to the epistemic goals can take many forms. If
we refer to the traditional Aristotelian classifications of virtues, which all have their
opposite vice, we can say that these people lack wisdom, and given that wisdom can be
either practical—knowing what the practical good is, or knowing how to live well—or
theoretical—knowing how to think well—they lack at least the latter. They do not
know, or have contempt for what is theoretically good. But this classification is too
wide, and would require a general definition of wisdom, a task in which I cannot
engage here.27 But the proper opposite of the fool may not be wise. It may well be, as
Musil (1937) suggests, the person who is modest, sober, or who exemplifies the virtue
of intellectual humility. In order to describe fully the kind of virtue and vice to which
this corresponds, we would have to deal with a whole range of other virtues, such as
open-mindedness (Adler 2004), intellectual honesty, or intellectual courage (Baehr
2011). I shall here focus on a narrower set of traits.
A responsibilist virtue-theoretic framework is clearly relevant for the epistemic vice
of folly or foolishness. What I have called epistemic indifference above is a case of what
Battaly (2010, 2014) calls epistemic self-indulgence and epistemic insensitivity which
she defines thus:
A person will be epistemically self-indulgent only if he either: (i) desires, consumes, and enjoys
appropriate and inappropriate epistemic objects; or (ii) desires, consumes, and enjoys epis-
temic objects at appropriate and inappropriate times; or (iii) desires and enjoys epistemic
objects too frequently, or to an inappropriately high degree, or consumes too much of them.
Finally, a person will be epistemically insensible only if he either: (a) chooses not to desire,
consume, or enjoy some epistemic objects that it would be appropriate to enjoy; or (b) chooses
not to desire, consume, or enjoy epistemic objects on some occasions when it would be appro-
priate to do so; or (c) chooses to desire and enjoy epistemic objects too seldom, or to an inap-
propriately low degree, or to consume too little of them. (Battaly 2010: 224)
25
Musil (1937), Mulligan (2014).
26
In a way—but I am not going to argue for this here—they are one of the main subject matters of liter-
ature. They are present today among the kinds of thinkers who prefer to have a lot of obscure but
deep-sounding ideas rather than a small set of clear ideas, and who prefer obscurantism to clarification and
provide what we may call fake scholarship. See e.g., Elster (2012). Elster distinguishes “soft obscurantism,”
by which he means a certain kind of postmodernist scholarship from “hard obscurantism,” of which he
finds a lot of samples in contemporary social sciences. Both have the characteristic of being fake
scholarship.
27
See in particular Whitcomb (2011), Baehr (2014).
214 Pascal Engel
Battaly explicitly uses the Aristotelian framework, defining first epistemic goods, such
as truth and knowledge, and the kind of activities of inquiry which are appropriate for
such goods. An epistemically temperate agent is one who has the appropriate reliable
traits or virtues adapted to these ends, and the kind of belief-forming activities associ-
ated to these. An epistemic self-indulgent or indifferent agent is one who does not have
such reliable traits, and someone who chooses not to desire the appropriate epistemic
objects. Battaly’s examples are trivia about Paris Hilton or the color of Angelina Jolie’s
dress, which, in ordinary circumstances at least, are assumed to be both unhelpful and
useless. Indeed the belief-forming activities have to be specified contextually in many
cases: collecting trivia about Paris Hilton could, in some circumstances (e.g. if one is a
police investigator on the scene of a crime involving Paris Hilton) become relevant and
useful. But we assume here that the belief formation in question is a more or less
enduring trait, not relative to specific circumstances. So the behavior of the epistemic
self-indulgent agent is largely voluntary, and does not pertain only to low-level reliable
vices, but also to high-level vices regulating inquiry. A fool in general is someone who
does not engage in the appropriate belief-forming activities. The question whether
such behavior is not only epistemic but also moral—or has moral implications—is
open. But in a number of cases, our verdict on such characters as the epistemically
self-indulgent person (the bullshitter, the snob, the intellectually lazy, the wit) is both
epistemic and ethical: we blame the wit or the snob just as we blame characters like
Oblomov, who is subject to a permanent acedia. The fool’s kind of vice is both theoreti-
cal and practical (actually the German Torheit, as Schopenhauer (1909: 50) reminds
us, designates a form of practical, rather than theoretical stupidity, unlike Dummheit.
Two objections, however, may be raised against this characterization of foolishness
as an epistemic vice. Both will help us in characterizing better the kind of epistemic
vice we are after. I have defined the fool as someone who is indifferent to the epistemic
goal of truth and knowledge or who does not have respect for these goals. There are,
however, fools who do have respect for these goals. Thus Flaubert’s Bouvard and
Pécuchet are very respectful of knowledge and truth. The two “idiots,” as Flaubert calls
them, are autodidacts who have a lot of respect for knowledge: they read encyclopedias
and books for every kind of activity that they want to engage in, and they actually learn
many things, although their knowledge is never appropriately put into practice. They
have the proper conception of the epistemic objects, their inquiry is directed at these
objects, but they fail miserably at each attempt—for instance with farming, gardening,
learning mathematics, metaphysics. They are not stupid in the sense of being inapt or
intellectually impaired. Shall we say that they are not fools? Actually, although they
have the proper conception of the epistemic goal, they are fools, in Sosa’s sense that
they are not adroit nor accurate, not putting their competence at the service of the rele-
vant activities necessary for their inquiry. They acquire a lot of truths, without under-
standing what these are truths for. This can happen when, like with Sartre’s “autodidact”
(whose character may have been modeled upon those of Flaubert’s fools) people desire
too much to reach the epistemic goal. As Aristotle insists, the excess of a virtue can be a
vice, like in the case of the great-souled or magnificent man (megalopsuchos) who is
pleased to help others and to gain their recognition, which reinforces his self-love
(Nicomachean Ethics, 4, 3) or indeed in the case of the periergon and of the polyprag-
mon who is idly curious and does not mind his own business.28 The Pharisee is also a
character who has too much desire for the good. He actually desires the good as such,
because it is the Good.29 Pharisiaism consists in aiming at virtue for the sake of being
virtuous. The reason why it is a vice is that one does not choose to be virtuous in order
to make it the case that one is virtuous. One is virtuous as the product of one’s actions.
The good and the true must not be aimed at as goals. They must supervene on correct
belief. Compare: one does not fall in love because one wants to fall in love. One falls in
love, period. One does not believe in God because one intends to make it the case that
one believes in God. One believes in God, period.30 Similarly we can identify a kind of
Pharisee in the epistemic domain, the epistemic Pharisee. A man may be convinced
that truth is the ultimate goal of inquiry, and can sincerely intend to seek truth (aim at
truth) for truth’s sake. But if that man believes what is true only because he takes truth
to be the epistemic goal, or because he wants to bring himself to believe only what is
true because it is true, or because Truth is a Good Thing, he will not pursue truth in the
proper way. What is wrong with that kind of person is that he does not believe that a
given proposition is true because it is true (by his own lights), but because he wants to
believe the truth and takes believing that proposition as an instance of his goal of
believing the truth. Such a man may well be a Victorian character in the style of W. K.
Clifford, who takes truth to be the ultimate value on which an ethics of belief is based,
or a character like George Eliot’s Casaubon in Middlemarch (Eliot 1871). Of Casaubon,
the heroin, Dorothea, says that his goal in life is Truth: “To reconstruct a past world,
doubtless with a view to the highest purposes of truth—what a work to be in any way
present at, to assist in, though only as a lamp-holder!” This kind of person, the epis-
temic Pharisee or the one whom we can call the benighted truth-lover, wants to have
true beliefs, and may well get many, but he does not have true beliefs for the right kind of
reasons. He wants to be in the state of someone who has true belief. We can here use
Parfit (2011) and others’ distinction between content (or object)-given reasons and
attitude or state-given reasons. If you intend to believe that your boss’s tie is very ele-
gant—while it is actually ugly and vulgar—because acquiring such a belief will pro-
mote your career, you may acquire this belief as the result of some self-indoctrination
or by willingly ignoring the evidence. But your belief will be acquired only because you
have somehow managed to cause in yourself the state of having this belief about your
boss’s tie. You do not believe that for the right reason. If you had, spontaneously and
directly acquired the belief because you believed its content—sincerely thinking the tie
to be very elegant—then you would have the attitude of belief towards its content, for
28
See Leigh (2013).
29
This kind of character is well described by Max Scheler (1916). Cf Mulligan (2008). I am here again
indebted to Kevin Mulligan for drawing my attention to this feature, and for his unpublished work on it.
30
See Elster (1983) about “states that are essentially byproducts,” Hieronymi (2005).
216 Pascal Engel
the right reason, which is the evidence that would then support for belief.31 On a com-
mon conception of self-deception, a self-deceived person—the stock example of the
wife who sees that her husband is unfaithful but manages to believe that he is not—is
the one who intends to have, and manages to bring it about that she has, a certain
belief. This gives us the general pattern of attitudes which I have called foolish (the list
is not exhaustive), and which share with self-deception the same structure:
– Self deception: causing oneself to believe that P because one believes that not P
and desires that P
– Bullshitting: causing oneself to assert P without caring for P’s truth
– Idle curiosity: causing oneself to believe P without caring for P’s truth
– Snobbery: causing oneself to assert or believe P because P is held-true by others
– Gullibility: causing oneself to accept that P because P is said by others
– Sham reasoning: causing oneself to infer P because one likes P
– Pharisaism: causing oneself to believe P for the love of Truth or the Good
– Epistemic indifference: causing oneself to desire inappropriate epistemic objects.
On the basis of this taxonomy, I want to suggest that the kind of epistemic attitudes
(beliefs) which are not foolish—hence wise, or epistemically concerned—are all con-
tent-given attitudes, those which one does not acquire because one wants to bring about
that one is a certain state. The fool is the person who not only does not have the correct
epistemic attitude towards possible contents, but also who does not have the proper
conception of the right attitude that one ought to have with respect to epistemic objects.
I have, following Battaly’s neo-Aristotelian view, expressed myself in terms of “proper
epistemic objects” and in terms of “epistemic goals.” But the notions of appropriateness
and correctness suggest a kind of analysis of the relevant attitudes which is actually
different from the teleological and virtue theoretic view. It is the view that epistemic
(ethical) goals are better understood as the proposer objects of fitting attitudes rather
than as values (Scanlon 1998, Engel 2013b, 2013c). The fool is the one who either ignores
or does not accept the correctness norms for belief. This is why his kind of believing is
mock-believing.
The second objection that one may raise against the conception of folly or foolish-
ness proposed here is that, contrary to my hypothesis, it is not clear that being a fool is
such a bad thing, either epistemically or morally speaking. Indeed when Erasmus
speaks “in praise of folly,” he is being ironical, and means to defend wisdom. But one
may object that the frontiers between what is epistemically good and what is epistemi-
cally bad are not so sharp, and that it is not obvious that truth and knowledge are the
proper epistemic goals or the correctness conditions of all belief and inquiry. Such a
skeptic will reject the “epistemic essentialism” about epistemic aims or norms which
the present account of folly presupposes. Thus Hazlett (2013), who is such a skeptic,
31
I here assume that the distinction is fairly clear, but there is indeed a large literature on the legitimacy
of the distinction and on the “wrong kind of reasons” problem. See Hieronymi (2005), Engel (2013c).
asks: why are we so sure that self-deceived and wishful thinkings are epistemically
bad? Can’t they too contribute to inquiry? Hazlett argues that truth is not the proper
epistemic goal, and that behaviors like wishful thinking, in so far as they can do us
some good, are not systematically vicious or incorrect. He does not raise this question
explicitly for foolishness as I have tried to characterize it, but we could raise the ques-
tion. False beliefs can be evolutionarily advantageous in some contexts (Stich 1990).
Stupidity, lunacy, foolishness could well be good both functionally in general and con-
textually in many cases. Why should we say that the bullshitter, the sham reasoner, the
snob, or the garrulous person are adverse to truth and knowledge? After all about 90
percent of what we know comes from testimony. Why should such social attitudes be
intrinsically bad? If literature, especially in the Romantic period, is fascinated by the
figure of the idiot or the fool, it is not simply because they are the incarnation of igno-
rance or of the flight from knowledge, but because they seem to be, in many cases, very
close to the genius. The skeptic (or for that matter the naturalist about the epistemic
goal) resembles the Nietzschean in his defiance against any kind of essentialism about
belief or any view about the final or intrinsic value of truth belief. If he insists upon
doubting, in all circumstances, the value of reason and wisdom, he will be a kind of
Romantic, who shuns truth and knowledge, because they are truth or knowledge. If,
on the one hand, he insists that one is more creative, thought-provoking, disruptive,
when one is an idiot or a fool than when one tries to be modest or sober in one’s inquir-
ies, he will probably be the mirror image of the epistemic Pharisee, a benighted falsity
and lover of unreason. Negative Pharisiaism is Pharisiaism enough. If, on the other
hand, he just intends to deflate the ideals of truth and knowledge, and remind us that
one need not subscribe to an essentialist conception of belief as necessarily aiming at
truth and knowledge, the skeptic can subscribe to the conception of foolishness as
indifference to the truth goal. If, however, he accepts the idea that this kind of indiffer-
ence is a kind of vice, and that modesty and intellectual humility are virtues, and if he
does not want to reject any kind of virtue epistemology characterization of these men-
tal habits, he will have to accept the kind of epistemic essentialism about the aim or
about the correct norms of belief that he pretends to oppose.32
5. Two Kinds of Stupidity?

Let us take stock. I have described two kinds of epistemic vice which can only loosely
be associated to the broadly common genus of stupidity. One is a cognitive deficit,
which is associated either to a lack of knowledge and more specifically to a lack of
32
Hazlett (2012) defines intellectual humility as a “disposition not to adopt epistemically improper
higher order epistemic attitudes, and to adopt (in the right way, in the right situations) epistemically proper
higher-order attitudes”; Hazlett (2013), however, seems to defend a form of skepticism and relativism (and
certainly a form of anti-realism) about epistemic norms and values. So his view seems to be closer to those
of the critics of virtue epistemology who would consider that epistemic (and moral) virtues do not have
reality outside variable and contextual social situations.
218 Pascal Engel
rational judgment. In so far as the notions of rationality and of intelligent judgment are
often elusive and contextual, this kind of intellectual stupidity is elusive too. It does not
mean that it does not denote any real property of individuals, for there are lots of
behaviors which are stupid in the sense of irrational, although it is not clear that they
denote a psychological natural kind. The intellectualist view has also the drawback of
characterizing stupidity along only one dimension, the cognitive one. But, as critics of
reliabilist epistemology have argued, to be stupid or not must be evaluated along
another dimension, with respect to the kind of epistemic goal that an individual pur-
sues, and with respect to his or her appreciation of that goal. There is a kind of stupidity
which does not consist in failing to reach the epistemic goal—be it truth, knowledge of
understanding—but in failing to have a proper conception of it, and if one does have
such an appreciation, in failing to respect it. What I have called, following the classical
tradition of the moralists folly or foolishness is not a cognitive deficit or a lack of under-
standing. It is rather a form of insensitivity or indifference to the intellectual values,
such that the individual refuses to engage in the proper kind of belief formation which
is characteristic of inquiry and which is governed by the norms of belief.
Although we could call both stupidity in the narrow intellectualist sense and folly
species of “stupidity” in the broad sense, these traits clearly are not the same kind of
epistemic vice. The first is most of the time involuntary, and involves unreliable dispo-
sitions at the level of basic competence, or errors in performance which flow from the
inaptness of the individual. It clearly belongs to “low-level” virtue epistemology. The
second is most of the time voluntary, and does not bear on competence or basic intelli-
gence: on the contrary, one of the distinguishing marks of epistemic indifference is that
it is an attitude which many learned and competent individuals (intelligent people)
can take. It is much more a personal character trait than a basic disposition. It is a high-
order epistemic attitude which involves not only the stance that one takes towards the
use of one’s low-level dispositions and skills but also the economy of inquiry and intel-
lectual research as a whole. And this attitude has clear ethical underpinnings. Hence its
proper study belongs to the “responsibilist” and “high-level” kind of virtue epistemol-
ogy. So shall we say that there are actually two different kinds of stupidity, which belong
to two different kinds of analysis, the former more narrowly “epistemological,” the lat-
ter more broadly “ethical” (or belonging to the ethics of belief)? Our taxonomy, correct
or not, does not solve the problem of the relationship between the two kinds of virtue
epistemology.
There is, however, much more continuity between the two kinds of epistemic vice,
stupidity in the narrow sense, and folly in the broad sense of epistemic indifference, if
we attend more carefully to the notion of judgment which was the starting point of the
intellectualist conception. In his most recent work Sosa recasts the distinction between
low-level reliabilist virtue epistemology and higher-level responsibilist virtue as the
distinction between “on one side intellectual virtues whose manifestation helps to put
you in a position to know, and on the other, intellectual virtues whose manifestation in
the correctness of a belief thereby constitutes a bit of knowledge” (Sosa 2015: chapter 2).
On his view of knowledge, a competence can constitute knowledge only if it is a

disposition to believe correctly, which is in turn a disposition to succeed when one
aims to attain certain objectives. Now, although Sosa agrees that, at a certain level of
abstraction we can distinguish the competences which are constitutive of knowl-
edge from those which in some sense regulate inquiry, hence the two kinds of virtues,
it would be wrong to trace a sharp frontier between the two. For, in the first place, it
would be wrong to suppose that we could divorce completely the competences of the
first sort, which may not be reflective, from the kinds of evaluations and accomplish-
ments in inquiry of the second sort, which are reflective. It seems clear that compe-
tence and aptness at the low level is a precondition for the exercise of virtues of the
high level. A person whom we would be prepared to call wise in the latter sense but
who would not be competent and knowledgeable in the former sense is hard to
imagine. Indeed, as I have argued, there are competent people who are fools or epis-
temically indifferent, but it is hard to imagine that these people would attain much
knowledge. In the terms of Plato’s Meno (97c), they would not be able to retain their
true beliefs firmly in mind. In the second place, it is not evident that only the “high-
level” virtues manifest agency, while the low-level ones would be involuntary and not
subject to criticism or praise.
There is a kind of belief which is merely a disposition to hold true, and which can be
defined in purely functional terms, without appealing to any intentional attitude
towards a proposition taken by an agent to be true. But there is also another kind of
belief, more properly called judgment, which is a certain sort of affirmation in the
endeavor to get it right on whether P. As Sosa points out (Sosa 2015: chapter 2), this
kind of disposition can become an ability, and it can be exercised well or badly by indi-
viduals. The important point is that it can be exercised so in the absence of any concep-
tion of the epistemic goal. A thief can “aim at truth” in his routine judgments about, for
instance, the location of what he intends to rob, without any kind of desire to search for
the truth. Nevertheless, his capacity to judge is an intentional endeavor to get it right
on whether P.33
Sosa argues that the kind of “aiming at truth” that is manifested in the capacity to
judge is independent from the further “aiming at truth” as a goal of inquiry, because
the former is compatible with any kind of goal. This is perfectly compatible with the
attitude of the fool, who can be indifferent to the final or intrinsic value of truth, while
exercising his capacity of judging well—hence being competent and apt in his judg-
ments. But in so far as his exercising judgment involves his awareness of what it means
to “get it right” for a belief, he cannot but have an at least implicit knowledge of the
correctness conditions of his belief. So at the level of the exercise of his basic compe-
tence, he must have a conception of the norms of belief and of belief formation.
Consider now lack of judgment, in which, according to Kant, stupidity is supposed
to consist. There is a form of absence of judgment which consists in the spontaneous
33
For arguments to the effect that judgment is a kind of action see e.g. O’Brien and Soteriou (2009).
220 Pascal Engel
disposition to believe, without reflection. There is indeed no agency here, and those
who judge in this way and reflect afterwards—when they do so—and say to themselves
“how stupid I was” are looking for excuses rather than for reasons. But there is also a
kind of judging where agents are confusedly conscious that they are wrong, as in
self-deception, and possibly in a number of behaviors described above as leading to
epistemic indifference. A good sign of this is the tendency of such agents to rationalize,
that is to give reasons and justifications to their own behavior, reasons which they know
to be inadequate but which nevertheless manifest their recognition of the proper norms of
belief. Perhaps explicit and reflective epistemic indifference is a superior form of
rationalization. It may be a long way from there to become epistemically vicious, in the
sense of attending to the values and the norms of proper inquiry, and nevertheless
rejecting them. But even if the way is long, it is most probably a matter of degree and of
continuity between the exercise or failure of competence and the exercise or failure of
performance, hence between stupidity and folly.34
References
Adam, M., 1975. Essai sur la bêtise, Paris: Presses Universitaires de France.
Adler, J., 2004. “Reconciling Openmindedness and Belief.” Theory and Research in Education
2(2): 127–42.
Baehr, J., 2011. The Inquiring Mind. Oxford: Oxford University Press.
Baehr, J., 2014. “Sophia.” In K. Timpe and C. Boyd (eds) Virtues and their Vices, 303–23. Oxford:
Battaly, H., 2010. “Epistemic Self-indulgence.” Metaphilosophy 41(1–2): 214–34.
Battaly, H., 2014. “Varieties of Epistemic Vice.” In J. Matheson and R. Vitz (eds) The Ethics of
Belief, 51–76. Oxford: Oxford University Press.
Black, M., 1983. “The Prevalence of Humbug.” In The Prevalence of Humbug and Other Essays,
115–46. Ithaca: Cornell University Press.
Ceci, S. J., 1993. “Contextual Trends in Intellectual Development.” Developmental Review 13:
403–35.
Cohen, G. A., 2002. “Deeper into Bullshit.” In S. Buss and L. Overton (eds) Contours of Agency:
Themes from the Philosophy of Harry Frankfurt, 321–36. Cambridge MA: MIT Press.
Cohen, L. J., 1981. “Can Human Irrationality Be Demonstrated Experimentally?” Behavioural
and Brain Sciences 4: 317–70.
Davidson, D., 1980. Essays on Actions and Events. Oxford: Oxford University Press.
Deleuze, G., 1968. Différence et Répétition. Paris: Presses Universitaires de France, tr. Difference
and Repetition, New York: Columbia University Press, 1994.
34
I have read various versions of this article in talks and conferences in Geneva, Neuchâtel, and in Lund
in (2009, 2010, and 2011), and at the “Obscurantism and Bad Incentives” conference in Bogota in
November 2013 and in Paris in 2014. I thank Charles Larmore, Louis de Saussure, Pierre Barillet, Erik
Olsson, Kevin Mulligan, Antanas Mockus, Jon Elster and Olav Gjelsvik, Igor Douven and Chris Kelp for
their invitations, their comments, and criticisms. I thank also for their help two anonymous referees for
OUP. To Miguel Ángel Fernández I owe a great debt, for his kindness and angelic patience.
Elgin, C., 1988. “The Epistemic Efficacy of Stupidity.” Synthese 74: 297–311. Reprinted in
C. Elgin, Considered Judgment. Oxford: Oxford University Press, 1999.
Elgin, C., 2006. “From Knowledge to Understanding.” In S. Hetherington (ed.) Epistemology
Futures, 199–215. Oxford University Press, Oxford.
Elgin, C., 2009. “Is Understanding Factive?” In A. Haddock, A. Millar, and D. Pritchard (eds)
Epistemic Value, 322–30. Oxford: Oxford University Press.
Eliot, G., 1871. Middlemarch. London: Penguin, 2003.
Elster, J., 1983. Sour Grapes. Cambridge: Cambridge University Press.
Elster, J., 2012. “Hard and Soft Obscurantism in the Social Sciences.” Diogenes 58(1): 159–70.
Engel, P., 2013. “Sosa on the Normativity of Belief.” Philosophical Studies 166: 617–27.
Engel, P., 2013a. “In Defense of Normativism.” In T. Chan (ed.) The Aim of Belief, 32–63.
Engel, P., 2013b. “Correctness for Belief.” Proceedings of the Aristotelian Society, Supp. Volume
87(1): 199–216.
Engel, P., 2013c. “Belief and the Right Kind of Reason.” Teorema 32(3): 19–34.
Engel, P., 2014. “L’avenir du crétinisme.” In A. Reboul (ed.) Philosophical Papers Dedicated to
Kevin Mulligan, ii: 135–50. Berlin: Springer.
Erasmus 1511. Moriae Encomium, Amsterdam. Tr 1668, In Praise of Folly, London.
Frankfurt, H., 2005 (1988). On Bullshit, new edn. Princeton, NJ: Princeton University Press,
2005. Initially in The Importance of What We Care About, Cambridge: Cambridge University
Press, 1988.
Frankfurt, H., 2006. On Truth. New York, NY: Knopf.
Gigerenzer, G., 1996. “On Narrow Norms and Vague Heuristics: A Reply to Kahneman and
Tversky.” Psychological Review 103(3): 592–6.
Gjelsvik, O., 2006. “Bullshit Illuminated.” In J. Elster, O. Gjelsvik, A. Hylland, K.-O. Moene, and
H. O. Melberg (eds), Understanding Choice, Explaining Behaviour: Essays in Honour of Ole-
Jørgen Skog, 101–11. Oslo: Oslo Academic Press.
Goldman, A., and E. Olsson, 2009. “Reliabilism and the Value of Knowledge.” In A. Haddock,
A. Millar, and D. Pritchard (eds.) Epistemic Value. Oxford: Oxford University Press.
Greco, J., 2010. Achieving Knowledge. Cambridge: Cambridge University Press.
Grimm, S., 2009. “Epistemic Normativity.” In A. Haddock, A. Millar, and D. Pritchard (eds.)
Epistemic Value. Oxford: Oxford University Press.
Haack, S., 1998. Confessions of a Passionate Moderate. Chicago: Chicago University Press.
Harman, G., 2000. “The Nonexistence of Character Traits.” Proceedings of the Aristotelian
Society 1999–2000, 100: 223–6.
Hazlett, A., 2012. “Higher-order Epistemic Attitudes and Intellectual Humility.” Episteme 9 (3):
205–23.
Hazlett, A., 2013. A Luxury of the Understanding: On the Value of True Belief. Oxford: Oxford
University Press.
Hendricks, V., 2010. “Knowledge Transmissibility and Pluralistic Ignorance: A First Stab.”
Metaphilosophy 41(3): 279–91.
Hieronymi, P., 2005. “The Wrong Kind of Reason.” Journal of Philosophy 102(9): 437–57.
Hom, C., 2010. “Pejoratives.” Philosophy Compass 5(2): 164–85.
Hookway, C., 2003. “How to Be a Virtue Epistemologist.” In M. DePaul and L. Zagzebski (eds)
University Press.
222 Pascal Engel
Inhelder, B., and J. Piaget, 1955. De la logique de l’enfant à la logique de l’adolescent, Paris: Presses
Universitaires de France. Tr. The Growth of Logical Thinking, New York: Basic Books.
Kahneman, D., 2011. Thinking, Fast and Slow. New York: Farrr, Strauss and Giroux.
Kahneman, D., and A. Tversky, 1982. Judgment under Uncertainty, Heuristics and Biases,
Kant, E., 1781. Kritik der Reinen Vernunft. Tr. and ed. P. Guyer and A. Wood as Critique of Pure
Reason. Cambridge: Cambridge University Press.
Kerian, M. L., 2010. “The Vice of Snobbery: Aesthetic Knowledge, Justification and Virtue in
Art Appreciation.” Philosophical Quarterly 60(239): 243–63.
Kitcher, P., 1992. “The Naturalists Return.” Philosophical Review 101(1): 53–114.
Kvanvig, J. L., 2003. The Value of Knowledge and the Pursuit of Understanding. Cambridge:
La Bruyère, 1688. Les caractères. Tr. H. von Laune, The Characters, London, 1885.
Leavitt, S., and S. Dubner, 2005. Freakonomics. New York: HarperCollins.
Legrenzi, P., 2010. Non occore essere stupidi per fare schiochezze. Bologna: Il Mulino.
Leigh, M., 2013. From Polypragmon to Curiosus. Oxford: Oxford University Press.
Lepock, C., 2011. “Unifying the Intellectual Virtues.” Philosophy and Phenomenological
Research LXXXIII(1): 106–28.
Locke, J., 1710. An Essay on Human Understanding, ed. Nidditch, Oxford: Clarendon Press,
1979.
Malebranche, N., 1711. Entretiens sur la métaphysique et la religion, in Œuvres, ii, ed. Rodis
Lewis, Paris: Gallimard, Pléiade. Tr. Dialogues on Metaphysics and on Religion, Cambridge:
Cambridge University Press, 1997.
Molière, 1670. Le Bourgeois gentilhomme. Tr. The Middle Class Gentleman.
Montmarquet, J., 1993. Epistemic Virtue and Doxastic Responsibility. Lanham, MD: Rowman &
Littlefield.
Mulligan, K., 2008. “Scheler: Die Anatomie des Herzens oder was man alles fühlen kann.” In H.
Landweer and U. Renz (eds) Klassische Emotionstheorien von Platon bis Wittgenstein,
589–612. Berlin: de Gruyter.
Mulligan, K., 2014. “Foolishness, Stupidity and Cognitive Values.” The Monist: The Philosophy
of Robert Musil, ed. B. Nanay, 97(1): 66–8.
Musil, R., 1937. “Uber die Dummheit,” Vortrag auf Einladung des österreichischen Werkbunds
Gehalten in Wien am 11. März und wiederholt am 17. März 1937. In A. Frisé (ed.)
Gesammelte Werke, 9 vols. Reinbek bei Hamburg: Rowohlt Verlag. Tr. and ed. B. Pike and
D. S. Luft, Precision and Soul: Essays and Addresses. Chicago and London: University of
Chicago Press, 1990.
Nietzsche, F., 1901. Die Wille zur Mach. Tr. W. Kaufmann The Will to Power, New York: Vintage
Books, 1968.
Nisbett, R., and R. Ross, 1980. Human Inference: Strategies and Shortcomings of Human
Judgment. Englewood Cliffs, NJ: Prentice-Hall.
O’ Brien, L., and M. Soteriou (eds) 2009. Mental Action. Oxford: Oxford University Press.
Olin, L., and J. M. Doris, 2014. “Vicious Minds, Virtue Epistemology, Cognition, and
Scepticism.” Philosphical Studies 168: 665–92.
Olsson, E., 2008. “Knowledge, Truth, and Bullshit: Reflections on Frankfurt.” Midwest Studies
in Philosophy XXXII: 94–109.
Owens, D., 2002. “Epistemic Akrasia.” The Monist 85(3): 381–97.
Parfit, D., 2011. On What Matters, i and ii. Oxford: Oxford University Press.
Peirce, C. S., 1937. Collected Papers, ed. A. Burks and P. Weiss. Cambridge, MA: Harvard
University Press.
Pope, A., 1711. An Essay on Criticism. London.
Proctor, R. N. and L. Schiebinger, 2008. Agnotology: The Making and Unmaking of Ignorance.
Stanford: Stanford University Press.
Roger, A., 2008. Bréviaire de la bêtise. Paris: Gallimard.
Ryle, G., 1945–6. “Knowing How and Knowing That.” Proceedings of the Aristotelian Society,
New Series, 46: 1–16.
Sacks, O., 1985. The Man Who Mistook his Wife for a Hat. New York: Summit Books.
Scanlon, T., 1998. What We Owe to Each Other. Cambridge, MA: Harvard University Press.
Schaubroeck, K., and H. Maes, 2006. “Different Kinds and Aspects of Bullshit.” In G. Reisch and
G. Hardcastle (eds) A Companion to Bullshit, 171–81. Chicago and La Salle Illinois: Open
Court.
Scheler, M., 1916. “Der Formalismus in der Ethik und die materiale Wertethik,” in Gesammelte
Werke, Band 2 herausgegeben von Maria Scheler, Bern: Francke Verlag. Tr. M. S Frings and
R. L. Funk, Formalism in Ethics and Non-Formal Ethics of Values. Evanston, IL: Northwestern
University Press, 1973.
Schopenhauer, A., 1909. The World as Will and Idea, i. Tr. R. B. Haldane and J. Kemp. London:
Routledge and Kegan Paul.
Sosa, E., 2001. “For the Love of Truth?” In A. Fairweather and L. Zagzebski (eds) Virtue
Sosa, E., 2013. “Epistemic Agency.” Journal of Philosophy CX(11): 585–605.
Sosa, E., 2015. Judgment and Agency, Oxford: Oxford University Press.
Sperber, D., 2010. “The Guru Effect.” Review of Philosophy and Psychology 1: 583–5.
Sternberg, R., (ed.) 2002. Why Are Smart People so Stupid? New Haven and London: Yale
University Press.
Stich, S., 1985. “Could Man Be a Rational Animal?” Synthese 64(1): 115–35.
Stich, S., 1990. The Fragmentation of Reason. Cambridge, MA: MIT Press.
Thakeray, W. M., 1848. The Book of Snobs. London: Punch Office.
Todd, P., and G. Gigerenzer, 2007. “Mechanism of Ecological Rationality: Heuristics and
Environments That Make Us Smart.” In R. I. M. Dunbar and L. Barrett (eds) The Oxford
Handbook of Evolutionary Psychology, 197–210. Oxford: Oxford University Press.
Wason, P., 1968. “Reasoning about a Rule.” The Quarterly Journal of Experimental Psychology
20: 273–81.
Whitcomb, D., 2011. “Wisdom.” In S. Berneker and D. Pritchard. The Routledge Companion to
Epistemology, 95–105. London: Routledge.
Zagzebski, L., 1996. Virtues of the Mind. Cambridge: Cambridge University Press.
13
Epistemic Circularity and
Virtuous Coherence
Ram Neta
In Knowing Full Well, Ernest Sosa presents a comprehensive virtue epistemology, and
argues that it can solve a number of philosophical problems.1 In the final chapter of his
book, Sosa argues that his virtue epistemology can solve the problem of epistemic cir-
cularity. But what is the problem of epistemic circularity, and what would count as a
solution to it?
What epistemologists call “the problem of epistemic circularity” is often described
as the problem of understanding how it is possible for us to rely upon our own epis-
temic faculties in establishing the reliability of those very same faculties. But, put this
way, it’s not clear just what the “problem” is supposed to be. In establishing the reliabil-
ity of any measuring instrument, it is necessary to use other measuring instruments—
but it doesn’t seem to follow from this that one can never establish the reliability of
measuring instruments. Why then, should there be a problem about relying on our
own epistemic faculties to establish the reliability of those very faculties?
To see just what is apparently problematic about epistemic circularity, it might help
to focus on some specific cases of epistemically circular inferences, and to contrast
them with inferences that are not—at least not in the same way—epistemically circu-
lar. By doing this, we will be able to isolate what I will call the “problem of single-case
circularity.” Once we’ve isolated that problem, and discovered how to solve it, and then
confirmed our solution by applying it successfully to the solution of another related
problem, we will then be in a position to spell out what Sosa (along with many other
philosophers) takes the problem of epistemic circularity to be.
I. The Problem of Single-case Circularity

Consider the following hypothetical cases:
1
Sosa (2011).
epistemic circularity and virtuous coherence 225
(Normal Thermometer)
Your child complains of body aches, so you take her temperature. You find a thermom-
eter in the cupboard, you use the thermometer to take her temperature, and it reads
101.5 degrees Fahrenheit. Now you make the following inference, which I call “Normal
Thermometer”:
(Normal Thermometer—1) The thermometer indicates that her temperature is 101.5
degrees Fahrenheit.
(Normal Thermometer—2) The thermometer reading is accurate on this occasion.
(We can suppose that this premise is justified by inference from the long prior track
record of accuracy that you have found your thermometers to possess, and so inde-
pendently of Normal Thermometer—1.)
----------------------------------------------------------------------------------------------
(Normal Thermometer—3) Therefore, her temperature is 101.5 degrees Fahrenheit.
Obviously, Normal Thermometer is a good inference. Its premises are credible in the
situation I’ve described, and the conclusion follows from the premises. Normal
Thermometer is a kind of inference by means of which you might gain justification for
believing its conclusion (Normal Thermometer—3).
Compare the situation just described with another situation:
(Indirect Thermometer)
Your child complains of body aches, so you take her to the doctor. The doctor takes her
temperature, and tells you that it is 101.5. You bring her home, put her to bed, and then
find that you have a thermometer at home. Wondering whether this thermometer is
accurate, you use it to take her temperature. Your own thermometer also reads 101.5
degrees Fahrenheit. Now you make the following inference, which I call “Indirect
Thermometer”:
(Indirect Thermometer—1) The thermometer indicates that her temperature is 101.5
degrees Fahrenheit.
(Indirect Thermometer—2) Her temperature is 101.5 degrees Fahrenheit. (We can
suppose that this premise is justified by the doctor’s temperature reading, and so inde-
pendently of Indirect Thermometer—1.)
----------------------------------------------------------------------------------------------
(Indirect Thermometer—3) Therefore, the thermometer reading is accurate on this
occasion.
Once again, Indirect Thermometer is a good inference. Its premises are credible in the
situation I’ve described, and the conclusion follows from the premises. Like Normal
Thermometer, Indirect Thermometer is a kind of inference by means of which you
might gain justification for believing its conclusion (in this case, Indirect
Thermometer—3).
226 Ram Neta
Now contrast the two situations above with a third:

(Easy Thermometer)
101.5 degrees Fahrenheit. Now you make the following inference, which I call “Easy
Thermometer”:
(Easy Thermometer—1) The thermometer indicates a temperature of 101.5 degrees
Fahrenheit.
(Easy Thermometer—2) Her temperature is 101.5 degrees Fahrenheit. (Let’s suppose
that this is justified, but not on the basis of anything that does not include Easy
Thermometer—1.)
----------------------------------------------------------------------------------------------
(Easy Thermometer—3) The thermometer gives an accurate reading on this occasion.
In contrast to Normal Thermometer and Indirect Thermometer, Easy Thermometer
does not seem to be a good inference. Of course, the conclusion follows from the
premises, since the premises and the conclusion are the same in Easy Thermometer as
they are in Indirect Thermometer. And of course, the premises and conclusion are all
credible in the situation that I’ve described, since that situation is not different from
Normal Thermometer, in which all of three of the propositions at issue have credibility.
So why, then, is Easy Thermometer not as good an inference as Normal Thermometer
or Indirect Thermometer?
This is one specific version of the problem of single-case circularity: it is the problem of
understanding why Easy Thermometer is not a good inference—is not the kind of infer-
ence by means of which we could come to be justified in believing its conclusion—even
though Normal Thermometer and Indirect Thermometer both are good inferences. If we
can rely on a particular thermometer reading to find out what someone’s temperature is,
and we can also rely on information about someone’s temperature to find out that a particu-
lar thermometer reading is accurate, then why can’t we do both at once? Why can’t we rely
on a particular thermometer reading to establish the accuracy of that very reading?
The problem, of course, doesn’t arise only for thermometers, or even measuring
instruments more generally. Consider:
(Normal Memory)
You bring your sick child to the doctor, who asks how high her temperature was last
time you took it (which was yesterday). You seem to recall taking her temperature, and
its being 101.5. Now you make the following inference, which I call “Normal Memory”:
(Normal Memory—1) I seem to recall that her temperature was 101.5 degrees
Fahrenheit.
(Normal Memory—2) My apparent recall is accurate on this occasion. (We can
suppose that this premise is justified by inference from the long prior track record of
accuracy that you have found your recall to possess, and so independently of Normal
Memory—1.)
----------------------------------------------------------------------------------------------
(Normal Memory—3) Therefore, her temperature was 101.5 degrees Fahrenheit.
Obviously, Normal Memory is a good inference. Its premises are credible in the situa-
tion I’ve described, and the conclusion follows from the premises. Normal Memory is
a kind of inference by means of which you might gain justification for believing its
conclusion (Normal Memory—3).
Contrast the situation just described with another situation:
(Indirect Memory)
You bring your sick child to the doctor, who asks how high her temperature was last time
you took it (which was yesterday). You seem to recall taking her temperature, and its being
101.5. Then your child responds to the doctor herself, and announces that her tempera-
ture was 101.5. Now you make the following inference, which I call “Indirect Memory”:
(Indirect Memory—1) I seem to recall that her temperature was 101.5 degrees
Fahrenheit.
(Indirect Memory—2) Her temperature was 101.5 degrees Fahrenheit. (We can sup-
pose that this premise is justified by my child’s testimony, and so independently of
Indirect Memory—1.)
----------------------------------------------------------------------------------------------
(Indirect Memory—3) Therefore, my apparent recall is accurate on this occasion.
Once again, Indirect Memory is a good inference. Its premises are credible in the situa-
tion I’ve described, and the conclusion follows from the premises. Like Normal
Memory, Indirect Memory is a kind of inference by means of which you might gain
justification for believing its conclusion (in this case, Indirect Memory—3).
Now contrast the two situations above with a third:
(Easy Memory)
You bring your sick child to the doctor, who asks how high her temperature was last
time you took it (which was yesterday). You seem to recall taking her temperature, and
its being 101.5. Now you make the following inference, which I call “Easy Memory”:
(Easy Memory—1) I seem to recall that her temperature was 101.5 degrees Fahrenheit.
(Easy Memory—2) Her temperature was 101.5 degrees Fahrenheit. (Let’s suppose that
this is justified, but not on the basis of anything that does not include Easy Memory—1.)
----------------------------------------------------------------------------------------------
(Easy Memory—3) My apparent recall is accurate on this occasion.
In contrast to Normal Memory and Indirect Memory, Easy Memory does not seem
to be a good inference. Of course, the conclusion follows from the premises, since
228 Ram Neta
the premises and the conclusion are the same in Easy Memory as they are in Indirect
Memory. And of course, the premises and conclusion are all credible in the situa-
tion that I’ve described, since that situation is not different from Normal Memory,
in which all of three of the propositions at issue have credibility. So why, then, is
Easy Memory not as good an inference as Normal Memory or Indirect Memory?
This question is yet another version of the problem of single-case circularity. If we
can rely on an apparent memory to find out what someone’s temperature was, and
we can also rely on information about someone’s past temperature to find out that
an apparent memory is accurate, then why can’t we do both at once? Why can’t we
rely on an apparent memory to establish the accuracy of that very same apparent
memory?
Both of the questions I’ve raised above—the question why Easy Thermometer is not
as good an inference as Normal Thermometer or Indirect Thermometer, and the ques-
tion why Easy Memory is not as good an inference as Normal Memory or Indirect
Memory—both of these questions are versions of the problem of single-case circular-
ity. But the problem of single-case circularity doesn’t arise only for empirically justified
beliefs: it can also arise for a priori justified beliefs. Consider:
(Normal Calculation)
You need to compute the sum of several positive numbers. You perform the calcula-
tion, you arrive at the answer x. Now you make the following inference:
(Normal Calculation—1) My calculations indicate that the sum of these numbers is x.

(Normal Calculation—2) My calculations on this occasion are correct. (We can sup-
pose that this is justified by the long track record of accuracy that you have found your
calculations of the relevant kind to possess.)
----------------------------------------------------------------------------------------------
(Normal Calculation—3) The sum of these numbers is x.
Obviously, Normal Calculation is a good inference. Its premises are credible in the sit-
uation I’ve described, and the conclusion follows from the premises. Normal
Calculation is a kind of inference by means of which you might gain justification for
believing its conclusion (Normal Calculation—3).
(Indirect Calculation)
tion, you arrive at the answer x. Someone else also performs the same calculation,
arrives at the same answer x, and tells you so. Now you make the following inference:
(Indirect Calculation—1) My calculations indicate that the sum of these numbers is x.

(Indirect Calculation—2) The sum of these numbers is x. (We can suppose that this is
justified by someone else’s testimony concerning their own calculations.)
----------------------------------------------------------------------------------------------
(Indirect Calculation—3) My calculations on this occasion are correct.
Again, Indirect Calculation is a good inference. Its premises are credible in the situa-
tion I’ve described, and the conclusion follows from the premises. Indirect Calculation
is a kind of inference by means of which you might gain justification for believing its
conclusion (Indirect Calculation—3).
Now compare the two inferences I’ve described above with a third:
(Easy Calculation)
tion, you arrive at the answer x. Now you make the following inference:
(Easy Calculation—1) My calculations indicate that the sum of these numbers is x.
(Easy Calculation—2) The sum of these numbers is x. (Let’s suppose that this is justi-
fied, but not on the basis of anything that does not include Easy Calculation—1.)
----------------------------------------------------------------------------------------------
(Easy Calculation—3) My calculations on this occasion are correct.
In contrast to Normal Calculation and Indirect Calculation, Easy Calculation does not
seem to be a good inference. Of course, the conclusion follows from the premises,
since the premises and the conclusion are the same in Easy Calculation as they are in
Indirect Calculation. And of course, the premises and conclusion are all credible in the
situation that I’ve described, since that situation is not different from Normal
Calculation, in which all of three of the propositions at issue have credibility. So why,
then, is Easy Calculation not as good an inference as Normal Calculation or Indirect
Calculation? This question is yet another version of the problem of single-case circu-
larity. If we can rely on a calculation to find out what the sum of some numbers is, and
we can also rely on information about the sum of the numbers to find out that the cal-
culation is accurate, then why can’t we do both at once? Why can’t we rely on a calcula-
tion to establish the accuracy of that very calculation?
I have now raised the problem of single-case circularity using three different sets of
cases. In general, this is the problem of understanding why Easy inferences are not as good
at establishing their conclusions as Normal or Indirect inferences are, despite having the
same premises and conclusion as the latter and occurring in the same epistemic situations
as the former. Notice that, in all of the Easy inferences above (Easy Thermometer, Easy
Memory, and Easy Calculation) the second premise was justified on the basis of the first
premise. The second premise did not provide any corroboration of the information given
in the first premise. Indeed, in each inference, there was no corroboration of the informa-
tion given in the first premise, and this seems to be at least one problem that afflicts all of
the Easy inferences. Could this absence of corroboration be what explains why the Easy
inferences are not as good as the Normal inferences or the Indirect inferences?
No. Consider the following inferences, in each of which the second premise is justi-
fied solely on the basis of the first premise:
(Easy Calculation*) This case is just like Easy Calculation, except for the conclusion
reached.
230 Ram Neta
(Easy Calculation*—1) My calculations indicate that the sum of these numbers is x.

(Easy Calculation*—2) The sum of these numbers is x. (Let’s suppose that this is justi-
fied, but not on the basis of anything that does not include Easy Calculation*—1.)
----------------------------------------------------------------------------------------------
(Easy Calculation*—3) The sum of these numbers is greater than x/2.
(Easy Thermometer*) This case is just like Easy Thermometer, except for the conclu-
sion reached.
(Easy Thermometer*—1) The thermometer indicates a temperature of 101.5 degrees
Fahrenheit.
(Easy Thermometer*—2) Her temperature is 101.5 degrees Fahrenheit. (Let’s suppose
Thermometer*—1.)
----------------------------------------------------------------------------------------------
(Easy Thermometer*—3) She is running a fever.
Although each of the two inferences above is such that its second premise is justified
solely on the basis of its first premise, neither Easy Calculation* nor Easy Thermometer*
is a bad inference: each of them provides a way of gaining justification for believing its
conclusions to be true. So absence of corroboration of the support that the first prem-
ise gives to the second premise cannot be what is wrong with the Easy inferences, since
that feature is shared by the perfectly good Easy* inferences. What, then, could be
wrong with the Easy inferences?
One plausible answer to this question is suggested by Crispin Wright’s discussion of
transmission failure:2 in the Easy inferences, in order to justifiably believe the second
premise on the basis of the first premise, one must justifiably believe the conclusion on
some basis that is independent of the first premise, and so the inference from the first
and second premise to the conclusion cannot be what justifies one’s belief in the truth
of the conclusion. More generally, for any premise set:
Transmission Failure (TF): Necessarily, S’s competent deduction P therefore Q fails to
transmit doxastic justification from S’s belief that P to S’s belief that Q if S justifiably
believes that P at least partly in virtue of S’s justifiably believing that Q on some basis
independent of P.
Could TF provide a good explanation of what’s wrong with the Easy inferences?
Not as it stands. To see why it won’t work, consider these cases:
(Easy Calculation^) This case is just like Easy Calculation, except for the conclusion
reached.
(Easy Calculation^—1) My calculations indicate that the sum of these numbers is x.
(Easy Calculation^—2) The sum of these numbers is x. (Let’s suppose that this is justi-
fied, but not on the basis of anything that does not include Easy Calculation^—1.)
2
See Wright (1985).
----------------------------------------------------------------------------------------------
(Easy Calculation^—3) If there is some error in my calculations, that error is compen-
sated for by some other error, to produce a correct result.
(Easy Thermometer^) This case is just like Easy Thermometer, except for the conclu-
sion reached.
(Easy Thermometer^—1) The thermometer indicates a temperature of 101.5 degrees
Fahrenheit.
(Easy Thermometer^—2) Her temperature is 101.5 degrees Fahrenheit. (Let’s suppose
Thermometer^—1.)
----------------------------------------------------------------------------------------------
(Easy Thermometer^—3) Even if my thermometer is systematically unreliable, still,
on this occasion, it somehow managed to produce an accurate reading.
Neither of the inferences above can provide one with a way of gaining justification to
believe that its conclusion is true. And yet in neither case must one justifiably believe
the conclusion in order justifiably to believe the premises: indeed, one need not even
have the conceptual resources necessary to believe the conclusion in order to justifia-
bly believe the premises. How can we explain what goes wrong in the Easy inferences
and the Easy^ inferences listed above, without falsely predicting a similar fault in the
Easy* inferences?
Here is a proposal that alters the transmission failure proposal bruited just above, so
that it applies not to doxastic but rather to propositional justification. Let’s call it “prop-
ositional transmission failure”:
Propositional Transmission Failure (PTF): Necessarily, S’s competent deduction P
therefore Q fails to transmit (doxastic) justification from P to Q if S is justified in believ-
ing that P at least partly in virtue of something E, such that E is what makes S justified
in believing Q.3
PTF differs from TF by focusing not on S’s doxastic justifications, but rather on the
source of S’s propositional justifications. By doing so, it promises to explain what goes
wrong in the Easy^ inferences above: in each of the Easy^ inferences, even if one
doesn’t believe the conclusion at all, what gives one justification to believe the conclu-
sion must be something that is independent of the premises, and in virtue of which the
first premise supports the second premise. Could PTF, then, be the correct explanation
of what goes wrong in the Easy inferences?
Again, the answer is no. To see why not, consider the following cases:
(Easy Calculation#) Again, this case is just like Easy Calculation, except for the conclu-
sion reached.
3
I made the mistake of defending PTF in Neta (2012).
232 Ram Neta
(Easy Calculation#—1) My calculations indicate that the sum of these numbers is x.

(Easy Calculation#—2) The sum of these numbers is x. (Let’s suppose that this is justi-
fied, but not on the basis of anything that does not include Easy Calculation#—1.)
----------------------------------------------------------------------------------------------
(Easy Calculation#—3) My calculations are correct on this occasion, and the sum of
these numbers is greater than x/2.
(Easy Thermometer#) This case is just like Easy Thermometer, except for the conclu-
sion reached.
(Easy Thermometer#—1) The thermometer indicates a temperature of 101.5 degrees
Fahrenheit.
(Easy Thermometer#—2) Her temperature is 101.5 degrees Fahrenheit. (Let’s suppose
Thermometer#—1.)
----------------------------------------------------------------------------------------------
(Easy Thermometer#—3) My thermometer reading is accurate on this occasion, and
she is running a fever.
Neither of the inferences above can provide one with a way of gaining justification to
believe that its conclusion is true. And yet in neither case must one have a justification
for believing the conclusion that is independent of one’s justification for believing the
premises in order for one to be justified in believing the premises. In fact, while it is
possible in each case to be justified in believing the premises, we may suppose that it is
not possible to have a justification for believing the conclusion that is entirely inde-
pendent of one’s justification for believing the premises, and that is because, in each
case, one’s justification for believing the second conjunct of the conclusion derives
from one’s justification for believing the second premise. In each case, being justified in
believing the premises requires no more than that one have independent justification
for believing the first conjunct of the conclusion. PTF will not do.
You might think that these (Easy #) counterexamples to PTF can be explained away
by appeal to a principle according to which conjunction derive their justification only
from the justifications for their individual conjuncts: if such a principle were true, then
we could explain what goes wrong in each of the inferences above in a way that is con-
sistent with PTF, by claiming that the conclusions can derive their justification only
from the justifications for their individual conjuncts. Unfortunately for PTF, no such
principle is true. Consider the following:
You walk into the seminar room, expecting to be a few minutes late for the faculty meet-
ing. But when you walk in, you see that nobody else is there. Now you reason as follows:
Nobody else is in the seminar room.
My colleagues are: Susan, Laurie, Geoff, …
----------------------------------------------------------------------------------------------
Susan is not in the seminar room and Laurie is not in the seminar room and …
This inference does transmit justification, and the conclusion is a conjunction. But the
justification for the conclusion does not—or anyhow need not—depend on the justifi-
cation for the individual conjuncts. So the counterexamples to PTF cannot be
explained away by appeal to any principle concerning how conjunctions derive their
justification. PTF will not work to explain all the data.
But a principle very similar to PTF will work.
Generalized Transmission Failure (GTF): Necessarily, S’s competent deduction P there-
fore Q fails to transmit (doxastic) justification from P to Q if there is some E such that
(i) E is part of what makes S justified in believing Q, and
(ii) S is justified in believing that P in virtue of (i), but
(iii) (i) does not obtain in virtue of S’s being justified in believing that P.
GTF is not easy to state, but it is easier to picture. Think of a structure in which one part
(Q) is supported by another part (E), and perhaps also by other things as well.
Furthermore, E’s support for Q helps to support a third part (P), which may or may not
be supported by other things as well. But while E supports Q, and E’s supporting Q
helps to support P, the support that P has does not in turn help E to support Q. The
support relation goes from E to Q to P, and does not go in reverse; so the inference from
P to Q cannot help to justify Q. That is a way of picturing the situation that, according
to GTF, results in the epistemic badness of inference from P to Q (even when the infer-
ence from P to Q is sound, and so logically and semantically unimpeachable).
GTF explains what goes wrong in all of the Easy, Easy^, and Easy# inferences above.
In each of those inferences, your justification for believing the premises depends upon
there being something (E) that is at least part of what justifies your believing the con-
clusion, where the fact that E is part of what justifies your believing the conclusion does
not depend upon your being justified in believing the premises. Let’s go through the
examples one at a time to see this.
Fahrenheit.
Thermometer—1.)
----------------------------------------------------------------------------------------------
Here, you are justified in believing (Easy Thermometer—2) on the basis of (Easy
Thermometer—1) only in virtue of your having some independent source of justifica-
tion for believing (Easy Thermometer—3). Let that independent source of justification
be E. Then E is what makes you justified in believing (Easy Thermometer—3), you are
234 Ram Neta
justified in believing (Easy Thermometer—2) partly in virtue of E’s making you justi-
fied in believing (Easy Thermometer—3), but it is not the case that E makes you justi-
fied in believing (Easy Thermometer—3) in virtue of your being justified in believing
(Easy Thermometer—2).
(Easy Memory—1) I seem to recall that her temperature was 101.5 degrees Fahrenheit.
(Easy Memory—2) Her temperature was 101.5 degrees Fahrenheit. (Let’s suppose that
this is justified, but not on the basis of anything that does not include Easy Memory—1.)
----------------------------------------------------------------------------------------------
(Easy Memory—3) My apparent recall is accurate on this occasion.
Here again, you are justified in believing (Easy Memory—2) on the basis of (Easy
Memory—1) only in virtue of your having some independent source of justification
for believing (Easy Memory—3). Let that independent source of justification be E.
Then E is what makes you justified in believing (Easy Memory—3), you are justified in
believing (Easy Memory—2) in virtue of E’s making you justified in believing (Easy
Memory—3), but it is not the case that E makes you justified in believing (Easy
Memory—3) in virtue of your being justified in believing (Easy Memory—2). Clearly,
the same analysis applies to (Easy Calculation).
(Easy Calculation^—1) My calculations indicate that the sum of these numbers is x.
(Easy Calculation^—2) The sum of these numbers is x. (Let’s suppose that this is justi-
fied, but not on the basis of anything that does not include Easy Calculation^—1.)
----------------------------------------------------------------------------------------------
(Easy Calculation^—3) If there is some error in my calculations, that error is compen-
sated for by some other error, to produce a correct result.
Here, you are justified in believing (Easy Calculation^—2) on the basis of (Easy
Calculation^—1) only in virtue of your having some independent source of justifica-
tion for believing that your calculations are accurate on this occasion. Let that inde-
pendent source of justification be E. If E makes you justified in believing that your
calculations are accurate on this occasion, then (given the closure of propositional justi-
fication under very obvious entailment) E also makes you justified in believing (Easy
Calculation^—3). Furthermore, you are justified in believing (Easy Calculation^—2)
in virtue of E’s making you justified in believing (Easy Calculation^—3), but it is not the
case that E makes you justified in believing (Easy Calculation^—3) in virtue of your
being justified in believing (Easy Calculation^—2). Clearly, the same analysis applies to
(Easy Calculation).
Again, the same analysis applies to (Easy Thermometer^).
----------------------------------------------------------------------------------------------
Here, you are justified in believing (Easy Calculation#—2) on the basis of (Easy
Calculation#—1) only in virtue of your having some independent source of justifica-
tion for believing that your calculations are accurate on this occasion. Let that inde-
pendent source of justification be E. If E makes you justified in believing that your
calculations are accurate on this occasion, then E is also part of what makes you justi-
fied in believing (Easy Calculation#—3). Furthermore, you are justified in believing
(Easy Calculation#—2) in virtue of E’s helping to make you justified in believing (Easy
Calculation#—3), but it is not the case that E helps to make you justified in believing
(Easy Calculation#—3) in virtue of your being justified in believing (Easy
Calculation#—2). Again, the same analysis applies to (Easy Thermometer#).
So GTF explains what goes wrong in all of the Easy, Easy^, and Easy# inferences
above. It also explains why nothing goes wrong in the Normal, Indirect, or Easy* infer-
ences above. In the Normal and the Indirect inferences, your justification for believing
the premises does not depend upon anything else (other than the premises) justifying
you in believing the conclusion. And the same is true of the Easy* inferences. In all
those cases, your justification for believing the premises does not depend in any way
upon what helps to justify you in believing the conclusion. GTF sorts the cases above
perfectly.
Before proceeding to accept GTF, though, I’d like to consider an apparent counter-
example to it (due to an anonymous referee). Consider the following series of two
inferences:
(1) It seems that P
(2) P
(3) P or it doesn’t seem that P.
Suppose that S infers (3) from (2) simply by disjunction introduction. But note that (3)
is equivalent to
(3*) If it seems that P, then P.
Justification can transmit from (1) to (2) only in virtue of S’s having some independ-
ent justification—call it E—for (3*). But then E is what makes you justified in
believing (3*), you are justified in believing (2) partly in virtue of E’s making justified
in believing (3*), and it is not the case that E makes you justified in believing (3*) in
virtue of your being justified in believing (2): thus, the present objection concludes,
GTF makes the false prediction that S’s inference from (2) to (3) does not transmit
justification.
My reply to this objection is that, as always, whether the inference from (2) to (3)
transmits justification or not depends upon the source of the justification for (2). The
inference may very well transmit justification, but only if the source of one’s justification
236 Ram Neta
for (2) does not include one’s justification for (3*). Otherwise, the inference does not
transmit justification.
This fact is easy to miss when we consider cases of this kind schematically, as the
objection does, but it becomes much clearer when we consider particular cases.
Suppose, for instance, that you see what appears to be your colleague walking towards
you, about 50 feet away. Then you reason as follows:
(a) It looks to me as if my colleague is walking towards me.
(b) My colleague is walking towards me.
(c) If it looks to me as if my colleague is walking towards me, then she is.
Now, if you believe (b) solely on the basis of (a), then could you become more justified
than you already are in believing (c) by inferring it from (b)? It seems clear that infer-
ring (c) from (b) could not, under these circumstances, make you more justified in
believing (c) than you already were. You cannot become more justified in believing (c)
by inferring (c) from something whose sole justification depended on (c). And this is
precisely what GTF predicts.
In sum, the problem of single-case circularity is the problem of understanding
what’s wrong with the Easy inferences listed above: GTF provides a solution to the
problem of single-case circularity. In the next section, I will introduce a new version of
the problem of epistemic circularity—distinct from the problem of single-case circu-
larity—and show that GTF solves that new problem as well.
II. The Problem of Bootstrapping Circularity

The problem of single-case circularity is a problem concerning valid deductive infer-
ences. In section I, I argued that GTF solves the problem. But there is a different prob-
lem of epistemic circularity that I would like to introduce in this section. The point of
introducing this different problem is not merely to broaden our sense of the various
forms that epistemic circularity can take, but also to provide further support for GTF.
For, as I will argue in this section, GTF solves this new problem.
I’ll begin by reminding us of the problem of single-case circularity, using the follow-
ing simple case:
(Easy Testimony)
You are wandering around St. Andrews, looking for Market Street. You ask an unre-
markable stranger how to get to Market Street, and she tells you that it is two blocks
straight ahead and one block left. Now you make the following inference, which I call
“Easy Testimony”:
(Easy Testimony—1) The stranger said that Market Street is two blocks straight ahead
and one block left.
(Easy Testimony—2) Market Street is two blocks straight ahead and one block left.
(Let’s suppose that this is justified, but not on the basis of anything that does not include
Easy Testimony—1.)
----------------------------------------------------------------------------------------------
(Easy Testimony—3) Therefore, the stranger’s testimony is accurate on this occasion.
Easy Testimony is not a good inference: it is not the kind of inference by means of
which you might gain justification for believing its conclusion (Easy Testimony—3).
And this is just as GTF would predict, for your justification for believing (Easy
Testimony—2) depends upon your having some independent source of justification
for believing (Easy Testimony—3), but not vice-versa.
Now compare Easy Testimony with another case:
(Bootstrapping Testimony)
You are wandering around St. Andrews, looking for Market Street. You ask an unre-
markable stranger how to get to Market Street, and she tells you that it is two blocks
straight ahead and one block left. It turns out that she is walking there too, and as you
walk with her, you talk some more. She tells you that her name is “Denise,” that she has
been traveling around Scotland for over a month now, that she lives in St. Louis,
Missouri, though she was born in Johannesburg, that she’s just had her thirtieth birth-
day last week, and that her favorite gift was a bottle of Scotch.
(Bootstrapping Testimony—1) The stranger said that Market Street is two blocks
straight ahead and one block left, and Market Street is two blocks straight ahead and
one block left (where the second conjunct is justified, but not on the basis of anything
that does not include the first conjunct).
(Bootstrapping Testimony—2) The stranger said that her name is “Denise”, and her
name is “Denise” (where the second conjunct is justified, but not on the basis of any-
thing that does not include the first conjunct).
(Bootstrapping Testimony—3) The stranger said that she has been traveling around
Scotland for over a month now, and she has been traveling around Scotland for over a
month now (where the second conjunct is justified, but not on the basis of anything
that does not include the first conjunct).
(Bootstrapping Testimony—4) The stranger said that she lives in St. Louis Missouri,
and she does live in St. Louis, Missouri (where the second conjunct is justified, but not
on the basis of anything that does not include the first conjunct).
(Bootstrapping Testimony—5) The stranger tells you that she was born in
Johannesburg, and she was born in Johannesburg (where the second conjunct is justi-
fied, but not on the basis of anything that does not include the first conjunct).
(Bootstrapping Testimony—6) The stranger said that her thirtieth birthday was last
week, and her thirtieth birthday was last week (where the second conjunct is justified,
but not on the basis of anything that does not include the first conjunct).
238 Ram Neta
(Bootstrapping Testimony—7) The stranger said that her favorite gift was a bottle of
Scotch, and her favorite gift was a bottle of Scotch (where the second conjunct is justi-
fied, but not on the basis of anything that does not include the first conjunct).
----------------------------------------------------------------------------------------------
(Bootstrapping Testimony—8) Therefore, the stranger is a reliable testifier.
Even if Bootstrapping Testimony has the form of a good inductive inference—even if its
premises provide a sufficient number of sufficiently representative cases to w arrant a
conclusion of reliability—it is still a bad inference. That is to say, it is not the kind of
inference by means of which someone could come to be justified in believing the conclu-
sion. This is not because the conclusion is not one that we can become justified in believ-
ing, nor is it because the conclusion is not one that we can become justified in believing
on the basis of an inductive inference from cases. It is just that the conclusion is not one
that we can become justified in believing on the basis of this inductive inference from
cases. Why is this?
Notice that GTF can explain why this is. In the inference above, your justification for
believing each premise depends upon your justification for believing its second conjunct.
And while your justification for believing its second conjunct depended upon its first
conjunct, it also depended upon your having some independent justification to regard
the testifier as reliable. In other words, your justification for believing the premises
depended upon your having some independent justification for believing the conclusion
(e.g., that the stranger is widely regarded by others as reliable). But your having that inde-
pendent justification for believing the conclusion did not also depend upon your having
justification for believing the premises: even if your interlocutor’s assertions had been
different, you would still have had that independent justification for believing her testi-
mony (whatever it was, so long as it was not implausible given what else you knew about
her). GTF, in short, explains what makes Bootstrapping Testimony a bad inference.
But now consider another inductive case:
(Super-Bootstrapping Newspaper)
You trust the New York Times to be a reliable source of news. You have good reason to
believe, and you do believe, that most of what is reported in the Times is true. But you
wonder: just how reliable is the Times? In particular, might it be perfectly reliable? You
don’t know, but you decide to find out, and so you go through the Times reading each
statement until you’ve read all N statements. Then you reason as follows:
(Super-Bootstrapping Newspaper—1) The Times reported that p, and p (where the
second conjunct is justified solely on the basis of the first, along with your justified
belief that the Times is reliable).
(Super-Bootstrapping Newspaper—2) The Times reported that q, and q (where the
second conjunct is justified solely on the basis of the first, along with your justified
belief that the Times is reliable).
….
….
----------------------------------------------------------------------------------------------
(Super-Bootstrapping Newspaper—N + 1) The Times is perfectly reliable.

Again, even if Super-Bootstrapping Newspaper has the form of a good inductive infer-
ence—even if its premises provide a sufficient number of sufficiently representative cases
to warrant a conclusion of super-reliability—it is still a bad inference. That is to say, it is not
the kind of inference by means of which someone could come to be justified in believing
the conclusion. This is not because the conclusion is not one that we can become justified
in believing, nor is it because the conclusion is not one that we can become justified in
believing on the basis of an inductive inference from cases. It is just that the conclusion is
not one that we can become justified in believing on the basis of this inductive inference
from cases. Why is this? It is not obvious how to use GTF to explain the problem with
Super-Bootstrapping Newspaper, since the conclusion of that inference is stronger than
the background needed to justify any particular premise.
Nonetheless, the conclusion of that inference is not stronger than the background
needed to justify the conjunction of premises. It’s one thing to believe the Times with
respect to a particular claim it reports. It’s another thing to believe the Times with
respect to the conjunction of claims it reports. The latter does not require simply jus-
tified trust in the Times: I have justified trust in lots of sources, but still am not justi-
fied in believing the conjunction of everything they report. (To take an obvious
example: I am justified in trusting my own cognitive powers, but should I therefore
believe that the conjunction of all the propositions I believe is true?) In order to be
justified in believing the conjunction of everything that the Times reports, I would
already have to have some independent justification for believing that the Times is not
simply reliable, but perfectly reliable. In short, GTF explains this case as well as the
earlier bootstrapping case.
I conclude my argument for GTF. In the next section, we’ll see that, while GTF solves
the problems of single-case circularity and bootstrapping circularity, it raises a further
question, and it is this further question that constitutes the more general problem of
epistemic circularity with which Sosa is concerned.
III. GTF and the Regress of Justifiers

According to GTF, the Easy inferences above are bad because to some extent the justi-
fication of their premises depends upon the justification of their conclusion, not vice-
versa. This means that, if the premises of those inferences are justified, then that is
because the conclusions are independently justified. But if we ask what it is that pro-
vides this independent justification for the conclusions, we may seem to be launched
on a potentially vicious regress.
To see the regress that I have in mind here, let’s focus on a particular example. Recall,
for instance:
(Easy Thermometer)
240 Ram Neta
101.5 degrees Fahrenheit. Now you make the following inference, which I call “Easy
Thermometer”:
Fahrenheit.
Thermometer—1.)
----------------------------------------------------------------------------------------------
According to GTF, Easy Thermometer is not a good inference because the premises are
justified only by virtue of the conclusion’s being independently justified. Now, there are
lots of different things that could provide justification for the conclusion (Easy
Thermometer—3) independently of the specific premises listed above. But let’s refer to
whatever factor it is that provides this independent justification for (Easy
Thermometer—3) as I. Now, if I justifies you in believing (Easy Thermometer—3),
then are you justified in believing that I indicates the truth of (Easy Thermometer—3)?
Either you are or you aren’t. If you aren’t so justified, then let’s say that I is an “external”
justifier: a justifer your having of which does not require you to be aware that it indi-
cates the truth of the proposition it justifies. Suppose, for the moment, that I is not an
external justifier, and that you are justified in believing that I indicates the truth of
(Easy Thermometer—3). In that case, you are justified in believing both (Easy
Thermometer—3), and that I indicates the truth of (Easy Thermometer—3). If you are
justified in believing both of these propositions, then can you make the following
inference:
(Easy I)
(Easy I—1) (Easy Thermometer—3).
(Easy I—2) I indicates that (Easy Thermometer—3) is true.
----------------------------------------------------------------------------------------------
(Easy I—3) I gives an accurate indication on this occasion.
Whatever the value of I, the inference above is a version of the Easy inferences
diagnosed above. Although it is valid, it cannot transmit justification from prem-
ises to conclusion any better than they can. GTF tells us that this must be because
the conclusion (Easy I—3) requires justification independent of the premises.
Let’s call the thing that provides this independent justification I’. Now, if I’ justifies
you in believing (Easy I—3), then are you justified in believing that I’ indicates the
truth of (Easy I—3)? Either you are or you aren’t. If you aren’t so justified, then I’ is
an external justifier. Suppose that I’ is not an external justifier, and that you are
justified in believing that I’ indicates the truth of (Easy I—3). In that case, you are
justified in believing both (Easy I—3), and that I’ indicates the truth of (Easy I—3).
If you are justified in believing both of these propositions, then can you make the
following inference:
(Easy I’)
(Easy I’—1) (Easy I—3)
(Easy I’—2) I’ indicates that (Easy I—3) is true.
----------------------------------------------------------------------------------------------
(Easy I’—3) I’ gives an accurate indication on this occasion.
Whatever the value of I’, the inference above is, again, an Easy inference: although it is
valid, it cannot transmit justification from premises to conclusion. GTF tells us that
this must be because the conclusion (Easy I’—3) requires justification independent of
the premises. Let’s call the thing that provides this independent justification I’’. Now, if
I’’ justifies you in believing (Easy I’—3), then are you justified in believing that I’’ indi-
cates the truth of (Easy I’—3)? Either you are or you aren’t. If you aren’t so justified,
then I’’ is an external justifier. Suppose that I’’ is not an external justifier, and that you
are justified in believing that I’’ indicates the truth of (Easy I’—3). In that case, you are
justified in believing both (Easy I’—3), and that I’’ indicates the truth of (Easy I’—3). If
you are justified in believing both of these propositions . . . regress looms.
In response to this regress of justifiers, we have four options: (i) We could claim that
the regress ends with an external justifier, that is, a justifier your having which does not
guarantee that you are justified in believing that it indicates the truth of the proposi-
tion it justifies. (ii) We could claim that the regress ends with a justifier that vouchsafes
its own accuracy, that is, a justifier that not only justifies you in believing some first-or-
der proposition p, but simultaneously justifies you in believing that that very justifier
accurately indicates the truth of p on this occasion. (iii) We could claim that the regress
goes in a circle. Or (iv) we could claim that the regress goes on forever, without repeat-
ing. Since justification can be overdetermined, these options are not mutually exclu-
sive: we could have one justificatory regress that terminates in an external justifier, and
another justificatory regress for the very same proposition that goes on ad infinitum.
The general problem of epistemic circularity with which Sosa and other epistemolo-
gists are concerned is the problem of how to respond to this regress, a regress that is
generated by presupposing GTF.
In the next section, I will describe Sosa’s response to this regress, which is a combi-
nation of (i) and (iii). I will argue that this response runs afoul of GTF, the very presup-
position of the regress problem. In the final section, I will argue that we can capture the
insights of Sosa’s account, and also respect GTF, without having to accept Sosa’s own
version of (iii).
IV. Sosa’s Attempted Solution to the Problem of

Epistemic Circularity
In the final chapter of his Knowing Full Well, Ernest Sosa addresses this general prob-
lem of epistemic circularity.4 In what follows, I critically assess his solution to it, and
4
Sosa (2011: chapter 8).
242 Ram Neta
then describe an emendation to his solution that would avoid the problems encoun-
tered by Sosa’s own solution, but would still be consistent with Sosa’s overall epistemo-
logical picture. To sketch Sosa’s own picture, I begin by considering:
(Easy Vision)
You are looking at a red table in front of you in normal lighting. It looks red to you.
Now you reason as follows:
(Easy Vision—1) It looks me to as if there is a red table before me.
(Easy Vision—2) There is a red table before me (justified, but not on the basis of any-
thing that does not include Easy Vision—1).
----------------------------------------------------------------------------------------------
(Easy Vision—3) My current visual experience is veridical (at least with respect to
there being a red table before me).
Once again, the inference Easy Vision, though sound, is not a good inference: while I
can easily become justified in believing the conclusion of the inference, I cannot
become justified in believing it on the basis of this inference. GTF explains why: it is
because my justifiedness in believing the premises depends on some independent jus-
tification I have for believing the conclusion, but not vice-versa. But where does this
independent justification come from?
According to Sosa, perceptual experience normally justifies us in believing such
things as the second premise of the inference above only in light of our justified pre-
supposition (perhaps only implicit, and perhaps never subject to reflective scrutiny)
that perceptual experience is typically veridical. But then, Sosa claims, this general
presupposition itself can be justified at two levels, and in a different way at each level.
At the animal level, the presupposition is justified by virtue of the “reliability [of the
faculty] that enables the harvest of needful information”.5 Such “animal justification” is
external, in the sense that the factor that justifies you in the presupposition is not one
that you need be in any way aware of, and so not one that you need be justified in
believing anything about. A fortiori, it is not one that you need be justified in believing
indicates the truth of the proposition that it justifies.
Sosa regards such animal justification as causally necessary in, but not sufficient for,
the development of a higher level of justification that Sosa calls “reflective.” At the
reflective level, the presupposition of the typical veridicality of perceptual experience
is justified in a way that involves circularity, but a kind of circularity that Sosa regards
as virtuous, not vicious. I will quote Sosa’s description of the kind of circularity
involved in reflective justification of the presupposition, because it is on the basis of
this description that I take Sosa’s overall solution to the problem of circularity to
include a version of (iii) above. Here is what Sosa writes:
5
Sosa (2011: 149).
Epistemic justification works more like a web than like a pipe that transmits the juice of justifi-
cation or warrant. Justified beliefs are nodes of a web properly attached to the environing world
through perception and memory. Take an intricate spider’s web with its many nodes, attached
at various points to various surfaces. The position of each node might then depend causally (to
some extent, perhaps to a small extent) on the positions of the other nodes. Here there is dis-
tributive dependence on each and also collective dependence on all.
That explains a web model for belief (though beliefs also occupy an important dynamical,
historical dimension, one that requires a more complex web model). Any given belief node is
in place through its connections with the other nodes, including that original given node.
Through the basing of beliefs on other beliefs and on experiences, a rational web is woven, each
member of which is upheld in part (perhaps in miniscule part) by others, directly or indirectly.
There is no apparent reason why such basing should be regarded as either causally or norma-
tively asymmetrical, no reason why many beliefs could not constitute webs in which each node
is based partly on the others. Each might thus gain its epistemic status through such relations
to the others, where the whole web is also attached to the world through causal mechanisms of
perception and memory.
Reflective endorsement may now take its place in the web with no apparent special problems.
Through our growing knowledge of ourselves and of the world around us and of the relation
between the two, we come to see our modes of rational basing and other belief acquisition as
sufficiently reliable. This enables us to endorse such modes reflectively as truth-reliable, of a
sort to lend epistemic justification to our commitments and beliefs. True, when we modify an
epistemic commitment, whether implicit or explicit, we do so based on beliefs acquired
through commitments already in place, prominently those involved in perceptual uptake.
There is hence an inevitable circle in how we come to modify and hold perceptual commit-
ments, whether implicit or explicit. We hold them, and sustain them over time, based on
continuing observations, which are themselves based on the now installed, and perhaps mod-
ified, commitments. No special vise pertains to the nodes of our web constituted by these
commitments.6
Sosa’s story seems to be this: believers begin their cognitive lives (and, in most cases,
complete them as well) by having their regress of justifiers terminate with an external
justifier—in particular, some facts about the kind of information needed by that
believer. But some believers can competently reflect on the relation between their own
beliefs and the things represented by those beliefs, and by engaging in such reflection,
they can manage to have their regress of justifiers not terminate at all, but rather go on
indefinitely in a circle. There are two versions of this story: on one version, suggested
by some passages in Sosa’s writings, reflective creatures do not lose their animal justifi-
cations, but retain them while also acquiring a new kind of justification as well—reflec-
tive justification. On the other version, suggested by other passages in Sosa’s writings,
animal justifications are not supplemented with, but are rather replaced by, reflective
justifications. I am not sure which of these views is Sosa’s considered view, and I’m also
not sure what considerations might decide between them.
6
Sosa (2011: 150–1).
244 Ram Neta
My objection to Sosa’s story can be raised no matter which of these two versions of
the story we accept. For my objection concerns how to reconcile the coherentist ele-
ment in Sosa’s story with GTF. Recall GTF:
Necessarily, S’s competent deduction P therefore Q fails to transmit (doxastic) justi-
fication from P to Q if there is some E such that:
(ii) S is justified in believing that P in virtue of (i), but
(iii) (i) does not obtain in virtue of S’s being justified in believing that P.
If this principle is to explain the failure of the Easy inferences, then those inferences
must be such that there is something that independently justifies us in believing their
conclusion, and our justification for believing the premises is in virtue of that, but not
vice-versa. But, if justification is a web in the nodes of which “there is distributive
dependence on each and also collective dependence on all”, then how can clause (iii) of
GTF above ever be satisfied? How, on Sosa’s story, can it ever fail to be the case that
what justifies S in believing the conclusion of an inference does not to some extent do
so in virtue of S’s justifiedness in believing the premises? How can asymmetric rela-
tions of epistemic dependence ever obtain, if (either version of) Sosa’s story is true?
And if such asymmetric relations of dependence do not ever obtain, then, by GTF, a
competent deduction never fails to transmit justification from premises to conclusion.
But that is obviously wrong: the Easy inferences above obviously do fail to transmit
justification from premises to conclusion. So: either GTF is wrong, or else Sosa’s coher-
entist epistemology is wrong. Since sections I and II above argued for GTF (by showing
that it provides the best explanation of cases of transmission failure), I conclude that
Sosa’s coherentist epistemology is wrong.
Could we accept Sosa’s coherentist by modifying GTF so that it allowed for some
directions of epistemic dependency to be more significant than others, as follows:
(GTF—modified) Necessarily, S’s competent deduction P therefore Q fails to transmit
(doxastic) justification from P to Q if there is some E such that
(ii) S is justified in believing that P (to some significant degree) in virtue of (i), but
(iii) (i) obtains to no more than a slight degree in virtue of S’s being justified in
believing that P.
While Sosa’s coherentist epistemology is consistent with GTF—modified, GTF—
modified is not true. Recall one of our cases above:
----------------------------------------------------------------------------------------------
Here, you are justified in believing (Easy Calculation#—2) on the basis of (Easy
Calculation#—1) only in virtue of your having some independent source of justifica-
tion (E) for believing that your calculations are accurate on this occasion. If E makes
you justified in believing that your calculations are accurate on this occasion, then E is
also part of what makes you justified in believing (Easy Calculation#—3). Furthermore,
you are justified in believing (Easy Calculation#—2) in virtue of E’s helping to make
you justified in believing (Easy Calculation#—3). But does E help at all to make you
justified in believing (Easy Calculation#—3) in virtue of your being justified in believ-
ing (Easy Calculation#—2)? No. E justifies you in believing (Easy Calculation#—3)
simply in virtue of justifying the first conjunct of (Easy Calculation#—3), and the justi-
fication of that conjunct does not depend at all on your justification for (Easy
Calculation#—2). To suppose, as Sosa would have to, that E helps to make you justified
in believing (Easy Calculation#—3) in virtue of your being justified in believing (Easy
Calculation#—2) is to suppose the existence of support relations for which there is no
evidence at all, independently of Sosa’s own theory.
Of course, if the content of (Easy Calculation#—2) were highly implausible given
other things you are justified in believing, then that would defeat your justification for
(Easy Calculation#—3). But it does not follow from that that, when (Easy
Calculation#—2) is not implausible, it plays any role in rendering (Easy
Calculation#—3) justified. GTF—modified will not work, and so Sosa’s coherentist
epistemology cannot be rendered consistent with the correct account of the transmis-
sion failure displayed by the Easy inferences.
In the next section, we’ll see what can be salvaged from Sosa’s epistemological pic-
ture. It turns out that the answer is: quite a bit. But the cost of salvaging it in the way
that I propose doing will be to move from a coherentist epistemology to a foundation-
alist epistemology in which coherence plays an important role.
V. A Foundationalist Solution to the Problem of

Epistemic Circularity
If we do not accept Sosa’s account of the regress of justifiers, then how should we
respond to this regress? In order to develop my answer to this question, let me begin by
making a point about an agent’s total evidence. Whatever exactly our total empirical
evidence consists in, it will be a conjunction of propositions, each of which is such that
we are empirically justified in believing it. And there are going to be many particular
propositions in our evidence set which are such that, in order to be justified in believ-
ing those particular propositions, we will also have to be justified in believing in the
veracity of the source of that particular proposition’s justification. (Specifically, this
246 Ram Neta
will be true at least of those evidence propositions which, like the second premise in
each of the Easy inferences, we are justified in believing on account of reasons that we
can identify.) But how can we be justified in believing in the veracity of that source on
the occasion in question, without relying upon the very pieces of evidence produced
by that source on that occasion?
This can happen if our total evidence contains pieces of evidence from a variety of
sources or occasions, where the deliverances of each corroborate the deliverances of
the others. If you have evidence from any one cognitive source and occasion of which
you are aware, then you also have evidence that does not derive from that same source
and occasion. So, with these pieces in place, I can finally state my solution to the prob-
lem of epistemic circularity as follows: in order for me to justifiably believe some
propositions as a result of the operation of some cognitive source of which I am aware,
I must be justified in believing in the veracity of that particular cognitive source on
that particular occasion. But in order to be justified in believing in the veracity of that
particular cognitive source on that particular occasion, my total empirical evidence
must make it highly probable that the source in question is veracious on that occasion.
In order for my total empirical evidence to make it highly probable that the source in
question is veracious on that occasion, my total empirical evidence must contain
pieces of evidence that are not from the source in question or not from the occasion in
question, and those other pieces of evidence must corroborate the evidence I get from
the source in question on that occasion. But I am justified in believing each of the
particular pieces of my total evidence (at least those particular pieces that I am justi-
fied in believing in virtue of the operation of some cognitive source of which I am
aware) only in virtue of being justified in believing some propositions about the
veracity of its source which I am in turn justified in believing only in virtue of my total
evidence. In short, I am justified in believing each proposition in my total evidence
only by virtue of being justified in believing some conjunction of evidence proposi-
tions. To sum up the present view in a doubly misleading way: I am justified in believ-
ing each particular evidence proposition only because I am justified in believing all of
them.7
This last formulation is misleading in two ways. First, it misleadingly suggests that
my justification for believing each particular evidence proposition is somehow infer-
ential, as if I infer each evidence proposition from the conjunction of my total evi-
dence. But this is wrong. There is a difference between justifiably inferring p from q, on
the one hand, and being justified in believing p partly in virtue of being justified in
believing q, on the other. The former is a species of the latter, but the latter is a much
broader category. For instance, I typically infer the conclusion of a mathematical proof
simply from the mathematical premises of that very proof, and not from any
non-mathematical propositions concerning my execution of the proof, and yet I am
7
In Neta (2008), I defend the propositional conception of evidence that I merely assume in the present
essay.
justified in believing the conclusion of the proof only in virtue of my being justified in
believing some non-mathematical propositions concerning my execution of the proof
(e.g., I was not careless when I did it). Similarly, I do not infer the particular proposi-
tions in my evidence set from their conjunction, but this does not mean that I am not
justified in believing those particular evidence propositions in virtue of being justified
in believing some conjunction of them.
But I also do not want to suggest that there is some specific conjunction of proposi-
tions in my evidence set such that my justifiably believing any single proposition in my
evidence set depends in any way (inferential or otherwise) upon my being justified in
believing that specific conjunction. Rather, what my justifiably believing any single
proposition in my evidence set depends upon is there being some conjunction or other
of evidence propositions that includes the single proposition in question, and includes
other propositions justified by other sources than the source that justifies the former
proposition, such that I am justified in believing that conjunction.
So the overall picture that I am proposing is as follows. My evidence set consists of
a conjunction of empirical propositions, each of which I am (non-inferentially) justified
in believing and each of which I am justified in believing partly in virtue of being jus-
tified in believing the others (or any others that could equally well constitute an evi-
dence set that included that particular conjunct). The evidentiary status of each
evidence proposition thus depends on its coherence with the rest of the propositions
in my evidence set: no set of propositions could constitute an agent’s evidence set
unless it was sufficiently coherent. By adopting this position, we can solve the prob-
lem of easy knowledge by appeal to transmission failure, and we can do so without
forsaking empiricism, and without running into any insurmountable problems of
epistemic circularity. And notice that it is very close to Sosa’s picture, without being
committed to his problematic claim that epistemic dependence relations are always
symmetric.
But have we really avoided the problem of epistemic circularity altogether? I said
that we corroborate the evidence that we get from each source at a moment by appeal-
ing to other sources or other moments that provide some of our other evidence. But we
cannot corroborate the whole of our empirical evidence. So what justifies me in believ-
ing that my total empirical evidence is not systematically misleading?
Contrary to those “conservatives”8 who claim that we are a priori justified in
believing that our total empirical evidence is not systematically misleading, I claim
that what justifies me in believing that my total empirical evidence is not systemati-
cally misleading is simply: my total empirical evidence. My total empirical evidence
justifies me in believing quite a few things about the world, including the following
claims of this paragraph. What evidence someone has is a result of impacts upon
their sensory systems. Such impacts are interrelated in such complicated ways that
it would be very difficult to make them systematically misleading (i.e., misleading
8
Wright (2004).
248 Ram Neta
in a way that left no trace in one’s evidence set itself). At the present time, I have no
evidence of the existence of anything that can perform such a difficult task, and a
great deal of evidence that no such thing exists. (Perhaps scientists who have more
evidence than I do, and schizophrenics who have less evidence than I do, ought
both to be less confident than I am about the non-misleading character of their total
evidence.) My total empirical evidence justifies me in believing that my total evi-
dence is not systematically misleading. Perhaps it is misleading here and there, but
it is not so misleading as to make it impossible for me to correct it by appeal to my
total evidence.
Given what I’ve said above, how should we respond to the regress of justifiers
described above? Here, we have plenty of latitude. We could, for instance, accept either
(i) or (ii), and treat the conjunction of all of one’s evidence propositions at a time as a
single foundationally justified proposition (either external or not).9 While GTF forces
us to reject the view that Sosa himself espouses, we can grant much of what Sosa says in
emphasizing the role of coherence in justification.
References
Neta, R., 2008. “What Evidence Do You Have?” British Journal for the Philosophy of Science 59:
89–119.
Neta, R., 2012. “Easy Knowledge, Transmission Failure, and Empricism.” Oxford Studies in
Epistemology 4: 166–84.
Wright, C., 1985. “Facts and Certainty.” Proceedings of the British Academy 71: 429–72.
Wright, C., 2004. “Warrant for Nothing (and Foundations for Free)?” Proceedings of the
Aristotelian Society 78: 167–212.
9
This is my suggestion in Neta (2008).
Index
abilities: epistemic circularity

and being reliably successful 66–7, 70–2, a foundationalist solution to the problem
130–1 of 245–8
cognitive 72–7 single-case 224–36
defective excercises of 68–9 epistemic dependence:
fallibility of 65, 73–7 negative 5–6, 35–42
perceptual-recognitional 74–7 positive 5–6, 42–5
success-rate 70 epistemic Twin-Earth 35–8
“success thesis” about 63
see also competences foolishness
achievement as epistemic indifference or
knowledge as 55, 126–9 insensitivity 211–20
see also attainment Frankfurt, Harry 109, 121, 209–10
adequateness 131
adroitness 4, 125 Geach, Peter 72 n. 14
amplitude 129 Gettier problem 127, 129
animal knowledge 152–3 Goldman, Alvin 85, 86 nn. 6 and 7, 87, 89–91,
aptness 4–7, 10, 125 93–4, 114–15
as affected by near error and luck 20–4
Austin, John L. 66, 67 Hazlett, Alan 216–17
attainment Hume, David 109 n. 9, 157
knowledge as 129–31
intellectual autonomy
Battaly, Heather 213–14, 216 and the value of reflection 172–4, 191–4
beliefs as performances 177–81 as endorsement of one’s own epistemic
goals 192–4
cognitive essentialism 88–90 intuitions
coherence as intellectual seemings 139–40
and the problem of epistemic circularity 241–8 as providing a priori justification 143–8
as a source of reflective knowledge 160–3
competences: judgment
epistemic 27–8, 44–5, 77–80 as a second-order endeavor 24–8
structure of 3–4, 26–7
see also abilities Kant, Immanuel 201
knowledge
dispositions 28–9 and abnormal circumstances 56–7
cognitive 26, 33–4, 38–41, 78–9, 130–1 and luck 4–5, 23–4, 37–41
finks vs masks 26 as action-like 108–9
the value of 120–2, 126, 129, 207–8
easy knowledge tracking account of 110–14
and transmission-failure 230–6
the problem of 224–30 Locke, John 212
Sosa’s treatment of the problem of 158–60 luck
Elgin, Catherine 205–8 and possession of competences 23–4
epistemic agency
and intellectual autonomy 13–14, 172–4 McDowell, John 68–9, 77 n. 16
as improving reliability 168–72
its role in second-order belief 174–6 New Evil Demon Problem
knowledge as requiring 9–10, 110–18 as a problem for externalist’s
epistemic bootstrapping accounts of justification 8–9, 88,
and generalized transmission-failure 236–9 153–7
250 index
Peirce, Charles S. 210 Sosa’s view on possesion of competences

performances in unfavorable environments
hard and easy 68–70 77–80
Plato 23, 126, 219 Sosa’s solution to the problem of epistemic
Pollock, John 85, 86, 88 n. 10, 89 circularity 241–5
Pope, Alexander 212 stupidity
as a misuse of reason 204–8
Raz, Joseph 73 n. 15 intellectualist conception of 200–4
reflection
value of 168–76, 191–4 transmission-failure
reflective knowledge and generalized transmission-failure
and philosophical assurance 157–60 233–6
as better than animal knowledge 184–91 and propositional
reliabilism transmission-failure 231–3
absolutist actual-world 93–4
indexicalist actual-world 94–7 Velleman, David 107, 110 n. 12
normal-conditions 100–4 virtue epistemology
normal-worlds 90–2 anti-luck 45–7
character-based 2, 198
safety condition for knowledge dispositional robust 31–4
absolute 53 responsabilist 198
varieties of 52–5 see also virtue reliabilism
relation with aptness 6–7, 41–2, 45–6, 55–7 virtue reliabilism 1–2, 198
relative 54
Schopenhauer, Arthur 204, 206 n. 13, 214 Wittgenstein, Ludwig 31 n. 1, 64 n. 5
Sosa’s view of second-order knowledge Wright, Crispin 230
and its connection to epistemic agency 174–6
internalist objections to 160–5 Zagzebski, Linda 34 n. 7, 130, 211 n. 20

Miguel Ángel Fernández Vargas - Performance Epistemology - Foundations and Applications-Oxford University Press (2016)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Miguel Ángel Fernández Vargas - Performance Epistemology - Foundations and Applications-Oxford University Press (2016)

Uploaded by

Copyright:

Available Formats

Performance Epistemology

Introduction—Performance Epistemology: Foundations and Applications 1

Part I. Foundational Issues

Part II. Applications and Problems

12. The Epistemology of Stupidity 196

Paul A. Boghossian, New York University

Baron Reed, Northwestern University

when in relevant circumstances, with a sufficient degree of reliability”; many authors

notion of attributability in play is in need of a deeper explanation than it has received

performance-based epistemology on the other, is what is to count, and why, as a satis-

performance-based epistemology as a form of second-order incompetence to properly

B. How Does Near-actual Simulation Affect

C. Why It Is More Complicated

Competence comes of course in degrees, as does, correspondingly, aptness. Consider

D. Judgment and Judgmental Belief

competently or reliably enough. But there is a contextual element here, depending on

E. The Structure of Competences (and Abilities)

c. What then is required for possession of a competence? What is required for

F. Competences, Epistemic and Otherwise

G. The Nature and Epistemology of Competences

2. Sosa’s Dispositional Robust Virtue Epistemology

meet a virtue-theoretic condition—as robust virtue epistemology. Sosa’s proposal is of

3. DRVE and Negative Epistemic Dependence

Twin-S’s perceptual apparatus, as exercised in that environment, is thus equally relia-

Earth Epistemic Twin Earth

Global: H2O Global: H2O

Regional : H2O Regional : XYZ

Local : H2O Local : H2O

condition designed to deal with knowledge-undermining epistemic luck is thus explic-

4. DRVE and Positive Epistemic Dependence

5. DRVE versus Anti-luck Virtue Epistemology

or causal-explanatory (Greco-style) route. On the latter rendering, knowledge is safe

Part One: Varieties of Safety

The way to read this passage, I suggest, is as follows:

But, of course, that just is a weak, ability-relative safety condition, restricted to

Part Two: How to Think about Virtue and Safety

Adding the other conditions that Sosa requires for knowledge:

Part Three: Dreamers and Jokesters

This approach to perceptual knowledge is threatened, however, by Sosa’s case of the

2. The Ability to Ride a Bicycle and Kindred Abilities

3. Having an Ability and Being Reliably Successful

4. Hard Performances and Completable Performances

ability or power or capacity is inherently liable not to produce success, on occasion,

(1´) The success thesis extends to perceptual-recognitional abilities: one exercises

e­ xercise it. Clearly, it is possible to have a perceptual-recognitional ability and, due to

provide no decisive reason to invoke the notion of a defective exercise as McDowell

6. Sosa and Competence

I. Being in the Right

II. The Correct Procedures Conception

Here’s how John Pollock put the idea:

Here’s how Richard Feldman put the idea:

III. The Reliabilist Programme

My audience disagrees over the significance of this fact.

Simple reliabilism is a “vertical” mind-to-world account of justification.9 Beliefs are

IV. Cognitive Essentialism

VI. Normal-worlds Reliabilism

Here is how Goldman defines normal worlds:

Normal-worlds reliabilism apparently avoids the counterexample:

Couldn’t there be another kind of psychological capacity in another kind of cognitive

VII. Actual-world Reliabilism

ACTUAL-WORLD RELIABILISM: In all possible worlds W, a belief is prima facie justified in

VIII. Absolutist Actual-world Reliabilism

Introduction—Performance Epistemology: Foundations and Applications 1

12. The Epistemology of Stupidity 196

B. How Does Near-actual Simulation Affect

e xercise it. Clearly, it is possible to have a perceptual-recognitional ability and, due to

IV. Sosa’s Attempted Solution to the Problem of

V. A Foundationalist Solution to the Problem of