Van Fraassen Bas C. (2008) - Scientific Representation Paradoxes of Perspective PDF

Scientific Representation: Paradoxes of Perspective
This page intentionally left blank

Scientific Representation:
Paradoxes of Perspective
Bas C. van Fraassen
CLARENDON PRESS · OXFORD

1
Great Clarendon Street, Oxford  
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 in
Oxford New York
Auckland Cape Town Dar es Salaam Hong Kong Karachi
Kuala Lumpur Madrid Melbourne Mexico City Nairobi
New Delhi Shanghai Taipei Toronto
With offices in
Argentina Austria Brazil Chile Czech Republic France Greece
Guatemala Hungary Italy Japan Poland Portugal Singapore
South Korea Switzerland Thailand Turkey Ukraine Vietnam
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
 Bas C. van Fraassen 2008
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2008
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, 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 book in any other binding or cover
and you must impose the same condition on any acquirer
British Library Cataloguing in Publication Data
Data available
Library of Congress Cataloging in Publication Data
Data available
Typeset by Laserwords Private Limited, Chennai, India
Printed in Great Britain
on acid-free paper by
Biddles Ltd, King’s Lynn, Norfolk
ISBN 978–0–19–927822–0
10 9 8 7 6 5 4 3 2 1
for
Janine Blanc Peschard
and the memory of
Dina Landman van Fraassen
Preface
When I began to rewrite the Locke Lectures I gave in Oxford in 2001

I found the comments I received leading me into new paths, some quite
unexpected. My main concern had been, and remained, the possibilities
for an empiricist version of structuralism in philosophy of science. But
how I came to conceive of those possibilities was altered first of all
by closer contact with the revivals of transcendentalist and neo-Kantian
thought, and secondly by the lively and growing interest I encountered in
technology, instruments, and experimental practices both in contrast with,
and complementary to, philosophical reflection on scientific theories.
Not all of my debts will appear explicitly in the text. I am thankful
especially for the great good fortune I had to meet Isabelle Peschard,
my main interlocutor on both these subjects. My thanks also to the
writings, helpful discussion, and correspondence of Michel Bitbol, Michael
Friedman, Alan Richardson, and Thomas Ryckman for more insight
into the neo-Kantian tradition, and to Mieke Boon, Nancy Cartwright,
Margaret Morrison, and Hans Radder for discussions of instruments and
experimentation. These debts are in addition to many debts, accumulated
with respect to structural realism and surrounding topics, to Jeffrey Bub,
Jeremy Butterfield, Otávio Bueno, Chris Fuchs, Hans Halvorson, James
Ladyman, Bradley Monton, Carlo Rovelli, and Simon Saunders. On the
subject of representation in the sciences, as will be quite clear, I have
substantial debts to Ronald Giere and Paul Teller. In addition, Bradley
Monton and Paul Teller went through an early manuscript version with a
fine tooth comb and made many valuable detailed comments on the text. I
am painfully aware that I owe more debts to more people than I can relate
here. But my greater debt, beyond words, is to Isabelle Peschard.
The main addition to the locke lectures comes in Part Two, on measure-
ment as representation, which is also the more technical part of the book.
While in any logical ordering this material precedes the third and fourth
parts, the less patient reader may wish to skip ahead to them.
viii 
The generous sabbatical and leave policy of Princeton University, a

splendid year at the Center for Advanced Study in the Behavioral Sciences
in Stanford, and the hospitality enjoyed at All Souls and Magdalen Colleges
in Oxford, the CREA (Centre de Recherche en Epistémologie Appliquée)
in Paris, and the University of Twente in the Netherlands, as well as Senior
Scholar Award SES-0549002 from the National Science Foundation, made
this work possible. My special thanks to Ralph Walker, Jean Petitot, and
Philip Brey for making me welcome at their institutions in Oxford, Paris,
and Twente respectively.
Bas C. van Fraassen
List of Figures
2.1. Reflection and Refraction 44

3.1. Perspective Altimetry 62
3.2. Window and Checkerboard 63
3.3. Dürer, The Draughtsmen of the Lute 65
3.4. Frame of Reference . Perspective 67
3.5. Speed in Perspective 68
3.6. Cross Ratio Invariance 73
4.1. Geometry of the Rainbow 102
4.2. Image Categories 104
6.1. Measurement Schema 147
6.2. Coherence of Measurement 153
8.1. Adequacy as Symmetry 196
10.1. Putnam’s Paradox 230
12.1. Copernicus’s Model of Retrograde Motion 288
13.1. Failure of Supervenience 293
Contents
Preface vii
List of Figures ix
Introduction: the ‘picture theory of science’ 1
Part I. Representation
. Representation of, Representation As 11

The value of distortion 12
How does a representation represent? 15
What’s in a photo? 20
What is a representation then? 22
Appearance to the intellect: illumination as embedding 29
In conclusion 30
. Imaging, Picturing, and Scaling 33

Modes of representation 33
What distinguishes a picture? 36
Mathematical imagery, distortion through abstraction 39
Scale models and virtuous distortion 49
Conclusion about imaging and scaling 56
. Pictorial Perspective and the Indexical 59

Pictorial perspective and the Art of Measuring 60
Perspective versus Descartes’s frames of reference 66
Mapping and perspectival self-location 75
What is in a map? 82
Visual perspective and the metaphor 84
Concluding empiricist postscript 86
xii 
Part II. Windows, Engines, and Measurement
. A Window on the Invisible World (?) 93

Instrumentation’s diversity of roles 94
Engines of creation: engendering new phenomena 100
The microscope’s public hallucinations 101
Objections to this view of ‘observation by instruments’ 105
Experimentation’s diversity of roles 111
. The Problem of Coordination 115

Coordination: a historical context 116
The problem of coordination reconceived 121
Mach on the history of the thermometer 125
Poincaré’s analysis of time measurement 130
Observables coordinated: two morals 137
. Measurement as Representation: . The Physical Correlate 141

Physical conditions of possibility for measurement 141
General theory of measurement 147
What is not measurement 156
. Measurement as Representation: . Information 157

What is measurement—number-assigning? 158
The scale as logical space 164
Data models and surface models 166
The over-arching concept for measurement 172
What is a measurement outcome? 179
Relating the views ‘from above’ and ‘from within’ 184
Part III. Structure and Perspective
. From the Bildtheorie of Science to Paradox 191

The Bildtheorie controversy 191
Representation: the problem for structuralism 204
 xiii
. The Longest Journey: Bertrand Russell 213

Prolegomena to Russell’s conversion to structuralism 213
Russell’s structuralist turn 217
Conclusion 223
. Carnap’s Lost World and Putnam’s Paradox 225

Carnap: Der Logische Aufbau der Welt 225
Putnam’s Paradox 229
Staying with Putnam: the Paradox dissolved 232
. An Empiricist Structuralism 237

What could be an empiricist structuralism? 237
The fundamental remaining problem for a structuralist view
of science 239
The two main dangers for an empiricist 244
The problem in concrete setting revisited and dissolved 253
Return to our epistemological question 261
Part IV. Appearance and Reality
. Appearance vs. Reality in the Sciences 269

Appearance and reality: the real and unreal problem 270
Appearance versus reality at the birth of modern science 270
Three putative completeness criteria 276
Appearance vs. reality: A deeper Criterion 280
Phenomena versus appearances 283
Three-faceted representation 288
. Rejecting the Appearance from Reality Criterion 291

The supervenience of mind challenge 292
The Great Leibnizian Escape move 296
The quantum mechanics challenge 297
Exploring the case of quantum mechanics 300
Supervenience? 304
An empiricist view 304
xiv 
APPENDICES
Appendix to CH . Models and theories as representations 309

Appendix to CH . Quantum peculiarities: fuzzy observables 312
Appendix to CH . Surface models and their embeddings 315
Appendix to CH . Retreat (?) from The Scientific Image 317
Notes to Appendices 320

Bibliography 322
Notes 345
Index 399
Introduction: the ‘picture theory
of science’
The Bildtheorie—‘picture theory of science’—formed the frame for much

discussion and controversy among physicists in the decades around the year
1900. In retrospect, it had close connections with philosophers’ attempts to
forge what we would now call a structuralist view of science. The debates
and projects of those years foreshadowed the debates some half a century
later, closer to our time, over scientific realism and structural realism.
To understand science we need to approach it from many directions. I
will focus on one aspect that I take to be central to the scientific enterprise:
representation of the empirical phenomena, by means of artifacts, both
physical and mathematical. The position that success in this respect is the
defining aim of the empirical sciences is an empiricist theme from which I
will not depart. In fact, this focus on representation fits the empiricist theme
very well. While we are now, it seems to me, much more demanding
in what we expect of philosophical accounts of science, I will take the
Bildtheorie as exemplar and inspiration. But there will be some major
differences between the view of science that I shall advocate and what was
then known under that name.
First of all, the Bildtheorie view was phrased in terms of mental pictures.
Thus Boltzmann formulates the goal of physical theory as ‘‘constructing
a picture of the external world that exists purely internally’’ (Boltzmann
1905, 77; 1974, 33) and called the product of scientific theorizing an inner
picture or mental construction (‘‘inneres Bild’’, ‘‘gedankliche Konstruc-
tion’’; Boltzmann 1974, 106; 165–6.). In an Encyclopedia Britannica article
he spelled it out:
2 
On this view our thoughts stand to things in the same relation as models to the
objects they represent. The essence of the process is the attachment of one concept
having a definite content to each thing, but without implying complete similarity
between thing and thought; for naturally we can know but little of the resemblance
of our thoughts to the things to which we attach them. (Boltzmann 1974, 214.)
I will have no truck with mental representation, in any sense.1 The view
Boltzmann expresses here, a view in philosophy of mind or language,
has nothing to contribute to our understanding of scientific representa-
tion—not to mention that it threw some of the discussion then back into
the Cartesian problem of the external world, to no good purpose.
Scientific representation is—as Boltzmann’s own examples in that article
amply show—by means of artifacts both concrete (graphs, scale models,
computer monitor displays, and the like) and abstract (mathematical models,
needed especially when the infinite on infinitesimal play a role). It is on
these artifacts, their use, and the characteristics that are germane to the
roles they play in this use, that we must focus. The reservations about how
the represented must be like its representation, which Boltzmann expresses
about thought, pertain equally to these artifacts, and are pertinent to any
view of how a science relates to its domain of application.2
Secondly, it is not only to our understanding of theories and their
models that representation is relevant. The achievement of theoretical
representation is mediated by measurement and experimentation, in the
course of which many forms of representation are involved as well. Scientific
representation is not exhausted by a study of the role of theory or theoretical
models. To complete our understanding of scientific representation we must
equally approach measurement, its instrumental character and its role. I
will argue that measuring, just as well as theorizing, is representing. The
representing in question also need not be, and in general is not, a case of
mimesis; rather, measuring locates the target in a theoretically constructed
logical space. In this respect I shall make common cause with views
currently found in philosophy of technology.
Thirdly, the analysis of measurement as well as of the conditions of
use for theoretical models can be completed only through a reflection on
indexicality. Since at least the time of Poincaré, Einstein, and Bohr it is
a commonplace that a measurement outcome does not display what the
measured entity is like, but what it ‘looks like’ in the measurement set-up.
 3
That point does not go nearly far enough. It serves, however, to announce
the introduction of relationality, perspective, intensionality, intentionality,
and the essential indexical into the discussion of science, though it stops far
short of presenting their full role.
Debates in philosophy of science take place in the context of much wider
tensions and oppositions in epistemology and metaphysics. When, in Part
Three, I come to the theme of structuralism I will begin with protagonists
in the historical Bildtheorie debate, and show how it was already implicitly
challenged by an apparent paradox at the very heart of its conception
of science. I will then take the message about the role of indexicality in
scientific representation to the paradoxes that beset, bedeviled, or otherwise
preoccupied Bertrand Russell, Rudolf Carnap, Hilary Putnam, and David
Lewis. For my part I will propose an empiricist structuralism, in contrast
to structural realism, as a view of science that can stand up to these
challenges. Then, in the last part, I will address the troubling relations that
the appearances bear to the world’s scientific image.
This, so far, is the general outline of what I am setting out to do. But
it may help to add here something about my own starting point. I try
to be an empiricist, and as I understand that tradition (what it is, and
what it could be in days to come) it involves a common sense realism in
which reference to observable phenomena is unproblematic: rocks, seas,
stars, persons, bicycles . . . . Empiricism also involves certain philosophical
attitudes: to take the empirical sciences as a paradigm of rational inquiry,
and to resist the demands for further explanation that lead to metaphysical
extensions of the sciences. There is within these constraints a good deal of
leeway for different sorts of empiricist positions. For my part, specifically,
I add a certain view of science, that the basic aim—equivalently, the
base-line criterion of success—is empirical adequacy rather than overall
truth, and that acceptance of a scientific theory has a pragmatic dimension
(to guide action and research) but need involve no more belief than that
the theory is empirically adequate.3 While this will undoubtedly shape my
discussion, I have tried to write as much as possible of this book in a
way that does not trade on the differences between this view of science
(‘constructive empiricism’) and its contraries (‘scientific realisms’). What
scientific representation is and how it works is everyone’s concern, and
there we may find a large area where more general philosophical differences
need make no difference.4
PA RT I
Representation
Aristotle was trading on a typical ambiguity when he wrote ‘‘Tragedy
is a representation of a serious, complete action which has magnitude,
in embellished speech . . . ; represented by people acting and not by
narration. . . .’’ (Poetics 49b25). We can read his statement as describing
either the poet’s activity or the poet’s product. It is in the activity of
representation that representations are produced. This is not an accidental
equivocation. We lose our topic altogether if we attempt to ask ‘‘what
is a representation?’’ and tacitly take just one or the other aspect into
account; for in fact we cannot understand either in isolation. That applies
to scientific representation as well.
There is a vast and recently rapidly increasing literature on representation
both in general and in philosophy of science. Let me express at once my
accord with the approach advocated by Mauricio Suarez:
I propose that we adopt from the start a deflationary attitude and strategy towards
scientific representation, in analogy to deflationary or minimalist conceptions of
truth, or contextualist analyses of knowledge. [. . .] Representation is not the kind
of notion that requires a theory to elucidate it: there are no necessary and sufficient
conditions for it. We can at best aim to describe its most general features. . . .1
This does require some decisions about what we can and cannot take for
granted, to be used in this description, as already understood. While not
trying to define representation or to reduce it to something else, we will
have to place it in a context where we know our way around.
The first question to broach is this: how, or to what extent, is represent-
ation related to resemblance or likeness-making? This venerable question
occupies the first two chapters. Not all, but certainly many forms of repres-
entation do trade on likeness, likeness in some respects, selective likeness. That
is not what makes them representations; it is part of what defines them as the
sort of representation they are, and may figure in what constitutes success.
But even these tend to trade equally on unlikeness, distortion, addition. A
representation is made with a purpose or goal in mind, governed by criteria
of adequacy pertaining to that goal, which guide its means, medium, and
selectivity. Hence there is even in those cases no general valid inference
8   : 
from what the representation is like to what the represented is like overall. Not
surprisingly, empiricist views of science will differ from scientific realist
views on where they locate the selective likeness and unlikeness.
The second question concerns perspective. The third chapter will exa-
mine and elaborate on the enigmatic but now oft-seen contention that
scientific representation is perspectival. In their general use, the words
‘‘perspective’’ and ‘‘perspectival’’ are largely metaphorical. The literal use
appears only when we assert, for example, that artists in the Renaissance
began to draw and paint in perspective.2 As I shall elaborate, the pertinent
point about this technique is Albrecht Dürer’s: drawing in perspective
is a measurement technique.3 The art of perspectival drawing is an art of
measuring. It is a technique for rendering a systematically selective likeness,
yielding information in desired respects, and it provides an initial paradigm
for measurement in general.
We are perennially plagued by the shifting uses and senses of even our
most common terms. In the long history of tension between physics and
astronomy before the modern period, ‘‘saving the phenomena’’ refers to
the appearances of the celestial bodies and their motions to the astronomer,
that is, in the outcomes of the astronomer’s measurements. Those celestial
bodies and their motions are one and all observable, unlike e.g. the
postulated crystalline spheres. But when Kant takes on this terminology of
‘‘appearance’’ and ‘‘phenomenon’’, he entwines their meanings so deeply
in his transcendentalist philosophy that we find ourselves as it were in a
different language. While I have not done so before, I will here make
a terminological distinction between these two words, though neither
will have Kant’s meaning. Phenomena will be observable entities (objects,
events, processes). Thus ‘‘observable phenomenon’’ is redundant in my
usage. Appearances will be the contents of observation or measurement
outcomes. The celestial motions of concern to the ancient or medieval
astronomer were all phenomena, in my sense, but Copernicus insisted that
they had confused what those phenomena are like with how they appear to
the earthly observer. Thus he distinguished, in my terms, the appearances
(in the measurements made by the astronomers) from the phenomena
(which they observed and measured).
Regimenting the terminology in this way, or in any way at all, will chafe
on some common usage. But it will align with other usage no less common.
For example, combustion, St. Elmo’s fire, lightning, and the aurora borealis
  :  9
are all commonly named as phenomena; and whereas Giorgione was so called
because of his size, Mars is called the Red Planet not because of its color but
because of its reddish appearance as seen or photographed from the Earth.4
With this regimentation we will have available a distinct terminology to
honor the insight that what measurement shows is not directly what the
measured is like but how it appears in that particular measurement set-up.
There are other techniques besides perspectival drawing, such as those
of the cartographer, that can provide a paradigm for our conception of
measurement and modelling. Perspective comes in there as well, though in
a quite different way, when we examine the conditions for the possibility of use
of these representations, for prediction and manipulation in practice. The
lessons drawn from these seminal examples illuminate not only the painter’s
art but the construction and use of models in science and technology.
Much of Part I will come by way of prolegomenon to further study,
marshalling telling cases to illuminate representation—and what will be
its use? Firstly, to remove the blinders that could focus us naïvely on
the idea that what is represented is simply like what is presented in the
representation. Equally, to save us from the opposite error, of assuming a
total independence of the represented from the content of its representation.
Likeness in contextually selective fashion is important to scientific practice.
The world, the world that our science is of , is the world depicted in science,
and what is depicted there, is the content of its theoretical representations;
but there is less to this equation than meets the eye—and thereby hangs
a tale. . . .
1
Representation Of,
Representation As
The most naïve view of representation might perhaps be put something

like this: ‘‘A represents B if and only if A appreciably resembles B’’.
Vestiges of this view, with assorted refinements, persist in most writing
on representation. Yet more error could hardly be compressed into so
short a formula. (Goodman 1976: 3–4)
Nelson Goodman was quite right to say so. The budget of examples
and counter-examples that prove it—we will look at some below—have
largely been with us since almost the beginning of philosophy. But there
must be a reason if the idea he disparages, that resemblance is crucial
to representation, is so persistently seductive. That perfect likeness is an
ideal pursued in visual imagery, at least, has much historical support. Pliny
described the painter Zeuxis’ grapes as so lifelike that birds tried to eat
them—while Zeuxis in turn was fooled by Parrhasios who painted a
curtain with such trompe l’oeil perfection that Zeuxis asked him to pull
the curtain aside in order to show his painting.1 We can cite the art of
our time as well: hyperrealism in recent painting, such as Donald Jacot’s or
Jacques Bodin’s, is surely admired in part for excellence of that kind. So
while representation cannot be equated with the presentation of a likeness,
and resemblance to what is represented is not crucial to representation as
such, resemblance does play a role inviting our attention.
12  : 
The value of distortion2

That even visual, pictorial or plastic, representation is not a matter of
producing accurate ‘copies’ exactly like their originals is already clear in
Plato’s dialogues.
Socrates and his young students once had a visitor from Elea whom they
invited to take over Socrates’ usual role in their ongoing seminar. The
visitor agreed, choosing Theaetetus as interlocutor, and soon has him tied
in knots on the subject of representation. There are real things, says the
Eleatic Stranger, and there are images of real things—some are artifacts like
paintings or sculptures, others natural like dreams, shadows, or reflections
(Sophist 265e–266d). Aren’t these images copies of what they are images of?
Theatetus had already expressed a view on copy-making: ‘‘What in the
world would we say a copy is, sir, except something that’s made similar
to a true thing and is another thing that’s like it?’’ (Sophist 239d–240a).
But does that apply to images in general? The Stranger reminds him that a
sculptor may need to distort in order to represent something successfully. To
make an exact likeness with respect to shape would require preserving the
proportions of length, breadth, and depth of the original, and the colors of
the parts as well. Those who sculpt or draw very large works don’t do that:
‘‘If they reproduced the true proportions . . . , the upper parts would appear
smaller than they should, and the lower parts would appear larger, because
we see the upper parts from farther away. . . .’’ (Sophist 235d–236a).
We may wonder whether Plato is referring to actual examples generally
discussed in Athens. There is a story, though its provenance not entirely
clear, related by Ernst Gombrich, concerning two sculptors of the fifth
century :
The Athenians intending to consecrate an excellent image of Minerva upon a high
pillar, set Phidias and Alcamenes to work, meaning to chuse the better of the two.
Aclamenes being nothing at all skilled in Geometry and in the Optickes made
the goddesse wonderfull faire to the eye of them that saw her hard by. Phidias
on the contrary . . . did consider that the whole shape of his image should change
according to the height of the appointed place, and therefore made her lips wide
open, her nose somewhat out of order and all the rest accordingly . . . when these
two images were afterwards brought to light and compared, Phidias was in great
danger to have been stoned by the whole multitude, until the statues were at
 ,   13
length set on high. For Alcamenes his sweet and diligent strokes being drowned,
and Phidias his disfigured and distorted hardnesse being vanished by the height
of the place, made Alcamenes to be laughed at, and Phidias to be much more
esteemed. (Gombrich 1960: 191)
As Roger Shepard (1990) has studied and richly illustrated in our own time,
this point is general, and does not just apply to sculptors of images to be
seen from below. Even in an ordinary drawing, if you want two differently
oriented parallelograms to look congruent, you have to make one larger
than the other.
It seems then that distortion, infidelity, lack of resemblance in some
respect, may in general be crucial to the success of a representation.3 This
does not rule out that resemblance in some other respect may be required.
Yet even when that is the case—and it may be a special case—the choice
of those respects in which resemblance or a specific kind of distortion is
required, and those for which just anything at all will do, will have to be
seen as crucial as well.
There must be a cautionary tale here for how we are to understand
scientific representation. It may be natural to take a successful representation
to be a likeness of what it represents—but much hinges here on what
the criteria of success were when it was made. One sort of success is
precisely what Copernicus was taken to have as against Ptolemy: that his
theory displayed the real structure of the cosmos.4 In general though, can
we infer from success of a representation, in respects that we can directly
appreciate, to the conclusion that it bears a structural resemblance to what
is represented? The examples of how distortion may be crucial to successful
representation (in view of the purpose of the representing) should certainly
give us pause. But now we are running ahead of the story.
Caricature and misrepresentation

Successful representation may require deliberate departures from resemb-
lance. It does not follow that likeness will always be irrelevant to successful
representation. Certainly the Eleatic Stranger’s example does not show that,
since even the sculptor of statues placed on high must ensure resemblance
in some respect.
But now consider another side to the role of distortion. Misrepresentation
is a species of representation after all: a caricature of Mrs. Thatcher may
14  : 
misrepresent her as draconian, but it certainly does represent her, and not
her sister or her pet dragon or whatever else she may have. Yet even if we
take the caricature to represent her because of some carefully introduced
resemblance there, we can declare it a misrepresentation by insisting that
it represents her as something she is not. A caricature may represent a rather
tall man as short (as a well known cartoon depicts Supreme Court Justice
Clarence Thomas as very small compared to the chair he occupies), but it
represents that man, and not someone that it resembles more as to height.
A caricature misrepresents on purpose, to convey a message that is clear
enough in context but is to be gleaned in a quite indirect fashion. My
two examples are both visual representations, but they are not visually
accurate, nor does their visual inaccuracy serve to produce a visually
accurate appearance to a properly placed eye (as in the Sophist’s sculptor’s
case), and yet, neither is their inaccuracy accidental.5
So distortion—departure from resemblance—which may be crucial to
accurate representation in certain cases, is in other cases the vehicle of
effective misrepresentation. Resemblance in some particular respect may
be the vehicle of reference: we recognize the caricature as being of Mrs.
Thatcher because of resemblance in certain respects. It may also be the
means of attribution or misattribution of some characteristic: we take the
caricature to represent her as draconian because of some likeness to a
dragon, which is actually an unlikeness to her. She is represented, and she is
represented as thus or so: the drawing is of her, and depicts her as thus or so.6
But a list of likenesses and unlikenesses does not tell us this much—why,
for example, is this not a caricature of a dragon as Thatcherian?
Let us look at another drawing, say, Spott’s drawing of Bismarck as a
peacock.7 Is this drawing a misrepresentation? There we broach a ques-
tion of truth that certainly is not settled in terms of visual accuracy or
inaccuracy—even if both reference and attribution were effected by select-
ive uses of resemblance and non-resemblance. The judgment whether
this is an illuminating caricature that conveys a truth about him, or
amounts to a falsehood, a lie, a misrepresentation, is not settled by geo-
metrical relations between the line drawing and Bismarck’s appearance to
the eye.
If we just focus on resemblance in some respect as the core notion in
representation then it is at best puzzling that distortion might be needed
for effective representation. But if the resemblances are just a means to an
 ,   15
end there is no puzzle. The sculptor wants the object he makes to have a
certain use, and he chooses the way in which the proportions of the object
are related to those of the original—the ways in which they are like and
the ways in which they are unlike —so as make that use possible. There have
(of course!) been efforts aimed at naturalizing representation, such as Fred
Dretske’s account of information-bearing as correlation, but these tend to
founder specifically on the issue of misrepresentation.8 Misrepresentation is
a species of representation. If the relationship ‘X represents Y’ were to lie in
a resemblance or correlation or other such structural relation between the
two, what would misrepresentation be?
Suppose I say that the caricature that depicts Mrs. T as draconian
misrepresents her. Then my assertion has as first part that it does represent
Mrs. T as draconian, but as second part not only that it is unlike her with
respect to shape, but that it depicts her as something she is not. To say
that it misrepresents her with respect to shape is to say that rather than
resembling her it depicts her as resembling (being like) something which,
according to me, she is not like. To say that it is a caricature, however, is
to say that it purveys an interpretative attribute, something that the picture
can convey only by drawing on a social context, not just on what is ‘in’
the picture taken by itself.
So what is accuracy? The evaluation of a representation as accurate or
inaccurate is highly context-dependent. A subway map, for example, is
typically not to scale, but only shows topological structure. Relative to its
typical use and our typical need, it is accurate; with a change in use or
need, it would at once have to be classified as inaccurate. Similarly, in one
political context, or relative to a certain kind of evaluation, a caricature may
rightly be judged to be accurate, in another misleading or blatantly false.
How does a representation represent?

We confront here the general question of how an item such as a picture
can correctly represent, misrepresent, caricature, flatter, or revile its subject.
Note well that the answer will not be also an answer to the question of
what representation is. The question here addressed arises rather if we take it
for granted that something is playing the representational role, and want to
know just how it plays that role.
16  : 
Nelson Goodman’s Languages of Art, the twentieth century’s seminal

text on the matter, characterizes even pictures as being like statements,
depicting a subject (the referent) as thus or so (as with a predicate).9 By
drawing the of/as distinction I was already more or less following him in
this view. That something is a picture of Bismarck does not imply that
Bismarck was in every respect the way he looks in the picture, it does
depict Bismarck as thus or so. Goodman translates this into: Depiction is [also]
predication. Since pictures denote things they are, in that respect, like names;
but since they depict those things as thus or so, they are also predicates
applied to what they denote. The content of this picture is the same sort of
thing—or should be talked about in the same sort of way—as the content
of a predicate or a sentence applying a predicate to the subject of that
painting. To say that X represents Y as F, that is taken to have as guiding
example something like
‘‘Snow is white’’ represents snow as colored
Putting these two points together it seems quite apt to liken a picture to a
sentence, rather than to either a name or a predicate. For in a sentence we
typically also see a subject to which a predicate is being applied—the subject
may be real enough, but it may or may not be correctly characterized. If the
sentence is complex, we may distinguish respects in which the predication
is accurate and ways in which it is not, just as we do with a painting.
Denotation, as Goodman understands it, is of something real, and the
relation is extensional. Pictorial predication need meet neither of the con-
ditions placed on denotation. If the painting depicts Thackeray as the
author of Waverley we cannot infer that Waverly exists, nor that it had a
single author if it does. But in addition, if Scott is the author of Waverley
we cannot even say that the painting depicts Thackeray as Scott, except as a
joke. Neither ‘‘predicates’’ nor ‘‘depicts as’’ is an extensional term.10
Goodman’s most controversial thesis is that denotation and predication
are also in the case of pictures entirely independent of each other. ‘‘The
denotation of a picture no more determines its kind’’, he writes, ‘‘than
the kind of picture determines its denotation’’.11 (He uses the ‘‘kind’’
terminology to distinguish pictures that predicate differently.) This thesis is
controversial since it is hard to accept that a picture could fail to convey
anything correct or true about something and still be a picture of that
thing.12 Instead, we would typically think that we can identify what it is a
 ,   17
picture of by looking at how things are depicted in it! But there are limiting
cases where this can be drawn into doubt.13
Undeniably, though, Goodman has brought into the limelight the strong
analogy: a picture is a picture of something, and depicts that something as
thus or so, and so is in that respect similar to how a verbal description is a
description of something, and describes that something as being thus or so.
We need now to see how we can go beyond this analogy.
One way in which Goodman did bring in resemblance was through
the intricate notion of exemplification. If a hardware store clerk or interior
decorator shows you color swatches or fabric samples, those exemplify the
property of interest in the following sense: they both have and refer to that
property (Goodman 1976: 45–68 ; Elgin 1996: 171–83). Obviously their use
is to represent to you what your wall or floor will look like if you choose
the corresponding paint or carpet. Representation by exemplification
involves likeness, but much more than likeness. In these examples visual
resemblance plays a role: in fact, the color swatch has the same color,
and in that crucial respect resembles, the paint that it represents to the
customer. This only works if the relevant visual resemblance is highlighted
in some way, has a status unlike that of the many other resemblances—that
is the point of taking exemplification of the relevant property, rather than
mere instantiation, as the vehicle by which representation is achieved.14
The relevant visual resemblance must be highlighted in some way in that
context, so as to bestow that status—here we have strayed from semantics
into pragmatics.
But do not equate even this beautifully articulated relationship with
picturing or imaging! As Goodman later pointed out, the word ‘‘word’’
both refers to and has the property of being a word, so it exemplifies that
property, but is not a picture or image. (Goodman 1987–8: 419)
Asymmetry of representation
There is an asymmetry in representation that resemblance does not have.
This is a much repeated point, made to show that resemblance is not the
right criterion for representation. Resemblance could not be the crucial
clue to representation, it is said, for even if representation did require
resemblance to its target, the target would then resemble its representation
but not represent it. While resemblance is indeed not the right criterion,
the argument from asymmetry is not all that strong.
18  : 
First of all, we do tend to use terms like ‘‘resemble’’ and ‘‘looks like’’ and
such asymmetrically in certain contexts, because the subject of a statement
tends to select the focus or contrast. Of Rosemary’s baby they said ‘‘He has
his father’s eyes’’. Hard to think of their saying about the father ‘‘He has his
baby’s eyes’’. The retort may be that literally the noticed resemblance goes
both ways; but ‘‘literally’’ may here just mean ‘‘if you ignore the context’’.
Probably we can find a context in which someone may, without oddity,
assert not that a given picture is an exact likeness of me but that I am its
exact likeness, or something to that effect. But that would not show that
resembles is symmetric in general, let alone in every context in which it
comes in.
That literally resemblance must go both ways—that literally speaking it
is both reflexive and symmetric, while representation is neither—is most
likely based on the simple idea that to resemble (in some respect) is to have
a property (of the pertinent kind) in common. But that is too simplistic
a construal anyway. The production of a photo involves a ‘collapsing’
of shades of color and of three-dimensional spatial structure into two
dimensions. If that counts as pertinent resemblance, then this is a relation
of homomorphism rather than isomorphism, yet central to modeling. That A is
a homomorphic image of B certainly does not entail that B is that of A.15
But this contrary point too is weaker than it seems. If A is a homomorphic
image of B then there is a reduction of B, modulo some equivalence
relation, to which A is isomorphic—so we might say that in this respect
B resembles A just as A resembles B. So there is no strong argument,
as far as I can see, based on any clear asymmetry to banish resemblance
from our topic, nor one to make it relevant to representation in general.
What does remain, as needs to be emphasized, is that certain modes or
forms of representation (but not all) do trade on selective (and not arbitrary)
resemblances for their effect, efficacy, and usefulness, and that this typically
goes in one direction only.
Resemblance in discord with representation

In examples of picturing, of visual representation, resemblance tends to
spring to the eye. Both reference and attribution can be achieved in other
ways, however, even in visual representation, rather than by means of
carefully selected resemblances. And conversely, even there, what is repres-
ented, and how it is represented, cannot in general be deduced simply by
 ,   19
attending to resemblances and non-resemblances. To show this, Goodman

mentions the painting of the Duke of Wellington which everyone agreed
resembled the Duke’s brother much better.
Socrates argues the point by means of a look at the extreme case, where
the resemblance is greatest, in his discussion with Cratylus about verbal
representation:
I quite agree with you [Cratylus] that words should as far as possible resemble
things, but I fear that this dragging in of resemblance . . . is a shabby thing, which
has to be supplemented by the mechanical aid of convention with a view to
correctness. (Cratylus 435c)
Socrates’ thought experiment leading to this remark has a quite contem-

porary ring, if we replace gods (as is usual now) with mad scientists. In his
discussion with Cratylus, correctness and accuracy are being characterized
in terms of greater resemblance:
[ . . . ] in pictures you may either give all the appropriate colors and figures, or
you may not give them all—some may be wanting—or there may be too many
or too much—may there not?
[ . . . ] And he who gives all gives a perfect picture or figure, and he who takes
away or adds also gives a picture or figure, but not a good one. (Cratylus 431c)
Here the ‘‘mere resemblance’’ view of representation appears on Socrates’

lips, somewhat surprisingly; but Socrates himself will soon put us right on
this. When Cratylus draws the parallel with language too simplemindedly,
the discussion immediately shifts into a subtler gear. Socrates replies, in
partial contradiction to the above:
[ . . . ] I should say rather that the image, if expressing in every point the entire
reality, would no longer be an image.
Let us suppose the existence of two objects. One of them shall be Cratylus, and
the other the image of Cratylus, and we will suppose, further, that some god makes
not only a representation such as a painter would make of your outward form and
color, but also creates an inward organization like yours, having the same warmth
and softness, and into this infuses motion, and soul, and mind, such as you have,
and in a word copies all your qualities, and places them by you in another form.
Would you say this was Cratylus and the image of Cratylus, or that there were
two Cratyluses?
Cratylus I should say there were two Cratyluses.
20  : 
Socrates Then you see, my friend, that we must find some other principle of truth
in images. . . . Do you not perceive that images are very far from having qualities
which are the exact counterpart of the realities which they represent? (Cratylus
432a–d)
That Cratylus does not grant that the copy, made by the god to duplicate
Cratylus entirely, is an image of Cratylus shows at the very least that
resemblance is not sufficient to make for representation. But the example
shows much more—we need to explore it in detail.
What’s in a photo?
Just to assert of something that it is a representation, or that it represents, or
that it represents something, is woefully elliptic and invites obscurity and
confusion. Our full locution must in the general case be at least of form ‘‘X
represents Y as F’’, as in ‘‘the caricature represents (depicts) Mrs. Thatcher
as draconian’’.16 Here X is a representation, Y its referent, and F a predicate
that X depicts Y as instantiating. The ‘‘as . . .’’ locution is fascinating, a
difficult topic in the analysis of language, and we will need to carefully
distinguish how its intensionality is connected with the relationality and
intentionality of representation. But this form too is still too brief to allow
all needed distinctions.
Why didn’t Cratylus agree that the god would have made an image of
him? The god would, in Socrates’ example, have made a perfect replica,
more perfect than any statue made by a human sculptor. The replica would
certainly, if properly displayed, have created the appearance of Cratylus’
being there. So as far as the later classification in the Sophist goes, the
god would have acted both as likeness-maker and as appearance-creator.
Yet Cratylus demurs: the god would not have made an image of him
but would have made ‘another Cratylus’. What are we to make of this
intuition?
A more contemporary example will show the inherent ambiguity in play
here about what represents what, with what, and to whom. Imagine: I have
acquired a famous photograph of the Eiffel Tower, Au Pont de l’Alma by
Doisneau. It hangs on my wall, but I scan it and print the scanned image.
This print is an image too—what does it represent? The Eiffel Tower seen
from the Pont de l’Alma, or the famous photograph?
 ,   21
There is no single immediately and obviously right answer; there couldn’t

be. It depends on what I do with the thing. If I send it to you from Paris
as a postcard, with the single note ‘‘Wish you were here!’’, then it is itself
a photo of the Eiffel Tower. If I insert it into the book I am writing about
photography, then it represents the famous photo by Doisneau. There are
still other possibilities.17 In other words, if it is an image of something at all
then what it is an image of depends on the use, on what I use it to represent.
So the question what does it represent? must in this case be taken as elliptic
for what is it being used to represent?
This term ‘‘use’’ can assimilate ‘‘make’’ and ‘‘take’’: the caricaturist made
the caricature to depict Mrs. Thatcher as draconian while I, seeing the
caricature, take it to depict Mrs. Thatcher as draconian, and display it to
you so as to depict her to you as draconian. There may also be a disparity:
the boy in William Golding’s Free Fall made drawings of hills and forests
but his teacher takes the drawings as pornographic. All of this falls under
‘‘use’’ in the general sense in which we say that in pragmatics we are meant
to study the relation not just of symbols to things, but the three-term
relation between symbol, user, and thing. Our fuller locution—shortened
in different ways depending on what is taken for granted in context—must
be ‘‘Z uses X to depict Y as F’’.18
I have put this in individualistic terms: you or I can use something
to represent something, though of course to communicate or convey
anything at all that way—be it factual information, feeling, intention,
or command—we must let each other in on how something is being
represented. (As Goodman put it, the most probative question is not what
is art? but when is art?) Since communication presupposes community to
some significant extent, this will be possible only in a context where some
modes of representation are already held and understood in common.
Within discussions in which the institutionalization of relevant symbolry
is already taken for granted, the point about use may be put in terms of
function or role. And although these terms make sense also for an individual
creation of a symbol, the role will generally be one specifiable within a
manifold of roles, a ‘system of representation’, a ‘language of art’. Thus
Ned Block writes
What any representation represents, and how it represents . . . depends on the
system of representation within which it functions. (Block 1983: 511)
22  : 
My point is not just that what represents what is relative to a system of represent-
ation. Rather my point is that you can’t tell for sure whether you are looking at
a representation at all just by looking. . . . One has to determine how the thing
functions. (Ibid.: 512)19
Relativity to systems, ‘languages’, recalls Goodman of course; and this
applies very well to pictures, as is well enough understood when different
styles of representation are studied.20
This notion of system, or of function if understood as a role in a
system, sounds still quite impersonal, but we must understand it in terms of
pragmatics, referring to contexts of use, broadly construed. The context-
sensitivity does not go away when (in different ways) Nelson Goodman
and Ned Block say that you need to know which system of representation the
item belongs to. For since the same item will in different contexts belong to
different such systems, we then need the relevant contextual factors which
determine that.
What is really to be emphasized here is the way in which individuals and
groups, though relying on some pre-existing communally understood form
of representation (a universal qualification of all communal activity), create
new representations and new modes of representation. When Descartes
created his method of coordinates, it is not as if he was just using an
already extant way of representing spatial shapes and motion. But it is
true that in his initiative, to use known numerical equations in this way,
he bestowed a role on already familiar equations that they had not had
before. Unlike a moment’s poetic depiction quickly lost to history, this
act engendered a mode of representation fundamental to all subsequent
science.
What is a representation then?

Look back now at Socrates, Cratylus, and the god they imagine. Did the
god make an image of Cratylus or did he not make a representation of
anything, but a clone? That depends. Cratylus was too hasty in his response!
Did this god go on to display what he made to the Olympic throng as a
perfect image of Greek manhood? Or did he display it as an example of
his prowess at creature-making? Or did he do neither, but press the replica
into personal service, since he couldn’t have Cratylus himself?
 ,   23
What is represented, and how it is represented, is not determined by

the colors, lines, shapes in the representing object alone. Whether or not
A represents B, and whether or not it represents the represented item as
C, depends largely, and sometimes only, on the way in which A is being
used. ‘‘Use’’ must here be understood to encompass many contextual
factors: the intention of the creator, the coding conventions extant in the
community, the way in which an audience or viewer takes it, the ways
in which the representing object is displayed, and so forth. To understand
representation we must therefore look to the practice of representing, to
how representation is a matter of use; and this involves attention first of
all to the users in a broad sense of ‘‘use’’. That is the main thing to be
concluded both from our discussion of caricature and misrepresentation,
and from the Cratylus and Eiffel Tower photo examples.
There is no representation except in the sense that some things are used, made,
or taken, to represent some things as thus or so.
I do not advocate a theory of representation, and this could not possibly be
offered as such since that would be circular. But if I did, I think this would
be its Hauptsatz.21
What does this exclude from the category of representations? That
depends of course on precisely what ‘‘used, made, or taken’’ means. And that
in turn depends on what is required if this Hauptsatz is to solve or dissolve
puzzles about representation. (For example, the puzzles that result if one
begins with the thought that resemblances—likenesses, correlations—will
determine, by themselves, what the representor represents.) What we can
conclude, at least, is that use, in the appropriate sense, must determine the
selection of likenesses and unlikenesses which may, in their different ways,
play a role in determining what the thing is a representation of, and how it
represents that. Moreover, the selection cannot be mute: in the pertinent
context, this selection and the precise role it plays, the selection must be
salient, so the use must be such as to highlight that selection.22
The use is what bestows the relevant role or function on the item
used. There are uses of the terms ‘‘use’’, ‘‘make’’, ‘‘take’’ which imply
no intentionality: the car uses gasoline, the tornado took the life of my
neighbor, the Ice Age glaciers made these valleys. But our puzzles about
what representation is do not disappear unless ‘‘use’’ and its cognates are
understood here in the sense in which they presuppose intentional activity.
24  : 
That said, I will just write ‘‘use’’ for use, make, or take understood in this
sense.
If that Hauptsatz is understood in this sense, then it places some immediate
limits on the range of representation. Firstly, at least if taken entirely literally,
it has no room for the notion of mental images or mental representations,
whether taken to be brain states or something more ephemeral—for no
such things, if they exist at all, are used or put to use, or taken in one way
or another.23 At least, not in the relevant sense: we can conceive of a
brain surgeon bestowing a representational role on the patients’ brain states,
but not of a person bestowing roles on his or her own brain states—or,
presumably, on whatever could count as mental states.24
Secondly, this conception leaves no room for ‘representation in nature’,
in the sense of ‘naturally produced’ representations that have nothing to
do with conscious or cognitive activity or communication. The Eleatic
Stranger gave a whole series of examples of copies, but it is not clear that
they are all images in the sense of representations:
Things in dreams, and appearances that arise by themselves during the day. They’re
shadows when darkness appears in firelight, and they’re reflections when a thing’s
own light and the light of something else come together around bright, smooth
surfaces and produce an appearance that looks the reverse of the way the thing
looks from straight ahead. [ . . . ] And what about human expertise? We say
housebuilding makes a house itself and drawing makes a different one, like a
human dream made for people who are awake. (Sophist 266c–d)
Most shadows and reflections that occur in nature are not being used or
taken, let alone made, by anyone to do anything. (The Balinese shadow
puppet theater is an exception.) So by our Hauptsatz, they are generally
not representations. Nor is the track left in sand by an ant in some desert
long, long ago in a galaxy far, far away. . . . not even if it has the shape of
our word ‘‘Coca Cola’’.25
A black mark on a rock does not refer, represent, or mean anything
unless it has a role, or has bestowed on it a role, in some practice—no
matter whether it is a simple stroke or a complex pattern. Nor is it sufficient
that it has the sort of shape, coloring, etc. that would place it in a certain
role if encountered or produced in a certain cultural context, by persons
belonging or assimilated there, if in fact it does not bear any relation to such
a context.
 ,   25
But a natural object can represent, just as it can play other roles, namely
if we bestow such a role on it. Imagine I am using a stone, found on the
ground, to hammer in a tent peg. I am using it as a hammer—it is my
hammer now, I have bestowed the hammer role on it. The hammers we
buy, in contrast, are manufactured precisely to play this role—they are
manufactured artifacts. The stone was not made for that, and it is not an
object that I created, constructed, or assembled. Nevertheless it is now a
physical object with a function—that is to say, an artifact. There is an
analogy here to ‘objets trouvés’, natural objects ‘made’, without physical
modification, into works of art. All of this applies mutatis mutandis when I
use the stone to represent, for example, a certain stateman’s heart: I bestow
a role on the stone for it to play, I give the stone a function for it to serve.
What is in a photo? What is in a picture? This question has the misleading
form of ‘‘What is in a box?’’ We won’t get much further by taking this
form at face value and giving an answer with the correlative form, such as
‘‘What is in the representation is its content’’. That is just a verbal answer,
conveying nothing by itself. To call an object a picture at all is to relate it to use.
As an analogous example we can think of Herbert Mead’s reflections
on the teacup (McCarthy 1984). If there were no people there would
be no teacups, even if there were teacup-shaped objects. For ‘‘there are
teacups’’ implies ‘‘there are things used to drink tea from’’ which in
turn implies ‘‘there are tea-drinkers’’. By ignoring the contextuality of
representation, the fact that we are dealing with about-ness, and that what
the representation is about is a function of its use, we could land ourselves
in useless metaphysical byways. If we were to ask ‘‘What is in a picture?’’
while taking the picture simply to be the physical object and with no
relation to anything that can bestow meaning, the answer would have to
be ‘‘Nothing!’’
The notion of use, the emphasis on the pragmatics rather than syntax
or semantics of representation in general, I will give pride of place in
the understanding of scientific representation. But does that exclude too
much? That a particular person at a specific time uses or takes or presents
something to represent something else is a very local event. Could it really
be a general condition on representation that something so specific has to
happen? In a comment on similar ‘‘intentional’’ views of what constitutes
representation, Mauricio Suarez suggests that it will hamstring the idea that
theories represent:
26  : 
on the intentional account of representation a theory cannot ‘represent’ a phe-

nomenon that hasn’t been observed. For a theory cannot be intended for a
phenomenon that hasn’t yet been established. (Suarez 1999: 82)26
The objection, if valid, would not just apply to a theory or a model, it

would imply that nothing can be intended (let alone used) to represent
something that has not entered our acquaintance, or something that we do
not know to exist.
The objection is presumably not that there can’t be a representation of
something that we have not already encountered. A meteorological model,
found for example on a weather forecasting website, does in fact provide us
with a representation—more or less accurate, or not accurate at all—of the
weather in the next five or ten days. Is the objection then that we cannot
be said to use or take or present this meteorological model to represent the
coming week’s weather? But we do use it, and the viewers so take it. So
could the objection rather be that this model could have been or provided
such a representation although it could not have been—to use Suarez’s
exact words—intended for the actual meteorological phenomena, which are
still in the future? It does not seem so, it seems that it was intended precisely
to represent those (as yet unknown) phenomena.
Relation, intention, intension

Representation is a relational notion, if we go by the form of assertions that
attribute this status. ‘‘Tragedy is a representation of an action. . . .’’ Here is
a painting of a picnic on the grass, the statue over there represents justice,
that graph depicts the growth of a bacteria colony. In each case we have a
subject term, a relational predicate, and a term for the second relatum.
But we are not dealing with something as simple as a relation of physical
contact or impact or proximity. First of all, the second relatum may not
be real. In fact, to say that something is a painting of a picnic does not at
all imply that there is a real picnic which that painting depicts. ‘‘The Mona
Lisa is a portrait of Mary Magdalene’’: this assertion purports to mention
two real things and a relationship between them. Perhaps it does. But the
important point is that the form by itself cannot reveal this. For the assertion
may be true still if it turns out that Mary Magdalene was not a real historical
character. Even if she was, the Mona Lisa could at best depict how Da Vinci
imagined her to be, which we can’t necessarily equate with depicting her.27
 ,   27
The ‘‘of’’ that marks the relation of representing object to what is

represented is like the one familiar from Brentano’s characterization of
mental acts as ‘directed’, intentional. Intentionality we also see in semantic
discourse when we say that ‘‘Zeus’’ is the name of a god, for example.
Representation is intentional in the sense of relating to epistemic intention,
in the sense of being about something, in just the way that reference (by
someone) and predication (by someone) are. But just as thought can be
directed in this sense at what is not present, not experienced, not known, or
even non-existent, so can any use of something to represent something. By
so using it, the user bestows a role, the role of representing such and such
as thus or so. If for example I draw a graph and present it as representing
the rate of bacterial growth under certain conditions, then by virtue of
that very act, what the graph represents is the bacterial growth rate under
those conditions—period. It is equally apt to say that I represented that
growth rate as thus or so—and it would be apt to say that if, instead of
drawing the graph, I had displayed the equation of which the graphed
function is a solution. And so forth, mutatis mutandis, for the case in which
I display a function that has such graphs as output for inputs about ambient
conditions of bacteria colonies, or state a theory that describes a family of
such functions with a further free variable for the type of bacteria. . . .
Given this intentionality it is perhaps not surprising that in the case of
representation, the relations can change with context of use. The very
same object or shape can be used to represent different things in different
contexts, and in other contexts not represent at all.28
The expression ‘‘A represents B’’ when used all by itself is misleading.
It is easy to get into confusions when the relational character of a term is
suppressed. To illustrate: every woman is a daughter and every daughter is
a woman, so why is being a daughter not the same as being a woman? Precisely
because to be a daughter is to be daughter of someone. Analogously,
to represent something is to represent something as thus or so. The
complexities appear in force when we extend these assertions to the three-
place relation ‘‘A represents B as C’’. Simone de Beauvoir depicted herself
as a dutiful daughter, but not as a dutiful woman.29 All and only creatures
with hearts are creatures with kidneys—yet to represent something as
having a heart is not the same as representing it as having kidneys. And so
forth. This ‘opacity’, the resistance to substitutivity of identity, is a mark
not only of the intentionality of thought, but of intensionality in discourse.30
28  : 
We see this in modal contexts: 9 = the number of planets, but it does

not follow that the number of planets is necessarily greater than 7. Also in
oratio obliqua: the mathematics teacher who taught us that 7 < 9 did not
tell us that 7 is smaller than the number of planets.31 Representation of an
object as the evening star is an activity that is intentional, in the way that
mental acts are traditionally said to be, precisely because to do so is not the
same as representing it as the morning star—even though that is the very
same object.
So assertions to the effect that something represents, are intensional. This
is primarily a point about language, but is closely related to the point that
representation itself (the activity) is intentional, both in Brentano’s sense and
in the common sense of the term.
Ordinary discourse does not mark the distinctions we are making here, or
not very well. In analytic philosophy language has been regimented to some
extent to do so. Thus ‘‘He said of Mrs. Thatcher that she was draconian’’
asserts a relation of the speaker to the real Mrs. Thatcher, while ‘‘He
said that Mrs. Thatcher was draconian’’ does not, in this regimented form
of discourse. If Mrs. Thatcher was the Prime Minister, the first sentence
implies ‘‘He said of the Prime Minister that she was draconian’’ but the
second does not imply that. The role here given to the ‘‘of’’ locution marks
an artificial verbal distinction (even if not without roots in prior general
usage), but such artifice can be useful when confusion threatens.32
Where do the intensionality and intentionality come from? To under-
stand this is as important for representation in science as in the arts. The
answer lies of course in our Hauptsatz, that there is no representation except
in the sense that some things are used, made, or taken, to represent some things
as thus or so. But even this does not suffice by itself, for it does not make
explicit what all is involved in the use, by way of value, purpose, aim,
and yes, intention. Ronald Giere spells this out concisely for scientific
representation and one further contextual factor, purpose:
If we think of representation as a relationship, it should be a relationship with
more than two components. One component should be the agents, the scientists
who do the representing. Because scientists are intentional agents with goals and
purposes I propose explicitly to provide a space for purposes in my understanding
of representational practices in science. So we are looking at a relationship with
roughly the following form: S uses X to represent W for purposes P. (Giere
2006: 60)
 ,   29
But the point is quite general. The spelling out can only go so far, because
the notion of representing has (suffers from?) variable polyadicity: for every
such specification we add there will be another one.
Appearance to the intellect: illumination as embedding

We have been concentrating mainly on appearance to the senses, but when
viewing a painting or movie, reading a story, or watching a computer
simulation, we may well ask ‘‘I can see what is happening, but what
is really going on?’’ Quite often however we do not even get to this
question: our active, agile imaginations have already supplied, assumed,
or conjectured a pattern behind the displayed events. On other occasions
we press to find or construct a model in which the random or puzzling
appearance is sublimated in a well-behaved structure, of which it is the
surface. In such a case we arrive at a representation of the original as
embedded in a larger whole, and thereby made to satisfy certain demands
of the intellect.
There certainly are examples of this in the visual arts. A picture of an
apparently levitating woman may be immediately recognizable as of the
Virgin Mary because the figure’s feet are on a crescent moon and she is
surrounded by jubilant angels. The embedding structure presents the story
in which the event is embedded. But something similar can be said of the
event depicted in Jacques-Louis David’s Oath of the Horatii, though it does
not draw on such pervasively common knowledge. The lamentations of
the seated women in the background show how much more is going on
than the event depicted taken in itself; they help to give meaning to what
is happening through an embedding in a larger story.33
But the most straightforward examples come from the history of mod-
eling in mechanics. Think for a moment of a swinging pendulum.
Determinism requires that if the same state occurs at different times, it
is always followed by the same succeeding states. At its lowest point, the
bob has its maximum speed—but location and speed can be the same at
two different moments, and the location a moment later be different. If
we keep only those two factors in our description, we have an apparent
violation of determinism. The remedy is simple: enlarge the description by
replacing speed by velocity, which is speed + direction. The velocity at
30  : 
the lowest point is not always the same; when it is, so are the immediately
succeeding locations.
All these features are still in the domain of kinematics. Descartes’s
famous aim for mechanics was that it should be deterministic but only
have ‘quantities of extension’, that is, kinematic parameters. When Leibniz
and Newton each came up with counter-examples, they also introduced
new, non-kinematic (dynamic) parameters to ‘‘fill out the picture’’. The
apparently indeterministic kinematic behavior is embedded in a model
that has additional parameters—such as masses and forces—which is
deterministic after all.
These examples provide the pattern for ‘hidden variable’ interpretations
of apparent indeterminism. Reichenbach argued that such added parameters
might not correspond to anything real, and that physics could forego
satisfying the demand for determinism. But as a good empiricist, he offered
this as methodological advice: neither demand them nor ban them from
modeling. The touch stone would be the usefulness of such models for
empirical prediction.
For us, the point here is simply that, for a given purpose, the best repres-
entation might well be one that embeds its target in a larger structure. So we
can add addition of ‘surplus structure’ to distortion and the trading on selective
unlikenesses, to our catalogue of means for representation achieved by
departures from mimesis. Hence, again, there is no universally valid infer-
ence from what the best representation is like to what the represented is like.
In conclusion
A scientific, technical, or artistic representation is an artifact. As such, it is
both an object or event in nature, that we can regard purely through the
physicist’s or chemist’s or mathematician’s eyes. But it is at the same time
something constituted as a cultural object, through its role or function,
bestowed upon it in practice. Just what the representation is, or what
is represented and how, is not determined entirely—and often enough,
hardly at all—either by what is ‘in’ the natural object or by its physical or
structural relations to other things.
When resemblance is the vehicle of representation, for example, the
representation relation derives from selective resemblances and selective
 ,   31
non-resemblance, but what the selections are must be somehow high-

lighted. If the selection or the highlighting is indicated by signs placed
in the artifact itself, these too need to be meaningful to play their role,
and so the task of identification is pushed back but reappears as essentially
unchanged.
Thus what determines the representation relationship, with all its poly-
adicity, can at best be a relation of what is in it to factors neither in the
artifact itself nor in what is being represented. In the examples and puzzles here
examined, the extra factors characterized use, practice, and context, and these
form the proper basis for generalization there. That is not the end of the
matter, representation is not to be subjected to definition: it is inexhaustible
as a subject.
2
Imaging, Picturing, and Scaling
Despite the arguments against taking resemblance to be the clue to

representation, there are many cases where what the representation refers
to, or what it attributes, is conveyed effectively by displaying a salient
resemblance.1 Examples come readily from the visual and plastic arts,
but there are relevant representational techniques commonly used in the
sciences as well.
Resemblance comes in, not when we are answering the question What is
representation?, but rather when we address How does this or that representation
represent, and how does it succeed? In the case of representation by pictures,
scale models, diagrams, and maps, and many other examples, the initial
answer to the latter is By selective resemblance and selective (even systematic)
non-resemblance. To fix terminology that I have used informally so far let us
reserve the terms imaging and imagery for those cases.
Modes of representation
Everything resembles everything else in many ways, so an effective use
of resemblance must always be selective. This can only be effective if
the selection itself is understood or conveyed. That point applies more
generally: even if resemblance is not the vehicle, whatever features of the
representing entity are instrumental to the representing must be somehow
highlighted there. It is not sufficient for the representor just to have them!
Secondly, even if we recognize a role played by a resemblance, this
resemblance need not be with respect to any visually or even perceptually
detectable features. Additionally, resemblance need not consist in sameness
of properties, but can also be at higher levels. This was a theme emphasized
and elaborated especially by Wilfrid Sellars in his discussion of theories
34  : 
and models.2 Resemblance may consist in having properties in common,

or instead in having properties that have properties in common with
relevant properties in what is represented. That is, the representor may
have properties which form a structure resembling a structure formed by
the properties of the represented, and so on up the hierarchy of types.
We should honor these distinct categories with distinct names. Rep-
resentation in general has under it the special case of representation that
trades for its success on some (specific) resemblance, or on multiple
resemblances. For this special case I will use the terms image, imaging, and
imagery—which in turn has under it the sort of imaging that manages to
represent in part by trading on some visually detectable resemblance—visual
imagery.3
When Galileo introduced his primary qualities as the properties to be
solely considered as primitive in scientific description, he both narrowed
and extrapolated from this base of visually detectable resemblances. The
list that became more or less standard in his century comprised just the
quantities of spatial and temporal extension and their combinations—hence
of space, time, and motion, the kinematic quantities—but were not qualified
by the limitations of human perception. So let us set imaging which manages
to represent in part by trading on resemblances with respect to kinematic
quantities side by side with visual imagery. Call it kinematic imagery. Trading
on resemblance is a very broad category, so we may come across other sorts
still besides visual and kinematic imagery. These as well may admit of both
simple and ‘higher level’ (structural) varieties.
Overlapping these categories of representation that trade on selective
resemblances lies still a further salient case, which shares some crucial
features found in visual perspective, a development which in art we
associate specifically with Renaissance painting. Perspective involves (as
we shall explore further below) such features as occlusion, marginal distortion,
texture-fading. For cases of imagery in which such features of perspective
are present I’ll use the terms picture and picturing—these can include cases
of kinematic and visual but perhaps also still further forms of imagery.
Obviously none of these categories are hereby precisely defined, we will
have to explore all of them further. But there are two caveats we must
emphasize. The first, already noted several times now, relates to the level
of resemblance, and the second to the reality or non-reality of what is
represented.
, ,   35
Resemblance, as I said, can be higher order: the spatial structure of a set

of letters on a page may be the same as the temporal structure of a set of
events named by those letters. The use of visual or kinematic imagery to
depict things that are not visual or kinematic is rife, and not excluded by
our notion of imagery. For the resemblance of some structure to a visually
recognizable structure may be precisely on the level of structure, not on
the level of features that only visible things can have.
This point may threaten to trivialize the notion of imagery. For it does
not take much by way of intellectual gymnastics to find some minimal
relevant resemblance in any two things at all. But the threat does not seem
to me very serious, for actualizing it would quickly turn into something
no one could really take seriously.4 Does the word ‘‘animal’’ resemble
anything that is not a word, does it resemble anything having to do with
animals? Perhaps so, but it will not be anything that plays a role in the
representation of animals by that word. The Cratylus’ attempt to see verbal
representation as hinging on resemblances of not just words but syllables
and even letters to what is represented, I take to be a choice bit of Socratic
irony, that reduces the entire idea to absurdity.
Second caveat: a representation may be of something non-existent,
non-actual. So a representation that trades on resemblance may in fact not
resemble any real thing, any person living or dead, or actually occurring
event. Le Déjeuner sur l’herbe is immediately seen as a picture of a picnic
because of a certain kind of resemblance to real scenes of the picnic-kind,
but it is still a painting of something that did not happen.5 The Judgment of
Paris, whether by Rubens, Renoir, or as Nazi propaganda by Ivo Saliger,
trades for its success on resemblances. Not resemblance to Paris or to the
goddesses, who may not exist, nor even to any real scene that ever happened,
but on resemblances to human judgments, bodies, clothes, and the settings
of such human scenes. So to call something an image, a visual image, or a
picture will not imply anything about the reality of what it depicts.6
This is a point obviously of some importance to an empiricist view of
scientific modeling as representation. That an image trades on resemblance,
on any level, does not imply that it resembles what it represents, nor that
there is something that it resembles, nor even that there exists something
that it represents. Fundamental to the understanding of representation in all
contexts is this fact, that images which represent something unreal have their
importance, their role, their effect in the context in which they function.
36  : 
What distinguishes a picture?

At first blush, it is easy to see what is in a picture—such as that familiar
one, of a picnic, in curiously dishabillé condition. What is in Le Déjeuner sur
l’herbe? A picnic. But how does that distinguish the picture from a verbal
expression or icon? After all, what does the phrase Le Déjeuner sur l’herbe
describe? A picnic. Despite the sameness in answer to these questions, there
is a difference.
As I have regimented the terms, above, picturing is a case of imagery,
that is, representation that trades on resemblance, distinguished by bearing
hallmarks of perspective. So it is the latter we should now try to isolate.
One popular response is that a picture can’t help but be specific. A
sentence that says that George is in the garden need have no information
about whether he is standing, sitting, or lying down. But a picture of
George in the garden, it is said, can’t very well be neutral on that.
Dominic Lopes calls this a ‘‘myth of specificity’’.7 In fact a picture need
not be determinate in any particular way. A picture of a written word might
not be of any specific word; the word might be illegible in the picture. A
picture of George in the garden might just show his hand emerging out
of some shrubbery, without revealing his posture. A few brush strokes can
suffice to depict George as in the garden, in tears, or in love.
But there are two ways in which pictures are peculiar in the way they
represent. Both ways have to do with what is accessible to vision. The first
relates to resemblance, the second does not.
First of all, as Dominic Lopes emphasizes: pictures are unlike other sorts
of representation precisely because there is a way in which they do literally
‘‘look like’’ what they depict. Pictures are physical objects; among their
physical properties there is a privileged set of visual properties, those by
which pictures represent their subjects—such as line, shading, color, and
visible texture. But note how this point is to be qualified. The line, shading,
color, and texture in the picture may not match, and in fact generally must
not match, those of the represented scene. This is essentially the same point
as the Eleatic Stranger’s about large statues on high pedestals: to create
a life-likeness, to create the right appearance, distortion is needed. That
applies to color as well as shape or proportion. The technique of chiaroscuro
was accordingly developed in the Renaissance along with perspective. This
, ,   37
is only the beginning of the systematic mis-matching necessary in a picture

to make it ‘look like’ the original.
In addition, as we have seen, on many counts the resemblances and
non-resemblances can easily give wrong clues both to what is depicted and
what is attributed to it. A picture can misrepresent its subject, attributing
properties to it that it does not have, whether by accident or on purpose.
To that extent the idea that pictures represent because they ‘‘just look like’’
their subjects is indeed a mistake. All representation is selective and the
selectivity is crucial to what is depicted, but for pictures, the selection is
subject to very specific constraints. The selection is not a choice simply to
render some aspects and be non-committal as to others. That sort of choice
is also made when we describe something in words.
The crucial difference—here we come to the second distinguishing
point—appears when we notice that in picturing, the selection of one
aspect may force the picture not to include certain others. There is first of all
occlusion, which is closely related to perspective: to depict a situation from
a given point entails that some objects will be in front of, and hence hide,
certain other objects. Thus revealing what things are like from one angle
is incompatible with simultaneously revealing the values of certain other
parameters. This point has been exploited well in technical description of
painting. But the topic has also been explored as crucial to machine design
and drawing (cf. Rothbart 2003: 242–4). John Hyman introduced the
terms ‘‘occlusion shape’’ and ‘‘relative occlusion sizes’’: an occlusion shape of
an object is its outline, relative to a line of sight.8 It is the shape of an object
as seen from a particular point of view. The occlusion size of an object is
the area that the object occludes from view from a particular viewpoint. It
depends on the actual size of the object and its distance from the observer.
As the observer moves, the occlusion shapes and the occlusion sizes of the
objects around him change. But this is not something subjective. These
terms concern the shapes and sizes that are projected from a particular
viewpoint on a plane perpendicular to the line of sight. As Hyman points
out, we may be mistaken about them and our mistake ‘‘can be corrected
by measurement and geometrical calculation’’.
Dominic Lopes develops this notion so as to distinguish picturing from
other modes of representation. Revealing what things are like from one
angle is incompatible with simultaneously revealing the values of certain
other parameters. Besides ‘‘occlusion’’ there are also:
38  : 
Grain: more distant objects, textures are not as finely depicted as

near; and in fact there is a minimum to the fine-grainedness of a given
picture.
Angle: even with multiple views, there is a limit to the number of
angles and distances from which an object can be depicted
Marginal distortion: this derives from the limitation of the ‘view’,
which can take in only a circumscribed range, and picturing reaches its
limits of reproduction near the limits of that range. Marginal distortion
is clearest in the pinhole camera picture—that ordinary photos do not
show this, is precisely because the camera, like the Sophist’s sculpture
of large statues, adjusts the ‘copy’ so as to create a more ‘faithful’
appearance.
Angle and marginal distortion are topics that belong under the heading
of specifically spatial perspective, and we will discuss that further below.
These characteristics are, for example, the ones drawn on in Ronald Giere’s
account of astronomical observation, to argue that what is gained from the
instruments is perspectival.9 All four of the characteristics mentioned are
plausibly grouped under the heading of perspectivity, but we cannot hope
for an explicit definition here. The notion of perspective is undoubtedly
what Wittgenstein called a cluster-concept, so that no specific set of
characteristics is both necessary and sufficient for its application.10
Lopes introduces an important further characteristic to the cluster. He
adapts here some terminology from Ned Block and Daniel Dennett.11 A
representation is committal with respect to some property F if it represents
its subject as having that property, also if it represents it as not having that
property, and not otherwise. If it simply does not go into the matter of
whether the subject has that property, then it is inexplicitly non-committal
with respect to F. But finally, a representation is explicitly non-committal with
regard to this property if it represents its subject as having some property (or
properties) that preclude the representation from being committal in that
respect. It is crucial to the notion of picturing that being committal in one
respect will preclude being committal in some other respect—in the sense
that it will force being explicitly non-committal. This point elaborates on
perspectivity: perspectives can be ‘‘ ‘aufgehoben’ in a higher unity’’, if I may
use an expression from a much earlier time, but they cannot be simply
combined as parts of a third perspective.
, ,   39
Does this allow us to differentiate picturing from describing? Given a

particular style of representation, it may not be possible to add more to a
picture and still let it remain a single picture as opposed to a gallery. But
what about collage? And what about cubism? Or ‘impossible’ pictures, like
Escher’s? Lopes maintains, and this seems correct, that although different
styles differ in precisely what they make possible in this respect, the same
point about the explicitly non-committal will apply mutatis mutandis. There
is still in each case a choice to represent the subject as thus or so, and this
precludes representing it as having certain other properties, which could
have been selected for depiction in another picture in the same style.12
What is still not obvious, in the catalogue of characteristics we have now
gone through, is that they capture the notion of perspective. When we
think of a picture as being drawn from one point of view (the location
of the eye and direction of vision), we are attending to its alternatives:
thinking of it as set in a ‘horizon’ of other perspectives on the same objects.
Occlusion is connected with this only if we have a sense that it can be
varied so that other objects come to light, other objects are occluded.
Similarly with what the picture is non-committal about: this is connected
with perspective only if we can imagine a shift in what is excluded and
included. It would be a mistake to concentrate on what is actually in a
picture, taken by itself, if we want to say what it is for an image to be
perspectival.13
Thus, as to the hallmarks of perspective: the characteristics listed will
not suffice if applied ‘piecemeal’. The content of the picture must be
related to a ‘horizon’ of alternatives that we can think of as coming from
‘different points of view’, if these characteristics are to count as marks of
perspectivity—and the explicit or implicit reference to such a horizons of
alternatives is what is most important in the concept of perspectivity.14
Mathematical imagery, distortion

through abstraction
Visual imagery and kinematic imagery are so-called because of the category
of features with respect to which they trade on selective resemblance.
Mathematical imagery, on the other hand, is so-called first of all because
it is imagery—i.e. representation trading on selective resemblance—and
40  : 
secondly because the representor is a mathematical object. While there is

no implication therein of perspectivity, mathematical imaging too involves
in general necessary or inevitable distortion, in both simple and subtle ways.
Of course the story is apocryphal, that a professional gambler funded
a mathematician to analyze horse-racing, and was thoroughly unhappy
with the report which began ‘‘Let each horse be a perfect sphere, rolling
along a Euclidean straight line . . .’’. But is that so far from real examples
of mathematical modeling? Consider the example in dimensional analysis
used to model the motion of a cloud of small, electrically charged oil
droplets in air under the influence of an electric and a gravitational
field—reminiscent of Millikan’s famous experiment to measure the charge
of the electron—which begins with the simplifying assumptions
• the oil in the droplets is in thermodynamic equilibrium with the oil
vapor in the air, and no further evaporation or condensation occurs
• the oil droplets are so small that surface tension effects dominate the
distortion of the droplets by the forces acting on it
• the electric charge is distributed with spherical symmetry over each
droplet
and so forth, such as that the droplets’ acceleration is negligible. . . . If we
are to understand mathematical modeling in general, we had better see such
simplifying assumptions—granted to be most likely false in fact—as the
norm. The question, though, is whether this is just a matter of human limit-
ations, inessential to mathematical modeling as such, or whether distortion
of any sort is inevitable in principle.
Mathematical statuary
It will have been obvious that the criteria for distinguishing pictures from
other visual representations do not imply two-dimensionality. The word
‘‘picture’’ no doubt connotes, in common usage, representation on a
plane surface. But a statue, as we saw at once in the Sophist example, is
subject to the sort of distortion practiced in painting to produce a visually
faithful image. There is necessary occlusion and the statue is explicitly
non-committal with respect to certain features that we don’t even think
about in the case of painting. That is easily seen when we compare, say,
the Venus of Milo with the Anatomical Venus in Florence’s Museo della
Specola.
, ,   41
If we are to bring these concepts to the study of scientific representation

we must look to how they can be applied more generally beyond painting,
drawing, photography, holography, or sculpture.
Descartes’s analytic geometry, Newton’s and Leibniz’s differential and
integral calculus, and the subsequent developments in descriptive geometry
and analysis provide, on an abstract level, resources for representation so
perfect that they tend to engender oblivion to the distortions on which
they trade—and oblivion as well to the necessary sacrifices of perfection in
practice. To counteract this, let us begin with a Cartesian dream of abstract
perfection, and then consider how abstraction itself blurs the real.
When Galileo said that the Book of Nature is written in the language of
mathematics, he was referring to geometry and geometric figures; shortly
afterward Descartes founded analytic geometry in which these figures can
be equally represented by numerical functions. That was an enormous
step, in which our spatially structured world came to be represented
algebraically—and one might equally say that almost all its qualities were
so to speak ‘spatialized’. Let me illustrate this by introducing the notion of
a mathematical statue.15
Here is a man, we’ll call him Kurtz, who stands at the precise intersection
of the Equator and the Greenwich meridian. He is of a certain height, no
more than 2 meters from head to toe. Our task: to construct a statue of
this man, as accurately as possible with respect to size and shape. No plaster
concoction will do him great justice. Let us define a function K as follows.
Its domain is the set of triples of real numbers x,y,z. The value of K
equals always 1 or 0, with this condition:
K(x,y,z) = 1 if and only if Kurtz’s body occupies a region including
latitude x, longitude y and distance z meters from the center of the
Earth.
This function has value 1 on a region that precisely fits Kurtz’s body. In
analytic geometry, this function describes a solid, three-dimensional figure,
and indeed, within analytic geometry there is not much difference between
figure and function. I offer this to you as a statue, invisible to be sure, but
more accurate with respect to size and shape than any plaster or bronze
could be.
Such mathematical statues are the objects on which the new scientists
of the modern era practiced their craft. Of course they are much more
42  : 
versatile than I have indicated yet.16 Thermodynamic study lets in a fourth

parameter:
K(x,y,z,T) = 1 if and only if Kurtz’s body occupies a region including
latitude x, longitude y and distance z meters from the center of the
Earth, and T is the temperature in degrees Kelvin at that point.
No Earthly museum contains a statue with this internal temperature
correspondence to Mr. Kurtz! Kinematics lets in still more:
K(x,y,z,T,t) = 1 if and only if at time t, Kurtz’s body occupies a
region including latitude x, longitude y and distance z meters from
the center of the Earth and T is the temperature in degrees Kelvin at
that point.
With time come trajectories; with Newton we add in masses and forces.
Now the mathematical statuary can be thought of as figures in, or functions
defined on, higher dimensional spaces—configuration spaces and phase
spaces. . . .
Trouble at the interface

Of course we were idealizing! Who would think that there is an objective,
sharp division between the geometric points inside and those outside a
human body?
But this idealizing fiction is the be-all and end-all of the seventeenth-
century geometric representation of nature, continued in the next several
centuries of rational mechanics.17 Yet it also leads inevitably to its own
limits, where retrenchment from the idealization becomes imperative.
Indeed, this process in which deliberate idealization brings us to its own
limits, and thereby defines a new problematic for the scientist, imparts a
new impetus for scientific progress.
As illustration let us look back to geometric optics in its simplest form.18
There light is treated as a set of rays, emanating from a source, propagating
through transparent media according to three simple principles:
• the law of rectilinear propagation, that light rays propagating through a
homogeneous transparent medium propagate in straight lines
• the law of reflection, which governs the rebound of light rays from
reflecting surfaces
, ,   43
• the law of refraction, concerning the behavior of light rays as they

traverse a sharp boundary between two different transparent media
(e.g., air and glass).
Hero of Alexandria established the law of reflection on the basis of a
principle of economy in nature: that light will always follow the shortest
path as it moves from surface to surface.19
Proposition . If the light is unobstructed, it will travel in a straight
line.20
Proposition . (Law of reflection) If light is reflected from a surface
(such as water or a mirror), the angle of reflection will equal the angle
of incidence.
By convention, the angle of incidence is taken to be the angle between the
ray and the normal, i.e. the line perpendicular to the surface at the point of
incidence; similarly for the angle of reflection.
Hero’s principle of economy, in which economy of action is identified
as following the shortest path, is fine when the light is traveling through
the same homogeneous medium all the way. When the ray travels through
different media, say air and glass or air and water, it will follow the shortest
path in each but change direction when it moves from one into another.
This refraction depends on the density of the media. Suppose light strikes a
water/air surface at an angle. Again we draw the normal, i.e. the perpendicu-
lar line at the point of incidence. The empirically ascertained findings were:
The light entering the denser medium is refracted toward the normal;
if entering from the denser medium into the less dense, it is refracted
away from the normal, to the same extent.
Ptolemy, in the second century  treated this phenomenon systematically,
but his work was lost to the Middle Ages until it became available, via
Arab scholars, in the twelfth century. The Arab mathematician Alhazen also
discussed refraction systematically and stated the above ca. 1100. The quant-
itative description we are about to present was found in the seventeenth
century.
Let R be chosen so as to be as far from the point of incidence as Q.
Draw perpendiculars from Q and R to the surface (assumed flat), to meet
that surface in Q and R .
44  : 
Proposition . The ratio Q P:PR is a constant, independent of

the location of P, and depending only on the nature of the two
media—their Refractive Index.
This Proposition is called Snell’s (or Snel’s) law after its discoverer in the
seventeenth century.21
Figure 2.1. Reflection and refraction
But now let us look at a range of phenomena that lies just where
reflection and refraction compete. (Today it is easy to see this illustrated
on the internet by computer simulation.22 ) Light is refracted if it strikes
the surface at a shallow enough angle; but it is reflected, if it arrives at a
sufficiently steep angle. What happens when a light source is moved so as to
change the angle of incidence? Precisely where does the one phenomenon
end, or the other begin?
In the diagram, let QP equal PR, but consider various values for the
angle QPQ . Let Q move downward toward the water surface; the angle of
refraction away from the normal becomes larger, R moves up and to the left
in our diagram. But noticing that the distance Q P is always larger than R P,
what happens when Q and Q coincide? What happens then as the source
moves still further? The only answer within this theory is that the light is at
some point (at the critical angle) no longer refracted but completely reflected,
, ,   45
and that when that happens there is suddenly a big jump to a distance
below the surface—a singularity, a discontinuity in nature! Perhaps even a
contradiction in the theory given the traditional principle accepted, at the
time when Willebrord Snel formulated his law, that ‘nature makes no leaps’.
Is there really such an enormous discontinuity in nature at this point? At
the level of observation open to a swimmer or fish, this phenomenon can
be found. But extrapolation to what happens ‘in the small’ is not valid; this
is just the point where the model gives out, where the idealization reaches
its limit of admissibility. The phenomena for which geometric optics works
do include the more easily observable ones studied early in the history of
optics. But there is no infinitely thin precise demarcation between water
and air, and in any case, ignoring the wave character of light will only yield
adequate results even for the observable phenomena in a limited range.
Infinitely perfectible idealization?

The assumption of continuity in all natural processes is no longer in force.
That does not nullify the lesson illustrated by the above ‘trouble at the
interface’, however. Agreed, we cannot demonstrate that in principle, as a
matter of logic, mathematical modeling must inevitably be a distortion of
what is modeled, although models actually constructed cannot have the
perfection reachable in principle. But on the other hand, the conviction
that perfect modeling is possible in principle—what Paul Teller calls the
‘‘perfect model model’’—does not have an a priori justification either!23
One conviction which supports the ‘perfect model model’ is that
however vague our ordinary language is, there is absolutely no vagueness
in mathematics. This support too loses its plausibility, however, if we look
not just to pure mathematics but to assertions made by way of application.
There the fascinatingly creative changes in mathematics itself belie the
idea. Since the scientific description of the world is couched throughout
in mathematical language, we can put it this way: the scientific image
itself harbors vagueness and ambiguity, at each historical moment of its
development—but this only comes to light in retrospect.
Consider this beer glass on the table: each has a shape. What that shape is,
precisely, we do not know. In the heady early days of the mathematization
of our world picture it could be assumed that this shape is [described
by] a precise function of the spatial coordinates. The edge of this table
could be thought of as a straight line, hence [described by] a function
46  : 
of form y = ax + b. But of course, the edge of a table is not perfectly

straight. . . . If eventually the table, the beer glass, and their environment
are re-conceived as assemblies of classical particles, they still occupy precise
regions of space. These regions can be similarly represented by functions
on the spatial coordinates. It may be a bit arbitrary exactly which particle
assembly is the glass at any given time—but upon any such arbitrary,
admissible choice, the table and glass have a definite shape.24 So now these
objects can be represented by a mathematical model in the same way but
more accurately than before —though now definitely only ‘in principle’, not
in practice! The shape is accurately modeled by a suitable mathematical
function; what that function is, we do not know, but there must be such
a function.
We are speaking here of the continuum of classical mathematics which
has equal use for the representation of each primary quality: length,
duration, shape, size, number, mass, velocity, what have you. The equation
of the primary quality shape with geometric shape is in effect the assertion
that a certain representation is completely adequate. But now we must ask:
what exactly is this representation? Not only the question as to what shape
the glass has, but that question is continually answered differently.
In the nineteenth century, mathematics developed to the point where
it was sensible to ask: is this shape an analytic function?25 Or is it only
‘smooth’, i.e. infinitely differentiable? There is no question but that, as a
reconstruction of the world picture of Galileo, Descartes, and Newton, we
can choose either option. They had not said that every physical magnitude
in nature is an analytic function, but they had not conceived of any
alternative. Nothing would have been lost from the subject as developed
at that time if we thought of the functions then discussed in this way,
the functions describing the primary qualities of real physical things, as all
analytic—nor if we thought of them as not necessarily analytic. Nor is
there any kind of experimental evidence to cite in favor of one or the other
option. The description was open, indeterminate in that respect.
To see how far such nineteenth-century questions are beyond those that
arose in the seventeenth century, reflect on what Descartes created when
he created analytic geometry. When Pascal took issue with Descartes, it
was because he felt the need for the existence of points given only as limits
of infinite sequences, while Descartes was willing, within mathematics, to
countenance only finitary constructions.26 By the time the mathematicians
, ,   47
could ask whether all functions are analytic or even continuous, and could
contemplate negative answers thereto, that controversy was long past.
To go even further, after the development of measure theory Birkhoff and
von Neumann pointed out that when classical mechanics solves problems
about systems with given precise configurations, we can construe it as using
conveniently simplified descriptions. More realistic, they suggested, would
be the description that results if we transform the precise descriptions by
identifying regions that differ only by sets of measure zero.27 Their reasons
for thinking of that as more realistic may or may not be cogent, but it
suffices here to note the conceptual possibility. That is, after the time of
Lebesgue we can look back to the older description of nature and we
have the new option for how to conceive mathematically of the shapes of
things.28
You will realize that I am simply giving examples of how, in many
ways, we must in retrospect look upon the scientific image inherited from
the older generation as open, vague, ambiguous in the light of our new
understanding (that is: in the light of alternatives not previously conceived).
What is the shape of this beer glass really? What was it in the Galilean,
Cartesian, Newtonian scientific image? In each case the presupposition that
it was one item in a certain class gives way to (i) the conditional that this
was so if that shape was correctly represented by some item in that class,
and (ii) the realization that there are other candidates.
As it is for shape, so it is for each primary quality, represented by the
mathematics of the continuum. Indeed, we need to cast our net more widely
still, if we want to find all the ways in which we could now understand
the scientific image fashioned in the seventeenth century. There is no such
thing as the classical continuum, if that is meant to be the continuum on
which the classical (= modern, 17th century) scientific image was erected
originally. Cantor, Brouwer, and Weyl had equal right to regard it as
erected on their continua, which are very different. Of course, today we
will use ‘‘the classical continuum’’ to refer to the subject of real number
theory as it now exists in main stream mathematics. That is the politics of
linguistic usage. But even in what we now call classical mathematics, recall
that we have the option of saying that quotient constructions are more
accurate (and the simple use of real numbers merely a convenient artifice),
as Birkhoff and von Neumann suggested. What would you like the shape
of the beer glass to be?29
48  : 
So, what is the shape of the beer glass in the scientific image? What is
meant by the assertion that its shape is one of the surfaces in the mathematical
representation of nature?
The openness of scientific description here come to light is irremediable.
Of course, every time we outline a range of alternatives for ourselves,
we can ascend our private throne—are we not all kings and pontiffs in
realms of the mind?—and assert that one of these alternatives is the one
true story of the world. When the range of alternatives is refined by
new conceptual developments—or simply by having our attention drawn
upward by logical reflection—we can choose a new option and make yet
another declaration ex cathedra. Arbitrary perhaps, but as definite as can be,
by choice. What we cannot pretend is to be non-arbitrary, or to close
our text once and for all. Yet the form of understanding is always one of
presumed objectivity and univocity. The scientific image is as replete with
uncashed and ultimately uncashable promissory notes as the manifest image.
Any practical context brings its own standards of appropriate precision, so
it is neither proper nor practical to keep this open-endedness constantly
salient, but to acknowledge that is not to deny it.
Distortion by statistical abstraction (Simpson’s paradox)

Abstraction just removes some factors or parameters, and leaves the relations
among the remainder intact, isn’t that so? Just think of a set of premises:
remove some, and the implications of the remainder, taken by itself, is
precisely what it was before the removal. Think of a color photo; remove
the color so that it becomes a photo in black and white: all features that are
independent of color, such as relative size and angles, remain the same. So
why ever think that abstraction can distort?
Simpson’s paradox in statistics gives the lie to this rhetorical question.
Here is an example: the civil servants in a given city claimed that lighting
and ventilation were seriously affecting their well-being and productivity.
The city hired a statistician who showed conclusively by means of sampling
that the productivity among workers in ill-lit and ill-ventilated spaces
was no less than that among workers in general (or in better lit, better
ventilated spaces)—the productivity level was the same in both groups. So
the complaint was concluded to be baseless.
Eventually a new study was done, and the second statistician broke the
data down by looking separately at women and at men. She showed clearly
, ,   49
that among women, the productivity was less for workers in ill-lit and
ill-ventilated spaces than elsewhere. She also showed that among men, the
productivity was less for workers in ill-lit and ill-ventilated spaces than
elsewhere!
So relevance of working conditions did not show up until there was
a subdivision by this third factor (gender). How is this possible? That
is precisely Simpson’s paradox: correlations can be washed out, or on
the other hand brought to light, by averaging in different ways. Here
is the solution to the puzzle: under all conditions the women were more
productive than men working under the same conditions, but the women
were predominantly working in poor conditions.30
The first statistician abstracted from gender, and by this very means
produced a ‘picture’ which was misleading, and even conveyed a falsehood.
The appearance for him, that is, the outcome in his measurement set-up,
has to be assessed as ‘‘how it looks in that set-up’’ or ‘‘relative to that
set-up’’. His abstracting from gender would have been fine if that factor
had been irrelevant to level of productivity under various conditions. On
the other hand, we say now that it was relevant only because at a lesser level
of abstraction in this respect, the correlation is different. (Mind you, there
is no guarantee that subdividing further, by other features besides gender,
won’t undo the correlation again!) In this particular case we can say that
by abstraction he produced a distorted picture of the reality, a picture in
which lighting and ventilation were irrelevant to productivity. Abstraction
is harmless only under very strict conditions of pertinence.
Scale models and virtuous distortion

The nearest to three-dimensional pictorial representation in use in science
is surely the scale model31 —can it be conceived of in the terms proper
to picturing? As we will see, scaling is not simple reduction or increase
in size in all dimensions. In that sense, useful scaling trades not just on
the obvious resemblances in shape but on distortion, both resemblance
and non-resemblance being selective in a way dictated by the purpose at
hand. The scaling cannot be ‘proportionate’ in all respects. The pertinent
question is whether there will be a sufficiently accurate resemblance in all
relevant respects for the purpose at hand.
50  : 
A presumption that this is always possible, and that the relevance will be
transparently perceivable, has had a strong grip on the physical imagination.
The idea of testing a hypothesis at a different scale tends to be immediately
convincing. Think for example of the experiment proposed by Galileo
concerning buoyancy. The conventional wisdom, which he disputed, was
that a ship will ride higher in the water in the open sea than in port, due to
the amount of water below it. This is difficult to test directly because of the
choppy waves on the high seas. So Galileo proposed to place a small vessel
in a shallow tank and load it with lead pellets until the addition of just
one more pellet would make it sink—and then to repeat this procedure
in a much larger body of (quiet) water.32 Doesn’t this proposal design a
definitive test of the hypothesis?33
To take a more critical attitude, we must recognize that for any concept
there are boundary cases where guidance by our usage so far dwindles and
eventually gives out. Scale modeling displays the characteristics of picturing,
by relying on selective resemblance to achieve its aim, but in a way that is
subject to inevitable occlusion or distortion.
Scaling as picturing
A scale model represents, and yields information about what it is a
model of, by selective resemblance. Are there such necessary limitations in
that case, analogous to occlusion, marginal distortion; is scaling explicitly
non-committal?
Consider a scale model of an airplane. It has the same shape overall, but
with the proportions reduced by a multiplicative factor, say 0.0001. Will it
fly? Not necessarily. If it is to fly, to mention just one factor, it must have
something to propel it; but its size limits necessarily what that can be. For
example, the relation to air resistance will be quite different at this scale:
the air, after all, has not been similarly scaled down in any way! There
are other reasons, as we will see below—‘the same shape’ is a deceptively
simple concept.34
Scale models can be produced for the sheer aesthetic pleasure of it, but
more typically they serve in studies meant to design the very things of
which they are meant to be the scaled down versions. This use and its
subtleties were brought out clearly in the Second Day of Galileo’s Two
New Sciences.35 His calculations involved an error, but his principles were
, ,   51
correct. In modern terms we can summarize his conclusions easily for

a cylindrical beam with constant density. Its strength decreases with its
cross-sectional area, which is proportional to the square of its radius. But
the mass is proportional to its volume, that is, to the cube of its radius. So
the strength to mass ratio of such beams with the same density becomes
N times less when the beam’s size is increased by a factor of N. Beyond
a certain point, the mass can no longer be supported, and the structure
collapses under its own weight. As Galileo observes, large ships taken out of
water are in danger of breaking for just that reason, and he gives examples
for optimal bodily structure:
. . . nor can nature produce trees of extraordinary size because the branches would
break down under their own weight; so also it would be impossible to build up
the bony structures of men, horses, or other animals so as to hold together and
perform their normal functions if these animals were to be increased enormously
in height. . .
We can observe conversely that if a reduced structure is to remain feasibly

like its original, some other features besides its size must be scaled as well,
and not proportionately but appropriately.36
Principles of Similitude and Approximation

Roughly speaking, a scale model of X is an object which is structurally
similar to X but suitably smaller. ‘‘Similar’’ and ‘‘structurally’’ have their
usual context dependence as much as does ‘‘smaller’’—in any particular
case, the goal implicit in ‘‘suitable’’ will determine the contextual parameters
for each. And still roughly speaking: there are two assumptions in force
when conclusions about the target are drawn from characteristics of a scale
model.
The first is that structurally similar objects will display the same behavior
in structurally similar circumstances. This was glamorized by Richard
Tolman in 1914 as his Principle of Similitude, more accurately called a
principle of dimensional homogeneity, rather poetically expressed on a
cosmic scale:
The fundamental entities out of which the physical universe is constructed are of
such a nature that from them a miniature universe could be constructed exactly
similar in every respect to the present universe. (Tolman 1914, 1915)
52  : 
So phrased it is a thesis in ontology; on the methodological side it could

perhaps correspond to something like ‘‘All laws of physics are to be, and
all measurable effects are to be conceived of as, invariant under scale
transformations of any kind’’.
Amazingly this principle, which occasioned a good deal of response in
the literature at the time, appears here in this evangelical form decades
after the advent of Planck’s quantum, also, almost ten years after Einstein’s
study of the photoelectric effect, and several years after Bohr’s model of the
atom. By this time it is certainly surprising to see the conviction that scale is
essentially irrelevant to physical modeling. But Tolman is trying to capture
a correct principle of scale invariance, though one that needs considerably
more sophisticated formulation.
If we doubt Tolman’s principle, then inferences from scale models are
just inferences from false assumptions. But there are useful fictions! The
second principle in force is a Principle of Approximation. The centrality
of this idea in applied science was highlighted by Reichenbach.37 Think
of how Newton proceeds to deduce the laws of motion for our solar
system. Keeping his basic laws of mechanics as foundation, he adds the
law of gravitation to describe this universe, then he adds that there is one
sun and six planets to describe our solar system, and finally he adds that
there is one moon to describe the more immediate gravitational forces
on our planet Earth. Newton demonstrates that from a very idealized,
simplified description of the solar system, something approximating the
known phenomena follows. Well, so what? What’s the point of deriving
true conclusions from a false premise? This is the precise point where we
notice this deep assumption at work:
if certain conditions follow from the ideal case, then approximately
those conditions will follow from an approximation to the ideal case.
Mutatis mutandis, this assumption is in force when we have recourse to
any model that we do not presume to be more than approximately similar
to what it represents, but especially in the case of laboratory simulations.
However, we cannot take this blithely as a context-independent meth-
odological principle, given its phrasing with this sort of generality. The
approximation to the ideal case, just as well as the similarity of a scale model
to its original, is similarity in certain respects, with other aspects ignored as
irrelevant for all practical purposes—which means, for the purposes at hand
, ,   53
in the context of application. Whether that ignoring will be vindicated is

an empirical question; we can’t very well decide it on principle. In fact,
Reichenbach shows by illustration in statistical mechanics that even small
departures in approximation can have widely divergent consequences—a
point now popularly familiar from Chaos theory.
Fine, but the study of scale models, and in general studies that seem
to be inspired by these rough and ready ‘principles’ of similarity and
approximation, are often useful, practical, and truly vindicated. So what are
the facts of the matter, the real constraints on such modeling?
Dimensions and invariance

We can glean these from the critiques of Tolman published in a seminal
paper by E. Buckingham and a critical article by Percy W. Bridgman, the
physicist famous for originating operationalism as a philosophy of science.38
The applications that Tolman outlined for his principle, both authors argue
to be derivable in dimensional analysis, a technique whose development
started with Fourier. I’ll explain some of the basic ideas and then display
the example of the screw propeller, which Buckingham analyzed in detail.
Fourier had extended the geometrical notion of dimension to the now
familiar general concept of physical dimensions, so that not just length,
area, and volume but also mass, force, temperature, charge and the like are
included:
[E]very undetermined magnitude or constant has one dimension proper to itself,
and . . . the terms of one and the same equation could not be compared, if they
had not the same exponent of dimension. We have introduced this consideration
into the theory of heat, in order to make our definitions more exact, and to serve
to verify the analysis; it is derived from primary notions on quantities; for which
reason, in geometry and mechanics, it is the equivalent of the fundamental lemmas
which the Greeks have left us without proof. In the analytical theory of heat,
every equation (E) expresses a necessary relation between the existing magnitudes
[length] x, [time] t, [temperature] v, [capacity for heat] c, [surface conducibility] h,
[specific conducibility] K. This relation depends in no respect on the choice of the
unit of length, which from its very nature is contingent, that is to say, if we took a
different unit to measure the linear dimensions, the equation (E) would still be the
same.39
This same passage, which introduces the general conception, also introduces
the idea of ‘‘dimensional homogeneity’’ and the importance of invariance
54  : 
under scale transformations for the fundamental equations of physical

theory.40
An equation must be dimensionally homogeneous to make sense: the
dimension of the quantity on the left must be the same as that on the right.
That is just the common place that you can’t add apples and oranges except
in the sense that you can take them both as fruits and count them that way.
In more complicated cases, this homogeneity has to be checked.
Take the equation of the distance covered s calculated from time t,
velocity v, and acceleration a:
s = vt + (1/2)at2
Does that make sense? Here distance has the dimension of length, call it L;
velocity has the dimension of length divided by time (T) and acceleration
the dimension that has velocity divided by time. We must first check that
on the right-hand side we are not trying to add apples and oranges, but
rather things of the same sort. We check this by replacing each of the
parameters by its dimension alone, while multiplying replacements of the
terms with each other:
(L/T)T
(L/T)(1/T). T.T
and then treat those dimensions algebraically the same as numbers. The
result, after cancelling the Ts against each other, is L in both cases. Thus
we are adding two like quantities, and the quantity denoted by the right
side of the equation has dimension L. But the dimension of s is also L, so
we have the required match. Without detailed scrutiny this may look like
a calculation by rules of thumb far removed from rigor, but I will leave the
detailed justification of dimensional analysis techniques to other sources.
The second requirement, recall, is that of invariance under scale trans-
formations. To achieve invariance under such transformations—rewriting
equations stated in terms of certain quantities in terms of others—is pre-
cisely served in dimensional analysis by the search for ‘‘dimensionless’’
quantities as sole constituents for the fundamental equations.41
As illustration we can begin with the familiar cgs system of units in
mechanics: centimeter for length, gram for mass, and second for time. A
different scale belonging to the same class of systems of units is one defined
by multiplying each unit by a positive number. This is the form of a scale
, ,   55
transformation. If we choose the numbers 100, 1000, 1 for this role, we define
the MKS system, with the units meter, kilogram, second—thus producing
another system belonging to the same class of systems of units. The basic
invariance requirement is now that to be significant, an equation must have
the same form regardless of which member of the class of systems of units
is chosen. This requirement was obviously respected well before Fourier,
let alone before the subject of dimensionless analysis matured. Newton’s
famous F ≈ ma does not depend on its validity on a particular choice of
units, and would not be famous if it did.42
A dimensionless number —more accurately speaking, a dimensionless para-
meter —of the class is a quantity that has the same value in every system of
units in the class. That is, it is an invariant of the set of admissible trans-
formations, which are precisely the scale transformations. The Hauptsatz of
dimensionless analysis, prominent in Buckingham’s article, says in part that
it is always possible to shift to a dimensionless representation.43
The screw propeller

Susan Sterrett points out the relevance of how the Wright brothers and
their colleagues in the field were frustrated when they tried to extrapolate
the behavior of children’s flying toys to a larger scale. Making the object
‘the same but larger’ ruined its capacity to fly—why? Wasn’t the toy a
scale model for their construction? The fact is that here, as much as in the
examples the Sophist pointed to, what counts as pertinent resemblance is
not at all obvious in the way that ‘looking alike’ is obvious.
Buckingham, partly in service of his critique of Tolman, analyzed the
case in detail. In the studies of the screw propeller, which had of course
been started for ships but were then crucial to the development of the
airplane, both rough and ready ‘principles’ can be seen in a rigorous form.
The thrust F of the propeller of given shape and immersion is taken to
depend only on the diameter D, the speed of advance S, the number N of
revolutions per unit time, the density and viscosity of the liquid, and the
acceleration due to gravity.
So suppose that a smaller propeller is meant to be a good scale model of
a large one with respect to thrust —in contrast with some other effects, here
regarded as ignorable side effects (noise, shape of the wake, . . .). Then the
equation which expresses F in terms of those other quantities must be the
same for both cases, provided that the ratios that specify the shape and
56  : 
immersion of the propeller stay the same. Any set of kinds of quantity that
furnish the basic units for this dynamics can be changed in any ratios what-
soever without affecting this. How does this serve to guide a practical study?
The obvious thing to do is to make the smaller propeller geometrically
similar to the original, to immerse it similarly, and to construct the propellers
so that the ‘angle of attack’ of the blades on the water is the same. Is that
enough to ensure that we can get information about the thrust of the
original large propeller from the behavior of its scale model? It suffices only
if one can completely control similarity in the effects of gravity, density,
and viscosity. It is easy to see that for extremely small propellers those
effects will be significant, and for larger but still small ones the difference is
after all only a matter of degree.
So in practice . . . one has to resort to some approximation. And here
special conditions can be experimentally investigated to see what can and
what cannot be ignored at various scales. For example, the pertinent
mechanical behavior in very turbulent motion does not vary much with
the viscosity of the fluid. And similarity in the effects of gravity will be
approached when the ratio between the two speeds of advance approximates
the ratio of the squares of the two diameters. Deep immersion will also
prevent significant effects due to disturbance of the liquid surface. (Notice
though that these are all matters of degree, and the purpose at hand
may require a specific level of accuracy.) With all of this supposed under
sufficient control, the ratio of the thrust of the small propeller to that of the
large one will be (DS/D S )2 , where the primes indicate the diameter and
speed of advance of the large propeller, and not (DS/D S ), that is, not the
ratio by which this product was altered in construction.
Conclusion about imaging and scaling

Imaging, recall, is representation that is effected through resemblance. Our
discussion of mathematical statuary ended with the conclusion that we can
see that mathematical representation of nature so far always involved some
features that, in retrospect, with hindsight, we saw as necessary failures in
resemblance. That does not imply that mathematical representations of the
sort now available are also thus necessarily deficient, though some humility
, ,   57
in even that respect is appropriate. But in application, the practice of actual

construction of models of situations, the idea of ‘perfect’ modeling is so far
from realistic that it can certainly not be maintained.
The resemblance of even so ‘obvious’ an example of a scale model to its
original is, as we have just seen, not nearly as simple as may strike the eye
at first glance. True, a scale model of a vessel or propeller under study in a
naval or aeronautics laboratory will ‘look like’ a real one. But for it to have
any use at all for the purpose at hand, there must be a delicately achieved
pertinent similarity (to a pertinent degree of approximation) between the
situation of these propellers revolving ‘similarly’ when ‘similarly’ immersed.
This may well, and typically does, come at the price of dissimilarity: as
Galileo already appreciated, the scaling must be different for different
parameters. The ‘‘selective’’ in ‘‘selective resemblance’’ is a delicate, highly
nuanced, contextually sensitive qualification—and this point is general: it
pertains to all pictorial representation.
3
Pictorial Perspective
and the Indexical
Imagery, as defined above, is pictorial exactly if it bears hallmarks of

perspectivity. In the notion of perspective, as so often, we have a cluster
concept, with multiple criterial hallmarks. There is no defining common set
of characteristics, only family resemblances among the instances. Whether
or not something is aptly called perspectival depends on whether some
appropriate subset of these hallmarks are present, but what amounts to
‘‘appropriate’’ we cannot delimit precisely either.
The hallmarks listed above were occlusion, marginal distortion, texture-
fading (grain), angle, and with special importance, explicit non-commitment and
the ‘‘horizon of alternatives’’. These are all characteristics that relate to the
content of the representation. But there is another notion closely connected
to perspective which does not appear here. This does not pertain simply
to content, but to how we relate to it; it comes to light when we very
naturally think that for the painter or photographer, a picture is showing
how the pictured scene ‘‘looks from here’’. The painter’s eye is located
with respect to the content of the painting in a way that he himself can
express with ‘‘this is how it looks to me from here’’. The viewer may
naturally say that this is how the scene ‘‘looks from there’’. If I say, for
example, that the photo shows the town as seen from the top of the church
tower, that indicates something like ‘‘that is how it would look to me if I
were on top of the church tower’’. These are indexical statements, with the
words ‘‘here’’ and ‘‘there’’ playing the context-sensitive role.
For a critic describing a painting this may not be relevant. While s/he
may refer to the painter’s view, no special interest may attach either the
painter’s subjective situation or what the critic’s own would be. But that
60  : 
changes when we turn to representations subject to different norms and

use. If we look to the painting or photograph to help us get around in
the town, that does require us to locate ourselves with respect to the
view presented by that picture. Since scientific representations are typically
produced so as to serve some such practical end, at least in principle,
this connection between perspective and the indexical becomes important
there. The connection shows up in the sciences, for example, when talk
of frames of reference is conducted in terms of observers (whose frames
they are, so to speak). We need to look closely into both the character of
perspective and the role of indexical judgments (such as self-attributions
and self-locations) to see whether that is just an irrelevant heuristic or
whether it brings to fore a fundamental connection between perspective,
measurement, and theoretical representation.
Pictorial Perspective and the Art of Measuring

la pittura è una specie de natural filosofia, perché l’imita la quantità e qualità, la
forma e virtù delle cose naturali.1
The histories of perspective in painting, measurement, geometry and

technology are thoroughly entangled.2 Geometry is so-called because it
began as the art of ‘earth-measuring’, and in Dutch its name is still ‘‘the art
of measuring’’ (meetkunde). But as we’ll see, a famous treatise on perspective
was called by that name as well.
Examining some of this history it will become quite clear that a picture
in the modern perspectival style is essentially the outcome of a measuring
procedure. Conversely, a measurement outcome is, in paradigm cases at
least, a pictorial representation of the object measured. This paradigm
example I will extrapolate subsequently: measurement falls squarely under
the heading of representation, and measurement outcomes are at a certain
stage to be conceived of as trading on selective resemblances in just the way
that perspectival picturing does.
Astrolabe and triangle

How do you measure the width of a river while remaining on the shore?
Or the height of a tower while remaining on the ground?
     61
The first clue to the answer is of course that the two problems are
essentially the same, related by a simple rotation from horizontal to vertical.
The second clue is also geometrical: in a right triangle, the ratio of height
to base is determined by the ‘‘angle of sight’’ along the hypotenuse. At this
point today’s student reaches for trigonometry, but these practical problems
were solved long before that was available.
By the end of the Roman Empire in the fourth century much of Greek
mathematics was lost to the West, not to return there from Arabic sources
until almost 800 years later. During these centuries practical surveying, archi-
tecture, and the scholarly study of practical geometry did continue, however.
The practical techniques of the Roman surveyors survived, were preserved,
collated, and taught among both artisans and cleric-scholars.3 A representat-
ive text, the practical geometry manual of Hugh of St. Victor, in the century
just before Euclid’s Elements became available again, is divided into three
parts: altimetry, planimetry, and cosmimetry (measurement relating to the
earth, to the sun, and to other aspects of the cosmos). In retrospect we see the
methods there presented as justifiable within geometry and geometric optics,
but what is taught there is simply the practical technology of measurement.
The instruments designed for this use—cross staff, quadrant, and even
the astrolabe introduced into the West about a century before this manual’s
date—consist basically of a ruler with a sighting device (alidade) at the cen-
ter, and part of a circle on which degrees are marked. These had significant
use in navigation, but let us here concentrate on land-measurement.4 The
surveyor measuring the height of a tower, for example, adjusts the alidade
until he can see the top through the two apertures. The angle thus formed
determines the ratio of the height of the tower to the distance from the
tower. Determines how?
Though most of the mathematic theory was lacking, the astrolabe can
be manually calibrated on relatively small similar triangles. This presumes
understanding of geometric similarity; sufficiently much of Euclidean
geometry was retained to understand this.
The distance from the tower may itself not be measurable directly if it
is far away, so Hugh’s manual gives several forms of ‘two station’ methods
to use. Suppose that the astrolabe sighting is done at two points P and Q,
at an unknown distance from the tower. Measure the distance between P
and Q, and a few practical steps, starting with the two alidade readings, will
yield the height of the tower:
62  : 
Figure 3.1. Perspective Altimetry
Let h be the height of the tower. The direct measurement by astrolabe at

P gives the first ratio A = h/PT. The reading at Q gives the second ratio
B = h/QT. Only the distance PQ is measured directly. From these three
data, the height h can be calculated directly.5
We will soon see this same configuration again.
Alberti’s De Pictura
When Alberti wrote his monograph on the technique of perspectival
drawing in 1435 a great deal of Greek mathematics and geometric optics
had been assimilated in the years since Hugh’s manual of practical geometry.
But his way of writing was not so different from Hugh’s, because Alberti
was also a surveyor applying those practical arts as well as thinking about
the theory behind them.6 Since the practical geometry manuals focused on
geometric figures created by physical objects and lines of sight, they were
an obvious source for his study of perspective. His great innovation was to
think of the visual cone or pyramid cut by the picture plane:
[Painters] should understand that, when they draw lines around a surface, and fill
the parts they have drawn with colours, their sole object is the representation
on this one surface of many different surfaces, just as though this surface which
they colour were so transparent and like glass, that the visual pyramid passed right
through it. . . . (Alberti 1991: Book I, 48)
     63
Art historian S. Y. Edgerton refers to Alberti’s invention as ‘‘Windows

1435’’. A painter drawing from life is as it were drawing on a window
through which he is seeing his subject. In fact of course, the plane on
which he draws (the canvas) is not the plane (imaginary window pane)
cutting his visual pyramid. But he will succeed in accurately rendering his
subject if what he produces is precisely what it would be if he did draw on
that ‘window’ plane.
There was an elementary exercise for this skill whose sign we see in
many paintings of that era: the checkerboard floor or pavement. Francesco
Rosselli’s Supplice de Savonarola (c. 1498), for example, depicts the central
square in Florence precisely with such a checkerboard pavement seen in
one-point perspective. The following illustrations show respectively how
the painter is imagined to see and paint, and what it is that he sees as it will
appear on the picture plane:
Figure 3.2. Window and Checkerboard
Notice how the left-hand diagram is really just the one above of the
‘‘two station’’ altimetry, flipped horizontally, but with the picture plane
and some other sightlines added. The painter’s eye corresponds in the
geometry to the top of the tower of the earlier illustration. So the point of
view is the opposite, as it were, but the geometry involved is the same.
Alberti’s concern was with technology. He began his development of
perspectival drawing techniques by making a box with a small eye-hole
in one side.7 In the box there was a checker board laid horizontally on
the bottom. Let us call what the checkerboard looked like, if viewed
through the peephole, the checkerboard appearance. (That is to say, the
box floor’s appearance in observation or measurement made through the
64  : 
peephole.) Studying this set-up in various ways he could make a drawing

which, if placed upright in the box at a certain point, presented the same
checkerboard appearance to the eye. In fact, viewers could not distinguish
between the two when looking through the peephole.8
Looking back at the second illustration you can see that parallel lines
orthogonal to the picture plane converge to a point on the horizon, while
lines in the other two orthogonal directions remain in place. So the lines
between the tiles that are parallel to the picture plane remain parallel to
each other; similarly the up-down lines remain vertical. This is one-point
linear perspective. In two-point perspective parallelism is preserved in only
one of the three directions; but this is hard to find in the history of
painting till much later. We can think of one-point perspective as the style
of representation that captures what is seen in the Alberti experimental
situation—the unmoving single eye at the peephole. We can equally think
of it as the style in which of three orthogonal directions in space, two are
preserved in effect in a grid of parallel lines, while in the third direction all
straight lines converge to a point. Masaccio’s Trinity, Botticelli’s Episodes
in the Life of Lucretia and Episodes in the Life of Virginia, as well as the
philosophers’ favorite, Raphael’s School of Athens, can be mentioned as
examples of this style of visual representation.
The Art of Measuring: mechanization of perspective

Alberti’s technique was applied by contemporary painters, though not
nearly as rigorously as either Alberti or Vasari would have it. On the other
hand, it was of use also to architecture, technical drawing, and machine
design. Alberti’s geometric and practical studies of perspective include ways
to ‘mechanize’ the process of perspectival drawing:
So attention should be devoted to circumscription; and to do this well, I believe
nothing more convenient can be found than the veil . . . whose usage I was the
first to discover. It is like this: a veil loosely woven of fine thread . . . divided up
by thicker threads into as many parallel square sections as you like, and stretched
on a frame. I set this up between the eye and the object to be represented, so
that the visual pyramid passes through the loose weave of the veil. (Alberti 1991,
Book II, 65)
His ‘‘veil’’was precisely the painterly window we have been discussing,

but now realized as a practical technological artifact. The veil with its grid
     65
is a measuring instrument, designed to measure not such simple quantities

as length or weight but—as I shall discuss further below—cross ratios,
projective structure.
This is explicitly recognized in Albrecht Dürer’s treatise, where the tech-
nique is presented in a part entitled Unterweysung der Messung—‘‘Teaching
of Measurement’’, generally translated as ‘‘Art of Measurement’’. The
mathematically precise and practical character of this way of rendering the
appearances implied its possible mechanization.9 The basis was in effect a
very careful and systematic form of measurement, in which certain geometric
features are faithfully captured on the picture plane. This way of under-
standing the episode is supported by a look at the machines that Dürer
designed to produce perspectivally correct drawings, which show how far
this can go.10 In his most advanced artifact even the human element consists
of fully determined mechanical motions. Measurement is precisely what this
was, as basis for pictorial representation. The content of such a visual
perspective is the content of a complex, technically advanced measurement
outcome.
Figure 3.3. Dürer, the Draughtsmen of the Lute

66  : 
The heart of an experiment is typically a sort of measurement: the

set-up produces or lends itself to a phenomenon that is meant to provide
information about the character of some target object, event or process.
The artificially produced or isolated phenomenon is treated as providing
data about the target, to provide us with a ‘view’ of it. Dürer’s machines
do exactly that, in elementary fashion: they produce drawings that provide
us with a view of the object from a given vantage point.
Perspective versus Descartes’s frames of reference

Alberti knew that his method worked, in the sense that it produced a realistic view
of three dimensional objects. What he didn’t know was why it worked. [Projective
geometry] presents the key mathematical concept behind Alberti’s method, the
cross ratio.11
To understand perspective mathematically we have to go on to seventeenth-

century geometry. But there another style of representation, more familiar
to the scientist, was developed as well: Descartes’s analytic geometry which
coordinatizes space with a generalized rectangular frame.
We must take care to distinguish these, for perspective and perspectival
painting are often drawn on as metaphors for the new ‘spatialized’ or
‘geometrized’ representation of states and motions introduced in the sev-
enteenth century. At first blush that fits: description of a ship or its motion
in a geometric frame of reference relates that structure to an origin and
an orientation in space. But the origin of such a frame is nothing like the
position of the eye in perspective, and the directions in space are there
only as a conventional orientation, introduced to provide a user-friendly
description. So that metaphor, however appealing and pervasive, needs to
be kept at arm’s length.
Illustrations suffice to show how different Descartes’ representation is
from that of a painter, even in structure. Both drawings depict a cube.
But in the first, in none of the three directions do parallel lines converge.
Parallel lines are depicted there as parallel, and equal line segments are
depicted as equal. For this representation pertains to Euclidean geometry
with its characteristic symmetries. A coordinate system is indeed a specific
     67
Figure 3.4. Frame of Reference . Perspective
representation of the space in the manifold R3 of triples of real numbers.

But coordinate transformations, moving us from one system into another,
preserve both distances and parallelism. The move from one visual per-
spective into another does not! Despite the common features and despite
the attraction of the metaphor, depiction in a Cartesian frame of reference
is clearly not literally a perspectival representation of anything.
My drawings do both utilize projection. When I drew the cube in the
Cartesian frame I could not show it to you in its proper three dimensions,
but had to project that on a plane surface as well, and in that respect
our two figures are similar. In the first drawing, projection is orthogonal.
Orthogonal projections occur naturally as well: the sun’s rays are parallel
(on our scale, within our modest degree of accuracy) so a shadow on a
properly positioned wall is an orthogonal projection. Lines converging to
an ever more distant vanishing point approach a set of parallel lines—thus
in orthogonal projection we have as it were an eye ‘at infinity’, but of
course not at any particular point at infinity.
The difference between Cartesian frames of reference and visual perspect-
ives comes out even more clearly if we allow for a ‘motion picture’. Galileo’s
diagrams strike us still as faltering first few steps. For example, in his study
of falling bodies and of projectiles we find separate diagrams for the time
axis and for different spatial distance axes. Descartes’s geometry combines
the spatial dimensions and links the geometric figures to algebraic equations
in a way to us so familiar that we can readily read his treatise on the subject.
68  : 
For an illustration, consider two cars rolling along a broad one-way road,
one on the left and one on the right. The road consists of equal squares,
and the two cars reach the end of each square simultaneously. We can
imagine a spatial frame of reference that has the left border of the road as Y
axis and the orthogonal border of the first square as X-axis. The two cars
are moving in parallel, with the same velocity, and are thus depicted in a
geometric representation of this situation.
Now a painter sets up his easel at beginning of the road, and he has his eye
precisely on the Y axis. He sees the two cars ready to start, the time is t = 0.
The left and right borders of the road are the lines X = 0 and X = 1. Of
course the painter sees these converging in the distance. To picture this, let
us modify the diagram showing the checkerboard floor drawing. The line
X = 0 is straight, orthogonal to his easel, but the line X = 1 slants in the X-
negative direction, meeting X = 0 on the horizon. The cars begin to move.
He sees them reach each horizontal line simultaneously. But the right-hand
car is moving along the hypotenuse of a perceived triangle, so covers a larger
perceived distance than the left-hand car in the same time interval.
Figure 3.5. Speed in Perspective
Within the painter’s perspective, the right hand car is moving faster than
the car on the left. If he were making a motion picture, or simply taking
notes of where in his visual field the cars are at t = 1, t = 2, etc. he
would be making a measurement of the velocities, but the content of his
measurement would be what the motions look like and not what they are.
Knowing the geometry of this space and the laws of projection, he can of
course draw on registered relations between the two perceived motions to
     69
obtain information about what the motions are really like. These motions
are observable processes, they have determinate speeds and directions which
are in fact the same, although the appearances (which are the contents of
the measurement outcome) are different.12
The ‘view from nowhere’ and the indexical

The above should be a caution against talk of ‘‘perspective’’ when discussing
coordinates and frames of reference. But why then is such talk so pervasively
prevalent?
Perspective is a cluster concept, and while careful not to extend its use so
broadly as to make it practically vacuous, we should also take care not to
narrow our use so much as to lose touch with common discourse. In fact, I
was ignoring the connection with the indexical, remarked on at the outset,
and we should now bring it back in.
First of all, it is true that despite its particularity due to an arbitrary choice
of origin and direction, we can think of the Cartesian representation of
extension, duration, and motion as embodying a ‘zero-point perspective’:
the ‘view from nowhere’ or ‘the point of view of no one in particular,’
to use some famous phrases.13 That is precisely because the chosen frames
of reference, the coordinate systems, are inessential to the geometry taken
in and by itself. The Cartesian representation is God-like; or so at least
Leibniz eventually depicted it:
the distinction between the appearance bodies have with respect to us and with
respect to God is, in a certain way, like that between a drawing in perspective and
a ground plan. For there are different drawings in perspective, depending upon
the position of the viewer, while a ground plan or geometrical representation is
unique. Indeed, God sees things exactly as they are in accordance with geometrical
truth, although he also knows how everything appears to everything else, and so
he eminently contains in himself all other appearances.14
But this Leibnizian God has no need of representations either, nor of

anything else crucial to us, finite and situated observers. The coordinate
system plays no role for God, and plays no role in geometry conceived in
pure abstraction—but what role does it play for us?
We can see the shift back and forth between these two ways of ‘seeing’
the world in how theories are presented in physics. In his 1905 paper that
introduced the Special Theory of Relativity Einstein wrote:
70  : 
Examples . . . lead to the conjecture that . . . the same laws of electrodynamics and
optics will be valid for all coordinate systems in which the equations of mechanics
hold. . . . We will raise this conjecture (whose content will hereafter be called
‘‘the principle of relativity’’) to the status of a postulate and shall also introduce
another postulate, which is only seemingly incompatible with it, namely that light
always propagates in empty space with a definite velocity V that is independent
of the state of motion of the emitting body. These two postulates suffice for the
attainment of a simple and consistent electrodynamics of moving bodies based on
Maxwell’s theory for bodies at rest. (Einstein 1905/2005: 124)
This passage concentrates on the ‘‘objective’’, that is on what is invariant,

valid for all frames of reference—which makes reference to frames of
reference inessential. But Einstein’s exposition thereafter takes on a different
form:
If we wish to describe the motion of a material point, we give the values of its
coordinates as functions of the time. Now we must bear carefully in mind that a
mathematical description of this kind has no physical meaning unless we are quite
clear as to what we understand by ‘‘time’’. We have to take into account that all our
judgments in which time plays a part are always judgments of simultaneous events.
If, for instance, I say, ‘‘That train arrives here at 7 o’clock,’’ I mean something like
this: ‘‘The pointing of the small hand of my watch to 7 and the arrival of the train
are simultaneous events.’’ (Ibid.)
This shift from the ‘God’s eye view’ to a frame of reference identified
with an observer equipped with clock and measuring stick continues in the
well-known thought experiment he then presents for the measurement of
a moving rigid rod. We now inquire about the length of the moving rod,
which we imagine to be ascertained by the following two operations:
(a) The observer moves together with the aforementioned measuring
rod and the rigid rod to be measured, and measures the length of the
rod by laying out the measuring rod in the same way as if the rod to
be measured, the observer, and the measuring rod were all at rest.
(b) Using clocks at rest and synchronous in the rest system . . . , the
observer determines at which points of the rest system the beginning
and end of the rod to be measured are located at some given time t.
The distance between these two points, measured with the rod used
     71
before—but now at rest—is also a length that we can call the ‘‘length
of the rod.’’ (Ibid., 128)
Not long after Einstein’s creation of this new theory, Minkowski recognized
that Einstein had in effect displayed an elegant new mathematical entity,
distinct from those utilized in classical models. Light paths are to be
represented by curves in this space along which the space-time interval
equals zero . . . motions of bodies by paths on which the points have
time-like separation . . . and so forth. That is to say, according to the new
theory, we are to use this mathematical object in that way to represent the
natural phenomena in this domain.
At this point observers and frames of reference are left behind. Neither
perception nor individual cognition is a salient topic of inquiry in the
context of use of Minkowski space for the representation of rigid motion,
electric current, magnetic field, and transmission of light. Frames of ref-
erence can be thought of as attached to any material body or to none;
certainly talk of relatively moving observers is de trop.
But if we focus on the theoretical models in and by themselves, we
are ignoring the use they have. If someone is to use Einstein’s theory to
predict the behavior of electrically charged bodies in motion, bodies with
which s/he is directly concerned, choice of a coordinate system correlated
to a defined physical frame of reference is required. The user must leave
the God-like reflections on the structure of space-time behind in order to
apply the implications of those reflections to his or her actual situation. The
physical world picture in abstracto is as far removed as possible from this use,
it embodies, in Eddington’s words, ‘‘the view of no one in particular’’. But
to put this picture to use, something must be done by the user, and this is
where choice of reference frame comes in. Hence Weyl’s words are equally
apt, when he refers to coordinate systems as ‘‘the unavoidable residuum of
the ego’s annihilation’’.15
In sum then, the use of ‘‘perspective’’ and ‘‘perspectival’’ in connec-
tion with depictions of events in varying frames of reference cannot be
banished completely. There is a close connection between them and
the acts of observers locating themselves with respect to the theoretical
model—acts of self-location, expressible by the actor in such terms as
72  : 
‘‘I am here, and this is how it looks from here’’. I will come back to
this after we have looked at some more general examples of use and
self-location.
Geometric projection and perspective

The Cartesian style of scientific representation is meant to be of the
phenomena as they are, and not as they appear to the observer. But how
the phenomena appear in observation or in measurement—their appear-
ances—must also be covered.16 There too, all that is crucial to the actual
information captured is in the invariants. What is invariant in the move from
one visual perspective to another? The subject of projection and perspectiv-
ity belongs now to projective geometry, which became an autonomous
subject only in the nineteenth century. But to the extent relevant to our
present discussion it was developed far enough within Euclidean geometry
in the seventeenth century, notably by Pascal and Desargues.17
We can ask the question ‘‘What is invariant in the move from one visual
perspective to another?’’ in a different way, if we take for granted that the
invariants are informative about the represented landscape. How, or to what
extent, is perspectival drawing mimetic? To what extent, or in what fashion, is the
drawing a ‘copy’ of what is drawn? Alberti had already noted that the ‘visual
pyramid’ is replete with similar triangles, so that certain proportions are
preserved as the objects are imaged on the painterly ‘window’.18 But this
pertains to proportions on lines parallel to that ‘window’, and does not get
us all that far.
There are relations however that are invariant under projection, and
will therefore be the same in all perspectives on an object regardless of the
position of the eye (or camera). The main projective invariant is the cross
ratio, a four-term metric relationship. It is not actually an easy relationship
to grasp for the visual imagination!
Consider a sequence of four points A, B, C, D on a line, in that order.19
Let us call A and B the major points and C, D the minor points. Then the
cross ratio CROSS(A, B; C, D) is the ratio of two distance ratios, namely of
the minor points to the major ones: (CA/CB) : (DA/DB).20
Here we have the correct generalization of the checkerboard pattern
template that we saw above: the cross ratios of 4-tuples of points on
the ‘windows’ and on the ‘ground’, that correspond to each other by the
projection from the ‘eye’, are the same. In the checkerboard floor, the
     73
Figure 3.6. Cross Ratio Invariance
distances between adjacent points are equal, so the cross ratio is 4/3—a
special case that is easily visualized and reproduced by the painter.21
Take any point P and draw lines connecting P with the four points in
question. It can be quickly established that the cross ratio is the same for all
projections of those four points onto other straight lines.22 In the diagram,
where point P now plays the role of ‘eye’, CA/CB : DA/DB = CA /Cb
: dA /db, and similarly for other sections of the pencil of projection lines.
This picture we will revert to below to show how taking a photograph can
be used to make a distance measurement.
The invariant content of a perspectival image is the structure ‘common’
to the images produced on any ‘window’ that cuts the pencil of projection
lines. That ‘common’ structure is the structure entirely captured in a
description of the cross ratios between such sequences of collinear points, in
addition to the basic relationships of incidence and order of points and lines.
It is this invariant content that carries the information in the perspectival
image that is independent of the choice of origin and orientation—except
of course that no one ‘window’ can contain an image of more than a finite
part of a selected half of the space ‘on its other side’.
Measuring with perspective

Alberti’s ‘demonstrations’ as well as the famous display by Brunelleschi of
his painting of the Florence Baptistery, had, as Feyerabend pointed out,
all the characteristics of a scientific ‘demonstration experiment’.23 What
they demonstrated was that the new techniques in painting correctly and
74  : 
accurately ‘latched onto’ certain aspects of geometric structure. The rough

and ready way to say this is ‘‘they got the proportions right’’, thinking first
of all of the proportions in the human body, and secondly in architectural
plans. But ‘‘proportion’’ is neither specific enough nor general enough,
as we have just seen. We can’t equate the preservation of cross ratios
with preserving proportions—though that is what it is in simple cases.
What is true is that the spatial structure is being correctly and accurately
ascertained in certain respects—and this sort of ascertaining is precisely what
measurement is all about.24
When Dürer called his seminal work on perspective Unterweysung der

Messung, that was not idiosyncrasy but insight. To begin its exploitation, let
us see precisely how the perspectival drawing is a measurement, revealing
quantities in the way any measurement outcome is meant to do. Imagine
you have such a drawing—or a pinhole camera photo—of a long straight
road, on which you see a person as well as the mile 1 and mile 2 markers.25
How far is the person from the mile markers?
We can use the above diagram. Let the line A, B, C, D be the drawing
(or the photo, still in place in the camera, if you like), while that person
and the markers are at A , B , and C on the line that represents the road.
(The line A bCd is not relevant here.) The only qualification to be entered
concerning the cross ratio concerns infinity. Imagine a fourth point D on
our line A B C placed somewhere beyond C and then moved further
and further outward. As it moves, the line PD begins to approach PD,
the horizontal line that never reaches the ground. In the case of a painter
looking through his window, the line PD is the one at eye level, going to
the horizon.
The mathematical conceit to be introduced at this point is that the two
parallel lines ‘‘meet in a point at infinity’’. No need to reify that fiction: we
can remain with the familiar and much more elementary tactic of taking
the limit.26 The degenerate cross ratio CR(A , B , C ) turns out, on this
construal, to be just C A /C B . The invariance of the cross ratio under
projection still applies: CR(A , B ; C ) so defined is numerically equal to
the original CR (A, B, C, D) from which it is this degenerate projection.
The cross ratio of those initial four points therefore equals precisely the
ratio C A /C B . But the denominator C B is just the distance between the
two mile markers, i.e. one mile. The quantities in the cross ratio CR(A, B;
     75
C, D) on the other hand can be found by measuring them in the drawing or

photo itself . Therefore the distance between the person at A and the second
mile marker equals a determinable quantity in the produced image.27 In
this fashion we shall thus have measured a distance on the road by means of
that drawing or pinhole camera operation. The measurement outcome, which
‘indicates’ the value of that quantity is precisely the photo thus produced:
The image = the outcome of a measurement, of a certain metric quantity,
performed on the depicted situation.
Conversely, the outcome of this measurement is the final state of the film
which represents the object thus measured with the value of the measured
quantity precisely indicated on its ‘face’.
As in any typical measurement, our procedure here does not convert
pure raw data into information in an assumption-free or presupposition-less
fashion! That the horizon is ‘at eye level’, and that the distance between the
two markers is precisely one mile, is assumed in our taking the photo to be
representing the situation in the way that we do. In this simple case those
assumptions can be independently checked by observation. In other cases
they are themselves supplied by a theory, with no such easy access. As Bacon
said, experience must be literate; seeing what the measurement outcome
reveals requires being able to ‘read’ it. Reading is always reading under
the aegis of our entire prior state of opinion plus current suppositions—in
that sense, all our reading, even in the general sense of taking notice of
measurement results of any sort, is always theory-laden.
Mapping and Perspectival Self-Location

The development of the art of Perspectiva into new branches of geometry
greatly favored the evolution of cartography. In maps we have scale models
of terrain, but projected onto a plane, thus producing occlusion of a sort
not inherent to three-dimensional imaging. Maps do not usually have
an obvious perspective; but we see perspectivity when, for example, the
curvature of the earth makes marginal distortion inevitable as a result of
this projection that lowers the dimensionality. A map too is the product
of a measuring procedure, but they bring to light a much more important
point about ‘point of view’, essentially independent of these limitations in
76  : 
cartography. The point extends to all varieties of modeling, but is made

salient by the sense in which use enters the concept of ‘map’ from the
beginning. A map is not only an object used to represent features of a
terrain, it is an object for the use of the industrial designer, the navigator,
and most of all the traveler, to plan and direct action. This brings us to an
aspect of scientific representation not touched on so far, though implicit
in the discussion of perspective, crucial to its overall understanding: its
indexicality.28
Representations have their uses

Use enters the very concept of representation, as we have seen: to be a
representation is to be something used or taken to represent something.
But in addition, representations have their uses. They are typically produced
for a certain use, with a certain purpose or goal. An artist may paint, sculpt,
or select an object to represent scenes of war for example, for display
to the public, either simply for their appreciation or because s/he ‘has
something to say’. A cartographer draws a map to represent the structure
of the Eurasian railroad system, for use by its management for planning or
alternatively for use by the public for getting around the continent. Bohr
creates his 1913 model of the atom to explain certain spectrographic data. A
map is close kin to a scale model; Rutherford’s atom was perhaps thought
of as a scale model (with dimensions dilated rather than contracted); Bohr’s
atom registered its resistance to being thusly viewed immediately.
While there is great diversity in these examples of the use of representa-
tions, many have in common the goal to convey information. The activity
of representation is successful in that case only if the recipients are able to
receive that information through their ‘viewing’ of the representation. Our
discussion of the propeller, for example, displayed a theoretical analysis;
how can it guide real propeller construction? In science the original creation
of a model may have been a purely theoretical activity, but eventually it
provides input for an application, where conditional predictions made on
the basis of that model feed into planning and action.
What are the conditions of possibility for success in this use?
Mapping: the cartographer’s art

A map is a graphical representation of geographical or astronomical features,
but this may range from a sketch of a subway system, to an interactive,
     77
zoomable, or animated map on a computer which constantly changes

in front of the eyes.29 The familiar Mercator projection maps come
theoretically under the same heading as perspectival drawing. A 3-D
animation that gives the viewer the sensation of flying over the surface of
Venus does too, but at some considerable remove from the primitive form
we looked at above. The word map’s application has of course spread by
metaphorical and analogical extension. So in addition to the above clear
cases there are representations of much more abstract landscapes, like the
chromosome map of an animal species, the color map showing variations
in hue, intensity, saturation, luminance, shade, and tint, or the site-map
of a website to guide one to locations in virtual reality—not to mention
the map of sociological theory,30 the human intelligence map,31 and the
mediology-theory map.32
The general concept of a map is not so different from that of a model,
though the one is extrapolated from a graph with spatial similarity to certain
features of a landscape, and the other from a table-top contraption. But
there is a significant difference in the connotations about the use for which
they are designed.
A map is designed to help one get around in the landscape it depicts.
What are the conditions, pertaining to map and user, of the possibility of
such use?
A meteorological model The cartographer’s art is as old as philosophy or

science itself—but let us choose a contemporary instance that gives us maps
with bells and whistles. The Aviation Model (AVN) for numerical weather
prediction is not something to display on a table top, it is a computer
program and what it produces as specific individual representations—in
effect, weather maps—are displayed on a computer monitor.
The AVN is one of the older national models in use by the US National
Weather Service, developed mainly to aid in forecasting for aviation. Its
resolution is about 100 km, and it tends to perform better than the other
models in certain weather situations, such as a strong low pressure area near
the East Coast of the USA. It has a forecast product, the Aviation Model
Output Statistics, which provides detailed weather information in three
hourly increments spanning the 48 hours after the Aviation Model is run.
The forecast weather information includes temperature, dew point, wind
speed and direction, cloud cover and obstructions to visibility, probability
78  : 
of precipitation, precipitation form (rain, sleet, freezing rain, or snow), and

probability of severe thunderstorms.
Model output statistics are a blend of raw model output (data taken
directly from the model grid) and statistical analyses of previous forecasts
for the region. A forecast is first taken from the AVN weather model,
but that is then adjusted on the basis of equations that relate the forecast
output of the AVN to what is actually observed at the forecast location.
The result is to skew the model’s forecast output to make it more accurate
for the location for which a forecast is being produced. The final ‘‘skewed’’
product is the AVN MOS product presented as weather forecast.
The AVN itself requires input to be run at all, of course: namely initial
conditions and lateral boundary conditions obtained from operational
weather centers in the relevant area. We can summarize what the model
does in the following form:
The model presents a space of possible states and their evolution over
time—the input locates the weather forecaster in that space, at the
outset of the forecasting process.
Notice what the forecaster, if s/he is also its user, has to provide in order to
create expectations that may be tangibly fulfilled: something to the effect
of ‘‘I am now here in the space of possible states’’.33
Using maps: Kant’s point34

What we have just seen in the case of the AVN, the entry of an indexical
when its product is put to use, is entirely analogous to what someone using
any map has to do to make the map useful for finding his or her way
around. What is crucial to any application is a judgment to the effect ‘‘I am
now here on the map’’.35
Suppose for example that I am lost, and buy a map. Maps normally do
not have an arrow labeled ‘‘You are here’’. But even if the map does have
that, the problem is really the same: I have to locate where I am with
respect to that arrow. Imagine I found this map lying in the gutter. Imagine
instructions about the significance of such labeled arrows: ‘‘If you stand in
front of a map under condition C, then. . .’’. You still need to supply the
indexical premise ‘‘I am in front of this map in condition C’’.
The extra information needed by the user to use the map cannot be
there, already encoded in that map. When I do have that extra information,
     79
I can express it by pointing to a spot on the map and saying ‘‘I am there’’—a
self-ascription of location on the map.
It is not as if there is an object or event that is indescribable, ineffable,
beyond the reach of objective or impersonal description. This act of self-
description too can be described and the information can be included on a
bigger map (with the label ‘‘location of vF’s map-reading at time t’’). But
then what I need to use this new map is still a self-ascription of location
with respect to it. The problem of practical use has not been altered.
With this new map, I can go on to self-ascribe a location by the different
words ‘‘I am vF and it is now t’’. An attempt to replace or eliminate
these self-ascriptions leads to an infinite regress, using an infinite series of
maps. But even given the accuracy of the whole infinite series of maps, the
regress does not succeed in eliminating the need for self-ascription. For I
will still be lost, unless I can locate myself with respect either to at least
one of these maps or to the series as a whole, and this I can only do by
asserting a self-ascription which is not deducible from the accuracy of those
maps.
Besides self-location in this narrow sense of ‘placing oneself ’, orientation
with respect to direction is equally crucial. An example will immediately
help to continue our discussion of measurement from magnitudes (distances)
to less metric features. Imagine being in New York with a city map. You
go on the subway; at one stop you check the station sign, and it says ‘‘18th
Street’’. This allows you to point to a spot on the map, labeled as 18th
Street station on the line. You exit and find yourself on 18th Street, at
the corner of 7th Avenue; but which way is West? Being unfamiliar with
the buildings, and too far from the Hudson to see it, you walk one block
along the street to see the name of the avenue that crosses it there. Is it 6th
or 8th? Now you are oriented with respect to direction too. This involved
two measurements: checking the station sign, or the initial street signs as you
exited, located you in a spot on the map. Walking one block to take a new
sighting was a second measurement which allowed you to orient yourself
with respect to direction.
Indeed, we can take that as our paradigm situation for how we can
draw on science in action or practice. These measurements had as func-
tion to locate and orient you with respect to a ‘public’ representation of
the object or situation of interest. That means: they were the practical,
instrumental means for arriving at those crucial indexical judgments that
80  : 
you needed to make use of the map. Although there are both simpler
and more complex cases of measurement, this case is paradigmatic in
what it reveals about measurement and the use of ‘public’—generally
theoretical—representations.
It was Kant who placed the inevitable indexicality of application center stage
in thinking about how experience, understanding, and reason are related.36
His reflections on this subject came at a crucial stage in his philosophical
development, in his early essay ‘‘Concerning the ultimate ground of the
differentiation of directions in space’’:
No matter how well I may know the order of the compass points, I can only
determine directions by reference to them, if I am aware of whether this order
runs from right to left or from left to right, and the most precise map of the
heavens . . . would not enable me [without this orientation] to infer . . . on which
side of the horizon I ought to look for the sunrise.37
. . . the most precise map of the heavens . . . would not enable me . . . to

infer . . . on which side of the horizon I ought to expect the sun to rise if it
did not, in addition to specifying the positions of the stars relative to each other,
also specify the direction by reference to the position of the chart relative to my
hands. (Ibid.)
There is here a precise and perfect analogy between theory, model, and map. The
point about the map, made already by Kant, applies to a model of any sort.
The activity of representation is successful only if the recipients are able
to receive that information through their ‘viewing’ of the representation.
But what are the conditions of possibility for this reception? The recipient
must be in some pertinent sense able to relate him or herself, his or her
current situation, to the representation.
There is one objection to Kant’s point which cannot be simply dismissed.
Might each location in the cosmos not have a uniquely defining description in
terms that are not in any way indexical or context-dependent? And if so,
could that description not be everywhere substituted by a person at that
location for the words ‘‘I, here, now’’? Could Kant’s thesis not be evaded
in that way, at least in principle? This objection I will take up later, by
connecting it with the attempts to develop a purely structuralist view of
science by Russell and Carnap, and relate it to arguments such as ‘‘Putnam’s
Paradox’’. For now we will look into the implications of Kant’s thesis at a
less abstract level.
     81
GPS, The automated ‘self-locating’ map (?) Does Kant’s analysis still apply,
now that we have Global Positioning Systems (GPS)? GPS satellites transmit
signals, which the GPS receivers receive passively; the receivers do not
transmit data at all—hence no data about their own location. Having such
a receiver we obtain talking, self-locating ‘‘maps’’ that don’t need anything
from us to help them. When installed in an automobile, the system is
designed to show an instinctively recognizable display to drivers allowing
them to see where they are in relation to surrounding topography. In
advanced versions, the interface includes a bird’s-eye view of adjoining
streets and buildings.
So how does this work? The auto receives signals from several GPS
satellites, with data about the satellite’s location and the time of emission.
These signals, sent simultaneously by the satellites, arrive at the receiver at
different times due to the difference in distance. From these differences the
receiver’s location in the coordinate system defined by the satellites can
be determined. The local map coordinates are defined in that coordinate
system, and this map also shows the roads and buildings situated in the
neighborhood of that location.
Well, is anything missing? Yes: the driver has to make the judgment
that the moving point on the map display is his own location. If s/he
does not know how the system works, that is obvious. But even if the
design is understood, a judgment tacitly intervenes; after all, only a slight
malfunction in the computerized calculations could result in a significant
displacement. The case is not so different from a person standing in front
of a map which has on it an arrow with the legend ‘‘You are here’’. That
person has to make the judgment ‘‘I am now at the location indicated by
that arrow’’. This judgment could very easily be false, for example, if the
wrong map had been affixed at that place. Suppose the GPS system makes
the sound ‘‘Right turn, 500 feet’’. I have to supply the judgment that I am
now in the right location with respect to the source of this sound—that it
is not, for example, received as a radio signal from the GPS that is currently
guiding my cousin in his new VW, sent to impress me.
Imagine the automation of ship navigation along similar lines. In the old
days the mate took his astrolabe and reported the elevation and compass
direction for certain stars; then the captain took those data and the data
from the chronometer to the maps on his table. There he plotted the data
and arrived at a location on the map; on the basis of this he issued his
82  : 
orders for the course to be steered. All of this except the very last step can
be automated—the GPS can collect the data that will play the role of what
mate and astrolabe provided before. But if the captain is to issue his orders,
when he sees a cross appearing on the map depicted on his computer
monitor, he has to add ‘‘we are now there, and the ship is currently moving
in that direction’’. The situation has not changed in principle, just become
less labor intensive.
It will perhaps be objected that the captain’s role can also be automated,
assigned to an ‘automatic pilot’, integrated with the GPS. But then, who is
the user? The person who programs the automatic pilot with a course to
steer, and then turns it on—in the full confidence that the map point taken
as input is precisely where s/he is at that moment. So again, the situation
has not changed in principle for the user.
What is in a map?
The short answer is ‘‘Nothing!’’. That is, if we take the physical object
by itself, considered entirely without reference to use, to us, then there is
nothing in it to determine its semantic content. But we can expand on this
point. Even relative to the conventions in force in our community or society, for
pictorial representation of terrain, there is some information—taking this
in a broad sense—that cannot be in the map itself.
So what is in a map? The topic of self-ascription belongs to pragmatics
and not to semantics. That is a fancy way to say that what the self-ascription
does cannot be equated with, but adds a crucial step to, the content of a map
or to the bare impersonal belief that the map ‘‘fits’’ the world.
The bearing of this puzzle: we must generalize this to scientific represent-
ation per se. The body of ‘pure’, ‘fundamental’ science, in the sense of the
totality of accepted scientific information, can in principle be written in
coordinate free, context-independent form. That is possible for theoretical
science, even if it includes the history of the universe or the evolution
of biological species on earth. But to draw on this body of science in
technology—whether in practical applications or even to test it or use it to
explain something, or add to it through research—the scientist or scientific
community must supply something extra, which does not come with that
body of science, but serves to locate the user with respect to it.38
     83
Let me put this again, somewhat differently, in terms of models. ‘‘Model’’

is a metaphor, whose base is the simply constructed table-top model. We
use this metaphor when we talk of cosmological models, Hilbert space
models, and the like. We could have used the word ‘‘map’’, and made
maps the base of our metaphor equally well.
Suppose now that science gives us a model which putatively represents
the world in full detail. Suppose even we believe that this is so. Suppose
we regard ourselves as knowing that it is so. Then still, before we can go
on to use that model, to make predictions and build bridges, we must
locate ourselves with respect to that model. So apparently we need to have
something in addition to what science has given us here. The extra is the
self-ascription of location.
Have we now landed in a dilemma for our view of science as paradig-
matically objective? If we say that the self-ascription is a simple, objective
statement of fact, then science is inevitably doomed to be objectively incom-
plete. If instead we say it is something irreducibly subjective, then we have
also admitted a limit to objectivity, we have let subjectivity into science.
But the threat we sense here takes its force from equivocation. Something
more than what is contained in the printed map, physically constructed
model, or computer monitor display is needed. But what is this ‘‘more’’?
Not a mysteriously different sort of fact which cannot be encoded on a
map! The scientific story can be complete in the sense of describing all the
facts, including that someone does or does not have the ‘‘extra’’ needed for
him or her to draw on a particular bit of science. It is just that describing the
having of it is no substitute for the having!
We will just have to admit a non-pejorative sense of ‘‘subjective’’, if the
essential indexical has to be labeled as something subjective. Whatever sense
we give it, we may be charged with giving a special role to consciousness
in science. But that implication is there only in the premise that there is use
of science. To the extent that use implies consciousness and agency, this
premise does so too—what more need be said?
Self-location in a wider sense

We can broaden the concept of self-location. Suppose I see letters on a
piece of paper. If I take them to constitute a text in my own language, I am
locating myself with respect to what I have before me—to be contrasted
with taking them as a text in another language. In very simple cases there
84  : 
can easily be such a contrast: suppose an Italian speaker sees the inscription
‘‘burro’’. He may well take it as being in his own language, hence a sign for
butter; but he could take it to be Spanish, to refer to a donkey. In a more
complex case, Sherlock Holmes sees what is ostensibly a simple innocuous
message in English but declares it to be a message in code having a quite
different meaning.
At first blush, such inscriptions are far from maps, and what happens
here is far from the use of a map to locate oneself in a landscape. But the
mere understanding of the inscription as a text requires relating it to one’s
own language—either through taking it to be an expression in one’s own
language or through a translation procedure.39
Language may seem too special an example here: it is the seat of meaning
if anything is. But to call, classify, something as a map or a model is to
locate it in what Wilfrid Sellars called ‘‘the space of reasons’’—at least as
this phrase is now broadly understood. By itself this is not yet self-location.
It is just to classify the item as having semantic content, and as having a role
in reasoned discourse and in practices subject to norms of rationality. We
can to some extent separate our understanding of the item, in the sense of
grasping its semantic content, from the understanding of our own situation
that comes with locating oneself ‘in’ or ‘with respect to’ the item. But the
latter comes in train, so to speak.
This applies tellingly to any observation report made (as it usually is) in
theory-laden language. If I say ‘‘Lo, phlogiston escaping!’’ I am locating my
own situation in the logical space provided by phlogiston theory. If instead
on the same occasion I say ‘‘Rapid oxidation happening!’’ I am locating my
situation in a logical space provided by modern chemistry. If Lavoisier had
heard someone shout the former, it would have been helplessly academic
of him to reflect that he was hearing just another false statement. If you
hear me say either of the above, you are well advised to flee the room, if
you take yourself to be with me, that is, if you are inclined to echo my
self-locating act (even if you will echo it in a different-theory-laden form).
Visual perspective and the metaphor

I have resisted the ‘‘perspective’’ terminology in this discussion of self-
location, all the way from the introduction of Kant’s point to Lavoisier’s
     85
predicament in the crowded theater where someone shouts ‘‘Lo, phlogis-

ton escaping!’’. Perspectivity and indexicality are closely related, but the
distinction is of equal importance. The terms ‘‘perspective’’ and ‘‘per-
spectival’’ are central to how we understand scientific representation. For
this very reason it is all the more important to constrain their use, to keep
their meaning from spreading too broadly, though equally important not
to narrow them myopically.
The most literal use pertains to the modes of visual perspective that
became important with the innovations of Renaissance painting. What are
its characteristics? First of all, a visual perspective has an origin—the painter’s
eye in one-point perspective being the paradigm example. Secondly, it has
an orientation: the direction in which this eye or the camera is looking.
Thirdly, the content of this visual perspective is expressible in an indexical
judgment, ‘‘that is how it is from here, looking in this direction, with
that my left, this my right, yonder my above. . .’’. Fourthly, there is the
systematic spatial distortion due to projection, so that lines orthogonal to
the plane, that together with the eye defines the orientation, intersect.
Fifthly, it is subject to the limitations of occlusion, marginal distortion, and
degradation of the grain (coarsening of the threshold of distinction), and
we take the content as belonging to a range of alternatives.
The prevalent use of ‘‘perspective’’ for frames of reference correspond-
ing to geometric coordinate systems can be misleading. But we must not
object too much. This use is a first step metaphorical extension only (in one
direction, I would like to say, if only to illustrate the metaphor). But the
geometric or physical frame of reference thus conceived is a depersonaliza-
tion of visual perspective, relinquishing all but origin and orientation, which
can be arbitrarily chosen. So I think it best to restrain even this extension
of the term, or at least emphasize that it needs to be used with much care.
Now we have come across another use that does not come directly
under the heading of visual perspective. If I stand in front of a map and say
‘‘I am here, looking in this direction’’ (pointing to a spot and indicating a
direction in the map) I am placing myself in a perspective within the depic-
ted landscape. . . . To speak this way is to allow ourselves that term when
the first, second, and third features of visual perspective are there. We have
therefore less reason to object to this use of ‘‘perspective’’ than in the pre-
vious case. Nevertheless it will be important to distinguish the literal visual
perspective of eye or camera from each of them, and they from each other.
86  : 
The question will come up a few times below, with either just origin and
orientation or indexicality at issue, whether to spread these terms still further
in application. For the most part I will resist that; the important relations
between representation and perspective will be obscured if we allow much
broader use. This despite the already current broad usage in everyday
language, which can’t always be resisted without irritating the ear.40
The question is least simple when it comes to verbal description. When
is a description perspectival? When it is a description from a certain point
of view—but ‘‘point of view’’ is subject to the same fluctuations in use
as ‘‘perspective’’. A hallmark we can look for in the case of language
is the occurrence of indexical terms or phrases. Among these I include
demonstratives such as ‘‘this’’, ‘‘those yonder’’, as well as the more obvious
‘‘I’’, ‘‘you’’, ‘‘here’’, ‘‘now’’. But the indexicality is sometimes hidden, not
apparent in the surface structure, so we do not have a perfect criterion
here. On the other hand I would resist calling a description perspectival on
the sole ground that some reference to the individual or communal users’
experience is indispensable to understanding its terms. That would take
in even such ostensibly paradigm ‘impersonal’ examples as ‘‘for any body
of gas, pressure will alter with temperature if volume is constant’’. In the
actual construction and use of specific models in practice, we may discern
a specific perspective. But on the other hand, general scientific theories,
in their ‘official’ formulation, are not perspectival descriptions, and their
models—if we consider the entire range of models for a given theory—are
generally not perspectival representations.41
When it comes to terminology, let’s practice tolerance. Whenever it
matters, we should refer to the specific characteristics themselves, not simply
to the perspectivity of which they can serve as hallmarks. Restraint to the
extent indicated here is especially preferable in a technical philosophical
context, while at the same time it is easy enough to understand the wider
usage so commonly found.
Concluding Empiricist Postscript

From an empiricist point of view, the task that science sets itself is to
represent the observable phenomena. What lessons can we draw now for
the aspirant empiricist philosopher of science ?
     87
The idea of representing phenomena need not, and if practical purposes

are kept in mind, must not be restricted to one of copying. But more
than that, as we have seen, representation useful for particular purposes
will involve selective distortion, and representation is closely involved with
useful misrepresentation. Even when likeness is crucial to the purpose,
we must look for likeness only in respects that serve the representation’s
purpose—and only to the extent that they do so. Given that the aim
of science is to provide empirically adequate theories about what the
world is like, we should conclude that wherever the representation does
trade on likeness, the general rule of selectivity targets the observable
phenomena.
A model often contains much that does not correspond to any observable
feature in the domain. Then, from an empiricist point of view, the model’s
structure must be taken to reveal structure in the observable phenomena,
while the rest of the model must be serving that purpose indirectly. It may
be practically as well as theoretically useful to think of the phenomena as
embedded in a larger—and largely unobservable—structure. This we can
quite easily illustrate with examples in the visual arts as well. In the sciences,
perhaps the longest surviving example is the familiar Zodiac, with the stars
arranged in constellations that are memorable as embeddings in such figures
as the Ram, the Scales, the Fish (Aries, Libra, Pisces). In mathematical
physics the illustrations are easily found, but much more abstruse, as one
sees how useful it is to embed structures to be studied in larger spaces.42
Mnemonics—and perhaps realist philosophies—can then be served by
pictorial glosses on those spaces.
But there are two more points that spring to the eye here. First of all, a
view of science would hardly be empiricist if it ignored measurement, which
is science’s main initial access to the phenomena. Spatial measurement
is explicitly perspectival, and its deliverances relate to scientific models
precisely in the way that visual perspectives relate to physical space.
Second, this turns into a general point: as Hermann Weyl put it graphically,
there will be, even in the most theoretical sciences, an ‘‘ineliminable residue
of the annihilation of the ego’’ to provide the conditions for relating the
theoretical models to specific empirical situations. All the revolutionary
developments in the theories of space and time as well as the upheavals
in atomic physics testify to that. The former brought frames of reference
to center stage, the latter engendered what is in fact precisely called ‘‘the
88  : 
measurement problem’’ as fulcrum for philosophical interpretation. But

thereby hangs a tale. . . .43
Secondly, a view of science would hardly be empiricist if it ignored the
uses of science, as a resource for praxis. How are theories and models drawn
on to communicate information about what things are like, to guide our
expectations in practical affairs, to design instruments and technological
devices, to find our way around in the world ? Here especially enters the
indispensability of indexical judgment, even in contexts where the term
‘‘perspective’’ may not be apt. In the end, as we shall see, that entry of
the essential indexical will prove to be the crucial clue to how science can
depict at all.
PA RT I I
Windows, Engines,
and Measurement
Scientific theories represent how things are, doing so mainly by presenting
a range of models as candidate representations of the phenomena. That
is right, but it is a view of science, so to speak, ‘from above’; it is a
view focusing on what is achieved when it has been achieved. There is a
long journey from the initial encounter with nature to the achievement
of an even temporarily stable representation. That journey is hugely
oversimplified if described—as it used to be, and still sometimes is1 —as
simple collection of data by observation or measurement followed by
invention of theories to account for what was found. Following upon both
recent and not so recent inquiries into scientific experimental practice,
instrumentation, and measurement we can pursue the question of just how
that task of representing the phenomena is achieved, and how it is to be
understood.
A measurement is at the same time a physical interaction and a meaningful
information gathering process. In the examination of measurement in Part I
I already announced the thesis that I mean to argue:
measurement falls squarely under the heading of representation, and

measurement outcomes are at a certain stage to be conceived of as
trading on selective resemblances in just the way that perspectival
picturing does.
The words at a certain stage, perhaps puzzling in their first occurrence, refer
to the end of that long journey from the initial encounter with nature
to the achievement of an even temporarily stable representation. We will
look at this process both from within the journey itself and from above, that
is, from the point of view achieved at its end.
Perspectival drawing provided us with a paradigm example of meas-
urement. The process of drawing produces a representation of the drawn
object, which is selectively like that object; the likeness is at once at a rather
high level of abstraction and yet springing to the eye. While the information
92   : , ,  
about spatial configuration is captured in an invariant relationship that is

quite difficult to formulate in words or equations, it is conveyed to us in
a user-friendly fashion. The example is paradigmatic also in that it displays
so clearly that the representation (the measurement outcome) shows not
what the object is like ‘in itself’ but what it ‘looks like’ in that measurement
set-up. The user of the utilized measurement instrumentation must express
the outcome in a judgment of form ‘‘that is how it is from here’’. And
finally, the coin has another side: it is precisely by a process engendering a
judgment of that form—that is to say, by a measurement!—that any model
becomes usable at all.
But this paradigm example of measurement also hides much, precisely
because of its beauty, simplicity, familiarity, and narrow scope. While
its development was a great theoretical and technological advance in the
history of our culture, it has been assimilated to such an extent that the
advance has become all but invisible. For all forms of measurement, and all
the roles it plays in relating theory to fact, familiarity breeds obliviousness.
To understand what measurement is, we need to examine how systematic
inquiry builds continually on what has been previously achieved, both in
theory and in technological practice, and thus changing the conditions
under which given interactions count as measurements at all, or as meas-
urements of particular quantities. So that journey of inquiry, from initial
steps to stable representation, is to be scrutinized. Nevertheless, this inquiry
will be unproductive unless guided by the sciences’ categorical imperative,
that is, the achievement of theories and models that do adequately, and
sufficiently accurately, represent the phenomena under study.
4
A Window on the Invisible
World (?)1
Constantijn Huygens was ecstatic when he had looked into Cornelis

Drebbel’s microscope and spoke of a ‘‘new theater of nature, another
world’’—he exhorted painters to ‘‘portray the most minute objects and
insects [thus seen] with a finer pencil, and then to compile these drawings
into a book to be given the title of the New World’’.2
The microscope is perhaps the best example of an instrument as an aid to
the senses. Detection by means of instruments is to be distinguished from
observation, in the sense in which I use that term: observation is perception,
and perception is something possible for us, if at all, without instruments.
This simple and commonplace distinction does not, of course, settle any
questions in epistemology. Rather, it raises the further question: how is this
use of instruments to be conceived then? As Huygens does, as a window
through which we can see into another realm? Does Huygens’s reaction to
the optical microscope remain as intuitively tempting when it comes to the
advanced technology now involved in experimentation? Just what is it that
we do by means of those instruments that are so typically taken to disclose
the unobservable?
If we simply put some white powder to the tip of the tongue to check
whether it is salt or sugar, we are making an observation, conducting a prim-
itive experiment, and in effect performing a measurement without instruments.
But we cannot take this simple case as very revealing. In general there is no
such simple relation between observation, experiment, and measurement.
This is in part because of the complexity of the instrumentation involved.
But it is also because measurements occur only as special elements of the
experimental procedure by which objects are deliberately placed in unusu-
al, artificially designed conditions—conditions in which they are made
94  : , ,  
to respond to the questions put to them. That intricate construction of

a well-designed instrumental set-up for experimentation is what we must
inspect first, to understand the intricacies of measurement in general. How
shall we understand it, if not through Constantijn Huygens’s so charmingly
naïve eyes?
Instrumentation’s diversity of roles

Instrumentation plays many roles, and we cannot place all its roles in
experiment and other laboratory uses under one simple heading. The
sorts of measurement that I have brought up so far as examples involve
instruments in what Michael Heidelberger classifies as the representative
role.3 This role the instruments have in relation to a theoretical con-
text: ‘‘the goal is to represent symbolically in an instrument the relations
between natural phenomena’’ (ibid.). The examples he gives are clocks,
balances, and measuring rods. An oscilloscope is also illustrative of this
role. Relative to a theory much more technologically advanced instruments
can play this role as well. The Scanning Tunneling Electron Microscope
used in nano-technology provides a good example of such an advanced
representative role, which can be understood as such—that is, as playing
this role—only relative to theory. But a caution is in order: given the
need for theory as intermediary, we must remind ourselves that repres-
entation need not be pictorial representation; and when the results are
presented in pictorial form there is a question as to how they are to be
taken.4
Instrumentation has several more roles. Just which role given instruments
are playing in a particular experimental set-up is a non-trivial question,
the answer to which is also typically theory-laden. Heidelberger points
to an imitative role, played when instrumentation produces phenomena,
in controlled artifacts, meant to mimic effects ‘‘as they appear in nature
without human intervention’’. As example he mentions an apparatus used
to simulate production of an enzyme in an organism. A fountain, so ori-
ented as to produce a ‘‘rainbow’’ when there are no rain clouds, plays that
role too. So did Otto von Guerrike’s electrical generator to produce imit-
ation lightning. Galison and Assmus tellingly—and surprisingly—locate
Wilson’s early development of the cloud chamber in a tradition of mimetic
      ⁽  ⁾ 95
experimentation in nineteenth-century science in geology and metereology

(see Galison and Assmus 1989).
When a phenomenon created artificially is taken to imitate a naturally
occurring phenomenon, a substantial theoretical claim is involved. The pur-
pose of mimetic experimentation is to learn something about the natural
phenomena, such as lava flow, lightning, rainbows, or cloud formation by
studying an artifact constructed in tightly controlled circumstances. This
presupposes that the two are indeed alike in relevant respects, and these
respects may not at all be entailed by visual or other noticeable similarities.
More accurately: it is sub specie the eventually accepted theory that they are
relevantly the same—the working hypothesis that they are the same is
posited in aid of that eventual, hoped for, theoretical achievement.
Similar to, but more remarkable than, such purported imitation is the use
of instruments ‘‘to produce phenomena that normally do not appear in the
realm of human experience’’, called productive by Heidelberger, manufacture
by Boon.5 Those may indeed have never occurred before—or occurred
in that form—anywhere in nature. Using his electrical generator, Von
Guericke found that he could make a sulfur ball glow. While electrolumin-
escence appears in nature under very different circumstances, the strange
happening that involved a relationship between luminescence, rotation,
friction, and sulphur was a new phenomenon.
In such examples it is not unnatural, even if sometimes confusing, to
speak of discovery. A new phenomenon is produced, but the important
news is that it occurs, and putatively always occurs, under certain gen-
eral conditions, which may also be realized in nature—if that this is so,
then that is a discovery. This is the sense in which we say, for example,
that Faraday discovered electro-magnetic induction. But Faraday’s own
description gives us a different impression of the work, when he presented
these results to the Royal Society in 1831. There Faraday described first of
all the induction of momentary currents in a wire when he connected or
disconnected an adjacent wire to a battery, or when he moved the wire
near the battery (‘‘volta-electric induction’’) Then he described the further
effects obtained by inserting iron in the helices of wire, and how currents
were induced by the movement of magnets brought nearby (‘‘magneto-
electric induction’’).6 Thus his report does not have the ‘discovery’ form;
it is a report of the means by which he created the phenomenal effects he
reports.
96  : , ,  
All three roles, representative, imitative, and productive, are played by

instrumentation in experimental, observational, and modeling work in the
laboratory.7 Following upon this taxonomy of roles, we must nuance the
account of experimentation accordingly. But first a strong caveat: these
three roles are easily confused, the experiments and instrumental procedures
can easily be misconceived, or conceived in plausible but critically dubit-
able ways. We need to distinguish them carefully and resist easy or familiar
categories before we can properly locate measurement in this technological
plethora.
‘‘Observation by instruments’’: our bewitching metaphors
Strange words simply puzzle us; ordinary words convey what we know already;
it is from metaphor that we can get hold of something new and fresh. (Aristotle,
Rhetoric 1410b10–13)
Two metaphors guide our thinking about instrumentation in experimental

and other scientific inquiry. The first is that of a window on the invisible world:
an instrument opens a window for us into a world beyond the directly
observable. In the opening quote from Constantijn Huygens we saw him
taking the microscope to alter the observable/unobservable boundary.
Whether we follow him in this, whether we extend the meanings of words
like ‘‘see’’, ‘‘perceive’’, ‘‘observe’’ to ‘‘see through a microscope’’ and the
like, or whether we refuse this extended usage (as I do), does not affect our
sense of novelty. In either case, there is a significant extension, of some
sort, to empirical inquiry. If we grant that these instruments function as
windows into realms not accessible directly to the senses then we are directly
confronted in experience by something that lies beyond our feeble reach. If
we do not regard them as playing this ‘window’ role, we must still do justice
to their novelty; and here another prevalent metaphor comes into its own.
The second metaphor is well conveyed by saying that our instruments
are engines of creation. They create new observable phenomena, ones that
may never have happened in nature, playing Heidelberger’s productive
role, only sometimes to imitate nature but always to teach us more about
nature.8 Once a new phenomenon has been created, it takes its place in
nature—for we and our efforts are part of nature. Those new phenomena
are themselves observable, and become part of our world. So they become
part of what science is meant to ‘save’.
      ⁽  ⁾ 97
These are valuable metaphors, each aptly guiding some of our thought
about instruments. They do not exhaust the subject: somewhere in between
the two lies instrumentation conceived of as creating copies, scale models,
likenesses, or ‘significant similarities’ of natural phenomena on the scale of
the directly visible. Peter Dear (1995: 159–60) gives as striking example
William Gilbert’s work on the magnet in 1600. His artifact is a spherical
magnet made on a lathe from a lodestone. He calls it a ‘‘little earth’’, a
terrella, and while we spontaneously read this as a metaphor, Gilbert makes
it clear that he means this very literally: the earth is a spherical magnet, and
what he has created here is the very same thing on a scale where we can
handle it.9 The instrumentation is here conceived of—in terms that echo
Dear’s own distinction—as having a mimetic function, to be contrasted
with ‘‘the semiotic, focusing on signs and representation’’. This is close to
the ‘‘windows’’ metaphor, in that the set-up is claimed to show ‘‘what
things are really like’’, but also close to the ‘‘engines’’ metaphor, because
hinging on the production of novel effects. So here the two metaphors
overlap.
We can speak aptly of representation in all three cases, and the bound-
aries between the three are blurry, in act if not in conception. In the
case of optical instruments we can readily see all three illustrated. Firstly,
the images produced by lenses are copies, literal likenesses, if the lenses
function as windows upon their domain. Secondly, whether or not that
is so, the images produced by lenses are themselves (artificially produced)
phenomena, that have seen their uses in ancient magic and ritual as well
as in modern science. Thirdly, phenomena instrumentally produced may
in some cases be scaled-down or scaled-up versions of naturally occurring
phenomena that are too small or too large to be grasped in percep-
tion, or be asserted to be ‘significantly similar’ to postulated processes in
nature.
But there is one very important disconnect between the second concep-
tion and the other two. Creation of new phenomena is not in general a case
of mimesis. It may indeed provide information about ‘natural’ phenomena,
but need not be doing so by presenting us with a likeness. Instrumentation
in science, both in experimentation and in application, has many roles.
One salient role that I want to emphasize here, that is important as different
theoretical leanings compete, is the production of new phenomena that all
theories in a given domain must account for if they are to compete successfully
98  : , ,  
there at all. That role is played by all that is novel in inquiries into nature
by means of instrumentation. Paradoxically, that is precisely the one that
tends to favor the ‘window’ metaphor, although it is most clearly the role
in which the instruments are engines of creation, not channels for passive
observation!
What better way to challenge a rival theory than to produce a new
phenomenon to be accounted for? The new phenomena are not created for
nothing or with no consequences: theory must submit itself to their tribunal.
There are many famous examples of such creation of new phenomena to
play this role. Hertz himself reported on the importance of his experiments
in electromagnetism in this way:
What we here indicate as having been accomplished by the experiments is
accomplished independently of the correctness of particular theories. Nevertheless,
there is an obvious connection between the experiments and the theory in
connection with which they were undertaken. Since the year 1861 science has
been in possession of a theory which Maxwell constructed upon Faraday’s views,
and which we therefore call the Faraday–Maxwell theory. This theory affirms
the possibility of the class of phenomena here discovered just as positively as the
remaining electrical theories are compelled to deny it. (Hertz 1962: 19)
The well-known story of Poisson’s challenge to Augustin Fresnel’s prize
essay furnishes a ready example (even if not generally told with perfect
historical accuracy).10 Fresnel submitted his essay for a scientific prize
competition on the diffraction of light, basing his analysis on the wave
theory. Poisson, a member of the panel which evaluated the essay, pointed
out that Fresnel’s analysis had a strange consequence. It implied that the
center of the shadow of a circular disc would have a bright spot. This
consequence contradicted the Newtonian corpuscular theory of light, and
seemed in any case highly unlikely. But Dominique Arago, the panel’s
chairman, did the experiment and the bright spot appeared! Despite
himself Poisson had designed an experiment that created an observed
phenomenon—new in what these scientists on the panel had seen, though
not new in nature—which only the wave theory was (at that point) able
to explain, and which was entirely upsetting to its great rival.11
The ‘‘window into the invisible world’’ metaphor has dominated modern
philosophical thinking about science as much as the ‘‘mirror of nature’’
metaphor dominated modern epistemology and metaphysics. It will serve
us better to dislodge or at least weaken its grip on our philosophical
      ⁽  ⁾ 99
discourse, and to think of experimentation in terms of a literal enlargement

of the observable world, by the creation of new observable phenomena, rather
than a metaphorical extension of our senses.12
The microscope as window on the invisible world

The contention that our epistemic situation changed at a certain point, just
because of instruments of observation, seemed to come along naturally with
the construction of those instruments, though the idea was then as new
as the instruments themselves.13 The technological change started after
the Renaissance with the development precisely of optical instruments:
the microscope and telescope. Catherine Wilson (1995: 57–65, 218–20)
describes how enthusiastic the English scientists of the seventeenth century
were about this achievement.
One writer on the microscope, Charleton, remarked that of course
the ancients, lacking optical instruments, would have greeted the atomic
hypothesis with skepticism—but our situation is quite different! Robert
Hooke writes in the preface of his Micrographia (1665) that with the aid of
the microscope we will discover:
the subtilty of the composition of Bodies, the structure of their parts, the various
texture of their matter, the instruments and manner of their inward motions, and
all the other possible appearances of things . . . . we may perhaps be inabled to
discern all the secret workings of Nature, . . . manag’d by Wheels, and Engines,
and Springs . . . .
And Hooke’s contemporary Henry Power’s Experimental Philosophy (1664)
says in his preface:
We might hope, ere long, to see the Magnetic Effluviums of the Lodestone, the
Solary Atoms of Light . . . , the springy Particles of Air, the constant and voluntary
motion of the Atoms of all fluid Bodies, and those infinite, insensible Corpuscles
which daily produce those prodigious (though common) effects among us.14
Notice that he takes the optical microscope to reveal even the composition
of light! Besides this radical misunderstanding of what the microscope can
do, several other errors both of fact and of principle would soon become
evident.
If appearances are what appear to us then, by definition, we never do
see beyond the appearances . . . ! This insight, clear enough in Locke and
Berkeley in the next century, could be the slogan for our entire discussion.
100  : , ,  
When caught unawares we do still spontaneously think about it in the

same way as those seventeenth-century enthusiasts, however. We simply
tend to add into this family of windows such devices as electron micro-
scopes, spectroscopes, MRI scanners, particle accelerators and so forth. In
fact, each of these devices creates new phenomena, truly humanly observable
phenomena. Whether or not one subscribes to the idea that these devices
extend the range of our vision, it is indisputable that they serve for the
systematic creation of new phenomena—new phenomena that must also
be saved by our theories, suffice to refute theories to be discarded, and
serve to gather empirical information.
Engines of creation: engendering new phenomena

Even for the optical microscope I am offering a change in view, by favoring
and emphasizing the ‘‘creation’’ metaphor over the ‘‘window’’ metaphor.
Though valuable as an heuristic guide, to take the ‘‘window’’ metaphor
literally acts as a brake on the possibilities for interpretation. Suspend for
a moment the question whether any of these instruments provide access
to the unobservable. All those instruments which purport to represent the
invisible mimetically, can be thought of, not as windows into the nether
world, but as productive experimental arrangements. As always, interpretative
options allow for alternatives. Even the clearest examples of instrumentation
that play a productive role can, conversely, also be reconceived as imitative.
The possible conflation does not lie along a one-way street. But the one
function we can all agree on for instruments used in science—for it is a
function served either way—is that of creating new observable phenomena
to be saved.
Consider the nineteenth-century ether theory that brought light under
the canopy of electromagnetism. Following Faraday’s early experiments,
place iron filings on a sheet of paper above an electromagnet; then turn
the magnet on. Suddenly the filings arrange themselves in a pattern of
lines of force. What happened? According to the theory, when you pressed
the switch, you affected the state of the very fine all-pervasive medium,
the ether. This new state is manifest in the arrangement of the filings.
In that ether theory the phenomena of electricity, magnetism, light, and
illumination are assimilated to the same pattern of explanation. Turning
      ⁽  ⁾ 101
on the light affects and alters the state of this ethereal medium. This
altered state is manifest in the illumination and visibility of the objects in
this room.
From a realistic point of view you can describe Faraday’s device as
a detector for the electromagnetic field, as a window allowing us to
look upon those mysterious lines of force in the ether. In theory it is now
certainly of a piece with optical phenomena. At the phenomenological level
it was at the time of a piece with the well-known visible phenomena of
magnetic attraction. The behavior of lodestone and iron had been recreated,
artificially produced—in a like yet different form—in the laboratory. But
look, we have here also a phenomenon that had never occurred before in
history: pressing one object and thereby instantaneously altering a ‘dust’
distribution on another object into such a regular pattern as this. Faraday
had created a new phenomenon. But this event had at once great theoretical
value—for it was a controlled phenomenon so closely related to known
natural phenomena of electricity and magnetism that it was required to be
accounted for in any future theory on offer in this area of physics.
Faraday had constructed an engine to produce a fascinating variety of
new phenomena. The story continues. When representing the distribution
of electric charge as analogous to what happens when iron filings are
scattered in the vicinity of a bar magnet, Faraday and Barlow displayed
the change in the position of the filings as a continuous function of their
position relative to the poles of a magnet. The examples can be multiplied:
brought to light are salient regularities in the new observable phenomena
that theory must account for.
The microscope’s public hallucinations

I submit that the analogous point is to be made even for the simplest optical
microscope. Microscopes too create phenomena, to be accounted for by
our theories, they too can be conceived of as engines of creation rather
than windows upon the invisible world.15 To say what they create, I must
direct attention to a special sort of phenomenon that nature also produces
spontaneously. The main examples are certainly—but not only—optical
phenomena. They include reflections in the water, mirages in the desert,
and rainbows.16 The subject is ancient: Plato’s characters in the Sophist
102  : , ,  
struggled with the questions of how or in what sense reflections in water

are real or unreal, and Aristotle began the theory of the rainbow in his
Meteorology.
The first point about these, as about light itself, is that we use nouns. In
fact we use count nouns; we talk about all of them as if they were things.
The second is that the phenomena themselves show that we are wrong to
do so. They refuse to allow us to represent them to ourselves as things, or
even as properties of things in any straightforward way.
Consider the rainbow. We realize pretty soon that there is no real
material shining arch standing above the earth, although at first it looks that
way. As a second guess we might think that certain parts of the clouds or
haze are colored. But that cannot be maintained because if we move, we
see the rainbow in a different location on that cloud or haze background.
Figure 4.1. Geometry of the Rainbow
In fact, we realize then that our usual way of speaking involves us in

falsehoods. I see a rainbow and you say you see it too. See what too? How
can you be seeing the rainbow I see, when yours is located in a different
place? Nor are they simply in a different place in our respective visual
fields, in the way the clouds are. For if that were so, we would see the
colors ‘attached’ to the same part of the cloud, modulo parallax. If on the
other hand I say there are two rainbows, and you agree, we are not even
counting the same things. In fact, we are not counting things at all.
      ⁽  ⁾ 103
But thirdly, we are not hallucinating. Hallucinations are private, sub-

jective. These rainbow observations are like hallucinations, in that they
are not of real things. But they are unlike hallucinations because they are
public. Nature creates public hallucinations. So public, in fact, that the camera
captures them as well! The observations are scientifically significant in part
because they can also be made indirectly, so to speak, with the camera as
instrument.
Let me put all this a bit more technically. In some sense we do represent
shadows, moving spots of light, reflections, mirages, and rainbows as things.
But the phenomena themselves refuse to let us maintain this representation.
The reason is that certain crucial invariances are lacking. If the rainbow
were a thing, the various observations and photos would all locate it in
the same place in space, at any given time. However, there are significant
invariants here that dreams and after-images lack. In the case of rainbows
this invariance is found in the relations between sun, cloud, and location
of the eye or camera. The subtended angle is always 42 degrees, with that
location (of eye or camera) between sun and cloud. The larger situation
exhibits invariances that allow us to represent it—the situation taken as a
whole, not the rainbow!—to ourselves as a structure in nature independent
of our subjective experiences.
Just to prevent a possible confusion, let us note a difference between
two differences. Think of a tree by a pond, reflected in the water. There
is a difference between a reflection and a rainbow that makes the one a
‘picture’ of something real and the other not. That difference is not the
difference between a real thing and an image. Different observers also do
not see the reflection in the water in the same place. For the reflection
too, the important invariance lies in the relation between light source,
water, and eye or camera. But there is a further factor in this invariant set
of relationships: the tree. The rainbow and reflection are alike in being
consistently mis-classified as things. But they are unlike in that we can
classify the latter, and not the former, as ‘copy’ or ‘picture’ of a real object
or event that it resembles in a certain respect.17
A catalogue of images
Nature creates these public hallucinations. Already in ancient times, concave
mirrors and lenses were used to do the same, namely to create (artificial)
public hallucinations. Even a concave mirror and a lighted candle can
104  : , ,  
produce a ‘stand alone’ image that is under some viewing conditions as

‘real’ as a hologram.18 It took more than a thousand years before this art
of imitating nature provided the resources for a systematic exploration of
nature.
That is how I suggest we should understand the microscope as well. Van
Leeuwenhoek did with his lenses essentially the very same sort of thing
as Newton did with his prisms—namely, imitate the ability of nature to
create public hallucinations. There are objections to this, especially for the
simplest cases, such as van Leeuwenhoek’s magnifying glass or the optical
microscopes we remember from high school biology classes. I will take
those up below; let’s try for a systematic overview first. For images I want
to display a division into several kinds and sub-kinds. For mnemonic ease
I’ll supply a diagram, with an elaboration on its labels.
Graven Public Private

Images Hallucinations Images
‘‘COPY’’- NOT
QUALIFIED ‘‘COPY’’-
QUALIFIED
painting reflection rainbow after-image

photo shadow mirage dream
sculpture fata morgana hallucination
〈microscope
image〉
Figure 4.2. Image Categories
(Graven Images) On one side are the images which are in fact things, such
as paintings and photos.
(Private Images) On the other extreme are the purely subjective ones
like after-images, dreams, and private hallucinations. These are personal,
not shared, not publicly accessible. Indeed, we are pretty clearly dealing
      ⁽  ⁾ 105
there with discourse that reifies certain experiences which are ‘‘as if ’’ one
is seeing or hearing.
(Public Hallucinations) In between these two are a whole gallery of images
which are not things, but are also not purely subjective, because they can be
captured on photographs: reflections in the water, mirror images, mirages,
rainbows. For those I will use the term ‘‘public hallucinations’’.
Some of these public hallucinations are actually ‘‘of ’’ real things: e.g. the
reflection of a tree in the water. When you see the reflection of a tree in
water you are not seeing a thing; a reflection is not nothing, it is something,
but it is not a thing, not a material object.19 That is clear enough because of
the way it moves when you move, quite differently from, say, a log floating
in the water. Some public hallucinations are not ‘‘of’’ real things; e.g. the
rainbow. But of those which are not, some—only some—would still lend
themselves to being conceived of or identified as pictures of real things. If
an image would so lend itself I’ll call it ‘‘ ‘copy’-qualified’’ (following the
Sophist’s distinction between making copies and creating appearances). But
of any ‘‘copy’’-qualified image we can still ask: is it really of something real,
or is it not? That is always a question of fact transcending the experience
itself.
Objections to this view of ‘observation by

instruments’
There are two sorts of immediate objection to seeing all such instrument-
ation used in experimental inquiry as productive rather than mimetic.
The first is theoretical, and points to the basis in the data obtained for
inference to real entities ‘perceived through’ the instrument. The second
is phenomenological: looking into a microscope just does not give us the
experience of seeing an image, it is phenomenologically just like looking
through a window or telescope.
Let us take the phenomenological objection first; it has been made
strongly and tellingly by Paul Teller (2001b): if you have your eye glued to
the microscope you do not have the experience of seeing an image. You
have the experience, for example, of seeing paramecia on a slide, not of
seeing paramecia-images. This supports the conclusion that when we look
106  : , ,  
into a microscope we are seeing the things that we experience ourselves as

seeing—in fact what is seeing except detecting through visual experience?
I agree to the phenomenology, but not to the conclusion.
Much instrumentation that offered us the opportunity for such new
experiences—so experientially like looking through windows—has tended
to disappear from scientific use. Output from instrumentation tends now
generally to be publicly perceivable print-outs or computer displays. We
don’t tend to think of this as introducing a loss; on the contrary—so we
should not attach too much importance to this change. The microscope’s
output can be sent into a scanner which transmits to a computer or
projector—then we see the paramecia on the wall or the monitor. We
are having a different sort of experience then, for we say after only a little
urging that we are seeing an image.
Even when the working of the simple optical microscope is explained,
the instructor will set up lenses and show how images are optically
produced; no need to pay much attention to the eye or its physiology for
this explanation to be quite satisfying. So it seems, surely, that nothing
about the status of the microscope or its deliverances can follow from
concentration on one of these experimental arrangements to the exclusion
of the others.
But I want to add to this demurral some further points about the
phenomenology as well. Just how different is the experience of looking
through a microscope from looking at a rainbow? First of all, no more
do we have the experience of seeing a rainbow-image when we, as one
says, see a rainbow. Secondly, we must be very careful about how we take
such expressions as ‘‘have the experience of seeing an image’’. That sort
of assertion has the same surface form as ‘‘have the experience of seeing
a boat’’—but remember, images are not things, and be wary of drawing a
parallel between the two propositions.
Don’t we, it may be objected, sometimes have the experience of seeing
an image, as distinct from seeing a real thing, and aren’t those kinds of
experiences phenomenologically different?
Yes, we do, and they are. Recall Macbeth’s ‘‘Is this a dagger which
I see before me?’’, reporting on his experience which is like seeing a
dagger although there is no dagger there. There was literally nothing—no
real thing—that he saw.20 That does not contradict his having had the
experience. So the question that we really have to answer is this: Under
      ⁽  ⁾ 107
what conditions do we correctly classify someone’s experience phenomen-

ologically as the experience of seeing an image of something? I envision three
cases:
Assertion: ‘‘S/he is having the experience of seeing an image
of X’’.
Cases:
Graven Images. S/he is seeing a real thing that we classify as a ‘picture’
of an X. (We have a case in point not only in a display
of paintings or statues but also if e.g. a microscope is
hooked up to a projector or monitor and s/he sees
the screen optically transformed into a picture).
Illusion-A. S/he judges that s/he is seeing a real X, while we take
him/her to be having an illusion or hallucination,
whether private or public.
Illusion-B. S/he judges that it is as if s/he is seeing a real X while
s/he herself takes that to be an illusion.
There are certainly phenomenological differences between these three sorts
of cases, as well as between them and cases of ‘real’ seeing. And the type
Illusion-B could be subdivided further: s/he might still be either right or
wrong in her judgment that it is only as if s/he is seeing a real X. We
might sometimes have to add: s/he is having the experience of seeing an
X-image, but is mistaken: what is actually happening, contrary to what she
thinks, is that s/he is seeing a real X. Sometimes one can’t be sure whether
something is illusory or not. The question of how to classify it becomes a
factual question, and we may or may not be content to remain agnostic
about it.
Briefly then, ‘‘seeing as an image’’ is a code for a classification of exper-
iences that refers both to the spontaneous judgment on the experiencer’s
part, and to what is really happening to that person.
What then, precisely, is Teller pointing out about looking into a
microscope? In the experience he describes he spontaneously judges that
he is seeing e.g. real paramecia, and that he has no inclination to correct
that judgment as illusory. He is quite rightly contrasting that sort of
experience with such experiences as those of seeing rainbows, mirages, or
reflections when they are not delusory. In other words, he contrasts it with
Illusion-B, namely with experiences in which the spontaneous judgment
108  : , ,  
includes a classification of that very experience as a misrepresentation or (public or

private) hallucination. This genuine phenomenological difference pertains,
however, in the first instance not to what is really happening to him, but to
his response to what is happening to him. That is, if we describe experience
as having two sides (what is happening to one that one is aware of, and the
judgment expressing what one takes that to be) it pertains not to the first
but to the second side of experience. So questions as to the first side—e.g.
the question whether the experience of ‘seeing’ in a microscope is or is
not a public hallucination—is not settled by his report. For that further
question is a theoretical question about what happens in a microscope.
Turning then to the theoretical objection: it is certainly the case that
we can represent the images produced by the microscope to ourselves as
images of real things (with the same structure as those images). In fact,
we can represent what we see as indirectly observed real things behind
the microscope’s lenses.21 Part of the reason, as Ian Hacking and Wesley
Salmon pointed out in different ways, is that there are strong correlations
between what the various instruments bring to light (so to speak) when
trained on the same microscopic objects. We are invited, in effect, to
engage in inference to ‘the best explanation’ or to ‘the common cause’,
so as to posit unobservable objects or events to explain those pervasive
correlations.
I agree with the core of their argument but not its conclusion. Having
argued so often against the epistemic status of such patterns of inference,
I prefer to leave that part of the issue aside here.22 The similarity I am
pointing out to the rainbow (and which I propose as basis for a possible way
of thinking about microscopes) is not a similar lack of invariances that could
be interpreted as invariances in possible relations to real objects. The success
of that family of instruments (microscope, electron microscope, radio
telescope) does indeed derive in part from the possibility of representing
their products as images of real things existing independently of any
relations to those instruments. The similarity I want to point to is just this:
their products are images; they are optically produced, publicly inspectable
images. It is these images that are like the rainbow (they cannot themselves
be represented as independent things).
The difference, that we cannot think of the rainbow as an image of a
real arch, while we can think of the microscope image as of a similarly
structured object, is important but irrelevant for my present argument. The
      ⁽  ⁾ 109
point I am making is that the microscope need not be thought of as a

window, but is most certainly an engine creating new optical phenomena. It
is accurate to say of what we see in the microscope that we are ‘‘seeing an
image’’ (like ‘‘seeing a reflection’’, ‘‘seeing a rainbow’’), and that the image
could be either a copy of a real thing not visible to the naked eye or a mere
public hallucination. I suggest that it is moreover accurate and in fact more
illuminating to keep neutrality in this respect and just think of the images
themselves as a public hallucination.
But secondly, what are the practical implications? To keep neutrality
in this respect does not prevent us from gathering empirically attestable
information by means of the microscope, or to base e.g. medical advice on
what the microscope shows us. We should recognize that a false contrast
is made if we oppose ‘‘merely’’ producing images to producing something
informative about the objects with which the instruments are placed in
interaction.
Specifically, the important correlations between the output of different
instruments in similar situations, and between the outputs of the same
instrument in a situation subject to variation, do not need to be explained
by unobservable external causes in order to be intelligible or useful. We
are indeed able to change the images produced in predictable ways, in the
case of e.g. microscopes, other than by changing our own relation to the
instrument. This ability derives from our latching onto significant regularities
in the phenomena, including now among these the objects and events under
study, the instruments, and the relations between the two.23
Hard and soft core science

What I have proposed here is contrary to the way microscopy and similar
instrumentation are usually presented, and it is easy to imagine further
misgivings. When a new physical, observable artifact, a radiograph or cloud
chamber track, is produced the rationale is clearer: we have the newly
produced artifact, and why do we say that it represents something else,
except sub specie a certain theory? But in the case of the simple optical
microscope we have no such physical artifact, we only have the images.
But images are observable phenomena, even if they are not objects, as
their capture on film shows. There is some linguistic regimentation or
articulation to be done: the statement, ‘‘There is an image’’ stands in for for
a long description of a set-up in which certain physical phenomena—such
110  : , ,  
as blackenings of photographic film—will happen. Without stretching

ourselves very far, we can report on our sightings made by means of a
microscope in the same way as we report our rainbow-observations.24
Rainbows are not objects, events or processes. Our use of the noun
hides from view the more sophisticated understanding that we’ll now
immediately display when pressed. But our common way of speaking has
not actually changed, and it need not change, it is just fine for all practical
purposes—it is just fine FAPP, as they say in philosophy of physics.25 I think
we can relate to our experiences with microscopes in the same way. Our
common way of speaking, about seeing things through the microscope,
need not change, as long as it is properly understood. Confusions of the
two need to be kept at bay. Specifically, in any ordinary way of speaking
it is not correct to say that we have the experience of seeing a rainbow-image,
nor of a paramecium image. In ordinary language the correct report is that
we have the experience of seeing a rainbow and paramecium. As long as
ordinary discourse is not filtered through some theory it does not imply
that those are objects.26
What about the observable/unobservable distinction then? The main
points of our discussion are not much affected by just where precisely the
line is drawn. I draw the line this side of things only appearing in optical
microscope images, but won’t really mind very much if you take this
option only, for example, for the electron microscope. After all, optical
microscopes don’t reveal all that much of the cosmos, no matter how
veridical or accurate their images are. The empiricist point is not lost if the line
is drawn in a somewhat different way from the way I draw it. The point would
be lost only if no such line drawing was to be considered relevant to our
understanding of science.
Personally I see such a line appearing in a number of contexts, not solely
in the debate over theoretically postulated entities. Here is a quote from
Steven Weinberg, in his 1998 review of Kuhn’s work:
It is important to keep straight what does and does not change in scientific
revolutions, a distinction that is not made in [The Structure of Scientific Revolutions].
There is a ‘‘hard’’ part of modern physical theories (‘‘hard’’ meaning not difficult,
but durable, like bones in paleontology or potsherds in archeology) that usually
consists of the equations themselves, together with some understandings about
what the symbols mean operationally and about the sorts of phenomena to which
they apply. Then there is a ‘‘soft’’ part; it is the vision of reality that we use to
      ⁽  ⁾ 111
explain to ourselves why the equations work. The soft part does change: we no
longer believe in Maxwell’s ether, and we know there is more to nature than
Newton’s particles and forces.
The changes in the soft part of scientific theories also produce changes in our
understanding of the conditions under which the hard part is a good approximation.
But after our theories reach their mature forms, their hard parts represent permanent
accomplishments.(Weinberg 1998: 50)
Weinberg is not exactly known for sympathy with philosophy.27 But I

can submit a challenge to philosophers, in Weinberg’s terms: if you are going
to distinguish between a hard and soft part of science, in some such way, tell us how
you draw the line.28
You can’t get out of this by pointing out that there is a continuum
on which the line is drawn, or that the line will be drawn differently in
different contexts, historical or social. For that is the case for all or almost
all distinctions we make, and does not make those distinctions unreal or
unimportant for understanding.29
Experimentation’s diversity of roles

In distinguishing the various roles of instrumentation in experiment we
automatically induce a corollary taxonomy of experiments, distinguishing
for example mimetic experimentation from the creation of new phenom-
ena. But there is another way to classify experiments, namely in terms of
the roles they play in the development of high level theories.
Philosophers of science, unlike philosophers of technology or historians
of science, used typically to concentrate on the function of experimenta-
tion as hypothesis testing. The experimenter reads over the theoretician’s
shoulder, and designs experiments to test whether the theorist has not gone
too far and made the theory empirically inadequate. This characterization
is simple and appealing. It is regrettably over-simple for many reasons, by
now amply exposed in the literature, although sufficiently many famous
examples illustrate it to explain the traditional philosophical fascination.
In contrast to the hypothesis testing role, there is another function of
experimentation, generally also described in the language of discovery,
but actually an essential ingredient in the joint evolution of experimental
112  : , ,  
practice and theory. We may describe it as theory writing by other means.

One example is Millikan’s work on the charge of the electron. Popularly
presented, he measured the charge. More accurately, he found the charge
to be assigned to the electron in the then developing atomic theory, by
measurements made on the motion of oil droplets drifting down in the air
between two plates, which he could connect and disconnect with a battery.
By friction with the air, the droplets could acquire an electrostatic charge;
and Millikan observed their drifting behavior, calculating their apparent
charges from their motion. After some ‘smoothing’ of the data. naturally,
he could point to a number of which all apparent charges were integral
multiples, and offer that as the charge of a single electron.
‘‘That number was something he discovered. Remarkably the number of
such charges per Faraday equals Avogadro’s number! No one could have
predicted that!’’ This way of announcing his achievement certainly sounds
as if by carefully designed experiment we can discover facts about the
unobservable entities behind the phenomena.
But this makes it also sound like a fortuitous discovery, as if Millikan
had stumbled on it. Quite to the contrary, Millikan approached the
experiment within a developing theoretical context. Instead we should see
the experiment as able to fill a blank in the theory then under development.
The theory is written, so to say, step by step. At some point, the principles
laid down so far imply that the electron has a negative charge. A blank is
left for the magnitude of the charge. If the theory is to be continued at this
point, there are two ways to go on. The theorist could hypothesize a value,
ask the experimentalist to design a test, offer a second guess, test again. But
that is not how it goes in such a case: the experimental apparatus writes a number
in the blank. What I mean is: in this case the experiment shows that unless a
certain number (or a number not outside a certain interval) is written in the
blank, the theory will become empirically inadequate. For the experiment
has shown by actual example that no other number will do; that is the
sense in which it has filled in the blank. So regarded, experimentation is
the continuation of theory construction by other means. Recalling the famous
Clausewitz view of war and diplomacy, I call this the ‘‘Clausewitz doctrine
of experimentation’’. It makes the language of construction, rather than of
discovery, appropriate for experimentation as much as for theorizing.
One way in which things may go badly for a theory is if the numbers
placed in various blanks by such experimental procedures do not bear the
      ⁽  ⁾ 113
relations that the theory says they should. So when Perrin showed that
about a dozen experimental ways of ascertaining Avogadro’s number all
gave the same result, his work gave atomic theory a stability that it could
not have attained if the results had been different. Similarly for Thomson’s
multiple ways of measuring the electron’s mass to charge ratio.30 In these
two cases too, the experiments were approached within a developing
theoretical framework that needed consistent empirical grounding for its
theoretical parameters.
For instruments we now have a cross classification to draw on: a
diversity of roles of instrumentation in experiment, and a diversity of
roles of experiment in theorizing. In the former, measurement shows up
as a single role among others, the representative role. In experiments,
however, measurement is crucially involved in all cases. All the examples
of production of new phenomena involved set-ups in which multiple
measurements were being made as well. Although it may seem as if
electron microscopes, spectroscopes, spectrophotometers, x-ray cameras,
refractometers, polarimeters, cloud chambers, scintillators, and the like have
been relegated to a ‘lesser’ status in my account than the scientific realist
would give them, there is no implication at all that the role of measurement
has vanished from the scene. What needs still to be elaborated, however,
is an account of measurement itself, as it appears in the process of joint
experimental-theoretical development and as it appears through the eyes
of the eventually achieved theory.
5
The Problem of Coordination
A scientific theory is typically presented with laws, principles, or equations

that involve terms specific to that theory:
s = vt + (1/2)at2 , PV = rT, F = ma, action = reaction
to mention only a few of the classical standbys. Although these terms are,
taken literally, symbols for functions with specified mathematical character,
they are often pronounced as if they were nouns already familiar before
the theory’s introduction: ‘‘distance’’, ‘‘velocity’’, ‘‘time’’, ‘‘pressure’’,
‘‘temperature’’, and so forth. That nomenclature is introduced more by
way of informal commentary than explanation, and certainly does not
define the theoretical terms.1 But the choice of familiar words does signify
something: they point to the sort of data to enter, and the sorts of
measurements that can help to determine the values of those functions.
The theory would remain a piece of pure mathematics, and not an empirical
theory at all, if its terms were not linked to measurement procedures. But
what is this linkage?
That question, which turns out to bring many further questions and
complexities in its train, poses what was once generally known as the
‘problem of coordination’. The term had appeared in Mach’s writings on
mechanics and thermodynamics, was salient in the discussion of the relation
between mathematical and physical geometry that extended from the
nineteenth century into the twentieth, and came to special prominence
through the writings of Schlick and Reichenbach when logical empiricism
was beginning to break with the neo-Kantian tradition. Since then it has
not been so much in use, but the problem to understand just how a
scientific theory is more than its mathematical guise reappeared at every
juncture in the subsequent history.
116  : , ,  
The questions What counts as a measurement of (physical quantity) X? and

What is (that physical quantity) X? cannot be answered independently of each
other. To echo another such realization, I am not ashamed to admit that
this brings us to the famed ‘hermeneutic circle’ (Eco 1992: 64). We shall
examine this apparent circularity by focusing on the one hand on its more
abstract consideration by Reichenbach, and on the other hand the practical
response in history examined by Mach and Poincaré.
Coordination: a historical context

Assigning a value, a location on a scale or in a larger space, is precisely
what Ernst Mach called ‘‘coordination’’.2 After a historical survey of how
thermometry developed (which we will look at in more detail below),
Mach began a critical discussion of how the involved choices had been
made. Precisely in the midst of this critique he introduces the term
‘‘coordination’’:
That number which, conformably to any chosen principle of coordination, is
coordinated with a volume indication of the thermoscope, and consequently
uniquely with a state of heat, is called the temperature of that state. (Mach
1986: 52)
How much would already have to have been in place before this could
be offered to identify the quantity temperature? It can make sense only
at a point when there is no longer any ambiguity about what counts as a
thermoscope or thermometer, or any doubt about whether the volume of
the thermometric substance is uniquely correlated with a ‘state of heat’, or
about what is meant by that.
Mach is very clear that this is at best a late stage in the historical
development. He sees clearly that it involves both idealization with respect
to the stability in the instruments that count as thermoscopes and choices
whose status needs to be critically examined. But some leeway for choice
is brought into the open precisely in the phrase that introduces the term
‘‘coordination’’. He continues
The temperature numbers are dependent on the principle of coordination, t = f(v),
where v is the thermoscopic volume, and, consequently, for the same state of heat
they will vary greatly according to the principle adopted. (Ibid.)
    117
As he points out, different such principles of coordination (choices of

function f ) were introduced. Galileo made the temperature numbers
correspond to increments starting from a chosen initial volume:
t = N if and only if the volume equals v0 (1 + Nα)
while Dalton’s function had the form
t = Nif and only if the volume equals v0 βN
There is a simple mathematical transformation that leads us from one to
the other (even if not as simple as a transformation from Fahrenheit to
Celsius!), so although the two are different they are, one might say, not
fundamentally different. When, however, Amontons and Lambert choose
the pressure of a mass of gas of constant volume as temperature index, they
are choosing a quite different principle of coordination. In the numbers
assigned their method of thermometry is not markedly different from those
made by Galileo’s scale. But in this case, the relationship between the scales
is not a matter of mathematics or logic—it is due to relations in empirical
and physical contingent fact.
So what is meant by coordination? Just for now let us take thermoscopes
and the indicated physical correlations between heating, expansion, and
pressure for granted. Then the initial statement by Mach identifies the
physical magnitude (parameter, quantity, observable) temperature. Since he
identifies its value as ‘‘dependent on the principle of coordination’’, hence
relative to a choice among coordinating principles, we must take it that
only what is invariant under these choices is significant. This might seem to
relegate the coordinating principles themselves to insignificance. But that
is not so: in the absence of any such coordinating principle, attributions
of temperature would have no empirical content at all, no reference in
what is already established as measurable. To put it briefly: on the one
hand, coordination requires for its possibility some recognized empirical
regularities, while on the other hand it is required for the new theoretical
assertions to have any empirical content at all.
The term ‘‘coordination’’ was central in Schlick’s Algemeine Erkennt-
nislehre, where the account of the process of cognition is based on idea of
coordination (Zuordnung).3 While I will ignore his generalization as only
tangentially relevant to our subject, its historical role underlines the impact
of this problem on the philosophy of science that was then developing.
118  : , ,  
Here Poincaré (whose analysis of this problem for time measurement we

will look into below) had been a central figure, and almost every discussion
had to orient itself with respect to his conventionalism. Einstein took an
active part, and in fact uses the terminology of coordination quite explicitly
in his 1921 essay, ‘‘Geometry and Experience’’:
It is clear that the system of concepts of axiomatic geometry alone cannot make
any assertion as to the behavior of those objects of reality which we designate as
practically rigid bodies. To be able to make such assertions, geometry must be
stripped of its merely logical-formal character by the coordination of experienceable
objects of reality with the empty conceptual schemata of axiomatic geometry.4
But the radical, foundational taking by the horns we find in Reichenbach’s

1920 book, The Theory of Relativity and A Priori Knowledge. There we see him
landing in serious difficulty with the question of how terms in the theory
of space, time, and motion—whose primary reference is to mathematical
objects—can be linked to items or aspects of the physical. The reason for
his difficulties lie precisely in the radicalism of his philosophical approach.
Coordination and Reichenbach’s problem

Mach’s discussion had encountered no difficulty in principle because he
stated his principles of coordination in a context where certain other
physical parameters and their measurement were assumed to be given. The
measurement of length or volume and volume change, for example, is
simply taken for granted in the passage I quoted.
In 1920 Reichenbach, on the other hand, realizing that he is addressing
the most fundamental level of physical theory, allows himself no such
luxury. He wants to find a general coordination of mathematical spaces
and their structure with physical relations. In his later Philosophy of Space
and Time he gives as first example how units of length can receive their
coordination: ‘‘a meter is the forty-millionth part of the circumference
of the earth’’( Reichenbach 1958: 15). But the recurring example of the
concept of a straight line, or more generally a geodesic, as having as
physical correlate a light ray, or the path of a freely falling body, is not
equally easy to understand. Nor is the idea that congruence relations have
as correlate coincidence with transported rigid bodies.5 In these examples
it is quite unclear how to identify the physical correlate without using any
geometric or kinematic terms. How are we to describe rigidity or free
    119
fall without using the language of geometry, or of mathematical physics

in general? In the example that most preoccupied him about the theory
of relativity, non-simultaneity is to be thus related to non-connectability
through any signals, whether by light emission and reflection or by material
transport. Not only the modal character of ‘‘connectable’’ but also the
required identity-over-time (genidentity) of the material are as puzzlingly
theoretical as any of the terms in physical geometry. And so we find him
perplexed. He writes in 1920:6
It is characteristic of modern physics to represent all processes in terms of mathematical
equations. But the close connection between the two sciences must not blur their
essential difference. (Reichenbach 1965: 34)
The mathematical object of knowledge is uniquely determined by the axioms and
definitions of mathematics. (Ibid.)
The physical object cannot be determined by axioms and definitions. It is a thing
of the real world, not an object of the logical world of mathematics. Offhand it
looks as if the method of representing physical events by mathematical equations
is the same as that of mathematics. Physics has developed the method of defining
one magnitude in terms of others by relating them to more and more general
magnitudes and by ultimately arriving at ‘‘axioms’’, that is, the fundamental
equations of physics. Yet what is obtained in this fashion is just a system of
mathematical relations. What is lacking in such system is a statement regarding the
significance of physics, the assertion that the system of equations is true for reality.
(Ibid.: 36)
So how can empirical significance be achieved? The examples he has in

mind, as we just saw, are the use of rigid bodies as choice to set the
relation of spatial congruence (in effect, to measure length), the choice of
a light ray path in vacuo as physical correlate for geodesics, the choice of a
certain periodic process as setting the unit of time. Question: how are these
physical correlates to be identified without use of geometric or kinematic
terms?
Look back to Mach’s examples of Galileo’s and Dalton’s choices for
temperature scaling. These choices are expressed in the above equations in
which one variable is set equal to a function of another variable. So is the
choice in question simply a choice of a function? A function relating what
to what? Isn’t a function a mathematical object itself, defined in terms of a
relation between mathematical objects? So Reichenbach writes:
120  : , ,  
The coördination performed in a physical proposition is very peculiar. It differs

distinctly from other kinds of coördination. For example, if two sets of points
are given, we establish a correspondence between them by coördinating to every
point of one set a point of the other set. For this purpose, the elements of each
set must be defined; that is, for each element there must exist another definition
in addition to that which determines the coördination to the other set. Such
definitions are lacking on one side of the coördination dealing with the cognition
of reality. Although the equations, that is, the conceptual side of the coördination,
are uniquely defined, the ‘‘real’’ is not. On the contrary, the ‘‘real’’ is defined by
coördination to the equations. (Ibid.: 37–8; my italics)7
Here Reichenbach was imagining, and discounting, the following naïve

sort of reply:
what is called for is simply a function, a mapping, between mathematical

objects and physical objects or processes—what is puzzling about that?
The reason he discounts it is because to define a function we need to

have the domain and range identified first—and the question at issue was
precisely how that can be done without presupposing that we already have
a physical-mathematical relation on hand.
So what does Reichenbach mean, when he seems to point to a solution
with the words ‘‘On the contrary, the ‘‘real’’ is defined by coordination
to the equations’’? We are here at the historical point where the split
between the neo-Kantian tradition and logical empiricism begins, at least in
Reichenbach’s intellectual life.8 In this early book, Reichenbach argues that
at least one essential element of the Kantian a priori can still be maintained.
So to effect the necessary coordination of abstract mathematical structure
to concrete empirical reality he posited a special class of mathematic-
al–physical principles—‘‘coordinating principles’’ or ‘‘axioms of coordin-
ation’’—whose role is precisely to insure those conditions of possibility.
These principles, Reichenbach argues at that point, are to be taken as given
or imposed a priori, and so to be sharply distinguished from mere empirical
laws (‘axioms of connection’).9 In this respect then he follows Kant’s origin-
al conception, though he adds the innovation that the a priori is historically
relative, the coordinating principles must be changed as the sciences develop.
As we should guess from Mach’s example, Reichenbach can coordinate
a mathematical representation with physical objects, events, and processes
only in a context where something is already given that will make that
    121
possible. Thus we must doubt whether Reichenbach really did, by this

reconception of the a priori as historically relative, provide something
given-in-context that makes it possible to link mathematical objects to
physical magnitudes. How could principles, or having principles, provide
the conditions in which we can relate something mathematical to something
physical?10 What exactly are the required conditions—isn’t that the question
to be answered before we can think about what would make those
conditions possible?
In fact, it seems that Reichenbach himself was not able to maintain
this reaction to his problem of coordination. At a somewhat later point
in his writings he offers ‘‘coordinative definitions’’ instead of axioms of
coordination. This was, I think, definitely not meant to be just a verbal
change: he means to trade at that point on the conventionality of definition,
the possibility to bring the definition—and therefore, supposedly, the
coordination—into being by an act of stipulation. At the same time he
means to be still actively inquiring into the empirical conditions under
which such definitions can play the requisite role. What he did not do
is change the ahistorical setting of his problem: the coordination is still
apparently to be conceived of as possible in the absence of any previous
such coordination. But is that possible at all? We have to ask more or less
the same question again as before: how can such coordinative definitions
be meaningfully introduced except in a historical context where there are
some prior coordinations already in place?11 I submit that they cannot.
The problem of coordination reconceived

How should we see the problem of coordination now? Not as a problem
so posed as to preclude its own solution! Nor does it seem that we get very
far by posing it as a problem that is solved by choosing a particular object
to be designated as standard of unit length, or light-ray paths as physical
correlate for geodesics.
Coordination must determine how measurement can establish a value
for what is measured. But that it can do so appears to presuppose
an understanding both of the measurement procedure and of what is
measured—that is, of the terms between which that relationship is estab-
lished through coordination. So the prior question at issue is precisely:
122  : , ,  
What is measured? In the assertion ‘‘the length of A was measured’’, or ‘‘the

temperature of B was measured’’ or ‘‘the duration of C was measured’’,
what exactly is the referent of the subject term?
This question will look different depending on where we take our
vantage point. There are actually two distinct vantage points common
in the philosophy of science literature—not just about measurement but
about experiments, models, theories, . . . But the divide becomes especially
salient in the ways measurement is discussed.
• We have no problem identifying what is measured by a particular
procedure if we look at it ‘from above’, that is, in retrospect, from
within a theory that is already stable and established, and deals with
that physical magnitude.
Looking at the matter from this vantage point we say that what was being
measured all along, however well or poorly, in the centuries leading up to
that achievement, was precisely what we say it is now.
• Nor do we have a problem in our discussion of measurement if we
look at it ‘from within’, that is, if we inquire into the introduction
of measurement procedures at historical stages where there were already
measuring procedures for certain other physical magnitudes taken as given.
The examples of Mach on temperature and Poincaré on time measurement,
that we will look at below, illustrate this very well.
It is only if we try to understand coordination without locating ourselves
either in the ahistorical ‘above’ or the historical ‘within’, but pretend to a
‘view from nowhere’, that we set ourselves an impossible problem—or fall
into metaphysical metaphors.
If in our retrospective view today we want to describe the historical
process, we can of course only speak our own language, the language we
have today. And today it is a truism that
an item has temperature T if the stable procedure that we now
designate as temperature measurement, when applied to this item, has
outcome value T.
So if today we have to say what the earlier activities we focus on—such
as Galileo constructing a sort of thermometer—were all about, and have
to answer what was the intended referent, while measurement procedures
    123
and theories were still evolving, our answer had better be coherent with
what we now take that referent to be. So we can only say:
all along they were referring to the magnitude identifiable as measured
by the procedures that would eventually be accepted as measur-
ing it.
So yes, it makes sense for us to say that Galileo’s activity was about
measuring temperature, although we have to add immediately (speaking in
our own voice, using our own words) that his apparatus did not really do
that, that there was a mistake involved in his set-up.
Notice how this answer echoes Reichenbach’s statement that what is
measured, and thus through measurement coordinated with a theoretical
concept, is not defined independently of that coordination, but defined by
the coordination. At first blush this formulation too may sound circular,
even viciously circular, if not debilitatingly anthropocentric or egocentric.
But it is not. What would make it viciously circular is only Reichenbach’s
tacit assumption that what was being done could in principle be done
without any historically prior achievements along such lines. But to
understand this properly we must follow the process of coordinating
both in its general form and in specific examples, to show how the joint
evolution of measurement and theory can happen and can come to a stable
resting point.
What is measured?
Can we identify the parameters that are measured by means of the meas-
uring procedure itself alone? Well, what does ‘‘alone’’ mean here? Does it
mean ‘‘with no theoretical background in which to operate’’? But with no
theoretical background, very few procedures classify as measurement pro-
cedures. For example, as Mach and later Poincaré emphasized, procedures
for measuring mass presuppose something about the items to which they
are applied, e.g. that they are subject to the law that action = reaction.12
This is precisely why Mach, and later Cassirer, Schlick, and Reichenbach,
in their different ways, concerned themselves so much with the problem
of coordination.13 But it is also through measurement results that theories
gain empirical support, so there must be some trade-off, or some subtleties
to how empirical support can be gained, if we are not to find ourselves lost
in a circular argument.
124  : , ,  
The idea that the parameters can be identified simply through the
measurement procedures was behind Bridgman’s operationalism, and we
often see its strong appeal still in scientific writings. But the specification
of the measured parameter even in terms of the (final, stable) measuring
procedure can never amount to a complete identification or definition. For
the procedure will classify only when it is actually applied, and even then
only establish order among the items to which it is applied, at the times
when it is applied.14 To try and stick with an ‘operationalist’ view of what
is measured would land us in the absurdity that a process or object has no
character at all while it is not being measured.
In practice a theory eventually emerges which encompasses the meas-
urement procedure itself as well as the items measured, and provides the
coordination. Thus in the case of temperature, the kinetic theory which
was developed after thermometry had been developing for two centuries,
provided the parameter which then was identified as precisely what was
measured by the thermometer. So a definite identification, a complete
definition, of the measured parameter is possible but only through, at the
hands of, and relative to the theory offered and finally accepted to account
for the stability of the measurement procedure.15 We need to see this
development in its historical perspective, and to recognize and distinguish
two facts:
(a) an empirical fact that has been discovered: namely, the very stability
in the procedures found in this historical development, and the
reliability of the predictions concerning these and their correlation
with other measurement procedures derived from the mature theory
in which they are now theoretically embedded.
(b) a historical fact: namely, that choices in the development of these
measuring procedures went hand in hand with the development
of the theory in question, so that we cannot identify an aspect of
nature that is measured if we refer only to those empirical procedures
without using concepts provided by the theory.16 What now counts
as simple passive measurement is a hard-won achievement.17
What were the choices involved? What epistemic status can we assign to
these choices? We will not gain much clarity if we continue simply in the
current abstract vein. We need to look at the actual history to get any grip
on this at all.
    125
Mach on the history of the thermometer18

As Mach emphasized, it was already a matter of actual practice, before
modern science turned its attention to the question of a quantitative
measure, to evaluate bodies as hotter or colder or of equal temperature
on the basis of common experience. But judgments based on perception
do not yield an unambiguous ordering—as was clear from many common
examples. Iron and wood never feel equally hot or equally cold no matter
how long they are left undisturbed in contact, for example. Evaluations
of hotter and colder tend to be reversed if the perceptions are reversed in
order. As Poincaré would later say for time measurement, psychologists
could ignore this but not the physicist: some grading procedure with results
that could fit into a single scale was desirable.
To pursue this desideratum is already a choice. Almost from the very
beginning, the aim was to establish a numerical scale. The ideas of hot, cold,
as hot as, and hotter than were to be transformed into the idea of a quantity
with the same general structure as height or volume or speed.
The air thermometer

The word ‘‘thermometer’’ came into use in the mid-seventeenth century.
In the first half of that century there had already been widespread use of
instruments that responded to temperature changes, based on the expansion
and contraction of air. Galileo is sometimes credited with the invention;
certainly there were examples in Italy in his time. Retrospectively they are
called ‘‘air thermometers’’. A flask with a long neck is vertically inserted into
a container of water, with the opening downward. As the flask becomes
colder, the water in the neck rises. Similarly, when the apparatus becomes
warmer, the water goes down. The differences in water height were at first
measured with compasses, but then divisions were marked with numbers
attached. Records were made of how low and how high the water would
be in winter and summer, or during the night and the day.
The very set-up cries out for a simple numerical scale: how high is the
column of water in the tube? Changes in that height indicate precisely the
change in volume of the contained air. The guess is that the air expands
with increase in temperature and contracts as it becomes colder. This guess
does not presuppose that we already know how to measure how much
126  : , ,  
warmer or colder it gets: the rough and ready comparison by the senses
will do for that. But if the change in height of the water column, correlated
with the volume of enclosed air, can be used as indicating how much hotter
or colder the air became, then we will have a numerical scale induced
by the most basic, most venerable form of already familiar measurement:
length.
Is there an empirical question involved in this choice? What cannot be
ascertained without the use of a good thermometer is whether the change
in volume is quantitatively proportional to the change in temperature. There
will certainly be empirical questions to come, but this is not one. Prior to
the construction of a thermometer, there is no thermometer to settle that question! So
here the scientist confronts a choice, one that may be provisional to start,
with hope of vindication of some sort or other later on: to use changes in
volume as quantitative measure of how much hotter or colder one thing is
than another.
There are a number of drawbacks in this arrangement, if we try to think
about it in current terms as measuring temperature. The water and the glass
are also relevantly affected by temperature changes, even if not to the same
degree as the air. The initial temperature of the apparatus will also play a
role. But most important from that point of view is the effect on the water
level of changes in atmospheric pressure—the instrument is, we might say
now, part thermometer and part barometer. This could not even be realized
or taken into account until the notion of atmospheric pressure was in play.
The history of the barometer began during the same century, and like
that of the thermometer was initially beset by theoretical disputes about
the possibility of a vacuum. Hence the problem of effects of atmospheric
pressure on the air thermometer did not appear till it could appear. The
problem that the air thermometer was actually a mixture of thermometer
and barometer was in fact pointed out by Blaise Pascal in mid-seventeenth
century, after his barometric experiment on Puy-de-Dôme.
The liquid thermometer

How was this problem to be avoided; how could one have an instrument to
gauge temperature differences that was not ‘‘part’’ barometer? In the second
half of the century the use of the liquid thermometer became common:
a liquid in a glass tube, filled completely when at very high temperature
and hermetically closed. When it cools to room temperature, the liquid no
    127
longer fills the whole tube—arguably, and at the time contentiously, one
now witnesses the creation of a vacuum. (Questions on that score were
crucial then; I shall ignore them here.) A common form had its length
divided into equal parts. Detailed instructions recorded at the Accademia
del Cimento in Florence, specified that in a good construction snow or ice
will not make the liquid column descend below the 20 degree mark, nor the
hottest day dilate it beyond 80 degrees.19 Since the tube is entirely closed,
this arrangement eliminates the effect of atmospheric pressure, though not
the effects of temperature changes on the liquid or the glass.20
The assumption or choice of definition that differences in temperature
are proportional to resulting differences in volume continues, though now
instantiated to liquids. Two obvious questions: do different liquids agree
with each other in this respect? and do they indeed always expand when, by
other common criteria, they are becoming warmer? These questions did not
receive very satisfactory answers. The early makers of liquid thermometers
used quite a variety of different liquids. Water was not a good candidate,
since its volume does not keep contracting as temperature drops toward
freezing.21 None of the liquids is of any use below its freezing point
anyway. But in the regions well above their freezing point it was at first
thought that they all expanded and contracted in the same proportion when
put in the same environment controlled for effects other than temperature
variation—an appearance which in turn was refuted by Dalton for the case
of high temperatures.22
The results are thus sensitive to the choice of liquid, partly because
their response to temperature changes varies at very low and very high
temperatures in a certain range—and the liquid state does not persist below
or above a certain point. Even in ‘good’ ranges, the expansion in volume
is not proportional for different liquids. Dalton wrote, a good century into
this evolution, that liquids were found to expand unequally, ‘‘but no two
of them alike. Mercury has appeared to have the least variation’’.23 But
even for the otherwise very good choice of mercury, the effects on the
glass tube are not negligible in comparison to the effects on the liquid. The
expansion of mercury is only seven times that of glass.
The gas law and kinetic theory

These findings occasioned a return to the gas thermometer, especially
in view of the discovery and refinement of what is now often known
128  : , ,  
as the ideal gas law (Boyle’s Law, the Boyle-Mariotte Law, the Law of
Charles and Gay Lussac, Charles’s Law—a multiplicity of names partly
due to nationalist partiality), PV = rT, with r a constant characterizing
the gas. If pressure is kept constant, the volume varies proportionally with
the temperature, for all gas in the same way—conversely, if the volume
is constant, the change in pressure is proportionate to the change in
temperature. Obviously the experimental findings could at best hint at this
relationship before thermometry had been stabilized. But of gases, unlike
of liquids, it could not be said that they expand unequally ‘‘but no two of
them alike’’.
Guillaume Amontons pointed out, in effect, that if a law of this form is
correct, then the temperature scale can become a ratio scale: it will have an
‘absolute zero’, marked by the theoretical condition of zero pressure.24
Is this law correct? To begin its correctness is supported by experimental
findings, made with thermometers of the previous generation, accepted
as a rough guide while recognized as inadequate. Then, as confidence in
the kinetic theory increases, theory cuts the Gordian knot with respect to all
questions about how the scale is to be fixed. The gas law is supported
by its incorporation in the kinetic theory—and refined as well, so that
it also becomes itself qualified there as having been only an empirical
approximation to what really happens. But the scale is now anchored in
prior scales for mechanical parameters. Once T is identified with the mean
of a mechanical parameter, the postulated mechanical behavior—‘‘the type
of motion we call heat’’, in Clausius’s incomparable phrase—is now the
referent of what is measured.
We can also put it the other way around: the parameter, identified by
the eventually stabilized procedures for its measurement, is now classified
by the theory as one aspect of the logical space that the theory provides for
location of items in its domain.
The history of thermometry did not stop here, but this much will suffice
to make the important points concerning coordination.
Choices and conventions

The new type of instruments that emerged in this evolution, before the
kinetic theory rewrote the story of what has happened, were selected as
correct for thermometry in a process that involved three main choices.
Mach’s history of the subject emphasized perspicuously the sense and
    129
extent of these choices involved in how the scientists arrived at their

results.
The first choice, which we have been emphasizing from the beginning,
concerns what is to be taken as representing the quantitative change in
temperature. The choice of change in volume passes the initial test of
rough agreement with the admittedly imprecise deliverances of experience
in familiar circumstances. The vindication of such an initial choice in
the future is of course hostage to fortune. The second is the choice of
thermoscopic substance, to answer ‘‘change of volume of what?’’ And the
third is just what above, at the outset of our discussion, we saw Mach
calling the ‘‘principle of coordination’’ for assigning numbers, that is, for
setting the scale. (op. cit: ch. II, sections 11–12).
There are coherence constraints on the choice of thermoscopic substance.
We have seen that if it had been decided to stay with liquids, there would
have been quite a lot of arbitrariness in this choice, first of all because they
do not appear to expand at the same rate even in the same range. Secondly,
different liquids would have to be chosen for different ranges, as they near
their different freezing or boiling temperatures. The gas law saved the day:
it seemed that different gases expand at the same rate as each other under
the same conditions. Again there are limitations, since real gases too could
liquefy and solidify at different temperatures.
The last choice, to set the scale, may now seem trivial. But it is not.
The significant choice involved is best seen in the discussion of whether
there is an absolute zero, a minimum to the temperature scale. The main
convention, historically, goes back as far as Galileo: to take the melting
point of ice and the boiling point of water as selecting fixed points on the
scale. We have already seen that the change in volume of water will not
be a convenient one overall. Using the gas law, for a particular gas, we
can define the temperature scale either in terms of that for volume at a
constant pressure or in terms of that for pressure at a constant volume. So
Amontons proposed the pressure p of a mass of gas at constant volume, and
chose as reference point the pressure p0 at the melting point of ice. Then
the temperature scale can be defined with an arbitrary scaling constant k as
t = kp/p0 . The numbers k and the pressure scale can be chosen so that the
difference between the temperature of melting ice and that of boiling water
is 100, and there is an absolute zero of temperature for p = 0, whether this
can be attained in nature or not.
130  : , ,  
This is a very natural way to proceed, but if it is thought that the lower
bound is a fiction, one could choose a different principle of coordination,
not bounded below any more than bounded above. Mach points out that
Dalton had in effect chosen such a scale, though of course that is not the
one that caught on.
The choice is settled later by the absorption of the entire subject of
inquiry into physical theory. In the kinetic theory the magnitude with
which temperature is identified has an absolute zero point. But it is well to
note that this was not part of the empirical facts guiding the evolution of
thermometry, though the kinetic theory was developed in a way that was
accountable to that evolution of empirical procedures.
The very simple operations we now perform with thermometers are
an obvious example of ‘passive’ measurement—in our theories we have
the interaction as well as the object and apparatus completely classified,
‘well understood’. This stage is an achievement. The story of thermometry
is a success story, and such success was by no means guaranteed a priori.
The conditions of possibility for such a successful coordination—here we
are at Reichenbach’s question—we have now seen to involve empir-
ical regularities that are contingent and choices subject to conditions of
coherence.
Poincaré’s analysis of time measurement

The second half of Poincaré’s essay ‘‘The measure of time’’ is the more
famous because of its connection with special relativity.25 But I will
concentrate here on the first half, where Poincaré begins with the problem
that we do not and cannot have a direct intuition of the equality of successive
time intervals (equality of duration of successive processes).26 This is not a
psychological point. Two successive periods of a clock cannot be compared
by placing them temporally side by side, that is why direct perception can’t
verify whether they lasted equally long. This had been a recurring topic of
discussion; Poincaré mentions Andrade but across the Channel too it was a
topic among the British Idealists and for the young Russell.27 Psychologists
may ignore this, Poincaré writes, but the physicist cannot.
In the case of two sticks we can check to see whether they are equally long
(at a given time) by placing them side by side; that is we can check spatial
    131
congruence (at that time) by an operation that effects spatial coincidence

(at that time). We can check whether two clocks run in synchrony during
a certain interval if we place them in spatial coincidence. These procedures
do not suffice for checking whether two sticks distant from each other in
time or space are of equal length, nor whether distant clocks are running in
tandem, nor whether a clock’s rate in one time interval is the same as some
clock’s rate in a disjoint time interval. But in physics, criteria for spatial and
temporal congruence are needed. Poincaré is concentrating on this need.
What measures duration is a clock, and physics needs a type or class of
processes that will play the role of standard clocks. What type or class to
choose? One answer might be: the ones that really measure time, that is,
mark out equal intervals for processes that really take equally long. While
certain philosophers or scientists might count this demand as intelligible,
it must be admitted that there could be no experimental test to check on
it.28 We cannot compare two successive processes with respect to duration
except with a clock; but clocks present successive processes that are meant
to be equal in duration. This is similar to Mach’s point about thermometry:
whether the melting of ice always happens at the same temperature, or
the volume of a substance expands in proportion to temperature increase,
can be checked only with something functioning as a thermometer—and
thus cannot be ascertained in order to check whether thermometers are
‘mirroring’ temperature.
So Poincaré presents the choice as a decision: to begin, he says, it was
decided to ‘‘admit by definition’’ that successive periods of a pendulum had
equal duration. This is not a historical claim about conscious or explicit
decisions, let alone about the actual practices of time reckoning before the
modern era. The clock mechanism driven by a descending weight was
in use before Christiaan Huygens and Galileo independently invented the
pendulum clock. Poincaré is writing a ‘‘just so’’ story, a way to understand
how choice not subject to direct empirical check is involved, no matter
which mechanism is settled on to begin. Some choice—which may be
subject to equally chosen constraints but not to empirical verification—is
needed to provide an initial starting point.
Coherence conditions
But it certainly is the sort of choice that is also subject to strong coherence
constraints, which are not a matter of choice. The class of processes chosen
132  : , ,  
as initial standard for equal duration must be such that if two are in spatial
coincidence, they run at the same rate, they run in synchrony. The decision to
choose the pendulum, in such bare form, cannot be sustainable in practice,
precisely because it does not have this coherence. Different pendulums will
disagree due to alteration in temperature, friction, barometric pressure, etc.,
and possibly beside these due to at that time unknown or uncontrollable
factors. Comparing the behavior of different pendulums does not display
stable synchrony. In 1672 Richer took a pendulum clock from Paris,
latitude 49◦ North, to Cayenne in French Guyana, latitude 5◦ North.29
He found that he had to change the length of the pendulum to make
it agree with standard time reckoning, precisely because as Huygens and
later Newton were able to explain, the force of gravity is not the same
everywhere on earth. So the next development, according to Poincaré,
was to correct mechanical clocks such as pendulums by the sidereal clock,
which is based on the passing of a star across the meridian—the unit then
being in effect the duration of one rotation of the earth. This too is a
decision, and this time it was made explicitly.30
Yet this adoption of the sidereal clock was subject to a similar objection,
if in a somewhat more theoretical guise: the movement of the seas will
act like a brake, slowing down the rotation of the earth. Far from being
a merely academic or skeptical doubt, this point was used to explain an
apparent acceleration in the movement of the moon.
Poincaré wishes to reveal by these examples two problems that arise in
developing a measurement procedure for duration. The first is the initial
one, illustrated with the pendulum: we cannot place successive processes
side by side so as to check whether their endpoints coincide in time. So
there is no independent means for checking whether successive stages of a
single process are of equal duration: the question makes sense only after
we have accepted one such process as ‘running evenly’.31 As said above for
temperature,
duration = the magnitude identifiable as measured by the clocks that
would eventually be accepted as measuring it.
This insight requires us to move all our concern to coherence constraints that
put bounds on what we can choose to regard as ‘running evenly’.
In another context, to illustrate that definitions are not totally arbitrary
but in general have factual presuppositions, Poincaré gave the example of
    133
the light-clock. Suppose we design a device in which light is emitted at

one end of a meter-long path, reflected at the end, and absorbed again at
the starting point. The interval between emission and absorption could be
taken as the unit of duration. Now we carry this device and many faithful
copies all over the place to be used as clocks. Any objection to this? Yes:
in nineteenth-century physics it is entailed that these clocks will not give
consistent results, because even rotating them with respect to each other
will result in violations of synchrony. No: after the Michelson-Morley
experiments, or more accurately, after Einstein’s special theory of relativity
was accepted, these light-clocks are perfectly acceptable in principle: there
is no necessary incoherence.
The second problem is brought out by the astronomers’ rejection of
the equality of sidereal days. Their rationale for this is an application of
Newtonian physics to the earth and its seas. The application of this physics
presupposes that the reference for its time parameter has been fixed. Fixing
that parameter by taking sidereal days to be equal runs into trouble precisely
with the theory’s representation of the very processes that define the sidereal
day. So Poincaré diagnoses the astronomers’ practice as follows: they define
duration in such a way as to ensure that Newton’s laws come out correct.32 But
this decision, that the time parameter should be fixed so that Newton’s
laws will come out true, does not suffice by itself! Newton’s laws could be
re-expressed in terms of different ways of measuring time; they just would
not have the same simple form.
Suppose for example that with a certain adopted standard of measurement
and its time scale t, these laws are satisfied. Now adopt a new scale T, which
is a logarithmic function of t. It would be silly to reflect that Newton’s laws
as originally stated are not satisfied if t is replaced by T: the thing to do is
to translate Newton’s laws by means of the transformation of t into T. The
result will be an empirically adequate theory if the original was—it will
just be not nearly as simple or user-friendly. So in rejecting the sidereal day
as time unit, the astronomers, according to Poincaré adopted not just the
above criterion but the following more complete one: Time is to be defined
in such a way that the equations of mechanics will be as simple as possible.33
Coherent rivalry and choice

This quick conclusion we’ll have to look at critically in a moment, for
surely Poincaré is going a bit fast here! Simplicity is perhaps only a code
134  : , ,  
here for a cluster of pragmatic desiderata for our theories. But we can add
that the issue became concrete later on due to a proposition advanced in
cosmology by Arthur Milne in the 1930s.34
By that time the corrected astronomical time reckoning that had become
standard had been supplemented at least at the theoretical level with atomic
clocks. It was part of the accepted theory that atoms of each chemical
element and compound absorb and emit electromagnetic radiation at
their own characteristic frequencies. As reckoned by the sidereal standard,
these resonances had been taken to be inherently stable over time and
space. In Milne’s time there was as yet no practical application—the first
atomic clock was not built till 1949. Recall now that there are coherence
constraints: the class of ‘‘standard’’ clocks must be such that whenever two
are in coincidence, they run at the same rate, they run in synchrony. In
principle there could be two such classes, disjoint from each other, and not
combinable in a way that satisfies this constraint. Milne proposed that this is
indeed the case, and that astronomical (sidereal) clocks run differently from
atomic clocks. The disparity is negligible outside cosmology. The relation
between them, he submitted, was an exponential function, so that the two
give us entirely different pictures of cosmic history: for the one, the past is
finite and for the other it is infinite.
Although a lively topic at one stage in the development of cosmology,
the idea did not have any continued impact. What it illustrates and is
important, though, is the point of principle already made by Poincaré. If
the class of measurement interactions (of one sort or another) cannot have
an ‘‘if and only if ’’ definition but only be required to satisfy certain coherence
conditions, then the possibility of such disparities (and hence leeway for
convention or choice) is entailed.
Einstein’s critique; an example with length

In his largely sympathetic discussion of Poincaré’s views, Einstein famously
added one telling criticism.35 While choice is involved in setting up the
principles of coordination, this sort of choice can enter at many different
points, and therefore it is not correct to single out any one element as a
stipulation, definition, or convention.
This becomes clearer when we consider the development of measure-
ment processes for more than just one observable. Without some element
of choice or convention, coordination is inevitably circular. So far I have,
    135
in all the examples, taken length measurement for granted. But suppose we
consider length in the same way as we considered temperature and time
measurement, where the principles of coordination presupposed length
measurement as given. In order to ensure standards of length comparison
that would remain applicable across time, another such history of coordin-
ation unfolded. One telling example occurred early on in a proposal by
Huygens. After his invention of the pendulum clock, and his appreciation
of how its period is related to latitude, he proposed that the unit of length
should be one third of the length of a pendulum with a one second period
at the latitude of Paris—not the English foot, not the pied du Roi, which
was 16/15 of the English foot, but the pes horarius. Here time measurement
is presupposed for the coordination of length.
While this proposal did not prove practical, we see a similar entanglement
of several principles of coordination at the most familiarly known stage:
the construction of the standard meter embodied in the platinum bar to be
kept in the French State Archives. Iron copies were distributed to other
countries, and needed to be sent to France every ten years to be compared
with the primus inter pares copy kept in Sèvres. So we see two decisions
installed as definitive: that such metal bars remain congruent to themselves
under transport, and so provide a standard for congruence with respect
to length, and that a particular such bar defines the unit of length. But
what precisely was the convention? It was decided at the Convention
of the Meter in 1875, and was precisely this: the meter is defined to be
the distance between the midpoints of the ends of the mètre des archives
at the temperature of melting ice. But is the temperature of melting ice
always the same? The convention presupposes temperature measurement,
just as the conventions for temperature measurement presuppose length
measurement. The circle can be broken by entering a choice at one point
or another, but there is no specially privileged point for entering a choice.
Thus Einstein separated Poincaré’s rather simplistic pointing to specific
‘conventions’ or ‘definitions’ from the real insight into the inevitable
holism in the jointly constructed theory and coordination. As Ryckman
observes, Einstein’s admission of Poincaré’s viewpoint ‘‘is not a concession
to conventionalism, that is to the freedom to choose any geometry we
like, but to the inevitable epistemological holism of a theory in principle
capable of explaining its own measurement appliances, and so its ties to
observation’’ (Ryckman 2005: 64). But this holism will isolate the theory
136  : , ,  
from what it is about unless the coordination is effective. This can be

come about in many different ways, with an element of choice appearing
at different points—just not without entering any choices at all. Let’s
summarize it this way: any way you slice the cake into equal pieces, some
choice is involved, but you can’t single out one slice as the unique result
of choice and the others as compelled.
Poincaré was simplistic also in his emphasis on simplicity as playing a
decisive role in the choices made along the way. There are many desiderata
for theories in science. The bottom line is certainly and only empirical
adequacy. Pragmatic considerations such as simplicity provide desiderata as
well, but they are various and diverse. There is a practical need to align the
continuation of physical theory in form with what is in place already. Thus
the desire for explanation, or for conformity to certain favored patterns such
as common cause models, could outweigh simplicity. None of this qualifies
the important point, that the crucial and necessary coherence constraints
on what qualifies something to play the role of physical correlate to meas-
urement leave much leeway—so that choice is inevitably involved—and
that there is no independent recourse beyond coherence.
To be sure, this requirement of coherence is not simply one for logical
consistency. Whether a sort of mechanism can be used to define the family
of standard clocks depends on empirical regularities that may or may not
obtain. The central coherence condition on the family of standard clocks,
recall, was that if two are in coincidence, they run at the same rate, they run in
synchrony. That is a matter of empirical fact. But if two sorts of mechanism
satisfy this condition, there is no matter of fact as to which runs evenly, and
a choice or convention alone can decide on one of them.
There is a natural drive once the role of such choices is recognized, to
reformulate theory in a format independent of those choices. The most sali-
ent example is the ever greater generality with respect to coordinatization,
which had begun earlier in mechanics, and the later extreme freedom in
this respect aimed for in general relativity and even more in the space-time
theories of Weyl and Eddington.36 The other great example is the advent
of group-theoretic formulations, in which invariance under as wide a class
of transformations as possible is the guiding aim. But eventually these
theoretical representations, which abstract from applicable models to that
extent, have to be brought back into applicable form, and there choices
determining the particularity of that form come back into their own.
    137
Observables coordinated: two morals

There are at least two morals to be drawn here, and the first is an anti-
foundational moral we can draw from Reichenbach’s impasse. The rules
or principles of coordination that can be introduced to define particular
sorts of measurement cannot even be formulated except in a context where
some forms of measurement are already accepted and in place. In the use
of astrolabe and ruler to measure distant lengths the local measurement
of lengths and angles was taken for granted. In the construction of
thermometry, the measurement of volume and weight, and later pressure,
was already in place. In the measurement of duration, measurement
determining pendulum periods, and later on astronomical alignments, was
presupposed as already possible throughout. There is no presuppositionless
starting point for coordination. We are, to adapt Neurath’s metaphor, sailors
engaged in a continuing construction, renovation, alteration, and repair of
the ways in which we measure our ship while already at sea.
Reichenbach, I would say, erred on the side of philosophical prudence,
precision, and caution, because he realized the crucial importance of
bestowing genuine empirical content on the equations in a theoretical
description. A case of philosophical scruples . . . but scruples, as the Church
emphasizes, are also a source of error.37
There is a second moral. The opposite error is more tempting, and less
excusable. It is the error that Reichenbach described and from which he
rightly shied away (even if too far). For, as he saw, the natural temptation
in response to the problem of coordination is simply to impose a parallel
vocabulary and declare victory. One might say, for example,
a point in Minkowski space corresponds to a real or physical space-
time point, which is a compact convex part of the  with zero
measure.38
Reichenbach would be quite right to retort that this parlance is meaningful
only after the problem of coordination is solved. It cannot be a solution!
For this ‘‘solution’’ begins by thinking of the  as a structure, a
set of elements with certain relations between those elements, described
precisely in the terms we use in geometry. So it takes for granted that we
can represent the  mathematically. And what is that, if not to say
138  : , ,  
that there is a certain ‘representing’ relationship between the  and

some mathematical structure?—thus assuming precisely what was to be
illuminated.39
Reichenbach offers another familiar example as illustration. Boyle’s Law
is an equation: PV = rT. To give this physical meaning, its terms must be
‘coordinated’ with physical quantities, pressure, volume, and temperature
of gases. But what is, for example, the temperature of a gas? It changes with
time, and differs from one body of gas to another, so isn’t it a function that
maps bodies of gas and times into the set of real numbers? But if I say that,
am I not already taking for granted that I have at my disposal a description
of bodies of gas and times? Could this be a description that does not use
terms relating to spatial boundaries and numerical comparison? I can try
to cut the Gordian knot by saying ‘There exist physical quantities P*, V*,
and T* which pertain to bodies of gas, and when ‘‘P’’, ‘‘V’’, and ‘‘T’’ refer
to these respectively then ‘‘PV = rT’’ is true’. That is to do just what I
said above: to create a parallel vocabulary and declare victory. But it is an
empty victory.40
What that sort of naïve response ignores is that there is no independent
epistemic access to the parameters to be measured—no access independent
of measurement, that is. How could one decide, before a detailed theory
is in place, whether or not the changing height of a column of mercury
mirrors the changing temperature, except by use of a thermometer?41 So,
before thermometry has been established, how could one decide that at all?
The answer is that the question is absurd.42 A metaphysician may postulate
that there are such physical magnitudes P*, V*, and T*, but s/he is building
castles in the air if the next step involves some god-like vision or ontological
telescope to compare their values with what our instruments show.
Instead, as Mach’s and Poincaré’s analyses show, measurement practice
and theory evolve together in a thoroughly entangled way. Somewhat
hesitantly one might say that the measured parameter—or at the very least,
its concept—is constituted in the course of this historical development.43
Choices are made, and once made may encounter resistance, whether
in experiment or in theory-writing or (more usually) in combination of
the two—or else vindicated by smooth progress on both fronts. Mach
oversimplified by keeping theory as much in the background as possible,
and Poincaré oversimplified by suggesting that it is mainly a matter of
submitting definitions, in such a way as to keep theory as simple as possible.
    139
But their simplifications, even if going too far, honor the insight that the
parameter that is measured is identified in the historical process by the
envisioned eventual stable measuring practice, while it is differently identified
in retrospect by the theory that draws on that history for its credentials.44
Within the vantage point of the accepted theory, once such stability has
been achieved, we can speak meaningfully of the accuracy with which a
given instrument gauges a given observable. At that late point, when the
parameter has found its place in the theory that has emerged now, it is of
course a characteristic which is no longer defined by measurement, but by
its role in the theory. That is right and legitimate also from an empiricist
point of view, for this discourse draws on how objects, phenomena, and
instruments are classified by (or classified relative to) the accepted theory.45
It does not presuppose an impossible god-like view in which nature and
theory and measurement practice are all accessed independently of each
other and compared to see how they are related ‘in reality’. The two ways
of looking at the matter we must combine in a synoptic vision. The first
is ‘from within’ the historical process in which measurement procedures
and theory are stabilized. The second is ‘from above’ with the theoretical
description of the domain including the measurement interactions already
in hand, a stage achieved in that historical process but no longer involving
any explicit reference to its own history.46
We are not done with measurement, even with this realization. There is
still a whole battery of questions about measurement that are not answered
by our escape from the simplistic notion of number-assigning. On the one
hand, a measurement procedure is a physical interaction. On the other
hand, the measurement outcome ‘locates’ the object or process measured
in a certain space of possible states. The physical and intentional aspects of
measurements are to be looked into separately, before we bring them back
into a synoptic view of how theory relates to experiment and observation.
6
Measurement as Representation:
1. The Physical Correlate
We have looked at measurement ‘from within’, and will now look ‘from
above’: that is, try for a view that we can have after the measurement
procedures, concepts, and theories have stabilized. This resting point in
the conceptual and scientific development is, to be sure, fleeting and
momentary only—yet marking a context in which some things can
be taken for granted pro tem, where it makes sense to ask about the
world-picture of currently accepted physical theory.
For the question What is measured? the most direct answer is Physical
magnitudes that characterize the objects measured. At the point where the
pertinent theory is stably established, what the relevant physical magnitudes
are and how they are measured is itself specified theoretically. We cannot
separate the questions What is measured? and What is a measurement?, and both
of them have, eventually, specific answers from within the pertinent theory,
which will classify certain interactions as measurements and their final stages
as outcomes. The interaction in question I will call the physical correlate
of the measurement. The criterion for what sorts of interactions can be
measurements will be, roughly speaking, that the outcome must represent
the target in a certain fashion—, selectively resembling it at a certain level
of abstraction, according to the theory—it is a representation criterion.
Physical conditions of possibility for measurement

Recall for a moment Nelson Goodman’s example of the Duke of Welling-
ton’s portrait that resembled the Duke’s brother more than the Duke. It
shows that what the painting is a painting of is not settled by resemblance.
142  : , ,  
Granting that, let’s ask though how we should classify this example, while
thinking of how Dürer’s Art of Measuring depicts the perspectival painter’s
art. Perhaps the portrait resembled the brother more, because the brother
was in fact the artist’s model while painting the portrait? That would not
have been so remarkable—perhaps the artist had noticed how much the
brothers did resemble each other, and the Duke was not always available
for sitting. Not only not remarkable: in no way threatening the status of
the painting as the Duke’s portrait. To take a more extreme case, Dante
Gabriel Rossetti’s painting The Girlhood of Mary Virgin depicts the Blessed
Virgin in the house of her parents. But the model for Mary was Christina
Rossetti, while Mrs. Gabriele Rossetti was the model for St. Anne. That
does not make the painting a family portrait of the Rossettis. It is without
a single doubt a picture of the Virgin Mary’s childhood, although the
painted figures—equally undoubtedly—resembled Christina and Frances
Mary Rossetti more closely than either the Blessed Virgin or her mother.
What if the artist commissioned by the Duke of Wellington had not only
used the Duke’s brother as model, but had used one of Dürer’s machines
for drawing, so as to produce the painting? In that case we say that the
drawing is a measurement outcome, but the measurement was surely not a
measurement of the Duke! So now the ‘‘of’’ in the two contexts diverge:
we will still grant the result the status of portrait of the Duke, however it
was made, but not that the operation carried out was a measurement of the
Duke.1
While the criteria for what a portrait is a portrait of are various, cultural
context-dependent, and sensitive to social factors, that is not quite so for
measurement, however similar they may be in other respects. In the case of
measurement the physical conditions of the object-instrument interaction
have more weight, and are in fact the subject for a good deal of foundational
research in the sciences themselves.
Focusing on the physical correlate

The answer What is measured are physical magnitudes that characterize the objects
measured can be elaborated, after science has reached the requisite stability,
in the form:
What is measured is a physical system of a certain sort. Being of a
certain sort, this system is characterized by certain parameters, and a
  :  143
measurement made upon this system is a measurement of one of those

parameters that characterize it.
For at that stable point in the joint evolution of theory and practice there
is a theoretical classification in place, of the objects that can be measured,
before those measurements are made. Even the humble substances to be
found on my breakfast table today are already so classified, by currently
accepted theory, that it makes sense to ask of any one of them whether it
is a sodium compound or whether it contains electrolytes; and there are
measurement outcomes that would answer the question.
While Reichenbach focused on the physical correlate of such theoretical
notions as geodesic, we can now raise questions about the physical correlate
of measurement per se. Measurement is an operation by whose means
we gather information; but this is of course done by way of a physical
interaction between apparatus and object, to play the information-providing
role. Measurement always involves a physical interaction between ‘object’
and ‘apparatus’; that interaction we can call the physical correlate of the
measurement.2 This distinction, between the measurement and the physical
interaction by whose means it is accomplished, comes into force when we
ask just what the theory, speaking in its own terms, must provide by way
of a representation of the measuring process. If that interaction is in the
theory’s domain, the theoretical description will be of this interaction in
the same terms as any other physical interaction, and involve no terms that
signify anything intensional or intentional.
At some point below we shall narrow our focus to the measurement
outcome, which is the information provided by the final physical state
of the apparatus in that interaction. There the same distinction needs to
be made: the final physical state of the apparatus is the physical correlate of
the measurement outcome. That is not a matter of meaning, but concerns a
physical precondition for meaning. No information is gathered at all unless
the outcome means something; but it can’t mean anything relevant unless
it appears at the end of a suitably structured physical interaction, and takes
a particular sort of physical shape.
Let us concentrate on the case in which physical theory is taken broadly
enough so that this interaction is in its proper domain. We take it for
granted then that the theory provides a representation of that physical
interaction, in physical terms.
144  : , ,  
We can put this partly in terms of language. A claim of the form ‘‘This
is an X-measurement of quantity M pertaining to S’’ makes sense only
in a context where the object measured is already classified as a system
characterized by quantity M. To so describe an object is already to classify
by theory. Therefore the claim is theoretical or at least theory-laden, and
has to be treated as such.3 It is a claim which will change in content and
in truth conditions as our accepted theories change, since our classification
of physical systems changes along with the theories. Here too we see a
difference in the views from within and from above. From the posterior point
of view the new theoretical claims are not displacements but refinements of
the old—precisely because the old claims are now seen as (usually in some
way imperfect) versions of less precise claims formulable in the new theory.4
This context-dependence of the claim may not be visible when we
discuss a measurement procedure that was stabilized so long ago that
we now take the description of the relevant objects and apparatus for
granted. No one today would think of the use of a thermometer as
being a meaningful practice only within certain theory-laden contexts. The
practical use of these instruments has attained such stability that it is for now
at least not at all sensitive to changes in the background theories. But that
context-dependence springs to the eye in cases where the relevant theory
is relatively new and has some claims to descriptive completeness—such
as relativity theory or quantum mechanics.5 In both of these, the behavior
of the measurement apparatus is indeed within the domain of the theory
itself, and the criteria for qualifying as [physical correlate of] measurement
are to be presented in the terms of the theory—with consequent questions
of coherence and consistency inevitably forced on us.6
Philosophical reactions to this point have included a good deal of
extremism. There has been outright denial, as in the operationalist and early
positivist delusion that we could have a hygienic ‘observation language’ in
which the measurement operations and their outcomes could be described
free from all theoretical content. Though we still sometimes see the logical
positivist and logical empiricist circles of Vienna and Berlin identified with
this point of view, it was in fact quickly abandoned there. The critique by
Hanson, Toulmin, and Kuhn in the fifties and sixties removed it from the
philosophical scene altogether—or should have.
The opposite extreme in denial came then, ostensibly in a famous
chapter by Kuhn: scientists in different centuries live in different worlds.
  :  145
That extreme is still worth some scrutiny even now. Understood in its
most blatant literal sense, it involves a denial that there is any such thing as
the physical interaction prior to or independent of the theoretical context
in which it is classified as such or such a measurement. Just to say this shows
it up for what it is; but for that very reason, shows that this was not what
was meant. The real point is that occurrence of the physical interaction is
one thing, its playing the role of representation of the value of a certain
parameter is another; and today everyone understands that.7
To escape from both extremes, we need to separate quite clearly the his-
torical process, examined in the preceding chapter, in which measurement
and theory are jointly stabilized, from the retrospective characterization of
measurement in the eventual established physical theory.
The question of coherence

How is the measurement interaction represented in the theory? The answer will
be different depending on which measurement interactions are meant, and
which theory. Stepping back from the specific cases for now, we ask the
more reflective question: what must the answer be like, if it is to be satisfactory?
The main criterion must surely be this:
the theoretical characterization of the measurement situations is
required to be coherent with the claims about the existence of
measurement outcomes, their relation to what is measured, and their
function as sources of information.
For this coherence it is required that the theoretical characterization of the
measurement interaction allows for a coherent story about how its outcomes
provide information about what is being measured. These coherence
conditions we need to spell out when we characterize the physical correlate
of measurement, in a context where it is assumed that both sides of the
interaction have already been well understood and classified.
Recall that ‘‘This is an X-measurement of quantity M pertaining to S’’
makes sense only when the object measured is already classified as a system
characterized by quantity M. This will be so if the context of discourse is
governed by a theory; and it is in this kind of context that we can sense
a challenge to consistency or coherence. What does this theory, or an
accepted background theory, imply about the measuring process whence
the theory reaps its credentials?
146  : , ,  
On the simplest view, the outcome of a measurement of quantity M

on system S reveals the value that M has at the time the measurement is
made. Even in the most familiar cases, however, this must be qualified.
The process takes time, so is what is revealed: the value of M at the outset,
or at the end, or . . . ? The final reading on a thermometer placed in a
cup of tea we can take to correspond to the tea’s final temperature. Why?
Because the theoretical description of the interaction implies, via a pertinent
conservation principle, that the two end up with the same temperature.
The coherence of this reading of the thermometer is thus underwritten
by theory. But the question that was most likely being asked was how
hot the tea was? In other quite familiar cases, measurements are destructive:
the patient does not survive the test [unaltered], the photon is absorbed,
the metal sample fractures or is vaporized. Then the measurement does
not reveal the final state of the object measured, since the object exists no
more. But the test result should at least show whether, for example, the
patient did have one illness or another.
Measurement can interfere or destroy—and perhaps the most direct
indication it typically yields is of the state of the object at the end. But
such a process will not serve its function as measurement (rather than, say, as
preparation) unless some information can be gleaned about what the target
was like before the measurement—isn’t it?8 So, granting that the outcome
has its significance as such only in a theoretical context, let us ask: what
information can be derived from the measurement outcome, concerning the
initial state of the object, sub specie the theory? In the case of the thermometer
the question would be: given the final reading, what can be inferred via
thermodynamics about the tea’s initial thermal state?
It looks like circularity threatens at once: to infer the initial temperature
of the tea, we’d have to know the initial temperature of the thermometer,
among other things. But the circle is easily broken: thermometry already
relies on some identified fixed point, such as the freezing point of water,
and the thermometer can be prepared to be initially in thermal equilibrium
with such a fixed reference point. Can the circularity be so easily evaded
in general? I think not. The familiarity of these classical examples and the
stability long since achieved in their treatment may tempt us to assume
that, but in fact, there are puzzles to be faced.
  :  147
General theory of measurement

The greatest felt tension between the measurement procedures that we
know by acquaintance and their description in a theory is found in
quantum mechanics. Not surprisingly, therefore, that is where we find
also the most extensive treatment of the physical side of measurement.
Under the heading of ‘‘quantum theory of measurement’’ we do not
find solutions to philosophical problems.9 Instead we find there simplified
and idealized but straightforward accounts of the physical [part of the]
process. These accounts have a general form that has considerable claim
to universal applicability in the empirical sciences. I shall take that form,
depicted non-technically, as my central example of theoretical descriptions
of measurement.
Initial measurement set-up

A proper measurement process must involve, besides the object on which
the measurement is to be performed, a separate system to play the role
of measurement apparatus. The procedure is meant to address a particular
parameter, property, quantity, or observable (essentially interchangeable terms)
that pertains to the object. The apparatus is to be coupled to the object in
such a way that a process will occur that counts as a measurement.
Figure 6.1. Measurement Schema
What are the criteria? If that process is a measurement of property A,

say, then there must be a corresponding property B, of the apparatus,
148  : , ,  
whose final value will be the outcome. That process has to be governed
by an equation which has to guarantee something about properties A and
B and that final value—but classical intuitions beware! In addition, the
system denoted as apparatus has to be such that it will play this role quite
independently of what state the object is in—give it any object in any state
and it will play its role.10
What must the governing equation guarantee, in order that this process
really be the measurement of some property pertaining to the object? In
a classical context we might say something like: if the object is thus or so
then the end-state of the apparatus has to indicate precisely that.11 This
assumes that the parameter measured already has a definite value before the
measurement, a value that can be revealed by the measurement outcome.
But this ‘‘revelation’’ assumption is at best controversial and certainly
generally rejected in foundational studies of quantum mechanics.
Measurement does not ‘reveal’ (?)

Complacency about measurement was sorely tried as the quantum theory
developed. There are now many arguments in the literature to show
how that theory requires a revision in our demands for what constitutes
measurement. Instead of repeating these here, I will illustrate such reasoning
with a simplified example that displays the sort of cautions we must
entertain.12
Consider a strange situation characterized by three physical parameters,
A, B, and C. The data are that, for each, measurement yields values 0 and
1 with equal frequency. We are able to measure them two at a time, and
always find opposite values. Here is a simple theory that generalizes on
these data:
For X, Y = A, B, C: the probability that X has value 1 equals 0.5, as
does the probability that it has value 0. But the probability that X and
Y both have value 1 equals 0 if X = Y
But this simple theory is self-contradictory. For it assigns probability 1 to
both of the following:
No two of A, B, C have the same value
Each of A, B, C has value 0 or 1
  :  149
It is impossible for both to be true! What went wrong? The data listed
were about measurement outcomes, while the extrapolated probabilities
were absolute and unconditional. The proper generalization must give due
attention to the fact that what was found was how things appear in the
measurement set-up:
For X, Y = A, B, C: the probability that X has value 1, conditional on
its being measured equals 0.5, as does the probability that it has value
0. But the probability that X and Y both have value 1, on the condition
that the two are measured together, equals 0 if X = Y
Now there is no contradiction, but rather the consequence that A, B, C
cannot all be measured together. So that is our solution.
How good a solution is it? We want to ask, surely, what value C has
while A and B have values 1 and 0 respectively. But how this question is to
be understood depends on what we take for granted here. Can we take for
granted that when not measured, A, B, C still can only have values 1 or 0?
Or could they have other values when not measured? Or perhaps have no
value at all? And secondly, can we take for granted that if a measurement
of A shows value x, then at that moment, A has value x? Or that it had
value x at the beginning of the measurement?
Could we perhaps reason like this? Suppose A is measured and the
outcome is 1. Now we can predict that a measurement of one of the
others, B or C, will have outcome 0, with certainty. On this basis, can we
assert that B and C already have value 0 at this moment? If we do, we will
have to add that joint measurements of B and C that are actually carried
out are systematically deceptive, for they never show them having the same
value.
Before seeing an example like this, or at least before having any inclination
to take it seriously, various assumptions involved in such reasoning would
likely have been taken for granted when thinking about measurement.
Classical intuitions (if such beings exist) suggest two postulates:
Value Definiteness: Each physical parameter always has some value,
namely one of the values which may be found
by measurement.
Veracity in Measurement: Measurement of a parameter faithfully reveals
the value it really has.13
150  : , ,  
These two postulates can be consistently added to our above story, but then
they imply that either measurement is subject to or involves a systematic
alteration, or else some sort of conspiracy in nature constrains when
measurements are made: when A and B both have value 1, we are lucky or
clever enough not to measure the two of them!
Perhaps we should put it this way. The conjunction of these postulates
would be an attempt to say that the world is basically the same, whether
things are being measured or not. But given the above story, the two
postulates are both true only if things are not basically the same whether
things are being measured or not! So the attempt fails: some difference
between measured and unmeasured world will have to be admitted as a
possibility here.
Paul Feyerabend’s (1958) name for the postulate of Value Definiteness
was Classical Principle C. This principle must be rejected, he argued.
From Bohr to Feynman, physicists have expressed similar opinions: an
observable (measurable parameter) might not have a specific value outside
the context of measurement. However, the second postulate—Veracity in
Measurement —has also been much looked upon as a candidate for rejection
or revision. To keep the first postulate and reject the second, by means of
an explanation through uncontrollable disturbance by measurement, would
not be a happy option. It would imply some sort of conspiracy again: if
A and B do sometimes both have value 1, how does the ‘‘uncontrollable’’
disturbance in measurement carefully and systematically hide that fact?
Finally, rejecting Value Definiteness would by itself already imply a
weakening of Veracity in Measurement. For if, at a certain time, parameter
A has no value, and is measured, then this measurement yields a value as
outcome, but clearly does not reveal a value.
When we specify what counts as a measurement of A, we describe a
physical arrangement which must have one of two outcomes (indicator
values), in this case 0 or 1. For this to be a measurement, hence to play a
role in information gathering, it must surely do something that is revealing
about what is measured? But what sort of information it does yield, and
how much, we shall have to consider very carefully.
General theory form

What the apparatus is like at the end must reflect some pertinent feature of
what the object was like initially. There must be a transfer of some character
  :  151
of the object’s initial state to the apparatus’ final state.14 Otherwise there
would be no way to use this process to gather information about the object
on which the measurement is performed. How this ‘transfer’ requirement
is to be made precise will of course be different in different theories.
For the general form we must allow for the case in which the relation
between physical state and measurement outcome is only characterizable
in terms of probabilities. A deterministic theory can be thought of as the
special case in which all probabilities are zero or one. The measurement
situation modeled as theory prescribes, when the apparatus is itself in the
theory’s domain of application, must include a specification of the following
factors:15
a family M of observables (physical magnitudes) each with a range of
possible values;
a set S of states—physical states of both the system measured and of
the measuring system;
a stochastic response function P s m for each m in M and s in S, which is a
probability measure on the range of m; with P s m to be interpreted as
the probability that a measurement of m will yield a value in E, if performed
when the state is s.
Suppose now that one sort of process represented in the theory is that
of the interaction that qualifies as measurement of an observable. The
situation depicted then involves two systems, the object measured S and
the apparatus R by which it is measured. Together S and R constitute a
larger system, a ‘two-body’ system, S + R. The family of observables must
then include some that pertain just to that object S, some that pertain just
to the apparatus R, and some that pertain to both at once, that is, to system
S + R—and similarly for the states. But there must also be a constraint on
how this situation evolves, as the two objects are coupled and interact.
Classifying R as an apparatus, for measuring observable A for example
(an observable pertaining to the object to be measured) entails that this
interaction will take a certain form, which qualifies the designation as
‘‘apparatus for measuring A’’. In fact, this interaction must be such that
something pertinent about the initial state of the object is reflected in the final state of
the apparatus. So imagine the apparatus as having a dial with a pointer—as
a system it is characterized in part by an observable B, the pointer observable,
152  : , ,  
whose possible values are precisely the numbers which the pointer can
indicate on that dial. Now the criterial condition, in its strictest form, must
be this:
Criterion for the Physical Correlate of Measurement: PB fin (E) =
PA init (E)
where fin is the final state of the apparatus, and init is the initial state of
the object on which A was being measured. That is, the final state of the
apparatus must reflect, in its probabilities pertaining to pointer observable
B, the probabilities pertaining to measured observable A in the initial state
of the measured object.16
This includes as a special case that the pointer observable B on the
apparatus would most certainly show e.g. value 17 if inspected at the end,
on the supposition that the measured object was initially in a state in which
observable A ‘most definitely’ had value 17. To this extent then the old
criterion of Veracity (or revelation) is being honored still.
The coherence constraint

But notice the qualifying clause ‘‘if inspected’’ and the qualifying quotation
marks that we found it necessary to insert in the preceding paragraph.
For the probabilities PB fin (E) and PA init (E) are understood as conditional
probabilities—conditional on measurement!
At this point one begins to have the sense that there may be ways into
the enchanted wood but no ways out. What does the Criterion for the
Physical Correlate of Measurement come to when we look at it in this
light?
Firstly, the theory will have implications for results of other measure-
ments (of observables A , A , etc. ) on systems with the same sort of
preceding history, depending on the results from the A-measurements.
This is one way in which inconsistencies could arise with the theory when
different measurements are made, so we have here an empirical coherence
condition. This condition depends so much, however, on the specifics of
the theory in question that it may not be possible to say more about it in
general.
Secondly, the theory must imply that if two other apparatus were coupled
to these systems, to the object at the beginning and to the apparatus at the
end, the probabilities of the corresponding outcomes would be the same.
  :  153
Figure 6.2. Coherence of Measurement
That is, suppose that we had two further apparatus, R1 and R2, the
former one also used to measure A on system S and the latter to measure
B on apparatus R. Then the Criterion’s being satisfied mutatis mutandis for
all measurement interactions will guarantee the coherence of the ‘exterior’
measurements with the ‘interior measurements’.
For a comparison of the probabilities pertaining to R1 and R2 should
show an accord that ‘reveals’ that the accord between S and R demanded
by our Criterion was satisfied. The theory must predict that if a second
independent measurement was made to secure the premise that the object
had been prepared in a certain initial state, and a third measurement were
made of observable B on the apparatus end-state, then the demanded
accord would be found with the requisite probabilities.
This is therefore a coherence constraint on the theory: it must first of all
have this internal harmony in what it predicts, but secondly, comparisons
of results in the two kinds of set-ups must be empirically vindicated. So
here we have a coherence condition that is partly internal consistency and
partly empirical. The theory is to satisfy that coherence constraint—and
that is the most we can ask of it. Put in terms of the sort of description
the theory provides for physical systems in general, this ‘physical theory of
measurement’ is not plagued by any sort of circularity.17 At the same time,
of course, it does not pretend to aim at a definition of measurement in
terms that have nothing to do with measurement!
In practice, the level at which a theory confronts experience is not that
of raw data taken from individual measurement outcomes, but of the ‘data
154  : , ,  
models’ constructed on their basis, and the further smoothing of the data
models in which for example sequences or discrete variables are replaced
by continuous functions.18 But the conceptual problems—such as the
‘measurement problem’ of quantum mechanics—refers to the individual
outcomes of measurement interactions, modeled in the above fashion. I will
return to this distinction between data and data models below. The point
is that, however we conceive of this, the above coherence constraint will
have to apply to how the data from measurement are to be accommodated
by the theory’s theoretical models.
Veracity reconsidered
The criterion imposed on the physical correlate of measurement is as
strong as possible, given the general form of a physical theory that was
under consideration. That form, in turn, was kept very permissive, so as to
allow for the form that quantum mechanics took when it was formulated
definitively c. 1925. As a result, the conditions on measurement had to
allow violations of the two ‘classical’ principles that we had noted: Value
Definiteness and Veracity in Measurement.
As to Value Definiteness, nothing in the Criterion for the physical
correlate of measurement precludes the observable A to have no value (or
only an ‘unsharp’ or ‘fuzzy’ value) at the outset. In fact it is not even
implied that the pointer observable B will have a specific one of its possible
values in the final apparatus state. All that is implied is that if measurements
be made, to measure those observables, the possible values will appear as
outcomes with certain probabilities. Value definiteness is not implied in any
sense, way, or form.19 Veracity is implied only in the very much weakened
form of accord among conditional probabilities. But if the theory specifies
nothing beyond those conditional probabilities then no stronger criterion
can even be formulated for the physical correlate of measurement, to the
extent that the theory can cover that.
If we now return to the empirical assertions, we are not bound to stay
within the theoretical description, and we can refer freely to the actual
outcomes of measurements, as typically summarized in data models. These
data models are constructed from the raw data that are actually gathered,
so we are dealing here with actual frequencies, and probabilities that have
a good fit to those frequencies. Suppose now that such a ‘summary’ is
pretty well a picture of a state in a theoretical model, to the extent of
  :  155
displaying probabilities derivable from that state. Then the theory may, and
generally does, have implications for how that state evolves in time.20 Thus
predictions can be made about what will be found if new measurements
are made in that situation at a later time.
As an example, we can take the Stern–Gerlach apparatus, named after
Otto Stern and Walther Gerlach’s famous 1920 experiment. There is a
classical picture behind the idea of the experiment: imagine a ‘beam’ of
particles being emitted, with a particle being like a classical dipole with
two halves of charge spinning quickly. In a magnetic field, such a particle
will begin to precess. This way the particle’s position becomes perfectly
correlated with its spin value. That is also the result, mutatis mutandis, with
the situation described in quantum mechanics. When the spin in a given
direction can have only two values, the beam is split into two separate
beams, ‘upper’ and ‘lower’. The early attempts to realize the experiment
encountered many difficulties (cf. Bretislav and Herschbach 2003). In the
first realization a beam of silver atoms (produced in an oven, at temperature
1000 ◦ C) was collimated by two narrow slits (0.03 mm wide) and traversed
a deflecting magnet 3.5 cm long with field strength about 0.1 tesla and
gradient 10 tesla/cm. The splitting achieved was only 0.2 mm, and there
were doubts as to the data obtained. (Of course the set-up is now described
in rather more ideal terms when it is used to illustrate quantum properties.)
The apparatus can be rotated, so as to measure spin in any direction.
Thus data on different spin observables can be collected on some samples
produced in the oven, and on the basis of the frequencies in those samples,
it is possible to infer—via the theoretical description and classification
of this process—just what state is prepared by that source.21 Then the
proportions of the output in the two channels in later measurements can
be predicted on that basis.
It would be illegitimate to conclude that the silver atoms exiting in
the upper channel were prepared in the oven in a state of spin-up.
Rather, the oven prepares a beam of atoms in a state which is such that
the probabilities of a Stern–Gerlach measurement having outcome up or
outcome down are definite, with the result that the relative proportions
in the two channels are definite. But still we can see now that Veracity is
honored at some appropriate level. The outcome does not reveal a prior
state for an individual silver atom, but the frequencies in the outcome do
give information about the prior state in which the source prepares what it
156  : , ,  
sends out. If it were not so, the role of measurement would not be played
at all, since the outcomes would not serve to provide information about
what the measurement is performed on. In practical terms it is precisely the
source on which the measurement, taken as a whole, is performed.
What is not measurement

In discussing the physical correlate of measurement we are displaying
physical preconditions, conditions for the possibility of measurement. These
do not define what measurement is, and in fact there will always be many
interactions conceivable in nature that satisfy those conditions but are not
measurements. For measurement is information gathering, a measurement
outcome is something that has meaning, is in fact a representation of what
is measured, and that point does not reduce to a physical condition.
In the foundational literature on physics, this distinction is mostly
honored by neglect—reasonably so, since the focus there is on what is
within the intended domain of application of physics. In the sense of
‘‘measurement’’, as the term is typically used there, there are (according to
the pertinent theory) measurements going on all the time in the stratosphere
and at the sub-atomic level, far away from what we can do or use. So in this
usage of the term, there is no reference to practice or to our information
gathering.
On the macroscopic level too we can think of processes that connect
two situations separated in time or space. These could be so correlated
that it would be possible in principle to get information about the one by
inspecting the other—provided of course we knew of that correlation! But
that something could be done does not mean that it is done. That something
could be assigned a representational role does not mean that it has one.
There are contexts where the distinction between a measurement and its
physical correlate can be neglected, but philosophical reflection is not one.
7
Measurement as Representation:
2. Information
A measurement outcome is something physical: an event, the end-state of

the apparatus, or an object (photo, graph, list of numbers) produced by the
measurement process. On the other hand, measurement is information-
gathering, so a measurement outcome has a meaning (i.e. information con-
tent). The information provided is ‘‘read’’ off the measurement outcome,
which is a physical state or event, but one with meaning to the literate.1
The term ‘‘measurement’’ is an endorsing term. If we call something
a measurement, we imply that there is something correct or valuable in
the way it yields a representation (even if on this particular occasion it
went wrong). Measuring is an operation by which we can produce or
gather information; and here ‘‘information’’ too has an endorsing sense.
The communication engineer’s neutral usage of ‘‘information’’ has begun
to modify common usage to some extent. But even now it would still be
puzzling or provocative to hear ‘‘We get information from observation,
measurement, fictions, lies, and popular astrology.’’ If we insist that meas-
urement is information gathering, we mean in part that fictionalizing or
speculating or guessing is not measurement.
Measurement is an operation, using something that functions as an
instrument, to gather information. The instrument is being used, in
Heidelberger’s terminology noted above, in a representative role, that
is, the operation has as outcome a representation of the object or situation
operated on.2 Since many sorts of operations can have such outcomes,
this notion needs to be narrowed. Attention to more complex measure-
ment situations, that need to be assimilated via a series of steps in model
building, leads—I shall argue—to the need for a broad concept in which
measurement is itself a specific form of [self-]location.
158  : , ,  
What is measurement—number-assigning ?
We do find simplistic answers sometimes even in places where it matters.3
Lord Kelvin, with a well-deserved reputation as scourge of purported
sciences outside the ken of physics, wrote:
I often say that when you can measure what you are speaking about and can
express it in numbers you know something about it; but when you cannot
measure it, when you cannot express it in numbers, your knowledge is of a
meager and unsatisfactory kind [and] you have scarcely . . . advanced to the stage
of science . . . . (Thomson, Lord Kelvin, 1891: 80-1)
This sort of view was developed more systematically by Norman Campbell

(Campbell 1920, 1928, 1943) in his influential writings on measurement.
More historically important to the subject is S. S. Stevens’s seminal paper
for the behavioral sciences in Science 1946, which appears at first sight to
express Kelvin’s view, for he writes
[W]e may say that measurement, in the broadest sense, is defined as the assignment
of numerals to objects or events according to rules. (Stevens 1946: 677)
[W]e may venture to suggest by way of conclusion that the most liberal and useful
definition of measurement is . . . ‘the assignment of numerals to things so as to
represent facts and conventions about them’. (Ibid., 680)
Stevens’s article was responding to a commission which had attempted to

arrive at a general, comprehensive account of measurement. The focus
in that historically important paper was quite narrow: the mathematical
foundations of measurement, in the sense of a study of conditions under
which an algebraic structure is embeddable in the real number continuum.4 Almost
all of his discussion pertains solely to measurement scales and the sort of
invariance that is required for meaningful scaling, and is only superficially
linked to either the process or the role of measuring in scientific practice.
The algebraic structures in question are described informally in terms of
physical addition operations, but this is kept skimpy and abstract. Stevens
is able to maintain those slogan formulations only by, on the one hand,
extending the concept of number beyond its normal use, and on the other
hand, by restricting himself to the abstraction involved in a mathematical
point of view.
  :  159
In fact the idea of an essential link between measurement and number-

assigning comes from taking a myopic view of mathematics and an equally
myopic view of measurement.
Mathematics is not the study of numbers. Numbers are merely a particular
if useful and user-friendly instance of mathematical structure. Granted, the
numbers’ salience is not the only reason to think of measurement mainly in
terms of numerical outcomes. Some paradigmatic examples of measurement
are indeed numerical. But as Dürer’s Treatise on Measuring already brought
out, we will have to look beyond those.
Kinds of scales
The guiding idea for the study of measurement scales is that the grading
must be thought of as reflecting characteristics of operations on something
physical—operations that can plausibly be called measurements in some
‘ordinary’ sense. Presumably Stevens was referring to this with ‘‘according
to rules’’, though what sort of rules for operating on real physical things
count as measurement recipes was left aside. The constraints on the possible
outcomes that he describes are designed so as to guarantee that what can be
measured is precisely what can be represented as graded on a real-number
scale.
Mathematically speaking there are many structures other than the real
number continuum. Even in the field called ‘‘foundations of measurement
theory’’, the conception is already so generalized that the standard numerical
version can only be seen as a particular case. The authors in that field
are aware of the limits of numerical representation, and Stevens himself
introduced the now commonly found distinctions:
nominal measurement is the assignment of (numerical) labels without
implying any algebraic structure;
ordinal measurement assigns a rank ordering;
interval measurement is ranking on a scale where only the intervals
between elements are numerically comparable;
ratio measurement is ranking on such a scale where there is also a
minimum, and the ranking can be represented by non-negative
numbers with the ratios between these numbers reflecting a
physical relationship as well.5
160  : , ,  
Each of these categories has its examples. The Mohs hardness scale for
minerals is the typical example of an ordinal scale.6 If we ignore abso-
lute temperature then our thermometers, whether Fahrenheit or Celsius,
provide an interval scale—the Kelvin scale for temperature with its absolute
zero is a ratio scale.7
But Stevens’s taxonomy is not exactly a table of categories supported
by a transcendental deduction! It looks nicely hierarchical: we can suppose
that ordinal measurement must be a special case of nominal measurement
with the labels reflecting the ordering. Then each category besides the
nominal presumes an ordering which is linear: two assigned labels x and y
must either be the same or have one greater than the other.
As was much emphasized in nineteenth-century discussions, notably by
Mach and Poincaré even this ordering requires a contingent empirical
regularity for its coherence.8 To be able to order at all, even in its most
minimal logical sense, one needs at least a criterion of equality. Suppose
this be specified, and suppose that by this criterion A and B are both equal
to C. Does it follow that by the same criterion A will be equal to B? No,
it does not, at least not logically, if the criterion refers to a performable
physical test. But the relationship of equality, that is of having the same
position in the ordering, must be transitive or the ordering falls apart.
(Similarly of course for the relationship of greater than or after if applicable).
What criterion is proposed is up to the proposer, and if there is a a variety
of plausible candidates for the criterion then this may be a matter of choice
or even convention. But whether a proposed criterion can be adopted will
depend in part on contingent empirical regularities, on pain of incoherence
in practice.
Returning now to ordinal measurement: the numerical relationships
between the assigned numbers are not all of them significant. If we rank-
order some items by assigning the numbers 1 to 10, merely on the basis
of some ‘greater than’ relationship, there is no significance implied by the
fact that 7 is as much greater than 5 as 5 is than 3, or that 6 is 3 times as
great as 2.
What is significant is what is invariant under ‘re-scaling’. This is not an
explanation; it is a remark that connects various notions connected with
each other. Take the example of a nominal scale: the members of a soccer
team are numbered 1 through 11. It is important only that different players
receive different numbers, the ordering does not matter. That is equivalent
  :  161
to saying that all permutations of this assignment result in equally admissible

numberings: each such transformation is an admissible re-scaling. But what
decides what is and is not important, what are precisely the relevant features
to be preserved in all and only the admissible transformations of the scale, is
not decided by mathematical characteristics. It is decided by two factors: the
measurer’s purpose and the relationships that can be accessed instrumentally
or practically on which to base the assignment. These factors are outside
the domain of the mathematical theory of measurement.
Significance and invariance

There is an important difference between what we might call ‘‘scale invari-
ance’’ and ‘‘scaling invariance’’. In the preceding Part, in the discussion
of scale models, we saw that a perfectly proportional decrease or increase
in geometric shape may be quite the wrong thing for scale modeling,
depending on what is meant to be studied or predicted. This is a point
about ‘‘active’’ transformations, in which e.g. something literally of great
size is replaced by something of small size for the purpose of some research.
But scale transformations such as the permutation of a nominal assignment, or
change from centimeters to inches in length measurement, are just changes
in bookkeeping. The books must be kept straight—what is important
must be properly represented; but this is a purpose-dependent norm for
representation, not something telling about nature.
Significance is to be equated with invariance. That is, the procedure by
which the numbers are assigned may favor only one large class of ways
to assign them, as against other assignments, and only what is common
to the favored class matters. This was an important insight exploited by
Stevens: the types of scales are to be distinguished in terms of the families
of transformations under which they are invariant.9
Interval measurement illustrates this well. An admissible transformation is
in this case a combination of a dilation (in a sense that includes contraction)
and a translation. As familiar example take the Fahrenheit (F) and Celsius
(C) ‘‘centigrade’’ scales for temperature. Suppose that the high today is
41 ◦ F and yesterday’s was 32 ◦ F. Does that mean that the temperature
today is more than 20 per cent higher than it was yesterday? The scale
transformation from F to C is this: F = 32+(9/5)C. So today the high
was 5 ◦ C, and yesterday it was 0 ◦ C. The ratio of today’s temperature
to yesterday has no invariant sense. But in ratios of intervals we find an
162  : , ,  
invariant. So suppose that on the third day it is 50 degrees Fahrenheit, so

we say that the increase was the same each day. Then in Celsius, carefully
calculated, we find the same: the increase from 0 to 5 was followed by an
increase from 5 to 10.
The rules for measuring temperature allow a large class of number
assignments. In fact, the only constraint on this a class of numerical
assignments is that it is closed under operations of translation and dilation.
That means: if Y is an admissible scale then aY+b is too.
So what significance can a thermometer reading have? The answer, as
we also mentioned, was that interval ratios are invariant:
(aY2 + b - aY1 - b) / (aY4 + b - aY3 - b) = (Y2 - Y1)/ (Y4 - Y3)
There is a way to express the significant part of what is found in a
temperature measurement with single numbers. That is to choose one
interval that can be stably fixed—e.g. the freezing and boiling point of
water—as reference, and pay attention only to the ratio of the intervals in
which that is divided by the temperature reading. If the numbers assigned
to those fixed points are 0 and 100, then the reading will be numerically
equal to the percentage selected of the referent. Thus the Celsius scale is the
most convenient: a reading of 15 ◦ C indicates saliently the proportion (the
interval from 0 to the reading point)/(the reference interval) as 15 per cent.
The choice of fixed reference points is conventional, subject only to
the possibility of stabilizing the measuring procedure to that extent—and
thereby hangs a tale.
The connection between invariance and significance (often referred to,
somewhat confusingly, as ‘‘meaningfulness’’) was studied at length in the
literature on measurement. There is an important connection between this
subject and the technique of dimensional analysis which we encountered
in the discussion of scale models. The seminal article by Duncan Luce
(1978) begins with the question why the laws in physics are required to
be dimensionally invariant, that is, invariant under the admissible scale
transformations. As focus he takes the case of measurement scales:
It is pretty well agreed that the problem of meaningfulness within a single measured
attribute is closely tied to knowing how two different, but equally acceptable,
numerical representations of the qualitative structure are related. Meaningful
  :  163
numerical statements are those that remain invariant under permissible changes in
the representation. That, however, is very similar to saying, in the more complex
case of physical laws, that the representation must be dimensionally invariant. So,
on the face of it, there appears to be a close relationship between meaningful
statements in a single attribute and dimensionally invariant laws stated in terms of
several attributes. (Luce 1978: p. 3)
The result he proves is in effect that features of a relational structure

are invariant under the symmetries of that structure precisely if they are
representable by a dimensionally invariant numerical function.10
Approximative measurement
We need not go far afield to find that linear ordering in a scale is too
restrictive, in general. After all, a measurement outcome is not infinitely
precise: the length of the table is registered e.g. as 100 plus or minus 1 cm.
So here the real outcome is not a number but an interval. These intervals are
ordered by inclusion, which is a partial rather than linear ordering. We can
introduce a notion of ‘strictly greater’, e.g. by the definition that one interval
is greater than another if all its elements are greater than all of the other’s
elements. But of course it does not follow then that the ranks are the same
if neither is strictly greater than the other—so the ordering is not linear.
If this practical point is granted, we have already left behind the idea
that measurement outcomes can always be represented as points on a
linear scale. For the class of regions—however delimited—that can be
indicated as found locations in this way is not a linearly ordered class. It
is a class partially ordered by set inclusion, or by set inclusion modulo
differences of measure zero, or some such relation. We should not call this
‘‘locating on an interval scale’’—that term already has an established use,
as we saw above. Rather, in this case, the object is located in the space
of intervals, or in the larger space of ‘Borel sets’ generated by countable
meet and join operations. This is the range most typically encountered
where measurement results are not assumed to be ‘point-like sharp’. There
an elementary form of statement, which can be either that of a theoretical
assertion or of a measurement outcome, relates a physical parameter to a
set of its possible values, which is itself linked to a defined region in a much
larger state space specified by a theory.11
164  : , ,  
The scale as logical space

Consider a simple theory of gases which characterizes them in terms of
three parameters, P(ressure), V(olume), and T(emperature). This theory
provides a three-dimensional mathematical space, and measurement will
locate given bodies of gas in regions of this space.
The first point to notice is that this use of measurement makes sense
only in the context of the theory in question, which already provides the
general framework for classifying the items to be located in that space.
Secondly, this is an example in which the classification is very simple and
particular. In general, measurement of an item classified as being in the
domain of a particular theory will locate that item rather indefinitely as
somewhere in a space common to a whole family of models provided by
that theory. The farthest but still illuminating generalization of this point
was made by Wittgenstein, inspired by his study of thermodynamics and
mechanics:
1.13 The facts in logical space are the world.
2.013 Every thing is, as it were, in a space of possible atomic facts. I can think of
this space as empty, but not of the thing without the space.
2.0131 A spatial object must lie in infinite space. (A point in space is an argument
place.)
A speck in a visual field need not be red, but it must have a colour; it has, so to
speak, a colour space round it. A tone must have a pitch, the object of the sense
of touch a hardness, etc.
2.202 The picture represents a possible state of affairs in logical space.
The HSB color space, with dimensions hue, brightness, and saturation is
a good example of a logical space, but so is the PVT space in elementary
gas theory, phase space in classical mechanics, Hilbert space in quantum
mechanics; space and time themselves also serve as examples.12 I sub-
mit the following generalization as the proper concept of a measurement
operation:
measurement is an operation that locates an item (already classified as
in the domain of a given theory) in a logical space (provided by the
theory to represent a range of possible states or characteristics of such
items).
  :  165
The act of measurement is an act—performed in accordance with certain operational

rules—of locating an item in a logical space.
While the terms vary, descriptions of actual measurement procedures
typically bear this out. This has been better appreciated in philosophy of
technology than elsewhere. Thus Davis Baird:
Measurement presupposes representation, for measuring something locates it in an
ordered space of possible measurement outcomes. (Baird 2004: 12).
When discussing the thermometer and, in more detail, the Indicator

Diagram for a steam engine, Baird uses the phrase ‘‘field of possibilities’’:
‘‘we have on the thermometer’s glass tube a scale that displays the field
of possibilities that we embrace with the thermometer’’. (Ibid., 2003: 50)
In the steam engine case, the indicator is an instrument that produces a
simultaneous trace of the pressure and volume inside the working cylinder
as the engine runs through its cycle:
The instrument harnesses an instance of material agency, the behavior of pressure
and volume in a steam engine cylinder as it goes through its cycle. [. . .] At the
same time, the indicator presents information. A field of possibilities is constructed
in terms of the pressure-volume graph on which the indicator ‘‘writes’’. (Ibid.: 52)
As a special case, the logical space can be a scale, which may indicate
the location as being in a certain region of a larger space. Thus a pressure
measurement locates a gas in a region of the larger PVT space, a momentum
measurement locates a body in a region of its phase space. So the locating
is typically not in an exact point, but in a region. We already saw this
in connection with imprecise measurement, which assigns not numbers
but intervals, hence takes its assignments from a partial rather than linear
ordering.13 But this holds more generally, since what is measured is usually
only some aspect of that ‘field of possibilities’.
Thus measurement is an act of locating an item in a logical space. The
converse does not hold: you can locate me in the logical space of astrology
simply by asserting that my Sun sign is Aries. Above I added ‘‘performed in
accordance with certain operational rules’’, but by itself that only points to
another question. The astrologer’s or soothsayers’s or visionary’s operational
rules may not count as yielding genuine measurement. What precisely is
needed? That is precisely the heart of the ‘problem of coordination’, hence
requires looking into the joint evolution of theory and measurement. But
166  : , ,  
we can add that once a stable theory has been achieved, the distinction
between what is and is not genuine measurement will be answered relative
to that theory.
Here is a good example: as Henry Margenau and Adolf Grunbaum
discussed, there are certainly procedures that look like simultaneous
position-velocity measurements of particles at any scale. But quantum
theory classifies them as having no such significance—for no operational
outcome can reveal characteristics that, according to the theory, the system
cannot have (cf. Grunbaum 1957: 713–15).
Data models and surface models

It is typical of discussions of measurement in foundations of physics to
retain the simplifying picture of a measurement as a single operation, a
single interaction with the object to be measured. In actual practice, as
pointed out above, that is quite unrealistic.
There are measurements that can be done in one shot: when the nurse
takes the patient’s blood pressure she does not make a series of measurements
and then calculate the mean and standard deviation. But that is in part
because no very great accuracy is required here, and in repeated results there
is unlikely to be a spread of more than a few points. In the general case this
tolerance of inaccuracy cannot be maintained, and a measurement will take
the more complex form. The result obtained in one operation we may call
a datum, the results obtained on one such occasion the data; these data are
subjected to a statistical analysis to reach the official measurement outcome.14
Blood pressure, weight and mass are still very simple examples, where
the outcome may be reported as a single number, or pair of numbers (e.g.
mean and standard deviation). Other, still quite familiar, cases display greater
complexity. On the weather forecast website I consult I can find a graph
depicting yesterday’s temperature plotted against time. This was constructed
from data gathered at various stations in the region, at various times during
the day—this graph is a smoothed-out summary of the information that
emerged from all these data, it is a data model. The question about the
daytime temperatures in this region of one day ago is answered with a
measurement outcome, certainly—but that is the graph in question, which is
a data model constructed from an analysis of the raw data.15
  :  167
That this is how the results of measurement, and the complexity of their
relation to theory, must be conceived was an early and continuing theme
in Patrick Suppes’ work:
exact analysis of the relation between empirical theories and relevant data calls
for a hierarchy of models of different logical type. Generally speaking, in pure
mathematics the comparison of models involves comparison of two models of
the same logical type, as in the assertion of representation theorems. A radically
different situation often obtains in the comparison of theory and experiment.
Theoretical notions are used in the theory which have no direct observable
analogue in the experimental data. (Suppes 1962: 253)
Our words have a sometimes disturbing plasticity. At one of the local

small stations a particular reading is made at a particular time: that is a
measurement. But the total process of all those single data collections plus
the statistical analysis plus the systematic ‘summarizing’ that finally yields
the graph—that also is a measurement. It is the latter that takes a more
scientifically significant form. What is important is that in both cases, the
primitive single data collection and the complex complete measurement
with a stable result, the outcome must be regarded this way: this is what
the object looks like in this measurement set-up. And in both cases, the object
measured is thereby located in a logical space, characteristically associated
with the type of operation involved.
Suppes continued with ‘‘In addition, it is common for models of a
theory to contain continuous functions or infinite sequences although the
confirming data are highly discrete and finitistic in character.’’ The graph or
other summary of the data found will be abstracted into a mathematically
idealized form before it reaches the theoretician’s desk. Let’s call that
processed artifact a surface model. Although usage does not restrict the term
‘‘data model’’ to summary representations of frequencies, let us take that as
our example. I distinguish data model and surface model in this way:
• the data model summarizes the relative frequencies found

• the surface model ‘smoothes’—in fact ‘idealizes’—this summary still
further so as to replace the relative frequency counts by measures with
a continuous range of values.
The term ‘‘data model’’ is often used in the more general sense that does not
distinguish summarized relative frequencies from probability measures, and
168  : , ,  
I will not insist everywhere pedantically on this distinction. Probabilities

relate to relative frequencies—a subject all by itself—but the frequencies
alone are what can be found in nature.16 Any procedure for replacing a
finite set of data by a graph furnishes an example, and such procedures are
common throughout the sciences, facilitated by easily available software.
But the replacement is a step in a long process from hands-on or mechanical
data gathering to the eventual mathematical structures that can confront
the proffered theoretical models.17
The complexity of the journey from raw data to graphical representation
is also well illustrated by Ronald Giere’s examination of astronomical
observation. The data gathered by optical, radio, and gamma ray detection
in today’s astronomical observatories are combined and transformed in a
process that yields visual representations color of comets, stars, nebulae,
and galaxies. Giere examines how astronomical data, gathered in the form
of black and white photos, are processed to yield images in color of a
nebula, through a process originally invented by James Clerk Maxwell.
Each individual photo captures just one aspect, but attention to the filtering
by which it is obtained allows it to contribute to the reproduction of color
in the images that are the final product. As Giere emphasizes,
the images presented . . . are conclusions. These images present a picture that is
continuous, or at least very fine-grained. The actual data cannot be that fine-
grained. The data are made up of individual events recorded in various detectors
at different times and processed by various physical and computational means. The
images are constructed using those data, but go beyond the data. (Giere 2006: 48)
Recall that just now we are looking at measurement ‘from above’, that is,
at this construction guided by accepted theory, rather than ‘from within’.
Surface models and their embeddings

A theory provides, in essence, a set of models. The ‘‘in essence’’ signals
much that must be delicately expanded and qualified; I will leave this
aside for the moment.18 These models—the theoretical models—are
provided in the first instance to fit observed and observable phenomena.
Since the description of these phenomena is in practice already by means of
models—the ‘data models’ or ‘surface models’, we can put the requirement
as follows: the data or surface models must ideally be isomorphically
embeddable in theoretical models.19
  :  169
Since the advent of quantum theory much thought has been given to the
form that any possible surface model must take. Consider an experimental
situation of a quite simple structure, involving several alternative measuring
arrangements, a classification of possible outcomes, and some probabilities
extrapolated from (imagined) observed frequencies. Then a surface model
can be thought of as specified by three factors:
(i) two sets of observable conditions:
(a) a set of realizable measurement choices—call it PRC, and
(b) a set of possible outcomes—call it PRS;
(ii) the surface state P; this is a function which assigns probabilities of
outcomes in PRS, conditional on measurements in PRC.
So P is defined on at least part of PRC × PRS and its values are real
numbers in the interval [0, 1]. This structure is subject to certain minimal
conditions which must guarantee that P is mathematically extendible to a
classical probability function.20 The numbers assigned by the surface state
we can call surface probabilities.
What about theoretical models? We already looked at this briefly above.
These need to be conceived without prejudice in favor of determinism or
causal modeling. The theoretical model could specify, in general,
a family M of observables (physical magnitudes) each with a range
of possible values;
a set S of states;
and a stochastic response function P s m for each m in M and s in S, which
is a probability measure on the range of m.
The number P s m is to be interpreted as the model’s specification of the
probability that a measurement of m will yield a value in E, if performed
when the state is s. From this we can at once see more or less what it shall
mean for such a theoretical model to fit a surface model. But not quite yet:
it only tells us the probabilities of surface phenomena, on the supposition
of a measurement and of a state. The latter is again something theoret-
ical, behind the phenomena. A stringent notion of ‘fitting’ could go as
follows:
A theoretical model MT fits an experimental model ME just in case
MT has some state s such that the function P s m contains the surface
170  : , ,  
state of ME, relative to the given identification of the measurement

setups as measurement of the physical magnitudes m.
The situation is thus, so to speak, represented ‘‘from below’’ and ‘‘from
above’’. In the surface model we have a representation that can be prepared
in the laboratory or observatory, without recourse to the theory and its
theoretical models.21 But the theoretician has a look at the same situation
‘from above’, by specifying how it can be represented by a theoretical
model. A way to evaluate what s/he provides along this line is presented
by the above notion of fit. Let us look at a concrete illustration.
The surface model of an EPR experiment

As example I want to take a much discussed and described experiment
that was inspired by the Einstein–Podolski–Rosen (‘‘EPR’’) paradox,
proposed by David Bohm, and eventually carried out by Alain Aspect.
There is no better or more generally accessible presentation of this sort
of experiment than the article David Mermin wrote for The Journal of
Philosophy (1981). This article makes very clear that there are two, quite
independent descriptions of the experiment, one ‘from above’ (in quantum
theory terms) and one ‘from below’ that can be understood without
knowledge of the theory.
As Mermin writes, it is not beyond present technology to mass produce
the experimental set-up in a form to be sold in local drugstores and
operated by anyone.22 There are three unconnected devices, a transmitter
and two detectors. The latter we may call L (for left) and R. Each has a
dial with (say) three settings or orientations. When the detectors are turned
on and the transmitter button is pushed, each detector flashes a light,
which is either red or green. For ease of presentation, let’s introduce some
notation:
L1: the proposition that L has been given the first setting.
Lx: the proposition that L has been given the xth setting: x = 1, 2, 3.
The experiments have each two distinct possible outcomes, the red and
green light flashes, which we may represent by the numbers zero and one:
L30: the proposition that L has the third setting and outcome zero.
Lx0: the proposition that L has the xth setting (x = 1, 2, 3) and outcome
zero.
  :  171
Lxa: the proposition that L has the xth setting and outcome a (a = 1, 0).
After a particular run on this apparatus, some of these propositions can

receive a score of T (true) or F (false). Propositions about settings other
than the actual one receive no score—they are irrelevant.
Suppose for example that for the first run, each apparatus was placed in
the first setting; L had outcome 1 and R had outcome 0. An experimental
report looks, in part, as follows:
Proposition Score
L1 T
L2 F
R1 T
L10 F
L20 No score
R10 T
L20 received no score. Since it is the proposition that L was placed in
setting 2 and had outcome 0, it could have been given F, simply on the basis
that L2 was false. But L20 is useless information for it records an outcome
for an arrangement that was not actualized, and so does not appear in the
experimental report.
This single report is in any case not likely to come to the theoretician’s
desk. What reaches him rather is like:
(S) With initial preparation X, the probability of outcome Lxa, given

setting Lx, equals p.
(L) For all initial preparations, the probability of (Lxa & Rxa), given
settings Lx and Rx, equals zero.
which is an extrapolation of a summary of many reports of the above sort,

and of the frequency counts made on their basis.
There is a gulf not only between the finite set of data points and
the extrapolation to relative frequencies, but also between the latter and
probabilities.
Accordingly there is a distinction between two intellectual artefacts
that lie between the collection of specific, concrete results on the one
hand and what the theorist takes as his or her input on the other. The
reported relative frequencies, in their own summarized form constitute
172  : , ,  
the data model obtained from repeated applications of a single measurement

procedure. What the theory confronts is abstracted from many data models.
The abstracting is an idealizing, an extrapolation to a form that could
not be reached in actual practice. So the reports (S) and (L) and their
cognates, describe an idealized but not yet theoretical structure—the surface
model.
Patrick Suppes pointed out that through construction of data models
the experimentalist is in general bringing the theoretician small relational
structures, constructed carefully from selected data. In the specific examples
that Suppes mentions the little structures are algebras; hence he calls them
empirical algebras.23 Literature on the foundations of quantum mechanics
typically points to such small structures that represent data, but they are
not always algebras. They are more generally partial algebras, or just
partially ordered sets (‘‘posets’’) with some relations and/or operations.
In the Appendix I will illustrate this, but the above already suffices to
show how we need to think of the deliverances of experimentation and
measurement as taking a much more general form than simple number
assignments.
The over-arching concept for measurement

The bounds we set to the form that surface models can have, how we
conceive that form, has changed historically. The bounds to the structure
a logical space, supplied by a theory to delineate its models, can have, that
has changed historically as well.
All the examples we have examined of measuring procedures are certainly
cases of grading, in a generalized sense: they serve to classify items as in
a certain respect greater, less, or equal. But as we also saw, this does not
establish that the scale must be the real number continuum, nor even that
the order is linear. The range may be an algebra, a lattice, or even more
rudimentary, a poset. Generalizing on such examples, we can still conclude
what I announced at the outset: By measuring we assign the item a location in
a logical space.
So locating something in a logical space is the over-arching concept
under which all actions of measurement can be arrayed. This is the only
  :  173
stable stopping point we have found in the successive generalization of the

notion of number-assigning.
Let’s emphasize again that we cannot equate the two, however. Not every
act even of locating something correctly in a logical space is the expression
of an actual measurement outcome.24 To say for example that our solar
system is a Newtonian mechanical ten-body system, that is to locate our
earth, the visible sun, and our sister planets in a logical space: the space of
Newtonian models, which is the logical space of Newtonian physics. The
bare assertion does not state a measurement result. But it is meaningful in
part because it is not unconnected with what measurement can show. Can
the assertion be taken as expressing a possible measurement outcome? In this
case, certainly. A very large family of measurements normally so-called can
be viewed as a ‘battery’, as a single complex measurement, and it can be
so viewed within a certain theoretical context. In this complex outcome,
our sun and planets ‘look like’ a Newtonian system. There are, however,
more abstruse assertions within the language of contemporary physics that
are much harder to see as thus connected with measurement, even sub
specie the pertinent theory. There is certainly a general concern in the
empirical sciences that the conditions of application of their concepts be
closely connected with experimentally realizable conditions, but there is
no rigid limit to how loose this ‘‘closely’’ can be.
Perspectival effects within logical space

While I do not want to stretch the term ‘‘perspective’’ very far beyond its
narrowly literal meaning, there are places in this discussion of measurement
where it seems apt enough. The aptness depends on context, and specifically
on what theories count as accepted in that context.
If a thermometer is used by one person to locate the air in his room on
the Fahrenheit scale, and by another to locate it in the space of possible
mean kinetic energies of its molecules, the two are locating the same thing,
by means of the same instrument, in two different logical spaces. That we
should not at once call a change of perspective, though it certainly marks a
change in ways of thinking about the air.
What if a theory then equates temperature and mean kinetic energy? In
that case we should say that relative to the theory it is appropriate to call this
a change of perspective. This qualification is important, though it may be
left tacit in a context where the theory has been entirely accepted.
174  : , ,  
Quite a different case is posed by two different measurement procedures

that are such as to assign always (error and vagueness aside) the same
locations in the same space. As an example consider several methods for
measuring velocity of a fluid, by pressure probe (use of pressure drop across
a nozzle—Bernoulli effect), hot wire anemometer (heat transfer from a
heated wire, due to convection heat transfer), and laser Doppler velocimetry
(use of light scattering from small seed particles, Doppler shift). These can all
be calibrated so as to give the same reading for the fluid velocity; there is no
difference even in scaling as there is with the Fahrenheit and Celsius scales.
It does not seem right to say here ‘‘there are differences in how things look
from here’’, but rather ‘‘things look the same from here as from there’’.
The moment we introduce this ‘‘from here and from there’’ distinction,
however, we also have reason to warrant the perspective appellation.
What if two persons use different but similar thermometers to locate the
same room’s air on the Fahrenheit scale and Celsius scale respectively? We
can’t say that they use different methods of measurement, nor that they are
measuring different parameters. Abstractly speaking we still have a choice:
should we speak of two logical spaces, or of different coordinate systems
in one space? But the latter choice is here so natural, that it almost seems
forced on us. It would certainly not be outré to speak of a difference in
perspective.
The two thermometers assign locations on two different scales, and these
can in principle be taken as two logical spaces. But there is a transformation
that systematically connects the readings on the one with the readings on
the other. Given knowledge of that transformation, we immediately have a
concept of a single logical space in which the two scales are, in effect, two
coordinate systems. The transformation is linear; we immediately conceive
of all the admissible temperature scales as connected to the Fahrenheit
scale by linear transformations. With this way of looking at the matter, we
are inclined to think of the use of Fahrenheit and Celsius thermometers
providing different perspectives on the same magnitude, whose logical
space is a more abstract structure.
We can see a similarity to the use of radar by two inertially moving
traffic policemen, in motion relative to each other, to determine the speed
of one and the same car. Their readings are different, for what they obtain
is the car’s speed relative to their own car—but whether they know it or
not, there is a transformation that translates their different readings into one
  :  175
another. The two policemen we say, surely, have before them the contents
of two perspectives within the same space—literally in this case.
At the same time we should recognize that relative to the pertinent theory
there is no real significance to a choice between a family of spaces related
by a group of transformations and a family of coordinatizations of a single
space. Little but bookkeeping ease is involved in the choice between
speaking of different spaces, transformable one into another, and different
coordinate systems imposed on a single space. Yet if we think in the former
terms it seems less apt to speak of differences of perspective. Perhaps the
resistance is the more reasonable, the more we recognize that there is high
theory in play; and the one important thing is to pay attention to what
theories are playing the background role.
Reconciling two views: perspective and invariance
For measurement the distinction is essential between the ‘giving’ of an object through
individual exhibition on the one side, in conceptual ways on the other. The latter is only
possible relative to objects that must be immediately exhibited. That is why a theory
of relativity is perforce always involved in measurement.25
This is Herman Weyl’s rather cryptic elaboration on how measurement

necessitates reference to something playing the role of a coordinate system
associated with the perspective embodied in the measurement set-up. But
there is a good deal of measurement in which this attributed relativity or
perspectivity does not readily spring to the eye. What are we to make of
measurements that yield outcomes ostensibly independent of differentiating
features of the measurement arrangement?
Imagine a container with a movable piston, a pressure gauge and
thermometer attached, and a scale on the outside that indicates the volume
for the various positions of the piston. Three measurements are being
carried out on this object, and together they locate its content in the three-
dimensional Pressure-Volume-Temperature manifold—itself represented
perhaps by the space of triples of non-negative real numbers. The outcome
registered at some particular time, locates the container contents in that
space.
This illustrates well the conclusion, that measurement is an action of
locating the measured object in some logical space. But what has happened
now to the much emphasized insight that a measurement outcome is after
176  : , ,  
all only a representation of the target, and in general does not show what
that is like but only what it ‘looks like’ in that measurement set-up?
Let us honor these two views of what measurement does with the names
Measuring is Locating and Measurement is Perspectival. Are they in tension
with each other at all? One small point may help: what is perspectival
is not the action of measuring but the contents of the measurement
outcome, and locating is an action, not a content. Action and outcome
are two different kinds of things. But this distinction does not go all
that far: the measurement outcome does after all represent the target as
located in a certain logical space. If we understand Measuring is Locating as
meaning just that, we are back with two takes on the same thing, on the
outcome.
If we are to call measurement perspectival, we need to qualify and
elaborate if we are to arrive at an accurately made point. Let’s begin with
a classic passage in which Poincaré insists on the outcome’s relativity to
set-up:
Lorentz could have accounted for the facts by supposing that the velocity of light is
greater in the direction of the earth’s motion than in the perpendicular direction.
He preferred to admit that the velocity is the same in the two directions, but that
bodies are smaller in the former than in the latter. If the surfaces of the waves of
light had undergone the same deformations as material bodies, we should never
have perceived the Lorentz–Fitzgerald deformation. In the one case as in the other,
there can be no question of absolute magnitude, but of the measurement of that
magnitude by means of some instrument. This instrument may be a yard-measure
or the path traversed by light. It is only the relation of the magnitude to the
instrument that we measure, and if this relation is altered, we have no means of
knowing whether it is the magnitude or the instrument that has changed.26
‘‘It is only the relation of the magnitude to the instrument that we

measure’’—that is either an overstatement or misleadingly put, though
it emphasizes something important. The outcome of the measurement
operation is a representation of the target, but it represents the object as it
appears in that measurement set-up. However, it does not follow that the
appearance there is different from the appearance of the same thing in other
such set-ups. For after all there are invariants too! In the case Poincaré is
discussing, the parameters measured are not invariant, they have different
values in differently moving inertial frames. But there are parameters, even
  :  177
ones definable from the results of local distance and time measurement
outcomes, that have the same value in different frames.27
Note well though that the invariance we are now discussing is not the
invariance that was cited as required for significance above. There we were
concerned with the transformations that connect all the members of an
admissible family of scales. Now we are discussing parameters for which the
value registered in a measurement outcome is the same under admissible
variations in the measurement set-up. As an example imagine again the
speed of a car measured by radar from a moving police car. This speed can
be registered on a scale of miles per hour or kilometers per minute, and so
forth. Leaving aside data about the police car’s own speed relative to the
road surface, what is registered is the speed of the target relative to the radar
source. If the policeman drives at a different speed himself, that relative
speed of the target will be different: it is not an invariant. If on the other
hand the radar is used to measure the target’s acceleration, the result will
be the same for any speed the police car may have. More precisely: where
Newtonian mechanics applies, the acceleration is the same in all inertial
frames, for all inertially moving measurement set-ups, while the speed is not.
There are therefore measurement outcomes that have no relativity left.
Generally, these are after instrumental outputs have been processed with
paper and pencil operations, with final outcome deduced relative to a
theory. This is an important point: a measurement and its outcome can
be complex, and include calculations and input from a model or theory.
Such a procedure still fits the general idea of an operation performed so
as to create a representation of the object; one that locates it in a certain
logical space, with a location that it does not have a priori. To see how the
activity signified by the slogan Measuring as Locating intertwines with the
fact that Measurement is Perspectival, let us take a look at some examples of
how ‘simple’ measurement outcomes are combined in such way to yield
an outcome of a thus created ‘complex’ measurement.
Think once more of celestial navigation in the days of sailing ships. Nav-
igating consists in locating oneself and guiding oneself from one location
to another. The first part is the measurement whose outcome will govern
that self-guidance.
What is one locating oneself in, in this case? In the grid, apparently
first proposed by Hipparcus: basically our system of latitude and longitude
which was also Ptolemy’s. In Ptolemy’s coordinate system as in ours, the
178  : , ,  
equator is at 0 and the North Pole at 90 degrees; similarly 360 degrees

of longitude cover the earth, though 0 was not at Greenwich. For the
ship’s longitude to be determined, it would be sufficient to know what
time it was at the same moment—e.g. local high noon—at 0 longitude,
since the Earth rotates 15 degrees per hour. Precise time reckoning was a
problem—let’s leave this aside here.
To determine the latitude by means of local measurements was pos-
sible, however. The astrolabe that I mentioned before was succeeded in
the eighteenth century by the sextant, to measure the elevation (angular
altitude above the horizon) of celestial bodies. The elevation of the sun at
noon, is a straightforward function of latitude. In the Northern hemisphere,
the elevation of the North Star above the horizon is equal to the latitude. So
imagine a sailing ship in the eighteenth century, with a sailor using a sextant
to measure the elevation of the North Star. This places the North Star in a
coordinatized Euclidean plane. Of course we do not read this as an ‘‘abso-
lute’’ feature, it is its location relative to the observation set-up. The number
found by the sailor is reported to the captain, who puts this to use to locate
the ship on the earth. That is, he uses the reported number to locate the ship
on the map, at a particular latitude—which together with his chronometer
reading and his calculations of longitude yield a single point on that map.
Note the difference: at this point what for the sailor was the operation of
finding the elevation there of the North Star has become for the captain a
position measurement performed on the boat. The outcome of the complex
measurement completed by the captain’s manipulation and combination of
data being the result of a deduction from that elevation value, relative to
premises that the captain already had at his disposal, including some that are
very theoretical. This measurement locates the boat in the geographic grid;
it can be recorded without adding a ‘‘from here’’. The longitude, latitude,
and time can now be entered in the log without any such ‘relative’ marker.
As a second example think of three cameras fixed on a grid, taking
photos of an ancient coin. In each photo the object has an elliptical shape.
In the simple set-up defined by one of the cameras, the shape measurement
outcome is ‘‘elliptical, with axes . . .’’. But in the three-camera set-up,
complemented with a little computer to process these three individual
outcomes, there is a big measurement going on, with an outcome such as
e.g. ‘‘circular’’. This too is a shape measurement. But here the outcome
seems not to be relative to the cameras.
  :  179
Yet this example brings out how context-dependent these judgments of

relative/non-relative are. We might have to see something relative where
we did not notice it before. In this example I probably tricked you into
imagining this as an inertial frame in which objects and cameras are at rest
with respect to each other. Shape is not a relativistic invariant, and if you
think about it again in the new context that I have now introduced into
our discourse, what was classified as not relative is reclassified as relative to
something after all.
But all the shape measurements located the object in the same logic-
al space of shapes. There is no contradiction when it is noted that the
location of the object in that logical space is relative to the measurement
set-up—and that adding in some other data and bits of theory will yield
more complex measurement procedures in which invariants are measured.
Finally, single simple measurements may be combined into complex ones
over time. Imagine a Stern–Gerlach apparatus with the usual two channels
and screens at their ends. When the first black spot appears, and it is on the
upper screen, we can say ‘‘looks like a particle in a definite spin-up state was
prepared’’. After, say, a hundred spots one might say ‘‘looks like this source
is preparing particles in a state in which 1/3 register as spin-up and 2/3 as
spin-down’’. There is no falsehood here, nor any contradiction, for the two
sorts of states mentioned can look the same in a single measurement. After
a long run, the ‘‘looks like’’ can disappear: the outcome is that the source
prepares particles in a superposition or mixture of spin-up and spin-down
eigenstates . . .’’ etc. This can be the output of a program on a computer
coupled to the apparatus; obviously certain assumptions or forms of infer-
ence are built into the algorithm that defines the program. The outcome in
this case locates the source output with respect to a certain Hilbert space.
What is a measurement outcome?

We cannot distinguish what measurement is by attending solely to the
physical correlate of measurement. The representational role is exhibited
only if we attend to how the measurement outcome is a representation of
what is measured.
A measurement is a physical interaction, set up by agents, in a way
that allows them to gather information.28 The outcome of a measurement
180  : , ,  
provides a representation of the entity (object, event, process) meas-

ured, selectively, by displaying values of some physical parameters that—
according to the theory governing this context—characterize that object.
It is a specific kind of representation, with some of its features inherited
from the nature of representation in general and some others peculiar
to it. The various strands in the way we think of measurement are to be
disentangled—though without implying for a moment that the distinctions
drawn are more than distinctions of reason, or that the different aspects we
can distinguish could exist all on their own.
To disentangle some of the strands in the way we think of measurement,
we can begin by emphasizing again the dual character, common to all that
properly belongs under the heading of technology: that we describe the
processes and objects involved both in physical and in functional terms.
What have we found so far? First of all there is the relationality in
the concept of measurement outcome. An outcome is an outcome of
something, of a process, which must be of a sort that satisfies various
stringent conditions. This is already correct on the purely physical side.
But however smoothly the story goes, of how certain implied correlations
qualify a physical process as being usable for gathering information, there
is still more to it. First of all, more is in play than a comparison of amounts
of information transmitted, and more than a correlation between physical
states of object and apparatus. The measurement was designed to answer
specific questions, and the information provided by the outcome is relevant
to their answer. Notions such as relevance and reference, as well as the
questions that define the context, are at best assumed as given when a
technical measure of the amount of information is applied.
The information borne by a measurement outcome refers to a specific
object or process (what was measured) and is selective with respect to
characteristics of that referent. Such very simple examples as, the inform-
ation provided by a reading of a thermometer (concerning the ambient
temperature) or a pressure gauge (concerning tire pressure), may suggest
that information transmission is something that can be explicated in purely
physical terms.29 But if a physical item is classified as the outcome of a
measurement (of a specific kind) then it is classified as a representation, and
therefore in effect classified under an intensional concept. The outcome
provides a representation of the measured item, but also represents it as
thus or so.
  :  181
We can see this as soon as the phrase ‘‘carries the information that . . .’’
is employed. To adapt an old example from Frege: to carry the information
that the Evening Star is within 15 degrees of the sun is not to carry
the information that the Morning Star is thus, although these are the
same object, and although therefore correlation with the position of the
Evening Star is automatically correlation with the position of the Morning
Star.
The intensionality of the concept of a measurement outcome consists in
the fact that it is something that has meaning. In reporting the outcome
one says, for example, that the pressure was [found to be] 17 psi; that
report is a sentence expressing a proposition. Even if, given the background
knowledge or opinion about the whole set-up, the pressure was necessarily
17 psi if and only if 17 = rT/V, the outcome was not that the value of rT/V
was 17—not even in the context defined by that background knowledge
or opinion.30
That the concept is intensional is not to be confused with its being
intentional. Literally, ‘‘intentional’’ refers to intention, but we take it
broadly to include purpose, goal, role, and function. To classify something
as a measurement outcome is to classify it as playing a certain role, namely
as the outcome of a process with a definite function. This is entirely in line
with the reflection that the activity of measurement belongs to technology,
and technological concepts have this dual character, of referring to physical
entities but partly (and essentially so) in terms of function.31
Fourthly, we must insist on the indexicality of the measurement outcome
content in general. That is easy enough to spot when the measurement
outcome is the indexical proposition that the iceberg is located 17 leagues to
the North-East. But it is an especially significant feature when the context
is more, rather than less, theory-laden.
Suppose for a moment that I take a pressure reading on the tire of my
car. The outcome can be reported simply and precisely as attributing a
feature to that tire. Where is the indexicality in that? But think of how
different a role this report plays from the assertion, written in the very same
words, in a historical account of what someone or other did somewhere.
For this outcome to be something useful for me, I must appreciate that
by means of this measurement operation I have also located my own
situation—which involves this tire—as having a place in the theoretical
region of pressure-graded objects.
182  : , ,  
Although this sort of representation is indexical in the same sense that

‘first person’ discourse is indexical (cf. Perry 1993), it is not ‘subjective’ in
contrast to ‘public’ or ‘inter-subjective’. Measurement outcomes are public
and they are intersubjectively accessible; this is crucial to the methodo-
logical requirement of reproducibility in scientific experimentation. The
measurement outcome content typically involves indexical reference to a
particular vantage point, but this vantage point is a publicly ascertainable
feature of the measurement set-up.
This fourth characteristic, the indexicality, is one not shared by rep-
resentation in general. It derives from the specific function which this
particular method of representation has.
There are further features that relate the measurement outcome to
specific kinds of representation, notably to imaging and picturing. Recall
that by imaging I mean representation that trades on selective resemblances,
and that the special case of picturing was introduced above as follows:
Overlapping these categories of representation that trade on selective resemblances

lies still a further salient case, which shares some crucial features found in visual
perspective, a development which in art we associate specifically with Renaissance
painting. Perspective involves . . . such features as occlusion, marginal distortion,
texture-fading. For cases of imagery in which such features of perspective are present
I’ll use the terms picture and picturing —these can include cases of kinematic and
visual but perhaps also still further forms of imagery.
The most easily recognized cases of picturing are ones in which the resemb-
lance is not at a very high or abstract level: it is just a sharing of selected
properties. The subway map shares the topological structure of the subway
system, say; it is a picture of that system. But resemblance, we recall, can
be higher order: the spatial structure of a set of letters on a page may be
the same as the temporal structure of a set of events named by those letters.
The use of visual or kinematic imagery to depict things that are not visual
or kinematic is rife, and not excluded by our notion of imagery. So a meas-
urement outcome may well purport to give us information by means of a
selective resemblance to what is measured, although the pointer-observable
may be of a very different character than the measured observable.
Indeed, the Criterion for the Physical Correlate of Measurement entails
that resemblance at such a structural level is required to be implied by the
  :  183
theoretical classification of something as a measurement. So this, the

trading on selective resemblance at some level, is our fifth characteristic.
And finally, there is the sense in which measurement outcomes are
perspectival. This has been explored extensively by Ronald Giere (2004;
2006), whose work has placed this topic at center stage.
There is historic precedent. Poincaré (1897: ch. 1) emphasized strongly
the ‘perspectival’ (‘relational’ in the sense of relative rather than derived)
character of the outcome. Poincaré overstated it perhaps (‘‘It is only the
relation of the magnitude to the instrument that we measure’’), but the
point is clear. What the outcome reveals is not directly what the measured
object is like, but what it ‘looks like’ in that measurement set-up. The latter
point was exploited saliently by Einstein in his 1905 STR paper, where
measurements of time and of length are taken to yield only data explicitly
relative to the state of motion of the measurement set-up.32
To assess this question of perspectivity properly we must take stock of its
hallmarks, such as occlusion, marginal distortion, and texture-fading. Literally,
marginal distortion is distortion in proportions toward the outer edges
of a perspectival drawing or painting. In a more general sense, marginal
distortion is distortion that is the result of the limits inherent in a given
mode of representation. Such would seem to be a feature of almost any
sort of measurement at all, given that instruments have a limited range,
and become inevitably less reliable and less exact near the limits of that
range. A liquid thermometer is not to be used near the freezing and
boiling points of that liquid, for example. Perhaps a simple detection of
presence or absence, with a yes–no answer, would be a counterexample,
but then only if the means of detection could not possibly give uncertain
results—an unlikely case. In the context of quantum mechanics we have a
more extreme sort of example. That an object’s location may, as far as its
state is concerned, be probabilistically smeared out over the whole of space,
while we can only get our measuring apparatus to ask whether or not it
is in this room (or similarly small finite region), means that information
gatherable beyond a narrow range is maximally indefinite—that is marginal
distortion.
Occlusion and texture-fading are easy to spot in the sort of measurement
that is paradigmatic for our concept, namely the sort of spatial measurement
I discussed with reference to Hugh of St. Victor, Alberti, Brunelleschi, and
Dürer.
184  : , ,  
It is tempting to see these same concepts as applying in quantum

mechanics due to the impossibility of simultaneous measurement of non-
commuting observables. If one observable is measured, the extent of
information that can be gathered with respect to incompatible observables
is drastically reined in: that remind us of occlusion. That the observable
really measured, when a measurement is officially designed to measure a
sharp observable, will necessarily be a fuzzy version of that observable—that
sounds like texture-fading. But to so use the terms is at best an analogical
extension of their literal meaning. Occlusion means hiding —changing
something so that it can’t be measured is therefore not occlusion.33 In the
case of quantum mechanics it is certainly not the case, for example, that the
measured value of momentum implies a real but hidden value of position.
On the other hand, if a system is in an eigenstate of an observable, and
an incompatible observable is measured, then that state has changed. So
we have two reasons not to call this occlusion. As to texture-fading: the
fuzzying discussed for quantum measurement does not have anything to
do with distance, which was what concerns texture-fading in pictures. But
though the analogy is far from complete, it is not far-fetched to analogically
extend the notions of occlusion and fading in this way.
Relating the views ‘from above’ and ‘from within’

As long as we consider what happens in measurement purely from the
theoretical point of view, the only criteria are theory-internal. The criteria
of adequacy cannot go beyond coherence. That is why the theoretical point
of view remains empty unless the problem of coordination is also faced and
taken into account. ‘‘Concepts without percepts are empty’’—this Kantian
motto transposes to the empiricist point that theory without coordination is
empty. To understand how theoretical parameters become coordinated, we
have to look into the historical process in which measurement procedures
and theory evolve together, in a thoroughly entangled way. Nor can this
evolution start in anything but prior meaningful discourse relating to what
eventually will be delineated as the theory’s domain.
The scientific realist will interpret this process, in which practice and
theory jointly stabilize, as establishing that the theory latches on to the
blueprint for the universe.34 But the one empirical fact is precisely this:
  :  185
that the practice, both experimental and theoretical, stabilizes, and that
‘nature cooperates’ to the extent that, perhaps temporarily, no or less
resistance is experienced in this practice.35 For the time being, at least,
the expectations engendered by empirical predictions are satisfied, the
retrospective evaluation says that thus guided empirical judgment has been
well calibrated.
PA RT I I I
Structure and Perspective
In the preceding parts we have uncovered some conditions for the possibility
of scientific representation. Viewed in one particular way, all of these
conditions can be brought under one heading: the crucial role of use
and practice. Although description in language is at best one mode of
representation, this crucial link to use and praxis points us toward the study
of pragmatics: the study of language in which word-thing relations are seen
as abstractions from word-thing-user relationships.1
The asymmetry of representation and the possibility of misrepresentation,
for example, we saw to derive from use rather than from independently
specifiable relations between representor and represented. Nothing is a
representation unless it has a certain kind of role in use and practice. In
addition, besides their status as representation deriving from use, some
representors have a use, which they can have only in a context in which
indexicals and self-reference are available. While I gave maps as paradigmatic
example, this use is central to all the practical sciences, where scientific
representations are drawn on so as to apply scientific knowledge in practice.
This is well illustrated also by the problem of coordination, which was
seen to be unsolvable except in a context where some coordinations are
already achieved and present. Coordination, which assimilates theoretical
terms to the language in use, is not to be understood as a completely
explicit or conscious historical process. We cannot think of theoretical or
other newly introduced terms as made subject to principles of coordination,
except in a context where it is already possible to rely on other terms,
‘old’ terms, as ‘already coordinated’, as meaningful.2 Meaning and use must
indeed be bestowed on newly introduced terms, but this makes sense only
if we think of them as introduced into an already extant language, into our
own language in use.3
The distinction between what is newly introduced theory, and what is
language in which the instruments and measurements are already described,
is historically conditioned. The phrase ‘‘we already knew how to describe’’
signals reliance on our own language at that historical moment. How could
it be otherwise? There is no moment outside history, and at each moment
190   :   
in history we not only can but must rely on the language in which we
conduct our business, the language we live in. This is Neurath’s insight
about mariners at sea, but extended to the conditions of possibility of
scientific representation in every respect.
The epistemic situation as here described has seen responses not just in
philosophy of science but in metaphysics, and not just in past centuries
but in our own time. We will follow the resulting problématique as it
developed through the twentieth century and into our own, starting with
the Bildtheorie of Hertz and Boltzmann as precursor to structuralism in the
philosophy of science. Structuralism about science is, roughly speaking, the
contention that scientific representation is of structure only. The obvious
question, what is structure and what is not?, is the first that any advocate of
structuralism must answer. The answers have tended to dissolve into vacuity
or inconsistency when pressed to precision—as we shall see, going through
Russell’s, Carnap’s, and Putnam’s arguments as structuralist conceptions
emerged, faltered, and took hold.
Structuralism unfortunately involved, during most of its problem-beset
history, another attempt to achieve the simultaneous interpretation of all
language and theory without relying on our prior language-in-use. Seen
in this way, reminiscent of Reichenbach’s attempt to conceive of and
solve the problem of coordination in an extreme form, structuralism too
pursued an impossible ideal.4 But just as we can see a real and viable process
of coordination behind Reichenbach’s reach for the ‘unconditioned’, so it
seems to me that we can see a genuine and viable sense in which structuralist
views of science are right, at heart.5
I shall advocate a version, an empiricist version, of structuralism. Once
again, the redeeming clues are to be found in pragmatics. The empiricist
view I propose will, I hope, do justice to the strong structuralist trend
found in philosophy of science without subordinating it to any form of
metaphysical realism, and without giving in to the attendant illusions of
reason.
8
From the Bildtheorie of Science
to Paradox
The arguments by scientists and philosophers in the decades just before

and after 1900 concerned not only the directions that science should take,
but also the very idea of what science was or was to be. As I shall present
it, the advocacy of the Bildtheorie, the view that what science gives us is
representations, important as it is in its own right, is also integral to the origins
of what is now known as structuralism in the philosophy of science.
Structuralism about science—the thesis that a science represents only
structure in its domain—became increasingly and recurrently a salient
theme in twentieth-century philosophy of science. This development
had two motivating philosophical controversies in its past. One was in
the philosophy of mathematics, and specifically of geometry, where the
conception of a theory of space was gradually replaced by the conception
of geometry as a branch of abstract mathematics. The other, which we will
consider first, occurred in the debates among physicists about the status and
use of models, where we see a clear foreshadowing of later debates over
scientific and structural realism.
The Bildtheorie controversy

One word of caution, however. At times the debates about how to conceive
of science, the scientific practice, its products, and its criteria of adequacy,
became entangled with those about the scientific acceptability of atomic
theory. It is important not to confuse or conflate the two. The protagonists
of the Bildtheorie1 in the time of structuralism’s infancy can be found on
both sides of the atoms debate. Mach, antagonist to atomic theory, plays
192  :   
an important role. So did some of Hertz’s misgivings about atomistic

representation of the phenomena, but Hertz’s work was instrumental to
the development of atomic theory. Boltzmann, a major protagonist of the
Bildtheorie, was an enthusiastic advocate and developer of the kinetic theory
and statistical mechanics. There is undeniably a distinctive anti-scientific-
realist tenor to the Bildtheorie conception. But the anti-realism takes a
sophisticated form, not to be confused with philosophical prejudice against
theories that postulate unobservables.2
Planck against the heretics

On December 9, 1908 Max Planck addressed the Student Corps of the
Faculty of Natural Sciences of the University of Leiden. His announced
topic was The Unity of the Physical World-Picture, but his intent included
a polemic against those scientists who had turned against realism in the
past fifty years. His target was mainly Ernst Mach, but he says that this
heresy ‘‘enjoys great popularity, particularly in circles of natural scientists’’
(ibid.: 129). In Planck’s eyes those so misguided had forsaken the faith of
their fathers. He speaks against them with passion:
When the great masters of exact research contributed their ideas to science: when
Nicolaus Copernicus tore the earth from the center of the universe, when Johannes
Kepler formulated the laws named after him, when Isaac Newton discovered general
gravitation, when your great countryman Christian Huygens put forward the wave
theory of light, and when Michael Faraday created the foundations of electro-
dynamics . . . [Mach’s] economical point of view was surely the very last thing
which steeled the resolve of these men in their battle against traditional views and
towering authorities. Nein! . . . it was their rock-solid belief in the reality of their
world picture. (Planck 1992: 131)3
Planck considered Mach’s heresy to be a mistaken if understandable
‘‘philosophical manifestation of unavoidable disenchantment’’ when the
mechanical world view began to disintegrate.
But there is some irony in this episode, as we shall see. Perhaps
unwittingly, very likely without any conscious polemical purpose, Planck
casts his opposition to Mach’s view of science in a way that leaves the
issue entirely entangled with the issue of the reality of atoms. His more
general discussion of the aim and structure of physical theory is in effect
interrupted by the long diatribe, in the last section, against Mach’s views
on atomism.4
  BILDTHEORIE     193
What complicates the disentanglement further is that Planck himself uses

the terms of the Bildtheorie of science, which are more adapted to less realist
views than his own. Not just in the title of his lecture but throughout,
Planck speaks of physical theories as pictures and of the product of science as
a whole as a world-picture. This is the language of the Bildtheorie of science,
it honors the view of science as representation that was also the common
coin of his opponents. While the term Weltbild was already used in much
philosophical literature, the introduction of this term into a view of how
we are to understand the scientific enterprise is generally attributed to
Hertz.5
But it is possible to see the real opposition between Planck’s and Mach’s
views on science, as well as, and properly distinguished from, the opposition
between their views on the reality of atoms—by carefully selecting first of
all the question that Planck himself calls the most fundamental:
What do we really mean when we speak of a physical world-picture? Is it merely

a convenient but basically arbitrary intellectual concept, or should we take the
opposite view, that it reflects actual natural processes quite independent of us?
(Planck 1970: 4)
The phrase ‘‘basically arbitrary’’ is certainly not innocent of polemics; that

is certainly not how his opponents would speak. We find Planck’s view of
science, its aim and structure, spelled out more fully in the last section of
the lecture. He begins by spelling out how he sees the history of modern
physics as a story of progress:
the old system of physics [sometime before 1900] was not like a single picture, but
more like a whole picture gallery; since every class of natural phenomena had its
own picture. And these different pictures did not all hang together; one could take
any one of them away, without affecting the others. In the future world-picture,
this will not be possible. Each one is an indispensable component of the whole
and, as such, has a specific meaning for observed nature; while, conversely, every
observable physical phenomenon must find its precisely appropriate place in the
picture. (Ibid.: 21–2)
The crucial point follows this immediately: ‘‘In this respect, it differs
essentially from ordinary pictures, which certainly need to correspond to
the original in some particulars, but not in all—a distinction to which, in
my opinion, physicists have not hitherto paid enough attention’’ (ibid.: 22).
194  :   
On the contrary, that is precisely what the rivals in this view of science
explicitly contradict, when they speak in terms of ‘‘pictures’’ along the
lines made prevalent by e.g. Hertz and Boltzmann. Indeed, Planck is here
rejecting, in effect, the very core of the Bildtheorie while keeping its picture
terminology. For there is no point in emphasizing that science presents us
with representations of natural phenomena, if not to convey that success
in science will consist in constructing an image of nature that is adequate
in certain respects and trades on resemblance at best in part, as opposed to
constructing a true and accurate copy.
That this passage is not meant simply as a bit of futurology, a vision of
the best conceivable future, but an expression of what Planck takes to be
the defining aim of science, is then made clear:
A constant, unified world-picture is, as I have tried to show, the fixed goal which
true natural science, in all its forms, is perpetually approaching; and in physics we
may justly claim that our present world-picture, although it shimmers with the most
varied colors imparted by the individuality of the researcher, nevertheless contains
certain features which can never be effaced by any revolution, either in nature
or in the human mind. This constant element, independent of every human (and
indeed of every intellectual) individuality, is what we call ‘‘the Real’’. (Ibid.: 25)
The passage I quoted at the outset follows now, and just after that there
comes, at least to our eyes, a curious ending to his polemics. He wishes
simultaneously to withdraw from the ‘representation’ or ‘picture’ view and
to embrace it on a higher (deeper?) level:
Those great men did not speak about their ‘‘world-picture’’; they spoke about
‘‘the world’’ or about ‘‘Nature’’ itself. Now, is there any recognizable difference
between their ‘‘world’’ and our ‘‘world-picture of the future’’? Surely not! For the
fact that no method exists for proving such a difference was made the common
property of all thinkers by Immanuel Kant. (Ibid.)
With this uncertain—faltering?—clarion call on behalf of scientific realism

still in our ears let us proceed to the more thorough-going views of science
as representation.
Maxwell, Hertz, Boltzmann, Mach

Planck directs himself primarily against Mach when he insists that the
scientific world-picture must ‘‘[differ] essentially from ordinary pictures,
which certainly need to correspond to the original in some particulars, but
  BILDTHEORIE     195
not in all’’. But it is in Boltzmann’s writings that we see the contrary view
most clearly.6
Boltzmann presents his own point of view as deriving mainly from
Maxwell and Hertz, two of the heroes of the then recent achievements
in electromagnetism.7 Maxwell’s writings are not exactly unambiguous. In
fact he is often taken as postulating the reality of the ether and of the elec-
tromagnetic waves in the ether, while sometimes despairing of any purely
mechanical theory of their character. However, as Boltzmann emphasizes,
Maxwell speaks of the envisaged mechanisms as merely analogies, partial
analogies, that allow us to get an imaginative grasp on the equations. The
equations must on the one hand fit the observed magnetic, electrical, and
optical phenomena, and on the other hand allow of some understanding
of the theory as a description of a physical process. But as far as description
goes, we receive mainly analogies with other forms of material propagation,
diffusion, and interaction—with gases, fluids, and heat. Maxwell himself
cautions us against thinking of this as a true description of reality behind
the phenomena:
By a judicious use of this analogy [between Fourier’s equations of heat conduction
and the equations of the electrostatic field] . . . the progress of physics has been
greatly assisted. In order to avoid the dangers of crude hypotheses we must study
the true nature of analogies of this kind. We must not conclude from the partial
similarity of some of the relations of the phenomena of heat and electricity that
there is any real similarity between the causes of these phenomena. The similarity
is a similarity between relations, not a similarity between things related. (Maxwell
1881: 51–2)
We have noted before that imagery may trade on ‘higher-order’ resemb-

lance: not a sharing of properties, but of relational structure. That the
models provided by a science may have that sort of less direct relationship
to the phenomena becomes a guiding theme for structuralisms in the next
century.8
Then, as Boltzmann sees it, Hertz makes a virtue of necessity and asserts
this as a way to understand the scientific enterprise as a whole. Thus Hertz
writes (and Boltzmann cites this passage):
If we wish to lend more color to the theory, there is nothing to prevent us
from supplementing all this and aiding our powers of imagination by concrete
representations of the various conceptions as to the nature of electric polarization,
196  :   
the electric current, etc. But scientific accuracy requires of us that we should in no
wise confuse the simple and homely figure, as it is presented to us by nature, with
the gay garment which we use to clothe it. Of our own free will we can make no
change whatever in the form of the one, but the cut and color of the other we can
choose as we please. (Hertz 1962: 28)
Indeed, with Hertz we begin to have such an emphasis on the representa-

tions and their adequacy to the experimental facts as sole anchor, that we
can quite understand Planck’s sense that the represented world is mostly
counted as well-lost for love of theory:
We form for ourselves inner pictures or symbols of external objects; and the form
which we give them is such that the necessary consequences of the pictures in
thought are always the pictures of the necessary consequences in nature of the
things pictured . . . .
The pictures which we here speak of are our conceptions of things. With
the things themselves they are in conformity in one important respect, namely in
satisfying the above requirement. For our purpose it is not necessary that they
should be in conformity with the things in any other respect whatever.9
In the first part of this passage we see the relationship pattern characteristic
of how a symmetry requirement is to be satisfied:10
picture 1 picture 2
event 1 event 2
Figure 8.1. Adequacy as Symmetry
The evolution at the top is a logical deduction relative to assumed

conditions; the evolution at the bottom is the regularity in nature in those
conditions, and the downward arrows stand for the pertinent ‘picturing’
relation.
We can illustrate the idea with our earlier example of the Aviation
Model (AVN) for numerical weather prediction. The data are entered
concerning current and past conditions, and the program generates a model
of the current meteorological state of the region, and calculates its forecast
  BILDTHEORIE     197
evolution. This calculation provides models of the region for the next five
or so days, and Hertz’s constraint requires (naturally!) that those accurately
represent the conditions on those days—with the models’ success gauged
by some measure of accuracy. In this sort of example, there are few if
any hidden parameters characterizing unobservable entities, but the pattern
Hertz displays is general. It can be cited equally well as a constraint on
hortatory astrology (in which natal charts are progressed to forecast the
native’s life history) as on quantum electrodynamics, both of which are
replete with parameters characterizing unseen influences. Not to say, of
course, that this properly imposed constraint is satisfied in all cases.
Hertz’s constraint is a crucial condition for the objectivity of scientific
representation. Far from mere exercises of the imagination, merely adding
levels of fantasy to the known empirical realities, scientific representations
must allow us to go reliably from what we know to what we will or
can encounter further on. This is the empirical constraint on scientific
theorizing, here phrased in a form easily seen to be appropriate to a
structuralist (as opposed to a naïve realist) conception of ‘picturing’ by
means of models.
Alongside of this positive contribution to an understanding of how an
abstract theoretical science can provide representations of the empirical
phenomena, there was a good deal of polemics in the air. Boltzmann,
though always on the side of those advocating the kinetic and atomic
theories nevertheless, lecturing in 1899, expressed the heretics’ philosophical
point of view most trenchantly:
We know how . . . to obtain a useful picture of the world of appearance. What
the real cause for the fact that the world of appearance runs its course in just this
way may be; what may be hidden behind the world of appearance, propelling it,
as it were—such investigations we do not consider to be of the task of natural
science. (Boltzmann 1905a: 252)
Finally, we may note Mach’s reaction to Planck’s criticisms of this train

of thought. Just as Boltzmann does in this last passage, so Mach attributes
those realist misgivings to metaphysical dreams by which philosophers have
infected physicists from time to time:
In any case, physicists have nothing to seek ‘beyond the appearances’. Whether
philosophers will always find it necessary to affirm something real . . . whose relations
may only be recognized in the wholly abstract form of equations, may be left
198  :   
entirely for the philosophers to decide. [. . .] Hopefully, physicists of the 20th

century will not let their investigations be disturbed by such meddling! (Mach
[1910/1992], 124–5)
So each side depicts the other as having strayed from the true concerns of
natural science into mistaken philosophical conceptions of their common
enterprise.
The ‘‘picture’’ and ‘‘image’’ imagery became pervasive in philosophically
reflective writing on physics by physicists, increasingly so during the
controversies over the interpretation of quantum mechanics in the late
1920s and 1930s. Bohr’s insistence on the use of the complementary
wave and particle pictures—neither of which can be regarded as faithfully
mimetic representations, precisely because they are mutually exclusive—is
too well known to bear repetition. But Erwin Schrödinger, who rejected
wave-particle duality, wrote in the same terms. Heisenberg’s uncertainty
relations, he wrote, ‘‘changed our conception [of . . . and] even what is
to be understood by a physical world image.’’11 This usage continues
throughout his writings:
we do give a complete description, continuous in space and time without leaving
any gaps, conforming to the classical ideal—a description of something. But we
do not claim that this ‘something’ is the observed or observable facts; and still less
do we claim that we thus describe what nature (matter, radiation, etc.) really is.
In fact we use this picture (the so-called wave picture) in full knowledge that it is
neither.’’12
We can certainly see major strands of anti-realism in all these writings, but
notice that it is not simplistic anti-realism—in fact, most of these writers
were actively involved in developing the new physics, including atomic
theory and quantum theory.
The Bildtheorie view of science takes a general form that is compatible, for
example, with what we know later under the name of structural realism. If
what science gives us by way of theories and models is to be conceived of as
pictures, as representations, then the question is opened as to just what the
relevant and appropriate criteria of adequacy are for them. An extreme view
would be that a good representation of that sort is one that corresponds
in every respect to what it is representing. That is not what even a quite
extreme scientific realist would say, since there is always a good deal of
mathematical artifice present. But it is approached, later in the twentieth
  BILDTHEORIE     199
century by various forms of scientific realism (Salmon, Boyd, or Leplin

for example, or to a lesser extent ‘‘entity realism’’), and by the criterion
for corresponding ‘‘elements of reality’’ in the paper on the completeness
of quantum mechanics by Einstein, Podolsky, and Rosen. But a different
criterion is found in the varieties of structuralism in the philosophy of
science that we shall go on to examine—criteria to the effect that scientific
models trade for their success on resemblance with respect to structure
alone. And there is also of course a still more liberal empiricist conception
according to which the base line criterion is just empirical adequacy. Each
of these is an answer to the sort of question which, in our initial discussion,
we saw proper to the identification of adequate representation, which needs
to take into account what is the purpose or aim and specifically what is at
stake in the representational practice.
Representational Options and Realism: Descartes, Hertz, Poincaré, Duhem

What options were coming into play at this point? The equations come
with a narrative about unobservable entities, but this narrative is classified
as providing us with a useful representation, that only needs to fit the
regularities in what is observed. This view, though newly salient at the time,
is not without historical precedent; witness Descartes’s stand at the end of
the Fourth Book of his Principles of Philosophy:
Prop. 204. That, touching the things which our senses do not perceive, it is
sufficient to explain how they can be . . . ( Descartes 1959: 210)
This is his reply to the objection that his theories may fit the phenomena
without being true, something he concedes quite readily:
. . . just as the same artisan can make two clocks, which, though they both
equally well indicate the time, and are not different in outward appearance, have
nevertheless nothing resembling in the composition of their wheels; so doubtless
the Supreme Maker of things has an infinity of diverse means at his disposal, by
each of which he could have made all the things of this world to appear as we see
them, without it being possible for the human mind to know which of all these
means he chose to employ. (Ibid.)
But the aim of science is misunderstood if this is taken as an objection:

I believe that I have done all that was required, if the causes I have assigned
are such that their effects accurately correspond to all the phenomena of nature,
200  :   
without determining whether it is by these or by others that they are actually

produced. And it will be sufficient for the use of life to know the causes thus
imagined, for medicine, mechanics, and in general all the arts to which the
knowledge of physics is of service, have for their end only those effects that are
sensible, and that are accordingly to be reckoned among the phenomena of nature.
(Ibid.)
In support, Descartes cites Aristotle’s On Meteors: ‘‘We consider a satisfactory

explanation of phenomena inaccessible to observation to have been given
when our account of them is free from impossibilities.’’13
How should we think of scientific representation if conceived of in this
manner? The guiding idea is something like a painting of real terrestrial
events surrounded by indications to show not just what is happening
but what is really going on. Demons and angels perhaps, or imagined
natural but invisible mechanisms made visible by the artist’s prerogat-
ive. The corresponding view of its aim or criterion of success would
then be that the painting is ‘‘right’’ if it is historically accurate and fits
the terrestrial events into a possible intelligible scenario that accounts for
them. The criteria of intelligibility are our own—requiring some sort
of causal pattern perhaps, satisfying e.g. mechanistic views of interaction.
Since intelligibility does not imply knowledge of the truth, those criter-
ia can be satisfied without demanding that the scenario is actually true
overall.14
The implied agnosticism, or even skepticism, is mostly downplayed in
Boltzmann’s writings. While seemingly insisting that not so much is needed
for science, his texts leave it open that one could also believe the entire
‘picture’ to be accurate in all respects.15 The skepticism is clearer in Hertz’s
verdict, later repeated by Poincaré, on classical electromagnetism. Hertz’s
early book Die Constitution der Materie already introduced his emphasis on
representation as the task and method of the exact sciences.16 How can
there be an atomic theory, to cover the macroscopic phenomena of e.g.
heat or elasticity? Any properties we can ascribe to the atoms will have to be
among the only ones we know, hence properties characterizing observable
macroscopic objects. But no conception of atoms as just small versions of
familiar kinds will be adequate for atomic theory; in fact, he calls any such
conception a logical error. Thus, at least, Hertz sees the issue—expressing
a concern which may not have seemed pressing to his contemporaries still
in the thrall of mechanical explanation.
  BILDTHEORIE     201
We can at the same time see that the entanglement with the question of
the scientific acceptability of atomic theory is only contingently involved
with this, as a currently salient example of scientific representation. Not
only was Hertz’s work crucial to the establishment of that theory, but his
sentiment concerning the theory’s representations is echoed almost literally
half a century or so later by Heisenberg:
The atom of modern physics can be symbolized only through a partial differential
equation in an abstract space of many dimensions . . . . All its qualities are inferen-
tial; no material properties can be directly attributed to it. That is to say, any picture
of the atom that our imagination is able to invent is for that reason defective.
An understanding of the atomic world in that primary sensuous fashion . . . is
impossible.17
Symbolized, hence represented, yes; but not mimetically represented, not

picturable in the familiar parameters of macroscopic observation. From
Hertz to Heisenberg, the complexities of modeling the unobservable, once
even Newton’s ‘‘instant distant correlation’’ version of mechanics gives
out, drives the liberalization of the sense of representation in which science
is taken to represent the investigated phenomena.
Whether or not Hertz’s argument here was telling, the issue led him
to the view that all we can have for atomic physics is a theory in which
the magnitudes (theoretical parameters) are connected to each other and
to macroscopic phenomena by mathematical equations. It does not have
seemed necessary to him, as far as success in science is concerned, that
some further or deeper meaning needs to be ascribed to those parameters.
But we must not understand this to mean that the theoretical parameters
are allowed to be empirically ungrounded! They need to be connected to
measurement, relative to the theory, in a way that allows for determination
of their values in principle. This remained a problem for another few
decades, but its appreciation goes hand in hand with Hertz’s conviction
that what is essential in the mathematical equations in which the theory
is formulated is that they provide a representation of relations between
macroscopically observable magnitudes.
In the Introduction to his book Electric Waves, Hertz described his own
presentation as but one possible representation of Maxwell’s electromag-
netic theory, simplifying it ‘‘as far as possible by eliminating . . . those
portions which could be dispensed with, inasmuch as they could not
202  :   
affect any possible phenomena’’. This presentation is to be contrasted with

Maxwell’s presentation of the theory and its representation as a limiting
case in Helmholtz’s theory, but Hertz insists that they all have ‘‘substantially
the same inner significance’’. He continues, in a familiar passage:
This common significance of the different modes of representation (and others
can certainly be found) appears to me to be the undying part of Maxwell’s work.
This, and not Maxwell’s peculiar conceptions or methods, would I designate as
‘‘Maxwell’s theory.’’ To the question, ‘‘What is Maxwell’s theory?’’ I know of
no shorter or more definite answer than the following:—Maxwell’s theory is
Maxwell’s system of equations. Every theory which leads to the same system of
equations, and therefore comprises the same possible phenomena, I would consider
as being a form or special case of Maxwell’s theory . . . .18
and this affects how he envisages scientific methodology:

It is true we cannot a priori demand from nature simplicity, nor can we judge what
in her opinion is simple. But with regard to images of our own creation we can
lay down requirements. We are justified in deciding that if our images are well
adapted to the things, the actual relations of the things must be represented by
simple relations between the images. (Ibid.: 23)
Maxwell himself had attempted to prove the existence of models of his

equations in mechanics. The theory of the ether was a sustained attempt
to provide them with a concrete mechanical underpinning (cf. Stein 1989:
61–2). When Maxwell had his theory fully worked out, he discarded the
earlier rather primitive ether models and tried to subsume his theory under
the generalized dynamics of Lagrange, which deals with mechanical systems
whose internal constitution is not fully specified.19 In Hertz’s, and later
Poincaré’s, verdict we recognize a definitive goodbye to the interrelation of
matter and ether as a live topic in physics.20
Hertz’s last book, The Principles of Mechanics Presented in a New Form
represents physical processes by embedding them in a larger structure
consisting of ‘hidden masses’, but no forces. When the kinetic energy of a
macroscopic system apparently changes, in a way normally accounted for
by postulating forces, the effect is instead accounted for within mechanics
by reference to the motions of the hidden masses. But these hidden masses
consist of mass-points, entities with zero extension, and so the theory does
not appear to afford a literal realistic construal. The Principles of Mechanics
is divided into two books, and the basic theory developed in the first is
  BILDTHEORIE     203
applied to what could be observable processes through something like what

Reichenbach would later have called coordination. Is there anything, we
might ask, that goes beyond coordination? This is precisely where Hertz
writes what I quoted above: ‘‘We form for ourselves inner pictures or
symbols of external objects; and the form which we give them is such that
the necessary consequences of the pictures in thought are always the pictures
of the necessary consequences in nature of the things pictured . . . . The
pictures which we here speak of are our conceptions of things. With the
things themselves they are in conformity in one important respect, namely
in satisfying the above requirement. For our purpose it is not necessary
that they should be in conformity with the things in any other respect
whatever.’’21
A decade or so later, close to the date of Planck’s scorching attack on
the heretics of his generation, Poincaré and Duhem were publishing views
in harmony with Hertz’s and Boltzmann’s Bildtheorie of science. Duhem
made the concept of representation the cornerstone of his view of science.
After a thorough critique of more metaphysical and realist conceptions in
the first chapter of his The Aim and Structure of Physical Theory, he presents
his own:
A physical theory is not an explanation. It is a system of mathematical propositions,
deduced from a small number of principles, which aim to represent as simply, as
completely, and as exactly as possible a set of experimental laws. (Duhem 1962: 19)
What could be the use of theory thus conceived? Success with respect to
simplicity, completeness, and exactitude in a representation would speak for
itself, one might say. But something more is needed to show how we can
conceive of familiar ways of drawing on theory in prediction, application,
and practice.
Duhem responds by elaborating on this conception as allowing us to see
theory as providing a taxonomy in which to locate the regularities, objects,
and processes that are of practical concern. To begin with the regularities:
‘‘Theory is not solely an economical representation of experimental laws;
it is also a classification of these laws’’ (ibid., 23), thus bringing order and
internal connections to what would otherwise be a jumble of empirical
generalizations. Secondly, as illustrated by his discussion of taxonomy
in zoology, the theoretical descriptions of the materials at hand are to
be understood as classification. This is a radio: anyone can see that. If
204  :   
we add that it is a device that transforms electromagnetic waves into

sound-waves, we have not—on this view—expressed a belief in the
reality of electromagnetic waves. Instead, we have pronounced the object’s
classification sub specie physical theory, we have located it in the space of
theoretical kinds.
Poincaré’s views on science were generally tending toward what
Planck considers the great heresy, though stated with caution and dip-
lomacy. In the spirit of Hertz’s view of science as non-mimetic rep-
resentation, Poincaré speaks in Science and Hypothesis of ‘‘images we
substituted for the real objects which Nature will hide for ever from
our eyes. The true relations between these real objects are the only
reality we can attain . . . .’’22 The same point occurs already in the
preface, p.xxiv: ‘‘the aim of science is not things themselves, as the
dogmatists in their simplicity imagine, but the relations between things;
outside those relations there is no reality knowable’’. Of the principle
of conservation of energy, for example, he writes that if we try to
enunciate it in full generality, ‘‘we see it vanish, so to speak, and
nothing is left but this—there is something which remains constant. . . .’’
(Ibid.: 132). This begins to sound quite extreme with respect to what we
can possibly know, but the denial is leavened with an insistence on a hard
core of increasing empirical knowledge:
No theory seemed more solid than Fresnel’s . . . . Nevertheless, we now prefer
Maxwell’s. Does that mean that Fresnel’s work was in vain? No, because Fresnel’s
aim was not to know whether there really is an ether, whether it consists of atoms,
whether these atoms really move in one sense or another; it was to predict optical
phenomena.23
Here Poincaré’s view is quite in harmony with Duhem’s, that the aim of
physical theory is to systematize experimental laws, by whatever theoretical
means lend themselves to that, and without any implication of reality for
the theoretical parameters.
Representation: the problem for structuralism

There is nevertheless something paradoxical in all this. If science describes
nature, Maxwell’s equations must form a theory about what something
  BILDTHEORIE     205
is like. Mustn’t the theory also say what that something is? Here is the
dilemma:
• if Maxwell’s equations are statements, what do they say?
• if they are not statements, how can they amount to a theory at all?
If we leave aside the more instrumentalist (non-statement) options, we can
discern here two not very well-distinguished alternatives. The first is that
no, there is no ether, no mechanical medium subject to wave disturbances,
but yes, there is something, it is the electromagnetic field itself, which
is a thing, and it is not the shape or form of something else. Today
that is an often expressed view, perhaps not always clearly distinguished
from simple rejection of the classical ether: Fields in empty space have
physical reality; the medium that supports them does not (Mermin 1998:
753).
On this option, there is no puzzle, just a new ontology, some new and
previously inconceivable furniture for the world. The second alternative
is more agnostic, and so presumably closer to how Poincaré shies away
from claims to knowledge of the unobservable. It could be expressed like
this:24
• The equations only describe a form or structure—if that is the form
or structure of something, then that something is an unknown entity.
• The field is first of all an abstract entity (mathematical: e.g. a function
assigning values to points in space), though we can of course also
give the name ‘‘field’’ to whatever it is—if anything—that bears this
structure.
• That unknown bearer might well have other properties, just as
ordinary things have properties beside their shape. But the theory
does not describe those.
• Science abstracts, it presents us with the structural skeleton of nature
only.
To begin even this sounds rather reactionary, just when we have discarded
the ether and its frustratingly elusive qualities, there is reference again to
something, whatever it is, that ‘‘bears’’ the field after all. But in retrospect
we can understand it as the beginning of sustained attempts to develop a
structuralist view of science, and we shall follow this attempt through several
successive stages.25
206  :   
Unfortunately, these attempts ran into heavy weather. The problems to

come show their first signs at this time; they will erupt into full-blown
paradox within two decades, and not disappear before our own day, if
then.
Duhem’s problem for Poincaré

On Duhem’s view, as we saw, a physical theory is ‘‘a system of mathematical
propositions . . . which aim to represent a set of experimental laws’’ . But
how can mathematical propositions represent experimental laws? The
latter have terms applying to physical objects and processes, the former
do not. Here we recognize what a decade or so later Reichenbach
presents as the problem of coordination. In the examples Duhem takes
up, the theoretical propositions have terms which—though standing for
e.g. real number valued functions—look like names of properties: ‘‘mass’’,
‘‘charge’’, ‘‘wavelength’’, ‘‘kinetic energy’’.
In atomic theory, for example, such terms appear to be applied to entities
to which no experimental law refers, at least as Duhem sees it. But the
problem is not peculiar to that case. Indeed, as Duhem forcefully points out,
even the experimental laws involve terminology that is meaningful only
because of its links with certain theoretical roles, and not simply because of
their extension among the experimentally manipulated or inspected objects.
The language in which this classification is pronounced is a theory-laden
language:
The role of the scientist is not limited to creating a clear and precise language
in which to express concrete facts; rather it is the case that the creation of this
language presupposes the creation of a physical theory. (Ibid: 151).
and it is in some ways amazing that what Duhem taught there needed to
be understood all over again a half century later.
We are some distance here from a set of pure mathematical propositions,
and indeed, the sort of taxonomy that a scientific theory provides for
its domain does not seem to be something that could be presented in
the language of pure mathematics. So Duhem attempts an account of
the scientist’s theoretical language. It is distinguished by its special jargon,
foreign to the discourse of every day, and so we may be tempted to
assimilate it to the technical language found in a practical art or craft. ‘‘That
would be a mistake,’’ he writes, and continues:
  BILDTHEORIE     207
I am on a sailing ship. I hear the officer on watch shout out the order: ‘‘All hands,
tackle the halyard and bowlines everywhere!’’ A stranger to things of the sea, I do
not understand these words, but I see the men on ship run to posts assigned in
advance, grab hold of specific ropes, and pull on them in regular order. The words
uttered by the officer indicate to them very specific and concrete objects, arousing
in their mind the idea of a known manipulation to be performed. Such, for the
initiated, is the effect of technical language.
Quite different is the language of the physicist. Suppose the following sentence
is pronounced to a physicist: ‘‘If we increase the pressure by so many atmospheres,
we increase the electromotive force of a battery by so many volts.’’ It is indeed
true that the initiated person who knows the theories of physics can translate this
statement into facts and can do the experiment whose result is thus expressed, but
the noteworthy point is that he can do it in an infinity of different ways. ( Duhem
1962: 148–9)
On this matter, Duhem criticized Poincaré, who expressed too simplistic

a view of scientific description. Poincaré wrote that a scientific fact ‘‘is
nothing but a brute fact stated in a convenient language’’, and likens the
relation of that convenient language to more familiar discourse to the relation
between French and German. Duhem sees a much weaker relation between
theoretical description and observable phenomenon:
Between an abstract symbol and a concrete fact there may be a correspondence,
but there cannot be complete parity; the abstract symbol cannot be the adequate
representation of the concrete fact, the concrete fact cannot be the exact realization
of the abstract symbol; the abstract and symbolic formula by which a physicist
expresses the concrete facts he has observed in the course of an experiment cannot
be the exact equivalent or the faithful story of these observations. (Ibid.: 151)
So now we can see the theory of scientific representation, of theoretical

description and mathematical picturing, getting into deep water, as Duhem
begins to press this chasm between representation and what is represented,
which is not bridged by a simple reference relation.26 While we cannot say
that he has as yet perceived the deep perplexities which this forebodes, we
can’t help but sense the foreboding.
Weyl on isomorphism
So we have seen that while Duhem starts by regarding a physical theory
as a set of mathematical propositions, this quickly turns out to be not a set
of propositions in pure mathematics but in a scientific language involving
208  :   
mathematical symbolism. The character of that language is left more as a

challenge than as a solved problem.
Poincaré, though unguarded in the passages Duhem criticizes, was much
more engaged in the foundations of mathematics and of physical geometry.
For him theoretical physics has a pure part, which is pure mathematics,
connected with scientific practice through what he calls conventions—e.g.
to fix the extension of ‘‘congruent’’ in nature. These conventions are what
Poincaré offers for the task of coordination, as we saw before. It is precisely
because he takes the problem of coordination to be solved, by this means,
that Hertz’s dictum that Maxwell’s theory is just Maxwell’s equations could
be one we can hear on Poincaré’s lips.
It would be anachronistic to attribute either to Duhem or to Poin-
caré a clear sense of just how much of a problem this skates over,
how deep a problem is appearing here for the structuralist program.
Scientific theories are, or provide, pictures, representations; they do so
by drawing on the resources of mathematics; but just how do those ‘pic-
tures’, those mathematical constructions, represent what they represent?27 This
is the problem of coordination not in its practical form, in which it is
solved again and again in scientific practice, but in the more abstract
form to be faced when structuralism is developed into a general view
of science—and the threat it poses is that it is possibly not solvable
at all.
This problem appeared explicitly when Bertrand Russell and Rudolf
Carnap developed their aspirant-structuralist views in the late 1920s,
which we will examine in some detail below. Before delving into the
more philosophical contexts in which those writers operate, we can find
the problem displayed graphically in a lecture by another philosoph-
er–physicist. Hermann Weyl expressed the fundamental insight as follows
in 1934:28
A science can never determine its subject-matter except up to isomorphic repres-
entation. The idea of isomorphism indicates the self-understood, insurmountable
barrier of knowledge. [. . .T]oward the ‘‘nature’’ of its objects science maintains
complete indifference. (Weyl 1934: 19)
The initial assertion is clearly based on two basic convictions:

• that scientific representation is mathematical, and
• that in mathematics no distinction cuts across structural sameness.
  BILDTHEORIE     209
The former is just the structuralist conviction we have witnessed so clearly

in the mathematical physicists of the late nineteenth and early twentieth
century. The latter is a take on what mathematics is and does, expressed by
one of its most philosophically profound masters.
Care is needed with Weyl’s point. If two structures are isomorphic,
but are parts of a third structure, then they may be distinguishable by
their relations to the whole. The Euclidean plane contains many mutually
congruent triangles. In addition, ‘‘isomorphic’’ often has a restricted sense
in context: ‘‘isomorphic groups’’ can mean ‘‘structures related by a group
isomorphism’’, but though the same as far as the group operations are
concerned they may be quite different in other respects. If however we
consider two structures, taken by themselves, and an isomorphism which
preserves all the definable structural aspects, then indeed they are not
discernible in any mathematically representable way.
But surely we can point to differences between domains with the same
structure? Weyl illustrates this with the example of a color space and an
isomorphic geometric object. Just a few years before his writing the CIE
1931 color space was created by the International Commission on Illumination,
based on experiments on human color vision performed in the 1920s.
This color space is a region in the projective plane. If we can nevertheless
distinguish the one from the other, or from other attribute spaces with that
structure, doesn’t that mean that we can know more than what science, so
conceived, can deliver? Weyl accompanies his point about this limitation
with an immediate characterization of the ‘‘something else’’ which is then
left un-represented.
This—for example what distinguishes the colors from the points of the projective
plane—one can only know in immediate alive intuition. (Ibid.)
When two objects are represented by isomorphic structures—hence, when

their representation presents no difference between them—we may still
know that they are distinct and in un-represented ways different objects.29 In
practice there would always be further differences that could be represented
as well (by enlarging the mathematical model), but Weyl clearly wants us
to think of the limiting case where there is nothing more to do in the
modeling. Even then it seems we must grant that we could know that the
two represented objects are distinct. But Weyl’s solution to how this is
possible, appealing to our immediately alive intuition, may not be entailed.
210  :   
What could the additional knowledge be? Weyl’s assertion sounds

paradoxical, it sounds diminishing with respect to scientific knowledge.
We seem to be left with four equally unpalatable alternatives:
• that either the point about isomorphism and mathematics is mistaken,
or
• that scientific representation is not at bottom mathematical represent-
ation alone, or
• that science is necessarily incomplete in a way we can know it to be
incomplete, or
• that those apparent differences to us, cutting across isomorphism, are
illusory.
In his comment about immediate alive intuition, Weyl appears to opt for
the second, or perhaps the third, alternative. But on either of these, we face
a perplexing epistemological question: Is there something that I could know to
be the case, and which is not expressed by a proposition that could be part of some
scientific theory?
The problem with Mary

Weyl’s paradox is quite abstract, in comparison with the discussions of the
Bildtheorie physicists. Perhaps his brief illustration of the isomorphism of the
color space to the projective plane is abstruse as well. But this example, to
which he offers his ‘‘intuitionist’’ philosophical response,
• what distinguishes the colors from the points of the projective
plane—one can only know in immediate alive intuition
has a narrative version that was proposed in philosophy of mind for a quite
different purpose. It is known, after its creator, as Frank Jackson’s Mary
Problem; and here is its statement in David Lodge’s novel Thinks:
That’s Frank Jackson’s Mary, the colour scientist. The idea is that she’s been
born and raised and educated in a totally monochrome environment. She knows
absolutely everything there is to know about colour in scientific terms—for
example, the various wavelengths that stimulate the eye in colour recognition—but
she has never actually seen any colours. Notice there are no mirrors in her room,
so she can’t see the pigmentation of her own face, eyes, or hair, and the rest of her
body is covered. Then one day she’s allowed out of the room, and the first thing
she sees is, say, a red rose. (Lodge 2001: 53)
  BILDTHEORIE     211
Having all the knowledge about color that theoretical physics and
physiology can provide, will Mary be able to classify the rose as red—and
if so, how?
Weyl would point out that the projective plane has many symmetries,
so that there are in effect many isomorphisms between the color space
and that mathematical structure. Transform an isomorphism appropriately
and you generate another isomorphism. Like Locke’s ‘‘inverted spectrum’’
problem, only worse!
So if what Mary knows is just this structure, that is certainly not enough
to identify which color is which. If her knowledge includes differentiating
connections of the colors with things outside the color space—e.g. of
what is filtered by a ruby—then the same problem reappears in principle.
If all she knows is the structure of the entire domain, colors and rubies
and whatever else may be included, that will not be enough to single out
specific features. Even if they happen e.g. to have a unique location in the
spatial configuration, that may not be one identifiable by Mary in her own
(though only partially accessible) frame of reference.
Can’t Mary do better than a disembodied mind endowed with a complete
mathematical representation of nature? She can indeed. Somewhat later in
the novel we read an imagined sequel, about the day when Mary will be
allowed to come out:
She glanced down at her own hands . . . sheathed in serviceable pigskin which
she was permitted to remove only at night, in total darkness, with the assistance
of the blind maidservant Lucy, thus preventing any inadvertent glimpse of the
pearly pinkness that—so she understood—tinted the translucent plates covering
the dorsal surfaces of her finger-ends.
Well, she would soon be able to see her fingernails along with many other
things . . . . (Ibid.: 154–5)
When the gloves are removed in broad daylight, she will see her fingernails,
and know where they are located in the color space, because she knew
already how to identify her nails and had already the knowledge that they
are pink.30 Suddenly that theoretical object, the color space, will be subject
to coordination (to that extent), she will be able to say ‘‘this is pink’’, and if
she can gather some more such clues, soon know her practical way around
the colored world. The change is that, at the point of success, she will
be able to locate any pink object she sees in the same region of the color
212  :   
space—be able to say ‘‘this, in front of me, is there in the color space’’, and
the many symmetries of the projective plane will no longer respect what
she can discriminate.
This sounds like a way to success, but note how crucially it relies on
the indexical: what Mary must be able to say at this point is ‘‘these are my
fingernails; my nails are pink’’. So let us repeat the question that we posed
for the conclusion of Weyl’s reasoning, which was:
Is there something that I could know to be the case, and which is not expressed
by a proposition that could be part of some scientific theory?
There is some sense, undoubtedly, in which Mary can answer this question
affirmatively for herself—but in what sense?31 In the solution here ima-
gined, the crux is the self-attribution of a location in color space. Is that
the only way? This question we will pursue throughout Part III, and we
will return to it explicitly at the end.
9
The Longest Journey: Bertrand
Russell
The first sustained, rigorous development of a structuralist view of sci-

ence appeared at the hands of Bertrand Russell, in whose writings the
philosophical motivation precedes a precise formulation drawing on math-
ematical logic. The appeal lay perhaps mainly in the motivation, though in
retrospect his arguments were couched in a by then passé ‘mirror of nature’
view of mind. The rigor came from his founding of theoretical physics in
a mathematics constructed along logicist lines, which is also precisely what
proved his undoing.
Prolegomena to Russell’s conversion to structuralism

Bertrand Russell provides an excellent example of a philosopher who ini-
tially resisted structuralist leanings, insisting specifically on a quite contrary
view. This connected directly with philosophy of mathematics, because on
this early view, he took geometry to be not simply abstract mathematics
but the theory of physical space.
Lobachevsky and Helmholtz

When non-Euclidean geometries were first created, the question arose
whether there could be an empirical test to decide whether space is
Euclidean. In principle it seemed easy enough. For example, in Euclidean
space the interior angles of a triangle add up to 180 degrees. Gauss’s
earlier studies of surfaces with negative curvature could be applied here. In
hyperbolic geometry, in a three-sided figure whose sides are arcs of minimal
length, the interior angles sum to less than 180 degrees, the defect increasing
214  :   
with the area. But the extent of this defect depends on the curvature of
the surface, which may be very small in which case the differences from
a surface of zero curvature (Euclidean plane) would show up significantly
only in very large areas.
So Lobachevsky looked into astronomical data. He suggested that one
might ‘‘investigate a stellar triangle for an experimental resolution of the
question.’’ The ‘‘stellar triangle’’ he proposed was the star Sirius and two
different positions of the Earth at six-month intervals. But if there was such
a defect in the sum of the interior angles, it was still within the limits of
measurement error.
Indeed, for any defect you can name and any size, there would be
a curvature constant small enough to guarantee that the defect would
still be within those limits. So, if there are also limits on the size of the
regions we can inspect in this way, in the course of human history, the
difference between Euclidean and hyperbolic space may be beyond what
such measurements can reveal.
But there is a deeper problem. The measurement results will also depend
on what we take as measurement standard, and what presuppositions are
in place with that choice. This point can be graphically (if imperfectly)
illustrated by von Helmholtz’s examples of mirror-universes. Helmholtz
imagined that we are making measurements to determine whether our
space is Euclidean. Do the interior angles of a triangle add up to 180
degrees? Suppose they do. At the same time he imagines that we are
reflected in a huge concave mirror.
In that mirror we see little people moving around with rulers and ‘doing
the same thing’ as we do. Of course, they get the same results, and announce
that they live in a Euclidean space. We want to disagree; they are moving
around on a concave surface, and their measuring rods change length as
they work.
All geometrical measurements . . . with regularly varying images of real instruments

would yield exactly the same results as in the outer world. . . . I do not see how men
in the mirror are to discover [the in-]correctness of Euclid’s axioms. (Helmholtz
1956: 661)
The reflection in the mirror is a perfect image of what is reflected. If we do

not look at any relation of these two sides to ourselves (or any third object
or background in which they are located) then they have precisely the
  :   215
same structure. Thus this example illustrates an isomorphism. Two objects

which are isomorphic are mathematically no different from each other. We
my overlook this if we, very naturally, think of the two as parts of a larger
structure (or as related to ourselves as onlookers). In that larger context
such parts may indeed be distinguishable, namely by their relations to other
parts or to the whole. But if we are dealing with the world as a whole and
a model of the world—as in a theory of physical space—then we have no
such recourse.
Mathematicians’ reaction
One response to the birth of non-Euclidean geometry had thus been that
the question might be subject to empirical test. This had not fared well,
in view of this relativity to the measurement standard, and was by no
means the only or even the most acceptable idea about this new subject
at the time. Another main reaction was that to understand non-Euclidean
geometries they had to be interpreted within Euclidean geometry.1 To
be intelligible they should be readable as strangely worded descriptions of
parts or aspects of Euclidean space. That this is possible was shown initially
for the hyperbolic plane by Beltrami in his ‘‘Saggio di Interpretatione della
Geometria Non-Euclidia’’ (1868).2 Beltrami himself expressed his goal as
maintaining Euclidean geometry as the one true theory of space.
Felix Klein perceived the limits of this effort and first attempted to
interpret both Euclidean and non-Euclidean geometries within projective
geometry. But very soon through Klein’s work, and even more radically
through Riemann’s, there came into being such a cornucopia of geometries
that any true theory of space underlying all of them could be very little
more than pure logic.
The mathematicians’ response was that none were privileged, all geo-
metries were on a par and certain to remain parts of mathematics. But how
do they relate to the physical world which we investigate empirically? With
the focus shifted to interpretation, through such puzzles as Helmholtz’s
mirror worlds, that problem has now taken on a very different shape. It is
no longer a straightforward question that could be settled by measurement.
Klein’s idea that a single projective space could be the domain of either
a Euclidean or non-Euclidean geometry introduced a new way of viewing
the matter. Thus with Klein, and also Lie, we get a different feeling:
they display the Euclidean and non-Euclidean geometries as pertaining
216  :   
to a single space, the space of projective geometry, but with different

metrics—or if you like, different congruence relations. Different congruence
relations, corresponding in practice to different measurement standards,
imposed on the same space give it a different geometry. We can see an
initial locus here for the problem of coordination.
The mathematicians’ attitude tied in well with Helmholtz’s illustration:
with different measurement techniques, we may be mapping out different
congruence relations. To find out whether two different objects are
congruent we would presumably do something like this: transport a meter-
stick from one to the other. But that we have designated this instrument
as a meter-stick, hence as establishing congruence, is a choice, and may
have alternatives. Noting then how this ties in with Helmholtz’s mirror
imaging, and the latter with Weyl’s point, we can now see how pervasively
Weyl’s paradox shows up. Within the realm of mathematical structure, we
cannot make the distinctions clearly used in empirical scientific practice.
The physicist—or more likely, the realist philosopher of science—‘‘wishes
to find out what things are like’’. But the mathematician says ‘‘that makes
sense only while taking for granted your choices when you accepted
or created measurement standards and kept in place a whole host of
presuppositions—apart from that, there are no real differences to be
found’’.
Bertrand Russell: There must be ‘‘a fact of the matter’’!

When Bertrand Russell wrote his Essay on Geometry he followed Klein’s lead
in this reconstruction of the different geometries, which ranged Euclidean
and non-Euclidean geometries under projective geometry. He even gave
what purported to be an a priori, or perhaps transcendental, argument
that physical space must be of this sort—in modern terms, this means that
physical space must have constant curvature.3 But Russell added that there
must be a unique real relation of congruence.
Given that, it follows that there is a fact of the matter, whether space is
really Euclidean or not. Geometry, at least the geometry used in physics,
must be according to Russell a non-vacuous theory of real spatial relations.
But how are those real relations identified? Russell maintains that we are
directly acquainted with them, through intuition. Having been pressed to
elaborate on this view by Poincaré’s review in 1899, Russell dismissed the
question.4 That is, he said, like asking me to spell the letter ‘‘A’’.5
  :   217
The view that Russell was resisting may aptly be called an early form
of structuralism, though it is anachronistic to name it thus.6 Russell insisted
to the contrary that there was something more to be known, and in
fact captured in scientific theories representing nature, than mathemat-
ics affords—something more than mathematical structure. (At times he
took this to an almost absurd extreme, asking even ‘‘wherein . . . lies the
plausibility of the notion that all points are alike?’’7 ) Certainly, for the
mathematician, all congruence relations, and all possible denotata of their
primitive terms, have equal status—but for the empirical sciences there
must be more, there must be a fact of the matter that goes beyond what
mathematicians can describe.
Russell’s structuralist turn

When physics describes the world, how much does it describe? Perhaps
there are real spatial relations in contrast to gerrymandered artificial metrics,
as Russell had been contending. Which ones are denoted by the geometric
language used in physics seems not to matter, in principle anyway. But
then, better to say perhaps that science does not describe nature down
to that level of detail. Better to say, perhaps, that science describes only
structure without content.
We see Russell taking this line in his Problems of Philosophy (1912).
Indeed, he elaborates there a structuralist position that is uniform with
respect to the entire content of the physical sciences. Let us begin with
what he says about space and geometry, and then note how this is extended
to the ether, electromagnetic waves, atoms, and so forth. Russell insists
that to understand a proposition (so that it may be capable of being judged
true or false) we must be acquainted with all its ultimate constituents.
This includes certain concrete individual entities but also the properties
and relations expressed by its predicates. However, we are acquainted only
with those things which are part of our direct experience.8 Physics speaks
of things well beyond the ken of direct encounter in experience, things in
the External World. Unfortunately, it is exactly that External World which
science purports to describe. How is this possible?
Russell could have said that science describes nature simply by saying
that there exist entities, with which we are not acquainted, but which have
218  :   
the same properties and stand in the same relations, as enter our direct
experience. Beginning with the theory of space, he appears to say precisely
that at first:
If, as science and common sense assume, there is one public all-embracing physical
space in which physical objects are, the relative positions of physical objects in
physical space must more or less correspond to the relative positions of sense-date
in our private spaces. There is no difficulty in supposing this to be the case. [. . .]
thus we may assume that there is a physical space in which physical objects have
spatial relations corresponding to those which the corresponding sense-data have
in our private spaces. It is this physical space which is dealt with in geometry and
assumed in physics and astronomy. (Russell 1912/1997: 30–1)
The reason he has given for this is that the objects are postulated in the first
place as causes to explain our sensations, and causation presupposes spatial
contiguity. But in that motivating discussion he was moving back and
forth between his rather strange ontology and the common sense picture.
Attempting to be more conscientious we see the gloss fading rapidly:
Assuming that there is physical space, and that it does thus correspond to private
spaces, what can we know about it? We can know only what is required in order to
secure the correspondence. That is to say, we can know nothing of what it is like
in itself, but we can know the sort of arrangement of physical objects which results
from their spatial relations. We can know, for example that the earth and moon
and sun are in one straight line during an eclipse, though we cannot know what a
physical straight line is in itself. . . . Thus we come to know much more about the
relations of distances in physical space than about the distances themselves. . . . We
can know the properties of the relations required to preserve the correspondence
with sense-data. . . . (Russell 1912/1997: 31–2)
Note well that now we are asserted to know, not the distances or other
spatial relations between bodies, but only the properties of those relations. That
is, we know what the abstract structure is. If we ‘‘cannot know what a
physical straight line is’’ we certainly cannot know such other relations in
physical space as congruence. All we know about that, presumably, is what
the axioms of a geometry can say about congruence.
Has the ‘‘real property’’ realism adopted in response to von Helmholtz,
Klein, and Poincaré been given up? That is not clear. What has been
given up certainly is any pretence to knowledge of those real properties
themselves.
  :   219
Every analogy with familiar things, like waves in water, planets, and
billiard balls was already heavy with disanalogies. So Russell says: we
can only infer the properties of the properties, and the properties of the
relations—the type of structure:
We can know the properties of the relations required to preserve the correspond-
ence with sense data, but we cannot know the nature of the terms between which
the relations hold . . . . [A]lthough the relations of physical objects have all sorts
of knowable properties, . . . the physical objects themselves remain unknown in
their intrinsic nature . . . .’’ (Russell 1912: 32, 34)
The Analysis of Matter (1927) makes this precise: this structure is exactly, no
more and no less, what can be described in terms of mathematical logic.
The logic in question is strong, and today we would see it as higher order
logic or set theory. But still, how little this is! Science is now interpreted
as saying that the entities stand in relations which have such properties as
transitivity, reflexivity, etc. but as giving no further clue as to what those
relations are.
. . . whenever we infer from perceptions it is only structure that we can validly infer;
and structure is what can be expressed by mathematical logic. The only legitimate
attitude about the physical world seems to be one of complete agnosticism as
regards all but its mathematical properties. (Russell 1927: 254, 270)
Newman’s objection to Russell

The mathematician M. H. A. Newman made the crucial critical point in a
review article concerning The Analysis of Matter:9
. . . it is meaningless to speak of the structure of a mere collection of things . . . .
Further, no important information about the aggregate A, except its cardinal
number, is contained in the statement that there exists a system of relations, with
A as field, whose structure is an assigned one. For given any aggregate A, a system
of relations between its members can be found having any assigned structure
compatible with the cardinal number of A. (Newman p. 140)
We can illustrate Newman’s point quite easily. Suppose I have seven

neighbors, and I insist that they instantiate a model M I have of a rigidly
hierarchical social structure.
This model M we can envisage as follows: it has seven elements and
a ‘directly under’ relation. To visualize it, its elements are the general,
220  :   
directly under him two colonels, and directly under each colonel there are
two majors. How can I say that my neighbors instantiate this hierarchical
structure? To justify that I have to define a relation which has the same
properties as the relation directly under has in my model. So I arbitrarily label
my seven neighbors as follows: 1, 10, 11, 100, 101, 110 111. Then I define
the relation as follows: neighbor 1 does not bear to anything; any
neighbor with a label of form Y1 or Y0 bears to the neighbor labeled
Y; and that is all.
Of course there is no sense to the idea that I have discovered a
hierarchical structure in my neighborhood by carrying out this ‘‘pencil
and paper’’ operation. But if I were to say that ‘‘my set of neigh-
bors is hierarchically structured in the fashion of model M’’ means only
that there is a relation on that set which has the same properties
as the directly under relation in M, then the trivializing result follows.
For there is indeed such a relation on that set of neighbors provided
only there are seven of them. In general, equality of size between
two sets means just that there is a one-to-one correspondence between
them, and that correspondence can be used to single out a copy in
the one set of any relational structure there may be displayed in the
other.10
You can easily see that this is von Helmholtz’s move with concave
mirrors, repeated at a more abstract level. The very reasons that drove
Russell originally to ‘‘real property’’ realism, and then later drove him
away from that epistemologically uncomfortable position, have returned to
plague him again.
Russell capitulates
Russell capitulated. In a letter to Newman, he reverted to the ‘‘real
property’’ realism of his early days. The only difference is an update from
real spatial relations to real spatio-temporal relations:
Dear Newman, [. . .] It was quite clear to me, as I read your article, that I had
not really intended to say what in fact I did say, that nothing is known about
the physical world except its structure. I had always assumed spatio-temporal
continuity with the world of percepts, that is to say, I had assumed that there might
be co-punctuality between percepts and non-percepts. . . . And co-punctuality I
regarded as a relation which might exist among percepts and is itself perceptible.
(Russell 1968: 176)
  :   221
He had always assumed that? On the contrary, that is precisely what he

shied away from in The Problems of Philosophy and precisely why he tried
for an extreme structuralism in The Analysis of Matter.
However that may be, Russell’s response clearly re-introduces his earlier
epistemically charged ‘‘real property’’ realism, slightly updated, with the
claim that we are directly acquainted with certain crucial spatio-temporal
relations. Science does not merely assert that there are relations with the
requisite formal character needed to give sufficient complexity to the causes
it postulates. On the contrary, it asserts that (to put it succinctly)
those entities postulated, and known at best by description, bear
certain relations which we know by direct acquaintance (both to each
other and to entities that we know by direct acquaintance).
Russell does not tell Newman very much, and mentions only one such
spatio-temporal relation—contiguity, ‘co-punctuality’—that is instantiated
by items of direct acquaintance. But he probably had in mind the con-
struction of space-time from relations among events in which he followed
Whitehead’s approach, so may have intended all significant space-time
relations or at least all those that can be instantiated in experience.
Disentangling Russell’s response

We need first to isolate within Russell’s response the part that is essential
and sufficient for countering Newman’s objection. Newman himself had
pointed out an obvious minimalist way to deal with his point. He suggested
that Russell could repair his position by distinguishing between ‘important’
and ‘unimportant’ relations. It would not matter just what distinction is
meant by this, as long as there is one. Then the repair would go as follows:
Relations fall into two classes, important and unimportant. The important
relations in nature instantiate the following structure. . . .
Newman’s argument is then blocked. Given a one-to-one mapping from
graph G to set S, any given relations on G will of course have images in
S. But there is no guarantee that their images will be important relations.
Hence the repaired assertion is contingent, possibly true and possibly false,
and hence informative.
But this is just a logical maneuver. Without any information about
that supposedly (postulatedly?) factual distinction between the important
222  :   
and unimportant, we have no grip at all on what the assertion says about
the world.11 Newman in fact derided his own suggestion as a counsel of
despair.
One way to take Russell is that he accepted Newman’s suggested way
out but added something so as to regain informativeness. He added that
the ‘important’ relations are exactly those with which we have direct
acquaintance. This ties in with the epistemology in his earlier Problems of
Philosophy: to understand a proposition requires acquaintance with each of
its constituents. Since we do understand some general propositions it follows
from this that we are directly acquainted with certain Universals—all the
ones that appear as constituents in the propositions we understand. But are
we acquainted with all the arbitrary relations Newman recognizes? That
does not follow. For example, we understand the proposition that every
set has a well-ordering. That entails acquaintance with the (higher order)
Universal being a well-ordering, but not acquaintance with e.g. a particular
well-ordering of the real numbers. So within Russell’s epistemology there
was indeed a natural division into important and unimportant relations to
which he could appeal.
To complete the repair then, Russell has to say that a scientific theory
tells us something about the structure of certain relations with which
we are acquainted, instantiated by those postulated entities in nature with
which we are not acquainted, and about whose qualitative properties we have no
idea.12
But now we can also see that the repair is still not finished. There is
the danger that it does not go far enough, and may still leave us with
very uninformative scientific theories, so two tasks remain. Task One:
those ‘‘certain relations with which we are acquainted’’ must be specified
to some extent. Russell clearly saw this, for here we arrive at Russell’s
specific addition to Newman’s way out: ‘‘ spatio-temporal continuity [of
the physical world] with the world of percepts’’. At the very least the
terms ostensibly spatio-temporal in the theory denote the very spatio-
temporal relations with which we are directly acquainted, and they relate
the postulated entities both to each other and to the observed entities of
our direct acquaintance. But this is a postulate: Russell is telling us that
the ostensibly spatio-temporal terms are univocal, that they denote the
same relations when they appear in physical theory as when we use them
ordinarily.
  :   223
Task Two: to show that this sort of construal of theories, with that
elaboration added, now has the resources to be sufficiently informative.
Suppose that the space-time geometry of the physical world is indeed
as science says it is, but when it comes to any other relations and prop-
erties attributed to those physical things out there, science can only
specify the formal structure. Then won’t Newman’s argument apply
again?13
The assertion that there are certain relations with a given formal structure
on those things will be automatically true, provided only it is logically com-
patible with the space-time geometry in question, plus some assumptions
about cardinalities. To put it the other way around, any additional physics
beyond the space-time geometry might then give information only about
cardinalities. That is equally disastrous unless the entirety of physics reduces
to providing space-time models. Such was the dream of W. K. Clifton’s
‘‘space theory of matter’’, revived eventually in ‘‘geometrodynamics’’, and
may still be alive in some programmes in the foundations of physics.
But it is certainly not something that we would be ready to accept as a
priori certain, I would think, nor something to which we would want to
indissolubly connect a philosophical position.
Conclusion
So Russell’s extreme structuralism collapsed and his repair, through a
reversion to ‘‘real property’’ realism, does not obviously save him. In
effect, it takes him back to the point where he had hoped that some
admixture of structuralism with realism might make his peace with von
Helmholtz’s lights and mirrors. But there is something striking in his repair,
that alters the view he presents of science quite radically. In The Analysis
of Matter scientific theories are presented as being completely formulable,
without loss, in the language of pure mathematics. Such a formulation
would involve no direct reference to what we encounter in experience,
let alone indexicality, self-reference, or self-location. Reference to actual
individuals would be achieved entirely by objective description. This aspect
of his structuralist view of science is lost in the repair, when our direct
acquaintance with certain entities separates what science is about from what
logical gerrymandering concocts.
224  :   
This story has its sequel in writings by Rudolf Carnap, Hilary Putnam,
and David Lewis, where we will recognize the themes from Helmholtz,
Russell, and Newman in a new setting. The problems are transposed
there to a context governed by concerns in philosophy of language and
analytic metaphysics. They become clearer there and so, if anything, more
devastating. But when the problems become clearer, so do the possibilities
of solution. Is there after all a viable form of structuralism about science?
I shall argue for a specifically empiricist structuralism, which escapes
trivialization by recourse to resources that we have been encountering all
along the way: the role of indexical and ostensive reference.
10
Carnap’s Lost World and
Putnam’s Paradox
Rudolf Carnap’s most famous work purveys a structuralist philosophy of

science, and runs into essentially the same problem as Russell faced—the
problem that Newman pointed out. But unlike Russell, Carnap exhibited a
fluctuating awareness of the difficulty besetting his programme, and of the
limited options his epistemology allows for escape.1 The option he finally
chooses is in the pattern that Newman suggested, somewhat ironically, to
Russell, but Carnap attempts simultaneously to refer to experience and
to claim that the notions he needs are experience-independent. The basic
problem returns after some decades when Hilary Putnam puts it to good
use in his seminal critique of metaphysical realism. In both cases I shall
argue that the solution—or rather, dissolution—of the problem is possible
with the introduction of indexical reference, and I shall explore the role of
the indexical further in the next chapter where I will discuss structuralism
independently of this history.
Carnap: Der Logische Aufbau der Welt

Carnap published his Logische Aufbau two years after Russell’s Analysis of
Matter, and begins Part Two with the announcement
we shall maintain and seek to establish the thesis that science deals only with the
description of structural properties of objects.
And he immediately specifies what this will amount to:
A property description indicates the properties . . . , while a relation descrip-
tion . . . does not make any assertion about the objects as individuals.’’ (section 10,
p. 19)
226  :   
There is a certain type of relation description which we shall call structure description.
Unlike relation descriptions, these not only leave the properties of the individual
elements of the range unmentioned, they do not even specify the relations
themselves which hold between these elements. In a structure description, only
the structure of the relation is indicated, i.e. the totality of its formal properties.
(section 11, p. 21)
The crucial problem

The problem that we saw so clearly spelled out by Hermann Weyl, then
appears in the next section:
Thus, our thesis, namely that scientific statements relate only to structural proper-
ties, amounts to the assertion that scientific statements speak only of forms without
stating what the elements and the relations of these forms are. Superficially, this
seems to be a paradoxical assertion . . . in empirical science, one ought to know
whether one speaks of persons or villages. This is the decisive point: empirical science
must be in a position to distinguish these various entities. . . . (section 12, p. 23)
How does Carnap deal with this problem? He admits that it looks as if the
use of definite descriptions will be successful only if eventually it relies
on some ostensive description—that is, some recourse to pointing or other
indexical or demonstrative form of reference. He takes that look to be
deceptive:
However, we shall presently see that, within any object domain, a unique system
of definite descriptions is in principle possible, even without the aid of ostensive
description. (pp. 24–5)
But his elaboration immediately lets us doubt this irenic assurance:

It is of especial importance to consider the possibility of such a system for the
totality of all objects of knowledge. Even in this case it is not possible to make an
a priori decision. But we shall see later that any intersubjective, rational science
presupposes this possibility. (p. 25)
What we see here is a vacillation between several possibilities. (A) Is nature

so structured that everything can be uniquely identified by means of a
description that captures only that structure? He has an illustration: the
stations in the Eurasian railroad system can be each uniquely identified
by listing just the railroad connections between them. But if this system
had some global symmetries, then the identifications would not be unique.
 ’     ’  227
What if we bring in more structural features, besides the railroad connection

relation? Can we be sure that if we bring in enough—when we are at
the level of ‘‘a system for the totality of all objects of knowledge’’ we will
not be plagued with such non-uniqueness? Here we come to (B), ‘‘any
intersubjective, rational science presupposes this possibility’’. If that is so
we have a transcendental justification for the assertion that knowledge of
structure includes all the knowledge there is to be had.
Disturbingly, Carnap throws out yet a third option in this Part of the
Aufbau: that (C) we should adopt an ontology in which isomorphism
implies identity. Quite a move, for a father of logical positivism and the
author of Pseudoproblems in Philosophy!
Return to the problem

In Part Four of the Aufbau Carnap returns to this problem.2 He has
been envisaging reconstructions of the body of science, or other parts
of knowledge, and it seems that he views the structuralist programme
announced in Part One of the Aufbau in the same way as Russell at this
point.
A purely structural statement must contain only logical symbols . . . . Thus . . . the
problem arises whether it is possible to complete this formalization by eliminating
from the statements of science these basic relations . . . . That this elimination is possible
becomes obvious . . . . (section 153)
The elimination must replace relation descriptions by structure descriptions,

as we saw above, that is to say, by assertions that there are relations having
certain properties, and instantiated in certain ways.
But now (section 154) he notices a difficulty—in effect the very same
difficulty that M. H. A. Newman pointed out to Russell just a little later. If
a statement to the effect that there are relations having certain properties and
are instantiated in certain ways is framed entirely in logico-mathematical
terms then it can be satisfied in any set of sufficient size. For the relations
in question can be taken to be suitably chosen sets of ordered pairs, triples,
and the like, and the suitable choice is possible provided enough elements
are available.
Thus Carnap writes in section 154, about the argument that he had given
to support his claim that the elimination is possible, that he had assumed
that:
228  :   
. . . after a replacement of one set of basic relations by another, the constructional

formulas of the system would not remain applicable . . . . However, our assumption
is justified only if the new relation extensions are not arbitrary, unconnected pair
lists, but if we require of them that they correspond to some experienceable,
‘‘natural’’ relations (to give a preliminary, vague expression).
If no such requirement is made, then there are certainly other relation extensions
for which all constructional formulas can be produced.
In other words, he admits that if the ‘suitable choice’ of sets of ordered

pairs is unconstrained, if they can be ‘unconnected pair lists’, then a theory
so framed will be true provided only there are enough elements to choose
the pairs from. Practically all empirical content would be lost, in such a
structuralist reconstruction of a scientific theory!
In contrast to relations of this sort [unconnected lists of pairs], we wish to call
relation extensions which correspond to experienceable, ‘‘natural’’ relations founded
relation extensions.
So one of the properties that must be attributed to the relations mentioned

in a structure description is this ‘foundedness’. A scientific theory will be
reconstructed as an assertion to the effect that there are founded relations
instantiated in a certain pattern.3
As Newman pointed out, any distinction of this logical form blocks the
trivialing argument. In that sense, Carnap has solved his problem, formally
speaking. But does this not give up entirely on the structuralist programme
he had announced? That a relation is ‘experienceable’, ‘natural’ is not
something that can be formulated using only logical vocabulary.
Carnap sees this very well, and makes what seems in retrospect a
desperate move. He submits, though guardedly, that this ‘‘founded’’ is itself
a new logical term: ‘‘It is perhaps permissible, because of this generality,
to envisage foundedness as a basic concept of logic.’’ But this concept was
introduced in an analysis of how one might add non-logical content or
reference to a structure description!4
The two sides of Weyl’s coin

As noted above, we see two problems for the concept of science as a
representation by means of mathematical models. First of all as Weyl says
explicitly, there are distinct but isomorphic structures discernable in nature,
 ’     ’  229
but mathematical models will capture only the structure that is common
to them. Thus to have specific knowledge of one of such a pair, we
must know more than that it is adequately represented by some such
mathematical structure. Secondly, as Klein and Helmholtz had pointed out,
given any significant limitation on what is observable or detectable by us,
there will be many non-isomorphic structures that fit what we do observe
or detect. I said above that these are two sides of one coin. In what sense is
that so?
Russell reacted to the latter problem by insisting that we have know-
ledge that goes beyond what can be mathematically represented: we are
acquainted with the physical congruence relation, whose mathematical rep-
resentation is just one of many congruence relations that can be imposed
on a projective space. One of these relations has a privileged epistemic
status, it is the one that ‘enters’ our experience.
Now we see Carnap opting for a precisely similar solution to the
former problem. He assigns a privileged epistemic status to certain relations,
and connects that status with what can be experienced. We hear an
echo here of the ‘problem of coordination’ which due to Schlick’s and
Reichenbach’s discussions must have been salient in Carnap’s world. The
problem Carnap has encountered pertains to the connection between
any structural representation and what it represents; preoccupied with
coordination, it is seen as relating specifically to the theory-experience
relation. We see the essential problem in much more general form when
Hilary Putnam takes it up some four decades later.
Putnam’s Paradox
At the American Philosophical Association in 1976, Putnam produced his
most famous argument against metaphysical realism. He called it his model-
theoretic argument. David Lewis called it Putnam’s Paradox, because, he
claimed, this argument would show that almost any theory at all is true.
For Lewis, therefore, the argument can be cited in support of a stronger
realism than the realism that Putnam targeted, namely a stronger realism
which can resist the argument. I will begin with what I shall call the core
of the argument.5
230  :   
The argument
Suppositions:
1.   has infinitely many pieces
2. Theory T is consistent
3. Theory T says that there are infinitely many things
Exactly one meta-mathematical result will be required in the proof; after
that it will proceed precisely like Newman’s argument against Russell. That
result is the Loewenheim–Skolem–Tarski–Vaught theorem, which shows
that if a theory has an infinite model then it has models of every infinite
cardinality.6
1. Since T is consistent and has (only) infinite models, it has a model of
the same cardinality as  .
2. Let be a one-to-one correspondence between the elements of that
model and the pieces of the world. (This correspondence exists by
definition, given that the two sets have the same cardinality.)
3. Each term of the theory will have an extension in the model; DO:
assign to each term precisely the image under of its extension in
the model. Take that image to be its extension in  .
4. Since all the theorems of T are true in the model (i.e. given the terms’
extensions in the model) all of them will also be true in  
when given those images as extensions in  .
5. So T is a true theory about  .
Figure 10.1. Putnam’s Paradox
An example, an objection, and a ‘way out’

Let’s think about an example of a simple theory and see how Putnam’s
argument applies. I will call this the Water Theory.
 ’     ’  231
A1) There are infinitely many distinct bodies of water

A2) There are infinitely many things that are not water
Assume that   is or can be regarded as having infinitely many

things in it. Then there is an extension that can be assigned to the phrase
‘‘body of water’’ such that both A1 and A2 come out true.
That is not very surprising! But as Putnam suggests, the theory can
contain an extra axiom for every assertion we want to make—or if you
like, every one that is ever in the fullness of time observed to be true—of
the form ‘‘X is a body of water’’ or ‘‘Y is not a body of water’’. Then there
will still exist an assignment of referents to the terms ‘‘X’’, ‘‘Y’’ and the
like, and an extension to ‘‘body of water’’ which will make A1, A2, and all
the extra axioms true. And so forth: for the argument is very general, and
does not rely on any specific assumptions about the content of the theory.
Whatever constraints we want to introduce Putnam challenges us to make
explicit; then he will add them to the theory under consideration . . . all
those constraints are after all ‘‘just more theory’’.
David Lewis then offered the following way out. When Putnam writes
‘‘Map the individuals of M one-to-one into the pieces of  ,
and use the mapping to define relations of M directly in  ’’
Lewis insists that we must add ‘‘in such a way that the defined relations
in   are natural relations’’. The ‘‘it’s just more theory’’ objection
is evaded, because the distinction between natural and unnatural relations
does not enter into the formulation of the theory. Rather, it is a distinction
in nature which must be respected. Must be respected how? Respected in
our view of just what it is that is asserted by a scientific theory, that is, in
our spelling out of the truth conditions for that theory. We can understand
the theory as a non-trivial assertion about what things are like only on an
interpretation of the terms as referring to natural relations.
The terminological link with the passage I quoted from Carnap about
foundedness is not accidental: we have here essentially the same form
of solution to essentially the same problem faced by the two offered
conceptions of science.
What is that postulated distinction in nature? Plato is said to have touched
on it with his famous phrase ‘‘carving nature at the joints’’. One sometimes
hears an item of faith: the relations that will ultimately be described in
successful science are in fact precisely the natural ones. But this seems
232  :   
almost empty! The argument seems to be that to understand what a theory

says we must interpret it as being about the natural relations, to assert that
it is true is to say that it is instantiated in natural relations, and so, it follows
at once that if science being ultimately entirely successful means being true,
then the relations it will describe will be natural. For this argument to be
logically compelling, ‘‘natural’’ cannot be independently meaningful; but if
it is not independently meaningful, the conclusion has no bite at all.
In any case, what cannot be meant here is that the relations that will
ultimately be described in structural terms are in fact precisely the natural
ones—unless one were to count ‘‘natural’’ as structural, more or less as
Carnap wanted to count ‘‘founded’’ as logical, and with no better reason.
Staying with Putnam: the Paradox dissolved

Putnam himself did not present the argument as a paradox, but as an
argument against what he called metaphysical realism. As I read him,
Putnam clearly signaled how the apparent paradox dissolves upon scrutiny.
The puzzle only begins with the reflection that Putnam established his
conclusion for theories in a very large class of languages, perhaps even all
languages (if that makes sense). The crucial next step is to add that our own
language—the very language in which we state his argument and develop
our scientific theories—is one of those. Infer then that his conclusion is
true of our own language. There you have it!
Trying to block this universal instantiation looks absurd, doesn’t it? But
the conclusion, that our ideal theories cannot be false, looks equally absurd.
So what can we say in retort? To our dismay it seems we have no recourse
but to say, with the metaphysicians: Putnam’s argument pertains only to
languages lacking the semantic glue to stick their words firmly to their
referents—so, our language must be different, our words must be somehow
guaranteed to refer to natural relations only. We seem to have no other
way out.
But that is not so. We should make a sharp distinction between how
Putnam’s argument applies to an arbitrary language under study, and how
we could apply it ourselves to our own language. An arbitrary language
under study is in need of an interpretation. If we can keep the pragmatic
dimension as fixed or given for the time being, that interpretation must
 ’     ’  233
certainly have at least as core an assignment of extensions to its singular and

general terms. But what is an assignment? The term ‘‘assignment’’ carries
the connotation of ‘‘something having been assigned by someone’’, but
Putnam’s argument actually just concerns functions.
The text of Putnam’s argument needs to be read very carefully. It
proceeds by means of instructions to us to do something, and includes
as implicature that we can do them. This may look like an inessential
feature of the argument, but actually shows precisely what we can resist.
We are apparently being led through a chain of reasoning destined to the
conclusion that we can regard the  as having a certain structure.
There exists a model of the ’s cardinality, there exists a function
that will map the one onto the other . . . Select one of each! Under what
conditions can we do that?
‘‘We can regard the  as having that structure.’’ That is an assertion
about us; and what is its basis? The basis is the argument that there exists
a certain function which maps the model and  one-to-one onto
each other. Fine, that function exists—but what does that have to do with
us? We have an interpretation for the given language only if we can define or
identify such a function. To do that we must be able to describe both the
function’s domain and its range, hence both the syntax of that language
and  , as well as the way in which the former gets mapped on to
the latter.
To illustrate the illegitimate slippage from ‘‘there exists’’ to ‘‘we have’,
let us look at two examples, one quite practical and the other abstract.
Here I have an apparently blank sheet of paper. It is the best sort of paper,
heavy bond, paper with character. Looking carefully at it, I see striations
and marks, hardly discernible but definitely there, and upon consideration
realize that they are as complex and manifold as the nighttime sky. There
exists a function that maps the main landmarks and streets of Paris more or
less continuously into these marks and striations, in a sufficiently detailed
way to rival even the best commercially available maps.7 So, this sheet is
in fact an exquisitely detailed map of the city of Paris: how much will you
pay me for it?
I assert the existence of this function. Its existence follows from some
simple but plausible assumptions about the small imperfections that all paper
made from wood must have.8 These assumptions are not nearly as abstruse
as those going into the proof of the Loewenheim–Skolem–Tarski–Vaught
234  :   
theorem. Yet you balk . . . You have a good point; you would not call this
sheet a map because in fact, although the function exists, it is not true that
you can regard this sheet as a map of Paris.
To illustrate the same problem and at the same time the ordinary,
acceptable use in which the slippage from ‘‘there exists’’ to ‘‘we have’’
causes no problems, take this simple problem in geometry:
The Euclidean sphere of radius 1 can be coordinatized as the set of
points (x,y,z) in R3 satisfying x2 + y2 + z2 = 1. Determine the distance
along great circles between two arbitrary points on the sphere as a
function of the coordinates.
What is asserted in the opening sentence? That there exists a suitable
mapping of the sphere into the set of triples of real numbers. The ‘‘can’’
in ‘‘can be coordinatized’’ has no literal significance, for how could we do
it? How could we select a point to be assigned (0,0,1) for example? In fact,
the sphere as a mathematical entity has perfect rotational symmetry, so any
description of any point on it applies equally to all points—unless it is a
description that relates some points to things outside the sphere, in which
case we are not considering the sphere by itself.9
Going back now to the context of Putnam’s argument we must similarly
conclude the following. As long as we are not given an independent
description of both the domain and range of an interpretation, we do
not have any such interpretation, nor any way to identify one. Given an
independent description of the interpretation’s domain and range, however,
whether the theory is true under that interpretation depends entirely on
how the mapping is defined, using those descriptions. Remember here
the Water Theory I gave above: if we do not have our own resources to
describe both this theory (its language and axioms) and the world, then we
can’t have an interpretation of it. But if we have those resources, then we
can insist that all admissible interpretations must assign water as referent
to the word ‘‘water’’—and upon that interpretation, the theory’s truth or
falsity is a contingent matter.
On my reading, Putnam spells this out quite clearly. In ‘‘Realism and
Reason’’, where the argument was first introduced, he defuses one point
with the laconic ‘‘because the world is not describable independently of
our description’’ (p. 496). And in ‘‘Two philosophical perspectives’’ he is
very explicit: ‘‘This simply states in mathematical language the intuitive
 ’     ’  235
fact that to single out a correspondence between two domains one needs
some independent access to both domains’’ (Putnam 1981: 74).
It is of no earthly relevance to know that there exists some such function
such that it, if only it could be described, would furnish an interpretation
under which the theory is true. The only question of interest for us, if we
do really have such an interpretation, is whether the theory is true under
that interpretation.
What happens if we want to apply Putnam’s argument to our own
theories formulated in our own language? We shall be able to grasp such
a theory if we can grasp an interpretation of the language in which
it is formulated (i.e. our language!). Well, as noted, we can grasp an
interpretation—i.e. function linking words to parts of  —only
if we can identify and describe that function. But we cannot do that unless
we can independently describe  . So Putnam’s model theoretic
argument, if applied to our own language, meets up with a dilemma:
(A) if we cannot describe the relevant elements of  , neither
can we describe/define/identify any function that assigns extensions
to our predicates in  ;
(B) if we can describe those elements of   then we can also
distinguish between right and wrong assignments of extensions to
our predicates in  .
For example, on alternative (B), we would insist that a right assignment of
an extension to ‘‘water’’ is water, and that all other assignments are wrong.
That the theory is true on some other such assignment is simply irrelevant
to whether it is true.
I take it that this is also how Putnam meant the apparent paradox to
be resolved: by noting the dilemma posed by (A) and (B). The response
is Wittgensteinian, in that it focuses on us, on our use of theories and
representations, and brings to light the impasses we reach when we abstract
obliviously from use to use-independent concepts. In terms of our prior
discussion, we must emphasize the crucial role of the indexical here. A
theory says nothing to us unless we can locate ourselves, in our own
language, with respect to its content.
11
An Empiricist Structuralism
The mechanization of the world-picture culminating in the seventeenth

century had as equally dramatic sequel the mathematization of the world-
picture that culminated in the twentieth.1 The former led quite reasonably
to a new conception of science itself, contrasting sharply with its conception
in the Aristotelian tradition. The latter transition has a rightful claim upon
us to develop views of science so as to do justice to its revolutionary impact
on the sciences themselves. It is this claim that I see honored in the century
long attempts to develop structuralist views of science, the recurrent defeats
of such views notwithstanding.2 Accordingly I shall argue here that the
defeats are not inevitable but that structuralism finds its proper articulation
only in an empiricist setting.
What could be an empiricist structuralism?

When structuralist views are expressed intuitively they come in such slogans
as ‘‘all that science describes is structure’’ or even ‘‘all that we can know
is structure’’. Such slogans are provocative, and meet immediately with
resistance. Doesn’t science tell us also what water really is, namely H2 O,
that when water is cooled it will first contract in volume and then at
a certain point begin to expand, that it will refract light and dissolve
sugar . . . ? None of this sounds at first blush like a description of the
structure of water but rather of what it is and what it is like. Such examples
are easily multiplied. By choosing water, I chose an example that is as
telling for the empiricist as for the scientific realist, though the former may
discount that phrase ‘‘what water really is’’. The latter, on the other hand,
will insist that similar examples of the unobservable abound: science tells
us what electric currents really are, what atoms are really like, how the
mass-energy distribution curves space, and so forth.
238  :   
At first sight, then, structuralist sentiments are greeted with an incredulous

stare. But in an empiricist setting, scientific theories are viewed precisely
as theories, not as our sole wherewithal for getting around in the world.
On the contrary, theories are artifacts, constructed to aid us in planning
and understanding—not to usurp those functions. Accordingly, the stories
about nature, about what things are like, which spell out a way the world
could possibly be like for such a theory to be true, take on a lesser role.
They allow us to move around in the theory, to exercise the imagination,
even to get to the point intellectually where we can draw qualitative
consequences via the theory without actual calculation.3
The two poles of scientific understanding, for the empiricist, are the
observable phenomena on the one hand and the theoretical models on
the other. The former are the target of scientific representation and
the latter its vehicle. But those theoretical models are abstract struc-
tures, even in the case of the practical sciences such as materials science,
geology, and biology—let alone in the advanced forms of physics. All
abstract structures are mathematical structures, in the contemporary sense
of ‘‘mathematical’’, which is not restricted to the traditional number-
oriented forms. And mathematical structures, as Weyl so emphatically
pointed out, are not distinguished beyond isomorphism—to know the
structure of a mathematical object is to know all there is to know. As
we have seen, this poses a problem for any structuralist view of scientific
knowledge.
Essential to an empiricist structuralism is the following core construal of the
slogan that all we know is structure:
I. Science represents the empirical phenomena as embeddable in certain
abstract structures (theoretical models).
II. Those abstract structures are describable only up to structural iso-
morphism.
In the empiricist version, the structuralist slogan is clearly and substantially
qualified. It does not take reliance on theory for us to know many things
about water—about observable phenomena in general—and so the slogan
must be read as meaning, at best, that all we know through science is structure.
Nor does the formulation as I. and II. imply that there is in nature, or
in the phenomena, a form/content or structure/quality distinction to be
drawn. The structuralism in ‘‘empiricist structuralism’’ refers solely to the
   239
thesis that all scientific representation is at heart mathematical. Empiricist

structuralism is a view not of what nature is like but of what science is.
This modesty leaves it squarely placed, however, in the range targeted by
Weyl’s paradox, Putnam’s paradox, Newman’s argument against Russell,
and Carnap’s puzzles over how science (as he conceived it) could be about
the phenomena at all. But its modesty does remove some of the problems.
I will now attempt to isolate the fundamental problem that was at the heart
of all these conundrums.
The fundamental remaining problem for a

structuralist view of science
Bertrand Russell and Rudolf Carnap are only two of the philosophers
who attempted to develop a structuralist conception of science. Some
later developments are explicitly so named (viz. Stegmueller’s ‘structuralist
view of theories’) while others did not adopt the name, but are clearly
of the same ilk (e.g. the ‘semantic view’ of theories). Starting with John
Worrall’s work in the mid-Eighties, Structural Realism made great inroads
into philosophy of science at many levels.4 Each of these is implicitly (and
sometimes explicitly) challenged through the puzzles and paradoxes that
we saw for the earlier advocates.5
As will have been apparent, I see the correct response to those challenges
as emerging with proper clarity only in the discussion surrounding Putnam’s
‘model-theoretic argument’. The form of this response is not linked differ-
ently to the different forms of structuralism. Bertrand Russell’s response to
Newman was not totally clear, and could be understood as simply accepting
Newman’s invitation to postulate a distinction that separates relations into
important and unimportant. But Russell, like Carnap, insisted on linking that
separation to (one’s own) experience, which is—as I see it—a definite step
toward the crucial clue of self-reference. Only in the discussion of Putnam’s
Paradox did this role of indexicality become entirely clear. Attention to the
indexical, which we found to be indispensable already in the use of maps,
models, and theories at an early point in reflection on science and practice,
is crucial to defuse the challenge from Newman’s argument and its variants.
What is needed to complete the task is to grasp a fundamental underlying
problem about representation. It is not one to be stated in just a few words
240  :   
if we are to feel its proper impact, but we can bring it out properly
in successive stages. To do so, I will begin by revisiting Reichenbach’s
problem of coordination in a more abstract form.
How can an abstract entity represent?

The empiricist structuralism formulated in theses I. and II. above looks pretty
well like precisely the view we saw Weyl as expressing, and for him it led
at once to what I called ‘‘Weyl’s paradox’’. The basic perplexity emerges
in its purest form when we ask just how Principle I. is to be understood.
What does it mean, to embed the phenomena in an abstract structure? Or
to represent them by doing so?6 We can understand how one mathematical
structure can be embedded in another—the word ‘‘embedding’’ refers here
to a certain kind of function. So Principle I. can mean this: that a scientific
theory will first of all represent the phenomena by means of mathematical
structures, and then show how those structures fit into larger ones, the
theoretical models.
So far so good—that relates well to the discussion we had earlier of
data models, surface models, and theoretical models. But it presupposes
that we understand the first part of this two-part process: representation
of the phenomena, which are concrete, by mathematical structures, which
are abstract. Hence the most fundamental question is this:
How can an abstract entity, such as a mathematical structure, represent
something that is not abstract, something in nature?
The perplexities appeared clearly in the passages, that we already discussed
in Part Two, from Reichenbach’s 1920 discussion of coordination. Recall
how perplexed Reichenbach sounded:
The mathematical object of knowledge is uniquely determined by the axioms and
definitions of mathematics. (Reichenbach 1920/1965: 34)
The physical object cannot be determined by axioms and definitions. It is a thing
of the real world, not an object of the logical world of mathematics. Offhand it
looks as if the method of representing physical events by mathematical equations
is the same as that of mathematics. Physics has developed the method of defining
one magnitude in terms of others by relating them to more and more general
magnitudes and by ultimately arriving at ‘‘axioms’’, that is, the fundamental
equations of physics. Yet what is obtained in this fashion is just a system of
   241
mathematical relations. What is lacking in such system is a statement regarding the

significance of physics, the assertion that the system of equations is true for reality.
(Ibid.: 36)
Indeed, a theory in mathematical physics is, if just taken in and by itself, a

mathematically formulated theory, in just the way that e.g. Euclidean and
hyperbolic geometry are. So taken, it is only a theory about mathematical
objects. How could it have more content, to make it something different
from pure mathematics?
No use, it would seem, to just add an extra axiom such as ‘‘the above
axioms are true of reality’’. At least at first sight, that would be ‘‘just more
theory’’, as Putnam would say, and the question Why is this more than
just mathematics? applies—or would seem to apply—equally to the newly
extended theory. If words are all we have . . . how can we answer this
at all?
The ‘offhand’ realist response

What did Reichenbach mean with his above remark ‘‘Offhand it looks as
if the method of representing physical events by mathematical equations is
the same as that of mathematics’’? He imagines in effect the following naïve
reply: what is called for is simply a function, a mapping, between mathematical
objects and physical objects or processes—so what is puzzling about that?
True, we have no difficulty with mappings between two sets of math-
ematical objects. But notice: to define such a mapping we must identify
its domain and its range, plus the relation that effects the coordination
between them.
For example, if two sets of points are given, we establish a correspondence between
them by coördinating to every point of one set a point of the other set. For this
purpose, the elements of each set must be defined; that is, for each element there
must exist another definition in addition to that which determines the coördination
to the other set. Such definitions are lacking on one side of the coördination dealing
with the cognition of reality. Although the equations, that is, the conceptual side
of the coördination, are uniquely defined, the ‘‘real’’ is not. (Ibid.: 37)7
So here is the problem baldly stated. If the target is not a mathematical

object then we do not have a well-defined range for the function, so how
can we speak of an embedding or isomorphism or homomorphism or
whatever between that target and some mathematical object?8
242  :   
Vacuity of the ‘offhand’ realist response

The natural temptation is simply to impose a parallel vocabulary and
declare victory. To illustrate this earlier on, recall, I took Reichenbach’s
own example of Boyle’s Law, PV = rT. To give physical meaning to this
equation its terms must be co-ordinated with physical quantities. But what
is, for example, the temperature of a gas? It changes with time, and differs
from one body of gas to another, so isn’t it a function that maps times and
bodies of gas into the set of real numbers?
This suggests that we could answer the question of how ‘‘PV = rT’’ is
more than just a mathematical theory by asserting that it ‘‘mirrors reality’’.
Suppose we just answer with ‘‘There exist physical quantities P*, V*, and
T* which pertain to bodies of gas, and when ‘P’, ‘V’, and ‘T’ refer to these
respectively then ‘PV = rT’ is true’’.
If this is added as a postulate to the theory, we can only be mystified as
to what wheels it turns at all! Although verbally apt, all this does is to create
a parallel vocabulary and declare victory—but it is an empty victory.
The metaphysical realist’s response depicts nature as itself a relational
structure in precisely the same way that a mathematical object is a structure.
On this view, if the mathematical model represents reality, it does so in
the sense that it is a picture or copy—selective at best, but accurate within its
selectivity—of the structure that is there. As one might say then,
there is no problem, because this depicted physical system is a set of
parts connected by a specific family of relations—so of course there
are functions defined on this set of parts with range in the model and
vice versa.
A little uncharitably we could summarize this as: a physical system is
an abstract entity with concrete parts or elements—uncharitably, since
a physical system is meant to be something physical, not something
abstract.9
But a function that relates A and B must have a set as its domain. If A
is e.g. a thunderstorm or a cloud chamber—a physical process, event, or
object—then A is not a set. Fine, the realist can answer, but A has parts,
and the function’s domain is the set of these parts. No reason why the
elements of the set need be mathematical entities after all. Moreover there
are specific relations between these parts; and these relations have as their
extensions sets of sequences in that domain. The function provides a proper
   243
matching provided the images of these relations are relevant relations in

the model.
(*) the function relates two structured sets, call them S(A) and B:
S(A) = SA1, SA2. where:
SA1 = the set of parts of A,
SA2 = the family of sets which are extensions of relations
on these parts
B is a mathematical object, representable correspondingly in the same
general form Set, Relations.
This follows the most general format of standard mathematical modeling:

the format of relational systems, the subject for what Whitehead called
Universal Algebra.10
What is taken for granted here is the relation between A, the physical
entity, and S(A), a relevant mathematical representation of A. But why
is the relation between A and S(A) any different from the one we asked
the question about, namely the relation between A and B? Isn’t S(A) an
abstract entity used to represent A? We cannot very well answer ‘‘how can
an abstract mathematical structure represent a concrete physical entity?’’ by saying
‘‘this is possible if we assume the latter is represented by some other mathematical
object’’.
Let us not be hasty. First of all, the realist can reply, there is no question
but that the sets S(A), SA1, SA2 are real if A is real. When we say that B
is an adequate representation of A we mean simply that S(A) is isomorphic
to or embeddable in or homomorphically mappable into B. So far, so
good—we can agree to all that (modulo whatever view of mathematics we
have), this use of an elementary part of set theory must be legitimate. But
the realist’s next problem is that this sounds like there is more uniqueness
than there is—that there is less selectivity in what S(A) can be than there
must in fact be.11 The respect in which B represents A may be one thing
or another: that depends precisely on how we ‘divide up’ this entity A and
which aspects of its relational structure we select—i.e. what we choose for
SA1 and SA2. What indicates here the relationship between A and S(A)
is so far no more than our description or denotation of SA1 as ‘‘the set
of parts of A’’, etc. The word ‘‘the’’ is however not justified here, it is a
misnomer, given that A can be divided up in various ways.
244  :   
And just what is this dividing up? It is nothing more nor less than the act
of representing A as having SA1 as its set of parts—i.e. of A as consisting of the
members of SA1.
Would all be well for the realist response (which thinks of the coordin-
ation as a simple two-place relation between the two items) if there
were only a single, unique way of representing A in this manner? The
question is moot. For, as a matter of fact, and as we have seen before,
as long as A can be represented as having a set of parts of the same
size as that of B, some such relationship as (*) above will certainly exist
between A and B! That is precisely the point that (by now, famously)
trivialized Russell’s structuralism and drives Putnam’s model theoretic
argument.
The realist has one final gambit, namely the one that we already saw
in David Lewis’s introduction of natural classes. S/he insists that there is
an essentially unique privileged way of representing: ‘‘carving nature at the
joints’’. There is an objective distinction ‘in nature’ between on the one
hand arbitrary or gerrymandered and on the other hand natural divisions of
A into a set of parts, and similarly between arbitrary and natural relations
between those parts. So A is already a highly structured entity, and the
abstract relational system we need is a precise copy thereof. We could now
ask about this ‘copy’ relation; but there seems no need to press that. For
what precisely is this gambit, this insistent assertion?
It is a postulate (if meaningful at all). What can it possibly mean? The
word ‘‘natural’’, its crucial term, derives its meaning solely from the role
this postulate plays in completing the realist response. Honi soit qui mal
y pense they’ll tell us.12 Unless we subscribe to this sort of metaphysics,
and something like a substantial correspondence theory of truth and
representation, the realist maintains, we cannot make sense of our own
practice.13 But their point will be moot, and the postulate superfluous, if
the problem dissolves upon scrutiny, as I shall now argue.
The two main dangers for an empiricist

Let us look at the problem afresh, without reference to how a scientific or
metaphysical realist might approach it. The problem of coordination, taken
in its most abstract form, is the problem of understanding what makes for
   245
a representation, what the relation can be between a representation and

what it represents. There is an ambiguity to be avoided, which we looked
at briefly in the Part I. A painting of an angel has representational content: it
depicts an angel. Whether it also has a referent, whether there is something
real that it depicts, is another question: that depends in part on whether
angels exist. What is at issue in the problem of coordination is the relation
between a representation and its referent.
Under this heading we can address how such a specific item as a geodesic
in a mathematical space relates to something physical—but that makes sense
only in a context already replete with theory and previously established
language or other form of representation. The philosophical impulse to
consider a problem in its purest possible form must be respected too, and
so we see the problem successively generalized—but this can go too far,
possibly to the point of losing sense in the end.
There is clearly a practice of representing by means of abstract, and not
just concrete, artifacts. This practice we can examine, to see just what
mappings, embeddings, or structural comparisons are involved in scientific
representation. If these, however, remain on an abstract or mathematical
level, then we still have to face the problem of how this practice can relate
to the phenomena.
For familiar natural phenomena such as daybreak and starlight, electric
light or electromagnetic effects, science provides us with models. Often
the initially startling event is that a model, of a sort we have been trusting,
does not save them. The question how an abstract structure can represent
something, transposed to such a context, is just this: how, or in what sense,
can such an abstract entity as a model ‘‘save’’ or fail to ‘‘save’’ this concrete
phenomenon? What is the pertinent relation that holds or does not hold
between the mathematical structure described by our equations and that
natural or artificially produced process?
There are really two cases here, hidden by my easy use of ‘‘us’’
and ‘‘our’’. Consider two scientists, one who offers a model for how
the frog population density varied in the Netherlands in the twentieth
century, and another who offers a quite similar model for the dinosaur
population density in a larger area at a much earlier time. For their models
to be adequate—hence, for their theories to be true—those observable
phenomena must be embeddable in their models. But the cases are different,
for the first scientist can draw on actual measurement results of the relevant
246  :   
population density, and the second cannot. So for the first we face the
question of what the adequacy of a model—the conditions under which
it accurately represents its domain—amounts to in a concrete practical
setting. For the second scientist we have to understand how a phenomenon
somewhere and somewhen, which is not encountered in human experience
or targeted in actual measurements or observations, can be said to ‘‘fit’’ a
theoretical model. There are of course many examples of the latter sort,
some much more recondite or abstruse than dinosaurs—think for example
of extra-galactic phenomena, even ones that a theory classifies as beyond
our ‘event-horizon’ so that we are precluded from ever gathering even
very indirect evidence about them.
Metaphysical postulates of the sort we just discussed above, concerning
‘structure in nature’, seem to me to be specially inspired by the latter case,
of the phenomena that are not encountered in our practice. On the one
hand, theories have many models that are never actually constructed by
any one—in the sense that a mathematician can say that a certain equation
has many solutions that are never actually written down by anyone. On
the other hand, those phenomena are never actually represented, in any
concrete way by anyone. With human participation entirely foreign to the
context, what can constitute adequacy or truth except a direct theory-model
to nature relation? How can we assert that those unknown phenomena
fit models of our theories, except in a sense that implies the metaphysical
realist’s postulation of ‘structure in nature’, consisting of universals or the
like? That rhetorical question present the first of the main dangers for the
empiricist that I want to take up here.
The second danger to the empiricist, implicit in discussion of how theor-
ies or theoretical models can represent phenomena when those phenomena
have actually been subject to measurement, is subtler. There the pertinent
‘matching’ seems so obvious and clear: the theoretical model and the data
model are both abstract structures, and it is not difficult to understand how
the latter may be embeddable in the former. But a theoretician at a desk,
matching his theories’ equations to graphs or density functions or the like,
is not someone directly in touch with nature. So doesn’t a reflection that
focuses on the data models for assessing empirical adequacy, lose contact
with reality altogether?
   247
Phenomena far outside experience

Theories represent the phenomena just in case their models, in some sense,
‘‘share the same structure’’ with those phenomena—that, in slogan form,
is what is called the semantic view of theories. My own variant upon this
theme is that the phenomena are, from a theoretical point of view, small,
arbitrary, and chaotic—even nasty, brutish, and short, one might say—but
can be understood as embeddable in beautifully simple but much larger
mathematical models. Embedding, that means displaying an isomorphism
to selected parts of those models. Here is the argument to present the
first challenge. For a phenomenon to be embeddable in a model, that
means that it is isomorphic to a part of that model. So the two, the
phenomenon and the relevant model part must have the same structure.
Therefore, the phenomenon must have a structure, and this shared structure
is obviously not itself a physical, concrete individual—so what is implied
here is something of the order of realism about universals.
What are we to make of this argument? As I see it, it is closely related
to similar arguments that convict us of realism about universals or of
postulating ‘truth-makers’ of some sort, as soon as we talk of truth. The
plausibility of the argument comes from our having to speak in a general
way when abstracting from particular cases. But the argument loses its
plausibility if we just look carefully at a particular case, and then reflect that
what we say about the general case is simply about all the particular cases
in a summary fashion.
Let’s begin with a simple example: during a period of optimal conditions
a certain bacteria colony grows exponentially, doubling in size in equal
intervals of time. The equation for exponential growth is N(t) = N(0)ekt ,
with k a constant reflecting the doubling time. It is easy to calculate the
doubling time τ: 2 = ekτ , so τ = ln(2)/k.
The solutions for this equation are specific functions defined on some
interval (t, t ) for specific values of N(0) and k, with range in the real
numbers. This is an example of a very simple theory (the equation) and its
models (the solutions of that equation). But the growth is discontinuous:
even bacteria do not reproduce continuously in the mathematical sense. So
the ‘empirical substructures’ of these models are the sequences N(0), N(τ),
N(2τ), . . . . .
248  :   
Now we consider a real phenomenon: the actual number of bacteria in a

certain colony located in Antarctica in a certain interval of 12 hours several
million years before humanity emerged on earth. Its growth during that
interval: does it fit one of these models? What does that mean?
At the beginning of those 12 hours there was a certain number N (0) of
bacteria in the colony and at each time t = τ, 2τ, . . . within those hours a
number N (τ). Is the function N actually one of the empirical substructures
of one of those models? Yes or no? If yes, then the phenomena fit that
model, and hence the theory, in the sense required.
In this discussion there was at no time an implied assertion that there
exists a structure, in the sense of a universal or similar abstract entity, that
the phenomenon instantiates, in addition to the phenomenon itself. Or let
us put it this way: there was no such implication, unless all predicative
statements have such implications.
We are here at a point of contact with our discussion above of the
notion of truth. If all predicative statements had such implications, that
would mean, for example, that ‘‘Snow is white’’ is true only if in addition
to snow there exists also a universal, whiteness, to which snow bears a certain
relation. I take it that is not the case: ‘‘Snow is white’’ implies the existence
of nothing at all other than snow, and similarly, the statement that there
were at time t 500 bacteria in this colony implies the existence of nothing
at all beyond the bacteria and the colony.14
Rejection of the correspondence theory of truth may be familiar from
much other philosophical literature, but a few—even cursory—remarks
may still help to underline this point. ‘‘Snow is white’’ is true. Does this
imply that besides snow, there is also whiteness? The case for Yes! begins
with the remark that snow cannot by itself make the statement true, because
the existence of snow does not imply that snow is white . . . snow could
have existed and have had some other color. So (?) there must be something
in addition to snow, whose ‘cooperation’ with the also needed snow will
make the statement true. I am carrying coals to Newcastle by responding
to this (Quine’s ‘‘On what there is’’ appeared long ago): all that is needed
for ‘‘Snow is white’’ to be true is that snow be white, and that does not
imply the existence of anything but snow.
But then, the objection goes, what about sentences that are not particular
but general, like ‘‘All statements made by Francis Bacon just after eating
oatmeal were true’’? While in the above particular example, the use of
   249
‘‘true’’ succeeds only in echoing the statement to which it is attributed,

so that it is dispensable, that is not the case with such a general form.
Correct; but nothing more is needed in nature for the general statement to
be true except all that is needed for each of the particular statements, made
by Bacon just after eating oatmeal, to be true. And none of those needed
metaphysical underpinning, unless they were metaphysical themselves. (We
needn’t deny that metaphysical statements need metaphysical grounds, but
that is not the fault of truth talk.) There is a lot more said in the discussions
about truth, but I’ll leave it to the literature; the main point suffices
here.15
The same point applies to isomorphism. Let’s take a particular case of
a concrete physical, observable object: this table top is metrically isomorphic
to a Euclidean square. That is true, but simply because this table top is
square—c’est tout! It is true because the top’s sides are of equal length and
the angles between them are right angles. It could be paraphrased as ‘‘the
table top instantiates the Euclidean square Form’’, but the cash value of the
assertion carries no metaphysical commitment: it is just that the table top is
square.
When we want to discuss something of greater generality, such as a
theory’s empirical adequacy, we must speak of all phenomena of a certain
sort and how they relate to mathematical models. But (just as for the
example of Francis Bacon’s statements) nothing more is needed for the
general statement than whatever is needed for the particular cases falling
under it. And that does not include ‘structure in nature’ in a sense that goes
beyond the simple point that some table tops are square, some birds are
fast, some battles go to the strong, and so forth.
Another question remains, though, about representation. There is no
such thing as representation apart from or independent of our practice.
So how can we say something like: this theory accurately represents that
bacterial growth phenomenon (in those 12 hours long, long ago, and far, far
away) although the relevant model was never constructed and the bacterial
colony was certainly not encountered in human practice?16
The structural relationship between the model in question and the
phenomenon, that we just described, does not suffice to make the model
a representation of the phenomenon. This was amply argued in our
initial discussion of representation. If a model were offered to represent
the phenomenon, that structural relation would determine whether the
250  :   
model was adequate with respect to its purpose, for example, whether
the representation was accurate with respect to the rate of growth in
volume or in numbers, as measured by this sort of clock or that, . . . . By
hypothesis, no model for this particular phenomenon was ever offered, so
the point is moot. Yet we would like to say that if the equation does have
such a solution—equivalently, if the theory has such a model—then that
(equation, theory) does correctly represent that phenomenon.
It seems to me that the only points to be made in response to this are
verbal. If the theory is offered, that amounts to the offer of a range of
structures—the structures we call models of the theory—as candidates for
the representation of the phenomena in its domain. If this range contains a
candidate that would satisfy the structural constraint—if the phenomenon
is embeddable in it, understood in the innocuous sense explained by the
above examples—then the theory is empirically adequate (and indeed true,
if the domain contains only observable phenomena).
Perhaps it would have been better if the word ‘‘model’’ had not been
adopted by logicians to apply to structures never offered in practice. For
undoubtedly, in many contexts, something is called a model only if it
is a representation, and the sense in which any solution of an equation
is a model of the theory expressed by that equation certainly does not
have that meaning. But it is too late to regiment our language so as to
correct that, and we will just need to be sensitive to usage in different
contexts.
The problem in concrete setting

We turn now to the second case: the phenomena are encountered in
practice, measurements are made, data models constructed, and so forth.
How exactly are theories confronted with the phenomena given by
measurement and experimental results in field work or laboratory practice?
Ronald Giere and Paul Teller have both discussed this with reference to
detailed accounts of empirical research.17 I will also take a specific example,
to discuss here in terms we set out before.
The observable phenomenon makes its appearance to us first of all in the
outcome of a specific measurement, or large set of such measurements—or
at slight remove, in a data model constructed from these individual out-
comes, or at a slightly further remove yet, in the surface model constructed
by extrapolating the patterns in the data model to something finer than our
   251
instruments can register. In such a scientific inquiry, from the gathering of

raw data to the achievement of something that can properly confront the
pertinent theory, we see moves toward increasingly abstract representation.
Well before the theorist displays his or her wares, experimentalists are
already constructing models of the data to convey the outcomes of their
inquiries. Describing the situation in this way, three ‘stages’ are displayed:
the observable phenomena subjected to measurement, the appearances
they present in progressively more abstract ‘outcomes’, and finally the
theory whose theoretical models are accountable to those—already quite
abstract—deliverances of scientific inquiry.
Young’s 1802 double slit experiments constituted, together with Fresnel’s
‘‘light spot’’ in a shadow, the crucial phenomena that Newton’s particle
models for light could not ‘save’. That was their significance at the time;
now we see much more in them, for there are more sensitive versions of
Young’s experimental design. Specifically, somewhat more than a hundred
years later, once the photo electric effect was discovered and its quantum
character appreciated, such diffraction experiments needed to be revisited
and reinterpreted.
The reinterpretation focuses on a variation, not performable or even
contemplated in Young’s time, in which the source is so slow that only
a single black spot appears on the screen during non-negligibly large time
intervals. (Experiments in which individual photons were isolated could
actually not be performed till the last quarter of the twentieth century.)
If carried out in this form, the interference pattern will still appear, but
only gradually. Similarly of course for each of the two separate single slit
experiments resulting with either of the slits closed.
So let us imagine what the data could be. Think of the screen divided
into N small areas, indexed as x = 1, 2, 3, . . . and consider for each the
proportion of hits in that area after the first n hits on the screen. Then the
data gathered are recorded in three relative frequency distributions rel(n,x),
relA (n,x), relB (n,x). For example, relA (50,17) denotes the proportion of
hits, among the first 50 hits, that occurred in area 17 when only slit A was
open.
This summary is the outcome of the specifically carried out experiment,
though ‘policed’ since the recording may have been subject to noise or
random errors or inaccuracies; it is the data model. But it is certainly not
yet what any theory about the phenomenon must confront. A ‘‘curve
252  :   
smoothing’’ program replaces the three relative frequency functions with

three probability functions p, pA, pB, the result being what I prefer
to call a surface model. (This involves an empirical hypothesis—future
repetitions of the experiment could conceivably yield a bad fit with these
functions.)
The surface model is p, pA, pB
Theoretical model T1: p is a mixture of pA and pB. Empirically refuted.
Theoretical model T2. p, pA, pB are determined by a geometric
probability model, in the way of quantum mechanics. This works.
Our diagnosis of this procedure: the theory to phenomena relation displayed
here is an embedding of one mathematical structure in another one. For the
data model—or, more accurately, the surface model—which represents
the appearances, is itself a mathematical structure.
So there is indeed a ‘matching’ of structures involved; but is a ‘matching’
of two mathematical structures, namely the theoretical model and the data
model. At this point, we can be sure, the metaphysical realist will ask for the
next step: so then what is the relation between data model and the phenomena it
models? But the dialectical situation is different now that we are in a concrete
setting, different from the bacteria colony in the Antarctica example: here
the measurements are really made and the modeling is explicitly a case of
representation.
The form of the metaphysician’s question insinuates a substantive pre-
supposition: that there is a relation between data model and phenomena,
which determines whether the data model represents the phenomena, and
which has nothing to do with anything but the two of them. Once again
we find ourselves with an idea akin to, of a piece with, the correspond-
ence theory of truth, the idea that there is a user-independent relationship
between words and things that determines whether a sentence is true or
false. Such an idea cannot be carried through without postulating a good
deal of ontological flora and fauna beyond concrete individuals. But we
have discussed this issue sufficiently above, we don’t need to repeat the
argument against such presuppositions.
What we need to ask ourselves now, though, is precisely how can we
phrase the question we had originally, about how an abstract entity can
represent something physical, so as to make sense of it in this concrete
context.
   253
The problem in concrete setting revisited and

dissolved
How can we answer the question of how a theory or model relates to the
phenomena by pointing to a relation between theoretical and data models,
both of them abstract entities? The answer has to be that the data model
represent the phenomena; but why does that not just push the problem
one step back? The short answer is this:
construction of a data model is precisely the selective relevant depiction
of the phenomena by the user of the theory required for the possibility
of representation of the phenomenon.
But this answer needs to be filled out
As we have just seen, when a theoretical model is said to represent certain
phenomena, there is indeed reference to a matching, namely between parts
of the theoretical models and the relevant data models—both of them
abstract entities. Note now, the crucial word in this sentence: the punch
comes in the word ‘‘relevant’’. There is nothing in an abstract structure itself
that can determine that it is the relevant data model, to be matched by the theory. A
particular data model is relevant because it was constructed on the basis of
results gathered in a certain way, selected by specific criteria of relevance,
on certain occasions, in a practical experimental or observational setting,
designed for that purpose.18
Looking into the model by itself we will find nothing to imply that it is
or is not a representation of e.g. a bacterial colony or tides in a sea inlet. (In
the same way there is nothing in a printed map taken in itself that can make
it a map of Austin, TX instead of, say, a kabalistic charm.)19 Rather, the
relation to be identified is not a 2-place relation between data model and
phenomenon but a 3-place relation that involves the user. Representation
is a relation between the abstract structure and the phenomena constituted
by the user. Nothing represents anything except in the sense of being used
or taken to do that job or play that role for us.
We can also reach this conclusion by a different route, thinking first of
the phenomena. Psillos writes quite correctly:
Ergo, the structure of a domain is a relative notion. It depends, and varies with, the
properties and relations that characterise the domain. Differently put, a domain has
no inherent structure, unless some properties and relations are imposed on it. Or,
254  :   
two classes A and B may be structured by relations R and R respectively in such a

way that they are isomorphic, but they may be also structured by relations Q and
Q in such a way that they are not isomorphic. (Psillos 2006: 562–3)
That is, the phenomenon, what it is like taken by itself, does not determine
which structures are data models for it—that depends on our selective
attention to the phenomenon, and our decisions in attending to certain
aspects, to represent them in certain ways and to a certain extent.20,21
Have we now lost or sidestepped Reichenbach’s problem, or have we
rather landed in it ourselves? Have we in effect succumbed to a post modern
il n’y a pas de ‘hors-texte’? No, but the way to see that we have not ‘lost the
world’ will explain why the sea-change I propose is of the sort that tends
to be described as a Wittgensteinian move.
An example: the metaphysician meets practice

To understand how the above maneuver does not simply push Reichen-
bach’s question one step back, we have to think of the assertions made in
context, by the scientists involved. Something will come to light if we listen
to the assertions that such a scientist would initially make in the first person.
Let us take a very simple example that puts the discussion into first-person
discourse, by imagining ourselves in the scientist’s role. The example I’ll
choose will accordingly not require the agent’s being very advanced in
a particular science. Best to have the theory in question deliver models
no more complicated than the data models engendered by an empirical
inquiry.
Imagine a township council meeting where I have represented the deer
population growth in Princeton by means of a graph. We have been
discussing a theory, call it T, about the Princeton environment for deer,
including luscious gardens as well as the council’s culling instinct, its
tendency to experiment with birth-control measures for the local fauna,
and the like. I point out that theory T provides models that fit very well
with the structure displayed in the graphical representation—call it S—that
I have just presented. Now someone notices that I am pointing to a match
between two abstract mathematical structures (a model of T and a graph), and
expresses doubts about whether we are staying in touch with reality:
: Yes, T fits well with this graph, your representation S,
but does T fit the actual deer population growth in Princeton?
   255
Well, this is Princeton. Still, to begin at least I do not realize that the
questioner is a metaphysician. So to begin I take this as a legitimate
request for more information about how I arrived at the graph. I describe
the measurements that sampled values of various parameters over time,
how we had set up stations for deer observation, the procedures fol-
lowed and the precautions taken against sampling bias. The challenger
responds:
Yes, I understand, I can see that you carried out those procedures
diligently and responsibly, and that the outcomes are summarized
properly in your graph. But the question remains: theory T fits the
summarized outcomes of your measurement procedures, but does T
fit the actual deer population growth in Princeton?
Now I am beginning to see that the question relates to a rather more
foundational worry, and I think that perhaps I can guess what it is. So
I explain that I did take guidance from the theory, that the parameters
measured were precisely those that T treats as relevant to population growth
in this sort of case. I add that although the measurement procedures were
mostly standard in this scientific area of inquiry, it was also in light of
T that I took for granted in some cases that values obtained for some
parameters gave information about the values of certain parameters less
accessible to direct measurement. But I add that none of this biased the
inquiry so as to guarantee that the results would be in accordance with
T. The measurement outcomes could have been quite different, while the
actual outcomes were, as I just showed, in accordance with T.
To the challenger this signals a complete misunderstanding of his concern.
I took his worry to be about a problem in epistemology! But his interest
is in what is really the case, and how we can understand how things really
are, independently of our knowledge and of any procedures followed to
gain knowledge. So he tries to convey that to me:
I understand that in this case your claim to knowledge about the
deer population growth in Princeton is warranted. But there is still
the real deer population growth, which is something in the world,
distinct from anything in your graph, distinct from anything in the
content of your warranted knowledge claim, distinct from the object
of knowledge that you have constituted in your practice (put it how
256  :   
you will)—and that, the real deer population growth, is what we

want theory T to match!
Although I can see the logical leeway on which he trades there is no leeway
for me in this context, short of withdrawing my graph altogether. Since this
is my representation of the deer population growth, there is for me no
difference between the question whether T fits the graph and the question
whether T fits the deer population growth. If I were to opt for a denial or
even a doubt, though without withdrawing my graph, I would in effect be
offering a reply of form:
The deer population growth in Princeton is thus or so, but the
sentence ‘‘The deer population growth in Princeton is thus or so’’ is
not true, for all I know or believe.
The first conjunct would be the content of the graph which I am presenting
as representation of the deer population growth, and the second conjunct
would express my doubt that this graph does correctly represent the deer
population growth.
Such a response would be as paradoxical as any of Moore’s Paradox
forms, like ‘‘It isn’t so, but I believe that it is’’ or ‘‘It is so, but I do
not believe that it is’’. In fact, I will have become incoherent if I let this
challenge lead me into any such concession.
Notice here that, unlike me, the challenger would not be (and is not)
inconsistent or incoherent. He may be doubting that my data model is a
good one, that it correctly and accurately represents the deer population
growth. For him that is logically independent of whether the graph was
correctly constructed from results to be taken seriously as veridical. But
that my data model does have that status is precisely the content of my
assertion when I present the graph with the words ‘‘here you see the deer
population growth in Princeton’’.
His challenge does not offer any evidence to the contrary. But what if I,
disturbed by his question, take it as an injunction to doubt my own memory
or competence, and I take that seriously? I will not have identified his
metaphysical concern, but heard him as speaking in this scientific context,
where the task of coordination is past. (The theoretical terms are already
closely connected with the practical procedures that define the relevant
   257
kinds of measurement.) If I do, I will first of all be bracketing, taking back,

putting on hold, my presentation of the graph.
What I cannot do is to both present the graph as representing something
and say that perhaps it doesn’t represent that at all. This example brings out
the crucial point:
in the chain [theory]–[data model]–[reality] the last link is one that is

expressed in indexical judgments.
The assertion that a given graph represents the phenomenon I have been
observing and measuring is on a par with the assertion that a certain spot on
a map is the point where I am. The conditions of correctness in both cases
pertain to the use to which the structure—whether abstract or physical—is
being put, and how. When I presented the graph to the audience interested
in deer population growth in their borough, I was saying ‘‘we are here’’ in
a logical space of possible growth patterns. And as Kant explained—as we
discussed in Part I—the ability to self-attribute a position with respect to
the representation is the condition of possibility of use of that representation.
The ‘link’ to reality

When Newton claimed that his theory of gravitation fit the phenomena,
he meant in part that his equations entailed (under certain simplifying
assumptions) Kepler’s laws of planetary motion.22 The latter he took as
descriptions of those phenomena, the planetary motions. What Kepler’s
laws gave him was in effect what I have called a surface model, a structure
constructed from data painstakingly amassed by astronomical observations.
The matching Newton demonstrated was therefore between mathematical
structures: between a substructure of his model of the solar system and that
Keplerian structure.
As long as we restrict our attention to this process, we are leaving aside
the final question: but how does the last structure taken into account relate
to what is meant to be the target of all this representing?
The answer, brought to light already in our initial discussion of maps
and models in use, is that at this last step we come to what can only be
expressed in indexical judgments. We can use a map to get around the
region in which we find ourselves if and only if that map is the thing we
258  :   
use and locate ourselves in to represent the pertinent features of that region.
This point transposes to the use of any model, and pertains specifically to
the assessment of whether a theory succeeds in correctly representing what
it is meant to represent. Take any such assertion as
The exponential function represents in smoothed summary form the growth of
the investigated bacterial colony.
Despite appearances, that is an indexical statement: the phrase ‘‘the invest-
igated bacterial colony’’ receives its reference from the context in which
it is used. The assertion is not made true by anything that can be formally
described within semantics, understood as limited to word-thing relations,
ignoring the role of the user and context of use.23
Nor does the phenomenon, what it is like, taken by itself, determine
which structures are data models for it. That depends on our selective
attention to the phenomenon, and our decisions in attending to certain
aspects, to represent them in certain ways and to a certain extent.24 So once
again we arrive at the conclusion: there is nothing useful to be found in
2-place structure-phenomenon relations alone when we try to understand
representation. Anything we see by way of such relations is something
abstracted from the 3-place relation of use of something by someone to represent
something as thus or so.
The ‘Loss of Reality’ objection dissolved

Perhaps, by now, the second danger for the empiricist in this context has
been entirely and successfully evaded. I hope so. But at the risk of sounding
completely de trop let’s see the challenge in its most strident guise:
Oh, so you say that the only ‘matching’ is between data models
and theoretical models. Hence the theory does not confront the
observable phenomena, those things, events, and processes out there,
but only certain representations of them. Empirical adequacy is not
adequacy to the phenomena pure and simple, but to the phenomena
as described!
An empiricist account of what the sciences are all about must absolutely
answer this objection.25 Let us therefore honor it with a special name: the
Loss of Reality Objection.
   259
The empiricist reply must be, in effect, the step that leaves the entire
game of metaphysics behind, and frees us forever from its illusionary charm
and glamour. But just because it is the step out of that so insidiously
enchanted forest into realistic common sense, it will have to be a very
simple one. On the one hand, we must immediately admit that:
the claim that the theory is adequate to the phenomena is not the same
as the claim that it is adequate to the phenomena as represented by
someone (nor as represented by everyone, or anyone).
After all the phenomena are actual objects, events, and processes, while
the representations that we or the scientific community construct of them
are the products of our independent intellectual activity. Yet on the other
hand it is clear that
if we try to check a claim of adequacy, then we will compare one
representation or description with another —namely, the theoretical model
and the data model.
What are we to make of these two points, taken together?
For us the claims
(A) that the theory is adequate to the phenomena and the claim
(B) that it is adequate to the phenomena as represented, i.e. as
represented by us,
are indeed the same!
That (A) and (B) are the same for us is a pragmatic tautology. That it holds
depends crucially on who the indexical word ‘‘us’’ functions to denote
in an assertion. Appreciating that the equivalence for us is a pragmatic
tautology removes the basis for the Loss of Reality objection.26
A pragmatic tautology is a statement which is logically contingent, but
undeniable nevertheless. Similarly, a pragmatic contradiction is a state-
ment that is logically contingent, but cannot be asserted. That is possible,
because the logical contingency pertains to its content, while deniability
or assertability is a concept pertaining to use. Assertion, denial, calling
into question, and the like are actions by a language user. The semant-
ic status (e.g. expresses something contingent) and the pragmatic status
260  :   
(e.g. is deniable) diverge quite obviously when the statement has some
indexical or context-dependent element or feature.
The classic example is Moore’s Paradox, which I already cited above.
But in the cases we are concerned with, there is often no explicitly indexical
word or phrase to be seen. That does not settle the matter. Consider for
example ‘‘There are no statements’’. This is logically contingent since there
are statements in the actual world but there are also logically possible worlds
where there are none. Yet it is not assertable, since its assertion would
be a statement—someone asserting it would stand convicted of falsity by
that very act. In any context of use, the statement is false in the world of
which that context is a part. Yet the statement does not have in it such an
indexical as ‘‘I’’ which plays the crucial role in the usual examples, as in
Moore’s Paradox.27
Similar remarks apply to Tarski’s famous schema ‘‘ The sentence ‘—’
is true if and only if—’’. Suppose we fill it with ‘‘Grows is’’ or with
‘‘Grnunkj’’. Then we do not obtain a truth, in the first case because we
do not have ‘‘Grows is’’ as a sentence of our own language, and in the
second because we do not acknowledge the word ‘‘Grnunkj’’ as a word of
our own language at all. The distinction expressed with the indexical ‘‘our
own’’ is crucial, but is not shown by the words in the sentences under
discussion. We can make the distinction in a general way by asking about
sentences both whether they could be true and whether they can be asserted
by us. If the world had been such that our language had developed so that
the word ‘‘snow’’ had been a word for grass, then ‘‘Snow is white’’ would
not have been a true sentence, though snow would still have been white.
So the Tarskian equivalence is not a necessarily true statement, but its
contraries are not assertable by us for sentences in our own actual language.
How does this apply here? We can sum up the relevant point quite
simply: in a context in which a given model is someone’s representation of
a phenomenon, there is for that person no difference between the question
whether a theory fits that representation and the question whether that theory fits
the phenomenon.28
Think of what would happen to Moore’s Paradox if we treated pragmatic
inconsistencies like logical ones. We would appeal to the logical principle
that if (A & B) is inconsistent then A implies the denial of B and
conversely. So from the incoherence of ‘‘I believe that it is raining in
Peking, but it isn’t’’ we would infer ‘‘If I believe that it is raining in Peking
   261
then it is’’. From the inconsistency in Moore’s Paradox we would proceed

to to the general certainty of It I believe that something is so, then it is, crediting
ourselves with clairvoyance.
To postulate metaphysical structure in re, or in nature, to save structur-
alism would be like ‘saving’ Moore’s Paradox by postulating clairvoyance.
Return to our epistemological question

After introducing Weyl’s text and examining the four options that it
appeared to offer us, I said that we face here a perplexing epistemological
question:
Is there something that I could know to be the case, and which is not expressed
by a proposition that could be part of some scientific theory?
The answer is YES: something expressed only by an indexical proposition.
For to use a theory or model, to base predictions on it, we have to
locate ourselves with respect to it. This applies specifically if a theory is
assessed by recourse to a data model or surface model. In this assessment
the mathematical structure is matched, but the relevance of the matching
consists precisely in the user’s relation to that structure. When using a
model to find our way around in the world we have to be able to say, for
example, that the phenomenon we are presently witnessing is classified in the
theory as oxidizing, or as phlogiston escape, or the like. We have to locate
our situation in the theory’s logical space, in a way that is similar to our
‘‘We are here’’ with respect to a map.
This implies no relevant incompleteness in the theory or model itself. To
make this point about science is not to deny that science is (in principle?)
so complete that every fact, every phenomenon, is completely represented
(representable?) within science. A New York subway map or Paris Metro
map is not incomplete because it comes without a ‘‘you are here’’ tag. It is
selective, it neither does nor purports to represent more than certain aspects
of the topological structure of the system—but that it does completely.
PA RT I V
Appearance and Reality
To begin this last part, let’s have a moment to take stock. A science presents
us with representations of the phenomena through artifacts, both abstract,
such as theories and mathematical models, and concrete such as graphs,
tables, charts, and ‘table-top’ models. These representations do not form a
haphazard compilation though any unity, in the range of representations
made available, is visible mainly at the more abstract levels.
The insistence on this unity as a hallmark of science has been with us in
philosophy since the beginning. In Aristotle we see a remarkable parallel in
his views on drama and on physics.1 The Physics presents us with a view
of the structure of nature and natural processes, and also, in conjunction
with the Posterior Analytics, of the structure of the science that deals with
nature. The Poetics presents the structure of the human condition and of
human action as they are depicted in tragedies, but also of the structure
of those tragedies, that dramatize human existence, themselves. This easily
noticed parallel actually extends from this form of the books to how they
depict their subjects.
A tragedy, writes Aristotle, is a representation of an action . . . . In a
well-constructed tragic plot, if some part were transposed or removed
the whole would be disrupted and disturbed. The action is to form a
tightly structured causal process, flowing jointly from the character of the
protagonists and the conditions in which they find themselves. The element
of chance is to be shunned and if needed at all, kept off stage. The events
‘‘should result from the actual structure of the plot, so it happens that they
arise either by necessity or by probability as a result of the preceding events.
It makes a great difference whether these [events] happen because of those
or [only] after those.’’ (Poetics 52a17–22) Human action and fate must thus
be represented as subject to a causal structure that suffices to make them
intelligible to us.
This kind of intelligibility is precisely what Aristotle asks for in a scientific
theory. The natural path of inquiry ‘‘leads from what is familiar or evident
to us to what is by nature clear or conclusive’’. Starting with what is naturally
obscure but apparent to us we must seek the reasons why, and thereby
266   :   
advance to what is intelligible in itself. ‘‘All inquiry aims at knowledge; but

we cannot claim to know a subject matter until we have grasped the ‘why’
of it, that is, its fundamental explanation.’’2
Aristotle himself seems to see the parallelism very well. When in the
Physics he comes to what he considers a bad theory (the theory of evolution
by natural selection and chance variation, as it happens!) he makes fun of it.
It does not meet his standard for scientific knowledge, for it does not ‘‘deal
adequately with the ‘why’. . . . in terms of each type of explanatory factor’’.
And he emphasizes that again in the Metaphysics: ‘‘But the phenomena show
that nature is not a series of episodes, like a bad tragedy’’.3
This conception of science has not only a venerable history, it has
exercised its grip on the philosophical imagination down to the present
day. A well-constructed scientific theory will tell a story, a narrative in
which the why is as clearly explained as the what, and we come to understand
not only ‘what happens’ but ‘what is really going on’. The question at
issue for us today, in philosophy of science, is precisely to what extent this
conception can still be a guide to the understanding of the sciences we now
have.
Theoretical scientific representation is in general very far from a simple
picturing. Creations of almost unimaginable beauty enter theoretical mod-
eling, and their exploration is understandably irresistible to philosopher
and mathematician alike.4 But—and this may not be just an empiricist
sentiment—these explorations need to be supplemented with inquiry into
the modeling relation itself. At first blush, a model typically represents the
phenomena as embedded in a larger nature or reality. Whether we take
this at face value or bracket the question of reality, we need to investigate
the relations of this theoretically postulated reality to the phenomena, and
to the appearances of those phenomena in measurement.
Scientific knowledge is objective, in a sense that implies maximal inter-
subjectivity. This intuitive conviction, which empiricists, transcendentalists,
and scientific realists all share, rightly remains regardless of the admitted
perspectivity in observation and measurement. But ‘‘objectivity’’ is a term
with many antonyms, and therefore diverse meanings, so the very formu-
lation of this thought invites equivocation. I will begin with the ideal of
objectivity as it appears in putative completeness criteria that philosophy
has purported to find in physical theorizing. Some prominent such criteria,
once honored, bit the dust in recent history. One, much loved and honored
  :    267
in modern times, still sings its siren song to today’s philosopher; I shall call
it the Appearance from Reality Criterion. But I shall argue that it must follow
the others into the dust, if we are to appreciate the new key in which
the sciences are now composed. Once rejected (as, I shall argue, it has
effectively been rejected already in scientific practice) we have the freedom
to follow the contemporary abstract structural forms now prevalent in the
advanced sciences without the unbearable constraint to satisfy a ‘‘realist’’
imagination.
12
Appearance vs. Reality
in the Sciences
If science offers a representation of nature, what precisely does it represent?

Paintings and photos depict things as they appear to us in perception. In
contrast a scientific theory may be said to depict things as they are. The
differences between how things appear to us and how they are depicted by
a scientific theory can certainly be striking.
To urge the intellectual understanding’s superiority over the senses, those
differences became fodder for the skeptical arguments Descartes adapted in
the Meditations. His skepticism was not directed at inferences blithely made
to go beyond the senses, but at trust in the senses themselves. Yet in the
end we can only evaluate the accuracy of any representation by attending
to how the represented object appears to us, whether in direct observation
or in measurement. So even if a scientific theory or model represents
how things really are, the scientific account is not finished unless it entails
correctly what their appearance will be like under realizable conditions.
The phrase ‘‘not finished unless’’ points to a notion or ideal of com-
pleteness. The ideal of completeness here in play—even if we merely
see completeness as a regulative ideal, perhaps not humanly achiev-
able—challenges a science to represent the appearances as well as the
theoretically postulated reality.
This challenge, so very clear to the seventeenth century scientist-
philosophers, is perennial. It was posed dramatically for our time by
Einstein and his colleagues in ‘‘Can quantum-mechanical description of
physical reality be considered complete?’’. This is a challenge that can
precipitate radical, revolutionary change in the sciences. As I will argue
here, these radical changes radiate outward from the content of the new
theories to the very methodology of science itself and to the conception of
what science is to be.
270  :   
Appearance and Reality: the real

and unreal problem
Even within philosophy the appearance/reality contrast has had somewhat
of an undeservedly bad press. When F. H. Bradley wrote his famous
Appearance and Reality he was not aiming to install that contrast as a
dichotomy, but to break out of it. He was in fact criticizing what he saw
as all the forms of realist metaphysics engendered in the seventeenth and
eighteenth centuries. In chapter XII he concludes:
We have found, so far, that we have not been able to arrive at reality. The various
ways, in which things have been taken up, have all failed to give more than mere
appearance. Whatever we have tried [to conceive as beyond the appearances] has
turned out to be something which, on investigation, has been proved to contradict
itself. (Bradley 1930: 110)
Where there is no contrast to be found there is no concept to apply. Bradley
is referring here specifically to the so-called Problem of the External World
that so bedeviled modern philosophy. What he concluded was that it is
a problem with no solution—hence a problem to be escaped rather than
solved. He was right. The problem of appearance and reality, as posed in
modern philosophy, was unsolvable. Accordingly, the problem itself must
have been based on a mistake, a mistaken presupposition hidden in its
origin or formulation.
But as so often happens in philosophy, this pseudo-problem was arrived
at in response to a real problem—in fact a real problem that faced the
physicist-philosophers of the early modern era. How was the philosophical
problem created, and how did it displace the real problem?1
Appearance versus reality at the birth

of modern science
Early in the seventeenth century the newly emerging modern sciences
faced a challenge on many fronts. How is the scientific characterization of
reality, whether in astronomy, or in dynamics, about the constitution of
material objects and processes, to be reconciled with the very different way
in which things appear to us?
 .     271
There were striking successes and glaring gaps. Copernicus’s achievement

was a paradigmatic success: the apparent planetary motions could be
explained, by means of geometric transformations, on the basis of his sun-
centered model. But what about the appearance that the earth does not
move? It was not part of Copernicus’s achievement, to have shown to his
contemporaries that this appearance was compatible with his hypothesis.
For in the physics of his time the motion described was taken to entail all
sorts of force effects that were not observed. Why aren’t the clouds left
behind? Why do cannon balls travel equally far, regardless of the direction
in which they are fired?
While Copernicus was not in a position to address these questions,
Galileo takes the disparity as an indictment of Aristotelian physics and
begins to develop a dynamics compatible with Copernicus’s cosmology. By
the time of Descartes it is widely accepted that such a dynamics can indeed
be developed, and the appearance of rest for loose objects on a moving
earth is no longer puzzling—a striking success achieved in the course of a
single century. Below I will return to Copernicus’s success, which began
this historic process, but I want to do so as illustration for just what Galileo
and Descartes saw as a real problem for the emerging natural science.
In The Assayer Galileo addressed directly the problem of what later
came to be called ‘‘secondary qualities’’, qualities that he did not want to
have any fundamental role in the new scientific image.2 Galileo insisted
on a strict discipline to be observed in the new sciences, to eschew the
multiplication of entities or features in service of explanation. This strict
discipline, in his insistence, implied the demand to couch theories in
terms of notions that could be mathematically represented. The qualities
that this allowed were measurable quantities—the ‘‘primary’’ qualities,
in our later terminology—and they were few.3 Even so, by some later
standards, Galileo admitted too many. Descartes reduced the list to just
attributes definable in terms of spatial and temporal extension; others wanted
additions, such as impenetrability, but ascetic scarcity was maintained. The
world as it appears to us is colorful, noisy, smelly, tasty . . . the world
described purely in terms of those primary qualities is none of that.
Galileo’s Assayer
Galileo famously promised that the colors, smells, and sounds in the
experienced world would be fully explained by a physics among whose
272  :   
descriptive parameters those qualities were not allowed. With typical

rhetorical flair he makes this promise as if it is already certain to be satisfied.
Giving as sole reason that he cannot imagine the contrary (!) he submits
that the perceived qualities fall into those that really belong to the objects
and those that are ‘mere names’:
. . . whenever I conceive any material or corporeal substance, I immediately feel
the need to think of it as bounded, and as having this or that shape; as being
large or small in relation to other things, and in some specific place at any given
time; as being in motion or at rest; as touching or not touching some other body;
and as being one in number, or few, or many. From these conditions I cannot
separate such a substance by any stretch of my imagination. But that it must be
white or red, bitter or sweet, noisy or silent, and of sweet or foul odor, my mind
does not feel compelled to bring in as necessary accompaniments [. . . .] Hence I
think that tastes, odors, colors, and so on are no more than mere names so far as
the object in which we place them is concerned, and that they reside only in the
consciousness. Hence if the living creature were removed, all these qualities would
be wiped away and annihilated. But since we have imposed upon them special
names, distinct from those of the other and real qualities mentioned previously,
we wish to believe that they really exist as actually different from those.4
He continues with some examples of how interaction with something

material can induce different sensations in different parts of the body, and
he offers possible explanations of those sensations in terms of tiny particles
and motions of the air affecting the sense organs. The colorful, tasty, smelly,
and noisy appearances are thus promised to be shown produced as interactional
events, in which the relata are in principle completely characterized in terms
of primary qualities. Nor is he shy to claim success: ‘‘Having shown that
many sensations which are supposed to be qualities residing in external
objects have no real existence save in us, and outside ourselves are mere
names. . . .’’ (Ibid.: 277).
While there is no implication here that the occurrence of secondary
qualities is something beyond the sciences’ domain of application, the
insistence on their ‘‘residence in the sensitive body’’ suggests an irrelevance
for basic physics. This distinction between primary and secondary qualities
coincided with one familiar to the Aristotelian tradition. In the latter, the
peculiar sensibles are those—such as color and smell—perceptible through a
single sense, while the common sensibles—such as shape and volume—are
those detectable through several senses. We can determine a shape either
 .     273
by looking or by feeling, for example. It can hardly be a coincidence that

the primary qualities of Galileo are more or less the common sensibles. And
there is a plausible rationale for that: something detectible only through a
single sense has something too specific, too specifically sense-dependent,
to lend itself to being given a sense-independent status as well.
Moreover, as is always remarked in this connection, there were already
quantitative measures available at least for the paradigm examples of primary
qualities—attributes of shape, volume, number—that were important to
the experimental methods now being developed. Galileo’s famous injunc-
tion that the Book of Nature is written in geometry, and Bacon’s that it
needs mathematical literacy to be read, is crucial to this moment.5 The
rhetoric focused on what was to be counted as real, what as mere appear-
ance. But there was—more importantly in my view—a methodological
reason for Galileo’s insistence on the restriction of descriptions of nature to
a small number of basic terms. Scientific discipline was to preclude the sort
of multiplication of concepts that might so easily provide merely ‘‘verbal’’
rather than real solutions to the problems posed by natural phenomena.
Really, all the above considerations are broadly methodological. The ques-
tion for us, in retrospect, is whether to applaud or decry the metaphysical
claims offered to buttress the methodological reasons.
Galileo’s promises, that the appearances couched in secondary qualities
would be entirely explained, were not empty. There were solid achieve-
ments behind them already, even if still rather modest in our eyes. Those
achievements accumulated into awesome riches by the late nineteenth
century. To mention but one salient example, beloved of philosophers:
combustion of sodium samples is an observable process (phenomenon)
that can be exhaustively described in the terms of basic physics, but this
description can also be utilized to explain how that process produces
a yellow appearance to the human eye, and of course, to a camera or
spectrophotometer.
We can say that, indeed; but how much does it mean? How much it
means depends crucially on what ‘‘explain’’ means, and on the criteria for
what counts as an explanation.
Descartes’s radical transposition

Both Galileo and Descartes were inclined to wave their hands toward
a reduction of all qualities to the primary ones. But Descartes, going
274  :   
beyond Galileo’s relegation of secondary qualities to ‘‘residence in the

sensitive body’’, initiated a strategy that bedeviled both philosophy and
science for some centuries to come. Using well known, in fact ancient,
skeptical arguments and techniques he made a radical move that would,
if successful, take the problem off the shoulders of physics altogether
and place it squarely on metaphysics. In his metaphysical world-picture
we are acquainted directly only with our ideas, our subjectively formed
representations, but have no logical warrant for thinking that what there
is outside the mind is correctly mirrored within.6 When he sought some
assurance for the belief that there is such a mirroring, this was limited to
mirroring of the primary qualities alone. What of the rest?
In Descartes’s posthumous work The World, or Treatise on Light he
purports to lay the foundation of a world-picture entirely transparent to
the human understanding. Here (unlike in Galileo) we do see criteria
for explanation, or understanding. They are the criteria of the mechanistic
philosophy which insisted that all interaction must be by mechanical action-
by-contact. The set of primary qualities is refined to just the attributes that
play a role in mechanical explanation: attributes of extension (spatial and
temporal). The primary qualities are therefore quantities, that is, numerically
representable magnitudes, entirely and adequately represented in geometry
(specifically in kinematics, that is, the geometry of space, time, and motion)
But in the Discourse and Meditations, besides this positive effort to construct
a complete physics, the programme is defended by the use of traditional
skeptical arguments directed especially against any claims to reality of the
secondary qualities. Perception by means of any single sense is subject to
many distortions and illusions. The analysis of warmth and material falsity
in Meditations III provides a good example of this:
As belonging to the class of things that are clearly apprehended, I recognize the
following, viz, magnitude or extension in length, breadth, and depth; figure, which
results from the termination of extension; situation, which bodies of diverse figures
preserve with reference to each other; and motion or the change of situation; to
which may be added substance, duration, and number. But with regard to light,
colors, sounds, odors, tastes, heat, cold, and the other tactile qualities, they are
thought with so much obscurity and confusion, that I cannot determine even
whether they are true or false; in other words, whether or not the ideas I have of
these qualities are in truth the ideas of real objects. For although I before remarked
that it is only in judgments that formal falsity, or falsity properly so called, can
 .     275
be met with, there may nevertheless be found in ideas a certain material falsity,
which arises when they represent what is nothing as if it were something. Thus,
for example, the ideas I have of cold and heat are so far from being clear and
distinct, that I am unable from them to discover whether cold is only the privation
of heat, or heat the privation of cold; or whether they are or are not real qualities:
and since, ideas being as it were images there can be none that does not seem
to us to represent some object, the idea which represents cold as something real
and positive will not improperly be called false, if it be correct to say that cold is
nothing but a privation of heat; and so in other cases.7
A more famous passage in Meditations II (sections 11–13) does too. The

warmed bit of wax, which loses its secondary qualities and undergoes
simultaneous changes in its primary qualities, is nevertheless known to be
the same—giving the lie to the impression ‘‘that the wax is known by
the act of sight’’ and establishing that some facts ostensibly gained through
experience are actually not given through the senses at all.
The problem of reconciling the scientific image with the appearances to
us is thus subsumed under the general problem of skepticism, with those
appearances located in the individual mind. The problem initially faced in
the sciences was thus transposed into one pertaining to mind and matter. The
way this problem is ‘‘solved’’ involves large concessions to skepticism with
respect to perception and judgment. Descartes finds a guarantee for our
adequate representation with respect to primary qualities but the rest are
merely subjective dressing. The gain? This certainly freed the emerging
new physics, but it would leave the emerging modern philosophy holding
the bag.
The cost? It isn’t just the secondary qualities that end up solely in the
mind, on Descartes’s construal. All that is known directly turns out to be
precisely what is in the mind. And what is there, separate from the putatively
known material entities, can at best claim to be a mirroring—with so far
no guarantee to be veridical—of whatever has those primary qualities.
Neither the primary qualities nor their bearers are directly accessible. A
proof of God’s existence and of His main attributes is needed to bridge the
gap. The problem of reconciling the appearances with the new sciences’
ascetic world-picture, in which material has only primary qualities, has
been putatively solved, the task no longer weighs heavily on the shoulders
of natural philosophy. But this came at the cost of introducing the much
more difficult—in fact, in the end, unsolvable—‘‘Problem of the External
276  :   
World’’. Philosophy landed itself in an unsolvable problem, a problem so

designed, in effect, as to be unsolvable, hence not one that makes sense at all.
If indeed these are the gain and loss, we are tempted to charge this natural
philosopher, our René Descartes, with philosophical leger-de-main.
Three putative completeness criteria

To understand the sciences, to achieve a synoptic vision in which the
manifest and scientific images both receive their due, we must take
the opposite tack from all of modern philosophy: ignore Descartes’s
transposition of the problem from nature/science to mind/philosophy,
and face squarely the accountability of the physical sciences to the public
appearances.
‘‘Reality’’ and ‘‘Appearance’’ are philosophically loaded terms. In our
context they are to be understood quite prosaically. The theoretically
postulated reality of science may be a world consisting of atoms, while
the appearances that the theories are meant to save are the contents of
measurement outcomes. Recall that it is too limited to think of these
outcomes as simply numbers or sentences! The yellow color of the flames,
the spatial configuration on a monitor display or in a projection made by
a ‘drawing machine’ or camera obscura, the smell of a gas sample, can
all count as outcomes to be reported. The smelly, colorful, noisy things
around us, such as apples and horses, sunsets and storms, are all to be
accommodated.
‘‘Appearance’’ does not refer here to subjectively experienced impres-
sions. All those colorful, noisy things are public, and so are the appearances:
the dictate of repeatability ensures that scientifically admissible experimental
results are public. Neither does ‘‘reality’’ refer to Kantian things-in-
themselves or to a Cartesian external world: all those smelly, colorful,
noisy things—and the colorful, noisy, smelly measurement outcomes—are
most certainly real. The theoretically postulated entities of a scientific the-
ory may be real as well; but in our context of discussion their theoretical
status is to be kept in mind.
What has science traditionally been held accountable for? If science
means to provide a representation of nature, criteria for its success must
be related to that task, and must appear concretely in scientific theory
 .     277
choice and evaluation. Such criteria have often been explicitly formu-
lated. Often enough they were not just debated among philosophers, but
loomed centrally in famous episodes in the sciences themselves. Most
striking, though, is how time and again science has refused submission
to ideals imposed from outside, or even from its own past. Scientific
progress may and does involve rejection of previously proclaimed criter-
ia. We shall examine some here, related to necessity, determinism, and
causality.
The historical pattern

While flaunting previously upheld criteria is remarkable, so is the typical
aftermath of such flauntings: sustained reactionary philosophical efforts at
restoration.
The pattern seems to be this. A certain criterion of completeness is held
up as an ideal for science. Perhaps it is even said to be satisfied already,
details apart; and the scientifically minded glory in its rule. Embattled by
new empirical findings, scientists violate and reject the criterion, and gain
a startling success by their own (new) lights. The success is hailed as a
triumph over a now discredited philosophical ideal. But then the reaction
sets in as well. There is a sense that in ignoring past demands, science
has betrayed its trust—now it ‘‘only describes and does not explain’’,
now it ‘‘only puts in mathematical form, but leaves nature unintelligible’’.
These were the Aristotelian complaints about Cartesian mechanics and
cosmology, the Cartesians’ complaints about Newton’s physics as mere
mathematization, and Huygens’s complaint that Newton’s theory was not
‘mechanical’, that is, did not explain gravitational phenomena.8 Nearer our
own time there were the complaints about Einstein’s discarding of the ether
which was argued to leave optical phenomena inexplicable, and lately the
Bohmians’ complaints about the Copenhagen school in quantum theory.9
When empirical successes are scored at the cost of violating respected and
revered criteria of theoretical success, many philosophers and reflective
scientists strive mightily to reinstate the rejected criteria, or show them to
be ‘essentially’ or ‘really’ satisfiable after all.
Necessity, determinism, causality

As first example consider the Aristotelian ideal that science must explain how
things happen, by demonstrating that they must happen in the way they do. That
278  :   
ideal requires that regularities in the phenomena derive from universal

necessary principles. Galileo, Gassendi, Boyle, Descartes, and Newton
consciously and explicitly refuse to take on this Aristotelian task for science,
or to accept it as criterion for scientific success. Indeed, they claim that the
modern era’s scientific success derives largely from their rejection of that
tradition.
But already in other passages at the hands of these very same writers,
we see the ideal, and that criterion, advocated in an even stronger form!
There is talk of laws of nature, not all that clearly distinguished from those
lambasted and ridiculed constraints of Aristotelian physics. Even more
extreme sentiments take hold: the regularities must derive from not just
natural but logical necessity.10 This sentiment is sometimes encountered
still, not so much among philosophers but in physicists’ dreams of a final
theory so logically airtight as to admit of no conceivable alternative, one
that would be grasped as true when understood at all.11
More salient in the development of modern natural science, however, is
the turn to mechanism as a guiding philosophy, thoroughly entangled with
a conviction of determinism in nature. Descartes and Boyle were main
advocates of the view of nature as run entirely by means of mechanical
action by contact, and their advocacy was nothing short of evangelistic.
Thus Boyle describes the world as a ‘‘self-moving engine’’, a ‘‘vast machine’’
running ‘‘by the meer contrivance of brute matter managed by certain laws
of local motion’’.12 When that vision was thoroughly diluted—or more
accurately, left behind—by Newton’s introduction of gravitational action
at a distance, it was the insistence on determinism that remained. Thus,
shortly before the revolutions that would shake physics, Maxwell wrote,
with an echo of the mechanical philosophy’s inspiration still:
When any phenomenon can be described as an example of some general prin-
ciple which is applicable to other phenomena, that phenomenon is said to be
explained . . . . On the other hand, when a physical phenomenon can be com-
pletely described as a change in the configuration and motion of a material system,
the dynamical explanation of that phenomenon is said to be complete. (Maxwell
1890: vol. ii: 418)
The conviction that a scientific account is complete only if it is deterministic

was thereafter strongly supported within the Kantian tradition. For as Kant
 .     279
saw it, the very coherence of experience requires that it takes a form of
experiencing ourselves as living in a spatio-temporally definite causal order.
The context in which physics was changing around 1900 included thus a
strong conviction, inherited from classical physics and modern philosophy,
that all phenomena in nature derive from an underlying deterministic phys-
ics. Determinism had become a criterion of completeness: any apparent
gap in determinism so far is filled by statistical laws, but the statistical
probabilities can only be a measure of ignorance. Poincaré begins one
chapter in his Science and Hypothesis of 1905 with ‘‘No doubt the reader
will be astonished to find reflections on the calculus of probabilities in such
a volume as this. What has that calculus to do with physical science?’’
But the question of the possibility of indeterminism was already salient
and in the air before it was related in any way to the quantum, due to
different views of statistical thermodynamics (cf. Stöltzner 1999).13 Max
Planck discusses that too in his famous 1908 Leiden lecture. There he
praised Boltzmann for ‘‘the emancipation of the concept of entropy from
the human art of experimentation’’ (p. 14), but notes that the price seemed
to be to relegate the second law of thermodynamics to a merely statistical
regularity that admits exceptions.
Boltzmann has drawn therefrom the conclusion that such strange events contra-
dicting the second law of thermodynamics could well occur in nature, and he
accordingly left some room for them in his physical world view. To my mind, this
is, however, a matter in which one does not have to comply with him. For, a nature
in which such events happen . . . would no longer be our nature. . . . (Ibid.: 15)
This is a strong profession of faith in determinism, but the mere fact that
it seemed appropriate to express this shows how the question could not
be ignored. Planck had an explicit antagonist in Franz Serafin Exner, who
defended an empiricism along Mach’s lines, rejecting as meaningless any
speculation as to whether there exist some unobservable deterministic laws
behind the phenomena (cf. Stöltzner 2002). So Exner emphasized that if
some domain appears to be subject to deterministic laws, these could be
the macroscopic limit of indeterministic basic laws governing nature at a
microscopic level. This theme was later taken up by Hans Reichenbach,
who coupled it with his view of how an indeterministic world could still
be ‘lawlike’ through his principle of the common cause.
280  :   
The challenge to determinism was the first, foremost, and most vis-
ible philosophical confrontation to arrive for quantum mechanics. The
probabilistic resources of classical statistical mechanics had been newly adap-
ted in such a way that, as it seemed then, no grounding in an underlying
deterministic mechanics was possible. A vocal part of the physics community
was averse to seeking out the logical possibilities here—some rejected expli-
citly the task of finding or displaying such hidden mechanisms.14 Nature
is indeterministic, or at least it can be or may be—and if that is so,
determinism is a mistaken completeness criterion for theory.
Now Reichenbach, who did much to provide a rationale for this
rejection of determinism, introduced an apparently weaker but still sub-
stantive new completeness criterion: the Common Cause Principle.15 This
third principle is satisfied by the causal models of general use in the social
sciences, and for many purposes in the natural sciences as well. They are
models in which all pervasive correlations derive from common causes
(in a technical, probabilistically definable sense). But the demonstration
in the 1960s and later that quantum mechanics violates Bell’s inequalities
shows that the new physics was riding rough-shod even over this third
criterion.16,17
However that may be, I’ll now turn to a fourth completeness criterion
that seems compatible with these rejections, and appears to be quite gen-
erally accepted at least among philosophers and by the general public.
But it too was one clearly, emphatically, and explicitly rejected by the
Copenhagen physicists.18
Appearance vs. Reality: A deeper Criterion

As we saw, physics in the modern era depicts nature as being quite different
from how it appears. The theoretically postulated reality is different from
the appearances. The disparity became truly salient when Galileo and
Gassendi embraced the atomism revived in the Renaissance. Those atoms
have only primary properties such as shape, volume, and number; the
appearances are colorful, noisy, smelly, and tasty. The disparity is certainly
not diminished by Descartes’s further restriction of the real attributes of
matter to extension in space and time. Newton, rejecting this ontological
asceticism, introduced forces and mass, but these seemed occult additions
 .     281
to many of his contemporaries; they certainly did not return closer to the
familiar world of experientia.
But at the same time, as we saw illustrated above, Galileo accepted a
commensurate criterion of completeness, that physics must explain how those
appearances are produced in reality. This amounts to a stringent demand: that
this noisy, colorful, smelly but tasty world of appearance be fully explained
in terms of the attributes that science explicitly counts among its significant
parameters.
Science is understood to be incomplete until and unless it meets that
demand: I call this the Appearance from Reality Criterion.
Does this Criterion govern and guide the scientific enterprise as a
whole? Examples abound to make us immediately sympathetic to that
idea. We credit science with adequate and satisfactory explanations of how
many familiar phenomena are produced: how ash is produced when we
burn a cigarette or some logs, how methane is naturally produced in a
swamp. In the example of how a flame is turned yellow when a sodi-
um sample is inserted, an aspect of visual appearance is explained. But
we must look carefully into just what is demanded, how far it can be
pushed, and what resistances it may have encountered since those heady
early days.
What the Appearance from Reality Criterion requires

In the examples I gave, scientific representation of nature is shown to
include the appearances in a very specific, particular way. Their ‘deriv-
ation’ does not just amount to showing that they fit into the theoretical
representation. That would be a minimal requirement; just the one that
Cardinal Bellarmini suggested to Galileo as solely relevant to the Coper-
nican theory. In contrast, the ‘derivation’ must be a demonstration that,
and how, these appearances are produced as a proper part of the depicted
reality. This stringency in the demand, for such a derivation/explanation,
has remained a continuing theme in scientific realist writing. Jared Leplin
writes:
A theory is not simply an empirical law or generalization to the effect that certain
observable phenomena occur, but an explanation of their occurrence that provides
some mechanism to produce them, or some deeper principles to which their
production is reducible. (Leplin 1997: 15)19
282  :   
The telling phrase is ‘‘provides some mechanism to produce them’’. The

appearances are to be explained as produced in the world depicted by
fundamental physics.
Just what is this requirement? Below we will take up a paradigm example,
in Copernicus’s explanation of the astronomical appearances. The mech-
anism to produce them is described completely in terms of kinematics and
geometric optics. The appearances there—like all the examples I listed
above—are public, intersubjective, not belonging peculiarly to individual
experience but displayed in publicly accessible form. Recall the familiar
example of a yellow color in the flame. What is produced, when a sodium
sample is inserted in a flame, is light of a wavelength, measurable and
recordable mechanically as well as humanly visible, in the ‘yellow’ part of
the color space.20
The requirement of production does not preclude indeterministic or
stochastic accounts.21 Imagine that the explanation of how a cigar-
ette burns by recourse to statistical thermodynamics were qualified by
the hypothesis that there are random small fluctuations which modify
the underlying mechanics. That would still leave us with a derivation
(in the appropriate sense) of the burning process.22 A means of pro-
duction subject to small random perturbations is not an inexplicable
miracle, and a stochastic dynamics is not a mere ‘‘black-box’’ predictor; it
provides a derivation of the phenomenon through the details of action and
interaction.
The Appearance from Reality Criterion applies tellingly, however,
whenever there is a felt gap in the story: the appearances must clearly
derive from what is really going on. The gap is not filled if one simply
adds a postulate to the effect that the Appearances are produced in some
specific way, without displaying that way.23 For it is easy enough, but
entirely uninformative, to wave a hand at some supposed relation between
the theorizing and measuring agent and those aspects of nature that are
measured and represented, and then to claim that this accounts for how
the Appearances differ from what science ostensibly describes. If left as
a mere claim that does not suffice, and certainly does not satisfy the
Criterion.
A look back to the three earlier completeness criteria show how they all
hinge on involving modality into our understanding of science. According
 .     283
to Aristotle science must show why the phenomena and their regularities
must be the way they are, deriving from principles that are universal and
necessary. The seventeenth-century conception of laws of nature, only just
emerging from their theological gestation, delineate what is necessity in
nature. The Appearance from Reality Criterion is similar in this respect:
it too is a demand for explanation which is satisfiable only by connections
deeper than brute or factual regularity.
Therefore, when it is claimed that to be complete physical science needs
to derive the appearances from that reality, the term ‘‘derive’’ cannot here
just mean ‘‘deduce’’ or ‘‘predict’’. Required is a connection of the order of
explanation through necessity and/or causal mechanisms to be displayed,
which produce the appearances. When is this demand not met? It is not
met if science should simply issue successful predictions of measurement
outcomes. Prediction does not suffice by itself even if the prediction is by
means of systematic rules of calculation, from the state of nature theoret-
ically described. For calculational and predictive success does not ipso facto
imply an explanation of why and how the appearances are produced.
Phenomena versus appearances

That empirical science must ‘‘save the phenomena’’ is an ancient creed.
The criterion that I call ‘‘Appearance from Reality’’ is however a stronger
dictate. The phrase ‘‘to save the phenomena’’ is often rendered colloquially
as ‘‘to save the appearances’’ and ‘‘appearances’’ is often used synonymous
with ‘‘phenomena’’. As emphasized from the outset, I use these terms
so as to mark a crucial distinction. Phenomena are observable entities
(objects, events, processes, . . .) of any sort, appearances are the contents of
measurement outcomes.24
Both are terms with a past. On philosophical lips they are often loaded,
unfortunately, with connotations linking them to mind and thought which
are not in any way relevant to my usage. The verbal distinction can for
example be found in Kant’s Critique of Pure Reason.
Still less must phenomenon and appearance be held to be identical. For truth or
illusory appearance does not reside in the object, in so far as it is intuited, but in
the judgement upon the object, in so far as it is thought.25
284  :   
Both the way in which each of these notions is understood here, and the
way the distinction is marked, have to do with perception and perceptual
illusions or mere impressions. Nothing in my usage of these two terms
refers to these factors, unless very indirectly. How an observable object
or process (phenomenon) appears in the outcomes of the measurements is
itself an objective fact, a public, intersubjectively accessible fact. But there
is a similarity nevertheless to Kant’s insistence on—to coin a phrase—the
greater objectivity of the phenomenon. That is, we do have to insist on
the distinction: the appearance is determined jointly by the measurement
set-up (involving both apparatus and the system to which it is applied), the
experimental practice, and the theoretical conceptual framework in which
the target and measurement procedure are classified, characterized, and
understood.
While usage is diverse, I do not think that my usage is egregious
with respect to the historical scientific literature. For example Hertz,
commenting on his research on the effect of ultra-violet light, says in the
Introduction to his Electrical Waves ‘‘Inasmuch as a certain acquaintance
with the phenomenon is required in investigating the oscillations, I have
reprinted the communication relating to it [here]’’. The reprinted paper
to which he refers begins as follows (and there we can see what the
phenomenon in question is):
In a series of experiments on the effects of resonance between very rapid electric

oscillations . . . two electric sparks were produced by the same discharge of an
induction coil, and therefore simultaneously. One of these, the spark A, was the
discharge-spark of the induction-coil, and served to excite the primary oscillation.
The second, spark B, belonged to the induced or secondary oscillation. The latter
was not very luminous . . . I occasionally enclosed the spark B in a dark case
so as more easily to make the observations; and in so doing I observed that the
maximum spark-length became decidedly smaller inside the case than it was before.
On removing in succession the various parts of the case, it was seen that the only
portion of it which exercised this prejudicial effect was that which screened the
spark B from the spark A. [. . .] A phenomenon so remarkable called for closer
investigation. (Hertz 1962: 63)
‘‘Appearance’’ I reserve strictly for the contents of (possible) measurement

outcomes. Phenomena are observable, but their appearance, that is to
say, what they look like in given measurement or observation set-ups, is to be
 .     285
distinguished from them as much as any person’s appearance is to be

distinguished from that person.26
As an especially good example for our present concerns I will take the
distinction between planetary orbits and their appearances to the terrestrial
observer.27 The Renaissance astronomers’ data detailed what they saw
night after night, hence the appearances, the contents of their measurement
outcomes. But Ptolemaic and Copernican astronomy, accountable to those
same appearances, described the phenomena differently.
Am I right to say that the planetary orbits are observable processes, while
what the astronomers cited were their appearances? I do not think there is
a difference in principle between this case and the way in which we must
distinguish what Mount Everest is from how Mount Everest looks from the
North, the South, and various other compass directions. That mountain
is undoubtedly an observable object—only weird philosophical jargon
could decide differently—though of course we infer its shape from what
its appearances are in telemetry data and photographs taken from various
positions. That data and inferences are needed to arrive at an accurate
description of the thing does not mean that it is not something observable,
and only a strange departure from common sense would lead us to the
idea that e.g. mountains are only theoretically postulated entities.28 The
planetary orbits differ from the mountain only in that measurements over
time are needed, while in the case of a mountain a number of simultaneous
measurements would do as well.
There were two developments in techniques of representation before
Galileo that we can see as feeding into the kinematic representation
developed in his century. The first was that of linear one-point perspective
in painting, and the second Copernicus’s and Tycho’s mastery of trans-
forming geometric models in astronomy so as to shift the center taken as
‘at rest’. Both concentrated on how a description of the visual appearance
from particular vantage points can be derived from a reality admitting of
many different vantage points. Both grew from the subject of Perspectiva,
a mélange of geometry, optics, and practical drafting techniques, and both
were steps on the way to projective and descriptive geometry. But the
sorts of representation that perspectival drawing and geometric technique
provided were more than superficially different.
For, if I may put this anachronistically, the second dealt not with
perspective but with transformations of frames of reference in Euclidean
286  :   
space and its consonant kinematics. This technique was mastered in practice
by the time of Copernicus, though formalized only by the end of the
seventeenth century. In contrast, the study and perfection of perspectival
drawing gave rise to the very different subject of projective geometry. That
too saw its first rigorous development in the seventeenth century, but was
then neglected, until coming into its own (with a unified treatment of
Euclidean and non-Euclidean geometries) in the nineteenth century.
This is not an incidental historical point. Both forms of representation
tend to be called ‘‘perspectival’’ and both tend to be thought of as
depicting the appearances. But they accomplish very different tasks. I
honor the difference between them with distinct terminologies. The
geometric representation of e.g. planets and planetary motion depicts the
phenomena. The content of a perspectival drawing of the same events is their
appearance, equivalently, the drawing depicts how they appear from certain
vantage point.29
At the risk of annoying by repetition: the phenomena, in the sense of the
observable parts of the world, whether objects, events, or processes, the
sciences must save (in the ancient phrase). These admit also of objective
and indeed purely theoretical description, which does not link their reality
to contexts of observation or to acts of measurement. What the latter, the
measurements, provide are how those observable parts of the world appear
to us in the corresponding measurement set-ups. That, the content of a
measurement outcome, is an appearance, and the sciences are accountable
to that as well.
How Copernicus saved the appearances
In the first book I set forth the entire distribution of the spheres together with
the motions which I attribute to the earth, so that this book contains, as it were,
the general structure of the universe. Then in the remaining books I correlate the
motions of the other planets and of all the spheres with the movement of the earth
so that I may thereby determine to what extent the motions and appearances of
the other planets and spheres can be saved if they are correlated with the earth’s
motions. (Copernicus, De Revolutionibus, Preface.)
Here we have the paradigm example of this distinction, playing a crucial

role at the very beginning of the modern era.
 .     287
Copernicus described the planetary motions. These motions are observ-

able; they can be observed and also be registered on film, whether from
Earth or from a satellite or from outer space. Copernicus depicted them
clearly in a geometric model. They certainly cannot be identified with
the appearance recorded by any such specific means as views through a
telescope, successive photos over a period of many days, film, or video
recording. The recording is always from a specific vantage point, and
that point is arbitrary, it has no privileged status either in nature or in
Copernicus’s model. What is the content of such a recording then? It is the
appearance of the planetary motions, for a photo, film, painting, or drawing
displays how the recorded object, event, or process ‘looks’ from the chosen
vantage point.
Specifically, the planetary ‘retrograde’ motions are, according to Coper-
nicus, (mere) appearances. Mercury’s retrograde motion is a good example
of an appearance. Literally speaking this is something that does not hap-
pen—it does not exist! Mercury never turns back to reverse its orbital
direction. But that is how it appears to the observer or the film camera,
viewing from Earth. Mercury’s motion is an observable phenomenon, but
Mercury’s retrograde motion is an appearance.
The Ptolemaic system, concretely depicted by an armillary on a table
top, represents the motions of the stars and planets in the frame of reference
of the earth. Copernicus’s system of the world represents the motions of
the stars and planets in the frame of reference of fixed stars, with the ‘mean
sun’ as center. But because Copernicus devised his system by a sort of
‘transcription’ of Ptolemy’s, he can point out that observations from the
earth in his system will deliver the same data.30
Consider what is utilized in this ‘pointing out’. First of all, there is
what the system represents directly, the postulated ‘‘general structure of
the universe’’, as he wrote in the Preface. Secondly, there is the geometric
optics, based on the postulate that unobstructed light travels in straight lines
with infinite speed. Thirdly—and here we make contact with the visual
arts—the appearances to be saved are identified with the projections through
a point on the earth of the celestial motions by those straight light-lines.
The appearances, thus conceived, change with time. There were no
motion pictures, but of course one could construct a series of stills, and one
could furthermore combine these into a single picture of a motion over
288  :   
time. This is the birth of modern kinematics together with kinematic, as

opposed to static geometric, representation. The most striking illustration to
Copernicus’s contemporaries as to us, is the new explanation of retrograde
motion of the planets. The following composite diagram shows both the
postulated kinematics and the light-lines that project the motions onto an
earthbound window, so as to derive the appearances from the real motions.
Figure 12.1. Copernicus’s Model of Retrograde Motion
The above diagram (from Cohen 1960: 39, figure 10) graphically presents
the Copernican explanation of the apparent retrograde motion of an inferior
planet such as Mercury or Venus, by depicting how its motion would look
from a slower moving Earth also orbiting the sun. Copernicus’s model
represents the observable phenomena, that is, certain processes in space and
time. What the Copernican does in order to credential his representation is,
in effect, to explain by means of geometric optics and projective geometry
how the visual appearances (content of outcomes of measurements made by
astronomers) are produced from reality. Copernicus can demonstrate that
his theory ‘saves’ certain phenomena by showing how the visual appearances
derive (via kinematics and optics) from what he postulates concerning them.
Three-faceted representation
There was another discipline in Copernicus’s time, located between astro-
nomy and physics. It carried the melodious name of Theorica.31 The aim of
 .     289
this enterprise was to construct physical, mechanical models of the heavens

that would explain the celestial phenomena. The crystalline spheres are of
course well known from the long uneasy history of attempts to recon-
cile ancient and medieval physics and astronomy. It was very difficult to
construct a physically possible cosmos along these lines that would save
the phenomena described by the astronomers, but arguably not impossible.
The attempt ended effectively with Tycho Brahe’s observations on comets
apparently moving unobstructed through the spheres. Although the theories
offered in Theorica along these lines failed, they were worthy forerunners
of later physical cosmologies like Newton’s System of the World with its
universal gravitational force. The models of Theorica, as well as Newton’s,
go well beyond Ptolemy’s or Copernicus’s description of the observable
phenomena. They depict a postulated ‘‘Reality’’, a nature structured in
unobservable ways and populated by largely unobservable entities of which
the phenomena are the observable parts.
The physical sciences give us representations of nature, and scientific
representation is in general three-faceted. From a purely foundational point
of view, the theoretical models that depict the ‘underlying reality’ are
the main thing. But some elements or substructures of those models are
meant to represent the observable phenomena—the empirical substructures.
Finally, it is a requirement for the theory to at least predict, and if possible
to ‘‘derive’’ the appearances, that is to say, the contents of measurement
outcomes. This division corresponds to three ostensibly different domains:
[1] Theoretically postulated reality
— Micro structure, forces, fields, global space-time structures
[2] The observable phenomena
— Macro objects, motions, tangible and visible bodies, . . .
[3] The appearances
— Measurement outcomes, ‘‘how things look’’ in observational context
The phenomena can be measured and observed in many different ways.
How they appear in the measurement outcomes will vary from one way to
another—and from one occasion to another—specifically because those
outcomes provide perspectives on the phenomena. So to say that the theory
must save the phenomena is not the same as saying that the theory must be in
accord with the experimental and observational results. There is certainly
a close connection! If the measurement is well designed for its purpose,
290  :   
what is to be found at its conclusion—such as pointer positions on gauges,

print-outs, numbers, computer monitor displays, or what have you—will
be especially telling when different theories try to achieve their designated
ends. But we need to keep remembering: the measurement outcome shows
not how the phenomena are but how they look.
In the modern era, the era of what we now call classical physics, each
level has a certain completeness. When frames of reference come into their
own, we have eventually a three-level representation: there is the world
[1] as described in co-ordinate independent terms,
then the world
[2] as described in a given frame of reference (co-ordinatization),
and finally
[3] as it looks from a given vantage point with specific orientation.
The first form of representation admits of many of the second sort, and the
second of many of the third sort. The higher level is uniquely determined,
in the classical world picture, by the collection of those at the next level plus
the transformations that connect them.32 There is the important difference,
however, that a single representation on the first or second level contains
everything in a way that the third most definitely cannot. The depiction on
that third level, that is, of the visual appearances—the true measurement
outcomes—is always limited, just as the observer cannot see what is behind
his back, so any measurement set-up at all displays only what is inside
its range.
With the end of the modern period in physics, the advent of relativity and
quantum theory, the conception of a physical theory and of the connections
between postulated reality, phenomena, and appearances underwent radical
change. It is at this point that more drastic challenges appear for the
Appearance from Reality Criterion.
13
Rejecting the Appearance from
Reality Criterion
What should be our take on the Appearance from Reality Criterion today?
We need first of all to explore the conceptual possibilities, the ways in
which a science could feasibly come to violate the Criterion, and then ask
how these relate to practice in recent science.
Although my main concern is with the natural sciences, it is instructive
to see how the issue is broached in analytic philosophy of mind and
its take on cognitive science. The question of how the psychological
phenomena relate to the physical processes in brain, body, or body plus
environment—our new version of the mind–body problem—has seen
a series of answers more or less culminating in the supervenience thesis.
This move points us readily to logical and philosophical moves available
when outright reduction seems out of reach. In fact, that thesis echoes a
move familiar from another historical episode: Leibniz’s reconciliation of
contingency with the principle of sufficient reason, and both episodes can
serve us as guide to other such perplexities.
For recent scientific practice I shall argue that the Copenhagen develop-
ment of quantum theory exemplifies a clear rejection of the Criterion. The
famous Measurement Problem in the philosophy of quantum mechanics is
not a problem from an empiricist point of view.1 What it marks, I shall
argue, is the methodological rejection of the Appearance from Reality
Criterion in this new science.2 The rejection may not be unique in the
history of science, but is brought home to us inescapably by the advent of
the new quantum theory.
Even if that theory is superseded (or if fundamental physics develops
in a accordance with a new interpretation under which the Criterion can
be satisfied) our view of science must be forever modified in the light
of this historical episode. The Appearance from Reality Criterion was
292  :   
rejected in practice, in an episode that can only be acknowledged as one

of genuine scientific advance, in the face of theoretical and experimental
problems. In this respect it is in the same boat as determinism: even if in the
future a completely adequate deterministic physics emerges, the history of
twentieth physics will have shown that determinism is not a completeness
requirement for science as such.3 Since these criteria were violated in
actual practice it is possible—and in fact incumbent on us—to conceive
of science as not so constrained.
The supervenience of mind challenge

Cognition and philosophy of mind are far from my proper concern here.
But they offer a remarkable illustration of how the Appearance from Reality
Criterion can be challenged. Not only that: the challenge comes in this
case within a self-proclaimed naturalist, physicalist, ‘scientific’ approach to
the so-called mind–body problem in philosophy.
Are psychology or cognitive science in principle autonomous or must
they be reducible to fundamental physics? That is hardly a practical question
for a working scientist. There is no such reduction even for current materials
science or chemistry, let alone physiology—not to speak of any part of
psychology. But we can ask about reducibility in principle, and that has been
a central question in the story of physicalism in twentieth-century analytic
philosophy of mind.
In the 1950s, U.T. Place offered the hypothesis that certain events
and processes traditionally classified as mental (for example, sensation)
are identical with events and processes in the brain (Place 1956). He
called this the materialist hypothesis. As he pointed out, it is in principle
falsifiable, namely if the described mental events and processes have a
greater complexity than brain events and processes.4
Not only this position, but every claim concerning reduction of
the psychological to the physical, bit the dust in the course of the
ensuing philosophical debates. The eventual claim, introduced to save
materialism or physicalism in principle, became the much weaker one
that, though irreducible, mental phenomena supervene on physical
reality.5
What this means, roughly presented, is that the actual psychological
phenomena could not be different without a difference in the physical state
      293
(of the organism, or possibly the organism plus its natural environment, or
possibly of the entire universe).
Leaving aside the more metaphysical notions that are often involved in
the idea of supervenience, we can draw one clear consequence from it
that will reinforce this reading (though in slightly more technical terms).
Suppose that P and M are two languages, one truly physicalist and the
other mentalist—whatever that means, whatever might suffice for the
one to describe everything in terms aptly used in physics and for the other
to describe psychological phenomena (as well) in terms of psychology or
cognitive science. What is implied here by the thesis that the psychological
supervenes on the physical?
(a) To say that we have no possible reduction but do have super-
venience implies that M cannot be translated into P sentence by
sentence, nor paragraph by paragraph, nor even that the definable sets
of sentences of M are translatable into definable sets of sentences of P.
That is the ‘‘no reduction’’ claim.
I will explain the restriction to what is definable below. Taking that
for granted for now, what does the claim to supervenience entail? To
understand that, let’s think of all the possible situations that can be described
(however partially) in both languages. We can depict them this way if we
assume no special relation between the two languages:
Figure 13.1. Failure of Supervenience

294  :   
Here the vertical bars depict the situations that satisfy a given maximally
complete description P(1), P(2), . . . in the physicalist language P. The
slanted bars depict those that satisfy given maximally complete descriptions
M(1), M(2), . . . available in the language M.
The letters ‘‘X’’ and ‘‘Y’’ name two specific situations: Both X and Y
satisfy P(1), while X satisfies M(1) but Y satisfies M(2). Now you can see
immediately that if there are such situations, then M does not supervene
on P! For suppose we are in situation X. Then if we would say ‘‘if the
mental phenomena were different then something physical would also be
different’’, the possibility of situation Y would contradict our claim.
So if we want to redraw the picture, so as to present the case in which
the mental is supervenient on the physical, then the lines that divide
complete descriptions in M would have to coincide with certain lines that
divide complete descriptions in P. In other words, roughly put: a complete
description in M has to be in effect a disjunction of complete descriptions
in P.
Note well: if we rule out reducibility, then those disjunctions cannot
be finite or recursively specifiable etc. That is why I wrote ‘‘roughly
put’’—I’ve stretched the word ‘‘disjunction’’ beyond its normal use. The
point is clear though: at the level of indefinable sets of possibilities, a
possibility in principle describable in M can be equivalent to a family of
possibilities in principle describable in P. But in principle describable does not, in
that case, refer to what is humanly or mechanically definable or computable
or recursively specifiable, and so does not imply reducibility. We can sum
up this conclusion, taking a little liberty with the notion of disjunction, in
this way:
(b) The syntactically complete descriptions of everything (to the extent
possible in M) correspond disjunctively to sets of sentences in P, as far
as truth conditions are concerned.
Putting the first and second point together shows perspicuously that the
correspondence just claimed must be between sets of sentences that are not
definable.6 Obviously no deduction could mirror this correspondence.
It would be beside the point to stop here to inquire into the plausibility
of this claim. What is important for us, once again, is a meta-issue. On this
view, that the mental or psychological only supervenes on the physical, what
becomes of the science of cognitive psychology? By classifying psychological
      295
phenomena—the subject of the science of cognitive psychology—as

irreducible, this position implies a pertinent autonomy for that science, its
independence from fundamental physics. For the ‘‘supervenience without
reduction’’ claim explicitly entails that no mechanism can be displayed
even in principle for the production of the (mental) Appearances from the
(supposedly physical) Reality. Those appearances depend on the physical in
the minimal sense that the appearances could not be otherwise without the
physical state being different. But they are not derivable in the relevant sense.
The claim of supervenience without reducibility implies that science
is not, will not, and cannot be complete in the sense of deriving the
psychological phenomena from the postulated physical reality. On the basis
that ought implies can we must then also conclude that science as a whole
is not required to be complete in that sense. We have here a conception of
the sciences that specifically exempts them from satisfying the Appearance
from Reality Criterion. For this conception envisages the development of a
cognitive science and psychology in the absence of any physicalist account of
how the pertinent phenomena are ‘derived’ or ‘produced’ in physical terms,
while maintaining that the physical description is in principle complete.
Just to prevent any misunderstanding: in displaying this conception I
am not objecting here to the claim that the psychological phenomena
supervene on the physical. Nor am I even maintaining that this claim fails
to adequately explain why those phenomena are or even must be what they
are.7 The point is rather that in the philosophy of mind it became popular,
though perhaps without explicitly noticing this, to reject the Appearance
from Reality Criterion as a completeness criterion for science as a whole.8
Once we see this logical leeway, we can ask how the supervenience
thesis could be maintained elsewhere. The possibility is then no longer
outré, and can be taken equally seriously, for example, for the various
‘levels’ of inorganic nature.
It may be objected finally that the supervenience claim is, by its very
character, neither refutable nor confirmable by empirical evidence, and that
therefore it is an empty addition to scientific psychology or cognitive science.
That may be so. But it is actually possible to have grounds to reject a super-
venience claim, given (besides any empirical evidence) certain other com-
pleteness claims that could be made for a scientific theory. So the claim of
supervenience cannot be part of every sort of interpretative view of science.
This we will see concretely illustrated in the case of quantum mechanics.
296  :   
The Great Leibnizian Escape move

The ‘‘supervenience without reducibility’’ claim asserts a connection in
nature which cannot be displayed by means of a theoretical deduc-
tion. Deduction is after all an operation to be carried out with certain
resources—linguistic, logical, mathematical—and these resources have
their limits. While these limits are the proper domain of meta-mathematics,
developed only in the last century, they were vaguely perceived already in
the seventeenth century.
Here Leibniz stands out as aware of possible limits to science. Like
Descartes he seems to have initially harbored the dream of a complete
theory of everything whose principles can be known a priori. But after a
certain point his vision changes. Then he begins to distinguish between
necessary and contingent propositions. As he characterizes them, the former
can be proved in a finite number of steps by reducing them through analysis
of the involved concepts to identical propositions or primary principles,
while the analysis of contingent propositions goes on ad infinitum. This is
why we cannot know the truth of contingent propositions a priori. God
alone can know this, not because he can complete the required infinite
analysis, but rather because he intuits the whole analysis with one glance.
The criterion for distinguishing necessary from contingent truths emerges from the
following feature, which only those who have in them a tincture of mathematics
will easily understand: in the case of necessary truths an identical equation will be
reached by carrying the analysis sufficiently far, which amounts to demonstrating
the truth with geometrical rigor; whereas in the case of contingent truths the
analysis proceeds to infinity, with reasons given for reasons, in such a way that
there is never a complete demonstration although the underlying reason for the
truth is always there, perfectly understood only by God, who, with one stroke of
thought, goes through the whole infinite series.9
Ignoring the overreaching rationalist ambition, we can transpose this view

to the relationship—as Leibniz envisages it—between nature as described
in fundamental physics and the appearances (whether physical measurement
outcomes or psychological phenomena). The assertion here is that given
the information that the physical state is thus or so, there is a strict
entailment of what the appearances must be. However, as we know
from meta-mathematics, not all entailments are capturable by definable
      297
consequence operations. So taken, this is precisely the ‘‘supervenience

without reduction’’ view, reading deduction for reduction.
As I emphasized early on, mere deducibility would also not ipso facto
satisfy the Appearance from Reality Criterion. But it is certainly a necessary
condition for success by this criterion. Remember after all that we are
not discussing criteria for God’s creation, nor for the structure of reality!
Our concern is with completeness criteria for the sciences in practice, which
evolve within the resources humanly available.
So analytic philosophy of mind is in effect offering cognitive science
the Great Leibnizian Escape move: there is a logical reduction of the
phenomena to the real, but it is not graspable by a finite mind (read: ‘‘not
definable by finitary or even recursive means . . .’’).
We will now examine the case of quantum mechanics in some detail.
The Leibnizian Escape move becomes pertinent only if we are already
willing to reject the demand that the Appearance from Reality Criterion
poses for science as a whole. But once we are so willing, we must not shy
away from it in the natural sciences either. If we are already willing to
grant that the appearances may be relatively autonomous, constrained by
but not derivable from the postulated theoretical reality, then we can next
ask whether, within a given theory, even something like a supervenience
thesis is tenable.
The quantum mechanics challenge

In quantum mechanics the physical systems are characterized by states
that change over time and physical quantities (‘‘observables’’) which are
represented by operators on the state space. The theory was developed
quite far before there was clarity on what is its vehicle for prediction.
The requisite rule was first introduced for the special case of scattering
by Max Born, and so is still called the Born Rule. We can state it in
several ways, but each takes the form of a prediction of outcomes, conditional
on the performance of a measurement. In one form it allows calculation of an
expectation value, which pertains to a weighted average in a large number
of measurements.
If observable A is measured on a system in quantum state ψ, the
expectation value of the outcome is ψ, Aψ
298  :   
In a simpler form, it specifies the probability of any given possible outcome

in a single measurement:
If observable A is measured on a system in quantum state ψ, and ψr
is the eigenstate of A corresponding to its possible value r, then the
probability of outcome r is (ψr . ψ)2
The details of calculation, and the simplification involved in these formu-
lations, have seen much discussion, but for our purposes we can leave these
aside for the most part.10
The dialectic that creates the problem

This Rule includes two terms that do not obviously belong to the vocabu-
lary of quantum mechanical—or even physical—description: ‘‘measured’’
and ‘‘outcome’’. As we have discussed at some length, measurement has a
physical correlate, and we must assume that this, rather than measurement
as such, is what the Rule is about. But then of course the question arises:
just what is that physical referent? The puzzle begins in all seriousness when
we ask what possible answers there could be within the quantum theory
itself.
1. Suppose first of all that ‘‘outcome’’ and/or ‘‘measurement’’ do not
refer to quantum states and/or their evolution at all.11
Then the theory is certainly incomplete with respect to what happens in
nature, no matter how complete within its own domain.
2. Suppose secondly that ‘‘outcome’’ and ‘‘measurement’’ refer respect-
ively to a final quantum state of the apparatus, and to the evolution
of the quantum state of apparatus + object during the interaction.
This supposition allows just three possible cases:
• the dynamics is incomplete, or
• it is complete and then the Born Rule is either
i. superfluous or
ii. inconsistent with it.12
The reason that it allows only these three is that quantum dynamics is
deterministic: the quantum state of an isolated system (whether Apparatus +
      299
Object, or Apparatus + Object + Environment) evolves deterministically.13

So if the dynamics is complete and correct, then the Born Rule must be
superfluous if ‘outcome’ and ‘measurement’ are supposed to be in the
domain of this dynamics.
The Born Rule cannot be deduced from the dynamics, so it is not
superfluous.14 We are thus left with a classic dilemma between incomplete-
ness and inconsistency; that is (one way to present) the famous Measurement
Problem.
States, appearances, and phenomena

Let us see for a moment how we can describe the situation in the terms of
the previous chapter. What matters is that measurement outcomes—in the
sense of the measurement outcome contents, rather than the outcomes as
observable physical events—are a prime examples of what we are to classify
as appearances. The quantum states are then the theoretically described
reality. The state of the system, among the possible states postulated by the
theory, provides the basis for prediction.
What about the middle term, the phenomena? These would be the
observable objects on which measurements are made, or any observable
events and processes targeted for quantum-mechanical explanation. This
includes the instruments involved in measurement, the physical events
that are the final states of such apparatus (which are meaningful as meas-
urement outcomes), and the entire set-up which is characterized as a
measurement.15
The Born Rule is one of conditional prediction. What it predicts is
what the appearances will be—with specified probabilities—under certain
conditions. The Copenhagen physicists astonished not only traditional
philosophers but also their colleagues by not recognizing, indeed refusing
to acknowledge, any need to close the apparent gaps in explanation. For
by itself the Born Rule certainly does not give any information about how
those appearances are produced. Can we, by looking into quantum theory,
find an answer to the question of how the measurement outcome comes
about? Does this scientific theory specify, explicitly or implicitly, a process, whether
deterministic or stochastic, by which this appearance is produced?
This is of course the question whether the Appearance from Real-
ity Criterion is satisfied. To answer that, the first task is to identify
300  :   
measurements and outcomes among the processes and events describable

in quantum mechanical terms. Having done that, along the lines displayed
more abstractly in the general theory of measurement, we can ask how
the Born Rule relates to the dynamical laws of that theory. It is then,
precisely, that we run headlong into the (in)famous Measurement Problem
of quantum mechanics.
Can the measurement outcome be ‘derived’ in the way we have been
discussing? What that means is this:
can we describe the process in which observable A is measured on a
system in quantum state ψ by a suitable apparatus, starting in its ‘ready
state’ ϕ and interacting with that system through a certain interval of
time, so as to show how the outcome is produced?
Even if we amend the last word ‘‘produced’’ to the phrase ‘‘produced with
probability . . .’’ there turns out to be a problem, as we’ll see. The reason
is, to announce it briefly, that the theoretical description of this interaction,
in quantum theoretical terms just does not seem to provide a place for the
specific outcome in question.16
Exploring the case of quantum mechanics

In quantum theory we have, it is often said, the most successful theory
in the history of science. As far as prediction goes, the riches gained have
been beyond the dreams of avarice. . . . But what about the explanation or
‘derivation’ of the predicted appearances?
Measurement in quantum mechanics17

Let’s begin by recalling the general format of a physical theory, that we
used in discussion of the general theory of measurement:
• a family M of observables (physical magnitudes) each with a range of
possible values;
• a set S of states;
• and a stochastic response function Psm for each m in M and s in S, which
is a probability measure on the range of m.
      301
The number Psm (E) is to be interpreted as the probability that a measurement

of m will yield a value in E, if performed when the state is s. In quantum
mechanics we must first of all respect the distinction between pure and
mixed states:
If s and u are states of a system X, and 0 < c < 1, then there is a
state v of S such that for all observables A pertaining to X, and all
intervals E,
PAv (E) = cPAs (E) + (1-c)PAu (E).
We call v a mixture of s and u in proportions (‘weights’) c and 1-c. If
there are distinct states of which v is a mixture, then it is a mixed state,
and if not then it is pure.
The Greek letters ψ, ϕ, . . . are customarily used for pure states. Usually
mixtures are introduced, to begin at least, to represent ignorance—and
they are suited to doing so even in quantum mechanics. But in that theory
that is not their only role, and in general we cannot equate ‘‘X is in a
mixture of ψ and ϕ’’ with ‘‘X is really either in ψ or ϕ’’.
Secondly, we need to pay attention to how systems can be parts of
other systems; for example, if X and Y are systems, so is their com-
posite X + Y. The states of a composite are related to those of its
parts, but are not determined by it (‘‘quantum holism’’). To explain that
we have to distinguish mixtures from another sort of combination of
states:
pure state ϕ of X is a superposition of pure states {ϕ(1), . . ., ϕ(N)} of X
if and only if for all intervals E and all observables A pertaining to X:
if PAϕ(1) (E) = 1 and . . . and PAϕ(N) (E) = 1 then PAϕ (E) = 1.
Note that mixtures of pure states are not pure, while superpositions are.
And finally we need to introduce a way in which composite pure states
can be constructed:
If systems X and Y in states ψ, ϕ, respectively have not interacted,
then the state of X + Y is the tensor product ψ ⊗ ϕ
But in general, and specifically if they have been interacting, the state
of X + Y is some superposition of states of that form (We speak then
302  :   
of ‘‘entanglement’’, ‘‘entangled state’’.) We’ll look at this again when we

come to the end-state of a measurement interaction.
Now we can describe the standard form of a measurement process in
quantum mechanics. The Apparatus to measure observable A starts in its
ready state ψ, say, and the Object on which A is being measured starts in its
initial state init. At this moment the system Apparatus + Object is in state
ψ ⊗ init. Now the two are coupled, and we recall that the criterion for the
physical correlate of measurement must be satisfied:
Criterion for the Physical Correlate of Measurement: PBfin (E) = PAinit (E)
where fin is the final state of the apparatus and B the ‘pointer observable’.
The form of interaction, or in other words, the evolution of the system
Apparatus + Object during the appropriate interval, must guarantee that
this will be so. It must guarantee that, regardless of what the Object’s
initial state init was (within the range within which the Apparatus can
operate).
There is a special case, when the outcome of a measurement is certain,
for which we can now introduce some shorthand notation (in the style of
Dirac):
When PA s ({r}) = 1, then s is called an eigenstate of A, corresponding
to eigenvalue r, and is denoted |A, r.
To satisfy the Criterion for the physical correlate of measurement, we need
at least the following to be the case:
If init is an eigenstate of A, |A, r then fin must be a corresponding
eigenstate |B, r of B.
It is also required, in the standard form of measurement, that an eigenstate
of A is not disturbed or changed by measurement of A (although other
states can be so changed). So this means that the evolution of Apparatus +
Object must be of such a character that
|ψ ⊗ |A, r evolves into |B, r ⊗ |A, r, for any eigenvalue r of A
And now the peculiarities of quantum theory take over. The fact that
evolution of system is, in the absence of external disturbances, unitary
guarantees the following:
      303
If |ϕ is a superposition of {|A, r(1), . . ., |A, r(N)} then |ψ ⊗ |ϕ

evolves into a superposition of {|B, r(1) ⊗ |A, r(1), . . ., |B, r(n) ⊗
|A, r(N)}
We are almost home; there is a relevant theorem that derives states for parts
from states of the whole. At this point one can deduce the character of the
Apparatus’s final state fin:
fin is a mixture of {|B, r(l), . . . , |B, r(N)}
This may look very nice! Couldn’t we say that this means that fin really
is one of the pure states {|B, r(1), . . ., |B, r(N)}? Couldn’t we add that
the measurement outcome is that A was measured to have value r(J) just
precisely if fin was |B, r(J)?
The adder in the grass appears precisely here. The answer is NO, we
cannot say that. The mixed state in which the Apparatus ends up is not
identical with any pure state.18 If we knew that an object was really in
some pure state, but did not know which, then it would be fine to attribute
a mixed state to that object, and use that as a basis for prediction. But the
converse is not the case; that a system ends up in a mixed state cannot be
equated in general with its really ending up in some pure state. Mixed
states are sui generis. And what is more, it is precisely states produced by
interactions such as happen in measurement that are demonstrably not
‘‘ignorance cases’’. This point can be richly illustrated with recombination
experiments.19 The idea for such experiments was originally suggested by
Eugene Wigner.20 In such an experiment, the reconstituted beam could be
once more in a coherent state. So a physical measurement interaction need
not, in itself, be such that the coherence is destroyed by the separation
and recombination. Therefore no information of entanglement was lost, as
would happen in a collapse. So now the perplexity can be stated as follows:
There is nothing we can see in the quantum mechanical description

of the end of the interaction that could be offered as equivalent to
the statement ‘‘the outcome of the measurement of A was the Jth
eigenvalue of A’’.
Yet it would seem that any single measurement would have one such
definite outcome.
304  :   
Supervenience?
Could the distinction between supervenience and reduction introduce
some leeway here? Suppose that the quantum mechanical representation
of nature, including dynamics, is complete, and that our statement ‘‘A is
measured on X during interval ’’ is true if and only if the pertinent
quantum states and their evolution during belong to some set depending
on A, X and ; call it S(A, X, ). So far that supposition is just a
stipulation, harmless under this supposition. But one could add: we humans
are quite good at recognizing whether or not the real situation is thus, but
S(A, X, ) is not definable within the language of quantum mechanics.
There are in general many such indescribable, undefinable sets for any
theoretical language no matter how rich we make it. Can we tenably add
that at the same time the relevant sets are describable perfectly well in the
ordinary language used also by laboratory assistants fairly ignorant of the
theory? Yes, we can, at least logically speaking, for there is the option that
the discourse to which ‘‘A is measured on X during interval ’’ belongs
may be precisely supervenient on, but irreducible to, that theoretical language.
But in fact, this simple move does not by itself bring us out of the
woods. While the logician may explore this loophole in abstract terms,
the physicist has a serious obstacle in store. For it certainly seems, both in
practice and in theory, that the physical conditions of measurement and
the physical correlate of the measurement outcome (the final state of the
apparatus) are perfectly well formulable in quantum mechanical terms. So
the obvious question is whether that physical correlate could be the same
even if the measurement outcome contents (the appearance) were different.
And that, unfortunately, seems to be obviously so: all the cases in which a
measurement of A has outcome value j, for j = 1, 2, 3, . . ., on a particular
occasion when none of them were certain, are cases in which the final state
of the apparatus is the same mixture of its possible pointer states. So the
appearances do not supervene on the physical state.21
An empiricist view
The view that I will now advocate will not be surprising given the way
I stated the perplexity in the preceding paragraph. But it does require
      305
appreciating an attitude toward theory that has been mainly absent from
foundational studies of quantum mechanics.
For unqualified adequacy of the theory, what is required is that the
surface models of phenomena fit properly with or into the theoretical
models. The surface models will provide probability functions for events
that are classified as outcomes in situations classified as measurements
of given observables. Those probability functions need to be parts of
the theoretically specified Born probabilities for the same situation as
theoretically represented in terms of possible states and evolutions.22 The
matching required is between two families of probability functions, not
between the individual events summarized in the surface model and the
states represented in the theoretical model.
But is the identification of actual events in measurement situations—so
classified in practice—and states in the theoretical models for those situ-
ations not presupposed by the question of whether the surface models fit
with the theoretical models?
There is a good question hidden in this challenge, but also a presup-
position of its own. We can view the situation ‘‘from within’’ and point
out that in the laboratory practice, as it has evolved in fact, a given process
is classified as a measurement of a certain sort. Then, viewing the matter
‘‘from above’’, the theory must be able to classify it as a physical correlate of
that sort of measurement. That will restrict the class of pertinent theoretical
models. The observable said to be measured is to be represented by an
operator of a particular kind—Hermitean, or at least bounded, etc.—on a
given state space, and the apparatus as capable of a certain range of states
represented by elements of such a space, and so forth. But none of this entails
that what happens in the actual situation must be displayed as entirely identifiable
in the theoretical model.23 The most stringent demand that can be made here
is that the relative frequencies of certain events in this sort of situation
must have a good fit to probability functions, extrapolated from them in
surface models, which are identifiable as parts of corresponding probability
functions in the theoretical models.
When this demand is met—whether strictly or to some approxima-
tion—the theory is borne out by the experimental results, and can be
used to make predictions. Take for example the Stern–Gerlach experiment
that we briefly looked at above. Suppose that with a given source, the
frequency counts in the upper and lower channel show a proportion 2:1.
306  :   
Starting with the assumption that the usual sort of theoretical model fits this
situation, the range of states possibly prepared by the sources is restricted
by the condition
PAs ({‘up’}) : PAs ({‘down’}) = 2:1
where A is the pertinent observable (spin in vertical direction, say). This

means that if the initial state is pure, it is a superposition of two eigenstates
with coefficients whose squares are 2/3 and 1/3. Alternatively the state
could be 2:1 mixture of those eigenstates. A further measurement (for
spin along a different direction) can decide between these two possibilities.
With the information about the state prepared by the source, we can go to
the theoretician for a prediction of outcomes if the magnet array is rotated
with respect to the source, or if a new such array replaces the counter at
the end of the upper channel, and so forth.
The prediction will be that the frequency counts fit the relevant
new Born probabilities in a theoretical model that satisfies the above
condition.
And that is all.
This point was summarized brilliantly if cryptically by Hertz in the
famous passage, that I already mentioned in connection with the Bild-
theorie of science, where he discusses modeling, in his Prinzipien der
Mechanik. The representation provided by the theory must be such that
the ‘‘intellectually necessary consequences’’ of the representation must
in turn represent the ‘‘naturally necessary consequences’’ of the system.
Now, with our eyes on quantum theory, we can read this in a new
light:
We form for ourselves inner pictures or symbols of external objects; and the form
which we give them is such that the necessary consequences of the pictures in
thought are always the pictures of the necessary consequences in nature of the
things pictured . . . .
The pictures which we here speak of are our conceptions of things. With
the things themselves they are in conformity in one important respect, namely in
satisfying the above requirement. For our purpose it is not necessary that they
should be in conformity with the things in any other respect whatever. (Hertz
1894/1956: 1–2; see also p. 177)
      307
Anti-empiricist confusion
There is a charge sometimes made against quantum theory itself, as typically
presented in physics textbooks, but also against any empiricist view of
physics. That is that the theory is made out to be about measurements
rather than about nature, about what people [will] actually find rather than
about what happens. But that was not what physics was meant to be! After
all, it is said, there are events in the stratosphere, where no measuring
instruments are present, and physics applies to those events as well.
Certainly it does! But the objection trades on a stunted impression of
what is meant, either by such a textbook presentation of the theory, or by
an empiricist reading of it. First of all, the observable phenomena do not
include just actual measurements and their outcomes. They include all the
observable entities—objects, events, processes—that there are, have been, or
will be, whether observed or measured or not. Secondly, the theory provides a
representation of the whole of nature, which includes both observable and
unobservable parts. An empiricist will emphasize the term ‘‘representation’’
here, and recall all the ways in which a representation may only partially,
and perhaps distortedly, mirror anything real.
So the empiricist construal does not contradict the assertion that the
theory is about all that can happen, and not just about measurements.
Thirdly, the predictions to whose accuracy the theory is directly accountable
are predictions of what the appearances will be when something appears, i.e.
is measured. But it would be an oxymoron (or anthropocentric metaphor)
to speak of accountability to what cannot be checked in practice, to what
cannot be accounted.
In the case of quantum theory it is sometimes objected that we can check,
by actual measurement as well as with our own eyes, that the pointer on
a given dial coincides with a particular number, whereas any state we can
ascribe to that object will not be localized in a finite region of space for more
than an instant. The conclusion drawn is that observable things are after all
not accommodated in the theory: the correct representation of the pointer
is as an object wholly in a small spatial region for an extended period of
time and this we find nowhere in the quantum theoretical representation.
But this objection confuses the three levels: Theorica, phenomena, and
appearances. What is subjected to a position measurement, may well be
an observable object, yes. But the way it appears when it is photographed
308  :   
or measured in some other way, is its representation by a measurement

outcome. It is not incumbent on the theory to represent it in the same way.
(In other words, the Appearance from Reality Criterion is not an inherent
constraint on physical theory, but one that is at best imposed on it.) It is
incumbent on the theory only to predict what its appearances will be like,
and that it can do via the Born Rule to the extent needed in empirical
applications: by providing probabilities and expectation values.
This will still be very puzzling if you think of ‘Theorica’ as telling it like
it is, if you identify reality with the theoretically postulated reality. Think
of this, however, as an empiricist must: the theoretical representation, in
which no object state is localized in a finite region of space for more than
an instant, is only theory. As is all of our theoretical science. The reality to
which it is accountable is only the observable part of the world, and that
implies for us that what it is in practice directly accountable to are the
appearances—the outcomes of the measurements and observations that are
actually made.
Advocates of the Appearance from Reality Criterion will not be satisfied
with quantum mechanics. Some of the resistance to that theory may be
explainable in this way. Even more so the attempts to provide the theory
with an interpretation that restores obedience to the Criterion at some
‘deeper’ level. Conversely, irenic acceptance of the theory—such as that
which we have seen prevalent in the physics community, throughout
the last century—would seem to signal an attitude content without any
sustained attempt to satisfy that Criterion. This, it seems to me, should
allow us to draw the right moral about what are and what are not norms
that govern scientific practice.
APPENDIX TO CHAPTER 1
Models and theories as

representations
Throughout this book, when I discuss theories and models I remain within what is
known as the semantic approach to (or, semantic view of) science. That approach
has lately been contested by various writers (Suarez 2003, Frigg 2006, and Morrison
2007, for example) on the basis that the concept of model it involves is not in
harmony with what models are. The tension displayed in these critiques, I submit,
is not actually due so much to a different concept of models as to different interests
in the same subject.
On the semantic view, a theory offers us a large range of models—in fact,
while a theory may have many different formulations, its set of models is what is
important.1 If a theory is advocated then the claim made is that these models can be
used to represent the phenomena, and to represent them accurately. A model can
(be used to) represent a given phenomenon accurately only if it has a substructure
isomorphic to that phenomenon.2 (That structural relationship to the phenomenon
is of course not what makes it a representation, but what makes it accurate: it is
its role in use that bestows the representational role.) A theory may therefore be
taken to represent its domain as thus or so in the sense that the models it makes
available for the representation of phenomena in that domain are thus or so.
One topic that was never entirely settled in this literature is just what a theory is.
Patrick Suppes, one of the fathers of the approach, identified Newtonian particle
mechanics with the set, which he defined, of Newtonian particle systems—in
effect, with the theory’s set of models. This was inspired by the form of con-
temporary mathematics: Euclidean geometry is simply the study of the set of
Euclidean spaces, and presenting that geometry consists in nothing more than a
definition of that set (cf. Blumenthal 1980). But of course, Euclidean geometry is a
mathematical theory, while Newtonian mechanics is a physical theory, so it seems
that something is left out.
This was quickly enough appreciated, and various additions were made. Thus I
added (van Fraassen 1970) a semi-interpreted language of ‘‘elementary statements’’;
Ronald Giere (1985, 1988) identified a theory with a combination of a theoretical
definition (of the set of models) and theoretical hypotheses (relating the models
to the domain of application). In a longer discussion (van Fraassen 1991: ch. 1)
I discussed Giere’s scheme in the light of some criticisms by Nancy Cartwright.
While not settling on any official definition of what a theory is, I emphasized that
310 
it must be the sort of thing that can be believed, disbelieved, doubted, and so forth.
Thompson 2007 is the latest comprehensive account of the development of the
semantic approach, with special emphasis on its uses (in Thompson 1989 and Lloyd
1984, 1986, 1987, 1994) to theories in biology, and contrasts various conceptions of
theories and models. Lately Margaret Morrison (2007) has emphasized that despite
the centrality of models for the understanding of science, important roles remain
for theories, though also without offering an ‘‘official’’ codification of just what a
theory is.
Nancy Cartwright, Towfic Shomar, and Mauricio Suarez (1995) developed a
more instrumental view of theories, as providing tools for the construction of
models. At first blush it seems that our conceptions of theory and model must be
quite different, if on their view a model is something constructed and a theory a
tool. But in fact I think that a closer look reveals that our conceptions are entirely
compatible; what we concentrate on signals differences in interest and focus.
The sense in which a theory offers or presents us with a family of models—the
theoretical models—is just the sense in which a set of equations presents us with
the set of its own solutions. In many cases, no solutions to a given equation are
historically found or constructed for a very long time . . . though mathematically
speaking, they exist all along. When the equations formulate a scientific theory,
their solutions are the models of that theory. In this sense, Newtonian mechanics
had in its range of models already solutions to the three-body problem, though the
scientists following Newton, not being logically omniscient, could not see what
it was.3
As Suarez and Cartwright (forthcoming) discuss, the process of actually and
historically arriving at a working model can take several forms. Only rarely is
it a matter of simply instantiating the theory to some specific values of some
parameters. In some cases, emphasized especially in Ronald Giere’s elaboration of
the semantic view, the process is one of ‘‘de-idealization’’. That is, the scientist
wishing to construct a model of an actual situation or phenomenon begins with a
very idealized model, and then adds more factors characterizing the real situation
until a good enough representation of the phenomena is achieved. An example
would be to model a real pendulum starting with a model of one in which friction
is ignored, thus an ideal case, and then add force terms that correspond to the
friction in the actual situation.
But there are also processes of model construction in which we see imaginative
leaps that are not like that—de-idealization is not at all the only tactic to be found
in model construction. For example, the construction of a model of superconduct-
ivity by the London brothers—a much discussed example, and central to those
critical discussions of the semantic view—involved a change from an analogy with
ferromagnets to an analogy with diamagnets.
Nevertheless, what is arrived at is a model of the background theory. If the ideal
pendulum is a Newtonian model, and it is de-idealized, then the result is still a
Newtonian model. For if it is not, then it is a counter-example to Newton’s theory,
    311
and hence not a de-idealization. Historically this could happen as well—with the
resulting model not being a de-idealization and not a solution to the problem
as originally defined—and sometimes famously does. For example, there seemed
after all to be no way to arrive at such a de-idealization that matched the actual
advance in the perihelion of Mercury. But the arrival of such an anomaly signals
revolutionary change.
On the other hand, if the imaginative leap succeeds in solving the originally posed
problem, where de-idealization guided by an earlier analogy did not work, then
again the success consists in arriving at a model of the situation that is not a counter-
example to the background theory. Thus, as Suarez and Cartwright also emphasize,
the London brothers’ model of superconductivity is still a model of Maxwell’s
electrodynamics, the background theory within which they were working.
So what accounts for the apparent tension in our two very different discussions
of modeling and representing? We find the answer in the precise complaint that
Suarez and Cartwright voice. To display the differences, they give their own
‘‘structural account’’ of the superconductivity story, which is quite like that in
various versions of the semantic approach and then write:
this framework can account for the piecemeal borrowing characteristic of the practice of
modeling precisely because it leaves out of the description everything that is of interest for
our thesis in ‘‘The Toolbox’’, namely the actual reasons scientists advance for building a
new model. The framework leaves out precisely the Londons’ reasoning in deriving the
London model out of the acceleration equation theory—which is what we have been
claiming is crucial to assess the theory-driven view. Hence we see that an appropriate
structural characterisation of this practice can at best describe correctly the products of the
modeling practice (i.e. the models themselves); while necessarily leaving out the intellectual
processes that lead to those models. (loc. cit.)
Quite right! Structural relations among models form a subject far removed from
the intellectual processes that lead to those models in the actual course of scientific
practice. So what is important depends on what is of one’s interest. Accordingly,
if one’s interest is not in those structural relations but in the intellectual processes
that lead to those models, then the semantic view of theories is nowhere near
enough to pursue one’s interests.
Both interests are important, it seems to me; leaving out either one we would
leave much of science un-understood. But the basic concepts of theory and
model—as opposed to the historical and intellectual dealing with formulations of
theories, or on the contrary, construction of models—do not seem to me very
different in the two approaches to understanding science.
In Part II I take up this divergence in interests with respect to measurement,
which is also studied in two very different—not incompatible but complement-
ary—ways: ‘from above’ and ‘from within’. That is a specific, and in some ways
simpler, instance of the same issue, since measurement too is representation.
Quantum peculiarities: fuzzy

observables
The conclusions about measurement in Part II are general, though quantum
mechanical examples motivated the leeway to be left for general conditions
of measurement. In the foundations of quantum mechanics we find furthermore
specific features that we cannot attribute to physical theory generally. They provide
further grist for our mill if we want to press the point that a measurement outcome
does not show what the target is like, as opposed to what it ‘‘looks like’’ in that
particular measurement set-up. In classical contexts the significance of this point
is minimized, by the presumed possibilities of so varying and combining simple
measurements into complex joint measurements that the limitations are effectively
transcended.
It is now customary in the literature on quantum theory to distinguish ‘‘sharp’’
observables from a generalization that encompasses also ‘‘fuzzy’’ or ‘‘unsharp’’ or
‘‘smeared out’’ observables.4 The former, which are the familiar sort, have (in their
simplest form) real numbers as their possible values. We shall restrict ourselves to
these for a bit, and then go to the more general category.
We consider to begin an observable that is discrete: its set of possible values is
not a continuum but a discrete set of values. If E is a set of possible values of sharp
discrete observable A then there is a state s such that the conditional probability
PsA (E) = 1. So if A can possibly have a certain value, or value in a given range, then
there is a measurement situation in which that will appear with certainty. When E
contains just a single possible value of A then such a state s is called an eigenstate of
A. But it is quite easy to find sharp observables A and B which are incompatible
in the sense that they have no eigenstate in common. Then there is in that case no
measurement apparatus or set up such that the Criterion for the Physical Correlate
of both A and B measurements could be satisfied simultaneously—A and B are
mutually incompatible.5
This incompatibility relation will have a familiar sound, because of the well-
known uncertainty relations for position and momentum which are often
the first curiosity of quantum mechanics anyone comes across. The famil-
iar sound is somewhat deceptive. For position and momentum observables
are indeed sharp observables with real number values; however, they are not
    313
discrete but continuous, they have no eigenstates. Still the somewhat weaker
requirement:
If E is an interval of possible values of sharp observable A then there is for
each positive number δ < 1 a state s such that the conditional probability
PAs (E) > δ
holds for them as well. Their incompatibility means now of course that this will
not be the case for both in the same states for the same probabilities—intuitively,
δ will have to decrease for one if it increases for the other.
Here we recall the notions of occlusion and ‘‘explicitly non-committal’’ pic-
turing. The extent to which measurement of one observable can be revealing,
however indirectly, will limit the extent to which it can be revealing about certain
other observables. (The analogy is fragile, however: if the unmeasured observable
has no value, then its value is not hidden from sight!6 ) That applies to both discrete
and continuous sharp observables.
But what is then actually, really measured in a measurement of position or
momentum? It is entirely unrealistic to think that the outcome could be a real
number clearly distinct from any nearby numbers, or that in a series of measure-
ments there could be a clear distinction between a frequency of 1 and of δ for any
and all δ < 1. So either the outcomes of real position or momentum measurements
should be reconceived, or we should reconceive just what observables are being
measured. It is the latter option that arrives with the generalized observables
allowing for ‘unsharpness’.
I will explain the basic idea here, but for the simple case of an observ-
able whose possible values form a finite set RA = {v1 , . . . , vN }. The func-
tion PAs assigns a probability to each of these values (the probability that it
would be found if A were measured on an object in state s). But now sup-
pose that these values as not being sharply distinguishable in measurement, so
that if a measurement outcome indicates value v3 , for example, we should
conclude that we have ‘really’ found values v2 , v3 , or v4 , with probabilit-
ies p2 , p3 , or p4 . Or, to be more operational in the gloss we give it: if
the outcome indicates value v3 , for example, we expect that immediately
repeated measurements would find v2 , v3 , or v4 , with probabilities p2 , p3 , or
p4 .
Let me put this a bit more generally. The function p, which is thus associated
with value v3 , has brothers, so to speak, similarly associated with each of the other
values in RA. Let’s therefore write, not simply p, but p[v3 ]. This association is a
confidence mapping. Then we can think of identifying another observable B such
that PBs assigns to v3 the probability PAs (v2 ) + PAs (v3 ) + PAs (v4 ), and similarly for the
other values and associated confidence mappings. Then B is called a fuzzy version
of A.7
For B to be a fuzzy version of A, the indicated relationship has to hold of
course for each of the values in RA and their associated smearings out by a particular
314 
confidence mapping; and this has to hold for all states s. This may not make sense,
intuitively. One way to think of it is to regard B as having values that are not
numbers but probability distributions over numbers (possible values of A)—such
values are sometimes called ‘‘fuzzy sets’’.8 The fuzzy versions of sharp observables
are in general not sharp, and in quantum theory are represented by a different sort
of operator.
A theorem on this subject establishes that in any standard measurement of an
observable, in the way that was described above, with the Criterion for a Physical
Correlate of Measurement being satisfied, what is really measured is always a fuzzy
version of an observable.9 To see this, the Criterion for the Physical Correlate
of Measurement is applied in reverse, as it were. Imagine that the specifications
of a measurement set-up, including the pointer observable B and the unitary
operator that governs the measurement interaction, are given. Now ask: what is
the observable A such that the equation
B
Pfin (E) = Pinit
A
(E)
is satisfied, for all initial states init? The found answer is that observable A,
so identified, is then a fuzzy version of some sharp observable, and the two
are the same only if A is discrete. In the case of continuous observables
such as position and momentum, that fuzzy version can not be the same as
the ‘original’, that is, as the observable that is putatively the target of the
measurement.10
Surface models and their

embeddings
As mentioned in the text, literature on the foundations of quantum mechanics
typically focuses on small structures that represent data, and they do not in general
have such a familiar form as e.g. Stevens was considering. They are not always
algebras, but more generally partial algebras, or just partially ordered sets (posets)with
some relations and/or operations. That generalization is needed because of the
presence of incompatible observables, which is foreign to classical physics. The
reason for the more general, more liberal, notion of the structures that can emerge
from measurement is that here one is forced to attend to measurements which
require incompatible set-ups and are not even in principle, not even in the ideal
limit, combinable into single measurements.
 (continued from the section The surface model of an EPR experiment in
Chapter 7)
One way to report the findings in the experiments above is to note that in
no surface model so obtained do Lxa and Rxa both receive the score T when
they can receive an informative score at all (i.e., when the preconditions Lx
and Rx obtain). We then call Lxa and Rxa orthogonal.
If we keep fixed this generalization about all obtainable surface models, then
Rxl must receive score T when Lx0 does (modulo probability zero), again
when the informative scoring conditions obtain; and we call that implication.
The latter is a partial ordering, naturally symbolized as ‘‘≤’’. So what we
found above, the family of propositions that register the possible experimental
outcomes, is a partially ordered set with an orthogonality relation.
Reflection on this form of representation leads to assertions of the form: ‘‘all
data models can take the form . . . ’’ and there are various proposals for what
this mandatory general form must be. It is important to note that structures
taking this form may not in general be ‘classical’. How can we view the measuring
procedure, thus conceived, as locating the system investigated in a logical space?
.
A. R. Marlow (1978, 1980) used the concept of dual poset: a partially ordered
set with zero element and a relation of orthogonality, plus a single operation,
duality, symbolized as *. This operation has the properties that x** = x*,
316 
x ≤ y only if y* ≤ x*, and x is orthogonal to y if and only if x ≤ y*. Duality

is a generalization of the idea of a complement or negation.
In the world of mathematical entities there are many dual posets. What must a
theoretical framework be like if it is to provide models that can have even very
strange dual posets of experimental propositions embedded?
The embedding must be good, that is, we must be able to see in the theoretical
model all the significant features of those ‘empirical algebras’.
Think back for example to the above type of experimental report, for an EPR
experiment, labeled S. Such a report starts ‘‘With initial preparation X, . . . ’’ and
then mentions probabilities. These probabilities characterize what is called the state
prepared by procedure X. Such states look like fragments of ordinary probability
functions, in that they assign probabilities only to propositions for which the
informative scoring conditions obtain. (For example, a probability is assigned to
L20 only conditional on L2.) If the theoretical model is to accommodate these
probabilities then they must be ‘visible’ in a certain sense in the computational
structure of that model.
 . Here is Marlow’s theorem. It requires two preliminary

definitions:
A probability function on a dual poset is any function f with the properties that
it assigns 0 to the zero element, that f (x*) = 1 − f (x), and that f (x) ≤ f (y)
if x implies y.
A base for the dual poset is a set of elements that contains either x or x* for
each element x, but does not contain elements orthogonal to each other.11
The theorem says now that if we have a dual poset and a base, we can embed the
poset in the algebra of projection operators on a Hilbert space, in such a way that:
(i) duality becomes orthocomplementation (as defined on these projections),
(ii) the partial ordering is preserved for elements within the base,
(iii) each probability function on the poset can be associated with a vector and
becomes calculable by means of the familiar trace computation used in
quantum mechanics.
This is a beautiful result, and one could take it to provide good reason to think
that we have here a plausible filling out of the contention ‘‘all data models can
take the form . . . ’’.
A good reason is not necessarily conclusive, though, and actually in some ways
the theorem is less powerful than it looks (the implication order is preserved only
within the base!) and in some ways less than informative (there are enormous
Hilbert spaces with room to embed almost anything).12
What the theorem does illustrate very well is the form in which phenomena,
depicted in data models, are located in logical spaces provided by theoretical
modeling.
Retreat (?) from The Scientific

Image
Readers of The Scientific Image may wonder how the view of science outlined in
Part IV, on Appearance and Reality (but specifically as pertaining to quantum
mechanics) relates to the view advanced there. Not surprisingly: it was with respect
to probabilistic theories in science—and hence, with respect to recent physical
theory überhaupt—that I had a change in view. At the earlier stage I subscribed
to a moderate frequentist interpretation of probability. I was under the impression
that empirical adequacy could be defined for probabilistic theories more or less in the
same way as for the rest. That impression was mistaken. Let me rehearse some of
the details here. To begin there was the following general pattern, to which I still
hold, if not taken too strictly:
Science aims to give us theories which are empirically adequate; and acceptance of a
theory involves as belief only that it is empirically adequate. This is the statement of
the anti-realist position I advocate; I shall call it constructive empiricism. (1980: 10)
What was meant by ‘‘empirically adequate’’ was explained first of all for what
had been the mainstay of modern physics until, arguably, the twentieth century:
theories in which no probabilities occur. As main example I took Newton’s
mechanics and theory of gravity, as depicted in the semantic approach. I will here
quote the representative passage, but let’s remark at once that I did not at the time
distinguish clearly the observable phenomena from the appearances:
The ‘apparent motions’ form relational structures defined by measuring relative distances,
time intervals, and angles of separation. For brevity, let us call these relational structures
appearances. In the mathematical model provided by Newton’s theory, bodies are located in
Absolute Space, in which they have real or absolute motions. But within these models we
can define structures that are meant to be exact reflections of those appearances, and are, as
Newton says, identifiable as differences between true motions. These structures, defined in
terms of the relevant relations between absolute locations and absolute times, which are the
appropriate parts of Newton’s models, I shall call motions, borrowing Simon’s term. (Later I
shall use the more general term empirical substructures.)
When Newton claims empirical adequacy for his theory, he is claiming that his theory
has some model such that all actual appearances are identifiable with (isomorphic to) motions in
that model. (This refers of course to all actual appearances throughout the history of the
universe, and whether in fact observed or not.) (Ibid.: 45)
318 
Indeed, Newton’s ‘apparent motions’ are relative motions, and they are precisely
as they can be determined to be by measuring relative distances, time intervals,
and angles of separation. So although we can draw the distinction between
phenomenon and appearance at the conceptual level, that distinction does not in
this case make a difference in practice.
What then of the case where the theory involves probabilities? Thinking of the
chi-squared tables and the like whereby statisticians assess the ‘fit’ of probabilistic
hypotheses to actual frequencies, I thought that we had to deal there with only
a small and manageable extension of the same relationship. This ‘fit’ extended
‘embed’, I thought. I proceeded to outline a ‘modal frequency interpretation’
of probability which related theoretical models to families of frequencies, so as
to show how and where the ‘fitting’ might be located as a relation between
phenomena and theoretical model. The failure of this approach was brought
home to me by the writings of Joseph Hanna (1983, 1984) and correspondence
with Zeno Swijtink. During the next decade in Princeton I came to abandon
any objective notion of probability, and to move toward something like Richard
Jeffrey’s radical probabilism in epistemology. ‘‘Belief ’’ was no longer the operative
term, but ‘‘opinion’’, the latter conceived of largely (though not entirely) in terms
of subjective probability. But now, how to understand probability in physics?
My understanding of the formal character of physical theory did not change:
a physical theory may have in it irreducible probability, which is truly the
new modality of science. What had to change rather was now to conceive of
the epistemic attitude of acceptance of a theory as empirically adequate. The new
understanding I proposed of how we relate epistemically to the probabilities
offered to us by a physical theory appeared then in Laws and Symmetry.
Belief must be replaced by the nuances of gradated opinion, modeled as personal probability.
The question becomes then: if we accept a theory, how do the probabilities it offers
guide our personal expectation? The answer I shall now begin to elaborate is: in the form of
Miller’s Principle. That is what constitutes acceptance. For someone who totally believes the
theory, that guidance will involve all theoretical probabilities. For the scientist qua scientist
(as described by empiricism) only the theory’s probabilities for observable phenomena will
play this guiding role. ‘Objective chance’ is in either case the honorary epithet we give to
the probabilities in theories we accept. (1989: 194)
To formalize this, I took my cue from Chaim Gaifman’s introduction of the idea
of expertise as guiding clue:
If P is my personal probability function, then q is an expert function for me concerning family F
of propositions exactly if P(A | q(A) = x) = x for all propositions A in family F. (Ibid: 198)
(This relationship is generalized in various ways for less simple cases.) The point is
that an expert is someone or something that guides and constrains one’s opinion
in a certain range. Thus, for example, to say that one accepts contemporary physics
without qualification, can now mean that one takes this physics as expert on
    319
the probabilities of observable events to which it assigns a probability. Outright

implications about what observable entities are like or how they behave or interact
can be treated as propositions which the theory ascribes probability 1, and thus
covered as a limiting case.13
When, as in the case of quantum theory, the probabilities provided are for
measurement outcomes, given that such measurements are made, there is nothing
in the theory to guide one’s opinion about the observable phenomena as opposed
to the appearances. That does not mean that someone who accepts quantum theory
cannot have opinions about the phenomena! Quite the contrary, we have such
opinions (and practices or policies to update them) already in place before we learn
quantum theory. The guidance we get from the theory is only about how things
will appear in situations classified relative to the theory as measurement set-ups of
certain kinds. That guidance will be vindicated, in the context of that classification,
if the measurement outcomes comply with the induced expectations.
Notes to Appendices
1 The semantic approach, or the semantic view of theories, has a number of

different versions and I try here for a weak formulation that does not distinguish
between them.
2 The exact meaning to be attached to this phrase, which has also been subject
to critique, is explored in Chapter 11.

3
This does not imply that there is, for each such problem, an exact solution
that scientists can find, even in principle, for there are mathematical structures not
describable by any finite or recursively specifiable sets of equations.
4 The former, familiar, ones are those represented by self-adjoint operators,
which correspond to projection valued measures, while the latter are represented
by a larger class of operators corresponding to a normalized positive operator
measure. For details see e.g. Busch and Lahti 1996; Bush, Lahti, and Mittelstaedt
1966; and more recently Dalla Chiara, Giuntini, and Greechie 2004.
5 We must be careful in how we understand this. Neither operational incom-
patibility nor operational compatibility offers us any logically sufficient or necessary

clue to theoretical significance. In discussing incompatibility of observables, we
are looking at the matter from above, that is, as classified within the theory. As
Grünbaum 1957 begins by pointing out, both in Heisenberg’s and Margenau’s
accounts we can find consistent operational prescriptions for the simultaneous
ascription of position and velocity. There is no logical contradiction, for example,
in taking the data of a sequence of position and time measurements on particles
emitted from a given source and ascribing a velocity during the relevant interval
by means of the classical formula, dividing distance covered by time elapsed (‘‘time
of flight measurements’’). But such ascriptions do not have the value that assertions
about physical magnitudes are meant to have, and those processes are not classified
as measurement processes by the theory. The prediction of a future position on
the basis of such an ascription of a current position and velocity will, according to
the quantum theory itself , not be verified.
   321
6 Thanks to Angela-Adeline Mendelovici for insisting on this. The uncertainty

principle bears a relation to occlusion, but only in the way that the quantum
mechanical ‘correspondence principle’ ever does, rupturing the very concepts that
it appeals to as analogues.
7
At this point, do not equate observables with what is represented by Hermitean
operators: admitting unsharp observables requires admitting a larger class of
operators for their representation (positive operator valued measures).
8 The term ‘‘fuzzy set’’ is due to Zadeh 1965. Cf. Heinonen 2005: 15–17.
9 Busch and Lahti 1996; see further Heinonen, Lahti, and Ylinen 2004, and the
discussion in Heinonen, 2005: 18.

10 See especially Busch and Lahti 1996: section 3.
11
Note that 0 is orthogonal to itself, and so is not in a base. Every set with
the second property can be extended to one that has the first as well. In view of
how we characterized orthogonality and implication at the outset, it follows that
any set of elements that have all received score T on a particular occasion cannot
have mutually orthogonal elements among its members. So what is obtained in
any experimental set-up of the above sort is part of a base.
12 Marlow realized this very well and attempted to add postulates to narrow
down this embarras de richesse so as to recover empirical content . My purpose here

is not so to draw attention to his specific program, but to illustrate the general
structure of measurement understood as the task of locating items (situations) in
logical spaces supplied by theories.
13 If I were to expand on this view now, I would draw on Stephen Leeds’
insightful [1984], which develops a related way to think about the quantum theory.
Bibliography
Addison, J. W., L. Henkin, and A. Tarski (eds.) [1965] The Theory of Models.
Amsterdam: North-Holland.
Aerts, D. (ed.) [1999] Einstein Meets Magritte: The White Book—An Interdisciplinary
Reflection. Dordrecht: Kluwer Academic Publishers.
Alberti, Leon Battista [1991] On Painting. Tr. C. Grayson, with an Introduction
and Notes by Martin Kemp. London: Penguin Books.
Alpers, Svetlana [1983] The Art of Describing: Dutch Art in the Seventeenth Century.
Chicago: University of Chicago Press.
Anderson, Ronald [1993] ‘‘The Referees’ Assessment of Faraday’s Electromagnetic
Induction Paper of 1831’’. Notes and Records of the Royal Society of London 47:
243–56.
Armstrong, David M. [1993] A Materialist Theory of Mind. London: Routledge &
Kegan Paul, 1968; revised edition with new preface, New York: Routledge.
and Norman Malcolm [1984] Consciousness and Causality. Oxford: Blackwell.
Asquith, Peter D. and Philip Kitcher (eds.) [1984] PSA 1984., vol. i. East Lansing,
MI: Philosophy of Science Association.
Bacon, Francis [1994] Novum Organum, With Other Parts of the Great Instauration.
Tr. and ed. P. Urbach and J. Gibson. Chicago: Open Court, 1994.
Baigrie, B. [1996] Picturing Knowledge: Historical and Philosophical Problems Concerning
the Use of Art in Science. Toronto: University of Toronto Press.
Baird, Davis [2003] ‘‘Thing knowledge: Outline of a materialist theory of know-
ledge’’, pp. 39–67 in Radder 2003.
[2004] Thing Knowledge: A Philosophy of Scientific Instruments. Berkeley:
University of California Press.
Barbour, Julian B. [2001] The Discovery of Dynamics: A Study from a Machian Point of
View of the Discovery and the Structure of Dynamical Theories. New York: Oxford
University Press.
 323
Barenblatt, Grigory I. [2003] Scaling. Cambridge: Cambridge University Press.

Barrow, John D. [1991] Theories Of Everything: The Quest for Ultimate Explanation.
Oxford: Oxford University Press.
Bartels, Andreas [2006] ‘‘Defending the structural concept of representation’’,
Theoria 55: 7–19.
Bell, J. S. [1990] ‘‘Against Measurements’’, pp. 17–32 in Miller (1990).
Benoist, Jocelyn [2006] ‘‘Intentionality As A Relation And Not, And The Inten-
tionality Of Perception’’, ms.
Berg-Hildebrand, Andreas and Christian Suhm [2006] Bas C. van Fraassen, The
Fortunes of Empiricism. Frankfurt: ontos verlag.
Bird, Alexander [2007] Nature’s Metaphysics: Laws and Properties. Oxford: Oxford
University Press.
Birkhoff, G. and J. von Neumann [1936] ‘‘The logic of quantum mechanics’’,
Annals of Mathematics 37: 823–43. Reprinted pp. 1–26 in Hooker 1975.
Bitbol, Michel [forthcoming a] ‘‘La structure quantique de la connaissance indi-
viduelle et sociale’’. In Bitbol forthcoming b.
[1996] Schrödinger’s Philosophy of Quantum Mechanics. Dordrecht: Kluwer
Academic Publishers.
(ed.) [forthcoming b] Théorie Quantique Et Sciences Humaines. Paris: CNRS
Editions.
Blackmore, John [1999] ‘‘Boltzmann and Epistemology’’, Synthese 119: 157–89.
(ed.) [1955] Ludwig Boltzmann His Later Life and Philosophy 1900–1906, A
Documentary History. Dordrecht: Kluwer Academic Publishers.
(ed.) [1992] Ernst Mach—A Deeper Look. Boston Studies in the Philosophy of
Science, vol. 143. Dordrecht: Kluwer Academic Publishers.
S. Tanaka, and R. Itagaki, (eds.) [2001] Ernst Mach’s Vienna 1895–1930. Or
Phenomenalism as Philosophy of Science. Dordrecht: Kluwer Academic Publishers.
Block, Ned [1981] Imagery. Cambridge, Mass.: MIT Press.
[1983a] ‘‘Mental pictures and cognitive science’’, Philosophical Review 92:
499–541.
[1983b] ‘‘The photographic fallacy in the debate about mental imagery’’,
Noûs 17: 651–61.
Blumenthal, Leonard M. [1980] A Modern View of Geometry. 2nd edn. New York:
Dover.
Bogen, James and James Woodward [1988] ‘‘Saving the phenomena’’, Philosophical
Review 97: 303–52.
Boghossian, Paul and Christopher Peacocke (eds.) [2000] New Essays on the A
Priori. Oxford: Oxford University Press.
Bohm, David [1957] Causality and Chance in Modern Physics. London: Routledge
& Kegan Paul, 1957.
324 
Boltzmann, Ludwig [1902] ‘‘Model’’ (his Encyclopedia Britannica article of 1902),

reprinted in Boltzmann 1974.
[1905] Populaere Schriften. Leipzig: J. A. Barth.
[1960] ‘‘Theories as representations’’ (trans. of pp. 253–69 in L. Boltzmann
1905), pp. 245–5 in Danto and Morgenbesser 1960.
[1974] Theoretical physics and philosophical problems: selected writings. Ed. Brian
McGuinness. Dordrecht: Reidel.
Boon, Mieke [2004] ‘‘Technological instruments in scientific experimentation’’,
International Studies in the Philosophy of Science, 18: 221–30.
Boyle, Robert [1772] The Works of the Honourable Robert Boyle: in Six Volumes. Ed.
Thomas Birch, London.
Brading, K. and E. Castellani (eds.) [2003] Symmetries in Physics: Philosophical
Reflections. Cambridge: Cambridge University Press.
Bradley, F. H. [1930] Appearance and Reality. Oxford: Oxford University Press.
Braunstein, S. L. (ed.) [1998] Quantum computation: where do we want to go tomorrow?
Weinheim: Wiley VCH.
Bretislav, Friedrich and Dudley Herschbach [2003] ‘‘Stern and Gerlach: How a
Bad Cigar Helped Reorient Atomic Physics’’, Physics Today 56: 53–9.
Bridgman, P. W. [1916] ‘‘Tolman’s principle of similitude’’, Physical Review 8:
423–31.
Brush, Stephen G. [1976] The Kind of Motion We Call Heat; Book I: Physics and the
Atomists. Amsterdam: N-Holland.
Bub, Jeffrey [2005] ‘‘Quantum Mechanics is About Quantum Information’’.
Foundations of Physics 35: 541–60.
Buckingham, E. [1914] ‘‘On physically similar systems; illustrations of the use of
dimensional equations’’, Physical Review 4: 345–76.
Bueno, Otavio [1999b] ‘‘What is structural empiricism? Scientific change in an
empiricist setting’’, Erkenntnis, 50: 59–85.
[1999a] ‘‘Empiricism, Conservativeness and Quasi-Truth’’, Philosophy of Sci-
ence, 66: S474–S485.
Busch, Paul, Pekka J. Lahti, and Peter Mittelstaedt [1996] The Quantum Theory of
Measurement 2nd rev. edn. Springer–Verlag Telos.
Busch, Paul and Pekka J. Lahti [1996] ‘‘The standard model of quantum measure-
ment theory: history and applications’’, Foundations of Physics 26: 875–93.
Campbell, Norman Robert [1920] Physics. The Elements. Cambridge: The Univer-
sity Press.
[1928] An Account Of The Principles Of Measurement And Calculation. London:
Longmans, Green.
[1943] ‘‘The measurement of dynamical magnitudes’’, Proc. Phys. Soc. 55:
204–10.
 325
Cantor, Geoffrey [1989] ‘‘The rhetoric of experiment’’, pp. 159–80 in Gooding,

Pinch, and Schaffer 1989.
Cartwright, Nancy Delaney [1983a] How the Laws of Physics Lie. Oxford: Oxford
University Press.
[1983b] ‘‘How the measurement problem is an artefact of the mathematics’’,
pp. 163–216 of Cartwright 1983a.
[1999] The Dappled World: A Study of the Boundaries of Science. Cambridge:
Cambridge University Press.
T. Shomar and M. Suarez [1995] ‘‘The tool box of science’’, pp. 137–49 in
Herfel, Krajewski, Niiniluoto, and Wojcicki 1995.
Cassirer, Ernst [1950] The Problem of Knowledge: Philosophy, Science, and History Since
Hegel. New Haven: Yale University Press.
[1998a] ‘‘Galilean particles: an example of constitution of objects’’, in Castel-
lani 1998b, 181–94.
(ed.) [1998b] Interpreting Bodies: Classical and Quantum Objects in Modern
Physics. Princeton, N.J.: Princeton University Press.
Chang, Hasok [2004] Inventing Temperature: Measurement and Scientific Progress.
Oxford: Oxford University Press.
Chevalley, Catherine [1995] Pascal: Contingence et Probabilité. Paris: Presses Uni-
versitaires de France.
Churchland, Paul M. and Clifford A. Hooker (eds.) [1985] Images of Science:
Essays on Realism and Empiricism, with a Reply by Bas C. van Fraassen. Chicago:
University of Chicago Press.
Coffa, Alberto [1986] ‘‘From geometry to tolerance: sources of conventionalism
in nineteenth-century geometry’’, in Colodny 1986, 3–70.
Cohen, I. Bernard [1960] The Birth of a New Physics. Rev. edn. New York: W. W.
Norton.
Cohen, Robert and Max Wartofsky (eds.) [1965] Boston Studies in the Philosophy of
Science, vol. ii. New York: Humanities Press.
Colodny, Robert G. [1986] From Quarks to Quasars. Philosophical Problems of Modern
Physics. Pittsburgh: University of Pittsburgh Press.
Condillac, Étienne Bonnot de [1949] Traité des systèmes (1749/1798), Œuvres
philosophiques de Condillac., vol. i. Paris: Presses Universitaires de France.
Cooper, David (ed.) [1992] A Companion to Aesthetics. Oxford: Blackwell.
Coxeter, H. S. M. and S. L. Greitzer [1967] Geometry Revisited. Washington, DC:
Math. Assoc.
Crombie, A. [1994] Styles of Scientific Thinking in the European Tradition. 3 vols.
London: Duckworth.
326 
Cushing, James T. [1994] Quantum Mechanics : Historical Contingency and the

Copenhagen Hegemony. Chicago: University of Chicago Press.
[1998] Philosophical Concepts in Physics: the Historical Relation Between Philosophy
and Scientific Theories. Cambridge: Cambridge University Press.
and Ernan McMullen (eds.) [1989] Philosophical Consequences of Quantum
Theory. Notre Dame: University of Notre Dame Press.
D’Agostino, Salvo [2004] ‘‘The Bild Conception of Physical Theory: Helmholtz,
Hertz, and Schrödinger’’, Physics in Perspective 6: 372–89.
Da Costa, Newton C. A. and S. French [1990] ‘‘The model-theoretic approach in
the philosophy of science’’, Philosophy of Science 57: 248–65.
Dalla Chiara, Marisa, Roberto Giuntini, and Richard Greechie [2004] Reasoning
in Quantum Theory: Sharp and Unsharp Quantum Logics. Berlin: Springer.
Danto, Arthur and Sidney Morgenbesser (eds.) [1960] Philosophy of Science. New
York: Meridian Books.
Darrigol, Olivier [2003] ‘‘Number and measure: Hermann von Helmholtz at the
crossroads of mathematics, physics, and philosophy’’, Studies in the History and
Philosophy of Science 34: 515–73.
de Regt, Henk W. [1999] ‘‘Ludwig Boltzmann’s Bildtheorie and scientific under-
standing’’, Synthese 119: 113–34.
[2001] ‘‘Erwin Schrödinger’’, pp. 85–104 in Blackmore, 2001.
[2005] ‘‘Scientific Realism in Action: Molecular Models and Boltzmann’s
Bildtheorie’’, Erkenntnis 63: 205–30.
and Dennis Dieks [2005]) ‘‘A Contextual Approach to Scientific Under-
standing’’, Synthese 144: 137–70.
Dear, Peter [1995] Discipline and Experience. Chicago: University of Chicago Press.
Demopoulos, William [2003] ‘‘On The Rational Reconstruction Of Our Theor-
etical Knowledge’’, British Journal for the Philosophy of Science 54: 371–403.
and Michael Friedman [1985] ‘‘Critical Notice: Bertrand Russell’s The
Analysis of Matter: Its historical context and contemporary interest’’, Philosophy
of Science 52: 621–39.
Dennett, Daniel C. and John Haugeland, [1987] ‘‘Intentionality’’, in R. L. Gregory
(ed.) The Oxford Companion to the Mind, Oxford: Oxford University Press.
Derksen, A. A. [2004] ‘‘Occlusion shapes and sizes in a theory of depiction’’, British
Journal of Aesthetics. 44: 319–41.
Descartes, René [1959]The Meditations and Selections from the Principles of René
Descartes. Tr. and ed. J. Veitch. La Salle, Ill.: Open Court.
Dijksterhuis, E. J [1969] The Mechanization of the World Picture. Oxford: Oxford
University Press.
 327
Dijksterhuis, Fokko Jan [2004] Lenses and Waves: Christiaan Huygens and the
Mathematical Science of Optics in the Seventeenth Century. Dordrecht: Springer.
Drake, Stillman [1957] Discoveries and Opinions of Galileo. New York: Doubleday
& Co.
Dretske, F. [1981] Knowledge and the Flow of Information. Cambridge, Mass.: MIT
Press.
[1991] Naturalizing the Mind . Cambridge, Mass.: MIT Press.
Ducheyne, Steffen [2005] ‘‘Lessons from Galileo: the Pragmatic Model of Shared
Characteristics of Scientific Representation’’, Philosophia Naturalis 42: 213–34.
Duhem, Pierre [1962] The Aim and Structure of Physical Theory. Tr. Philip P. Wiener
of La Théorie Physique, son Objet, sa Structure. Paris: Marcel Rivière & Cie, 1914.
New York: Atheneum.
Eco, Umberto et al. [1992] Interpretation and Overinterpretation. Cambridge: Cam-
bridge University Press.
Edgerton, Samuel Y. [1975] Renaissance Rediscovery of Linear Perspective. New York:
Basic Books.
[1991] The Heritage of Giotto’s Geometry—Art and Science on the Eve of the
Scientific Revolution. Ithaca: Cornell University Press.
Einstein, Albert [1905/2005] ‘‘On the electrodynamics of moving bodies’’, pp.
123–60 in J. Stachel 2005.
[1988] ‘‘Remarks to the Essays Appearing in this Collective Volume’’, pp.
663–88 in Schilpp 1988.
[2004a] ‘‘Geometry and Experience’’, pp. 12–48 in Einstein 2004.
[2004b] Sidelights on Relativity. Whitefish, Mont.: Kessinger Publishing.
and L. Infeld [1938] The Evolution of Physics. New York: Simon and Schuster.
Elgin, Catherine Z [1996] Considered Judgment. Princeton, N.J.: Princeton Univer-
sity Press.
[2006] ‘‘Exemplification, Idealization, and Understanding’’, ms.
Ellis, Brian [1966] Basic Concepts of Measurement. Cambridge: Cambridge University
Press.
Falmagne, J.-Cl. and L. Narens, [1983] ‘‘Scales and Meaningfulness of Quantitative
Laws’’, Synthese, 54: 287–325.
Feyerabend, Paul [1958] ‘‘Reichenbach’s Interpretation of Quantum Mechanics’’,
Philosophical Studies 9: 49–59; reprinted pp. 109-22 in Hooker 1975.
[2001] Conquest of Abundance: A Tale of Abstraction versus the Richness of Being.
Rep. edn. Chicago: University Of Chicago Press.
Files, Craig [1996] ‘‘Goodman’s rejection of resemblance’’, British Journal of
Aesthetics 36: 398–412.
328 
Fine, Arthur [1986] The Shaky Game : Einstein, Realism and the Quantum Theory.
Chicago: University of Chicago Press.
[1998] ‘‘The Viewpoint of No-One in Particular’’, Proceedings of the American
Philosophical Association 72: 9-20.
Fourier, J. B. J. [1952] Analytical Theory of Heat, in Great Books of the Western
World., 45: 161–251, Chicago: Encyclopaedia Britannica.
Freeland, Cynthia [2007] ‘‘Portraits in painting and photography’’, Philosophical
Studies 135: 95–109.
Freeland, G. and A. Corones, (eds.) [1999] 1543 and All That: Image and Word,
Change and Continuity in the Proto-Scientific Revolution. Dordrecht: Springer.
Freistadt, H.[1957] ‘‘The Causal Formulation of Quantum Mechanics of Particles,’’
Nuovo Cimento, Suppl. to vol. 5: 1–70.
French, P. et al. (eds.) [1979] Midwest Studies in Philosophy, vol. iv, Metaphysics,
Minneapolis: University of Minnesota Press.
French, Steven [2003] ‘‘A Model-Theoretic Account of Representation (Or, I
Don’t Know Much About Art . . . But I Know It Involves Isomorphism)’’,
Friedman, Michael [1987] ‘‘Carnap’s Aufbau Reconsidered’’, Noûs 21: 521–45,
repr. as ch. 5 of his 1999.
[2000] ‘‘Transcendental philosophy and a priori knowledge: a neo-Kantian
perspective’’, pp. 367–83 in Boghossian and Peacocke 2000.
[2002] ‘‘Physics, Philosophy, and the Foundations of Geometry’’, Diálogos
79: 121–42.
Frigg, Roman [2006] ‘‘Scientific Representation and the Semantic View of
Theories’’, Theoria 21: 49–65.
Fuchs, Christopher A. [1997] ‘‘Information gain vs. state disturbance in quantum
theory’’, Fortschr. Phys, 46: 535–65. Repr. pp. 229–59 in Braunstein 1998.
[2002] ‘‘Quantum Mechanics as Quantum Information (and only a little
more)’’, arXiv:quant-ph/0205039 v1 8 May.
Galilei, Galileo [1957] The Assayer. Eng. tr. in Drake 1957.
[1964–1966] Le Opere. Firenze: G. Barberi.
[1974] Two New Sciences. Tr. Stillman Drake. Milwaukee: University of
Wisconsin Press.
[1994] Lettera a Fortunio Liceti, gennaio 1641, Opere, XVIII; Engl. tr. in
Crombie 1994, vol. i: 585.
[2005] Discourse on Bodies in Water. Tr. Thomas Salusbury; with an introduc-
tion by Stilman Drake. New York: Dover Publications.
Galison, P. and A. Assmus [1989] ‘‘Artificial clouds, real particles’’, pp. 224–74 in
Gooding, Pinch, and Schaffer 1989.
 329
Georgalis, Nicholas [2005] The Primacy of the Subjective: Foundations for a Unified
Theory of Mind and Language. Cambridge, Mass.: MIT Press.
George, A. (ed.) [1953] Louis de Brolie physician et penseur. Paris: Albin Michel.
Ghins, Michel [1992] ‘‘Scientific realism and invariance’’, Philosophical Issues, 2:
249–62.
[1998] ‘‘Van Fraassen’s constructive empiricism, symmetry requirements and
scientific realism’’, Logique et analyse 164: 327–42.
Giere, R. [1985] ‘‘Constructive realism’’, pp. 75–98 in Churchland and Hooker
1985.
[1988] Explaining Science. Chicago: University of Chicago Press.
[1988] Explaining Science. Chicago: University of Chicago Press.
[2006] Scientific Perspectivism. Chicago: University of Chicago Press.
[1996] ‘‘Visual models and scientific judgement’’, pp. 269–302 in Baigrie
1996.
[1999] Science Without Laws. Chicago: University of Chicago Press.
Glymour, Clark [1999] ‘‘A mind is a terrible thing to waste’’, review of Kim 1998.
Gombrich, Ernst [1960] Art and illusion; a study in the psychology of pictorial
representation. New York: Pantheon Books.
Good, Irving John, Alan James Mayne, and John Maynard Smith (eds.) [1961] The
Scientist Speculates. London: Heinemann.
Gooding, D., T. Pinch, and S. Schaffer [1989] The Uses of Experiment: Studies in
the Natural Sciences. Cambridge: Cambridge University Press.
Goodman, Nelson [1968] Languages of Art: An Approach to a Theory of Symbols. 2nd
edn. Indianapolis: Hackett Publishing Company, 1976.
[1987–8] ‘‘On what should not be said about representation’’, Journal of
Aesthetics and Art Criticism 46: 419.
and Catherine Z. Elgin [1988] Reconceptions in Philosophy and Other Arts and
Sciences. London: Routledge.
Gopnik, Blake [1995] Pictorial Mimesis in Cinquecento Italy, 1500–1568: Texts, Visual
Rhetorics, and a Roman Test-Case. A doctoral thesis for the University of Oxford,
Wolfson College.
Graßhoff, Gerd, Samuel Portmann, and Adrian Wüthrich [2005] ‘‘Minimal
Assumption Derivation of a Bell-type Inequality’’, British Journal for the Philosophy
of Science 56: 663–80.
Greenstein, George and Arthur G. Zajonc [1997] The Quantum Challenge: Modern
Research on the Foundations of Quantum. Sudbury, Mass.: Jones & Bartlett
Publishers.
330 
Gregory, R. L. (ed.) [1987] The Oxford Companion to the Mind. Oxford: Oxford
University Press.
Grünbaum, Adolf [1952] ‘‘Some Highlights of Modern Cosmology and Cos-
mogony’’, The Review of Metaphysics 3: 489–91.
[1954] ‘‘E. A. Milne’s Scales of Time’’, The British Journal for the Philosophy of
Science 4: 329–31.
[1957] ‘‘Complementarity in Quantum Physics and its Philosophical Gener-
alization’’, Journal of Philosophy 54: 713–27.
[1968] Geometry and Chronometry in Philosophical Perspective. Minneapolis:
University of Minnesota Press.
Hacking, Ian [1981] ‘‘Do we see through a microscope?’’, Pacific Philosophical
Quarterly 62, pp. 305–22; repr. pp. 132–52 in Churchland and Hooker 1985.
[1983] Representing and Intervening: Introductory Topics in the Philosophy of
Natural Science. Cambridge: Cambridge University Press.
[2006] Another New World is Being Constructed Right Now: the Ultracold.
Max-Planck-Institut für Wissenschaftsgeschichte, Preprint 316, 2006.
Hagen, Margaret A. [1986] Varieties of Realism: Geometries of Representational Art.
Cambridge: Cambridge University Press.
Hanna, J. F. [1983] ‘‘Empirical adequacy’’, Philosophy of Science 50: 1–34.
Hanna, J. F. [1984]; ‘‘On the empirical adequacy of composite statistical hypo-
theses’’, pp. 73–80 in Asquith and Kitcher 1984.
Harleman, Donald R. F. [1982] ‘‘Hydrothermal Analysis of Lakes and Reservoirs’’,
Journal of the Hydraulics Division 108: 301–25.
Harms, William F. [1998] ‘‘The Use of Information Theory in Epistemology’’,
Harré, Rom ‘‘The materiality of instruments in a metaphysics for experiments’’,
pp. 19–38 in Radder 2003.
Heidelberger, Michael [2003] ‘‘Theory-Ladenness and Scientific Instruments in
Experimentation’’, pp. 138–151 in Radder 2003.
Heinonen, Teikko, Pekka Lahti, and Kari Ylinen [2004] ‘‘Covariant fuzzy observ-
ables and coarse-graining’’, Reports on Mathematical Physics 53: 425–41.
Heinonen, Teiko [2005] Imprecise Measurements in Quantum Mechanics. Turkey:
Annales Universitatis Turkuensis.
Heisenberg, Werner [1945] ‘‘Zur Geschichte der physikalischen Naturerklärung’’,
Wandlungen in den Grundlagen der Naturwissenschaft. 6. Aufl. Leibzig: S. Hirzel.
Helmholtz, Hermann von [1956] ‘‘Ueber den Ursprung und die Bedeutung der
geometrischen Axiome’’. Tr. J. R. Newman, ‘‘On the origin and significance
of geometrical axioms’’, pp. 647–68 in Newman, James 1956.
 331
Herfel, W., W. Krajewski, I. Niiniluoto, and R. Wojcicki (eds.) [1995] Theories

And Models In Scientific Processes. Amsterdam: Rodopi.
Hertz, Heinrich [1956] The Principles of Mechanics. Tr. D. E. Jones and J. T. Walley
of Die Prinzipien der Mechanik in neuem Zusammenhange dargestellt, 1894. New
York: Dover Publications.
[1962] Electric Waves: Being Researches on the Propagation of Electric Action
with Finite Velocity Through Space, 1892. Tr. D. E. Jones. New York: Dover
Publications.
Hilgevoort, Jan (ed.) [1994] Physics and Our View of the World. Cambridge:
Homann, S. J., Frederick A. [1991] Hugh of St. Victor’s Practical Geometry. Milwau-
kee: Marquette University Press.
Hooke, Robert [1665] Micrographia: Or Some Physiological Descriptions of Minute
Bodies Made by Magnifying Glasses. With Observations and Inquiries Thereupon. 1st
edn. London: J. Martyn and J. Allestry.
Hooker, Clifford [1975] The Logico-Algebraic Approach to Quantum Mechanics.
Dordrecht: Reidel.
Hopkins, Robert [1994] ‘‘Resemblance and Misrepresentation’’, Mind 103:
421–38.
Hornung, Hans G. [2006] Dimensional Analysis : Examples of the Use of Symmetry.
Mineola, New York: Dover Publications.
Howard, Don A. [2004] ‘‘Who Invented the Copenhagen Interpretation? A Study
in Mythology’’, Philosophy of Science 71: 669–82.
Hugh of St. Victor [1991] Practical Geometry. Tr. with introd. Frederick A. Homann,
S. J. Milwaukee: Marquette University Press.
Hughes, R. I. G. [1997] ‘‘Models and Representation’’, Philosophy of Science, 64:
S325–S336.
Hyder, David Jalal [2003] ‘‘Kantian Metaphysics And Hertzian Mechanics’’, pp.
35–46 in Stadler 2003.
and Heinz Lübbig [2000] ‘‘Questions Suspended in the Ether,’’ in Nature
404: 223–4.
Hyman, John [1992] ‘‘Perspective’’, pp. 323–7 in Cooper 1992.
[2000] ‘‘Pictorial Art and Visual Experience’’, British Journal of Aesthetics, 40:
21–45.
Ismael, Jenann [1999] ‘‘Science and the Phenomenal’’, Philosophy of Science
66: 351–69.
[2007] The Situated Self . Oxford: Oxford University Press.
Jauch, J. M. [1968] Foundations of Quantum Mechanics. New York: Addison-Wesley.
Jauernig, Anja [2004] Leibniz Freed From Every Flaw: A Kantian Reads Leibnizian
Metaphysics. Diss., Princeton University.
332 
Jourdain, Philip [1918] The Philosophy of Mr. B*rtr*nd R*ss*ll. London: George
Allen & Unwin.
Kant, Immanuel [1848] Critique of Pure Reason. Tr. Francis Haywood. London :
William Pickesing.
[1850] Critique of Pure Reason. Tr. J. M. D. Meiklejohn. London: Henry
G. Bohn.
[1983a] ‘‘Von dem ersten Grunde des Unterschiedes der Gegenden im
Raume’’, pp. 991–1000 in Kant 1983b.
[1983b] Werke in Zehn Baenden, (her. W. Weischedel), vol. 2: Vorkritische
Schriften bis 1768, Zweiter Teil. Darmstadt: Wissenschaftliche Buchgesellschaft.
[1992] ‘‘Concerning the ultimate ground of the differentiation of directions
in space’’, tr. of Kant 1983a in Walford 1992.
Kelly, Sean Dorrance [1998] ‘‘What Makes Perceptual Content Non-
Conceptual?’’, Electronic Journal of Analytic Philosophy, Issue 6.
Kemp, Martin [1990] The Science of Art: Optical themes in Western art from Brunelleschi
to Seurat. New Haven: Yale University Press.
Kim, Jaegwon [1998] Mind in a Physical World. Cambridge, Mass.: MIT Press.
Klein, Carsten [2003] ‘‘Coordination And Convention In Hans Reichenbach’s
Philosophy Of Space’’, pp. 109–20 in Stadler 2003.
Körner, S. (ed.) [1962] Observation and Interpretation in the Philosophy of Physics: a
symposium of philosophers and physicists. New York: Academic Press, 1957. Repr.
New York: Dover.
Koslow, A. [1968] ‘‘Mach’s Concept Of Mass: Program And Definition’’, Synthese
18: 216–33.
Koyré, Alexandre [1965a] ‘‘Huygens and Leibniz on Universal Attraction’’, pp.
115–38 in his 1965b.
[1965b] Newtonian Studies. Chicago: University of Chicago Press.
Kroes, Peter [2000] ‘‘Engineering Design and the Empirical Turn in the Philosophy
of Technology’’, pp. 19–43 in Kroes and A. Meijers 2000.
[2003] ‘‘Physics, Experiments, and the Concept of Nature’’, pp. 68–86 in
Radder 2003.
and A. Meijers (ed.) [2000] The Empirical Turn in the Philosophy of Technology.
Amsterdam: Elsevier Science Ltd.
Kubovy, Michael [1986] The Psychology of Perspective and Renaissance Art. Cambridge:
Kulvicki, John ‘‘Any Way You Slice It: The Viewpoint Independence of Pictorial
Content’’, with comments by Dilworth, on http://www.interdisciplines.org/
artcognition/papers/9/version/original
 333
Ladyman, James [1998a] Structural Realism and the Model-Theoretic Approach to Physical
Theories. Diss., University of Leeds.
[1998b] ‘‘Structural realism: epistemology or metaphysics?’’, Studies in the
History and Philosophy of Science, 29A: 409–24.
Lange, Marc [forthcoming] ‘‘Dimensional Explanation’’.
Latour, Bruno [1999a] ‘‘Circulating Reference: Sampling the Soil in the Amazon
Forest’’, pp. 24–79 in his 1999b.
[1999b] Pandora’s Hope: Essays on the Reality of Science Studies. Cambridge,
Mass.: Harvard University Press.
Laudisa, Federico and Carlo Rovelli, ‘‘Relational Quantum Mechanics’’, in
the Stanford Encyclopedia of Philosophy, http://plato.stanford.edu/entries/qm-
relational/.
Leeds, Stephen [1978] ‘‘Theories of Reference and of Truth,’’ Erkenntnis 13:
111–29.
[1984] ‘‘Chance, Realism, Quantum Mechanics’’, Journal of Philosophy 81:
567–78.
[1994] ‘‘Constructive Empiricism,’’ Synthese 101: 187–221.
[1995] ‘‘Truth, Correspondence, and Success,’’ Philosophical Studies 79: 1–36.
[forthcoming] ‘‘Correspondence Truth and Scientific Realism’’ (forthcoming
in Synthese).
and R. Healey [1996] ‘‘A Note on van Fraassen’s Modal Interpretation of
Quantum Mechanics’’, Philosophy of Science 63: 91–104.
Leibniz, G. W. [1989] Philosophical Essays. Tr. Roger Ariew and Daniel Garber.
Indianapolis: Hacket Publishing Co.
Leplin, Jared [1997] A Novel Defense of Scientific Realism. New York: Oxford
University Press.
Leroux, Jean [2001] ‘‘‘Picture theories’ as forerunners of the semantic approach
to scientific theories,’’ International Studies in the Philosophy of Science, 15:
189–97.
Lewis, David K. [1970] ‘‘How to define theoretical terms’’, Journal of Philosophy
67: 427–46; repr. as chapter six of his Philosophical Papers (Oxford, 1986).
[1979] ‘‘Attitudes de dicto and de se’’, Philosophical Review 88: 513–43.
[1984] ‘‘Putnam’s Paradox’’, Australasian Journal of Philosophy 62: 221–36.
[2001] ‘‘Forget about the ‘correspondence theory of truth’’, Analysis 61:
275–80.
Lipton, Peter [2004] Inference to the Best Explanation. 2nd edn. London: Routledge.
Lloyd, Elizabeth [1986] ‘‘Thinking about Models in Evolutionary Theory’’, Philo-
sophica 37:87–100.
[1987] ‘‘Confirmation of Ecological and Evolutionary Models’’, Biology and
Philosophy 2: 277–93.
334 
Lloyd, Elizabeth [1984] ‘‘A Semantic Approach to the Structure of Population

Genetics’’, Philosophy of Science 51:242–64.
[1994] The Structure and Confirmation of Evolutionary Theory. 2nd edn.
Princeton, N.J.: Princeton University Press.
Locke, John [1959] An Essay Concerning Human Understanding. New York: Dover
Publications
Lodge, David [2001] Thinks. New York: Viking.
Loemker, L. E. [1976] Leibniz: Philosophical Papers and Letters: A Selection. 2nd edn.
Dordrecht: Reidel Publishing Co.
Loewer, Barry [1982] ‘‘Review: F. Dretske, Knowledge and the Flow of Inform-
ation’’, Philosophy of Science 49: 297–300.
Lopes, Dominic [1996] Understanding Pictures. Oxford: Oxford University Press.
Luce, Duncan [1978] ‘‘Dimensionally invariant laws correspond to meaningful
qualitative relations’’, Philosophy of Science 45: 1–16.
McCarthy, E. Doyle [1984] ‘‘Toward a Sociology of the Physical World: George
Herbert Mead on Physical Objects’’, Studies in Symbolic Interaction 5: 105–21.
McGinn, Colin [1997] Review of Dretske 1991. Noûs 31: 528–37.
Mach, Ernst [1986] Principles of the Theory of Heat. Tr. B. McGuinness of Die
Prinzipien der Wärmelehre (Leipzig: Barth, 1896). Dordrecht: Reidel.
[1992]: ‘‘Sensory elements and scientific concepts’’, tr. J. Blackmore of
‘‘Sinnliche Elemente und naturwissenschaftliche Begriffe’’, Archiv für Physiologie,
136 (1910): 263–74, in Blackmore [1992].
McLaughlin, R. (ed.) [1982] What? Where? When? Why? Essays in honour of Wesley
Salmon. Dordrecht: Reidel.
McLendon, Hiram J. [1955] ‘‘Uses of similarity of structure in contemporary
philosophy’’, Mind 64: 79–95.
Maddy, Penelope [2007] Second Philosophy. Oxford: Oxford University Press.
Magnani, M., Thagard P. and N. Nersessian, (eds.) [1999] Model-Based Reasoning
in Scientific Discovery. Dordrecht: Kluwer Academic Publishers.
Marlow, A. R. [1978] ‘‘Quantum Theory and Hilbert Space’’, Journal of Mathem-
atical Phvsics 19: 1841–1846.
[1980a] ‘‘An Extended Quantum Mechanical Embedding Theorem’’, pp.
71–7 in Marlow 1980b.
(ed) [1980b] Quantum Theorv and Gravitation. New York: Academic Press.
Mates, Benson [1986] The Philosophy of Leibniz: Metaphysics and Language. New
York: Oxford University Press.
Matthen, Mohan and Christopher Stephens (eds.) [2007] Handbook of the Philosophy
of Science, Volume 2: Philosophy of Biology. New York: Elsevier.
 335
Maxwell, J. C. [1881] Elementary Treatise on Electricity. Oxford: Oxford University

Press.
[1890] The Scientific Papers of James Clerk Maxwell. Cambridge: Cambridge
University Press.
Mermin, N. David [1981] ‘‘Quantum Mysteries for Anyone’’, The Journal of
Philosophy 78: 397–408.
[1998] ‘‘What Is Quantum Mechanics Trying to Tell Us?’’, American Journal
of Physics 66: 753–67.
Michael, M. and J. O’Leary-Hawthorne (eds.) [1994] Philosophy in Mind: The Place
of Philosophy in the Study of Mind. Dordrecht: Kluwer Academic Publishers.
Michell, J. [1993] ‘‘The origins of the representational theory of measurement:
Helmholtz, Hölder, and Russell’’, Studies in the History and Philosophy of Science
24: 185–206.
Middleton, W. E. Knowles [1966] A History of the Thermometer and its Use in
Metereology. Baltimore: Johns Hopkins Press.
Miller, A. I. (ed.) [1990] Sixty-Two Years of Uncertainty. New York: Plenum Press.
Milne, E. A. [1948] Kinematic Relativity. Oxford: Oxford University Press.
Monton, Bradley [2008] Images of Empiricism. Oxford: Oxford University Press.
Morgan, Mary S. and Margaret Morrison (eds.) [1999] Models as Mediators:
Perspectives on Natural and Social Science. Cambridge: Cambridge University
Press.
Morgenbesser, Sidney (ed.) [1967] Philosophy of Science Today. New York: Basic
Books.
(ed.) [1967] Philosophy of Science Today. New York: Basic Books.
Morrison, Margaret [2007] ‘‘Where Have All the Theories Gone?’’, Philosophy of
Science 74: 195–228.
[1999] ‘‘Models as autonomous agents’’, pp. 38–65 in Morgan and Morrison
1999.
Nagel, Ernest, P. Suppes, and A. Tarski (eds.) [1962] Logic, Methodology and the
Philosophy of Science: Proceedings of the 1960 International Conference. Stanford:
Stanford University Press.
Nagel, Thomas [1986] The View From Nowhere. Oxford: Oxford University Press.
Narens, Louis and R. D. Luce [1986] ‘‘Measurement: the theory of numerical
assignments’’, Psychological Bulletin 99: 166–80.
Nelson, Edward [1967] Dynamical Theories of Brownian Motion. Princeton, N. J.:
Princeton University Press.
Newman, James R. [1956] The World of Mathematics. New York: Simon and
Schuster.
336 
Newman, M. H. A. [1928] ‘‘Mr. Russell’s causal theory of perception’’, Mind 

37: 137–48.
Nola, Robert and Howard Sankey (eds.) [2000] After Popper, Kuhn and Feyerabend.
Boston: Kluwer Academic Publishers.
Norton, John [2000] ‘‘How We Know about Electrons’’, pp. 67–97 in Nola and
Sankey 2000.
[2003] ‘‘A material theory of induction’’, Philosophy of Science 70: 647–70.
Nozick, Robert [1998]: ‘‘Invariance and objectivity’’, Proceedings of the American
Philosophical Association 72: 21–48.
Nyhof, John [1988] ‘‘Philosophical objections to the kinetic theory’’, British Journal
for the Philosophy of Science 39: 81–109.
Olmsted, John W. [1942-43] ‘‘The Scientific Expedition of John Richer to
Cayenne (1672–73),’’ Isis 34: 117–28.
Panovsky, Erwin [1945] Albrecht Dürer. 2nd edn. Princeton: Princeton University
Press.
[1955] Meaning in the Visual Arts. Chicago: University of Chicago Press.
[1956] ‘‘Dürer as mathematician’’, excerpt from Panovsky 1945, pp. 603–21
in vol. i of James Newman 1956.
[1991] Perspective as Symbolic Form. New York: Zone Books.
Papineau, David [1996] The Philosophy of Science. New York: Oxford University
Press.
Parkinson, G. H. R. [1970] Leibniz on Human Freedom (Studia Leibnitiana.
Sonderheft). Stuttgart: Franz Steiner Verlag.
Perry, John [1979] ‘‘The Problem of the Essential Indexical’’, Noûs 13: 3–21.
Peschard, Isabelle F. [2007a] ‘‘Non-Passivity of Perceptual Experience’’, ms.
[2007b] ‘‘Participation of the Public in Science: Towards a New Kind of
Scientific Practice’’, Human Affairs 17, 138–153.
and Patrice Le Gal [1996] ‘‘Coupled wakes of cylinders’’, Physical Review
Letters 77: 3122–5.
Peterson, Mark [2002] ‘‘Galileo’s Discovery of Scaling Laws’’, American Journal of
Physics 70: 575–80.
Pitt, Joseph C. [2005] ‘‘When is an Image not an Image?’’, Techné: Research in
Philosophy and Technology 8: 24–33.
Place U.T. [1956] ‘‘Is consciousness a brain process?’’, British Journal of Psychology
47: 44–50.
Planck, Max [1970] ‘‘The Unity of the Physical World-Picture’’, pp. 1–27 in
Toulmin 1970.
[1994] A Survey of Physical Theory. New York: Dover
 337
[1992]: ‘Die Einheit des physikalischen Weltbildes’, Physikalisches Zeitschrift

10 (1909): 62–75; sect. 4 trans. in Blackmore 1992.
Poincaré, Henri [1952] Science and Hypothesis. Tr. W. J. Greenstreet. New York:
Dover Publications.
[1968] La Science et l’Hypothèse. Paris: Flammarion.
[1970] The Value of Science. Paris: Flammarion.
[2003] Science and Method. New York: Dover.
Pooley, Oliver [2003] ‘‘Handedness, parity violation, and the reality of space’’, pp.
250–80 in Brading and Castellani 2003
Portides, Demetris [forthcoming] ‘‘Scientific models and the semantic view of
theories’’, Philosophy of Science.
Power, Henry [1664] Experimental Philosophy, in Three Books: Containing New
Experiments: Microscopical, Mercurial, Magnetical. 1st edn. London: T. Roycroft,
for John Martin and James Allestry.
Psillos, Stathis [1995] ‘‘Is structural realism the best of both worlds?’’, Dialectica 49:
15–46.
[2006] ‘‘The Structure, the Whole Structure, and Nothing But the Structure’’,
and Martin Curd (eds.) [2008] The Routledge Companion to the Philosophy of
Science. London: Routledge.
Putnam, Hilary [1976] ‘‘Realism and reason’’, Presidential Address to the Eastern
Division of the American Philosophical Association, December 1976. Proceedings
and Addresses of the American Philosophical Association. 50: 483–498; repr. in his
1978, pp. 123–40.
[1978] Meaning and the Moral Sciences. New York: Routledge.
[1981] Reason, Truth and History. Cambridge, Mass.: Cambridge University
Press.
[1992] Renewing Philosophy. Cambridge Mass.: Harvard University Press.
Quine, W. V. O. [1951] ‘‘Two dogmas of empiricism’’, The Philosophical Review
60: 20–43.
Radder, Hans (ed.) [2003] The Philosophy of Scientific Experimentation Pittsburgh:
University of Pittsburgh Press.
Reichenbach, Hans [1958] The Philosophy of Space and Time. Tr. Maria Reichen-
bach and John Freund, of Hans Reichenbach, Philosophie der Raum-Zeit-Lehre.
Berlin/Leipzig: Walter de Gruyter 1928. New York: Dover Publications.
[1959a] Modern Philosophy of Science. Tr. M. Reichenbach. London: Routledge
& Kegan Paul.
[1959b] ‘‘The present state of the discussion on relativity’’, pp. 1–45 in
Reichenbach 1959. Trans. of ‘‘Der gegenwaertige Stand der Relativitaets-
diskussion’’, pp. 342–403 in Reichenbach 1979, vol. 3.
338 
Reichenbach, Hans [1965] The Theory of Relativity and A Priori Knowledge. Tr.
Maria Reichenbach. Berkeley: University of California Press.
[1979] Gesammelte Werke. A. Kamlah and M. Reichenbach (eds.), Braunsch-
weig: Vieweg.
[1991] The Direction of Time. Berkeley: University of California Press, 1956;
new edn. with foreword by Hilary Putnam, 1991.
Rosen, Gideon [1994] ‘‘Objectivity and Modern Idealism: What is the Question?’’,
pp. 277–319 in Michael and O’Leary-Hawthorne 1994.
Rothbart, Daniel [2003] ‘‘Designing instruments and the design of nature’’, pp.
236–54 in Radder 2003.
Rothschild, Daniel ‘‘Structuralism and reference’’, http://www.columbia.edu/
∼dhr2107/structureshort.pdf.
Rovelli, Carlo [1996] ‘‘Relational Quantum Mechanics’’, International Journal of
Theoretical Physics, 35: 1637–78.
Roxburgh, Ian W. [1977] ‘‘The Conventionality of Uniform Time’’, The British
Journal for the Philosophy of Science 28: 172–7.
Russell, Bertrand [1899] Sur les axiomes de la géométrie , Revue de la
métaphysique et de morale, 7: 684–707.
[1901] ‘‘Is position in time and space absolute or relative?’’, Mind  10:
293–317.
[1927]: The Analysis of Matter, London: Allen & Unwin.
[1959]: Problems of Philosophy. Oxford: Oxford University Press.
[1993] Our Knowledge of the External World: As a Field for Scientific Method in
Philosophy. London: Routledge.
[1996] An Essay on the Geometry. Foundations of London: Routledge.
Ryckman, Thomas [2005] The Reign of Relativity. Oxford: Oxford University
Press.
Schilpp, Paul Arthur [1988] Albert Einstein: Philosopher-Scientist. 3rd rev. edn.,
Chicago, IL.: Open Court Publishing Company.
Schrödinger, Erwin [1928] ‘‘Neue Wege in der Physik’’, Elektrische Nachrichtentech-
nik, 5: 485–8.
[1929] ‘‘Neue Wege in der Physik’’, Elektrotechnische Zeitschrift 50: 15–16.
[1951] Science and Humanism: Physics in Our Time. Cambridge: Cambridge
University Press.
[1953] ‘‘The meaning of wave mechanics’’, pp. 16–30 in André George (ed.)
Louis de Brolie physician et penseur. Paris : Albin Michel.
Schwartz, Robert [1980] ‘‘Imagery—There’s More to It than Meets the Eye’’,
PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association, vol.
2: 285–301.
 339
Sellars, Wilfrid [1965] ‘‘Scientific Realism or Irenic Instrumentalism: A Critique

of Nagel and Feyerabend on Theoretical Explanation’’, 171–204 in Cohen and
Wartofsky 1965.
Sfendoni-Mentzou, D. (ed.) [2000] Aristotle and Contemporary Science., vol. 1. New
York: Peter Lang.
Shepard, Roger N. [1990] Mind sights: Original visual illusions, ambiguities, and other
anomalies, with a commentary on the play of mind in perception and art. New York:
W. H. Freeman.
Sismondo, Sergio and Nicholas Chrisman [2001] ‘‘Deflationary Metaphysics and
the Natures of Maps’’, Philosophy of Science. 68: S38–S49.
Solovay, R. M. [1965] ‘‘2âleph0 can be anything it ought to be’’, p. 435 in
Addison, Henkin, and Tarski 1965.
Stachel, John [2005] Einstein’s Miraculous Year. Princeton, N.J.: Princeton Univer-
sity Press.
Stadler, Friedrich [2003] The Vienna Circle and Logical Empirism. Re-evaluation and
Future Perspectives. Secaucus, N.J.: Kluwer Academic Publishers.
Stein, Howard [1989]: ‘‘Yes, but . . . —Some skeptical remarks on realism and
anti-realism’’, Dialectica, 43: 47–65.
Steinberg, Leo and Samuel Edgerton [1987] ‘‘How shall this be?: reflections on
Filippo Lippi’s Annunciation in London’’, Artibus et historiae VIII: 25–53.
Sterrett, Susan G. [2002]‘‘Physical Models and Fundamental Laws: Using One
Piece of the World to Tell About Another’’, Mind and Society 5: 51–66.
[2005] Wittgenstein Flies a Kite: A Story of Models of Wings and Models of the
World. Upper Saddle River, N.J.: Pi Press.
[2006] ‘‘Models of machines and models of phenomena’’, International Studies
in the Philosophy of Science 20 (2006): 69–80.
Stevens, S. S. [1946] ‘‘On the theory of scales of measurement’’, Science 103:
667–80.
Stöltzner, Michael [1999] ‘‘Vienna Indeterminism: Mach, Boltzmann, Exner’’,
Synthese 119: 85–111.
[2002] ‘‘Vienna Indeterminism II: From Exner’s Synthesis to Frank and von
Mises’’, philsci-archive.pitt.edu/archive.
[2003] Causality, Realism and the Two Strands of Boltzmann’s Legacy (1896–1936).
Dissertation, University of Bielefeld 2003. (available http://bieson.ub.uni-
bielefeld.de/volltexte/2005/694/pdf/netpubdiss.pdf)
Suarez, Mauricio [1999] ‘‘Theories, Models, and Representations’’, pp. 75–83 in
Magnani, Nersessian, and Thagard 1999.
[2003] ‘‘Scientific representation: Against similarity and isomorphism’’, Inter-
national Studies in the Philosophy of Science, 17: 225–44.
340 
Suarez, Mauricio [2004] ‘‘An Inferential Conception of Scientific Representation’’,

and N. Cartwright [forthcoming] ‘‘Theories: Tools vs. Models’’, forthcoming
in Studies in the History and Philosophy of Modern Physics.
Suppe, Frederick [1993] ‘‘Credentialing scientific claims’’, Perspectives on Science 1:
153–203.
Suppes, Patrick [1957] Introduction to Logic. Princeton, N.J.: Van Nostrand.
[1960] ‘‘A Comparison of the Meaning and Uses of Models in Mathematics
and the Empirical Sciences’’, Synthese 12: 287–301.
[1962] ‘‘Models of Data’’, pp. 252–61 in Nagel, Suppes, and Tarski 1962.
[1967] ‘‘What is a Scientific Theory?’’, pp. 55–67 in Morgenbesser 1967.
Suppes, Patrick, David M. Krantz, R. Duncan Luce, and Amos Tversky [1989]
Foundations of Measurement., vol. i. New York: Academic Press.
Teller, Paul [2001a] ‘‘Twilight of the Perfect Model Model’’, Erkenntnis 55:
393–415.
[2001b] ‘‘Whither Constructive Empiricism?’’, Philosophical Studies 106:
123–50.
[2008] ‘‘Representation in Science’’, pp. 435–41 in Psillos and Curd 2008.
Temple, G. [1948] The General Principles of Quantum Theory. London: Methuen.
Thomas, David A. [2006] The Mathematics of Perspective: An Introduction
to the Cross Ratio; posted on the web at: http://www.math-ed.com/
Resources/cr/index.htm
Thompson, Paul [1983] ‘‘The Structure of Evolutionary Theory: A Semantic
Approach’’, Studies in History and Philosophy of Science 14: 215–29.
[1986] ‘‘The Interaction of Theories and the Semantic Conception of
Evolutionary Theory’’, Philosophica 37: 73–86.
[2007] ‘‘Formalisations of Evolutionary Biology,’’ pp. 497–523 in Matthen
and Stephens.
Thomson, Wm. Lord Kelvin [1891] Popular Lectures and Addresses., vol. i. London:
Macmillan.
Timpson, Chris [2004] Quantum Information Theory and the Foundations of Quantum
Mechanics. Diss., Oxford University.
Tolman, R. C. [1914-15]‘‘The principle of similitude’’, Physical Review 3: 244–55
and 6: 219–33.
Toraldo di Francia, G. (ed.) [1979] Problems in the Foundations of Physics. Amsterdam:
North-Holland Publishers.
Torretti, Roberto [1984] Philosophy of Geometry from Riemann to Poincaré. Dordrecht:
Reidel.
 341
[1999] The Philosophy of Physics. Cambridge: Cambridge University Press.

Toulmin, Stephen [1970] Physical Reality: Philosophical Essays on Twentieth-Century
Physics. New York: Harper & Row.
van Fraassen, Bas C. [1970] ‘‘On the Extension of Beth’s Semantics of Physical
Theories’’, Philosophy of Science 37: 325–34.
[1974] ‘‘The Einstein-Podolski-Rosen Paradox’’, Synthese 29: 291–309.
[1979] ‘‘Foundations of Probability: A Modal Frequency Interpretation’’, pp.
344–94 in Toraldo di Francia 1979.
[1980] The Scientific Image. Oxford: Oxford University Press.
[1982a] ‘‘Rational Belief and the Common Cause Principle’’, pp. 193–209
in McLaughlin 1982.
[1982b] ‘‘The Charybdis of Realism: Epistemological Implications of Bell’s
Inequality’’, Synthese 5, 25–38, reprinted in Cushing and McMullen 1989 with
new Appendix on explanation.
[1985] Introduction to the Philosophy of Time and Space. 2nd edn. New York:
Columbia University Press.
[1989] Laws and Symmetry, Oxford: Oxford University Press.
[1991] Quantum Mechanics: An Empiricist View. Oxford: Oxford University
Press.
[1993]: ‘‘From vicious circle to infinite regress, and back again’’, pp. 6–29
in D. Hull, M. Forbes, and K. Ohkruhlik (eds.) PSA 1992, vol. 2. Evanston:
Northwestern University Press.
[1994] ‘‘Interpretation of science: science as interpretation’’, pp. 169–87 in J.
Hilgevoort 1994.
[1997a] ‘‘Modal interpretation of repeated measurement: reply to Leeds and
Healey’’, Philosophy of Science 64, 669–76.
[1997b] ‘‘Putnam’s Paradox: Metaphysical Realism Revamped and Evaded’’,
Philosophical Perspectives, 11: 17–42.
[1997c] ‘‘Structure and perspective: philosophical perplexity and paradox’’,
pp. 511–30 in M. L. Dalla Chiara et al. (eds.) (1997) Logic and Scientific Methods.
vol i. Dordrecht: Kluwer Academic Publisher.
[1998] Review: J. Bub, Interpreting the Quantum World. Foundations of Physics
28 (1998): 683–9.
[1999] ‘‘The Manifest Image and the Scientific Image’’, pp. 29–52 in Aerts
1999.
[2000a] ‘‘The sham victory of abstraction’’ (Review of Feyerabend, Conquest
of Abundance), Times Literary Supplement 5073: June 23, 2000: 10–11.
[2000b] ‘‘The theory of tragedy and of science: does nature have narrative
structure?’’, pp. 31–59 in Sfendoni-Mentzou 2000.
342 
van Fraassen, Bas C. [2001] ‘‘Constructive Empiricism Now’’, Philosophical Studies

106: 151–70.
[2002]: The Empirical Stance. New Haven: Yale University Press.
[2004] ‘‘Transcendence of the Ego: The Non-Existent Knight’’, Ratio (new
series) xvii, 453–77.
[2005a] ‘‘Appearance versus Reality as a Scientific Problem’’, Philosophical
Exchange 35, 34–67.
[2005b] ‘‘Wouldn’t it be lovely? Explanation and Scientific Realism’’, Review
Symposium for Lipton, Inference to the Best Explanation, 2nd edn. Metascience 14,
344–52.
[2006a] ‘‘Representation: the Problem for Structuralism’’, Philosophy of Science
73: 536–47.
[2006b] ‘‘Structure: its substance and shadow’’, British Journal for the Philosophy
of Science 57: 275–307
[2007] ‘‘The Constitutive A Priori’’ Review of Ryckman, The Reign of
Relativity., Metascience 16, 407–19.
[forthcoming] ‘‘Rovelli’s World’’, Foundations of Science.
and Isabelle Peschard [2008] ‘‘Identity over Time: Objectively, Subjectively’’,
Philosophical Quarterly, 58: 15–35.
Varadarajan, V.S. [1968] Geometry of Quantum Theory., vol. i. New York: van
Nostrand.
Vargish, Thomas and Delo E. Mook [1999] Inside Modernism : Relativity Theory,
Cubism, Narrative. New Haven: Yale University Press.
Veltman, Kim H. [1988] Review of Michael Kubovy The Psychology of Perspect-
ive and Renissance Art., in Journal of the History of the Behavioral Sciences 24:
251–3.
and Kenneth D. Keele [1986] Studies in Leonardo da Vinci, I: Linear Perspective
and the Visual Dimensions of Science and Art. Muenchen: Deutscher Kunstverlag.
Visser, Henk [1999] ‘‘Boltzmann and Wittgenstein or How Pictures Became
Linguistic’’, Synthese 119: 135–56.
von Neumann, J. [1955] Mathematical Foundations of Quantum Mechanics. Princeton,
N.J.: Princeton University Press.
Vuillemin, Nathalie [2005] ‘‘Hypothèse et fiction : les relations complexes de deux
discours. Quelques remarques sur les stratégies discursives de J.-B. Robinet dans
la philosophie de son temps’’, Comètes, Revue des Littératures d’Ancien Régime 2
(October 2005) http://www.cometes.org
Walford, David (tr. and ed.) [1992] Immanuel Kant: Theoretical Philosophy, 1755–1770.
Cambridge: Cambridge University Press.
Weinberg, Steven [1988] ‘‘The Revolution That Didn’t Happen’’, The New York
Review of Books XLV, 15 (October 8, 1998): 50.
 343
Weschler, Lawrence ‘‘Through The Looking Glass—Further adventures in optic-

ality with David Hockney’’, on-line art magazine ArtKrush.
Weyl, Hermann [1931/1950] The Theory of Groups and Quantum Mechanics. Tr. H.
P. Robertson, New York: Dover Publications.
[1934] Mind and Nature. Philadelphia: University of Pennsylvania Press.
[1952] Symmetry. Princeton, N.J.: Princeton University Press.
[1953] Space, Time, Matter. New York: Dover.
Whitehead, A. N. [1898] A Treatise on Universal Algebra. Cambridge: Cambridge
University Press.
Whitrow, G. J. [1954] ‘‘E. A. Milne’s Scales of Time’’, The British Journal for the
Philosophy of Science. 5: 151.
[1961] The Natural Philosophy of Time. London: Nelson and Sons.
Wigner, Eugene Paul [1961] ‘‘Remarks on the mind-body question’’, pp. 284–302
in Good, Mayne, and Smith 1961.
[1963] ‘‘The Problem of Measurement’’, American Journal of Physics 31: 6–15.
[1970] Symmetries and Reflections—Scientific Essays. Boston Mass.: MIT Press.
[1995] Collected Works. Part B. Historical,Philosophical, and Socio-Political Papers.
Jagdish Mehra (ed.), vol. 6: Philosophical Reflections and Syntheses (annotated by
Gérard G. Emch) Berlin: Springer.
Wilce, Alexander [forthcoming] ‘‘Formalism and Interpretation in Quantum
Theory’’, Foundations of Physics.
Williams, Bernard [1978] Descartes: The Project of Pure Enquiry. Harlow, U.K.:
Penguin.
Williams, Bernard [1985] Ethics and the Limits of Philosophy. Cambridge, Mass.:
Harvard University Press.
Wilson, Catherine [1995] The Invisible World: Early Modern Philosophy and the
Invention of the Microscope. Princeton, N.J.: Princeton University Press.
Wilson, Margaret [1984] ‘‘Skepticism without indubitability’’, Journal of Philosophy
81: 537–44; repr. as chapter 1 in her 1999.
[1999] Ideas and Mechanism. Princeton, N.J.: Princeton University Press, 1999.
Wimsatt, William C. [1990] ‘‘Taming the Dimensions-Visualizations in Science’’,
PSA: Proceedings of the Biennial Meeting of the Philosophy of Science Association, vol.
2: 111–35.
Worrall, John [1989a] ‘‘Fresnel, Poisson and the White Spot: the role of successful
predictions in the acceptance of scientific theories’’, pp.135–58 in Gooding,
Pinch, and Schaffer 1989.
[1989b] ‘‘Structural realism: the best of both worlds?’’, Dialectica, 43: 99–124;
repr. pp. 139–65 in David Papineau, The Philosophy of Science. New York:
Oxford University Press, 1996.
344 
Zadeh, Lofti [1965] ‘‘Fuzzy sets’’, Information and Control 8: 338–53.

Zahar, Elie [2001]: Poincaré’s Philosophy: From Conventionalism to Phenomenology.
Indianapolis: Chicago, IL.: Open Court.
Zajonc, Arthur [1995] Catching the Light: The Entwined History of Light and Mind.,
new edition New York: Oxford University Press.
Zalta, Edward N. [1988] Intensional Logic and Metaphysics of Intentionality. Boston,
Mass.: MIT Press.
Zola, Émile [1867] ‘‘Une nouvelle manière en peinture—Edouard Manet’’, Revue
du XIXe Siècle, January 1: 34.
Notes
Introduction
1 As I shall characterize representation below, ‘‘mental representation’’ is an
oxymoron.
2 Boltzmann’s way of thinking about this, and especially the various pragmatic
aspects of representation with models, is continued and elaborated in Teller 2008.

3
This describes unqualified acceptance; in practice, acceptance will come with
restrictions and qualifications, and belief will come in degrees.
4 While I will not rehearse or respond to arguments for or against empiricist
views here, I want to thank the editors of two recent volumes to allow me my say:
Andreas Berg-Hildebrand and Christian Suhm 2006, and Bradley Monton 2007.
Part I: Representation
1 Suarez 2004, 770. This appears to be a change of mind from his earlier ‘‘I
take it that a substantive theory of scientific representation ought to provide us
with necessary and sufficient conditions for a source to represent a target’’ (Suarez
2003: 226). While Suarez 2004 contrasts his view to such approaches as R. I. G.
Hughes’s influential (1997), his own ‘‘inferential conception’’ still came very close
to providing just the sort of theory that he dismisses (in the passage I quoted) as
not to the point. See further Ducheyne (2005) which emphasizes that a model
represents for a person if that person accepts certain things. While not wishing
either to propose rivals to their accounts or to adjudicate between them, I have
learned much from these, as well as from such other recent writings as French
2003.
2 For the etymology, and for what became the stable concept in art history, see
Panovsky 1991, 27.

3 We shall take a close look at how Alberti and Dürer, among others, developed
this point, as well as how it was further developed in projective geometry.

346  :  
4 See for instance the photo of Mars made through the Hubble tele-
scope, at http://www.jpl.nasa.gov/news/features.cfm?feature=533. Ian Hacking
has recently studied the use of such terms in the history of science, and has emphas-
ized that the term ‘‘phenomenon’’ typically denotes things classed as remarkable,
unusual, or amazing. What amazes are often individual occurrences but also often
processes with many instances. Some of this appears in Hacking 2006 (see p. 32,
section C1 ‘‘The creation of phenomena’’) issued in connection with his Carl
Friedrich von Weizsäcker-Vorlesungen, Hamburg 2007.
1. Representation Of, Representation As

1
In Pliny the Elder’s Natural History, xxxv. Available on-line http://www.
perseus.tufts.edu/cgi-bin/ptext?doc=Perseus%3Atext%3A1999.02.0137&query
=toc:head%3D%232431
2 For the point that many recent objections to ‘resemblance’ or ‘isomorph-
ism’ views of representation (or as necessary and/or sufficient conditions for

representation) were already offered in Plato’s Sophist and Cratylus, see my 2000b.
3
Once the Eleatic Stranger’s point is appreciated, examples abound. A painting
or drawing, however realistic, achieves its aim by means that are strictly different
from what it is a ‘likeness’ of—think only of the cross-hatching to render shadows,
or of the strokes of thick pigment in a fresco, noticeable when inspected from
close up.
4 We should add: and that he had as well correctly spelled out the distorting
relationship that produces quite different appearances to the astronomer’s eye.

5
The issue of misrepresentation, and how it requires adjusting any view of
representation that trades on resemblance, is dealt with for the case of sculpture
(but with farther reaching conclusions) by Hopkins 1994.
6 For an exploration of how representation-as is crucial to how models in science
represent what they model, see Hughes1997 and my 1994.

7 Reportedly in E. Spott, Bismarck: A Book of Mistakes. 1883; I have found no
trace of this book.

8
For an analysis of the difficulty see Harms 1998: 482, 485.
9
For further development of Goodman’s view, with special application to
scientific representation see Elgin 2006.
10 I leave aside Goodman’s attempt to accommodate them somehow in exten-
sional discourse.
11 Goodman 1976: 26. This phrasing derives from his constant attempt to
observe a strict nominalism: a depiction of Scott as wise he would call a ‘‘wise’’-

picture, and say that that was the kind of picture that it was.
 ,   347
12 For a clear and balanced exposition of an alternative, including a defense of

a moderate ‘resemblance’ view of pictorial content, see Files 1996.
13
The main moral I want to draw is that this account is not far from the
view that sentences and pictures have after all the same kind of thing as content:
propositions, to use the common term for the content of a sentence. Goodman’s
nominalism would of course stand in the way of this sort of formulation, but it can
be construed innocuously, without violence to anti-ontological scruples.
14 See further Elgin 2006, which explores this with special reference to scientific
experimentation and models.

15
Bartels 2006. These terms are also context-sensitive: an isomorphism is a one-
to-one onto function that preserves ‘‘pertinent’’ structure, while a homomorphism
is a many-to-one onto function that does so. The term ‘‘pertinent’’ gets its content
from the context, where one sort of structure or another is under study.
16 For the moment I am staying with representations of specific, real things;
later we may have to look seriously at such examples as ‘‘X represents a man
holding a candle/ Santa Claus delivering presents/phlogiston as escaping rapidly’’
and the like.
17 ‘‘Once we take on board the distinction between bare bones and fleshed
out content, we are in a position to notice an important feature of pictorial

representations. In the late 70s, Sherry Levine produced a controversial series of
photographs. She made photos of some Walker Evans photographs that were
basically indistinguishable from Evans’ originals. This was controversial because
she displayed her photos as her own work. In a sense, they are her own work: she
made photos of Evans’ photographs while Evans made photos of life in the rural
United States. In another sense, their similarity to Evans’ originals makes them
seem like mere copies. Controversy aside, this sheds light on an interesting feature
of pictures. Many different scenes, like a chair and a jumble of line segments,
can result in the same kind of photograph or linear perspective picture from any
given point of view. Levine’s photos show that one potential subject for any given
picture is a plane that has shapes and colors indistinguishable from the picture itself.
So, in a sense, the rural US, a well-designed Hollywood set, or simply a photo
like Evans’s could result in a photo like Levine’s.’’ (John Kulvicki, ‘‘Any Way
You Slice It: The Viewpoint Independence of Pictorial Content’’, with comments
by Dilworth, on http://www.interdisciplines.org/artcognition/papers/9/version/
original).
18 This form is still restricted; as we will see, it needs extra contextual parameters,
such as the purpose for which the representation is made or which it is made
to serve. That is especially relevant for scientific representation; see e.g. Giere
2006: 60.
348  :  
19 See further sections 3 and 4 of Schwartz 1980. That what makes something
a representation is the fact that it is pressed into representational service by
representation users is also a theme emphasized in Paul Teller’s 2001a.
20 The standard reference here is Hagen 1986.
21 Compare Georgalis 2005: 128–9: ‘‘For any item r to represent a particular
item t, there must be a conscious agent s to whom r represents t. I call this the
fundamental fact of representation.’’ My statement goes beyond this, since I imply
that there is also a representing by (and not just to) an agent. However, there is a
limiting case, in which one and the same agent plays both roles, for example when
someone spontaneously takes an encountered natural object or event to represent
something.
22 cf. Georgalis 2005: 122: ‘‘Since there is no necessary connection between
a physical item serving as a representation and that which it represents, if the

representation is to do its job, . . . somehow the uniqueness of what is represented
must be secured’’—and see his pages 123–9 for where this point leads.
23
I take this to be consonant with the literal use of ‘‘use’’. When we say that
a car’s engine uses gasoline there is no implication of agency or community, and
perhaps one part of a brain could be said to use another in that sense. But I
take those uses of ‘‘use’’ to be at best derivative from the literal use. The narrow
concept of representation, due to that narrowly construed use of ‘‘use’’ will not,
in my view, hamper our discussion of scientific representation.
24 The emphasis on use, as here understood, implies community: there is no
such thing as essentially private representation any more than private language,
except in the sense in which private uses can exist as derived from or parasitic on
communal practices.
25 The example is essentially Putnam’s. As to the main point, the Eleatic
Stranger has a line on this: he has prefaced the list by saying that he regards the
natural ones as divine workmanship—so perhaps he thought of shadows as really
made in order to represent the objects casting them. What about things in dreams?
It’s a nice conceit, that paintings are dreams made for people who are awake—a
conceit echoed by the idea of films as dreams that money can buy, and indeed by
the film ‘‘Dreams That Money Can Buy’’. But as an example to show that there
are ‘representations in nature’ it begs the question of what dreams are.
26 This is part of an objection to the semantic approach. Suarez takes that
approach to involve a view of what representation is—mistakenly, in my opinion.

In this paper he also refers to my view of representation as I presented it in my
2000b, but sees that as an addition to what he gleans from the semantic approach.
27
Cf. Goodman 1968: 21–6. Since Goodman kept denotation at the heart of
his account of representation, he devoted considerable attention to the case of
, ,   349
representations that do not represent real things, hence have nothing to denote.
His gloss on this will do for our purposes.
28
The term ‘‘relation’’ may in fact not be the most apt to explain intentionality.
As Benoist 2006 emphasizes, Brentano who defines the intentionality of mental
activity by the statement that it is characterized as such by ‘‘the relation to
something as object’’ drew a distinction between this case and a genuine relation
(which must be between real things), and says eventually that intentionality is not
a relation but something relation-like: kein Relatives, aber ein Relativliches.
29
In more linguistic terms, the context created by ‘‘represents’’ is ‘referentially
opaque’, for the premise that S and T are the same or apply to the same things
does not license the inference from ‘‘X represents Y as being S’’ to ‘‘X represents
Y as being T’’.
30 I use ‘‘intensional’’ for terms, expressions, forms of discourse, the criteria
being essentially those discussed under this heading by Quine: opacity to reference,
resistance to substitutivity of identicals, and so forth. See for instance The University
of Alberta Dictionary of Cognitive Science, ‘‘ ‘Intentional’ is not to be confused with
‘intensional’ spelled with an ‘s’, the latter of which refers to the meaning of a
term, (along with ‘extensional’)’’ (though that dictionary entry gives too narrow a
meaning to ‘‘intentional’’). For ‘‘intention’’ see Dennett and Haugeland 1987. For
a metaphysical approach, of the sort that I contest here, see Zalta 1988.
31 It is best and clearest to think of this as a point about language, displayed
by focusing on what is and what is not to be inferred in particular examples. But

the same examples, presented in the ‘‘material mode’’ (as Carnap would say) serve
to make the point about what these assertions display as (putative) fact. I will not
bother to obsessively distinguish formal and material mode.
32 Note however that even the regimented verbal distinction does not hinge
on the word ‘‘of’’ alone. ‘‘His speech included a description of Mrs. Thatcher as
draconian’’ is not covered by the convention, it is still ambiguous in the usual way,
and the substitution of ‘‘the then Prime Minister’’ might or might not change the
truth-value of this sentence.
33
In aesthetics, discernment of that level of meaning is the subject of icono-
graphy—see Panovsky, 1955: 26–54.
2. Imaging, Picturing, and Scaling

1 Even in Goodman’s theory, the important case of representation by exem-
plification hinges on a highlighted resemblance. The relevance of such sorts of
representation for science as well as for art is strongly argued in French 2003.
2
See especially Sellars 1965: 180–2.
350  :  
3 For a detailed discussion of the use of visual imagery in the sciences, see
Wimsatt 1990.
4
This does come up in debates in metaphysics: must there be a basis in nature
for objective judgments of similarity, or does the similarity we privilege in our
judgments relate to our concerns, values, and practice? That will be a topic for
another time.
5 There are many paintings to which my description applies, but of course I
mean Manet’s. Thus Zola 1867: ‘‘Les peintres, surtout Edouard Manet, qui est
un peintre analyste, n’ont pas cette préoccupation du sujet qui tourmente la foule
avant tout; le sujet pour eux est un prétexte à peindre tandis que pour la foule le
sujet seul existe. Ainsi, assurément, la femme nue du Déjeuner sur l’herbe n’est là
que pour fournir à l’artiste l’occasion de peindre un peu de chair. Ce qu’il faut voir
dans le tableau, ce n’est pas un déjeuner sur l’herbe, c’est le paysage entier, avec ses
vigueurs et ses finesses, avec ses premiers plans si larges, si solides, et ses fonds d’une
délicatesse si légère; c’est cette chair ferme modelée à grands pans de lumière, ces
étoffes souples et fortes, et surtout cette délicieuse silhouette de femme en chemise
qui fait dans le fond, une adorable tache blanche au milieu des feuilles vertes, c’est
enfin cet ensemble vaste, plein d’air, ce coin de la nature rendu avec une simplicité
si juste, toute cette page admirable dans laquelle un artiste a mis tous les éléments
particuliers et rares qui étaient en lui.’’
6 Recall the earlier reference to Nelson Goodman’s discussion of this, his 1976.
7 Section 6.1 (pp. 112–17) of Lopes 1996.
8
Hyman 2000 and 1992; see further Derksen 2004.
9
Chapter 3 of Giere 2006; see specifically pp. 48–9.
10 Wittgenstein proposed the idea of cluster concepts to highlight classification
by ‘‘family resemblance’’: concepts characterized by a network of more or less

loosely interconnected properties. Application of the concept draws on some parts
of the cluster; there is no general limit to this conceptual ‘‘plasticity’’. Wittgenstein’s
main example is the idea of a game: ‘‘How should we explain to someone what a
game is? I imagine that we should describe games to him, and we might add: ‘This
and similar things are called games’ ‘‘(Philosophical Investigations, para. 69) . . . . ‘‘But
this is not ignorance. We do not know the boundaries because none have been
drawn . . . We can draw a boundary for a special purpose’’ (Ibid).
11 Lopes: 118; Block 1983, and Introduction to Block 1981 where he discusses
Fodor’s and Dennett’s views on this.

12 Although not willing to offer a precise characterization of pictorial as opposed
to descriptional representation, Ned Block has an instructive example of how the

distinction is made in practice:
Consider what goes into a computer by way of storing a picture of, say, a vertical line.
Consider a matrix n squares wide. Each of the squares can be light or dark. A vertical line
, ,   351
would be represented if each lighted square is n squares distant from the first lighted square
(counting by row, as on a calendar). The descriptionalist sees the line representation as a set
of sentences. If n = 7, the set of sentences could be: ‘1 is dark’, ‘2 is light’, ‘3 is dark’, . . . . ‘9
is light’, ‘10 is dark’, and so on . . . .
The pictorialist, on the other hand, sees the representation of a line . . . as like the matrix
itself, not the corresponding set of sentences.
But what does the distinction come to between the set of sentences and the matrix itself?
Aren’t they just different ways of inscribing the same representation with the same semantic
properties?
I say ‘‘No.’’ The key is the way the representations function. Consider how the descrip-
tionalist would rotate his line. A small counterclockwise rotation could be accomplished if
the first lighted square stayed lit, the next number of a lighted square was increased by 1,
the next by 2, the next by 3, and so on. In terms of the example, 2 would stay lit, 10 would
be lit instead of 9, 18 instead of 16, 26 instead of 23, etc. So the descriptionalist’s new set of
sentences would be ‘1 is dark’, ‘2 is light’, . . . ‘10 is light’, . . . ‘18 is light’, . . . ‘26 is light’,
and so on. The important point is that the computer’s ‘‘rotation’’ calculation just involves
the numbers, not the arrangement of the numbered squares. The matrix display is for us
and plays no role in what the computer does. From the point of view of the computer’s
calculations, the squares could as well be arranged in a line or a circle rather than in a matrix.
Real live computer graphics works this way. The machine manipulates numbers. The
programmers think of the numbers as numbers of cells in matrices, and they put in numbers
that correspond to matrices of visual interest. They program the computer to operate
on the numbers in ways that correspond to visually interesting or useful matrix changes.
These correspondences are what makes the computer’s number crunching graphical, but
the correspondences play no role at all in the number crunching itself. (1981: 53)
This is illuminating, though I don’t think it can stand without qualification either.
Note that the functioning in question is not what is meant when we say that a
certain array of numbers or dots functions as a representation. The functioning on
which Block focuses is relevant here, but must be distinguished: it concerns the
transformation of one symbol in a system of symbols into another symbol.
13
Thanks to Isabelle Peschard for this criticism.
14
This insight may be best honored by attention to global differences between
pictorial and other systems of representation. That is, in fact, how Goodman
approaches the matter in his Languages of Art. According to Goodman, as we have
seen, pictures denote and predicate. But what distinguishes pictorial systems from
other denotational systems (such as systems of verbal description) are features such
as denseness that make them more like analog systems, like diagrams and maps. Cf.
Goodman 1976: 194–8; Goodman and Elgin, 1988 ch 7. Pictorial symbol systems
are syntactically and semantically dense. That is, any two pictures, no matter how
similar, could be depicting different things or depicting the same things in different
ways. Cf. Goodman 1976: 226–7. That is not at all true of verbal description,
where small differences typically make no difference at all. An analog measuring
device is similar in that respect: two different pointer positions, no matter how
352  :  
close together, indicate different values. In diagrams or maps too a small difference
in lines or positions can be significant. For example in maps made to scale the
merest movement of a border or contour line depicts a different possible situation
in the mapped region. Nevertheless, there is a distinction also between pictures
and diagrams or maps. Goodman takes that difference to be syntactic, that is, it
does not have to do with reference or attribution but solely with the composition.
The difference is only a matter of degree, however: pictorial symbol systems
are more replete. In a painting a much larger set of features—color, thickness,
intensity, contrast, etc.—is relevant to what it denotes and what it predicates than
in diagrams or even the most detailed topographical maps. This is a difference of
degree that I will not build into my terminology here, for it will not be pertinent
to distinctions I will need for different modes of scientific representation. So I will
describe diagrams, maps, etchings, and paintings all as pictures, and their use as
picturing.
15 Adam Elga has suggested an appropriate form of presentation: we can here
write S = {Set of Points}, Assignment of co-ordinates (x,y,z), Content occupation

function K . This is a model of an ‘occupied’ space, of a world [within just four
parameters, of course, a bit skimpy].
16 But they do assume, very unrealistically, that ordinary macroscopic objects
are as sharply defined as e.g. classical atoms. What is the shape of this beer glass?
Descartes’s answer (translated into our current terms) would have been ‘‘an analytic
function’’. Why do scientists and philosophers tend to be impatient with the point
that such mathematical modeling is unrealistic? The answer is presumably that
they have long since appreciated it, and prefer to focus on the practical usefulness
of the models—a reasonable attitude, even if we can’t afford it when we study
scientific representation as a subject in its own right. Below I will give an example
from classical optics to show how this sort of idealization can lead to paradox, and
thence to changes in theory.
17 See further my 1999.
18
I take this simple example from the history of geometric optics, on its way
to becoming physical optics in the modern sense. For a fascinating study of this
historical stage see Fokko Jan Dijksterhuis 2004.
19 Hero’s fundamental assumption about nature was not precisely true. It is true
under rather limited conditions; correct in the very simple case of reflection by a
plane mirror set in a homogeneous medium like ordinary room temperature air.
Later it was modified: first by suggesting that light takes the path that takes least
time, and later still further.
20
This simple result has far reaching consequences too. Consider the following
problem set by Aristotle: if a circular light set in the ceiling shines through a square
, ,   353
hole in the floor, what is the shape of the light spot on the basement floor? (This
question is, as it were, a precursor to the discussion of the camera obscura constructed
by the Arab mathematician Alhazen in the tenth century.)
21 Willebrord Snel van Royen (Netherlands, early seventeenth century). It is
also called the Law of Sines because, in modern terminology, the ratio of the sines
of the angles of incidence and refraction equals the ratio of velocities in the two
media.
22 For animated demo see http://micro.magnet.fsu.edu/primer/java/refraction/
criticalangle/index.html
23
Teller 2001a; see further his 2008.
24 I am staying here with classical physics, which suffices to make the current
point about vagueness in mathematical-physical representation; the same point

could be made for later physics, mutatis mutandis.
25 i.e. locally determined by a convergent power series.
26 For Descartes, a geometrical problem is always a problem of construction
with a ruler and compass; only when such a construction can be achieved with
a continuous movement can we have a clear and distinct idea of the geometrical
solution. The exclusion of the infinite is a constant theme in Descartes; cf. Principles
I, Proposition 26. In attributing a contrary opinion to Pascal we don’t mean of
course that he had the conception developed in point set theory after 1900; though
he went very far in his rejection of the apparent paradoxes of the completed
infinite. Cf. Chevalley 1995.
27
Birkhoff and von Neumann 1936 point out that areas that differ by measure
zero are equivalent as far as anything instrumentally measurable is concerned. A
quotient construction is a reduction modulo some equivalence relation, that is,
replacement of elements by their equivalence classes. In the case of intervals, for
example, (0,1) and [0,1], (0,1], [0,1) all differ just by measure zero, and would be
treated as the same.
28 At that point we can either say or deny Birkhoff and von Neumann’s
contention that the calculations can go on as usual, but the shape is correctly
represented not by one region in geometric 3-space, but by an object in the
quotient construction that identifies regions modulo differences of measure zero. If
we go on to meta-mathematics we can extend this critique much further: anything
beyond finite, combinatorial mathematics is susceptible to multiple ‘‘unintended’’
and ‘‘non-standard’’ interpretations, introduced by pointing to very different sorts
of mathematical structures.
29
We may be reminded here of Solovay’s 1965 ‘‘2âleph0 can be anything
it ought to be’’. The foundations of mathematics galloped away with its own
subject, shanghaied it, abducted it (to hear some mathematicians’ complaints about
354  :  
meta-mathematics). The categoricity proofs for geometry, in the early 1900s,

suggest that at least your text will be invulnerable to deconstruction if you say:
‘‘the shape of this glass is a surface in a geometric 3-space defined by these axioms’’.
But if analytic geometry is conceived of as part of set-theory (in the general
set-theoretic reconstruction of mathematics) then these categoricity proofs have a
very limited bearing. They establish only that within each model of set-theory,
those geometric 3-spaces are isomorphic. So the assertion which looked like a
totally closed text, is once more to be regarded as open, vague, and subject to
many alternative construals.
30
To make this concrete, imagine a very small situation, involving only 4 men
and 7 women. Under good conditions the women produce 8 items per hour
and the men 4. Under bad conditions the women produce 4 items per hour and
the men 2. But two men and two women work in good conditions, with two
men and five women assigned to bad working conditions. In the bad workplace,
the production is 2(2) + 5(4) = 24 items per hour. In the good workplace, the
production is 2(4) + 2(8) = 24 as well, precisely the same. So the abstract picture
shows no correlation between production and working conditions.
31 For this topic I have benefited greatly from Sterrett 2002, 2005, and 2006.
32 See Galileo 1964–1966, vol. iv, p. 756, cited in Torretti 1999: 3.
33 The following similar passage in Galileo’s Discourse on Floating Bodies shows
well how easily and naturally he proposes experiments to be done on a small scale
so as to test hypotheses that pertain as well to much larger bodies: ‘‘The third
difficulty in the doctrine of Archimedes was, that he could not render a reason
whence it arose that a piece of Wood, and a Vessel of Wood, which otherwise
floats, goes to the bottom, if filled with Water. Signor Buonamico hath supposed
that a Vessel of Wood, and of Wood that by nature swims, . . . goes to the bottom
if it be filled with water . . . but I . . . dare in defense of Archimedes deny this
experiment, being certain that piece of Wood which by its nature sinks not in
Water, shall not sinke though it be turned and converted into the forme of any
Vessel whatsoever, and then filled with Water: and he that would readily see the
Experiment in some other tractable Matter, and that is easily reduced into several
Figures, may take pure Wax, and making it first into a Ball or other solid Figure,
let him adde to it so much Lead as shall just carry it to the bottome, so that being
a graine less it could not be able to sinke it, and making it afterwards into the
form of a Dish, and filling it with Water, he shall finde that without the said Lead
it shall not sinke, and that with the Lead, it shall descend with much slowness;
& in short he shall satisfie himself, that the Water included makes no alteration.’’
(Galileo 2005: 21)
, ,   355
34 Specifically the example of airplane modeling is held up by Boltzmann in his

Encyclopedia article ‘‘Model’’, as illustrating how alteration in spatial dimensions
can alter other physical characteristics: ‘‘a flying machine, which when made on a
small scale is able to support its own weight, loses its power when its dimensions
are increased’’ (1904: 220). For extended discussion see Sterrett, Wittgenstein Flies
a Kite.
35 For analysis and historical context, see Peterson 2002.
36 These calculations suggest that a nano-machine would not have to worry
about gravity, almost like an ordinary machine in weightless conditions.

37
I refer here to Reichenbach’s striking and little appreciated demonstration that
indeterminism was all along consistent with the success of classical physics. He gives
an example from statistical mechanics to show that the classical deterministic laws
can, with appropriate initial conditions, lead to behavior that is, by measurement,
indistinguishable from genuine indeterminism. The example is exactly one in
which what I call here the Approximation Principle fails. See Reichenbach 1991:
93–5.
38
Buckingham1914; Bridgman1916.
39 quote from sections 160–1 of Fourier 1952.
40 Dimensional analysis, which we can think of as a special application of
symmetry arguments (see e. g. Hornung 2006) has the appearance sometimes

of generating far-reaching empirical consequences from a priori reasoning—a
beautiful topic but which we can only touch on here. For a literate and literary
but rigorous treatment see Barenblatt, 2003; for rare philosophical discussions
see e.g. Ellis1966: 127–51 and Sterrett 2006: 127–8, 135–8, 180–202; Lange
[forthcoming].
41 See further Hornung, op.cit.; for the dimensionless parameters as invariants
of transformation groups see Barenblatt 2003: 94–6. The seminal paper relating
dimensional invariance to more general notions concerning symmetry, with special
reference to measurement scales, is Luce 1978.
42
The law is that force is proportional to mass times acceleration; the more usual
specific form F=ma does depend on choosing the units such that one unit of force
gives a unit mass unit acceleration.
43 For a precise exposition, see Barenblatt 2003: section 1.2.2. This requirement
of scale invariance for what is significant is generalized to invariance under any

‘‘rendition’’ or definition of some quantities in terms of others, such as ‘‘velocity’’
as ‘‘distance/time’’. Buckingham gave the example of two sets of quantities that
can furnish the basic units for mechanics: Force, density, and linear speed can be
expressed in terms of mass, length, and time; but the converse is also the case
(Buckingham, op. cit: 348–9, with force to have dimensions MLT−2 , density
ML−3 , and speed LT−1 ).
356  :  
3. Pictorial Perspective and the Indexical

1 ‘‘Painting is a kind of natural philosophy, because it imitates the quantity,
quality, form and character of natural objects’’, Pino, Dialogo di pittura, Trattati 1:
109; cited Gopnik1995: 99; see further his pp. 95–102 on how Leonardo da Vinci
and his contemporaries saw the relation between the sciences and the visual arts.
2
For general studies on the development of geometry, optics, and perspective
in art with reference to their role in the birth of the new sciences, see Edgerton
1975, 1991 and Freeland and Corones 1999.
3 For a synopsis of the relevant history from roughly Heron and Ptolemy to
the end of the eighteenth century, see the introduction in Homann 1991, which
is a translation, with an introduction and notes, of Hugh of St. Victor’s Practica
Geometrica, dated circa 1120 .
4
The history of astronomical measurement as aid in navigation is part of the
history of Perspectiva. Ptolemy’s planisphere is a geometric construction in which
the eye is located at the South Pole, and instead of an intersecting plane just the
Equator—an intersecting line—is used in the projection. The astrolabe improved
on this by adding lines of latitude and longitude drawn from a second point
of view, for an observer located at a certain latitude. This involved therefore a
projection of the celestial sphere on the Equatorial plane. Hence astrolabes could
be used to show how the sky looks at a specific place at a given time. Typical uses
of the astrolabe include finding the time during the day or night, finding the time
of a celestial event such as sunrise or sunset and as a handy reference of celestial
positions. Cf. Veltman and Keele 1986: 42–4.
5 Since PT = h/A and QT = h/B, it follows that the measured distance PQ,
which is just the difference between them, must be h times the difference between
1/A and 1/B. But both PQ and that difference are known from the measurement
results, so h can be calculated directly from the known numbers. For example if
PQ is 100 ft, A = 1 and B = 2 then h = 200 feet. As a historical note, Thales
was credited with calculating the height of the pyramids from the lengths of their
shadows, in the fifth century . One suggestion, however, is that Thales probably
did not prove any general theorem on similar triangles, but noticed (empirically)
that at a given time of day, for every object, the ratio of height to shadow was
the same. So he could measure a nearby stick and its shadow directly, and also the
length of the pyramid shadow—and then the pyramid height would be the only
unknown factor, so could be calculated.
6 In the case of Alberti too we see at least in practice a close connection
between his studies of measurement and of perspective. His Ludi matematici applies
mathematics to the measurement of distances, dimensions, and weights. The
     357
Elementa picturae describes some geometric figures and projections. Immediately

after De Pictura Alberti produced his Descriptio urbis Romae in which he explained
triangulation surveying and included a table of sightings he had made of monuments
in Rome. There he details how he has used a surveying disk similar to an astrolabe.
His De Statua deals with proportions in the human body and how to replicate
them in sculpture. See further the Introduction by Martin Kemp to Alberti 1991.
7 Although Alberti begins with the modest disclaimer that he writes as a painter
rather than as a mathematician, his monograph begins in Book I with a good bit
of geometry and treats his subject in a mathematically sophisticated way.
8
Reputedly, Alberti went on to considerably more sophisticated examples. As
was reported at the time, ‘‘the pictures, which were contained in a very small box,
were seen through a tiny aperture. There you were able to see very high mountains
and broad landscapes around a wide bay of sea, and, furthermore, regions removed
very daintily from sight, so remote as not to be clearly seen by the viewer. He
called these things ‘demonstrations’ ’’ (from the anonymous Vita in Opere vulgari,
ed. Bounucci, pp. cii-ciii; cited in Alberti, p. 98, Explanatory Note 15).
9
In this passage and below—see also the passage quoted below from Leibniz
for a comparison—I am using the term ‘‘appearance’’ judiciously, in a way that
does not equate it with ‘‘phenomenon’’ but allows a distinction between an
observable phenomenon and its appearance (the way it ‘looks’ in the content of a
measurement outcome). Recall the remarks on this distinction in the Introduction
to Part I.
10
See Panovsky 1956. While these mechanical aides and machines may or may
not have been actually made or used, their design brings out clearly how the
perspectival drawing technique is actually the recording, in a sophisticated visual
display, of the outcome of a complex spatial measurement.
11 From Part I of Thomas 2006.
12 A crucial complication comes in when it is realized by the end of the
seventeenth century that light rays do not provide an instantaneous connection.

While this point goes to stage center with Einstein’s famous 1905 discussion, it
is well enough developed in astronomy throughout the modern period. When
focusing on the unmoving or instantaneous case, one does have the luxury of
ignoring the finite speed of light and thinking of all the connections between
object and screen effected simultaneously. Certainly inaccuracies due to this will
not show up in a rustic landscape or family portrait, but it could even be ignored in
the early debates of whether or not Copernicus saves the appearances of planetary
motion. Descartes still offered the possibility that light is indeed an instantaneous
connection. By the end of the seventeenth century, however, the finite speed
of light was known, and was taken into account in the derivation. So by then
358  :  
it is possible to speak, at least in principle, of true perspectival geometry and

kinematics.
13
Bernard Williams 1978 introduced ‘‘the absolute conception’’; Tom Nagel’s
phrase ‘‘the view from nowhere’’ is also the title of his 1989; Eddington coined
‘‘the point of view of no one in particular’’.
14 Notes for the letter to Des Bosses, February 5, 1712: 199–200 in Leibniz
1989, cited in Jauernig 2004, chapter 1. The passage begins with ‘‘If bodies are
phenomena and judged in accordance with how they appear to us, they will not
be real since they will appear differently to different people. And so the reality of
bodies, of space, of motion, and of time seems to consist in the fact that they are
phenomena of God, that is, the object of his knowledge by intuition.’’
15 See Ryckman 2005: 128–35 on Weyl (especially the quoted passages on
page 134), and on Eddington, 183–4.

16 Recall, from the Introduction to Part I, the distinction between ‘‘phenomen-
on’’ and ‘‘appearance’’.

17
I’ll keep within that context for now. The subject can certainly be pursued
within Euclidean geometry, especially as rendered analytic by Descartes and
supplemented soon thereafter by the infinitesimal calculus of Newton and Leibniz.
So it is perhaps not so surprising that after a good start by Desargues and Pascal
the subject of projective geometry itself languished and was only completed more
than a century later.
18 A few simple results will illustrate this. If two line segments, one k times
as long as the other, lie on a line parallel to the painterly window then their
projections on that window will also be in proportion k:. If two line segments
parallel to the window are equal in size, with one edge on the central line of
sight, but the one is at a distance from the eye m times that of the other, then its
projection on the window will be /m as large as that of the other.
19 We focus here on the cross ratio of collinear points. There is also a dual, the
cross ratio of a quadruple of lines intersecting in a point.

20
Also called anharmonic ratio, and anharmonic section. There are a number
of different conventions and definitions in use in the literature; I use here the one
of Coxeter and Greitzer 1967: Section 5.2, 107–8. None of the main points made
here hinge on this choice. Also, the four points can of course be ordered in six
different ways, so as a set they have six cross ratios. Given any one of these, the
others can be calculated from it, so we can just concentrate on one.
21 Let the distance between the equidistant adjacent points be called 1 unit;
then CA/CB = 2/1 and DA/DB = 3/2, so their ratio is 4/3. If the painter placed
the points meant to correspond to these with e.g. distances AB = 9, BC = 3,
     359
CD = 1, the cross ratio on his picture plane would be 12/3 divided by 13/4,
which is pretty close to 4/3, perhaps close enough for the purpose.
22
Because the cross ratio is equal to the same magnitude with the sines of the
corresponding angles replacing the line segments:
CA/CB : DA/DB = sin(CPA)/sin(CPB) : sin(DPA)/sin(DPB)
But those angles, and hence the sines, are the same if we look instead at the
projected points A , b, C, d for example. So that must also equal CA /Cb : dA /db.
The proof is elementary, using the fact that the area of any triangle equals on the
one hand half of the height times the base, and on the other hand the product of
two sides times the sine of the angle between them. All the relevant triangles in
the diagram with vertex P have the same height.
23 A demonstration experiment, to be distinguished from experiments that test
hypotheses on the one hand, and from experimental exploration on the other.
For Brunelleschi’s ‘experiment’ see for example Zajonc 1995. Whatever process
Brunelleschi actually used, it produced a painting that provides a view of the
Baptistery from a certain vantage point, and is thus properly conceived of as a
measurement of the Baptistery. See also Feyerabend 2001: 89–115 in his chapter
‘‘Brunelleschi and the invention of perspective’’, and my review thereof 2000a.
24 In the case of Alberti too we see at least in practice a close connection
between his studies of measurement and of perspective. His Ludi matematici applies
mathematics to the measurement of distances, dimensions, and weights. The
Elementa picturae describes some geometric figures and projections. In Descriptio
urbis Romae he details how he has used a surveying disk similar to an astrolabe; De
Statua deals with proportions in the human body and how to replicate them in
sculpture (see Martin Kemp’s Introduction to Alberti 1991).
25 For this example see Geometry Forum Articles at http://www.geom.uiuc.
edu/docs/forum/, ‘‘Photo Puzzles’’.

26
Having but three points A , B , C to relate to the ‘eye’ at P, we can take the
degenerate cross ratio CR(A , B , C ) to be the limit of the numbers CR(A , B ,
C , D ) as D moves along the ground farther and farther away beyond C . That
limit exists because the difference between segments D A and D B diminished
then toward zero, so that the ratio D A /D B tends toward 1. CR(A , B , C )
turns out, on this construal, to be just C A /C B .
27
In this example the person is at A and the distance between B and C is one
mile. The cross ratio CR(A, B, C, D) is found by measurements in the photo, and
equals C A /C B , but C B = 1 (in miles).
28 To see this idea developed in a different key, see Ismael 2007.
29 For philosophical background see especially Giere 2006: ch. 4; my 1993,
sect. 1; Sismondo and Chrisman 2001; Ismael 1999.

360  :  
30 http://www.hewett.norfolk.sch.uk/CURRIC/soc/theory.htm
31
http://www.indiana.edu/∼intell/map.shtml
32
‘‘Mediology: a metatheory for analyzing media and institutions’’; see
http://www.georgetown.edu/faculty/irvinem/CCTP748/mediology-map.html
33 We can relate this to Lewis 1979, where he construes all assertion as self-
ascription of properties, and writes ‘‘What happens when he believes a proposition,

say the proposition that cyanoacrylate glue dissolves in acetone? Answer: he locates
himself in a region of logical space’’ (p. 518). This must be understood, in his
context, in terms we need not share, e.g. that a region of logical space is a set
of possible worlds; but it is not difficult to assimilate his general conception of
assertion and belief in our own terms.
34 In view of the strictures I will relate below, I try to be careful not to use
‘‘perspective’’ terminology in this section—while self-location is a crucial hallmark

of perspective, the notion of perspective involves more than that, and we need to
be careful to distinguish the precise characteristic under discussion.
35
There are obvious connections here with the literature begun with Perry
1979. Much of that literature focuses however on belief in general, which will not
be the focus here; the closer relation is to Lewis 1979 which certainly concerns
belief but only as one example in a large range that includes bare assertion.
36 See further Pooley 2003; van Fraassen and Peschard 2008: Part II.
37 Compare page 367 of the translation on pp. 361–72 of Kant 1992. The
full text is this: ‘‘Wenn ich auch noch so gut die Ordnung der Abteilungen
des Horizonts weiss, so kann ich doch die Gegenden darnach nur bestimmen,
indem ich mir bewusst bin, nach welcher Hand diese Ordnung fortlaufe, und die
allergenaueste Himmelskarte, wenn ausser der Lage der Sterne untereinander nicht
noch durch die Stellung des Abrisses gegen meine Hände die Gegend determiniert
würde, so genau wie ich sie auch in Gedanken hätte, würde mich doch nicht
in den Stand setzen, aus einer bekannten Gegend, z. E. Norden, zu wissen, auf
welcher Seite des Horizonts ich den Sonnenaufgang zu suchen hätte.’’ (Kant 1983:
995/6)
38
I hesitate to use such terms as ‘‘applied science’’ and was hesitant to insert the
word ‘‘pure’’ into this discussion: these points about indexicality pertain equally
well to experimental research in pursuit of theoretical goals.
39 We can think of Nelson Goodman as making this into a general point for
the viewing of art works as well, with his introduction of the concept of languages
of art.
40
To give but a few examples from the Oxford English Dictionary: ‘‘ F.
MYERS Catholic Thoughts IV. xxxv. 359 Clearly no method can be satisfactory
but that which preserves the perspective of history true.  H. DRUMMOND
      ⁽⁾ 361
Lowell Lect. Ascent Man 11 Evolution . . . has thrown the universe into a fresh
perspective.  J. GOULD & W. L. KOLB Dict. Social Sci. 262/1 There has
been much discussion from many perspectives as to the origins and ‘‘causes’’ of
fascism.  Review No. 53. 21/1 Aiming for a 100 per cent safety record is the
right thing to do, not just from an ethical standpoint but also from a hard-nosed
business perspective.  W. JAMES Let. 12 Dec. in R. B. Perry Thought &
Char. W. James (1935) I. 727 Metaphors and epigrams which, witty and striking
and perspective-suggesting as they often are, . . . may be in danger of having the
changes rung on them too long.’’
41
But note my qualifiers ‘‘in general’’ and ‘‘in their official formulation’’—I
am not ruling out examples that bespeak the contrary of the general claim.
42 To give one example from epistemology: as De Finetti emphasized, subjective
probability is more realistically modeled by finitely additive functions. But that is

not a very tractable space, so one closes it by adding in limit points for certain kinds
of sequences, thus embedding it in the much more tractable space of probability
measures.
43
The tale I mean is largely told in Ryckman 2005, though from a transcend-
entalist rather than empiricist point of view.
Part II: Windows, Engines, and Measurement

1 cf. Maddy 2007, p. 2, describing a character who is thoroughly at home in
current science ‘‘She uses what we typically describe with our rough and ready
term ‘scientific methods’, but again without any definitive way of characterizing
exactly what that term entails. She simply begins from commonsense perception
and proceeds from there to systematic observation, active experimentation, theory
formation and testing, working all the while to assess, correct, and improve her
methods as she goes.’’
4. A Window on the Invisible World (?)

1 I have here benefited greatly (although our use of the terms ‘‘representa-
tion[al]’’ and ‘‘image’’ is not quite the same) from Pitt 2005.
2 Quoted Alpers 1983: 7. The Latin text makes clear, in the preceding sentences,
that Constantijn Huygens (not to be confused with his more famous son Christiaan
Huygens) conceives of the experience with the microscope as a case of seeing.
3
Heidelberger 2003. For an illuminating discussion that relates Heidelberger’s
conceptions to those of Baird and Harré, see Boon 2004.
4 This is precisely what Pitt 2005 discusses critically, addressing how the results
are presented in nano-technology literature. Contrast his discussion with the view
argued in Hacking 1985 and my reply to that paper in the same volume.
362  :  
5 See Boon, op. cit. Nancy Cartwright’s [1999] concept of a nomological engine
was certainly also one of the factors inspiring me to see putative ‘observation by
instruments’ as involving engines of creation, although her concept is at the same
time different from, as well as related to, that of instruments as providing engines
to produce phenomena. Specifically, her notion of a nomological engine cannot
be understood apart from some grasp of her notions of capacity and necessity. But
the examples are of the creation of processes that instantiate certain patterns with a
faithfulness achievable in the laboratory and not found in an uncontrolled natural
setting.
6
See further Anderson 1993.
7 In summary, Mieke Boon offers the following classification that covers
Heidelberger’s among others: ‘‘Simplifying somewhat, I assume that the categor-

izations proposed by these authors share enough features to be merged into three
types, which I shall call Measure, Model, and Manufacture. Measure is a category
of instruments that measure, represent or detect certain features or parameters of an
object, process or natural state. Model is a type of laboratory systems designed to
function as a model of either natural or technological objects, processes or systems.
Manufacture is a type of apparatus that produces a phenomenon that is either
conjectured from a new theory or not as yet theoretically understood.’’ (Boon
2004: 223)
8 The phrase ‘‘creating new phenomena’’ comes from Hacking 1983. But
although his illustrations of this idea are grist to my mill, it is clear also that there
are real differences in what we mean by this phrase. As I use the words, ‘‘observable
phenomena’’ is just an emphatic way of saying ‘‘phenomena’’, for by that word I
refer to all and only observable, i.e. perceptible, objects, events, and processes. See
further the discussion of phenomena versus appearances in Part IV.
9 Gilbert is said to have demonstrated to Queen Elizabeth I his theory of the
Earth’s magnetism: placing a small compass at various places around the terrella,
Gilbert showed that it always pointed north–south—offering this as a model to
explain why on Earth a compass points north–south. (Actually, even if the Earth’s
core is iron, it is too hot to be a magnet.)
10 For discussion both of the experiment and complaints about its poor use in
reflections on methodology, see Worrall 1989 and Cantor 1989.

11 Cantor 1989 remarks that the phenomenon had actually been observed
almost a century earlier. Its use to illustrate how new phenomena are created by
experiment does not depend on Poisson being the first one to create it, of course.
Even less does its being new have anything to do with the methodological value or
lack of value of novelty in predictions, if only because being new does not imply
that it should have been surprising.
      ⁽⁾ 363
12 This way of presenting the story of an experiment, emphasizing creation

rather than observation, may seem to underplay the role of measurement in
empirical research. It does not: when such experiments are done—and today it is
rare for them to consist of a single clearly recognizable observable phenomenon that
can play such a role—the result takes the typical form of a summary presentation of
measurement results. So we will not have a clear and synoptic view of experiment
and its relation to theory until we have clarity on the topic of measurement itself.
For further comparison and views on the roles of creation as well as inspection of
phenomena see for example Galison and Assmus 1989.
13
The idea that scientific instrumentation undermines the observable/ unob-
servable distinction, touted especially with the advent of scientific realism in the
1960s, is thus not exactly new!
14 In addition to acknowledging my debt to Catherine Wilson’s The Invisible
World, I would like to thank the University of Oklahoma, and especially Dr. Kerry
V. Magruder, for access to the History of Science Collections to obtain material
from Hooke 1665 and Power 1664.
15
Although the distinctions drawn to classify instruments in science seem
usually to go against this assimilation, I find this suggestion for the microscope
also in Michael Heidelberger’s ‘‘Roentgen’s apparatus was, as we might say,
unconditionally productive, but there are other productive instruments that produce
known phenomena—although in circumstances where they have not appeared
before. I am thinking of instruments, like microscopes or telescopes . . . .’’ (op. cit:
146/147).
16
The rainbow is unlike reflections in the water because it is not the image
of some real arch. That is important to illuminate the point below. But the more
important feature is the status they both share with mirages (and share, I will argue,
with microscope images), which makes them ‘‘public hallucinations’’.
17 The qualities and structure of the rainbow do reveal something about the
light source; yet the rainbow is not a copy or picture of that source.
18
Hence the anecdotal evidence, in ancient sources, of the appearances of
gods and spirits, which could have been produced by projecting images made by
concave mirrors into smoke in a darkened room.
19 In such ‘quantifier’ locutions as ‘‘something’’ or ‘‘there is such a thing as’’
or ‘‘everything’’, the word ‘‘thing’’ does not occur with any substantive meaning,
but is a sort of pronomial device. In elementary logic we paraphrase ‘‘Something
is . . . ’’ as ‘‘There is x such that x . . .’’ and ‘‘Everything . . . is—’’ we render as
‘‘(All x)(if x is . . .then x is —)’’. The word ‘‘thing’’ has disappeared. Two of
the three occurrences of ‘‘x’’ correspond there to the relative pronoun ‘‘it’’. But
the first occurrence of ‘‘x’’, corresponding to the ‘‘thing’’ part of ‘‘Everything’’,
364  :  
does not play a different role from the others. (This is clearer in combinatory
logic: a universally quantified sentence says that a certain predicate has universal
application.) In venerable terminology, use of ‘‘thing’’ in ‘‘something’’ is not
categorematic but syncategorematic.
20 Let me explain how I understand this. My experiences are the events that
happen to me of which I am aware. Such events have two sides, so to say: what
really happens to me and the spontaneous judgment I make in response, which
classifies that event in some way. In good cases the two coincide, but often they
do not. For example, I trip over a marmot but take it to be a cat. What happened
to me was that I tripped over a marmot, but I ‘experienced it as’ tripping over a
cat. See further my 2002: 134–6.
21 The most sustained objection along this line is that of Hacking 1981, which
actually also details tellingly the differences between ‘‘seeing through’’ an optical
microscope and a magnifying glass, but still argues for the realist conclusion. See
further my reply ‘‘Ad Ian Hacking’’ in the same volume.
22
Besides the above reply to Hacking 1985, see e.g. my 1982a, and the critique
of Inference to the Best Explanation in my 1989, chapter 6, and in my review of
Lipton 2005.
23 While I cannot go further with this here, I mean to echo at least some aspect
of the transcendentalist story of the constitution of theoretical objects when such

correlations appear.
24 The discussion of images, such as are projected on screens or by reflections
in water, certainly introduces a sort of observation report that I did not have in
mind while writing The Scientific Image. Then I was thinking quite simply in terms
of a classification of objects, events, and processes as observable and unobservable.
25 John Bell 1990 said famously that ‘‘ordinary quantum mechanics is just
fine FAPP’’ (introducing this abbreviation for ‘‘for all practical purposes’’), while
criticizing interpretations of the theory as unsatisfactory if they remain at that level.
26 To continue the preceding footnote about experience: to have the experience
of seeing a paramecium means here to have the spontaneous judgment that one
is seeing a paramecium, in response to what is happening to one, namely to have
one’s eye pressed to the eye-piece of a microscope.
27 His footnote at that point indicates, however, that he wrote this passage
in response to a comment by a philosopher, Christopher Hitchcock. In any

case, Weinberg’s opinion of philosophy should not make us discount his own
philosophizing.
28
Once you do, however you do it, we will see that you have a view of science
instantiating the same pattern as constructive empiricism.
29 I won’t elaborate on this here, but empiricist positions on philosophy are
stances, and the lines drawn for any useful distinctions are sensitive to value,
    365
purpose, use, and other contextual factors. With respect to science, as to other
such important, pervasive aspects of our culture and civilization, the values an
empiricist pursues require rendering it intelligible without the need for metaphysical
underpinnings.
30 For a contrasting, realist way of appreciating this harmony, see Norton 2000
and 2003.
5. The Problem of Coordination

1 The contrary opinion, that the new terms do admit of explicit definitions that
use only old terms, was argued by David Lewis 1970; for critique see my 1997b.
2 For Mach see further below; the reference is to ch. II, especially sections 11
and 12, of Mach 1986.

3
Michael Friedman 2002 describes Moritz Schlick’s Space and Time in Contem-
porary Physics (1917) as follows: ‘‘Schlick portrays the variably curved space-time
of general relativity as an entirely abstract, entirely non-intuitive ‘conceptual con-
struction’, which can only be related to experience and to the physical world by an
entirely abstract, entirely non-intuitive relation of ‘designation’ or ‘coordination’
by which the purely mathematical ‘conceptual construction’ . . . can then receive
empirical content by being interpreted in terms of physical measurement.’’ (p.
136). See further the extensive discussion in Ryckman, 2005, sections 2.4.1 and
3.2. Thanks also to Anja Jauernig for helpful discussion of how Schlick used and
understood ‘‘coordination’’ in his Algemeine Erkenntnislehre.
4 This well-known essay is illuminatingly analyzed in Ryckman (who cites this
passage) 2005, section 3.3.

5 For a sustained critique of equivocations in Reichenbach’s discussion of
coordination in The Philosophy of Space and Time, see Klein 2003.

6
The chapter from which I quote is called ‘‘Cognition as Coördination’’.
7
The set-theoretic context he chose to make this point is not at all essential to
the point. Suppose we use only algebraic terms and ask about two entities whether
they are isomorphic, say, or whether the second can be homorphically embedded
in the first. As thus stated, the question has no sense. At best it is elliptic, for the
terms are context-dependent. The question receives different answers depending
on the parameters selected as contextually relevant. Take for instance a particular
Hilbert space and ask whether the family of a certain kind of Hermitian operators
on it is isomorphic to a given permutation group. The former may indeed form
a group, under certain operations, and considered only as such, be isomorphic to
the latter. But obviously by considering it only as such we are ignoring a great
deal—the permutation group defined in terms of shuffling operations on a domain
has nothing to do with Hilbert space.
366  :  
So the assertion or denial of isomorphism depends on a certain selection on

our part. In the case of two mathematical objects we can make the selection in a
straightforward way, since they are already ‘given’ in a format which lends itself to
us. Given a particular Hilbert space and a family of operators on it singled out by
some equations, the relevant questions can obviously be formulated: for example,
does this family contain an element I such that for all its members X, IX = XI
= X? But how do I formulate questions of this sort for a part of nature, without
using a selective description of it that already rests on a ‘mathematization’?
8
I am deeply indebted to the writings of Michael Friedman, Flavia Padovani,
and Thomas Ryckman for my understanding of this episode (though of course
they are not to be reproached for any misunderstanding on my part).
9 In the case of classical mechanics, the axioms of coordination are Newton’s
laws of motion, and his law of universal gravitation (which can only be formulated
in the context where the former are given) is an axiom of connection.
10 It is instructive here to see how, almost just in passing, Reichenbach touches
on ways to back this up: ‘‘Not only the totality of real things is coordinated to
the total system of equations, but individual things are coordinated to individual
equations. The real must always be regarded as given by some perception. By
calling the earth a sphere, we are coordinating the mathematical figure of a sphere
to certain visual and tactile perceptions that we call ‘perceptual images of the earth,’
according to a coordination on a more primitive level. If we speak of Boyle’s gas
law, we coordinate the formula p.V=R.T to certain perceptions, some of which
we call direct perceptions of gases (such as the feeling of air on the skin) and some
of which we call indirect perceptions (such as the position of the pointer of a
manometer). The fact that our sense organs mediate between concepts and reality
is inherent in human nature and cannot be refuted by any metaphysical doctrine.’’
(op. cit. page 37). But under what conditions do we have a right to call those visual
and tactile perceptions ‘‘perceptual images of the earth’’? It seems to me we cannot
classify them as such unless we are already able to describe perceptual images by
themselves on the one hand, and the earth on the other. If that condition is not
met, coordinatization seems to amount to no more than equating one otherwise
meaningless noise to another.
11 There is a good sidelight on the problem in Campbell 1943 where he discusses
‘‘on what experiments the Newtonian theory of dynamics is most suitably based,
and in particular whether quantity of matter, mass, and force can be measured
independently of that theory.’’ That the independent measurement procedures
he discusses involve reliance on previous coordination of some parameters with
measurement procedures is all too clear. Nor do his arguments meet the points
made by Poincaré and Sneed (see note below).
    367
12 Poincaré pointed out that to measure the mass of an object, in the sense of
Newtonian physics, we must presuppose the object to be a Newtonian mechanical
system, to which Newton’s laws of motion apply. Joseph Sneed later made this
insight the basis of his theoretical/non-theoretical distinction.
13 Mach provided what he called a definition of mass, for example, but to say
that is to use his own words; by our present understanding of ‘‘definition’’ he did
not do that—we should rather read it as another example of attempts to spell out
a coordination. Patrick Suppes’ comment (1957: 298) is strictly speaking correct,
but for a more balanced exposition of what Mach was up to, see Koslow 1968.
14
What about modality: the object has such and such a feature if a measurement
would have outcome so and so, if the measurement were made? This pushes the
problem back a step: how do such modal assertions receive empirical content?
15 Once that theory is accepted, it begins to infect the language in use; the
language in which the measurement procedures themselves are described becomes

new-theory-laden, and of course after a while what had been the new, novel theory
sinks into the background of research (until and unless its empirical adequacy is
drawn into doubt).
16 This sort of consideration has sometimes been seen to support scientific
realism against empiricism. But that is a confusion: the eventually established

theory classifies the observable measurement processes and set-ups in theoretical
terms, and that classification is the assumed starting point when predictions are
to be made via the theory from the measurement results obtained there. There
is nothing un-empiricist in the remark that the ink in this book, for example, is
classified in current physics as having a certain kind of molecular structure, and
that predictions about its observable fading can be made on that basis.
17 Cf. Peschard 2007.
18 For the early history I take the main narrative details from Middleton 1966,
which enlarges on and corrects the history that Mach himself provides in Mach
1986. There are now more up-to-date treatments (see for instance Chang 2004)
but the main philosophical points I wish to take up are all in Mach’s treatise. As
I will say also of Poincaré, Mach is presenting us with a ‘‘just so’’ re-creation of
the history, to make the philosophical points, even though as conscientious as he
could be with respect to the historical details.
19 The entire passage from the records of the Accademia is quoted in Middleton,
pp. 33–4.
20 That what is measured is still, in our terms, a combination of mutually isolable
factors, remains true though. Middleton writes about Hooke’s liquid thermometers:
‘‘While it was an excellent attempt, it suffered from several disabilities: spirit of
wine is not a well-defined substance, its properties varying rapidly with the
368  :  
amount of water in it; Hooke’s choice of the freezing point of water, rather than
the melting point of ice, was unfortunate; and he did not make any allowance for
the difference in expansion of brass—or silver—and glass. It may be noted that
Réaumur inherited the first two of these sources of error.’’ (op. cit. page 46)
21 Liquid water has its maximum density at 3.98 ◦ C, and expands both if the
temperature is decreased and if it is increased at that point. At exactly 3.98 ◦ C,

the thermal expansion coefficient of water is actually zero. Plotting density against
temperature yields a graph that is a parabola.
22
Cf. Mach 1986: ch. II, section 8–10, 14.
23
quoted Mach 1986, ch. II, section 14, 54.
24 Amontons investigated the relationship between pressure and temperature
in gases near the end of the seventeenth century, and his results led him to
the speculation that a sufficient reduction in temperature would lead to the
disappearance of pressure. As Hasok Chang points out, this is a different notion
of ‘‘absolute temperature’’ from that initially introduced by Kelvin, though later
the two were assimilated. Kelvin was originally concerned to give an independent
standard for equality of temperature intervals, not tied to the choice of one
‘‘standard’’ thermometric substance. For this purpose he drew on Carnot’s work,
using the measure of work equated with the descent from a given higher
temperature to a lower one: a specific amount of work could mark a difference of
one degree, so that ‘‘all degrees have the same value’’. See Chang 2004: 173–86.
25 Poincaré, H. The Value of Science, ch. II. ‘‘The measure of time’’. The page
numbers cited below are for the French edition, Flammarion 1970 (original 1905).
26
This begins section III; I am ignoring the passages in which Poincaré
concentrates on psychology.
27 Cf. my 1970: ch. III-2-c, pages 78–81 concerning Bosanquet, Russell, and
Russell’s debate with Poincaré on this subject. In 1897 Russell still states at least
that ‘‘No day can be brought into temporal coincidence with any other day . . . .;
we are therefore reduced to the arbitrary assumption that some motion or set of
motions, given us in experience, is uniform’’ (Russell 1996: section 151, 155), but
his views expressed in that debate shortly afterward were already quite different.
See further Grünbaum 1968: 44f.
28 This is analogous to the question whether the spatial congruence relation is
‘‘objective’’ or ‘‘intrinsic’’, a ‘‘matter of fact’’, as e.g. Russell maintained at one

point in exchanges with Poincaré (as we will discuss in connection with the roots
of Structuralism).
29
For an account of the entire expedition and its results, see Olmsted 1942–43.
30
‘‘On admet, par un définition nouvelle substituée . . . que deux rotations
complètes de la terre autour de son axe ont même durée.’’ (p. 43) This was in
    369
keeping with the ancient form of time reckoning that specified units of time
in astronomical terms. Note that the 24-hour period thus defined, in terms of
passage of a fixed star across the meridian is not equal to the older one defined
in terms of the sun’s passage across that meridian, with an average difference of
almost 4 minutes. The latter, however, are not equal to each other if reckoned
in sidereal time; ‘sun days’ are not the same during the different seasons of the
year.
31 In section IV he considers the suggestion that the case of the pendulum
illustrates that some general principle may be thought to justify a definition of

this sort, namely the principle that identical phenomena have the same duration.
But, as he points out, this is empty unless it has been spelled out what we count
as identical phenomena, and if that has been specified then the ‘principle’ is an
empirical assumption(Ibid., 44–5).
32 ‘‘ils définissent la durée de la façon suivante: le temps doit être défini de telle
façon que la loi de Newton et celle des forces vives soient vérifiées.’’ (Ibid. 46)
This could perhaps be said more appropriately about Euler; see my 1985, section
III-2-a. 73–4.
33 ‘‘Le temps doit être défini de telle façon que les équations de la mécanique
soient aussi simple que possible’’ (Ibid.).

34 E. A. Milne first made his suggestion in 1937; see his 1948: 22; Whitrow
1961: 46, 247; Roxburgh 1977. See also the discussion in Grünbaum 1952 followed
by Grünbaum 1954 and Whitrow 1954: 151; Grünbaum 1968: 18–19.
35
As Fine 1986 showed clearly, Einstein’s views are not univocal; in this
connection I would cite specifically his 2004a and the debate he imagined between
Reichenbach and Poincaré (Einstein 1988: 676–79).
36 See Ryckman 2005 and my 2007.
37 From the Latin ‘‘scrupulum’’ meaning a small sharp, or pointed, stone: an
unfounded apprehension and therefore unwarranted fear that something is a sin

when in fact it is not.
38
Would it be too polemical to call this ‘‘naïve scientific realism’’?
39
See further below for this problem in a more abstract setting, where it takes
a form that threatens paradox for any Structuralist conception of science.
40 I have been using Putnam’s (perhaps somewhat derogatory) term 
here, and I take the response that we are presently exploring to be precisely of the
sort that Putnam called metaphysical realism. What I call Reichenbach’s problem
is certainly not unrelated to the problems raised by Putnam in his celebrated
‘‘model-theoretic argument’’, which we shall have occasion to examine as well.
41
Not to mention the required previous coordination for the classifications of
height and mercury, for example.
370  :  
42 In the technical sense of the logic of questions—the subject of what Nuel

Belnap called the theorem of the fifth gymnosophist: Ask a foolish question, get a
foolish answer.
43 This is neo-Kantian terminology, of course, and could easily be misunder-
stood if taken out of context. The assertion that temperature, as we now understand
it, is constituted in the historical development starting in the time of Galileo does
not imply, for example, that it is not now correct to say that there were days
before his time when the temperature in someone’s cellar dropped to 4 ◦ Celsius.
Compare the carefully evenhanded treatment in Bitbol [forthcoming a] of Latour’s
discussion of whether Ramses II died of tuberculosis.
44 This is not to deny that the theoretical representation goes beyond what can
appear in any physically possible measurement. In a classical physics text there is

no objection to stating a problem about a particle with mass 1/π grams. But there
is no physically possible series of mass measurements that literally converges to that
value, only ones that will not contradict it.
45
Just for now, let me remark on the Duhemian view that what theories
provide are new classifications for observable things. To say, for example, that
light is a wave disturbance in the ether is to classify observable optical effects in
the taxonomy provided by classical electrodynamics. To say that these observable
effects are thus classified (correctly, in that taxonomy) is not to imply that the ether
is real. Peter Lipton seems to me to be confused on this point, when he suggests
that a constructive empiricist must be involved in such simple self-contradictions
as ‘‘my computer is on the table, which is a swarm of particles, but there are no
particles’’ (2004: 146).
46 By writing in terms of a single theory cum measurement practice, stabilized
in this historical process, I have also simplified the matter. There is typically a broad
range of theories involved; when eventually the community achieves stability in
the measurement of a theoretically identified parameter, that will generally involve
concordance between a variety of different procedures, for which the harmony in
their results is explained theoretically by their counting as measurements of the
same quantity. The main points, about how theory and practice are entangled in
this process, and how that involves elements of choice as well as appreciation of
empirical regularities, remain the same.
6. Measurement as Representation: 1
1 Perhaps we’ll be less happy, however, to call it a painting of the Duke than a
portrait of the Duke. On this distinction see Freeland 2007.
  :  371
2 Relevant to quantum mechanics is the question whether a theoretical descrip-

tion of this physical correlate must suffice to tell us how the object ‘looked’ in the
outcome. I’ll bracket that question for now, but return to it later.
3 That a claim is theory-laden does not mean that it presupposes the truth of a
theory, only that theoretical terms are used in its formulation. For example, ‘‘this
powder is classified in chemistry as sodium nitrate’’ is theory-laden but does not
imply anything about whether the chemical theory is true or false.
4 Cf. Weinberg’s 1998 discussion of mass, and my 2002: 115–16.
5
Even if not equally salient, the situation is not in principle different in
earlier physics. The thermometer, for example, is itself an object in the domain
of thermodynamics, and the theory is required, for its coherence, to admit a
description of the thermometer-object interaction that satisfies the relevant criteria
for measurement.
6 Those questions can appear also even if the theory is well established and
accepted; however, they will play a visible role mainly if the measurements are
being made at the theoretical threshold of accuracy and precision. Examples include
the gravitational disturbance by ‘test particles’ in general relativity, which can be
lessened by reducing size, while in quantum mechanics reducing size increases
uncertainty.
7 Perhaps I am being too charitable. If this extreme reaction ever had any
plausibility, that may have been partly because it is generally only in a new
theoretical context that the relevant interactions first occur. But if the point needs
to be hammered home, take a simple example: you and I inspect a paper with
black marks on it, I recognize it as a Greek word and can read it while you have
no idea what it is. Are we seeing the same thing? Undoubtedly. If there is any
doubt, I will ask you to take a pin and trace the black marks. You may not do it
in the order in which I would, but you certainly trace the same marks—so you
are seeing those very same black marks. If it is objected that here I chose a context
in which you and I do share pertinent concepts and apply them in the same way
to the object before us, I answer that this is inevitable. No matter how far back
we go, with examples of comparative ignorance or lack of similar education, there
is a common background sufficient to ensure communication. Or, if you like, if
there is not, as with perhaps an alien intelligence, then the question of whether
we are seeing the same thing doesn’t arise at all. There is no intelligible Robinson
Crusoe state without prior knowledge or opinion, a tabula rasa waiting to be
inscribed by bare experience. But that we see the same thing does not mean that
you can see it as writing, or as Greek, let alone read it. So for me this object
is classified, is assigned a location in a logical space, which may not be available
to you.
372  :  
8 The question is rhetorical of course; in many discussions of quantum mech-

anics the main criterion applied is just that the outcome should predict what
outcome a new measurement made immediately afterward will have. The desire
to gather information about what preceded the measurement is sometimes ruled
out of court altogether. My rhetorical question is not meant to indicate a taking of
sides.
9 Cf. Busch, Lahti, and Mittelstaedt 1996 and my 1991, chapter 7.
10 This is admittedly a bit of an idealization, which we could ameliorate by
limiting to certain ranges of states etc.

11 As Ronald Giere has perspicuously argued, the example of color vision should
already tell you that things aren’t so simple. For do we really want something like
this: if the object has the property that the color objectivist account mentions then the apparatus
will end up with the property that the color subjectivist account mentions? One of the
objections to the objectivist account is that there is no physical property, describable
without reference to observers, which equates in this way to the subjectivist’s
observer’s property. There is no physical property that the object has if and only if it
looks blue to the observer. The most troubling phenomena for the color objectivist
are the contextuality and constancy of color. Observers will report different colors
when the object is placed against relevantly different backgrounds; on the other
hand they will continue to report the same color while the object is subjected to
different lighting conditions. There is a further difficulty in the common inability
to identify colors independently of certain objects or kinds of objects that have
them. Seeing a blue steel ball and a blue woolly sweater, we can always continue
to doubt whether they really have the same color (shade of blue). Sean Kelly 1998:
The second kind of dependency—the dependency of a perceived property on the object
it’s perceived to be a property of—is shown by Peacocke’s example of the height of the
window and the height of the arch . . . and also by Merleau-Ponty’s equivalent claim that
‘the blue of the carpet would not be the same blue were it not a woolly blue’. The basic idea
is that when I perceive a property like height or color, what I see is not some independently
determinable property that any other object could share; rather what I see is a dependent
aspect of the object I’m seeing now. (paragraph 26).
12 I first heard this sort of example from Simon Kochen in a seminar in
Princeton.
13 Also referred to in the literature as ‘‘faithful measurement’’.
14
The first sentence of this section was phrased more carefully than the second
(which is a standard formulation), because of differences in interpretation with
respect to the notion of state. (See for example my [forthcoming] on Rovelli’s
Relational Quantum Mechanics.) It is more interpretation-neutral to discuss the
issue in terms of physical quantities (observables) and their values only, than in
terms of states. But since most discussions are in the standard formulation which
  :  373
takes the notion of physical state for granted, I will ignore this complication except
when very pertinent. This remark applies also to subsequent chapters.
15
These factors can be identified in many alternative ways, when the math-
ematical foundations of a theory are presented. The state is typically something
quite theoretical. In some presentations, however, the state s is simply identified
by (or with, or through) the set {Psm : m is an observable}. On the other hand,
sometimes the observables are simply identified by (or with, or through) the way
in which states assign those probabilities. Also, a simplification is typically achieved
by thinking of all observables as ‘‘made up’’ of simple ones, whose possible values
are real numbers, so that the variable E in Psm (E) can just range over the Borel sets
of real numbers (that is, the sets formed from intervals by infinitary intersections
and unions). We can proceed on a level that abstracts from these different forms of
representation of basic theory structure.
16 The theory must imply more than that the criterion is satisfied in a particular
case; it must imply that this is always so for such cases. In quantum theory this
means that the Hamiltonian which governs this sort of interaction is such as to
guarantee that relation between initial and final states. This point is crucial to
eliminate limiting cases as counterexamples; see the discussion of an objection by
Jon Dorling in my 1991: 221.
17 What I have here related in very general terms is provided in detail for
quantum mechanics in the references above. The reader may naturally wonder
whether our account of measurement so far helps to solve or dissolve the famous
‘‘measurement problem’’ of quantum mechanics. The answer is Yes and No.
On the one hand, the account which takes for granted that surface models
are produced from realizable experimental conditions, and views the empirical
content of a theory as consisting in claims as to how and to what extent these
surface models can be accommodated by the theoretical models, does not run
into conceptual difficulties. On the other hand, in quantum mechanics there is a
core problem—not touched here at all—that can be formulated for the physical
correlate of measurement taken in and by itself (without regard to the ‘‘information
gathering’’ connotation of the word measurement). This core problem is one to
which we will return under the heading of ‘appearance and reality’ in theoretical
description.
18 I call the replacements ‘‘surface models’’ in contrast to the data models; I will
return to this subject later.

19 Veracity in Measurement does of course imply that Value Definiteness is
satisfied at least to the extent that any parameter has a definite value when it is
measured, and that means always, if any occasion at all would in principle allow it
to be measured. But the Criterion for the Physical Correlate of Measurement does
not have any such implications by itself.
374  :  
20 Not without premises going beyond the one that the surface model is thus
related to that state—what are needed are also premises about the sorts of systems
involved, and their relations (general conditions formulated in terms of e.g. the
Hamiltonians). On the assumption that the theory applies, however, these too
are extrapolated from, or made with support of, previous measurement results in
experimental situations.
21 By sequencing such Stern–Gerlach apparatus it is easily shown that the
process does not leave the state unaffected; hence data have to be collected on
different samples of the original beam.
7. Measurement as Representation: 2
1 Francis Bacon insisted, in his clarion call to the founders of the new sciences,
that experience must become literate (Bacon 1994: Novum Organum I:101). By this
he emphatically did not mean to include the sort of bias that practically every
form of literacy has riding along with it. Yet bias is inevitable, both for good and
for bad, and it is not incidental. If we can read at all, our responses are shaped
by presuppositions and assumptions, prior opinion, conditioning, learning, not to
mention strong intellectual commitments and norms that govern our selectivity.
We read measurement outcomes through theoretically-schooled eyes; and this is appropriate
as well as inevitable, while also inevitably hostage to the fortunes of later learning
that shows those spectacles’ distortions.
2 For the purpose at hand we can ignore some complexities, but they will
be remembered from the first chapter. How the object is represented by the
measurement outcome depends on the theoretical context in which it is read:
meaning is not independent of reading. Recall the example of ‘‘burro’’, read
as a word in Italian or as a Spanish word—a symbol has its meaning not
absolutely but due to its role in a (contextually) given language, symbol system, or
representational framework, which in scientific contexts is determined by a theory
governing discourse and practice.
3 Both here and elsewhere I am using illustrations from the history of meas-
urement; it is not to my purpose to present that history. The parts I draw on

are chosen to serve as motivating illustrations for ingredients in the account of
measurement that I am presenting, to some extent in contrast to views that have
been proposed earlier. For the history of the concept of measurement, especially
in the nineteenth century, see e.g. Michell 1993 and Darrigol 2003; but note
the sometimes contentious philosophical differences in these and related historical
literature.
4 Historically this may be said to begin with Hölder in 1901; for definitive
results see Suppes, Krantz, Luce, and Tversky 1989.

  :  375
5 At first blush at least there are numerical scales which do not fit in any of
these four categories. For example, an earthquake ranked 6 in the Richter scale is
10 times stronger than one ranked 5, which in turn is 10 times stronger than one
ranked 4. But this is a matter of presentation: the magnitude on the Richer scale is
the logarithm (to base 10) of a magnitude on an underlying ratio scale, namely the
logarithm of the combined horizontal amplitude of the largest displacement from
zero on a seismometer output. Thanks to Richard Otte for raising this point.
6 The Mohs scale of hardness had its difficulties. Mohs, in developing his scale
for the hardness of minerals, ranked minerals relative to each other by the relation
‘‘scratches’’. He selected ten minerals to represent particular points on the scale
and assigned them numerals from 1 to 10. The operation of scratching does not
give more than an ordinal significance to these numbers. Mohs had assumed that
his relation of ‘‘scratches’’ was transitive and asymmetrical and that the equivalence
indicated by not being able to scratch one another was transitive and symmetric.
This assumption ran into a problem when it was found that some minerals could
not scratch each other, yet differed with respect to scratching a third mineral.
But this does not refute that an ordinal scale could be defined in terms of the
classes of minerals scratchable and minerals not scratchable. That is, minerals could
be classified as equally hard if they can scratch all the same minerals, and can be
scratched by all the same minerals. Yet here too there is an empirical assumption,
if it is asserted that this leaves no ambiguity in the ordering. Later attempts to
arrive at a measurement scale for hardness by other operations such as microscopic
measurement of the depth of a scratch made by a diamond under constant pressure
or the amount of work done in grinding away a certain weight or volume of
material, allowed for more quantitative comparisons. Note well: the real issue here
was to find ways to arrive at a stable ordering induced by outcomes of physical
procedures applied to the minerals.
7 Suppose that x, y, z are three Fahrenheit temperatures, and x , y , z the
corresponding three Celsius temperatures. If y = Nx, it certainly does not follow

that y = Nx . But the ratio of |y-x| to |z-x| equals the ratio of |y -x | to |z -x |. So
the usual measurement of temperature is on an interval scale, not a ratio scale.
8 For discussion see for example Darrigoll 2003, section 1.4 and 3.1.6.
9 On Stevens’s proposal, measurement scales are classified according to the
mathematical group structure of the group of admissible transformations. For

variants and later developments, see Ellis 1966: 58–67, Narens and Luce 1986.
10 See further Falmagne and Narens 1983.
11
This point is most salient in literature on the foundations of quantum
mechanics, originally emphasized in Birkhoff, and von Neumann 1936. They also
suggested the innovation of reducing modulo differences of measure zero.
12 Two qualifications: mechanics will provide different families of models for
376  :  
systems with different numbers of degrees of freedom, and the relevant phase
spaces are thus not common to all models. Models can be grouped by common
state spaces (see Lloyd 1994: 19–20 and 35). Secondly, the conception of time,
space, or space-time as a logical space for which I argued in my 1970 is of course
controversial, being at odds with substantivalist theories in this area.
13 This becomes especially pertinent when the ascription of an ‘unsharp’ value
in quantum theory does not arise simply because of ignorance of the ‘real’ sharp
value (see Appendix pp. 312–314).
14
A well-known paper by James Bogen and James Woodward (1988)—but in
a way that results in a terminology more confusing than the simplifications they
criticize. It is not grammatical, it seems to me, to say that data are observed—a
datum is the content of the outcome of a measurement. Nor does it seem to me to
properly respect the history of the word ‘‘phenomenon’’ to say that phenomena are
typically not observable. Admittedly, regimenting language always has its leeway;
here I will concentrate on the word ‘‘data’’, and in a later chapter address usage of
‘‘phenomena’’ and ‘‘appearances’’.
15
The analysis may be made by the experimenter, or already automatically by
the instrument which may issue e.g. a graph rather than a set of points. Thanks to
Todd Harris and Paul Teller for pointing this out.
16 I am taking for granted here agreement that a frequency interpretation
of probability—in contrast even to the modal frequency interpretation—is not

feasible; see e.g. my 1980: ch. 6 section. 4 for critique. See further my 1979 for a
discussion, in connection with Reichenbach’s view, of how geometric probability
cannot be regarded as a simple extrapolation of relative frequency in the long run.
17 Latour 1999 provides an illuminating narrative for this process, from digging
up the soil samples to disseminating the smooth graphs and annotated maps, in a
study of the changing boundaries between savannah and forest.
18 The semantic view of theories to which this statement refers is easily
misconstrued if conveyed by such a short, slogan-like statement. A model is a

mathematical structure; but that is like saying that a sentence is a black mark on a
page—true, but not a sufficient guide to what is meant when we say that a theory
provides a set of models for representation of the phenomena. I spelled this out to
some extent in my 1991, sections I, 2–4; see further e.g. my 2006a.
19 Here, as almost everywhere in this book, I concentrate on the ideal case of
exactness. For detailed considerations of how this can miscarry, and how we are
to relate to the inexact representations we can have in practice, see Teller 2008, as
well as his earlier 2001a.
20
This does not ignore the possibility that some measurement choices in
PRC are mutually incompatible; in fact, that would make it easy for the family
  :  377
of conditional probability functions P(. . .|C), C in PRC, to be embedded in a

classical probability function.
21
The design of the experimental or measurement set-up, in contrast to its
physical construction, is however guided by theory: in that sense, theoretical
models and data models constrain each other’s form. Thanks to Isabelle Peschard
for insisting on this point.
22 For a more extensive discussion and references to the literature, see my 1974
and 1982b.
23
An algebra is a mathematical structure consisting of a set of elements and a
collection of operators on those elements—though the term is variously defined
in different contexts in mathematics, so as to narrow the meaning (e.g. an algebra
is a vector space with a bilinear multiplication operation).
24 As for representation in general, so for measurement: it would be useless,
perhaps fatuous, to try for a definition. But in this case too, we can come to an
understanding of the subject by eliciting its general features and placing them in
context.
25
Weyl 1953: 8–9; quoted and discussed in Ryckman 2005: 134.
26 From section I of ‘‘The Relativity of Space’’ in Poincaré 1897.
27 Important to keep in mind that ‘‘invariant’’ is not an absolute term, that it is
context-sensitive: invariant under what family of transformations? But the context

may specify that, as it does in these examples.
28 At the risk of boredom, let’s remind ourselves that the discussion of meas-
urement interactions in physics concern the physical correlates of measurement,

and that the type of physical interaction there so characterized may well have,
according to the theory, instances far beyond what human agents can construct,
including e.g. interactions at the Planck level or in the stratosphere. The term
‘‘measurement’’ in that context refers solely to the general physical constraints
required for an interaction to be of the sort that can be the physical correlate of a
measurement properly so-called.
29
Cf. e.g. Dretske 1981 for an attempt to explicate information content in this
way, and its severe critique by e.g. Timpson 2004: ch. 1; Loewer 1982; McGinn
1997.
30 When two terms with the same referent cannot be substituted salva veritate,
the context is ‘referentially opaque’, a mark of intensionality, non-extensionality .

31 cf. Kroes 2000: 29; Kroes 2003: 74–8; Peschard, 2007b.
32 The point returns with a vengeance in many reactions (not only Bohr’s!) to
the Einstein–Podolsky–Rosen critique of quantum mechanics. Lately this point

has taken pride of place again in approaches to quantum theory such as Rovelli’s
378  :  
relational quantum mechanics as well as information-theoretic approaches (Fuchs

2002, Bub 2005, Timpson 2004).
33
Thanks for this point to Angela-Adeline Mendelovici.
34 I am paraphrasing John Worrall here.
35 For exploration of this as an analogy to be drawn on in the philosophy of
perception, see Peschard, 2007a.
Part III: Structure and Perspective

1 This is the sense in which Charles Morris introduced the term ‘‘pragmatics’’
in the International Encyclopedia of Unified Science.
2
To call the old terms ‘‘already coordinated’’ is only a metaphor if they were
not previously at some definite historical moment explicitly subject to that process.
Further below I will take up Reichenbach’s problem of coordination in a more
abstract form as well.
3 While perhaps every categorematic term we have today was new at some
previous stage of our history, it does not follow that there was or could have been
a moment when meaningful language was created from nothing. With scientific
terms being our concern alone, we do not need to speculate on how language came
into the world; we must only become clear on how new terms can find their use
when explicitly introduced for practical and theoretical reasons. For the way David
Lewis responded quite differently to the basic problems in philosophy of science
(though mainly as posed by Carnap and Putnam) see my detailed examination of
his take on this issue in my 1997b.
4 Much of the problématique encountered here is found, mutatis mutandis, in
the ‘Ramsey sentence’ literature. I will not take that up, but see my 1997b, where
I examine David Lewis’ attempt to define theoretical terms by adapting Ramsey’s
move.
5 My phrase ‘‘reach for the unconditioned’’ is of course meant to echo Kant’s
diagnosis of what he calls the transcendental illusions, the illusions of reason: such
illusions are made inevitable by reason’s tendency to seek the unconditioned, that
is, to carry a series of ideas or questions or arguments to their ‘logical conclusion’
even when their completion would lie clearly beyond the bounds of sense.
8. From the Bildtheorie of Science to Paradox

1 For the name ‘‘Bildtheorie’’ and specifically the development of this view by
Boltzmann see especially Blackmore1999; Henk de Regt 1999 and 2005; Visser
1999; section 3.4 of Stöltzner 2003. See also D’Agostino 2004 and Leroux, 2001.
The entire period with all the relevant dramatis personae is surveyed in chapter V
  BILDTHEORIE     379
of Cassirer 1950. Today the term seems to be used mainly for the study of visual
media, both in aesthetics and in communication theory, but in philosophy also for
Wittgenstein’s ‘‘picture theory’’ of meaning in the Tractatus.
2 Nyhof 1988 argues convincingly against the claim that Boltzmann later
converted to, or at any point held, a purely positivistic or instrumentalist conception

of science, as was sometimes claimed. (This article is otherwise permeated with a
philosophy of science that, as I see it, tends to color his account of the period.)
I would add here also that for constructive empiricism, which is a view of what
science is and not a view about what exists, the question of whether atoms are real
is a bit of a red herring, just as for Boltzmann’s philosophy of science.
3 Translations of the entire paper can be found in Planck 1994 and in Toulmin
1970.
4 This sort of conflation of the philosophical issue of scientific realism with
the question of the reality of unobservable entities is a recurrent problem in the

literature, and accounts for some of the puzzlement philosophers on both sides of
the debate tend to express concerning the others’ views.
5
Early in the nineteenth century, the term still meant something like ‘‘look of
the world’’, and not ‘‘worldview’’. But by the time Dilthey introduces its most
famous use, he can only make a rather subtle distinction between Weltbild and
Weltanschauung. (Thanks to Anja Jauernig for help with the history and etymology.)
In Planck’s lecture it is not so clear whether he is attempting to depict the world
as, according to him, it is depicted in physical theory or rather what he takes to be
the Weltanschauung proper to the physical sciences.
6
Though Boltzmann was undoubtedly indebted to and inspired by Mach’s
more radical view, he argues convincingly that his own ‘picturing’ view of science
is more informative than Mach’s phenomenalism—see e.g., his 1902.
7 For Boltzmann’s attribution of this view to Maxwell, see especially Boltzmann
1974: 217–19; for his references to Hertz, 214, 225.

8 For an early appreciation of the role of higher-order similarities in structural
representation, see Sellars 1965: 180–2.

9
‘‘Wir machen uns Scheinbilder der äußeren Gegenstände, und zwar machen
wir sie von einer solchen Art, dass die denknotwendigen Folgen der Bilder
stets wieder die Bilder seien von den naturnotwendigen Folgen der abgebildeten
Gegenstände’’ Hertz [1894/1956]: 1–2; see also p. 2–3 and 177, and discussion by
Leroux 2001.
10 The diagram will, I hope, make clear what is meant, but see further my
1989: 258–61 concerning symmetry.

380  :  
11 Schrödinger 1929: 16; cited D’Agostino 2004: 381; see also Schrödinger
1928.
12
Schrödinger 1951: 40; cited Bitbol 1996: 29; and see further Bitbol 1996,
passim, and Schrödinger 1953. The first part of this passage does not attribute
continuous spatio-temporal trajectories for particles; it refers solely to the fact that
the wave function of a system is defined with spatial coordinates (though it is not
a wave in 3-space).
13 The reference is to Aristotle, Meteorology I, 7. In the next section (proposition
205), Descartes submits that under these conditions we can have ‘‘moral certainty’’
(‘‘a certainty sufficient for the conduct of life’’). This could certainly be cited
as presaging (an admirably modest version of) the Rule of Inference to the Best
Explanation.
14 In medieval paintings of the Annunciation we sometimes see an attempt to
assimilate the action of the Holy Spirit to optical or mechanical interaction; see for
example Steinberg and Edgerton 1987 on Filippo Lippi’s Annunciation in London.
15
For textual support see Blackmore 1999. Blackmore, a prominent Boltzmann
scholar, is very unsympathetic to the anti-realist leanings and makes a point of
showing that Boltzmann is careful to hedge his bets on this account.
16 Here I am much indebted to Hyder and Lübbig 2000 and Hyder 2003.
17 Heisenberg 1945, p. 36; cited in translation in Cassirer, op. cit., p. 117.
There is a clear echo here of Bohr’s repeated insistence that the wave function is
no more than a summary of what will be observed in measurement arrangements
which themselves are described in our common language in use before the advent
of atomic physics.
18 Hertz 1962: 20–1.
19 When Poincaré later says that Maxwell had shown that there must exist
mechanical models of electromagnetism, he presumably thought that this sub-

sumption under the generalized Lagrangian mechanics was successful. This already
derives its sense at best from several successive weakenings of what can count as a
mechanical model.
20
Note however the vagaries of linguistic change in science. Stein (1989:
57) makes a good case for holding that the retention of the word ‘‘atom’’ and
discarding of ‘‘ether’’ was historically arbitrary: ‘‘our own physics teaches us that
there is nothing that has all the properties posited by nineteenth-century physicists
for the ether or for atoms; but that, on the other hand, in both instances rather
important parts of the nineteenth century theories are correct.’’
21
Hertz 1956: 1–3; see also p. 177.
22
Poincaré 1952: 161; in view of Poincaré’s literate and literary style we
may well think of historical precedent for this sentiment in e.g. Condillac’s
  BILDTHEORIE     381
‘‘[C]omment nous assurer que les principes que nous imaginerions, sont ceux-
mêmes de la nature? Et sur quel fondement voudrions–nous qu’elle ne sache faire
les choses qu’elle nous cache, que de la manière qu’elle fait celles qu’elle nous
découvre ? Il n’y a point d’analogie qui puisse nous faire deviner ses secrets ; et,
vraisemblablement, si elle nous les révéloit elle-même, nous verrions un monde
tout différent de ce que nous voyons. [. . .] C’est que l’imagination voit tout ce qu’il
lui plaît, et rien de plus’’ (Condillac 1749/1798/1949: 197–8; cited and discussed in
Vuillemin 2005).
23
My translation; ‘‘Nulle théorie ne semblait plus solide que celle de Fres-
nel . . . Cependant, on lui préfère maintenant celle de Maxwell. Cela veut-il dire
que l’œuvre de Fresnel a été vaine? Non, car le but de Fresnel n’était pas de
savoir s’il y a réellement un éther, s’il est ou non formé d’atomes, si ces atomes
se meuvent réellement dans tel ou tel sens ; c’était de prévoir les phénomènes
optiques’’ Poincaré 1968: 173.
24 This view, expressed by John Worrall, was named ‘‘epistemic structuralism’’
by James Ladyman.
25
There is an often mentioned bit of support for such views, already alluded
to by Maxwell’s remarks about fruitful analogies across different areas of physics.
Important equations tend to recur in many places. They tend to identify recurrent
patterns in nature, found not once but many times. Often a new process is first
described in analogy to an old one, with the equations transposed or reinterpreted.
Heat diffusion and gas diffusion are analogous, the harmonic oscillator crops up
everywhere. . . . So the equations omit the distinguishing characteristics. As a
reason for structuralism, this observation does not show much at all. For whenever
we see the same equations describing two scientific subjects, we also see science
describing the differentiating characteristics. If we didn’t, we wouldn’t have an
example to give! The point that such equations describe at once many different
processes needs serious reflection, but it is not much of an argument for any
general view.
26
It is exactly the difference he outlines between the technical language of
a practical profession or craft, and scientific language: only the former can we
interpret by identifying the technical terms’ referents (Duhem 1962: 147–53).
27 As I see it, this applies to all the sciences, not just physics: biology and
the social sciences too involve construction of models that are essentially abstract,
hence mathematical, structures. I realize that this view rests on a view of what
mathematics is, how it encompasses all abstract representation used in science.
28
This insight, and the problem it raised, is found quite explicitly in Carnap’s
Aufbau (some five years before Weyl’s lecture) but with enormous resistance to
its import (as we’ll see below). The famous objection by Newman to Russell’s
382  :  
structuralism, at about the same time, trades on a specific instance of what Weyl
states here with complete generality.
29
We will need to re-examine this point carefully below, since assertions of
isomorphism are context-sensitive—a group isomorphism between two mathem-
atical objects certainly does not imply that they are identical, for example, for
although they are both groups, they may have other characteristics as well. Weyl’s
point is telling only when there are no mathematically describable features left out
of account in the isomorphic mapping.
30
In the novel it seems she has simply been told that her fingernails are
pink. She could have also done an experiment to see what a spectrometer would
show about the light reflected from her fingernails, so that she would know the
wavelength of the reflected light. Such variations do not affect the problem.
31 The crucial clue here, as in our discussion of the use of maps and models in
Part I, I see in the word ‘‘this’’ in Mary’s ‘‘this is pink’’. For a similar line, with
respect to the philosophy of mind issues for which the Mary example was devised,
see Ismael 1999.
9. The Longest Journey: Bertrand Russell

1 In this section I am drawing on the more extensive account by Torretti 1984:
52, 63–4, 74, 149–52.
2
The result was not perfectly satisfactory. Poincaré’s models of the hyperbolic
plane (see note below) are considerably more user-friendly. Beltrami himself
asserted that no similar interpretation is possible for hyperbolic 3-space in Euclidean
3-space, presumably because he had provided a Euclidean surface extended in three
dimensions to model the hyperbolic plane. The model also has singularities, and
Hilbert proved later that the hyperbolic plane cannot be mapped isometrically
onto a Euclidean surface with no such singularities.
3
Russell, 1897: chs II and III. At the end of ch. II he writes ‘‘Spaces without
a space-constant, . . . spaces, that is, which are not homogeneous throughout, we
found logically unsound and impossible to know, and therefore to be condemned
à priori. The constructive proof of this thesis will form the argument of the
following chapter.’’
4 Poincaré reviewed the book in 1899 and Russell replied (see reference in the
next note). Russell’s reply (similar to Frege’s point of view, see below), is in effect
that the axioms of geometry are statements capable of being true or false, which
presupposes that their terms designate; what those designata are like determines
the truth value.
5 ‘‘La question de M. Poincaré me place dans la situation désavantageuse d’un
écolier à qui l’on demanderait d’épeler la lettre A, en lui défendant d’employer

  :   383
cette lettre dans sa réponse. Si cet écolier était mathématicien, il répondrait tout
bonnement : A est la lettre qui précède B; et si on lui demandait d’épeler B, il
dirait que c’est la lettre qui suit A. Mais s’il sait vraiment ce que c’est qu’épeler, il
renoncera à la tache, de désespoir.’’ (Russell 1899: 701)
6 Frege elaborated his similar response in a controversy with Hilbert during the
years 1899–1906. (See Coffa 1986; Torretti 1984: ch. 3, section 2.10.) The theory
of space must be non-vacuous and true. But the question of truth can’t arise unless
the primitive terms have an independent meaning which fixes their reference.
They certainly cannot get that meaning from the axioms or theorems, as Hilbert
asserted. For those axioms and theorems are incapable of having a truth value
unless their terms have referents. Pressed to explain how we identify those real
spatial relations, Frege also retreated to intuition or direct acquaintance: ‘‘I give the
name of axioms to propositions which are true, but which are not demonstrated,
because their knowledge proceeds from a source which is not logical, which we
may call space intuition’’ (in correspondence with Hilbert; quoted Torretti 1984:
235). This was of course just the sort of thing which Poincaré and Hilbert explicitly
professed to not understand.
7 Russell 1901; see specifically pp. 313–14; this passage was spoofed in Philip
Jourdain’s The Philosophy of Mr. B*rtr*nd R*ss*ll ( Jourdain 1918: 84–5).

8 As far as Russell’s view at this point goes for concrete entities, that includes
only sense data (and possibly the self ), but that point does not play much of a role
here. The objects postulated by physics are beyond the reach of experience; that is
the only relevant point here.
9
The importance of this article, its devastating import, and its relation to more
recent discussions of realism, was pointed out by Demopoulos and Friedman.
10 We can also easily put the point in terms of models and equations. Suppose
some equations have a model in which there are N distinct entities (where N
may of course be uncountably infinite). Choose a set of the same cardinality in
the world. Because same cardinality implies the existence of a correspondence, we
have an implicit transfer of the relations in the model to that chosen set. Therefore
the world satisfies those equations! Provided only the world’s size is large enough,
experimentation is superfluous. As David Lewis realized (and as we shall take up
below), Newman’s point becomes simply Putnam’s model-theoretic argument,
once we phrase it in terms of models and equations.
11 Or so it seems. In the next chapter we will see how David Lewis denied this
very assertion.
12
Since we are here speaking of a repair being carried out within the context
of Russell’s own epistemology, note that the only things with which we are
acquainted there are our sense data and those Universals (and possibly our selves).
384  :  
But we can presumably adapt the same moves to a more liberal view according to
which we are acquainted with all the observed phenomena as usually understood.
13
In the discussion of Putnam’s paradox, we will see this question returning in
a more general way. The answer is yes, the same problem does return even if one
retreats from Russell’s extremism with respect to non-logical terms.
10. Carnap’s Lost World and Putnam’s Paradox

1 Michael Friedman, addresses the same sections of the Logische Aufbau that I
shall examine here, in Friedman 1987: sections II and III. Friedman’s purpose is
not to examine the perils of structuralism but to explore Carnap’s epistemology
and his relations to the neo-Kantian tradition (continued in ch. 6 of his 1999).
Here he makes clear just why Carnap, in his pursuit of objectivity for knowledge,
cannot take the option of allowing the reference or extension of theoretical terms
in science to rest on ostension. That this option is excluded for Carnap I’ll take for
granted here.
2 I want to thank Daniel Rothschild for directing me to this and for much
helpful discussion. See his insightful paper ‘‘Structuralism and reference’’.

3 Here we see the pattern of reaction that Newman suggested to Russell:
to view a scientific theory as quantifying only over a selection of relations, the

‘‘important’’ ones in some sense. At the same time we see Carnap, like Russell,
indicating a selection that is somehow connected to the possibility of experience,
though with so little elaboration that it does not do much more than pay lip service
to what an empiricist should honor.
4 In his own commentary on this move, Michael Friedman calls it an
‘‘extraordinary suggestion’’ (1999: 103), and asks ‘‘But what can the ‘experi-
enceable, ‘natural’ relations’ be except precisely those relations somehow available
for ostension?’’ (ibid.).
5
I’ll quote here the argument as it was originally presented to the APA in 1976
(in which that core is not isolated): ‘‘So let T1 be an ideal theory, by our lights.
Lifting restrictions to our actual all-too-finite powers, we can imagine T1 to have
every property except objective truth—which is left open—that we like. E. g. T1 can
be imagined complete, consistent, to predict correctly all observation sentences (as
far as we can tell), to meet whatever ‘operational constraints’ there are . . . to be
‘beautiful’, ‘simple’, ‘plausible’, etc. . . .
I imagine that   has (or can be broken into) infinitely many pieces.
I also assume T1 says there are infinitely many things (so in this respect T1 is
‘objectively right’ about  ). Now T1 is consistent . . . and has (only)
infinite models. So by the completeness theorem . . ., T1 has a model of every
infinite cardinality. Pick a model M of the same cardinality as THE WORLD.
   385
Map the individuals of M one-to-one into the pieces of  , and use the
mapping to define relations of M directly in  . The result is a satisfaction
relation SAT—a ‘correspondence’ between the terms of [the language] L and
sets of pieces of  —such that theory T1 comes out true—true of 
—provided we just interpret ‘true’ as TRUE(SAT). So whatever becomes
of the claim that even the ideal theory T1 might really be false?’’ (Putnam 1976:
485; Putnam 1978: 125–6)
6 This theorem was initially debated in the context of philosophy of math-
ematics, in connection with Skolem’s relativism. One spoke there also of the
Loewenheim–Skolem paradox.
7 Thanks to Jenann Ismael for striking examples of this sort.
8 It might be objected that the map, like Paris itself, has much additional
structure, while a model has only the structure it displays. That is not a good
disanalogy, for the model also displays a selection of the structure it has. Suppose
the model is D, F with D its domain and F a family of sets and relations on D.
Then F is of course a selection from the set of sets of n-tuples, for various numbers
n, of members of D. All those other sets of sets there are selectively excluded
from F.
9 This is just the sort of example that must have led Leibniz to agree that
abstract entities can violate his Identity of Indiscernibles principle.
11. An Empiricist Structuralism

1 I begin with the title of E. J Dijksterhuis’s 1969 magisterial work that stops
with Newton; recall though that the ‘‘world-picture’’ phrase was current only in
that later tradition.
2 See my 2006b for a discussion of the recent Structural Realisms, due to John
Worrall and James Ladyman, and a defense of an empiricist structuralism with

reference to the criteria that Worrall laid down for a successful realism.
3 That ability is according to Dieks and De Regt 2004 the hallmark of
understanding a scientific theory.

4 For exposition and critique see my 2006b. See further Ladyman 1998a, 1998b;
Da Costa and French 1990; Bueno 1999a, 1999b.

5 Newman’s argument against Russell’s structuralism, which we saw also at the
heart of Carnap’s difficulties and of Putnam’s Paradox, is explicitly drawn on in the

critique of the semantic view in general, and constructive empiricism in particular,
in Demopoulos 2003.
6 The semantic view of theories runs into severe difficulties if these notions are
construed either naively, in a metaphysical way, or too closely on the pattern of

the earlier syntactic view. Constructive empiricism, which was formulated in the
386  :  
context of the semantic view of theories, shares this vulnerability (cf. Demopoulos
2003). I began to face these difficulties in my 1997a, 1997b, 2001, and mean to
complete their dissolution here.
7 Easy to see, I think, the echo of this passage in Reichenbach’s student
Putnam’s ‘‘This simply states in mathematical language the intuitive fact that to
single out a correspondence between two domains one needs some independent
access to both domains’’ which I quoted above.
8 Mea culpa: in The Scientific Image constructive empiricism was presented in
the framework of the semantic view of theories, but seemingly in the shape
of the above ‘‘offhand’’ realist response. See for instance ch. 3 section 9, p. 64
where I define empirical adequacy using unquestioningly the idea that concrete
observable entities (the appearances or phenomena) can be isomorphic to abstract
ones (substructures of models). Demopoulos 2003 comments on this passage. That
this ‘‘offhand’’ way of talking is most readily interpreted in metaphysical terms is
clear also in comments by such acute and careful readers as Stathis Psillos (2006,
remarking on correspondence and structures in re) and Michel Ghins (Ghins 1998:
328). Rather than try to excuse or explain my obliviousness to these issues at that
time, I tried to do better in my 1997a, 1997b, 2001, and I will try to do better here.
9 Admittedly, this conception is quite a natural one to find in the foundations
of physics, for physical systems in the intended domain of a theory are conceived of
as highly structured, and the entire discussion is targeted on mathematical models.
10 In his preface Whitehead says that he took the term from an earlier use:
‘‘The general name to be given to the subject has caused me much thought: that
finally adopted, Universal Algebra, has been used somewhat in this signification by
Sylvester in a paper, Lectures on the Principles of Universal Algebra, published in the
American Journal of Mathematics, vol. vi., 1884. This paper however, apart from the
suggestiveness of its title, deals explicitly only with matrices.’’ (Whitehead 1898)
11 This point is made and explored by Psillos 2006.
12 The motto of the British Order of the Garter. When Edward III danced
with the Countess of Salisbury her garter fell off. Edward said ‘‘Honi soit qui mal
y pense,’’ (‘‘shame on whoever thinks ill of it’’) and tied the garter around his own
leg.
13 It seems to me, though, that this line of thought is disappearing from the
scene: Stephen Leeds (1978, 1994, 1995, forthcoming ) shows that the scientific
realist does not need to rely on a correspondence theory of truth; lest we think
that metaphysical realists must all need it, there is David Lewis 2001.
14
I won’t do much to argue that here; if a reference is needed beyond the
above discussions of the correspondence theory of truth, my own favourite is
Quine’s ‘‘On what there is’’.
   387
15 See e.g. the papers by Leeds and Lewis cited above, or the enormous further
literature on the subject.
16
This is essentially the point raised by Mauricio Suarez which I took up in
Chapter I-1.
17 Giere 1988: chs. 3, 7, 8; Giere 1999: chs. 6, 7; Teller 2001, discussed in my
2001.
18 There is an echo here of the main point of virtue epistemology: for some
information held to count as knowledge, not only its content but the history of
its acquisition is crucial. The point is different but the re-direction of attention, to
how the item is arrived at, is the same.
19 It may be objected that a given model may not have the requisite complexity
to represent a particular phenomenon. But representation is selective: the most

we can say here is that a given model may not have the requisite complexity
to represent certain aspects of a particular phenomenon in a particular way.
Nevertheless, that does qualify my assertion; take it as thus amended.
20
This assumes of course that we have ways of describing the phenomena—we
already have a language that we live in, as I said. We must once again resist the
‘just born Robinson Crusoe self’ picture, in which we are confronted by a world
which is neither thus nor so, as yet indescribable by us, etc.
21 Perhaps Reichenbach means something like this when he says that reality
(i.e., the phenomenon that we are describing) is first defined by the coordination.
But it could only be ‘something like this’, since this way of putting it seems to
imply that what the models represent is something like the physical objects as described
rather than the physical objects. The ‘‘as’’ in that phrase is the traditional ‘‘qua’’,
and one is meant to both identify and distinguish the referents of ‘‘the X, which
is F’’ and ‘‘the X qua F’’—not a resource the later empiricism can draw on, to
put it mildly. I want to thank Anja Jauernig for help in my effort to understand
Reichenbach’s 1920 attempt to forge a modified neo-Kantian view of the matter.
22 The simplifying assumptions are strong of course: consider just two bodies,
the mass of one of them much less than the other’s, and with the distance between
them remaining limited. Newton’s laws admit other solutions even for the two-
body system and also quickly allowed corrections to Kepler’s laws, taking into
account that the smaller mass is not negligible even for a planet in relation to the
sun.
23 That is not to deny that there is a pertinent 3-place relation that can be
described, assigned a set as its extension, and so forth: namely a relation between
the scientist, the bacteria colony, and the data model this scientist constructed.
The important point is that an indexical sentence is not meaning-equivalent to an
388  :  
non-indexical one, except within or relative to a context of use, and it can have
uses which cannot be served by any non-indexical sentence.
24
As Psillos 2006 also emphasized.
25 A response entirely at odds with empiricist scruples would be to escape
into metaphysics, by postulating some nexus to link representations to what they

are representations of, independent of our fragile practices, hoping to make our
tenuous grasp on reality secure.
26 As analogy consider the two questions ‘‘Is snow white?’’ and ‘‘Is our sentence
‘Snow is white’ true?’’ They are not the same question. After all ‘‘Snow’’ could
have been our word for grass, water, or wine. But given that this sentence is our
sentence, as things actually are, the questions amount to the same thing for us. That
is why for us, who speak this language, ‘‘The sentence ‘Snow is white’ is true if
and only if snow is white’’ is a pragmatic tautology.
27 It may be easy to see the similarity to Descartes’s ‘‘I think’’, for which he
demonstrates not the necessity or truth but the indubitability for the speaker of
the ‘‘I’’, so that the sentence fits well into the category of pragmatic tautologies. It
seems to me that the same holds for Putnam’s ‘‘I am not a brain in a vat’’ in view
of Putnam’s own form of refutation of this sort of scepticism.
28 Now we can see why the offhand realist responds sounds plausible at
first, because it does get something right—namely that in the end there is no
problem, precisely because we can (a) correctly describe relevant parts of nature
and mathematical objects, and (b) say how they are related to each other. But this
plausibility hides the mistake of replacing ‘‘we can’’ with a relation independent of
the user (the ‘‘we’’) and ignores the selectivity exercised by the user for the user’s
specific purpose.
Part IV: Appearance and Reality

1
For a more detailed discussion of the parallels between Aristotle’s Poetics and
Physics see my 2000b.
2 Physics I, 184a15–21, 194b19–20; the discussion of a poorly constructed
theory that I mention next is found at 198b10–35.

3 Metaphysics N3, 1090b, tr.: Ross I want to thank Fran O’Rourke for pointing
me to this passage.
4 I do not mean to underrate the way in which mathematical modeling too
trades on resemblance. But the extent to which mimesis is involved at all diminishes
continuously as we move our view successively from table-top models to textbook
illustrations, from there to mathematical modeling, and eventually to the geometric
spaces that appear in relativistic cosmology and quantum field theory.
 .     389
12. Appearance vs. Reality in the Sciences

1 Cf. Margaret Wilson 1984. I gratefully acknowledge my debt to her analysis.
2 For the seminal text for later discussions and the now standard terminology,
see John Locke, Essay Concerning Human Understanding, bk. 2, ch. 8, sections 9, 10.
3
A distinction also made by Galileo’s contemporary Gassendi, with similar
reference to ancient atomism.
4 The Assayer, section 48; p. 274 in Drake 1957.
5 Also not Galileo’s terminology. Galileo wrote ‘‘Philosophy is written in this
grand book, the universe, which stands continually open to our gaze. But the book
cannot be understood unless one first learns to comprehend the language and read
the letters in which it is composed. It is written in the language of mathematics,
and its characters are triangles, circles and others geometric figures without which
it is humanly impossible to understand a single word of it; without these, one
wanders about in a dark labyrinth.’’ Il Saggiatore (1623), in Drake 1957: 237–8.
Similarly, ‘‘The book of philosophy is that which stands perpetually open before
our eyes, but because it is written in characters different from those of our alphabet
it cannot be read by every body; and the characters of this book are triangles,
squares, circles, spheres, cones, pyramids and other mathematical figures fittest for
this sort of reading.’’ Lettera a Fortunio Liceti, gennaio 1641 in Crombie 1994: vol.
i, 585.
6 This quaint conviction beset much of early empiricism as well. Locke writes
that the mind ‘‘hath no other immediate object but its own ideas’’ (An Essay
Concerning Human Understanding, Bk IV, i, I); Berkeley that ‘‘the objects of human
knowledge are either ideas actually imprinted on the senses, or else such as are
perceived by attending to the passions and operations of the mind’’ (Principles of
Human Knowledge, Part I, I); Hume that ‘‘[a]ll the perceptions of the human mind
resolve themselves into impressions and ideas’’ (Treatise of Human Nature, I, i, I).
7 Meditation III, section. 19, tr. John Veitch (Descartes 1959).
8 See for example ‘‘Huygens and Leibniz on Universal Attraction’’, pp. 115–38
in Koyré 1965.
9 For the former cf. ch. 2 of Vargish and Mook 1999; for the latter, Cushing
1994.
10
That logically necessary connection may not be finitary, may in fact be
inaccessible to any finite mind—as Leibniz made explicit (therefore not within
even the potential reach of the physical sciences)—but be graspable only by
the divine mind. There are many ambiguities in these developments. Sometimes
Descartes and Leibniz do sound as if there can be only one world, of logical
necessity. The popular version would go like this: from the concept of God it
390  :  
follows that he would not create a world at all, if among all the conceivable ones
there was not a best one (Leibniz’s Theodicy) or one uniquely transparent to the
human mind (Descartes’s posthumous The World). But at other points the claim is
that although the regularities derive with logical necessity from the laws of nature,
those laws characterize a selection from the realm of conceivable possible worlds
which has no further rationale, at least within the context of even these ‘‘theories
of everything’’. Notice that we have here, in effect, the first ‘‘supervenience
without reduction’’ claim, for reduction would require finitary reasoning but the
demonstrative link is claimed to be non-finitary. I will return to questions of
supervenience below.
11 ‘‘Some seekers after the theory of Everything would seem to be hoping
that the uniqueness and completeness of some Particular mathematical Theory

will make it the only logically consistent description of the world . . .’’ (Barrow
1991: 202). In philosophy too there are arguments currently that the laws of
nature are necessary after all (although necessity need not be thus related to logical
consistency): cf. Bird 2007.
12
Boyle 1772: vol. v, 162.
13 See Stöltzner 1999.
14 Sociologists of science have (in)famously explored social and political factors
in post-World War One Europe to explain the readiness of prominent physicists

to embrace indeterminism. Looking back after four decades of discussion of Bell’s
Inequalities and surrounding issues, we must certainly also see that readiness as
(perhaps fortuitously) prescient.
15
Cf. my 1982a, and references therein. Under certain conditions this criterion
actually demands determinism, as I show there. But from the example of Bohmian
mechanics, we can see as well that satisfaction of this criterion is not logically
implied by determinism; see further notes below. For more on Reichenbach’s
conception see e.g. Graßhoff, Portmann, and Wüthrich, 2005.
16 Bohm’s mechanics is deterministic but also violates the sorts of locality
constraints required by Reichenbach’s criterion; surprisingly it presents a picture

of ‘determinism without causality’ so to speak. For early discussion of Bohm’s
alternative, see Grünbaum 1957: 715–16n., and for an early sympathetic survey of
this alternative and its links to other work then, see Freistadt 1957. For its more
recent and quite spectacular revival in philosophy of physics see Cushing 1994.
17 This point, like so much else in the interpretation of quantum mechanics is
tendentious and controversial. I do not want to enter into that debate just now.
The controversies illustrate, in any case, the denouement of the historical pattern
that I discern. The new and revolutionary success in science came with a ostensible
rejection of a prevalent completeness criterion for science, but in the aftermath
 .     391
many strove (and strive!) valiantly to reinstate the criterion and to show that its
rejection was not logically required.
18
There is some disparity between what Bohr and his colleagues, students, and
followers originally professed and what almost the entire physics community came
to subscribe to under the name of ‘‘Copenhagen interpretation’’; see Howard 2004.
We could say the same for any revolutionary stance, whether social, religious,
or intellectual. What I will insist on is that with all its ambiguity intact, the
Copenhagen view was intellectually and scientifically revolutionary.
19
See further e.g. p. 80. After maintaining this view strongly throughout most
of the book, Leplin includes a section in which he raises the possibility that recent
and current theories in physics will not be in accord.
20 There is now a quite extensive literature on how this point about wavelength
of reflected light is a far cry from the nuanced and delicate account of human
visual experience. The critique I am in the course of offering here is independent
of that; it targets not the degree of success in such explanations of specific
appearances, but the conception of methodology that involves those putative
criteria of completeness.
21 Attempts to provide such accounts which could perhaps provide such
explanations even today include work by Stephen L. Adler and his colleagues on
Generalized Quantum Dynamics, as well as earlier work by e.g. Nelson 1967.
Such accounts, if successful, can satisfy the Appearance from Reality Criterion no
less than deterministic theories.
22
This is too simple an idea of course, but is not in essence too far from the
work by Adler and Nelson mentioned above. Besides these I am thinking here of
the Ghirardi, Rimini, and Weber (GRW) version of quantum mechanics, which
introduces a ‘Lucretian swerve’ into Schrödinger’s deterministic equation.
23 Wigner’s 1961 appeal to consciousness in quantum mechanics, to ‘explain’
collapse of the wave-packet falls under this heading, as do the idealist and
(quasi-)instrumentalist accounts which Grünbaum depicted as prevalent forms of
easy anti-realism among scientists (Grünbaum 1957: 717–19).
24
I have only slowly come to see the importance of marking such a distinction.
In The Scientific Image I did not make this distinction either carefully or clearly.
The chapter on saving the phenomena introduces ‘‘appearances’’ to denote what
Newton called ‘‘apparent motions’’, identifying them as ‘‘relational structures
defined by measuring relative distances, time intervals, and angles of separation’’(p.
45). I would now refer to those relational structures as data models. Data models
are the summarizing refinement of the contents of a battery of measurements,
typically, so this is not far from my present usage. But in the passages that follow
there, the reference seems from time to time to be just to observable entities, i.e.
392  :  
phenomena rather than appearances in my current stricter usage. Thus Paul Teller
rightly writes ‘‘First is the idea of. . .phenomena (which means, the observable
process and structures). . .’’ (SI 3). ‘‘I take van Fraassen to use ‘phenomena’
and ‘appearances’ interchangeably. (SI 45, 64) I will understand phenomena and
appearances as that which we can observe, or that of which we can become
perceptually aware, without the use of instruments’’ (Teller 2001: 125–6).
25 Kant, Critique of Pure Reason, first para. of Transcendental Dialectic, Introduction
section I. The quote is from Meiklejohn’s translation (Kant 1850: 209), which had
many new editions well into The twentieth century. The translation by Francis
Haywood (Kant 1848: 234) is practically the same. The German original is ‘‘Noch
Weniger dürlen Erscheinung and Schein für einerlei gehalten Werden’’.
26 There is lots more to be said about the terms of course. The term
phenomenon is often enough applied to microphysical processes that are not

observable—against the explicit stipulations by Bohr, which Wheeler formulated
as ‘‘No phenomenon is a phenomenon unless it is an observed phenomenon’’. We
should also note that like many other nouns ‘‘phenomenon’’ has both a generic
and a specific use—a specific effect produced in a laboratory at a particular time is
a phenomenon, but so are oxidation, ebbing, planetary motion, and so forth. In
the generic use, as I understand the term, it refers to classes of observable entities.
See further Hacking 2006.
27 This makes the point I want to make simple. The entire discussion of
modern science that follows here could be more informatively developed around
e.g. Newton’s system of the world—but I think we would lose the forest for the
trees.
28 Amazingly, this common sense realism is often lacking among the most
soi-disant ‘‘realist’’ philosophers. In their wish to defend a requirement to believe

in the reality of unobservable entities it has not been rare to see the tu quoque that
we all believe in such things as mountains although, as Quine famously claimed,
‘‘physical objects are epistemologically, to the gods of Homer [. . .]Both sorts of
entities enter our conception only as cultural posits. The myth of physical objects is
epistemologically superior to most in that it has proved more efficacious than other
myths as a device for working a manageable structure into the flux of experience.’’
(Quine 1951: 42). Surely this idea rests subliminally on the ‘homunculus’ view of
the subject, modeling our epistemic and doxastic activity on the case of a little
man in a booth reading only a ticker tape and inferring, postulating, speculating
on market activity?
29
On the face of it, my introducing the noun ‘‘appearance’’ may suggest that
I will postulate the existence of a special sort of entity, denoted by that noun. I
am certainly not doing that; this sort of game in analytic metaphysics is very far
      393
from my mind. That a penny looks elliptical to me if I hold it up in a certain way

does not imply that there exists besides the penny also an elliptical penny-look; it
implies the existence of me and of the penny, and that is all. Nor does it imply any
mistake on my part: I know very well that circular things can look elliptical if held
at a certain angle.
30 For a clear and informative description of Copernicus’s geometric ‘‘tran-
scription’’ see Barbour, 2001: 216.

31 Thus Regiomontanus published the Theoricae novae planetarum of his former
master Georg Peurbach. Another example is Pedro Nunes (1502–78), Tratado da

sphera com a theorica do sol e da lua (1537). An anonymous, Theorica Planetarum
was recently offered for sale on the internet with the advertisement ‘‘One of
evidently only three copies known of a richly illustrated astronomical handbook
filled with colored diagrams and movable volvelles that show the persistence of
the Ptolemaic world view as expressed in his Almagest, which remained a corner-
stone of astronomical thought even after the discoveries of Nicolaus Copernicus
(1475–1543). The present manuscript probably served as a demonstration text for
a teacher-astronomer.’’
32 We may reasonably suspect that the conviction, that this is so, helped to
inspire the ‘‘construction’’ programmes of Russell’s Our Knowledge of the External

World and Carnap’s Aufbau.
13. Rejecting the Appearance from Reality Criterion

1 Thus I agree with Nancy Cartwright (1983: 163–216) that the measurement
problem is an artefact of the formalism (though we do not have quite the same
diagnosis).
2 As with every viewpoint taken on this issue, it is possible in retrospect to
see it presaged in earlier ones. Some of what I will argue is certainly along the
line of how Michel Bitbol depicts Schrödinger’s responses to the measurement
problem: ‘‘But what was really needed was a full acceptance of the parallelism
between the time-development of the holistic wave-function (object + apparatus)
and the sequence of macroscopic events, rather than an new blend of the old idea
of a causal interaction which takes place between objects and apparatuses in order
to produce the events.’’ (Bitbol 1996: 123).
3
I emphasize this because arguably, the Appearance from Reality Criterion is
not violated in Bohmian mechanics, where position is the only genuine observable,
evolution of states is deterministic, and all measurements, of any sort, are recon-
structed as in the end just position measurements. Even if such a pattern should
overtake all of physics in the coming century (however unlikely that may be),
the methodological point would stand: the Criterion cannot be said to have been
394  :  
in force throughout the history of science, hence is not one to be discerned as

binding in scientific practice.
4
David Armstrong took a similar position in the 1960s; see Armstrong 1993.
5 There is an enormous literature on this subject, with different notions of
supervenience distinguished, most of them intelligible only in the context of

substantial metaphysical presuppositions. But there is a basic notion that relates
languages (e.g. the language of physics and the language of psychology), which can
be understood with more minimal background and without violating empiricist
scruples. I would describe it roughly and informally as follows: L supervenes on L
if for every set of sentences of L there is a set of sentences of L such that the truth
values in the former being different requires some difference in the truth values in
the latter. This implies an abstract form of translatability of L into L , but if we add
non-reducibility, that means that there is a translation only for the angels, not by
any humanly or mechanically feasible means. For a longer but still brief discussion
see my 2004.
6
I am drawing here on some notions from logic and foundations of mathem-
atics. Definable sets include here sets definable by recursion; weaker and stronger
notions of definability are available to differentiate various strengths in claims of
[non-]reducibility.
7 There are of course complaints from the side of philosophy of science about
philosophy of mind shenanigans; see for instance Glymour’s gleefully critical 1999
review of Kim 1998.
8
As a metaphysical postulate this supervenience claim presumably gives some
emotional comfort to the materialist/physicalist. While science is here admitted to
be incapable of showing this, the world is still conceived as how the physicalist
would like it to be.
9 Quoted by Mates 1986: 108–9. For this part, and for the following references,
I am thoroughly indebted to Anja Jauernig’s dissertation (Princeton 2003). Leibniz’s

On Freedom:
But in the case of contingent truths, even though the predicate is in the subject, this can
never be demonstrated of it, nor can the proposition ever be reduced to an equation or
identity. Instead, the analysis proceeds to infinity, God alone seeing—not, indeed, the end
of the analysis, since it has no end—but the connexion of terms or the inclusion of the
predicate in the subject, for he sees whatever is in the series (Parkinson 1970: 109).
The Monadology has:

There are two kinds of truths, truths of reasoning and truths of fact. Truths of reasoning are
necessary, and their opposite is impossible. Truths of fact are contingent, and their opposite
is possible. When a truth is necessary, the reason for it can be found by analysis, resolving
it into more simple ideas and truths until we reach the primitive. It is thus that speculative
theorems and rules of practice in mathematics are reduced by analysis to definitions, axioms,
      395
and postulates. There are, finally, simple ideas which cannot be defined, and there are
also axioms and postulates, or, in brief, primitive principles, which cannot be proved and
need no proof. And these are identical propositions whose opposites contain an explicit
contradiction. (Loemker 1975: 1050).
10 For simplicity of exposition I assume that A is discrete and not degenerate:

there is a single unit eigenvector for each eigenvalue, and the possible values
constitute a countable set of eigenvalues. Let me repeat another caution: I am using
a quite standard formulation here, which takes not only the notion of observable
but also that of physical state for granted. This is the form in which the issues will
be most familiar to most readers. It also seems to me that the same issues will arise
with more or less the same impact in a more interpretation-neutral formulation,
where it is allowed that local information (e.g. due to previous measurements) or
even subjective probabilities play a part in the assignment of states. See for example
Rovelli 1996 and my [forthcoming].
11 Notice that I use ‘‘refer’’ to set up the dilemma; the function-dependence in
meaning of these terms is left aside.

12 There is no way to rule out the possibility of inconsistency, of course: even
the mathematics within which quantum theory is formulated does not allow of a
consistency proof.
13
Evolution of an isolated system obeys Schrödinger’s equation: there is a
group of unitary operators { Ud : d in R } such that the pure states ψ(t) evolve
under the action of these operators: ψ(t+d) = Ud ψ(t).
14 We must distinguish here the actual historical development of the quantum
theory from interpretative additions and extrapolations of recent years. The ‘‘bare’’
theory is empirically empty without the Born Rule, but there have been attempts
to deduce the Born Rule from the basic theory supplemented with assumptions
involved in certain interpretations of the theory, such as the ‘‘many worlds’’
interpretation, GRW, or Bohmian mechanics. These sorts of assumptions were
either entirely absent in the development of quantum theory or roundly rejected
by the main physicists involved, so this does not affect the claim that this historical
episode in physics involved a rejection of the Appearance from Reality Criterion.
15 We may note here that there are certainly set-ups that are not neatly
dissectible into object measured and measuring apparatus. Rom Harré 2003 aptly
coined the terms ‘‘Bohrian artifacts’’ and ‘‘apparatus/world complexes’’ to designate
the peculiarities of such set-ups: ‘‘Let us call the apparatus/world complexes that
scientists, engineers, gardeners, and cooks bring into being Bohrian artifacts.
Properly manipulated they bring into existence phenomena that do not exist as
such in the wild [. . . .] In the famous Bohr–Einstein debate around the EPR
paradox, it is possible to see the outlines of Bohr’s account of experimental physics.
396  :  
While Einstein is insisting that for every distinct symbol in a theoretical discourse
there must be a corresponding state in the world . . . Bohr . . . is concerned with
the concrete apparatus and its relation to the world as part of the world. An
apparatus is not something transcendent to the world . . . . The apparatus and the
neighboring part of the world in which it is embedded constitute one thing.’’
(pp. 28, 29.) Despite this indissoluble entanglement, when the set-up is classified
as a measurement, then its outcome is classified as representing something in that
apparatus/world complex.
16
That is (another way to state) the Measurement Problem! This has seen many
offered ‘solutions’ and ‘dissolutions’ and much debate between their advocates, in
its now almost century long history. The literature on this subject is enormous.
For an older detailed treatment see my 1991; for a perspicuous recent discussion
that highlights the points that I will take up here, but in a general probabilistic
setting, see Wilce, forthcoming.
17 See the caution in a previous note about the reliance here on a standard
formulation of the subject matter. I meant to introduce all the basic notions needed
to understand the Measurement Problem, but at this point (and some others) I
included assertions that I am not justifying here. They are not egregious however;
the justification can be found at many places in the philosophical literature on
quantum mechanics, at almost any level of (relative) (non-) technicality.
18 The quick argument sketch is as follows: if the apparatus ends up in one of
the pure states |B,r> then it is not true that the Apparatus + Object ends up in a
superposition involving more than one such eigenstate of B–which contradicts the
conclusion about how the composite system has evolved. See for example Eugene
Wigner’s seminal 1963, specifically pp. 11–12.
19 See for instance Greenstein and Zajonc 1997.
20 Wigner, op. cit.; see specifically pp. 10–12, where Wigner discusses the
separation of an x-polarized beam of spin-1/2 atoms into two z-polarized beams

by a Stern-Gerlach apparatus, and asks what would result if one subsequently
merged the two beams, whether the reconstituted beam could be in a coherent,
x-polarized spin state.
21 No such point as this is incontrovertible: the literature contains so many
interpretations that almost all logically available niches are occupied and almost
every point will fail on some offered interpretation of quantum mechanics. But
there certainly are salient interpretations on which supervenience fails, and that
without locating the measurement outcomes in consciousness or other non-physical
realm. I am thinking here especially of the entire range of modal interpretations
of quantum theory. See my 2005a. What is not satisfactory, it seems to me, is to
appeal to decoherence, for that does not remove the problem in principle.
      397
22 As always, in practical contexts this is a matter of satisfactory approximations;

as is customary, the condition here stated relates to the ideal case.
23
This was the insight that has driven all ‘hidden variable’ and ‘modal’ inter-
pretations of quantum mechanics—though those also go beyond the minimum
point that is needed for my argument (namely by offering a separate identification
of the events in the measurement situation in terms of certain theoretical para-
meters, whether taken from quantum theory itself (as in modal interpretations) or
added as alien quantities absent from the theory altogether).
Index
abstract structures 120, 238, 240–1, 242–3 determinism 278–80

and phenomena 245–6, 248–50 necessity and 277–8
see also data models; theoretical models and phenomena 283–8, 317
abstraction 48–9, 251 primary/secondary qualities 271–6
accuracy 12, 15, 19, 139, 269, 309: see also and reality 270–6
approximation; distortion approximation 52–3, 56, 57, 111, 128, 163
admissible rescaling 160–1 Arago, Dominique 98
admissible transformations 55, 161–3, 174 Aristotle 7, 96, 200, 265–6, 283, 352 n. 20
air thermometers 125–6 primary/secondary qualities 271–3
Alberti, Leon Battista 62–4, 72 rainbows 102
Alhazen 43 artifacts 25, 94, 238, 245, 265, 395 n. 15:
alidades 61 see also models
altimetry 61–2 Aspect, Alain 170
Amontons, Guillaume 117, 128, 129 Assmus, A. 94–5
analogies 17, 25, 27, 101, 195, 380 n. 22, astrolabes 61–2
381 n. 25 astronomical clocks 132, 134
and disanalogies 219, 385 n. 8 astronomy 8, 168, 214, 218, 257, 289
maps 77–80 planetary motion 285–8
models and 310–11 sidereal days 133–4
and observables 184, 313 asymmetry 17–18, 189
analytic geometry 41, 46, 66–7, 353 n. 29 atomic clocks 134
anti-empiricism 307–8 atomic theory 112–13, 191–2, 198,
anti-realism 198 200–1, 206
apparatus 94–5, 112, 170–1, 362 n. 7, 393 atomism 280
n. 2, 395 n. 15 Aviation Model (AVN) (weather
measurement 143, 144, 298–9, 304 prediction) 77–8, 196–7
measurement outcomes 180, 183 Avogadro’s number 112–13
measurement theory 147–8, 150–3, 154
in quantum mechanics 300, 302–3, Bacon, Francis 75, 273, 374 n. 1
Stern–Gerlach 155, 179 Baird, David 165
thermometry 123, 125–6, 130 barometers 126
see also instrumentation Barrow, John D. 390 n. 11
Appearance from Reality Criterion 281–3, Beauvoir, Simone de 27
291–2 Bell, John 364 n. 25
and cognitive psychology 292–5 Bellarmini, Cardinal 281
and deducibility 296–7 Belnap, Nuel 370 n. 42
and philosophy of mind 292–5 Beltrami, Eugenio 215
and quantum mechanics 291–2, Benoist, Jocelyn 349 n. 28
297–300, 308 Berkeley, George 389 n. 6
and reducibility 292–5 Bildtheorie
appearances 8–9, 29 Boltzmann 1–2, 195–6, 197
Appearance from Reality controversy over 191–204
Criterion 281–3, 291–300, 308 Hertz 195–7
completeness criteria 276–83 Mach 197–8
400 
Bildtheorie (cont.) measurement procedures and 123, 124,

Maxwell 195 128, 139, 164, 172
Planck 192–5, 196 of psychological phenomena 292, 294–5
Birkhoff, G. 47 theory and 111, 139, 143–5, 164, 203,
Bitbol, Michel 370 n. 43, 380 n.12, 246, 261, 319
393 n. 2 Clausewitz doctrine of experimentation
Block, Ned 21–2, 38, 350 n. 12 112
Bogen, James 376 n. 14 Clausius, Rudolf 128
Bohm, David 390 n.15 and 16, 393 n. 3, Clifton, W. K. 223
395 n. 14 clocks 130–6
Bohr, Niels 76, 380 n. 17, 392 n. 26, cloud chambers 94–5, 113, 242
395 n. 15 cluster concepts 38
Boltzmann, Ludwig 192, 200, 279 cognitive psychology 292–5
and Bildtheorie 1–2, 195–6, 197 coherence 122–3, 136, 144, 145–6, 160,
scale models 355 n. 34 184, 303
Boon, Mieke 95, 362 n. 7, 361 n. 3 of experience 278–9
Born rule 297–300, 308 of measurement 153
Botticelli, Sandro 64 coherence conditions 130, 131–3, 134,
Boyd, Richard 199 136, 145, 152, 153
Boyle, Robert 278 coherence constraints 129, 131, 132, 134,
Boyle’s Law 128, 138, 242 136, 152–4
Bradley, F. H. 270 color vision 209, 372 n. 11
Brahe, Tycho 289 committal 38: see also non-committal
Brentano, Franz 349 n. 28 common cause 108, 136, 279–80
Bridgman, Percy W. 53 Condillac, Étienne Bonnot de 380 n. 22
Brunelleschi, Filippo 73, 183 congruence 13, 118, 130–1, 135, 216–17,
Bub, Jeffrey 378 n. 32 218
Buckingham, E. 53, 55 constructive empiricism 317, 385 n. 6,
Bueno, OtÆvio 385 n. 4 386 n. 8
contingent truth 296, 394 n. 9
Convention of the Meter 135
Campbell, Norman 158, 366 n. 11 conventions 19, 23, 43, 66, 82
Cantor, Geoffrey 362 n. 11 choice and 128–30, 134–6, 158, 160,
caricature 13–15 162
Carnap, Rudolf 225–9 and coordination 208
cartography 75–82 coordinating principles 117, 120
Cartwright, Nancy 310–11, 362 n. 5, coordination 241, 244–5
393 n. 1 historical context 116–21
Cassirer, Ernst 123, 379 n. 17 and measurement 136–7
Celsius scale 160, 161–2, 174 problem of 121–4
Chang, Hasok 368 n. 24 thermometers 125–30
Charles’s Law 128 time measurement 130–6
Charleton, Walter 99 coordinative definitions 121
chiaroscuro 36 coordinatization 136, 175, 178, 234,
classical mathematics 47 366 n. 10
classification 103, 113, 151, 155, 179, Copernicus, Nicolaus 8, 13, 271, 286–8
203–4, 211 cosmology 134, 277, 289
of experiences 106–8, Cratylus 19–20, 22, 35
language and 84, 206 Criterion for Physical Correlate of
and measurement outcomes 180–1, Measurement 182–3, 302, 312, 314
182–3, 299, 305 cross ratios 66, 72–3, 74–5
 401
Dalton, John 117, 127, 130 Euclidean geometry 61, 66–7, 213–14,
data models 166, 167–8, 172, 251–2, 215–16, 234, 285–6, 309
391 n. 24 exemplification 17
and phenomena 252–9 Exner, Franz Serafin 279
David, Jacques-Louis 29 experimentation: roles of 111–13
de Beauvoir, Simone, see Beauvoir, explicitly non-committal
Simone de representations 38–9, 40, 50, 313
Dear, Peter 97
deducibility 296–7 Faraday, Michael 95, 101
Dennett, Daniel 38 Feyerabend, Paul 73, 150, 359 n. 23
denotation 16, 348 n. 27 Feynman, Richard 150
depiction 16 De Finetti, Bruno 361 n. 42
Desargues, Gérard 72 Fine, Arthur 369 n. 35
Descartes, René 22, 30, 199–200, 269, Fourier, J. B. J. 53
271, 278, 280 frames of reference 66–70
analytic geometry 41, 46, 66–7 Frege, Gottlob 383 n. 6
frames of reference 66–70 French, Steven 345 n. 1, 349 n. 1, 368 n.
logical necessity 389 n. 10 25, 385 n. 4
primary/secondary qualities 273–5 Fresnel, Augustin 98
determinism 169, 292 Friedman, Michael 365 n. 3, 384 n. 1 and 4
and appearances/reality 278–80 Frigg, Roman 309
in mechanics 29–30 Fuchs, Christopher 378 n. 32
dilation 161–2 function 79, 181
dimensional analysis 53–5, 355 n. 40–3 of experimentation 111
distortion 12–15, 36–7, 38, 40, 183 indexicality and 182
Doisneau, Robert 20–21 instrumentation and 94–100, 157
dreams 24 mimetic 97
Dretske, Fred 15 use and 21–2, 23, 25, 30
Duhem, Pierre 203, 206–7 fuzzy observables 184, 312, 313–14,
Dürer, Albrecht 8, 65–6, 142 321 n. 7
fuzzy values 154, 376 n. 13
Eco, Umberto 116
Eddington, Arthur 71, 136
Edgerton, S. Y. 63, 356 n. 2, 380 n. 14, Galileo Galilei 41, 67, 278, 280, 281
351 n. 14 buoyancy experiment 50–1
Einstein, Albert 118, 134–5, 183, 395 n. 15 primary/secondary qualities 34, 271–3
principle of relativity 69–71 scaling 50–1
Einstein–Podolski–Rosen (EPR) thermometry 117, 123, 125
experiment 170–1, 299 n. 15, 315–16 Galison, P. 94–5
Elga, Adam 352 n. 15 gas law 127–8, 129
Elgin, Catherine Z. 17, 346 n. 9, 347 n. 14 Gassendi, Pierre 278, 280
embedding 29–30, 87, 168–72, 240, 247, geometric optics 42–5
252, 316 geometry
empirical adequacy 3, 136, 199, 246, 249, analytic 41, 46, 66–7, 353 n. 29
258, 317 Euclidean 61, 66–7, 213–14, 215–16,
empiricism 3, 304–6 234, 285–6, 309
empiricist structuralism 237–9 hyperbolic 213–14, 215
epistemology 222 non-Euclidean 213, 215–16
Escher, M. C. 39 projective 66, 72–3, 74–5,
essential indexical 3, 83, 88 215–16, 286
ether theory 100–1 Georgalis, Nicholas 348 n. 21, 348 n. 23
402 
Gerlach, Walther 155 kinematic 34–5, 39, 182

Stern-Gerlach apparatus, 179, 305–6 mathematical 39–49
Giere, Ronald 28, 168, 183, 309, 372 n. 11 visual 11, 34, 35, 39, 40, 182
Gilbert, William 97 images 22, 35, 101–10, 198, 202, 204, 221
Global Positioning Systems (GPS) 81–2 categories of 103–5
Glymour, Clark 394 n. 7 as copies 12–13, 18–21, 24, 97
Golding, William 21 distortion and 40
Gombrich, Ernst 12–13 hallucinations 101–5, 107–9
Goodman, Nelson 11, 19, 351 n. 14 instrumentation and 97, 105–6, 108–9,
on art 16, 21 110, 168
denotation 16, 348 n. 27 Loewenheim–Skolem–Tarski–Vaught
exemplification 17 theorem 230
GPS (Global Positioning Systems) 81–2 measurement and 75
Grünbaum, Adolf 166, 320 n. 5 mental 24
Guericke, Otto von, see von Guericke, mirror 214
Otto in models 242–3
and perspective 39
Hacking, Ian 108, 346 n. 4 (Part I) perspectival 73
hallucinations 101–5, 107–9 scientific 45, 47–8, 271, 275, 276
Hanson, N. R. 144 imaging: and scaling 56–7
Harré, Rom 395 n. 15 implication 315–16
Hauptsätze 23–4, 28, 55 incoherence 133, 160, 256, 260–1
Heidelberger, Michael 94, 95, 363 n. 15 indeterminism 279–80, 355 n. 37
Heisenberg, Werner 201 indexicality 59, 181–2, 239, 259–60, 261
Helmholtz, Hermann von 214, 229 and maps 77–8, 79–80, 257–8
hermeneutic circle 116 and perspective 60, 85–6
Hero of Alexander 43 information 21, 36, 75, 76, 80
Hertz, Heinrich 98, 192, 200–3, 284, 306 data models and 166
Bildtheorie controversy 195–7 experiments and 66
Weltbild 193 maps and 78–9, 82
hidden variables 30, 397 n. 23 measurement and 91, 143, 145, 146,
higher order resemblance 33–4, 35, 182, 150–1, 155–6, 157, 179, 180–1,
195 182, 183–4
Hilbert, David 382 n. 2, 383 n. 6 perspectival drawing and 8, 91–2
Hipparcus 177 perspective and 68–9, 72–3
homomorphism 18 scale models and 50, 56
Hooke, Robert 99 instrumentation 363 n. 15
horizon of alternatives 39, 59 astrolabes 61–2
Hughes, R. I. G. 345 n. 1, 346 n. 6 barometers 126
Hugh of St Victor 61 microscopes 93, 99–107, 108–9, 110
Hume, David 389 n. 6 observation by 93, 105–11
Huygens, Christiaan 131–2, 135, 195 observation metaphors 93, 96–9
Huygens, Constantijn 93–4, 96 roles of 94–100, 157
Hyman, John 37 thermometers 125–30, 144, 371 n. 5
hyperbolic geometry 213–14, 215 intensionality 27–8, 181
hyperrealism 11 intentionality 27–8, 181
interval measurement 159, 160, 161–2
invariance 52, 117, 136
ideal gas law 127–8 cross ratios 72–3, 74–5
image categories 104–5 dimensions and 53–5
imagery and perspective 72–3, 91–2, 175–9
 403
phenomena and 103, 108 Locke, John 389 n. 6

and scaling 158, 160–3 Lodge, David 210, 211
invariants 70, 103, 161–2, 176, 177, 179 Loemker, L. E. 394 n. 9
cross ratios 72–3, 74–5 Loewenheim–Skolem–Tarski-Vaught
irreducibility 304 theorem 230
Ismael, Jenann 359 n. 28 and n. 29, 382 n. logical necessity 278
3, 385 n. 7 logical space 164–6, 172–9
isomorphism 18, 214–15, 238, 247, 249, perspectival effects in 173–5
365 n. 7 Lopes, Dominic 36, 37, 38, 39
Weyl and 208–10, 211 Luce, Duncan 162–3
Jackson’s Mary Problem 210–12

Mach, Ernst 116–17, 123, 138, 191–2
Jauernig, Anja 358 n. 14, 365 n.3, 379 n. 5,
Bildtheorie controversy 197–8
387 n. 21, 394 n.9
thermometers 125–30
Maddy, Penelope 361 n. 1 (Part II)
Kant, Immanuel 8, 80, 257, 278–9, 283–4 maps 76–84
Kelly, Sean 372 n. 11 and indexicality 77–8, 79–80, 257–8
Kelvin, Lord (William Thomson) 113, 158, Margenau, Henry 166
368 n. 24 marginal distortion 34, 38, 183
kinematic imagery 34–5, 39, 182 Marlow, A. R. 315–16
kinematics 30, 34–5, 42, 274, 282, 285–6, Masaccio 64
287–8 mathematical imagery 39–49
kinetic theory 127–8, 130 distortion and 40
Klein, Felix 215, 229 geometric optics 42–5
Kuhn, Thomas 144 mathematical statues 41–2
Kulvicki, John 347 n. 17 Simpson’s paradox 48–9
mathematical statues 41–2
Maxwell, J. C. 195, 202, 278
Ladyman, James 381 n. 24, 385 n.2, Mead, Herbert 25
385 n.4 measurement 2–3, 91–2
Lambert, Joseph 117 altimetry 61–2
Lange, Marc 355 n. 40 approximative 163
language 83–4, 86 coherence and 145–6, 152–4
scientific 206–8 coordination and 136–7
theory-laden 84, 206 definition of 157
Latour, Bruno 370 n. 43, 376 n. 17 general theory of 147–56
Law of Charles and Gay Lussac 128 interval 159, 160, 161–2
Leeds, Stephen 321 n. 13, 386 n. 13, and logical space 164–6, 172–9, 179
387 n. 15 nominal 159, 160–1
Leibniz, G. W. 30, 69, 296, 389 n. 10 ordinal 159, 160
Leplin, Jared 199, 281 and perspective 8, 60–6, 73–5
Levine, Sherry 347 n. 17 and physical correlate 142–5, 156
Lewis, David 229, 231, 360 n. 33, 365 n. 1, in quantum mechanics 300–3
378 n. 3 and 4, 383 n. 10 scale/scaling invariance 161–3
light-clocks 132–3 significance 161–3
linear perspective 63–4 of time 130–6
linear one-point perspective 285, 286 types of 158–63
Lipton, Peter 370 n. 45 measurement, general theory of 147–56
liquid thermometers 126–7 coherence constraint 152–4
Lobachevsky, Nikolai 214 form of 150–2
404 
measurement, general theory of (cont.) non-simultaneity 119

initial set-up 147–8 Nyhof, John 379 n. 2
Value Definiteness 149–50, 154
Veracity in Measurement 149–50,
154–6, 373 n. 19 observables 137–9, 297–8, 300–2,
measurement outcomes 91, 157, 179–84 305–6
measurement procedures 123–4 fuzzy 184, 312, 313–14, 321 n. 7
measurement scales 159 and measurement 117, 134, 147, 150,
mechanics: determinism in 29–30 151–5, 182, 305
Mendelovici, Angela-Adeline 378 n.33, and models 169, 305
321 n. 6 sharp 184, 312–13
mental representation 24 observation 93
mental acts as 27, 28 instrumentation and 96, 100
see also Bildtheorie models and 87, 168–9
Mermin, David 170, 205 observation language 144
Michelson-Morley experiments 133 occlusion 34, 37, 39, 184, 313
microscopes 93, 99–107, 108–9, 110 one-point linear perspective 64
Middleton, W. E. Knowles 367 n. 20 opacity 27–8
Millikan, Robert Andrews 112 optics: geometric 42–5
Milne, Arthur 134 ordinal measurement 159, 160
mimetic experimentation 94–5 orthogonality 315–16
Minkowski, Hermann 71 Otte, Richard 375 n. 5
misrepresentation 13–15
modality 28, 119, 282–3, 318
models Padovani, Flavia 366 n. 8
data 166, 167–8, 172, 251–9, paintings 12, 141–2
391 n. 24 Panovsky 345 n.2, 349, n. 33, 357 n.10
geometric 285–6 Parkinson, G. H. R. 394 n. 9
theoretical 238, 240, 245–6, 248–50 Pascal, Blaise 46, 72, 126, 353 n. 26
and theories 309–11 pendulum clocks 131–2
surface 166–72, 240, 250–2, 257, pendulums 29, 131–2, 135, 137, 310
305, 315–16 Perrin, Jean Baptiste 113
Monton, Bradley 345 n. 4 Perry, John 182
Mohs hardness scale 160 Perspectiva 75, 285, 356 n. 4
Moore’s Paradox 260–1 perspectival drawing 62–5, 91–2
Morrison, Margaret 310 linear one-point 285, 286
as measurement 8
perspective 34
Nagel, Thomas 358 n. 13 (p.69) angle 38
nano-technology 94 and indexicality 60, 85–6
navigation 177–8 and invariance 72–3, 91–2, 175–9
necessary truth 296, 394 n. 9 linear 63–4
necessity 277–8 and logical space 173–5
neo-Kantian tradition 115, 120 marginal distortion 34, 38, 183
Neurath, Otto 137, 190 and measurement 60–6, 73–5, 183
Newman, M. H. A. 219, 222 occlusion 34, 37, 39, 184, 313
Newton, Isaac 30, 52, 257, 278, 280–1, one-point 64
317–18 texture-fading 34, 38, 184
nominal measurement 159, 160–1 two-point 64
non-committal 37, 38–9, 40, 50, 313 view from nowhere 69–72, 122
non-Euclidean geometry 213, 215–16 visual 84–6
 405
Peschard, Isabelle 351 n. 13, 360 n. 36, 367 Putnam’s Paradox (model-theoretic
n. 17, 377 n.21, 377 n.31, 378 n. 35 argument) 229–35
phenomena 8–9 dissolution of 232–5
and abstract structures 245–6, 249–50
and appearances 283–8, 317 quadrants 61
and data models 252–9 quantum theory 144, 155, 164, 183, 197,
outside experience 247–50 198, 199, 277
and theories 250–2, 259–61 and Appearance from Reality
philosophy of mind: and Appearance from Criterion 291, 297–300, 308
Reality Criterion 292–5 and data models 172,
photographs 18, 20–2, 178–9 and determinism 279–80
physical correlates 118–19, 121, 136, 179, and empiricism 306–7, 308
298, 304, 305 and fuzzy observables 184, 313–14
Criterion for Physical Correlate of and measurement 147–8, 154, 166, 184,
Measurement 182–3, 302, 312, 314 291, 300–3, 319
physical conditions for measurement and sharp observables 184, 312–13
141–6 and surface models 169, 170, 315, 316
theory of measurement 147–56 and supervenience 304
picture plane 62–4, 65, 358 n. 21 and theoretical models 252
picture theory of science, see Bildtheorie Quine, W. V. O. 392 n. 28
pictures 35–9
picturing 34, 182, 313
Pino, Paolo 356 n. 1 rainbows 102–3, 110
Place, U. T. 292 Raphael Sanzio 64
Planck, Max 192–5, 196, 279 ratio measurement 159, 160
planetary motion 8, 271, 286–8 ratio scales 128, 160
planisphere 356 n. 4 re-scaling 160–1
Plato 12, 24, 101–2, 231 realism 198–204, 229, 241–4
Pliny the Elder 11, 346 n. 1 real property 220–1
Poincaré, Henri 204, 279, 367 n. 12, scientific 198–9
380 n. 19 structural 198
and coordination 118, 208 reality 270–6
and measurement 125, 130–6, 138, Appearance from Reality
176, 183 Criterion 281–3, 291–300, 308
Poisson, Siméon 98 completeness criteria 276–83
Power, Henry 99 determinism 278–80
pragmatic contradictions 259 necessity and 277–8
pragmatic tautologies 259 primary/secondary qualities 271–6
pragmatics 3, 17, 21–2, 25, 82, 189, 190, rectilinear propagation 42
232–3, 259–60 reducibility 292–5
predication 16 reflection 42, 43–5
prediction 283 in water 101–2, 103, 105
primary/secondary qualities 34, 271–6 refraction 43, 44–5
Principle of Approximation 52–3 Reichenbach, Hans 30, 52–3,
Principle of Similitude 51–2 240–1, 279
probability 305, 317–19 common cause principle 280
projective geometry 215–16, 286 and coordination 118–21, 123, 136–7,
cross ratios 66, 72–3, 74–5 387 n. 21
Psillos, Stathis 253–4 relationality 26, 225–9, 231–2,
Ptolemy 43, 356 n. 4 242–3
Putnam, Hilary 229–35, 386 n. 7 relativity, principle of 69–71
406 
Renoir, Pierre-Auguste 35 Schlick, Moritz 117

representation 7–9, 35 Schrödinger, Erwin 198
asymmetry of 17–18, 189 scientific realism 8, 113, 184, 194, 198–9,
committal/non-committal 38, 39 237, 266, 281
and distortion 14 screw propeller 55–6
intentional 27 sculpture 12, 38, 41–2
modes of 33–5 selective likeness/unlikeness 7–8, 9, 14,
and referents 244–5 18, 23, 30–1, 33, 34, 141
and resemblance 14, 17–19, 30–1, drawing and 91
33–4, 35 imagery and 39, 182
and structuralism 204–12 and measurement 60, 91, 179–80,
use and 22–6, 28 182–3
resemblance 11–13, 35 scale models and 49–50, 57
caricature and 14 selectivity 7–8, 37, 76, 87
and distortion 14 data models and 253–4
measurement and 192–3 maps and 261
misrepresentation and 14 mathematical models and 242, 243, 247
paintings and 141–2 self-ascription 79, 82, 83
and representation 14, 17–19, 30–1, self-location 78–84, 85
33–4, 35 Sellars, Wilfrid 33–4, 84, 349 n. 2, 379 n. 8
Socrates on 19–20 semantics 17, 27, 84, 232, 258, 259–60,
Richer, Jean 132 semantic view of theories 239, 247,
Richter scale 375 n. 5 309–11
Rosselli, Francesco 63 set theory 353 n. 29
Rossetti, Dante Gabriel 142 sextants 178
Rothschild, Daniel 384 n. 2 sharp observables 184, 312–13
Rovelli, Carlo 372 n. 14, 377 n. 32, Shepard, Roger 13
395 n. 10 Shomar, Towfic 310
Rubens, Peter Paul 35 sidereal clocks 132, 134
Russell, Bertrand 213, 216–17, 229, 239, similitude, principle of 51–2
368 n. 27, 368 n. 28 simplicity 133–4, 136, 138–9, 202, 203
epistemology 222 Simpson’s paradox 48–9
and real property realism 220–1 simultaneity 37, 68, 70, 81, 165, 166, 312
and structuralism 217–23 sines, law of 353 n. 21
Rutherford, Ernest 76 Sneed, Joseph 367 n. 12
Ryckman, Thomas 135, 358 n. 15, Snel van Royen, Willebrord 353 n. 21
361 n. 43 Snell’s (Snel’s) law 44
Socrates 12, 19–20, 24
Saliger, Ivo 35 Solovay, R. M. 353 n. 29
Salmon, Wesley 108, 199 space of reasons 84
saving the phenomena 8 Spott, E. 14
scale: as logical space 164–6 standard meter 135
scale models 49–50 Stern, Otto 155
scale transformation 52, 53–5, 161, 162 Stern–Gerlach apparatus 155, 179, 305–6
scaling Sterrett, Susan 55
and imaging 56–7 Stevens, S. S. 158, 159, 161
and invariance 54–5, 160–3 stochastic response function 151, 169, 300
as picturing 50–1 structural realism 198
scaling invariance 54–5, 160–3 structuralism 191
Scanning Tunneling Electron Carnap and 225–9
Microscope 94 empiricist 237–9
 407
Newman and 219 Thompson, Paul 310

and representation 204–12 Thomson, William (Lord Kelvin) 113, 158,
Russell and 217–23 368 n. 24
Suarez, Mauricio 7, 25–6, 310–11 time 34, 68, 70–1, 87, 115, 156, 164, 242
supervenience in equations 247–8
without deducibility 296–7 self-location and 81, 178
and irreducibility 304 and space 280, 288
without reducibility 292–5 space–time 71, 137, 221, 223, 289
Suppes, Patrick 167, 172, 309 unit of 119
surface model 166–172, 240, 250–252, see also kinematics; time measurement
257, 305, 315–316 time measurement 118, 122, 125, 130–6,
surface models 167, 168–72, 252, 315–16 176–7, 183, 317–18
symmetry 18, 40, 66, 163, 196, 211, 212, clocks 131–4, 199
226, 234 Timpson, Chris 377 n. 29, 378 n. 32
synchrony 131–3, 134, 136 Tolman, Richard 51–2
syntax 26, 233, 294, 351 n. 14, 385 n. 6 Toulmin, Stephen 144
tragedy 7, 26, 265–6
Tarski, Alfred 260 transcendentalism 8, 266
Teller, Paul 45, 105, 107–8, 391 n. 24 transformations 55, 161–3, 174
temperature 53, 86, 116–17, 122–3, 124, translation 84, 161–2, 394 n. 5
131, 138 truth 144, 200, 234, 246, 260, 283
Fahrenheit/Celsius scales 161–2, 174 contingent 296, 394 n. 9
of gases 138, 242 correspondence theory of 244, 248–9,
and length measurement 135 252
and mean kinetic energy 173 images and 14, 20
pendulums and 132 necessary 296, 394 n. 9
statues and 42 secondary qualities and 274
thermometers 125–30, 146, 160, 175, two-point perspective 64
180
weather forecasts 77, 166 unsharp observables 312, 313–14,
texture-fading 34, 38, 184 321 n. 7
Thales 356 n. 5 unsharp values 154, 376 n. 13
Theaetetus 12
theoretical models 238, 240 Value Definiteness 149–50, 154,
and phenomena 245–6, 248–50 373 n. 19
Theorica 288–90, 308 van Leeuwenhoek, Antonie 104
theories Vasari, Giorgio 64
de-idealization of 310–11 Veracity in Measurement 149–50, 152,
and models 309–11 154–6, 373 n. 19
and phenomena 250–2, 259–61 verbal description 17, 86
semantic view of 309–11 view from nowhere 69–72, 122
theory-laden language 84, 206 visual imagery 11, 34, 35, 39, 40, 182
theory-ladenness 75, 94, 144, 181 visual perspective 84–6, 182
thermodynamics 42, 279 von Guericke, Otto 94, 95
thermometers 125–30, 144, 146, 160, 173, von Helmholz, see Helmholtz, von
175, 180, 371 n. 5 von Neumann, J. 47
thermometry 116–17, 123
thermometers 125–30, 144, 146, 160,
173, 175, 180, 371 n. 5 Weinberg, Steven 110–11
Thomas, David A. 357 n. 11 Weltbild 141, 192–5, 237, 274, 275
408 
Weyl, Hermann 71, 136, 175, 228–9 Worrall, John 239, 362 n. 10, 378 n. 34,
and isomorphism 208–10, 211 381 n. 24, 385 n. 2
Whitehead, A. N. 243
Wigner, Eugene 303 Young, Thomas 251
Williams, Bernard 358 n. 13
Wilson, Catherine 94–5, 99
Wilson, Margaret 389 n. 1 Zajonc, Arthur 359 n. 23, 396 n. 19
Wittgenstein, Ludwig 38, 164 zero-point perspective 69
Woodward, James 376 n. 14 Zeuxis 11
world-picture 141, 192–5, 237, 274, 275 Zola, Émile 350 n. 5

Van Fraassen Bas C. (2008) - Scientific Representation Paradoxes of Perspective PDF

Uploaded by

Copyright:

Available Formats

You might also like

Van Fraassen Bas C. (2008) - Scientific Representation Paradoxes of Perspective PDF

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

Van Fraassen Bas C. (2008) - Scientific Representation Paradoxes of Perspective PDF

Uploaded by

Copyright:

Available Formats

Scientiﬁc Representation: Paradoxes of Perspective

This page intentionally left blank

Bas C. van Fraassen

CLARENDON PRESS · OXFORD

When I began to rewrite the Locke Lectures I gave in Oxford in 2001

The generous sabbatical and leave policy of Princeton University, a

2.1. Reﬂection and Refraction 44

Introduction: the ‘picture theory of science’ 1

. Representation of, Representation As 11

. Imaging, Picturing, and Scaling 33

. Pictorial Perspective and the Indexical 59

Part II. Windows, Engines, and Measurement

. A Window on the Invisible World (?) 93

. The Problem of Coordination 115

. Measurement as Representation: . The Physical Correlate 141

. Measurement as Representation: . Information 157

Part III. Structure and Perspective

. From the Bildtheorie of Science to Paradox 191

. The Longest Journey: Bertrand Russell 213

. Carnap’s Lost World and Putnam’s Paradox 225

. An Empiricist Structuralism 237

Part IV. Appearance and Reality

. Appearance vs. Reality in the Sciences 269

. Rejecting the Appearance from Reality Criterion 291

Appendix to CH . Models and theories as representations 309

Notes to Appendices 320

The Bildtheorie—‘picture theory of science’—formed the frame for much

The most naïve view of representation might perhaps be put something

The value of distortion2

Caricature and misrepresentation

How does a representation represent?

Nelson Goodman’s Languages of Art, the twentieth century’s seminal

Resemblance in discord with representation

attending to resemblances and non-resemblances. To show this, Goodman

Socrates’ thought experiment leading to this remark has a quite contem-

Here the ‘‘mere resemblance’’ view of representation appears on Socrates’

There is no single immediately and obviously right answer; there couldn’t

What is a representation then?

What is represented, and how it is represented, is not determined by

on the intentional account of representation a theory cannot ‘represent’ a phe-

The objection, if valid, would not just apply to a theory or a model, it

Relation, intention, intension

The ‘‘of’’ that marks the relation of representing object to what is

We see this in modal contexts: 9 = the number of planets, but it does

Appearance to the intellect: illumination as embedding

non-resemblance, but what the selections are must be somehow high-

Despite the arguments against taking resemblance to be the clue to

and models.2 Resemblance may consist in having properties in common,

Resemblance, as I said, can be higher order: the spatial structure of a set

What distinguishes a picture?

is only the beginning of the systematic mis-matching necessary in a picture

Grain: more distant objects, textures are not as ﬁnely depicted as

Does this allow us to differentiate picturing from describing? Given a

Mathematical imagery, distortion

secondly because the representor is a mathematical object. While there is

If we are to bring these concepts to the study of scientiﬁc representation

versatile than I have indicated yet.16 Thermodynamic study lets in a fourth

Trouble at the interface