REALISM BETWEEN METAPHYSICS AND SCIENCE A DISSERTATION … · The aim of my dissertation is to understand what the realism debate in the philosophy of science should be about, and,

REALISM BETWEEN METAPHYSICS AND SCIENCE

A DISSERTATION

SUBMITTED TO THE DEPARTMENT OF PHILOSOPHY

AND THE COMMITTEE ON GRADUATE STUDIES

OF STANFORD UNIVERSITY

IN PARTIAL FULFILLMENT OF THE REQUIREMENTS

FOR THE DEGREE OF

DOCTOR OF PHILOSOPHY

Johanna Wolff

July 2010

http://creativecommons.org/licenses/by-nc-sa/3.0/us/

This dissertation is online at: http://purl.stanford.edu/dx641bh8212

© 2010 by Johanna Wolff. All Rights Reserved.

Re-distributed by Stanford University under license with the author.

This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 United States License.

ii



http://purl.stanford.edu/dx641bh8212

I certify that I have read this dissertation and that, in my opinion, it is fully adequatein scope and quality as a dissertation for the degree of Doctor of Philosophy.

Alexis Burgess, Primary Adviser


Thomas Ryckman, Co-Adviser


Nadeem Hussain


Helen Longino

Approved for the Stanford University Committee on Graduate Studies.

Patricia J. Gumport, Vice Provost Graduate Education

This signature page was generated electronically upon submission of this dissertation in electronic format. An original signed hard copy of the signature page is on file inUniversity Archives.

iii

Abstract

“Realism is dead”, said Arthur Fine a quarter of a century ago. What Fine was trying to bury

wasn’t just realism, but debates about realism or antirealism in the philosophy of science.

The aim of my dissertation is to understand what the realism debate in the philosophy of

science should be about, and, hopefully, to revive interest in the debate.

To achieve this goal, a number of questions need to be addressed: What is the target of

realism and antirealism in the philosophy of science, and what does it mean to be a realist

or an antirealist? Is realism the only position in this debate which implies a commitment

to metaphysics, or are both realism and antirealism equally metaphysical positions? Is

there really anything to be debated, or should we just be quietists about realism debates in

philosophy of science?

My answer will be that, in its current form, the realism debate in philosophy of science

is a skeptical debate, that is, antirealism is offered exclusively as an epistemic challenge to

realism. This leads to an unsatisfactory stalemate between realism and antirealism, which

often prompts a kind of quietism about these debates. While I reject this quietism as in-

sufficiently supported by argument, I concede that the debate in its current shape is not

satisfactory either. Instead I propose to change the target of realism debates away from

claims about unobservables and towards modal claims made in the sciences. Debates about

the latter, I argue, can go beyond skeptical challenges to include semantic and metaphysical

questions as well. Switching the target in this way makes the debate about realism in the

philosophy of science more like realism debates in other fields.

iv

Acknowledgements

Like most academic endeavors, this dissertation would not have been possible without

the help and support of many people. I am very grateful to my advisors and committee

members Tom Ryckman, Alexis Burgess, Helen Longino, and Nadeem Hussain, for their

sheer endless patience with my attempts to bang my head against several walls at once.

I would also like to thank my former committee members, Michael Friedman and Ken

Taylor, who gave me helpful advice in the early stages of this project.

I am grateful to many members of the philosophy department at Stanford, but I would in

particular like to thank Chris Bobonich and Lanier Anderson for their support and advice,

as well as Jesse Alama, Alexei Angelides, Tal Glezer, Dan Giberman, Tomohiro Hoshi,

Alistair Isaac, Samuel Kahn, Micah Lewin, Peyton McElroy, Teru Miyake, Assaf Sharon,

Quayshawn Spencer and Ben Wolfson for relentless criticisms of parts of earlier versions

of this project.

Beyond the Stanford community, I would like to thank Hans Halvorson for generously

taking the time to explain to me the basics of quantum field theory, as well as Katherine

Brading, James Ladyman, and George Smith. I would also like to thank Johannes Haag

and Karl-Georg Niebergall who sent me on this way.

Finally I would like to thank the Andrew W. Mellow Foundation and the American

Council of Learned Societies, who supported my research through a Dissertation Comple-

tion Fellowship as part of the Andrew W. Mellon/ACLS Early Career Fellowship Program

during the academic year 2009/10.

v

Contents

Abstract iv

Acknowledgements v

1 Introduction 1

2 The Carnap - Quine debate 102.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Carnap’s anti-metaphysical stance . . . . . . . . . . . . . . . . . . . . . . 11

2.2.1 Carnap’s position in ESO . . . . . . . . . . . . . . . . . . . . . . . 12

2.2.2 How is Carnap’s position supposed to work? . . . . . . . . . . . . 16

2.3 A closer look at some aspects of Carnap’s view . . . . . . . . . . . . . . . 18

2.3.1 Agreement about evidence . . . . . . . . . . . . . . . . . . . . . . 18

2.3.2 Deflationism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.4 Non-Cognitivism and Quietism . . . . . . . . . . . . . . . . . . . . . . . . 24

2.5 The practical/epistemic distinction criticized . . . . . . . . . . . . . . . . . 29

2.5.1 Arbitrary Frameworks . . . . . . . . . . . . . . . . . . . . . . . . 29

2.5.2 Existence questions and pragmatism . . . . . . . . . . . . . . . . . 33

2.6 Yes/No questions and pragmatism . . . . . . . . . . . . . . . . . . . . . . 36

2.6.1 Chemical elements . . . . . . . . . . . . . . . . . . . . . . . . . . 37

2.6.2 Charged particles . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

vi

3 Realism debates in philosophy of science 443.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.2 The skeptical challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

3.2.1 Constructive Empiricism . . . . . . . . . . . . . . . . . . . . . . . 45

3.2.2 The pessimistic meta-induction argument . . . . . . . . . . . . . . 48

3.3 A new quietism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.4 Should we settle for realist quietism? . . . . . . . . . . . . . . . . . . . . . 57

3.4.1 Three worries about quietism . . . . . . . . . . . . . . . . . . . . . 57

3.4.2 Moving to the exit . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.5 Changing the target - modal claims . . . . . . . . . . . . . . . . . . . . . . 66

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

4 Where does structural realism fit? 734.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

4.2 An alternative to standard scientific realism . . . . . . . . . . . . . . . . . 76

4.3 A thesis in metaphysics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

4.3.1 What is ontic structural realism? . . . . . . . . . . . . . . . . . . . 80

4.3.2 ESR as a response to PMI . . . . . . . . . . . . . . . . . . . . . . 83

4.3.3 Theories and Ramseyfication . . . . . . . . . . . . . . . . . . . . . 84

4.3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4.3.5 The dialectic between ESR and OSR . . . . . . . . . . . . . . . . . 88

4.4 As the ‘ontology’ of particular theories . . . . . . . . . . . . . . . . . . . . 90

4.4.1 The argument from the interpretation of quantum mechanics . . . . 91

4.4.2 The argument from the interpretation of QFT . . . . . . . . . . . . 95

4.4.3 How well does this strategy work? . . . . . . . . . . . . . . . . . . 100

4.5 As a resolution of underdetermination problems . . . . . . . . . . . . . . . 101

4.5.1 The argument from multiple interpretation . . . . . . . . . . . . . . 102

4.5.2 Other cases of multiple interpretation? . . . . . . . . . . . . . . . . 104

4.6 A different debate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106

4.6.1 Objects and Structure - which one has priority? . . . . . . . . . . . 109

4.6.2 Structure and Modality . . . . . . . . . . . . . . . . . . . . . . . . 112

vii

4.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5 The varieties of necessity in physical laws 1155.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

5.2 Varieties of necessity and the place of laws . . . . . . . . . . . . . . . . . . 117

5.2.1 Lange’s view . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118

5.3 Some examples of varieties of laws . . . . . . . . . . . . . . . . . . . . . . 129

5.3.1 Electrons have negative charge . . . . . . . . . . . . . . . . . . . . 130

5.3.2 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . 131

5.3.3 What about other laws? . . . . . . . . . . . . . . . . . . . . . . . . 136

5.4 The necessity of laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138

5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

6 Conclusion 1466.1 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

6.2 The Answer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

viii

List of Figures

3.1 Realist and Antirealist options . . . . . . . . . . . . . . . . . . . . . . . . 69

4.1 What is the thesis of structural realism? . . . . . . . . . . . . . . . . . . . 75

4.2 The Aharonov-Bohm Effect . . . . . . . . . . . . . . . . . . . . . . . . . . 98

ix

Chapter 1

Introduction: Realism betweenmetaphysics and science

The aim of my dissertation is to understand what realism debates in philosophy of science

should be about. This question is of interest not just to philosophers of science, but also to

metaphysicians, since it bears directly on the question of the relationship between science

and philosophy of science and metaphysics.

To achieve this goal, a number of questions need to be addressed: What is the target of

realism and antirealism in the philosophy of science, and what does it mean to be a realist

or an antirealist? Does only realism in this debate imply a commitment to metaphysics, or

are both realism and antirealism equally metaphysical positions? Is there really anything to

be debated, or should we just be quietists about realism debates in philosophy of science?

My answer will be that, in its current form, the realism debate in philosophy of science

is a skeptical debate, that is, antirealism is offered exclusively as an epistemic challenge to

realism. This leads to an unsatisfactory stalemate between realism and antirealism, which

often prompts a kind of quietism about these debates. While I reject this quietism as in-

sufficiently supported by argument, I concede that the debate in its current shape is not

satisfactory either. Instead I propose to change the target of realism debates away from

claims about unobservables towards modal claims made in the sciences. Debates about the

latter, I argue, can go beyond skeptical challenges to include semantic and metaphysical

questions as well. Switching the target in this way makes the debate about realism in the

1

CHAPTER 1. INTRODUCTION 2

philosophy of science more like realism debates in other fields.

Both realism and antirealism turn out to be metaphysical views in two different senses.

First, they are metaphysical because they are genuine philosophical positions – they arise

as answers to a distinctively philosophical project. Second, since they make (positive or

negative) claims about the sources of modality and the truth-makers of modal claims, they

involve various general metaphysical commitments, e.g., about the relations between modal

and non-modal claims, or even the ontology of properties. This is a welcome result, since

it allows a more cooperative relationship between metaphysics and philosophy of science,

and it suggests a way in which metaphysics might be ‘accountable to’ science.

A defining feature of analytic philosophy as a movement has been its admiration and

respect for the natural sciences. Early analytic philosophy hoped to make philosophy ‘sci-

entific’ through clarity, the use of formal methods, and the rejection of anything that looked

like speculative ‘metaphysics’. Proper philosophy was supposed to be philosophy of sci-

ence. Since then formal methods have become widespread, clarity continues to be praised,

although not always practiced, but philosophy is not just philosophy of science, and meta-

physics is a thriving discipline within analytic philosophy.

With metaphysics on the rise and philosophy of science, perhaps not in decline, but at

least somewhat marginalized, philosophers of science in particular should ask themselves

how their discipline relates to metaphysics. This question often provokes hostile or polem-

ical reactions. Some philosophers of science continue to see philosophy of science as an

enterprise in competition with metaphysics: philosophy of science is what we ought to do

instead of metaphysics. Other philosophers of science (Ladyman and Ross 2007) seem

more inclined to think that doing philosophy of science means doing metaphysics, and do-

ing metaphysics in the right way. Metaphysics, on their view, ought to be closely tied to

scientific theories and results, not some independent philosophical discipline.

To make matters worse, ‘metaphysics’ is used in more than one way in these debates:

sometimes what is meant is a certain kind of philosophical attitude, a certain kind of re-

alism, in other contexts what is meant are a priori investigations or claims, and at least

occasionally metaphysics is used as a label for particular philosophical positions.

Metaphysicians, on the other hand, have focused much of their energy on defending


metaphysics as a discipline against various ‘deflationists’, who, in the tradition of ordinary

language philosophy and (perhaps) logical positivism, attempt to dismiss metaphysical de-

bates as merely verbal disputes.1 While this is certainly a worthwhile and necessary debate

to have, it doesn’t really help to illuminate the relationship between metaphysics and phi-

losophy of science, or metaphysics and science, for that matter. As far as the latter is

concerned, many metaphysicians seem to think that metaphysics should take seriously the

results and theories of science, but this leaves room for interpretation. Does it mean that

science constrains what metaphysicians can say? Or can we just ‘read off’ our metaphysics

from our best scientific theories? Can metaphysics be in conflict with claims in science, or

are they somehow operating on different levels of investigation?

I will focus exclusively on the relationship between metaphysics and philosophy of

science, and not engage in the debate over whether ontological and other metaphysical

questions are ‘merely verbal’. I will also usually work with a relatively broad notion of

metaphysics, one that goes significantly beyond ‘ontological’ questions. This is necessary

not only to accommodate the vast range of positions philosophers on all sides of the debate

hold, but also to include the matter of realism and antirealism themselves, which on some

views are metaphysical positions.

It is one of the peculiarities of the realism debate in philosophy of science that two

different views of the debate about realism in the philosophy of science might count as

rejecting metaphysics: quietism and antirealism.

Since the most articulate antirealism in the philosophy of science is empiricism (these

days typically constructive empiricism), and since empiricism takes itself to be an anti-

metaphysical view, being a realist in the philosophy of science is often taken to mean:

being metaphysically committed. This is rather odd, since in most other realism/antirealism

debates, realism and antirealism alike are regarded as metaphysical positions, so being an

antirealist would count as no less metaphysical than being a realist. To avoid metaphysics

would mean to be a quietist.

Quietism is the view according to which a particular kind of debate is ill-conceived or1See (Hirsch 2008) for an articulation of the deflationist viewpoint, and (Sider 2009) for a response on

behalf of metaphysics. For a (somewhat polemical) defense of metaphysics against ‘linguistic philosophy’more broadly, see (Williamson 2007).


misguided, and should hence be avoided altogether. A quietist might consider the debate

over realism as such a debate: pointless, wrongheaded and to be avoided. A quietist dis-

misses a particular kind of debate, whereas an antirealist takes a particular position within

that debate. Dismissing a debate altogether requires strong arguments and, as we shall see

in Chapters 2 and 3, proves to be difficult.

Both antirealism and quietism are sometimes offered as anti-metaphysical positions, but

we should be careful to distinguish the two. Notice, for example, that they are incompati-

ble. The antirealist takes the debate he is contributing to as at least intelligible, and perhaps

even worthwhile, whereas the quietist thinks of the debate as mistaken from the start. In

the philosophy of science, however, quietism and antirealism are not always clearly distin-

guished. One of the reasons for this is that both quietism and antirealism can be motivated

by empiricism.

Empiricism can be seen as a position within the debate over scientific realism, but it

has also been used to motivate the idea that debates about realism are ill-conceived. As

constructive empiricism, empiricism is an antirealist view in the scientific realism debate,

proposing that we take empirical adequacy as the aim of our scientific theorizing. For log-

ical positivism, on the other hand, empiricism served as a reason to reject realism debates

as ill-conceived because they lacked appropriate standards of evidence. In both forms em-

piricism can be seen as opposed to ‘metaphysics’, but notice that metaphysics in the first

case means any claim that goes beyond the phenomena, whereas metaphysics in the second

case means realism/antirealism debates themselves.

Quietism is not always motivated by empiricism, though. As we shall see, contempo-

rary quietists are frequently motivated by naturalism, not empiricism. Naturalism is the

view that philosophy is continuous with science, which is often understood to mean that

philosophers do not have access to methods beyond the methods used in science, or that

philosophy does not have a vantage point from which to judge science.2 A naturalisti-

cally motivated quietism can also seem like an anti-metaphysical position: metaphysics

is here understood as ‘first philosophy’, that is, as an attempt of philosophers to engage

in questions and debates of a distinctively philosophical nature. In a different sense of

2For the origin of naturalism in this sense see (Quine 1951) and (Quine 1969a). For a more recent defensesee (Maddy 2007).


metaphysics, however, naturalism is not anti-metaphysical at all: unlike the empiricist, the

naturalist does not take claims about unobservables or ontological commitment to unob-

servable entities to be problematic. Unlike empiricists, naturalists tend to be sympathetic

to realism. The empiricists’ skepticism is just a sign of their mistaken commitment to first

philosophy.

In the dissertation I will look both at empiricist and naturalistically motivated quietism.

I argue that neither empiricists nor naturalists succeed in presenting conclusive arguments

for quietism. At the same time, the continuing appeal of quietism in these debates has

to do with the peculiar form realism debates take in the philosophy of science: they are

at bottom skeptical debates. This reduces the number of realist and antirealist positions

available in these debates, and the realism and antirealism available seem rather unattractive

as philosophical positions. The lack of attractive positions and arguments makes quietism

seem appealing.

Despite this I suggest that we should not be quietists. To break out of the skeptical

stalemate, we need to change the target of realism/antirealism debates in the philosophy

of science from claims about unobservables to modal claims. Modal claims, unlike claims

about the unobservable, are problematic even for non-empiricists. What makes them true?

What is the source of their modality? How do we know them?

The status of modal claims has long been a matter for philosophical debate. The debate

over modal claims in the natural sciences can be traced back to at least Hume and Kant.

Modality also played an important role in the discussions of the Vienna Circle. So to

suggest that modal claims ought to be the target of realism debates in philosophy of science

is not an entirely new idea. The new debate over modal claims will differ in a number of

respects from its predecessors, however.

First, traditionally much of the debate about modal claims has been framed in terms

of causality, or ‘necessary connections’. I’m working with a broader notion of modality:

claims about what is necessary and possible, laws of nature, and perhaps even probability

claims.

The second main difference is that the traditional debates were not explicitly framed

as realism/antirealism debates, whereas I hope to articulate a debate in analogy with real-

ism/antirealism debates in meta-ethics and the philosophy of mathematics. Like the subject


matter of meta-ethics or the philosophy of mathematics, I propose that the status of modal

claims can and should be questioned from epistemic, metaphysical, and semantic perspec-

tives. Are modal claims, or modal claims of a particular type, true or false? How can we

tell the true from the false ones? Are certain kinds of claims, like laws of nature, neces-

sary? This differs from the empiricist way of motivating the debate as a primarily epistemic

debate.

Finally, the target of the debate as I propose it are not modal claims in general, but

modal claims in science. This raises further questions about the relationship between (po-

tentially) different forms of modality, including metaphysical, logical, and mathematical

necessity on the one hand, and the modal status of claims in the sciences on the other. Here

philosophy of science will benefit from the progress that has been made in characterizing

the former forms of necessity.

The project in this dissertation is to articulate not an answer to an existing debate, but

to change an existing debate. What, one might ask, is the point of such a change: what are

the advantages of switching the target of realism debates in the philosophy of science?

The change in debate I argue for in this dissertation has several advantages. As already

mentioned, the new debate promises to be more fruitful than the skeptical debate in which

we seem to have reached a stalemate. Since modal claims raise questions not just for

empiricists, but for everyone, it is a debate that more parties have an interest to engage

in. Moreover, certain kinds of modal claims, such as laws of nature and perhaps also

probability claims, are already subject to significant debate. Changing the realism debate

in the way I’m suggesting in the dissertation will help to put these specialized debates into

an overall framework that will allow a clearer statement of the various options and their

realist or antirealist implications.

A second advantage of changing the debate is that it makes clearer what the relationship

between realism debates and metaphysics in the philosophy of science is. In the philoso-

phy of science, frequently realism, but not antirealism, is regarded as metaphysics. The

debate I’m proposing here is more symmetrical. Both realism and antirealism alike are

metaphysical positions in the two senses I described above: they are answers to distinc-

tively philosophical questions, and they offer views about the nature of modality, whether


realist or antirealist. The relationship of philosophy to science is not that of ‘reading off’

metaphysics or ontology from science. Metaphysics as a discipline contributes to this en-

terprise by providing a rich background of arguments over modality, whereas philosophy of

science will typically provide analyses of particular kinds of modal claims and their func-

tion in scientific theorizing. This new interaction and cooperation between metaphysics

and philosophy of science will hopefully replace the outdated conflict between the two

disciplines.

Finally, the new debate is able to make sense of more recent contributions to the realism

debate in philosophy of science like structural realism. Structural realism has emerged over

the past two decades as a contender in the realism debate in philosophy of science, but with

increasing popularity of structural realism as a position it has become more difficult to say

just what structural realists claim, or even what kind of a thesis they are offering. Structural

realism oscillates between a position within debates in philosophy of science and a much

grander metaphysical thesis. Changing the debate over realism in the way I proposed will

allow us to position structural realism more clearly. Changing the target of the realism de-

bate in philosophy of science has a number of advantages, then, some of which go beyond

philosophy of science.

Since this dissertation does not proceed by adding a move or a position to an already

existing debate, the format will be somewhat atypical. Much of the philosophical work in

this dissertation consists in unravelling various debates that seem to bear on the question of

how science, metaphysics, and the question of realism are related. Accordingly a lot of the

discussion will take the form of pursuing different ways in which particular debates have

been understood. My aim is to show that there is a serious question about how to understand

realism and antirealism in the philosophy of science, and in what sense such positions are

themselves metaphysical or come with distinctively metaphysical commitments. I do not

argue for either realism or antirealism ‘about science’. Instead I try to articulate just what

such a view might amount to.

Similarly, since the dissertation aims to reconnect philosophy of science and meta-

physics, I discuss literature from both fields, even though the emphasis will be on work

classified as philosophy of science. While there is much overlap in the terminology used in


these fields, there will occasionally be differences in how particular key expressions such

as ‘metaphysical necessity’ are used, which I will clarify where they occur.

The dissertation has four main chapters, and a short conclusion. Each chapter is largely

self-contained, with occasional references to results in other chapters. There are a number

of recurring themes, though: the attitude with which philosophers should approach science,

the continuity of scientific and philosophical questions, the conflict between empiricism

and naturalism, and of course the relationship between science and metaphysics.

In Chapter 2 I look at the locus classicus for the debate between empiricism and natural-

ism: the Carnap - Quine debate. Much of that chapter will be devoted to the articulation of

the anti-metaphysical stance Carnap offers in Empiricism, Semantics, and Ontology. This

stance, I suggest, is a form of quietism, not antirealism. Carnap’s quietism is supported

by a combination of verificationism, deflationism, and non-cognitivism. I argue that Car-

nap’s verificationism is not well defended, that Quine’s objections against his deflationism

are successful, and that his non-cognitivism is interesting, but not enough to motivate qui-

etism once it is isolated from its verificationist and deflationist companions. At the end of

the chapter I nevertheless defend one aspect of Carnap’s position against Quine’s holism.

While this is not enough to defend the anti-metaphysical position Carnap offered, it does

shed some light on the idea that realism questions when formulated as existence questions

might be problematic.

Chapter 3 takes on the contemporary debate over scientific realism. I show that the

debate in its current form is ultimately a skeptical debate, and neither the realism nor the

antirealism that emerge from it are compelling options. I also argue that the quietism that

develops in response to the skeptical debate fails to offer conclusive reasons for abandoning

the debate. I then propose a change in the target of realism debates in the philosophy of

science, away from claims about unobservable entities towards modal claims. I sketch out

a variety of realist and antirealist options in such a debate, without defending any particular

one.

Chapter 4 is an attempt to understand how structural realism, a recent addition to the

realism literature, fits into the debate. I distinguish four different branches of structural

realism, and argue that as it is currently presented, there is no single position ‘structural

realism’, but a collection of positions bearing a certain family resemblance to one another.


Moreover, in the particular debates to which different branches of structural realism are

offered as contributions, structural realism usually doesn’t do any better than its local com-

petitors. I suggest that we can make more sense of structural realism if we see it as a con-

tribution to the new debate over scientific realism I suggest in Chapter 2. Ontic structural

realists offer a particular realist stance within the debate over the status of modal claims

in science. The difficulties they face in articulating and defending that particular realist

position suggests that in order to articulate views within the new debate over realism in the

philosophy of science, philosophers of science have to engage in a number of metaphysical

disputes as well.

The interconnection between philosophy of science and metaphysics explored in Chap-

ter 3 is not the only point of contact between the two in this new debate. Chapter 5 investi-

gates the question of whether there might be different types of necessity or modality among

modal claims in science, and in particular among laws of physics. While conservation laws

in modern field theories can be said to be metaphysically necessary, other laws have to be

accounted for differently. I propose that the necessity of such laws reflects their status as

rules, not a variety of alethic modality.

I do not opt for either a form of realism or a form of antirealism about modal claims of

science in this dissertation. My goal is not to position myself within the new debate, but

to suggest that this is the shape the realism debate in philosophy of science should take. A

debate about the status of modal claims promises to be fruitful, and it opens the debate for

interactions with broader questions in metaphysics. The old division between metaphysics

and philosophy of science has to be given up, for the benefit of both.

Chapter 2

The Carnap - Quine debate

2.1 Introduction

In this chapter I will look at what is often considered the turning point in the relationship

between analytic philosophy and metaphysics: the Carnap - Quine debate. Rudolf Carnap

is rightly regarded as an ‘anti-metaphysician’ par excellence, but what is the metaphysics

he was opposed to, and why did he oppose it?

Initially much of Carnap’s opposition against metaphysics was opposition to ‘non-

analytic’ philosophers, famously against Martin Heidegger’s philosophy (Carnap 1932).

Of more interest to contemporary analytic philosophers are Carnap’s later attempts to avoid

metaphysics, especially his arguments in Empiricism, Semantics, and Ontology (ESO)

(Carnap 1956). In this chapter I will be concerned with Carnap’s position on metaphysics

as presented in ESO, and its relevance for contemporary debates about metaphysics and

realism.

The main target of Carnap’s anti-metaphysical position in his later philosophy are re-

alism/antirealism debates about science and mathematics. Carnap seems to take what we

would now call a quietist attitude towards such debates: he would prefer to stay out of the

debate altogether. Carnap’s quietism appears to be motivated by a combination of verifi-

cationism, deflationism, and a certain kind of non-cognitivism. In sections two and three I

will show how these different strands of anti-metaphysics come together in Carnap’s later

work, and raise worries about the different aspects of his anti-metaphysical position.

10

CHAPTER 2. THE CARNAP - QUINE DEBATE 11

I then discuss in section four whether there is still something to the non-cognitivism

about realism debates about, even if we give up the verificationist and deflationist elements

of Carnap’s thought. While I suggest that this is possible, I argue that it doesn’t show that

we should be tolerant, that is, quietism doesn’t follow from non-cognitivism.

Finally I look at one aspect of Quine’s criticism of Carnap, the question whether there is

anything to the idea that ontological questions are practical, and that this suggests that they

are ‘non-cognitive’ in some sense. Quine argues that all our reasons are ultimately practical,

and accordingly that there is no important philosophical distinction between ‘practical’ (or

pragmatic) and ‘evidential’ (or epistemic) reasons. I suggest that even if Quine is right

about that, there remains a distinction between yes/no questions and other questions.

2.2 Carnap’s anti-metaphysical stance

Empiricism, Semantics and Ontology is widely regarded as Carnap’s most definitive state-

ment of his views on ontology.1 The purpose of this section is to gain a clear understanding

of Carnap’s anti-metaphysical stance as it is presented in this particular paper and surround-

ing works. I will argue that Carnap does not introduce a two-fold or four-fold2 distinction

among different questions of existence, but instead offers a variety of interpretations of

one and the same question (or of examples of the same type of question). Some of these

interpretations make the question meaningful, but there is one intended understanding of

the question Carnap rejects as meaningless – the question only philosophers ask (ESO,

p. 207). I would like to call this understanding the ‘metaphysical’ understanding of the

question, and the question when asked with this understanding in mind the ‘metaphysical

question’. The central question in what follows will be: what is this question and why does

Carnap reject it? My claim is that Carnap’s strategy in ESO is to show that metaphysicial

questions and answers have conditions that are never jointly fulfilled.

1Carnap himself, in a reply in the Schilpp volume, refers to ESO as “the paper on ontology” (Schilpp1963)

2(Bird 1995) suggests that the distinction Carnap makes in ESO is a four-fold, not just a two-fold one.


2.2.1 Carnap’s position in ESO

The immediate purpose of ESO is to allow empiricists in general and Carnap in particular

to talk about abstract objects, such as propositions, without abandoning their empiricist

stance (ESO, p. 206). This is important for Carnap, since he wants to accept science

and mathematics, both of which seem to come with commitments to entities that aren’t

observable, while maintaining his empiricism. Carnap’s strategy is to show that debates

between realists and antirealists about science and mathematics are ill conceived. Unlike a

nominalist, Carnap does not attempt to explain away abstract entities, but instead attempts

to show that the apparent worry for empiricism is based on an ill understood question, and

can hence be safely ignored. The question at issue is of the form “Are there Xs”, where ‘X’

is a sortal.3 Carnap discusses a couple of them: “Are there numbers?”, “Are there (really)

space-time points?”.

Carnap’s discussion has often4 been understood along the following lines: first Carnap

distinguishes internal and external questions, then he goes on to distinguish among the

external questions between theoretical and practical. The question then becomes: which

one of these are the sortal questions mentioned above?

I think this is the wrong approach to ESO, despite the fact that Carnap himself in-

vites it, when he begins his discussion with the distinction between internal and external

questions. Carnap indeed distinguishes internal and external, theoretical and practical (or

non-cognitive) questions, and questions of degree from yes-no questions. It is not the case,

however, that the sortal questions are one of these questions. Instead, Carnap uses his

distinctions in an attempt to make sense of the sortal questions, only to arrive at the conclu-

sion that the questions, as asked by philosophers, do not make sense. The different ways of

thinking about, for example “Are there (really) space-time points?” are not different ques-

tions, but different interpretations of the same question, as becomes very clear in Carnap’s

discussion:5

A question like “Are there (really) space-time points?” is ambiguous. It

3At least the questions he considers in ESO are such that ‘X’ is a sortal. I will come back to this point inmy discussion of Quine’s objections to Carnap’s position.

4See (Bird 1995), for example.5It is unsurprising that Bird should find this passage “puzzling” (Bird 1995, 59).


may be meant as an internal question; then the affirmative answer is, of course,analytic and trivial. Or it may be meant in the external sense: “Shall we intro-duce such and such forms into our language?”; in this case it is not a theoreticalbut a practical question, a matter of decision rather than assertion, and hencethe proposed formulation would be misleading. Or finally, it may be meantin the following sense: “Are our experiences such that the use of the linguisticforms in question will be expedient and fruitful?” This is a theoretical questionof a factual, empirical nature. But it concerns a matter of degree; therefore aformulation in the form “real or not?” would be inadequate. (ESO, p. 213)

Here Carnap gives us what he takes to be the full list of meaningful, unproblematic in-

terpretations available for this question, but unfortunately the interpretation the metaphys-

ically inclined philosopher is after, is not among them. What the metaphysically inclined

philosopher wanted to ask was an external, theoretical, yes-no question, something like:

independent of the framework of space-time points, is it the case that space-time points

exist, or not? – This question is external because appeal to the framework of space-time

points is explicitly denied, it seems to be theoretical in virtue of its form and it purports to

be a yes-no question, in virtue of being a question about the existence of objects, which is

generally not considered to be a matter of degree.6

Why does Carnap reject this question? I will argue that Carnap takes his argument to

have shown that the conditions – external, theoretical, yes/no question – cannot be jointly

fulfilled. Crucially, this argument depends on a connection between evidence dependence

and internal questions, which Carnap only makes explicit in the last part of the paper.

It is clear7 that the metaphysical question cannot be an internal question, particular or

general, because the metaphysically inclined philosopher will be perfectly happy to grant,

for example, that once you have accepted the framework of numbers, the answer to the

question “Are there numbers?” will have to be affirmative. (ESO, p. 209) So the metaphys-

ical question is one to be asked prior to or in any case independent of the acceptance of

the framework. This means, crucially, that no evidence from within the framework can be

brought to bear on the question whether or not numbers exist, as is clear from Carnap’s

6This is true at least for the modern debate. Medieval philosophers might well have recognized degreesof existence or reality, but it seems this is not what’s at issue in the present debate.

7Carnap’s argument to the effect that the metaphysical question is supposed to be all of the above takesthe form of offering a potential metaphysical opponent possible interpretations of his question.


point that, once you accept a statement like “five is a number”, “there are numbers” follows

immediately.8

More interesting is Carnap’s discussion of why such an external question can be un-

derstood meaningfully only as a practical, not a theoretical question. The question the

philosopher wants to ask about existence leads to endless controversy “[a]nd it cannot be

solved because it is framed in a wrong way. To be real in the scientific sense means to be

an element of the system; hence this concept cannot be meaningfully applied to the system

itself.” (ESO, p. 207)

As Susan Haack (Haack 1976) notes, Carnap seems to be begging the question here.

After all, the philosopher might be asking about ‘real’ in the philosophical sense; to claim

that ‘real’ just means ‘being an element of the system’ seems to be just what the philosopher

wants to deny. If this were Carnap’s only objection to the philosopher’s question, he indeed

wouldn’t have much of an argument. I suggest, however, that Carnap has a more interesting

reason for rejecting the claim that external existence questions should be meaningfully

understood as theoretical questions.

I cannot think of any possible evidence that would be regarded as relevantby both philosophers [i.e. nominalists and realists], and therefore, if actuallyfound, would decide the controversy or at least make one of the theses moreprobable than the other. [. . . ] Therefore I feel compelled to regard the exter-nal question as a pseudo-question, until both parties to the controversy offera common interpretation of the question as a cognitive question; this wouldinvolve an indication of possible evidence regarded as relevant by both sides.(ESO p. 219, my emphasis)

A pseudo-question (at least in ESO) is “one disguised in the form of a theoretical ques-

tion while in fact it is non-theoretical;” (ESO, p. 209) This suggests that Carnap’s reason

for rejecting the idea that this external question can be a theoretical one is the lack of agree-

ment between the disputing parties about what would count as possible evidence one way

or the other. The problem with the question as asked is that answering it seems to require a

standard for evidence that is not being provided.

8This line of thought continues to play an important role in contemporary meta-ontology as well (Yablo2000), (Hofweber 2009).


It is tempting to read9 this claim as a skeptical challenge, i.e., a claim that whatever the

answer to these questions may be, it so happens that we do not have access to it. There are

numbers or there are not, but unfortunately we cannot find out.

This, however, is not the right way to understand Carnap. His complaint is not that

no actual evidence has been presented, but that it is not even clear what could count as

evidence. That’s why it is so important that he demands agreement about potential evidence

between the disputing parties. A controversy is only cognitive once the disputing parties

have agreed about what will count as evidence for the respective claims. Unless such

an agreement has been reached, it is always possible for one party to present something

(empirical or logical “data”) as evidence for their claim, and for the other party to accept

the logical or empirical ‘data’ presented, but to deny that these count as evidence for the

opposing claim.10

Note also that Carnap does not require the evidence in question to be conclusive, which

again suggests that Carnap is not making a skeptical point. All he requires is that some

agreement is achieved about which evidence will be regarded as rendering one hypothesis

more likely than the other.

Carnap is suggesting that, in order to understand the question as a theoretical one, we

would have to understand potential answers to it as assertions, and in order to understand

potential answers as assertions, we would have to know what would count as evidence for

(or against) the assertion. If it is unclear that anything counts as possible evidence either

way, no assertion has been made.

The skeptic, by contrast, wants to say that we have no reason to believe or assert that

material things (for example) exist, because we do not have evidence for their existence

(every piece of evidence presented is also compatible with their non-existence in skeptical

scenarios). Realists, on the other hand, want to claim the opposite.

Carnap denies either position the status of an assertion; all that they can meaningfully be

9Barry Stroud suggests such a reading, when he says: “The only reason Carnap has got for declaring theskeptical conclusion meaningless is that the philosopher’s ‘statement’ of the existence of the external worldis neither confirmable nor disconfirmable in experience. But the skeptical philosopher has precisely the samereason for declaring the truth of skepticism – all possible experience is equally compatible with the existenceand with the non-existence of the external world.” (Stroud 1984, 179)

10I will later come back to the question whether we should accept the idea that agreement about possibleevidence is needed for assertion.


are recommendations to accept or not accept a certain framework (that of material things,

for example), but that is not the same as making an existence claim. Denying a statement

the status of an assertion is not the same as denying the statement. This is why Carnap

insists that his position is neither nominalism nor realism.

2.2.2 How is Carnap’s position supposed to work?

So Carnap’s position in ESO is not a form of skepticism. But some parts of the argument

still sound like verificationism, which would seem even worse.11 Why do we need agreed

standards of evidence to make an assertion?

This is not the strong verificationism of the Vienna period, where non-verifiable state-

ments were simply rejected as meaningless. In the argument in ESO, it is not meaning but

assertion that depends on the availability of evidence.12 This, together with the reading of

ESO according to which different interpretations of a certain kind of question are offered,

helps to explain why metaphysical questions appear to be perfectly intelligible. There are

several perfectly fine questions in the vicinity of the problematic metaphysical questions,

like the internal question or the external pragmatic question, which leads us to think that

we understand what the metaphysician is asking for and that an answer to that question is

in fact possible. Carnap’s answer now is that we may seem to understand the metaphysi-

cian’s claim just fine, but it cannot qualify as an assertion. It is instead (insofar as we are

attempting an adequate response to it) something like a proposal to adopt the framework in

question, within which we can then go about making assertions.13

Even so, we might still find the argument too verificationist. Even if Carnap can now

11Worse, since verificationism is often charged with incoherence, or at the very least regarded as highlyimplausible and not well motivated.

12In a broad sense of meaning we may of course wish to include the question whether a statement orutterance counts as an assertion as a question about the meaning of the utterance. The contrast betweenmeaning and assertion here is merely meant to emphasize the difference between Carnap’s earlier views(Carnap 1932), according to which particular words or sentence-structures were meaningless in the sense ofbeing unintelligible, and the view under discussion. For even if we want to include the question of whetheran utterance counts as an assertion as an aspect of its meaning, to say that something is not an assertion is notthe same as saying that it is unintelligible. This is what distinguishes Carnap’s later views from his earlierviews.

13The idea to treat questions of metaphysics and other apparently meaningless subjects as practical ratherthan theoretical can already be found in (Carnap 1934). The questions metaphysics poses require a practicaldecision, not a theoretical assertion.


explain why the metaphysician’s question seems intelligible, it still seems odd to claim

that the answers to the question cannot be assertions, and that the question must hence

be understood as a practical, and not a theoretical question. Why is evidence, or at least

agreement about standards of evidence, required for assertion?

Carnap’s argument in ESO relies on several different components. Metaphysical ques-

tions are pseudo-questions, because they appear to be ‘theoretical’ questions, when the only

available external questions are practical questions. The latter claim relies on what looks

like a verificationist sentiment: to make an assertion we need to have agreed standards

of evidence. There are two ways we might want to reject that verificationist sentiment.

We might reject the verificationist connection between assertion and agreed standards of

evidence, or we might insist that agreed standards of evidence are in fact available even

‘outside of’ frameworks.

But even if Carnap were right to insist that the original metaphysical question was a

pseudo-question in some problematic sense, Carnap will also need an argument why the

metaphysician cannot just opt for the practical question as a re-interpretation of the question

they originally asked. Could Carnap be happy with that? The worry, as we shall see in more

detail below, is that if Carnap’s overall position was supposed to be a form of quietism, then

turning the debate into a practical one does not lead to the quietist’s desired result: to stop

the debate altogether.

Crucial for this latter point of view is his Principle of Tolerance which Carnap recom-

mends at the end of ESO:

Let us grant to those who work in any special field of investigation the freedomto use any form of expression which seems useful to them; the work in the fieldwill sooner or later lead to the elimination of those forms which have no usefulfunction. Let us be cautious in making assertions and critical in examiningthem, but tolerant in permitting linguistic forms. (ESO, p. 221, emphasisoriginal)

Why should we be tolerant? As we shall see in the next section, Tolerance is motivated

by a certain kind of lingering deflationism about what it means to introduce a framework,

a deflationism Quine is going to criticize.


2.3 A closer look at some aspects of Carnap’s view

The verificationist core of Carnap’s position – the link between a sentence’s status as an

assertion and agreement on standards of evidence – will seem sufficiently problematic to

many philosophers today to be reason enough to reject Carnap’s position. But since Car-

nap’s position is really a complicated combination of various different, potentially ques-

tionable ideas, rejecting Carnap’s view on suspicion of verificationism might not be the

most illuminating approach. In this section I want to set aside worries about Carnap’s ver-

ificationism and instead focus on some of the other potentially problematic aspects of his

view. The first of these will be the question whether Carnap can defend the claim that

outside the framework, there is no agreement about evidence. The second will address the

question of whether tolerance can be justified through a certain form of deflationism about

our commitments in adopting frameworks.

2.3.1 Agreement about evidence

One response to Carnap’s argument is to argue that there are standards of evidence available

outside of conceptual frameworks. The metaphysician’s response could be an attempt to

find a way to specify criteria acceptable to both parties for what will count as evidence

supporting or undermining the respective claims.

I suggest that we should think of this as setting up a meta-framework or a meta-

language, in which the acceptability of a given framework can be discussed. Not only

does Carnap not want to deny the possibility of meta-languages, he in fact requires that

such meta-languages be possible for discussing the question of whether or not to adopt

the framework in question. But the metaphysical question cannot be a theoretical question

even in a meta-language, according to Carnap. Can this point be defended?

In such a meta-language we can discuss not only whether or not to adopt a particular

language but also certain features of the language, such as simplicity and fruitfulness. So

why couldn’t the metaphysician appeal to the fruitfulness (and other nice features) of the

language in question and use this as evidence for realism?

However, it would be wrong to describe the situation by saying: “The fact ofthe efficiency of the thing language is confirming evidence for the reality of


the thing world”; we should rather say instead: “This fact makes it advisableto accept the thing language”. (ESO, p. 208)

The move Carnap proposes here is now widely seen as a standard antirealist move: to

reject the idea that ‘pragmatic’ virtues like fruitfulness or simplicity could serve as evidence

for the truth of a theory is a key ingredient to many antirealist arguments and positions in

the philosophy of science. Realists, by contrast, would of course want to claim just that:

the fruitfulness of a particular language is indeed evidence for the existence of the entities

the language talks about.14 It would seem that Carnap is simply rejecting the evidence the

metaphysician is trying to provide for his claim. Once more he seems dangerously close to

begging the question.

(Haack 1976) notes that rejecting the efficiency of a framework as evidence for the

existence of its objects seems to commit Carnap to a strong instrumentalism. But in light

of what Carnap says about the necessary agreement over evidence between the opposing

parties, Carnap might hope to get away with not taking a stance on the matter himself.

Instead he could just point out that as long as instrumentalists, the immediate opponents of

realists, are unwilling to accept the efficiency of a framework as evidence for the reality of

its objects, no agreement about admissible evidence has been reached between the opposing

parties, and hence the conditions for assertion have not been fulfilled.

This still seems less than satisfactory. For one, the mere fact that some party to the

dispute is unwilling to grant certain data as evidence for certain kinds of claims seems

too weak to warrant taking the entire dispute to be non-assertoric.15 It seems that if we

are going to draw any conclusions about the status of the dispute from the fact that some

parties to the dispute don’t acknowledge some of the evidence presented as legitimate, we

should at least find that the parties in question have reason to do so. As soon as we do that,

though, we will have to take a stance on the matter whether we think pragmatic virtues are14And similarly, perhaps, why couldn’t an antirealist use nasty features of the language as evidence against

realism?15A different worry might be who is to count as a legitimate party to the dispute. To the best of my

knowledge, Carnap does not address that question anywhere. I take it he is primarily interested in respondingto particular philosophical debates, in which the two opposing sides are already fairly well defined (realists- instrumentalists, platonists - nominalists). Whether or not this point can be generalized to a full-blowntreatment of any disagreement between any number of parties depends on a more thorough treatment of therequirement for agreement and on what constraints can be placed on who counts as a legitimate party to thedispute (standards of reasonableness, for instance).


evidence or not. The neutrality the Carnapian quietist had been hoping for doesn’t seem

attainable.

2.3.2 Deflationism

Another argument Carnap sometimes seems to invoke to support the idea that we at least

don’t need any external justification for adopting a framework is a certain kind of deflation-

ism about the commitments that come with adopting a framework. The deflationist line of

thought fits better with Carnap’s earlier writings, but hints of it show up in ESO as well.

The main idea is that adopting a framework is ‘ontologically innocent’ in that in accepting

the framework one is not thereby asserting the reality of the objects involved.

In contrast to this view [i.e. that of philosophers who seek ontological justi-fication before the introduction of the framework], we take the position thatthe introduction of the new ways of speaking does not need any theoreticaljustification because it does not imply any assertion of reality. (ESO p. 214)

This passage is important, since it casts some light both on why Carnap thinks that

the questions are merely practical, and why he thinks we should be tolerant. He takes

the acceptance of a new conceptual framework to be matter of introducing ‘new ways of

speaking’. Whether or not we should accept a new way of speaking will indeed seem like

a decision that is up to us, and it may also seem that there isn’t much more to the matter

than preferences, so why not be tolerant? Nothing much seems to hang on it.

This idea goes back to some of Carnap’s earlier ideas developed in The Logical Syntax

of Language (LSL). There he first introduces the idea of Tolerance:

Principle of Tolerance: It is not our business to set up prohibitions, but toarrive at conventions. [. . . ] In logic, there are no morals. Everyone is atliberty to build up his own logic, i.e. his own form of language, as he wishes.All that is required for him is that, if he wishes to discuss it, he must state hismethods clearly, and give syntactical rules instead of philosophical arguments.(LSL, p. 51/2, emphasis original)

The idea there is that because a logic is just a syntactical language, there is no true

logic, and accordingly there is no point in trying to prevent anybody from setting up their


logical frameworks as they please. If adopting a framework is like adopting a logic in

the sense of LSL, then it might seem that just as we were supposed to be tolerant about

logic, we ought to be tolerant about frameworks. Not only are the questions about the

fruitfulness and related qualities of frameworks questions of degree, they are also, and

crucially, questions about the frameworks, not the entities within the framework. This fits

Carnap’s distinction in LSL between two kinds of theoretical questions: “object-questions”

and “logical questions” (Carnap 1937, 277). Logical questions do not directly refer to

objects. So from the standpoint of Syntax, Carnap can say that to accept a framework is to

accept a certain way of speaking, but that any attempt to talk about this way of speaking

from outside of the framework is to talk about the expressions used in the framework, not

the objects these expressions might refer to.

At this point Quine’s first objection arises. Quine’s point is that accepting a conceptual

framework is not just a matter of introducing a new way of speaking, or just introducing

some new symbols. Introducing a new framework means to take on some commitments:

ontological commitments.

Carnap tries to resist this commitment by continuing to speak of frameworks as new

ways of speaking. But this line of response is less satisfactory at the time of ESO than it had

been during the Syntax period. Carnap had since then come to accept Tarski’s conception

of truth and the semantics associated with it. Accordingly the frameworks of ESO are no

longer just the pure calculi Carnap had offered in LSL, and he seems to oscillate between

regarding languages as pure calculi and interpreted systems.16 In other words, it is not quite

clear from ESO alone whether we are to understand accepting a framework as accepting

just a calculus, or also a domain of objects over which the variables used in that language

quantify. (ESO p. 213/4) It seems that Carnap has to accept the latter, which would seem

to force him to accept ontological commitment in a Quinean fashion. Quine had published

On What There Is two years before the publication of ESO and Carnap indeed refers to

it in a footnote (ESO p. 215). In that paper, Quine famously introduces his criterion for

ontological commitment as follows:

[A] more explicit standard whereby to decide what ontology a given theory orform of discourse is committed to: a theory is committed to those and only

16That Carnap felt forced to clarify their status for the reprinted version seems to indicate this.


those entities to which the bound variables of the theory must be capable ofreferring in order that the affirmations made in the theory are true.(Quine 1948,13/4)

So if Carnap is willing to talk about truth in terms of interpretations, then why doesn’t

he want to acknowledge ontological commitments? He is now clearly committed to more

than just uninterpreted symbols.

What’s at issue between Carnap and Quine is what commitments come with accepting

a new conceptual framework, and it seems that once we start to talk about truth, the idea

that the only commitments are syntactical no longer works.

Suppose we accept then, that introducing a new conceptual framework is not just a

matter of introducing a new logical form, but also involves certain commitments; for the

time being let’s take them to be ontological commitments. Does that mean that the question

whether to accept a framework is no longer a practical one?

Not obviously, since whether or not to commit in certain ways still seems like the sort of

thing we might have to make a decision about. But also, not all of Carnap’s argument rests

on the idea that introducing a new framework is just introducing a new way of speaking, that

is, on Carnap’s deflationism. There are general consideration of what it means to introduce

a theoretical concept that don’t rely on this idea. They might help us to understand why

Carnap thinks that there is not going to be any evidence available outside of the framework,

which might justify calling the external question a practical question.

One important reason seems to be that Carnap doesn’t think we know what we are

talking about once we ‘step outside of the framework’.

In fact, however, all that can accurately be said about atoms or the field isimplicitly contained in the physical laws of the theories in question. (ESO, p.211)

As soon as we try to talk about electrons or atoms outside of the framework (that is:

the physical theory) by which they have been introduced, we lack both potential evidence

for any of the claims we want to make about them, as well as any understanding of the

terms themselves. The terms are understood only through the complicated structure of the

framework/theory they are a part of, which contains correspondence rules linking them to


observations and testable claims. The terms are interpreted as part of the framework, not

independent of it. This is why “[t]here is no answer to the question: ‘Exactly what is an

electron?’ ”(Carnap 1966, 234).

The framework through which they are adopted exhausts what can be said about them,

but not only because talking about frameworks means talking about languages. In order

for the question to be a theoretical one, it seems that it must be possible to give an answer

that is an assertion or a theoretical claim. But this is precisely what Carnap denies, for an

assertion must be capable of being backed up with evidence. (ESO, p. 218) The assertions

seem to hold only within their conditions of evidence, i.e. their framework. Taken out of

that context of potential justification or falsification they will cease to be assertions and

become misleading pseudo-statements.

Carnap’s argument that the question is a practical and not a theoretical question does

not depend on his idea that to introduce a new theoretical term or conceptual framework is

just a matter of introducing new forms of speaking. It does, however, still depend on the

idea that to make an assertion, we need to be able to present evidence. The verificationist

flavor of Carnap’s argument is still strong, and we might still feel inclined to reject it for

that reason.

In summary: Carnap’s argument for the idea that the questions asked by the meta-

physician really (upon closer inspection) are pseudo-questions because they are practical

questions in a theoretical guise relies on largely unpersuasive verificationist arguments.

The verificationism comes in whenever Carnap tries to argue that questions from outside

the theoretical frameworks have to be practical. As we have seen, neither the claim that evi-

dence is needed for assertion, nor the claim that there is no evidence outside of frameworks

were especially convincing.

This might seem to settle the score as far as Carnap’s anti-metaphysical stance is con-

cerned, but I think it is worthwhile to pursue independently the question whether there is

something to the idea that ‘metaphysical questions’ are in some sense practical. For even if

Carnap’s arguments to that effect remain unpersuasive, there could be other arguments that

don’t rely on verificationism. In the next section I will detour through a recent attempt to

argue that philosophical or metaphysical positions are something other than assertions or

claims.


2.4 Non-Cognitivism and Quietism

Carnap is not the only empiricist to suggest that debates about philosophical views like

empiricism, realism, and so forth should be understood as practical proposals, not theo-

retical claims. Bas van Fraassen has recently suggested that we should take philosophical

positions to be stances, which he takes to mean attitudes that include elements that aren’t

beliefs. Looking at van Fraassen’s suggestion will help us to separate out the verificationist

from the non-cognitivist elements of Carnap’s quietism.

Van Fraassen (van Fraassen 2002) suggests that empiricism as a philosophical position

should be understood as a stance rather than a thesis about reality. Initially this move serves

as van Fraassen’s answer to the self-application problem commonly raised for empiricist

critics of metaphysics. If an empiricist critic wants to maintain, for example, that only ex-

perience can give us information about reality, the metaphysician might question this very

claim on empiricist grounds, since it appears to be a non-empirical claim about reality.17

But van Fraassen’s argument goes beyond the self-application worry. Taking “only expe-

rience can give us knowledge about reality” as a factual claim means that its contraries

will also have to be treated as factual claims. But that undermines the radical nature of the

empiricist’s criticism of metaphysics. The empiricist wanted to reject the entire game, not

make a move in the game.

By turning empiricism into a stance as opposed to a thesis, van Fraassen hopes to es-

cape this type of objection. The empiricist should not be understood as making a claim

about reality at all, instead her empiricist critique of metaphysics should be understood

as displaying an attitude or taking a stance. Unfortunately van Fraassen is not very clear

about what exactly a stance is: “A philosophical position can consist in a stance (attitude,

commitment, approach, a cluster of such – possibly including some propositional attitudes

such as beliefs as well). Such a stance can of course be expressed, and may involve or pre-

suppose some beliefs as well, but cannot be simply equated with having beliefs or making

assertions about what there is.” (van Fraassen 2002, 47/48).18

17A related way of putting this, which van Fraassen likes to use, is this: Empiricism admires naturalscience for its methodology, which in particular involves not rejecting any factual hypothesis out of hand.But if empiricism, and hence its contraries, are understood as factual claims, holding on to empiricism wouldviolate one of empiricism’s crucial principles: never to reject a factual claim out of hand.

18It has become fairly common in the literature to take ‘stance’ to mean something like ‘epistemic policy’


Taking a stance, according to van Fraassen, is something different from having a belief

or making a claim. This difference is supposed to help the empiricist articulate her position

without being subject to self-refutation worries. If in articulating her empiricism she is not

making a claim, then the position she articulates should not be criticized for failing to live

up to the standards she proposes for claims or assertions.

There is a flat-footed response to this move. The metaphysician might simply decide

not to take the stance proposed. If what the empiricist is doing is to propose that we take a

certain stance, the metaphysician might simply decline to follow the proposal.

On its own, this doesn’t make for an especially good response. After all, it is not

the case that simply ignoring or dismissing a proposal makes for an appropriate response

to proposals in general. Similarly, if we take somebody to display an attitude we don’t

share, we are not in all cases content to just leave it at that, especially not if the attitude in

question is a negative attitude towards a view we ourselves hold. But there is more to the

metaphysician’s reply.

The move from claim to stance, or belief to attitude, tends to suggest a weakening

of the justificatory status of the positions in question. For the claim that a position is

a (mere) stance seems to be a move to require less justification for it: it is not a claim

about reality, so don’t ask me for the kind of evidence you would require for a claim about

reality. That invites the thought that stances can be easily rejected or dismissed. And given

that taking a stance is explicitly characterized as partially evaluative attitude, doesn’t that

mean that taking a stance ultimately is just a matter of which values we happen to hold?

If taking a stance means that no reasons for or against taking a particular stance can be

given, or that taking a stance amounts to nothing more than expressing a preference, the

metaphysician would seem to be right in dismissing the empiricist critique. Even if the

move to stances protects the empiricist from immediate reductio, is there still room for

criticism of metaphysics?

It seems indeed that empiricism has become less attractive. If the only way we can

coherently articulate empiricism is by giving up the idea that it is a claim and acknowledge

that it is at least in part a matter of having certain values and attitudes, empiricism would

(Teller 2004). Van Fraassen does not endorse this understanding of ‘stance’ because he thinks that stancesare not just epistemic, but the notion itself remains vague.(van Fraassen 2004)


seem to be less attractive compared with other philosophical positions, which at least don’t

seem to be self-contradictory even when understood as theses.

But van Fraassen, like other empiricists before him, wants to go further. Not only should

we understand empiricism to be such a stance, but philosophical positions in general are

stances, not claims about reality. Van Fraassen discusses materialism in particular as an

example of a position that is a stance rather than a thesis about reality. Making use of

Hempel’s dilemma,19 he concludes that any given materialism is always bound up with

the state of empirical/scientific knowledge of the time and as such has to be abandoned as

science advances, but that the spirit of materialism survives such changes in content. This

shows, according to van Fraassen, that materialism cannot be regarded as a thesis, but must

be viewed as a stance. As a thesis it is almost certainly false, and it is unclear how it could

be said to be same thesis across time. Van Fraassen concludes that regarding materialism as

a thesis is philosophy in bad faith – the materialist doesn’t realize that she is in fact taking

a stance, displaying an attitude, not making claims about reality.20 Philosophical positions

do not consist in having certain beliefs about what is or is not the case. Instead they involve

both epistemic and evaluative attitudes.

If we find van Fraassen’s discussion convincing21, this would seem to even out the score

somewhat, but it still leaves the empiricist with a number of problems. For one, it seems

that if this line of argument is going to be successful, the empiricist will have to show for

any position, not just materialism, that it in fact turns out to be a stance. It is unclear how

far this line of argument can be extended. Hempel’s dilemma is a problem for philosophers

who want to understand science as metaphysics, but what about the traditionalists, who

take metaphysics to be an independent discipline? If that’s also supposed to be a stance,

19Hempel’s dilemma (Hempel 1965) is typically taken to be a problem for physicalism (or materialism)as a metaphysical thesis in the following way: either the physical is what current physical theories quantifyover/are committed to, in which case physicalism will be (and has been) proven false by novel physicaltheories. Or the physical is whatever our final physical theory will say it is, but since we have no idea whatsuch a theory will be like, and in particular what parts, if any, of our current physical theories will still bevalid in such a final theory, there is not much we can say on the basis of physicalism so understood.

20This seems to echo Carnap’s suggestion that the question I’ve called ‘the metaphysical question’ is apseudo-question because it appears to be a theoretical question when it is in fact a practical question.

21And that’s big if, since it is not clear that the problems resulting from Hempel’s dilemma show thatmaterialism is best interpreted as a stance. The dilemma might just show that materialism in particular is nota particularly plausible philosophical position.


van Fraassen will have to provide a different argument for why their position doesn’t work

as a thesis, but does work as a stance.22 It would be much better for van Fraassen to have a

general argument for the idea that philosophical positions in general should be understand

as stances rather than theses.23

This points to the most serious problem with van Fraassen’s strategy. Suppose a philoso-

pher accepts van Fraassen’s idea that philosophical positions are stances.24 Does that imply

that any stance is acceptable, as long as the philosopher taking it acknowledges that it is a

stance? If so, how much bite does this position have as an anti-metaphysical stance? For it

seems that debates in metaphysics can happily continue, as long as the participants remem-

ber to point out that they are taking stances. And in fact van Fraassen seems to suggest

that holders of different stances (like materialism and empiricism) should be encouraged to

pursue their different stances (van Fraassen 2002, 61). So how is there still any criticism of

metaphysics left, once we’ve moved to the level of stances?

To put it in Carnapian terms: it seems that the kind of debate empiricists like Carnap

and van Fraassen were trying to avoid by calling the questions in that debate ‘practical’

and the answers ‘stances’ or ‘proposals’, can continue. There is no immediate need to be

tolerant, just because the positions are now defended as stances rather than claims.

Anjan Chakravartty (Chakravartty 2007b) argues that the radical criticism of meta-

physics van Fraassen seems to have in mind is not available to the empiricist even after

the move to stances.25 Since the only criteria van Fraassen offers for choosing stances

are some constraints on rationality and whatever sets of values we might hold, any radical

22Alternatively van Fraassen might of course attempt to argue that traditional metaphysics doesn’t evenwork as a stance. But that, too, would require an independent argument.

23Van Fraassen might respond that different philosophical positions are supposed to be contraries, and thathence if one of them can be shown to be a stance, the others must be stances as well. But it is unclear thatthis is strong enough to convince the metaphysician who thinks that the fact that her position doesn’t haveto be formulated as a stance makes it more appealing. Furthermore, it is not clear that the different stancesvan Fraassen outlines are indeed contraries. (Ladyman and Ross 2007) seem to think that empiricism andmaterialism can be combined into a scientistic stance.

24(Ladyman and Ross 2007) seem to be willing to do so. And Jeffrey Poland’s physicalism as a pro-gram (Poland 1994) might be understood as making a similar move. So at least among the science-inspiredmetaphysicians there is a willingness to accept that move.

25Similar worries are raised by (Lipton 2004), who suggests that more than rationality needs to enter as acriterion across stances, for example the consequences of taking a particular stance.


criticism of metaphysics will turn out to be question-begging. Van Fraassen’s own epis-

temic voluntarism26 suggests that different stances are rationally permissible, and so the

criterion for choosing stances that is supposed to hold across stances, rationality, will be

exhausted without having eliminated all but one stance as viable option. This is crucial

for van Fraassen’s voluntarist epistemology. But then how will the empiricist criticize the

metaphysician for taking a different stance? For Chakravarrty that means that the empiricist

will even have to give up criticizing the metaphysician for being ‘in bad faith’. For the idea

that something pursued from a particular stance might lead to ‘Truth’, while rejected by the

empiricist, might be endorsed, as part of their stance, by the metaphysician. So the empiri-

cist’s criticism of metaphysics as philosophy in bad faith will also be question-begging.

Van Fraassen’s answer to this charge remains unsatisfying. In his original proposal he

merely points out that most disputes ‘out in the real world’ involve disputes about values,

later on he insists that whatever relativism one might be committed to at the level of stances

is not ‘disabling’ relativism. It is possible to acknowledge that others have other values

while standing by one’s own. But even if one didn’t think that acknowledging that taking

a stance is evaluative undermines whatever stance one is taking, it still seems clear that the

criticism of metaphysics will be lost. If all the empiricists can say to metaphysicians is that

they don’t like playing their game, it seems that the metaphysicians can continue playing

that game quite undisturbed.

The suspicion is of course that empiricists like Carnap (and perhaps also van Fraassen)

think they have resolved the debate once they have shown the question to be practical

because they think that practical questions are non-cognitive. But as the debate in meta-

ethics has amply demonstrated, one might well hold that the utterances in a particular

debate are non-cognitive, while maintaining that there is a real debate going on. So merely

pointing out that a question is a practical one, or that the positions defended are really

stances is not going to make the debate go away. It might be a different kind of debate, but

it’s still a genuine debate.

If the only criticism empiricists make of metaphysical debates is that metaphysical

26An epistemic voluntarist will not attempt to give a theory of knowledge or cognition, where this isunderstood as an attempt to describe how we actually acquire knowledge. Instead, the voluntarist will focuson the decision to accept or reject positions that present themselves as live options at a given point in anindividual’s life.


questions are really practical questions, not ‘theoretical’ ones, then it seems there is noth-

ing wrong with metaphysics. At most we have to adjust the way we conduct metaphysical

debates slightly, to show our awareness of the practical nature of the questions, but that’s

all. Quietism doesn’t follow from non-cognitivism.

In the next section we will encounter a very different argument to the effect that practi-

cal or pragmatic questions are worth debating. Quine seems to suggest that we have to take

practical or pragmatic differences seriously, since all disagreements are essentially resolved

pragmatically. If there is no distinction to be drawn between theoretical and practical ques-

tions, then it seems that the idea that we could somehow be tolerant about the latter seems

hopeless.

2.5 The practical/epistemic distinction criticized

Quine offers a number of different criticisms of Carnap’s philosophy, the most famous

being his rejection of Carnap’s analytic – synthetic distinction. While that distinction has

a role to play for Carnap’s anti-metaphysical stance in mathematics and logic, I focus here

on an objection which in a way presents a much more serious problem for Carnap’s anti-

metaphysical endeavor, since it applies to his rejection of metaphysical consequences of

physics as well. The objection is that Carnap’s distinction between theoretical and practical

questions cannot be drawn in a non-arbitrary way. This undermines the sharp line Carnap

wants to draw between internal and external questions.

2.5.1 Arbitrary Frameworks

Quine directly criticizes the distinction between internal and external questions in On Car-

nap’s Views On Ontology (Quine 1976). There Quine suggests that Carnap’s distinction

depends on a distinction between subclass and category questions, which in turn is en-

tirely arbitrary and trivial. It is often held that Quine’s criticisms miss the mark.27 In

27Those who argue that Quine’s criticisms are beside the point (Haack 1976), (Bird 1995), point out thatCarnap’s position allows for both internal and external general (or category) questions and it seems that thereis no reason for him to deny the possibility of a particular (or subclass) external question. It so happensthat metaphysical questions tend to take a very general form, so we should not be surprised to find Carnap


this section I explain what I take to be the significance of Quine’s objection for Carnap’s

anti-metaphysical program.

Quine takes Carnap’s position in Empiricism, Semantics and Ontology to be essentially

continuous with some of Carnap’s earlier attempts to criticize metaphysics, in particular

his discussion of universal words (Allworter) in The Logical Syntax of Language (Carnap

1937) (sections 76 and 77).28. There Carnap characterizes a universal word as one that

“expresses a property (or relation) which belongs analytically to all the objects of a genus”

(Carnap 1937, 239). Crucially, in Syntax, these words are dispensable – instead of using

the word, a special kind of variable can be introduced: “If, as is usual in the symbolic

languages, different kinds of variables were used for the different genera of substitution

values, the addition of a universal word would be superfluous.” (Carnap 1937, 295) Philo-

sophical problems seem to arise because of a misuse of universal words in material mode.29

By translating material mode sentences into formal mode, the universal words in question

disappear in favor of a special type of variable. One might, for example, introduce a rule

according to which all greek letters are supposed to be variables standing for numbers.

Thereby one no longer needs a predicate for number in one’s language, since it will be

clear from just looking at the variable-type (e.g., greek letters) that a particular sentence

in the language concerns numbers. But one still needs a predicate for, say, prime-number,

since not all greek letters stand for prime numbers.

This suggestion depends on a strict distinction between genera (numbers) and species

(prime-number), where the former can be transformed away by using a certain type of

variable, whereas the latter are interesting predicates yielding true sentences only for some

values. It is this distinction Quine wants to capture with his subclass – category distinc-

tion.30

primarily concerned with external general questions. But Carnap takes all external questions to be practical,not just general external questions, so the category – subclass distinction has nothing to do with the internal– external distinction. An exception to this is (Alspector-Kelly 2001), who also thinks that Quine’s objectionis significant.

28See also (Friedman 1999)29In The Logical Syntax of Language, Carnap distinguishes between material mode and formal mode

sentences. Sentences in material mode are about objects, whereas sentences in formal mode are about ex-pressions, for example: “Five is a number” is in material mode, as opposed to “‘Five’ is a number-word.”,which is in formal mode.

30This is clear from Quine’s discussion in (Quine 1969c), where he says about universal words: “Theywere a special breed of universally true predicates, ones that are universally true by the sheer meanings of


It begins to appear, then, that Carnap’s dichotomy of questions of existence isa dichotomy between questions of the form “Are there so-and-so’s?” where theso-and-so’s purport to exhaust the range of a particular style of bound variables,and questions of the form “Are there so-and-so’s?” where the so-and-so’s donot purport to exhaust the range of a particular style of bound variables. Letme call the former questions category questions, and the latter ones subclassquestions. I need this new terminology because Carnap’s terms ‘external’ and‘internal’ draw a somewhat different distinction which is derivative from thedistinction between category questions and subclass questions. (Quine 1976,207)

Quine then goes on to argue that any distinction between subclass and category questions

will be arbitrary: “Even the question whether there are classes, or whether there are physi-

cal objects, becomes a subclass question if our language uses a single style of variables to

range over both sorts of entities.” (Quine 1976, 208) So where Carnap hopes to find an im-

portant difference between two kinds of questions, Quine only sees an arbitrary distinction.

This relates to Carnap’s argument in Empiricism, Semantics, and Ontology in the fol-

lowing way. As I argued in the first half of the chapter, Carnap’s distinction between in-

ternal and external questions is a distinction between evidentially confirming claims about

entities within a framework, and non-evidentially accepting certain claims about very ab-

stract entities, as part of accepting a framework. But if there is no way to distinguish, in

principle, between these two kinds of claims, because any claim or question about entities

can be regarded either as a case of talking about a language with a smaller domain (the

prime-numbers), where the predicate (‘prime-number’) is exhaustive, or a language with a

larger domain (the numbers), where it is not, it looks like Carnap has a problem. Carnap

does not want to say that every time we accept a new individual object, we set up a new

framework. Instead he wants to say, at least in some cases, that we’ve found the new object

to exist within a framework.

Suppose we are working within the framework of the “thing world”31 and one day we

their words and no thanks to nature. In his [Carnap’s] later writing this doctrine of universal words takesthe form of a distinction between “internal” questions, in which a theory comes to grips with facts about theworld, and “external” questions, in which people come to grips with the relative merit of theories.” (Quine1969c, 52). Similarly in (Quine 1969b).

31‘Thing world’ is Carnap’s expression for a language the basic concept of which is the notion of aneveryday material object. It is to be understood in contrast to a language that acknowledges as basic onlysense impressions or experiences, and material objects only as derivative.


observe a new comet. We will presumably give the object a name within the framework,

which already contains comets, rather than accepting a whole new framework just to ac-

commodate the newly found comet. Similarly, even with new kinds of objects, a new

species of frogs, say, we do not need a whole new framework, but can accept the new

species within an already accepted framework of animals. If so, what about electrons?

Atoms? Electromagnetic fields? Are these subclasses of objects that we discover, or kinds

of entities we introduce through the adoption of a framework?

So it looks like there is something to the Quinean complaint that Carnap’s distinction

between internal and external questions indeed depends on some distinction between cat-

egory and subclass questions. The problem is that Carnap wants to distinguish between

adopting a framework and with it new entities, and accepting an existence claim within a

framework. The latter is to be based on evidence as it occurs within the framework, whereas

the former cannot be based on such evidence. That seems to force Carnap to draw a sharp

distinction between very general kinds of entities (numbers, things, space-time points) and

less general kinds of entities (prime numbers, comets, frogs), otherwise it seems the distinc-

tion between accepting a framework and accepting a kind of entity within that framework

becomes blurry. But can such a distinction be drawn in a non-arbitrary fashion? This is

Quine’s real challenge to Carnap, and this is the sense in which the internal – external dis-

tinction is derivative of the category – subclass distinction: it depends on the notion of a

framework and the distinction between accepting a framework and its objects vs accepting

objects within a framework. As Quine puts it at the end of On Carnap’s Views on Ontology:

Carnap maintains that ontological questions, and likewise questions of logicalor mathematical principle, are questions not of fact but of choosing a conve-nient conceptual scheme or framework for science; and with this I agree onlyif the same be conceded for every scientific hypothesis.(Quine 1976, 211, myemphasis)32

Quine’s objection, then, is this: There is no point at which we are forced to say, in this

case we are choosing a new framework, but in that case we keep our previous framework,32Similar sentiments are voiced also in Two Dogmas (Quine 1951) and On what there is (Quine 1948). In

fact this seems to have been Quine’s main objection even at the first reading of ESO: on the back of Carnap’sletter accompanying the manuscript of ESO, Quine notes: “Say frameworkhood is a matter of degree, &reconciliation ensues.” (Carnap and Quine 1990, 417)


we are just accepting a new entity within that framework. According to Quine, there is

no principled way to distinguish the two cases. If so, Quine can accept Carnap’s claim

that questions about whether or not to adopt a particular framework are practical, while

maintaining that tolerance is not an option. For to adopt a scientific hypothesis is always a

pragmatic decision, but surely we don’t want to say that we should be tolerant about which

hypotheses we adopt.

Carnap’s anti-metaphysical stance is a form of quietism. To argue for that quietism,

Carnap had to invoke a number of distinctions among kinds of questions (and to some

extent kinds of answers) that he hoped would allow him to avoid metaphysical questions

altogether. As we’ve seen, two of these distinctions, the distinction between internal and

external questions, and the distinction between practical and theoretical questions are not

well founded. Carnap’s idea had been that the internal questions are ‘theoretical’ and ex-

ternal questions are ‘practical’, but this distinction cannot be upheld, since it relies on

verificationism, and on a clear delineation of frameworks, which does not seem possible.

In the last section I will return to the remaining distinction Carnap uses, the distinction

between yes/no questions and questions of degree. But before that, let’s have a look at the

Quinean alternative regarding existence questions.

2.5.2 Existence questions and pragmatism

Quine’s own view is not entirely different from Carnap’s. Like Carnap, Quine is an empiri-

cist. But Quine wants his empiricism to be more holistic and more pragmatic.

Holism in this context is to be contrasted with reductionism. Reductionism, for Quine,

is the idea that each statement can receive confirmation or ‘infirmation’ independent of

others. This idea needs to be rejected in favor of a holistic view of confirmation according

to which “our statements about the external world face the tribunal of sense experience not

individually but only as a corporate body.” (Quine 1951, 41)

Support for this holistic understanding of confirmation is usually taken to come from

Pierre Duhem’s observation that scientific hypotheses are typically tested under a number

of ‘background’ assumptions.(Duhem 1906) If the evidence in a given situation seems to go


against the hypothesis in question, it is open to us to reject one or more of the background

assumptions instead of giving up the hypothesis. This view is often called the “Quine-

Duhem thesis”.

Carnap himself, however, seems to have accepted this Duhemian point as early as The

Logical Syntax of Language.33 So in what way is Carnap’s position less holistic and prag-

matist than Quine’s? There are two differences: Carnap wants to distinguish between log-

ical statements and statements that are empirical – this is the famous analytic-synthetic

distinction; and he wants to distinguish between two ways of changing our beliefs: prag-

matic and epistemic.

Quine of course rejects the analytic – synthetic distinction. What makes his empiricism

holistic and more pragmatic, however, is the rejection of the second distinction. Quine

sometimes presents his view as radical in that he takes every statement, including logical

ones, to be open to revision. But while this is a radical view, it is one shared by Carnap.

Carnap would be perfectly happy to concede that we can even change the logic we use

if we find that it does not suit our purposes. So the claim that no statement is immune to

revision is not in itself something Carnap would disagree with. Carnap, contra Quine, holds

that there are two ways of revising one’s beliefs: epistemic, on the basis of evidence, and

pragmatic, in accordance with our purposes.34 The former happens within what Carnap

calls frameworks, the latter with regard to adopting or rejecting a framework. So while

Carnap wants to distinguish between changes of framework and belief change within a

given framework, for Quine any change is a pragmatic matter, based on considerations of

simplicity of the overall belief system.

Recall that Carnap tied existence questions to internal questions, which he understood

as theoretical questions, in contrast to questions about what frameworks to accept, which

33“Further, it is, in general, impossible to test even a single hypothetical sentence. In the case of a singlesentence of this kind, there are in general no suitable L-consequences of the form of protocol-sentences;hence for the deduction of sentences having the form of protocol-sentences the remaining hypotheses mustalso be used. Thus the test applies, at bottom, not to a single hypothesis but to the whole system of physics asa system of hypotheses (Duhem, Poincare).”(Carnap 1937, 318)

34A further, less dramatic difference is that Quine is very focussed on one particular purpose, organizingour experience, whereas Carnap is willing to accept a number of different purposes that could lead one toadopt different frameworks; for instance if avoidance of contradictions is the main purpose, we might go fora weaker intuitionist logic, but if we want to do physics, we will want to use more powerful classical logic.


were external questions and with pragmatic answers. If for Quine all questions are ulti-

mately pragmatic, what does he say about existence questions and answers?

For Quine, objects in the sciences, but also everyday objects and mythical figures such

as Homeric gods, are posits, introduced by us to simplify and organize our experience.

Epistemologically these [i.e. objects of science and abstract objects] are mythson the same footing with physical objects and gods, neither better nor worseexcept for differences in the degree to which they expedite our dealings withsense experiences.(Quine 1951, 45)

Quine accordingly denies that there are two kinds of existence questions: “Ontological

questions, under this view, are on a par with questions of natural science.”(Quine 1951, 45)

Questions about the existence of numbers are to be treated exactly like questions about the

existence of a particular species of frogs in Scandinavia. Note, though, that Quine is not

suggesting that we discover either that numbers exists or that this species of frogs exists –

instead both are posited, that is, they are introduced for pragmatic reasons.

Because all of our beliefs are held on the basis of pragmatic considerations, and be-

cause all the entities we accept are posited, there is no difference between claims about

frameworks and claims within frameworks, and hence, metaphysics and science. As Quine

himself puts it: “One effect of abandoning them [the two dogmas of empiricism, i.e. the

analytic-synthetic distinction and reductionism] is, as we shall see, a blurring of the sup-

posed boundary between speculative metaphysics and natural science.” (Quine 1951, 20)35

Quine’s holistic, pragmatist empiricism, then, while not advocating metaphysics in the

traditional sense of a priori inquiry into fundamental reality, in effect undermines Carnap’s

attempt to reject the metaphysician’s question, since it can be interpreted as a practical

question (by Carnap’s own lights), and pragmatic reasons are all the reasons we can ever

give for introducing new entities, so there is no reason to reject the metaphysician’s question

just because it’s practical. Quine’s criticism comes from the direction opposite to the one

a traditional metaphysician would take. For the traditional metaphysician would want to35And in contemporary metaphysics, which was deeply influenced by Quine, this move is indeed used to

justify the introduction of entities. David Lewis, for example, writes: “I begin the first chapter by reviewingthe many ways in which systematic philosophy goes more easily if we may presuppose modal realism [i.e.the view that a plurality of possible worlds exists] in our analyses. I take this to be a good reason to think thatmodal realism is true, just as the utility of set theory in mathematics is a good reason to believe that there aresets.” (Lewis 1986a, vii, my emphasis).


claim that far from being pragmatic, even Carnap’s external questions are epistemic in the

sense of receiving evidence-based yes-no answers. Quine’s criticism puts Carnap in an

uncomfortable middle-ground between all-out pragmatism on the one hand, and traditional

metaphysics on the other.

Whether this kind of pragmatist holism can really be regarded as a plausible alternative

will be the subject of the next section. It should be noted however, that even if holism does

not provide a fully adequate alternative, this does not mean Carnap is right. In particular,

Quine’s objections might still undermine Carnap’s program.

2.6 Yes/No questions and pragmatism

Quine’s extreme holism36 has been criticized in a number of different ways in the litera-

ture.37 Given that Quine claims to respect the authority of the natural sciences, the best

strategy to criticize Quine is to show that his holistic position is not sensitive to the practice

of the natural sciences. I will focus on the issue of positing entities, and argue, that we

need to distinguish positing from discovering entities. I will look at two examples: the first

is a case of the prediction of a particular entity of a class of already accepted entities and

the later discovery of that entity; the second is a case of introducing a new kind of entity

for explanatory purposes, introduced as a fundamental principle of a particular theory and

later as explanation of a particular set of experimental results.

These examples will be interesting, since they show that our reasons for accepting a

particular (kind of) entity, even if ultimately ‘pragmatic’ are of more than one type, and

that the form of the question we might ask to which the acceptance of the entity is an

answer, can come in more than one variety. I suggest that this indicates that there is a

distinction of interest to be drawn between yes/no questions and questions of degree, but

that this distinction is not correlated with a distinction between evidential and pragmatic

reasons, as Carnap had suggested.

36Quine may have changed his mind about the extent of his holism in his later philosophy. (Quine 1995)Since I am primarily interested in Quine’s response to Carnap, not Quine’s own philosophy, I shall focus onhis position in “Two Dogmas”.

37See for example (Sober 2000) and (Friedman 1997).


2.6.1 Chemical elements

When Mendeleev developed what we now know as the periodic table of the elements,

he left certain positions open for elements yet to be discovered, and concluded his paper

introducing the periodic table with predictions about the existence of those elements and

some of their properties.38 In 1875, when Lecoq de Boisbaudran detected novel spectral

lines in a particular zinc blend, we want to say that he discovered Gallium. His discovery,

together with that of Scandium 1879 and Germanium 1886, served as confirmation of the

periodic system developed by Mendeleev, because these newly found elements had the

properties predicted by Mendeleev’s table. (Scerri 2007)39 At least prima facie it looks like

Mendeleev may count as ‘positing’ these novel elements, whereas Lecoq de Boisbaudran

and others discovered them.

The pattern of prediction and discovery just described may sound too much like text-

book science, but even if we’ve long come to the conclusion that textbooks don’t tell us the

full story about science, we shouldn’t simply dismiss such examples. What seems to be

going on here is that a kind of entity is initially described on the basis of a new theory (the

periodic table) and later on something is identified as fulfilling the description given on the

basis of the theory. Whether we want to call such a theoretical description a ‘posit’ or not, it

seems that the two steps have to be clearly distinguished. Once such a theoretical descrip-

tion has been given, it is then a yes/no question whether something fulfilling the conditions

has been found. If we accept that the measurements and observations made show that the

conditions have been fulfilled, the theoretically described entity has been discovered.

Are the reasons for accepting the newly discovered entity pragmatic or epistemic?

Quine would of course suggest that they are at least ultimately pragmatic. At the end

of the day, what persuades us to accept the new kind of entity is that it makes our dealing

38“6. The discovery of numerous unknown elements is still to be expected, for instance, of elements similarto Al and Si having atomic weights from 65-75.” (Boorse and Motz 1966, 312)

39As noted in the obituary notice of the Royal Society of London: “How some of his [i.e. Mendeleev’s]predictions have been verified by the discovery of gallium, of scandium and of germanium, which correspondto Mendeleeff’s theoretical elements, ekaluminium, ekarboron, and ekasilicon, is matter of common knowl-edge, and supplies a complete justification of the scheme.” cf. (Boorse and Motz 1966, 302) This shouldnot be taken to mean that the discovery of those elements were the only confirmation Mendeleev’s systemreceived.


with experience overall, and the experiment in particular, easier. This type of analysis sug-

gests that there is no interesting distinction to be drawn in terms of the kinds of reasons we

might have for accepting a particular kind of entity, and as an argument against Carnap’s

position, that might well seem appropriate. But it doesn’t mean that there isn’t a distinction

in the vicinity that is worth drawing: a distinction between a yes/no question and various

other questions that might also lead to the introduction of new kinds of entities. One such

yes/no question was put forward by Mendeleev, and answered by Lecoq de Boisbaudran.

We shall see other kinds of questions that lead to the introduction of entities in the next

example.

There are of course reasons why this case is a particularly clear example of predic-

tion and discovery: that there are chemical elements was uncontroversial even before

Mendeleev, his prediction concerned specific instances of such elements (including pre-

dictions of their physical and chemical properties) that had not yet been isolated. Similarly,

there already existed a number of accepted methods to identify elements, like spectrum

analysis. So it was fairly clear what would count as having discovered the elements in

question. The next example is different in that it involves the introduction of a new entity

for explanatory purposes. The entity in question is the electron.

2.6.2 Charged particles

The situation is somewhat different when the introduction of an altogether new type of

entity is at issue, one that is not already a member of a wider class of already established

entities such as chemical elements. One example of such an entity is the electron. Electrons

were important because they were supposed to be an altogether novel entity: a charged,

massive particle.40 The concept of massive particles was of course familiar, but the idea

that the same entity could be both, the carrier of mass and electric charge, was a novel one.

The idea that electric charge might come in discrete bits can be found in a number of

physical theories in the 19th century. I will focus on H. A. Lorentz’s account. He explicitly

introduces the idea that all charges are carried by particles that also have mass.41 This

40I use ‘massive’ in the sense of contemporary physics, where bodies with a non-zero rest mass are called‘massive’ in contrast to photons, which have zero rest mass.

41See Stein (Stein 1983) This differs from, for example, Weber’s theory of electric atoms, in that Lorentz’s


idea is introduced by Lorentz specifically as first principle of the theory he was going to

develop, not on the basis of any particular evidence, nor was the nature of these particles

to be further explained or elucidated by mechanical models.(McCormmach 1970) Lorentz

initially made no attempt to specify properties of his charged, massive particles, which

would have allowed experimental confirmation. The idea that charge was to be carried by

ordinary matter was a introduced as a first principle of his unifying view, not as an empirical

hypothesis.

This is interesting in contrast to the example of chemical elements. In contrast to the

case of chemical elements discussed above, the properties of the new kind of entity were

not specified, and so Lorentz could not make predictions of the kind Mendeleev had to offer.

It also meant that Thomson’s discovery of the electron was not a confirmation of Lorentz’s

theory, whereas the discovery of the elements Mendeleev predicted was confirmation for

his periodic table. Lorentz’s introduction of the concept of a massive charged particle seems

to match the Quinean idea of positing better than Medeleev’s prediction of new elements.

For Lorentz’s reasons for thinking that there should be such things seem to match the kind

of pragmatic considerations Quine has in mind: by introducing the concept of a massive,

charged particle, Lorentz was able to unify previous theories.

What about J.J. Thomson’s discovery of the electron? Thomson’s cathode-ray experi-

ments are traditionally viewed as discovery or identification of the electron. Nevertheless,

this case differs in some respect from the example discussed above. Unlike in the case of

chemical elements, there were no established methods to isolate subatomic particles, that

is, experiments the particular outcome of which would show that a certain kind of sub-

atomic particle existed. Thomson’s experiments suggested that cathode rays consisted of

negatively charged particles, and he indeed concluded not only that cathode rays consist

of charged particles, but furthermore that these particles were in turn constituents of atoms

and indeed the only such constituents. Thomson’s reasoning may be characterized as an

inference to the best explanation: he performed experiments the results of which were best

explained by the assumption that cathode rays consist of negatively charged particles of

either very small mass or very large charge. Not all of his wide-ranging conclusions turned

particle are simply particles of ordinary, massive matter, that also carry charges, not some special kind ofparticle. See: (McCormmach 1970)


out to be justified – we no longer believe that electrons are the only constituents of atoms.

Thomson’s discovery was not the discovery of the particle Lorentz had predicted. So

the case is not analogous to the case of chemical elements. Both Thomson’s and Lorentz’s

reasoning have ‘pragmatic’ aspects to them, but the purpose is a different one: theoretical

unification in one case, explanation of experimental results in the other.

The questions and answers in the second example are not yes/no questions and answers.

Lorentz’s question was of the form ‘suppose we make the following assumption, how does

this help to unify the theory?’. Thomson’s question, on the other hand, was ‘what do

cathode rays consist of?’. Mendeleev’s prediction, by contrast, raised a yes/no question: do

such and such elements exist, or not?

What the examples suggest is that the reasons we might have for introducing new enti-

ties into our theories can vary considerably. Quine’s suggestion that ultimately all reasons

are pragmatic might be right, and at any rate, it seems plausible that there is at least no

clear distinction to be drawn between pragmatic and other reasons for introducing entities.

There is no distinction between pragmatic and epistemic use of data as evidence.

There is a different distinction, though, and one Carnap also appeals to, between yes/no

questions and other questions. Only some questions in science are of the yes/no variety. We

can introduce entities into theories both in response to yes/no questions, and in response to

other questions, as the examples above suggest. But interestingly, existence questions are

always of the yes/no variety. To ask ‘Do Fs exist?’ is to ask a yes/no question. This is a

distinction not between the reasons we have for answering a question one way or another,

but a distinction of questions according to their semantics, in a broad sense of semantics.

If what we care about in establishing a kind of anti-metaphysical stance is not rejecting

existence claims, but existence questions, then pointing out that existence questions are

yes/no questions might seem like a promising move. For if asking a yes/no question has

presuppositions that other questions don’t have, then it might well turn out that yes/no

questions cannot always be asked, and hence existence questions cannot always be asked.

But such a strategy has to be distinguished clearly from rejecting existence claims, which

as we have seen can be (part of) the answer to questions other than yes/no questions.

What might these presuppositions be? The example of chemical elements suggests that

one such presupposition might be something like ‘belonging to an already accepted kind’.


Asking yes/no questions about particular chemical elements was possible because there

was a background understanding of what chemical elements are, and what it would take to

isolate and demonstrate the existence of particular such elements. Could there have been a

yes/no question about chemical elements as such? Now it seems we are close to the kind of

analysis Carnap is tempted to give. For what would a potential answer to the question look

like? Presumably the answer would initially be something like: of course, gold, hydrogen

and uranium are chemical elements. They exist. So chemical elements exist. We don’t

have to follow Carnap in characterizing these questions as ‘trivial internal questions’, but

it seems these questions and answers are different, and much less interesting, compared to

the question: do elements with such-and-such features exist?

The distinction that seems to be lurking here is the distinction between subclass and

category questions we saw Quine criticize earlier. Something seems to change when we

move from questions about particular chemical elements to questions about chemical ele-

ments in general. What changes are not what kinds of reasons (epistemic or pragmatic) we

can offer for the answers we might give, but how interesting a yes/no question about F is.

2.7 Conclusion

The debate between Carnap and Quine continues to serve as a point of reference in con-

temporary meta-ontology.42 As the discussion in this chapter has made clear, however, the

debate between Carnap and Quine about ontology is a complicated one, involving several

layers of agreement and disagreement. Quine criticizes Carnap’s views on two main points:

first Quine points out that introducing theoretical terms in science and philosophy is more

than just the introduction of a new form of speaking or a new uninterpreted symbol. Since

Carnap accepts Tarski semantics, he can no longer fall back on the idea of a purely syntactic

language he held earlier. Quine’s challenge that Carnap seems ‘ontologically committed’

to abstract entities holds up to scrutiny.

Similarly, Quine’s arguments that several of the distinctions Carnap invokes to distin-

guish scientific questions (the good kind) from metaphysics (the confused kind) are not

sharp enough to provide the kind of radical distinction between science and metaphysics

42See for example (Eklund 2009), (Price 2009).


Carnap had hoped for, are correct. It doesn’t follow from this that we ought to give up

on these distinctions altogether, but the use Carnap wanted to put these distinctions to is

undermined by Quine’s arguments.

Quine’s own radical holism won’t do either, though. This became especially clear in

the question of positing entities. Quine offers the idea of ‘positing entities’ as an alternative

account of what it means to introduce a theoretical concept. Carnap had suggested that to

introduce a theoretical concept just meant to introduce a new linguistic form. That, Quine

suggests, is too deflationary. Quine’s own account, however, isn’t quite correct either. For

on his account it is not possible to distinguish introducing the theoretical concept from dis-

covering the entity that ‘fits the bill’. But this is a distinction we seen to need to describe

what’s going on in science. This suggests that there is an important distinction Quine’s

thoroughgoing holism fails to capture. That distinction is not the distinction between prag-

matic and evidential reasons for introducing entities. It is the distinction that singles out

yes/no questions from other kinds of questions.

What does any of this mean for metaphysics? It is important to keep in mind that what

is criticized as ‘metaphysics’ by Carnap in his later philosophy are debates about realism

and antirealism with respect to mathematics and science. Carnap does not wish to engage

in a realism/antirealism debate about either science or mathematics, but Quine challenges

him: Carnap has to engage in such a debate to sort out the (apparent) conflict between his

acceptance of mathematics and science (including in particular the theoretical, i.e., non-

empirical posits of theses sciences) and his empiricism. Carnap’s attempt to deflate that

commitment fails.

If we take the debate between Carnap and Quine to be about the possibility of realism

and antirealism about science and mathematics, it may seem as though Carnap actually

holds the more metaphysics friendly position. For Carnap allows that there might be philo-

sophical questions over and above questions asked in science, since he says that there is at

least one kind of external question that makes sense: the pragmatic question as to whether

or not to adopt a particular theoretical concept. This may not be the question the meta-

physician had originally wanted to ask, but it seems to be a question we can reasonably ask

as philosophers, and one might see the realist and antirealist as disagreeing on the matter,

one choosing to adopt the framework, the other refraining from it. In such a case Carnap


suggests that we ought to be tolerant, but we don’t have to accept this kind of quietism. We

may still try to persuade the other side to follow our own decision. The mere fact that it is

a practical, not a theoretical question doesn’t mean we cannot argue about it. We can, after

all, argue about decisions. Quinean naturalism, as we shall see in the next chapter, doesn’t

really leave much room for even such practical disagreement, since it just seems to demand

that we accept whatever science tells us to accept.

Chapter 3

The death and resurrection of realismdebates in philosophy of science

3.1 Introduction

In the last chapter we saw how Carnap tried to resist certain kinds of realism and antirealism

debates about mathematics and science. Carnap’s position was not that of an antirealist (a

nominalist or an instrumentalist), but a form of quietism: he didn’t want to engage in the

debates to which nominalism or instrumentalism are possible answers. We also saw that

Carnap’s attempt was unsuccessful.

In this chapter I want to look at some developments in the debate over realism in the

philosophy of science since the end of logical positivism. My approach is not going to

be chronological, I’m interested in the fact that this debate developed in a rather peculiar

fashion compared to realism debates in other areas of philosophy.

Unlike in meta-ethics, or even in the philosophy of mathematics, opposition to realism

in the philosophy of science has largely been offered in the form of skeptical challenges.

This is unusual. In other areas of philosophy, challenges to realism about a discourse

typically either question the truth or truth-aptness of the claims made in that discourse, or

question the existence of the entities the discourse is (apparently) about.1 The challenges

1For discussion of these two aspects of realism debates see (Brock and Mares 2007).

44

CHAPTER 3. REALISM DEBATES IN PHILOSOPHY OF SCIENCE 45

take on the semantics or the ontology of the discourse, not just the epistemology.2

Matters are different in realism debates in the philosophy of science. While the empiri-

cism of the logical positivists involved semantic as well as epistemic claims, contemporary

empiricists do not appeal to semantics to articulate opposition to realism. Since they also

try to avoid outright ontological or metaphysical claims, what remains is the epistemic di-

mension of the resistance to realism. Similarly, the most successful antirealist argument in

the philosophy of science, the pessimistic meta-induction argument, is an epistemic argu-

ment: science has been wrong in the past, so we have no reason to think it is not wrong

today.

In the next section I will look at the two main skeptical challenges to realism in the

philosophy of science. The first such challenge comes from constructive empiricism, the

second from the pessimistic meta-induction argument. Much has been said about both,

and I will not go into every detail.3 My interest is not in defending or objecting to either

challenge, I’m interested articulating the skeptical nature of the challenges, and to some

extent also how they shape the realism that opposes them.

3.2 The skeptical challenges

I will look at two skeptical challenges in this section, one a full-fledged non-realist view,

the other an argument against realism. We shall see that both of them are skeptical in

nature, that is, they call into question the epistemic status of scientific theories, not their

semantics or ontology. Surprisingly, perhaps, it is not entirely clear how the two challenges

are related.

3.2.1 Constructive Empiricism

Bas van Fraassen’s constructive empiricism is on all counts the most developed view in op-

position to realism in the philosophy of science. Like traditional empiricists, van Fraassen

2Sometimes worries about our epistemic access to the putative entities of the discourse can fuel antirealistsemantics of course, for example in the philosophy of mathematics.

3For a detailed discussion of these challenges and a realist response, see (Psillos 1999).


wants to draw a distinction between claims about observable and claims about unobserv-

able entities. Unlike earlier, instrumentalist empiricisms, however, his empiricism does

not draw a semantic (or ontological) distinction between these two types of claims, only

an epistemic one. The difference between claims about unobservable entities and claims

about observable ones is not that the former have a different semantics than the latter, or

that unobservable entities are metaphysically suspect, but only that we are well advised to

withhold belief in unobservable entities.4

To withhold belief in unobservable entities is not to reject the theories that make claims

about such entities, but to take a weaker epistemic stance than belief. Van Fraassen suggests

that we can still accept such theories, but to accept them merely means to believe that they

are empirically adequate, that is, that they get the phenomena right, not that the claims

about unobservable entities are true.5 To repeat, this is not because the semantics of claims

about unobservable entities are in any way different from claims about observables, but

because we should epistemically distinguish between the two. Van Fraassen suggests that

doing so is adequate to science in that it gives an appropriate characterization of the ‘aim

of science’.6

Van Fraassen’s empiricism makes two advances over traditional empiricist positions:

one is that the observable/unobservable distinction is to be drawn by science, not philoso-

phy. It is our scientific theories that determine what counts as observable and what doesn’t.

This helps against the common criticism that the observable/unobservable distinction is

difficult to draw. The second, more important innovation is to make the position entirely an

epistemic one, dropping the instrumentalist semantics of earlier empiricism.

4“But the point stands: even if observability has nothing to do with existence (is, indeed, too anthro-pocentric for that), it may still have much to do with proper epistemic attitude to science.” (van Fraassen1980, 19)

5There is a bit of a debate about just what goes into the attitude of acceptance as opposed to belief.(Rosen1994)

6“Science aims to give us theories which are empirically adequate and acceptance of a theory involves asbelief only that it is empirically adequate. This is the statement of the anti-realist position I advocate; I shallcall it constructive empiricism”.(van Fraassen 1980, 12) Note that van Fraassen articulates antirealism as aview about the ‘aim of science’. Van Fraassen’s position is not meant to be revisionary - ordinary scientificpractice and ‘the aim of science’ are meant to be adequately characterized by van Fraassen’s approach, notchanged. For discussion of what it means to speak of ‘the aim of science’ see (Rosen 1994) and (van Fraassen1994) .


Van Fraassen’s view has been criticized both as not in fact adequate to scientific prac-

tice, and also as insufficiently motivated, even if it were adequate.

It might not be adequate to scientific practice, since scientific practice involves com-

mitments towards abstract entities, which seem on the same footing as unobservable en-

tities insofar as, if the empiricist has epistemic scruples about the latter, he ought to have

epistemic scruples about the former. Moreover, constructive empiricism as a view about

theories is itself committed to abstract entities, viz. theories, and if skepticism about un-

observables is warranted, and theories as abstract entities are unobservables, then it seems

that constructive empiricism is a far reaching skepticism.(Rosen 1994)

The second reason constructive empiricism might not be adequate to scientific practice

is that just like claims about unobservable objects like electrons go beyond the immediate

delivery of experience, so do claims about what’s possible, impossible, or probable.7 Modal

claims, which are ubiquitous in science, are nevertheless suspect from an empiricist’s per-

spective, as van Fraassen himself recognizes.(van Fraassen 1989) But not only does this

provoke some worries about whether skepticism about modal claims might be inadequate

to scientific practice, there is also the worry that constructive empiricism itself relies on

modal claims that go beyond the deliverances of experience. Observable has modal com-

mitments, and even the characterization of observation and experience might require modal

notions like causality.(van Fraassen 1989)

Even aside from these worries about the adequacy of van Fraassen’s view, there is a

general worry about the motivation and force behind the constructive empiricist’s recom-

mendations. Constructive empiricism seems attractive as position about scientific theories

only if there is a strong background commitment to empiricism, the view that only expe-

rience gives us information or knowledge about the world.(Rosen 1994) As we’ve seen in

the previous chapter, van Fraassen defends empiricism only as a stance, not a thesis, and

acknowledges that nobody can be persuaded to accept empiricism on the basis of ratio-

nal argument alone.(van Fraassen 2002) This, I concluded, is not very satisfying, and this

worry remains in the background as a concern about constructive empiricism as a form

7As Gideon Rosen points out, not all modal claims go beyond the deliverances of experience, since atleast some claims about what’s possible do follow from claims about what’s actual: if the moon in fact orbitsthe earth, then that is also a possibility. Similarly for claims about what’s observable: if it has been observed,then it is observable.(Rosen 1994)


of antirealism. For anybody already committed, if only in a weak way, to some kind of

unreflective realism about science will not feel compelled by the possibility of the less

committal stance van Fraassen has on offer, even if that stance were judged to be adequate

to scientific practice.

If constructive empiricism is the most developed antirealist position in the philosophy

of science, antirealism in philosophy of science is a lot less forceful or tempting than an-

tirealist positions in other areas. In other realism/antirealism debates there are of course

worries about the adequacy of antirealism as well, but in many other areas, antirealism

seems better motivated. The problem with antirealism in the philosophy of science, as we

shall see in more detail in sections three and four, is that the entities that make scientific

claims true or false are not especially suspect, metaphysically speaking, and they are only

mildly epistemically suspect. Finally, the semantics of claims containing terms referring

to them seem perfectly straightforward. That makes it much more difficult to motivate

antirealism.

3.2.2 The pessimistic meta-induction argument

If constructive empiricism is the most developed antirealist position in the philosophy of

science, the most developed argument against realism is the so-called pessimistic meta-

induction argument (PMI). One immediate point to notice is that the PMI argument is not

an argument for constructive empiricism, as one might have expected, but only an argument

against realism, or perhaps indeed only an argument against a certain argument for realism.8

The PMI argument is based on ideas derived from Kuhn’s treatment of scientific revo-

lutions9, but was solidified by Larry Laudan.(Laudan 1984) The argument, in outline, goes

like this:

8It is actually not entirely clear how the PMI argument and constructive empiricism should go together.Constructive empiricism builds its argument on the (alleged) difference in epistemic status between what wecan and cannot observe. The PMI, on the other hand, relies on the (allegedly) poor track-record of scientifictheories in the past.

9It is not clear that anything like this argument actually strictly speaking follows from Kuhn’s accountof scientific revolutions (Kuhn 1962), nor that Kuhn himself subscribed to the pessimistic meta-inductionargument.


1. Many theories that were empirically successful in the past were overturned later (inthe sense of being now regarded as false).

2. There is no reason to think that our own successful theories are substantially differentfrom our past theories.

3. Hence we should also expect that many of our own successful theories will turn outto be false.10

Laudan supports the first premise with a number of examples from the history of science

that seem to be cases of successful theories that are now considered false (or of theoretical

terms that are now considered not to refer). Accordingly, much of the debate has focussed

on these cases. Scientific realists have tried to argue that some of the examples are not

actually cases of well-established theories (or part of mature science), or that they weren’t

really successful theories.(Psillos 1999) Both parts of this strategy are meant to undermine

the strength of the induction: if the basis for the pessimistic induction is much smaller than

Laudan and others suggest, then it seems that even if some past theories turn out to be false,

we might still have reason to believe that our current best theories might be true.

One common worry about both aspects of this strategy is whether realists succeed in

spelling out ‘mature’ and ‘successful’ in a way that doesn’t cherry pick theories that we

continue to hold true. Even realists, it seems, have to concede that Fresnel’s theory of light

was empirically successful given its surprise prediction of a bright spot in the center of the

shadow of a disk, and yet it is now considered false.11

Still, even if some cases of overturned successful theories remain, shrinking the in-

duction basis means that the pessimistic meta-induction argument is less persuasive as an

induction. The original intent of the argument was to show that based on the large number

of past theories that turned out to be false, we have not only no reason to believe that our

current most successful theories are true, we indeed have reason to think that many of them

are false. In this way, the PMI could be seen as an argument in favor of antirealism. But if

10An analogous argument is sometimes made for theoretical terms of our theories: theoretical terms ofprevious theories turned out not to refer, and hence we should expect the same to happen to the theoreticalterms in our current theories.

11Realists typically try to deal with this particular case by suggesting that the parts of the theory nowconsidered false were somehow not required for that prediction. This marks a second type of strategy, theselective realism strategy. For discussion see (Psillos 1999).


the induction basis is much smaller than the argument originally made it out to be, then the

PMI fails to convince as a positive argument for antirealism. It no longer gives us reason

to believe that our best theories are likely to be false.

Suppose realists are right to claim that the number of cases of truly successful but

false theories is smaller than the PMI initially suggested. Can the PMI still serve as an

argument against realism, even if it doesn’t establish antirealism? It can, I think, if the

argument is seen as a direct response to the no-miracles argument for realism. The no-

miracles argument suggests that the empirical success of scientific theories would be a

miracle, unless we adopt a realist stance, that is, take those theories to be true or at least

approximately true.(Smart 1963) The PMI argument is a response in that it suggests that

empirical success is not an indicator of truth. The success of theories that later turned out

to be false shows that realism in the sense of taking the theories to be approximately true

is not warranted on the basis of the no-miracles argument. Truth cannot be what explains

the empirical success of theories. As an argument against realism, or more precisely, as

an argument against a certain kind of argument in favor of realism, that would seem to

be enough. For it suggests at least that a theory’s empirical success is not always best

explained by its truth, and that the inference from success to truth is hence not reliable

This fits with the general skeptical strategy many antirealists in the philosophy of sci-

ence tend to follow: realists are seen as wanting to draw stronger conclusions from the

empirical success of scientific theories than appears warranted. For the PMI argument

in particular that means that past examples of successful-but-false theories are meant to

undermine our belief that our current theories are true. For if that belief was based on the

empirical success of our current theories, the PMI can be understood to show that empirical

success can be achieved even by false theories.

So the PMI is strongest when taken as an argument directed against not realism, but

against an argument for realism. All that needs to be shown for this is that there are some

counterexamples, that is, some cases of successful theories that turned out to be false. Of

course there is still a debate to be had about whether a putative counterexample really

amounts to a counterexample, and the realist strategies outlined above can be employed to

reject such putative counterexamples. But it is a lot more difficult to show that there are

no counterexamples to the thought that empirical success is best explained by the truth of


the theories in question, than to attack the grander claim that most past theories are now

considered false, and that there is no difference between our current theories and those past

false ones. I will assume that there are at least a few cases of theories that are very much

like our contemporary theories, and were quite successful, but turned out to be false.

What happens if this has been conceded? It follows that the argument from empirical

success to truth of the theory can no longer be considered sound.12 Does this mean that

realism couldn’t be true? No, because for all that’s been said it might still be that in some,

indeed in many cases, theories are not merely empirically successful, but true. All that has

been shown is that a certain feature of theories, their empirical success, is not in general a

reliable indicator of them also possessing another feature, truth. But it hasn’t been shown

that empirical success is never an indicator of truth.

A realist strategy might be to find a further criterion that separates the empirically suc-

cessful, but false theories from those that are empirically successful and true. The problem

with this strategy is that typical candidates for such criteria, e.g. simplicity and fruitfulness,

are often suspected to be too subjective to be indicators of truth. They are considered ‘mere

pragmatic virtues’, and (hence) not epistemically significant. What is ‘simple’ or ‘fruitful’

seems to depend too heavily on what humans (or indeed perhaps particular groups of hu-

mans) find simple or fruitful, and it seems rather optimistic to hope that there is a close

connection between what humans find simple and fruitful and what is in fact the case.13

This doesn’t show, of course, that theories we find simple and fruitful couldn’t be true, nor

does it show that we shouldn’t accept such theories over complicated ones, but it suggests

that we are not compelled to infer that they are true.

Even if we don’t follow an explicitly realist strategy, we might ask just what position

12Note that this is not a point about the validity (or lack thereof) of abductive reasoning in general, whichis often a central point of disagreement between realists and antirealists in the philosophy of science. Whatis in question here is not the general validity of inferences to the ‘best’ explanation, but its applicability tothe case of empirically successful theories. The antirealist doesn’t have to reject IBE in principle to makethe point that, if truth is not always a suitable explanation for the empirical success of theories becausesuccessful theories are sometimes false, then the particular application of IBE the realist had in mind, namelyto infer from empirical success to truth, is not sound. Since we don’t know for a fact of any particular theorythat is empirically successful and true, but know of a number of theories that are empirically successful butfalse, truth is not the best explanation for empirical success. But to repeat, this point can be made withoutquestioning the validity of IBE in other cases.

13This is the point at which general considerations about the validity of abductive reasoning have theirplace.


we are supposed to adopt in light of the PMI argument. For if we take the PMI argument

to be an argument against a particular argument in favor of realism, and not an argument

in favor of antirealism, we haven’t actually been offered a reason to be antirealists about

scientific theories either. The PMI as I’ve interpreted it just now merely leaves us at a point

where we can’t infer that our current theories are true, without giving us reason to think

that they are false. Nor does the PMI on its own develop much of an alternative view of

science. Unlike van Fraassen who tries to establish a full-blown view of scientific theories,

the PMI doesn’t offer an alternative picture.

Where does this leave us? If the PMI successfully undermines the no-miracles argu-

ment, but without establishing an antirealist position itself, it seems that we find ourselves

in a realism/anti-realism debate without (positive) arguments. The no-miracles argument as

an argument for realism is undermined by the pessimistic meta-induction argument, which

on its own doesn’t yield antirealism. Similarly, van Fraassen’s constructive empiricism was

only supported as a non-realist position compatible with scientific practice, not as the only

way to make sense of scientific practice.

This stalemate between the two arguments has led a number of philosophers to give

up on realism and antirealism debates in the philosophy of science. The feeling that there

is something wrong with the debate that led to this stalemate finds its expression in what

might be called a quietist realism. In the next section I will look a little more closely at this

new quietism.

3.3 A new quietism

The motivation behind Carnap’s quietism, which we looked at in the previous chapter,

was verificationism. What was supposed to be wrong with debates about realism was

that it wasn’t even clear what could count as evidence to decide the matter, which Carnap

thought undermined its status as a genuine debate. The new quietism I will look at in this

section doesn’t come with explicitly verificationist commitments.14 Instead the worry is

14I say explicitly, since Fine at one point (Fine 1986, p.134) does seem to suggest that at least a certainkind of realism debate doesn’t seem to have clear evidential standards, but this is not his main worry aboutthose debates.


perhaps best formulated as a worry about what it means to be a realist (or an antirealist, but

especially a realist) about science.

Arthur Fine, who’s ‘natural ontological attitude’ (NOA) is the most well-known ‘realist

quietism’, opposes a realism that hopes to ‘add on’ anything to what Fine calls the ‘core

position’.

[I]t seems to me that both the realist and the antirealist must toe what I havebeen calling ‘the homely line.’ That is, they must both accept the certifiedresults of science as on par with more homely and familiarly supported claims.That is not to say that one party (or the other) cannot distinguish more fromless well-confirmed claims at home or in science; nor that one cannot single outsome particular mode of inference (such as inference to the best explanation)and worry over its reliability, both at home and away. It is just that one mustmaintain parity. Let us say, then, that both realist and antirealist accept theresults of scientific investigations as ‘true’, on par with more homely truths. (Irealize that some antirealists would rather use a different word, but no matter.)And call this acceptance of scientific truths the “core position”. (Fine 1986,128)

Fine later on famously calls this core position the ‘natural ontological attitude (NOA)’

and suggests that this is the attitude we should take towards science. NOA, according to

Fine, is neither realism nor antirealism, and indeed both realism and antirealism are add-

ons to the core position of NOA that we should simply refuse to make either way. So

far Fine, but what is this core position and just how does it actually relate to realism and

antirealism? Is it really the case that the NOA forms a core position acceptable to both

realists and antirealists?

The realism (and to some extent also the antirealism) Fine had in mind when proposing

NOA were realism/antirealism debates about the appropriate notion of truth, with realists

insisting on a robust, correspondence notion of truth, and antirealists offering various ‘epis-

temic’ construals of truth.15 These are not realism and antirealism as I’ve considered them

here, nor is this how realism and antirealism in the philosophy of science are primarily

understood today.16 If these are indeed the realism and antirealism to which Fine’s own

15Putnam’s (Putnam 1979) and Dummett’s (Dummett 1963) formulations of antirealism are cases in point.In neither case is the idea of an epistemic construal of truth is tied to scientific theories in particular, though.

16There is of course an ongoing debate about truth and realism, but this has largely shifted away from


position is an alternative, then Fine’s position is best understood as a kind of deflationism

about truth.17 I will not pursue the prospects for Fine’s position as an alternative to these

positions, though, since the debate about truth, and how realism debates might be articu-

lated in terms of truth, goes far beyond questions of realism in the philosophy of science.

Moreover, Fine’s own ‘core position’ as stated above doesn’t appear to be a position about

the notion of truth, so it is unclear that we should really understand Fine as attempting to

make a contribution to the debate about realism and truth.(Psillos 1999)

The more interesting question for our purposes is whether Fine’s position is indeed ac-

ceptable to both realists and antirealists in the philosophy of science as I’ve described them

above. And there we find that the only point about which (most) contemporary realists

and antirealists in the philosophy of science agree with each other and with Fine is that the

semantics of ordinary claims and scientific claims are on a par. As we’ve seen, empiri-

cists have moved away from instrumentalist analyses of the semantics of claims about the

unobservable and can hence accept the core position as a view of how to understand the

semantics of ordinary and scientific claims. Realists of course have always insisted on that

anyway, so in that sense realists and antirealists will agree with each other and with Fine.

But as we’ve also seen, there still appears to be some amount of disagreement about the

epistemic status of those two kinds of claims.

Constructive empiricists do not think that ordinary and scientific claims are epistemi-

cally on a par – although they of course draw that distinction as one between claims about

observable and unobservable entities. Here we then have a point of disagreement between

antirealists and the core position. Realists on the other hand do of course accept the core

position, both as a point about the semantics and as a point about the epistemology of

claims about unobservables. That makes the core position at least on the face of it more

realist than antirealist, and it has frequently been interpreted in this way.(Musgrave 1989)

philosophy of science. In the more general debate the main question is whether we could be pluralists abouttruth in a certain way, such that different discourses might require slightly different truth predicates, someof which might indeed be construed epistemically. See (Wright 1992) for discussion. It should be noted,though, that even among friends of pluralism, science is usually not regarded as a paradigmatic discourse forwhich an epistemic truth predicate would be adequate. In the philosophy of science, the focus is very muchon the PMI argument and various realist responses to it. See (Psillos 1999) for discussion of the latter.

17Fine himself prefers to say that he holds a no-theory view of truth, not a deflationism about truth. For thepurposes here that won’t matter.


What, if anything, does the realist add to the core position? Table pounding, is what Fine

seems to think, but that seems a bit unfair.

I think, however, that NOA differs from standard scientific realism in at least one im-

portant way. Not because NOA’s attitude towards truth is deflationist, as Fine seems to

suggest, but because NOA doesn’t take on the skeptical challenge implied in contemporary

antirealisms. NOA’s position is that neither a general skeptical nor a general anti-skeptical

policy will do. Instead we should recognize distinctions between well-confirmed and less

well-confirmed claims, but this attitude is again one we should have both for claims of sci-

ence and ordinary claims. The big difference between standard scientific realism and NOA

is not that the latter is ‘antirealist’ in any way, but that the NOAer is unwilling to meet the

skeptical challenges put forward by antirealism. NOA is not a defense of realism against

skepticism. Standard scientific realism tries to be just that.

This becomes clear when we compare NOA with its recent, more naturalistic cousin,

Penelope Maddy’s ‘Second Philosophy’. Maddy rejects constructive empiricism as too

skeptical, but like Fine she doesn’t attempt to meet the challenge head-on, but instead

thinks it ought to be dismissed.

What’s at work here [in the standard realist’s response to constructive empiri-cism] is the endearing if misguided human tendency to rise to a skeptical chal-lenge. Though van Fraassen isn’t a radical skeptic – he believes he has hands– he puts the Second Philosopher in a familiar position: [. . . ] he asks her tojustify something – in this case her belief in atoms rather than hands – withoutusing any of her tried and true methods for settling such questions. This isa challenge the Second Philosopher grants she cannot meet – justify p with-out using the methods you have developed and honed for such justifications –but she doesn’t take this to undercut her original grounds. [. . . ] The Realist, incontrast, feels that something must be said in reply, which leads to his distortedpicture of scientific method, that is, to his lack of faith in ordinary evidence.(Maddy 2007, 311)

The second philosopher, like the NOAer, refuses to answer the skeptic, but doesn’t

view the persistence of skeptical worries as a threat to her own attitude towards science.

She takes herself to be nevertheless entitled to her own belief in the results of science, at

least to the extent to which these results are justified by the evidence provided by standard

scientific methods.


NOA and Second Philosophy are positions we end up with if we don’t already share the

general empiricist outlook that was required to motivated constructive empiricism, but also

don’t share the desire of the standard scientific realist to answer the skeptic directly. Where

the standard scientific realist aims to show that antirealism (including skeptical antireal-

ism) is unwarranted, NOA is just an attempt to articulate something like a default position

towards claims of science. If Fine is right and what he calls the core position is indeed a

natural or default attitude, then the burden of proof is on the antirealist to show that there

is something wrong with this default view.18 But since the motivation for constructive em-

piricism requires empiricism in general, it is difficult to see how the constructive empiricist

should really go about offering the NOAer or Second Philosopher a compelling alternative.

NOA is a position aimed not at winning the realism debate about science, but ending

it. NOA concedes too much to the realist to be acceptable to the constructive empiricist,

but antirealism proves too weak to undermine NOA. This is true not just for constructive

empiricism, but also for the PMI argument. For NOA, unlike standard scientific realism,

does not offer anything like the no-miracles argument by way of defense of the realism

that comes with NOA.19 And if my suggestions in the previous section are correct, then the

PMI is an argument against realism only insofar as it undermines a particular argument for

realism. Since NOA’s plausibility does not depend on this particular argument in favor of

realism, it is difficult to see how the PMI could undermine NOA itself.

This seems to leave us with what I have called a realist quietism, or, if we already have

some background empiricist convictions, perhaps with constructive empiricism.20 Since

constructive empiricism itself is coupled with the idea that it is not a compulsory view, it

seems that we find ourselves without any actual realism/antirealism debate in the philoso-

phy of science.

18Why on the antirealist and not on the realist? Largely because the realist as we have seen can indeedaccept the core position, whereas the antirealist of the constructive empiricist variety cannot. The constructiveempiricist has to motivate the idea that epistemically, ordinary claims and claims about unobservables are noton a par.

19Fine himself actually criticizes the motivations behind the no-miracles argument as offering the wrongtarget for explanation in the first place, and realism as incapable of offering any explanation to the realquestion.(Fine 1986)

20Assuming, in the latter case, that answers to the difficulties outlined earlier can indeed be found.


3.4 Should we settle for realist quietism?

For some philosophers this may be a welcome result. Dropping realism/antirealism debates

about science might seem like getting rid of a superfluous, outdated part of philosophy of

science. Haven’t we long since understood that instead of developing grand views of the

nature of science in general, we ought to pay close attention to the details of particular de-

velopments and questions in different scientific disciplines? Interesting as these questions

are, I have to confess that I am not content to leave it at that.

3.4.1 Three worries about quietism

One reason is that I worry that this kind of quietism quickly turns into too modest a view

of the role philosophy can play vis-a-vis the sciences. This is especially clear in the case

of Second Philosophy. Second Philosophers are “immersed” in scientific practice and form

their beliefs about observables or unobservables in very much the same way and on the

basis of the same evidence as the practicing scientists.

There are two worries about such an attitude. One is that there might not be all that

much left to do for philosophers that couldn’t be done, and indeed perhaps done better, by

scientists themselves.21 The second worry is that in immersing herself in scientific practice,

the second philosopher is no longer in a position to question the implicit and explicit norms

that govern that very practice. While one might sympathize with the naturalist sentiment

that the idea of first philosophy as a judge over and above science is arrogant and outdated,

one might nevertheless hope that philosophy is special in that it doesn’t accept without

question the norms that govern any particular practice, including in particular various sci-

entific practices. The philosopher’s stance may not be above science, but it might still be

sufficiently outside any particular practice to question the norms that govern that particular

practice. If that’s so, then why couldn’t such a stance include both raising and respond-

ing to skeptical challenges to science in general and certain scientific claims in particular?

One worry about realist quietism is then that it might simply not leave enough room for

21Maddy suggests that philosophers can still be useful both to put together scientific evidence from differentscientific disciplines that is required to answer questions about topics such as perception, and also to reflecton scientific method and the role of logic and mathematics in that method. Overall she envisions a closecollaboration between scientists and philosophers. (Maddy 2007, 115f)


philosophy, including in particular for philosophy of science.

A second reason not to be content with giving up on realism debates is that it seems

to cut off philosophy of science from other areas of philosophy, such as metaphysics and

epistemology. There is a general sentiment that metaphysics, if practiced at all, should be

closely connected to what’s going on in science, although the exact relationship between

science and metaphysics remains debated.22 It would be helpful to have a general view

about how metaphysics should take into account theories in science, but one of the ques-

tions that we would want an answer to in presenting such a view would be what the status of

scientific claims and theories is supposed to be. The unreflective, default realism that comes

with NOA doesn’t seem to be suitable for answering the question of how metaphysicians

should approach results from science in their own attempts to come up with theories. If it

remains possible to understand scientific theories as merely aiming to be empirically ade-

quate, how should a metaphysician approach, say, the standard model of particle physics?

As long as quantum mechanics remains interpretatively underdetermined, how can it be

used to constrain what metaphysicians can say about causation?

None of this is likely to bother the committed quietist, since she will probably also be

unenthused about metaphysics, but it should perhaps bother us as philosophers of science.

For many philosophers of science would like to see metaphysicians pay close attention to

what’s going on in science, but it seems to bridge the gap between the unreflective default

realism and debates in metaphysics about, say, the nature of time, we need to be engaged

in debates about the epistemic and semantic status of propositions in science.

One last reason not to be content to give up on realism/antirealism debates in the phi-

losophy of science is the thought that perhaps the reason the debate ended up as a purely

skeptical debate is due in part to a somewhat accidental historical development, and that

there is something to be disputed after all. Unlike the verificationist quietism we encoun-

tered in the previous chapter, the quietism we’ve arrived at here seems to derive largely from

a certain frustration with the state of the debate, not from an attempt to provide conclusive

arguments against the possibility of such a debate. Given that we’ve seen how difficult it

proved to provide such arguments for verificationists, maybe we shouldn’t let the stalemate

in the debate deter us from pursuing realism debates.

22See for example (Hawley 2006).


With Carnap, Fine, and Maddy we’ve encountered three slightly different attempts to

motivate quietism as a stance towards realism/antirealism debates in the philosophy of

science. Despite their differences, all these quietisms are aimed at dismissing or avoiding a

certain kind of debate altogether, although only Maddy’s view is explicitly directed against

the realism debate as a skeptical debate. Let’s distinguish two kinds of motivations for

such views: in Carnap’s case, the worry is explicitly that there are no standards of evidence

to evaluate the realist’s or antirealist’s claims. Fine seems to have a similar concern, but

doesn’t conclude explicitly that this makes the debate meaningless.23 Maddy, on the other

hand, worries that the skeptical challenge cannot be met either by philosophy or any other

discipline, and suggests that we should (hence?) not let our realism be undermined by it.

We could call these two strands verificationism and naturalism respectively.

Verificationism, as we saw in the previous chapter, is not a basis on which to rest qui-

etism. But naturalism also has shortcomings. Maddy’s argument against various skeptical

challenges seems to come down to her lack of interest in such questions.24 This leaves

the Second Philosopher in a position not too different from that of both the NOAer and,

ironically, also the constructive empiricist. Just as constructive empiricism needs general

empiricism as a background motivation, but lacks the resources to persuade anybody lack-

ing this motivation, Maddy’s Second Philosopher will have difficulties persuading either

the hard-nosed skeptic or the committed scientific realist.

What we have then, it seems, is a serious philosophical impasse, but one that affects

quietism just as much as either realism or antirealism. The realist has nothing to offer that

would persuade the skeptic, the skeptic has nothing to offer that would persuade the NOAer

or Second Philosopher, and the Second Philosopher has nothing to offer to either the realist

who wants to answer the skeptic, or the skeptic who questions realism. If the impasse

between the realist and the skeptic has been the motivation for quietist realism, then the

23Fine writes: “NOA pretends to no resources for settling these disputes [i.e. realism debates], for NOAtake to heart the great lesson of twentieth-century analytic and continental philosophy, namely, that thereare no general methodological or philosophical resources for deciding such things. The mistake common torealism and all the antirealisms alike is their commitment to the existence of such nonexistent resources.”(Fine1986, 134)

24Writing about Carnap’s philosophy Maddy writes “[T]he methods Carnap employs at his higher level areexplicitly drawn from among those of ordinary science: the tools of modern mathematical logic. The catch– what loses the Second Philosopher – is that he applies those ordinary scientific methods in the service of apeculiarly philosophical project that from her perspective has no discernible point.”(Maddy 2007, 82)


fact that quietism finds itself in a similar impasse with regard to the other two views is

going to be especially frustrating to the quietist. Is there a way out?

3.4.2 Moving to the exit

As we’re looking at the three contestants – realists, empiricists, and quietists – unsuccess-

fully trying to move the other parties down their path, we should perhaps ask just why it

is that the debate over realism take this peculiar form in the philosophy of science, and

why quietism continues to be such a tempting option. It will be instructive to compare

the motivations for antirealism in the philosophy of science with those in other areas of

philosophy.

The motivations for antirealism in the philosophy of science can be distinguished into

two broad types: empiricism and a less well defined but strong sentiment that there is

something wrong with any robust conception of realism. The clearest attempt at arguing

against realism from the latter perspective are Putnam’s attempts to articulate internal re-

alism.(Putnam 1979) The problem with the latter motivation is not only that it tends to be

somewhat vague – what exactly is the ‘metaphysical realism’ Putnam holds so little hope

for? – it is also that it is aimed far beyond the question of realism in the philosophy of

science. ‘Metaphysical realism’ goes wrong not just as a view of what scientific theories

may hope to achieve, but any theory we might form, and indeed, anything we might say

in any language we might develop. As I mentioned earlier, there is of course an ongoing

debate about the possibility of different conceptions of truth and global antirealism, but it

has moved quite a bit away from the issue of realism in the philosophy of science.

The global antirealism of the kind Putnam offers can be contrasted with a number of

local realisms and antirealisms, that is, realism and antirealism restricted to particular ar-

eas of discourse such as ethics, mathematics, and science.25 Local realism is interesting

because it can be used to contrast different areas of discourse: one might, for example, be

a realist about science, but an antirealist about ethics. To argue for such a position one has

to show that there is some relevant difference between the two discourses, such that one,

but not the other, warrants realism. For the rest of this chapter I will be concerned with the

25Just how to individuate these discourses is debatable.


question of whether and how local antirealism about science might be a possibility, while

setting aside the question of the possibility of global antirealism.

Local antirealism about some subject matter or discourse D will typically gain support

by contrasting D with some other discourse D’. Typical examples include, of course, con-

trasting ethics with ordinary empirical judgments, or perhaps even with scientific claims,

or contrasting claims about numbers with claims about chairs and tables. These local an-

tirealisms usually point to similarities in the ‘semantic surface structure’ of propositions in

those discourses, which suggests that we should understand what it would be for them to

be true in the same general way (and that often means Tarski semantics, of course). An-

tirealism is then motivated by pointing to epistemic and metaphysical problems with the

‘truth-makers’ for such claims. Possible solutions to these problems include accepting the

metaphysically suspect entities needed as truth-makers, finding different truth-makers, or

reinterpreting the semantics of the claims in such a way as to avoid commitment to those

problematic truth-makers.26

The closest thing to a local antirealism we have in the philosophy of science is con-

structive empiricism: constructive empiricists contrast discourse about unobservables with

discourse about observables. Crucially, though, the problem for claims about unobserv-

ables is a purely epistemic one, and as we have seen, the solution constructive empiricism

offers to this problem also targets exclusively our epistemic attitude to those claims, not

their semantics. This limits the possible ways of articulating antirealism in the philosophy

of science, which, as we saw, leaves antirealism in the philosophy of science rather weak.

In this section I want to look at some of the reasons for this, and suggest that we give up on

taking claims about the unobservable as target for antirealism in the philosophy of science.

Both in meta-ethics and in the philosophy of mathematics, antirealism is commonly

motivated by the thought that the entities in virtue of which the ethical and mathematical

claims are true, if they are true, are metaphysically suspect. In the case of mathematics

because the entities are abstract, in the case of ethics because they seem to have funny

26Talk of ‘metaphysically suspect’ entities is of course somewhat vague. A fair gloss might by that themetaphysical status concerns the identity conditions of the entities in question. Entities with unclear or bizarreidentity conditions are metaphysically more suspect than those for which identity conditions are clear. As weshall see in the next chapter, questions about the identity conditions of quantum objects have been used toargue that these particles are metaphysically suspect.


motivational or normative features that other entities do not possess.27 This is perhaps a

crude way of motivating antirealism in either field, and of course many of the antirealisms

offered in meta-ethics or in the philosophy of mathematics take a more complicated form,

and are often argued for in a more sophisticated manner.

Nevertheless, it is teling that the idea that the entities which ‘make true’ the claims

of science are metaphysically suspect is not even part of the background motivation for

opposition to realism in the philosophy of science, let alone part of the official arguments.28

This is not surprising, really, since the objects of science are typically regarded as among

the metaphysically robust entities. Surely physical entities exist, if anything does, goes the

thought. Other entities, whether mental, abstract or ‘normative’ are typically characterized

as suspect in contrast to the entities of physics. I’m not suggesting that we endorse this

attitude, but in light of its prevalence it is hardly surprising that metaphysical ‘queerness’

of the entities in question, or the love for desert landscapes in general, are not among the

motivations for antirealism in the philosophy of science.

One might object, of course, that in the philosophy of mathematics the worries mo-

tivating antirealism are ultimately epistemic, not metaphysical. It is because it is puz-

zling how we should have knowledge about abstract entities, if mathematical entities are

indeed abstract entities in the relevant sense, that realism about mathematics is problem-

atic.(Benacerraf 1973) If we take this to be the real motivation for antirealism in the philos-

ophy of mathematics, then the worry seems much more similar to at least the motivations

for constructive empiricism.

There is still a difference, though, and one closely related to the metaphysical status

of the entities in question. In the philosophy of mathematics, the worries about the epis-

temic status of mathematical truths and the entities that make mathematical claims true

arise in light of the metaphysical status these entities apparently have. The worries about

unobservable entities that motivate constructive empiricism in the philosophy of science,

on the other hand, arise from concerns about our epistemic limitations as humans. What is

27The latter suggestion was made (notoriously) by John Mackie to motivate his ‘error theory’.(Mackie1977)

28With the exception perhaps of arguments about the ontology of quantum mechanics, where the apparentweirdness of the entities does play a role in the arguments motivating antirealism. We will encounter someof them in the next chapter. But it seems clear that those arguments are peculiar to quantum mechanics andnot applicable to all unobservable entities.


observable to us as humans is left for science to determine, but the thought is that whether

or not something is observable for us makes some kind of important epistemic difference.

This thought motivates the idea that we ought to refrain from ‘believing in’ electrons. What

accounts for the (supposed) epistemic difference between beliefs in electrons and beliefs in

tables has to with what we qua humans can observe, not with a difference in metaphysical

status between electrons and tables. The epistemic difficulties with abstract entities is that

they are causally inert, and hence cannot interact with us or anything else in the world. But

that’s not true of ‘unobservable’ physical entities like electrons. So if electrons are a typical

case of epistemically problematic unobservable entities, then it seems there is still a differ-

ence between the sense in which they might be epistemically problematic and the sense in

which numbers and other mathematical entities might be epistemically problematic.

A related point is that electrons, unlike numbers, are of course experimentally detected.

While experimental evidence for electrons, gravitational waves or black holes may be found

insufficient, or perhaps at least less compelling than the ‘direct’ observational evidence for

mice and tables, it is still experimental evidence. We conduct experiments to find out

whether or not there are gravitational waves. So far we’ve failed to detect them, so we

continue to try. Nothing remotely similar is going on in the case of numbers. Nobody is

working on experimental designs to detect numbers, sets, or any other entities typically

classified as abstract. The likely explanation for this is that there is somehow good reason

to expect that numbers and other abstract entities just aren’t the sort of thing that could

show up in an experimental test. In which case it seems that whatever the epistemic import

of a distinction between what we can observe and what we cannot observe, the distinction

between abstract entities and concrete entities has epistemic important that goes beyond

the import of observable/unobservable.29 Once more we are naturally moved to look at the

29As we saw in the previous chapter, following Quine philosophers have suggested that all empirical evi-dence is indirect, and that in this sense, there is empirical evidence for numbers in not too different a sensefrom the sense in which there is empirical evidence for electrons (and indeed also tables). I have alreadysuggested in the previous chapter that this is a bit too schematic as an account of how entities are introducedinto science. Nothing I’ve said here relies on denying that the success of theories can be a reason to believein the entities they seem to be committed to. What I deny is that there is no relevant difference between theempirical evidence gathered for electrons and the reasons we have to believe in numbers. The standard modelof particle physics posits the existence of a particle called the Higgs Boson. The standard model of particlephysics is very successful, so by Quinean standards, that would be enough to give us reason to believe in theexistence of Higgs Bosons. And yet, physicists at the LHC desperately hope to experimentally detect a Higgs


metaphysical difference to explain the difference in our behavior. Causally inert entities

don’t interact with measuring devices.

None of this is to deny that the realism/antirealism debates in meta-ethics and the phi-

losophy of mathematics have an epistemic dimension. The point is that they have more than

just an epistemic dimension. Worries about the epistemic access we have to the entities in

questions are motivated in part by the metaphysical status of the entities in question. These

debates have an ontological/metaphysical dimension of worrying about the metaphysical

status of the entities in these discourses as well as their epistemic status, and as we shall

see in a moment, they also often have a semantic dimension. By semantic dimension I

mean that positions in these debates often suggest that the characteristic utterances in the

discourse in question are to be understood in such a way as to avoid commitment to the

metaphysically (and perhaps also epistemically) problematic entities.30

The semantic dimension is especially prominent in meta-ethics, where we find posi-

tions like non-cognitivism that give a very complicated analysis of the semantics of ethical

(or perhaps in general normative) statements. But fictionalism in the philosophy of math-

ematics also seems like a semantic answer to the worries that might seem to come with

mathematical discourse.(Yablo 2002)

The closest we come to a semantic version of antirealism in the philosophy of science

is instrumentalism, the view according to which theories in science are not to be taken

literally as (sets of) propositions about what the world is like, but instead as instruments for

generating predictions.

Instrumentalism isn’t just a skeptical challenge - the instrumentalist is trying to sug-

gest that to interpret statements about unobservables as claims or propositions is a mistake.

That’s different from recommending a more cautious epistemic attitude. Verificationism

was of course an especially extreme version of this, declaring all non-empirical statements

Boson. At the same time, even if the usefulness of calculus in science is a reason to believe in functions, no-body even tries to imagine what it would take to find experimental proof of the existence of functions. Whatdoes a Quinean make of this difference? Quine is right to question the idea that only ‘direct’ observation canprovide reasons for believing in entities. But there still seems to be a difference in attitude towards entitiesthat are merely part of a successful theory or model, and entities which have been produced, identified, orregistered empirically.

30This use of semantic dimension is to be distinguished from the question whether, in general, realismdebates should be understood in terms of different conceptions of truth, rather than existence.


meaningless. This proved implausible. But even milder forms of instrumentalism don’t

seem especially well motivated when the target is the ontological commitment of claims

about unobservables. On the surface, claims about unobservable entities seem to have

exactly the same semantic structure as claims about observables. Metaphysically, unob-

servable entities don’t seem to be different from observable entities. So the only ground

for offering a different semantic analysis would seem to be epistemic. And motivating a

different semantics on the basis of a different epistemic status brings us right back to veri-

ficationism, especially if the difference is the difference between what we can observe and

what we cannot observe. Instrumentalism, then, has become a rather unpopular position

in the philosophy of science. Since instrumentalism is a much stronger antirealist posi-

tion, that’s a serious blow to antirealism in general, and the main reason antirealists are

apparently left with only skeptical options.

This leaves realists in the philosophy of science engaged in an argument with the skep-

tic, and as Maddy points out, that leads realists to appeal to inflated ideas about the success

of the scientific method. Since the challenge is skeptical, realists feel compelled to argue

that scientific theories are approximately true, that inference to the best explanation is a

valid form of inference, and that science is never really wrong.31 We don’t have to follow

Maddy and Fine in rejecting the skeptical debates altogether, but we may want to question

whether the skeptical debate is really what we want a realism/antirealism debate to be like.

Realism debates in other areas are not typically skeptical debates, and like other skeptical

debates, the structure of the debate in the philosophy of science makes both realism and

antirealism look like unattractive options. So if we want a realism debate in the philoso-

phy of science, we should try to find a way to make it a non-skeptical (or not exclusively

skeptical) debate.

I suggest that this can be achieved by changing what we take to be the target of our

realism debate. Traditionally, realism debates have at least two different targets: the exis-

tence of the entities in question, or the truth of a class of contested statements. Neither of

these quite works as a target for realism debates in the philosophy of science, though. Even

the constructive empiricist doesn’t outright deny the existence of unobservable entities,31What I have in mind here are the various arguments by realists to the effect that what’s responsible for

the empirical success of scientific theory is retained across theory change, or that theories that are false werenot part of mature science.


but remains agnostic. This allows constructive empiricists to sell their position as non-

metaphysical: they only make an epistemic recommendation, not a claim about what there

is (or isn’t). Similarly, if we take as the class of problematic statements the claims about

unobservables, then we once more seem limited to epistemic criticisms only. Constructive

empiricists explicitly don’t offer an irrealist analysis of the truth of these statements. Since

the problematic class of statements is characterized in epistemic terms – statements about

unobservables – attributing a different semantics to these statements will naturally try to

move from epistemic difficulties with these claims to semantic differences, and that smacks

too much of verificationism to have broad appeal.

In order for the debate to get its bite back, we need to find positions that include a

semantic dimension, positions that allow a semantic reconstruction of the contested claims

that would avoid problematic seeming commitments. But as we have seen, the ontological

commitments of scientific claims do not seem problematic in a principled way, unlike the

commitments of ethical or mathematical claims. So as long as we take the entities scientific

theories are about to be the target of our realism/antirealism debate, we might well feel that

not only is a semantic reconstruction of them avoiding such commitment not very plausible,

it is also ill-motivated.

What we are looking for, then, are claims made in science that do seem problematic,

and that could do with an analysis that avoids problematic commitments. In the next section

I will look at a kind of claim, or a type of commitment, that seem to come with scientific

theories, but which aren’t in the first instance ontological.

3.5 Changing the target - modal claims

Claims about unobservable entities are not the only kinds of claims empiricists (and other

antirealists, but empiricists in particular) are worried about. Any kind of claim that goes be-

yond the deliverances of experience is problematic from an empiricist’s perspective. That

includes, in particular, various modal claims many of which also go beyond what’s observ-

able.32 As claims that go beyond the deliverances of experience, modal claims have already

been one of the targets of (empiricist) antirealism in the philosophy of science. Empiricists

32Excluding again inferences to what is possible from what is actual.


traditionally reject the idea that there are laws of nature or necessary connections in the

world. Their complaint typically is that such connections or laws cannot be established

empirically. That is, the empiricist’s complaint is once again an epistemic one.33

Unlike in the case of ontological commitments or existence claims, however, I think

that modal claims allow for a much wider range of worries, metaphysical, epistemic, and

semantic (or logical) and are hence a better target for realism/antirealism debates in the

philosophy of science than the former. Modal claims are problematic not just because we

might worry about how we know them to be true, but also because we worry about what it is

that makes them true. To put the point slightly differently: modal claims seem epistemically

problematic in a way very similar to the way in which mathematical claims are epistem-

ically problematic. They both go in principle beyond the deliverances of experience, not

just in virtue of a (contingent) limitation of human perceptual capacities. That means they

are still an appropriate target for empiricist critics, but they are something not only em-

piricists should worry about. If modal claims have truth-makers, those truth-makers seem

potentially just as epistemically problematic as the potential truth-makers of mathematical

claims, if they have truth-makers. With respect to modal claims, then, it seems we have

a situation that promises to be much more analogous to the kinds of realism/antirealism

worries we see in other debates.

That modal claims are frequently regarded as problematic suggests that they might be

an appropriate target for philosophical debate, but are they really an appropriate target for

realism debates in the philosophy of science?

There is of course a general philosophical debate about the status of modality.34 In that

debate, the main concern is with what is called ‘absolute modality’, that is unrestricted

or unconditional modality. The worry about absolute modality is both metaphysical and

epistemic. Metaphysical in that there is a serious question about what the source of such

absolute modality might be. The epistemic worry is how we could have knowledge about

modal facts. How does this general debate apply to philosophy of science?

33The most famous empiricist critic of modal claims is of course Hume. Hume’s own projectivist analysisof necessary connections has fallen out of favor at least as far as laws of nature are concerned, but there arestill a number of empiricist critics of the idea of laws of nature.(van Fraassen 1989)

34For a recent collection of papers, see (Hale and Hoffmann 2010).


For philosophers of science the question will not just be about absolute modality. In-

stead the worry will be motivated by particular kinds of modal claims found in the sciences:

impossibility and possibility claims, laws of nature, and perhaps also certain kinds of prob-

ability claims. It is far from obvious that all of these should receive the same kind of

analysis. Nor should we assume that they are cases of ‘absolute’ modalities, and if they

aren’t, the question arises how they relate to absolute modalities.

Given that the candidates for truth-makers of modal claims are often regarded with sus-

picion, modal claims inside and outside of science are a potential target for non-propositional

semantic analysis. Just like in realism debates in other areas, proposals to analyze the se-

mantics of modal claims in a non-propositional fashion have to offer not only reasons to

question the semantics of modal claims, but also proposals for how to understand modal

claims instead.

In the philosophy of science, this task is especially interesting, since one of the stan-

dards for correctness or adequacy of such an alternative semantic analysis will be that it be

compatible with the way modal claims occur in the sciences. Since it is not only difficult

to determine, which claims are (covertly) modal claims, but also frequently contested what

role modal claims play in particular scientific theories, this is a point at which philosophers

of science will find themselves with more than enough work to do. What are the realist and

antirealist options that philosophers of science might go in for with respect to (particular

kinds of) modal claims?35

I’m suggesting that a first possibility for antirealism arises in asking whether modal

claims, or perhaps modal claims of a particular kind, have truth values at all. If that is de-

nied, antirealism of the instrumentalist variety seems to be an option. If scientific theories

(or at any rate: some of them) contain modal claims of the kind under discussion, then the

challenge for this kind of antirealism is to explain why these claims occur in scientific theo-

ries at all, that is, they need to explain their function for scientific theorizing, and of course,

just like non-cognitivists in meta-ethics they also need to explain why ethical statements

35I formulate these options as answers to questions about particular kinds of modal claims, since muchof the literature from which to draw examples for the positions in questions is concerned with one type ofmodal claims, laws of nature. Since I’m interested in modal claims more broadly construed, and since I don’twant to assume that all modal claims that might occur in scientific practice are of the same kind, I will herearticulate the possible positions as though they applied only to one kind of modal claim.


Do modal claims (of

type X) have truth values?

Are at least some of them (likely to be)

true?

What makes

them true?

Instrumentalism/Fictionalism

Error Theory.

Modal Facts.

Non-modal facts.

Are these facts

objective?

Humean Supervenience Lewis' style.

Humeanism Hume style

no

no

no

yes

yes

yes

What is the source of the modality?

The World.Logic/

Semantics.

What is the source of that modality?

The World. Us.

Figure 3.1: Realist and Antirealist options


on the surface look like regular propositional statements, even if their function is quite dif-

ferent. Traditional instrumentalism of course assigns theories, and laws in particular, the

function of being instruments for prediction. More recently Ronald Giere has suggested

that laws are rules for constructing models (Giere 1995).

If a certain kind of modal claims are taken to be capable of being true, a serious question

arises as to whether they are at least likely to be true. If we thought there was a principled

reason against taking laws to be true, we would end up with an error theory of laws. Nancy

Cartwright famously argued that we shouldn’t understand laws of nature to be statements,

since if we did, we would have to regard them as false (Cartwright 1983). It is not entirely

clear, though, how we should understand laws instead on Cartwright’s view, and it isn’t

clear that her ‘error theory’ of laws really holds up to scrutiny.36

What if we take a certain kind of modal claims to be not only capable of being true

or false, but also think that at least some of the modal claims we make in science are

not only capable of being true, but in fact true? Then the task is to find truth-makers for

these claims. Just as in the case of mathematics finding such truth-makers will have to be

balanced against our ability to know true from false modal claims.

It is at this point that the discussion in philosophy of science is interestingly different

from the discussion of modality in general, and also from discussions in meta-ethics, which

otherwise boasts a similarly wide range of realist/antirealist options.

The most straightforwardly realist option in response to the question ‘what makes

modal claims true?’ are modal facts. Just as claims about unobservables are made true

by facts about unobservables, modal claims are made true by modal facts. The question

that immediately arises from this, however, is of course the question we encountered above

as a central question of the general debate about modality: what is the source of modality

in modal facts? This question is analogous to the question ‘what is the source of norma-

tivity?’ for normative claims. In the philosophy of science, some answers to the question

of the source of the modality include necessitation relations (Armstrong 1983), but also,

36Interestingly, neither Giere nor Cartwright seem to want to ban modalities for science altogether. InsteadCartwright has recently proposed to understand modalities in terms of dispositions or capacities in nature(Cartwright 1994). This is somewhat surprising, since avoiding the attribution of unexplained capacitiesto entities was one of the main motivations for moving to the idea of laws of nature in the first place. SoCartwright’s proposal can really be seen as a move away from a certain way of understanding modalitytoward a different one, but it seems the modalities we are left with remain ‘real’ modalities.


more recently, dispositions or capacities (Ellis 2001). All of these locate the source of the

modality ‘in the world’, the modality is ‘real’ modality.

By contrast it has sometimes been suggested that all modality is really logical.37 This

pushes the question further back, though: what is the source of logical modality? Once

more, one answer is ‘the world’ – but this time there is a small advantage, since it is

plausible to suggest that logical necessity can be understood as true in all possible worlds,

and logical possibility can be understood as ‘true in at least some world’. This might

require realism about possible worlds38, but at least it avoids commitment to any special,

irreducible capacities or dispositions in the world. A second option, preferred by some

logical positivists, is of course to locate the source of the modality not in the world, but

‘in us’. The immediate next question for this option is of course just how we are able to

‘generate’ this modality.

Going back to the initial question ‘what makes modal claims true?’, there are also less

obvious answers. These answers suggest that what makes modal claims true are themselves

non-modal facts. We can then distinguish views that take those views to be subjective –

facts about our expectations or associative tendencies, for examples – from views that take

them to be objective. Versions of the former are offered by Alfred Ayer (Ayer 1956), and

perhaps also by Hume himself.39 The latter is famously held by David Lewis. While the

subjectivist camp is clearly antirealist, David Lewis calls himself a modal realist, because

he takes the truth-makers of modal claims to be possible worlds, about which he is a real-

ist.(Lewis 1986a)

Of course the positions I’ve charted here could have been charted differently. But what

I hope is clear even now is that taking modal claims as the target opens the space for a

much richer realism/antirealism debate than the skeptical debate we looked at earlier in the

chapter. Add to this that scientific theories might well be found to contain more than one

type of modal claim, and it should be clear that there is a lot of work to be done here.

In contrast with the skeptical debate I looked at in the first half of this chapter, the

debate about modal claims seems to offer a large number of realisms and antirealisms.

37This was of course a key logical positivist move.38This is of course debatable.39In light of the serious debate about how to interpret Hume’s views on causation it would surely be a

mistake to try to fit his views neatly into one of these boxes.


These positions involve semantic, epistemic, and ontological components, which allows

for a broader range of positions. Moreover, since in the case of modality, like in the case

of morality, there is a serious question about its source, subjectivism and objectivism are

among the possible realisms and antirealisms.

One might object that isolating modal claims is no easier than drawing a distinction

between what’s observable and unobservable. Indeed, some might argue, modal commit-

ments are intertwined with other commitments made by ostensibly non-modal claims. So

haven’t we just cast the net too widely? How will we know which claims to look at? These

questions will be addressed, somewhat indirectly, in the next two chapters. I argue that the

question whether and how modal claims might be intertwined with non-modal claims is an

important question we need to settle - and one of the important connecting points between

metaphysics and philosophy of science. In chapter four I argue that there is a variety of

different modalities in science.

3.6 Conclusion

I’ve argued in this chapter that the realism debate in philosophy of science as it has been

understood for the past couple of decades is at heart a skeptical debate. Moreover, it is

a skeptical debate that a number of philosophers are tempted to leave behind altogether.

Against this I have suggested that we should change the target of our realism/antirealism

debate in the philosophy of science from claims about unobservable entities to the modal

commitments that come with scientific theories. Without exploring any of the options in

detail I have suggested that such a shift in target opens up the space for a variety of realisms

and antirealism in the philosophy of science, and it offers a connection point for other areas

of philosophy such as metaphysics to interact with philosophers of science.

Chapter 4

Where does structural realism fit?

4.1 Introduction

In the last chapter I suggested that we should change the target of realism/antirealism de-

bates in the philosophy of science from claims about unobservable entities to modal claims.

In the past couple of years the debate about realism in the philosophy of science has seen

some new developments, in connection with the introduction of ‘structural realism’, or

more recently, ‘ontic structural realism’. In this chapter I will argue that structural realism

can be made sense of when we take the approach to realism debates in the philosophy of

science I suggested in the previous chapter.

Structural realism often1 comes with the promise of being an intermediate position

between realism and antirealism in the philosophy of science, but structural realism as it is

currently presented wants to be more than just a position in the realism/antirealism debate in

philosophy of science. It is also at times offered as a general metaphysical thesis: the claim

that ‘structure is all there is (fundamentally)’ (Ladyman and Ross 2007). At other times

it is offered as an account of the ontology of particular scientific theories like quantum

mechanics and quantum field theory (Lyre 2004). And finally it is sometimes offered as

the idea that structure should somehow replace not a particular ontology, but ontology in

general (French 1998). Often these different approaches are not clearly distinguished.

Even this brief overview shows that it is not only unclear what the thesis of structural

1See (Worrall 1989)

73

CHAPTER 4. WHERE DOES STRUCTURAL REALISM FIT? 74

realism is, but that it is actually not even clear what kind of a thesis structural realism is.

What debate(s) is structural realism contributing to? What does it say in these different de-

bates? These are the main questions of this chapter. I distinguish four different ‘branches’

of structural realism, each of which is not just a different particular claim, but even a dif-

ferent kind of thesis, that is, the different branches of structural realism are not all answers

to the same question, but answers to different questions. The diagram on page three illus-

trates the various branches of structural realism, together with the main arguments offered

in support of each branch.

I argue that structural realism actually doesn’t fare especially well on any single one the

different branches, that is, compared to its local competitors on any given branch, structural

realism is not doing better than the competitors on that branch, and in some cases it is doing

worse. Despite this, and despite the difficulties in stating what exactly structural realism

is, structural realism continues to receive a lot of attention.2 In the final section of this

chapter I suggest that we can make sense of structural realism if we see it as contributing

to yet a different debate, a debate that is part of reshaping what it means to be a realist or

an antirealist about science along the lines I suggested in the previous chapter.

2For recent contributions see (Ainsworth 2010), (French 2010).


an alternative to full blown

scientific realism

a thesis in metaphysics

a claim about the ontology of

particular theories

a means to avoiding

metaphysical underdetermination a position within

a new debate of realism in phil

science

What kind of thesis is structural realism?

only structure is preserved

across theory change

all there is (fundamentally)

is structurein QM and QFT the fundamental entities are not

individuals

to avoid underdetermination, change ontological

basis

metaphysical underdetermination

no miracles argument

pessimistic meta-induction

argument

quantum statistics argument

naturalism

all we can know from scientific theories is

structure

realism debates in phil science concern the modal

claims made in science

only structure is preserved across

theory changeRamseyfication

+

Figure 4.1: What is the thesis of structural realism?


4.2 Structural realism as an alternative to standard scien-

tific realism

In this section I look at the ancestor of the current forms of structural realism, John Wor-

rall’s proposal of structural realism as an intermediate position in the realism/antirealism

debate in philosophy of science. It will be worth having this view on the table, since many

of the recent versions of structural realism continue to invoke Worrall’s arguments as mo-

tivating and supporting their own versions of structural realism.

I will do three things in this section. First I show how Worrall’s view was supposed to

fit into the traditional realism/antirealism debate in the philosophy of science I discussed

in the previous chapter. Second I argue that structural realism is not doing better than its

immediate local competitors in that debate. Finally I suggest, following the discussion in

the previous chapter, that this debate is really an instance of the problem of skepticism, and

not a typical realism/antirealism debate.

John Worrall (Worrall 1989) proposes structural realism as a position that can provide

an answer to the two main opposing arguments in the scientific realism debate: the no-

miracles argument and the pessimistic meta-induction argument. As we saw in the previous

chapter, the point of debate between the two arguments is whether we have reason to believe

that empirically successful theories are true. One realist strategy in response to the PMI

argument is to isolate one element or aspect of empirically successful theories and to show

that this element or aspect is in fact retained even across theory change. The suggestion

then is that it was really this, still correct, element of the old theory that accounted for its

empirical success. This strategy might be called the ‘selective realism’ strategy.

John Worrall’s structural realism can be seen as a peculiar version of this strategy of

defending realism against the pessimistic meta-induction argument. Worrall’s suggestion

is that structural realism can give something to each side of the debate. In response to the

pessimistic meta-induction, structural realists acknowledge that theory change comes with

a change of ‘ontology’ or ‘content’, which explains the sense in which older theories can

now be seen as false. In response to the no-miracles argument, structural realists point to

‘structure’ as something that is preserved across theory change and suggest that scientific

theories are successful, because they ‘get the structure right’, even if they are wrong about


other things, like ‘content’ or ‘nature’ (Worrall 1989). In this way, Worrall’s structural real-

ism hopes to be an intermediate position between realism and antirealism, acknowledging

the force of both arguments. The idea is that what is identified as crucial for the success of

scientific theories is structure, and that structure is in fact preserved across theory change.3

The biggest challenge for this kind of structural realism is to articulate the distinction

between ‘structure’ and ‘content’ in a non-question-begging way.4 It obviously won’t do to

simply define ‘structure’ as whatever is in fact preserved across theory change, yet it seems

difficult to give a general account of structure that would actually allow us to say for our

current theories, which parts we can trust in, and which parts we cannot trust in.5

What Worrall had in mind as structure were the equations of the theory, e.g., Maxwell’s

equations. An important question for structural realism of this kind is then going to be:

why are equations special, and what do they say? In particular, in what sense can we claim

that the equations have been preserved? For we don’t just treat the equations as strings of

mathematical symbols, so it seems the mathematical formulae always also come with some

kind of ‘interpretation’.6 If so, then structural realists will have to say why it is fair to ignore

the different interpretations given to ‘Maxwell’s equations’ for the purpose of articulating

structural realism.7

Worrall’s brand of structural realism is now commonly called ‘epistemic structural re-

alism’, since it suggests that structure is just all we can know through scientific theories,

whereas the nature of things remains hidden from us (Ladyman 1998). Worrall’s formula-

tion of structural realism relies on a contrast between ‘structure’ and ‘nature’. The latter is3In contrast to standard scientific realists, structural realists do not require that all theoretical term of

successful theories refer.4See (Psillos 1999) for discussion. This is a problem we will encounter over and over in the discussion of

structural realism.5If we see the realists as responding to a skeptical challenge, it also won’t do to say that structure is

preserved, but we cannot know which parts of our current theory are structure. In the dialectic with theskeptic that wouldn’t do us any good, since it still seems that we don’t know which parts of the theory totrust, and which parts not to trust. So the epistemic caution that the skeptic urges still seems to have force.

6For discussions of different ‘interpretations’ of electromagnetism in classical and in quantum physics see(Hesse 1970) and (Belot 1998).

7Some structural realists have recently appealed to the semantic view of theories to make the notion ofstructure more precise. What is preserved across theory change is not the entire structure of the old theory,but instead partial structures. Within the semantic approach to theories, the partial continuity of structureacross theory change can be seen in ‘partial isomorphisms’ that hold between different successive theories.(French 2000) (French and Ladyman 2003b).


needed both to articulate what structure might be (in contrast to nature) and also to explain

the sense in which older theories got things wrong: they got things wrong with respect

to nature. Many contemporary structural realists dislike this epistemic humility, and pre-

fer what is often called ‘ontic structural realism’ (OSR), which in slogan form states that

‘structure is all there is’.8 OSR rejects the idea that there is somehow something forever

beyond our epistemic grasp as insufficiently naturalistic. More specifically, some ontic

structural realists suggest that at least in the case of quantum physics we have good reason

to think that structure really is all there is.9 Even if we prefer not to side with the ontic

structural realists, we might agree with their concern that epistemic structural realism leads

to humility. More committed scientific realists will be inclined to hope that science can

give us knowledge about more than structure.

Worrall’s structural realism is not the only position that claims this kind of middle

ground between realism and anti-realism. Entity realism claims that what is preserved

across theory change are not ‘structures’, but entities.10 An entity like the electron has

been a stable ingredient to any theory of matter ever since its discovery, and even across

the switch from classical to quantum mechanics. Entity realism is another form of se-

lective realism, that is a realism that allows for some parts of scientific theories to be

false, while leaving others intact, thereby accommodating both the no-miracles and the

pessimistic meta-induction argument. Van Fraassen has presented an intermediate kind of

empiricism in response, arguing that what is preserved are phenomena, which possess both

‘content’ and ‘structure’ (van Fraassen 2006). Anjan Chakravartty (Chakravartty 2007a)

has recently proposed a view he calls semirealism, which is yet another kind of selective

realism.11

Even if structural realists can overcome the difficulties of articulating what is meant

by structure, they face stiff competition from other selective realisms, and are certainly no

longer the only game in town when it comes to ‘intermediate’ positions. This plurality of

selective realisms shouldn’t come as a surprise. The question that has shaped this particular

8I will discuss the details of OSR in the next section.9Again, I will come back to some of the arguments based on quantum physics below.

10Entity realism has been proposed by (Cartwright 1983) and (Hacking 1983).11Since Chakravartty’s realism is a realism about concrete structures, some structural realists might be

inclined to count his view as a version of structural realism, although Chakravartty seems to see it as adistinguishing mark between structural realism and his own semirealism. (Chakravartty 2007a, chapter 3)


debate has been: ‘what is retained across theory change?’, and while the extreme positions,

‘everything’ and ‘nothing’, are not especially plausible, it seems that a plausible answer

might be, ‘bits and pieces of the old theory/ies’. The research done in response to the pes-

simistic meta-induction argument suggests overall that there is both change and continuity

in cases of theory change. The plurality of selective realisms seems to indicate, if anything,

that there might not be a single way of characterizing what is continuous across theory

changes. What is retained in any given theory change (or development) may very well be

best approached by looking at the particular development itself.

Structural realism in this early incarnation is a contribution to what has long been con-

sidered the ‘realism-antirealism’ debate in philosophy of science. This debate, as discussed

in the previous chapter, has been shaped by two arguments pulling in opposite directions,

the pessimistic meta-induction argument and the no-miracles argument. As I suggested in

the previous chapter, perhaps it is time to evaluate the idea that these are the questions we

should focus on in the debate about realism and antirealism in philosophy of science. If the

challenge to scientific realism is really a skeptical challenge, then we may want to look for

a different way of articulating what realism and antirealism amount to in the philosophy of

science. This is not to say that we should ignore the skeptical challenge, it is just to say that

we should take it for what it is: a challenge to the confidence we place in any particular

scientific claim. We are well-advised to be epistemically cautious, but that’s not the same

as being an antirealist.

4.3 Structural Realism as a thesis in metaphysics

In this section I look at a more recent way of understanding structural realism, as a general

ontological or metaphysical thesis. I will begin by sketching out the thesis of ontic struc-

tural realism, and the motivating idea, that all we can know through scientific theories is

structure. Then I will look in some detail at the arguments that are offered to support the

controversial idea that our scientific knowledge is somehow ‘structural’. Afterwards I look

at the dialectic between ontic structural realists and their local competitors on this branch

of structural realism, epistemic structural realists. I conclude that the dialectic with ESR

does not provide a good argument for OSR.


4.3.1 What is ontic structural realism?

Structural realism of this variety is usually called ontic structural realism, to distinguish

it from structural realism as the thesis in philosophy of science discussed in the previous

section12, but also to distinguish it from epistemic structural realism, its main competitor

on this branch. Ontic Structural Realism has been proposed by a number of philosophers,

but in particular by James Ladyman.

What is ontic structural realism?

Ontic Structural Realism (OSR) is the view that the world has an objectivemodal structure that is ontologically fundamental, in the sense of not super-vening on the intrinsic properties of a set of individuals. According to OSR,even the identity and individuality of objects depends on the relational struc-ture of the world. Hence, a first approximation to our metaphysics is: ‘Thereare no things. Structure is all there is.’ (Ladyman and Ross 2007, 130)

The first sentence suggests that ontic structural realists want to make an addition to

our ontology: ‘structure’ (more specifically: modal structure, see below) is part of our

fundamental ontology, it doesn’t just supervene on intrinsic properties. The second sen-

tence seems to go further, in that it suggests that not only is this additional modal structure

part of our fundamental ontology, but ‘objects’, whatever those might be, stand in some

kind of dependence-relationship to the structure. The last sentence, which contrary to the

suggestive ‘hence’ does not appear to follow from the first two, seems to go even further:

‘structure’ is not only a part of fundamental ontology, it is the only constituent of funda-

mental ontology. What seemed like an extension of our fundamental ontology really turns

out to be a replacement of what some philosophers might have taken to be our fundamental

ontology.

This eliminativist flavor seems strengthened a few lines down when Ladyman and

Ross suggest that their position “consists in the conjunction of eliminativism about self-

subsistent individuals, the view that relational structure is ontologically fundamental, and

structural realism (interpreted as the claim that science describes the objective modal struc-

ture of the world).”(Ladyman and Ross 2007, 130) One thing we need to check, then, is12Ontic structural realists often seem to see themselves as nevertheless continuing the tradition of Worrall,

and invoke the idea that structural realism is an intermediate position that can answer both the no-miraclesargument and the pessimistic meta-induction.(Ladyman and Ross 2007)


whether this conjunction forms a coherent position. As this last passage indicates, OSR

is a metaphysical thesis which combines eliminativism about objects with realism about

structures, together with a thesis in epistemology or philosophy of science: science gives

us knowledge of these modal structures. Indeed, as we shall see at various points below, it

is because science gives us knowledge of these structures, but not of objects, that we should

adopt the metaphysical theses involved in OSR.

Who are they arguing against? In the passages so far, ‘structure’ has been variously

contrasted with ‘intrinsic properties’, ‘objects’, ‘things’, and ‘self-subsistent individuals’.

In other passages, ontic structural realists turn against “traditional metaphysics”, which

they characterize as follows:

(i) There are individuals in spacetime whose existence is independent ofeach other. Facts about the identity and diversity of these individuals are de-termined independently of their relations to each other. (ii) Each has someproperties that are intrinsic to it. (iii) The relations between individuals otherthan their spatio-temporal relations supervene on the intrinsic properties ofthe relata (Humean supervenience). (iv) PII [the Principle of the Identity ofIndiscernables] is true, so there are some properties (perhaps including spatio-temporal properties) that distinguish each thing from every other thing, and theidentity and individuality of physical objects can be accounted for in purelyqualitative terms.(Ladyman and Ross 2007, 151)

This is the standard metaphysical view to which ontic structural realism is meant to be an

alternative. But what is presented here as standard is too specific. While (i)-(iv) may go

well together, it is not clear that all metaphysicians are committed to all four of them.13

Not all ‘traditional’ metaphysicians are committed to individual entities as basic ontology.

Nor does everyone accept the PII. One could have an ontology of properties (understood,

perhaps, as universals), or something less traditional, like facts, events, or processes. This

raises the question which of the positions in more ‘mainstream’ metaphysics should count

as clear opponents of ontic structural realism, and which might count as sympathetic to, or

at least compatible with, structural realism.

It will be helpful to treat Lewis’ ‘Humean Supervenience’ view as the main competitor

13Ladyman now acknowledges that this characterization does not do justice to all contemporary metaphys-ical views out there. (Ladyman, personal communication.)


to the structural realists’ metaphysical picture.14

Humean supervenience is named in honor of the great denier of necessary con-nections. It is the doctrine that all there is to the world is a vast mosaic of localmatters of particular fact, just one little thing and then another. (But it is no partof the thesis that these local matters are mental.) We have geometry: a systemof external relations of spatio-temporal distance between points. Maybe pointsof spacetime itself, maybe point-sized bits of matter or aether or fields, maybeboth. And at those points we have local qualities: perfectly natural intrinsicproperties which need nothing bigger than a point at which to be instantiated.For short: we have arrangements of qualities. And that is all. There is no dif-ference without difference in the arrangement of qualities. All else superveneson that. (Lewis 1986b, x)

Humean Supervenience is not committed to all of (i)-(iv) above, notice for instance

that Lewis is not, strictly speaking, committed to the existence of space-time points, let

alone space-time points as individuals. But what Lewis is clearly committed to is the idea

that structures, in particular modal structures, supervene on intrinsic properties, and this

is of course one of the things ontic structural realists are going to deny.15 This is at least

one clear point of disagreement. Indeed, if we accept the stronger statements of ontic

structural realism, we might interpret ontic structural realism as an attempt to invert the

supervenience relationship suggested by Lewis.16

What is the argument for ontic structural realism? Ontic structural realists invoke a

whole potpourri of arguments in favor of their view, including the alleged middle ground

structural realists can take in the realism/antirealism debate in the philosophy of science and

arguments from particular theories in physics. In this section I will look at a much more

narrow argument for ontic structural realism, which takes epistemic structural realism as

its starting point, and offers ontic structural realism as an improvement over the epistemic

version of structural realism.17

14OSR was initially offered as a position in the philosophy of science, as an improvement over Worrall’sstructural realism (see previous section). As a metaphysical thesis, however, it makes more sense to treat it asa competitor to Humean Supervenience, since Worrall’s view does not come with a particular metaphysicalclaim.

15Note though, that Lewis might accept spatio-temporal structure as independent. This point won’t berelevant to the discussion here, though.

16Just how, if at all, such an ‘inversion’ might be intelligible is the subject of the final section of this paper.17I will discuss arguments from particular theories in the section after this one.


The argument proceeds by way of a dialectic with the epistemic structural realist (ESR),

who holds that all we can know form science is structure. There are actually three potential

targets: Worrall’s view, as discussed in the previous section, Lewis’ Ramseyian Humility

view (Lewis 2009), and van Fraassen’s recent empiricist structuralism (van Fraassen 2008).

I will take Lewis to be the main competitor in this section, since neither Worrall nor van

Fraassen are interested in defending a particular metaphysical thesis. Once the idea that all

we can know from science is structure has been conceded, ontic structural realists insist that

naturalism requires that we accept that structure is all there is (Ladyman and Ross 2007,

131). Before we look at the dialectic between ESR and OSR, let’s look at the arguments

that are supposed to support the rather radical sounding claim that all we know through

science is structure. OSR’s strategy is to let ESR do the work of showing that all we can

know is structure, which leaves OSR only with the task of showing that OSR fares better

than ESR. This means, of course, that no matter how the dialectic between OSR and ESR

plays out, as an argument for OSR it is only as good as the arguments for ESR were to

begin with.

4.3.2 ESR as a response to PMI

One argument for the idea that all we know through science is structure comes from Wor-

rall’s answer to the traditional realism/anti-realism debate in philosophy of science. As we

saw in the last section, Worrall’s suggestion had been that to answer the pessimistic meta-

induction argument, structural realists concede that claims about ‘the nature of things’ or

‘ontology’ are lost in theory change, and can hence not be regarded as knowledge. This

loss is contrasted with continuity of ‘structure’ across theory change. If structure is retained

across theory change, that means that our theories give us knowledge of structure, since our

claims to knowledge of that structure aren’t undermined by the pessimistic meta-induction.

The pessimistic meta-induction undermines our claims to knowledge of ‘natures’ or ‘on-

tology’, since the latter aren’t preserved across theory change, and we hence have to admit

that just as we regard previous accounts of the ‘nature’ or ‘ontology’ of things as false, we

have to be skeptical of such claims as made by our own theories today. But since in the

past ‘structure’ has been preserved, there is no reason to be similarly skeptical about the


knowledge of ‘structure’ our theories might provide. Hence we should conclude that all we

can know from theories is ‘structure’.

As we saw in the previous section, there are a number of problems with ESR and with

this argument for ESR. In particular, there doesn’t seem to be a clear sense of ‘structure’

or what it is contrasted with, and accordingly the claim that all we can know is structure

itself remains unclear. Furthermore, as we also saw in the previous section, ESR is not the

only competitor when it comes to answering the conflict between PMI and the no-miracles

argument, which means that to argue for OSR from ESR, ontic structural realists will have

to show that OSR is a better response, not just than ESR, but also than any of the other

local competitors of ESR in this debate. Unfortunately, in the dialectic with Worrall’s ESR,

ontic structural realists typically rely on ESR as emerging as the winner of that earlier

debate (Ladyman and Ross 2007).

4.3.3 Theories and Ramseyfication

The second argument an epistemic structural realist might use to argue for the claim that

all we can know is structure has very little to do with worries about the pessimistic meta-

induction argument. Traditionally, structural realism was articulated using what is know

as Ramseyfication of a theory. Ramseyfication is often associated with the ‘syntactic ap-

proach’ to theories, which has fallen out of favor in philosophy of science, although it is

still used as a method in metaphysics. For our purposes it is important because Ramsey-

fication can give rise to an argument for the idea that all we can know through scientific

theories is structure.

To formulate the Ramsey sentence of a theory, we begin with a formalization of a theory

T, Π(O1,O2, . . .On; T1,T2, . . .Tm), in which we distinguish between O-terms and T-terms.18

We get the Ramsey sentence of the theory by replacing the T-terms by bound variables,

and quantifying existentially: ∃t1, t2, . . . tmΠ(O1,O2, . . . ,On; t1, t2, . . . , tm). Crucially, the

Ramsey sentence entails all the observable consequences of the theory, but without direct

18Traditionally O-terms were supposed to be observational terms, whereas T-Terms where meant to betheoretical. Following Lewis (Lewis 1970), many philosophers prefer to understand O-terms simply as the‘old’ vocabulary, that is, vocabulary that is somehow supposed to be understood independent of the theory.


reference to theoretical entities.19 Ramseyfication is not a way of denying the existence of

such entities, though, since we obviously still existentially quantify over them.

Ramseyfication was suggested as a way of articulating a form of structural realism by

Grover Maxwell (Maxwell 1970). From the Ramsey sentence it seems that all we can

know about the non-observable entities of a theory, those that theoretical terms refer to,

are formal or logical properties. The Ramsey sentence only tells us that there are entities

that stand in certain kinds of formal or logical relations, but it tells us nothing about the

entities themselves, other than that they stand in these relations. This only holds for the

theoretical part of the language, since of course the observable consequences are entailed by

the Ramsey sentence, and the observable terms are not replaced in the way the theoretical

terms are replaced.

Ramseyfication, then, is one way of articulating structural realism. Unobservable enti-

ties exist, but our knowledge of them is merely structural, that is, formal. Ramseyfication

will then seem to support the idea that all we can know through our theories is structure.

‘Structure’ in this context then means the (first-order) logical structure of a theory.

Ever since Russell first proposed this kind of approach, however, a fundamental worry

has been raised for this kind of account, known in the literature as ‘the Newman problem’

after M.A.H. Newman who first raised this worry (Newman 1928). The basic idea is this:

once ramseyfied, it seems that it is too easy for a theory to be true. Since nothing but

the most basic formal properties for the relations are specified, all kinds of collections of

objects might make a ramseyfied theory true. All that is required is that there is the right

number of them, that is, that the domain has the right cardinality for the theory.20 This

seems to place too few constraints on what it takes for a theory to be true.21

This objection is generally taken to undermine the use of Ramseyfication for the pur-

poses of articulating structural realism as a form of scientific realism (Cei and French 2006).

While the syntactic approach to theories has few supporters in the philosophy of science,

Ramseyfication is still alive and kicking as part of a metaphysical research program known

19The idea that direct reference to unobservables was something to be avoided of course had to do with thekind of empiricism many philosophers of science held in the first half of the 20th century. The details don’tmatter to the discussion here.

20This worry was forcefully made again by (Demopoulos and Friedman 1985), who expanded the problemsomewhat.

21There is an echo of this in Putnam’s paradox. See (Demopoulos and Friedman 1985) for discussion.


as the Canberra Plan (Braddon-Mitchell and Nola 2009). There the Ramsey sentence is no

longer taken to quantify over individual objects, but instead over the properties referred to

by the central T-terms of the theory we (philosophers) develop of a certain area of discourse.

The role-occupants in such a Ramsey sentence are not individual entities, but properties.

This use of Ramseyfication in metaphysics has sparked new interest in the Newman

problem and surrounding discussions. Initially Lewis defended the idea against Putnam’s

paradox (Putnam 1979) (which can be interpreted as a problem closely related to the New-

man problem) by introducing the idea of perfectly natural properties (Lewis 1984). That

means that the ‘roles’ in the Ramsey sentence are occupied by intrinsic properties, and in-

deed, for fundamental theories, the occupants will only be members of the elite class of

perfectly natural properties. This way Lewis seeks to avoid the triviality of adequacy worry

raised for Ramseyfication approaches. Theories that take only perfectly natural properties

to serve as role occupants in the Ramsey sentence are better than those that take gerryman-

dered properties. This way not just any theory will be true. The only theory that will be

true is the one that picks out the perfectly natural properties.

But now a different problem arises. Since the perfectly natural properties, by definition,

are intrinsic and not individuated just by the place they have in the Ramsey sentence, and

since Lewis concedes the possibility that there might be perfectly natural properties in other

possible worlds that aren’t part of the actual world, there is now a problem of multiple

realizability. Different perfectly natural properties might be playing a particular role in the

Ramsey sentence, but we wouldn’t know it.

Lewis turns this line of thought into a more thoroughgoing thesis of humility: “Quite

generally, to the extent that we know of the properties of things only as role-occupants,

we have not yet identified those properties. No amount of knowledge about what roles

are occupied will tell us which properties occupy which roles.” (Lewis 2009, 204) Lewis

then goes on to argue, using Ramseyfication, that we do in fact only know fundamental

properties22 as role-occupants, and hence we haven’t really identified them. Fundamental

properties are what Lewis called earlier ‘perfectly natural properties’. T-terms of a theory,

22A more controversial part of Lewis’ paper extends the argument to cover non-fundamental properties aswell. I will focus on fundamental properties here.


and only T-terms, denote fundamental properties. Given this and the idea of Ramseyfica-

tion, his argument goes like this:

Suppose [the Ramsey sentence of a theory] does indeed have multiple possiblerealizations, but only one of them is the actual realization. Then no possibleobservation can tell us which is actual, because whichever one is actual, theRamsey sentence will be true. There is indeed a true contingent propositionabout which of the possible realizations is actual, but we can never gain ev-idence for this proposition, and so can never know it. If there are multiplepossible realizations, Humility follows.(Lewis 2009, 207)

What is Lewis up to here? As soon as we ramseyfy a theory, the theory can be realized

by all sorts of different properties, provided they stand in the appropriate formal relations,

or in Lewis’ terms, provided they occupy certain kinds of roles. But those roles can be

occupied by different fundamental properties, provided they are of the appropriate category

(or syntactic type). Since the Ramsey sentence has all the same observable consequences

as the original theory, in the case of a complete theory this will mean that the Ramsey

sentence already entails anything that might possibly count as evidence for it, and hence

nothing remains to distinguish different realizations of the Ramsey sentence, that is, nothing

remains to distinguish a case where fundamental property P1 occupies role R from a case

where that role is occupied by fundamental property P2. Hence we only know fundamental

properties as role occupants, and accordingly, by the earlier considerations, we haven’t

‘identified’ them.

The argument crucially relies on the idea of quidditism: the fundamental properties are

not individuated by the role they play. Only with this assumption does it make sense to sug-

gest that there is a sense in which two different fundamental properties could occupy the

same role. If properties were individuated by their roles, then the two properties could not

differ. One of the reasons to accept this idea, though, is the first problem for the Ramseyfi-

cation approach. Lewis needed perfectly natural properties to be role occupants to get rid

of the triviality of adequacy problem. The cost seems to be that there is always something

beyond the reach of our knowledge.


4.3.4 Summary

We’ve now seen two very different arguments to the effect that all we can know through

scientific theories is structure. One of these arguments relied on the skeptical challenge to

our knowledge claims in science as voiced by the PMI, the other resulted from a certain

account of what it means to give a scientific theory: Ramseyfication. Since OSR only enters

into the debate after ESR has established its central contention that all we can know from

science is structure, OSR’s success is hostage to the success of these arguments.

Interestingly, some ontic structural realists have attacked both Worrall’s original re-

sponse to the PMI, and Ramseyfication (Cei and French 2006). They suggest that both

Worrall’s informal notion of structure, and the Ramsey-notion of structure have to be re-

placed by a different notion of structure, one emerging from the semantic approach to

theories. While the work that has been done to articulate the notion of ‘structure’ in a more

formal fashion by using the semantic approach has been helpful in illustrating what Wor-

rall’s original idea might amount to, switching to the semantic approach is not by itself an

argument in favor of structural realism of any variety. The semantic approach to theories

is used both by scientific realists like Ronald Giere, and by empiricists like van Fraassen.

At best, structural realists might find the semantic approach a useful tool to articulate their

position more clearly. But vis-a-vis the Ramseyfication approach, using the semantic ap-

proach instead might actually create a bit of a dialectical problem for the ontic structural

realists. For Ramseyfication, as we just saw, leads to the idea that all we can know from

science is structure. That claim is important for the ontic structural realist in arguing for

their position, even if they dislike the epistemic humility that seems to come with it. Unless

using the semantic view has similar consequences, the ontic structural realist cannot rely

on that claim as a launching pad for their own argument.

4.3.5 The dialectic between ESR and OSR

Epistemic Structural Realism (ESR), at least insofar as it is motivated from the considera-

tions discussed in section 4.3.3., seems to be a way of defending Humean Supervenience

as a metaphysical position. Humean Supervenience was the idea that only intrinsic proper-

ties subvene, everything else supervenes on them. It is thereby in direct competition with


OSR, which wants to claim that Humean Supervenience is false. OSR, as we shall see in

this section, sometimes tries to turn the considerations used by ESR to motivate the idea of

humility against Humean Supervenience.

Ontic structural realism rejects the humility involved in rescuing Humean Superve-

nience. It is not the case that there is something intrinsic or otherwise non-structural that

we are missing out on. If our theories don’t tell us about anything but structure, the ontic

structural realist insists that the conclusion we ought to draw from this is that structure is

all there is. The reason our theories only tell us about structure is that (ultimately) all there

is is structure. Indeed, the ontic structural realist insists that naturalism requires us to go

with this latter explanation, rather than with the postulation of forever unknown intrinsic

natures of things.(Ladyman and Ross 2007, 154) The intrinsic properties ESR introduces

as unknowables are just idle wheels.

How good are the reasons behind the impressive sounding rhetoric, though? For one,

naturalism minimally requires that metaphysicians don’t make claims that are in clear con-

flict with scientific claims. But ESR doesn’t seem to be in direct conflict with claims

made by science. Unsurprisingly, some structural realists opt for a much stronger nat-

uralism that requires not only that we formulate metaphysical theories compatible with

science, but that requires that any new metaphysical claim must be motivated by scien-

tific hypotheses.(Ladyman and Ross 2007, 37) It seems clear that there is some conceptual

space between this latter kind of naturalistic metaphysics, and a view according to which

metaphysicians should not make claims in direct conflict to claims in science.

More importantly, however, it is not clear that the intrinsic fundamental properties of

ESR are really idle wheels. After all, in characterizing what structure is, we need to be able

to contrast it with something that is not structure. Epistemic structural realists can do so

by contrasting the structure we can know with the intrinsic properties of things we cannot

know. But what are ontic structural realists going to say?

The idea that structure is all we can know through science was motivated, in part, by

examples of switching or exchanging intrinsic properties, the change in which didn’t make

a difference to the structure. The ontic structural realist is tempted to respond: well, if they

don’t make a difference to the structure, then that just goes to show that they don’t matter.


The differences in intrinsic properties which we could never know about that lead the epis-

temic structural realist to her humility just aren’t really differences at all. Ontic structural

realists then express this as the idea that there really aren’t any intrinsic properties after all.

There is only structure, and since we know about structure, there is nothing we cannot in

principle know about. Ontic structural realists then offer a kind of error theory for intrinsic

properties and individual objects: there really aren’t any such things. As we shall see be-

low, they often tend to combine this kind of eliminativism with a ‘reconceptualization’ of

the notion of object and perhaps also property, which is supposed to replace the notion of

individual object and intrinsic property.

But now it seems as though the ontic structural realist has kicked away the latter he was

still standing on. For the argument the epistemic structural realist had been using to get to

the idea that all we can know is structure relied on the distinction between a role within a

structure and what occupies that role. If we lose that distinction by individuating role oc-

cupants solely by the role they play within a particular structure, it seems that at it becomes

increasingly difficult to see why the knowledge we gain should be called structural. For

now we’ve just erased the distinction between structural and nonstructural knowledge.

So at least insofar as OSR is offered in the context of the dialectic with ESR, OSR un-

dermines what was meant to be ‘structural’ about ESR’s proposal, and the arguments that

were used to motivate ESR’s claim that all we can know is structure no longer seem avail-

able to support OSR. That doesn’t show that OSR is wrong or incoherent as a metaphysical

claim, but it shows that the arguments used to support ESR cannot be used as presupposi-

tions of the argument for ESR. OSR would do well to just reject ESR and find independent

arguments for OSR.

4.4 Structural Realism as a claim about the ontology of

particular theories

Let’s move on to the next branch. Perhaps ontic structural realism can be argued for in-

dependent of the arguments for ESR. Ontic structural realism is sometimes offered as an

account of the ontology of particular theories in physics. It will be helpful to take the


demand for ‘the ontology of the theory’ as a demand for a list of the ontological commit-

ments of the theory in something like the Quinean sense, although determining what these

commitments are is often not especially closely tied to the quantificational criterion. Find-

ing the ontology of theories in modern physics is a task philosophers of physics often see

themselves as working out (van Fraassen 1991). Structural realism can sometimes be seen

as providing an answer to that task or question (Lyre 2004). The theories in question are

quantum theories, in particular quantum mechanics and quantum field theory (QFT). Both

of these are a common target for philosophical ‘interpretation’. Unsurprisingly, structural

realists hold that the appropriate interpretation of these theories is that their ontology is

‘structure’.

4.4.1 The argument from the interpretation of quantum mechanics

One of the features of quantum mechanics that has attracted a lot of philosophical attention

are the statistics used in quantum mechanics to describe the behavior of particles. I will here

only focus on the different ways in which these statistics have been used to argue for ontic

structural realism – given the extensive debate this issue has received in the philosophical

literature, obviously at lot more could be said about it.23

The way it is typically presented, the argument goes like this.24 Quantum particles come

in two kinds, fermions and bosons, which obey different statistics: Fermi-Dirac statistics

and Bose-Einstein statistics. FD statistics and BE statistics differ from each other, as well

as from the statistics of classical statistical mechanics, Maxwell-Boltzmann statistics. This

difference can be brought out when we consider two particles of the same kind25, such as

two electrons, or two photons, and two different states, A and B. Labeling the particles 1

and 2, we expect four possible composite states for the composite system of particles 1 and

2:23Van Fraassen (van Fraassen 1991) gives an extensive treatment of the matter. For a critical perspective on

the metaphysical significance of different statistics for classical and quantum statistics, see (Huggett 1999).24Such a presentation can be found, for example, in (Ladyman and Ross 2007).25Particles are of the same kind if they have the same state-independent properties, such as mass, charge,

or spin. These properties can be said to define what kind of particle it is. Electrons, for example, have unitnegative charge, a spin value of 1/2 and mass: 9.109×10−31 kilograms or 0.511 MeV. These state-independentproperties are also sometimes regarded as a particles ‘intrinsic’ properties in the context of this literature.


1. Both particles in state A.

2. Both particles in state B.

3. Particle 1 in state A and particle 2 in state B.

4. Particle 2 in state A and particle 1 in state B.

These expectations are often said to correspond to classical Maxwell-Boltzmann statis-

tics.26 In quantum mechanics it turns out that we need different statistics, and indeed two

different ones. In none of these composite states is there any distinction between ‘particle

1’ and ‘particle 2’. The composite states differ only with respect to how many particles are

in a given state, not which particle is in which non-composite state (i.e. state A or state

B). The general lesson, it seems, is that particles are indistinguishable, that is, swapping

particles is not experimentally detectable. But this happens in slightly different ways for

fermions and bosons, and their respective statistics reflect this.

In the case of fermions like electrons, cases 1 and 2 can never obtain, and cases 3 and

4 are antisymmetrical, that is, the wavefunctions describing the two states differ only by a

sign. That leaves a probability of 1/2 for each 3 and 4. It is important to note, though, that

cases 3 and 4 are not experimentally distinguishable, since what is used to calculate exper-

imental outcomes is not the wavefunction, but its modulus square, and through squaring

differences in sign of course disappear.

Things are slightly different in the case of bosons. There cases 3 and 4 are symmetrical,

not antisymmetrical, that is the wavefunctions for 3 and 4 don’t differ even by a sign. This

also means that for bosons, cases 1 and 2 are not ‘forbidden’. Accordingly, BE statistics

instead assigns 1/3 to each case 1 and 2, and 1/3 to the state of: one particle in A, one

particle in B.

Another way of expressing this is to say that quantum statistics display permutation

invariance: switching around particles (or their labels) leaves the statistics unchanged.27

26But perhaps we should be more careful, see (Huggett 1999).27Switching the particles or switching the particle labels are two ways ‘to think about’ the permutations,

which may be called ‘active’ and ‘passive’ respectively. Given that we cannot actually pick one particleand switch it with another, it would be misleading to think that in one case we are carrying out a physicalpermutation, though. On the other hand, one might think that given permutation invariance, it is misleadingto speak of labeling the particles at all.


Quantum statistics doesn’t distinguish between the case of Particle 1 being in state A and

particle 2 being in state B, and the case of Particle 2 being in state A, and particle 1 being

in state B. As far as the physics is concerned, these permutations don’t matter.

It has been suggested that these statistics show that quantum particles violate Leibniz’s

Principle of the Identity of Indiscernibles (PII).28 According to this principle, if two entities

a and b share all properties, a and b are in fact identical. But quantum particles, it seems,

are indistinguishable, that is, they don’t differ in any property, without being identical. The

number of bosons in a given state makes a difference, even if they cannot be distinguished.

So they cannot be identical – or at any rate not numerically identical. But Leibniz’s princi-

ple seems to require this, since these particles have all properties in common. So quantum

particles violate Leibniz’s Principle.

How is this supposed to be an argument for structural realism as the appropriate on-

tology of quantum mechanics? One way in which the issues surrounding permutation

invariance have been used to support structural realism is by suggesting that the permu-

tation invariance shows that particles are not individuated over and above the position they

occupy within the structure.29 Particles are non-individuals, and since structural realists

interpret objects as not having individuality over and above the structure they are a part

of, the fact that the quantum state is the same under permutation of particles shows that

structural realism is the appropriate kind of ontology for this theory.

How should one respond to this argument?

One option is to refine Leibniz’s principle in a way that will allow us to distinguish the

particles after all. Different versions of this principle can be distinguished, depending on

how strong the requirement of having all properties in common is taken to be: does that

mean only intrinsic ones, or does it include extrinsic properties as well? The second is a

weaker version of the principle, since one might think that it is much more plausible to

28This suggestion is not new. Worries about the PII and quantum statistics go back to Reichenbach. (Re-ichenbach 1956) It has more recently been taken up by various philosophers of science (esp. Steven French),(Krause and French 1995), (French 1998) and is now widely regarded as at least one of the possible interpre-tation of the quantum statistics. (van Fraassen 1991), (Teller 1995).

29This is not how the argument was originally introduced - a point to which I will return in the nextsection. There is some amount of tension among structural realists about how to understand the lessons fromquantum statistics as an argument for structural realism. Ladyman at times (Ladyman and Ross 2007) seemsto subscribe to the argument as I present it in this section, whereas French (French 2010) continues to takethe argument as an argument from underdetermination (see next section).


claim that objects sharing all intrinsic as well as all extrinsic properties are the same object,

than to claim that objects merely having all intrinsic properties in common are the same.

Paradigmatic extrinsic properties (in this context) are spatio-temporal properties. Two ob-

jects, the thought goes, with all the same intrinsic properties may nevertheless be distinct

simply in virtue of being in different spatio-temporal locations. For classical objects, that

will solve the problem in most cases, but for the quantum case that won’t do. Quantum par-

ticles will require a different class of properties to be distinguishable, since unlike classical

objects, quantum particles lack definite space-time trajectories.

There are other ways to refine Leibniz’s principle, however. Simon Saunders, following

Quine, has offered a three-fold distinction between absolute, relative, and weak discernibil-

ity.(Saunders 2006) Absolute discernibility means that there is a formula such that it is true

of one object, but not the other, relative discernibility means that there is a formula such that

the objects can only occur in one order in that formula if they are to make it true, and weak

discernibility means that there is a two-place irreflexive relation the two objects satisfy.

This last notion of discernibility allows for fermions to be discernible, since fermions have

to satisfy Pauli’s exclusion principle, i.e., they have to be in a different state with respect

to at least one quantum number. So any two fermions will always satisfy some two-place

irreflexive relation, such as ‘being of opposite spin to’. This does not apply to bosons. For

in the case of bosons, not only is it the case that classes of bosons (like photons) share all

their intrinsic (state-independent) properties30, but they can also be in the same state. This

is reflected in the statistics explained above. So it would seem, then, that bosons violate

even very weak forms of the principle of identity of indiscernibles.

Ontic structural realists have responded to Saunders’ suggestion by embracing it as just

one way of being a structural realist. The individuals distinguished by weak discernibility

are “nothing over and above the nexus of relations in which they stand” (Ladyman and Ross

2007, 138). This suffices for ontic structural realism. This is a partial retreat from more

radical forms of OSR, according to which individuals don’t exist at all.31 The claim is now

merely that structures are what exists fundamentally, whereas individuals are somehow

30This use of ‘intrinsic’ is typical in the literature on this problem, see (French 1998).31This again also reflects a kind of tension between more eliminativist tendencies in structural realism,

which sometimes sound like a kind of ‘error theory’ regarding object-talk, and an understanding of OSRaccording to which objects aren’t so much eliminated as reconceptualized.


ontologically dependent on the structures they stand in.32

A different response to the argument, instead of refining Leibniz’s Principle, is to in-

troduce a property that will serve to distinguish the different particles. The property will

have to be non-dynamical, since the statistics suggest that there is no difference between

states where the particles have been switched. The empirically adequate statistics can be

obtained by taking all of the states we classically expect to be possible, but then to assume

that there is some hidden reason why some of them are inaccessible. For example, quantum

particles can be thought of as having certain non-dynamical intrinsic properties (some kind

of hidden variables), that make the possibility of the nonsymmetric states intelligible, and

the idea that particles are individuals is preserved. So while the physics doesn’t tell us, even

in principle, which particle is in which state, there is nevertheless a difference between the

case of: particle 1 in state A and particle 2 in state B and the case of: particle 2 in state A

and particle 1 in state B. French himself uses this underdetermination to argue for struc-

tural realism in a somewhat different way, as we shall see in the next section. However,

this underdetermination means that structural realists who embrace an argument along the

lines offered here have to concede that the considerations from physics do not compel us

to adopt OSR. The argument from indistinguishable particles, then, remains inconclusive,

even by their own lights (Ladyman and Ross 2007, 154).

4.4.2 The argument from the interpretation of QFT

As we have seen, the argument from indistinguishable particles can be used as an argument

for ontic structural realism insofar as one of the interpretative options, namely particles as

either non-individuals or in the case of fermions at best ‘thin’ individuals, seems to be a

structuralist option. In this section I look at ontic structural realism as an ontology for quan-

tum field theory. Quantum field theory is often appealed to as undermining the traditional

picture of fundamental reality to which structural realism is offered as an alternative. The

‘traditional picture’ is here once more Humean Supervenience, and structural realists are

not the only ones who object to Humean Supervenience on the basis of considerations from

particular physical theories. Tim Maudlin, for example, uses QFT to argue against Humean

32Earlier French and Ladyman (French and Ladyman 2003b) maintain that our difficulty with this idea isdue to our attachment to traditional (first-order) logic.


Supervenience, without subscribing to the label ‘structural realism’ (Maudlin 2007) .

The arguments from QFT depend upon particular ways of reading the formalism(s) of

quantum field theory. Finding ‘the ontology’ of quantum field theory requires interpreting

the formalisms of quantum field theory. Since OSR is offered as a claim about the ontology

of QFT, the arguments for OSR are usually committed to particular interpretations of QFT.

Ontic Structural Realists argue that QFT seems to suggest an ontology that is more rela-

tional, and more holistic then the traditional ontology would allow (French and Ladyman

2003b). The fundamental entities in QFT, structural realists argue, cannot be regarded as

substances with intrinsic properties, nor are the properties of the whole determined by the

properties of the parts. QFT shows that traditional ontology is wrong.

Field theory?

Sometimes it is suggested that the traditional ontology is already undermined by taking

QFT to be a field theory (French and Ladyman 2003b). It is not quite clear what the

argument is, however. Traditional ontology, seems to be the thought, would go well with

a particle theory, but not with a field theory. Classical particles can be seen as individual

entities with intrinsic properties that account for their behavior. They are localized in space

(-time). The properties of one particle do not depend on the properties of others. That all

looks very friendly to traditional metaphysics. But if our fundamental theories are field

theories, so the thought, the fundamental entities are fields, and fields are not individual

substances with intrinsic properties. So if traditional metaphysics is somehow committed

to tiny substances buzzing about in empty space, field theory will seem to be a problem.

The far more interesting target, however, is not this allegedly traditional picture, but

Humean Supervenience. And Humean Supervenience doesn’t have a problem with field

theory at all. For Humean Supervenience has a response to the considerations from field

theory. The appropriate candidates for individuals, if needed, are not fields, but instead the

space-time points on which the fields are defined. On this view fields aren’t anything over

and above collections of space-time points with certain physical quantities defined on them.

So in the end we would still have individuals with intrinsic properties at the fundamental

level, whereas fields and other structures supervene on them.33

33Recall that Humean Supervenience is primarily committed to the intrinsic properties, but is more flexible


The success or failure of such a position depends, on the one hand, on what views

are plausible views of the status of space-time (and space-time points), and on the other

hand on the status of quantum fields. For quantum fields, unlike classical fields, are not

defined at each point in space time, but rather ‘smeared out’ over regions of space-time.34

So even if we take space-time points to be our fundamental entities, it seems we will not

be able to say that each of them is an individual with its own set of intrinsic properties,

the field properties. But perhaps the friend of Humean Supervenience can settle for small,

well-defined space-time regions as the bearers of intrinsic properties instead?

The Aharanov-Bohm effect

Another way in which QFT might suggest an alternative to ‘traditional’ metaphysics (and to

Humean Supervenience in particular), are the interpretations of the Aharonov-Bohm effect.

A simple version of the AB-effect can be described as a modification of the standard two-

slit experiment in quantum mechanics. In the standard two-experiment, a beam of electrons

is send from a source through a mask with two openings (the two slits) to a screen. When

both slits are open, the screen will show an interference pattern.

A modification of this experiment exhibits the AB-effect (see diagram next page). In

this modification, a shielded solenoid is placed behind the mask. When an electric current

passes through the solenoid, a magnetic field will be generated, but since the solenoid

is shielded, the magnetic field remains inside the solenoid. Interestingly, however, if a

shielded solenoid is placed behind the mask and an electric current is sent through it, the

interference pattern on the screen still appears, but it will have shifted by a certain amount.

This shift is surprising, since the magnetic field inside the solenoid should have had no

influence on the electrons passing through the two-slits. The puzzle, then, is how to explain

the shift in the interference pattern.

At least two of the three standard interpretations35 of the Aharonov-Bohm effect seem

in terms of where those intrinsic properties are instantiated: space-time points, or ether, or fields.34This echoes a worry Michael Redhead (Redhead 1998) raises for Sunny Auyang’s treatment of quantum

field theory. (Auyang 1995) Auyang suggests an ontology of events which are located at particular points inspace-time, and accordingly faces the same problem.

35We could either treat the magnetic field as the only physically significant quantity, or we could in additiontake the electromagnetic potential as additional physically significant quantity, or we could treat holonomies


Source

Electron beam

solenoid

mask screen

shifted interference

pattern

original interference

pattern

Figure 4.2: The Aharonov-Bohm Effect


require some kind of non-locality, either as action-at-a-distance, or by forcing us to give up

the idea that physical quantities are assigned to space-time points, and instead attributing

them to loops on space-time.(Belot 1998) The latter, so called holonomy interpretation,

suggests not only that the relevant properties are no longer located at individual space-

time points, but perhaps even that basic entities are abstract mathematical objects (the

holonomies).36 Holger Lyre (Lyre 2004) seems to think that the holonomy interpretation is

committed to this, whereas Richard Healey (Healey 2007) thinks that the basic entities are

the holonomy properties, i.e., properties that are located in space-time on loops described

by the holonomies.37 Healey in fact hopes that this offers a much more Humean ontology

for QFT than many of the alternatives. For on his view, the holonomy properties come out

as intrinsic properties.

What does this mean for structural realism? On Lyre’s understanding of the holonomy

interpretation, it supports structural realism because the entities it is committed to are ab-

stract and because the properties are not located at space-time points but space-time regions.

This ‘non-separability’38 is holistic, which Lyre takes to be a point in favor of structural re-

alism. Healey, on the other hand, thinks that the holonomy interpretation comes to the

rescue of a more Humean picture, since the holonomy interpretation allows for intrinsic

qualitative properties to be located in well-defined space-time regions, even if the regions

are non-separable. So the same interpretation of the AB-effect is here used both for and

against structural realism.

This paradoxical situation arises because in one case, holism of a certain kind is seen as

characteristic of structural realism, whereas in the other case, structural realism is seen as a

view in conflict with intrinsic properties. The conflict with intrinsic properties of course ties

(or holonomy properties) as physically significant, but not the electromagnetic potential. Each of these possi-bilities has certain costs (requiring different notions of non-locality, taking gauge dependent properties to bephysically significant), but the formalism seems to allow for all of them.

36Holonomy is a phenomenon in differential geometry: parallel transport of vectors is path-dependent,meaning that parallel displacement of a vector v0 along a closed curved can result in a different vectorv f .(Frankel 2004)

37“The holonomy properties of a loop act on quantum particles by affecting qualitative intrinsic physicalproperties that attach to a particle on, or in arbitrarily small neighborhoods of, the entire loop.”(Healey 2007,110)

38Non-separability is contrasted with ‘separability’, the idea that intrinsic properties are defined at individ-ual space-time points.(Healey 2007, 46) The contrast between separability and non-separability is introducedto distinguish these issues from other questions about locality in quantum physics.


in directly with the discussion in the previous section between the epistemic and the ontic

structural realist, and is hence not peculiar to the discussion about quantum field theory.

What quantum field theory adds is a different kind of worry for Humean Supervenience,

namely a worry about the possibility of a new kind of non-locality, generated by the fact

that field properties might perhaps have to be regarded as no longer located on individual

space-time points. Other interpretations of the AB-effect introduce a different kind of non-

locality: the kind of non-locality that seems to violate special relativity by introducing

action at a distance. Either of these kinds of non-locality might potentially be a threat

to the Humean picture, but the non-locality implied in action at a distance is something

we already encounter in quantum entanglement. The non-locality that is at issue in the

Aharanov-Bohm effect might be of a different kind, it might non-separability in the sense

of locating properties not on points but larger regions of spacetime.

These worries are different from the general concerns that have been raised earlier (in-

cluding permutation symmetry in quantum statistics) in that they are worries that arise

specifically within particular theories, and at a point about which there is disagreement

even within physics. Permutation symmetry is just a matter of ruling out certain apparent

distinctions as physically irrelevant. Non-locality and non-separability on the other hand,

are genuine physical possibilities, and as of now we have no uncontested account of entan-

glement or the AB-effect within physics.

4.4.3 How well does this strategy work?

We’ve now seen two attempts to argue for OSR as an answer to the question ‘what is the

ontology of our best physical theories?’. There is a bit of a tension between taking ontic

structural realism to be a general position in metaphysics and taking it to be the answer to

the question of what the ontology of a particular theory is like. For in the latter case, what

we are doing is determine the ontological commitments of the theory, and of course, those

commitments may differ from what there actually is. Of course, being naturalists, ontic

structural realists will insist that the only way to answer the former the question is by going

through our best scientific theories. But unfortunately, as we have seen, it is far from clear

what the ontological commitments of these theories are. In the case of quantum field theory


we moreover have the problem of how to individuate the theory in the first place.39 So we

should perhaps be cautious before attempting to draw general metaphysical conclusions

from the ontology we interpret these theories to have.

Another worry about this line of argument for OSR is that there is no clear sense in

which structural realism is the same position when it is offered as the ontology of two (or

more) different theories. That would only be the case if what is meant by ‘structure’ when

offered as the correct ontology for quantum mechanics were the same as what is meant

by ‘structure’ in the interpretation of quantum field theory. But while there may be some

similarities (in both cases there are worries about non-locality of some form), it is far from

clear that this is in fact the case.

This holds even if we take the non-locality that seems to show up in quantum mechan-

ics as well as in quantum field theory to be the motivation for structural realism, and not

permutation symmetry, which seemed to be less a matter of what the ontology of a partic-

ular theory is like, and more a matter of what differences are important for physics. For

as I mentioned above, it is an open question of how to account for the non-locality that

seems to be involved in entanglement, and it is an open question of what exactly to make of

the AB-effect. There may be different kinds of non-locality, both of which might rule out

Humean Supervenience as a viable option, but that doesn’t mean they are both accounted

for by the same metaphysical view.

4.5 Structural realism as a resolution of underdetermina-

tion problems

Structural realists sometimes seem to suggest that what they are after, when they offer

structural realism in the context of such interpretation debates is not one position among

many that are offered as interpretations, but instead an attempt to avoid any commitment

to a particular interpretation. This style of argument was originally offered in response to

what was perceived as a ‘metaphysical underdetermination’ arising from the possibility of

interpreting quantum particles either as individuals or non-individuals.(French 1998) The

39For discussion of this problem see (Ruetsche 2002).


idea was to avoid the underdetermination altogether by giving up an ontology centered on

objects. Since then allegedly similar ‘underdetermination’ situations have been offered as

also requiring a structuralist treatment.

4.5.1 The argument from multiple interpretation

The first instance of this type of argument in favor of structural realism can be found in

Steven French’s discussion (French 1998) of quantum statistics vs classical statistics, dis-

cussed above. French argues that while Bose-Einstein and Fermi-Dirac statistics suggest

that quantum particles are somehow non-individuals, really we have a case of ‘underdeter-

mination of metaphysics by physics’, because it is possible to interpret the statistics in a

way that is consistent with the particles being individuals. Since these two metaphysical

possibilities are underdetermined by our physical theory, we should instead just be struc-

tural realists, leaving aside the question of which interpretation to prefer altogether. How

is that supposed to work? Here’s a statement of the argument by Ladyman.

We need to recognize the failure of our best theories to determine even the mostfundamental ontological characteristic of the purported entities they feature. Itis an ersatz form of realism that recommends belief in the existence of entitiesthat have such ambiguous metaphysical status. What is required is a shift to adifferent ontological basis altogether, one for which questions of individualitysimply do not arise.(Ladyman 1998, 419/20)

The perceived problem, for French and Ladyman, appears to be that quantum mechan-

ics (more specifically permutation symmetry) is compatible with an interpretation that takes

particles to be individuals as well as with an interpretive package that denies this. Failing

to determine the answer to this question is so bad, Ladyman seems to suggest, that it under-

mines standard scientific realism, since recommending that we believe in objects with ‘such

ambiguous metaphysical status’ is unacceptable. The structural realists’ remedy is to find

a way of being realists that avoids the question of individuality altogether: once we shift to

‘structure’ as our ontological basis, the two different interpretations are no longer alterna-

tives between which we have to choose. The strategy, then, is to avoid underdetermination

by finding a way to reject the underdetermined question.


This argument has been criticized in a number of different ways. One line of objection

questions whether there really is underdetermination.40 But other premises of the argument

have also been called into question. Katherine Brading and Alexander Skiles (Brading

and Skiles ), for instance, take issue with a number of other premisses of the argument,

including in particular the question whether we should really expect our theories to have an

answer to the question of individuality.

Finally, and this is a worry also offered by (Brading and Skiles ), if underdetermination

really warrants such radical conclusions, then shouldn’t structural realists themselves be

concerned that physics doesn’t decide, for examples, between different forms of structural

realism?41

Structural realists often suggest that underdetermination is bad because it seems to lead

to an epistemic humility they find unacceptable. As French puts it: “OSR is grounded

on the fundamental principle that we tailor our metaphysics to our epistemology” (French

2010). The idea, which we already encountered in the dialectic between ESR and OSR,

is that if our theories don’t answer a question, or give us knowledge about something, we

should change our metaphysics in such a way that the unknowable ‘goes away’ in some

way. But such a principle is plausible at best in cases where there is a principled problem

with our epistemic access. It doesn’t seem plausible at all when the claim is merely that,

for now, a question remains unanswered. And even in the in principle case, it is unclear that

we should change our metaphysics instead of our epistemology. The latter just seems like

extreme verificationism, which Ladyman and French may well accept, but it is unsurprising

that others choose not to follow their lead. The argument from underdetermination seems

less than fully persuasive, not just because the use of the particular physical theories raises

questions, but also because it seems to rely, ultimately, on rather questionable verificationist

principles.

40See (Pooley 2005) for details.41For a similar concern, see Ainsworth (Ainsworth 2010), who argues that only eliminativist OSR will

avoid the underdetermination altogether.


4.5.2 Other cases of multiple interpretation?

There is similar kind of argument for structural realism based on QFT. Holger Lyre (Lyre

2004) suggests that there are three different ways of interpreting the Aharonov-Bohm ef-

fect.42 We could either treat the magnetic field as the only physically significant quantity,

or we could in addition take the electromagnetic potential as physically significant quantity,

or we could treat holonomies (or holonomy properties) as physically significant, but not the

electromagnetic potential. Each of these possibilities has certain costs (requiring different

notions of non-locality, taking gauge dependent properties to be physically significant), but

the formalism seems to allow for all of them.

So like the quantum statistics example we seem to have a case where the formalism of

the theory doesn’t by itself give us a unique ‘metaphysics’ or ‘ontology’ of the situation.

This ‘underdetermination’, Lyre suggests, is itself a reason for structural realism, for the

structure of the theory remains the same across these different interpretations. The structure

Lyre has in mind is the mathematical or formal structure of the theory. Since the formalism

that allows for the different interpretations remains the same, there is a sense in which

the structure can be regarded as more reliable than any one of the interpretations. Notice,

though, that this is a different sense of structure from the one French and Ladyman have

been using.

There are two differences between these two cases, one subtle, the other less subtle.

The less subtle difference is that in the case of the different interpretations of the AB-effect,

it is unclear that the problematic questions of interpretations can be avoided by ‘switching

to a different ontological basis’. The suggestion to stick to the formalism has little to do

with reconceptualizing our notion of object, nor is it clear that taking an ‘objected centered’

view of ontology is responsible for the problems with the AB-effect in the first place. This

reaffirms the worry that the structural realists’ response to a particular underdetermination

problem isn’t all that helpful, since if underdetermination occurs in other situations as well,

and ones where the structuralists’ gambit doesn’t apply, then perhaps we need to find a

better response to underdetermination overall, and that response is unlikely to be structural

realism.42Similar options are presented by (Belot 1998).


The subtle difference between the two cases is that in the first case, we are not in doubt

about which entities, quantities, or states are physically significant: particle label permu-

tation is not physically significant. The underdetermination comes in when we ask about

what one might call ‘metaphysical significance’. For the two metaphysical alternatives be-

tween which the theory remains indecisive concern not anything that could be physically

significant: by definition, the properties that would give each particle individuality have to

be dynamically inert. The remaining question, it seems, is whether or not we have meta-

physical grounds to reject the idea of entities as non-individuals, and hence have to insist

on haecceitism. It is perhaps not surprising that physics alone doesn’t answer this question.

In the interpretation of the AB-effect, however, the question is precisely what to take as

physically significant quantities in the first place. Each interpretation offers different candi-

dates. The costs and benefits of choosing one interpretation over another are motivated by

other, more general, considerations from physics: physically significant quantities ought

to be gauge independent, certain kinds of non-locality are in conflict with field-theoretic

thinking in general, and special relativity in particular. So at least in the first instance,

the principles invoked to assess the costs and benefits of the different interpretations are

principles used in physics, not in metaphysics.43

The argument from multiple interpretations in the case of the AB-effect is then not

fully analogous to the argument as presented with respect to quantum statistics. Indeed,

the ‘adjust metaphysics to epistemology principle’ used to motivated the shift to structural

realism in the case of the quantum statistics argument seems even less plausible for the un-

derdetermination in the case of the AB-effect. For here it seems that further research might

well favor one of the interpretations over the others, and this particular underdetermination

might turn out to be merely temporary.

Despite the dissimilarities between the two arguments, there is a clear theme in com-

mon. In both cases different interpretations/ontologies are said to be underdetermined by

the formalism itself. At least Lyre furthermore explicitly suggests that what is the same

across different interpretations/ontologies is structure, and hence contrasts, once more,

43One might of course wonder just what the status of these physical principles themselves is. For they arenot, it seems, straightforwardly empirical facts; it is not just an observational fact that all physical quantitieshave to be gauge invariant, or that there is no action-at-a-distance. But that’s a topic for a different debate.


structure with interpretation/ontology. This suggests that this line of argument, from under-

determination of interpretation/ontology, is closer to the argument Worrall offered (which

I discussed in section 4.2.), in that it argues for structure as an alternative to ontology.

The force of the arguments is not the same of course, since the argument from history is

explicitly meant to address Kuhnian incommensurability worries, whereas the underdeter-

mination of interpretation of particular theories is not suited to do so - after all, there is no

theory change involved. In both cases, however, the contrast is drawn between structure

and ontology.

One worry is of course that structural realists cannot have it both ways, and should

hence make up their minds as to which of the two options they choose: is structural realism

a particular interpretation, or a proposal to abandon the search for interpretations?

A further concern is that in the latter case we might be losing some of the realist impetus

of structural realism. For the idea that we should stand back and refrain from committing

to any particular interpretation is a view shared, for instance, by van Fraassen. While van

Fraassen suggests that pursuing and developing different interpretations may be a worth-

while task for philosophers of science to engage in, he reserves the right as a constructive

empiricist to withhold belief in any particular interpretation offered, precisely because there

is always going to be a plurality of interpretations (van Fraassen 1991). Structural realists

don’t want to follow van Fraassen here, but it is unclear that their response has the same

broad appeal as van Fraassen’s. If it only works for particular cases of underdetermination,

then it doesn’t seem to answer to the larger concerns van Fraassen is addressing.

4.6 A different debate

We’ve now seen four different ways of understanding what it might mean to be a structural

realist. Structural realism has been offered as a thesis within the traditional debate over

scientific realism, as a metaphysical thesis, as a position regarding the ontology of particular

theories, and as an attempt to get out of the business of offering answers to certain kinds

of interpretive questions about the ontology of physical theories. While there seems to be

a kind of family resemblance between these different positions, it is clearly not the same

position that is under discussion in these different debates, nor is it clear that there could be


a single position that fits the bill in all of them.

It is also quite clear that the notion of ‘structure’ employed in articulating and arguing

for these different positions varies from case to case. Two senses of ‘structure’ can be found

at various points throughout the debate: formal structures and concrete structures.44

Understanding structure as formal is part of the older tradition of structural realism,

going back to Russell and Maxwell. It comes in two varieties: logical structure or set-

theoretical structure, depending on whether the syntactic or the semantic view of theories

is preferred. To be a realist about this kind of structure is to offer an answer to the ques-

tion: what is preserved across theory change, to which the answer is: structure (or partial

structure) in the sense of formalisms. This sense of structure is also easily contrasted with

interpretations that can be offered for particular formalisms.

If being formal is taken to be the main characteristic of structure, though, it is unclear

how structural realism can be offered as a metaphysical position, unless it is offered as an

extreme kind of platonism. For if the claim is that structure is all there is, and by structure

what is meant are mathematical or logical formalisms, then it seems that the claim is that

all there is are abstract entities of a certain peculiar kind. This would certainly be quite

an extreme metaphysical view, and also one that seems implausible as a view about what

physics and science in general are all about.45

Accordingly, ontic structural realism as a metaphysical view is typically presented as

view about concrete structures, where concrete structures are understood as modal or causal

structures (Ladyman and Ross 2007), (French 2010). While such concrete structures don’t

have the same strong platonist flavor, they seem rather odd as ‘ontology’ for physical the-

ories. Below I suggest a way to understand the metaphysics of structural realism vis-a-vis

more traditional metaphysical positions. But this switch from formal to concrete structures

means that certain kinds of arguments are no longer obviously available to structural re-

alists, and that the contrast between structure and interpretation or structure and ontology

is not longer available even on an intuitive level. For it is not clear that modal structures

remain the same across theory change, or that there is any sense in which different interpre-

tations of the same modal structure are available. So both the argument from the traditional44(Ainsworth 2010) also draws attention to this distinction.45For worries along these lines see (Cao 2003), for a response by structural realists see (French and Lady-

man 2003a).


realism debate, and the argument from underdetermination start to look inapplicable if

structural realism is taken to be realism about modal structures.

Given how unclear it is what structural realism is supposed to be, and given how little

promise the view seems to hold in many of the debates it appears to contribute to, it is

tempting to dismiss structural realism as ill-conceived. Structural realists have been trying

to contribute to three main debates: what aspect of science should we be realists about,

what are the ontological commitments of particular theories like quantum field theory, and

a general metaphysical debate about what our fundamental ontology looks like. As we’ve

seen there are tensions between answering all three of these debates with what seems to

be meant as a single position. It seems to me that we will be able to make more sense of

structural realism if we see it as a developing position: it begins with an attempt to articulate

a particular form of realism about science, which leads to a change in how to conceive what

realism and antirealism debates in science are understood to be about, and OSR is then

articulated as a particularly extreme position within the newly emerging debate.

Initially one might think the debate should be about whether or not the structures our

theories ascribe to phenomena exist or not. That is, initially one might try to change what

it is that we are realists with respect to from ontology to structure, while hanging on to the

idea that what it means to be a realist is to make an existence claim (and accordingly, what

it means to be an antirealist is to deny that existence claim). This doesn’t work so well.

For if it is the structures we are realists with respect to, and realism means accepting an

existence claim, then structural realism just means believing that structures exist. But since

it is very difficult to conceive how there could just be structure, without any non-structural

elements to stand in or occupy that structure, it seems that structure is at best an addition

to a more standard ontology of objects, without being the radical new view that structural

realists had been aiming for.46

If structural realists have to accept that structure and objects both exist, the next best

thing to articulate what it means to nevertheless be a realist about structure (as opposed

to objects) is to try the move according to which structures are more fundamental than

objects. We’ve seen in some of the quotes above that this is indeed a direction in which

some structural realists have moved. How is the priority or fundamentality of structures

46See discussion of OSR in section three above.


going to be cashed out?

4.6.1 Objects and Structure - which one has priority?

One way to understand the metaphysical debate structural realists are trying to contribute

to is one of priority. Are structures grounded in the objects (and their intrinsic properties)

that occupy roles within the structure, or are objects grounded in the structures they figure

in? Initially there seem to be three potential views: objects ground structures, structures

ground objects, structures and objects are equally fundamental.

Structural realism seems ambiguous between the second and the third view. There’s a

good reason for this, which can be brought out if we try to articulate the kind of dependence

relation that might hold between structures and objects. For it turns out that we might have

to formulate that relationship differently depending on whether we take objects or structures

to be fundamental.

When objects (and intrinsic properties) are fundamental, then structures can be taken to

supervene on them. Supervenience is cashed out in terms of what makes a difference: if A

supervenes on B, then for any change in A, there has to be a change in B. Supervenience

works well for the case of taking objects or intrinsic properties to be fundamental, since this

allows us to distinguish between a role and its occupant. In particular, it seems possible

to switch the occupants of a role, without changing the roles available for occupation.

This allows for the possibility of differences in the objects and/or their intrinsic properties,

without differences in the structure they are a part of.47

The same is not the case if structures are taken to be the supervenience basis. For

objects to supervene on structures, any change in objects would have to require a change in

structure. But as we’ve seen, that doesn’t seem to be the case. For while it is conceivable

how a change in objects (or intrinsic properties) could leave the structure unchanged, it is

unclear how the reverse could be true. Changes in objects or intrinsic properties can seem

to happen without changing the structure. So objects and their intrinsic properties cannot

supervene on structures. This means that if structural realists want to claim that objects or

intrinsic properties somehow depend on the structures they are in, this dependence cannot

47Recall that this had been one of the key ingredients to getting to the thesis of humility earlier.


be cashed out in terms of supervenience.

Structural realists attempt to characterize the dependence in terms of an ontological

dependence. Objects are just nodes in the structure, they cannot exist independently of the

structure, or they don’t exist over and above the structure. But it’s unclear whether that will

actually give the appropriate kind of dependence, as opposed to interdependence. For this

to be a one way dependence, structures would have to be able to exist without objects. But

if objects are just nodes in the structure, how could there be a structure without nodes? If

objects are nodes in a structure, it is hard to see how you could get rid of the nodes without

getting rid of the structure. You could, of course, get rid of a particular node, or add a

node, but each of these two operations would also seem to involve a change of structure.

The only operation that leaves the original structure intact would be to replace a particular

node with a different node in the same place of the structure. But if all there is to objects

is that they are nodes in the structure, such an exchange doesn’t even seem possible. Such

an exchange was intelligible as long as we thought of objects as more than just nodes, but

once we give that up, any change in objects also means a change in structure. So structural

realists, due to the way they characterize objects, seem to have blocked the one operation

that might help to explain the sense in which structure is more fundamental than objects.

It is only if objects are more than merely nodes of the structure that it makes sense to talk

about swapping nodes without changing structures; but now that’s because we are swapping

occupants of nodes, not nodes. Structural realism then has to be an interdependence view.

A third way in which structural realists might aim to articulate the priority of structure

over objects is along the lines of the notion of grounding. Grounding has recently been

used by a number of philosophers to articulate metaphysical priority relations.48 The way

to cash this out, given the way grounding has been discussed in the literature, would be to

say that propositions concerning (individual) objects or intrinsic properties are grounded in

propositions about structures.

Should structural realists decide to pursue this particular approach, however, they will

have to say a lot more. In particular, it seems that this approach rules out more elimina-

tivist versions of OSR, since grounding, unlike reduction, is non-eliminativist (Fine 2001).

Moreover, the main evidence for grounding claims seem to be intuitions we have about

48See for example (Schaffer 2009), (Fine 2001).


grounding - and to a certain extent grounding is supposed to be an explanation for these

intuitions (Fine 2001). The problem is that in the case of structures and objects, or struc-

tures and intrinsic properties, it would seem that insofar as we have any intuitions about

grounding, they point the other direction: from objects to structures, not vice versa. Fi-

nally, structural realists tend to be dismissive of the use of intuitions in determining our

metaphysics, which would make appeal to intuitions about grounding methodologically

unsuitable for structural realists. But given that grounding is commonly motivated as a

means for explaining our intuitions, it is unclear that grounding is appealing to a brand of

metaphysics that tries to ban the appeal to intuitions altogether (Ladyman and Ross 2007).

The contrast between Humean Supervenience and Structural Realism can then be re-

formulated as follows: Humeans believe that objects and their intrinsic properties are inde-

pendent of the structures (including external, modal, and other relations) they might stand

in, whereas structural realists believe that this contrast is misleading: the roles in a structure

and what occupies these roles stand in a much tighter relationship than the Humean wants

to admit. If we take the case of properties rather than objects, we could say that structural

realism is a close cousin of the causal view of properties.49 According to the causal view

of properties, properties are individuated by the causal relations they figure in, and only

those. This is in sharp contrast to the quidditism that comes with Humean Supervenience,

according to which different intrinsic properties can play the same causal roles. And, one

might add, according to which there are no necessary connections between the intrinsic

properties and the roles they might occupy.

While this helps to place structural realism within existing debates in metaphysics, it

does little to explain just what it means to be a structural realist. We know that the target

of the realism is structure, but we still don’t know what it means to be a realist, and neither

taking realism to be a matter or existence, nor taking it to be a matter of priority seemed

to offer an appropriate gloss. For in both cases the structural realist doesn’t end up with a

view that gives more weight to structure; at most, structure is now given equal weight. The

interdependence between structures and objects that was brought out by the considerations

49If, as I suggested above, structures are understood as concrete modal or causal structures, as opposed toabstract or formal structures.


in this section are of course interesting, and certainly make for a controversial metaphysical

view, but by themselves they don’t seem to make for a realism/antirealism debate.

4.6.2 Structure and Modality

A hint is provided by the characterization of ontic structural realism as a claim that objec-

tive modal structures are part of fundamental ontology (Ladyman and Ross 2007). That

suggests a somewhat different debate than the one we just looked at, although as we shall

see, there are also some connections between the two. Structural realists have asked us

to shift the target of realism debates from ontology to structure, and as we now see, the

structures in question were meant to be modal structures. Using the suggestions made in

the previous chapter, perhaps we should try to take OSR as an answer to the question: what

is the status of modal claims made by scientific theories?

With this understanding of what the target of realism is, what does it mean to be realist?

To be a realist about modal structures means to take them to be two things: objective, and

fundamental. Both of them are needed, since, as we saw in the last chapter, in this debate

there are a variety of opponents.

The claim that modal structures are fundamental accounts for the contrast between the

Humean account of modal claims and OSR’s account. For Humean Supervenience is of

course a view on which modal claims are objective50, but they hold in virtue of what is

true in other possible worlds (or perhaps: in virtue of what other possible worlds there are),

not in virtue of modal structures in the actual world. Possible worlds, in turn, are Humean

again, so once more modal structures are not fundamental, even though modal claims are

objective.51

It is no accident that OSR and Humean Supervenience should find themselves in these

two positions with respect to modality, and on the dependence vs independence camp with

respect to the previous debate. Humean Supervenience is all about the independence of

what something is (its identity) and what it can do or how it relates to other things. This

50As in, for example, the Mill-Ramsey-Lewis view of laws of nature.51This is not to say, of course, that there aren’t worries about the Humean account of modal claims like

laws. Here I’m only interested in what kind of a realist response the contemporary Humean response mightamount to.


drives the idea that there is no necessary connection between what something is and which

roles it might occupy (quidditism). OSR, on the other hand, wants to say that there is a

necessary connection between the identity of an occupant and the role it occupies. If OSR

is right that there is such an interdependence, then the Humean account of modal claims

is not going to work. At least there will be necessary connections between roles and their

occupants.

Even if OSR were to win the interdependence battle, realism about modal structures is

not the only option. For even once we accept that there are modal structures, there appears

to be another question: are these objective, or are they in some sense subjective? That is,

one might concede that there is a necessary connection between the roles and the occupants

of these roles, but one might wonder whether that connection in turn is independent of us

or dependent on us. In other words, here might be an opportunity for articulating realism

and antirealism as a matter of mind-dependence or independence. This fits with the way

in which one might expect to find realism and antirealism about modal claims in general,

where it seems that for any given modal claim there might be a debate whether the modality

is in some sense subjective, or whether it is objective, and in what ways.

Finally there is a debate to be had about the character of the modality involved. Is it out

there in the world, or is it somehow an artifact of our language or perhaps our theories and

their formalism? Here OSR in its contemporary form seems to opt for a robust ‘modality is

out there in the world’ kind of view. This is ironic, given that the more traditional account

of structural realism was a view according to which all the structure was meant to be formal

structure, suggesting a logical/formal view of modality.

OSR is then a position that opts, on the one hand, for interdependence of roles and

their occupants, and on the other hand for the objectivity of such modal structures. We’ve

seen the Humean go for independence on first. But there might be positions that opt for

interdependence and subjectivity, or independence and subjectivity as well. I won’t pursue

these here, since the main goal has been accomplished: we’ve finally come to see what

debate structural realism should be seen as contributing to.


4.7 Conclusion

In this chapter we’ve looked at structural realism, as a view that is often proclaimed in

the philosophy of science as a supposed way of being a realist about science. As we’ve

seen, there is no single thesis of structural realism, but a family of views that in fact make

contributions to different debates, some concerning realism about science, others of a more

broadly metaphysical nature. While the views didn’t do so well in the particular debates,

and don’t form a coherent single position, the fault lies not entirely with structural realism,

but also with the debates themselves. In the final section I suggested that structural realism

is best understood both as an attempt to change the topic of the scientific realism debates,

and as an attempt to articulate a particular position within that new, yet to be established

debate. That accounts for much of the tension within structural realism as it is currently

presented. I propose that we follow structural realism in its attempt to change the debate

and offer some suggestions about how the new debate might go and what different positions

in that debate might look like, without taking a stance myself. As we shall see in the next

chapter, we may in fact have to take more than a single stance, since different modal claims

of our theories might require different analyses.

Chapter 5

The varieties of necessity in physicallaws

5.1 Introduction

In the previous two chapters I suggested that we should take the modal claims of science

as the target for realism/antirealism debates in the philosophy of science. In this chapter

I look at a particular kind of modal claims in science, laws of physics. I suggest that to

understand the modal character of laws of physics, we may need more than one kind of

analysis of modal claims.

In his paper The varieties of necessity (Fine 2002), Kit Fine considers in passing the

idea that we might have to distinguish different kinds of necessity among laws of nature.

The goal of this chapter is to pursue this suggestion further to see whether there are indeed

varieties of necessities among laws of nature and, if so, to see what this tells us about

necessities in general.

I shall look only at laws of physics, for two reasons: one, it is fairly uncontested that

there are laws of physics1, second, if we can find a variety of necessities already just among

1Of course there are philosophers who object to the idea that there are laws of nature at all, including lawsof physics. But insofar as philosophers accept the idea that there are laws of nature or laws of science, theidea is usually that among the sciences, physics is the most likely to contain such laws. For the purposes ofthis chapter I will not enter into the debate as to whether there are laws of nature, but simply assume that atleast as far as physics is concerned, there are laws, and that the modal status of these laws is a topic of interest

115

CHAPTER 5. THE VARIETIES OF NECESSITY IN PHYSICAL LAWS 116

laws of physics, it seems we have already shown that there is more than one kind of ne-

cessity to laws of nature, even independent of the question how laws from other sciences

might differ from those of physics.

Furthermore, there is an independent motivation to look into the possibility of a variety

of necessity in the case of laws of physics, since within the philosophy of physics there is

a debate about the relationship between conservation laws and symmetry principles on the

one hand, and more ‘standard’ laws of physics on the other, and some philosophers have

suggested that conservation laws and other laws might differ in the degree to which they

are necessary.2

I am going to argue that laws of physics indeed come in a variety of necessities, and

that these necessities have to be understood as differing in kind, not just degree. This has

some interesting consequences for the question of how different forms of necessity, such as

narrowly logical, metaphysical, mathematical and natural necessity are to be understood.3

For if it is indeed the case that even within the laws of physics we have to distinguish dif-

ferent kinds of necessities, as opposed to merely different degrees, then overall the varieties

of necessity will have to be accounted for as species rather than degrees as well. Moreover,

distinguishing the different kinds of necessities we may find with respect to laws of physics

may help us shed some light on the strange modal status accorded to laws of nature, some-

times expressed in the paradoxical formula that laws are both, necessary and contingent.4

to philosophers.2Marc Lange, whose views I will look at in more detail below, takes symmetry principles to be meta-laws,

and conservation laws to be more necessary than other kinds of laws.(Lange 2009)3How to distinguish these different forms of necessity, and indeed the question of whether there really

are all these different kinds of necessity, is of course subject to debate. This is of course especially true ofmetaphysical and natural necessity. I will initially use ‘natural necessity’ as a kind of dummy designation forthe kind of necessity we are inclined to attribute to laws of nature, without committing myself either to theidea that there is indeed a distinct form of necessity reserved for laws of nature, or to any particular accountof what it might be. Only in the final section of this paper will I try to argue that we should indeed recognizea distinct form of necessity for laws of nature.

4This is clearly part of Fine’s motivation for raising the issue in the first place.


5.2 Varieties of necessity and the place of laws

Before we attempt to understand the potentially diverse forms of necessity among laws of

physics it will be helpful to get a better grip on the overall options for varieties of necessity

and the place of laws of nature among these different options, that is, on questions of

comparison between logical, metaphysical, and natural necessity.5

At first glance it might seem that there should be two options: monism or pluralism,

that is, either all necessary propositions are necessary in the same way, or they aren’t. As

we shall see, the latter position comes in two, not always clearly distinguished, forms: ne-

cessity comes in degrees, and necessity comes in different species. The difference between

a degree view and species view is that on a degree view, what it means to be necessary for

a proposition is the same for all propositions: they are all necessary in virtue of possessing

a certain feature. But unlike monism about necessity, a degree view of necessity allows

that the feature is possessed in different amounts. On a pluralistic view that takes necessity

to come in different kinds, by contrast, it is not a single feature that makes propositions

necessary, but different ones. Schematically one might put it this way: on a degree view, a

proposition is necessary in virtue of possessing feature N, and distinctions are to be made

between propositions that are more or less N/possess N to a greater or smaller extent. On

a kind view, by contrast, what makes this proposition necessary is possession of feature N,

and what makes that proposition necessary is possession of feature M.

The idea that there is only one form of necessity was especially popular in the first half

of the twentieth century, when logical positivists tried to argue that all necessity was really

logical necessity. Logical necessity, at least in the narrow sense envisioned by the logical

positivists, is too narrow to comprise all propositions we count as necessary.

A strategy for allowing for propositions other than the narrowly logical propositions to

be necessary is to make them necessary relative to some set of truths. If certain proposi-

tions are ‘held fixed’ then certain other propositions will be necessary. This relativization

strategy has been subject to a lot of criticism, and Fine clearly thinks there is no future

in it.(Fine 2002, 255) The general problem with this strategy is that such relative necessity

will seem too easy to achieve and hence trivial or insubstantial, since depending on what we5There are other forms of necessities, such as ethical, deontic, and epistemic necessities, which are often

thought to be yet more species of necessity. I will set those aside for most of the discussion.


are willing to allow into the set of propositions held fixed, any proposition could potentially

turn out to be necessary. If we attempt to define natural necessity in terms of metaphysical

necessity, say by demanding that propositions are naturally necessary iff they follow from

the laws, we need to show how the resulting ‘necessity’ is not just a trivial necessity of the

form: anything that is entailed by some preferred set of propositions is necessary relative

to those propositions.(Fine 2002)

Those trying to defend a form of monism about necessity will hence have to go for

the second strategy Fine considers: restriction or subsumption. Instead of starting with

the most narrow (or strictest) form of necessity, the second strategy is to start with the

broadest notion of necessity and to define the other notions of necessity by restricting them

in some way. The typical suggestion for such a broader notion is of course metaphysical

necessity, but there are those who suggest that the broadest form of necessity is in fact

natural necessity. One such view is Marc Lange’s view (Lange 2009). Lange’s official goal

is to argue for a species view of necessity, not a monism about necessity. But as we shall

see, the details of his arguments suggest that his view is better described as an attempt to

take natural necessity as the broadest notion of necessity and then using the strategy of

restriction to arrive at the other forms of necessity.

These views are examples of what it means to take a degree view of necessity. Like

monism, it is one feature that propositions have to have in order to count as necessary,

but unlike monism, not all propositions have that feature in the same way. For the top

down approach, where we start with the most narrow form of necessity, all other necessary

propositions have the feature at best conditionally. For the bottom up approach, all neces-

sary propositions have the feature unconditionally, but some possess it at greater strength.

As we shall see, Lange’s approach exemplifies this.

5.2.1 Lange’s view

The thesis Lange wishes to defend is that laws of nature possess their own species of neces-

sity, which Lange calls ‘natural necessity’, and that this species of necessity is just as much

a genuine necessity as logical or metaphysical necessity. Still, Lange doesn’t want to say

that natural necessity is just logical or metaphysical necessity, so in addition to establishing


that laws of nature indeed possess a necessity worthy the name, he also needs to show that

that necessity is distinct from logical necessity. I will present two objections to Lange’s

view. First I will show that Lange’s view ends up being a degree-view of the variety of ne-

cessity, not a species view. The main reason for this is that Lange envisions necessities to

stand in a strict hierarchy. Then I will argue that Lange in fact fails to establish the desired

necessity of laws vis-a-vis accidents. The second point is important for the purposes in this

paper since it helps to remind us of the problems with degree views of necessity for coping

with our modal intuitions about laws of nature. Despite my criticisms I think Lange’s view

is worth discussing, because the motivating thought behind much of Lange’s discussion is

very close to the central theme of this paper: the idea that there are varieties of necessity

even among the laws of nature themselves.

Let’s have a look at some of the details of Lange’s view. Lange’s strategy for showing

that laws, in contrast to accidents6, possess a special kind of necessity is to show that laws

stand in a special relationship to counterfactuals. As Lange himself acknowledges (Lange

2009, 12), this strategy has been tried before, but he hopes that his particular approach

can avoid some of the problems of previous approaches. Lange offers a principle he calls

‘Nomic Preservation’ to characterize this relationship:

NP: m is a law if and only if in any context, p � m holds for any p that islogically consistent with all of the n’s (taken together) where it is a law that n(that is to say, for any p that is logically consistent with the first order laws).(Lange 2009, 20)

The idea behind Nomic Preservation (NP) is that laws are invariant under a large range

of counterfactual suppositions, p, and it is this invariance that indicates their necessity.7

Intuitive as this idea seems, there are a number of problems in the details. NP is an attempt

to make the idea more specific. The most important feature (as we shall see) is the idea

that the counterfactual suppositions must be logically consistent not only with the partic-

ular law m whose counterfactual invariance we are ‘testing’, but with all the laws taken

6An accident, for Lange, is “a truth that does not follow from the natural laws (and the ‘broadly logical’truths) alone.” (Lange 2009, 5)

7As a motivating example Lange offers the following: “[E]ven if Bill Gates had possessed 23rd-centurytechnology, he would have failed had he tried to accelerate a body from rest to beyond the speed of light.”(Lange 2009, 8)


together. This is important, since any fact (law or not) is invariant under some counter-

factual suppositions, and no fact is invariant under a counterfactual supposition that is in

contradiction to it.8

Take a paradigmatic ‘accidents’ like “There are four books on my desk.” The counter-

factual supposition “There are exactly three books on my desk.” won’t preserve the original

claim, but many others will. Similarly, if the law under consideration is Coulomb’s law,

then using the counterfactual supposition that “the electrostatic force between two charged

particles is equal to the product of their charges over the cube of their distance” won’t pre-

serve the law. Both for laws and for accidents there will be counterfactual suppositions

under which they remain true, and such that they turn out to be false.

To help distinguish the invariance of laws under counterfactual conditions from the

invariance of accidents under counterfactual conditions, Lange demands that to qualify,

counterfactual suppositions must not only be consistent with the particular putative law m,

but with all the laws taken together. Putting this additional criterion on which counterfac-

tual suppositions have to be taken seriously will make laws more invariant than accidents,

Lange hopes, since he thinks that any counterfactual supposition under which a particular

law fails to be invariant will have to be in conflict with some law or other, even if it isn’t

in contradiction to the particular law we are ‘testing’. Accidents, on the other hand, get no

such special protection, and hence, Lange hopes, accidents will not be invariant even under

some counterfactual suppositions that aren’t in contradiction to them, whereas laws will be.

The strategy, then, is to show that laws, but not accidents, are especially stable or invariant

under counterfactual suppositions by finding ways of excluding as inadmissible any such

8Robert Stalnaker (Stalnaker 1968) and Timothy Williamson (Williamson 2005) suggest, by contrast,that metaphysical necessity could be defined (or perhaps at least ‘captured’) by the idea that a necessaryproposition is one that is true under any conditions, including in particular its own negation, that is, they accept¬p � p as a definition of (metaphysical) necessity. The idea is that a necessary proposition is one who’snegation is impossible, and so the conditional ¬p� p is vacuously true for p if p is necessary.(Williamson2005) This principle would appear to be one according to which it is precisely necessary propositions thathold even under counterfactual suppositions that are in contradiction to them. Whatever the merits of thisproposal, Lange rejects it (Lange 2009, 63,64) as implausible given the way we use counterfactuals withimpossible antecedents, and furthermore as at best capturing the strictest form of necessity, not a varietyof necessities. This second point is clearly correct. For any variety of necessity weaker than ‘truth in allpossible worlds’ ¬p� p is no longer vacuously true, since there will be some possible worlds in which ¬pactually holds, and of course then p doesn’t hold, let alone necessarily. So Lange at least cannot appeal tothis principle in his account of the relationship between laws and counterfactuals.


suppositions that might seem to threaten particular laws.

This of course introduces a problem of circularity, since now laws appear on both sides

of the biconditional in NP. Laws are involved in imposing constraints on which counter-

factual suppositions are going to be allowed, and they do so in a way that favors laws over

accidents, as Lange himself acknowledges (Lange 2009, 26). Before we turn to his attempt

of breaking out of this circularity, we need to introduce another important notion, that of a

sub-nomic fact.

The notion of a sub-nomic fact plays an important role in Lange’s overall account.

A sub-nomic fact is any fact that doesn’t explicitly concern lawhood (or ‘accidenthood’).

Lange illustrates this as follows: “That like charges repel is sub-nomic and a law, though

the fact that it is a law is not a sub-nomic fact.” (Lange 2009, 18) The idea, then, appears

to be that sub-nomic facts are all those facts that are independent of the distinction of facts

into laws and accidents. For the rest of the discussion of Lange’s views I will only be

concerned with sub-nomic facts.

To solve the problem of circularity indicated above, Lange appeals to the idea of “sub-

nomic stability”. “[A] set of sub-nomic truths is ‘sub-nomically’ stable if and only if what-

ever the conversational context, the set’s members would all still have held under every

sub-nomic counterfactual (or subjunctive) supposition that is logically consistent with the

set – even under however many such suppositions are nested.”9 (Lange 2009, 29)

As we’ve seen above, the problem with Nomic Preservation is that the set of sub-nomic

truths any counterfactual supposition was required to be logically consistent with to count

as an admissible test case had been the set of laws, and accordingly the notion of a law had

appeared on both sides of the bi-conditional. Lange now proposes to remedy that situation

by allowing any sub-nomically stable set to play that role. The key claim is of course

going to be that there is no sub-nomically stable set besides the set of laws.10 Laws get

to be special because they form the only sub-nomically stable set, or as we shall see in

a moment, because they form a hierarchy of sub-nomically stable sets. Accidents, so the

9Lange discusses the case of ‘nested’ counterfactuals in some detail, but we don’t need those details forthe discussion here.

10Lange considers the possibility that the set of all sub-nomic facts might also be stable, and modifies hisclaim accordingly to: no non-maximal set of sub-nomic facts is sub-nomically stable. I will set aside worriesabout maximal sets here.


claim, don’t form such sets.

At this point, the two objections I promised at the beginning of this section can be

raised. One objection is that despite his efforts, Lange fails to distinguish successfully

between accidents and laws, that is, the criterion of forming a sub-nomically stable sets is

not enough to single out the laws. The second objection is that Lange at best succeeds in

establishing a degree picture of necessity, not a genuine species picture.

I will begin with the second objection, since it is quite easy to see how Lange ends

up with a degree view of necessity. Lange’s picture of all sub-nomic facts is that of a

pyramid.(Lange 2009, 41) At the top of the pyramid are the broadly logical truths, below

them rank conservation laws, and below them all laws (Lange designates this set as Λ),

below which we find (perhaps) all sub-nomic truths.11 Each rank of the pyramid marks a

sub-nomically stable set, with Λ containing all the other ones (except of course the bottom

one, should it be stable) as sub-sets. This last point is crucial for Lange’s claim that only

the laws form a (non-maximal) sub-nomically stable set, other sub-nomically stable sets

are subsets of the set of laws Λ.

The set of all laws, Λ, contains not only the laws of nature, but also logical truths, meta-

physical truths, and conceptual truths.12 It forms the largest non-maximal sub-nomically

stable set.13 Sub-nomic stability relates to necessity as follows: “[F]or each variety of gen-

uine necessity, the sub-nomic truths possessing it form a sub-nomically stable set – and for

each sub-nomically stable set (except the set of all sub-nomic truths, if it be stable), there

is a variety of genuine necessity where the sub-nomic truths so necessary are exactly the

set’s members.”(Lange 2009, 75) Lange’s idea is that the variety of necessity both among

the laws of nature (in form of the distinction between laws and meta-laws) and the variety

11The set of all sub-nomic truths is again included in case the set of all sub-nomic truths turns out to bestable. Also note that all the truths in this pyramid are meant to be sub-nomic truths.

12Lange doesn’t offer definitions for any of these, but gives examples for each. An example for a conceptualnecessity is ‘All sisters are female.’, an example of a metaphysical necessity is ‘Red is a color.’. Langeuses ‘broadly necessary truth’ to capture any such necessity, plus mathematical, moral, and narrowly logicalnecessity. (Lange 2009, 45).

13This is the idea that makes Lange’s view a version of the strategy of ‘subsumption’, that is, the strategyof taking the largest set of necessities and restricting it to yield the narrower forms of necessity. Lange’scase is especially interesting since he attempts to take natural necessity as the broadest form of necessity, notmetaphysical necessity.


of necessity in general, including at least the distinction between broadly and narrowly log-

ical necessities, is to be captured by this hierarchy of sub-nomically stable sets, where the

smaller the set, the stronger the necessity. This identification of necessity with sub-nomic

stability, together with the view that sub-nomically stable sets are all subsets of Λ, leads to

a view according to which ‘higher’ necessities, that is, smaller sub-nomically stable sets,

always also possess the ‘lower’ level necessities, because they are a subset of all larger

sub-nomically stable sets.14

What makes the set of broadly logical truths and the set of natural laws alike isthat they both form stable sets. It is this commonality that makes both classesof truths “necessary”. Yet they remain distinct species of necessity. The laws ofnature are not as necessary as the broadly logical truths since they rank lowerin the pyramidal hierarchy of sub-nomically stable sets [. . . ]. That is, the rangeof counterfactual suppositions under which the laws must all be preserved, forthe set of laws to qualify as stable, is narrower than the range of counterfactualsuppositions under which the broadly logical truths must all be preserved, forthe set of broadly logical truths to qualify as stable.((Lange 2009, 77), myemphasis)

Lange seems to think that he needs the hierarchy to ensure that laws of nature qualify as

necessary in much the same way as logical truths, but since he doesn’t want to say they are

necessary in exactly the same way, he locates them on lower ranks on that hierarchy. Except

that this leaves us not with a species view of necessity, but a degree view of necessity. For

the only difference between the way in which laws of nature, mathematical truths, and

narrowly logical truths are necessary is in terms of how large the set of counterfactual

suppositions is, with respect to which they have to remain robust. For narrowly logical

truths that set is much larger than for laws of nature. But that means the necessity of laws

is not different in kind from the necessity of logical truths. They are both necessary in

the same way, their necessity can be measured along the same metric, they just possess

stronger and weaker versions of this necessity.15

14Even the smallest of the sub-nomically stable sets will be comparatively large, since in the case ofmathematical truths, they will presumably contain all propositions of the ‘2+2=4’ variety. The differencein size I have in mind is just that since the higher sub-nomically stable sets are subsets of the lower ordersub-nomically stable sets, the latter will be even larger.

15It is interesting to compare the pyramid picture Lange offers with the picture Sandra Mitchell (Mitchell


This might not be a bad thing, but it’s worth noting that Lange’s rhetoric seems to point

in a different direction, since he talks of species of necessity when really what he ends

up with seems like degrees, and in that he explicitly claims that his aim is “to discover

how genuine modalities differ from merely relative modalities without collapsing all of

the species of genuine modality into one.”(Lange 2009, 64) This aim, as we shall see,

proves hard to obtain. For with a degree view of necessity it is very easy to see only

logical truths as being genuinely necessary, with all other claims of necessity being relative

to some condition or other. Logical necessities will be those truths that hold under any

counterfactual suppositions, but all the other supposed necessities only hold under certain

kinds of counterfactual suppositions, that is, conditional on some restrictions as to what

counts as an admissible counterfactual supposition in the first place. Lange hopes to avoid

that ‘conditionalization’ or ‘relativization’ of the necessity of laws by using his notion of

sub-nomic stability. If sets of accidents cannot be sub-nomically stable, then it will still be

something significant about laws, perhaps justifying the label ‘necessity’ even if the picture

of necessities is one of degrees.

This of course brings me to my second objection, namely the objection that Lange does

not in fact succeed in finding a way to distinguish laws from accidents by appealing to sub-

nomic stability. Let’s take a look at Lange’s argument for the claim that only laws form

sub-nomically stable sets. Lange’s claim is that any set of sub-nomic truths containing at

least one accident will prove to be unstable.(Lange 2009, 32) He then considers the example

of a set containing the fact that all gold cubes are smaller than one cubic mile and all of its

logical consequences. The argument, which I reconstruct below, is supposed to show that

this set of sub-nomic facts, which contains at least one paradigmatic accident, is not stable,

and the idea is of course that the same strategy/argument applies to any other such sets.

Premise 1 “[I]t is not the case (in every conversational context) that all of theset’s members would still have been true had Bill Gates wanted to have a

2000) offers for measuring the contingency of laws. On her picture, the continuum spans from ideal lawswhich are contingent, universal, and true, to accidental generalizations. Closest to the ideal laws we findconservation laws, at the bottom we find a generalization like ‘All the coins in Goodman’s pocket are madeof copper.” Two observations are important: first, the measure is a continuous, and second, it’s a measure ofcontingency not necessity. But what Lange’s and Mitchell’s picture say about laws, and even how they orderthem, is quite similar. That shows that stratification by itself neither shows that laws are necessary as opposedto contingent, nor that we need a species rather than a degree picture.


large (exceeding one cubic mile) gold cube built.”

Premise 2 “Stability requires the set’s members all to be invariant under ev-ery sub-nomic counterfactual supposition that is logically consistent withthem all (taken together).”

Premise 3 “The only way for this set to be stable, despite failing to be pre-served under the supposition “Bill Gates wants to have a large gold cubebuilt,” is for that supposition to be logically inconsistent with the set.”

Conclusion A “To be stable, then, the set’s members must logically entail“Bill Gates never wants to have a large gold cube built,” so that the sup-position that Bill Gates wants to have a large gold cube built is logicallyinconsistent with the set’s members.”

Conclusion B “Since the set contains every sub-nomic logical consequenceof its members, the set must contain the fact that Bill Gates never wantsto have a large gold cube built.”

Premise 4 The problem can be generalized.16

Conclusion C “To prevent the set’s behavior under those suppositions fromrendering it unstable, further claims q must be admitted into the set.”

Premise 5 “The process snowballs until the set contains every sub-nomic truth.”

Conclusion E “Thus, no nonmaximal set containing accidents possesses sta-bility.”(Lange 2009, 32,33)

There are several problems with this argument, the biggest of them the question of how

to support premise 1 without begging the question. Support for that premise is rather slim.

It seems that Lange is just appealing to our intuition as to what would happen if Bill Gates

really wanted a large gold cube. It is not hard to see what that intuition is: if Bill Gates

really wanted a large gold cube, nothing would stop him from building one, and of course,

once he’s built one, it is no longer true that all gold cubes are smaller than one cubic mile.

What is crucial to make this example work is the intuition that nothing would stop him,

and accordingly that if he wants the gold cube, he will get it. If it turned out that Bill Gates

may want a large gold cube, but wouldn’t succeed in building one, no matter how much he

16Lange argues for this point by suggesting that even the improved set containing the claim that Bill Gatesnever wants a large gold cube is unstable, since it is threatened, for example, by the counterfactual suppositionthat Melinda Gates wants a large gold cube. To counter this threat to stability, the same procedure has to beemployed, and so for any suppositions of this form.


wanted to, the counterfactual supposition would not threaten the sub-nomic stability of the

set we are testing.

One worry is that the reason behind the intuition that nothing would stop Bill Gates

from building a large gold cube makes implicit reference to the laws of nature, and thereby

begs the question as to whether only laws form sub-nomically stable sets. The concern is

that what drives the intuition that, if Bill Gates wanted a large gold cube, he could build

one, is the thought that there is no law of nature that would get in the way. This intuition

will be especially salient if we compare it to the case of Bill Gates wanting a large uranium

cube or a perpetual motion machine.17 In the latter two cases our intuition will point in

the other direction. There is something in the way of Bill Gates getting his wish: a law

of nature. But if this is the intuition behind premise 1, it is hard to see how this argument

is any less circular or problematic than the principle of Nomic Preservation from before.

For now it seems that judgments about which sets are sub-nomically stable are implicitly

based on laws, even if they don’t explicitly mention them, and hence we haven’t really

made any progress over the problems with Nomic Preservation.18 This is compatible with

it in fact being the case that only laws form sub-nomically stable sets, but it means that this

argument doesn’t add much to that claim.

The other worry is that premise 1 might simply not be true. We may agree that there

is no law of nature standing in the way of Bill Gates’ wish, but that shouldn’t lead us to

the conclusion that nothing else does. While the gold cube, unlike the uranium cube, is

not going to chain react upon assembly, there might be problems even before anybody

gets to the point of building the cube, namely in acquiring the requisite amount of gold.

Should we expect there to be problems? Well, consider the following estimate: “Simi-

larly, it is estimated that all the gold ever mined in the world (160,000 tonnes as of 2007),

could be placed in a single cube roughly 60 ft. on a side, with a value of $3.68 trillion.”

(source: http://www.usagold.com/reference/properties.html) If that’s even roughly accu-

rate, it seems that there are in fact quite important constraints on how much gold could be

17The first example is of course implicit in the background, given the traditional supposed difference be-tween the fact that all gold cubes are smaller than one cubic mile, and the fact that all uranium cubes aresmaller than one cubic mile. The second example is explicitly suggested by Lange.

18Recall that Nomic Preservation was problematic because it seemed to circularly prefer laws over otherfacts.


assembled, even if Bill Gates wanted it really badly and we disregard market pressures that

would drive up the price of gold per ounce should anybody attempt to buy up all the gold

ever mined on this planet. So premise 1 might just be false. It might in fact be the case that

every member of the set containing the claim that all gold cubes are smaller than one cubic

mile remains true, even if Bill Gates really wants a large gold cube.19

At the bottom of the problem is that what looked like a neutral criterion that would allow

us to characterize what’s special about laws with respect to their relationship to counter-

factuals in a way that doesn’t already presuppose that laws are more ‘robust’ or ‘resilient’

to counterfactual suppositions than accidents turns out not to be neutral after all. Lange’s

whole strategy was to explain the necessity of laws in terms of their special relationship

with counterfactuals, which he characterized as a strong form of invariance under coun-

terfactuals. But if that invariance is only to be had if we already bring to bear, explicitly

or implicitly, the special modal status of laws, we haven’t characterized that special modal

status in terms of their relation to counterfactuals. Instead it begins to look as though the

relationship between laws and counterfactuals on the one hand, and accidents and coun-

terfactuals on the other is precisely not useful in characterizing what it means for laws to

be necessary, since the only thing we can use to distinguish laws from accidents is to give

priority to considerations of laws of nature over considerations of other factors. But what

we wanted in asking about the special modal status of laws of nature (if they indeed have

such special status) was, if anything, a justification for giving priority to laws in this way.

The problems with premise 1 also affect other parts of the argument. For now we might

want to question premise 3, since it seems that what needed to be added to the original19Lange is going to point out that of course there are conversational contexts in which sets containing

accidents are sub-nomically stable, but that such sets are not stable in all contexts, whereas sets containinglaws are. I have to confess I don’t see how laws get to be exempt from that problem. The conversationalcontext according to which Bill Gates, just like the rest of us, is in fact constrained by the amount of goldactually available seems a perfectly natural one – there’s nothing artificial about picking that fact as a relevantconsideration when it comes to the question whether Bill Gates’ wanting to build a large gold cube wouldlead the set containing the sub-nomic fact ‘All gold cubes are smaller than a cubic mile’ to be unstable. Andof course for any set of sub-nomic facts save the set of logical truths there are going to be conversationalcontexts in which the sets are no longer stable. The whole idea behind a view like Lange’s is to take laws tohave a necessity different from that of logical truths. But unlike logical truths, which are true in all possibleworlds, there are some worlds in which laws aren’t true, and accordingly there will be some conversationalcontexts in which we use counterfactual suppositions that contradict laws. Lange will try to insist that lawsstill take priority over accidents, but we don’t need to follow him there, or think that this marks an importantdifference, even if it were true.


set (containing the fact that all gold cubes are smaller than one cubic mile) was not a fact

that was logically inconsistent with Bill Gates wanting a large gold cube, it was a fact that

suggested that it would be impossible for Bill Gates to build a large gold cube, even if he

really really wanted to. Indeed, consider this set, and add to it the fact that “all the gold

ever mined in the world could be placed in a single cube roughly 60 ft. on a side, with

a value of $3.68 trillion.” That set seems like an excellent candidate for a sub-nomically

stable set containing an accident. If we wanted to be on the safe side, we could add the

fact that gold production is not going to increase significantly over the next hundred years.

(A pretty safe bet.) All of these facts would seem like paradigmatic accidents, but taken

together they seem sub-nomically stable.20 It is simply not the case that the only way to

make stable a set of sub-nomic facts is by adding facts that will disqualify the original

problematic supposition by making it logically inconsistent with the set.

Accordingly we should also question the claim that the problem can be reiterated so as

to lead, eventually, to the inclusion of all sub-nomic facts. Even in the original argument,

setting aside the worries raised about some its premisses, it wasn’t clear that this would be

the case. In the initial set-up all that seemed needed was the fact that nobody ever wants

to have a large gold cube built. Of course one might worry that somebody could come up

with a different counterfactual supposition that would once more threaten the set, but from

that it doesn’t follow that we would in the end have to include all sub-nomic facts.

Lange’s argument to the effect that the only sub-nomically stable set is the set of all

laws fails. This means that Lange has not succeeded in establishing a distinction between

laws and accidents on the grounds of their different relationship to counterfactuals. Since it

was their sub-nomic stability that was supposed to make laws bearers of genuine necessity,

if laws are not unique in being sub-nomically stable, sub-nomic stability fails as a criterion

for necessity. None of this shows that laws aren’t necessary, all that I’ve shown is that the

necessity of laws, should they be necessary, cannot consist in their sub-nomic stability.

I’ve discussed Lange’s approach in detail, since it is motivated (at least in part) by the

idea that laws might posses different forms of necessity. But on Lange’s view, this differ-

ence in necessity turned out to be a matter of degree, not kind. In light of the discussion

of Lange’s view we can now say somewhat more clearly what distinguishes degree views

20Notice though that the fact itself was perfectly sub-nomic - it didn’t concern lawhood.


of necessity from (genuine) species views of necessities. On a degree view, all forms of

necessity are connected, and ordered according to strength. That is, on a degree view, we

can pick any two necessary propositions and compare their relative strengths. On a genuine

species view of necessity that’s not going to be possible. A genuine species view of neces-

sity is one where in attributing to a proposition a necessity of species A, and attributing to

a different proposition a necessity of species B, we aren’t thereby in a position to order the

two propositions according to the relative strengths of their necessities. It is not as though

we thereby know which of the two propositions is ‘more’ necessary. Of course, one big

concern for a species view of necessity will be to explain why all of these different species

should indeed be species of the same genus. Lange errs in that his species aren’t really

species – to count as a species view, at least two species have to be on the same level but

different, they cannot just form a hierarchy.

It also shows that even the restriction approach to get distinct forms of necessity out of

a monist/degree view is threatened by the problem of triviality: even accidents can seem to

qualify as necessary on the criterion Lange employs.

5.3 Some examples of varieties of laws

Let’s go back to the idea we started out with, the idea that there might be a variety of

necessity among the laws of nature, or more specifically, among laws of physics. Fine’s

suggestion that there might be varieties of necessity among laws was motivated by the idea

that not all natural necessities are also metaphysical necessities. Recall that Fine thinks of

metaphysical identities as those propositions that are true in virtue of the identity of some

thing.21

Fine proposes that we should consider whether there might not be different kinds of

necessity at play when it comes to laws of nature: perhaps some of them are metaphysically

21This notion of metaphysical necessity is more specific than the notion used by Lange. I will focus onthe idea that a necessity is metaphysical if it is an identity of some kind. There is of course a debate asto whether such identities always, or at least sometimes hold in virtue of our theories or language, that is,somehow depend on us, as opposed to the nature of things in the world. While Fine is very sympathetic to thenotion of ‘real definitions’(Fine 1994), I will remain neutral with respect to this question. In this chapter I’monly interested in investigating whether some modal claims in physics hold in virtue of identities/definitions,whereas others don’t, not in the source of the necessity of such claims.


contingent, whereas others are metaphysically necessary. He suggests that this idea is

intuitively plausible: “That electrons have negative charge, for example, strikes one as

metaphysically necessary; it is partly definitive of what it is to be an electron that it should

have negative charge. But that light has a maximum velocity or that energy is conserved

strikes one as being at most naturally necessary. It is hard to see how it could be partly

definitive of what it is to be light that it should have a given maximum velocity, or partly

definitive of energy that it should be conserved.” (Fine 2002, 261)

For Fine, metaphysical necessity is “the sense of necessity that obtains in virtue of

the identity of things”(Fine 2002, 254). Accordingly, that electrons have negative charge

is a better candidate for metaphysical necessity than that there is a maximum speed, or

that this speed should be the speed of propagation of light. Suppose we accept this as

an appropriate gloss on metaphysical necessity. Should we follow Fine’s assessment that

other laws of nature (or physics) are not metaphysically necessary in that sense? Is there a

different sense of necessity that we can instead attribute to them? Just how should we think

about and assess the different forms of necessity we might encounter in laws of nature, and

what does this tell us about natural necessity and its relation to other forms of necessity? I

will look at a couple of examples of laws of physics, which are by no means exhaustive, but

which I believe nevertheless offer a good sample of different forms of necessity we might

encounter among laws of nature.

5.3.1 Electrons have negative charge

Let’s begin with the example Fine offers as a case of metaphysical necessity: ‘Electrons

have negative charge.’. While it is a bit odd to say that this is a law of physics, Fine’s

intuition that this is partly definitive of electrons seems correct. If we found a particle that

had the same mass as electrons and behaved like electrons in every other way, except that

it was positively charged, we wouldn’t call it an electron, and, no, we wouldn’t call it a

schmelectron either, we call it a positron. That’s pretty good evidence for taking ‘being

negatively charged’ to be partly definitive of what it means to be an electron.

Following Fine that means that the necessity involved in this case is metaphysical ne-

cessity: it is part of the identity of this kind of thing, electrons, that they are negatively


charged. Particles that aren’t negatively charged, even if everything else about them is the

same, aren’t electrons. So Fine is right, it is metaphysically necessary that electrons have

negative charge.

Notice also, though, that the proposition that electrons have negative charge isn’t ex-

actly a characteristic law of physics. Indeed, we may wonder whether we should classify it

as a law of physics at all. ‘Electrons have negative charge’ seems a lot more like ‘sisters are

female’ than like ‘F = ma’. So this might in fact not be a case of a law of physics that is

metaphysically necessary, but an example of a metaphysically necessary truth that happens

to be about certain kinds of particles, but is not thereby any more a law of physics than the

proposition that sisters are female is a law of human biology. In which case we haven’t yet

found a law of nature that’s metaphysically necessary, we’ve just found that there may be

metaphysical necessities that hold for kinds we take an interest in when we do physics. So

far at least it is not clear at all what the relationship between such metaphysical necessities

and laws of nature might be.

5.3.2 Conservation laws

Let’s look at a different candidate for natural laws, then, conservation laws. As I suggested

at the beginning, philosophers of physics often treat conservation laws as ‘special’, so let’s

see what we should say about their modal status. Such laws, Fine suggests, are at best

naturally necessary, but not metaphysically necessary, since it is “hard to see how it could

be partly definitive of energy that it should be conserved”.(Fine 2002, 261) I take it Fine

means for this to hold of other conserved quantities, such as momentum and charge, as well.

In what sense, if any, are conservation laws necessary? To deny that conservation laws are

metaphysically necessary means to deny that it is part of ‘the definition’ or a matter of

the ‘identity’ of the various conserved quantities, such as energy, angular momentum, or

electric charge, that they are conserved. But even if this might ‘intuitively’ seem right, it is

not clear that it is actually correct.

In contemporary physics conservations laws are closely tied to symmetry principles via


Noether’s theorems22, and the exact relationship between symmetry principles and conser-

vation laws and their status is a matter of much dispute in the philosophy of physics.(Brading

and Castellani 2003) Of course, the topic is a very rich one, and I will not have the space to

address all interesting aspects of the debate. Some points of the dispute bear quite directly

on the question of the modal status of conservation laws, though, and those will be the ones

I will be focussing on here. I will try to keep technicalities to a minimum, many of the

details can be found in the literature.23

In slogan form, Noether’s first theorem is often said to state that for each symmetry of

the equations of motion, there is a conserved quantity. This is of course just a slogan, and

we need to be more careful in stating just what the theorem says. In general, Noether’s

theorems apply only to systems for which the equations can be given a Lagrangian formu-

lation. The symmetries are continuous symmetries, taking solutions of the Euler-Lagrange

equations into solutions.

What Noether’s first theorem shows is that for a group of continuous global symmetry

transformations (given certain conditions) there are conserved (Noether-) currents.24 Given

certain boundary conditions, the existence of conserved currents implies that there are also

conserved quantities, commonly called ‘Noether-charges’. Noether-charges are conserved

as a matter of necessity, or if you like, by definition. Part of what it is to be a Noether-charge

is to be conserved.

None of this yet shows that energy, angular momentum, or electrical charge is con-

served by (metaphysical) necessity. For in order to show that any of the usual conserved

quantities are conserved we have to show that the conditions for the application of Noether’s

theorem are fulfilled, and (hence) that the quantity of interest turns out to be a Noether-

charge. In the classical case this is done by showing that the Euler-Lagrange equations of

motion hold. It is because the Euler-Lagrange equations hold that there is a conserved cur-

rent, and it is only using a further assumption about boundary conditions that we can make

the inference from the conserved current to the conserved quantities.(Brown and Holland

22Only certain kinds of symmetries (continuous ones), and not all conservation laws (only exact ones).23See especially (Brading and Brown 2003) and (Brown and Holland 2004).24Noether’s first theorem as she puts it reads: “If the integral I is invariant with respect to a Gρ, then ρ

linearly independent combinations of the Lagrange expressions become divergences – and from this, con-versely, invariance of I with respect to a Gρ will follow. The theorem holds good even in the limiting case ofinfinitely many parameters.” (Noether 1918)


2004) The conserved quantities of classical mechanics are Noether charges only because

the classical equations of motion are what they are.25 But whether or not the classical equa-

tions of motion hold is something that needs to be established, and there are problem cases.

In particular, it is certainly not a priori that they hold.

More generally, while conservation laws can be derived from the equations of motion,

unless we are prepared to say that the latter are metaphysically necessary, we have no rea-

son to think that classical conservation laws are metaphysically necessary. Given what

the equations of motion are, and that they hold where they do, it is indeed necessary that

the conservation laws hold, but that’s just a conditional necessity. That it is a mathemat-

ical necessity is shown by Noether’s theorem.26 That these are the correct equations of

motion, however, is a completely different matter. As far as deriving conservation laws

using Noether’s first theorem is concerned, it seems then we should conclude that while

it is a metaphysical necessity that Noether-charges are conserved, it is not a metaphysical

necessity that energy, linear momentum, or angular momentum is conserved.

Matters are somewhat different, however, if we take the relevant Noether theorem not to

be her first, but her second theorem, a possibility that has been much discussed in the recent

literature in the connection with the role of gauge symmetry.(Brading 2002) Noether’s

second theorem is concerned with local symmetries, not global symmetries.27

Again, there are many technical details, discussed for example in (Brading and Brown

2003) and in (Brading 2002), that I will skip over here. The key point for our purposes is

that the conservation of energy in general relativity or the conservation of electric charge

in quantum electrodynamics can be derived using local symmetries28 and Noether’s second

25To put this in terms of conserved quantities being Noether-charges may seem like a rather idiosyncraticway of putting this point. Doing so brings out the connection between ‘kinds’ and ‘laws’ that is at stake inthe broader debate of this chapter.

26Or shown most clearly by Noether’s theorem. That there is a connection between symmetries of theequations of motion and conservation laws had been observed before.

27Noether’s second theorem, which she proved in the same paper, is stated as follows: “If the integral Iis invariant with respect to a G∞ρ in which the arbitrary functions occur up to the σ-th derivative, then theresubsist ρ identity relationships between the Lagrange expressions and their derivatives up to the σ-th order.In this case also, the converse holds.”(Noether 1918) Notice that in this theorem, arbitrary functions occur inG.

28In the case of local symmetries, the variation takes continuously varied parameters, as opposed to con-stant parameters. Compare footnote above.


theorem.29 The problem is that if conservation laws are derived in this way, the conservation

of electric charge, for example, seems to follow merely as a matter of mathematics, with

no input from the ‘real’ physics, that is, from the equations of motion.30 If this is so,

it would seem as though the conservation of electric charge in quantum field theory and

the conservation of energy in general relativity hold as a matter of necessity, and indeed

‘metaphysical’ necessity as we’ve been using the term, since they simply hold in virtue of

mathematical identities, not because the equations of motion.

Katherine Brading and Harvey Brown (Brading and Brown 2003) try to resist the con-

clusion that conservation of electric charge and energy respectively are ‘merely’ mathemat-

ical, and point out that we don’t usually treat the conservation of electric charge as having

‘no physical significance’. As a solution to the problem they offer the idea that what looks

like a mathematical identity can be treated as physically significant (as opposed to merely

mathematical) as long as we treat the equation as relating two different fields. The identity

becomes a constraint on what the field equations can look like, and that is physically sig-

nificant. So they suggest that we can take conservation laws as physically significant even

where it is possible to derive them as a matter of pure mathematics.

I have nothing to add to the technical details of their discussion, but it seems to me that

two things are striking, and at least one of them is crucial for the discussion here. First,

Brading and Brown initially almost equate ‘empirical’ and ‘having physical significance’,

which is then contrasted with the merely mathematical identities that can be derived from

Noether’s second theorem. But this contrast doesn’t quite capture what seems problematic

about deriving conservation laws in this way. For how we come to know that a particular

conservation law holds is one thing, but what seems to be the deeper issue is in virtue of

what the conservation law holds.

In the case of Noether’s first theorem and global symmetries, it was clear that the con-

servation laws hold in virtue of the equations of motion. We shouldn’t think that this means

we came to know conservation laws in that case empirically, that is, through observation.

Conservation laws in classical physics hold in virtue of the equations of motion, which

29For discussion of the latter point and its historical context, see especially (Brading 2002).30This potential problem is the main concern of (Brading 2002) and (Brading and Brown 2003), but the

debate goes back to the early days of the general theory of relativity and Noether’s theorems.


themselves don’t hold in virtue of mathematics alone. In the case of using Noether’s sec-

ond theorem and local symmetries to derive the relevant conservation laws, that no longer

seems to be the case. For now we don’t require the equations of motion to hold in order for

the conservation laws to be derived. Now it seems that the conservation laws hold just in

virtue of mathematics. That is the disturbing feature of theories with local symmetries and

the use of Noether’s second theorem that is made possible through this. It is not a matter

of a priori vs. empirical knowledge of the conservation laws, but a question about in virtue

of what the conservation laws hold. Brading and Brown may be right that whether we take

conservation laws to be physically significant is a matter of further interpretation, but no

matter how we interpret the resulting law, it still looks like it holds in virtue of mathematics

alone.

This brings me to the second observation, which in many ways is more crucial to the

point of this section. If the conservation laws follow as a matter of pure mathematics, it

all of a sudden seems that conservation laws might be metaphysically necessary after all.

For whereas in the classical case we needed particular physical laws (the Euler-Langrange

equations of motion) to hold in order for the conservation laws to hold, for the modern field

theory case this no longer seems needed.

Indeed, if we follow modern parlance and call ‘charges’ the ‘generators of the local

symmetry groups’, (Martin 2003) it seems as though we should say that electric charge, for

example, is conserved in virtue of what charge is, not in virtue of something else, like the

equations of motion. Electric charge is conserved because it is a generator of a particular

continuous symmetry group, U(1), and the color charge of quarks is conserved because it is

the generator of a different symmetry group, S U(3). This is something that shows up with

theories with local symmetry principles, which are characteristic of modern field theories,

and it marks a genuine change from classical mechanics and its conservation laws. What

has changed, however, is not their epistemic status, but their modal status: they now seem

to be metaphysically necessary, they hold as partly definitive of the conserved quantities. So

it looks like in modern physics certain conservation laws hold as a matter of metaphysical

necessity.


5.3.3 What about other laws?

So far then we’ve found some metaphysically necessary laws, conservation laws in mod-

ern field theories, and if you count them as laws, also propositions like ‘electrons have

negative charge’. But while conservation laws certainly count as laws, they, too, might

seem somewhat atypical, especially in the case of modern gauge theories. What are typical

laws of physics? Newton’s law of gravitation, Coulomb’s law, or Maxwell’s equations,

Schrodinger’s equation . . . . Are these laws necessary, and if so, are they metaphysically

necessary?

The standard argument for their metaphysical necessity is that in order for the laws

to be different, the properties they hold of would also have to be different.31 The idea is

to model the argument for the necessity of laws of Kripke’s argument for the necessity of

propositions like ‘Gold is an element’.

A first step is to take properties as causal or dispositional, rather than categorical.32 The

idea is that properties are individuated by the causal powers they contribute to, and that

the collection of the causal powers a property contributes to is the essence of that property.

This establishes a necessary connection between properties and causal powers.33 A second

step is to take laws as relations between properties. If laws are then taken to describe the

causal features of things, it follows that if the laws are to be different, that means that the

causal features of things are different, and if the causal features of things are different,

that means that in this case the properties also have to be different, since properties are

individuated by the causal features they contribute to. That means that the laws could not

have been different without the properties having been different. Shoemaker (and others)

want to infer from this that laws are metaphysically necessary (or that causal necessity is

just a species of metaphysical necessity), since they hold in virtue of essential features of

properties. If the properties contribute to the causal powers they contribute to in every

possible world, since the causal powers they contribute to are essential to them, then the

31Versions of this argument date back to Shoemaker’s causal theory of properties (Shoemaker 1980), butit has more recently been defended by Brian Ellis under the heading of ‘scientific essentialism’ (Ellis 2001).

32This needs to be qualified. Only some defenders of this view treat properties as being causal in the senseof being reducible to the causal powers or features of things, whereas more careful proponents (Shoemaker1998) suggest only that the causal powers they contribute to individuate the properties by being essential tothem, without the properties being reducible to them.

33Recall also the discussion at the end of chapter three on structural realism and properties.


laws that are relations between properties are also the same in every possible world.34

It is the final step of this argument that Fine objects to. He acknowledges the Kripke

inspired argument that our ability to image that different laws could have held (such as

an inverse cube law for gravity as opposed to an inverse square law) shouldn’t lead us

to conclude that the laws aren’t metaphysically necessary, that is, the ‘conceivability’ of

different laws holding shouldn’t be mistaken for the (metaphysical) possibility of different

laws holding. What we were imagining when we thought we were imagining that a different

law holds for gravity really involved a property other than mass (customarily designated

‘schmass’), since the property mass and the laws that hold for it are necessarily connected.

Fine responds that while this may be so, it is unclear that we should conclude from this that

all natural necessities are metaphysically necessary. For if the causal account of properties

is right, then it would now seem to be a natural necessity that there is no schmass in our

world, but mass. But even if this is a natural necessity – because things in this world in

fact work according to the inverse square law of gravity, not the inverse cube law, thereby

‘excluding a certain sort of behavior’(Fine 2002, 258) – this natural necessity is not a

metaphysical necessity. For it seems (metaphysically) possible that there could have been

schmass rather than mass.

One way of putting Fine’s objection to the view that laws are metaphysically necessary

is that what the essentialist about laws establishes, if anything, is a necessary connection

between (certain) properties and (certain) laws. It may be essential to certain properties that

they figure in particular laws, and so it may be that a case where the laws would have been

different is really a case where the properties are different. But while there may be such a

necessary interdependence between properties and laws, this doesn’t show that either the

properties are (metaphysically) necessary, or that the laws are (metaphysically) necessary.35

Fine suggests that instead we should take this as an indication that there is a diversity

of necessity, according to which some laws are metaphysically necessary, whereas other

laws are metaphysically contingent, but perhaps have natural necessity. Some laws may

34Shoemaker is very careful to make this argument exclusively for causal features and causal laws, butothers, like Brian Ellis, speak more generally of dispositional properties, and laws of nature in general.

35Recall what I said about structural realism at the end of last chapter. The causal theory of properties mayshow that there is an interdependence of the elements of a structure and the roles they play in that structure.But that doesn’t yet tell us what the modal status of the structures themselves is, or whether should regard themodal structures as objective or subjective.


still come out metaphysically necessary, and as we’ve seen above, this may hold not just

for propositions like ‘electrons are negatively charged’, but in some circumstances also

for conservation laws. But other laws, Fine suggests, should be regarded as metaphysically

contingent – they could have been different. This contingency was traditionally captured by

the idea that the laws could have been different, and in light of the essentialists’ argument

it will have to be captured by the idea that different properties could have been actual.

If laws could have been different, does that mean they have a special kind of neces-

sity, natural necessity, that is somehow weaker than metaphysical necessity? If we take

it that we should think of laws as metaphysically contingent because we need a sense in

which different laws would have been possible, why not just opt for the view that laws are

contingent?

Primarily because laws seem to involve some kind of funny modality, given that we

use laws to rule out certain happenings as impossible, and that we take laws to ‘support

counterfactuals’. Laws seem to be necessary in some sense, although most of them are

not metaphysically necessary. The natural suggestion had been that laws are naturally

necessary, although it remained unclear what exactly one should say about this natural

necessity. Fine’s suggestion seems to be that natural necessity is just sui generis. But while

we might agree with Fine that whatever necessity we attribute to laws has to be different

from metaphysical necessity, we may want to know a bit more about that necessity. Even

if we can’t give a definition in terms of some other kind of necessity, it would still be nice

to understand somewhat better what that supposed necessity does, and how it relates or

doesn’t relate to other forms of necessity.

5.4 The necessity of laws

Here’s an attempt. We take laws to have some kind of necessity because they seem to

contain ‘musts’, that is, they demand or (more frequently) prohibit or ‘exclude certain

kinds of behavior’. Demands for, or prohibitions of, certain kinds of behavior are rules.

So maybe we should interpret the ‘musts’ in laws of physics as indication of their status as

rules, not as a sign for their metaphysical necessity. Why should we think so?

One way to see the appeal of this way of thinking about the necessity of laws comes


from the lessons we can learn from the problems with Lewis’ Humean Supervenience ac-

count of laws. It is instructive to see why Lewis’ approach runs into trouble, since Lewis’

view is commonly taken as the view that leaves laws more ‘contingent’ than any other view

out there. It is often objected to Lewis’ view that it leaves laws ‘too contingent’, and since

we’re having trouble understanding the sense in which laws are supposed to be necessary,

it will be interesting to see what is missing from Lewis’ view. What is missing from Lewis’

view is perhaps just what we need to characterize the elusive necessity of laws.

Lewis’ account of laws is a version of the regularity account of laws. For Lewis, laws

are contingent generalizations, not necessary truths about the world.(Lewis 1973, 73) But

not just any contingent generalization will count as a law, only those that are part of our

best system for organizing the facts in our world. The best system is the one that strikes

the best balance between simplicity and strength. Lawhood then means membership in the

best system for organizing the facts in a particular world, and since what the best system

is may differ from world to world, a generalization may be a law in one world, but only

a generalization in a different world. Unlike Lange, Lewis doesn’t think that laws always

take priority in our counterfactual reasoning, but any such priority has to be understood

in terms of the closeness relation between possible worlds. Worlds in which the laws are

the same will generally be closer than worlds where they are different. Whether or not

something is a law in a given world depends on the (collective) facts in that world, since

whether a generalization is part of the best system for organizing the facts in that world

depends on what the facts taken together are like. Laws, then, supervene on the particular

facts, that is, facts about the distribution of local qualities.(Lewis 1994)

Counterexamples to Lewis’ account typically take the form of presenting two possible

worlds in which everything that actually happens is the same, but in which different laws

hold.36 The plausibility of these examples is taken to show that, pace Lewis, laws don’t

supervene of facts. For on Lewis’ view every difference in laws has to be traceable to a

difference in the (actual) facts. That’s the whole point of supervenience. But if the facts in

the two worlds are the same, and yet we can make sense of the difference in laws, it seems

that laws do not supervene on facts. There can be differences in laws that aren’t traceable

36A frequently discussed counterexample of this kind is the ‘mirror argument’ offered by John Car-roll.(Carroll 1994)


to differences in facts.

It is interesting that Helen Beebee(Beebee 2000), in an attempt to defend the Humean

against this particular line of objection, ends up denying that the two worlds in fact differ

in laws. She claims that the counterexample to Humean Supervenience simply begs the

question against the Humean, since it presupposes a ‘governing conception of laws’ which

is precisely what the Humean wants to deny. But it is hard to see that this by itself is

going to suffice as a counterargument, since what is at issue is precisely what the status

of laws is: Humean or, if you like, ‘governing’. What the counterexample was meant to

suggest is that the Humean conception is worse off, since it can be made intelligible, as

the example illustrates, that laws may differ even where facts do not. It is true that the

counterexample includes an appeal to our intuitions, but that is not by itself illegitimate or

question-begging. What the counterexample shows is that it seems perfectly intelligible to

have laws differ where facts don’t differ. If Humean Supervenience is correct, that shouldn’t

even be intelligible. All that is needed is to make plausible the intuition that facts can be

the same when laws differ, for it shows that Humean Supervenience fails to capture what

laws are.

Nor is the problem just a matter of our intuitions going against Humean Supervenience.

We distinguish the two possible worlds that agree on the facts and disagree on the laws by

considering what would have happened, counterfactually. I argued above in the discussion

of Lange’s view, that counterfactual resilience does not distinguish laws from accidents.

So that we (often) keep laws fixed when reasoning about counterfactual situations does

not make laws necessary (in contrast to accidents). But Beebee’s proposal seems to make

it impossible to keep laws fixed across counterfactual situations, since now it seems that

every time we change the facts, we also change the laws. That’s going too far, because if

we adopt this strategy, it seems that we can no longer use laws to reason counterfactually

at all. Lewis original proposal, however, was meant to allow us to do just that.37 Even

if being used in counterfactual reasoning is not a distinguishing feature of laws, it is still

something that we thought we could do with laws, and if answering the counterexample to

37Again, Lewis doesn’t suggest that laws always take priority. But in reasoning about other possible worlds,other worlds where the laws are the same are generally closer to the actual world, than those where the lawsare different. These will be worlds where the laws are the same, but some of the facts are different. So onLewis’ account, worlds can definitely differ in fact while being the same with respect to the laws.


save Humean Supervenience requires giving up on that idea, that certainly counts against

Humean Supervenience as an account of laws. Supervenience originally required that if

there is a difference in laws between two possible worlds, that means there has to be a

difference in facts. But that’s not to be confused with the idea that any difference in fact

entails a difference in laws. The latter is not required by supervenience.

The problem with Lewis’ approach as brought out by this style of counterexample is

analogous to certain problems we observe in the case of rules. What makes these cases

seem intelligible, and indeed plausible, is that we can in general imagine two rules that

agree for any actual cases, but diverge for some of the non-actual cases. A finite pattern

might have been generated according to a vast number of different rules, that is, a large

number of different rules will yield results that match the finite sample of cases. So as long

as we are dealing with finite worlds, it seems that different laws holding in these worlds are

always possible, even in cases where the worlds agree on the facts.

The problem is that if we give up the idea that laws act as rules and take them to be

just facts, we can then no longer use laws in the way we do, i.e., as excluding certain types

of behavior or as supporting counterfactuals. It is because rules require certain kinds of

behavior (or as we shall see in a bit: certain solutions) for cases that haven’t yet happened,

or did not in fact happen, that laws understood as rules can be used to predict and reason

counterfactually.38 If laws are rules, then accounts like Humean Supervenience will fail

because there is always going to be more to the rule than the sum of actual cases, and

hence counterexamples of the ‘mirror-argument’ variety can be produced. If the Humean,

in response, gives up on the idea of laws as rules, and makes laws fully dependent on the

facts, laws can no longer be employed to reason counterfactually or rule out happenings as

impossible.

Suppose we understand the necessity of laws as marking their status as rules. Meta-

physical and logical necessity are different. If something is true in all possible worlds, in

virtue of being a logical truth or a mathematical or metaphysical identity, then it will hold

in any possible circumstance, but it is not clear that it functions as a rule. Both logical and

38This doesn’t explain why any particular laws are successfully used in this way, it only explains why theyhave the right ‘status’ to be used in this way.


metaphysical necessities were characterized in terms of what ‘makes them true’: metaphys-

ical necessity was understood as true in virtue of the identity of things, logical necessity as

true under substitution of all non-logical symbols. In general necessity was understood as

truth in all possible worlds. So far then necessity has been understood in virtue of particular

ways of being true. But perhaps that is not the right way for understanding the necessity of

laws. Perhaps the necessity of laws has to do with their status as rules, not with being true

in a particular way.

Something is not a rule in virtue of being true or false in a particular way, and it is not

the case that in order to be rule, a proposition has to be true in all possible worlds. The

California traffic laws are not true in all possible worlds, but they are still rules. Being true

in all possible worlds is one way of being necessary, or having a ‘must’ flavor, being a rule

is a different way. It is not obvious that the two are related.

My suggestion is that the ‘must’ in laws indicates they are more than just facts, but that

this ‘more’ is not that they are facts that are true in every possible world, but that they are

(also) rules. We should explore this possibility, because the attempt to explain the ‘must’

in laws in terms of their truth led to the attempt to subsume other necessities (like logical

and metaphysical necessity) under natural necessity as in Lange’s view, and as we’ve seen,

there are problems for such views.

But it is easy to see why it is so tempting to try, despite the problems: laws of nature,

unlike traffic laws, are supposed to be true or false. While ‘Bicyclists stop at stop signs.” is

really a command (it certainly doesn’t accurately describe the behavior of bicyclists), and

hence exempt from being true or false39, this is not obviously so for laws of nature, and we

might indeed be inclined to think that laws are also true (or false as the case may be). So

unlike the case of traffic laws, where we might be happy to accept that the must involved in

them is simply an expression of their status as rules, things aren’t as straightforward in the

case of laws of nature.

To mark this peculiarity we should separate the question of a law’s truth or falsity from

the question of the law’s necessity. What makes laws true or false is the way things are

around here, in this world. But that’s not what makes laws necessary. What makes laws

necessary is that they are rules. So laws have a dual character: in some respects they are

39At least on most accounts of the semantics of commands.


like facts, they are true or false in virtue of how things are with respect to this world. On

the other hand they are like rules in that they require things to be a certain way. How can

something be both fact-like and rule-like?

Here’s a tentative answer. Laws in physics in particular are typically formulated math-

ematically. It is their mathematical character that allows them to act as rules. For the laws

of physics are not truths of mathematics, that is, they are not true in virtue of mathematical

identities.40 The reason it nevertheless makes sense to formulate laws of physics in terms

of mathematics is, among other things, that mathematical equations have solutions beyond

the finite range of data that have actually been collected for a particular system of inter-

est. In offering a mathematical equation over simply offering a set of data, a physicist is

offering a rule that goes beyond the collected data. For an equation to qualify in the first

place, it of course has to match the data, that is, the data so far have to be solutions to the

equation.41 But being an equation, it will have solutions for which no data exist, and it is

with respect to such solutions that two equations may differ while agreeing for the data we

actually have.42 That makes an equation more than just a ‘short-hand’ for a set of data.43

Laws are not just equations, of course. Being equations accounts for their status as

rules, but it doesn’t account for their truth. For if the law qua equation were true or false,

what it would be true or false of would be numbers (or vectors, or tensors), but not the

‘things in the world’ we initially thought laws were telling us about. Equations divide

numbers (or more generally: set-theoretic structures, models) into those that ‘make them

true’, that is, are solutions, and those that ‘make them false’, that is, aren’t solutions to

the equation. Which models are solutions to a particular equation is entirely a matter of

mathematics and doesn’t depend on anything in particular about this world. In particular,

the Raleigh-Jeans law and Planck’s law both have solutions, so that by itself doesn’t tell us

anything about which one is true for black-body radiation. Being equations only accounts

40Perhaps with the exception of conservation laws in modern field theories, see discussion above.41Typically that’s only going to be true in an approximate fashion, that is, the data will approximately be

solutions to the equation.42Again, actual science is of course a bit more complicated than this cartoon-version. For example there

will be cases where one equation fits well with some set of data, but not with a different set, while anotherequation fits with the second, but not the first set of data. The competition between the Raleigh-Jeans law,Wien’s law, and Planck’s law to account for black-body radiation is a good example of this.

43The Mill-Ramsey-Lewis account of laws seems a bit too much like taking laws to be shorthand, since ontheir view, laws merely organize facts.


for the status of laws as rules.

In virtue of what are laws true or false? In virtue of how things are in this world, that

is, in virtue of the forces, particles, fields, that are actually found in this world. But that’s

just a matter of how things are in this world, not a matter of how things might be in other

worlds. What allows us to use laws in reasoning about what would have or could have

happened, is the fact that laws are equations the solutions of which we can calculate even

for cases where the numbers in question are not given by any actual data. Solutions to

equations are (prima facie) possibilities, although in some cases we may think that at least

in our world there isn’t any actual state of affairs corresponding to a particular solution,

given other things we believe about this world.

Is this enough for the felt necessity of laws? Does it do justice to the claim that laws

‘exclude’ certain kinds of behavior? I think it does, since laws by themselves rarely exclude

behavior. What they do is put a ‘price tag’ on what it would take to alter the behavior of

particular things. Take rocks. No law of nature prevents rocks from flying off into outer

space. Laws of nature instead tell us just how much energy we would have to use to make

the rock take off rather than fall back to the ground.44 Laws of nature can render certain

scenarios impossible, but they don’t do so all by themselves. It is impossible for a human to

throw a rock hard enough for it to fly off into outer space not because of some law of nature

alone, but also because of some facts about humans and the mass of the earth. When we

reason about possibilities in this way we typically leave the laws fixed and imagine (and

sometimes alter) the conditions in which the laws are applied (we use additional power

sources far beyond what a single human could do to ‘throw’ rockets into outer space), but

that shouldn’t lead us to think that the laws could not have been different.

The necessity of laws, then, should not be understood as definable in terms of other

necessities. Unlike metaphysical necessities, most laws45 have a ‘must’ character to them

because they are rules, not because they are true in all possible worlds or true in virtue of

some identity holding. Unlike other kinds of rules, however, laws are also true (or false),

44There are exceptions. Conservation laws are one, but as we’ve seen, those might in fact be metaphysicallynecessary. Setting the speed of light as a universal speed limit is another. But that’s also a rather funny kindof law.

45Excluding perhaps conservation laws, and propositions like ‘electrons have negative charge’, whicharen’t laws.


which explains why it has been so tempting to classify them with metaphysical or logical

necessity. But I think it makes more sense to acknowledge that laws of nature have their

own form of necessity, a necessity they posses in virtue of being rules, not in virtue of being

true. Laws could have been different, that is, in other worlds other equations might have

been appropriate, other quantities used. But given that a certain equation is a law, it is fixed

what its solutions are, and therefore, which states are possible or impossible. This explains

why laws are necessary and contingent. They are rules, and that makes them necessary, but

they could have been false, that makes them contingent.

5.5 Conclusion

In this chapter I’ve discussed the possibility of a variety of necessity among laws of nature.

I’ve suggested that at least some laws of nature, conservation laws in modern field theories,

might be regarded as metaphysically necessary. But while metaphysical necessity could be

understood in terms of special ways of being true, such as holding in virtue of an identity,

the felt necessity of laws cannot be accounted for in this way. Instead the necessity of most

laws of physics has to be understood as reflecting that they are (also) rules. Laws end up

having a funny status of being both rules and facts. I suggested that this dual character of

laws is not all that perplexing if we keep in mind that laws are formulated as mathematical

equations. This allows them to have the character of both rules and facts. It is this dual

character that accounts for the odd status of laws as both necessary and contingent. As

rules they have a necessary flavor, as facts they could have been false.

Chapter 6

Conclusion: Realism and antirealism asmetaphysics

The aim of my dissertation has been to understand what realism debates in the philosophy

of science should be about. I asked what realism and antirealism currently mean, and what

their target is. I furthermore asked how much metaphysics is involved in taking realist and

antirealist stances. What are the answers we’ve arrived at?

6.1 Summary

In my attempts to answer these questions I first looked at Carnap’s attempt to avoid meta-

physics by refusing to come down on the side of either realism or antirealism with respect

to particular scientific (or mathematical) entities. Realism and antirealism were here un-

derstood as existence claims about certain kinds of entities, like numbers, fields, or atoms.

Carnap saw both realism and antirealism as ‘metaphysics’, and offered what looked like a

quietist position. His reason seemed to be that no meaningful debate about the existence

or non-existence of these entities could be possible outside of the particular conceptual

frameworks in which they belonged. But in doing so Carnap seemed to ignore important

commitments that come with accepting such conceptual frameworks, viz. ontological com-

mitments. Moreover, the principle of Tolerance he suggested we apply in dealing with

146

CHAPTER 6. CONCLUSION 147

different frameworks didn’t seem compelling. Even if we accept that philosophical ques-

tions about realism and antirealism are in some sense practical or non-cognitive, quietism

does not follow.

Carnap’s quietism was motivated by his empiricism, but empiricism does not always

lead to quietism, nor is quietism always motivated by empiricism. In chapter three I looked

at the contemporary debate over realism in the philosophy of science. Starting from the two

most developed challenges to realism – constructive empiricism and the pessimistic-meta-

induction argument – I first showed that the debate is at heart a skeptical debate, since the

challenges to realism are exclusively epistemic. The debate in its current form finds itself

in a stalemate, where antirealists challenge the epistemically strong conclusions realists

like to draw, but without offering a compelling argument for their weaker alternatives. This

prompts what I called a ‘realist quietism’ – a view that is basically a form of realism, but

makes no attempt to respond to the skeptic. Unlike Carnap’s quietism, this quietism is not

motivated by empiricism, but naturalism, and it is closer to realism than to antirealism.

I argued that this quietism also lacks argumentative support, and remains unsatisfying as

a position. We find ourselves in this triple-stalemate that looks so unlike other realism

debates because the target of realism debates in the philosophy of science is characterized

in epistemic terms: ‘unobservables’ are the problematic entities, claims about them the

problematic statements of the debate. This exclusively epistemic characterization restricts

the available antirealist positions to skeptical ones. What is needed, I suggested, is a class

of statements that are problematic not just from an epistemic perspective. Modal claims

are such statements, and I argued that targeting modal claims in the sciences yields a more

diverse realism debate, which also more closely resembles the structure of realism debates

in other areas of philosophy.

In the fourth chapter I approached the matter from a different direction. I looked at

structural realism, a view that is often presented as a novel position in the realism debate

in the philosophy of science. The view itself turns out to be very difficult to pin down,

and it often seems unclear just what particular debate structural realists are trying to con-

tribute to. I followed structural realism along what I called different ‘branches’, that is, I

looked at how structural realism fares as a contribution to a number of different debates. I

argued that structural realism doesn’t do as well in many of these debates as some of its


proponents claim, and that the different contributions moreover make it difficult to see how

there could be a single position playing all the roles structural realists had hoped for. I

suggested that structural realists should be seen as trying to do very different things at the

same time: (1) shift the debate about realism in the philosophy of science from being con-

cerned with ontology to being concerned with structure, and in particular modal structure;

and (2) articulating a particular position within this new debate. The first step nicely fits

with the suggestion I made in chapter three: target modal claims qua modal claims, instead

of targeting claims about unobservables. The second step articulates a particular form of

realism about modal claims: modal claims are made true by irreducible modal structures

in the world. Ontic structural realists furthermore make the additional metaphysical claim

that non-modal or non-structural elements depend on structure. I suggested that this claim

is best understood as an interdependence view. While I side with structural realist in their

attempt to shift the debate, I believe there are a number of interesting positions available

within the new debate, and we should not hastily side with the structural realists.

With the idea in mind that the appropriate target for realism/antirealism debates in the

philosophy of science are the modal claims of scientific theories, I turned, in chapter five,

to a particular class of such modal claims: laws of physics. I argued that there is a genuine

variety of necessity amongst laws of physics. Some laws are metaphysically necessary

in the sense that they are identities, whereas for other laws we should account for their

necessity by regarding them as rules. Such laws, I suggested, have a dual character of

being both rules in virtue of being mathematical equations, and true (or false), in virtue

of being applied to phenomena in this world. Their truth (or falsity) is accounted for by

the way things are in the actual world, whereas their modal status is accounted for by the

fact that they are rules. With respect to the emerging debate about the modal status of

claims in science, this suggests that we may have to take a more differentiated view than a

straightforward realism or antirealism with respect to all modal claims in science.

6.2 The Answer

What then, is the answer to the two questions I started out with, the question of what the

debate about realism and antirealism in the philosophy of science should be about, and


whether answers to that questions inevitably involved taking a metaphysical stance?

The answer to the first question, I suggest, is that the target of debates about realism

in the philosophy of science are the modal claims of scientific theories. To think of the

target in this way doesn’t mean, however, that realism and antirealism have to be held with

respect to all modal claims of scientific theories. As I suggested in the last chapter, we may

target subsets of the modal claims of scientific theories and be realists or antirealists with

respect to them. A mix and match of realisms and antirealisms seems possible if we take

seriously the idea that there may be a variety of modalites involved in the modal claims of

science.

What it means to take a realist or an antirealist stance with respect to those claims

involves several components. The most realist position available takes modal claims to be

‘made true’ by irreducible modal facts (or perhaps ‘modal structures’), and the source of

the modality to be ‘in the world’. Ontic structural realism is a particular instance of such

a realist position, but there are others as well. This strong form of realism can be opposed

in a number of different ways. One might agree with the realists that modal claims have

truth values, but insist that they are made true by facts that are not irreducibly modal. Or

one might insist that the facts are irreducibly modal, but that the source of the modality

lies in us, not the world. Depending on how that is cashed out, modal claims will be

subjective as opposed to objective. There are even more radical non-realist positions: an

error theorist will suggest that while modal claims have truth values, we are systematically

mistaken about them – our modal claims are false. An instrumentalist about modal claims,

by contrast, will suggest that taking modal claims to have truth values is a mistake to begin

with. Modal claims serve as instruments for predictions, they aren’t propositions at all.

How plausible any of these positions end up being is of course a matter of debate – this

is the debate we ought to have as the realism debate about science. Modal claims are not

characterized in terms of their problematic epistemic features only. While the epistemology

of modal claims is indeed something to worry about, it is not the only reason to worry

about these claims. Unlike in the case of claims about unobservables, we here have reason

to question the semantics and metaphysics of these claims as well. Modal claims not only

go beyond what we can observe, there is also a question about what the source of modality

might be. What makes it necessary, as opposed to merely true, that no signal can travel


faster than the speed of light? It is precisely because modal claims raise these additional

concerns that there seem to be reasons to question their semantics and metaphysics as well

as their epistemology. And it is because this becomes a possibility that we find ourselves

with a plurality of non-realist positions. The debate looks more like a debate in meta-ethics,

which picks out certain kinds of claims in ordinary discourse, normative claims, and asks

what their status is. I here don’t side with any of these positions, I’m merely suggesting

that this is what such a new debate about realism in science might look like.

Now that we know what realism and antirealism in the philosophy of science are about,

and what it might mean to take realist and antirealist views, we can tackle the second ques-

tion, namely whether taking such a view always involves doing metaphysics. In the old

debate, the idea had been that only the realist takes a metaphysical view, since ‘belief in’

unobservables meant going beyond ‘the phenomena’, which by empiricist lights consti-

tutes doing metaphysics. On the other hand, quietists seemed to oppose both realism and

antirealism in these debates as too metaphysical.

Changing the debate in the way I suggested can help us understand better what is and

isn’t metaphysical about taking realist or antirealist views in philosophy of science. I sug-

gest that the debate is metaphysical in two distinct ways. First, both realisms and non-

realisms in the new debate are metaphysical positions in the sense that they are answers to

distinctively philosophical questions about claims made in science. Any analysis of modal

claims, whether as ‘made true by modal facts’ or as ‘instruments for prediction’ takes a

position that goes beyond what science ‘tells us’. Scientific practice acts as a constraint on

the adequacy of any such analysis, but it is far from obvious that science dictates a clear

answer – science doesn’t even ask the question. Unlike the naturalistically motivated qui-

etist, both realists and antirealists in this debate offer philosophical positions that are not

simply the outcome of accepting scientific results.

Do we have to engage in this debate, or could the debate simply be avoided? As we’ve

seen, quietism as the suggestion that everyone should just stop engaging in a particular

debate is difficult to defend. But isn’t it equally difficult to motivate the quietist to engage

in a particular debate? Here I think the new debate I am proposing stands a better chance

of success than the old debate. For the old debate, as I argued in chapter two, is a skeptical

debate. Skeptical debates can question the certainty or firmness of the knowledge claims


made. They are not very powerful against anybody who concedes that particular claims

they make might be wrong, but nevertheless insists that no matter how fallible their claims

might be, they are still better supported by the available evidence than others. This is the

position held by naturalist quietists like the Second Philosopher, and the skeptic has a hard

time convincing her that there is something she ought to worry about. But the debate I am

proposing is not an exclusively skeptical debate. It starts with questions that are not easily

dismissed: What does it mean to say p is a law? Or: This is physically impossible? If

we take science seriously and find modal claims in the sciences, then we ought to have an

answer to these questions. As I said, there are many potential answers to these questions,

and the answers will typically be constrained by the use made of such claims in scientific

practice, but the question seems legitimate and, unlike the empiricists’ suspicion against

claims about what’s unobservable, without any particular philosophical prejudice.

So taking a stance in the new debate will involve metaphysics at least in the sense of

taking a distinctively philosophical position, and this holds for realists and non-realists

alike. Moreover, while quietism of course always remains a possibility, it seems even more

difficult to motivate than it had been with respect to the old debate. I would like to suggest,

however, that there is second sense in which the new debate will involve metaphysics.

In the new debate, it seems that in order to articulate any one of the positions, we

need to do a lot of metaphysics. As the discussion of structural realism in Chapter 4 sug-

gested, we need to decide questions about the relation between objects and the structures

they stand in, or more generally, we need to say something about the relationship between

modal and non-modal facts, or, if we prefer instrumentalism or error theory, between or-

dinary and modal claims. Various forms of dependence and interdependence relationships

have been offered, and it seemed that, at least in some cases, the articulation of the position

required far reaching metaphysical commitments, including commitments about the ontol-

ogy of properties. Not all views will be equally metaphysically committed, but it seems

that everyone will have to make some metaphysical commitments.

A similar observation can be made in light of the discussion in Chapter 5. For there

we found that in taking a position with regard to modal claims in science, we need to say

something about how the modality or modalities involved in such claims relate to other

forms of modality, such as metaphysical or logical modality. This requires engaging in


characteristically ‘metaphysical’ debates: the articulation and defense of view as to what

these different forms of necessity and possibility amount to, and how they relate to each

other, e.g., whether one or more are reducible to others.

This should come as no surprise. In the old debate, antirealists could seem to get

away without metaphysical commitments precisely because the debate was exclusively

epistemic. The most radical antirealism was constructive empiricism, which is really a

form of agnosticism. The agnostic of course avoids any commitments either way, and can

thereby seem non-metaphysical, although not anti-metaphysical. Since the new debate is

no longer limited to just epistemic concerns, articulating the various positions also goes

beyond epistemic commitments.

This should not suggest that metaphysics simply takes over philosophy of science. We

also need to do a lot of philosophy of science, since to establish whether particular modal

claims in the sciences are accounted for by the position of our choice, we need to look very

carefully at the claims in question and the role they play in particular theories. This type

of work requires the kind of careful analysis of particular areas of science philosophers

of science typically engage in. The role and use of modal claims in the sciences places

conditions of adequacy on any account of modality in the sciences, and potentially beyond.

To take a stance in the realism/antirealism debate involves both metaphysics and phi-

losophy of science, then. But that’s not because science somehow always already involves

metaphysics, as is sometimes suggested. Science is just science. It is because in talking

about science from a philosophical point of view we need to do more than just science.

If we are going to take a realist or antirealist view, we are articulating and offering more

than just a restatement of claims already made in science. We are offering an account of

claims made in science, and offering such an account involves doing metaphysics. Both

philosophy of science and metaphysics would benefit from a more cooperative interaction

to address these matters.

Bibliography

Ainsworth, Peter Mark. 2010. “What is ontic structural realism?” Studies in History and

Philosophy of Modern Physics 41:50–57.

Alspector-Kelly, Marc. 2001. “On Quine on Carnap on Ontology.” Philosophical Studies

102:93–122.

Armstrong, D. M. 1983. What is a Law of Nature? Cambridge University Press.

Auyang, Sunny Y. 1995. How is Quantum Field Theory Possible? Oxford University

Press.

Ayer, A. J. 1956. “What is a Law of Nature?” Revue Internationale de Philosophie,

vol. 36.

Beebee, Helen. 2000. “The non-governing conception of laws of nature.” Philosophy and

Phenomenological Research 61 (3): 571–594.

Belot, Gordon. 1998. “Understanding Electromagnetism.” The British Journal for the

Philosophy of Science 49:531–555.

Benacerraf, Paul. 1973. “Mathematical Truth.” The Journal of Philosophy 70 (19): 661–

679.

Bird, Graham H. 1995. “Carnap and Quine: Internal and External Questions.” Erkenntnis

42:41–64.

Boorse, H. A., and L. Motz, eds. 1966. The World of the Atom. Volume I. Basic Books,

Inc.

Braddon-Mitchell, David, and R. Nola, eds. 2009. Conceptual Analysis and Philosophical

Naturalism. MIT Press.

153

Bibliography 154

Brading, K.A., and H. R. Brown. 2003. “Symmetries and Noether’s Theorems.” Chapter 5

of Symmetries in Physics: Philosophical Reflections, edited by Katherine Brading and

Elena Castellani, 89–109. Cambridge University Press.

Brading, Katherine. 2002. “Which symmetry? Noether, Weyl, and conservation of electric

charge.” Studies in History and Philosophy of Modern Physics 33 (1): 3–22.

Brading, Katherine, and Elena Castellani, eds. 2003. Symmetries in Physics: Philosophi-

cal Reflections. Cambridge University Press.

Brading, Katherine, and Alexander Skiles. “Underdetermination as a Path to Ontic Struc-

tural Realism.” manuscript.

Brock, Stuart, and Edwin Mares. 2007. Realism and Anti-Realism. McGill-Queen’s

University Press.

Brown, Harvey, and Peter Holland. 2004. “Dynamical vs. variational symmetries: under-

standing Noether’s first theorem.” Molecular Physics 102, no. 11 & 12.

Cao, Tian Yu. 2003. “Structural Realism and the interpretation of quantum field theory.”

Synthese 136 (1): 3–24.

Carnap, Rudolf. 1932. “Uberwindung der Metaphysik durch logische Analyse der

Sprache.” Erkenntnis 2 (4): 219–241.

. 1975 [1934]. “Theoretische Fragen und Praktische Entscheidungen.” In Logis-

cher Empirismus, der Wiener Kreis: ausgewahlte Texte mit einer Einleitung, edited by

Hubert Schleichert. Fink.

. 1937. Logical Syntax of Language. Kegan.

. 1956. “Empiricism, Semantics, and Ontology.” In Meaning and Necessity. The

University of Chicago Press.

. 1966. An Introduction to the Philosophy of Science. Edited by Martin Gardner.

Dover.

Carnap, Rudolf, and W. V. Quine. 1990. Dear Carnap, Dear Van – The Quine-Carnap

Correspondence and Related Work. Edited by Richard Creath. University of Califor-

nia Press.

Bibliography 155

Carroll, John. 1994. Laws of Nature. Cambridge University Press.

Cartwright, Nancy. 1983. How the laws of physics lie. Clarendon Press, Oxford.

. 1994. Nature’s Capacities and their measurement. Oxford University Press.

Cei, Angelo, and Steven French. 2006. “Looking for structure in all the wrong places:.”

Studies in History and Philosophy of Science 37:633–655.

Chakravartty, Anjan. 2007a. A metaphysics for scientific realism. Cambridge University

Press.

. 2007b. “Six Degrees of Speculation.” In Images of Empiricism, edited by Bradley

Monton. Oxford University Press.

Demopoulos, William, and Michael Friedman. 1985. “Bertrand Russell’s The Analysis

of Matter: Its Historical Context and Contemporary Interest.” Philosophy of Science

52 (4): 621–639.

Duhem, Pierre. 1982 [1906]. The Aim and Structure of Physical Theory. Edited by Philip

P. [trans.] Wiener. Princeton University Press.

Dummett, Michael. 1978 [1963]. “Realism.” In Truth and Other Enigmas. Harvard

University Press.

Eklund, Matti. 2009. “Carnap and Ontological Pluralism.” Chapter 4 of Metametaphysics

- New essays on the foundations of ontology, edited by David Chalmers, David Man-

ley, and Ryan Wasserman. Oxford University Press.

Ellis, Brian. 2001. Scientific Essentialism. Cambridge University Press.

Fine, Arthur. 1986. In The Natural Ontological Attitude. University of Chicago Press.

Fine, Kit. 1994. “Essence and Modality.” Philosophical Perspectives 8:1–16.

. 2001. “The Question of Realism.” Philosophers’ Imprint 1 (1): 1–30.

. 2002. “The Varieties of Necessity.” Chapter 6 of Conceivability and Possibility,

edited by Tamar Szabo Gendler and John Hawthorne. Oxford University Press.

Frankel, Theodore. 2004. The Geometry of Physics. Cambridge University Press.

Bibliography 156

French, Steven. 1998. “On the withering away of physical objects.” In Interpreting

Bodies, edited by Elena Castellani. Princeton University Press.

. 2000. “The Reasonable Effectiveness of Mathematics: Partial Structures and the

application of Group Theory to Physics.” Synthese 125:103–120.

. 2010. “The interdependence of structure, objects and dependence.” Synthese,

vol. online first (March).

French, Steven, and James Ladyman. 2003a. “The dissolution of objects: Between pla-

tonism and phenomenalism.” Synthese 136:73–77.

. 2003b. “Remodelling Structural Realism: Quantum Physics and the Metaphysics

of Structure.” Synthese 136:31–56.

Friedman, Michael. 1997. “Philosophical Naturalism.” Proceedings and Addresses of the

American Philosophical Association 71 (2): 5–21 (November).

. 1999. “Tolerance and Analyticity in Carnap’s Philosophy of Mathematics.” In

Reconsidering Logical Positivism. Cambridge University Press.

Giere, Ronald. 1995. “The Skeptical Perspective: Science Without Laws of Nature.” In

Essays on the Philosophical, Scientific, and Historical Dimensions, edited by F. Wein-

ert. Walter de Gruyter.

Haack, Susan. 1976. “Some Preliminaries To Ontology.” Journal of Philosophical Logic

5 (4): 457–474.

Hacking, Ian. 1983. Representing and Intervening. Cambridge University Press.

Hale, Bob, and Aviv Hoffmann, eds. 2010. Modality - Metaphysics, Logic, and Episte-

mology. Oxford University Press.

Hawley, Katherine. 2006. “Science as a guide to metaphysics.” Synthese 149, no. 3.

Healey, Richard. 2007. Gauging what’s real. Oxford University Press.

Hempel, Carl Gustav. 1965. Chapter Empiricist Criteria of Cognitive Significance: Prob-

lems and Changes of Aspects of Scientific Explanation, and Other Essays in the Phi-

losophy of Science. Free Press.

Hesse, Mary. 1970. Forces and Fields. Greenwood Press.

Bibliography 157

Hirsch, Eli. 2008. “Language, Ontology, Structure.” Nous 42 (3): 509–528.

Hofweber, Thomas. 2009. “Ambitious, yet modest metaphysics.” In Metametaphysics -

New essays on the foundations of ontology, edited by David Chalmers, David Manley,

and Ryan Wasserman. Oxford University Press.

Huggett, Nick. 1999. “Atomic Metaphysics.” The Journal of Philosophy 96 (1): 5–24.

Krause, D., and Steven French. 1995. “A formal framework for quantum non-

individuality.” Synthese 102 (1): 195–214.

Kuhn, Thomas S. 1996 [1962]. The Structure of Scientific Revolutions. University of

Chicago Press.

Ladyman, James. 1998. “What is structural realism?” Studies in History and Philosophy

of Science 29:409–24.

Ladyman, James, and Don Ross. 2007. Every thing must go. Oxford University Press.

Lange, Marc. 2009. Laws and Lawmakers - Science, Metaphysics, and the laws of nature.

Oxford University Press.

Laudan, Larry. 1984. Science and Values. University of California Press.

Lewis, David. 1970. “How to define theoretical terms.” Journal of Philosophy 67 (13):

427–446.

. 1973. Counterfactuals. Blackwell Publishing.

. 1984. “Putnam’s Paradox.” Australasian Journal of Philosophy 62 (3): 221–236.

. 1986a. On the Plurality of Worlds. Basil Blackwell.

. 1986b. Philosophical Papers Volume II. Oxford University Press.

. 1994. “Humean Supervenience Debugged.” Mind 103 (412): 473–490.

. 2009. “Ramseyan Humility.” In Conceptual Analysis and Philosophical Natu-

ralism, edited by David Braddon-Mitchell and Robert Nola. MIT Press.

Lipton, Peter. 2004. “Discussion - Epistemic Options.” Philosophical Studies 121, no. 2.

Lyre, Holger. 2004. “Holism and Structuralism in U(1) Gauge Theory.” Studies in History

and Philosophy of Modern Physics 35, no. 4.

Bibliography 158

Mackie, John L. 1977. Ethics: Inventing right and wrong. Harmondsworth.

Maddy, Penelope. 2007. Second Philosophy. Oxford University Press.

Martin, Christopher A. 2003. “On continuous symmetries and the foundations of modern

physics.” In Symmetries in Physics: Philosophical Reflections, edited by K.A. Brading

and Elena Castellani. Cambridge University Press.

Maudlin, Tim. 2007. The Metaphysics within Physics. Oxford University Press.

Maxwell, Grover. 1970. “Structural Realism and the meaning of theoretical terms.” In

Analyses of Theories and Methods of Physics and Psychology, edited by S. Winokur

and M. Radner, Volume 4. University of Minnesota Press.

McCormmach, Russell. 1970. “H. A. Lorentz and the Electromagnetic View of Nature.”

Isis 61, no. 4.

Mitchell, Sandra D. 2000. “Dimensions of Scientific Law.” Philosophy of Science 67 (2):

242–265.

Musgrave, Alan. 1989. “NOA’s Ark - Fine for Realism.” Philosophical Quarterly 39:383–

98.

Newman, M.H.A. 1928. “Mr. Russell’s Causal Theory of Perception.” Mind 5 (146):

26–43.

Noether, Emmy. 1918. “Invariante Variationsprobleme.” Nachr. d. Konig. Gesellsch. d.

Wiss. zu Gottingen, Math.-phys. Klasse, vol. Transl. by Tavel, M.A.

Poland, Jeffrey. 1994. Physicalism - The Philosophical Foundations. Clarendon Press,

Oxford.

Pooley, Oliver. 2005. “Points, Particles, and Structural Realism.” In The Structural Foun-

dations of Quantum Gravity, edited by S. French, D. Rickles, and J. Saatsi. Oxford

University Press.

Price, Huw. 2009. “Metaphysics after Carnap: The ghost who walks?” Chapter 11

of Metametaphysics - New essays on the foundations of ontology, edited by David

Chalmers, David Manley, and Ryan Wasserman. Oxford University Press.

Psillos, Stathis. 1999. Scientific Realism. Routledge.

Bibliography 159

Putnam, Hilary. 1979. “Realism and Reason.” In Meaning and the Moral Sciences.

Routledge.

Quine, W. V. 1980 [1948]. “On what there is.” In From a Logical Point of View, 3rd.

Harvard University Press.

. 1951. “Two Dogmas of Empiricism.” The Philosophical Review 60 (1): 20–43.

. 1969a. “Epistemology Naturalized.” Chapter 3 of Ontological Relativity, 69–90.

Columbia University Press.

. 1969b. “Existence and Quantification.” In Ontological Relativity. Columbia

University Press.

. 1969c. “Ontological Relativity.” In Ontological Relativity. Columbia University

Press.

. 1976. “Carnap’s views on ontology.” In The Ways of Paradox, Revised Edition.

Harvard University Press.

. 1995. From Stimulus to Science. Harvard University Press.

Redhead, Michael. 1998. “Review - S.Y. Auyang How is Quantum Field Theory Possi-

ble?” British Journal for the Philosophy of Science 49:499–507.

Reichenbach, Hans. 1956. The Direction of Time. University of California Press.

Rosen, Gideon. 1994. “What is Constructive Empiricism?” Philosophical Studies

74:143–178.

Ruetsche, Laura. 2002. “Interpreting Quantum Field Theory.” Philosophy of Science 69

(2): 348–378.

Saunders, Simon. 2006. “Are quantum particles objects?” Analysis 66 (289): 52–63.

Scerri, Eric R. 2007. The Periodic Table - Its Story and its Significance. Oxford University

Press.

Schaffer, Jonathan. 2009. “On What Grounds What.” Chapter 12 of Metametaphysics -

New essays on the foundations of ontology, edited by David Chalmers, David Manley,

and Ryan Wasserman. Oxford University Press.

Bibliography 160

Schilpp, Paul Arthur, ed. 1963. The Philosophy of Rudolf Carnap. Volume XI. Open

Court.

Shoemaker, S. 1980. “Causality and Properties.” In Time and Cause, edited by Peter van

Inwagen. D. Reidel Publishing Company.

Shoemaker, Sydney. 1998. “Causal and Metaphysical Necessity.” Pacific Philosophical

Quaterly 79:59–77.

Sider, Theodore. 2009. “Ontological Realism.” Chapter 13 of Metametaphysics - New

essays on the foundations of ontology, edited by David Chalmers, David Manley, and

Ryan Wasserman. Oxford University Press.

Smart, J. J. C. 1963. Philosophy and Scientific Realism. London RKP.

Sober, Elliott. 2000. “Quine’s “Two Dogmas”.” Proceedings of the Aristotelian Society,

vol. Supplementary Volume 74.

Stalnaker, Robert. 1968. “A theory of conditionals.” American Philosophical Quaterly

Monographs, vol. 2.

Stein, Howard. 1983. “After the Baltimore Lectures: Some Philosophical Reflections on

the Subsequent Development of Physics.” In Kelvin’s Baltimore Lectures and Mod-

ern Theoretical Physics - Historical and Philosophical Perspectives, edited by Robert

Kargon and Peter Achinstein. MIT Press.

Stroud, Barry. 1984. The Significance of Philosophical Scepticism. Oxford University

Press, USA.

Teller, P. 1995. An Interpretive Introduction to Quantum Field Theory. Princeton Univer-

sity Press.

. 2004. “Discussion - What is a stance?” Philosophical Studies 121, no. 2.

van Fraassen, Bas. 1980. The Scientific Image. Clarendon Press, Oxford.

. 1989. Laws and Symmetry. Clarendon Press, Oxford.

. 1991. Quantum Mechanics. Clarendon Press, Oxford.

. 1994. “Gideon Rosen on Constructive Empiricism.” Philosophical Studies

74:179–192.

Bibliography 161

. 2002. The Empirical Stance. Yale University Press.

. 2004. “Replies to discussion on The Empirical Stance.” Philosophical Studies

121:171–192.

. 2006. “Structure: It’s Shadow and Substance.” British Journal for the Philosophy

of Science 57:275–307.

. 2008. Scientific Representation: Paradoxes of Perspective. Oxford University

Press.

Williamson, Timothy. 2005. “Armchair Philosophy, Metaphysical Modality and Counter-

factual Thinking.” Proceedings of the Aristotelian Society, vol. 105.

. 2007. The Philosophy of Philosophy. Blackwell Publishing.

Worrall, John. 1989. “Structural Realism: The Best of Both Worlds?” Dialectica 43,

no. 1-2.

Wright, Crispin. 1992. Truth and Objectivity. Harvard University Press.

Yablo, Stephen. 2000. “Aprioricity and Existence.” In New Essays on the A priori, edited

by Paul Boghossian and Christopher Peacocke. Oxford University Press.

. 2002. “Go Figure: A Path Through Fictionalism.” Midwest Studies in Philoso-

phy, vol. 25.

Documents

REALISM BETWEEN METAPHYSICS AND SCIENCE A DISSERTATION … · The aim of my dissertation is to understand what the realism debate in the philosophy of science should be about, and,