7

gig

£69.? .
‘ .. .
4; HI

. =3" ‘
Ewan. .‘wumimg- .

:59. 5!...

.,. immiugninwrlf

:E.
5 .
. ‘ uh: fiadfiazfi if
.3 53.... 1. .
$4.3; .

. . .55.
2 .. g ‘3... . mu...
i 2.5!.“ ’5. T:
2. 2.3.3.9.
:“Lfl
3......
% 3r
:
url
11:“
3.23...”
:55.

L."
, ha.»
y . a a:
(mini.
A I. I
Q. Ira... ..
xvi.”
in?» .
6.. v!
3...}. ..2£.z.i.!¢as.. an
AS. I»... b 5
IT.“ '3; g
5 F3
4.

l .

39!. .l 11.
‘2... 3.2.

.4355»
§ 21...}. a

mem.

It... .. ..
£113.”..flui.
v. o);- v;

3...".
is: :2 .
ll:

.. i
gun!!! .r.
1%...- .3..." 5....xwauhifiéységuum
93:? 9.. . 5 MelaIVg!.tII..qhhHH
A .33.. if 3:371"?!
ll! .00.“. 3..
?. ..

.2.
I. .15..» l
(1.01.1350... .
2‘1?!) 3
11.1311: .

 

55:1.
7.... Qt:

T H 55‘ S " u IGAN STATE unwensmr UBRAR £8 "

lI/liflliifillzl!(I!WWI/l!HIIHHIIIIMHH('1 Hi

93 2048 9096
LIBRARY

Michigan State
University

 

 

 

 

 

This is to certify that the

dissertation entitled

A DEFENSE OF A PRAGMATIST VIEW
OF SCIENTIFIC LAWS

presented by

Roger Vajda

has been accepted towards fulfillment
of the requirements for

Ph . D . degree in Philosophy

/ a
Major professor

 

(12.714 2J9) ‘20 V6)

Date

 

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

11/00 cJCIHCDatoOmpes-pjd

A DEFENSE OF A PRAGMATIST VIEW OF SCIENTIFIC LAWS
By

Roger Vajda

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Philosophy Department
2000

ABSTRACT
A DEFENSE OF A PRAGMATIST VIEW OF SCIENTIFIC LAWS
By

Roger Vaj da

Laws of nature have, since Descartes, been seen as a central feature
of scientific inquiry and philosophical analysis. The standard empiricist
analysis of laws has come under challenge by realists, like Armstrong, who
believe the analysis leaves out important features we intuitively ascribe to
laws (e.g., necessity, explanatory power), and by antirealists like van
Fraassen who believe the very notion ensnares us in metaphysics. In my
dissertation, I defend the view against both the realists and van F raassen that
an empirically respectable notion of scientific law is available within the
framework of pragmatism, which relativizes lawhood to the choice of a
language and a set of experimental manipulations.

In chapter one, I argue that the philosophical problem of scientific
laws has a long history, and that van Fraassen has picked out some tensions
between views in that history. In chapter two, I analyze and criticize three
realists' views and Show that they have a flawed common strategy based on
incompatible presuppositions. In chapter three, I argue for my own view, a
pragmatist view that builds on the work of Dewey, Nagel, Goodman and
Skynns. I argue that replicability is the source of our confidence that
regularities are projectible (in Goodman's sense) and resilient (in Skyrm's
sense), and that this is the source of our confident assertion of
counterfactuals, our belief in the necessary connection between events, and
our belief that laws are empirically accessible. In chapter four, I answer

objections from the realists, from van Fraassen and from those

constructivists who find too much residual realism in my view. I there
defend the view that one can have objectivity without being a realist, and

relativity without being a relativist.

DEDICATION

To Louis and Joseph

ACKNOWLEDGEMENTS

I would like to thank the following people for the help they have
given. Martin Benjamin helped me see the central themes of pragmatism as
relevant to the philosophy of science. Kay Rout and Kitty Geisler helped the
stylistic revisions of chapters one and two. Herb Garelick commented on the
entire manuscript. Jo Anne Sorlie edited an earlier draft and gave helpful
advice on the final draft. I am especially in debt to my chair, Joseph Hanna
for his patient reading, and thoughtful discussion of all the drafts of this
dissertation. I would also like to thank my wife and two sons for the

sacrifices they made so that I could complete the work.

SUMMARY TABLE OF CONTENTS

CHAPTER ONE: THE PROBLEM OF SCIENTIFIC LAWS 1
CHAPTER TWO: THREE REALIST POSITIONS 43
CHAPTER THREE: A PRAGMATIST POSITION 80
CHAPTER FOUR: THREE SOLUTIONS IN SEARCH OF A PROBLEM 154
CHAPTER FIVE: CONCLUSION OF THESIS . 207

BIBLIOGRAPHY 21 1

vi

DETAILED TABLE OF CONTENTS

CHAPTER ONE: THE PROBLEM OF SCIENTIFIC LAWS
1.1 The Problem according to van Fraassen
1.2 Structure of the Chapter and Thesis

1.3 Origins of van Fraassen's Challenge

1.3.1 Van Fraassen's List
1.3.2 Sources of the Philosophical Problem
1.3.2.1 Ancient Greek Reflections on Change
1.3.2.2 Medieval Reflections on God's Rule over
Nature
1.3.2.3 Modern Reflections on Newtonian Mechanics
1.3.3 The Standard Empiricist Model of Laws
1.3.4 The Realist Challenge to the Standard Model
1.3.5 Van Fraassen's Antirealist Counterchallenge to
the Realists

1.4 Thesis: An empiricist can meet the burden of saying what
additional conditions a regularity must meet to be a law without
going realist.

CHAPTER TWO: THREE REALIST POSITIONS
2.1 Realism's Stake in Science
2.2 Three Realists Positions: a) Armstrong b) McCall

and c) Harre

vii

16

17

27

32

36

41

43

43

43

 

2.2.1 Laws as Relations between Universals (Armstrong) 43
2.2.2 Laws as Relations in Branched Space-Time (McCall) 48
2.2.3 Laws as Relations in a Natural Kind Hierarchy

(Harre) 49

2.3 An Analysis and Criticism of the Core Realist Assumptions 54

2.3.1 Internal Critique of the Realist Assumptions 58
2.3.2 External Critique of Realist Assumptions 72
2.4 Conclusion ' 79
CHAPTER THREE: A PRAGMATIST POSITION 80
3.] Laws as Constructions 82
3.1.1. Construction Relative to Language 82
3.1.2. The Role of Mathematics 87
3.2 Practical Conditions of Constructions 88
3.2.1. The Connection between Thought and Action 88
3.2.2 The Connection between Words and Action 90
3.2.3 The Social Character of Practice 91
3.2.4 The Practical Character of Science 92
3.2.5 The Continuum of the Practices of Science 93
3.2.6 The Test of Practice: Replicability 107
3.2.7 Laws as Constructed Relations between Measurable
Variables 108
3.3 Success Conditions for Constructions 112
3.3.1. Empirical Adequacy Insufficient 115
3.3.2. Consensus Insufficient 125

3.3.3. Reliable Prediction Required. 126

viii

3.3.4. Reliable Prediction from Experimental

Replicability 128
3.3.5. Reliability and Decision-Making 131
3.3.6. Laws and Action-Plans 132
3.3.7. Three Potential Problems 132
3.4 Answers to Important Parts of van Fraassen's List 136
3.5 The Remainder of List Mistaken 148
3.6 Summary 152

CHAPTER FOUR: THREE SOLUTIONS IN SEARCH OF A PROBLEM 154
4.1 Laws not Representations, but What Representations are About. 1 55

4.1.1 The Logical Problem from the Extension of 'Law' 156

4.1.2 An Objection from the Conceptual Role of Law 160
4.1.3 Response: Law-statements can be about the world
without being about laws 165
4.2 Laws as Socially Negotiated Rules. 166
4.2.1 Replication as Social Achievement 167

4.2.2 First Response: Algorithms are methodologically
useful. 173

4.2.3 Second Response: Consensus is prior to dissensus. 176

4.2.4 Third Response: Science is constructed at every level

without being constructions all the way down! 177
4.3 Laws as Doppelgéinger for Subjective Probabilities 178
4.3.1 Van Fraassen's Dual Strategy 180
4.3.2 The Problem with Induction 181

4.3.3 The Problem with Inference to the Best Explanation 184

4.3.4 The Problem with Pragmatism 188

ix

4.3.5 Van Fraassen's Position
4.3.6 My Response to the Objection and the Position
4.3.7 Van F raassen's Realism

4.4 Conclusion

CHAPTER FIVE: CONCLUSION OF THESIS

BIBLIOGRAPHY

191
192
196

205

207

211

Chapter One: The Problem of Scientific Laws

1.1 The Problem according to van Fraassen

In science, there are certain relations that have been called laws.
These include Newton's laws of motion, the law of gravitation, Ohm's law,
Snell's law, Hooke's law, Mendel's law, etc. There is nothing scientifically
problematic about these laws. Scientists understand both the conditions
under which these generalizations hold up and the conditions under which
these generalizations break down. There are, however, philosophical
problems involved in understanding the nature of these laws. The long
history of philosophical reflection on scientific laws need not be described in
detail here. An interesting twist that has occurred recently is worth
understanding, and serves as the occasion for this thesis. Bas Van Fraassen,
a respected philosopher of science, claims that the whole philosophical
problem of laws rests on a mistake. There are no such things as 'laws' as
philosophers have described them. All attempted accounts of 'laws' run up
against contrary demands. It would seem nothing could do all the work
expected of laws. But, van Fraassen reassures us, this result is not bad. We
do not need laws to be able to revise our empirical beliefs rationally, nor
does science need them to constrain theory choice. Rather, we can revise our
empirical beliefs in the light of new experience by updating the probabilities
we assign to them by Bayes's Theorem. Also, we can reason from the

symmetries of theoretical models in science to some of the traditional

conservation laws that have impressed philosophers more than they should.

Science does not need laws, but gets along nicely with symmetries.

1.2 The Structure of the Chapter and Thesis

It is the thesis of this dissertation that van Fraassen's challenge can
be met. I argue in chapter three that there are laws in the required sense.
Before I get there, however, I will show that van Fraassen's challenge needs
to be met (chapter one) and that the realists have not met it (chapter two).
Afier I argue for my position, I will need to address at least a few of the
remaining objections, lest the reader wonder whether I have failed to meet
the challenge also. That will be the business of chapter four.

In chapter one, I defend the claim that van Fraassen has identified
the philosophical problem of laws. He does so by drawing up a list of
' properties philosophers claimed for laws, and by showing the internal
tensions in that list. As he reflects on that list, he identifies two problems
that any successful analysis will have to solve, claims that both problems
cannot be solved at the same time, and concludes that there are no such
things as laws. So the problem of laws, as I call it, depends, for van
Fraassen, on these two other problems that arise from reflection on the list of
properties laws must have to do the jobs they do.

My problem is related to his two problems in the following way. My
problem looks at the whole list and asks if anything can be all those things;
van Fraassen breaks the list into two sub-lists and associates a distinct
problem with each of the two sublists of properties laws are expected to
have. He argues that things that do well in meeting the requirements on one
sub-list do poorly in meeting the requirements on the other. So one or the

other associated problem remains unsolved. The two problems have

independent interest, but are a way of motivating the claim that the entire list
is important. In short, if one or the other of van Fraassen's two problems
remains unsolved, my problem remains unsolved. Van F raassen maintains
that the problems are jointly insoluble. I argue that they are not only soluble,
but also solved.

Van Fraassen argues that any account of laws that does not solve his
two problems fails to satisfy the criteria for laws that philosophers have
considered important. The argument, to have any bite, requires that there be
substantial agreement about items in the list of properties. If there is no such
agreement, then the problems van F raassen sees with laws may not be that
serious.

I will show in this chapter that the problem, in fact, is of long
standing and that the list grew out of a long history of philosophical
reflection on natural regularities. There has been such reflection since the
presocratics search for the invariant element(s) in the variable. That it has
long been seen as an important problem is evident from the knots Leibniz
and Locke tie themselves in trying to give a satisfactory account of Newton's
laws. They all were working with a list of properties, perhaps implicit, but in
some cases (e.g., An'stotle, Berkeley) explicit. I will show that in the
twentieth century, something very much like van Fraassen's list was
presupposed by both the logical empiricist Hempel, and his realist
adversaries Armstrong, McCall and Harre. Thus I will show that van
Fraassen's challenge is not one that can be simply ignored.

Hence, in chapter one, I first give van Fraassen's list and its historical
roots. Then, I go on to give the standard empiricist analysis of Hempel, the
realist criticism of Hempel, and the realist alternatives to his view. After I

have given the historical context of van Fraassen's challenge, I summarize

the most important of van Fraassen's criticisms of the realist alternatives. I
conclude chapter one with the claim, which I defend in chapter three, that a
pragmatist position remains possible, even after one has eliminated the
logical empiricist, realist and antirealist views.

In chapter two, I will give an exposition, analysis and criticism of the
realist alternatives offered by Armstrong, McCall and Harre. These are not
the only realist positions being defended, but they are varied enough to make
any common elements they have interesting and worth commenting upon.
My goal in that chapter will not be to criticize the particular details of the
views, but to look for the larger core assumptions realist views like these
share.

In chapter three, I will develop a pragmatist analysis of laws that
draws on the work of Dewey, Nagel, Goodman, Quine and Skyrms. I will
argue that we arrive at the truth-conditions of the conditional claims
embodied in laws insofar as we are able to manipulate the conditions
expressed in the conditional claims. This view immediately connects up
truth and knowability, projectibility and reliability, and, finally,
measurement and mathematical models by making linguistic construction
subject to empirical test. To state my position briefly, I claim that we
construct the world so that its behavior is regular in particular ways, but that
because our constructions point forward to events which have not happened
yet, we may find that the world fails to behave the way we expect. We then
conclude that we have failed to order the world in that particular way. Our
construction has failed to generate a coherent practice because there is no
consistent context of use. For example, 'like reproduces like,‘ a central claim
of Aristotelian biology, fails in the light of genetic mutation and evolution.

Our attempt to find orderliness in the biological realm by Aristotle's

construction fails, because the expectations it sets up in us are not met in
experience. I

Our confidence that we can rely on the regularities as constructed,
and hence predict the future on their basis (and not simply explain, afier the
fact, why what happened did happen) points to the methodological processes
by which the constructions were made, a method which ensures that what is
constructed in the laboratory will have applicability outside the laboratory as
well. The methodological maxim that covers this might be put thus: if it is
going to fail, make it fail here where we can see it. Laws are those
regularities that do not fail under such conditions.

Chapter four looks at an array of objections that might be raised by
realists, constructivists and van Fraassen himself to my position. At their
core, each objection finds my position unstable, and thus indefensible, but
the point of instability is found in different places, and the destabilizing
forces are characterized in different ways. I cannot hope to anticipate all the
objections anyone might raise to any part of my analysis, but I hope the
objections II have chosen give a fair test of my position and a chance to
demonstrate the resources my position possesses that enable it to maintain
itself. In chapter five, I give a brief summary of the arguments for my thesis.
1.3 The Origins of van Fraassen's Challenge

In this section, I give van Fraassen's list, with a brief explanation of
the items (section 1.3.1). Then, I give a reconstruction of the historical
origins of the list, up to the nineteenth century (section 1.3.2). Then, in
successive sections, I give three twentieth century responses to the problem:
the logical empiricist view of Hempel, the realist views of Armstrong,

McCall and Harre, and the antirealist view of van Fraassen.

1.3.1 Van Fraassen's List

Central to van Fraassen's argument against the existence of laws is his list

of characteristics anything would need to meet in order to be a law. The list

is a philosopher's list, but it is not purely the product of speculative

imagination. Philosophers constructed the list after reflecting on scientific

practice and the items scientists have called 'laws.‘ I will here give the list,

with a brief gloss on the meaning of each requirement.

The list van Fraassen gives (Laws 25-3 8) is:

1) Universality: The law must hold under all relevantly similar

circumstances. All F's must be G's if it is a law that F's are G's.

2) Relations to Necessity: Laws involve necessity in the following

ways:

2a) Inference: One must be able to infer the current
observable state of affairs from the law and current

conditions if this is a case governed by this law.

2b) Intensionality: One cannot simply obtain the truth of the
observable state of affairs by detaching it from the law-
operator, because the state of affairs is located in an
intensional context. The logic involved in the inference in 2a)

is not truth-functional.

2c) Necessity Bestowed: Any necessary connection between

natural events derives its necessity from a law.

2d) Necessity Inherited: The view (which van Fraassen calls
a minority view) that for laws to confer necessity on their

instances, they must themselves be necessary truths.

3) Explanation: Laws must explain why their instances occurred.

4) Prediction and Confirmation: Laws must be connected to
observable states of affairs so that our belief in the law can be
increased by viewing positive instances and our belief that the state

of affairs will occur is increased by our belief in the law.

5) Counterfactuals and Objectivity: Our belief in the connection
between events related by law must be such that we would expect it
to hold in unobserved instances and even in impossible to observe
instances as well. This is part of what we mean by the 'mind-

independent' status of laws, i.e., their objectivity.

5a) This includes context-independence. Laws are not

strongly theory dependent or language dependent

5b) Laws are objective: They are discovered, not constructed.
6) Relation to Science: Laws are a central (perhaps the central)
feature of science. Without laws, science could not perform the
intellectual role it does play for us.
I will show next that the list is neither idiosyncratic (section 1.3.2) nor
innocent (section 1.3.3, 1.3.4, and 1.3.5).
1.3.2. Sources of the Philosophical Problem
The list is van Fraassen's own. It sets up the pincer movement of his
dialectic against realism, where he shows that important realist views find it
impossible to solve both of two problems that emerge from the list, what van

Fraassen calls the inference problem and the identification problem.

The identification problem is the problem of empirical access. How
do we identify laws in experience? This is a problem for realists who do not
identify laws with observable regularities.

The inference problem is the problem of necessary connection. Why
must something which satisfies the antecedent conditions stated in a law also
satisfy the consequent conditions? We need an account of laws that enables
us to see how laws help us to predict and explain the sequence of events that
we find in experience.

The problem seems to be that one can set up any number of analyses
that solve the inference problem that leave us unclear how to solve the
identification problem. It is the very point of the metaphysics deployed by
the realists to explain the necessary connection between antecedent and
consequent conditions in laws. But it is the very same metaphysics that
makes identifying laws in experience problematic for the realists. I think van
Fraassen is right, against the realists, on this matter, and will give in chapter
two my diagnosis of the weakness in realism that van F raassen rightly
criticizes. However, before I do so, I need to Show that the list is not the
invention of either the realists or van F raassen, but forms the historical
backdrop against which philosophers of the present work. Where did this list
come from? Why should we take it seriously? Why is it so difficult to find
an analysis that successfully meets all the conditions?
1.3.2.1. Ancient Greek Reflections on the Nature of Change

Edgar Zilsel, in his historical article "The Genesis of the Concept of
Physical Law,"1 has noted that the term 'law' was not clearly and

systematically used to describe universal empirical regularities before

 

1 This section of the chapter relies primarily on Zilsel, Needham and Ruby.

Descartes.2 Descartes' usage became standard only because Newton used it
similarly in his enormously influential work, PrincipiaMathemetjca.
While I do not challenge Zilsel's history of the term, I start my own
historical investigation earlier for two reasons. First, the concept is clearly
older (e.g., Galileo has the concept but not the term). Second, even where
the concept is absent, there are discussions among the ancient Greek
philosophers that are both relevant to our concept and influential in its
construction. There is a long tradition, stretching back to the presocratic
philosophers, that discusses natural progressions and natural processes.
Although they do not call empirical regularities 'laws', what they say about
such regularities is sufficiently analogous to what we would call laws to
make what was claimed relevant to our discussion. Furthermore, I want to
claim that the modern concept of scientific law was influenced by these
Greek philosophers. By the time we get to the mature work of Plato and
Aristotle, we are discussing views that had historical influence on the
modern discussions in Descartes and Newton. In fact, I want to claim that
most of the contours of our present philcscphlcal problem of laws can be
found in the work of Aristotle, even though the term 'law' is not used.
Gregory Vlastos, in his book, Blatc'sflniyerse, finds examples of
invariant natural uniformities in the work of both the pre-Socratic
physiologoi, (nature-philosophers such as Thales, Anaximander and
Empedocles) and Plato himself. Although they differed among themselves
whether the reason for the order was to be sought within nature, or outside

nature, in some transcendent Being or deity, on this issue they were agreed:

 

2 The qualification 'systematically' is important: Jane Ruby finds uses in the modern sense in
the fourteen and fifteenth centuries by Roger Bacon in optics, and Regiomontanus in
astronomy and mathematics.

the regularities were inviolable and inviolate. So the path of the sun was
fixed, for Heraclitus, in a manner which included an entirely rational
explanation for eclipses. And for Plato, the behavior of the sun over the
course of the year could be entirely explained in terms of two different
circular motions that explained the daily circuit of the sun in the plane of the
ecliptic (the Same) and the yearly movement from the solstice north of the
ecliptic to the solstice south of the ecliptic, and back again (the Different).
Vlastos also claims that Plato had a theory of the behavior of material
(sublunary) things based on an account of the material of which they were
made (the four elements) and the forms.

While Vlastos shows the important, empirical, non-speculative
elements in Plato's otherwise poetic and speculative account, Plato's
approach seems remote from Newton's and Descartes'. Much more congenial
seems to be the comprehensive, detailed philosophy of nature we find in
Aristotle. His work contains a great deal that is of interest for our study.

But, before we can appreciate the similarities of Aristotle's project to ours
(and identify the historical continuities that led modern scientists and
philosophers to apply his criteria to what we call laws), it is important to set
out, briefly, some of the historical context that set the problems Aristotle
was attempting to solve. The speculations of the nature philosophers
focused on what one thing could account for the variety of things we see in
nature. Water, air, fire and earth, which began as rival principles in
competitor accounts of matter, were later combined in the so-called four
element theory of matter.

Parmenides, and his disciple Zeno, argued that the whole explanatory
enterprise rested on a mistake. While their arguments have more often been

ignored than refuted, those arguments compelled both Plato and Aristotle to

10

give careful justifications of the explanatory enterprise. Parmenides argues
that change is incoherent because the very idea involves the notion of non-
being, which is incoherent as well. The real either is or it isn't. But non-
being cannot even be thought. So everything real is. Change seems to
involve the notion of non-being. If something were to change, it would not
be what it once was, and would not have been what it is now. But if so,
change seems to involve the idea of being and not being at the same time,
which would be incoherent (indeed, contradictory). So change is impossible.
And since motion is a species of change, motion is impossible also. Zeno's
paradoxes are intended as proofs of the impossibility of motion by showing
the absurd conclusions to which the assumption of motion leads.

The idea that the intelligible is the changeless becomes explicit with
Parmenides's challenge, and both Plato and Aristotle respond to that
challenge by partially agreeing with him. Change is problematic, unless it is
anchored in something that does not change. Explanation will require
intelligible principles, not inert material elements. Both Plato's theory of
ideal forms and Aristotle's theory of substantial forms are accounts of
change that meet Parmenides's challenge by accepting at least some of his
assumptions.

Aristotle seeks to find the principle of invariance within the
phenomena themselves. Aristotle, in his Physics, gives an account of change
that identifies the invariant elements of bodies in motion as the so-called
four causes: material, formal (essence), teleological and efficient. A rock
may grow hot or cold, but the material remains the same throughout and
supports the qualities of hot and cold. A rock falls naturally because it is
primarily made of earth (formal cause supervening on the material of prime

matter), and earth 'wants' to be near the center of the Earth (teleological

11

cause). A rock may experience violent motion if hurled (efficient cause).
Explanations of biological changes are more complex, but the same
explanatory strategy prevails: seek the principle of change in something that
does not change through the process. Organisms come into being because
matter is formed in various ways that mark out the organism as a member of
a distinct kind with distinct activities. Plants grow by adding matter. They
mature by reaching the goal set by the kind of plant they are. Animals move
because they have the sorts of souls (form) that can initiate motion. Humans
think because their forms include rationality, and it is the peculiar and
proper function of humans to think.

These are not what we would call laws, but neither are they irrelevant
to the discussion of laws. Many of our laws are conservation laws, laws that
specify some quantity that remains invariant through a process of change.
Conservation of charge, momentum, angular velocity, mass (in chemical
equations), energy, mass-energy (in relativity theory) are all direct
descendants of the Greek project to locate the invariant in the variable.

Aristotle seems to be the primary historical source for the following
parts of the problem of laws (cf. van Fraassen's list above):

1) Universality: explanatory principles must make contact with the

essential characteristics or the definition of the thing. An explanation

that latches onto accidental characteristics will fail to be truly
explanatory of the characteristic behavior of the species (whether
rocks, plants, animals or planets), because it will not have picked out
the features that are true of all or most of the species.

2) Relations of necessity: principles must be necessarily true, and the

relations between subject and predicate in the conclusion of a correct

theoretical syllogism must hold of necessity.

12

23) The relation between the characteristic properties and
behavior of the things and its essence is deductive. From the
definition of the essence and the claim that one has a
member, one can infer the appropriate properties.

2b) The major premise of a theoretical syllogism is grounded
in the essence, and therefore necessarily true. Being always or
for the most part true is necessary, but not sufficient, for a
proposition to be part of a scientific explanation of a natural
event. Both the conclusion and the major premise are
typically universally true, but one cannot interchange them
and still have an acceptable scientific explanation for
Aristotle. The difference is the role essences play in each.
The sorts of changes a substance can undergo, Aristotle
maintains, are determined by the essential features of the
substance, and it is only in virtue of the essential features that
one has a rational explanation for the change. In brief, the
connection between the subject term (or universal) and the
predicate term (or universal) in the conclusion requires a real,
necessary connection between the subject term and the
predicate term in the major premise that appeals to the
essence of the thing designated by the subject. Since the
major premise functions like a law in Aristotelian science,
this is equivalent to the claim that a law requires a connection
that is not merely universal but necessarily universal.

2c) Anything deduced from necessary principles is itself

necessary.

13

2d) One incontestable case of necessary connection between
subject and predicate indicates a necessary principle that lies
behind and explains the connection.
3) Principles must be invoked in any correct explanation for why
phenomena occur or why things have the properties or display the
behaviors they do.
4) The features of prediction and confirmation are not covered as
such, and this point will be important later in the chapter. However,
unlike Plato, who found the phenomenal distracting, Aristotle
believes that the road to principles starts with the phenomena. This
leads him to the following important epistemic claims:
i) That which is most knowable in itself is not necessarily
what is most knowable to us. We need to come to principles
through the senses, abstracting the features embodied in the
principles.
ii) There are no transcendent forms or essences. All forms or
essences are features of spatio-temporal substances that
govern actual processes of change and movement.
5) Counterfactuals are not covered by Aristotle, because the analysis
of the conditional awaited the work of Stoic logicians like
Chryssipus. However, again, Aristotle does have an analysis of
modal propositions that will prove important in the late nineteenth
century and early twentieth century developments. Objectivity is not
discussed as such, but is present in such realist locutions as
'knowable in itself, ',clear in itself. The essences are objective in the
sense of residing in the object, and only present in the subject after

being abstracted from perception. But,

14

5a) Principles are context—independent in that their truth is
necessary independent of the truth of principles governing
other areas of being For example, principles governing the
behavior of plants are independent of those governing human
beings except of course where human behavior is plant-like,
in growth, nutrition and reproduction. Since they are
language independent (that is, true whether anyone
formulated them in language or not) principles are context
independent in not being relative to a choice of a language.
Essences determine the one and only correct language.
5b) Principles are objective in that we discover them rather
than construct them. They are not relative to any beliefs we
may hold. A fortiori, they are not relative to any theories we
hold.

6. Aristotle places these principles right in the heart of science.

Science just is the discovery and deductive arrangement of natural

principles.

In conclusion, a substantial portion of the problem of laws emerged early in
the reflection on natural regularities, even before the modern notion of laws
emerged. In the light of those reflections, the modern notion of laws was
formulated. This is not an especially odd result. While other Greek scientists
were known in the Middle Ages, Aristotle towered over all the rest. It would
be a pardonable exaggeration if we were to claim that Aristotle just was
science for his medieval counterparts, whether Islamic, Jewish or Christian
(Cf: Singer 150,152). Aristotle set the terms of the debate even for those

attempting to break free of his manner of conceiving the sciences of physics,

15

chemistry, astronomy and biology. An important element was added during

the Middle Ages to which we now turn.

1.3.2.2. Medieval Reflections on God 's Rule over Nature

If Vlastos is right, the Greeks originated the notion of 'cosmos', our
idea of an ordered universe. This is not identical to our concept of a law-
governed universe. But given the content of Plato's Ilmaeus, the medieval
thinkers who reflected on the ancient Greek corpus of scientific and
philosophical writings were not wrong to find analogies between the Judeo-
Christian God, who created the universe, and the Platonic craftsman
demiurge, who fashioned the world in accordance with the ideal forms. Still,
Zilsel, Needham and Ruby all agree that among the ancient Greeks, 'law'
was simply not used to describe natural regularities until the Stoics, and
even there it was primarily with respect to the moral character of the
universe. The notion of 'natural law' came in during the Middle Ages as an
expression of the law written by God on men's hearts, as opposed to the laws
of custom or positive law.

Nevertheless, in the Bible are hints of the notion of law as describing
God's providential reign over nature as well as his moral rule over humanity.
Although Augustine had coined the phrase 'etemal law' centuries before to
cover God's rule over non-rational as well as rational beings, Aquinas still
viewed 'law' in this context as metaphorical.3 The medieval view combines

Christian faith in God's rule with the ancient Greek conception of order.

 

3 Aquinas, WW: II, 1 qu 90,91,93,95 qtd. in Zilsel 256-258; cf. Ruby 341. "
Even creatures without reason share eternal reason in their own way, like reasoning
creatures; but because reasoning creatures do it with reason and understanding their sharing
in the eternal law is itself law in the true sense, for law is a product of reason, as we have
said. Non-reasoning creatures share it in a non-reasoning way, and so in them it is a law

W-"

16

Thus the notion of law here is teleological as in Aristotle, rather than
descriptive as in the modern view. Nevertheless, we do, in this period,
according to J. Ruby, get some of the first uses of the word 'law' in the
modern sense to cover the laws of geometry, astronomy and optics. It would
become natural to describe the contrasts between the old views of physics
and astronomy and the newly emerging views of the coming centuries in

terms of the laws governing those phenomena in each view.

1. 3.2. 3. Modern Reflections on Newtonian Mechanics

While the history of the idea of 'law' is interesting and well told by
Zilsel, Needham, and Ruby we are concerned here with what philosophers
have made of the idea. The early history of the concept of 'law' was
complicated for a number of reasons. The concept was present without the
word. The word was present without the concept. Although the Greeks
lacked our concept of law, they developed analogous concepts such as 'order'
and 'principle.' Properties of 'order' and 'principle', were later transferred to
the concept of law when it was formulated in the Middle Ages, as we have
just seen. Our attempt to separate out a philosophical problem of laws, as
opposed to a scientific problem of laws, was complicated by the
considerable overlap between science and philosophy before the modern era.
We could not cleanly separate the scientific development of the concept
from the philosophical development because many of the same people were
involved in both (e.g., the physiologoi, Plato and Aristotle).

Now, the two histories cannot be neatly separated in the early
modern period either. Descartes and Newton were creators in both science
and philosophy. The distinction becomes clearer afier science and natural

philosophy separate in the nineteenth century. A further complication arises

l7

in the modern period. Modern philosophers are reacting to the concept of
'law' as much as articulating it. Interpretation and explication go hand in
hand, for example, in the eighteenth century reflections on law. Still,
philosophers would work out the implications of the Newtonian
'mechanization of the world picture' over the next couple centuries
(Dijksterhuis 3-5). Twentieth century philosophers would continue to work
on the implications of the notion of 'law' in the light of quantum mechanics
and relativity theory. Philosophers from Hempel to van Fraassen would
work out the explicit, detailed formulation of those implications captured in
van Fraassen's list.

The scientific revolution was very complex, and summary historical
descriptions of it are controversial. However, since our purpose here is to
examine historically important reflections of laws, the risk involved in
simplification is worth taking. Buchdahl, in MetaphysicsantheEhilcscphy
cfScience, a study of historically important metaphysical reflections of
early modern philosophers, gives his summary of the features of the
scientific revolution considered salient by those philosophers. Buchdahl
claims that during the scientific revolution, the following occurred:

i) the reduction of the anthropomorphic character of the world (e.g., the
embrace of heliocentrism); ii) the rejection of transcendental perspectives
and the growth of materialism (e.g., the revival of Epicureanism and
atomism; Descartes relegating biology, physics, astronomy to the realm of
matter; Galileo's refusal to consult the Bible rather than experiment on
matters of physics); iii) the removal of teleology from the world, and the
embrace of mechanistic explanations (e.g., Descartes explanation of the
motions of the world, the behavior of animals and the functions of the

human body in mechanistic terms); iv) experiment; vi) mathematical

18

formulations replacing qualitative explanations in terms of substantial forms
and occult properties (cf. the work of Kepler, Newton, Galileo and
Descartes) (Buchdahl 51-78).

If such an assessment is accurate, then we have substantial
agreement during the scientific revolution on the items on van F raassen's
list. Since there is some shifi in the content ascribed to those properties
required of laws, it is worth looking at the list once again to see the state of
the discussion during that period:

1) Universality: Both the heavens and earth are governed by a

common set of laws. Both humanity and nonhumanity are governed

by the same set of laws. Laws are the same in the past and present
and future.

2) Necessity: Instead of seeking necessity in a connection between

what happens and God's eternal will, modern natural philosophers

now locate necessity in the invariable concomitance and succession
of events (determinism). Kepler is interested in the mathematical
relation governing the periodicity of the planetary motions, not the

Prime Mover.

3) Explanation: What theologians of the Middle Ages called the

secondary cause is now the only cause mentioned in science.

Newton's dynamics explains Kepler's law of planetary motion by

appeal to central forces. This, rather than the perfection of the circle,

explains the approximately circular motion of the planets around the
sun. Good explanations mention the mechanism or mundane cause.

Explanations are secular and in terms of efficient cause rather than in

terms of teleology or design.

19

4) Prediction and confirmation: Attention is paid to experiment and
exact mathematical formulation. For example, Boyle shows concern
to control relevant variables in experiments and experimental design.
Also measurement comes to assume a central place with the work of
Tycho Brahe and Galileo.

5) Objectivity: The rejection of anthropomorphism and its
replacement by materialism and atomism were part of a new notion
of objectivity that made matter mind-independent, and physical
models unlike social or psychological models. This also led to the
notion that truths about the world were context-independent in the
sense of not being strongly conditioned by language, culture, history
or social structure. Francis Bacon, for example, defends the view that
the world is the way it is, no matter what the church, state or
philosophical school say it is.

6) Relation to Science. While many disagreed over the structure and
methods of science, all agreed that science aimed to uncover the laws
of nature: While Newton, Descartes, Boyle and Francis Bacon agreed
on little else, they agreed on that.

However, our having said that all this was implicit in claims made

during the scientific revolution, does not make our examination of the major

philosophers of the next few centuries superfluous. After all, at least part of

the problem may be that philosophers misunderstood or even rejected the

exact notion of laws that modern science was working with. Buchdahl goes

so far as to suggest that Locke, Berkeley, Hume, Descartes, Leibniz and

Kant hadsignificant reflections on laws of nature in their philosophies as

part of their reflection on the 'nature of the propositional link' between

subject and predicate in judgment (Buchdahl 51-61). For Locke, we have no

20

knowledge of what things are really like (their real essence, in Locke's
terms), only how things appear to be (i.e., their nominal essence). We locate
empirical laws, for Locke, in the connection between the subjects of
appearance and their regular properties.4 For Berkeley, perceptible things
just were the set of their appearances, and the laws were simply the set of
successive appearances.5 The cause was God. For Hume, cause itself gets
analyzed in terms of Berkeleian regularities, which we project into the future
because we have acquired the habit of expecting the sequence to hold up.6
There is no natural necessity -- only psychological associations. Some
psychological associations are better than others, however. Because the
possibility of error is everywhere, we need to apply some canons to make

sure the regularities we project are worthy of projection.7 This anticipates

 

4 As Buchdahl shows, Locke's case, far from being the easiest, may well be the most
difficult, because in places he seems to have held two different views on the matter. One
view, the necessitarian view, maintains that there are real necessary connections in nature,
and it is only our ignorance of the real essences of things on a microlevel that prevents us
from learning of them. The other view maintains that there is no necessity to know, but
rather only the connection we find in phenomena (with the nominal essence of things). But
this added complexity only reinforces my point: necessity was viewed by these philosophers
as puzzling and mysterious. Relevant passages are found in Locke 527; bk. 4, ch. l-3.

5 Berkeley, Principles 78-80; pt. 1, sect. 30-32: "The ideas of sense are more strong and
lively, and distinct than those of imagination; they have likewise a steadiness, order, and
coherence, and are not excited at random, as those which are the effects of human wills
often are, but in a regular train or series-the admirable connexion whereof sufficiently
testifies the wisdom and benevolence of its Author. Now the set of rules or established
methods wherein the Mind we depend on excites in us the ideas of sense, are called the
laws of nature ; and these we learn by experience, which teaches us that such and such ideas
are attended with such and such other ideas, in the ordinary course of things... all this we
know not by discovering any necessary connexion between our ideas, but only by
observation of the settled laws of nature, without which we should be all in uncertainty and
confusion, and a grown man no more know how to manage himself in the affairs of life than
an infant just born." He goes on to worry that those who reify such laws 'wander afier
second causes' rather than the Creator who is the First Cause. Cf. Berkeley, QeMcm .
6Hume, Treatise 10-13; bk 1, pt. 1, ch. 4.

Hume, Ireatlse 155; bk 1, pt. 3, sect. 14 "Of the Idea of Necessary Connexion."

7Hume, Imatise 173-176; bk 1, pt. 3, sect. 15 "Rules by which to judge of causes and
effects" " l. The cause and effect must be contiguous in space and time. 2. The cause must
be prior to the effect. 3. There must be a constant union between the cause and effect. 4. The
same cause always produces the same effect, and the same effect never arises but from the
same cause...5...that where several different objects produce the same effect, it must be by
means of some quality, which we discover to be common amongst them...6. The difference

21

Mill‘s later reflections on method.8 For Descartes, the laws were contingent
choices of God's that led tothe regular order in the world. Since God also
sustained that order, it was as immutable as God.9 Such order was not an
empirical one, but an intelligible one. Its laws could be grasped only by
close, attentive, methodical analysis. On the basis of such a methodical
analysis, Descartes believed that he had a priori proof of the laws of motion
and optics. Leibniz needs to make the laws secondary, since all that exists
are the monads, which neither change nor interact. Monads have all their
predicates by creation. For Leibniz, perhaps even more than for Descartes,
error was persistent and difficult to remove. We think the monads inert, for
example, because our thoughts are confused. We have taken appearance for
reality. In reality, all monads are self-moving agents. Nevertheless, Leibniz

did not believe all appearances deceive us. Some phenomena -- space, time

 

in the efl‘ects of two resembling objects must proceed from that particular, in which they
differ- — - 7. When any object encreases or diminishes with the encrease or diminution of its
cause, 'tis to be regarded as a compounded effect, derived from the union of several different
effects, which arise from different parts of the cause...8. an object which exists for any time
“1 {‘5 fill! perfection without any effect, is not the sole cause of that efi‘ect, but requires to be
assrsted by some other principle, which may forward its influence and operation." Mill's
canons; are much clearer and his illustrations better informed by the science of his day, but
the Pnnciples are very similar.

“what is called the uniformity of the course of nature...is itself a complex fact,
compounded of all the separate uniformities which exist in respect to single phenomena.
These phenomena when ascertained by what is regarded as a sufficient induction, we call, in
comfy-s0“ parlance, Laws of Nature. Scientifically speaking, that title is employed in a more
giggicted sense to designate uniformities when reduced to their simplest expression" Mill
9D ’ pt 3, ch. 4 "Laws of Nature."

De:scartes 1: 92-8; W. ch. 7 "The laws of nature of the new world." For example,

it foff-l‘tes says "...God continues to preserve [matter] in the same way that he created it. For
must 0W8 of necessity, from the mere fact that he continues thus to preserve it that there
actio be many changes in its parts which cannot, it seems to me, properly be attributed to the

e ‘1 Of God (because that action never changes), and which therefore I attribute to nature.

a rtiles by which these changes take place 1 call the 'laws of nature.” and further down the
p ge That is to say, with God always acting in the same way and consequently always
1:32:10 ing the same effect...I shall set out two or three principle rules according to which it
expo lfe thought that God causes the nature of this new world to operate." and thence, an

81non of his three laws of motion in Descartes 1: 92-3

say:°frtes 1: 240-3; "Principles of Philosophy", pt 2, sec. 36, 37, 39,40,42; Descartes also

see ’ From God's immutability we can also know certain rules or laws of nature, which are
0ndilt'y and particular causes of the various motions we see in particular bodies."

Des‘mrtu 1: 40)

22

and matter -- qualify as well-founded, even if not metaphysically basic.
Laws are such. Nevertheless, laws do describe the rational order we find in
the world for us, because we cannot grasp everything in their particularity as
God can.10 Kant goes one step further: lawfulness is a condition of the
possibility of experience for finite cognizers like ourselves, a condition we
impose because we are rational.

What does this thumbnail sketch of modern pronouncements on law
contribute to our understanding of the problem of laws? The answer is
complex. A great variety of ontological and epistemological statuses are
ascribed to laws by the modern philosophers we have just discussed. What is
the source of their disagreement? It seems some of this disagreement stems
from disagreement over how to accommodate elements of the new and old
views of the place of God and humanity in the universe, the nature of the
order found in the world, and the possibilities of human knowledge of that

order- Locke and Leibniz, for example, seem dubious about whether laws
belong in an ultimate accounting of the fumiture of the universe. Berkeley,
with his project of defeating the skeptic simply identifies laws with
empirical regularities. Descartes agrees that the skeptic must be defeated, but
locates the possibility in pure thought. Laws for both are near the
foundations of empirical knowledge. Insofar as they all solve the problem
by rrlaking our immediate awareness to be of our ideas, they solve the
knowability problem by putting at risk its objectivity. Each of them tries to
secure objectivity for laws in his own way: Descartes by indubitability of
clear and distinct ideas, Hume by method, and Berkeley by sequences of

Ideas that are not subject to our wills.

 

10
There are difficulties in Leibniz's view of laws. cf. Buchdahl 461-464

23

If we read what they say, however, with van F raassen's list in mind,
we find that they do indeed seem to take something like that list for granted.
In fact, they have re-interpreted some things we already found, in our
discussion of Aristotle, into what we now would call the standard
interpretation. Let us look at the list again and see how these early modern
philosophers have advanced the discussion:

1. Universality is taken for granted as a legitimate demand for

principles. However, in the light of skeptical objections, it is

problematic. The rationalist Descartes claims that laws are known a

priori. The empiricist Berkeley claims that laws are known by

immediately perception. But Leibniz and Hume, who agree with

Descartes and Berkeley on so much disagree with them here. The

laws, if knowable, are not known with certainty precisely because it

is hard to know if they are universally true, Hume would argue.

Furthermore, even if known, what is known lacks the universality of

a true metaphysics, Leibniz would argue, because it is not true of

what basically exists, the monads, but only how they must appear to

beings such as ourselves.

2. The necessity ascribed to laws, owing to their inviolability, is

problematic. Aristotle had seen the need for distinguishing physical

necessity from other types, but with the repudiation of Aristotelian
substance and the notion of essences, the nature of physical necessity
becomes mysterious. While Descartes and Leibniz locate the
necessity in God, Hume makes necessity a creature of the mind.

Whether necessity can be bestowed from God to creature or whether

the laws are necessary truths or not, is already matter for debate.

24

3. Inference becomes mathematical at this stage. Galileo talks about
the book of nature being written in the language of mathematics, but
it is far from clear what the justification is for applying mathematics
and mathematical reasoning to natural phenomena. An unanalyzed
Pythagoreanism seems to be the root of the mathematization of
physics that occurs in this period.

4. The problems of intensional contexts and opacity of reference
have not been developed yet. The assumption is that any idea or
perception can be had in common by anybody, and that language is
simply the expression of those ideas. However, all is not clear here.
Francis Bacon and, following him, Locke, already warned about the
distorting influence of language. Leibniz denies that our ideas can
always be made clear to others or to us.

5. That laws explain occurrences in the natural world does seem
taken for granted by all parties that accord the laws ontological
ultimacy. For Hume and Leibniz, however, the explanatoriness of
laws is an issue.

6. Prediction and Confirmation are rendered problematic by Hume,
who questions the justification of induction.

7. Counterfactuals are raised in unclear form with Descartes's
'hypothesis' about the origin of the world from chaos and Leibniz's
notion of possible worlds. In both cases, the issue is the explanation
of the current order of the world or the current set of laws. Descartes
argues that the laws are contingent on God's choice, but necessary
after he chooses them. Given this set of laws, our world will
inevitably evolve into its current state, regardless of its original state.

Laws have a hypothetical necessity. For Leibniz, however, there

25

could only be one set of laws. God had to choose the set of laws that
was the best. To have chosen differently would make God arbitrary
rather than the all-wise being he is. But this means that what the laws
are is not a subjective matter even for God. This may be the closest
Leibniz came to a formulation of the objectivity of laws. For
Berkeley, the objectivity of the world order depends on the fact that
we cannot change, by an act of our will, the ideas we have while
witnessing natural events. He contrasts this with imaginary ideas,
which we can change by an act of the will.

8. Context independence appears of several sorts. Thoughts and ideas
are independent of language and culture. We see with our own eyes
and think with our own minds. This makes the individual
independent of his or her history. Next, thoughts are independent of
each other. So, for example, the monads of Leibniz do not really
influence each other, Descartes' clear and distinct ideas do not
require each other, and Hume's associated events are independent.
This 'atomicity' of analytical units in the different philosophies is not
altogether unlike the behavior of individual physical atoms in
revived atomism. The behavior of one 'atom' has nothing to do with
the behavior of another, unless they collide. The laws, apparently,
need have no connection with each other, except, possibly, in the
mind of God.

9. The assumption they all do make is that science concerns itself
with laws. Some, like Hume, want to formulate similar laws for

moral philosophy based on the psychological laws of association.

26

1.3 .3. the Standard Empiricist Model of Laws
The previous section showed, I think, that matters were far from tidy

among philosophers reflecting on the nature of scientific laws. This should

hardly be surprising. Both empiricists and rationalists were trying to solve

problems of knowledge, and had set the standard at certainty. It is far from

clear, however, even with the tremendous achievement of Newton, that the
methods of experiment lead to certainty. Kant was not mistaken to make that
point. The emphasis on method in both traditions served to demonstrate that
it is easy to lose one's way in the world without systematic procedures.

The twentieth century brought with it a number of scientific
revolutions. Darwin's revolution finally won some major battles in the
universities. Sociology and psychology claimed that human social and
psychic life was fair game for scientific exploration. These revolutions
pushed the universality claim rather far, much farther than Descartes or
Newton would have liked, since they believed the mind was special. But,
after genetics and biochemistry discovered that there was nothing special
3130111: the chemicals involved in living tissues, there were attempts to
Progressively reduce the laws of psychology to the laws of physiology, then
to the laws of chemistry, and finally, to those of physics. Reductionism,
altho‘ngh not universally embraced, testifies to the belief that the laws of
physics are indeed universally true.

The revolutions that brought about renewed philosophical reflection
in the first half of the century were the revolutions in mathematics and
phySi cs: the rise of quantum mechanics, relativity theory, and mathematical

logic - These led to the work of the logical empiricists who gave a formal

treatiInent of the empiricist view of laws we found in Berkeley and Mill.

27

It is a feature of positivism, especially the positivism of Mach and
Duhem, that it is the role Of science to provide economic abbreviations of
experience. Since mathematical laws are brief in statement, but have a
potential infinity of instances, mathematical laws are paradigms of scientific
knowledge. When Hempel and Oppenheim gave their initial analysis of laws
in l 948, it was part of a larger analysis of scientific explanation. Laws were
important because explanations were seen, as in Aristotle, as deductions of
empirical states of affairs from general principles. To be able to do that, laws
needed to have the right formal structure and empirical content, and the
pro j ect Hempel and Oppenheim undertook was to state clearly the formal
and material conditions for lawhood. The problems this project encountered
are vvell known to those familiar with the troubled career of the Verifiability
Criterion of Meaning. Nevertheless, since the analysis Hempel and
Oppenheim gave explicitly formulates properties of laws that are implicit in
car] i er philosophical discussions of laws, the analysis is still valuable. And
since, for a time, it represented the standard view, both the reasons for which
it Was held and the reasons for which it was rejected, hold interest.

The analysis consisted of two parts. First, one needed to get the
forT1121] structure right. One gave the syntax for some formalized language
that Would enable one to pick out the well formed formulas and the correct
transformations of those formulas. Next, one gave the symbols in that
la“gllage semantic content. This enabled one to distinguish the function of
“”1115. Some did logical work, others designated empirical objects or their
p1“Denies. This sufficed to give one a lawlike statement, a statement that
could be either true or false. It is formal because one stipulates in advance
how one is going to use the symbols, and one can use them correctly even if

one does not know what those symbols mean. The second condition is not

28

formal, but material. The lawlike statement must be true. That cannot be
known apart from experience. So one would first set up some model
language L which allowed inferences in the lower functional calculus
Without identity. After introducing both the logical symbols and the
extralogical ones (constants, variables, predicates of arbitrary degree), one
then defined the atomic statements. From these, sentences of arbitrary
complexity could be built up by means of truth-functional connectives ('and',
'or', 'if...then', 'not'), and quantifiers ('all', 'some'). Once the syntax was in
pl ace, one would go on to define important semantic concepts, such as
'formal truth in L', and give the semantic constraints on substitution
instances for the constants (e.g., any physical object) and predicates (purely
qual itative). Having set up the syntax and semantics of the language in this

manner, Hempel defines a lawlike sentence in L as follows:
S is a fundamental lawlike sentence in L iff S is purely universal.
Simi larly, one defines a law in L as follows:

S is a fundamental law in L iff S is purely universal and true.

Hempel suggests that (x)[ P(x) ...Q(x)] would be the form that a law
would take in the formal language L.

Later, Hempel came to have some reservations about the confident
prol'louncements about the form of a law, since laws were not often
exIDl‘essed in formalized languages. 'This object is soluble in water,' he said,
is sit'rgular in form, although it expresses a law, and may also be expressed in

a generalized form.ll Nothing important hangs on this worry for our

\

29

purposes, since all parties to the dispute grant that many of the interesting
laws of science are expressed in mathematical form and wear their
generality on their face, so to speak. What is interesting for our purposes is
what parts of the traditional characteristics of laws are preserved, and what
parts are given up in this style of analysis.
1. Universality: The law is given a formal treatment as a universally
quantified statement in some formal language. If the statement is
true, its universality is guaranteed by its form.
2. Necessity: Relations to necessity are more problematic. The
necessity admitted is a purely syntactic one, that of the formal
deductive relationship between the general claim and its instances, or
between the explanation (the law and initial conditions) and the event
to be explained. Any form of physical necessity, in addition to this
logical necessity, is denied. Intensionality is denied as well: the
formal language is set up with unambiguous and precise semantics
(for both the logical symbols and the extra-logical ones) , so that the
truth-conditions of law statements are purely extensional.
3. Explanation: Explanatoriness is placed center stage in this view. It
is precisely laws which make scientific explanations distinctly
trustworthy. Again, syntax is central since the relation between
explanation and the events explained is conceived as a deductive
relationship between the sentences that express them.
4. Prediction and confirmation: It is the logical empiricist analysis
which makes both of these quantitative. Past empiricist views,
including early logical empiricist formulations, had trouble here, but
thanks especially to Popper's analysis of empirical strength, the

logical empiricists moved away from concerns for strict verification

30

and falsification towards more sophisticated views of partial
interpretation, confirmation, and statistical theories of data.

5. Counterfactuals: Nelson Goodman is the logical empiricist who
first discussed, explicitly, the connection between counterfactuals
and laws: it was a problem that required one first see why material
conditionals were not sufficient to capture all the content of laws.
Too many conditionals would qualify as laws, since it suffices for the
truth of a conditional that its antecedent is false. So, for example, if
no diamond was ever immersed in butter, the conditional 'if a
diamond is immersed in butter, then it dissolves.‘ is vacuously true,
but few of us would want to call it a law because of that. Some
stronger connection seems to be required than the truth-functional
one. We want to say that even if a diamond never is immersed in
butter, it is still false to say that if it had been so immersed, it would
have dissolved.

6. Objectivity was the point of formalizing logic and cutting it loose
from the psychologistic form it had acquired as the so-called laws of
thought. Formalism is objective because, once the syntax is clear,
decision procedures can be set up for significant fragments of the
language that are purely mechanical in their operation. No opinion is
involved in the inference 'p. . .q & p therefore q'. The inference is
decidable mechanically. Verification was seen as guaranteeing
intersubjectivity, since observation is publicly checkable and
challengeable. However much bias might operate in the context of
discovery and in the context of justification, its presence did not, by
itself, undermine the justifiedness of a law. The only relation that

counted was the logical relation between the claim expressing the

31

law and the claims expressing the evidence. Laws stood or fell only

in relation to the evidence, not in relation to the psychological states

involved in their formulation. Furthermore, what laws there were
was not a matter of logic either. Laws were part of the world, among
the 'fumiture of the universe', as contemporary metaphysicians might
say. Laws were discovered, not invented. Truth-conditions for
sentences were such that the truth of sentences varied independently

(so long as there were no syntactic relations between the sentences).

This is context-independence in van F raassen's sense: the
truth or probability we would assign to law claims is viewed as
independent of other beliefs we hold, including theoretical
commitments and more general background beliefs. ‘2 It is a view
that would be abandoned as Camap and Quine moved in the
direction of more holistic views of language and sentences.

1.3.4. the Realist Challenge to the Standard Model.

A protest arose among some philosophers that the standard
empiricist analysis of laws would not do. While this protest was apparently
of long standing,'3 the protest became more focused after Hempel and
Oppenheim and N. Goodman published their papers. Kneale protested that
the analysis missed the necessity of the regularities governed by laws.'4
Chisholm, criticizing Goodman's analysis of counterfactuals, suggested that

perhaps the atomism of the regularity analysis is at fault. Perhaps there is a

 

'2 Since for van Fraassen both explanation and counterfactuals require pragmatic conditions
(that is, reference to what the speaker and bearer know, don't know and believe about the
situation) as conditions of their acceptability, and hence, require contextual considerations in
their evaluation, laws, which are intimately tied to counterfactuals and explanations will
require context for their evaluation too. No theoretical model holds of the world without
some 'theoretical hypothesis' concerning the intended interpretation of the model.

'3 W.E. Johnson, aiming at Hume, in 1924 protested that a regularity view of laws missed
law's 'compulsiveness' cf. Johnson, pt.3.

t4 Kneale, "Natural"; "Universality"; Chisholm, "Contrary-To-Fact."

32

connection afier all between antecedent and consequent in counterfactual
conditionals.

It is in the late 1970's, however, that some rival analyses of laws are
offered. Three of these will be discussed in chapter two. I would like here to
sketch their complaints against the empiricist view, the nature of their
alternative analyses, and the reason they believe their views better describe
what has been identified in the tradition as 'laws of nature.'

Perhaps the most detailed critique came from D.M. Armstrong. For
the sake of brevity, I will focus here only on his critique. In the first five
chapters of his book, WW, Armstrong criticizes, by
turns, first, what he calls the 'naive regularity theory' and second, its
attempted sophistications. The 'naive' view identifies laws with
exceptionless empirical generalizations. It is naive because it makes no
attempt to distinguish accidental regularities from those we would regard as
laws. Such a view gets both the extensions and intensions of the term 'law'
wrong. An analysis gets the extensions wrong if something can be an
exceptionless regularity (he calls it a 'Humean regularity' ascribing such a
view to Hume) without being a law, or can be a law without being an
exceptionless regularity. That is, the analysis does not pick out all and only
instances of law, and consequently, gives neither a sufficient nor a necessary
set of conditions for lawhood. For examples of actual regularities that do
not correspond to laws, he shows how to construct Humean regularities that
apply to exactly one thing, or at most a finite number of things. For
examples of (logically possible) laws that are not Humean regularities, he
produces thought experiments of laws that are strictly local, laws that have

no instances (because their conditions are never fulfilled), and laws with

33

instances missing (because they are represented by a continuous function,
with an uncountable infinity of instances, and our samples are all finite).15
As Armstrong argues how the regularity view gets the intension (or
the meaning) of law wrong, he deploys concepts that remind us of items on
van Fraassen's list. Here are the subheadings of chapter 4 of Armstrong's
book, with my gloss showing where each item fits into van Fraassen's list:
1. Lack of Inner Connection: This corresponds to item 2 "relations of
necessity", especially 2c 'necessity bestowed' Armstrong denies this
element of laws is captured in the regularity analysis at all. (Since
Armstrong believes laws are contingently true, he rejects 2d). 2b.
'Intensionality' is captured in the whole chapter.
2. Laws of Nature as Principles of Explanation: Item 3
("Explanation") of van Fraassen's list.
3. Paradoxes of Confirmation: Item 4 here. Armstrong thinks the
regularity analysis is incomplete because it allows irrelevant
information to confirm a law.
4. The Problem of Counterfactuals: Item 5 on the list
("Counterfactuals and Objectivity"). Armstrong thinks the regularity
view cannot allow counterfactuals ever to be true. If laws supervene
on factual states of affairs, and would not be the same if the

constituent states of affairs were different, then any state of affairs

 

'5 The details are unimportant, but to give a flavor of the arguments here is how one
constructs a trivial Humean regularity that intuitively is not a law. First, take any object you
please, and give an exhaustive list of its properties. Some subset of that set will probably
suffice to distinguish that thing from every other thing in the universe. Second, make the
conjunction of those properties the antecedent, and the conjunction of the remaining
properties the consequent. One now has a true universal generalization (the antecedent could
after all describe more than one thing although in fact it does not) that is only true of one
thing, and no one would want to call this a law.

34

that is not the case is ruled out as physically impossible. If so, laws

do not support counterfactuals, but rule them out.

5. The Problem of Induction: Item 23 ("Inference") and item 4

("Prediction and Confirmation"). He thinks that the Humean view

founders on Hume's

Problem. Experienced instances can never license inferences to

unexperienced instances.

The items on van Fraassen's list that are missing from Armstrong's are
universality and relations to science. Armstrong grants universality to the
regularity theory because he agrees that universality is a property of laws
(although he thinks the regularity theorist is not well placed to defend the
claim that laws must be universal truths). Armstrong grants to the regularity
theorist as well that laws are an important part of science.

What does Armstrong offer in place of the regularity view of laws?
Armstrong maintains that laws are contingent relations of necessitation
between universals. This view focuses on the objective truth-makers of
scientific law claims and overcomes the subjectivity of sophisticated
regularity views that relativize laws to conditions of knowing them.

How, according to Armstrong, does this analysis succeed where the
regularity view fails? First, since the relationship is between the universal
properties of things, and not the particular things themselves, universality of
the law is guaranteed. Next, it is in virtue of this relationship between
universals that what is governed by laws happens necessarily. 'Necessitation'
guarantees that. The relationship is intensional. The antecedent gives the
reason for the consequent, not simply that both happen to be true. This
reason grounds the instances, and explains them. This account solves the

induction problem because it supplies a necessary connection between

35

natural events. Also, because universals are intelligible, they provide criteria
for picking out and identifying events as relevantly similar. With natural
necessity and criteria of identity, both prediction and relevant confirmation
are possible. We can now understand, Armstrong maintains, how laws can
support counterfactuals. Universals can have more instances than they do. If
some universal had additional instances, and if a law bound that universal to
a second universal, then the additional instances of the first universal would
also be instances of the second universal. Laws are thoroughly objective,
since they are relations between parts of the world, not our language or
thought. And laws are context-independent in all the relevant senses, since
they do not depend on theory, language or culture. Laws are even
independent of each other (except when we are dealing with hierarchical
relations between universals). Finally, Armstrong claims that science

discovers such laws, and that they constitute an important part of science.

1.3.5. Van Fraassen's Antirealist Counterchallenge to the Realists

Van Fraassen disagrees. As always, what would solve our problems
if true does not solve our problems tout court: our problems may in fact be
insoluble. Van Fraassen claims that the realists have not solved the problem
of laws at all because what they call laws do not have all the properties we
expect laws to have. To see this, van Fraassen argues, we need to see that the
requirements these properties place on would-be laws do not all pull in the
same direction. In fact, these properties fall into two groups, and the groups
of properties set requirements that pull in opposite directions. He labels
these requirements 'the problem of identification' and 'the problem of

inference.‘ Of these he says, "An easy solution to either spells serious

36

trouble from the other."(Van Fraassen, Laws 38) The simple explanation of
the problems is as follows:

1. "The problem of inference is simply this: that it is a law that A

should imply that A, on any acceptable account of laws." (3 8-9)

2. "To answer these queries one must identify the relevant sort of fact

about the world that gives 'law’ its sense; that is the problem of

identification." (39)

The inference problem is a problem for views, like Armstrong's, which
emphasize the intensional character of the law-operator 'it is a law that...’
The problem is: how does one go from 'it is a law that A' to the truth of A,
since the latter only gives information about events and sequences of events
in the world. To explain what happens, to ground predictions, one needs to
use laws as inference tickets.

On the older empiricist views, this would confront one with the
problem of induction, because all the evidence for a law would be old, and
the predictions would involve inferences beyond one's evidence set. The
realists come in at this point and say 'we can ground the inference and solve
the problem of induction by a more adequate theory of laws.’ Van Fraassen
agrees: most realist views solve the problem by construing 'it is a law that...’
as 'it is necessary that...‘ and use modal logic to infer 'A' from 'necessarily A.‘
(van Fraassen, Laws 39) Armstrong, however, has a problem here, because
his view that laws are contingent relations of necessitation between
universals leaves untouched the problem of the relation between their
instances, and it is the latter relation that needs to be clear if we are going to
be able to explain or predict why some event occurred in terms of its
correlation with antecedent conditions (Laws 96-99). After all, except in the

case of so-called singular causation, there are always antecedent conditions

37

of importance picked out in causal laws (and, a fortiori, in non-causal laws
like the universal gas law and Ohm's law). The problem would be solved
completely if the relationship between antecedent and consequent were one
of extensional inclusion: that just is what makes the class logic of the
Aristotelian syllogism work: if all humans are mortal, and all university
professors are human, then all university professors are mortal ( Laws 97).
But Armstrong claims that getting the extensions right is not enough. Even if
all F's were G's, it does not follow that it is a law that F's are G's. Something
more, and van Fraassen insists, something different, is required. But it is
hard to know what different kind of relationship could do the same job as
extensional inclusion. The problem of inference, therefore, requires one say
enough about the relationship to explain how it can license inferences. Van
Fraassen thinks that possible world semantics, similar to the version created
by D. Lewis, almost solves that problem by modal logic.16 D. Armstrong,
however, does not solve that problem. Calling a relationship one of
necessitation may make it sound like one has explicated 'necessity'. But
without more, this is simply a case of similar sounding words, and one
solves nothing by coining a word. (Remember the 'dormitive powers'
invoked in Moliere's Bcnrgeciseflemleman, as a supposed explanation for
why opium put one to sleep). To make the relationship primitive (that is,
incapable of analysis or definition in terms that do not involve universals or
modality) is to leave the problem of inference unsolved, because the
unpersuaded reader is lefi in the dark as to how the relations between

universals can have anything to do with the relations between empirically

‘

’6 Van Fraassen, Laws 45,64: "As we have found, on Lewis's account, the assertion that it

is a law that A entails that it is physically necessary that A. This meets one of our main
criteria," and "The inference problem is thus successfully handled." Cf. van Fraassen's
discussion of Pargetter and McCall in van Fraassen, Laws 65-93.

38

accessible events, since these two sets of relations are on different levels,
ontologically speaking.

The problem of identification moves in the opposite direction, from
phenomena to laws (Laws 72-73).17 The question here is how can we
identify the laws empirically. Presumably, for the realists Pargetter and
Lewis, once we have the laws, the problem of inference can be solved. But
the problem for all realists, including Armstrong, is that laws are not to be
identified with anything empirical, certainly not with empirical regularities.
So now the problem for the realists is similar to the problem for
theoreticians whose theories always outrun the empirical evidence; how
does one pick out, out of all the mutually incompatible views that are
indistinguishable empirically, which view is the correct view? For possible
world theorists, we can explain the necessity of laws if we can say that laws
are true in all the physically possible worlds. But to be able to say that, one
has to define the accessibility relationship 'physically possible relative to
world X'. However, there are an infinite number of relationships between
possible worlds. How do we identify the one we are looking for? By
description? That would not work because any description will leave a very
large class of isomorphic relations. But, for realists, the real relationship
must be unique. How then shall we pick it out? By pointing? But again,
because there are no causal connections between possible worlds, we cannot
point to entities in any world but our own. Van Fraassen believes that
Armstrong has his own version of this problem. He asks, how, on
Armstrong's view, does one determine which relationship is the relationship

01’ necessitation between universals? One cannot solve the problem by

\‘1 For discussion of the version of the problem for McCall's branched space-time model,
see Van Fraassen, Laws 74-79.

39

simple stipulation, by claiming there is only one relationship and that is the
one we are talking about. That would be to "baptize the relationship in
absentia," he says, which is not useful for those who are unpersuaded of the
existence of such and want to be convinced.

Van Fraassen believes that exactly similar difficulties arise for realist
views of objective probability, that are constructed to handle probabilistic
laws (Laws 76-86;151-214 esp. 160-170). The questions remain, which of
the empirically indistinguishable models contains the relationship of
objective chance (identification problem), and how does objective chance
help us set our subjective expectations so that the probability we assign to
occurrences matches what we empirically find to be the case? The strategy
of inference to the best explanation does not help here. In fact, that strategy
leads to incoherence in the set of personal probabilities we assign to beliefs,
as we systematically overweigh or underweigh evidence in a way that makes
us subject to Dutch Book problems (160-170, esp. 166-169).18

The upshot of van Fraassen's brief against realist theories of laws
can be stated briefly. Just because the philosopher can show the world
behaves as if laws were relations of necessitation between universals, or
relations between sets of possible worlds doesn't mean these really exist. For
us to claim that laws exist will require that we have clear criteria for telling
when we are confronted with a universal, for distinguishing one universal
from another, and for picking out the relationship between universals that
licenses the inferences we want. It is not enough to consistently describe the

relationship.

'8 The problem comes, van Fraassen claims, when one gives bonus points to hypotheses for
CXplanatory success beyond what they would derive from Bayesian conditionalizing of
belief on evidence.

40

Van Fraassen goes on in W to argue that not

only do we not have what realists would call laws, but we don't need them
either for rational belief or to do science. First, Van Fraassen explains how
we can have rational belief revision without laws. Realism leads to irrational
beliefs about objective probabilities because the beliefs are inconsistent.
Conversely, biting the bullet and being explicitly subjectivist about
assignment of probabilities allows one to calibrate belief by conditionalizing
on experience (129-214). Second, he shows that the great conservation laws
can be deduced from symmetry considerations, and that laws therefore are
entirely dispensable in science: science can and now does simply work with
the symmetries. Laws, in the classical sense of inviolable rules of the
possibility and impossibility of events, are nowhere in evidenCe in quantum
mechanics or relativity theory (215-348).19

1.4 Thesis

My thesis is that an empiricist can meet the burden of saying what
additional conditions a regularity must meet to be a law without embracing
realism.

It is the task of this thesis to show that van Fraassen exaggerates. We
can and do have in science laws that are not simply derivable from
symmetries of problem situations. It is important that we do because purely
formal laws give us no information about the world.

To make my case, therefore, I need to register a different diagnosis
of where the realist views go wrong and show how the challenge of the

realists and van Fraassen can be met in broadly empiricist terms. Part of my

 

‘9 Part three of van Fraassen book develops the notion of symmetry and its use in theory
constraint for geometry, mechanics and relativity theory; part four develops the notion of
symmetry relevant to probability theory.

41

work, therefore, is critical. In the next chapter, I discuss and criticize the
central presuppositions of three realist views. In chapter three, I develop my
alternative, a pragmatist view building on the work of those in the
pragmatist tradition who have analyzed scientific laws. In chapter four I
will discuss objections to my view that realists, constructivists, and van

Fraassen might raise about the stability of the position I defend.

42

 

Chapter Two: Three Realist Positions

2.1 Realism's Stake in Science

Realism seems to the reader to be an alternative preferable to the
verificationist empiricist view of science because the realist claims that
empiricists make science to be about us and our states of mind, whether
sensations or beliefs, when in fact science is about the world. Realists claim
to understand what science is about. I intend to show in this chapter that
realism is an unattractive view that cannot make good on its promises
because of the very assumptions on which it most insists. A fortiori, its view
of laws is unacceptable and should be rejected. Another view of laws is to be
preferred, one that squares better with our widely shared assumption of the
continuity and interactions between natural systems, including the systems
that we are. Such a view will be developed in chapter three.

2.2 Three Realists Positions: D. Armstrong, 8. McCall and R. Harre.

2. 2.1 Laws as Relations between Universals (Armstrong)

David Armstrong defends the view that laws are relations between
universals. He sets up the analysis by showing that there are certain
problems insoluble on the regularity view. 20 The most significant problems
are these: 1) The regularity view gets the extensions of laws wrong; there
can be regularities that are not laws and laws that are not regularities. 2) The

regularity view gets the intensions wrong: even if every law had a

 

20 This is similar to his strategy in arguing for universals. He first shows that there are
insuperable problems with all versions of nominalism, except possibly moderate
nominalism. He then argues for a realist view as having fewer problems and as having more
resources to handle the problems than nominalism could.

43

corresponding regularity, regularities do not handle the modality of laws,
i.e., the necessary connection between items related by law. 3) Attempts to
sophisticate the regularity account add an irrelevant subjective element: laws
may exist even if no conscious beings existed.21

What is at stake for Armstrong in an account of laws is their
objectivity. He wants an account that gives necessary and sufficient
conditions for lawhood and that furthermore gets the meaning of 'law' right.

Armstrong believes that to get laws right, one must do ontology. He
wants to find a real truth-maker, in virtue of which claims about laws are
correct. So, to him it is very important to make and maintain a firm
distinction between laws and law expressions. The phrase 'law expressions'
seems ambiguous between linguistic expressions (i.e., the laws in science
books) and manifestations of laws in instances.22 However, the ambiguity
doesn't matter because Armstrong believes each of the law expressions
covaries independently of laws and that both undermine the regularity view.
We will take up both in turn later, but the importance of the distinction is to
drive a wedge between the ontology or metaphysics of law, on the one hand,
and the epistemology of law, on the other hand.

The solution, for Armstrong, is to claim that laws are relations

between universals. What are universals? What kind of relations between

 

2' This again parallels Armstrong's argument against nominalist views of universals. The
predicate nominalist, the concept nominalist, the class nominalist, and the mereological
nominalist all tie universals to something external, but his intuitions tell him these can all
exist or fail to exist independently of each other. Armstrong believes that the Platonist, who
holds to transcendent universals, falls afoul of the same problem. Armstrong's immanent
universals, embedded in concrete states of affairs is the only view that characterizes
universals by something intrinsic. He believes that an account that gets the intensions right
automatically gets the extensions right.

22 See his discussion of iron and oaken laws, and the problem of infinitely qualified laws in
Armstrong, m 147-150; 27-29.

universals counts? How will claiming universals have this kind of relations
solve the problem of laws?

Universals are ways in which existing particulars might be.
Fundamental for Armstrong's ontology are states of affairs. States of affairs
comprise particulars and their associated properties. Each state of affairs is
located in some particular place in space-time, and collectively they
constitute the objectively existing truth-makers for claims made about things
in the world. The universals that exist all have instances. Universals are the
abstractable parts of a state of affairs. They are either properties of
particulars or relations between particulars. Unlike particulars, universals are
repeatable and can be realized in new instances. Why suppose universals
exist? For Armstrong, the decisive argument is that two objects that can
truly receive the same predicate must have some ontological ground for the
predication. That is, there must be some sense in which we can say two
nonidentical things are identical before we can correctly attribute the same
properties to them. Universals must be fully determinate. If some object has
a weight, there is only one weight that it has exactly. If it has a color (say
red), there is one shade of red it has, not another.

Universals can be related in more than one way. Clearly, they can be
related by subsumption or inclusion. More general concepts can embrace
less general ones. All mammals are animals. But this analytic or a priori
relationship of inclusion does not interest Armstrong. It may even be
problematic for him if all universals are determinate and none are
determinable.23 But the important relationships are contingent relations of

necessitation. He doesn't simply want it to be the case that two universals

 

23 Nevertheless, he believes universals may be analyzable in terms of other universals, and
that this may be true of every universal. There may be no primitive unanalyzable universals.

45

share all their instances. A nominalist about universals could endorse that
view and use it in a version of the regularity theory of laws. No, what
Armstrong is interested in is an intensional relation that is nevertheless
discovered a posteriori by science. He wants, for cases where it is a law that
F's are G's, that everything that is an F is also a G myimecfheingefl. It
is this 'in virtue of which makes the account intensional and makes the
universals essential to the account.

How does such an account solve the problem of laws? According to
Armstrong, it makes clear why the counterfactuals associated with laws are
true. He believes that mere regularities, even sophisticated ones, cannot
handle counterfactuality. A mere regularity suffices for lawhood, even if we
have reasons to think the correlation accidental. Absence of instances means
we have to regard an occurrence as nomically impossible. Both of these
conclusions are counterintuitive, Armstrong believes. Functional laws have
missing instances that we believe are nomically possible even if not actual.
Probabilistic laws tell us what might happen, even if it does not. The
presence of favorable instances does not tell us whether we are confronted
with a deterministic law, probabilistic law, or coincidence. Besides instances
we need the reasons they are instances, reasons that can sort out various
kinds of law (e. g., stochastic from deterministic) and laws from non-laws.
So, necessitation explains the necessity we associate with connections
between events made by laws. Necessitation is a contingent relationship.
Nomic necessity is not logical necessity, because another set of laws might
have been our laws. It is a primitive relationship in that there are no more
basic terms to which it might be reduced by analysis.

Now, 'necessitation' may sound like a relationship that guarantees

that if one has an F, one also has a G, that is, 'a is an F' entails 'a is a G'.

46

However, for Armstrong, this is not the case. The law might be a
probabilistic one, for example, and only give a probability that, say, half the
atoms of radium in this sample will decay in 10 seconds. For those laws, one
only has the probability of necessitation. Furthermore, there are some laws
that say what will occur W, and other laws that say what will
happen if conditions are right or if the operation of other laws does not
override or block the expression of this law. The former, the 'iron laws', as
well as the latter, 'oaken laws', both involve necessitation, even though they
differ in their expression.

'Necessitation' is an objective fact about the world, not to be
confused with the semantic relationship of predication. Similarly, Armstrong
claims that universals are objective constituents of the world, not to be
confused with predicates, which are constituents of a language. Universals
and predicates do not map unto each other in any simple fashion.

There may be universals without predicates. We only learn of a
universal through its concrete instances. Universals that have not yet been
discovered may nevertheless exist but lack their corresponding predicates.
For example, the universal 'radium' existed before we discovered it and
before we coined the corresponding predicate 'radium'. Radium samples
made discovery of the universal 'radium' and the coining of the predicate
'radium' possible. There are predicates with no corresponding universals.
Negative predicates and disjunctive predicates, for example, do not have
corresponding universals.

So the question naturally arises: how can one tell when empirical
propositions are true, if their truth depends on the existence of these
objective truth-makers, and there is no guarantee that our language can pick

those out? Armstrong's answer, uniformly throughout all his work, seems to

47

be that whatever exists has causal efficacy.24 So, science is the appropriate
vehicle for separating the predicates that map unto universals from those that

do not. Our best science tells us what the fumiture of the universe is.

2.2.2 Laws as Relations in Branched Space-Time (McCall)

Storrs McCall, in AMcdelcfjheJlnixerse, defends the view that the
universe is best depicted as a forward branching tree in space-time. The
forward direction is the future. The branches are physically possible events
or physically possible ways the world could go under the conditions at the
junction point. The unused branches are pruned off as the present moves
past the junction point along one branch or another. The intuition behind this
view is that at any given time there are different ways the history of the
world might go, but the past is settled.

In McCall's view, laws are the constant concomitance of universal-
instantiated event types. Thus, his branching structure has nodes from which
branches arise. If every A-type branch has only B-type branches above it,
then it is a law that all A's are B's. If every A-branch has B branches and
non-B branches above it, then it will be a probabilistic law that all A's are
followed by B's with a probability equal to the ratio of B branches to total
branches above A.25 If that probability is rational, the branches are finite.
McCall claims that for real-valued probability, one has a decenary tree, with
each branch being subdivided into ten sub-branches, and each sub-branch

subdivided again into ten sub-branches, and so on, to give the equivalent of

 

2‘ Hence, mental states exist if they are causally connected with the environment and
appropriate behaviors. Beliefs are dispositions to generate behavior. Knowledge is belief
which is nomically connected with the state of affairs that makes it a true belief. Universals
cause particulars to be a certain way. Causal language appears throughout Armstrong's
works.

25 This, of course, assumes that there is a fixed ratio that holds for this type-identical event.

48

the decimal expansion of a real number. However, branches are neither
diagrams nor propositions, but are real 'ways' in which the world might have
gone. After the fact, however, there is only one way the world could have
gone and uninstantiated branches are lopped off. Nothing 'decides' which
way the world will go. The world is indeterministic, and each branch is equi-
probable. Models of the world, therefore, are isomorphic to some portion of
the world tree. Law-statements are made true by the world tree, and are
equivalent with any other law-statement that has the same models (in the
above sense), even if they differ in wording.

Laws are objective, he claims, and do not fall prey to the worrisome
subjectivity he finds in David Lewis, whose view of laws counts on personal
judgments of simplicity, strength, and the optimal balance to determine the
laws.26 Rather, laws are not statements at all, but structures that possibly
change through time. Hence, McCall is willing to accept the possibility that
laws evolve through time, and that we might have more laws now than
before, a view reminiscent of Pierce, as Joe Hanna has reminded me.

2.2.3 Laws as Relations in a Natural Kind Hierarchy (Harre)

Harre describes his own view as a variation on the view that laws are
relations between universals, but his view is an interesting variation that
makes some new points. In his book, Lamfflatme, Harre claims that the
following problems arise with respect to laws. Laws are general in form and
seem to say what will always and necessarily take place. So the first problem
is about the content of laws. What are laws about? Actual events and

processes? Dispositions? About relations between universals, or properties?

 

25 Lewis says, "A contingent generalization is a law of nature if and only if it appears as a
theorem (or axiom) in each of the true deductive systems that achieves the best combination
of simplicity and strength" in Lewis, Ccumerfacmels 73. This is a variation on a theory of
laws given by Ramsey.

49

Or are they redescriptions, in scientific language, of facts and true by
convention? Each of these, Harre believes, has an aspect of the truth, and
there are historical examples that fit each.

In his attempt to capture what is true in each of these views, he looks
at the problematic character of universality, necessity and confirmation for
laws. Universality (or generality, as he prefers) admits of three dimensions:
substance-generality, experimental-conditions generality and generality of
time and space. So, for an example of the first dimension, whereas Newton's
laws of motion apply to every physical object, Ohm's law applies only to
conductive metals and solutions. For an example of the second dimension,
Boyle's law relating the pressure and volume of a gas is pretty general,
whereas Reynold's law relating aerodynamic drag to velocity is restricted to
conditions of laminar flow at low velocities. For an example of the third
dimension, laws of mammalian behavior would be restricted to times and
places mammals exist.

Harre believes that one reason actualism, the view, associated with
Mach, that laws are summaries of experience, is that such a view is too
simple in the advice it gives us when our laws turn out to have
counterexamples: reject the law as false and replace it by one that is true. He
claims that scientists properly treat counterexamples differently depending
on differences in theoretical models explaining the regularity. If the law is
taxonomic, for example 'all mammals bear their young live,‘ we can treat the
characteristic of 'bearing live young' as either definitional or nondefinitional
of the species 'mammal'. The difference shows up in how we handle
counterexamples to a law. Sometimes we allow contrary evidence to refute
the law and sometimes we allow counterexample to revise the law. If the

characteristic is definitional, and we have one clear counterexample, then the

50

law is refuted. If the characteristic is nondefinitional, then counterexamples
lead us to restrict the range of validity for the law. Some characteristics
named in the law can change their status on contrary evidence. So, if we
thought bearing live young part of the definition of 'mammal' then we have a
problem when confronted by the duck-billed platypus, which, in other
respects is clearly mammalian. Such an instance refutes the law if the
characteristic is taken as definitional--or forces us to reject the duck-billed
platypus as a mammal. Or, we could, as in fact we did in this case, hedge our
original claim and demote the status of the characteristic 'bearing live young'
to a nondefinitional one. We would now say that most mammals bear their
young live, and that other characteristics are definitional of 'mammal.‘
Alternatively, we could start with a characteristic that was not definitional,
and handle counterexamples by restricting the scope until one has a
characteristic that becomes definitional of the class of objects within its
scope. So, we handle deviations from Newton's laws of motion by restricting
the scope of those laws to point masses, rigid bodies, harmonic oscillators
and conservative systems in general where there are no losses due to
friction. Paradoxically, restricting the range of valid conditions in one way,
leads to an expansion of the range in another. No longer restricted to planets
and projectiles, Newton's laws of motion now apply to all material
substances. One makes Newton's laws definitional of what it is to be a
material body. This comes close to what Harre believes Poincare was doing
in making laws true by convention.

Non-taxonomic laws, like functional ones, can turn out to be
disguised identities or not. So heat and the average kinetic energy of
molecules are not simply correlated, they are identical. Nontaxonomic

nonidentities would include the functional laws, like Boyle's law. Contrary

51

evidence here is handled by restricting the scope of the experimental
conditions under which the law holds true: not at very low temperature or
very high pressure (we should add, not at extremely high temperature
eitherl). Harre also considers the possibility that some of our laws may not
have held in the first thousandths of a second after the big bang. This
possibility, if true, would be a restriction of the scope along the third
dimension of generality.

Modality is another matter. Harre rightly points out that Hume
believed there were no natural necessities, that necessity was a
psychological matter describing our expectation that the future would
resemble the past. Since Harre wants to defend a notion of natural necessity,
distinct from logical necessity and empirical regularity, he wants an
ontology that grounds the 'must' in the claim that laws describe what must
happen. He believes this is revealed in the sort of explanations scientists
give for why the regularity is a law. Mechanisms, for nontaxonomic laws,
and stable natures, for taxonomic laws, are set forth in theories. Theories are
both models and the hierarchy of natural kinds in which the model is
embedded. The idea is that the proposed mechanism is suggested by the
network of natural kinds to which an object belongs. For example, atoms are
ascribed properties like momentum and charge, based on their participation
in the natural kind 'physical body' and the fact that other physical bodies
have these properties. The hierarchy determines which analogies are natural
and which are not; the analogies, in turn, determine which models are
plausible and which are not. Models, of course, are on our side of the
ontological divide, along with language, mathematics and formal systems.
But they all have their ground in what is in the world, i.e., objects with their

properties and relations, including causal relations in various mechanisms.

52

Belief that there is a mechanism at work in producing the regularity is how
we separate accidental regularities from laws, Harre says. This is natural
necessity.

Hierarchies of natural kinds, furthermore, solve the two great
paradoxes of induction: the grue paradox and the paradox of the ravens,
Harre claims. The first is solved because we do not ascribe properties until
we are persuaded we have ascribed stable natures to things. Harre contends
that for an emerald to be grue would require it to be green when we observe
it and blue when we do not, but that requirement has some odd sort
consequences: A grue emerald that is viewed at some times and not others
would be switching back and forth between green and blue, something we
do not believe happens or would only believe if we had some explanation for
the switching.27 One wants to ascribe stable natures to things which only
change properties when caused to do so. "Such a scheme [as the realist
seeks] is elaborated until as many as possible of the observed empirical
irregularities and discontinuities in some field of interest are accounted for
in terms of theoretically defined stable natures." (Harre 106). The paradox of
the ravens is handled similarly. Contraposition does not preserve lawhood,
because there are no natural kinds picked out by the complementary class.
The universality of laws is not strict universality covering everything in the
universe, because there are presuppositions restricting the universe of
discourse. The claim lacks truth value if the presupposition is false, e. g., if

there are no ravens.

 

27 Harte here gives a common reading of Goodman's new problem of induction, but one that
I believe is mistaken. Grue emeralds do not switch back and forth; they remain grue unless
something causes them to change. Expectations are set in an exactly similar fashion to the
way our color words set out expectations. The point is that both are confirmed to exactly the
same degree by past evidence, yet both set different expectations for the future. In this
regard, the problem of grue is exactly the same as the problem 'which regularities do we
project into the future as laws'.

53

2.3 An Analysis and a Criticism of the Core Realist Presuppositions
Although but three out of the many various realist positions have
been set forth here, the variety is sufficient to make one wonder. What
makes these views all realist views? Is there anything like a set of core
assumptions held by all or most realists on the topic of laws?
I think I can find three:
1. An cntclcgical claim: The world is independent of the mind.
2. A semantic claim: Language has meaning because terms pick out
objects in the world, and because sentences pick out states of affairs
in the world. Truth and reference are independent of verification or
indeed of verifiability. Some realists, but not all, would say the
nonlogical terms in the theories of mature science refer.
3. An epistemclcgical claim: we know what kinds of things there are,
and what their properties are by the normal practices of science.
Some realists, but not all, use the success of science in prediction and
technological application as an argument that science gets the
structure of the real world right, approximately right, or right in the
long run.
A plausible fourth assumption, shared by many realists, if not most, is a
historical claim: scientific progress increasingly approximates the truth. I do
not list it here because I do not think it is really independent of claim three.
Rather, it simply tells us that the isomorphism between language and reality

is achieved only in the long run.28 Are these three core assumptions held by

 

28 My view does not need such a notion, since the focus is on the coherence of experience
under a set of descriptions and manipulations. I will discuss my more pragmatist notion of
truth in chapter three.

54

these realists? Let us look at some of the claims they make in various points
of their works.
In W, Armstrong says the following:

First, I assume the truth of a Realistic account of laws of nature.
That is to say, I assume that they exist independently of the minds
which attempt to grasp them. (7)

Laws of nature must therefore be sharply distinguished from law-
statements . Law statements may be true or (much more likely)
false. If they are true, then what makes them true is a law. (8)

It is clearly possible that there should be a universe governed by
laws, ...but one in which there were no minds. (63)

Our problem is now before us. There are no secure paradigms of
laws of nature. Consider contemporary natural science. It is perfectly
possible, epistemically possible, that we do not know a single law of
nature...the scientific theories that we now work with are obviously a
reasonable approximation to at least some of the real laws of nature
even the rough-and ready generalizations of pre-scientific practical
wisdom represent a reasonable degree of approximation to genuine
laws. (6)

In volume 2 of Uniletsalsandjcientiflcfiealism, Armstrong says the

following:

Philosophers are familiar with the idea that science attempts to
discover the laws of nature Further, philosophers are familiar with
the idea that it is a weariness and a labor to establish in any degree
what these law-like patterns are. But philosophers have tended to
assume that there is no particular difficulty in identifying the
universals themselves....What has to be realized, instead, is that
determining what universals there are is as much a matter for
laborious enquiry as determining how universals are linked in laws.
(The two enterprises are, of course, bound up with each other). (8-9,
parenthetic remarks in the original)

What properties or relations there are in the world is to be decided by
total science, that is, the sum total of all enquiries into the nature of
things. (Philosophy is part of total science, but a mere part and not
the most important part). ( 8; parenthetic remarks in the original)

Storrs McCall, in AMcdechtheflniyeme, says the following:

55

The universe tree can be regarded as a huge cosmic entity, depending
neither for its existence nor for its nature upon being cognized by a
conscious intelligence. (7)

Given a language, an assignment function can be found which links
items in this language to items in the universe model. (41)

[T]he truth of a law depends on the way universals are instantiated
in the branched model. (73)

On the basis of the branched model, how could such a law come to
be known? [R]epeated experimentation can reveal the mode of
distribution of a universal or repeatable B on a fan of branches above
A, even though on every experimental occasion only one of the
branches is actually inspected. (64)

Rom Harre says the following in Lawmfll‘iatme:

Those who, like myself, wish to defend the idea that there is a
genuine natural necessity and that it is a feature of the physical
world, are obliged to find some natural property or process, over and
above observed regularities of adjacent event, but bound up with
them, which embodies that natural necessity. (82)

[S]cientific Realism. In the form in which I wish to defend this idea
and make use of it in this book, I shall express the relationship
between theories and reality in terms of the concept of verisimilitude.
The verisimilitude of a theory is the degree to which the model a
theory describes resembles the reality it represents. (93)

The question of what the concepts of verisimilitude of models, the
plausibility of theories and the truth of scientific laws mean must be
distinguished from the question of how we know that one theory is
nearer the truth than another, one model more verisimilitudinous than
another and so on. The former or semantic questions can be
answered independently of and prior to attempts to answer the latter
or epistemic questions. (94)

[T]he theoretical context of any research programme depends on
some scheme of interlocking type-hierarchies. Such a scheme is
elaborated until as many as possible of the observed empirical
irregularities and discontinuities in some field of interest are
accounted for in terms of theoretically defined stable natures. (106)

In the reconstructive process through which the method of
production of the phenomenon is depersonalized and simplified, the
- complexity of the relation between an item of knowledge and its

56

source is systematically concealed....Why is there this difference in
the presentation of a law [from the actual complexities of the
experiment]? It is the role of the guiding hand of the actual
experimenter which is written out of the successive accounts. The
effect of this is to display the phenomenon as a natural occurrence
existing independently of Faraday, Boyle, Snell, Mendel or anyone
else. (3 7)

In Realismfiescyed, R. Harre and co-authors claim:

We agree that many of the problems associated with fending off
these recent anti-realist attacks result from a failure to separate
metaphysical, semantical and epistemological issues, and that
attempts to refine and make more precise the true nature of the realist
doctrine are certainly justified. (2)

If the terms of a theory do not refer to mind-independent entities how
does the theory reach out to reality? For example, what parts or
aspects of the world is a particular theory about? (3)

It seems pretty clear that these realists do hold the three assumptions
I have ascribed to them. They want to hold that mind, language and the
world are independent of each other in metaphysics, but come together in
epistemology, especially through the work of science. The independence
assumptions come through in their claims of the separability of
metaphysical, semantic and epistemological questions, and in the claim that
some of those questions can be settled without settling others. The
connection assumption comes through in their appeal to science and
scientific method to sort out which items exist and which do not.

Needing to be made explicit are:

1. What notion of independence is at work in assumptions one and

two?

2. What notion of connection is at work in assumption three, i.e.,

how can science bring into isomorphic relation, ontologically

independent structures like mind, language and the external world?

57

Can science give us a picture of the world that is independent of the

processes which brought that picture about?

In clarifying these notions of independence and connection, I shall
go through the list of core realist assumptions twice. The first time, in 2.3.1,
I will examine the internal consistency of the assumptions. The second time,
in 2.3.2, I will examine the list for the intrinsic plausibility of the
assumptions, given our commitment to modern science and naturalism.
2.3.] Internal Critique of the Realist Assumptions

Since in general it seems that all scientific realists are working with
something like these assumptions, we will, for the sake of compactness,
focus on Armstrong's answer to these questions, and then discuss the views
of McCall and Harre where they differ from Armstrong's to see if their
views are free from the problems I found in his view. My criticism of these
assumptions will not turn on any differences that may exist among these
three, or, so far as I can tell, any of the other scientific realists who are .
realists on the topic of laws (some of whom will be addressed in footnotes).

Armstrong has identified his assumptions for the reader in several
works. These are:

l) Realism about Laws: Laws are parts of the world.

2) Realism about Universals: Universals are features of the world.

3) Actualism: There are no uninstantiated laws or universals.

4) Naturalism: The totality of entities is nothing more than a space-

time system

5) Physicalism: The only particulars that the space-time system

contains are physical entities governed by nothing more that the laws

of physics in a completed physics)

58

6) Factualism: The hypothesis that all that there is, is a world of

states of affairs, a so-called Tractatus world) (What 5, 7-9; World 6).

These assumptions are not all independent of each other. Number
three and six are related: universals just are properties of some particular in
some state of affairs. How, if at all, does Armstrong's list of these six
assumptions hook up with my earlier list of three core realist assumptions?
What, if any light, do they shed on realism in general and on realism about
laws in particular?

Let us compare the lists. We will ignore Armstrong's numbers one
and two for the moment: clearly if he holds to realism about universals and
realism about laws, then he is a realist! However, since one can be a realist
with respect to different classes of entities, for the sake of clarity it is good
that Armstrong tells us what kind of realist he is. Still, describing himself as
a realist leaves us none the wiser about any assumptions on which his
realism depends. Some of the other assumptions in this list are relevant only
to his epistemology and irrelevant to his metaphysical realism. Numbers
four and five are specialized versions of assumption number three on my list
because they specify physics as the science of choice for getting our
ontology right. Number six sheds some light on why he holds a
correspondence theory of truth, which forges the link between language and
the world lefi independent in my second assumption. Assumption number
three on Armstrong's list links up reality and experience, left independent in
my first assumption. His assumption number four, naturalism, which locates
everything that exists in space-time, probably is best seen as a rationale for
my assumption three, the belief that science produces ontology. Since states

of affairs just are the propositional content of perception, assumption six

59

adds to my assumption three that the truth is knowable in principle, even if
coming to know it takes a lot of hard, scientific work.

So, Armstrong distances himself from other ways of explaining the
independence of mind and world (e. g., any mind-body or mind-matter
dualism, as in Descartes), or the independence of language and reality (e.g.,
positing an independent realm of meanings, as in Plato or Meinong), or the
connection mind and language come to have with reality (e. g., rational
dialectic in Plato). And in light of the success of modern science in
explaining the world, Armstrong's assumptions may seem entirely
reasonable. The question I want to ask is whether they combine into a
coherent position. Then I would like to generalize the problems found in
Armstrong's view to show there are good reasons not to embrace the three
assumptions held in common by the realists. Finally, I would like to show
how these weaknesses in the realists' position wreak havoc on their account
of laws.

The first point is this. It is understandable why a dualist, who resists
reduction of mind to matter, might want to insist on the ontological
independence of mind and world, but why would a materialist want to do
so? In particular, why would one committed to the theory of evolution want
to make such distinctions central to an analysis of scientific knowledge? To
be sure, what the world is like is not something we automatically know, and
one needs some way to describe the difference between what we think is the
case and what is actually the case. So, some form of epistemic realism might
be given some plausibility. But even here, natural selection seems to place
some constraints on what may successfully be believed, and so, it seems,
that it is precisely the non-independence, and even homogeneity, of mind

and matter that must be given some credit for the success of humans in

60

knowing the world. At least, so some of Armstrong's fellow-naturalists have
argued.

The second point is this. It is understandable that someone who holds
assumption one would also hold assumption two. Language, after all, is a
historical structure, created by conventions and practices of people who may
have entertained incorrect views of the world, and hence, created predicates
that do not pick out genuine natural classes.29 But, if that is so, assumption
three becomes problematic. How can science, an institution that emerges
from the same human social milieu as language, succeed in describing the
world as it is where millennia of human attempts to do so have failed. If
human history were simply a tissue of errors, the prospects for science
would be dim. Armstrong, of course believes no such thing. He agrees that
even common sense gets things approximately right.30 He likes the picture
that W. Sellars gives of the relationship between the manifest image of the
world available in common sense, and the scientific image, which gradually
clarifies and replaces the manifest image--but is be entitled to that picture?
One needs some account of how we can get things right at all before we can-
grant him that we get things right either pre-scientifically or scientifically.
The independence notions in assumptions one and two force a divide
between the world and ourselves not easily bridged.

The third point is this. One needs more in an account of science than
simple faith in the experimental method to weed out falsehoods. It is

difficult to see what help Armstrong expects from science, since, for all the

 

29 However, it is certainly not a requirement that realists go this way. Socrates defended a
doctrine of natural names against the conventionalist views of the sophists in Camus.

30 Cf. Armstrong : "They are real features of things, real joints in reality, some of which are
grasped in approximate fashion by everyday perception and others of which are uncovered a
posteriori by deep scientific investigation" in Armstrong, Mild 30.

61

voluminousness of his philosophical publications, his references to scientific
practice are brief and underdeveloped. His strategy is to establish by
metaphysical arguments that laws exist, and then leave it up to science to
discover which laws there are, since which laws there are is an empirical
matter. But it is important to see first, that settling the question of which
laws there are can be an empirical matter only because Armstrong stipulates
in his assumptions that every law that exists must have instances. Second, it
is important to see that merely saying science is competent to the task is
hand-waving without some account of how science can pull this off. From
my reading of various realists, this seems to be a problem none of them have
even adequately formulated, much less solved.“ Popper could talk this way
because the items compared were the same: statements with empirical
content.32 But the realist is interested in comparing very different sorts of
things: predicates and universals. Simple appeal to the practice of science

won't do, because the practice of science, while public and 'out in the world'

 

3' Dretske, for example, tentatively suggested that observations entail general statements as
well as particular ones, and some realists, like F. Suppe, thus argue that science is not really
inductive! But Dretske qualified his claim severely to say that we have solved the induction
problem only for the very special case where l) we already have general beliefs among our
background beliefs that permit the inference or 2) we believe a conditional where we take
the antecedent for granted (Dretske, Seeing 220-234, 233). Dretske, in "Laws of Nature",
suggests that laws are confirmed faster than their corresponding universal generalization, but
does not say how this occurs. Abduction does not help, because either it implies some
special intuition or capacity for getting things right, in which case it is not obvious that we
have any such capacity, or it rides coat-tails on perceptual abilities we already have, in
which case the problem of independence looms large. As we will see shortly, even the
realist who has the most to say about scientific method, Harre, has little to say about this
problem beyond what a coherentist like myself would say, and that will not solve the
epistemological problem generated by the ontological independence assumptions either. The
problem is deep: the realists are trying to conjoin older metaphysical views with modern
scientific views, and the clash does not become apparent until one asks, as I am asking here,
about the ontological status realists would assign to parts of science.

32 Even Popper has problems, however, making a connection between this dialectical view
of the epistemology implicit in scientific method, on the one hand, and his realist
metaphysical program involving verisimilitude and his platonist three-world ontology, on
the other hand. His failure is instructive for those who would back realist metaphysics in
philosophy of science.

62

so to speak, is linguistically mediated and related to all sorts of theoretical
beliefs held by the scientists, whether viewed individually or communally.
So, the practice of science is irretrievably bound up with precisely the items
-- mind and language -- that assumptions one and two tell us are independent
of reality! How is it going to be able, therefore, to build a bridge between
them?

The fourth point is this. The bare existence of natural classes would
not help us in science if we did not know which classes were natural. We
could not use them as a guide in theory selection, for example. All we are
really in a position to do in science is to say that the classifications we are
using have held up reasonably well in experience. Such judgments are
retrospective, revisable, and never guarantee that the classes we converge
upon are natural in the sense the realist requires. It could not be otherwise
since we are never in a position to inspect the relationship between our
representation of reality and reality itself by virtue of the independence
claim in assumption one. The only way to overcome this would be to make
the classes cause our beliefs in them. Armstrong does speak this way in a
crucial discussion of the epistemology of general propositions.33

How might Armstrong answer the objections? While apparently
willing to concede an implicit skepticism in his view, he goes on to say it is
a mitigated skepticism like Hume's. When he goes on to criticize the skeptic,
he does so in terms that show he does not recognize himself in the position
he criticizes. He claims that the skeptic, if sensible, is selective, and if

consistent, holds a position so unattractive as to pose little serious threat.

 

33 Armstrong: "The nature of arsenic itself ensures that it is poisonous....1n practice, this will
mean that the belief [i.e., the belief that all arsenic is poisonous], to be knowledge, must
have been brought into existence by the action of individual samples of arsenic. "
(Armstrong, Belief 203; emphasis original).

63

Positively, he argues that belief entails knowledge. One cannot believe
something unless one also has the capacity to know since beliefs are
candidates for knowledge (Belief 190-192, 217-219). Since Armstrong
believes there are no unrealized possibilities in the world, the possibility of
knowledge implies the existence of some actual knowledge.

How good are these replies? They beg the question. Let us leave
aside the question of how he could know when he actually gets the laws
right: Armstrong will say that knowing when we are right would involve
knowledge of knowledge, and is not required for simple knowledge (we can
know without even believing that we know). Furthermore, he would go on,
second-order knowledge, i.e., knowledge of knowledge, raises no new
problem, since it has the same structure, as first order knowledge. That is,
the state of affairs which the second-order belief picks out causes the
second-order belief and is nomically connected with it. Lest we spend too
much time on Armstrong's replies, let us see that the skeptical problem has
been side-stepped, not solved. He believes one can have conclusive reasons
for beliefs that close any gaps into which the skeptic could stick wedges.
Armstrong does not take the skeptic seriously because he fails to see how
deeply mired in skepticism his own position is. One could not support the
claim that knowledge involves a relationship to a mind-independent world
unless one already had the independence assumption. And one should not
allow Armstrong to slip in the word 'reliable', as he does in explaining how
general belief can be transformed into general knowledge, without some
account of how we can reliably form beliefs about a mind-independent
world.

Now, Armstrong does talk at times as though the causal efficacy of

properties caused beliefs in us of various sorts--and that once we have these

64

beliefs, we are in a position to notice resemblance (or, as he prefers to call it,
partial identity) of instances which we collect together in natural classes.
This turns out to be crucial to his account of how we acquire the general
(universal) notion of mass, length and color.34 Lest the reader be taken in by
the nominalist sounding program of building up resemblance classes from
particular instances, one should keep in mind that for Armstrong,
resemblances are objective features of the world, there to be discovered.
They have nothing to do with language--or even minds!

The appeal to causation, if it worked, would buy us perceptual access
to reality, but would not explain why science is especially qualified to carve
nature at the joints. In fact, one needs to be very careful how one phrases
one's causal theories lest one acquire insuperable problems in explaining
how error is possible. And at any rate, Armstrong wants something broader
than a causal theory of knowledge because of well-known problems with the
simple versions, and this is why he talks about nomic connections and
reliability.

We will come back to this appeal to causation later, because it is
important to see what problems such an appeal raises if one accepts the
realist assumptions. I would like to point out that Armstrong also reverses
the direction: science is successful, and since his account gives the best
explanation of its success, it is likely to be true. This is the 'inference to the
best explanation' strategy that has been widely criticized by Cartwright and
van Fraassen (Cartwright 44-53; 87-89; van Fraassen, IheScientificlmege,
19-40). How precisely does Armstrong's ontology of states of affairs explain

science's success?

 

34 And also, the meter, the kilogram! See Armstrong, World 47.

65

The following story about how science works is consistent with
things Armstrong does say about science, but nowhere explicitly set forth by

Armstrong:

Science seeks to get the geography and history of the world right,
i.e., the kinds of things that exist and their properties. Science does
so by isolating bits of the world and studying them closely. By
manipulating some of the conditions, it establishes which states of
affairs are causally linked with which others, and thus establishes the
causal laws. By seeing which properties are invariably linked
together, it establishes the universal laws. By seeing which properties
are sometimes linked, it establishes the probabilistic laws. The best
explanation for how science is able to do this is to posit laws as
relations between universals, by virtue of which these regularities
hold up. Similarly, judgments of perceptual similarity require that
two items be the same in a certain respect, and that is all Armstrong
intends by a universal. Universals combine with particulars in states
of affairs, and states of affairs combine to give the world. Science
converges on the right set of distinctions over time by the
accumulation of careful observations and precise measurements that

allow one to make a distinction only when there is a difference.

That seems a plausible realist story about how science works, and
there is enough of recognizable scientific method here to make it attractive.
Indeed, it resembles the story I will tell about science in the next chapter--
with some crucial differences. What could be wrong with this as an account

of how we know universals through science?

‘66

To answer that question, we need to see what we have done in
constructing such a story. We have made an inferential leap over the
ontological divide by claiming that the universals are not empirical items but
are empirically accessible, as is the case with the relations between
universals. So the perceptual means of science are up to the task of
establishing the laws of nature. But what does this amount to but canceling
one stipulation by another? What form of independence is left after we have
granted that all universals and their relations are instantiated and
perceptible? What form of independence is left when I stipulate science is
successful now, and don't hold out for some predicates (I know not which)
that correspond exactly to reality (I know not what) in some future science?
Adding the word 'approximation' doesn't help; it simply reminds us that the
realist doesn't want the independence to get out of hand, but wants to keep
reality nearby to actual science.

Armstrong, in W, attempts to address the
charges of van Fraassen and D. Lewis, that the notion of necessitation
between universals is not coherent by describing a thought experiment that
he claims leads to his view. He starts out with perceptible regularities of
properties, goes on to perception of single causation, then continues by
abstraction to the universally quantified material conditional, and finishes
with the second order relation between universals as the best hypothesis for
why the regularity holds up. Look at how much this response assumes: 1)
that we perceive instances of universals, 2) that we perceive causes, and 3)
that we can perceive second-order relations between abstract entities. If he is
using this in the extensional or transparent sense of 'perceive' (Dretske's non-
epistemic seeing, where one can see an X without seeing that it is an X) then

the truth of these assumptions rests on the truth of his metaphysical theory,

67

and therefore does not support it, on pain of circularity. If he is using
'perceive' in a nontransparent or opaque context (i.e., intensionally), then, by
his own hypothesis that beliefs and reality are separate matters, his argument
does not go through. What we have for evidence is on the subject side of the
subject-object divide. What we have is the way things appear; what we need
is the way things really are. Only if 'perceive' is treated as equivalent to
'know,’ and given his account of non-inferential knowledge, that is probably
the way he intended it to be taken, does his argument work. However, then
the premises become dubious. Do we know there are universals, or causes,
or second-order relations between abstract entities?

Are matters in any better repair in Harre's view? Unlike Armstrong,
Harre has written books on scientific method. One might expect, therefore, a
better account of how realistically construed laws are captured best by the
methods of science, but such expectations are disappointed. In spite of the
much greater attention to how hypotheses are formed and tested, one still is
none the wiser on the confidence Harre has that there is a natural stopping
point for revision, that the world really does have joints at which it can and
should be properly cut.

Unlike Armstrong, Harre is not a logical atomist but a holist of sorts.
But unlike Quine, whose holism is one of beliefs (his web of belief notion),
Harre is a holist with regard to natural kinds in the world. Mind and
language turn out to be two subsystems of a larger natural hierarchical
system of kinds. This holism is supposed to explain the use of models in
science, which are drawn from natural analogies based on location in the
hierarchy of natural kinds. And Harre also seeks an isomorphism between
the representing system and the represented system that leads to a family of

theories in sequence of increasing verisimilitude until the last theory, which

68

is the truth. Where does independence come in here, in this version of
naturalism? Both thought and language are already embedded in this web of
natural kinds and in interaction with it. And exactly the same problem
occurs here in its construction of natural kind classes. How does Harre's
version of the induction to the best explanation show, in a way that does not
beg the question at issue, that stable consensus indicates verisimilitude, or
that powerful technologies indicate truth? The argument, borrowed from
Hacking, that manipulation is a good argument for the reality of the
manipulated cannot prove what they would like it to prove, that is, that
manipulation reveals a thing in itself, with non-relative properties.

As for McCall, his view is odd enough in several respects that it
requires separate comment. His 'reality', an evolving, dynamic space-time
tree, is not unchangeable like natural kinds or universals. Now, I want to be
careful here. Universals and natural kinds are created and destroyed for both
Armstrong and Harre, since evolution occurs. But while the natural kind or
universal exists (i.e., has instances or examples), it remains changelessly the
same. And the evolution of the tree is a random matter; which way the
'world decides to go' when faced with a fork in the tree is indeterminate. So
the tree does not strictly speaking cause our beliefs, and we are left with
experience to tell us how to reconstruct accurately the state of the tree at any
given moment in the history of the universe. McCall, more than Harre or
Armstrong, relies heavily on the inference to the best explanation. He claims
that if his space-time tree existed, it could solve many of our philosophical
puzzles, including questions of free will, laws, decision, counterfactuals,
quantum mechanics, identity, and essence. Whether scientists should be
impressed with this list is not discussed. The explanation, however, is

entirely retrospective. The indeterrninism of the process of evolution makes

69

prediction impossible, and he admits his view still has Hume's problem. So
McCall not only has van Fraassen's identification problem (as have his
realist colleagues) but also has the inference problem as well. McCall seems
to have the worst of both worlds.

The general problem seems to be that realists are committed to
the idea that the world is what it is, regardless of our limitations or modes of
access to it. We want to say that it is, without saying what it is, or how we
know that it is or what it is. So, an ontological gap must be kept between the
world and the knower. But then the realists want to go on to bridge that gap
and say what the world is like (i.e., describe it in language) and give force to
that language (by way of claiming it is true, or approximately true, and that
we know it, partially know it, justifiably believe it, etc.). To bridge the
ontological gap, realists claim that science is able to put us in touch with the
real world. But that claim poses a problem for them, because all the means at
science's disposal are precisely the ones on our side of the ontological
divide.

This is precisely a problem because realism of all forms, thanks to
Parmenides, has the legacy of a strong distinction between appearance and
reality that threatens realist epistemology with deep skeptical worries, e.g.,
Cartesian demons. Once one has this 'divide' between subject and object of
knowledge, the relation between them is problematic, and with that relation,
the possibility of knowledge. It is for this reason that realists like Armstrong
and Dretske, when they define knowledge, focus on 'conclusive reasons' that
eliminate the very possibility of error.

One further point deserves comment: the use of the notions 'cause'
and 'causation' in realist accounts. Philosophers who are realists about laws

tend to be realists about cause as well. Pargetter and Tooley, for example,

70

develop detailed accounts of causation, because they must challenge the
Humean account of cause that makes a regularity theory of laws attractive
(Tooley; Bigelow and Pargetter 263-294). To be sure, not all laws are causal,
and some realists entertain the possibility of singular causation (with or
without singular laws). For realists, causation bridges the ontological divide
between minds and the external world. In a passage that shows this
assumption at work, Pargetter addresses the problem of mathematical
knowledge thus:
Part of the complaint about mathematical objects lies in the
allegation that they are not observable. The complaint traces back to
the deeper allegation: that numbers lack causal powers. They do not
cause anything... Our strategy, however, is to deny that
mathematical entities are idle. In fact, we believe that some
mathematical entities are observable after all. But we do not need to

establish observability here...Of more central concern for us here is
the more central question of causation (380).

What lies behind this preoccupation with causation for the realist?
One motivation is certainly epistemic. As Armstrong points out in his sketch
of his ontology, WIS, while it is important that
properties not be identified with powers, still "we think of [properties] as
bestowing powers upon the particulars that have them. A property that
bestows no power will not be easy to detect!" (69) Causation bridges the
ontological divide between minds and the external world by guaranteeing
that whatever really exists will have observable instances. But does this view
work? Not if causation is considered part of the external world and separate
from minds and language. If causation is separate from mind and language,
then we have again the appearance/reality problem that bedevils realism.

The effect and the cause are viewed as standing on opposite sides of the

71

ontological divide, and it is unclear what one can infer from the effect about
the cause. Some realists, like Dretske, try to finesse this point by talk about

information flowing from the world to the knowing mind, but this is simply
an empty metaphor if one cannot explain how such a transfer between such

radically different items can take place.

The issue, again, is not whether realists make any sense when they
talk of one system causing another system to be in a particular state. Clearly
such talk makes good sense. The only question is whether, given the realist's
commitment to the independence assumptions, it is good sense to which the
realist is entitled. Causal talk makes sense if it links items of experience, but
it transcends the bounds of sense when it tries to relate experience to
something that transcends experience.

I can now briefly state why realism cannot solve the problem of
laws. Laws are part of the structure of the universe for realists, not items of
experience or belief, which are subjective matters. Regularities are on our
side of the ontological divide. Assigning laws a causal or explanatory role
for regularities leaves one none the wiser about how such divided things can
be related. Since this criticism depends on nothing more than the core
assumptions the realists share, it is a criticism internal to realism.

2. 3.2 External Critique of Realist Assumptions

I would like in this section to address the following questions: 1)
What, from my perspective, has gone wrong? 2) What should we make of
their three core assumptions in the light of modern science and naturalism?
3)What makes realism seem plausible, and can I account for the part of
realism that seems to make practical good sense?

On question number one, I think what the realists fail to realize is

that they are simply constructing a language and failing to take into account

72

the role they are both individually and collectively playing in making such a
construction. They are doing so with materials they have drawn from
ordinary experience, the language-games we play in practical life that have
well-recognized public criteria of application. As Kantians (including
phenomenologist Husserl) would put the matter, they have failed to
recognize the extent to which they have constituted the objects of knowledge
(through objectification), or how subjects and objects of knowledge are
abstracted from an original relation in consciousness (or, as I would prefer,
in social practice). What they are doing is translating statements from the
scientific idiom into their preferred realist language and mistakenly taking
the translation for an explanation. This point is not terribly new. Nietzsche
complained that Plato had merely doubled the world, and that philosophers
have been two worlders ever since. Duhem complained that appeal to
explanation is metaphysical, because it attempts to pierce through the veil of
appearances to underlying reality.

The realists have, I think, adopted a language that makes sense for
certain sorts of projects. It makes sense to insist on the transcendence of
experience by reality if one wants to rescue reality from change and
corruption as Plato did, or to rescue God from being confused with the
forces of nature or the designs of humanity, as Augustine and Aquinas did.
It makes sense for a dualist to insist on a divide between mind and body.
These distinctions allow one to deploy a particular system of values, engage
in some actions and refrain from others in ways some traditional societies
have found satisfying. But such distinctions come with a price tag. Once one
opens a gap, one has a problem how to relate the items again, as shown by
the ancient problem of knowledge of Being, the medieval discussions of the

incomprehensibility of God, and the modern discussions of the existence of

73

the external world and other minds. I can throw in one more related
problem, the problem of deriving an 'ought' from an 'is', sometimes called
'the problem of relating facts and norrns'.

To answer question number two, the three core realist assumptions
are unattractive and should be rejected because they do not comport well
with either modern science or naturalism.. Furthermore, this is for reasons
that the realists should themselves admit, since they endorse modern science
and naturalism.

Modern science views its objects as relative to a particular
framework, whether a system of measurement (as in quantum mechanics), or
an inertial coordinate system (as in relativity theory). Fundamental particles
are events that occur in accelerators or other apparatus. Genes are relative to
assays, and animal populations are relative to schemes of classification. So
identity is modulo processes of identification, and invariance is modulo
processes of variation. Observer and observed are related before we abstract
out what belongs to the objects. So, objectivity is modulo processes of
objectification. The relativity of objects to frameworks of thought and action
is even more important in the social sciences, where the observer and the
observed interact in ways far from trivial. The assumption that minds and
the world are independent, core assumption number one for the realists, is
not in keeping with the spirit of these shifts in scientific thinking.

Naturalism places the human observer in the context of natural
systems as a naturally selected organism, related to other naturally selected
organisms, and in constant interchange with its environment. This continuity
of the human and nonhuman is eXactly the point of the naturalistic stance.

The importance of this point will be explored in the next chapter, where I

74

develop the pragmatist view of laws. Here I only intend to explore its
implications for the realist assumptions.

Let us go through the list of core realist assumptions I put forth in
section 2.3, and subjected to an internal critique in section 2.3.1, so that we
may now subject it to an external critique.

Assumption number one is unattractive. Without minds, there could
be no choice of an appropriate frame of reference, and hence, no
measurement or detection. Indeed, without concepts there could be no
evidence, because relevance is established only relative to a set of criteria.
Without a world to provide landmarks, the mind could not find its way
about. Minds are fundamentally connected with the world from the outset,
making sense of it from various points of view, expressed in language.
Science simply continues these sense-making activities in a more precise
and self-conscious manner. Within naturalism, the mind, language and the
objects of the same just are parts of a complicated system of relations. A
world free of all connection with human minds would be a world with which
we could have nothing to do. As Kant put it, without concepts, the world in
itself is an X, unknown and unknowable.

Assumption number two is also unattractive. A language containing
true utterances we may never verify is a language we cannot use. It is a
language detached from the moorings of sensible employment and hence,
language without meaning. The realist position, since it is formulated in
language, simply does not make sense! We do not know where to attach the
predicate 'real' when reality is both mind-independent and language-
independent. (Our ordinary use of the term is neither mind-independent nor
language-independent, but part of an evidence-essential language-game

where we have successfully uncovered the bogus). This is not simply a point

75

about the single word, 'real.' It is science's objective to formulate claims that
have some experiential import. Quarks are not explanatorily on a par with
Aristotelian occult properties. Otherwise, there would be no point to
building expensive and politically controversial superconducting
supercolliders. Even the most theoretical of theoretical entities demand some
empirical tether. Otherwise, we are not playing the scientific language-game,
but doing something else. For naturalism, there is no reason to treat language
any differently than other naturally selected systems to which it is related.
Distinctions need to be made, but they do not amount to wholesale
ontological divides.

Assumptions one and two combine to underwrite a distinction
between nature and culture, a distinction that divides the sciences into the
natural sciences and the social sciences. Not only naturalists, but everyone
should have problems with this way of dividing the world and science. As a
division of scientific labor, this is unproblematic. One cannot do everything.
The problems arise, however, because once the division is made, there is a
tendency to forget that divisions of labor are matters of convenience. With
that come reductionist social theories that draw the causal arrows in only one
direction. A case in point is in biological deterrninist arguments about
intelligence, as discussed in Gould's MismeasurecflMnn. Being realist about
intelligence led researchers C. Murray and R.Hermstein in the B_ell_C_mye to
consider intelligence a thing and to look to biology and genetics for its
cause, and to insist that politics and a particular view of educational and
social institutions had nothing to do with the results. Left unexplored are
questions of how different ways of measuring intelligence may have led to
different kinds of correlations than the ones Murray and Hemnstein

discussed, and how value judgments may have entered into choosing certain

76

measures as the most important measure. Similar points can be raised with
discussions of the 'real rate of inflation', and claims that public transfers
indexed to the rate of inflation have been based on too high estimates. If
policy formation (soft science) is going to defer to biology and genetics
(hard science), the public is entitled to know how soft hard science is and
how hard soft science is. A naturalist perspective will put into question the
evaluative overlay which the distinction between hard and soft science
inherits from an earlier dualism between natural and moral science.

Finally, assumption number three is unattractive as well. Not only
are the realist accounts of how science supports realism underdescribed,
those accounts fail to address the issue of motivation. If the objects of
science are as they claim, mind-independent and language-independent,
what could have motivated naturally selected organisms like ourselves to
engage in their pursuit? To say, as Aristotle does that "All men by nature
desire to know," will not do, because it leaves untouched the problem of
why anyone should have thought the procedures of science yield knowledge.
On the other hand, if inquiry starts with problems an organism has in
satisfying desires, or learning what to expect from its environment, in order
to better defend itself, for example, then one has an account of the organism
in relation to a given environment that has some chance of showing how
inquiry evolved into experimental science. To start with the concrete is not
to bar any claim to the abstract. However, to start with the abstract, and,
what is worse, to reify it, may block the road home to the concrete. Such a
view loses sight of our stake in the sciences, and the technical exploitability
of its results.

For all of these reasons, the assumptions of realism and the realist

view of laws should be rejected. To paraphrase Gorgias, the realist position

77

does not make sense, and even if it did, we could never know it, and even if
we could know it, we couldn't communicate it in a language-dependent
social institution like science. I have not proved that realism is false. If
realism is nonsense neither truth nor falsehood appropriately apply to it. But
I do claim that the position is so unattractive that it is not worth anyone's
time to try to articulate a sensible version of it.

Let us now answer the third question with which I began this section.
What part of the realist project can be salvaged from the debris? Is it all
loss? I think not. I intend to show in chapter three that we can make all the
valuable distinctions a realist would want, without making the realist
assumptions. Here, I would like to highlight a couple. One can salvage the
worry about the fallibility of belief within a non-realist naturalist framework.
Some of our attempts to relate to the world fail to succeed. Some languages
may be more useful than others. Next, objectivity may be salvaged as an
attempt to construct a world picture that is free from particular biases and
idiosyncrasies andthat has intersubjective validity. The connections to
interests and needs may attenuate, and the results may transcend and modify
the culture from which it arose.

Next, a sort of realism, relative to a language, seems defensible. We
can, after all, within experience, make distinctions between objects that we
treat separately. We can isolate systems of interest, at least partially, from
other systems in which they are embedded or to which they are related. And
invariance under sets of manipulations may qualify something as real
relative to the language-game in which such manipulations have point. So,
some form of relative or qualified independence of subjects and objects is
defensible. We are not, as anti-realists, in the night in which all cows are

black (to quote Hegel's blast against Schelling). In particular, I will show in

78

chapter three that one can be a holist and still have a place for a distinction

between laws and accidental correlations.

2.4 Conclusion

Realism about laws came as an answer to defects seen in the
regularity analysis offered by Hempel and Oppenheim. But realism seems to
be saddled with
insuperable problems of its own that make the position unattractive for those
holding a naturalistic worldview. In particular, the claims of ontological
distance seem at war with claims of epistemic closeness. Such claims open a
gap between the subject and object of knowledge that becomes problematic
to epistemic bridge-makers. Appeal to science to bridge the gap seems a
prayer of faith given that all the resources available to science lie on one side
of the divide and the goals of science on the other side of the divide. A view
that respects the continuity of the natural systems we are with the natural
systems we study looks more promising, and will be explored in the next

chapter.

79

Chapter Three: A Pragmatist Position

In the light of the discussion in chapters one and two, it seems that
what is required for a solution to the problem of laws is a way of drawing
the distinction between laws and accidental regularities or correlations that
does not raise insuperable epistemic problems. What I intend to do in this
chapter is what both the realists and van Fraassen claim cannot be done:
show how laws can be empirically accessible and still license inference to
future experience. In other words, I will solve simultaneously what van
Fraassen calls the identification problem and the inference problem.

The view to be defended here has been worked out within the
framework provided by American pragmatism.35 The central claim is that
laws are constructions that are part of our construction of a comprehensive
world-picture. While these include qualitative laws and taxonomic laws,
which in their own way are useful enough, I shall concern myself in this
thesis with an important subset of scientific laws, the quantitative,
experimental laws. Such laws are repeatable relationships between measured
quantities that are determined by our manipulations of the conditions under
which they are observed to hold. Why pay especial attention to these? I
shall pay especial attention to these because science tries to replace the
qualitative and the classificatory laws by quantitative ones when it can. I
shall later argue that the reason for this is that the drive for more informative
claims leads to the demand for greater precision in formulation, and
quantification allows that. We seek to measure, when we can, and to control

the processes of interest when we can, in order to develop more testable

 

35 The works relevant to my work here are: Dewey; Nagel, W; Goodman; Quine,
Bland; mm; SkyrmS-

80

claims. On this point, the pragmatist is in agreement with Popper. So, while
not dismissing the importance to science of other types of law, from this
point on I shall talk only about the quantitative ones, especially the ones that
are subject to experimental control.

Laws arise within the context of a set of scientific practices which
clarify and replace vague, imprecise articulations of regularities found in
common sense. We construct laws in ways that make the replacing
regularity more reliable as a basis of practice than the replaced regularity.
Indeed the central feature that science adds is the practical understanding of
the conditions under which regularities hold up or break down--precisely
what is needed to provide a defensible distinction between laws and
accidental correlations.

To be in a position to say what laws are, I will need to make some
general claims about language, minds and the world that will set my position
off from realist claims about the same. The structure of each subsection in
this chapter will be a movement from the general to the specific. I will first
develop general claims about language, practice and success, then specific
claims about them.

In this chapter I defend the view:

1. That laws are constructions embodied in some language,
especially mathematical language (section 3.1).

2. That the language is applied within a set of practices that measure
and monitor the relations between variables of interest (section 3.2).
3. That the constructs are placed in an environment in which they
may fail (section 3.3).

4. That the test scientific constructs need to pass is that of reliable
prediction (section 3.4).

81

5. That this view solves the problem of laws as we have developed it
in chapters one and two (section 3.5).

3.] Laws as Constructions
3.1.1. Constructions Relative to Language36

In this section, I argue that scientific laws are designed for a
particular kind of use, and as such need to be explicitly embodied in some
representation that we can grasp and use. This need not be language as we
usually think of it, but language-like objects (e. g., charts and diagrams) may
be used as well. The language-relativity of laws is significant: some laws
may even lack a translated version in other languages (e.g., 'grue' and laws
stating regularities of emeralds), because there is no change that requires a
law for its description.

In calling laws constructs, I seem to be making an elementary
blunder: there is language and then there is W, that to
which it refers or which it describes. But this is not a blunder, because one
cannot talk ahcut anything except within some set of distinctions that
describe the properties of that something and the contrasting properties
lacked by that same something. Meaning is not a function of isolated bits of
language uttered in a vacuum, but is rather within the fiamework provided
by a meaning space mapped out by a language. One cannot, for example, tell
what one means by 'cat' without some space of meaning that includes
'physical object,‘ and 'living being'. The language must allow concrete
particulars of the right sort. In addition, one needs further words to mark out
distinctive properties, so shape-words, color-words, and texture-words are
required to even introduce a word like 'cat' into a language that lacks such a

word. The semantic space mapped out by a language has to be the right

 

36 see Dewey chapter 3:"The Existential Matrix of Inquiry: Cultural" and Quine, Word.

82

shape, not only to allow definitions of undefined terms, but to be able to
characterize the class of objects at all. To be sure, some crucial distinctions
may be lacking, and one may be able to coin new words or expand the
meaning of some others by analogy and metaphor to talk about unexpected
events. Still, under such circumstances, one is deploying or re-deploying
linguistic resources one already has. Science never starts from a blank slate,
but within the range of a meaning-space constructed within ordinary
experience. The language as a whole is at work in the semantic reach of its
parts. So, linguistic atomism is untenable. One must work with some version
or other of holism.

Still, while one cannot talk about things that cannot be talked about,
may they nevertheless exist? Cannot one make a meaningful and serviceable
distinction between the meaning and the truth of claims? And if one can,
might laws hold 'behind our backs' so to speak, where we neither see nor
think about them? Might it not be of enormous practical importance that
there be laws, even if no one is prepared to notice them or explicitly express
them in language?

All of these questions have the same point: to drive a wedge between
verification conditions and meaning conditions. This is, as I said in chapter
two, part of the realist strategy: to make the truth of a claim independent of
ways of finding out whether the claim is true. A claim can be true even
though no one knows it to be true, and this applies to all claims whatsoever.
The point I would like to raise in reply is that the appearance of making
sense notwithstanding, the realist strategy produces nonsense. The joke
about the earnest German biologist and the flea comes to mind. Noticing that
the flea jumped for shorter and shorter distances as it progressively lost one

limb after another, the biologist was dumbfounded by the fact that the flea

83

suddenly refused to jump at all upon the removal of its final appendage.
Since the flea had been compliant up to this point, and zero was not on the
line formed by the other data points, the scientist concluded that the flea had
inexplicably gone deaf! The point of the joke is the same point being made
here: a question that makes sense in one context may utterly lack it in other
contexts. To ask what an utterance means apart from any individual's
experience may make sense. To ask what someone might mean by the
utterance apart from any possible experience, does not. One has deprived
such an utterance of any legs with which to jump.

Language marks out distinctions within experience, and makes
sensible applications within experience. To raise a general question of
whether all our standards of correct application might not after all be in error
is to use language in a way that denies the conditions of sensible utterance.
This is a version of the claim Kant made about the fundamental categories of
our thought: they make experience of objects possible; they cannot be used
to extend beyond experience to objects that transcend experience.
Wittgenstein's point that language requires publicly assessable criteria for
correct application holds for scientific utterance as well. One simply cannot
make sense of theoretical discourse apart from numerous connections with
talk about phenomena. The phenomena must be appropriately labeled,
described, measured and investigated. The language weaves experience into
a structure. This point does not turn on our ability to reduce the meaning of
individual terms or sentences to some so-called pure or neutral observation
language. The point is rather that if the language, be it theoretical,
observational or mixed, does not relate bits of experience with each other in

some coherent way, then we are certainly not dealing with the language of

84

science, and may not be dealing with a language at all. Our view of language
and linguistic meaning must be broadly verificationist.

Language helps us come to grips with the world cognitively. We
construct it to direct our noticings in different ways: perceptual contrasts,
monitoring the effects of our interventions and actions in the world, placing
an evaluative overlay on objects from the environment labeled fiiend or foe,
and so forth. We simply are not in a position to act intelligently in our
environment until we have constructed a space of meanings and values that
map out relations of relevance, appropriateness and importance.

In science in particular, we have additional reasons for demanding
explicitness and precision of statement. Laws in science are cognitive
devices constructed to enable us to predict the behavior of interesting
systems in the world of experience. To do so, we need to isolate the system
of interest. To do that we need to be clear about what belongs to the system
and what does not, and about which conditions are relevant to the behavior
and which are not. Do we know these things a priori? No, we do not. But
generalizations from ordinary experience form reasonable starting points for
science, and science goes on to make these generalizations more precise
(more on that later).

It is important that science render explicit normally tacit habits of
usage in order to make claims and terms precise. Attention to a habit
already in place puts one in a position to notice that the habit is not working
very well, and to construct a new habit that better serves in navigating the
world of experience. The metaphor of a 'map by which we steer' used by
Armstrong, originating in Ramsey, is an appropriate one for belief, but I
think Armstrong missed the pragmatic point: we are in the business of

constructing maps or representations of the world because we have practical

85

goals to meet. For example, the qualitative regularity that air gets 'thinner'
and air pressure goes down with increasing altitude may be serviceable
enough for mountain climbers wondering why they tire more easily at higher
altitudes, but it is not as useful as Boyle's law, which, with the addition of
some laws of respiratory physiology will tell airplane designers precisely
what altitude the pressurized cabin needs supplemental oxygen in order to
prevent passengers from passing out.

The position I am defending here may buy us a realism of a sort, but
it is an internal realism, to borrow H. Putnam's phrase. For the internal
realist, we can only sensibly make existence claims within the semantic
space mapped out by our language or some reasonable extension of our
language. Any plausible existence claim must cohere with the framework of
sense that we construct by the collection of our accepted theoretical and
factual beliefs. So construction of laws within a language involves
recognizing that our language is our world -- all the hard won beliefs we
have about the way the world is are formulated within our language and
modify our language. This is to deny, of course, a central tenet of realism: .
that minds (and their contents), language (and sentences formed in that
language), and the (external) world are separate metaphysically but
connected semantically. Mind and language cannot be 'about' the world
unless they are intimately involved with each other from the outset. But such
intimacy belies the claim that there is some large metaphysical distinction
between the three to defend.

Laws in particular connect up bits of experience by explicitly stating
the general relation between specific observable quantities in mathematical
form. Even qualitative laws need to specify the relation in question to be

taken seriously. The reason is at least partially to determine the area in

86

which confirmation or disconfirmation is to be sought. This explicitness
excludes many ways the world could be. It sets both a higher standard for
success and a lower standard for failure for the claimed generalizations.
Why this would be desirable needs explanation, but here we simply observe

that this is the effect of greater explicitness in formulation.

3.1.2 The Role of Mathematics

No constructivist account of laws can succeed without some account
of the role mathematics plays in the articulation of laws. After all, if laws
require mathematics, and if mathematics requires realism, then, as Pargetter
argued, one needs to explain why laws should not also require realist
analysis if they do not. The answer is brief: even if numbers were real and
their relations were real, it would not follow that the empirical relations
modeled by the numerical relations are also real. The empirical relations
may be constructs even if the numerical relations are not.

I think we need not even grant Pargetter that much. There is no
compelling reason to treat mathematics as other than a language with its own
set of rules for correct manipulation. There is a great deal of historical
evidence that mathematics incorporates, in abstract form, relations from
extramathematical systems: geometry from the measurement of physical
spatial relations, groups from transformations of various sorts, arithmetic
from practices of counting, etc. It doesn't matter that there may be
mathematical objects for which there are no experiential or physical
correlates, or that the natural model may not even be correct. For example,
physical geometry may not be Euclidean, even though the latter was
designed with the former as the intended model: once one interprets the

formal term 'line' as the 'least-time curve light traces out in a vacuum', there

87

is a fact of the matter about whether such lines behave like Euclidean lines
or like Riemannian lines. The connections with experience are numerous
enough to make Platonism an unattractive account of mathematical truth. To
be sure, mathematical abstractions do take on a life of their own with
unexpected consequences. But the point here is that realism is no more
required in an account of mathematical relations than it is for the empirical

laws they model.

3.2 Practical Conditions of Constructions

Wittgenstein described language-games as the interweaving of
language and action, and however apt a description that may be of language
and language subsystems in general, it certainly applies to the language of
science. One does not simply have distinctions or terms, but a range of
activities that involve determining the correct applications for specific cases,
with the values to be assigned. To anticipate points to be raised in the next
section, one wants to have not only the words, but the criteria for
successfully applying those words to experience. Later, I will give a theory
of measurement that accounts for the mathematical form and the

experimental verification of laws.

3.2.1 The Connection between Thought and Action

A point made by all pragmatists especially important here is that all
thought arises in a practical context in connection with problems a thinking
organism has in achieving particular ends. The problem may be more or less
organic in character, where for example the organism is hungry, in pain, or
up against a physical barrier of some sort that needs to be surmounted, and

there may be no direct way to satisfy the needs the organism has. Thinking

88

is an indirect way of satisfying the needs or solving the problems the
organism has, by looking for items in experience which have been reliably
associated with satisfactions in the past, e. g., a door or a ladder to surmount
the physical barrier of a wall. The problem may be of a distinctly social sort,
for example, looking for a coordination of activities of a group when its
members are in disagreement over how to proceed. The problem may be an
intellectual one, such as puzzling out the application of mathematical
resources worked out in different divisions of number theory to come up
with a solution to Fermat's last theorem. The point here is that so long as the
problem is a familiar one, for which we have the resources, habits already
acquired are sufficient, and no thinking occurs. Insofar as the problem has
novel aspects, however, thinking may be required for successful solution.
Explicit attention, as I said in the previous section, becomes required when
old habitual ways of proceeding do not work. In science, one goes out of
one's way to test habitual ways of proceeding to see where they break down,
in order to replace those with habits that are more successful in the
problematic situation. The indirect route of thought may become very
indirect indeed, as we will see when we come to the pragmatist account of
mathematics and logic, but the point remains that thought is rooted in the
practical and naturally finds its way back to the practical in the long run. The
applicability of scientific knowledge is what one would expect, on a
pragmatic reading of what science is about.

A number of points need to be made here. First, a thinker is an
embodied organism that constantly interacts with its environment. Action is
the part of the interaction for which the organism is the initiator. Second, the
interaction is for the sake of maintaining the organism in that environment:

survival and flourishing are the point of the actions. Third, any organism

89

that has survived has a repertoire of actions that seem to work reasonably
well in the environments it finds itself in. Insofar as they are performed
roughly the same way in roughly the same contexts, they are deemed
habitual. Fourth, any habit may fail to achieve its ends in a context that
superficially resembles the ones in which it has succeeded in the past. F ifih,
rational organisms, through the medium of representation, can turn the habit
itself into an object for scrutiny, in order to differentiate the contexts and
manner of its observance to improve its rate of success. Thought is precisely
the process by which the deviation between ends and results is observed and
corrected. In organisms without brains, similar results are achieved, if at all,
by random modification of the behavior until something works. With
organisms with brains, the trials can be internalized, so to speak, in

inferential chains tracing out the conditions and consequences of actions.

3. 2.2 The Connection between Words and Action

The next point is that this thinking is conducted in the medium of
language for language using organisms like ourselves. This point goes
beyond the point raised earlier, that language is for the sake of experience
and requires empirical applicability. Here the point is that language is part of
a way of being in the world, part of a way of life. The set of distinctions that
exists in our language is there for the actions it enables and the set of social
practices it sustains. The language needs to be at least rich enough in
distinctions to enable us to notice when the conditions are right to engage in
a particular practice. Those distinctions need to be available to speakers of
the language so that they can police its rules. The language need not be rich
enough to formulate those rules explicitly -- although natural languages like

English are -- but they must be rich enough to enable speakers to distinguish

90

superficially similar perceptual occasions where the distinction marks the
difference between success and failure (whether complete or partial) of the
enterprise.

The last point is that actions create the perceptual events that lead to
linguistic utterance as well as the other way around: action elicits utterance

and utterance elicits further action.

3.2.3 The Social Character of Practice

The next point to be made about practice is that practice is engaged
in by the group. Since language is a part of a practice, this implies that
language too is essentially social in character. Both points are important.

Practices are performed by groups, who train others to do the same
things in the same circumstances. As Wittgenstein has argued, the rules of a
way of life are caught as much as taught. Imitation is more effective, and
more frequent than explicit drill in formulated rules. Nevertheless, the
individual performance is positively or negatively reinforced depending on
whether it conforms or deviates from the group standards. Language, as a
specific subset of practices, accompanies other practices and is similarly
positively or negatively reinforced by the group. The group tries to sustain
itself in a given environment, by means of group habits of action and speech.

While individuals interact with environments to sustain themselves,
part of the environment is the practice of the social group in which it finds
itself. For this reason, survival tends to be at the level of the group.
Individuals tend to survive or succeed if their group does. One can put this
point in terms of institutions and speak in functionalist terms: institutions
that meet needs in the society in which they are created tend to survive. The

culture of the institution maintains itself. But its survival depends on what

91

other institutions and non-institutions are doing: if the work is redundant, it
may not be able to sustain itself. Furthermore, if the institution is in a
society where some fundamental needs are not being met, the institution
may die because the society does. It is hard to find uncontroversial examples
of each, but here are some suggestions: the electoral college may be an
example of an institution that dies for the first reason; the monarchy, an
example of the second; and the hunter-gatherer ways of the native American,
an example of the third. In each case, significant changes in the political and
social environment -- the grth of an educated electorate; the growth of
strong parliaments and a wealthy middle class; the invasion of hunting
grounds by numerous well armed and determined European settlers -- led to
the demise of institutions that succeeded well enough in a former
environment. Science, as an institution, faces pressure from similar shifts in
political, social and economic environments, and maintains itself because
the social practices of the group are able to 'deliver the goods' of reliable

predictions and control.

3. 2.4 The Practical Character of Science

The next point is to see that science is itself a set of practices that
tries to maintain itself in a particular environment. Everything will hinge on
how we characterize that environment, so much more needs to be said.
Nevertheless, for one committed to naturalism, the point should be clear:
science, as a sublanguage, is part of a natural system embedded in a set of
other natural systems, like minds and societies and ecosystems, and like
them, it tries to maintain itself in that environment. Distinguishing my view
from that of other constructivists who might be inclined to say the same

thing with very different import will need to await first a discussion of the

92

practices of science (3.2.5) and then some attention to the conditions of
success for constructive activity (3.3).

Here however, I simply want to make some general observations. In
its internal relations, the science community polices the rules of its practices.
It sets standards for education and credentialing, for publication and awards.
It sets the benchmark for quality of research and the uniformity of
nomenclature, all to ensure that its members do basically the same things the
same way. In its external relations with the rest of culture and society,
whether in seeking outward funding, or in doing public relations work with
the government or the community at large, the scientific community seeks to
ensure the continuity of its work and its institutions. It succeeds in both
because there is general agreement on the value of its products, a subject to
which we will return in section 3.3.

3.2.5 The Continuum of the Practices of Science37

When I claim that science is a language-using set of practices created
by embodied cognizers in interaction with the world, I am making some
general claims about the representations that occur in science: none of them

are autonomous, but all must sustain themselves in an environment created

 

37 The following list of scientific activities is my own, but it is by no means idiosyncratic.
To compare, Lenzen in "Procedures of Empirical Science" gives the following list: 1)
Observation (under which he subsumes: a. perception, b. counting c. measurement and d.
instrumental detection) 2. Systematization which includes classification, correlation,
successive approximation, successive definition, and statistical analysis Lenzen,
"Procedures" in Neurath, et. al. 1: 279-339; Ian Hacking, more recently and more
provocatively, has suggested that the activities in the laboratory simply put into play and try
to maintain a stable configuration of the following heterogeneous elements: 1. ideas,
including questions, background knowledge, systematic theory, topical hypotheses,
modeling of the apparatus; 2. things, including targets, sources of modification, detectors,
tools; and 3. marks and the manipulations of marks, including data, data reduction, data
analysis, interpretation. Hacking, "Self-Vindication" in Pickering (ed.), Science 29-65. I
think Gooding's criticism of both is appropriate: the agency of the experimenter is not given
its due in either view. The former focuses on logical relations between representations; the
latter, on equilibrium of elements; neither focuses on the activity and the learning associated
with active manipulation of the world.

93

by this constant interchange between cognizer and cognized. To make the
point explicit and raise its plausibility, I will now look at a range of activities
that occur within the sciences to show how they involve the manipulation of
the environment by the organism, and the modification of the organism to fit
the environment.

On each activity, I wish to show that a) any cognitive yield from this
activity comes only as a result of some interaction between the cognitive
agent and the system of interest (i.e., there is no such thing as a cognitive
free lunch) and b) that the product of such activity takes on a life of its own
and can only retain its worth if it can maintain itself in the various
environments into which it is placed within the context of scientific
practices.

(i) Observation

It is practically a banal truism that science involves the observation
of natural systems of various sorts. The point ceases to be banal, and
becomes controversial when the point is made that there is no observation
without change, often change induced by the observer. Least controversial is
that the change involves a change of the mental state of the observer. But
observation involves physical changes as well. The dye in the rods and cones
must undergo change. The lens must focus, and the head must rotate, to
bring things into view. Some such physical changes must take place in us,
the subjects of knowledge if we are to see anything at all. But in addition,
some physical changes in the object may be necessary. It may, for example,
need to be dyed, chipped, put into a culture, on a slide, under a lens, or into a
bubble chamber to be 'observed' at all. Even if the object is already visible as

is, it must, at least, change state to scatter photons.

94

All observation involves selection and distinction. We need to
distinguish object from background, and distinguish changes relevant to the
behavior of the object from those irrelevant. Observation is value-laden in
that there are preferred places and evaluative judgments involved in what we
choose to observe and what we choose to ignore. Observation is relative to
us. We constitute the physical frame of reference. Our sensory modalities
select what is perceptible and our way of life places an evaluative overlay
over all that we perceive. To put that out of play, in order to make
disinterested observation, is an achievement, one that requires effort and
attention.

Perception involves (at least rudimentary) conception. We have
already begun to think about perceptual objects when we have marked them
off from the surrounding phenomenal field--even before we have described
or classified it.

Observation involves action. Observation is far from a passive affair.
So observations cannot be made unless we produce the phenomena in a
suitable device, for example, a particle accelerator or an electron
microscope. Similarly we must prepare the specimen in suitable ways, for
example, bombard it with a particle beam, dye it with Gram stain. Even
naked eye observation is no straightforward matter, but requires training. It
requires skill to do well, a skill that scientific
community inculcates in new members. One must distinguish actual
observations from instrumental artifacts or optical illusions.

Observations occupy a semantic niche in which they must maintain
themselves against competitors. Plato claims to have observations showing
heavenly bodies were perfect spheres, and most ancient Greek astronomers

agreed in those observations, but those observations had competition.

95

Observations that the heavenly bodies has features inconsistent with smooth
spheres--mountains on the Moon and spots on the Sun-- observations
recorded by Galileo in Sidentanncins competed for the same semantic
niche and dislodged the observations of the ancients, and with those
observations, the ancient theory was dislodged as well. In fact, where the
ancient theory was absent, as in China the observations of sunspots were
made at a much earlier date!.

Observations are placed in a context where replicability is important.
The ideal would be to have a machine do it. And many chemical and
physical experiments do have electronic devices as detectors, which feed the
information directly to computers which record and process the data. Short
of that ideal, however, one wants to have the procedures replicable by any
suitably trained person. Ultimately, one needs to describe the conditions of
observing phenomena in such a way that any other researcher could make
the same observations. Once one can do so, they become detached from the
individual who first made them and take on a life of their own: they now
have to sustain relations of compatibility or incompatibility with other
observations and the observational consequences of other beliefs. Whether
they will succeed in so maintaining themselves is up for grabs. As the
Galileo example above shows, rival observations may be made in the field
that show the researcher failed to see what he or she claimed to see. It may
be that one notices differences that are irrelevant to the behavior of the
system, or fails to notice differences observed by others that are relevant to
the behavior of systems.
(ii) Description

Skilled observation already involves language, and the two

reciprocally modify each other. However, description goes further in

96

systematically characterizing some object or system, at least at the level of
phenomena. Description involves construction at several levels.

Like observation, description involves interaction with objects. For
example, one may need to turn a rock over, push it, crush it, or dissolve it to
generate a list of characteristics sufficient to identify some stone as being a
particular mineral. Even in field studies, description requires a special sort of
attention, a looking disciplined by the phenomenal features of the object.
Descriptions place objects in relevance classes of interest, which suggest
interesting properties the object may also have.

Description also involves interaction within language. As Dewey
usefully points out, once one has a linguistic designation for something, one
can trace out not only relations between events, but also inferential relations
between terms. This may mean embedding the term in a new environment.
To use Dewey's example, once we have the word for 'cloud' we not only
have the customary connection with rain, but also the possibility of
association with terms like 'temperature', 'pressure' and 'volume'. (Dewey 58-
9). The possibility of abstracting an item from one context and placing it in
another is a form of manipulation built on an analogy with transplanting
plants. The inferential component lies in the fact that as represented the item
can now be thought and its properties and behavior simulated in thought.
The notion of simulation has turned out to be very powerful indeed. The
Einstein-Podolsky-Rosen thought experiment and Monte Carlo simulations
of particle detector behavior have proved very important to modern

theoretical and experimental physics respectively. 38

 

38 For the former, see Kuhn, "A Function for Thought Experiments" in Kuhn, Essential 240-
265; for the latter, see Knorr Cetina, Enistemic 46-78.

97

 

If a term proves useful, it may enable scientists to forge connections
within their theoretical web of belief to make the web a simpler, more
coherent whole. It may enable us to explain more, to understand more. But if
so, it will do so by putting us in a position to predict successfully how a
system will behave. That leads to the next point.

Descriptions also involve predictions. When one describes something
as inanimate, a set of expectations is in place that prepare one to be surprised
if they are not met. Rival descriptions of systems of interest set expectations
different ways, and different behavior may lead one set of descriptions to
displace another. What one is prepared to notice is a function of
discriminations available in one's language. Those discriminations exist
alongside of practices that make use of them. So, the distinction between
'gram-negative' and 'gram-positive' enables one to distinguish bacteria based
on how those bacteria absorb the gram stain we use to make visible cellular
structures which are normally invisible under an optical microscope. This
distinction turns out to be useful, however, because the different bacteria are
differently related to particular kinds of disease, and so this procedure of
identifying bacteria turns out to be diagnostically useful.

(iii) Systematic Classifying

As I argued above, classifying is already involved in description.
Preparatory to constructing theories in a domain is the belief that some
things or items of experience 'belong together' in some way. We construct
classes for which we later construct accounts explaining the unifying
principle of the classification. As taxonomic disputes in evolutionary theory
indicate, classifying is not an innocent matter. Any classification admits of
rivals and needs to maintain itself in the field by considerations of fit into

some set of theoretical and empirical commitment already held in the

98

 

scientific community. Classifying objects or systems in a particular way
indicates ways of treating the object that are appropriate to the kind of thing
it is. Thus, if the 0-H group in a particular molecule is labeled an alcoholic
O-H, this will lead us to treat the substance differently than if we classify it
as an alkalinic O-H, as anyone who mistook sodium hydroxide for rubbing
alcohol would recognize!
A classification goes beyond mere description in the following ways.
i) It places this item in conjunction with other items not present, with which
we claim it has some similarity. The similarity grounds inferences to other'
structures, properties and behaviors this item may be expected to have in
common with other items in the class. Those inferences may not have been
in anybody's mind when the classification was first suggested, but may now
change the expectations of those who use it in ways that either strengthen or
weaken its value as a system of categories. It takes on a life of its own once
articulated and in use, in that it suggests theoretical moves and empirical
tests that may not pan out. Thus a whale, notwithstanding its superficial
resemblance to fish, is classified by us as a mammal because it bears its
young alive and suckles them. But, this classification leads us to expect
other common features if we were not already aware of them: whales are air-
breathers and warm-blooded. ii) a classification that has any claim to
'naturalness'39 will involve some theoretical analysis of the object that will
Sort out which items in a description are relevant to the behavior in question

and which are irrelevant. This point, too, is present in the example: it seems

A

39 We can still use such phrases as 'natural class' as long as we recognize what we are doing.
We are simply saying of one set of constructions that we are committed, in some way, to
saying the world is this way rather than another way, and that our willingness to commit that
way is at least partially a function of the theories we accept. This again is a realism of a sort,
but an internal realism constructed within a particular language with its associated set of
practices.

99

'natural' to us to regard whales as mammals precisely because we do believe
lungs and ovaries are features that only arose late in evolutionary history.

As should be pretty clear, we can classify items in a way that does
not hold up. We may assign a rock to one geological stratum, and then find
ourselves confused over the fossils it contains. We can assign an event in a
particle accelerator to a particular particle, and then find out it has the wrong
mass or charge or spin for that particle. We may find that not all members of
the class share the identifier property, and then need to give up that property
as the criterion of class membership. It may turn out that the class has no set
of properties widely shared by its members, and therefore lacks the
cohesiveness of a useful classification. Classifications ought to be treated as
hypotheses that may or may not prove useful.

We may find one or more properties shared by all members of a class
without concluding that the class is a natural class. We do not, at any point,
find classifications that are ready-made. The natural classes, which realists
assume, do not exist. All classes are constructed by us, and we judge the
degree of naturalness a particular collection of objects has by theoretical
considerations governing the connection between the objects and the

identifier property. 'All the objects in this room' may score low or high on
the degree of naturalness scale depending on our theory of how things came
to be in this room.

(iv) Measurement

All measurement involves interaction between systems to determine

the behavior of the empirical variable of interest. We need this information
before we can construct an appropriate numerical model for the empirical
system that will assign numbers to physical magnitudes and predict the

states of an empirical system by manipulating the numbers representing the

100

state variables (Suppes and Zinnes 4-8;17-19). So, for example, we learn
that weight is such that if we weigh on a pan balance two different
collections of objects and they balance, then any third set of objects that
balances either of the first two sets will also balance the other. So transitivity
holds for the notion of equal weight. It also holds for unequal weight: some
third set of objects heavier (or lighter) than either of the first two will go
down (or up) if either of those two sets of objects are put in the opposite pan
on a pan balance. Finally, there is the important property that for any two
sets of unequal weights, some object can be found that will make the lighter
set balance the heavier set. Those properties behave the same way as a
numerical model which is a subset of the real numbers under the operation
of addition. So the empirical system involving physical objects, an operation
called 'weighing' and an operation of concatenating sets of physical objects
(by placing them on the same pan of a pan balance) behaves like a set of real
numbers, ordered by the less than or equals relation and the operation of
addition. Furthermore, one can show that any other numerical representation
of this empirical system will be isomorphic with our first one. They will be
equivalent up to an equivalence transformation, in this case, multiplication
by a constant. So one can construct a scale for weight that will allow one to
use arithmetic on measurements. Nor is weighing restricted to pan balances.
If spring balances make the same objects equal, less than or greater than the
same objects so ordered by the pan balance, including for concatenations of
objects, then these two ways of measuring weight are equivalent. So, as one
can have equivalent scales with a different choice of units, so too one can
have equivalent empirical systems.

Here too, we construct. We construct the numerical relational

system. We construct the measuring devices. We engage in activities like

101

calibration, testing and refining measuring devices which solve the problem
of nonequivalence. A simple example may suffice. Pan balances do not
exhibit complete transitivity because there is a smallest discernible
difference for any empirical balance. This presents the following problem:
for a set of physical objects a, b and c, it may be the case that a is
indistinguishable from b in weight on a given pan balance, and b from c in
weight on the same pan balance, while a and c are distinguishable because
their difference exceeds the smallest discernible difference of that pan
balance. But if this is so, then the numerical relational model and the
empirical system of objects and pan balance are not isomorphic. We try to
construct ever more sensitive devices so that we can avoid the
inconsistencies obtained from the less fine grain measurement. 'Weighs the
same as' should be an equivalence relation.
(v) Experiment

Less needs to be said here: experiments are very obviously
constructed. The equipment, the design and the procedures used are all
human inventions. Experiments all involve manipulating some system by
manipulating variable conditions internal or external to it and then
monitoring the changes such manipulations produce in the system. For
example, we may produce genetic mutations either directly by irradiating the
cell with x-rays, or indirectly by putting dioxin, a known mutagen, in the
water surrounding the cell. Then the interest lies in how such mutations
affect the behavior of the cell. So, in experiments, we are manipulating the
world and watching how it responds to our manipulations. What is the
epistemic pay-off for such manipulations? What do experiments add to what

we already have in observation and measurement?

102

While it is true that the value of any measurable variable cannot be
determined without manipulation of the system in question, there are
manipulations which intentionally change and control the value of some
variables while observing the impact on other variables. One may change the
chemical environment of a bacterial culture and observe the impact on
growth. Similarly, one may change the rate of interest, and observe the
impact on consumer spending. Some of this variation is done to develop
understanding of what control of the variable means (Gooding 24-27).
Some is done because one does not know the shape of the relation of
covariation, or even what all the variables are.

(vi) Data Analysis

Data are not simply given but are constructed. Often, in the course of
an experiment, we sit down and decide whether we believe the data or not.
This may happen after the run is over, or in the course of the run, where the
readings may simply not make sense. We may check the calibration of the
detectors, see if operational protocols have been followed, and test the
instruments for defects, wear, alignment, etc. Why on earth would one do
this? Because we have expectations of what should happen, given our
theories about the behavior of the system in question. When those
expectations let us down, we have a problem that needs to be solved. Did we
do the right experiment? Did we perform it the right way? Were the
instruments the appropriate ones? Are they working properly? Each of these
questions sets us an investigative task. We cannot simply throw out the
points we don't like, because we need some justification for discarding data.
If the detector drifted off calibration, then one could argue that systematic

error in the detector invalidates the set of data points.

103

Data analysis may involve charting, graphing or trying to fit the
values into some sort of geometry of the relevant quality space: this involves
constructing a model for the data, which interacts with theoretical models of
the system. These constructions may fail: the data may simply not conform
to the model we choose, and the prediction based on the model may not

work.

(vii) Modeling

Modeling covers an array of activities. We may build physical
models by analogy with simple systems, e.g., billiard balls attached to
springs as a model of the thermal behavior of a solid. Or the models may be
very abstract, as with superstn'ng theory. To revisit a point already raised in
the discussion of measurement, one needs to be able to show that the
mathematical model is isomorphic or homomorphic with the empirical
system modeled, before one accepts it as a model. In order to do that for
measurement, one needs to construct a numerical relational model and an
empirical device with associated set of procedures for measurement, and
then show that they have the same relational properties. The same
considerations hold for more theoretical models invoking theoretical entities.
One still needs to 'get at' or identify (to use van Fraassen's language) the
relevant behavior that makes the systems relevantly similar.
(viii) Statistical Analysis

Data analysis requires statistical analysis because not all errors are
systematic. Some errors are random. A simple illustration of this is that one
may measure the length of a chair carefully ten times and get ten different
(but very close) results. The same holds true for measuring the temperature

of a glass of water with a mercury thermometer. This may seem odd, but our

104

conception of the universe is that there is a great deal of complexity to the
systems we are studying and their interactions with us, our devices and each
other. Thus, we occasionally find that our predictions of where the next data
point will be, based on our last one, fail. In such cases it sometimes is
possible to identify the reason for the failure in some feature of a system's
environment which is creating 'noise' in the detector. What we do is
construct a model of the environment that tells us why the data diverge from
the prediction generated by the model for the system of interest. Where we
cannot identify the source of the noise, we simply give a model of the data
that predicts the next data point will fall within a given range with a given
probability. Clearly, the claims are sensitive to which statistical model we
construct and the actual empirical behavior of the system in question in its
particular environment. Statistical models can fail just like other models.
(ix) Communication and Verification by the Community

Last, but not least in our (incomplete) survey of the practices of
science come communication and verification. All scientists publish their
results at some point or other. This was as true for Aristotle and Galileo as it
is today. And the point of publishing is to have the results publicly available
and publicly checked. Some advocates of the social studies of science (e.g.,
Gilbert and Mulkay) have focused on the form of the scientific report, and
the divergence from the informal spoken discourse of scientists. Although
these same authors have addressed the rhetorical features at the expense of
the logical features of scientific writing, it seems clear from the way these
reports are actually used in the scientific community that their primary
function is to enable quality control by the community over its results.
Results are never simply used nor are results simply disputed, as an

emphasis on either consensus formation and alliances, or dissensus

105

 

formation and controversy might suggest. Rather, a process of public
checking is initiated that involves examining procedure, instrumentation,
methods, conditions and materials used, as well as the 'reputations' of the
research team and lab for good work, and so on.

The sociologists have a point: a Nobel laureate is taken more
seriously than some other researcher, no matter how good the latter may be.
_ But, at the end of the day, not even Nobel prizewinners can pass off just any
project. Linus Pauling won a Nobel prize for chemistry, but the respect for
his work on protein structure did not carry over to his claims for vitamin C.
Notwithstanding the stature of Einstein, his claims against quantum
mechanics are not taken seriously in the physics community today.

This amounts to a distinguishing mark of scientific practice:
scientists check up on the work of their peers. This is not to say the system
of peer review always works. It does not work, for example, when, owing to
prejudice for or against the conclusions or the researchers, the accepted
canons of good experimental work are not used to evaluate the experiment.
But, while the system of peer review does not always work, it is nevertheless
important for the scientific community to subject its work to public display
and review. This is not the case with art, for example, where one writer or
painter need not be aware of the work of or criticisms of another artist.
There of course is peer review in the arts: juries, foundations for the arts,
etc., do employ artists to evaluate the work of fellow artists. Nevertheless,
no artist need alter their work to line it up with the recommendations of such
review--in fact, the cult of authenticity, where it is in force, would accuse an
artist who did change his or her work of lacking integrity. In the sciences,
however, it is another matter. One is encouraged to stick to one's guns there

only if one anticipates that further research will answer the worries

106

expressed by other researchers in the field. Vindication comes, if it comes at
all, in the form of experimental data that passes all the tests constructed by
the research community. The crucial test is replicability: can anyone else,
with similar materials and similar equipment repeat the experiment and get
the same results?

3.2. 6 The Test of Practice: Replicability

Why is this a reasonable test? What is at stake is whether one can
replicate the results of some piece of research. Repetition is the only
sufficient answer to the objection that a result was an artifact of an
instrumental set-up, or a researcher's expectations. Both of these points bear
on the objectivity of the results. It is not a condition of success, for example,
of symphonic performance. One may appreciate, indeed seek, different
interpretations of the same piece by Beethoven, for example. Nothing further
hangs on 'replication' of an artistic performance, except of course that it
ceases to be a performance of a particular piece if the minimal conditions of
the performance (right notes in the right order, by the right instruments, for
approximately the right length of time, etc.) are observed. Still, one may
positively value a performance played slightly faster or slightly slower, or
transcribed in a new key.

In replicating experiments, however, variation serves only one end:
to test the limits of validity of previous experimental results. 'Replication'
cannot mean exact repetition in any event, since at minimum, one is doing
runs at different times. But replication marks out a similarity class that
overlaps with experiments which extend the range or increase the precision
of the original experiment. One systematically varies the conditions to see if
the same curve captures the data. The law states the relevant variables. Now,

the point is to see how the outcomes vary with the variables. If the

107

 

covariation ceases, then one has not found the functional relation one is
seeking. If Boyle's law is checked at higher and lower temperatures, higher
and lower pressures, with the same or different apparatus, it is all with a
mind to checking the conditions under which the relationship between
pressure and volume hold. Has one captured all the interesting variables?
Does one know which variables are relevant (e. g., temperature, pressure,
volume, the amount of gas) and which irrelevant (e.g., the type of gas), and
whether, under some conditions, the irrelevant ones become relevant? For
example, at the critical point, at a different temperature and pressure for each
type of gas, Boyle's law breaks down.

The scientific community, wanting to eliminate certain types of
errors, will minimally demand the researcher repeat the experiment.
Successful repeatability, however, buys one another form of objectivity: the
results become exportable, not only from a particular lab but from labs in
general. This is valuable because, whereas replicability is a matter of the
scientific community coming to agreement that an experiment was properly
performed and the results credible, the repeatability of the results allows
scientific results passport to other communities as well, indeed, to culture as
a whole. I will say more on this point later.

3. 2. 7 Laws as Constructed Relations between Measurable Variables

In the limited sense that is central to this dissertation, a law is a
defeasible relation of covariation between measurable variables that models,
in mathematical terms, relations in some system of interest. It is a relation of
covariation: we cannot even determine it unless something in the world
changes, usually by our initiation. But it is a stable relation of covariation,
by which I mean it is invariant with respect to a particular set of

manipulations that both creates instances and fails to create counterinstances

108

even though we might have expected it to. I will have more about creating
counterexamples in the next section. The positive role of constructing
instances makes the manipulations a 'recipe' for creating ephemeral
phenomena (e.g., decay events) or stable products (e.g., polymer plastics).
The defeasible relationship between variables is one we can bring about
repeatedly. Replicability is an important part of the second order invariance
of first order relations of covariation that distinguishes an accidental
correlation from a law.

To say that laws are constructed within experience needs
clarification. There are views of experience that make it the subjective
contents in someone's head, or the pure present (the future and past only
available in representations of various sorts) or all laid out (tenselessly, as
observations past, present and future), or as untamable flux. None of these
views as they stand is particularly helpful. Experience has to include more
than the self or it will be imprisoned in the self. It must include more than
the present, or it will be imprisoned in the present. It must include more than
some static, eternal structure of observations past, present and future, or one
will never grasp the temporal character of experience. Experience must
include more than untamable flux or else one will never be able to construct
notions that make the movement of experience intelligible--or to set
expectations of what will happen that stand some chance of being useful. All
of these notions of experience are old, and caught up with the problematics
of the static and the changing. If formulated in these terms, the problem we
have set for ourselves cannot be solved, for the stability of dynamic patterns
of experience does not start with according priority to either the static or the
changing. Rather the stability that is sought is stability relative to a particular

set of conditions as we have ascertained them in a given environment. We

109

F}-

set the formulation up as we interact with systems of interest, and note under
which conditions the system behaves as expected according to the
formulation. Then we put the formulation under pressure by altering the
conditions to see if the formulation holds up. We call 'laws' only those
constructions that seem to hold up pretty well under systematic variation of
conditions of validity. So, for example, we are more confident in the claim
that a particular virus causes a disease when we can create the disease in a
healthy specimen by injecting the virus. We are more confident still when
we can cause the same disease in other species.
The conditions of the construction of laws account for some of the
features of laws.
i) These conditions of construction account for the use of simple
mathematical relations where the conditions of validity are
suppressed. Boyle's law, as we saw, is valid for atmospheric gases
under the normal range of temperature for atmospheric gases.
Although the variation of temperature is great from the tropics to the
poles, the variation remains within the scope of the law. Since such
variation in temperature does not affect the validity of the law,
temperature can be left as an unstated parameter in the formulation
of the law. Similarly, interpolation and extrapolation are 'covered' in
the sense that we have reasons to think the law continues to hold in
places where we have not checked it. (The details of the justification
for this confidence must await 3.3 on the justification of laws)
ii) These conditions of construction account for the functional form
of many laws, because the laws were arrived at, in many cases, by
varying one condition while monitoring others. Hence differential

equations frequently occur in formulations of laws.

110

up: ifuih 'I’A‘l'r x.

iii) These conditions of construction account for the fact that laws
often do not specify the units. For example, Boyle's law does not
specify whether volume is in cubic centimeters, liters, or cubic feet.
One can prove this does not matter, as Suppes and Zinnes show,
because any suitable scale will be related by a transformation
function. A similar argument can be given about the equivalence of
different suitable measurement processes for the same variable
(Suppes and Zinnes 8-10; 19-20).

iv) These conditions of construction account for the universality of
the law. This point may seem redundant in the light of point i, but it
is worth looking at the implications. First, replicability accounts for
universality with respect to time and place. Replicability involves a
denial that when and where an experiment took place is a condition
of the law holding. Absolute time and place do not occur in the
equations, only relative time and place do. Second, it accounts for
what Harre called in our previous chapter 'substance universality', his
paradigm example being Newton's laws of motion, which apply to
physical bodies generally, not just the moon or falling apples.
Replicability may, by semantic ascent or descent, widen or narrow
the domain of a law. Boyle's law originally applied only to air, but
we find that other gasses behave similarly. In fact, it has become part
of how we define a gas, so it combines with Charles's law into
something we call 'the nnimsal gas law.’ Similar extensions occur in
biology, where observations valid at the individual level are extended
to the species, the genus, the family, class, phylum or kingdom. In
the reverse direction, the laws of planetary motion are applied to

newly discovered asteroids and satellites to determine their

111

trajectory. The laws of their ballistic behavior are the same as the

laws of planetary motion. So, the domain of objects to which a law

properly applies is fixed by replication, and the universality is
relative to that domain.

v) Finally, these conditions of construction account for the inferential

role played by laws. The mathematical structure enabling deduction

applies to the empirical system because we set up the construction to
make it so.
3.3 Success Conditions for Constructions

I will argue in this section that the justification of law claims lies in
their usefulness. While no construction is guaranteed success, successful
constructs are very useful in performing the important roles of explanation,
prediction and control.

In the previous sections, I commented on how laws are constructed
and how they maintain themselves in a particular environment. Before I go
on to say how these constructs are justified, it is appropriate to say
something general about this notion of a construct 'maintaining itself in a
particular environment.’ What does this metaphor mean? Similarly, what can
it mean to say a construct 'takes on a life of its own'?

What I mean in this context by 'maintaining itself is that any new
term or claim will only be able to establish a consistent practice if it fits in to
a web of distinctions or claims already in place. In ecology, one would say
'finds its niche.’ A distinction can do so if there are no competing
distinctions: the phenomena may be new and require new vocabulary to
describe them. The competing distinctions may be vague and lump together
distinct phenomena. 'Heat' for example, in Bacon's usage, was too vague to

support any interesting theory of heat as we now understand it. Color words

112

may need to become more precise and definite to handle spectral analysis or
qualitative chemistry. Competitor descriptions that need to be dislodged may
be entrenched with other terms, concepts and claims that would need to be
revised or replaced if a new term were going to establish itself. For example,
'sunrise' as a description of the movement of the sun needs to be reduced in
theoretical status to mere appearance before one can simply say 'the earth
moves around the sun, not vice versa.‘

'Taking on a life of its own' involves the following features of
contingent group practice in the use of terms: 1) A term, once objectified,
may come to have a very different significance than the one the individual
author or even the original group of practitioners had in mind. The term may
be altered to fit in with contemporary practices and beliefs. 2) The term as
part of a process of inquiry may come to have different and, perhaps, more
definite, content as inquiry proceeds. The term may turn out unusable as
nature resists being constructed that way. Finally, and most importantly on
my view, 3) a term may be successfully embedded in new contexts. It may
be independently combined and employed with other terms and claims in
another language game. The term may do useful work in contexts other than
the one in which it originated. I have in mind here how laboratory
phenomena can be articulated in ways that make them useful in
technological application. For example, one may baptize terms such as
'atom' to describe chemical events, only to find not only the conception and
practices of chemistry change, but the proper description of atoms changes.
Dalton launched the term and the conceptual change, but the evolution of the
significance of the term was no longer in his hands once the term was

launched.

113

Both of these metaphors point to the complex and sometime very
indirect ways experience causes us to alter our practices. 'Maintaining itself
is a metaphor that captures my holism because every term requires a context.
The phrase 'life of its own' captures the objectivity of terms, because once
articulated in group practices, these terms can resist idiosyncratic use, or can
change values in the light of experiential consequences. Both of these points
differ from realist positions on meaning. For realists, it is the world, not the
group that determines correct usage. For the realists, it is the atomicity of the
world, not the articulation of the sign, that guarantees independent
combination. Still, it is important to say these things to show that 1) there is
an alternative to the view that makes meaning depend on reference that does
not make meaning 'something in the head.‘ The use theory of meaning places
meaning in group practice, with all the historical contingency that involves.
2) It is also important that holism need not commit us to deny the obvious:
that words and sentences can do roughly similar work in different linguistic
wholes.

Both points are important to both the construction and the testing of
constructs. A term or claim must make sense (fit into some practice) of some
group to be viable, but to be successful, it will need to be able to fit into the
practice of other groups as well. With some allowance for technicalisms that
are properly creatures of technical language games only, some terms or
paraphrases will have to lend themselves to common sense use. If the terms
or claims cannot help us in the way we think about the world, or change the
way we live, they have not succeeded in the sense of success important to
this thesis. I will explain what I mean by that by explaining how laws

succeed under testing.

114

Since laws are constructed under a set of conditions, there is already
testing occurring during construction, so the separation we make here is
somewhat artificial. Nevertheless, it is worthwhile for analytical purposes to
give construction and testing separate treatments, even if the same processes
are involved in both cases. We are looking at the same processes from
different perspectives when we ask alternately how laws are created and
sustain themselves in a given environment, on the one hand, and
how their stability contributes to their usefulness on the other hand. So, in
this section, I want to go on to show how constructs in general, and laws in
particular, are tested. This falls short of justification in the sense of
demonstrating the truth of such claims, at least if truth is construed as being
acceptable beyond any possibility of future revision, but it is still grounds
for rational acceptance. In particular, I want to argue the following:

1. Empirical adequacy is not enough (3.3.1).

2. Consensus is not enough (3.3.2).

3. Reliable prediction is required (3.3.3).

4. Reliable prediction comes of replicable experimental results

(3.3.4).

5. Reliability forms the basis for decision-making at both the level of

individual action and broader technological policy (3.3.5).

6. Laws are constructs on which we can build plans of action and

technical products that have some assurance of success (3.3.6).
3.3.1. Empirical Adequacy Insufficient

We live and act in time. We act based on experience, but we must
decide what to do in the light of events that have not happened yet. This is
the problem for the empiricist. The empiricist is right to insist that the world

is too complex to be known from the armchair: we must study the way

115

objects behave before we can be in any position to frame the laws governing
their behavior. But the future, which hasn't happened yet, is relevant to the
truth of claims about those laws.

One way to deal with the problem is by the concept of 'empirical
adequacy'. Construct the set of all observations, past, present and future that
bear on the truth of the claim. Van F raassen calls a theoretical claim that is
consistent with such a set 'empirically adequate.‘ Van Fraassen claims that
logical consistency and empirical adequacy are the most that an empiricist
can reasonably demand of a construct. I shall now argue that the pragmatist
should ask for more than this, and that this additional demand is reasonable
given the cognitive role we expect laws to play.

Let us grant" that, in some sense, truth is atemporal. We would agree
that if, in the future, events turned out differently than expected, we would
claim that the law we used to make the prediction was false and always had
been false. If we do not concede this, then prediction affords no real test of
laws. Still, there are two problems with van Fraassen's formulation. The first
problem is that it seems we could not know whether a claim is empirically
adequate, because we never have access to the set of observations as van
Fraassen describes it. The second problem is that 'consistency with
observation' is too weak a demand, given the requirement of
informativeness.

The first problem is that van F raassen at times talks as if this
atemporal class existed, whether anyone ever makes these observations or
not. While at times be carefully relativizes observability to technical means
available at the time, at other times he speaks as though these confirmers and
defeaters simply existed, whether they ever become part of experience or

not. This, I think, is problematic for the same reasons as the realist view of

116

truth is: once a wedge is driven between truth, and verifiability, meaning
becomes problematic. Since we are temporally and spatially located beings,
it is important that we be careful how we describe the relevance of future
evidence we do not yet possess. Van Fraassen's set of observations is an
atemporal set, and all our empirical means are temporal. We have a problem
of empirical access here.

The second problem is that consistency with observation can be
bought too cheaply. One simply weakens the theory. Instead of predicting
rain tomorrow afternoon, I predict it will rain some time this month. I save
the hypothesis, if I do, at the cost of content. I can weaken the hypothesis
into a tautology: either it will rain tomorrow or it will not. I now have a
hypothesis that is consistent with all the empirical data--and perfectly
useless as a law (Hanna, "Empirical Adequacy"). The law no longer tells us
anything definite about the world.

Now, it is one thing to say that we need empirical access to relevant
evidence, another to say how we can have it. If the future is relevant to the
truth of our claims, but we do not live in the future but the present, we have
two historically important philosophical problems we need to address. The
first is Hume's problem, that is, that past experience, as logically distinct
from future experience, cannot be the ground of a sound inference to the
latter. The second is Goodman's problem of 'grue': regularities are where you
find them, and you can find them anywhere.

The first problem is acknowledged by empiricists and is the reason
why they insist on the relevance of future evidence in the first place. It is
well known that accidental regularities, especially stochastic ones, like a run
of heads in coin tossing, may fall apart in time. One doesn't always know

when sufficient time has lapsed, and the negative induction from past claims

117

about laws has made us all fallibilists. We have learned over time that our
best formulated laws are often special, holding only under limited
circumstances within particular boundary conditions. The universal gas law,
for example, holds only for gases above their critical point. Empiricists have
attempted to solve Hume's problem by formulating inductive logics, and by
theories of statistical sampling and decision theory, but the problem of
justifying such procedures remains.

The second problem, however, is a problem even if we have a
satisfactory way of handling the first problem. Even if we were confident
that some regularities do hold up in the future we would still have the
problem of which regularities will hold up in the future. If an n-order curve
will go through all the data, there always exist curves of order greater than n
that also go through those points. The problem here is not that too few
regularities will hold up, but that too many will.

Both problems are problems because they put in question our ability
to make predictions. The first problem puts in question whether such a move
can be an inference, i.e., a justified conclusion, based on the evidence. The
second takes for granted that there can be justified inferences of this sort and
goes on to ask: which ones?

In my view the solution comes from two tests: 1) consistency with the
linguistic practice of the scientific community, that is, with the terms,
claims, laws, theories--and their logical implications, paraphrases and
assertibility conditions, and 2) the test of experimental practice that involves
the manipulation of the conditions under which the claims are assertible.

The first test comes from Goodman. Goodman solves what I have
called above the second problem, and what he calls the 'new riddle of

induction', by giving the conditions of projectibility. A regularity must

118

contain terms entrenched in our language (that is, having a history of
successful use), having examples, having no known counterexamples, and
fitting in with the structure of already accepted laws and theories. The
regularities we would consider laws must pass the test of coherence thus
broadly conceived. This includes a condition he calls cotenability. Two
sentences are cotenable if they are capable of being simultaneously true.
Some consistent pairs of sentences fail this test. For example, 'Joe is in
London', and 'Joe is in Paris' are logically consistent, even logically
compatible, but cannot both be asserted of the same person at the same time
for extralogical reasons: no person can be at two such widely separated
places at the same time.

Goodman's coherence test for lawhood is essentially backward
looking. Past success as evidenced by entrenchment is the best predictor of
future success.

Skyrms defines a second test regularities must pass to be laws. This
test looks forward. Skyrms, in W, develops a notion he calls
'resilience', which qualifies the regularities we should call laws by additional
tests. Regularities involve relations of covariation between observable
variables, observed under specific conditions. Any regularity that holds up
under the variation of some set of conditions is said to be resilient with
respect to that set of conditions.40

How do these views combine in my view? First, the question of
coherence with past expectations is important, since in the pragmatist view,

we always start in the middle. But the past is not simply observational but

 

40 Actually the technical definition involves the change in conditional probability of the
truth of the generalization conditional on variations of variables in the set of variables tested.
So, relative to that set, the regularity is resilient if the conditional probability does not
change at all, or only very slightly (with the standard of how slightly specified in advance)

119

theoretical as well. We need to be able to pick out the relevant features to
formulate a regularity, and that requires the use of language, a language
which ought to already have some success in dealing with experience. But
more is required. One needs to put pressure on the regularities in ways that
will make them fail, if they are going to fail, sooner rather than later. And
for this, one has at one's disposal, a repertoire of ways that regularities of
this type have failed in the past. This repertoire gets worked into
experimental methodology as self-checks that make sure the data and the
models are not flawed in certain specifiable ways. They may be flawed in
other ways not yet determined, but minimally, a model should at least not
fail in some of the ways past candidates have failed. Thus, one does things to
the apparatus to make sure it has held its calibration and doesn't have
peculiar flaws invalidating the results. One checks the sample for purity.
One checks other samples for comparable results. Learning to separate the
relevant from irrelevant variables is a process that may take a variety of
procedures and experimental set-ups, as anyone reading the series of
experiments Boyle actually performed can see.

How does my approach help with either Hume's problem or
Goodman's problem? How does reference to the logical and experimental
practices of the scientific community license inferences from limited data to
universal generalizations? How does such reference handle the problem of
underdetermination to pick out the function that best predicts the future
behavior of interesting systems?

My approach dissolves Hume's problem. There are no serious
skeptical doubts about whether induction is rational for the pragmatist, and if
we do not have such a doubt, we do not need to eliminate it. The problem is

not whether inductive procedures are rational. We project regularities and

120

have no reason to be dissatisfied with those projections so long as they
continue to work. This just is what we mean by rational inference, based on
the success of our best practices. regularities we succeed in projecting are
laws. Now, 'success' here, should not be glossed, as the realists might, by
'approximately true' but rather as 'true.' This is the only truth we know, that
which enables us to navigate through experience. Successful belief sets
expectations that are not frustrated, and we make these judgments from the
point of view of the present looking into the uncertain future, not from the
present looking over the certain past. So the first test, the test of coherence
with past practice, is different from a test of coherence based merely on
belief, insofar as practice already involves successful realization of the
belief. Practice thus undermines at least one of the possible ways in which
belief can go wrong, unrealizability. This point is important to make because
oftentimes, the terms or claims are formulated after the phenomenon has
been observed, and deployed to explain the past. But an explanation that
yields no predictions is justifiably suspect.

The second test, however, goes beyond the first insofar as we
explicitly fly to upset the regularity by playing with the conditions under
which it is observed. Inforrnally, in everyday experience, we do this when
walking on ice or unstable structures. We send 'test loads' by shifting our
weight in various ways, or only placing some of our weight on a spot, rather
than risk our whole bodies prematurely. The common sense notion behind
all this is: any number of times, I have been wrong about the soundness of
these structures, and better that they fail now when the risk is small than
later. In this fashion, we can localize the trouble, if trouble there be.

Similarly, regularities can fail any number of ways, and specifying

the conditions of failure actually renders the regularity itself more precise.

121

Hence, testing and attempting to fail regularities also is a way of picking out
the regularity, from the class of regularities allowable by the data so far,
which we find the most projectible. Playing with the conditions may allow
us to see the character of the functional dependence (more on this later), and
see if the best curve is, for example, a linear or quadratic function of the
variable. The most informative function will be the one that restricts the
outcomes the most, but it will be the one most vulnerable to upset when we
further determine data points.

An objection that might be raised at this point: by calling this
'success', am I not begging the question? Isn't the success of past practices
precisely what is at issue, and what is threatened by future evidence we do
not have yet? My response is: realizability is a real test, and continual
realizability is all that temporal beings like ourselves, without crystal balls,
can reasonably ask.

To see why this is a good response, to see why the pragmatist's rule
'don't worry about beliefs that are not causing you problems right now' is not
just ostrich-
epistemology in the face of insoluble problems, let us see the connection
between this answer and Reichenbach's pragmatic dissolution of the problem
of induction. That response goes as follows: if there are no laws, if, for
example, the relative frequency is not going to converge on some definite
value, then induction will fail, but so will every other attempt to find a
regularity. If there are regularities, if the series of coin tossings converge on
a definite result, if any other method can find the regularity, so can
induction. If crystal ball gazing reliably predicted the weather, and
sophisticated computer models did not, the inductivist would notice the

regularity between the prediction of the crystal ball gazing and the event

122

predicted and form a law connecting the two. If the crystal ball connected
two sets of conditions reliably, the inductivist could formulate the law sans
crystal ball. How does appeal to past practical success hook up with this
response? Past success implies that the language we have adopted and the
practices we employ had survival value. That is why we continue to use
them. But ultimately, the question of survival can only be made at the end of
time. So here is the dilemma. It may be that we are massively in error now
and will be extinct by the year 3000, if local conditions have been benign to
our error over the past 100,000 years. All our success may be apparent.
Here is the practical version of Reichenbach's response. Perhaps none of our
practices succeed, in which case no one will formulate regularities that hold
up and we will all fail. But if any practice succeeds, scientific practice will
succeed as well, because it builds on past successes. The worry is unreal
furthermore, because the fact that we are still here, living in a variety of
environments over a long stretch of time has to count as success if anything
does. What could the contrasting term to apparent success be if, ex
hypothesi, there is no such thing as real success? So long as our laws and
theories continue to assign high likelihood to whatever happens, we have all
the success we could reasonably demand of them. We can use them to
navigate through experience because we can base actions on their
predictions. They inform us, in non-trivial ways, about what will happen
next

The practice of scientists, therefore, in requiring accepted
terminology and procedures for scientific work, goes beyond empirical
adequacy in at least two ways. First, it is not enough for a term or a claim
that it retrospectively can account for all that has happened. It must be put

into practical use and prove its usefulness by continuing to make good sense

123

of experience over a period of time. If the distinction is idle or proves
unusable because its conditions of application are unclear, it will fail to
embed itself in any long-lived practice.

Terms already in the language have already proven themselves to
some extent. The conservative impulse in science, remarked by Kuhn in his
notion of 'norrnal science', has this pragmatic justification: although there is
no sense to talking, as a fallibilist, of absolute success, since we may be
wrong in any of our claims, it still makes sense to talk of relative success.
We can demand that scientific claims be W as common
sense claims, because common sense claims have proved their survival
value. Current science can demand that its successor be W
as it currently is, even if definite trade-offs are involved in the switch.
Similarly, terms can be replaced, but only by terms that with their associated
distinctions and claims do a better job. The group judges whether to adopt
the new term or not based on what the new term will enable the group to do.

It may not be immediately apparent, but the demand for practical
usableness, as a common sense measure of theoretical adequacy, amounts to
a demand for minimal informativeness. One needs to know the empirical
conditions under which to correctly deploy terms and claims.

Second, an interest in self-consciously improving empirical beliefs,
led scientists to develop the informal common-sense procedures for
checking beliefs into formal scientific methodologies of great power. Those
scientific methodologies, in turn, transform the vague hypotheses found in
common-sense theories into precise testable scientific theories. This quest
for improving belief leads to the search for more informative hypotheses and
precise predictions--the very same object sought in the quest for improved

practice. After all, we are better able to act when our plans are based on

124

reliable forecasts of the behavior of the systems with which we interact.
Mariners are better able to navigate when they have more informative maps
of the ocean.

Briefly, why does not my view have the problems I found with
empirical adequacy, i.e., empirical access to an atemporal set, and failure to
motivate the drive for informative, testable claims? First, while I admit the
relevance of future evidence, I do so by hedging my claims as defeasible.
Only experience is relevant to whether my claims work or not, but so long as
my claims continue to pass appropriate tests, I do not let any merely
theoretical possibility of error cause worries that I do not have. The claim is
true -- but I may revise that claim later. Second, mere consistency with
observation is not enough. Laws as guides to practice require the most
informative claims about the world we can sustain.

Empirical adequacy is not enough because even when it is forward
looking and predictive, and not simply assuming a view from nowhere (or
nowhen), it fails to give any reason why more informative laws are
preferable to less informative ones. If simple consistency is enough, then
tautologies will be fine, because they are consistent with anything that
happens, no matter what happens. One cannot pull any definite predictions
out of vague or vacuously true claims, so they are of no practical use. The
very reason we frame laws is to enable us to successfully navigate through
experience. So, in addition to empirical adequacy, one requires some
judgment by the scientific community that the regularity is realizable and
informative.

3.3.2. Consensus Insufficient
To bring in the judgments of the scientific community may seem to

give conservatives veto power over new formulations and approaches. Even

125

if standards evolve (as they surely do), the standards are set by the most
established group that retains the power to silence dissent. Some social
theorists of science (Latour, for example) seem to believe this is true. They
view science as the construction of networks of alliances where the strongest
rhetoric wins. The point I want to make here is that constructions may fail
even if scientists, with all the good will such consensus requires, agree on
the language and tests employed. The analogy with engineering is perfectly
appropriate here: even bridges constructed with the approval and careful
work of engineers sometimes fail. We find conditions or combinations of
conditions we had not considered frustrating the generalization we
employed. It may turn out that some simplifying assumption or idealization
we used has a more limited domain than we thought. We cannot conjure up
phenomena simply because such phenomena would fit our theory. We may
never produce either the phenomena that we would call 'superstrings' or
those that we might call Higgs particles, just as we were unable to produce
phenomena we would call ether drifi, cold fusion or polywater. Such terms
may turn out not to be useful because we cannot construct a consistent
context of use when there simply are no phenomena that consistently behave
as we would expect these to behave, no matter how much we play with the
conditions of observation. While there can be theory-motivated reasons for
saying our theory must be wrong, and while the point at which we give up
may be socially negotiable, it is simply false to say all is negotiation.
3. 3.3. Reliable Prediction is Required.

Thus, I believe the test of importance, to which we subject scientific
formulations of theory and law, is the forward looking test of reliable
prediction. This may disappoint those who expected something more than

Popper's criterion of falsifiability, but it is important to see why this is a

126

reasonable test having direct bearing on the usefulness of laws in grounding
practice. We have already pointed out that empirical adequacy without
inforrnativeness fails here, because if any formulation is empirically
adequate, so are all weaker versions. By 'going vague', we can save laws
from empirical falsification, but then they no longer support definite
predictions.

The point is that the functional relationship expressed in a law can be
characterized both positively and negatively. Positively, it is simply to say
that there are instances of the functional relationship that we have been able
to produce and that this functional relationship holds over a broad range of
environmental variation. In other words, this is simply to say that the
relationship is a reliable basis for action. Negatively, we are saying that there
is no condition or set of conditions we have been able to produce showing
that the functional relationship does not hold. It is precisely our inability to
produce conditions under which both variables vary independently of each
other that shows us we are dealing with a law rather than an accidental
correlation. We believe that for accidental correlations there is some
producible condition or set of conditions under which the functional
relationship would disappear.“ The collective experience of the scientific
community furnishes a repertoire of techniques for unmasking accidental
correlations, i.e., a set of experimentally controllable conditions, and a set of

techniques or manipulations for controlling them. We want to say that if a

 

‘1 One doesn't want to say "under which the functional relationship would change" since the
latter includes situations where one finds out that the law requires more complex
formulation. The law governing aerodynamic drag is a case in point: under low velocity,
aerodynamic drag is approximately a linear function ; under higher velocity, second and
third order terms turn out to have non-negligible contributions. One wants to say here not
that the correlation picked out in Reynold's law does not exist or is accidental; rather, one
wants to say that the simplifying assumptions do not hold when air flow becomes turbulent.
The boundary conditions for the simple law signal the need for the more complicated
formulation when the boundaries are transgressed.

127

generalization holds with respect to the potential defeaters producible by the
relevant set of techniques, that generalization has a prima facie, defeasible
claim to lawhood. So, for example, we know how to produce a mass of gold
and a mass of U235. We could bring together masses of a magnitude
showing that there is a law that masses of U235 above a certain critical limit
disintegrate, but masses of gold can be increased without limit (beyond the
available gold, of course). Thus, we can manipulatively produce a '
counterexample to one supposed law, breaking the symmetry that existed
when considering them only on syntactic or semantic grounds. Thus, the
objection of van Fraassen from syntax and semantics that laws are
indistinguishable from accidental regularities can be handled in a
straightforward way in my account. The pragmatics of laws differs from that
of accidental regularities.
3. 3. 4. Reliable Prediction from Experimental Replicability

That is the negative side. Failure to defeat a law claim gives a set of
conditions irrelevant to the holding of the regularity. We can say , at least,
that the generality is invariant with respect to variations within that set. On
the positive side, one wants to be able to reliably bring about the holding of
the regularity by producing the set of conditions relevant to its holding. To
do this, one needs a recipe for producing the phenomena in the laboratory.
This is what Gooding in ExpenmemAndthMakmgoflMeamng calls
"making skills disappear". The ingenuity, skill and learning of a particular
researcher needs to be incorporated into a device or instrument with
instructions for its proper use that can be followed by anyone to yield
consistent results. Gooding's example is Michael Faraday and his production
of an electric dynamo. The point is that skill is relative to individuals, but the

phenomena need to be publicly available or publicly producible before they

128

can be candidates for natural processes. Since I believe skill is part of a
group practice, I am not as concerned about the problem of individualism as
Gooding, but I do wonder how scientific products can be exported to the
non-scientific public. The first step, I think, is the one Gooding has
mentioned: one needs to embody the knowledge in devices and routines for
their regular production.

The next step is crucial. These recipes need to be able to be followed
outside the laboratory. The device must be dispensable, or usable outside
the laboratory. The usefulness of the result will depend on whether the same
(or similar) conditions produced in the laboratory by device cum procedure
can be produced outside the laboratory. This can, of course be true
vacuously if the conditions cannot be produced in the laboratory. So, we
need to restrict ourselves to phenomena that can be produced and leave to
one side phenomena that we think might be producible but are not within the
limits of our current theories or technology. Minimally, we want phenomena
whose production is feasible under current theory and technological
development. We want results exportable from the lab, after being produced
under conditions artificially in ways that we think the phenomena might
arise naturally, given our theoretical beliefs about the set of laws we believe
hold‘zand about how the environmental conditions normally vary.

Thus, the devices used by Boyle to determine the functional
covariation of pressure and volume at constant temperature of gas, are
unnecessary for the regularity to hold outside the laboratory, because we

have independent ways to measure the temperature and have both empirical

 

‘2 The apparent circularity in the appeal to the set of laws in the settling of some law is
simply a restatement of the holism implicit in the pragmatic view: any law needs to fit in
with the set of laws we already accept, and modifications need to be made to make the
whole set consistent when additions, subtractions and revisions take place.

129

and theoretical reasons for thinking that the correlation holds up under
normal temperature variation in the atmosphere. The regularity's holding
being independent of a specific set of manipulations, however, does not
mean it is independent of all manipulations. Some manipulation or other,
perhaps suggested by the common experience of mountain climbers, or by
mariners observing temperature shifts and wind patterns, accounts for the
intuitions that led to the scientific formulations of empirical and theoretical
beliefs about the normal variation in the atmosphere.

In some cases, the practical application of a regularity precedes its
theoretical articulation. For example, water pumps preceded the experiments
of Torricelli, Perrier and Boyle. Once we had the theoretical explanations,
we could understand the limitations of the water pump. We now understand
why a different device, working on different principles, is required to pump
water from a depth below 32 feet. Methods of starting fires preceded our
understanding of the chemical nature of combustion. But, once we have
Lavoisier's experiments we come to have other methods for making fires and
for putting fires out. In other cases, practical application has followed
theoretical articulation. For example, the lightening rod, a practical way to
prevent fires from lightening awaited an understanding of the electrical
nature of lightening, and Franklin used the latter to construct the former. In
both cases, the basic point is the same: we want lab results to have relevance
to events outside the laboratory. We want, in other words, to be able to claim
that if it consistently works in the laboratory ('in vitro'), one has good
grounds for believing it will hold outside the laboratory ('in vivo') as well.

Bare replication may seem like a poor candidate for a reason why
experimental results apply to extra-experimental situations, and usually we

do want more than simple repetition: we want extension of conditions,

130

variation of apparatus, test material, more precise measurements, etc. But
even simple repetition may constitute a severe test for an anomalous result.
It was enough to settle the questions of polywater and cold fusion, to
mention two recent controversial cases. The demand for replicability
captures our intuition that if we all do things the same way, we should get
the same results. replicability weeds out the idiosyncratic and replaces it
with routine procedure. Ideally, we would like the procedures to be
executable by a machine. Data reproducibility is the scientist's criterion of
intersubjective validity, the form objectivity takes in science.

The test of replicability is not always decisive. There are times when
we trust the irreproducible results of a reputable scientist over the inability
of his or her less reputable colleagues to duplicate the results. But for the
most part, the rule is: the more people who can get the same results, the
likelier that the results are trustworthy.

3. 3. 5. Reliability and Decision-Making

Reliability, in the sense of 'reliable basis for decisions at the level of
individual action and broader technological policy' is a natural extension of
the requirement of replicability: both are rooted in the intersubjectively
assessable 'world' we construct, and the same 'recipes' work in both the
internal scientific context and the external social context.

Such objective results are the useful bases of actions and policies on
both the individual and social level. Once Faraday constructed his
dynamometer, we could use it to generate electricity for personal and
business use in motor cars and offices. The science of electronics puts us in
a position to produce, modify, block electrical and magnetic phenomena for
various uses. The science of chemistry puts us in a position to construct

materials with various useful properties. The laws in these disciplines

131

describe the behavior of physical and chemical systems under modified
conditions in their environment. Thus, we can melt metals and alloy them
with various substances to improve strength, conductivity, malleability,
ductility and other useful properties. We can produce a pressure gradient that
forces fluid through a tube, or a potential difference that produces current in
an electric circuit. We can follow the course of some bacterium or some
macromolecule in a living system by altering it with a radio-active
component in a radio-immuno assay. Using the laws of chemistry, we can
exchange radioactive atoms for stable atoms in a cell's structure, and using
the laws of physics we can make a detector sensitive enough to follow the
cell with the radioactive tags.
3.3. 6. Laws and Action-Plans

The upshot of my view is this: we construct laws under conditions
observed while we interact with systems of interest. Those conditions allow
us to reproduce results when we have come up with a useful construction.
Reproducible results are useful because we reproduce them whenever we
want. We can exploit such results to devise systems and products to enhance
our quality of life. Laws are constructs on which we can build plans of
action and technical products that have some assurance of success.
3. 3. 7. Three Potential Problems

Are there any special problems with this view? Let us see how this
view handles the following three problems:

I. How do I justify the use of continuous functions?

11. How do I justify the selection of one curve from an infinite class

of data

equivalent ones?

111. How do I justify the use of idealizations?

132

I. How do I justify the use of continuous functions?

Any data set is always finite. A law, however, is often expressed in
mathematical form by differential or algebraic equations which have an
infinite set of potential instances. In fact, some equations have an
uncountable infinite set, since they have real valued functions for solutions.
How can we justify using continuous functions on experiential grounds?

We can and do by arranging that the conditions we sample 'cover'
the curve we think best captures the law. We sample at different points
within a given run, between runs, and between replications. As we test for
replicability, we enlarge the sample space to check in between and beyond
points checked before. We can systematically vary the sampling to
approximate the rationals: by deliberately checking in between points
checked in the past, we can produce an approximation, in a series, to a dense
ordering.

Now, we are not restricted to measurements with the same precision
as previous ones. Clearly, scientific measurement has become progressively
more accurate over the past three centuries, and if we were to hypothesize
that we could continue to refine measurements over infinite time, we could
come up, for any given data point, with a series of measurements that would
give a decimal expansion of the reals. This would be the operational
equivalent of a Cauchy sequence. All one needs to do, if one is an
empiricist, is to say that there is no empirical bar to thinking one could
continue to improve measurements in this way until there was no practical
difference between what one has empirically and what one has in the real
valued function. Nested sequences of measurements, with progressively
smaller widths of error, relative to the same level of confidence, can make

such an induction rational if not risk-free. In fact, if one assumes the series is

133

completeable, continuous functions are the only ones that will do, and the
conclusion is forced (Hanna, "Empirical Adequacy").

11. How do I justify the selection of one curve from an infinite class of
data equivalent ones?

Even if we are justified in using continuous functions, we still have a
problem. There are simply too many real valued functions that will do the
job, infinitely many in fact. This is not the case of course if we can complete
the series mentioned in the previous section: in that case we are left with
exactly one function that will do.43 In principle, the justification is complete
for such convergent series. But short of that, for any finite sequence of
measurements, even a convergent one, we are left with a nontrivial width of
values for any confidence level.44 How do we justify the choice of any one
in particular? The answer a pragmatist will want to make is that so long as
the curve chosen continues to give good predictions, that is so long as all
data turns out to be, within experimental error, on the curve we have done as
well as we could want. If it makes no difference in the expectations set
which curve is chosen, then the choice is one of indifference. We need to
keep in mind what I said earlier about the need for the most informative
curve justifiable by the data, since the precision of prediction and,
equivalently, its falsifiability, hinges on that. Nevertheless, we still are left
with a very large class of highly informative curves, and so long as the one

chosen continues to assign high likelihood to whatever occurs, we have no

 

43 Actually, even here we have an equivalence class of curves that differ from each other by
at most a countable set of points.

44 The confidence level, or level of significance, is, as mentioned before, the chance that the
next measurement will fall outside the interval of error for a given method of measurement
with a given device. If one insists on certainty, the width of the interval is infinite. Dropping
to 99% confidence gives one a manageable width, but more typical are the choice of 95% or
90% confidence levels.

134

reason to complain that perhaps we have chosen the wrong curve after all.
Such a worry does not arise.

Of course, it ceases to be a matter of indifference once we improve
the accuracy of measurement, and, as I said earlier, the desire in science for
maximally informative claims about the world drives the push for more
accurate measurements. What happens here is that the interval within which
we look as the most probable place for the next measured data point to be
found, relative to a given confidence interval, progressively shrinks as
accuracy improves. So long as the maximally informative curve continues to
lie within the interval allowed by our most accurate measurement, then the
curve is resilient in Skyrm's sense, and projectible in Goodman's sense, and
justified in my view.

111. How do I justify the use of idealizations?

How do we justify the use of laws that invoke entities that cannot
exist, for example, point masses, frictionless pulleys, infinite potential wells,
and perfectly rigid bodies? The answer again is: because we can construct a
series of physical systems where we progressively control the bothersome
variables and minimize them. I can reduce fiiction, and watch how the
system behaves under conditions of reduced friction. I can observe trends in
the behavior that I can extrapolate to the fictional case where friction is
eliminated altogether. The last term is a construct, never observed, but it
may turn out to be a useful approximation to the behavior of unideal
systems. Indeed, it often turns out that real systems, under a range of
conditions, behave very much like the ideal systems we talk about in our
laws. Real gases, in many cases, under normal pressure, temperature and
constructible volumes, behave very much like the ideal gasses in the ideal

gas law. Ball bearings rolling down smooth planed incline planes behave

135

like frictionless systems and bumper cars floating on carbon dioxide gas,

such as those used in classroom demonstrations by Feynman, behave even

more so (Feynman 10-5). So long as I can control the variables, I can make

the real systems behave as closely like the ideal systems as I wish, and that

is sufficient justification, I think, for constructing such idealized, simplified

systems in our notation.

3.4 Answers to Important Parts of Van Fraassen's List

It is time to see whether we have solved the original problem or not.

Let us look again at van F raassen's list and see how my view fares. The list

van Fraassen gives (again, with my explanations of the terms):

1) Universality: The law must hold under all relevantly similar
circumstances. All F's must be G's if it is a law that F's are 6'5.

2) Relations to Necessity: Laws involve necessity in the following

ways:

23) Inference: One must be able to infer the current
observable state of affairs from the law and current
conditions if this is a case governed by this law.

2b) Intensionality: One cannot simply obtain the truth of the
observable state of affairs by detaching it from the law-
operator, because the state of affairs is located in an
intensional context. The logic involved in the inference in 2a)
is not truth-functional.

2c) Necessity Bestowed: Any necessary connection between
natural events derives its necessity from a law.

2d) Necessity Inherited: The view (which van F raassen calls
a minority view) that for laws to confer necessity on their
instances, they must themselves be necessary truths.

3) Explanation: Laws must explain why their instances occurred.

4) Prediction and Confirmation: Laws must be connected to
observable states of affairs so that our belief in the law can be

136

increased by viewing positive instances and, so that our belief that
the state of affairs will occur is increased by our belief in the law.

5) Counterfactuals and Objectivity: Our belief in the connection
between events related by law must be such that we would expect it
to hold in unobserved instances and even impossible to observe
instances. This is part of what we mean by the 'mind-independent'
status of laws: their objectivity.

5a) This includes context-independence: Laws are not
strongly theory dependent or language dependent

5b) Laws are objective: They are discovered, not
constructed.

6) Relation to Science: Laws are a central (perhaps the central)
feature of science. Without laws, science could not perform the
intellectual role it does play for us.

1) Universality:

Universality is relativized to a set of experimentally produced
conditions and conditional variations. As Skyrrns would put the matter, the
regularity is resilient with respect to a give set of conditions. But this
universality is nevertheless quite useful. We have well-motivated reasons for
specifying the domain of objects, systems, events, phenomena for which the
law holds, based on the inability to eliminate the functional covariance by
manipulating the objects in various ways. This includes reasons for thinking
that location in space and time will always be irrelevant for any given law.
2) Relations to Necessity:

With Goodman, I want to say that if we can break into the circle of
concepts that includes: law, necessity, disposition, possibility and
counterfactuals anywhere, we can define the others in terms of the one
clearly understood concept. Projectible regularity explains and determines
the truth of counterfactuals, and those in turn give the following analysis of

necessity: invariable succession, not only factual but counterfactual. The

137

necessary is contrasted with the accidental, on my view, and we pry apart
accidental correlations by creating a condition or combination of conditions
where the functional relationship disappears. When we cannot do that, we
assent to the counterfactual and maintain the necessity of the occurrence.
But then, all we mean by necessity is that an event is to be expected based
on the set of laws in conjunction with the set of conditions prevailing at the
time.
2a) Inference

The heart of the matter is that we believe we can infer further
(future) facts from facts we already have. In particular, we believe that the
states and behavior of systems do not vary randomly, but in a manner that
we can determine by test and measurement. We can construct models from
which we can pull, by appropriate mathematical procedures, information
about the world verifiable by measurement. If I am right, we necessarily
solve both of van Fraassen's problems at the same time. We identify laws
with the construct because the conditions of construction include identifying
empirical states of affairs through observation, measurement, and
manipulation of the system. The inference problem is solved when we have
satisfied ourselves through appropriate tests that the construct is stable, and
therefore usable as the basis for action.
2b) Intensionality

Laws remain intensional, since co-extensive scientific terms may
differ in procedures by which we identify instances, and claims may differ in
the tests they have sustained. To use Dretske's example, it might be a law
that all diamonds have a refractive index of 2.419 and not be a law that all
minerals mined in kimberlite have a refractive index of 2.419, even though

all diamonds are mined in kimberlite and vice versa (Dretske, "Laws of

138

Nature", 250). This would be because we are more inclined to identify
diamonds by their structure, and refractive indices test structure, whereas we
think kimberlite should be identified by its location in a geological stratum,
and think kimberlite's history relevant to the appropriateness of the
designation. Indeed, while it makes sense to think we can create the
conditions under which artificial diamonds might be made, by, say,
compressing coal, we would think we could create kimberlite artificially
only if we could reproduce some of the historical geological conditions that
brought kimberlite into existence in the first place.
2c) Necessity Bestowed

Goodman maintained that once we have broken into the semantic
circle of law, possibility, necessity, disposition, counterfactual--we can
explain the remaining terms in terms of the clarified term. We broke into the
circle by means of the two notions of projectibility and resilience.
Projectibility might here be glossed in a way that gives us a handle on
necessity: we confidently believe this regularity carmot be broken, and that
the consequent always will follow on the antecedent. Resilience adds to this:
our confidence is born of failed efforts to produce a counterexample.
2d) Necessity Inherited

If this criterion maintains we have necessary laws, it is simply
mistaken. The problem is that such laws would hold in all possible worlds,
even worlds where the antecedent conditions do not exist and could not be
brought about by any set of manipulations. Assuming we can make sense of
this notion of possible worlds, the idea of a law that holds, even though there
are no instances of it ever, does seem exceedingly odd, but some realists
seem willing to contemplate such laws. The best sense I can make of the

notion is that conceivably, the world is such that there is only one set of

139

possible laws, and that, try as we might have to manipulate the conditions
that obtained at the big bang, our current set of laws would have been the
ones that would have resulted. If this is the claim, it seems to me unlikely to
be true. Had conditions differed but slightly, life would not have evolved,
and none of the laws governing biological things would have existed. Had
the conditions been but slightly different, large bodies such as stars and
planets would not have existed, and the law of planetary motion would not
have existed either. If, in other words, some laws emerge only when entities
of the requisite complexity exist, no such law would exist were the entity
non-existent. Since laws are the laws of behavior of some system or other, it
is very hard to make sense of a notion of behavior without a behaver. So, if
my view fails to meet this criterion, it is no defect in my account, since the
notion of a necessary law (outside logic or mathematics) seems incoherent.

A notion of law relative to a language and set of manipulations
delivers necessity where it counts: firm expectations of what will happen in
the world. Such a notion of law does not confer necessity on the law itself in
a way that would make the law hold independent of a choice of a language
or a set of manipulated conditions. A judgment of invariance under
systematic manipulation comes a posteriori, not a priori.
3) Explanation

One puzzle for realists is how an extensionalist'view can explain
anything. If a law just is a regularity, then it does not explain the regularity
but comprises it. The point here is: insofar as we may legitimately infer new
instances from a law, we at some level explain their occurrence. For a
pragmatist, we're simply imbedding an event in a coherent explanatory
context, and this is as legitimately an explanation as anything the realists

might put forth.

140

There is a deeper point, however, that involves rehabilitating a
principle now regarded in disrepute, the symmetry thesis that claims all
explanations are also predictions and all rational predictions involve
explanations. The point I want to raise here is that it is too easy to rationalize
after the fact based on some theory constructed just for that purpose. That, I
maintain, is something of the origin of the realist view of universals: they are
shadows of real things in the world taken for explanations for why things are
the way they are. The real test of an explanation, something that bears on its
falsifiability, is the ability to assign a high likelihood to every event that
happens, including those which, but for holding the explanation, we would
have assigned a low likelihood.

The point of the symmetry thesis is that an explanation, to be any
good, must be informative. If it is to yield any genuine understanding, it
must tell us where to direct our attention and where to direct our efforts, if
we are interested in a particular outcome. My view makes sense of the
demand for simplicity and the ban for the ad hoc in explanations. The
simplest explanation has the highest information content and is the most
falsifiable: if it works, and works consistently, it rules out a lot of ways the
world could be, and also a lot of possible actions as ineffective. Complicated
ad hoc explanations tend to be made after the fact, to repair some failed
explanatory attempt. Simple explanations are preferred because they give us
a small class of factors to manipulate in testing the law. The goodness of an
explanation lies in its ability to predict, consistently, whatever is going to
happen. An after the fact explanation may be no good for prediction
precisely because it is vague: it gives us nothing to do, and fits such a large
class of outcomes that its information content is low. On my account,

science is driven by the need for maximally informative accounts, since

141

 

 

those are of the most practical value. Increasingly accurate and precise
measurements give increasingly narrow windows for success: the law that
passes such tests is more useful than one not so tested.

4) Prediction and Confirmation

Van Fraassen, together with empiricists generally, tends to treat
prediction and confirmation on a par: theory is a black box with empirical
inputs and outputs. Prediction is the empirical output, which gets compared
with direct empirical observations of the system: if theory and observation
agree, the theory is confirmed by observation. This is overly simple, and not
all agree with it. Popper accepted the importance of prediction but rejected
any notion of confirmation. Some, like Camap and Hempel, tried to have
both.

Here in the context of a discussion of laws, the important point is
this: we have observed regularities which, as I have already said, are
themselves constructions, and then one has calculation enabling models,
some concrete, some formal and abstract; most mathematical, and we may
compare an observed course of events with a predicted one to see how well
they agree. I take it as a virtue of my view that it tells how the use of
mathematical models is consistent with an emphasis on empirical checking
of the behavior of modeled systems. There is a continuum of manipulations
from the experimental, through the quasi-experimental (i.e., linear regression
methods) to the nonexperimental, formal logical and mathematical
manipulations of symbols that involves no radical breaks. That one naturally
selected system can model another need not cause any surprise, and
naturalized epistemology need not entangle itself in the problems of
defending some sharp distinction between observation and theory. On this

view, prediction and confirmation are cognitive extensions of our continuous

142

interaction with our environment, which at this level involves transfer of
information.
5) Counterfactuals and Objectivity

Counterfactuals are claims about what might or would have been the
case if some aspects of the situation had been other than they were. They
have proved notoriously difficult to analyze. How can we extend the scope
of laws to govern not only the observed, but the unobserved and even
impossible to observe? One way is to notice the expectations laws set up.
For example, this book on my desk is not now falling. Have I refuted
Newton's law of universal gravitation? No. At present, the force of gravity is
balanced by the upward force of the desk (which in turn, is based on the
electrostatic forces of attraction and repulsion between molecules of the
desk, the book, the floor, the earth, etc.) But, if the forces were not balanced,
if the book were unsupported, then the book would be falling toward the
center of the earth at the average velocity of acceleration of gravity times the
time fallen, divided by two. An astounding claim! Why do I believe it? On
my view, we assent to counterfactuals of this sort, because we can bring
about the conditions where its truth is manifest: we can move the book so it
is no longer supported, and observe that it does indeed fall. So,
counterfactual truth rides parasitic on factual truth, and the bridge is made
by our constructive activity altering conditions.

This seems a bit too simple. Could Einstein really travel at the speed
of light to observe the beam of light? No, he could not. But thought-
experiments are internalized and abstract versions of our concrete
manipulations of real systems. And the test of the validity of a thought-
experiment is often whether there is some possible experimental realization

of the results arrived at by thought. Insofar as we cannot do this, our

143

intuitions embodied in the thought experiment are often shaky, and the truth
of the counterfactual claims are often indeterminate.
5a) Context-independence

I am inclined to think that this requirement ought to be given up. No
construct is context-independent in any absolute sense. All wind up
embedded in various contexts, whether linguistic, institutional or
experiential. Nevertheless, there is some remaining sense to context-
independence that'is important and worth commenting upon. This is the
sense in which a construct can be successfully embedded within a context
other than the context in which it was constructed. We want inoculations to
work not only for laboratory rats but for the people living outside the
laboratory who are prone to the disease. Science would be a pretty poor
affair if it were created only for the consumption of scientists. And some
work in the social studies of science seems to suggest that science is just
that. With my notion of reliability, we have a notion of scientific validity
that allows scientific results to lodge themselves into practical contexts
outside of science. Engineers can count on the work of physicists when
building suspension bridges. Inspectors can count on the standards of
workmanship created by engineers. Insurance agents can count on the work
of qualified inspectors. Politicians and citizens can count on the work of
physicists, engineers, inspectors, insurance agents and underwriters when
deciding to build and use bridges. There is a web of practices, united by
overlap of expertise, that mediates the formulation, testing and application of
scientific laws, and these multiple frames assure a limited sort of context
independence that can be quite far reaching in its effects: it can allow a
certain degree of transcendence of culture so that western science can be

appropriated by non-westem cultures.

144

Given my claims about how laws are language-dependent and
embedded in a set of practices, how is this possible? I think the short answer
is that strings of symbols can be subject to different interpretive contexts
imposed by the different sets of practices employed in different
environments. So, "F=ma" as a schema, gets construed differently in the
practice of pure physics than in applied physics, even though the same
physical models may have led to the formulation and testing of the law as in
the application of it. And this is simply because engineers follow different
practices than physicists, although there is some overlap. The engineer has
to be concerned with whether the devices will actually work in the
environments for which they are designed, and how to modify them if they
do not. The physicist takes for granted that some devices work, and tries to
elucidate the principles of their operation. The language used by them differs
in ways appropriate to the aims, methods and standards appropriate to their
set of practices, and some translation is required for them to communicate
with each other. But communicate they do, and strings of symbols like
"F=ma" are accorded different uses in the process of translation. The brief
answer is that bits of language can embed themselves in new linguistic
contexts and do useful work there. Insofar as this is possible for various
scientific laws, we have a form of context independence for laws.
5b) Objectivity

This, I believe is the crucial condition. Making laws items of
experience, even uniform experience, may sound like making them creatures
of subjectivity. Allowing that they are constructs may reinforce that
suspicion, since the act of construction seems a matter of arbitrary will. It is
up to us, is it not, to establish the conventions of language? This fear, I

believe, is unfounded. We are everywhere confronted with constraints on

145

choices, constraints that rarely eliminate all elements of arbitrariness, but
certainly prevent 'anything goes' as a serious description of scientific
activity.

I have interpreted 'objectivity' as intersubjective validity, and that
calls for some comment. The traditional realist notion of objectivity is as
'something that exists, apart from minds, apart from language, and would
exist even if the latter did not.‘ To call laws constructs seems to confuse the
concept with that to which the concept applies, and to confirse the epistemic
with the semantic or themetaphysical. I argued in chapter two that this
notion of objectivity has serious problems with it: it is, if not self-
contradictory, incoherent. To bring the notion down to intersubjectivity
raised the question of what coherence and consensus have to do with truth.

Here is my answer: while experience does not select the one and only
correct representation of the world (because there is no such thing), it does
resist being construed in certain ways. One can adopt a way of talking about
the world and try to live consistently within it, but one may not be able to
get away with it. To paraphrase Rorty: truth is what experience will let you
get away with (a test that may prove much stricter that what your colleagues
let you get away with!). To use James: any belief must run the gauntlet of
experience.

Objectivity is found at the level of the group: if the group tries to
construct the world in a certain way and fails, the group may cease to exist.
Since group survival is at stake, the group has an interest in ensuring
uniformity of practice. This may involve getting everyone to use the word
'quark' the way the experts do, or it may involve getting everyone to use the
word 'rights' or 'democracy' the way they do. The pay-off, in both cases, will

be that to do so makes experience consistent in the sense that things behave

146

 

the way we expect them to. If people do not use the word the same way, or
engage in practices inconsistent with the word or the system to which it
belongs, then we can say the construction failed. 'Democracy' taken one
way, may fail to take hold, or a society trying to maintain it may perish. The

expectations of regular behavior based on the language game employing

'democracy' in that sense (call it 'democracyt') may be frustrated, and we

simply say: 'democracyt' does not work. Similarly, the language game

involving 'phlogiston', the putative agent of combustion in the eighteenth
century majority view of chemistry, failed to maintain itself. This is
objectivity robust enough to explain theory change in terms not entirely
social or psychological, but not realist in the sense of theory- and language-
independent either. While laws still must find some niche within experience,
in experiential terms, the environmental forces that bring selective pressure
on laws are sufficient to unmask wishful thinking and to prevent
gerrymandered regularities from maintaining themselves.
6) Relation to Science

That there are laws in this sense in science, I take to be obvious. No
scientific textbook could be written that did not include mathematical
formulae, models of interesting systems, and general description in
qualitative terms of recurring and therefore predictable behavior. Certainly
in fundamental physics, to pick van F raassen's area for examples, laws are
the presupposition of both quantum mechanics and relativistic mechanics.
Maxwell's Laws were the source of Einstein's insight in relativity, and the
Wien-Stefan law of radiation was the source of Planck's. And the quantum
model of the atom articulated by Bohr would have been inconceivable
without the laws of spectroscopy embodied in the Balmer series, the Lyman

series, and others. And, not to raise too fine a point in this post-Kuhnian age,

147

the notion of a mechanics ingredient in both quantum and relativity theory
assumes the work of Kepler, Galileo and Newton embodied in the laws of
motion and gravitation.

Van F raassen's point about symmetry arguments being sufficient for
the formulation of conservation laws is correct, but beside the point. Not all
laws are conservation laws. Many lower level empirical laws are crucial in
setting our expectations about the behavior of specific systems we
encounter. Even the conservation laws require experience in the construction
of the language in which (possibly some) laws seem to fall out a priori as
symmetries of the problem-situation.

3.5 The Remainder of List Mistaken

A possible response to my claim to have solved the problem van
Fraassen lays out is that I have cheated: some of the central terms in the
conditions have been given non-standard meanings, and the problem I have
'solved' is not a problem anyone has. To make this objection more precise,
let us look at which conditions have been met straightforwardly, and which
required some 'tweaking.' The empirical virtues of confirmation, prediction,
and possibly, universality and relations to science, may possibly be
conceded by the realist opponents. But the logical and metaphysical virtues,
involving necessity, intensionality, inference, explanation, strong objectivity
(that is, complete independence of any conditioning by subjectivity) and
stronger notions of universality (Kant's 'strict universality') are left begging
in my account, unless one accepts the impoverished replacements I give as
'analyses' of the original notion. Although I shall address objections in the
next chapter, this one is too important to leave unanswered in this chapter,
for it strikes at the core of my claim to have solved the problem in any but a

Pickwickian sense.

148

 

First, let me grant the substance of the objection. My analyses of
these central terms do not contain all the content traditionally ascribed to the
notions. 'Necessity' for example, conceived as some compulsion or force
making events turn out as they do, is absent from my notion. My notion
reaches as far as, and no farther than, that of a regularity that supports
counterfactual claims. There is no mysterious 'physical necessity' lying
between logical necessity and cosmic accident that requires an account. We
are simply accounting for our expectation that certain regularities will hold
up in the future and others will not. The case is similar with possibility. We
can characterize possible outcomes as those which are not determined either
way by the set of laws we accept, but may, under some conditions, turn out
one way, and may, under other conditions, turn out the other way. So,
understanding what a law is helps us understand what possibility is.

Now, this answer may not satisfy advocates of the traditional notion
of necessity or possibility, but what of it? Any analysis of central disputed
terms will require some account of why certain parts of traditional usage are
regarded as more important than others and some argument why successful
analysis requires no more than that part. The strongest possible argument for
the latter is that the original notion is incoherent, and the defense of that
claim has already been given in chapter two. Necessity, objectivity and
universality as the realists construe them involve contradiction at the level of
presuppositions. So, if we're inclined to insist that terms be used as the
tradition or traditions have used them, then I insist the terms cannot be so
used. And if the problem involves such a commitment, then I admit neither I
nor anyone else has solved the problem because the problem so conceived is

insoluble, indeed cannot be coherently stated, much less solved.

149

But, if that is so, then this is a problem no one actually has. The real
problem, if I may speak thus, turns out to be one of having a language that
one cannot use because the conditions of correct use do not exist. The
realists have tried to construct the world in a certain way, and have failed. So
the solution of their problem would be to abandon that way of speaking and
look for a form of speech which at least prima facie is free of that problem.
So, if the translation of central terms into the idiom of pragmatism is
disallowed for some items on the list, I claim that those items are mistaken.

Nevertheless, I do not want to leave matters here. First, because I
think there is a real problem of understanding what laws are and how they
are able to fulfill their role as predictors of future behavior of systems we
encounter in experience, a problem that does not disappear with the
disappearance of realism. Second, I believe that these terms have perfectly
good uses that are not marred by contamination with realist assumptions,
and that they are not terms which belong to realism per se. Realists do not
have a monopoly on the use of 'necessity' and 'possibility': these are terms of
ordinary speech. And these uses may figure in an account of how laws can
perform the role the do. Third, I think that it still remains a task of the
defender of laws to show that one can give a coherent account, and it is not
beside the point to show that the central terms in an account can be
construed in a manner which makes the whole account coherent and
consistent. As a matter of fact, it may have turned out that central terms do
not have any uses that together make up a consistent semantic niche for a
term like 'laws'. 'Phlogiston' turned out to be such a term of scientific art:
there simply was no way to construct a coherent account that covered the
work such a notion would have to perform without, for example,

countenancing notions like 'negative weight', work we now believe the

150

 

 

addition of oxygen in some processes, or the release of hydrogen in other
processes, to do in chemical theory. But that is fine. We can make perfectly
good sense of combustion without it. It may have turned out similarly for the
notion of 'laws', but I suggest here that it did not turn out that way. Law-talk
marks a distinction in our treatment of constructed regularities, a distinction
that does useful work in science. It is a distinction that needs to be made
clear, and my account is an attempt to do so. If my account has succeeded,
then it has solved a problem we actually have: coming to terms with how
naturally selected organisms like ourselves, with limited cognitive access to
our environment, can successfully predict the future.

My answer, in brief, has been that we interact with our environment
in ways that establish and systematically try to fail regularities by employing
a repertoire of actions designed to construct defeaters for the regularity, at
just those points defeaters have been found for similar regularities. This is a
good test for a regularity we are willing to project into the future, since we
believe that many of the conditions within our experimental control, if not
connected by a law, vary independently of each other. No expectation of
covariation arises from such conditions of independence. So, systematic
interaction pries loose covariation that is simply accidental because this, in
the normal state of affairs, is precisely how such accidental covariation is
discovered to be accidental. So, the acceptability-conditions of
counterfactuals are constructed through this process of attempted failure of
laws. The failure to fail does not amount to a proof of truth, but proof is not
required. That which proves unproblematic, for the pragmatist, retains the
status of 'innocent until proved guilty.‘ The claim that maintains itself is not

treated as a rascal we haven't unmasked yet: that way lies madness and

151

skeptical regress. Rather, the claim is treated, prima facie and defeasibly, as
a claim on which we can base practical plans of action.

I have not in this chapter addressed all the objections and worries
that may have been created for the reader, nor will I in the next chapter. But
I do hope to answer at least some central objections to my presupposition,
and may thereby prevent a much greater body of objections, which I will not
address directly, from arising.

3.6 Summary

I believe a satisfactory account of laws requires taking the
naturalistic viewpoint seriously, which sees all natural systems maintaining
themselves in an environment, including those natural systems constituted
by our language and minds. The naturalistic viewpoint raises the question:
how can limited cognitive agents with our resources be capable of our
achievements? The answer given by pragmatists to this question is by our
constructive activity subject to our control over the environment of the
constructs. If the analogy with building in general is taken seriously, as I
suggest it should be, then we have the following points of analogy to notice.

1. Buildings do fail, with catastrophic results. Since they are

designed for use by human beings, we have an interest in studying

the ways those buildings fail, and incorporating the results in a

checklist.

2. Buildings do succeed, and prove, by their usefulness, worth the

risks in building them. We record the ways that have worked in a list

as well. The two lists become our building code, the set of standards
we employ to check the worthiness of structures built for human use.

3. Neither success nor failure makes any sense apart from interests,

purposes and actions of agents.

152

4. However, interests, purposes and actions do not guarantee success,
because the environment has a say. Buildings appropriate for one
environment may utterly fail in another environment and some
buildings may be unconstructable dreams of the planner.
Laws, too, are constructs, designed to enable us to form reliable expectations
of the behavior of systems of interest. The behavior must be of interest, but
being of interest is no guarantee we will succeed in capturing it in a formula
that enables reliable prediction. Some attempts to construct experience fail,
while others succeed. We build up cognitive repertoires by noticing the
conditions of success or failure when we can discern them. The task of
science is to systematically construct formulas that enable us to successfully
predict and control the behavior of natural systems, including ourselves.
Laws, so far from playing the subordinate role they have come to have with
respect to theory, actually provide the crucial criteria of practical importance
that theory must meet. Our speculative interests, whatever they may be, are
embedded in an overall practical framework of survival and flourishing.
This, I suggest solves the problem, or at least a real problem of laws.
Maintaining itself in certain environments does not remove a law claim from
any possibility of future revision or replacement. But a coherent fit between
construct and environment is a positive if fallible indicator of reliability, and

reliable predictive capacity is all that we can reasonably ask of laws.

153

Chapter Four: Three Solutions in Search of a Problem

Laws are predictors on everybody's books. That is a given. How
prediction is possible, or even whether it is possible, is a point on which
views sharply divide. And the divide is related to the ways these views
handle the problem of induction. On this question we find a remarkable
concordance between views as disparate as those of Armstrong, Collins and
van Fraassen, which we will analyze in this chapter. Why those three?
Armstrong's inclusion should be obvious. One can safely assume he would
not be persuaded by the arguments I gave against him in chapter two, or the
arguments I gave for my view in chapter three. So here I voice an objection I
believe he might make after having read what I have written so far. Collins
represents the social studies of science view. Since I have identified myself
as a constructivist of some sort, it is important to make clear what kind of
constructivist I am. Collins' rival conception of constructivism, which is
close to mine in many respects, allows me a chance to say where I believe
the differences are between my view and views based on more sociological
models. Van F raassen's inclusion should also be obvious: this thesis is
written as a response to his rejection of scientific laws. We are largely
agreed on the flaws with the realist position. How do we come to disagree? I
will address that by responding to an objection I believe van Fraassen would

make to my account.

154

In this chapter, I will examine three fimdamental objections that
three rival views of laws might raise against my account. The objection from
the realists comes first: laws cannot be representations, but must be what
those representations are about. The objection from the constructivists
comes next: expectations are uniform because they are subject to social
negotiation. Laws are no different in this regard than social rules. The
objection from van Fraassen comes last: expectations may be rational
although lawless. He argues that any attempt to say what laws would add,
epistemically, leads us directly into incoherence. The point I will argue in
this chapter is that laws may be reliable representations without begging any

important questions in semantics, scientific practice or epistemology.

4.1 Laws not Representations but What Representations are About

Armstrong argues that there must be objective grounds for correct
predications and truthful assertions. He lodges these in his objective states of
affairs and relations between them. Laws, if they are anything at all that
science discovers, must be part of the world. He has specific arguments for
why laws must not be identified with regularities. How might he adjust those
arguments (as surely he would) to show that laws are not constructions? He
would, I think, invoke a semantic argument from the nature of language as
representations of the objective world. Laws cannot be statements, but what
those statements are about. If laws were statements, they could not tie
together events as we believe laws do.

This argument would, I think, be supplemented by a separate logical
argument: laws and regularities, even exceptionless ones (and, indirectly,
constructions that formulate such regularities), cannot be identified because

they independently covary. We have exceptionless regularities where there

155

 

is no law, and we have laws where there is no corresponding exceptionless
regularity. This second argument, although important, I think is the less
important of the two. Certainly, if it works, it undermines my position. But
since I believe it is more easily answered than the first argument, I will take
the second argument first. The second argument amounts to a claim that my
view gets the extensions of the term 'law' wrong when I make laws to be
constructed regularities. The first argument amounts to the claim that I
mistake a real relation for a formal one.

4.1.1 The Logical Problem from the Extension of 'Law’

While the first argument marks out an ontological difference
between regularities and laws, the second argument marks out a logical
difference: regularities, even exceptionless ones, are logically distinct from
laws because they have different classes of instances. There can be laws that
do not have regularities as their expressions: probabilistic laws, for example,
do not. There can be regularities that are always true because their
conditions are never met, or only are met a finite number of times while
attended by their consequents. A list of all the regularities that hold up at
every time and every place ("Humean" regularities, in Armstrong's
terminology) would contain types and instances absent from a complete list
of laws. Furthermore, it is possible that the list of laws contains types and
instances missing from the list of Humean regularities. But if so, the
reduction of laws to regularities fails, and one cannot analyze laws as simply
omnitemporally and omnispatially true regularities.

The objection is problematic in that it treats regularities as true in the
past, present and future, assuming that we have the class of instances
including members that do not exist yet. This is to treat regularities

realistically, but to do so begs the very issue under debate. The pragmatist

156

wants to insist that it is important that we occupy a position in time and do
not have the future available to us except in the form of projections by laws.
Next, Armstrong refers to the view he has refuted as the naive regularity
view, and it is appropriate to respond to it by saying I do not hold the naive
version: it is important to my view that we hold not just exceptionless
regularities, but tested regularities.

So, it is appropriate to ask if Armstrong would have some further
objection to my sophistication of the naive regularity theory. He does. His
complaint, specifically directed at Skyrrns, is that laws might not be resilient
in the form I have described resilience in chapter three. Recall that the
conditional probability of the regularity on the testing conditions varies at
most by a wiggle within a particular interval, specified in advance. My gloss
is that we can count on the regularity holding up on reasonable variation of
environmental variables. What does Armstrong mean by "the laws might
not be resilient"? He claims that for probabilistic laws, the true probability is
consistent with instances or runs of arbitrary relative frequency. What is my
response to this? Again, the notion of 'true probability', 1 think begs the
crucial question: how do we know that a law is a probabilistic one? We need
not be committed to any claim about how fast a given mean of a sample of
instances will converge on some limiting frequency, so long as we are
prepared to say under which conditions we would give up the claim that
there is a definite limiting probability involved. And it is important to see
why we should insist on this: the ability to make predications, even with
probabilistic laws, depends crucially on our ability to determine the
probability that the consequent conditions will follow on the antecedent

conditions. If we cannot do that, it makes no sense to talk about a 'law' at all.

157

For my next point, let us look at the counterexamples Armstrong
constructs to show how Humean regularities are not laws, and vice versa,
and show what problems lurk in the counterexamples. Armstrong constructs
Humean regularities that no one would want to call laws because they have
only a single case and lack the repeatability we expect of laws. The recipe
for cooking up such a regularity involves the following steps:

1. Single out the set of properties sufficient to pick out exactly one

object.

2. Pick the other properties of that object, and make them a set.

3. Make the conjunction of the first set the antecedent and the

conjunction of the remaining set the consequent.

One now has a conditional that is general in form (no names or spatial or
temporal locations), is therefore reproducible, but in fact only has had, has,
and will have exactly one instance. These, however, will not be
counterinstances to my tested regularities, because it is built in to my notion
of 'test' that there must be at least two instances (the original and the
replication) and will usually be many more than two. The reason for
insisting on this lies in the purpose or point of formulating laws: regularities,
if that is the right name for them, that are exhausted by their observed
instances (see the discussion of Goodman, in chapter three) are of no use to
us for predicting future events.

What about counterinstances the other way, i.e., laws that are not
Humean regularities? I have already addressed the supposed inability of the
regularity view to handle probabilistic laws. What about the more general
problem, say of laws that turn out to be local laws and not laws that hold in
all times and places? Tooley's example, of a garden that grows only apples,

no matter what is planted, but contains no special properties that set it apart

158

from other gardens does support a generalization, 'everything planted in
garden x grows apples', but it lacks the proper spatial and temporal
generality (because of the definite description 'garden x') to be a Humean
regularity. My response would be, it is not a law either. It is built in to our
notion of replicability that it should not matter where or when the
experiment is performed, provided nothing of relevance to the results is
changed. The same experiment, performed under the same conditions should
give the same results. If it does not, it is not projectible. Now, perhaps, such
a strange phenomenon as 'apples always grow in garden x' can be treated as
a degenerate case, since it has temporal generality and spatial generality of a
limited sort: anywhere in garden x will work, still, it is a funny sort of law
and certainly not a paradigm case that should set the criteria for lawhood.
Regularities whose scope is less than all time and all space may still be
useful, so long as they are universal within the scope of their applicability .
But, without such universality, they would be useless as inference tickets
and powerless to predict.

This brings me to the larger issue. If the role of laws is to guide
actions with respect to effectiveness, then what is the problem with even the
untested Humean regularities, provided they are truly exceptionless?
Waiving aside the legitimate complaint we might make that in acting on a
regularity we are testing it after a fashion, even an untested regularity will
work if the consequent always follows on the antecedent, we desire the
consequent and can bring about the antecedent. Everything we could want
for effective action is present. The reason we insist on testing is because
untested regularities have often proved unreliable in the past. We have
learned how they failed and have some idea how to make them fail in

advance--before we committed life and resources to their use. But the

159

problem is not in the regularity as such, but in the conditions of its
observance. We have sampling errors, observational errors, error of
inference, and errors of formulation that we need to check before the
putative law is acceptable. We believe that there is a regularity if we can
produce instances of it. We believe it is an exceptionless regularity if we
cannot produce counterinstances of it. But if we were just given an
exceptionless regularity (never mind how) and knew it was exceptionless
(again, never mind how) -- where is the problem that needs to be solved?
The gap that Armstrong and the realists want to open between laws and
Humean regularities strikes the pragmatist as unreal. If the grounds for such
a distinction are the intuitions of the realists, the pragmatist would like to
suggest those intuitions are mistaken. I, at least, have no such intuitions. The
distinction I make between accidental generalities and laws is one that is
made entirely within experience and makes them disjoint subsets that
partition the set of experiential regularities. We get the extension of law
right by revising the laws we accept in the light of experience, both in
common life and the laboratory. But at the end of the day, we are left with
regularities, tested regularities it is true, but regularities nevertheless. The

pragmatist does not see a logical problem here that needs to be solved.

4.1.2 Objection from the Conceptual Role of Laws

The serious worry is that law statements do not do the same
conceptual work that laws do. Here we get at something very basic to the
realist view. Laws are real relations. They are part of the furniture of the
universe. They cannot be bits of language because language does not make
event B follow on event A in causal laws, and causal laws are the

fundamental laws. Law statements, if they are about anything, are about

160

 

laws, which makes law statements true if they are true. Or so claims
Armstrong.

This is part of his claim to have solved the problem of induction by
the appeal to realism. It is part of what he means when he claims that the
regularity view does not get the intensions of laws right, because such views
miss the inner connection between the items connected. How does his
solution to the problem of induction work?45 The argument is that:

1. One cannot be rational in one's expectations unless one can give a

reason for them.

2. A reason would be the same as an explanation of the events.

3. Regularities do not explain events, since they are simply sets of

events, but laws do explain events.

4. We don't know laws by induction, on pain of circularity (i.e., a

solution to the problem of induction may not assume induction in its

proof).

5. We don't know laws by deduction from evidence, since they do

not follow without other premises from that evidence. Further

premises would have the same problem ('how do we know such
general claims?') as laws, leading to regresses.

6. Therefore, if we know laws, we know them non-inferentially.

7. The only non-inferential knowledge we have is perceptual or

analytic.

8. Knowledge of laws is not perceptual.

 

45 I am relying on the analysis van Fraassen gives of Armstrong's argument in van Fraassen,
Laws 138-142 in the section "Armstrong's Justification of Induction." I have run together
what he calls 'two sub-arguments, and have taken some liberties on how I parse out the
premises, but the basic moves here were identified by van F raassen. For the passages in
Armstrong where these arguments occur, see Armstrong, What 54-59.

161

9. Whatever better explains constitutes a good reason for believing

something, so belief in the best explanation is rational by definition.

That is, it is analytic, therefore necessary and a priori.(The 'inference

to the best explanation' rule)

10. Laws better explain events than any rival hypothesis.

11. Therefore, we are rational to believe in laws without needing

induction for their justification.

12. But if we have laws, then the problem is solved: we have good

reason to expect some regularities to hold in the future as they have

in the past, although which ones those are is still an empirical matter.
Van F raassen believes the hinge of the argument for laws and for induction
is the 'inference to best explanation' rule, my premise nine. Accordingly, he
attacks the argument there. Since I am only concerned with the objection
this argument raises for my view of laws, I do not need to go into a detailed
criticism of the strategy, although I think I largely agree with what van
Fraassen says against the strategy.46

The fundamental objection to the regularity view coming out of the
realist argument for laws goes as follows: it is not that the regularity view
counts too many things as laws that intuitively are not laws, but that
regularities are the wrong sort of thing. We have supporting evidence in the
fact that, while laws are explanations, regularities are not, and, a fortiori,
expressions of regularities are even further removed: bits of language at best
express explanations.

What lies behind this realist objection? I think the following

expresses the realist intuition here. One must distinguish the manifestation

 

46 See the discussion of van Fraassen's criticism of the argument below.

162

..
l A- _‘A

of something from the thing itself. Language points to realities and should
not be mistaken for that to which it points. Regularities manifest laws, but
are distinct. Linguistic expressions may manifest laws, but they also are
distinct. Laws explain the regularities and give the truth-conditions for law
statements. The explanation arrow does not go from regularities to laws. The
regularity theorist errs in identifying laws with regularities, because the
regularity just is the set of events we would count as manifesting laws. But
those events cannot explain themselves. The connections between one set of
events cannot explain themselves, still less can those connections explain the
connections between other sets of events. Harvest may have always followed
planting, but that fact does not explain itself, and it certainly does not
explain any connection between future plantings and harvests. Something
different is needed. In this example, we need an explanation drawn from the
laws of biology to explain the connection between planting and harvest. The
regularity theorist seems to think we can argue from known events to
unknown events, a non-sequitur, or argue from the set of known plus
unknown events to unknown events, which is either trivially true, or
equivalent to the previously mentioned non-sequitur. In either event,
regularities cannot play the role of explanation, so they are not the right sort
of thing to be counted as laws. A fortiori, whatever logical connections there
may be between parts of language, that cannot be what anyone means by
'necessary connection' in this context. To suggest, as some have, that the
only necessity is verbal or logical necessity is to attribute to philosophers an
inability to distinguish between a sign and what it stands for.

I'm not inclined to quarrel, as van Fraassen does, with the premise
that rational expectations involve being able to give reasons or explanations

for why the person expects the past pattern to be repeated in the future. I

163

think some suitably weakened form of that premise is correct. We may not
know the reason why, or may not be aware that we are projecting a .
regularity, but when aware and challenged, we will usually give a reason
that includes or implies some statement of law as an explanation for why we
expect the regularity to hold up. My point of disagreement is that regularities
are not explanatory.“7 Class membership may provide important clues about
properties all members of the class share, and it may license inferences
based on analogy even if nothing like Armstrong's universals exist.
Analogical reasoning does not involve appeal to truth-makers because it
does not presuppose some set of properties all members of the set share.
Rather, it involves learning which properties members of the set share and
which they do not. The members are not viewed as firlly determinate
metaphysical realities, but as epistemically fuzzy objects whose nature is
being determined during a process of inquiry. The determination is carried
out in some language that may carve up the terrain differently than other
languages we might have used. We do not simply list the members and read
off their properties, because neither the set nor the properties are givens, but
must be constructed.

Our way of constructing the classes involves sorting out relevant
from irrelevant properties by manipulating conditions. We learn about how
these objects work simultaneously with our classifying them. The
regularities we wind up with are not the regularities we started out with: they
have become more determinate, more precise, and have already been
challenged to fail in various ways. For example, we learn a lot about nuclei

and nuclear processes while we sort out which materials are radioactive and

 

‘7 Van Fraassen has a very sharp challenge to my claim in his criticism of induction, which
I will need to address when we come to it below.

164

 

which not, and the concept of radioactivity is adjusted and modified while
the list of members is determined. Most concepts in physical science display
this open-texture as H. Putnam has noticed.

The problem is generated by Armstrong's atomism. He believes the
world is the facts, that is, states of affairs that are logically independent of
each other. So for him, an extensionalist would have to view laws as sets of
ordered pairs of events, which allows him to ask "what do these ordered
pairs have to do with each other if they do not share universals?" Now, for
me, individuation is relative to language. Are laws about events or about
processes? Or neither? There is no canonical minimum vocabulary in which
we have to state what laws govern. That is important. How we describe the
constituents of laws is part of our overall enterprise of deciding how to
describe the world. That we have some language that treats laws as
something other than a construct and maintains there is separate work for
them to do should be regarded as a historical comment on the languages we
have developed in the past. But before we accord them normative force, we
need to ask what we can do with such a distinction? The problem I have
continually come up against in trying to understand the realists is that we
have no use for unformulated laws. All our explanations occur in language.
Where else could they occur? We are trying to make sense of experiential
patterns.

4.1.3 Response: Law-statements can be about the world without being about
laws

The upshot of my response to the realists is this: law-statements can
be about the world without being about laws. Law-statements pick out
experiential episodes in a particular way that constitutes them both as

regularities and as laws--but this constitution is relative to a language. There

165

 

simply is no way around this relativity. But for all that, constructions can
and do guide expectations, just as constructed bridges and roads guide our
actions. Just as with bridges and roads, we do not invent the material of the
constructs ex nihilo: we are constructing experience, and find ourselves up
against the resistance of the world in trying to construct the world one way
rather than another. Still, we need to overcome the tendency to treat
successful constructs as 'having been there all along, waiting to be revealed',
because the constructional activity could have taken a different direction.
Roads can guide without revealing some deep world path that would have

been there even if humans and their designs had not existed.

4.2 Laws as Socially Negotiated Rules

Social constructivists in the social studies of science and technology
typically do not talk about laws, and when they do, they almost invariably
mean laws as part of some legal system. This lexical fact is interesting, and
deserves study. But here, I want to notice that although the word 'law' may
be absent, there are discussions in the social constructivist literature that are
relevant to the view I have put forth. Since H.M. Collins has put out an
entire book of essays where he explicitly and self-consciously addresses the
issues of replication and induction in scientific practice, it will be instructive
to see where we agree and where we part ways. We are both social
constructivists of some sort. The question is, where might we disagree in the
account I have given of laws, and how do I respond on the point of
disagreement? I will argue in this section that we disagree over the degree to
which constructive closure is a matter of social negotiation and that we
disagree over the extent to which the advance of science requires the

replacement of the tacit and inexplicit appeals to skill and group

166

membership by standardized and mechanically applicable procedures that
can be explicitly set forth in algorithms.
4.2.1 Replication as Social Achievement

After admitting that perception and the stability of perception
amount to the same thing, and claiming that order requires our ability to re-
identify something as the same thing we perceived before, Collins goes on to
ask where our expectation of perceptual stability and order comes from in
science. This is the problem of induction, but since the focus is on predictive
success, it lends itself to a discussion of scientific laws.48 This discussion of
Collins has direct bearing on our discussion of laws, because what he calls
the problem of order just is what I have called the problem of laws. After
looking at Wittgenstein's and Goodman's work, he settles on entrenchment
of central terms in a network of terms as giving the taken-for-granted
background for such agreement, although he claims to have gotten this
insight from M. Hesse rather than Quine, and he wants the coherence lodged
in social institutions, not probabilistic relations between beliefs. Next, he
comes to puzzle over how such a background of agreed upon meanings ever
gets challenged and changed (as it surely does in science) and again arrives
at the same conclusion I arrived at: "Press scientists and in the last resort
they will defend the validity of their claims by reference to the repeatability
of their observations or the replicability of their experiments...Repeatability,
or replicability (I will use the terms interchangeably), is the touchstone of
common sense philosophy of science." (Collins 19)

With such substantial agreement, the reader may well expect the

differences between Collins and me to be trivial. However, it is at just this

 

‘8 Since in the context, Collins discusses Hume's view of cause and laws, this interpretation
of Collins is one I believe he would accept.

167

point that Collins makes a critical break: "Replicability, the vanguard of
common sense theories of science turns out to be as much of a philosophical
and sociological puzzle as the problem of induction rather than a simple
straightforward test of certain knowledge. It is crucial to separate the simple
idea of replicability from the complexities of its practical accomplishment."
For "[o]nly in exceptional circumstances is there any reward to be gained
from repeating another's work. Science reserves its highest honors for those
who do things first, and a confirmation of another's work merely confirms
that the other is prizeworthy. A confirmation, if it is to be worth anything in
its own right, must be done in an elegant new way or in a manner that will
noticeably advance the state of the art." Collins discusses in this book the
difficulties associated with replicating advanced design lasers from one
laboratory to the next, with detecting gravity waves, and with detecting
paranormal events.

Collins uses these cases studies to establish the following five
propositions, which constitute the core of his 'empirical programme of

relativism' [EPOR]:
Proposition One: Transfer of skill-like knowledge is capricious.

Proposition Two: Skill-like knowledge travels best (or only) through
accomplished practitioners.

Proposition Three: Experimental ability has the character of a skill
that can be acquiredand developed with practice. Like a skill, it
cannot be fully explicated or absolutely established.

Proposition Four: Experimental ability is invisible in its passage and

in those who possess it.

Proposition Five: Proper working of the apparatus, parts of the
apparatus and the experimenter are defined by the ability to take part

168

in producing the proper experimental outcome. Other indicators
cannot be found.

Proposition Six: Scientists and others tend to believe in the
responsiveness of nature to manipulations directed by sets of
algorithm-like instructions. This gives the impression that carrying
out experiments is, literally, a formality. This belief, though it may
occasionally be suspended at times of difficulty, re-crystallizes
catastrophically upon the successful completion of an experiment.
(Collins 73,74,76)

In the postscript to his book, Collins sharply distinguishes two
models of science, the algorithmic view and the enculturation view. The
algorithmic view has science driven by a set of explicit, formal procedural
rules. The enculturation view has science driven by the acquisition and
transmission of skills among qualified experts. In the subsection,
"understanding the deconstruction of facts," Collins maintains: "the
perception of certainty is a matter of distance from the scene of the
crystallization, both in time and 'social space.‘ Certainty increases because
the details of the social processes that went into the creation of certainty
become invisible....Thenceforward certainty is maintained by continued
representations of the data in the style of facts.... The mere act of describing
an experiment as a piece of ordinary life reduces its power to convince"
(162).

The central idea of the enculturation model is that the scientific
taken-for-granted beliefs begin as controversial, and come to be accepted
only after the debate is closed. But the debate gets closed only through a
series of decisions about which practitioners are expert, which experiments
were correctly done, and which experiments count as replications. There are

no algorithms that could be executed by some machine to sort out the

169

scientists from non-scientists, skilled practitioners from unskilled, successful
experiments from unsuccessful, and significant results from insignificant
ones.49 Drawing heavily on the notion of 'tacit knowledge' from Polanyi, and
the notion of 'paradigm' from Kuhn, Collins makes the construction of
scientific fact turn on the authority of the scientific community.

Now, up to a point, there is little to disagree with here. The scientific
community identifies its skilled members and is responsible for what Ravetz
has called 'quality control' on research. The standards of proper techniques
are constructed and enforced by the community. The community establishes
when an experiment has been replicated or not. It also decides when the
weight of evidence is sufficient to close a particular debate. However, what
is missing here is what I have been calling 'the resistance of the world to
being constructed in certain ways.’ I conceive of this resistance as negative,
because it does not convey positive information. The world does not
suggest particular forms to us that correspond to its joints. I agree with
Collins that the realists are wrong when they insist it is our job to discover
pre-existing natural classes that are part of the furniture of the world.
Nevertheless, there are constraints on our constructions that are not simply
social. A team of Nobel laureates could bring it about that the scientific
community universally agreed on a particular set of constructs, and that set
of constructs could fail nevertheless.

This is the point I made in chapter three: consensus is not enough.
Many scholars working in the social studies of science (Collins is not alone
here) seem to believe that all agreement is achieved by negotiation between

social actors. For example, Gilbert and Mulkay, writing from the perspective

 

49 See"The Idea of Replication" in Collins 29-49; esp. 39-46.

170

of discourse analysis also argue that scientific facts are constructed by
consensus , or by the illhsjen of consensus, since agreement may be
achievable only by ignoring real differences of interpretation, community
membership, or ascribed competence (Gilbert and Mulkay 112-140). Some
scholars are willing to allow natural objects some say in how the
negotiations go,50 but others are deeply troubled by this metaphor.“ The
point here, again, is that the function of laws is to tell us how systems will
behave in the future, and this cannot be an entirely social matter. Even after
we grant the importance of language in shaping our perceptions in everyday
life and in science, not all conceptual frameworks are created equal, and
consequently some groups find their way of life under pressure from events
that fail to turn out as expected.

Healings may not heal, agricultural practices may produce deserts,
methods of diplomacy may lead to ruinous wars, and methods of distribution
of goods may lead to shortages and starvation. Some ways of doing things
simply do not work. I can say this without being ethnocentric. Our own
culture is certainly not immune to the charge of having institutions that do
not work the way they are supposed to work! Nor is it immediately clear,
when a problem arises, what the solution will be. None of the available
alternatives may be any better, or if better, practicable in that particular
context. The problem may be hard to diagnose: it may be not an isolated

practice that is at fault, but its combination with other practices. The Quine-

 

” See Latour and Callon's actor-network theory and their somewhat broad notion of an
'actant' in Latour, Seience and Callon, "Four Models of Dynamics in Science" in Jassanoff,
et.al.. 29-63 .

5' See Collins and Yearly, "Epistemological Chicken" in Pickering (ed.), Sejehee 301-326;
Latour's reply "Don't Throw out the Baby with the Bath School"; and Collins and Yearly
response "Journey Through Space" follow. Especially important here is the section
"Networks Revisited: Ants and Ors" where Collins and Yearly discuss their reservations
about actANT S rather than actORS. Pickering, Seienee 372-376.

171

Duhem problem has its counterpart in practice, and so it should if practice
and belief are hooked together as I have claimed in chapter three.

Now let us say Collins were inclined to grant that my point applied
to technology and engineering but not to concepts. It is unlikely that he
would, since his examples of replication failure include examples of
constructing devices that simply do not work (even though theory says they
should), like the TEA laser Harrison was trying to duplicate. And since the
Bath school of social studies in science seems committed to focusing on
agency and practice, and the contingency of scientific choices in general, I
find it unattractive to suggest a difference between theory and practice at this
point. Nevertheless, let us create an opponent, called pseudo-Collins, who
would want to make the following move: we have complete control over the
descriptions we give of an activity and its results, so we can make the world
safe for our versions by deploying appropriate rhetorical devices. If I am
reading them correctly, Gilbert and Mulkay are making moves of precisely
this sort. Scientists as social actors deploy the empiricist repertoire to
explain how they are just following the observational facts and the
contingent repertoire to explain how their opponents, though bright and
hardworking, through social and psychological distorting influences got
things wrong. They also can, when necessary, use the rhetorical device of
'the truth will out' to explain why the scientific community still hasn't backed
their theoretical explanation of the experimental events. In our opponent's
move, we can, following Quine and Duhem, make whatever adjustments we
need in the rest of our beliefs in order to save our version of the events.

This version misses something important: the empirical character of
science. What we want is more than having our views accepted by others.

We also want our beliefs to track experience and register appropriate

172

sensitivity to changes in our environment. The realists are not wrong to
suggest that empirical belief involves relation to the world, and this is a
point that is missed when all agreement comes down to consensus. Once
again, such a view must, in its nature, be entirely retrospective: we cannot
explain away recalcitrant data until we have already been disappointed in
our expectations. The trick, of course, is to get it right in advance of the data,
to predict what is going to be the case. And for this, we need to stop the
negotiation process and make some claim that risks being wrong. On this
matter, I side with Popper: if we never risk being wrong, to keep our options
in play and avoid any firm commitments, we simply are not doing science.
To be sure, this does not mean we never revisit our commitments or revise
our beliefs. This too occurs. But if we never expose our beliefs to
falsification, we never really know how solid our views are, and it becomes
treacherous to base any practical plans on such untested beliefs. I am
sympathetic to the claims of those in the social studies of science that we do
not simply read off of nature (or of our tables of data) how nature is
organized, and I agree that the actual story is more complicated than that.
Still, if science is not sensitive to experience in the appropriate way, it
simply becomes an elaborate game or ritual and not the source of reliable
belief I maintain it is. And if there are some in the social studies of science
who believe science is just an elaborate game, then I maintain they are
missing something basic in their views, that is, how science can fund
technology that works.
4. 2.2 First Response: Algorithms are Methodologically Useful

I would like to give a modest rehabilitation of the algorithmic view
Collins opposes. Collins opposes the algorithmic view of scientific decision-

making by appeal to L. Wittgenstein, M. Polanyi, and T. Kuhn with their

173

respective notions of 'following a rule', 'tacit knowledge' and 'paradigm'. The
role of skill in reproducing results and the difficulty in identifying skilled
members apart from their ability to produce the same results as other
recognizably skilled practitioners are seen as undermining the older
forrnalist views of a logic of science held by the logical empiricists R.
Camap and K. Popper.52 I, too, oppose this view. Scientific judgment is not
algorithmic, but involves multiple criteria, vague terms and plenty of scope
for disagreement on the weighting and application of criteria in scientific
conflict. The case studies found in the social studies of science literature
provide abundant evidence of the correctness of Kuhn's case against the
algorithmic view, so I agree with Collins because I agree with Kuhn here.
Nevertheless, there is something important that both Kuhn and his followers
are missing and here I must part company with them. The formation of
explicit procedures executable in the limit by a computer or a machine is an
important part of the progress of science towards greater objectivity and
publicity.

The reason for replacing qualitative descriptions with comparative
and quantitative ones is that the latter are more informative: there are more
ways such claims can turn out false, so they provide more information about
the world. Again, this is a point raised by Popper. Why do we want claims
that do not give much latitude for interpretation? Because to the extent that
we eliminate possible interpretations, we narrow the range of possible
disagreement when different persons agree. Most of us agree there are a lot

of trees on campus at Michigan State University. Fewer of us agree on the

 

52 Popper's claim to having always been an opponent of the logical positivists should be
taken seriously, but usually is not. See Hacking, Representing 43-44, for an explanation why
Popper is commonly lumped together with the logical positivists.

174

exact number. Since nothing practical hinges on that, we are content to leave
matters at the level of vague agreement. But what if the matter were one of
medical diagnosis, and different diagnoses led to quite different
incompatible therapies? Both doctors and patient have a stake in determining
the preferred description of the disease. In the interest in successful therapy,
neither can afford to leave matters vague: both will want the most
informative description justified by the evidence.

That the scientific community is looking for algorithms, even
imperfect ones, is shown by the replacements of human observers by
electronic detectors, human recorders by computers, and human hands by
robots. The rationale is a good one: we do not want the outcome to turn on
idiosyncrasies of individual scientists. We want the experimental procedure
to be performed the exact same way wherever possible. To be sure, this is an
ideal even when we talk of robots. Lubrication changes as a function of time
over a given run, signal strength may vary with random noise in the circuit,
replacement parts may not be identical, etc. Still, we are talking here as
always, about controlling variables that affect the outcome of experiments,
and with that, the reliability of results. The formulation of laws turns on the
ability to control conditions under which the uniformity holds. This means
that the trend towards explicitness is a trend that favors the formulation of
laws.

This leads to another related point. Insofar as we can replace
judgment with explicit procedure, we replace difficult to control variables by
ones we can better control. Insofar as we depend on the community of
science to identify skilled performers and skilled performances, we are
taking matters on trust. Now, since science is a human endeavor worked out

under conditions of ignorance, we cannot eliminate the need for trust

175

altogether. But it is important to see that the judgments within this system of
credit constitute promissory notes that may be difficult to fill. We trust the
competence of well-trained scientists from good institutions and we trust the
output of instruments constructed by reputable firms. But it would be better
to actually watch the scientist do the experiment, and it would be better to
have comparative studies of the performance characteristics of various
instruments under a variety of conditions. To some extent, the game really is
what D. Gooding suggested: to embody skill in the construction of a device
that will behave the same way for everyone. However, this requires us to be
more explicit about the conditions involved in the procedures, and this, as I
said above, involves specifying the laws relevant to the operations and their
outcomes.

Insofar as social studies of science unmask consensus as illusory, it
points out a problem that needs to be solved by the scientific community.
Intersubjective validity, if it is going to cut across group membership and
give non-scientists results they can use, requires reflection on the methods
used in an attempt to be as explicit as possible about what needs to be done
and how. This reflection is necessary to minimize differences due to
idiosyncrasy and incompetence. Nonstandard views and practices may
generate fruitful new ideas, but can also spell disaster for common
enterprises that require real agreement among the actors.

4. 2.3 Second Response: Consensus is prior to Dissensus.

I believe that the focus on scientific controversy has led some (like
Gilbert and Mulkay, T. Pinch) to exaggerate how much dissensus there
really is and to miss the deep agreements that link even those scientists who
disagree with each other. Not only must scientists agree widely about the

nature of the discipline and its procedures, but certain widely shared

176

agreements between scientists and non-scientists form the background for
the very formulation of controversial issues within the discipline. To put this
in the language of laws: if we did not have shared expectations of how
scientists and scientific devices behave, we would not be in a position to
notice when they don't behave the way we expected them to, i.e., when we
have views that disagree or data that does not fit our theories. In this
respect, I think they have missed something basic to Kuhn: the possibility,
even under conditions of normal science, of anomalies. The more explicit
we are in the formulation of laws and procedures, the more we are in control
of outcomes, because becoming explicit involves sorting out relevant from
irrelevant conditions. The problem of order turns out to include the scientist,
who is able to make his or her behavior more controlled as he or she makes
his or her procedures more explicit.

4 .2.4 Third Response: Science is constructed at every level without being
constructions all the way down .’

In sum, my response to constructivists who reject any independent
control by the world over belief is that science is constructed at every level,
but is not constructs all the way down. That is, there is no isolable aspect of
science that is free from constructional activity, as I argued in chapter two.
There is no 'given' content that can somehow be freed from all human dress,
so as to be viewed naked and unadomed. Every fact is a constructed fact.
Still, not every construct works. All are of the nature of hypotheses
tentatively floated, tested, and tried out in various environments. Since we
work out the details of empirical processes under conditions of imperfect
information, we may find ourselves under pressure to revise constructs when
they fail to track those processes. Firm belief does not eliminate anomalies.

It is only with firm belief that anomalies are possible.

177

The lesson for laws should be clear. Laws cannot be social rules like
which side of the road we drive upon, because the problems we face in the
world are not simple coordination problems (to steal a phrase from game
theory used with good effect by D. Lewis in Cenyentien). We are not the
only ones who police our constructs. The world does as well. At the very
least, the world holds veto power. However difficult it is to explain the point
without lapsing into metaphor, the point remains important: there must be
more involved in the formulation of laws than passing the inspection of
journal referees. Truth cannot simply be whatever our colleagues will let us
get away with. It must also be at least partially a matter of what the world

will let us get away with.

4.3 Laws as Doppelganger for Subjective Probabilities

Van F raassen and I agree at many points about the realists's
weaknesses. The question is, why does he give laws over to the realists and
say good riddance? Why does he believe laws dispensable? The answer will
turn out to be very complicated, and, like the others we have discussed, turns
on his views of induction and semantics.

I will start with his conclusion, which turns out to have two parts,
only one of which I have addressed so far. I will then work through his
arguments for the unanswered part of his conclusion, because those
arguments contain an important objection to my view. His conclusion is that
laws are impossible and superfluous. They are impossible because nothing
could have all the properties ascribed to laws by philoSOphers. I have already
addressed and answered that part of van Fraassen's conclusion in chapter
three. The pragmatist can supply items that will meet the appropriate

philosophical demands. The part of his conclusion I have not addressed is

178

the second part. Laws are superfluous because they duplicate the work
already done by subjective probabilities and arguments from the symmetries
of theoretical models. But like the doppelgéinger of literature, this is not a
twin, but an evil one that tempts us into metaphysics and incoherence. Laws
are not simply useless; they are impediments to clear understanding of good
empirical scientific practice. It is irrational to assign an event any higher
probability than a strict Bayesian would, but laws tempt us to do just that. If
van Fraassen is right, then I have a choice between being a rational
subjectivist or an incoherent objectivist. The pragmatist should repent of
fence-sitting, give up Truth and Laws as a realist dreams, and follow the
honest path of experience and reason.

How does van Fraassen argue for the claim that laws are irrational
and superfluous? In brief, he regards induction as impossible (and
unnecessary) and believes the meaning of scientific language is to be found
in language-independent models. Combined, these lead to a rejection of
laws: we cannot know what the future will be like even though we can (and
must!) talk as though we do. Those who claim we have knowledge of the
future based on laws are mistaken. The best we can do is project past
relative frequencies into the future. Van F raassen arrives at these views in
articulating a semantic theory of theories and a perrnissibility view of
rationality. Combined, they amount to the claim that rational beliefs are not
knowledge but are merely probable on the evidence. But this, it will turn out,
is simply a disguised version of the realist independence assumptions that I
criticized in chapter two. Because laws are not part of language, much less
parts of minds or mental states, they can play no role in guiding belief-

forrnation or actions. That's the claim in a nutshell.

179

The objection arises that any role playable by laws is already
performed by conditionalizing beliefs on evidence. If we attempt to do better
than the conditional probability by giving laws weight in the probabilities
we assign to events, we actually do worse because we violate the axioms of
probability and embrace incoherence. However, so long as we are consistent
in our assignments, we can assign whatever probabilities we like to events,
and then adjust our beliefs in the light of how well they predict what
happens. Van Fraassen is arguing against the view that without laws we can
have no rational expectations for the future. Since this is my belief, he is
arguing against my view. So it is important for us to look at the arguments
he gives to see if they directly address the claims I have made, or can be
modified to do so.

4.3.1 Van Fraassen’s Dual Strategy

In chapter 6 of Lamnifiymmehjes, van Fraassen argues against
views of both logical empiricists and realists (131-150). The logical
empiricists claim that the data are enough to license a move from particular
samples to general conclusions. The realists claim that although the data are
not enough, without laws, to yield universal conclusions, we have laws on
the basis of inference to the best explanation. Logical empiricists try to solve
the problem of induction by the construction of definite confirmation
functions that act as rules telling us which generalizations to accept. The
realists claim we employ abduction, rather than induction to get laws and
that laws constrain beliefs about the future. Van Fraassen has criticisms of
both.

The logical empiricists use an internal approach: the data themselves,
without external constraint, point to the conclusion. The realists use an

external approach: laws, existing separately from us, language and minds,

180

can control our beliefs (on one interpretation on the realist view)53 and set
our expectations so that we can draw particular conclusions (i.e., make
predictions) from sets of data. Both critiques given by van F raassen contain
potential objections to my view, even though I do not advocate an inductive
logic or inference to the best explanation. So after I give some of the details
of the argument van Fraassen gives, 1 will need to tease out particular
objections I think he might raise to my account of laws. I believe that for van
Fraassen to make a solid objection to pragmatism, he must share one or
more of the fundamental assumptions of the realists, but if he does so, he
becomes heir to all the headaches those assumptions bring.
4.3.2 The Problem with Induction

In his critique of induction (132-134), van Fraassen criticizes efforts
to construct a formal inductive logic with rules setting specific confidence
levels for scientific propositions based on the amount of support they receive
from observational propositions. He claims that any choice we make for the
rule will turn out either arbitrary or uninformative. He singles out the
straight-line rule as an example. The rule states that we should take the
relative frequency of an event as the probability of its future occurrence. If
the rule were good, it should hold in all situations. However, suppose we
apply it to the proposition 'this coin will turn up heads next time I toss it'
after we have already tossed it once with it coming up heads. By the
straight-line rule, we have a relative frequency of 1 (since one out of one
tosses turned up heads). Should we therefore assign the probability of l to

the proposition that the coin will turn up heads on toss number two? Of

 

53 Van Fraassen considers this causalist view as one of three possible interpretations to be
assigned to Dretske's claim about the remarkable confirmation of laws, but goes on to deny
that this is the interpretation Dretske had in mind.

181

course not, because we have experience of other coin tosses that is relevant.
If the coin is fair, it probably will not turn out only heads for, say, the next
ten tosses. But using the straight-line rule, we have no reason to expect that
our success in getting heads will be broken any time in the future. To avoid
this problem, we could go to a second order rule that judges the confidence
we attach to our first order claim about probabilities. But the same problem
occurs there. The alternative strategy is to avoid setting the probability
exactly, and say that as the sample size increases, the probability of the
relative frequency matching the true probability approaches one. But this
makes the rule more sensible at the expense of its informativeness. If, for the
sake of definiteness, I try to increase precision by specifying narrower and
narrower confidence levels, I gain inforrnativeness, but at the expense of
rational motivation. I cannot justify the choice of a confidence interval by
appeal to scientific practice, because there is no universally accepted
confidence interval that holds for all disciplines. So, no such rule using only
the data can guide our selection of probabilities to assign to propositions.

Do I get caught in this dilemma? The problematic assumption is that
there are objective chances to be discovered. But events do not have
probabilities associated with them apart from the way we describe them,
classify them or measure them.

Probability is relative to a practice. Within the practice we have a set
of distinctions that enable us to distinguish initial and final conditions, and
we also have a set of manipulations that allow us to bring about the
connection between the initial and final conditions expressed in a law. It
makes sense that probability is sensitive to the construction of a reference
class--even a realist like Salmon recognizes that--but the construction of a

reference class itself is relative to the procedures for determining its

182

membership. There are no objective reference classes out there waiting for
us to discover them.

The relativity of probability to a set of procedures can be seen even
in the simple example of tossing a coin. It makes no sense to describe the
'probability of a coin coming up heads' without specifying the procedure for
coin tossing, because, quantum effects aside, the ballistic behavior of a
tossed coin is deterministic and can be made as close to certainty as we like
with the appropriate control of conditions. Series of coin tossings based on
different procedures can converge on very different probabilities, and the
straight-line rule would yield different probabilities under these different
conditions. Furthermore, if we give up the assumption of determinism
(under which all events have a probability of one or zero), and assume that
natural processes and events are irreducibly stochastic, then probability
evolves over time, and the probability of the coin coming up heads may be
very different depending on whether we ask the question a minute before the
toss or a nanosecond before.

The matter of probability is tied to the notion of replicability. Any
stable probability we would assign to an event would require we can identify
what would be a replication of the experiment. As social constructivists
rightly point out, there is a large social element in the criteria of identity or
relevant similarity, to weaken the notion a bit for experiments that cover a
different range of conditions. Some questions, which I paraphrase from
Collins, must be addressed:

1) Did the same experiment contain the replication in successive
runs?

2) Was different apparatus used?

3) Was the experiment performed by equally competent researchers?

183

4) Was the experiment performed by friend or foe?

5) Was the experiment performed in a related or dubious field?

6) Was the data 'close’ to the expected curve? (Collins 39-44)

All of these criteria need to be negotiated, as the social constructivists
correctly claim. We weigh the trade-offs, but can only be vindicated in
retrospect. If the beliefs work, we made the right choices. If not, back to the
drawing board!

Now, we could stop here as far as the pragmatist is concerned. The
question I have is: would van Fraassen be content to let matters rest here? If
so, enough has been said: his objection was not directed against me and has
been sufficiently answered. If not, however, then we need to go on. What
more would he require? The question has some bite, because van Fraassen
does talk about the difference between acceptance and belief: we may accept
claims as empirically adequate that we do not believe to be true. This
distinction between truth and experience enables van Fraassen to enter into
dialogue with the realists, who in their independence assumptions mark
similar distinctions between reality and mind, between reality and linguistic
claims, and between truth and meaning. So the questions that need to be
addressed are, does van Fraassen allow 'objective chance' an existence
independent of 'subjective probability', and does he believe that it makes
sense to claim objective probability exists even if its existence cannot be
experientially verified? If so, his permissive view of rationality is too
permissive, and at war with his empiricism. I will give a fuller exposition of
these points in 4.3.4 and my response to them in 4.3.5.

4. 3.3 The Problem with Inference to the Best Explanation
Van Fraassen's criticism of the realist strategy, 'inference to the best

explanation', takes the following form (142-149; but also, 160-170). He

184

puzzles over what the strategy might mean and comes up with two
possibilities:

I) Privilege: We are disposed to form true beliefs.

2) Force Majeure: We are forced to choose among historically

available hypotheses.
He then considers two revisions of the principle that might make it more
palatable to its critics: make it simply a matter of raising the probability of
truth, or make it simply a rule of the rationality (and not the truth) of the
belief. He then criticizes each of these possibilities in turn. Privilege
suggests we have a knack for coming up with the right hypothesis, so that
the best explanation could be among the rivals considered. But, he goes on
to say, neither rationalism (a la Spinoza) nor naturalism (a la Darwin) can
really justify this claim. Against force majeure, he argues that being forced
to choose does not mean one is forced to accept the result as good. One may
have chosen the best of a bad lot. Some choices are simply ones that enables
us to get on with the work, without judging the merits beyond that of the
chosen means.

The sophisticated versions of the 'inference to the best explanation'
strategy sound close to what he attributes to Dretske and Armstrong. Dretske
claims that laws are confirmed faster than the corresponding universal
regularity, and that this suggests something other than induction is going on.
Van Fraassen finds this puzzling, but lodges the onus on some notion of
intrinsic explanatoriness. Otherwise, either the regularity is raised at exactly
the same rate as the corresponding law, or the result is incoherence because
bonus points given to laws that depart from the probabilities assigned by a
conditionalizer lead to Dutch book problems. Other versions of what

Dretske might mean fail even quicker. It is unclear how the mere existence

185

of laws could raise the probabilities we assign to our formulation of those
laws. Nor are matters improved if we raise probabilities based on the belief
that it is a law, because then the law is assigned a probability of one. Once
we have assigned it a probability of one, there is no longer any question of
evidence confirming the law faster than its corresponding regularity, since
the law is already certain. At any rate, the confirming power of evidence is
linked to the idea that something can be intrinsically explanatory. As one
might expect from the contextual character of the pragmatics of explanation
that van Fraassen endorses, van Fraassen finds this a puzzling notion.

Better is Armstrong's attempt to make 'induction is rational' an
analytic truth (Laws 138-142). Problematic about this move is the notion of
rational compulsion that seems to be required if inference to the best
explanation is going to qualify as a normative rule for formation of beliefs
and expectations about the future. While van Fraassen agrees it might be
rational to believe something based on such a rule, he believes it fails to be
rational if one insists on following this rule. Rationality is permissive, not
compulsive, for van Fraassen. Like William James, van F raassen believes
we are rational in our belief formation so long as we do nothing that
conflicts with rational constraints. But this differs from a claim that we must
believe what is rationally required, because nothing is rationally required
absolutely, but only relative to some set of beliefs we hold. It is rational to
consult the set of beliefs we already hold in constructing other beliefs, but
unaided rationality cannot supply some initial set of beliefs as Descartes,
Spinoza and some other rationalists supposed. So, even assuming we could
solve the problem of what counts as 'best' in the best explanation, it still
would be inappropriate to suggest that this supplies foundation principles for

knowledge.

186

Now I heartily endorse these criticisms. Any version of pragmatism
maintains that we always Start in the middle of things. All versions deny that
there are any absolute beginnings. Some versions question absolute endings
as well, even in the long run. Rationality works on the stuff we supply from
the space of ordinary life and common experience. Science refines the
process and makes it self-conscious, critical and self-modifying. There is no
Baron von Munchhausen here, lifting himself by his own ponytails from the
swamp of human contingency and historical situatedness. We do construct
explanations, including laws, to guide us through experience and to fund our
practical projects. Still, this can not properly be described as inference, and
laws are still very much with us on the plane of experience.

Would van Fraassen be satisfied with this, or would he still have
problems with my point of view? What more does he require? Once again,
empirical adequacy is an issue. Not only do we never get to absolute
beginnings, i.e., truth, and hence laws in the realist sense, but empirical
adequacy places demands on constructs that rob us of laws in the sense of
empirical regularities. What we have are regularities that have held up in the
past, with no guarantees that they will hold up in the future. It still makes
sense, van Fraassen thinks, to go with the regularities that have held in the
past since simple conditionalizing gives us that much. But all we ever really
get is belief in the probability of the regularity holding, not in the truth of the
regularity. So long as the regularity is empirically adequate, we have no
reason to complain. But still there is something missing (i.e., future data)
that makes us withhold belief in the truth of the claim. Instead, we accept the
claim without believing it. This view, that we accept scientific claims
instead of believing them, has problems and I will discuss those problems

below in 4.3.6.

187

4.3.4 The Problem with Pragmatism

Since I have already tipped my hand, the reader knows that van
Fraassen is working out some of the implications of a view that draws on
William James. That makes him a pragmatist of some sort. I am a pragmatist
of some sort. The reader may be beginning to wonder what there is left to
discuss: if van Fraassen and I agree that pragmatism is the way to go,
where's the beef?

Re-reading van Fraassen, I am tempted to ask the same question.
Still, certain places and claims he makes leave the nagging suspicion that we
are not very agreed at all. While van F raassen invokes the William James of
'The Will to Believe' in justification of his claims to rationality, he does not
quote the William James of 'Pragmatism' who identifies truth with what
works and who asks about the cash value (in terms of experience) of abstract
terms. The puzzle of omission continues with the absence of any discussion
of the view of E. Nagel54 and a somewhat cursory discussion of N. Goodman
(Laws 34-35). Here are two missed opportunities to show his endorsement
of pragmatism by favoring some of the pragmatist views of laws already
available. However, van F raassen seems unwilling to entertain a use theory
of laws, of which my view is one version. Why?

I suggest the answer lies in van Fraassen's commitments that make
the pragmatic features of laws irrelevant to their truth. Van Fraassen
develops a very different theory of truth in his semantic conception of
theories. His view, I will suggest later shares some common presuppositions

with the realist positions he rejects. Evidence that van F raassen would say

 

5‘ Unless one counts the following dismissal as discussion: "The culmination of Camap's

and Reichenbach's, and Hempel's attempts, which is found in Ernest Nagel's WIS
mm, was still strangely inconclusive. In retrospect they were dealing with modalities
which they could not reduce, saw no way to finesse, could not accept unreduced, and could

not banish." Laws 38.

188

these things comes from the following passages. In Ijhefiejentifielmage,

van F raassen distinguishes the realist view from two sorts of antirealist: the
first antirealist maintains that theories are not literally true or false. This
view he identifies as instrumentalist. The second antirealist view maintains
that theories are literally true or false, but we do not know which. All we
know is whether things are as if those theories are true, that is, whether they
are empirically adequate. This second antirealist view is van Fraassen's
view. The realist is like the theist, who believes scientific theories make
literal existence claims that aim at truth. The first type of antirealist is like
the atheist, who holds that scientific theories, when not simple summaries of
observations, have at best a metaphorical truth. Finally, the second type of
antirealist is like the agnostic who maintains that scientific theories aim at
literal truth, but we have no empirical means to know if they have reached
such truth.

Whom does van Fraassen include in his instrumentalist category, the
atheists? Does the category contain only the logical empiricists whose
suspicion of theoretical entities leads them to seek creative interpretations of
central terms in physical theory? That seems true, given other things he says.
He says, for example, '[l]ogical positivism even if one is quite charitable
about what counts as a development rather than a change of position, had a
rather spectacular crash" (IheSeientjflelmage 2). In another context, he
talks about how far Stegmuller has gone on the road to "logical positivism
and instrumentalism" (Laws 191). He explains 'instrumentalism' as a view
that refuses to take a doxastic attitude toward theories. It seems to me that
this term throws together some very different approaches. Surely the view
that attempts to reduce theoretical terms to observational ones, differs from

the view that replaces one set of theoretical terms (so-called folk-theory) by

189

another set, a strategy employed by eliminativists, who seem most nearly
like the atheist of whom van Fraassen speaks. These views in turn differ
dramatically from those views of instrumentalists like Dewey and Nagel.
They all agree in denying literal truth to theories, but they affirm very
different things. However, the choice of the term 'instrumentalist', a term of
choice for the pragmatist, suggests that pragmatism is included in the views
van F raassen intends to criticize.

What is the criticism, then, that van F raassen has of pragmatists,
including those who make truth a matter of what works in practice? The
criticism would be that they one and all have the wrong notion of truth.
Features of utterances important to the pragmatist (e.g., historical and social
context and the context of other beliefs held by the utterer or the listener)
along with features associated with the uses of the utterance (e.g.,
fruitfulness, simplicity, and coherence with other accepted theories) are
irrelevant to the truth of the claim. Truth is to be identified with an
isomorphism between one of the models of the theory and the system it
purports to represent. This, it should be noted, is exactly the formulation the
realists would give, a point to which we will return later.

Why should a use theory of laws not count? Again, based on
evidence from the book, pragmatists are not correctly playing the language
game associated with laws. Laws have been taken to be principles, or
axioms of science and knowledge. This the pragmatist can concede:
rationalist philosophers have, in the past, mistaken the aim of science to be
the articulation of a system of axioms from which the entirety of knowledge
can be deduced. And van Fraassen is right again when he says that axioms
are hard to come by and do not eliminate unintended models. But unless one

is a realist, there is no reason to insist on a unique set of axioms, and the

190

pragmatist is not interested in a structure that deductively yields all
knowledge. We can make useful predictions with rough analog models,
which are considerably easier to formulate than a set of axioms. Multiple,
incompatible models may be employed on the same phenomena for different
purposes. The pragmatist finds the insistence on axioms a hindrance if it
prohibits such sound scientific practices as these.

While van F raassen's critique of axiomatic approaches is interesting,
since it does address logical empiricist projects no longer pursued, it does
not really raise a difficulty for the pragmatist. The real battle, it would seem,
lies in the rivalry between the pragmatists' account of meaning and truth, and
van Fraassen's. I will now turn to the account van Fraassen gives, how it
influences his rejection of laws, and then go on to say why I am not
persuaded to give up laws with him.

4.3.5 Van F raassen 's Position

Let me summarize what van F raassen has said so far, then go on to
develop his view, contained in his chapter 'Toward a New Epistemology'
(Laws 151-182, esp. 151-155 and 170-176). First, van Fraassen rejects
inductive logic. There is no rule that leads from data to universal
generalizations, much less laws, supposing they differ. The data do not
determine the theory. Next, there are no external constraints, laws or
otherwise, constraining our belief, even in the limit, to force one choice
upon the scientist as the rational one. These two points combine to say
rationality does not compel belief. Rationality is permissive: so long as we
are not in direct conflict with other firmly held, observationally-based or
otherwise, beliefs, we are permitted to believe what we like. Construction is
like free enterprise, we can construct whatever the marketplace of ideas will

bear. Any models we construct of the data will outrun the data. The added

191

parts are either true or false in the claims about the world they support. But it
is not the business of science to settle such matters, so long as the models
give the correct empirical results.

In technical terms, as long as we can embed all the data into some
empirical sub-model of the theory, we have no grounds to complain about
the non-empirical sub-modelsof the theory. A theory is not to be identified
with sentences and their satisfaction conditions, but with classes of models
defining some model space to represent the data. In van Fraassen's view, we
face two constraints: logical consistency and empirical adequacy. The way
science proceeds, van Fraassen claims, is by noticing when the model spaces
we constructed are not wide enough to accommodate the data, by widening
the model space by constructing new models, and then by narrowing the
space once again by empirical testing through experimentation. Laws turn
out to be formal matters; they are symmetries of the models that reproduce
symmetries of the problem. Symmetry arguments from mechanical problems
suggest the conservation laws of physics (239-243). Symmetry
considerations lead to the axioms of probability theory (331-337). It is
precisely in noticing asymmetries that Dutch book problems arise, and these
asymmetries are removed by holding to the axioms of probability (338-348).
For this reason van Fraassen believes that even though there are no rules of
ampliative inference, should we wish to adopt a rule anyway, it ought to be
Bayes's Theorem, so that it conforms to the axioms of probability, indeed, is
a deductive consequence of them, and no rule not equivalent to that rule
will.

4. 3.6 My Response to the Objection and the Position
Van Fraassen's view of model construction and experimental model

elimination, is interesting, but I think wrong in implying that all we ever do

192

is theorize, and in defining experimentation as theory carried on by other
means. At this point the pragmatist must part ways with the antirealist,
because action is not simply an extension of thought. On the contrary,
thought is a form of action. All concepts are operationally defined because
concepts are constructed through interaction with the environment. The
connection between thinking and doing is far more intimate than allowed by
the realist or, apparently, the antirealist. In spite of van Fraassen's
constructivism and his appeal to the pragmatism of William James, his
distinction between truth-conditions and truth-irrelevant pragmatic
conditions is inimical to the pragmatist enterprise. This distinction is
equivalent to the distinctions the realist makes between truth-conditions and
verification or assertibility conditions.

Because the problem lies in what appears to be a natural
interpretation of the semantic conception of theory, our best course is to take
things slowly. The semantic conception of theories identifies theories with
their classes of models, and not with the language in which those same
theories are expressed. Models are language transcendent, and theories,
which are composed of models, are language transcendent as well. Consider
what the relationship between models and language truly is. In what sense,
for van F raassen, do we construct models? My guess is that we posit an
object to which we assign different properties. In symbolic form, we
construct a set of properties and claim that any object satisfying those will
be a model of the theory. No further constraints are placed on admissible
models.

Van Fraassen's semantic conception of theories assumes that there
are objects, along with their properties and relations, as parts of the world

' and that some of those objects-in-relation are candidate models. So far, this

193

is simply the extensionalist way of talking about predicates and their
meanings. Still, an important move being made here should be noticed.
While language enters by supplying the predicates, it plays no further role in
securing the objects. Hence, satisfaction conditions are viewed from the
perspective of a representational theory of truth, rather than another, such as
a coherence theory of truth. It does so because nothing further is said about
satisfaction conditions. The existence of a model is sufficient for the
meaningfulness of the set of predicates, even if no one is ever in a 'position
to verify that a model exists by producing one.55 There may be a model
without it being the case that anyone has a model. While there are antirealist
construals of this claim, it is a favorite rallying point for realists of all sorts.
In fact, a crucial but very small step from here leads directly to the realist
independence claims. The additional move that a model may exist without
being verifiable even in principle, when added to the semantic conception of
theory, is equivalent to the realist independence claims I identified in chapter
two.

The wedge between verification conditions and truth-conditions was
already driven by van Fraassen in his work IheSeientifieImage, when he
claimed there might be models that are true but not verifiable even in
principle. He then goes on to say that as scientists, we are only concerned
with the models we can verify. This is problematic. Language has meaning
because of public conditions for correct use. If language generates these
models, where does the surplus of meaning, over and above the publicly

verifiable conditions of correct use, come from? One can say, as van

 

55 While Quine eliminates logical subjects by bound variables and hence puts predicates in
control, it remains clear that he is offering an analysis of the substantive terms of a language
and of the boundaries of sense marked by a language. We are not given unlimited license to
supply substitutional instances, or posits, for which no corresponding stimulus could be
named.

194

Fraassen does say, that language is always already outrunning experience.“
But this will not do. All of these outrunnings ought to be accorded the
character of risky projections that may need to be revised in the light of
future experience. They are taken on credit, but their credit-worthiness has
not been certified. The point for the pragmatist, however, is that all the ways
in which language outruns experience can be seen in the semantic relations
between terms and the inferential consequences of those relations. Under the
pressure of experience, we revise our language, not in a piece-meal fashion
but in a fashion that acknowledges the ways in which our central terms are
related. In other words, the relationships, though part of a language and not
verifiable as such, are empirically revisable, because the indirect
confrontation with experience may reveal the web as not making useful
distinctions. The language may be too impoverished to enable us to account
for the phenomena. The pragmatist would want to insist that while terms
may not be directly verifiable or even verifiable in practice, we still must
retain the claim that terms are empirically revisable in principle.

So, the wedge van Fraassen wants to drive between truth and
verification requires a contrast between language and models that cannot be
made. Models outrun data only if language does. Models are revisable under
the pressure of empirical data only if language is. It would seem that
language generates models in a way that renders doubtful the language-
independent status van Fraassen accords to models.

This should have been clear from reflection on laws. Laws do not

exist only at the level of fully developed mathematical models like Boyle's

 

56 Quine does too with his distinction between posits, which are rich dense objects, and
sensory stimuli, which underdeterrnine posits. The difference is that Quine nowhere
suggests that there is any question of the truth of the semantic surplus, only of the usefulness
for economy of thought and expression.

195

law or the Schroedinger equation. One may prefer, as I have for much of this
thesis, to preserve the term 'law' for science. If so, the terminology can be
rendered consistent again by properly qualifying extended uses: 'common-
sense law', 'qualitative law', 'lawlike regularity'. The point, of course, is that
these play a similar cognitive role (though not as effectively) as scientific
laws: predictions of future occurrences. This is not simply analogy: it is in
keeping with Dewey's continuity principle to call them both 'laws.‘ Laws are
present in our posits in language, as Popper showed, even in our claims
about ordinary glasses of water: firm expectations of future behavior are part
of the meaning of attribution of class properties to a perceptual object,
indeed, even in the naming of them. The difference is in the degree of
precision involved in the predictions of future behavior.
4.3. 7 Van Fraassen 's Realism

It is appropriate that our final remarks should come back full circle to
our first remarks in chapter one. Van F raassen is right to say that there is a
philosophical problem, not a scientific problem, of laws. He is also right to
identify the source of the trouble as the fondness for speculative metaphysics
present in the philosophical reflections on the nature of science throughout
the history of Western philosophy. Where he is wrong, I think, is to identify
'law' as a creature of this speculative tendency in philosophy, rather than as a
genuine constituent of science needing clarification. His willingness to make
this identification, 1 think, betrays common ground van Fraassen shares with
the realists that enables him to have dialogue with them but that also
weakens his critique. The assumptions, I suggest, lurk in a particular
understanding of semantics that pulls him in the direction of realism, and a
particular understanding of rationality that frees him from that pull, but

leaves an opening for the realists to enter.

196

In this section I am going to argue the following: insofar as van
F raassen is consistently empiricist, he offers no arguments for abandoning
anything other than realist versions of 'scientific laws'. I wholeheartedly
endorse this rejection. Insofar as van Fraassen offers a strong argument
against pragmatist construals of law, however, he must do so by partially
embracing realist assumptions. Therefore he has offered no additional
reasons for dropping the project of constructing empiricist construals of
scientific laws. Either way, nothing new has been offered by van Fraassen,
despite the sophistication of his presentation, to undermine pragmatist
efforts to elucidate the concept of scientific law.

The empiricist van Fraassen asks how we can identify the entities
invoked by realists. This, I have argued, is a good question -- in fact, the
right question to pose. The empiricist van Fraassen asks how something
objective, like a law or objective chance, could affect our expectations, or
the personal probabilities we assign to events. I think this question is
unanswerable, given the realists's strong commitment to independence of the
world from both language and mind. To bring in causality is to bring in too
little too late, and, indeed, it is the wrong sort of solution, given the
ontological divide between cause as part of the world, and language and
minds on the other side. The realist has reduced the options to dogmatism or
skepticism.

The realist van Fraassen wants to distinguish belief from acceptance
and empirical adequacy from truth. These distinctions per se are not
objectionable: we can have good empirical reasons for making such
distinctions. But the way van Fraassen makes these distinctions shows that
he has already conceded some crucial ground to the realists. Both truth and

belief are construed as the realist would construe them: truth requires

197

isomorphism of model and reality, and belief is belief that the model is
isomorphic with reality. These are precisely the ways that the Wittgenstein
of the Itaetahts would characterize these terms. This is the way the realists
talk when they define the conditions under which a theory is to be counted
as true and should be believed. It funds the talk about verisimilitude we find
in Popper and Newton-Smith. It assumes it makes sense to say that there is a
model without any consideration of how anyone could ever know that was
the case.

How does it happen that van F raassen and the realist are in
agreement here? The analogy with religion may be at fault here. In religion,
the difference between realism and antirealism is framed in terms of belief
in a transcendent Being, God. Both agree the term is meaningful. They
disagree on the possibility of knowing whether such a being does exist. Both
differ from the atheist who often wants to claim that the question does not
make sense. However appropriate such a stance may be for theology, we
can't help but wonder whether it is at all appropriate for science.

Given this agreement between van F raassen and the realists on truth
and belief, how does van Fraassen come to be in disagreement with the
realists? Van F raassen weakens the concept of rational constraint to the
point that we are not forced to believe what our best theories claim.
Rationality as permissive, rather than coercive, stands center stage in the
constructive part of van Fraassen's constructive empiricism. And‘it comes
down to this: any internally consistent model is rationally permissible if also
consistent with the empirical data. This, I believe, leaves the gate open for
the wolves. It frees us from scruples that every empiricist should want to
maintain. We can invoke ridiculous entities in our theories, so long as

logical consistency and empirical adequacy are respected. But as a claim

198

 

about the plausibility of hypotheses, this simply will not do. Analogy needs
to be respected. Common sense ontologies are given a role in van Fraassen's
model of theory construction that ignores the important transactions in the
opposite direction. Science modifies common sense--and should! The
pragmatist view of reflective equilibrium acknowledges van Fraassen's point
about lack of privilege, without giving up the continuity of mutual
modification of our entire set of beliefs. As William James said, any new
belief must run the gauntlet of our other beliefs to pass muster. Any new
belief may be rejected--or may lead to the rejection or far reaching
modification of our other beliefs. The test ultimately lies in the predictive
success of the set of beliefs as revised. Since laws are 'the predictors par
excellence, the devices we explicitly construct for the sake of predictive
success, laws must continue to play a role in any epistemology endorsed by
a pragmatist.

Van Fraassen's permissive rationality is too permissive. It places no
bars on realist projects beyond logical consistency and empirical adequacy,
and these bars can be surmounted by weakening the content of theories. The
predictive failure--indeed the absence of any predictive element in the realist
proposals is a serious fault that van F raassen's critique must leave
untouched. Indeed, his view shares this weakness: experiment is simply seen
as a cognitive activity that requires machinery. Experiment is not part of our
active probing of the world producing defeaters and confirmers that would
not even exist without such active probing.

Van F raassen's models are not constructed in the important relevant
sense. Just as for Popper, the temporal introduction afforded by theoretical
construction does not affect the Platonic existence of the class of confirmers

and defeaters. Empirical adequacy, defined as consistency with all

199

observable events past present and future, invokes a set with as objective an
existence as any the realistwould want. The pragmatist wants construction
in a more robust sense, where there are no defeaters or confirmers until we
construct either them or the conditions for their existence.

We construct the initial conditions, the possible falsifying conditions,
and control the conditions under which the instances unfold. We adjust
beliefs in response to changes in the world, and we adjust the world in
response to changes in belief. The point about laws is as follows: laws are
not objective items trying to break in to our subjective consciousness as the
realists suppose. In one sense, there is nothing more to laws than van
F raassen's personal probabilities, provided, of course, that we make these the
property of a group rather than an individual. This is our best guess about
how events will go in the future based on evidence of the past. I have called
this the backward looking argument, and it is a point on which van F raassen
and I are in agreement. Nevertheless, van Fraassen relegates to the realm of
pragmatics the practical testing and projecting that shows we do indeed
believe these claims, and he claims pragmatics are epistemically irrelevant.
For the pragmatist, however, our willingness to lay everything on the line, so
to speak, and to stick out our epistemic necks to see if we can reproduce the
results and apply them in practical situations -- these are quite relevant to
whether we have true beliefs or not. This is the forward looking argument. It
is absent fi'om van Fraassen as well as the realists. The reason it is absent, I
suspect, is that the constraints either philosophical school is willing to place
on models are too weak. If some parts of a model are not verifiable even in
principle, we have little motivation to try to deliberately revise all parts of
our theoretical vocabulary in the light of experience. We can declare some

parts of our language permanently off-limits to empirical scrutiny. Space-

200

time trees seem to be of this sort. Or we can come up with duplicate
predicates that are verifiedwhenever their ordinary double is. Universals
strike me as being in this category. Either way, we have distinctions that do
not even in principle make a difference in experience. The central drama we
see is the Hume world: events happen in a regular way without announcing
the regularity they exhibit. The metaphysical scaffolding that is supposed to
make the events in the Hume world possible are all hidden behind an
impenetrable curtain or perhaps we do have metaphysical agents who come
from behind the curtain only to appear on stage disguised as actors we have
already seen before.

The pragmatist theory of meaning, that distinctions worth making
must make a difference to mark a difference, preserves the connection with
experience that makes the forward looking argument appropriate. It is for the
sake of making our lives go as well as possible that we attend to whether our
ways of interacting with the world are working or not. Indeed, we
deliberately try to fail our claims if a lot hangs on their truth practically.57
Van Fraassen lacks this element, because his timeless class of empirical
evidence makes the temporal character of human existence recede from view
and from consideration for the reasons why we engage in science. But the

motives of predicting the future and taming uncertainty drive scientific

 

57 Metaphors are no substitute for arguments, but someone may wonder how I would
describe the scene in terms of the stage-metaphor I just used? Do we see behind the curtain
by 'projection?' Are we the authors of the play? No, neither will work. One claims we see
the invisible, which is untrue. The other is simply idealism, which also is false. The reality is
more complicated. It is more a collaborative enterprise, like film-making, where screen-
writer, director, and actors all have input in how the film goes, and the original direction
given by the script may be challenged successfully when director or actor appeal to the need
for coherencies or consistencies of various sorts, e.g., keeping a character in type by adding
a typical action or omitting an atypical one. But at times the screen-writer will win, claiming
the director or actor failed to understand the complexity of the character or the situation.

201

progress and prevent science from being an idle intellectual game as some
argue metaphysics tends to be.

Insofar as van Fraassen shares the realist premises, he shares their
weaknesses. The ontological divide poses problems if we take it seriously,
even for those going on to say science should not concern itself with it. Real
classes of defeaters or confirmers raise the same problematics for van
Fraassen as objective chance raises for the realists. He is not in a stable or
strong position to dislodge other views from that ground, and has, I think
nothing new to offer against empiricist views other than the traditional
problems raised against the logical empiricist views. But, I have argued,
pragmatism, certainly in the form it appears in Quine and Skyrms, is free of
most, if not all, of these problems. In the absence of new objections based on
unproblematic assumptions, I believe the pragmatist view holds the field.

For all these reasons, I believe laws should be preserved. Realist
laws, even if it made sense to claim that they existed, would leave us none
the wiser on how to live. Van Fraassen is right about that. But van Fraassen,
by weakening his commitment to empiricism with realist moves in
semantics and epistemology, has deprived himself of a strong reply to the
realists.

His opting for the subjectivist view of probability as opposed to the
objective one of the realists, simply leaves him enmeshed with the same
problematics. It misses a perfectly good sense of objectivity to which he is
entitled, and which I have developed, that preserves laws as objective guides
for expectations of the future. It also misses what such objectivity adds to
Bayes's Theorem that does not lead to evaluative irrationality: the need for
realizing, in practical actions, the lawful regularities we believe hold.

Replication and reliable application do not change the probabilities we

202

assign to events, as Dretske might claim. Rather, they simply make clear
which probabilities we believe, and which we merely entertain with some
weaker doxastic attitude, perhaps acceptance, in van Fraassen's terms. We
believe what we act upon, provided we do not act tentatively or with
qualifications and hedges, as we do in the initial stages of formulating a law.
Once we are willing to export scientific laws for general use, we are
declaring it defeasibly true, since we may withdraw or revise the claim at a
later date, but true nevertheless. Van Fraassen's sophisticated probabilism
makes sense in extended theoretical construction, and in risky policy
discussions, where the trade-offs need to be evaluated carefully. But for
general living and the application of tested technologies, it makes no sense
to quarrel with the claim that people treat taken-for-granted belief as true
and treat taken-for-granted technology as working in the intended way. If we
were not able to treat some expectations as secure, we would not be in a
position to treat other expectations as less than secure. This is the same point
I raised against Collins who starts with dissensus rather than consensus. This
is something, by the way, that is very odd for a metal theorist of science to
do, who would agree, I should think, that we must have common ground to
even be in a position to disagree on anything. Perhaps we go too far if we
call such expectations 'certainties', although we would assign them personal
probabilities very close to, if not identical, to one. They are not certainties
because, in the light of past experience, we know that some of them, perhaps
all of them, will need revising in the future. Still, they are as close to
certainties as anything we ever have in our lives, because we are willing to
act on them without question or hesitation. They become the basis of our

habits, without which we could not live.

203

Ultimately, my quarrel with van F raassen lies here. Theoretical
revision takes place in the context of taken-for-granted beliefs and habits.
Laws are sharp formulations of beliefs that become the basis of sharp
formulations of practice, whether technical, like in the design of devices like
televisions and computers, and'in' technical practices, like medicine and
engineering, or non-technical, as in hygiene habits we practice without
thinking about their justification in medical theory and practice. Laws are
habits of thought we acquire through transactions with systems under study,
habits of thought that give rise to habits of practice if truly believed. We
cannot understand the connection between theory and practice if, like van
Fraassen, we withhold the connection of belief that united them. If 'belief
and 'firm expectation' are seen as equivalent, then the problem of laws
comes down to the problem of belief, and this problem cannot be
formulated, much less solved in van F raassen's terms. So, van Fraassen
banished laws from science when he banished belief. His commitment to
realist premises made the problem of laws insoluble.

Thus, I believe it is his partial commitment to realist assumptions
that has left van Fraassen bereft of good reasons for continuing the search
for an empiricist or
pragmatist account of scientific laws. But this commitment is inconsistent
with his empiricism. Insofar as he is an empiricist, I think we agree. Where
he departs from empiricism, he acquires the weaknesses and incoherencies
of the very views he opposes. Realism gives him a platform from which to
criticize my view, but it is a view or family of views that has troubles in the
foundations. If realism's assumptions are mutually inconsistent, as I have
argued, no coherent objection can be made based on that set of assumptions.

Hence, as I read him, either we agree, in which case there is no reason to

204

forgo laws, or we disagree. Because he is not in a position to coherently state
the nature of our disagreement, I still have no reason to forgo laws. Thus I
defend pragmatism in its view of scientific laws.

4.4 Conclusion

All three sets of objections seek to show I cannot have objectivity
and relativity at the same time. The realist seeks to show that the pragmatist
is really a relativist and subjectivist who cannot do justice to the causal
structure of the world. The social constructivist seeks to show that I am
really an inconsistent realist who realizes that science is social and involves
human choices, but cannot give up the last remnant of realism. I still want
nature to teach us, although the lessons are negative rather than positive.
Van F raassen seeks to show that I have not learned yet that science is not
about truth, but rational belief. I have argued that the only world we know is
a world for us, a world with which we interact and which we construct. All
three of my objectors, Armstrong, Collins and van Fraassen, miss the
importance of dynamic interaction with the world.

In response to the realists, I say that our conceptual framework is a
product of our interaction with the world. Our concepts and actions work
together to create a livable world. How they create such a world is as much a
function of our interests and aims as it is of the satisfactions the world
offers. When we forget how much we have contributed to the picture of the
world we have, we open ourselves to becoming ideological dupes of a
particular social agendum, as, I fear, is the case with the biodeterrninists. We
construct scientific laws for our own well-being.

In response to the social constructivists, I say the world does resist us
and our constructions sometimes fail. No species is an island. We all live in

an environment that offers us no guarantees of success. Consensus is not

205

necessarily secured by conquest, rhetorical or political. Consensus, like any
other construction, may fail. Laws,_like social conventions, are important
because we need taken-for-granted regularities for cooperative enterprises.
Laws, unlike social conventions, are important because they are taken-for-
granted regularities that are not endlessly negotiable. They mark out the
boundary of variation, and guide our actions as much by what they rule out
as by what they rule in.

In response to van Fraassen, I say that there is nothing to be gained
by distinguishing rational belief and knowledge, or acceptance and belief, if
the only point in making such a distinction is to avoid epistemic risk.
Scientific methodology reduces, but does not eliminate, the risk of error.
Commitment and willingness to act are the litmus test of belief for the
pragmatist, and being taken-for-granted is the sign of knowledge and truth
for both ordinary life and science. I believe not only is my view stable, but

also it is the only view of those considered that keeps related things together.

206

 

Chapter Five: Conclusion of Thesis

I have argued in this thesis that scientific laws are of great
importance to finite beings like ourselves, who occupy a limited stretch of
space and time, and, with limited resources, need to figure out what set of
practices will conduce to our well-being. We not only have a need for
generalizations that work locally, in the environment we happen to find
ourselves, but globally as well. Laws can do this because they pick out the
features that covary and distinguish them from the features that vary
independently. The problem of laws has been to figure out how laws can do
that. How can laws pick out stable regularities from unstable ones?

The realist seeks the answer in metaphysics. Universals, space-time
trees, and natural kind hierarchies have a stability lacking in their instances,
and having the right metaphysics gives us an explanation for the stable
regularities. Aristotle favored an answer of this sort, and modern realists,
like Armstrong, have worked out sophisticated versions of Aristotle's views.
The problem of laws arises because we have come to accept Aristotle's claim
that there is a natural necessity, distinct from logical necessity, that we need
to grasp if we are to understand laws. There's the rub. Modern realists reject
Aristotle's epistemology in favor of modern empirical science. Information
has lost its metaphorical status for us since we no longer believe we abstract
the form and have our consciousness shaped by it--in-formed by the form of
the object. We have a distinction, between mind and material object, that
grew out of religious and ethical convictions of a Socrates, who believed the
inner person was more important than the outer one. And we have modern
naturalistic science, that makes distinctions between different natural
systems for convenience sake, but is committed to the fundamental

similarity of all physical things. How do we bring Socratic distinction

207

together with naturalistic similarity? Can we? I have argued that realism
cannot save scientific laws because the items in its accounts cannot do the
work they are supposed to do. Universals and causes are on the wrong side .
of the ontological divide between minds and the world to do us any
epistemic good.

I have argued that it is best to start our reflections in the middle of
our inquiries, rather than seeking a mythic beginning in some ur-principles
that have been guiding us all along. Perhaps our paths are guarded by angels,
but that doesn't help us epistemically unless they speak. Our inquiries are the
efforts of finite, embodied cognitive beings. We muddle along as best we
can, given our limitations in time, energy and experience. We construct a
world useful to our needs by casting about for usefill descriptions that get us
somewhere. There may have been other distinctions we could have made
that would have done just as well. That doesn't matter. What matters is
whether the distinctions we have made bring order to experience. If we can
discern regular repeatable patterns in experience, we are in a position to take
advantage of them for our practical purposes. The trick is to pick the right
regularities. We accomplish the trick by trial and error until we see the ways
in which some regularities fail. Experimental methodology grows out of
common sense experience of discovering our mistakes. The demand for
replication in the laboratory, in a controlled environment, and reliable
application outside the laboratory, in a less controlled environment, is the
reasonable demand of those deceived in the past. Regularities become stable
in the process of making sure they hold up. A stable regularity is really all
one needs to infer the future from the past. A stable regularity is just the sort
of thing one can see, because it only becomes stable as we closely attend to

its conditions of validity. So, my view, 1 think, solves both the identification

208

problem and the inference problem as van F raassen describes them. I can tell
you which states of affairs pick out a law. A covariance among measurable
variables that we cannot upset in any of the standard ways, is a law. I can
also tell you why such covariances are inference tickets to future events.
Such covariances as pass the test are unlikely to be accidental.

If there are no guarantees, then of course I cannot guarantee my
position is stable. Nevertheless, I am unpersuaded by at least some of the
objections that might be raised to my position by realists, antirealists and
social constructivists. I am not persuaded by those who want to say that
either nature dictates to us or we to nature, but not both. Realism versus
relativism strikes me as part of the shop-wom problematics of an ancient
dualism. Nature or culture? That was the debate between the sophists in
Athens. We work out the distinction between the social and the natural
within experience, and should not treat complementary perspectives as
though they were antithetical alternatives. I am not impressed either with
van Fraassen's assumption that verificationism was simply a bad idea
created by the logical empiricists. They may have constructed a version
which failed, but the demand for verifiability seems to me to be part of
empiricism. That demand should only be abandoned if one is prepared to
suggest how else, besides through experience, we are to come to know about
the world.

Given the choice between a slow muddle of scientific inquiry, and
the jet-flight to Platonic heaven offered by the inference to the best
explanation, I prefer the muddle. At least the muddle has some chance of
getting us somewhere, whereas the speculative jet-flight is never going to
get off the ground. So, while I am grateful for the realist offer to solve my

problems, I think I am better off trying to solve them myself.

209

Laws are human constructions. When they work, they can be used by
anyone, anywhere at any time. Scientific laws are universal. Since there are
very few human institutions for which that is true, scientific laws are
remarkable testimony to the value of science for humanity. If human history
is viewed as one big coordination problem, we can see in laws 3 way
forward through ideological impasses. If that is so, then van Fraassen's

demand that we give laws up is one from which we should demure.

210

BIBLIOGRAPHY

211

10.

ll.

. Armstrong, D[avid] M[amet]. W3.

. Aronson, Jerrold, Rom Harre, and Eileen Cornell Way. Realism

Bibliography

Cambridge: Cambridge UP, 1973.

llnixersalaanificiemifieliealism Cambridge: Cambridge UP,
1978.

---. What is a Law at Nature? Cambridge: Cambridge UP, 1983.

---. A Wetld Qt States at Affairs. Cambridge: Cambridge UP, 1997.

WWW LaSalle 111.: Open
Court, 1995.

Barnes, Jonathan, ed. WW. Princeton:
Princeton UP, 1984.

 

Berkeley, George. 126.121.9111. Ed. and Trans. Douglas M. Jesseph.
Dordrecht: Kluwer, 1992.

---. jueiplemfflhmanfinewledge. Lasalle, 111.: Open Court, 1986.

Bigelow, John, and Robert Pargetter. W.
Cambridge: Cambridge UP, 1990.

Bronshtein, I. N., and K. A. Semendyayev. HandheekeflMathematiea.
New York: Van Nostrand Reinhold, 1985.

Buchdahl, Gert. Metaphmsandtlrefilulosonhuffimence
Cambridge, Mass.: MIT, 1969.

212

12.

13.

14.

15. ---

16.

17.

18.

19.

20.

21.

Burks, Arthur W. W. Chicago: U of Chicago P,
1977.

Burn, Edwin A. IheMmanhxsicaLEQundafionmeodemPhxsical
Seienee. Garden City, N.Y.: Doubleday, 1954.

Carnap, Rudolf. Meaningandfleeessiu. Chicago: U of Chicago P,
1956.

.Ehflosonhicalioundatlonsoffihxsics New York: Basic, 1966.

Chisholm, Roderick. “The Contrary-to-Fact Conditional.” Mind 55
(1946): 289-307.

. “Law Statements and Counterfactual Inference.” Analysis 15
(1955): 97-105.

Cochrane, R Mmmaflflmmmflmmhmnmm
eflfitandatds. Washington, D. C. National Bureau of Standards,

1966.

Collins, Harry. W. London: Sage, 1985.

Conant, James Bryant. “Robert Boyle's Experiments in Pneumatics.”

WWW Eds James
Bryant Conant and Leonard K. Nash. Vol. 1. Cambridge: Harvard

UP, 1957. 1-64.

Coyne, Gary. IhaLabnratcmHandhooknflMatefialafixnefiment
ahdleehnigue. Englewood Cliffs, NJ: Prentice Hall, 1992.

213

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

 

Descartes, R. IheBhrleonhrealfluungmeeseanes Eds. J.
Cottingham, R. Stoothoff, and D. Murdoch. Vol. 1. Cambridge:

Cambridge UP, 1984. 177-291.

---. “The World.” Descartes, Rhilesephieal. 1: 79-99.

Dewey, John. W. Carbondale, 111.: South
Illinois UP, 1991.

Dijksterhuis, E. J. MeehanizatieneflthLWetfiEiemLe. Princeton:
Princeton UP, 1986.

Dretske, Fred 1. “Laws of Nature.” Rhilesephuffieienee 44.2
(1977): 248-268.

---, W. Chicago: U of Chicago P, 1969.

Ellis, Brian. RasieQeneeptsefMeasmement. Cambridge: Cambridge
UP, 1968.

Feynman, Richard. Ihe_Charaeter_Qf_Ehysieal_Law. Cambridge: MIT,
1 967.

---, W. Vol. 1. Reading, Mass.: Addison-Wesley,
1963.

Gooding, David. ExperimentandtheMakingefiMeaning. Dordrecht:
Kluwer, 1990.

Goodman, Nelson. Eaet,_Eietien_and_F_QLeeast. 4rth. ed. Cambridge,
Mass.: Harvard UP, 1983.

214

33.

34.

35.

36.

37.

38. ---

39.

40.

41.

42.

43.

Hanna, Joseph. “Empirical Adequacy.” Ehllflmhimflficifinfl 50
(1983): 1-34.

. “Objective Homogeneity Relativized.” Rhilesephuffieienee l
(1986): 422-431.

Harre, Rom. Lawmflflatnre. London: Duckworth, 1994.

---. _arieties_9_f_Realistn. Oxford: Blackwell, 1986.

Hempel, Carl. Asneetseflfieientiflefixnlanatien. New York:
Macmillan, 1965.

. "Empiricist Criteria of Cognitive Significance." Hempel, Aspeets.
101-122.

Hesse, Mary. “Laws and Theories.” Eneyelenediaeflfihilesenhy.
Eds., Arthur Pap and Paul Edwards. Vol.4. New York: Macmillan,

1967. 404-410.

Hume, David. Treatiseenflnmanhlamre. Ed. P.H Nidditch. 2nd. ed.
Oxford: Clarendon, 1978.

Jackson, Frank ed. Cenditienals. Oxford: Oxford UP, 1991.

Jasanoff, Sheila, GeraldE. Markle, James C. Petersen, et al eds.

HandhmknfScimandIeshnolmSmdies Thousand Oaks:
Sage, 1995.

Kneale, William. “Natural Laws and Contrary-to-Fact Conditionals.”
Analysis 10 (1950): 121-125.

215

44.

45.

46.

47.

48.

49.

50.

51.

52.

53.

54..

55.

---. “Universality and Necessity.” Bdtishlenmalfiflhefihilgsenhy
eflSeienee 12 (1961): 89-102.

Knorr-Cetina, Karin. WWW:- Oxford: Pergamon,
1 980.

---. W. Cambridge: Harvard UP, 1999.

Latour, B. SeitmeejnAetien. Cambridge: Harvard UP, 1987.

Leibniz, Gottfried Wilhelm. WWW. Ed.
Leroy E. Loemker. Dordrecht: Kluwer, 1989.

Lenzen, Victor. “Procedures of Empirical Science.” in Neurath 1:
279-339.

Lewis, David. Cannterfaemals. Cambridge: Harvard UP, 1973.

---. “New Work for a Theory of Universals.” RanetsinMetanhysies
andEnistenielegy. Cambridge: Cambridge UP, 1999. 8-55.

---. anheflnmhmeflflquds. Oxford: Blackwell, 1986.

Maekie. J. L. WWW
Oxford: Clarendon, 1974.

McCall, Storrs. AMedeleflthellnis/erse. Oxford: Clarendon, 1996.

Mill, John Stuart. SystemeflLegie. New York: Harper, 1881.

216

56.

57.

58.

59.

60.

61.

62.

63.

65.

66.

Nagel, Ernest. W. Indianapolis: Hackett, 1979.

Needham, Jacob. “Human Laws of Nature in China and the West.”

W121 (Jan 1951) 3 30; 194 230

Neurath, Otto, Rudolf Camap, and Charles Morris, eds. International

Eneyelenediaefllnifletlfieienee. 2 vols. Chicago: U of Chicago P,
1955.

Pickering, Andrew. Manglegifltaetiee. Chicago: U of Chicago P,
1995.

---, ed . W. Chicago: U of Chicago P,
1992.

Plato. Cemnleteflqtks. Ed. John M. Cooper. Indianapolis: Hackett,
l 997.

Popper, Karl. IheLegieeflSeientifieDisegyem. New York: Harper,
1959.

Quine, W[illard] V[an] O[rmand]. “Ontological Relativity.”

QmeloeieaLRelatixinLandDJheLEssaxs New York Columbia
UP, 1969.

---. W. Cambridge: MIT P, 1960.

Reichenbach, Hans. Laws,_MedMeundfiMetfaemals. Berkeley:
U of California P, 1976.

Ruby, Jane. “The Origins of Scientific 'Law'.” LQumaleLtheHistem
endeas 47.3 (July 1986): 341-359.

217

67.

68.

69.

70.

71.

72.

73.

74.

75. ---

76.

77.

Sehwedt. Georg.Ihe_Essennal_Gu1de_tsLAnahm;_Chemm- New
York: Wiley, 1997.

Shugar, Gershon, and John A. Dean. IhefihemistlsReadeeferenee
Handhoek. New York: McGraw, 1990.

Singer, Charles. W. Mineola: Dover, 1997.

Skyrms Brian W
W. New Haven: Yale UP, 1980.

Suppes, Patrick and Joseph Zinnes. “Basic Measurement Theory.”

HandhmknfMathematicalehologx Eds R D Luce R..R
Bush andE. H. Galanter. Vol. 1. New York: Wiley, 1963: 3- 76.

Thomas Aquinas. Snmmalheolcgiaelatintextandfinglish

l . . 1 . 1' 1 1 . .
Cambridge: Blackfriars, 1964.

Tooley, Michael. CansanemARealistAnpi’eaeh. Oxford: Clarendon,
1987.

van Fraassen, Bas. LawsantLSymmetm. Oxford: Clarendon, 1989.

. IheSeientifielmage. Oxford: Clarendon, 1980.

Vlastos, Gregory. Elatejsllniyezse. Seattle: U of Washington P, 1975.

Walters, R. 8. “Laws of Science and Lawlike Statements.”

Eneyelepediaeffihilesonhy. Eds. Arthur Pap and Paul Edwards.
Vol. 4. New York: Macmillan, 1967. 410—414.

218

78.

79.

80.

81.

82.

Wartofsky. Marx W. WWW
New York: Macmillan, 1968.

Williams, Frederick W
QuamitatixeReseaLeh Fort Worth. Harcourt, 1992.

Wilson, E. Bright , Jr. IntrednetientefieientifieReseateh. New York:
Macmillan, 1952.

Wittgenstein, Ludwig. Ehilesenhiealjngestigatiens. Trans. G.E.M.
Anscombe. 3rd. ed. New York: Macmillan, 1953.

Zilsel, Edgar. “The Genesis of the Concept of Physical Law.”
EhilosnnhicalRexiewSl (1942)! 245- 279

219

"‘lllllllllllllllllli