University of Iowa

Iowa Research Online

Theses and Dissertations

2012

Simplicity in science
Daniel Benjamin Schulz
University of Iowa

Copyright 2012 Daniel B. Schulz

This dissertation is available at Iowa Research Online: http://ir.uiowa.edu/etd/2979
Recommended Citation
Schulz, Daniel Benjamin. "Simplicity in science." PhD (Doctor of Philosophy) thesis, University of Iowa, 2012.
http://ir.uiowa.edu/etd/2979.

Follow this and additional works at: http://ir.uiowa.edu/etd

Part of the Philosophy Commons

SIMPLICITY IN SCIENCE

by
Daniel Benjamin Schulz

An Abstract
Of a thesis submitted in partial fulfillment of the
requirements for the Doctor of Philosophy
degree in Philosophy
in the Graduate College of
The University of Iowa

May 2012

Thesis Supervisor: Associate Professor Evan Fales

1
ABSTRACT

This dissertation investigates the possibility of justifying simplicity principles in

science. The labor of these projects is organized into three chapters.
The first chapter introduces some of the key authors and issues in the history of
simplicity in science. This chapter also gives a detailed discussion of the work of the
19th century physicists Le Verrier and Newcomb who played a crucial role in setting the
stage for Einsteins theory of relativity. These examples are used to illustrate points in the
following chapters. However, they play a specific role in the first chapter to show serious
problems with a view defended by an important contemporary author, Richard Swinburne,
that one version of the principle of parsimony contributes to the probability that scientific
theories will be true.
The second chapter elucidates the problems involved in specifying and measuring
the simplicity of scientific hypotheses and theories. Several twentieth century authors like
Harold Jeffreys, Nelson Goodman, and Mario Bunge made significant contributions to the
clarification and organization of various kinds of simplicity criteria. When simplicity criteria are employed in a scientific methodology, we find that simplicity judgments of one
kind are always traded-off with simplicity judgments of another kind. We also find that
the scientific project involves a delicate balancing of several aims. This analysis renders
a valuable result: that some dogmas, in particular, the dogma that principles of parsimony
are the final court of appeal in scientific theory selection must be jettisoned. I also find that
it is misguided to ask the question of whether or not simplicity of some clearly specified

2
kind is related to the truth. In point of fact, the legitimate questions about the justification
of specific simplicity judgments in science are much more complex and nuanced than this.
This becomes clear when it is seen exactly how different simplicity criteria are related to
one another and to the various desiderata of science.
The third chapter investigates which argument forms may be available to justify
simplicity principles in science. The results from the second chapter are employed in two
ways. First, methodological principles stand in a tight-knit set of interrelations, so our
analysis of justificatory argument forms must incorporate the complexity of these relations.
Second, simplicity is extremely heterogeneous and since no conceptual reduction of all of
the various simplicity criteria is possible, justificatory arguments must deal with clusters
of interrelated principles. This result may have certain advantages and other disadvantages
for inductive, transcendental, or inference to the best explanation approaches to the justification of simplicity. My analysis shows what will and what will not work for these
possible approaches to the question of justification and shows what some of the systematic
and metaphilosophical commitments would have to be were philosophers to pursue this
project.
Abstract Approved:
Thesis Supervisor

Title and Department

Date

SIMPLICITY IN SCIENCE

by
Daniel Benjamin Schulz

A thesis submitted in partial fulfillment of the

requirements for the Doctor of Philosophy
degree in Philosophy
in the Graduate College of
The University of Iowa

May 2012

Thesis Supervisor: Associate Professor Evan Fales

Graduate College
The University of Iowa
Iowa City, Iowa

CERTIFICATE OF APPROVAL

PH.D. THESIS

This is to certify that the Ph.D. thesis of

Daniel Benjamin Schulz
has been approved by the Examining Committee for the
thesis requirement for the Doctor of Philosophy degree in
Philosophy at the May 2012 graduation.
Thesis committee:
Evan Fales, Thesis Supervisor

Richard Fumerton

Gregory Landini

Ali Hasan

Frederick Skiff

Never tell me the odds.

Han Solo, The Empire Strikes Back

ii

TABLE OF CONTENTS

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

vi

CHAPTER
1

INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

SIMPLICITY ON THE ROAD TO PERDITION . . . . . . . . . . . . . . . 11

2.1
2.2
2.3
2.4

2.5
2.6
2.7
3

Non-Deductive Inference . . . . . . . . . . .
Induction . . . . . . . . . . . . . . . . . . . .
Inference to the Best Explanation . . . . . . .
Motivations, Terminology and Distinctions . .
2.4.1 The Venerated Tradition of Simplicity
2.4.2 Ockhams Razor . . . . . . . . . . . .
2.4.3 The Anatomy of Parsimony . . . . . .
2.4.4 Ockham Revisited . . . . . . . . . . .
2.4.5 Hume Revisited . . . . . . . . . . . .
More Than Just Ockhams Razor . . . . . . .
Neptune and Vulcan . . . . . . . . . . . . . .
The Simplicity of the Ancients . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

14
20
32
34
36
42
44
49
50
53
56
78

THE COMPLEXITIES OF SIMPLICITY . . . . . . . . . . . . . . . . . . 87

3.1
3.2

Categorizing and Gauging Simplicity . . . . . . . . . . . .

3.1.1 Ontological Simplicity . . . . . . . . . . . . . . .
The Simplicity of Systems of Signs . . . . . . . . . . . . .
3.2.0.1 The Syntactical Simplicity of Predicates .
3.2.1 The Syntactical Simplicity of Formal Expressions .
3.2.2 The Simplicity of Theories . . . . . . . . . . . . .
3.2.3 Semantical Simplicity . . . . . . . . . . . . . . . .
3.2.4 The Semantical Simplicity of Expressions . . . . .
3.2.5 The Semantical Simplicity of Theories . . . . . . .
3.2.6 Epistemological Simplicity . . . . . . . . . . . . .
3.2.7 Pragmatical Simplicity . . . . . . . . . . . . . . .
3.2.7.1 Psychological Simplicity . . . . . . . . .
3.2.7.2 Notational Simplicity . . . . . . . . . . .
3.2.7.3 Algorithmic Simplicity . . . . . . . . . .
3.2.7.4 Experimental Simplicity . . . . . . . . .
3.2.7.5 Technical Simplicity . . . . . . . . . . .

iii

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

94
94
98
100
104
108
108
110
111
112
114
114
116
116
117
117

3.3

3.4

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.

118
119
122
129
131
134
136
149
153
160
161
162
169
175

SIMPLICITY AND JUSTIFICATION . . . . . . . . . . . . . . . . . . . . 178

4.1

4.2
4.3

4.4

3.2.8 Why Categorize Simplicity? . . . . . . . . . . .

Metascientific Criteria and Their Relations . . . . . . . .
3.3.1 Systematicity . . . . . . . . . . . . . . . . . . .
3.3.2 Linguistic Exactness . . . . . . . . . . . . . . .
3.3.3 Testability and Accuracy . . . . . . . . . . . . .
3.3.4 Accuracy . . . . . . . . . . . . . . . . . . . . .
3.3.5 Depth . . . . . . . . . . . . . . . . . . . . . . .
3.3.6 Representativeness . . . . . . . . . . . . . . . .
3.3.7 The Unity of the Sciences . . . . . . . . . . . . .
3.3.8 Other Relations . . . . . . . . . . . . . . . . . .
Evaluation . . . . . . . . . . . . . . . . . . . . . . . . .
3.4.1 Ockhams Razor and Conceptual Analysis . . . .
3.4.2 Avoiding Obsessions with Syntactical Simplicity
3.4.3 A Modest Suggestion . . . . . . . . . . . . . . .

Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.1.1 Clarifying Reduction in Science . . . . . . . . . . . . .
4.1.2 The Conceptual Reduction and Simplicity . . . . . . . .
Denial . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4.3.1 The Question of Circularity . . . . . . . . . . . . . . . .
4.3.1.1 Sameness and Change . . . . . . . . . . . . .
4.3.1.2 Reasons Not to Use Induction . . . . . . . . .
4.3.2 Indispensability . . . . . . . . . . . . . . . . . . . . . .
4.3.3 Richard Swinburnes Argument . . . . . . . . . . . . . .
4.3.3.1 Swinburnes Kinds of Simplicity . . . . . . . .
4.3.3.2 Criticisms of Swinburnes View . . . . . . . .
4.3.3.3 Piety and Swinburnes Examples . . . . . . .
4.3.4 Paul Churchlands Argument . . . . . . . . . . . . . . .
4.3.4.1 Criticisms of Churchlands View . . . . . . . .
Inference to the Best Explanation . . . . . . . . . . . . . . . . .
4.4.1 How IBE Might Go . . . . . . . . . . . . . . . . . . . .
4.4.2 Thagards Analysis . . . . . . . . . . . . . . . . . . . .
4.4.3 Circularity and IBE . . . . . . . . . . . . . . . . . . . .
4.4.4 Fumertons Challenge . . . . . . . . . . . . . . . . . . .
4.4.4.1 Structural Similarities with Metaepistemology
4.4.5 The Independence of IBE? . . . . . . . . . . . . . . . .

CONCLUSIONS

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

192
194
201
208
215
216
217
219
221
224
226
229
235
236
239
243
245
252
255
257
257
260

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263

APPENDIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270

iv

A SHERLOCK HOLMES AND THE SIGN OF THE FOUR . . . . . . . . . . 270

B SIMPLICITY AND PARSIMONY . . . . . . . . . . . . . . . . . . . . 274

C METASCIENTIFIC CRITERIA CONCEPTUAL MAP . . . . . . . . . . . 277

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284

LIST OF FIGURES

Figure
2.1

Data Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.2

Curve Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.1

The Counting Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

3.2

Simplicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

3.3

Conceptual Connectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

3.4

Systematicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126

3.5

Linguistic Exactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

3.6

Testability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133

3.7

Scrutability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

3.8

Accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

3.9

Depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149

3.10 Representativness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154

3.11 Unity of Science . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
3.12 Other Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

vi

1
CHAPTER 1
INTRODUCTION

Nearly every insightful author who has written on simplicity has made note of the
little irony that simplicity is notoriously complex. While there is widespread agreement that
there are things to be said about the methodological roles played by simplicity judgments
in science, there is very little agreement on what is to be said, and very little agreement
about how we ought to approach the relevant issues. In the cycles of human intellectual
history, simplicity has made more than one appearance. As the planetary gears of epistemology go round, the philosophy of science takes paths that sometimes track the motions
of classical epistemology and, sometimes wandering paths. Questions about simplicity are
approaching perigee in both fields once again, and I believe that this has something to do
with the popularity of investigations into computational modeling, database warehousing,
and curve-fitting. Simplicitys number has come up and I turn the spears of analysis upon
it to see what clarity might contribute to this tremendous and highly technical discourse.
Despite one hundred years trees sacrificed to contrary ink, I show that clarity about
simplicity is possible and that philosophical inquiries into the justification of simplicity
judgments are both motivated and far from being resolved. I offer a critique of the simplicity in science discourse which I hope will provide a useful framework for future inquiries into simplicity judgments generally. I conclude that the traditional study of human
knowledge is discontinuous with the study of scientific theory construction and appraisal
in important ways. If I am right about this, then some issues in the philosophy of science
may not need to wait for the skeptic to be given a final answer. However, the justification of

2
simplicity in science may not be one of those issues. My results show that questions about
the justification of simplicity judgments in science are at least not crippled by formal problems and my analysis lays the groundwork for new research programs. I wish to convince
scientists and philosophers to stop using the phrase, Occams Razor because it is both
vague and subject to misguided historical interpretations. More fundamentally, I will argue
that, although it is possible (and often tempting) to formulate principles of parsimony with
ceteris paribus clauses, they are impossible to apply in theory construction or evaluation.
My critique aims to convince philosophers to stop using these these mistaken terms and to
lay the groundwork for future investigations into the actual roles played by simplicity in
science.
Some have hoped that specific judgments of simplicity might guide us in ontology.
The notion of treating simplicity as an attribute or property has been associated with beauty,
perfection, and truth. Where simplicity is instantiated so are these others expected to be
instantiated as well. But I wonder why we should not expect beauty and perfection to be
found in that which hangs on the very edge of chaotic complexity. It does seem as if a great
artist, and perhaps a great engineer, would achieve the finest of works by balancing order at
the very threshold of chaos - unless, of course, the engineer could explain, in a few simple
words, why things worked out that way. The abstract impressionist or the jazz musician
seeks to get a feel for the patterns of sensation and then to make just one contribution to
the darn-near-chaotic ripples of the media - not too little not to be noticed - not too much
to destroy.
I am stuck. Is it simplicity or complexity that is a guide to knowledge about how

3
to make the best contribution to art or to science? I need only to think about one (either
simplicity or chaos) to notice that the one that I am thinking of seems to be governed by
the other. On the one hand, it seems that the world is complex, and my contribution (as an
artist) to the world is simple if it is well placed. Is it simple because it is one contribution
or because it is well placed? On the other hand, I wonder if a complex series of sensations
is simple in some respect - in that they all occur within one medium or in that they are
governed by few laws. In science, as in art, we seek to balance the complex with the simple.
In art this involves a science and in science I suspect an art. Simplicity judgments are a
mixed bag. We must distinguish them so that we can tell which are relevant to ontology
and which are not. Future projects can show which of the simplicity judgments that are
relevant to ontology are actually justified, if any are.
Sometimes simplicity is a word used to indicate easiness with respect to certain
physical or mental efforts. Perhaps the association of simplicity with low metal effort
explains why the word simple has been used to connote stupidity or ignorance. As it turns
out, the simple which indicates ignorance is totally incompatible with another association
often made with the word: that simplicity is a divine attribute, and therefore naturally
associated with omnipotence, omniscience and perfection. It asserts no mere tautology to
say that simplicity either is or is not a divine attribute. We must first disambiguate senses of
the word simplicity. Perhaps the or in this sentence is inclusive. Even if we restricted
our discussion of simplicity to the mind, we may still run the risk of equivocation. A
thought may be simple in some ways but complex in others.
Many have thought that our non-deductive inferences depend upon judgments of

4
simplicity. Many have also thought that if there is a solution to the problem of induction,
then that solution depends upon the success or failure of the project to justify some criterion
of simplicity. Naturally, a subset of these folks consists of those suspicious that no such
justification is available and that the problem of induction has no solution. This discussion
is focused on the roles of simplicity judgments in the construction and evaluation of scientific theories. Some might meet this with despair - that the philosophy of simplicity in
science is, at best only a partial guide to the problems of simplicity in traditional epistemology and metaphysics, and at worst it is no guide. I meet this conclusion with optimism
- progress on some issues in the philosophy of science would not wait for the skeptic to
be given a final answer. However, in the final analysis, it is possible that the questions of
traditional epistemology will be revealed to be more fundamental than questions about the
justification of simplicity in science. If that is the case, the so be it. We would have learned
something about the order of priority of our intellectual divisions of labor. Alternatively,
the philosophy of science and traditional epistemology may inform one another.
This project takes the approach that this may be the case, by investigating some of
the history of science and some of the arguments from the philosophy of science to see
if we can learn anything about the structure of the simplicity dialectic generally. I leave
open the question of whether or not we could have discovered any of the things that my
investigation reveals by searching the depths of our imaginations alone and say only that
this philosopher lacks the imagination to have generated these results without looking into
both the philosophy of science and the history of science.
The result that we can settle some issues in the philosophy of science without an-

5
swering the skeptic is a game-changer for some controversies in the philosophy of science.
The realism/antirealism controversy in the philosophy of science may be characterized
as an epistemological debate about whether or not the truth is a desideratum of science.
Philosophers like Richard Swinburne attempt to argue that simplicity is related to the truth
and if simplicity is not a desideratum of science, then science is non-rational. Swinburne
would agree with Paul Churchland on an additional point; that the scientific antirealist view
which does not include non-empirical criteria of theory choice (like simplicity) collapses
into skepticism. I do not engage the scientific realism versus antirealism controversy directly. However, some of these arguments are useful to present and evaluate because they
involve the claim that simplicity judgments are essential for the construction and evaluation
of our successful scientific theories. Swinburne and Churchland could be right that simplicity is essential for science. However, Swinburne has not yet successfully argued that
simplicity is related to the truth and neither Churchland nor Swinburne are able to show
that the mind has simplicity concepts. By analyzing the forms of their arguments, I show
that the final justification for simplicity judgments (if one is available at all) does depend
upon arguments in traditional epistemology, but that the scientific realism/anti-realism controversy does not collapse into the problem of skepticism. We need only to give a careful
analysis of scientific methodologies involving simplicity judgments to reach the conclusion
that an elucidation of science that takes empirical adequacy to be the sole criterion of theory acceptance (the scientific anti-realist view) is very implausible. This conclusion shows
that we need not level the playing field - so that everyone must answer the skeptic - in
order to decide some crucial things about how to elucidate science. I show how an analysis

6
of simplicity judgments and their interrelations to the other desiderata of scientific theory
construction and evaluation contribute to these conclusions.
My analysis contributes to projects that began in the early and middle parts of the
twentieth century to elucidate simplicity judgments. These efforts were spearheaded by
notable philosophers, scientists and mathematicians like Harold Jeffreys, Nelson Goodman
and Mario Bunge. In the contemporary literature, Jeffreys is cited far more frequently than
Goodman or Bunge, which I think is a serious oversight. Additionally, attempts to elucidate
simplicity seem to have been generally neglected after about 1970. One of the last attempts
at this was given by Elliott Sober in 1975. I believe that it is a shame that Goodman and
Bunge are not cited more often because their arguments might temper the speed at which
contemporary papers on curve-fitting fly into publication.
The fact that very little has been done to bring analytic clarity to the roles of simplicity in science is at once tantalizing and vexing. Why would simplicity judgments that
have long been recognized to play crucial roles in scientific theory construction and evaluation generally escape the efforts of analytic philosophers? I have no answer for this.
Scientists are certainly aware that simplicity judgments play roles in theory construction
and evaluation. Newton himself put forth four principles of simplicity. The astronomer
Simon Newcomb, who made important contributions to the scientific dialog that set the
stage for Einsteins papers on Special and General Relativity, says of his adjustments to the
Hall hypothesis (1895) that,
I can only remark that its simplicity and its general accord with all optical phenomena are such that it seems to me that it should be accepted, in the
absence of evidence against it ([50] p.63).

7
Newcomb also makes a general claim regarding simplicity in scientific methodology.
Should we aim simply at getting the best agreement with observations by
corrections more or less empirical to the theory? It seems to me very clear
that this question should be answered in the negative. No conclusions could be
drawn from future comparisons of such tables with observations, except after
reducing the tabular results to some consistent theory,... Our tables must be
founded on some perfectly consistent theory, as simple as possible, the elements of which shall be so chosen as best to represent the observations ([50]
p.64).
We should expect that the explicit invocation of principles of simplicity by scientists would draw philosophers with analytic ambitions to simplicity. We should expect that
the fundamental issues of simplicity in epistemology would draw philosophers. Draw them
simplicity does, but like vultures to something believed to be dead. The sheer volume of
articles published in the last one hundred years on simplicity in science is intimidating to
say the least. However, little of this literature aims to elucidate simplicity in science rather
than either to avoid questions about its justification or to point out that it has already been
settled that simplicity is indispensable for science. What is worse is that many contemporary authors believe that the problem of the role of simplicity in theory choice has been
successfully boiled down to the question about curve-fitting algorithms. This literature is
extremely technical and much of it is mathematical in nature. This perhaps scares some
people away while others it sucks in, tantalized by its mystifying rigor. It is not true that
the philosophical problems of simplicity have been dispatched to aesthetics or that they
have been analyzed in purely pragmatic terms. It is not true that the role of simplicity in
theory selection can be wholly analyzed using curve-fitting algorithms. It is frustrating that
contemporary journal articles are littered with vague and confused appeals to simplicity,

8
but we need not scuttle the literature for errors or oversights to find puzzling claims involving simplicity. Some of the most influential and careful thinkers have said things that
deserve some clarification.
Lavoisier said,
I have deduced all the explanations from a simple principle, that pure or
vital air is composed of a principle particular to it, which forms its base, and
which I have named the oxygen principle, combined with the matter of fire
and heat. Once this principle was admitted, the main difficulties of chemistry
appeared to dissipate and vanish, and all the phenomena were explained with
an astonishing simplicity. ([60])

It is a weird kind of thing to say. Lavoisier started with a simple principle and then
was astonished when simplicity was exemplified by his resulting theory. I do not know
why a scientist who has just struck upon the oxygen in - oxygen out theory of chemistry
would be astonished when a simple principle generates a theory that exemplifies simplicity
- simplicity in - simplicity out. Perhaps Lavoisier noticed that a principle of one kind of
simplicity could be used to construct a theory that exemplified another kind of simplicity.
That would be astonishing. This is precisely the sort of thing that requires clarification. I
aim to provide a framework of analysis that will aid in clarifying claims of this sort.
Another highly regarded author, J.J.C. Smart, makes use of the a principle involving
some form of simplicity in Sensations and Brain Processes (1959). Smart says,
The suggestion that I wish if possible to avoid is a different one [different
from the suggestion that saying that one is in pain replaces another behavior
like crying], namely that I am in pain is a genuine report, and what it reports
is an irreducibly psychical something. And similarly the suggestion I wish to
resist is also that to say I have a yellowish orange after-image is to report
something irreducibly psychical.
Why do I wish to resist this suggestion? Mainly because of Occams Razor.
([53] pg.141)

9
The term Occams Razor has a long history in science and it involves some appeal
to simplicity. It is not always clear what authors have in mind when they use the term.
Smart, however, perhaps can be clearly interpreted. Smart wishes to resist explanations
about the mind that float free of any causal laws. Smart may have in mind a principle of
simplicity about ad hoc hypotheses. However, when it comes to science, I am not sure how
many hypotheses that float free from causal laws would be permitted. My intuition is that
strict prohibition rather than simplicity governs this kind of reasoning in science.
Smart goes on to say that the hypothesis of substance dualism, offends against the
principles of parsimony and simplicity. ([53], pg.155) I wish to know what kind of offense
this is. If it is merely a misdemeanor, then I might just go in for it. After all, revolutions
arise when individuals risk breaking a few norms. However, if I could be shown that
violating the principle of simplicity meant that I was not doing science, then I would not
violate the principle while doing science.
For reasons like these, I believe that both philosophy and science will benefit from
a bit of conceptual clarification about simplicity and the roles it plays in scientific theory
construction and evaluation. I divide the labor of the analysis of simplicity in science
between three chapters.
Chapter one introduces the motivations for the project of justifying simplicity in
science and investigates several historically important arguments from both science and
philosophy to help distinguish key terms.
Chapter two applies the terms and distinctions from the first chapter to analyze the
roles played by simplicity in science. The first part of the chapter categorizes simplicity

10
judgments and introduces several of the problems involved in specifying and justifying
simplicity criteria. The second part of the chapter shows how these criteria, when they can
be clearly defined, are related to one another and to the other desiderata of science.
Chapter three investigates the justification of simplicity in science. The first part
of this chapter motivates the problem of justification by showing that it is very difficult to
avoid, and that the assertions made by many authors fail to demotivate an investigation into
the justification of simplicity in science. The remaining sections analyze argument forms
which might be employed to justify simplicity. My analysis shows what are the problems
with these argument forms and helps to reveal why, what intuition might first suggest are
problems, are not obviously problems. I conclude that inductive arguments, arguments involving an inference to the best explanation, and arguments involving an indispensability
principle are not obviously out of contention. However, I will only analyze the forms of
these arguments as they apply to the problem of simplicity in science. I will not actually
give the justificatory arguments, nor will I settle other issues having to do with the systematic reasons that philosophers might have for favoring or rejecting any of these argument
forms. As it turns out, these three argument forms are in a sort-of three way stand off.
Although I argue that none of them suffer from any obvious formal problems if applied to
justify simplicity in science, it is not clear which is more epistemologically fundamental,
or whether any depend upon some answer to the problem of skepticism.

11
CHAPTER 2
SIMPLICITY ON THE ROAD TO PERDITION

The philosophy of science literature on simplicity has grown into a hydra-headed

monster. The only way to deal with it is to hook any neck which presents itself most naturally and, from there, slash our way into the breast of the beast. We must start somewhere,
and the ride will never be less violent than it is in the opening. I suggest that the best way
to begin is to consider some very basic examples of the ways that simplicity is involved in
making inferences and then to consider a few more detailed historical case studies.
This chapter is divided into four sections. The first section introduces some of
the roles that simplicity judgments might play in non-deductive scientific reasoning. Sir
Arthur Conan Dyoles Sherlock Holmes novels are often referred to on this topic. I discuss
Doyles famous Sign of the Four to help illustrate some of the ways in which simplicity
judgments are involved in non-deductive inferences. The second section introduces the
fundamental motivations, terms and distinctions for a philosophy of simplicity. The philosophical problems of the principle of parsimony are introduced in this section. I also show
why it is difficult to bracket questions of metaphysics in analyzing or applying principles
of parsimony. The third section makes several suggestions about the roles that simplicity
judgments might play in theory construction and evaluation. These roles are not necessarily
involved with principles of parsimony. In the fourth section I show how deeply embedded
concepts of simplicity are in science. Many philosophers who work on issues involving
simplicity and parsimony have pointed to the Ancient Greeks when they think about the
archetypes for the philosophical dialectic involving these issues today. I find it useful to

12
organize the simplicity dialectic by doing just a bit of the genealogy of simplicity.
There is a secondary objective for this chapter. The chapter is dialectical in the
sense that it aims for conceptual clarification. In addition to introducing examples, terms
and motivations, I also launch a few arguments. I am critical of some of the popular approaches to a philosophy of simplicity. Many philosophers of science who investigate
issues involving simplicity appear to have come around to the view that progress in the
dialectic on simplicity in science has ground to a halt. I have wondered if this might be
due, in part, to how easily philosophical inquiry into simplicity is motivated. It is very easy
to see how judgments of simplicity could solve some of the biggest problems in philosophy
like: the problem of universals, the problem of induction, and questions about the nature
of the divine. Yet to solve these problems simplicity criteria would have to be clearly defined and the judgments of simplicity would have to be justified. Many philosophers are
overtaken by frustration on the issue of clarity and overtaken by skepticism on the issue
of justification. However the obsession with solving the big problems in philosophy might
just involve letting the smaller game slip away.
It is easy to gesture to a vast philosophical literature where opposing camps, bristling
with Ockhams bayonettes, charge one another with having committed the sins of complexity. Immaterialists and materialists might end up doing this on the issue of what an ordinary
sensible thing is. Berkeley, for example, appears to hold that it would be an unnecessary
complexity in Gods universe if material substance caused sensations when God can cause
sensations straight away. A materialist like the Bertrand Russell of The Problems of Philosophy might accuse the Idealist of introducing so many ad hoc hypotheses to systematize an

13
unwieldy metaphysics that it would be simpler to admit material substances to ontology.1
Trope theorists and universal theorists may stagnate in this bog as well. The trope
theorist might wish to keep an ontology free of extravagances like propositions or universals by positing only instances of properties. But the universal theorist charges that the
trope theorist is left either with the mystery of explaining what binds individual properties
together in a thing or with some universal binding relation. Perhaps the intuition is that
universals alone is a simpler view than tropes + binding relations or tropes + universals.
In the sciences this issue may be resurrected each time some party disputes the
foundations of an evolutionary theory committed to excluding final cause explanations,
and evolutionary theorists univocally snort, we need no additional causes. A similar
issue may have served as the basis of the dispute between Einstein and Heisenberg about
whether or not randomness needs to be introduced to the ontology of causation. Perhaps
these disputes are about whether or not we ought to prefer explanations that are simpler in
that they posit fewer kinds of causal mechanisms.
These are some of the heavyweight issues in metaphysics, epistemology, science
and theology no doubt! If arguments are not available to reduce disputes about simplicity
to disputes about other things then the philosophy of simplicity is well motivated. Yet these
exciting lines of inquiry may end up being a sort of distraction. It is easy to get swept
1 Some

of Russells claims appear to be obvious misinterpretations of Berkeley. So it is not

easy to see what Russell means when he says that, the way in which simplicity comes in from
supposing that there really are physical objects is easily seen. Russell goes on to say that every
principle of simplicity urges us to adopt the natural view, that there really are objects other than
ourselves and our sense-data which have an existence not dependent upon our perceiving them. I
base my interpretation of Russell on this passage and the fact that it is located in the Second Chapter
of Problems which argues against the idealism and immaterialism.

14
away by the desire to answer the skeptic or to solve the problem of universals and it is
especially easy to slip into these projects as we cavort in the neighboring dialectical space.
But the focus of this project must be to take a bite out of the philosophical problems of
simplicity; the problems involving the roles played by simplicity judgments in scientific
methodology. I have heard it said that when you are young it is fun to think that youll
generate the scorched earth argument that will leave the army of the skeptic decimated on
the battlefield. As you grow older, you settle for the skirmishes.2

2.1

Non-Deductive Inference

Elementary my dear Watson. Even people who have never read Sir Arthur Conan
Doyles novels know of the famous tag line uttered by the cocaine, heroine, nicotine, and
alcohol addicted genius detective who regularly shocked and amazed his side-kick Watson
with his forensic inferences. The line has always seemed to me like the sort of thing that
a bully would say. Doyle may have been employing a literary device to build the extreme
brilliance of the Holmes character by having him talk down to Watson, a well educated
physician. But for a non-fictional character, it would be rude to talk to ones friend in this
way. It has always struck me as if Holmes was happy to celebrate the fact that he had some
perhaps God-given access to a principle of inference which poor Watson (perhaps due to
his less fortunate breeding) lacked. Theres no point in explaining it. Its easy for me
but not for you, I understand Holmes as meaning. Were it Doyles plan to build such a
character, it would not be surprising given Holmes other traits since heroine and cocaine
2 Professor

Fales said something very much like this in my first graduate seminar in Epistemology. I cannot resist referring to it because I cannot think of a better way to put the point.

15
addicts are not famous for their politeness or humility e.g. Sid Vicious, Miles Davis, and
G.G. Allin.
Doyle often describes Holmes as performing one of his famous deductions. Given
his fabled intellectual brilliance, Holmes may well have known what Doyle did not: that
his arguments were not deductively valid. To illustrate this, consider an example from the
first chapter of The Sign of the Four (the entire selection is included in the appendix).
After briefly examining a watch Holmes makes the following statements which
are confirmed by Watson who happens to know the history of the watch. The watch has
recently been cleaned and the watch had recently been given to Watson by his brother. Big
Brother Watson had inherited the watch from their father many years ago.
Holmes says that he made these judgments for the following reasons: the H.W.
on the back suggests of the Watson family name, the initials are as old as the manufacture
date of the watch, it is customary for jewelry to be passed from father to oldest son, and as
Holmes recalls, Watsons father had been dead for many years.
Holmes then amazes Watson by saying of his partners late brother that,
He was a man of untidy habits very untidy and careless. He was left with
good prospects, but he threw away his chances, lived for some time in poverty
with occasional short intervals of prosperity, and finally, taking to drink, he
died. That is all I can gather.

In response to Watsons declaration that Holmes was correct in every particular,

Holmes says,
Ah, that is good luck. I could only say what was the balance of probability.
I did not at all expect to be so accurate.

16
Nevertheless, Holmes insists that he did not guess in drawing his conclusions. He
explains in a bit more detail the bases upon which his judgments were made. Watsons
brother was judged to be careless on the basis that the scratches and dings indicate to
Holmes that the expensive watch was kept in the pocket with keys and coins. Holmes adds
that Watsons brother must have then been careless in general. Holmes noticed several of
the sort of scratches which pawnbrokers customarily put on items to match with tickets and
inferred that Watsons brother was in financial trouble when he pawned the watch and in
better standing on those occasions when he redeemed it. Lastly, Holmes cites the many
scratches around the key hole as evidence of the unsteady hand of a drunkard who wound
the watch each night.
Before attempting to formalize Holmes inferences it is worth while to draw attention to what Holmes does not say. Many other details which could be relevant to the
inferences which Holmes draws are not specified. Holmes does not say anything about the
specific character of the watch scratches which are supposed to have been caused by keys
and coins. Keys and coins are of different shapes and are made of different materials. A
piece of iron can be cleaned with a brass brush, for example, and the soft brass will lightly
scratch the iron leaving a golden residue on the iron which looks different than a piece of
iron which has been brushed with aluminum. It is possible that some scratches might be
classified more precisely by saying something about the specific gold alloy that the watch
is made of and how various metals transfer material to that alloy. Coins of different sizes
might be expected to leave different sized impressions on the gold watch. An expert might
be able to distinguish between the marks left by a six pence piece and a shilling. Another

17
bit of specificity which Holmes does not give involves the possible idiosyncrasies of the
local pawn-brokers. Some, perhaps, are known to use more digits to match the tickets than
others do. In a way, the fact that Holmes omits these details contributes to the probability
that his inferences will be true. If he had been so bold as to hazard assertions about the kind
of money Big Brother Holmes spent or the specific pawn-brokers who handled the watch,
Holmes would have opened his hypotheses to possibility that some of his assertions might
be falsified. As it is, Holmes claims that he did not expect his assertions to be so accurate.
This could be due to the fact that Holmes could have been right about many of the general
facts about Watsons brother but wrong about some of the specifics.
Holmes has asserted three things about Watsons brother: that he was, in general,
careless, that he had experienced extreme ups and downs in financial security, and that he
was a drunk.

Holmes also explained what was the evidence for each of these assertions:

initials, scratches of various sorts, and customs common to his culture.

Deductively valid arguments are those arguments where the truth of the premises
guarantee the truth of the conclusion. Two classical rules of inference are modus ponens
and modus tollens:
1. If P then Q
2. P
3 Doyle

has Holmes offer a list of three inferences whose conclusions are: 1. Bad financial
times for BBW. 2. Good financial times for BBW. 3. BBW was a drunk. I find this list a bit odd
because financial ups and downs are a cycle which, if analyzed, amount to either ups or downs.
Drunkenness and gambling are the normal causes of this kind of financial history. Finally, though
not all alcoholics are careless, carelessness could be the cause of all of this. I have tried to carve up
the inferences as clearly as possible. At a certain level of analysis, it must be admitted that Doyle
probably does not provide the best case studies.

18
Therefore: Q
and
1. If P then Q
2. not-Q
Therefore: not-P .
But were Holmes to be giving arguments of this sort, his arguments would be, at
best, unsound because they would contain implausible or false premises. In order to get his
first two assertions by deductive inferences Holmes would need principles of the following
sort:
If a watch is scratched then it has been handled carelessly by the owner.
If a person is careless about one of his or her possessions then he or she is careless
with all possessions.
Principles such as these could be employed in a sequence of deductively valid arguments like the following:
1. If a watch is scratched then it has been handled carelessly by the owner.
2. The watch is scratched.
Therefore: The watch has been carelessly handled by the owner.
3. If a person is careless about one of his or her possessions then he or she is careless
with all possessions.
4. The pocket watch was one of the possessions belonging to Watsons brother.
5. Watsons brother was careless with his watch.

19
Therefore: Watsons brother was careless with all of his possessions.
The problems with this chain of reasoning are obvious. Firstly, these arguments
are unsound. Premises 1 and 3 are at best implausible and at worst false. A scratched
watch may not have been handled carelessly by the owner. After all, if Holmes is right that
the watch has a history of passing through the hands of pawnbrokers, then it might be the
case that the pawnbrokers handled the watch carelessly while it was in their possession. If
Holmes is right that drunkards mishandle the winding of watches in ways distinct from the
ways that cocaine and heroine addicts mishandle things, then there is still the possibility
that the live-in drunkard prostitute girl friend of Watsons brother was responsible both for
the key scratches and the frequent visits to the pawnbroker. Big Brother Watson may have
thought that the watch was in good hands and thus never got wise to its cycles. Doyles
readers might speculate about many other hypotheses which Holmes does not discuss in
connection with watch-scratches and wonder how it was that Holmes ruled out all but the
one he gave. Holmes never does consider the possibility that Big Brother Watson may have
been an epileptic and he never does consider that people other than Big Brother Watson may
have caused the scratches. Explanations for watch-scratches such as this might undermine
assertions about Watsons brother, his carelessness and his drunkenness.
Perhaps my initial versions of some of the premises are not the correct way to interpret Doyle. This could well lead to a worse situation for Doyles genius character, where the
arguments he offers are invalid by deductive standards. This would be the case if Holmes
was reasoning from effects to causes. I do not believe that the above principle stated in
premise 1 is really the correct interpretation of what Holmes asserted. I believe that Doyle

20
was attempting to model the inferences of Sherlock Holmes on the inferences characteristic
of scientific reasoning. One important component of scientific reasoning involves collecting data and then inferring what are the likely causes of the data. As a forensic investigator
Holmes does exactly this sort of thing. The phrase has been appearing in some of these
premises inverts what I believe to be an implicit causal relation which Holmes reasoning is
supposed to track. However, if these arguments are rewritten as expressing some principle
of causation and an inference from effect to cause then they are no longer merely unsound,
but they would be invalid by deductive standards of argument evaluation. If the correct
analysis of premise 1 is that it asserts a causal relation then it should read instead:
The careless handling of watch W causes watch W to be scratched.
But, if Holmes observes that the watch is scratched and infers that mishandling is
the cause, then he would have committed the fallacy of affirming the consequent. As stated
above, this is due to the fact that watch scratches can come about in many ways.
For these reasons, many readers of Sherlock Holmes novels have suggested that
Holmes was performing non-deductive inferences. The suggestions are that Holmes employed either the inductive inference or the inference to the best explanation.

2.2

Induction

The Eighteenth Century Scottish philosopher David Hume famously analyzed and
criticized inductive reasoning. Hume pointed out that many inferences essential to the
scientific process do appear to take the form taken by the argument:
1. If a watch is scratched then it has been handled carelessly by the owner.

21
2. The watch is observed to be scratched.
Therefore: The watch has been carelessly handled by the owner.
However, this argument should be analyzed differently so that it does not look like
an invalid deductive argument. Perhaps a better way to write this argument would be:
1. Past Observation Set 1: this watch has been carelessly handled by its owner and it
ended up scratched.
2. Past Observation Set 2: A different owner carelessly handled a watch and it ended
up scratched.
3. Past Observation Set 3: Yet another owner carelessly handled a watch and it ended
up scratched.
4. On the basis of uniform past experiences, I have come to expect that when a watch is
scratched, it has been caused by the careless handling of the owner.
5. I observe that the current watch before me is scratched.
Therefore: It is probable that the watch has been carelessly handled by the owner
(just like in every other case that I have observed)
This argument form would appear to fit Holmes reasoning. It may also be indicated
by Holmes references to what is customary and to what is on the balance of probability
that he is thinking of some principles which he has derived from more-or-less uniform cases
of past experience.
There are a couple of different ways to think about what was illuminated by Humes
analysis of the inductive argument form. One of Humes general projects may well have

22
been to partake in an Enlightenment Period movement to reign in an inflated view of human
cognitive powers which Europeans inherited from ancient thinkers.4 Some people, straddling ancient values and an optimism imparted by the successes of the scientific revolution
may have gotten to thinking that it was only a matter of time before the methodological
principles that have been so successful in science, would also solve the ancient puzzles of
metaphysics and perhaps even the Big One (the question of Gods existence) would be
settled as well. One unifying thread in Humes work may be his arguments that scientific
principles cannot offer answers of this sort. The theme of Humes Dialogues On Natural
Religion is to reveal the impieties of arguing for Gods existence on scientific principles
and this may be evidence that Humes arguments had specific Enlightenment characters as
their targets.
Hume suggests that we should turn the principles of science upon human thought.
But when Hume does this, he finds that the traditional problems of metaphysics are intractable. Hume argues that nothing in the content of sense observations themselves will
bestow upon us knowledge of the necessary connection between cause and effect. Nor do
sensations themselves reveal to us what kinds of things are causes. A pattern of repeated
observations will not, at some point, cough-up for the careful scientist the theory which
explains them. Hume can see no other explanation for how it is that humans carry out inductive inferences except that our volatile psychologies or habitual conditioning somehow
dispose us to form the expectation that the future will resemble the past in certain respects.
4 This

reading of Hume is suggested by Tom Beauchamp in his introduction to an edition of An

Enquiry Concerning Human Understanding.

23
It can be seen from the example above, that although watch scratches may have
been caused by the careless handling of watch owners in the past, there is no reason not
to expect watch scratches to arise due to the careless handling of watches by non-owners,
or by seizures or by micro-meteorites or.... This is the pattern of introducing competing
hypotheses loved by all skeptics. The pattern involves introducing alternative hypotheses
about the causes of observed effects, because induction cannot give us an argument that
any particular causes are any more likely to have preceded some observed effects than any
others. This is because the regularities of past experience alone give us no reason to expect
that things will not be different in the future. After all, past experience also informs us that
things change.
Humes result is discussed in the philosophy of science literature as the problem of
the underdetermination of theory by data. No accumulation of data (or cases or observations), would be sufficient to determine the theory that explains them. It is easy to think
about this situation with a geometric example. However, the phrase underdetermination of
theory by data can be a little misleading for the graphic example since it would be unusual
for a theory to consist of a single equation. For this example, it will be sufficient to discuss
the problem of the underdetermination concerning the functional relation between physical
variables by data. Since many theories contain such hypotheses it should be sufficient to
show how the problem arises for an essential component of theories. Consider a set of data
points represented on the Cartesian coordinate plane in figure 2.1.
Different functions describing the plot in figure 2.1 could be fit to the points. Consider the different graphs in figure 2.2, all of which are fit to the data points from 2.1.

24

Figure 2.1: Data Points

All of these graphs would be generated by different functions. Some of the functions appear to fit the data points more precisely (or we might say better) than others do.
Hypothesis h3 fits the data very well and it also has a few bumps and dips which do not correspond to any of the points of data presented. This helps to illustrate the fact that we could
fit infinitely many functions to a finite set of data points just by making the functions bumpy
in between points. Hypothesis h4 is a discontinuous or piecewise function. Infinitely many
functions of this sort could be generated as well to fit equally well any finite set of data
points.
Infinitely many hypotheses could be offered to fit the data. It is worth noting, however, that it would not be sufficient to say that a hypothesis is a scientific one just because
it fits some data. Scientific hypotheses involve several components. A scientific hypothesis
is never merely a mathematical equation. For one thing, the signs in the functional expression must be given a physical interpretation. Often the physical interpretation involves
positing underlying mechanisms which would, if the hypothesis is true, generate the data.
A functional hypotheses which fits some data may only show that the data is correlated.
But, correlation is not causation. Additionally, a scientific hypothesis will have to do much
more than merely fit and the data and offer a putative explanation of it. Many of the interre-

25

Figure 2.2: Curve Fitting

26
lated criteria that scientific hypotheses satisfy will be discussed in the next chapter. Here, I
wish to draw attention to the relation between the problem of induction and the fact that we
need more than just the ability to generate hypotheses that fit the data to justify the methods
employed to generate hypotheses.
Suppose that a hypothesis involving a mathematical description is given such that
data of this sort are caused in the way described by hypothesis hj . This hypothesis would be
competing in conceptual space with infinitely many other hypotheses which would cause
the data and also, with infinitely many hypotheses which would, if true, explain the disunity
of the data. Perhaps each datum is caused in a unique way and deserves a unique causal
explanation. Although it is not entirely clear to me exactly what Aristotle had in mind in his
famous proof of the unmoved mover in Book VIII of the Physics[4], he seems to suggest
that there may be infinitely many movers (which I take to mean causes) for states of affairs
which are realized in an infinite time interval. However, a crucial part of Aristotles argument is that some additional causal explanation must be given for what would unify these
unique movers in certain respects. My guess is that all movers would agree on basic things
like that effects always follow causes. Aristotles appeal to some additional principle about
unified causal explanations goes beyond merely explaining the sequences of events. I do
not know precisely what Aristotles additional principle was (or principles were). Perhaps
he had general methodological principles that endorsed unified explanations, or perhaps he
had reasons to posit just one mover because this best reconciled with his general systematic
thought - perhaps both. Whatever was the case, Aristotle had to reach beyond induction
alone to give a causal hypothesis.

27
The problem of the underdetermination of a hypothesis by data is twofold. 1) It
is difficult to see how one hypothesis can be selected when infinitely many curves will fit
the data; infinitely many smooth curves will fit the data and infinitely many discontinuous
curves will fit the data. 2) Even if we settle upon a mathematical or logical method by
which a single curve which fits the data may be selected over its competitors, this is still
not sufficient to justify the judgment that the data was all caused in the same way that
there is a unified causal explanation.
This is one way of illustrating the problem of induction in science. The problem
is one of determining which of infinitely many hypotheses to select for a given data set.
In addition to fit, what criteria must our scientific explanations satisfy? Aristotle appears
to argue in the Physics[4] that no hypotheses involving random causes will be physical
explanations, since he has argued that there must be some unifying causal explanation for
the laws of nature, and that chance (randomness) is not a cause.
There are two points which may serve to launch inquiry on the basis of these illustrations. It is a well entrenched feature of the scientific process that scientists collect data
and then fit curves to the collection. Firstly, it is often the assumption that curve-fitting
gives a function which does not merely describe correlations, but rather that the function
which is selected to fit that data will contribute to the formation of a causal hypothesis.
Secondly, scientists often fit curves to data imprecisely to some degree - this represented
better by h1 and h2 than by h3 or h4 in the examples above. But some scientific principles
must constrain the sloppiness of curve fitting. Hypotheses can get so vague that they are
never false. Fortune tellers have mastered the art of sloppy curve fitting. If Doyle success-

28
fully modeled the inferences of his character on the usual methods of science then he may
have left some clues for the reader that he did so. Holmes career as a forensic investigator
would demand of him that he resist the sirens song of the skeptic, and would force him to
reason from effects to causes and to choose just one explanatory hypothesis from infinitely
many possible hypotheses equally compatible with the data.
On this view of Holmes as an exemplary scientist, the questions we ought to ask
of Holmes are the questions that we ought to ask of all scientists. 1. What, if anything,
would justify hypotheses which track causation if induction cannot give to us an idea of the
necessary connection between causes and effects, but only an idea of their being correlated
in the past? 2. What, if anything, would justify selecting just one hypothesis from among
infinitely many that would also account for the data? 3. What, if anything, would justify a
certain degree of inaccuracy in hypothesis formation and what would constrain hypotheses
from becoming incorrigible due to vagueness?
The first two questions are the questions raised by Hume. The third question has
taken center stage in contemporary debates regarding curve-fitting algorithms and it is
likely that the popularity of computational modeling in science has contributed to the recent spotlighting of this topic. Judgments of simplicity may play roles in answering each
of these questions. However, simplicity has not yet been elucidated sufficiently to discuss
the relations between various types of simplicity judgments and their roles in constructing
and selecting hypotheses. Elucidating the various species of simplicity judgments and their
relations to the desiderata of science is the goal of the next chapter. The first two questions
have been taken by many to launch inquiry in classical epistemology.

29
Consider the two components of Humes criticism of induction. The first component involves pointing out the problems of the underdetermination of theory by data. The
second component involves pointing out that justifying induction on empirical principles
involves a circularity. Whether we reason from causes to effects or from effects to causes,
the problem of underdetermination by data presents a problem for non-deductive reasoning.
There is a version of inductive skepticism with some sobering consequences: the
problem of skepticism with respect to the external world. Just change the above examples
around a bit in order to generate this problem. Say that instead of data points we have a
collection of experiences and say that instead of algebraic hypotheses we have hypotheses
about the causes of our sense experiences. Now the trick is giving an argument that will
justify inferring that material objects cause our sense experiences instead of evil spirits, or
Matrix-style insect-shaped robots with sophisticated hardware and software systems5 , or
the mind itself in a dream state. All of these are hypotheses which would (at least it seems
to some philosophers) explain the sequence of our perceptions and this is the problem of
the underdetermination of theory by data. Additionally, it will do no good to say that in
every case in the past our sensations have been caused by material objects external to our
minds. We cannot get outside of experience, as it were, to check what are the usual causes
of experience. It seems that we have only past experience to go on to justify the method of
inference. This is the problem of circularity.
Now if it turned out that we had some idea of the necessary connection between one
cause-effect pair, we could reason from cause to effect in that kind of case. However, we
5 The

reference is to the 1999 film The Matrix

30
still would be unable to reason from effect to cause. This shows that both the problems of
underdetermination by data and the problem of circularity apply to reasoning from cause to
effect. But reasoning from effect to cause is crippled by the underdetermination of theory
by data alone.
The strategy may not be acceptable to all philosophers, but it may be possible to
keep some dialectical separation between the questions for the philosophy of science and
the general problem of skepticism. Some people may just give up on knowing anything
about what causes of our sensations (the dinge an sich) and content themselves to discover
whatever science might tell us about the general structure of the mind or the matrix or the
whatever it is that causes sensations. This acquiescence to the Cartesian skeptic will not
side-step all of the problems of induction. The questions about what justifies the selection
of one hypothesis over its competitors remain for science. Perhaps this turns into the question of which concepts or cognitive structures are in the mind. Some will also argue that
the Cartesian skeptic lurks at the terminus of even this dialectical path. That is, the question
of what justifies the methods or the principles by which we proceed to reveal what are the
fundamental cognitive structures which are somehow filled out by experience poses itself
even to thinkers like Berkeley or Kant.
Of course if there were some principled way to judge which, among competing hypotheses, is most likely to be true, then some (if not all) of the questions about what justifies
inductive inferences might be answered. The notion that simpler hypotheses are somehow
more likely to be true than more complex competitors has suggested itself to many thinkers.
Some people find it intuitively plausible that the so-called Real World Hypothesis is simpler

31
than the evil demon hypothesis. This might be correct for all I know, but I will not weigh in
on this field here. I can imagine how arguments could be given to show that one scientific
hypothesis is simpler than another. Formal apparatus may be available to gauge the relative complexity of logical or mathematical hypotheses. The problem for the philosophy of
science would be to show the that simplicity of a scientific hypothesis is somehow related
to the truth in such a way that it is made more likely than its competitors. This problem
is the same as the general problem of external world skepticism in that we cannot argue
on inductive principles that the truth is usually tracked by hypotheses which are simple in
this or that way, because such an argument would involve checking what the relationship
between simplicity and the truth is, but this is exactly what demands an argument.
Perhaps the fun of the Sherlock Holmes novels has to do with the fact that Holmes
is supposed to have a special divine gift, and like a modern Job or Moses, he wrestles
with his human imperfections. Perhaps Sherlock Holmes has immediate intuitive access
to the degree of simplicity of hypotheses. Perhaps he has an even more grand gift and has
knowledge of the relation between simplicity and the truth. Instead of telling us what this
relation between simplicity and the truth is, Holmes just laments the industrial-age version
of the problem of evil: the problem of boredom. Perhaps it is at least clear that there are
two lines of inquiry which might contribute to solving the problems of induction: 1. the
question of how to classify and gauge the simplicity of hypotheses, and 2. the question of
what the relationship between simplicity and the truth might be.

32
2.3

Inference to the Best Explanation

A different interpretation of Doyle is that Sherlock Holmes inferences are modeled

on the the style of argument which takes the form of an inference to the best explanation.
Very basically the inference to the best explanation might take the form:
1. Observation set O.
2. If hypothesis h were true, h would best explain O
Therefore: probably h.
This formulation is a bit too basic though, because this characterization may not
clearly distinguish this form of argument from induction. The second premise suggests
that h is compared to competitors so it should read If hypothesis h were true, then it
would be better than any of its competitors at explaining O. If the observation set turns
out to be just a bunch of uniform cases, then this argument may be indistinguishable from
induction. So, to make this argument form unique it should be open to contain observation
sets which are not necessarily uniform.
1. Observation set O.
2. Observation set O0 .
3. Observation set O00 .
4. If h were true then it would be better than all of the (infinitely many) hypotheses
which give a unified explanation for the bunch of observations which would otherwise be a totally disconnected set of coincidences. Therefore: h is probably true.

33
Considered this way the inference to the best explanation certainly appears to be
unique. If we come home and open the front door to see that the sliding glass door is
broken, there are muddy shoe prints going both directions between the broken glass and
the entertainment center and the high fidelity stereo is missing, then most of us make an
immediate inference as to what must be the explanation for all of this that the stereo has
been stolen.6 Most people immediately discount the neighbor with the tin foil hat who
interrupts his ravings about the Mayan calendar only to say that he has no idea what caused
the broken glass or the stereo to go missing, but the footprints are evidence in favor of his
alien invasion hypothesis. After calling the police to report the stolen electronics it may be
tempting to explain to the neighbor that his offering of the disunified hypothesis is, once
again, evidence that he is not rational. But this would be a waste of time. There is nothing
crazier than reasoning with the insane.
The example of a probable stereo theft is an example of a forensic inference. This
example can easily be extended to the type of inferences made by Sherlock Holmes. Holmes
may have a set of uniform experiences (say, observation set O) about how people customarily pass watches on to eldest male offspring. He may have another uniform set of experiences (observation set O0 ) involving the behavior of drunks and another (observations
set O00 ) involving the behavior of pawnbrokers and so on. Holmes may infer that the best
explanation for this, otherwise heterogeneous, set of data is that Watson had a reckless,
drunken older brother.
6 This

basic example has been presented to students in Richard Fumertons classes. I picked it
up as a teaching assistant.

34
Of course it does no good to let the word best bear the weight of all the really
nasty philosophical problems of non-deductive inference. Of course we might object that
the word best is philosophically loaded in this deployment. The introduction to philosophy student can easily challenge the view, Does best h mean that I like it? Does best
h mean that it gets me access to fancier parties? Does best h mean that h is more likely
to be true?
Despite the fact that philosophers may not yet have generated replies which will
satisfy even the most rudimentary objections to the argument form involving the inference
to the best explanation, the form may well capture what scientists actually do. For that
matter, the argument form may well capture what most of us (non-scientists) do on a daily
basis. Again, simplicity may play a key role in the analysis of the inference to the best
explanation. If, for some reason, simpler hypotheses are more likely to explain what unifies
a set of data, as opposed to leaving that data as a wild set of coincidences, then perhaps we
can justify judgments about what the best explanation for a given data set would be. The
philosophical questions regarding this type of inference are launched in the same way as
are the questions regarding induction: 1. How do we gauge simplicity? 2. What would be
the relation between simplicity and the truth? 3. Why would it be a mark of rationality or
of our physical explanations that they give unity to experience?

2.4

Motivations, Terminology and Distinctions

When I first tell people that do I research on the role of simplicity in science, two reactions are most common. One reaction involves something like a shoulder-shrug, I guess

35
I never thought about that being important to science. The other reaction is disgust, I
think there really is no philosophical problem on this matter. Both the shoulder-shrug and
the retreat to nave pragmatism indicate suspicions about the motivations for a philosophy
of simplicity. It will not satisfy people harboring such reservations to point to a long scientific tradition where in scientists have both employed and endorsed principles of parsimony.
Scientists may have been mistaken, unclear, talking past one another or philosophers may
have misinterpreted them thinking that simplicity played a role which no scientist took it
to play.
Judgments of simplicity might solve the problems of non-deductive inference. In
this way a philosophical investigation of simplicity is motivated at least minimally. But
three typical types of assertions are nevertheless given in resisting the philosophy of simplicity: 1) that simplicity is a hopelessly vague concept, 2) that simplicity reflects nothing
more than pragmatic or aesthetic values, and 3) that judgments of simplicity are unlikely
to be found playing any important role in the scientific method. But such claims demand
arguments! For my money, some of these claims might just be true. It would be nice to
know if arguments might be marshalled for or against these views.
Philosophers recognize three distinct projects for a philosophy of simplicity: 1)
clarification 2) justification 3) application 7 . There are philosophical reasons for considering what role, if any, some criterion of simplicity might play in governing non-deductive
inferential judgments. There are also reasons for considering the roles played by judgments
of simplicity in hypothesis construction and testing. Simplicity is subject to vagueness and
7 Alan

Baker parses the discourse this way in the Stanford Encyclopedia article on Simplicity[5]

36
therefore might stand to benefit from a bit of analytic clarity. So, the project of organizing, categorizing and clarifying species of simplicity well motivated. I intend to show that
it is very difficult either to ignore or to analyze away the roles of simplicity in scientific
methodology and this leaves the epistemological questions as the really interesting ones.
These issues will be discussed in the fourth chapter.
If simplicity can be clarified and categorized, then perhaps an epistemological project
is possible. The epistemological project is to investigate which species of simplicity, if any,
might be justified and if so, in which ways. Pragmatic justification for simplicity judgements in specific roles may be easy to come by. The real problem would be to show that
there is some relation between a particular species of simplicity and the truth.

2.4.1

The Venerated Tradition of Simplicity

I can at least guess why Christian Europeans cared about simplicity. Thinking about
their preference for simplicity reveals something about why so many Early Moderns were
systematic thinkers. The rough idea was that simplicity was a divine attribute. God made
the world resembling Gods nature and God made human minds also resembling Gods
nature such that humans could appreciate creation. I imagine that if simplicity is a feature
of human cognition and if it is also a feature of the world then there might be a causal
relation between features of the mind and features of the world. Such a relation would be
exactly what one would expect if simplicity judgments of some sort were justified.8
There may be a great deal to say just about the role of simplicity in the thought of
8A

proper philosophical investigation into the role of simplicity in the methodologies of Early
Modern thinkers is a project that I intend for the future.

37
some philosophers, but also a great deal might be said about the ways some philosophers
inherited methods involving simplicity and then changed them somehow. In other words,
there may be an illuminating story to be told about the morphing principles of parsimony in
the period of the Copernican revolution. Here, I will point out just a few examples to show
that major figures in this period probably saw the relationship between the human mind,
the world, and Gods simplicity as some fundamental commitment of theory construction.
Descartes thought that simplicity was a divine attribute and that the only way this
idea is caused to be in his mind is by God. In the Third Meditation Descartes considers
the possibility that his idea of God was brought about by partial causes. The supposition
here being, Descartes says that all the perfections are to be found somewhere in the
universe but not joined together in a single being, God. But Descartes abandons this
hypothesis because the divine attributes are a package deal so-to-speak. On the contrary,
says Descartes the unity, the simplicity, or the inseparability of all the attributes of God is
one of the most important perfections which I understand him to have. ([20]pg. 34)
Leibniz opens the Discourse on Metaphysics saying that God has all of the perfections and nature has some of them. He says this in a way which suggests that nature
resembles God.
The most widely accepted and meaningful notion we have of God is expressed well enough in these words, that God is an absolutely perfect being;
yet the consequences of these words are not sufficiently considered. And, to
penetrate more deeply into this matter, it is appropriate to remark that there are
several entirely different perfections in nature, that God possesses all of them
together, and that each of them belongs to him in the highest degree. ([44] I.)
Leibniz also appears to hold that the perfections of nature and of God can be known by
humans. Lloyd Strickland places Leibnizs views about simplicity in a larger historical

38
context also.
Now we will recall that richness of phenomena was only one of the elements in Leibnizs characterization of the most perfect world, the other being
simplicity, or simplicity of hypotheses. By identifying simplicity as a criterion
of worldly perfection Leibniz appropriated another idea already in circulation,
this time incurring a debt to his contemporary Nicolas Malebranche; for the
doctrine is present in many of Malebranches works, from the sprawling Search
after Truth (1674) onwards, though it was in that work that Leibniz apparently
first came upon it.
Interestingly, Malebranche touted the idea of simplicity for precisely the
same reason that some earlier thinkers had promoted the idea of plenitude: it
was the best way for a completely perfect being to express itself. In Malebranches view, God acts only for His glory, and is therefore determined to
will that work which could be produced and conserved in those ways which,
combined with that work, would honor Him more than any other work produced in any other way. Therefore, He formed the plan which would better
convey the character of His attributes, which would express more exactly the
qualities He possesses and glories in possessing. ([58] pg.67)
Bishop George Berkeley motivates his empiricism on the basis that God would
not have created a universe and yet leave humans ill-equipped to understand it. One of
Berkeleys most famous passages is the third entry from the introduction to A Treatise
Concerning the Principles of Human Knowledge.
But, perhaps, we may be too partial to ourselves in placing the fault originally in our faculties, and not rather in the wrong use we make of them. It
is a hard thing to suppose that right deductions from true principles should
ever end in consequences which cannot be maintained or made consistent. We
should believe that God has dealt more bountifully with the sons of men than
to give them a strong desire for that knowledge which he had placed quite out
of their reach. This were not agreeable to the wonted indulgent methods of
Providence, which, whatever appetites it may have implanted in the creatures,
doth usually furnish them with such means as, if rightly made use of, will not
fail to satisfy them. Upon the whole, I am inclined to think that the far greater
part, if not all, of those difficulties which have hitherto amused philosophers,
and blocked up the way to knowledge, are entirely owing to ourselvesthat we
have first raised a dust and then complain we cannot see.([6] pg. 6)

Additionally, Berkeley argues in various places that material substance would be a

39
needless extravagance of Gods creation since an all powerful being could certainly cause
sensations in us directly.
If therefore it were possible for bodies to exist without the mind, yet to hold
they do so, must needs be a very precarious opinion; since it is to suppose,
without any reason at all, that God has created innumerable beings that are
entirely useless, and serve to no manner of purpose. ([6] pg. 40)
Many of the major contributors to Western thought in the period of the scientific
revolution appear to have shared a certain family of views which involved the resemblance
of the simplicity of the universe to Gods simplicity and that Gods universe is knowable
for humans. If such a view is correct, then judgments of simplicity may at least be a guide
to true judgments of ontology. This view is the situation that we ought to question9 . Firstly,
Gods simplicity is a matter for serious theological debate. God might just have a view
of beauty more in common with Jackson Pollock or Charlie Parker. God may have slung
the world together in total defiance of structure so that the loveliness of creation is that it
hangs together on the very edge of chaos. Certain naturalistic arguments for the existence
of God would do better on this theology anyway. This is the sort of thing which Hume
points out in The Dialogues on Natural Religion. What some people see as an impious
result of the Design argument is that it makes God similar to humans. God may have
splattered the primordial stuff onto the canvas of very minimal laws just to impress deities
in other more simple universes. Some philosophers and theologians appear to embrace
this notion. William Paley appears to be committed to accepting that God is similar to
9 Elliot

Sober makes similar points about the history of the role of parsimony in science in Reconstructing the Past; Parsimony, Evolution, and Inference. Sober also says that in this century
the idea that parsimony must involve ontological commitments has fallen into disrepute, and that
parsimony can be treated as a purely methodological principle. He argues against this modern view.
See Sober pg.37

40
humans in some ways. He defends the Design Argument by arguing that the universe is a
teleological system just like a watch is. However, by the rules of analogical reasoning, the
inference would only be a strong one if God resembles human designers also 10 .
Pending the results from theologians about the best way to know what are the divine
attributes, it is not yet clear to me that positing a chaotic universe is an impiety. Even if
God is simple and so is Gods creation, then we still face philosophical puzzles. Unless
Spinoza is right and the universe and God are to be identified then the universe would at best
resemble Gods simplicity in some respects and not in others. The philosophical project
remains to establish which, if any, judgments of simplicity correspond to Gods plan. Even
on the chaotic universe hypothesis it is difficult to avoid considering simplicity. Saying that
the constraints for Gods post-modern styled universe were very minimal suggests a kind
of simplicity too, and we might even wonder if darn-near-to-chaos is in some way simpler
than a number of highly structured moral and physical laws.
There is also a naturalistic hypothesis which would be compatible with some simplicity judgments being justified. It has been said that the natural history of human kind is
red in tooth and claw but it is also spattered with clandestine affairs. Humans have always
been busy trying to outwit one another to secure power, wealth and preferable breeding
rights. Succeeding in the hominid breeding pool takes some rather clever maneuvering. On
this matter, I could not disagree more with Patricia Churchland: mere chance just wont
cut-it at my usual watering holes.11 Humans have become very successful at reading subtle
10 See

William L. Rowes Philosophy of Religion: An Introduction for a discussion of the various

versions of the Design Argument which have been defended.
11 Patricia

Churchland says that, from an evolutionary point of view, the principal function of

41
behavioral cues, anticipating responses and moving to head off competition for breeding. If
gripped for even a moment by the temptation to say that hominid courtship is not this ferocious, just stand on a street corner on any Friday night in a college town. The roving packs
of males are in the market for romance but their behavior is indistinguishable from what it
would be if they were going to a fight! Since clever maneuvering may play a key role in the
long run for who gets the breeding rights and who does not, there is reason to think of the
cognitive capacity for reasoning causally in the social arena as playing an important evolutionary role. On this hypothesis, next to Cleopatra, Cerano was a sissy. Perhaps, humans
might never be expected to have true beliefs about quantum physics, gravity, or even about
the shape of the Earth, but they might be expected to get very good at judging cause and
effect relations of some sort. Patricia Churchland appears to assume that evolution exhibits
alethic blindness, but this assumption is not justified in some obvious way.
The question of which simplicity judgments might be justified remains whether we
think that God made the world in a certain way or that humans evolved to navigate it in
the often horrendous ways that we do. On the eighth day could God have said, Let ye
who is the meanest proliferate? So, a naturalistic hypothesis may give reason to think that
human minds came to have features which correspond to the features of reality in such a
nervous systems is to enable the organism to move appropriately. Boiled down to essentials, a
nervous system enables the organism to succeed in the four Fs: feeding, fleeing, fighting, and
reproducing. The principal chore of nervous systems is to get the body parts where they should be
in order that the organism may survive. Insofar as representations serve that function, representation
are a good thing.(pg. 548-549) Churchlands account seems to leave out something very important.
It is reasonable to think that social interactions provided the radical evolutionary pressures necessary
for the human brain to get as folded as it did in what anthropologists tell us was a stunningly short
period of evolutionary time. Churchland argues in favor of a connectionist model of the mind in this
article, and this may still be correct.

42
way that some judgments are true judgments. A theistic hypothesis might do the trick also.
But, in either case, questions of the justification of simplicity judgments demand first the
clarification of simplicity.

2.4.2

Ockhams Razor

The first issue deserving clarification is the famous methodological principle known
as Ockhams Razor. For many years it has been rumored that Ockhams Razor (or Occams
Razor) is a fundamental principle of science. Roughly, the idea gestured at by the moniker
Ockhams Razor is that other things being equal, simpler hypotheses are more likely to
be true than are their competitors. Ockhams Razor is usually associated with the slogan
Entia non sunt multiplicanda praeter necessitatem (entities must not be multiplied beyond
necessity). J.J.C. Smart probably had in mind either some scientific principle, or at least
a heuristic principle, when he said that he wished to resist saying that statements about
subjective mental states are reports about irreducibly non-physical things mainly due to
Occams razor.[53]
I prefer to avoid using the term Ockhams Razor. It contributes confusion to an
already convoluted discourse. First of all, the principle is named after the Thirteenth Century Franciscan friar William of Ockham. William of Ockham never said Entia non sunt
multiplicanda praeter necessitatem. Not only is the slogan not to be found in Ockhams
work, but Roger Ariew has argued that this principle is incompatible with Ockhams theology [3]. On Ockhams theology, Gods omnipotence includes Gods ability to make all the
entities God wants to make.

43
Ockhams Razor is also subject to the vagueness that any principle of parsimony
is subject to. The principle involves an appeal to simplicity but the word simpler can
be interpreted in many different ways. The slogan often associated with Ockhams Razor
suggests a constraint of some sort placed on inferential judgments of ontology. It suggests
that we count the number of entities posited by a theory. The notion is that the principle
governs our judgment that the hypothesis which posits the smaller number of entities is the
hypothesis which is more likely to be true. It would be unusual, however, for a philosopher
to invoke Ockhams Razor intending that it be understood this way. Most of the time,
philosophers do not mean to count the number entities posited by a hypothesis, but the
number of kinds of entities posited by a hypothesis. Perhaps J.J.C. Smart intended this,
or it was a consequence of a more general principle. The universe may contain infinitely
many entities, but perhaps there are a finite number of kinds of entities. Similarly, there
may be infinitely many properties or relations instantiated in the universe, but only a finite
number of kinds of properties or relations. For these reasons I prefer to avoid using the
term Ockhams Razor altogether. Some careful philosophers (like Smart) probably do
have something precise in mind when they used the term, but I fear to say that some people
have fallen into careless habits. We might try to avoid propagating unnecessary confusion
by stating these principles as clearly as possible whenever they are discussed or invoked.
Ockhams Razor would be a version of what is more generally a principle of parsimony. I suggest that we analyze the more general concept.

44
2.4.3

The Anatomy of Parsimony

Suppose that we have a methodological principle like the following:

Principle: 2.4.1. Other things being equal, if hypothesis h1 is simpler than competing hypotheses h2 , h3 , ..., hn , then h1 is more likely to be true than its competitors.

Principle 1.2.1 is an example of what would be called a principle of parsimony.

Consider the following problems with this principle. Such a principle cannot be put into
practice. Even this formally presented principle is not precise enough to be employed.
The word simpler is subject to ambiguity. Also, it is not clear what would justify such
a principle that is, it is not clear why we ought to expect there to be any relationship
between simplicity and the truth. There is further a practical problem for the application of
this principle in science, since the ceteris paribus (other things being equal) clause is never
satisfied in actual scientific cases.
Sometimes the words parsimony and simplicity are treated as if they are interchangeable12 . I suggest that this should be avoided. See the Appendix for the detailed
defense of the view that these terms should be kept distinct. Roughly, the reason is that
parsimony is a methodological principle and simplicity is a state of affairs, or property. The
principle of parsimony involves a criterion of simplicity13 . But what does involves mean?
To make a particular principle of parsimony sufficiently clear to distinguish it from
other principles of parsimony and to make it possible to apply it to hypothesis selection,
12 Alan
13 This

Baker mentions this in the Stanford Encyclopedia article [5]

discussion follows closely Sobers second chapter in Reconstructing the Past; Parsimony,
Evolution, and Inference 2002[54]

45
simplicity criteria must be defined for a principle. So a few notes on criteria are essential for
this discussion. It is rather difficult to find philosophical discussion devoted to an analysis
of criteria, so I follow the lead of Stanley Cavell from the first chapter of The Claim of
Reason[15].
Firstly, criteria sometimes govern judgments, and sometimes judgments govern criteria. Secondly, criteria are object-specific. What makes a bridge unstable is not the same
as what makes a person the president of a club, or what makes one recognize a chess piece
as being a queen14 . Thirdly, Cavell says that criteria may be epistemic (they tell us what
counts as knowledge or as evidence) or they may be ontological (they tell us what a thing
is). I might add that criteria may also be pragmatic (telling us what is useful for some purpose or what requires the least amount of effort under certain circumstances). Perhaps also
there are aesthetic criteria. My guess is that utilitarians must at some point identify what
counts as suffering, so there may be phenomenological criteria as well.
Consider the way that criteria would govern judgments in our present case. The
principle of parsimony would govern non-deductive inferential judgments. Defining simplicity criteria for a principle of parsimony would tell us which features of our candidate
hypotheses are to be compared. There are two ways that judgments may govern criteria.
We may judge to what degree criteria are satisfied and we may also make judgments about
which criteria to define for a particular principle of parsimony.
A helpful way to start thinking about the clarification project in the philosophy of
simplicity is in terms of the clarification of simplicity criteria. Simplicity criteria define the
14 [15]

(pg.15)

46
bases upon which simplicity judgments are made. Our judgments of simplicity are comparative judgments. In comparing hypotheses, we have some simplicity criteria which specify
the objects of comparison. In order to apply a principle of parsimony at least one criterion
of simplicity must be defined. We compare the relevant features of the hypotheses in question and judge to what degree the criterion is satisfied in each case. Once the hypothesis
that is simpler than its competitors is identified the principle of parsimony would govern
the inferential judgment about which hypothesis is the most likely to be true.
Now the problem is that simpler than is subject to ambiguity. The reason is that
there are many different simplicity criteria which may be defined. One natural way to think
of the word simpler is that it is related to the word fewer. This suggest that simplicity
criteria could be defined on the basis of some counting project. But there are many ways
to count objects in the world and many ways to count the features of hypotheses. Consider
this short list of possible simplicity criteria:
fewer numbers of entities
fewer kinds of entities
fewer numbers of relations
fewer kinds of relations
fewer numbers of terms
fewer kinds of terms
fewer ad hoc posits
fewer axioms

47
fewer sentences
fewer logical connectives in sentences
fewer kinds of logical operators in a sentence (and, or, not)
fewer modal operators
fewer kinds of modal operators
fewer nested modal operators
This list is far from exhaustive. But in this way we can begin to see the massive
ambiguity which principles of parsimony might be subject to. There are many ways to
define simplicity criteria. This does not show that there is a deep philosophical problem
plaguing the analysis of principles of parsimony. We can just be clear about what the
simplicity criteria are. Nor are there any obvious formal problems involved in applying
principles of parsimony to theory construction or evaluation. It is not a problem to apply a
principle of parsimony in science provided that the simplicity criteria are clearly defined.
Isaac Newton, gave four rules of reasoning which he offered in Book III of the
Principia Mathematica.

15

1. We are to admit no more causes of natural things than such as are both true and
sufficient to explain their appearances. To this purpose the philosophers say that
Nature does nothing in vain, and more is in vain when less will serve; for Nature is
pleased with simplicity and affects not the pomp of superfluous causes.
2. Therefore to the same natural effects we must, as far as possible, assign the same
15 Reprinted

from Sober 1998 pg.52. Originally from Newton 1953, pg.3-5

48
causes. As to respiration in a man and in a beast, the descent of stones in Europe and
in America, the light of our culinary fire and of the sun, the reflection of light in the
earth and in the planets.
3. The qualities of bodies, which admit neither intensification nor remission of degrees,
and which are found to belong to all bodies within the reach of our experiments, are
to be esteemed the universal qualities of all bodies whatsoever. For since the qualities
of bodies are only known to us by experiments, we are to hold for universal all such
as universally agree with experiments, and such as are not liable to diminution can
never be quite taken away. We are certainly not to relinquish evidence of experiments
for the sake of dreams and vain fictions of our own devising; nor are we to recede
from the analogy of Nature, which is wont to be simple and always consonant to
itself.
4. In experimental philosophy we are to look upon propositions inferred by general
induction from phenomena as accurately or very nearly true, not withstanding any
contrary hypothesis that may be imagined, till such time as other phenomena occur
by which they may either be made more accurate or liable to exceptions. This rule
we must follow, that the argument of induction may not be evaded by hypotheses.
I will leave it to the Newton scholars to tell us exactly what Newton meant by each
of these assertions. Context does suggest some ways in which Newtons principles might
involve specific simplicity criteria. Perhaps (1) and (2) reveal a preference for theories involving fewer kinds of causes. Principle (4) may suggest a principle of parsimony which
selects against theories with the larger number of ad hoc hypotheses. The third principle

49
is left vague by this limited context alone. Principle (3) might suggest of some sort of
property nominalism, or perhaps simplicity criteria which select epistemic objects. Perhaps this principle encourages the scientist to posit few qualities of bodies that go beyond
what we access by way of sense experience. Consider again what has been said about the
general package-deal God might give to the world in creation. It would be expected that
the Early Modern philosophers mentioned earlier would think that certain criteria defining
what counts as knowledge or understanding would naturally be related to ontological criteria due to their theological commitments. It appears likely that Newton held views of this
sort as well.16
There is something else which is important to notice about Newton and parsimony.
This is that Newton is simultaneously committed to a universe which is simple in some
respects and complex in others. Newton gave just one law governing the motion of bodies
by gravitation. In taking there to be only one type of law governing gravitational motion and
in taking there to be many particles in the universe it is a logical consequence of Newtons
view that he was committed to plenitude with respect to instances of this relation. Every
particle stands in the F = G Mr2m relation to every other particle.
2.4.4

Ockham Revisited

It is difficult to interpret Ockahm as holding a razor that urged us to posit few numbers of entities. Newton did not hold such a principle either. How, then, did confusion
about Ockhams Razor arise? This is one of the subjects of Roger Ariews 1976 disserta16 Elliott

Sober discusses Newtons theological justification in Reconstructing the Past; Parsimony, Evolution, and Inference. (pg.53-54)

50
tion. Some people have identified Ockhams principle of absolute divine omnipotence with
Ockhams Razor. But this is a principle of possible plenitude. Ariew says that this principle
says that it was Ockahms view that, all things are possible for God save such as involve a
contradiction.([3] pg.6)
Most of Ockhams arguments are directed against John Duns Scotus. They are
theological arguments (and this point, for various reasons, has been muddled by historians).
However, Ockham did hold principles of parsimony. Ockham held the principle frustra
fit per plura quod potest fieri per pauciora. This means in vain we do by many that which
can be done by means of fewer. He also held the principle non est ponenda pluralitas
sine necessitate which means pluralities ought not be supposed without necessity. On the
surface these do seem like Entia non sunt multiplicanda praeter necessitatem but Ariew
argues that they are different.
Entia non sunt... appears to be a rule about entities (or real things) whereas
the others do not. At the very least, the other two formulations can lend themselves to another interpretation. They are not bound to entities. Frustra fit...
seems to be a rule about our explanations and pluralitas non est ponenda...
seems to be a rule about statements or concepts.([3] pg.17)

2.4.5

Hume Revisited

In thinking about how the general problem posed by the inductive skeptic, it is clear
how a principle of parsimony might be applied. Hume suggested variously that people form
some habit which explains how they make inferences, and that there may be some unity
of nature principle which governs inductive judgments. There is an apparent duality in
Humes account. On the one hand our inferences appear to be explained by habit and on
the other by a principle that experience, it would seem, cannot justify. Sober reconciles

51
the apparent duality of Humes account of induction saying that Hume does not say that
the unity of nature principle never plays a role in inductive reasoning. Perhaps people are
creatures of habit most of the time, but sometimes a principle of unity governs inductive inferences [54](pg.41). Unity is the limiting case (or lower bound) for simplicity judgments.
Unity is perfect simplicity. So, we can see how a principle of simplicity may be involved
in inductive reasoning.
I will mention briefly here the problem of justification for principles of parsimony.
In every case there is a very serious philosophical puzzle about what would justify a specific
principle of parsimony. Humes creatures of habit are irrational, and inductive arguments
to justify the unity of nature principle are viciously circular. I give a detailed analysis of the
types of arguments which might be given to justify principles of parsimony or the selection
of simplicity criteria in the fourth chapter. What is interesting to notice is that the inductive
skeptic, on this particular issue, does not force the epistemological project to the highest
philosophical priority but rather, pushes metaphysical issues into the foreground. This is
Sobers main point in Chapter 2 of Reconstructing the Past; Parsimony, Evolution, and
Inference. Principles of parsimony often make substantive claims about the way the world
is by involving ontological simplicity criteria. Even Humes unity of nature principle does
this and Humes view that people naturally form irrational habits suggests that there are
mechanisms of the mind which cause this irrational behavior. In either case the metaphysical issues are forced to the foreground. So, the project which philosophers are faced with
is one of coming up with a way to categorize and analyze simplicity criteria. Mario Bunge
has given us the best way to do this. How simplicity criteria are categorized and analyzed

52
is the subject of Chapter Three.
Another problem about principles of parsimony in science arises because it is very
difficult to give clear examples of the principle at work in science. It is notoriously difficult
to find a case where the ceteris paribus clause is satisfied. There is a problem getting a
handle on the notion of what it would be for hypotheses to be competitors in actual cases in
science. Often we expect that scientific hypotheses do not merely describe a data set, but
that they explain it. So, we can imagine that the phrase other things being equal suggests
that two competing hypotheses would both equally well explain the same data and there
would be no way of deciding which hypothesis to select other than by some appeal to
simplicity. Although this is conceivable, it is difficult to find unproblematic examples of
such a principle at work in the history of science.
This could mean that we should think about the work that scientists do in a different
way. Perhaps instead of selecting hypotheses over their competitors, scientists give arguments for why we ought to reject certain hypotheses. Scientists would not, for instance,
say that one hypothesis is acceptable because it explains the data and is simpler than its
competitors, despite the fact that it also entails absurdities.17 Also, scientists would not say
that although one hypothesis explains the data (in some sense of the word explains), the
explanation is scientific, although the hypothesis depends on the actions of gnomes or other
types of causes which are far removed from the accepted scientific ontology.
It is very difficult to find an example of a case where scientists had before them
17 Although,

to be fair to philosophers, scientists do not seem to think of things like singularities

or mysterious observer effects as absurdities.

53
a pair of different hypotheses which are perfectly acceptable scientific explanations for
exactly the same data and neither of which failed to reconcile with the body of science in
general, so that we are left only with an appeal to simplicity (in some respect or another)
to decide between them. It might turn out that no cases actually arise where judgments of
simplicity are the only court of appeal in selecting which of a set of genuinely competing
hypotheses science is to embrace.
Scientists, in a venerated tradition, have invoked and defended principles of parsimony. It is at least clear that there are many roles for principles of parsimony to play
in scientific methodology, although they would need to be precisely stated and justified.
So suspicions about the practical applicability of a principle of parsimony do not head off
the motivation for the analysis of principles of parsimony. This kind of thing occurred to
Kant when he read Hume. In just clarifying the principles that would govern rational inductive inferences we end up either doing traditional metaphysics or doing metaphysics of
mind. Kants Copernican Revolution is to recast the traditional study of metaphysics as the
science of mind.

2.5

More Than Just Ockhams Razor

It is easy to fantasize that some clearly defined and justified principle of parsimony
would solve the problems of non-deductive inference. Since non-deductive inferences play
prominent roles in scientific methodology it is tempting to think that some principle of
parsimony is at work in science. It is tempting to imagine that scientists are occasionally
confronted with hypotheses which, if true, would explain some data and it is tempting to

54
think that scientists occasionally select one hypothesis over a competitor for testing because
it is simpler, in some some sense, than its competitors. Nevertheless it is notoriously difficult to find unproblematic examples of the principle of parsimony in science. Two general
problems present themselves in finding examples of the principle of parsimony at work in
science. One problem is that the ceteris paribus (other things being equal) clause is rarely
satisfied. The other problem is that arguments which are meant to justify the positive selection of a hypothesis over its competitors are rarely given by scientists. Rather, scientific
arguments often appear to take forms which would not fit the principle of parsimony at all.
The reductio ad absurdum is doing a great deal of work for scientists and it would be an error to assume that judgments of simplicity are playing a role in scientific method if, in fact,
scientists are employing reductios instead. Many of the arguments which are paradigmatic
of good scientific practices either rule out certain hypotheses due to their failure to unify
with the body of accepted theories and data or they show that current theory and ontology leave some phenomena unexplained, thus opening the door for scientific revolutions.
Does this mean that it is difficult to find judgments of simplicity playing roles in scientific
methodology? It does not. Judgments of simplicity may play a variety of important roles
in scientific methodology. It is just that the role for simplicity is, in many cases, not one
which involves the form of the principle of parsimony sketched above. Philosophers could
speculate about the scientific process and imagine cases where the principle of parsimony
plays a role in hypothesis selection a priori. But the transworld travel agent may sell only
one-way tickets. It would be nice to have a few actual cases.
An example which tends to crop up in casual conversations is that the postulation of

55
the planet Neptune was simpler than other hypotheses which might explain Uranus orbital
data. Neptune was posited theoretically and then, a very short while later, its existence
was confirmed empirically. This example is assumed by some to illustrate the success of
scientific methodology. Furthermore, some people assume that the principle of parsimony
is the method which played the crucial role in this story. Fortunately, Richard Swinburne
has put this example into writing. So, in the spirit of engaging the polemic, I can engage
Swinburne on the use of this historical case.
Swinburne has managed to give a very lucid defense of a systematic view of simplicity in his very short book Simplicity as Evidence of Truth. Swinburne defends a principle of parsimony but there are other significant features of his view as well. Another part
of Swinburnes view involves a muti-faceted simplicity concept which can be filled out in
a variety of ways. In looking closely at the history of the discovery of Neptune, I cannot
figure out how Swinburnes principle of parsimony plays a role. However, simplicity may
play a multitude of roles in the construction and testing of scientific hypotheses. My guess
is that Swinburne would not be disheartened by these results. Swinburne might just say
that my research illustrates features of his view involving the multiple facets of simplicity. However, he should choose a different example. I investigate this important bit of
eighteenth century astronomy to criticise Swinburnes principle of parsimony and also to
supply the second and third chapters with rich examples.

56
2.6

Neptune and Vulcan

In 1845 it was known to scientists that Uranus did not follow a Keplerian orbital
path. Working independently, the French astronomer Urbain LeVerrier and the English
astronomer John Couch Adams each posited the existence of a planet with a specific mass
and with a particular orbital path which would interact gravitationally with Uranus in just
such a way so as to explain the data. LeVerriers calculations were sent to the Berlin
Observatory where, in 1846, Johann Gottfried Galle and Heinrich dArrest were able to
observe the planet now known as Neptune.
Swinburne argues in defense of a principle of parsimony: that other things being
equal, the simplest of competing theories is most probably true. Swinburne refers to the
story of LeVerriers famous successful hypothesis to help illustrate his view:
Assuming Newtons theory to be true, LeVerrier wondered why Uranus
moved along a path which was irregular in comparison with the path which
Newtonian theory seemed to predict for it. On the supposition that the only
other planets were the five known to the ancients together with Uranus, the irregularities were not to be expected. So LeVerrier postulated a seventh planet,
which pulled Uranus out of its regular orbit when close to it. He could have
postulated seven hundred heavenly bodies with a common centre of gravity at
the position at which he postulated Neptune, but the former supposition was
simpler (in virtue of the first facet which I described)18 . Or he could have put
forward a yet more complicated proposal, amending Newtons theory and postulating two further heavenly bodies at the same time. But of course he would
not have supposed that the evidence supported that complicated hypothesis.
19
([59] pg. 57)
At first blush some things seem to be intuitively clear about the example. Unless
18 The

word facet is Swinburnes term. It indicates that there are different ways of conceiving
of simplicity.
19 I

do not know what Swinburne had in mind. I do not see that LeVerrier was concerned with 5
planets. LeVerrier was already concerned with seven planets, one astroid belt, and one sun in his
studies of the perihelion advance of Mercury when he posited Neptune.

57
we wish to rework our successful scientific theories then it seems to be simpler to posit
one thing rather than seven hundred to explain specific data and to explain away anomalies.
Some people find one to be intuitively simpler than seven hundred because one is less than
seven hundred. Some people also find not reworking our successful scientific theories to
be simpler than reworking them in the sense that it is easier not to rework entire theories.
Most people likely to read the works of philosophers like Swinburne have been
taught that there is a planet called Neptune. I learned about it from an arrogant elementary school science teacher responsible for a mixed bag of facts, poisonous errors and
infelicities. Since my initial introduction to the word Neptune art, popular culture, and
media sources have used the word often together with the words science and scientists
in ways which have reinforced the things that I have come to believe about Neptune. My
guess is that most readers of Swinburne have a similar history associated with their use of
the word Neptune and with the notion of its participation in the history of science. And
although very few of us have actually worked out LeVerriers calculations and although few
of us have actually viewed Neptune through a telescope, most people are fairly confident
that the story of the discovery of Neptune is a story of scientific success. I suppose that
the rough idea about why this would be a story of scientific success is that the existence of
Neptune would have been predicted by mathematical theory before it was observed.
Despite these intuitive footholds, I am not convinced that this case has been described clearly enough to serve in any regard relating either to simplicity or to science.
Attempting to employ this example in the service of simplicity in science may be a joust
with a pivoting target on the short end of a trebuchet. The example must be spelled out in a

58
bit more detail if there is to be any hope that it will not immediately supply its own counter
examples. There are three basic reasons to reconsider how this example might work: two
are historical and one is philosophical.
All planets wander from their Keplerian paths (depending on how accurately we
demand that the data fit the mathematical predictions). Some planets diverge from their
Keplerian orbital shapes more than others. LeVerrier famously studied two of the more
poorly behaved planets: Mercury and Uranus. LeVerrier was actually engaged in a study
of Mercury prior to his study of Uranus. The successful confirmation of the existence of
Neptune was thought by some also to be confirmation of the methods LeVerrier employed in
his studies of Mercury. Mercury, which was notoriously difficult to observe due to its small
size and nearness to the sun accumulated non-Keplerian data as telescope construction
methods evolved and as generations of astronomers went to painstaking lengths to take and
record precise observations. LeVerrier was the scientist who gave the argument that there
was something unique about the advance in Mercurys perihelion; something which could
not be explained only by Mercurys gravitational interactions with the other known planets
in the solar system. LeVerrier posited additional planetary matter to explain this just as he
did to explain the perturbation in Uranus orbit. Eventually LeVerrier endorsed the view
that the planet Vulcan which some people (not all of whom ought to be called scientists)
already thought to be located between the Sun and Mercury was part of the inter-planetary
matter that he had posited.
On the basis of the same sources of authority which lead most people today to
believe that Neptune exists, it is also believed that Vulcan does not. The reason that we

59
might be suspicious about the strength of this example is that it is possible that LeVerrier
came to posit a roughly correct hypothesis about the orbital shape and mass of Neptune,
but stumbled upon it by methods which we might think of today as non-scientific. Even
during LeVerriers lifetime there were scientists critical of his methods. Emmanuel Liais
was very suspicious of the inter-Mercurial matter hypothesis and critical of LeVerriers
methods, saying of the apparent success of the Neptune hypothesis, To Galle therefore,
and not to LeVerrier, the honour of the discovery, as to Newton and not to the apple, that of
universal gravitation.[?] p.26
If this sort of criticism turns out to be on point then we should not accept the story
of the discovery of Neptune as a scientific success story, but rather as a fable about the
treasures that can occasionally be acquired by blindly groping about. Naturally a tale of
non-scientific groping is not the sort of tale appropriate for the job which philosophers like
Swinburne intend for it.
Another problem with the story is that hindsight indicates that one of LeVerriers
assumptions may be false. The Newtonian inverse square law of gravity has been replaced.
The accepted theory is now General Relativity which describes the shape of space-time
by a set of equations: Einsteins field equations. Today gravitational motion is explained
by General Relativity rather than by Newtons inverse square law. It is now accepted that
the additional (roughly) 4000 per century

20

of advance in Mercurys perihelion which is

not explained by interactions with other planets is explained by General Relativity. From
calculated 3900 of perihelion advance not explained by gravitational interactions with
other planetary matter, but Simon Newcomb later calculated the value to be 4300 per century.
20 LeVerrier

60
a contemporary perspective, the advance in Mercurys perihelion might appear to be a
symptom of the fact that the inverse square law of gravitation was not exactly correct.

21

I would not wish to defend the principle that other things being equal the conjunction of
simplicity and at least one false assumption generates theories which are most probably
true. So, a reason to proceed carefully is so that the example does not do better to illustrate
this principle rather than the sort of principle defended by Swinburne. Also, the fact that
science has embraced a more complicated mathematical description of gravity might do
better for the view that complexity (rather than simplicity) is a guide to truth.

22

I am not quite sure what to make of Swinburnes suggestion that the simplicity
exemplified by 1 < 700 is a guide to the truth of the theory committed only to 1 thing
rather than 700 things. If 1 and 700 are both numbers, then it is not at all obvious why
they are not equally simple. It is the tradition to offer scientific theories in general form
and so we might not be surprised to see a theory where x and y are the variables which
contingently take the values of 1 and 700 on some interpretations. The variables x and y
might be said to be equally simple.
Also, as a historical matter, it was not the case that LeVerrier posited his there is
one Neptune hypothesis and suggested tests for it in competition with some there are
700 things hypothesis. Even though I realize that philosophers are supposed to be able
to imagine empirically equivalent hypotheses, I am never quite sure how far I might fly on
21 Some

scientists have even given arguments that the shape of space-time can be derived from
the advance in Mercurys perihelion and the Legrangian [50] p.166
22 Of

course Swinburne knows that GR is more complicated in some sense than Newtonian mechanics. He discusses this in his book. I discuss his view in more detail in Chapter 3

61
the wings of imagination before I express only ignorance about laws or initial conditions
relevant to the scientific discourse. If I assume the Newtonian law of gravitation to be true
and then imagine seven hundred objects with a center of mass located on an orbit the same
as the one occupied by Neptune, then I am just imagining a big, less-dense-Neptune. Now
only the density of Neptune is disputed by the rival theories of which I have conceived.
The densities of Neptune1 in hypothesis1 and Neptune2 in hypothesis2 are not relevant
to Uranus goofy orbit since Newtonian mechanics models masses as point masses. The
example could be tweaked so as to remove this triviality. If it were also suggested that the
odd flip-flopping about of irregular objects with a common orbital center of mass helped to
explain additional features of Uranus behavior (perhaps some nutation in its precession23 )
or some other dynamic in the solar system (perhaps an advance or retrograde motion in the
perihelion of yet another planet) then the one thing hypothesis would not exactly be a
competitor with the seven hundred thing hypothesis because they would predict different
data. In this case the ceteris paribus clause of the principle of parsimony is not satisfied. If
perhaps the view that the laws of motion are invariant under Galilean transformations were
changed or if we conjure theories with different space-time shapes or different numbers
of dimensions of space-time then we might imagine a seven hundred thing hypothesis
which is a genuine competitor with the one object hypothesis, but that would not be
holding other things equal either, and thus would be a poor way to advance Swinburnes
project.
A closer look at the historical details of LeVerriers work may help to clarify this
23 A

nutation is a periodic oscillation in the precession of the axis of a planet.

62
story enough to permit us to reconsider the relevance (or irrelevance) of certain features of
the example to the role of simplicity judgments in scientific methodology. N.T. Roseveare
authored the book, Mercurys Perihelion: from LeVerrier to Einstein, the purpose of which
was to counter the murky things that have been said about the efforts from the 1840s to
the early 1900s to account for the non-Keplerian orbital data recorded for certain planets.
Roseveare holds that what had been said was lacking in crucial details [50] and that only by
hindsight does it seem as if the issues which led to the overturning of the inverse square law
of gravity are clear. The problems faced by astronomers and physicists during this roughly
one hundred year period were complex and difficult, and Roseveare holds that some accounts have missed crucial details. Roseveares historical investigation of this period helps
to shed some light on the scientific process and some of the roles for simplicity judgments
in it.
Questions ought to be raised about which of LeVerriers methods are thought to
be exemplary scientific methods. In this regard, LeVerriers studies of both Mercury and
Uranus are relevant. Questions also ought to be raised about what LeVerriers contributions
to science might be. In this regard both his methods and his arguments are relevant. On
this issue, I think that LeVerriers arguments involving the orbital data on Mercury supply
especially rich material for examples. The reason is that LeVerriers arguments involving
the perihelion advance of Mercury fueled the next one hundred years of scientific debate
about general theories of gravitation and eventually took center stage in discussions about
General Relativity.
LeVerrier and John Couch Adams both made their initial predictions about the or-

63
bital position of the planet presumed to be perturbing Uranus in accordance with Bodes
Law. Bodes Law states that each solar system body will occupy an orbit with a semi-major
axis located at a distance from the sun twice the distance of the semi-major axis of the orbital body just inside of it. Bodes Law (or the Titus-Bode Law) is based on the series:
0,3,6,12,24,48... . Adding 4 to each number and then dividing each by 10 gives Astronomical Units for the distances of the semi-major axes of the planets from the sun (where Earth
is 1.00au from the sun). Planets seemed to fit roughly Bodes Law until the discovery of
Neptune. As it turned out, the telescopic data which confirmed the existence of Neptune
contributed to the dislodging of Bodes Law from the corpus of accepted scientific principles. Apparently, for many scientists, the semi-major axis of Neptunes orbit a bit too
roughly fit Bodes Law predictions of its distance from the Sun. As an extra twist to this
story, many nineteenth century scientists argued against hypotheses which posited matter
between Mercury and the Sun on the basis that such hypotheses would violate Bodes Law!
Multi-bodied gravitational problems are very difficult to solve mathematically. There
is no general solution to the three body problem in Newtonian mechanics. The positions
and velocities for each body at some time step all depend upon one another. There is no
way to peek at Uranus future position without knowing what Neptunes future position
at that same time is. But this depends upon Uranus position at that time. Still mathematicians have worked out methods to model multi-bodied systems and these calculations are
quite laborious to do by hand.
Consider the sort of skill that LeVerrier had mastered. Newtons equation for the
gravitational force between two masses M and m1 separated by a radius r1 is,

64

F = G

M m1
r.
r12

(2.1)

Where r is a unit vector in the direction of the radius between from the first mass
to the second. This shows that the force of gravity acts in the direction opposite the unit
vector.
If we wish to know the perturbing force on a planet m1 we need another equation
of the form,
F = S + P,

(2.2)

where S is the force between the sun and the planet and P is the perturbing force.
If the perturbation is caused by a third mass m2 at distance d from m1 then these
equations combine to give,

F = G

m1 m2
M m1
r + G 2 d.
2
r1
d

(2.3)

If m2 is not a moon of m1 but rather another planet orbiting M then another equation is needed for the total force on m2 ,

F 0 = G

M m2
m1 m2
r + G 2 d.
2
r2
d

(2.4)

Substituting Newtons Second Law F = m

a for F and F 0 in equations 2.3 and 2.4
canceling terms and factoring we have,

a
1 = G


M
m2
r + 2 d
r12
d

(2.5)

65


a
2 = G


M
m1
r + 2 d .
r22
d

(2.6)

Acceleration is the first derivative of velocity with respect to time and the second
derivative of position with respect to time. So equations 2.5 and 2.6 can be used to describe
the velocity and position of m1 and m2 . However, at this point a few decisions must be
made about how to proceed. We must decide whether spherical or Cartesian coordinates
will be used. Assume Cartesian coordinates. Then it is convenient to place M (the sun) at
the origin and generate the following equations which can be solved for velocity,

vx
21

vy
21

+
t

vz
21

= G

x21

M
m2
p
+p 2
2
2
2
2
+ y1 + z1
x1 + y1 + z1 + x22 + y22 + z22

(2.7)

vx
22 + vy
22 + vz
22
= G
t

M
m1
p
+p 2
2
2
2
2
2
x2 + y2 + z2
x1 + y1 + z1 + x22 + y22 + z22

(2.8)
and for position,
2

p
x21 + y12 + z12
= G
t2

p
x22 + y22 + z22
= G
t2

M
m2
p
+p 2
2
2
2
x1 + y1 + z1
x1 + y12 + z12 + x22 + y22 + z22

M
m1
p
+p 2
2
2
2
x2 + y2 + z2
x1 + y12 + z12 + x22 + y22 + z22

!
(2.9)

!
. (2.10)

In order to integrate 2.9 and 2.10 we need some values for the position. However,
in order to get the relevant hx, y, zi values for some time step we need to know the velocity. But the velocities depend upon the positions, and the velocities and the positions of

66
each planet depend upon the position of the other. This illustrates why there is no general
solution for this sort of problem. Nevertheless, mathematicians have worked out ways to
approximate solutions to these differential equations. These sorts of methods involve getting solutions to much simpler equations, then using those values to feed back into the more
complex equations. Such methods have been called the method of variation of elements or
the finite element method. We might, for example, treat each planet as if it is only in a
two bodied system, solve for the position at some t and then use those results to go back
and start the first round of rough calculations, knowing that the initial values could not be
exactly correct.24
Now it might be easy to appreciate what LeVerrier accomplished. He worked out
the solutions not for a three bodied problem (like the example given) but for a 8 bodied
problem (because the was concerned with 7 planets and the sun)! LeVerrier demonstrated
his great skill as a mathematician by working out these complicated partial differential
equations using the finite element method. LeVerriers results on Mercurys orbit were
included in nautical almanacs and astronomical ephemerides internationally for over forty
years.
LeVerrier employed Newtonian mechanics in an argument which contributed to
refuting Newtonian mechanics. This is the sort of method which, I think, deserves more
focus from philosophers of science. It is a method which generates an argument that current
24 LeVerrier probably used spherical coordinates to work out his solutions.

This is what one would

expect given that LeVerrier was interested in Mercurys perihelion advance. That is, he would want
to know what the change in the changing angle swept out by Mercury in its orbit was over long
periods of time. But the choice of coordinate system does not alter the point about the mathematics
- that there is no solution to the three bodied problem. I find it a bit less difficult to illustrate how
the interdependent variables arise in the three bodied problem using Cartesian coordinates.

67
theory and ontology leave some phenomena unexplained. LeVerrier gave the argument that
the advance in Mercurys perihelion could not be accounted for by gravitational interactions
with the known planets alone. Later the ability to explain Mercurys advance was thought
to be one of the explanatory virtues of General Relativity. Just like all astronomers at the
time, LeVerrier was aware of the two general strategies for pursuing a scientific explanation
of Mercurys behavior: 1) posit changes in the amount or configuration of planetary matter
in the solar system, 2) reject the inverse square law of gravity. Scientists could see that the
equation for the perturbing force,

F = S + P,
might be expressed by a law involving only two bodies like,

F = G

M1 m1
M1 m2
r + G 3 r.
2
r1
r2

Rosveare says,
that laws differing from the inverse square law gave precessing orbits was
known long before Mercurys anomalous advance was discovered. Such results may be found in Newtons Principia, the principle work in which Newton
put forward his gravitational theory.([50] pg. 12)
In 1745 Alexis Clairaut suggested laws differing from the inverse square law in this
way. Yet LeVerrier resisted challenging the inverse square law, saying that,
If the tables [of Mercurys positions] do not strictly agree with the group of
observations, we will certainly not be tempted into charging the law of universal gravitation with inadequacy. These days, this principle has acquired such a
degree of certainty that we would not allow it to be altered; if it meets an event
which cannot be explained completely, it is not the principle itself which takes
the blame but rather some inaccuracy in the working or some material cause

68
whose existence has escaped us. Unfortunately, the consequences of the principle of gravitation have not been deduced in many particulars with a sufficient
rigour: we will not be able to decide, when faced with a disagreement between
observation and theory, whether this results completely from analytical errors
or whether it is due in part to the imperfection of our knowledge of celestial
physics. ([50] pg. 21)
Roseveare suggests that social forces may give another reason for scientists to be
reluctant to challenge the inverse square law of gravitation. He says that,
Only a highly reputable mathematician would have had sufficient stature
for his results to stand up against the authority of Newton. ([50] pg 38)
So a principle common in LeVerriers study of Mercury and his study of Uranus
was the principle that it is better to posit more planetary matter than to muck with Newtons
laws when explanations are constructed to explain the orbital data of these planets. To
explain the behavior of Uranus, LeVerrier favored a hypothesis positing only one other
planet. However to explain Mercurys behavior LeVerrier favored a multiple bodies (or an
astroid belt) hypothesis. Even when LeVerrier endorsed certain claims which would have
suggested confirmation for his hypothesis (that the transit of an inter-Mercurial object had
been observed) he still held that there would have to be more than one such object. It is
interesting to note that it appears to be consistent with LeVerriers methodology to posit
one thing or hundreds of things to explain certain data, since the number of things was
not a basis upon which LeVerrier expected the truth of his hypotheses to depend. Rather,
LeVerrier was committed to explaining data with a very high degree of precision without
changing general laws of motion. Perhaps a more subtle commitment is that LeVerrier
posited more of the same kind of stuff with which scientists were already familiar. He
posited matter either balled up in a planet or distributed in rings to explain specific data and

69
yet never forwarded hypotheses involving gnomes, spirits, ethers, or any other sort of stuff
different from that which planets are thought to be made of.
Although scientists were aware of the logically possible strategies for constructing
hypotheses to explain Mercurys perihelion advance, it took several years for an array of
competitors to be published. LeVerrier himself suggested several hypotheses involving
inter-Mercurial matter, but other approaches were reflected in the hypotheses given over
the next thirty to forty years. In the late nineteenth century a brilliant astronomer, Simon
Newcomb, collected these hypotheses and constructed arguments regarding the plausibility
of each. The following is a list of the hypotheses with condensed versions of the reasons
given by scientists that each hypothesis was unlikely to be true.
1. Inter-Mercurial planet (Le Verrier): Le Verrier thought that this hypothesis was implausible because a planet of sufficient mass to account for the advance would have
been easily observable in transit and no observations of a planet this large had been
made.
2. Inter-Mercurial astroid ring (Le Verrier): Le Verrier also thought that the transit of astroids would be observable. He ended up endorsing contentious claims made by some
astronomers to have observed objects of this sort. Even so, Le Verrier knew that astronomers had not recorded the sort of data that one would expect on the astroid-ring
hypothesis because the observations (mistakenly made of sun spots not planetiods)
were of objects too small to account for Mercurys advance. He may have endorsed
this view in hopes that more planetoids would be observed.
3. Inter-Mercurial rings (Le Verrier): It was known that on certain intervals of Saturns

70
orbit the rings are at such an inclination with respect to Earth that they were not
visible to Earth-bound observers. Inter-Mercurial rings might not be visible from
Earth. Le Verrier argued against this hypothesis concluding that in order to be at the
correct inclination to cause the advance in Mercurys perihelion the rings would be
observable from Earth.
4. Zodiacal Light Matter: There is a pancake shaped haze of dust around the sun which
diminishes as the distance from the sun increases. The evidence of this dust can be
viewed at certain times of the year on very dark nights as a glow in the night sky.
It was suggested that this haze of matter might interact gravitationally with Mercury
so as to account for the advance in Mercurys perihelion, but Simon Newcomb argued that the zodiacal cloud would cause a braking action on the orbits of Venus
and Earth causing retrograde motion in their perihelions. This was not consistent
with the orbital data on Earth and Venus. ([50] p. 48) Later scientists revised the
zodiacal cloud hypothesis to account for Mercurys perihelion advance and obtained
very good results. This later hypothesis was accepted as the correct explanation until
after General Relativity when in 1918 Harold Jeffreys argued that the particles in the
zodiacal cloud could not be large enough to produce the effect due to the way the
way that the particles reflect light and given the sizes they would have to be in order
not to get blown out of the solar system by solar winds.
5. Oblate Sun: It is natural to think that if there are not inter-Mercurial rings, then
perhaps the sun is oblate. However, arguments had been given by the end of the
nineteenth century that the sun was actually slightly prolate. ([50] p. 46-47)

71
6. Non-inverse square law (Newton): As Newton realized, such laws might well account for Mercurys orbit, but they would not describe terrestrial gravitational motion
(objects falling near Earth) ([50]p.51).
7. Dynamic non-inverse square laws (Zollner and Tisserand, 1872): The suggestion was
made that gravitational laws may be like Webers velocity-dependent laws of electrodynamics: F =

GM m
{1
r2

1
h2

dr 2
dt

2r2 d2 r
};
h2 dt2

where h was a constant similar to

a constant used in electro-dynamics. Scientists raised objections to Webers electrodynamics at the time, so basing a theory of gravity on a hotly contested theory of
electro-dynamics raised immediate objections. Newcomb noted that the choice of
h to get a theory which accounted for orbital data would be a value of h different
from the one accepted in electro-dynamics. Newcomb neither accepted nor rejected
this. The choice of h which accounted for Mercurys perihelion advance was

2c

([50]p.44). Another type of objection involved the suggestion that if the space were
infinite then infinite forces might be possible ([50] p. 130). This was viewed as
an absurdity since matter was not observed to be moving at velocities approaching
infinite velocities25 .
8. Non-inverse square law (Asaph Hall, 1894): Newton had shown that the angle between successive perihelia was = 2

bc
mbnc

where b, c, n, m are constants which

might be chosen for a force law of the form: F =

brm crn
.
r3

Hall chose n =

2.0000001574 so that it accounted for Mercurys advance. The law seemed to work
25 I

do not know why scientists thought that this was absurd or why they expected to observe
things moving with near infinite velocities.

72
well to describe the orbits of other planets as well. Newcomb even appears to entertain the simplicity of Halls hypothesis relative to other non-inverse square laws as a
possible virtue saying that,
This hypothesis seems to me much more simple an unobjectionable
than those which suppose the force to be a more or less complicated function of the relative velocity of the bodies. ([50] pg.51)
However, Newcomb went on to argue that if this law were applied to the data on the
motion of the moon, then the moon would have a center of gravity offset from its
geometrical center. Many scientists saw this result as absurd since it would change
what they thought about how planets travel orbital paths, and would lead to apparently absurd conclusions like that the moon may well have an atmosphere and life
collected on the side away from the Earth. ([50]p.51-55)
One interesting feature of the methodologies of LeVerrier and Newcomb is that
they rule-out, by way of argument, far more hypotheses than they defend. Newcomb was
a well respected scientist and his arguments were taken quite seriously by the scientific
community. One way to see the result from the contemporary perspective might be to
say that it is not surprising that all eight hypotheses are false because none of them is
General Relativity. However, another way to see this situation is to notice that Newcomb
gave arguments showing the explanatory failure of each of these hypotheses, and thereby
motivating an investigation into what would be an acceptable explanation for the anomalous
advance in Mercurys perihelion, thus helping to set the stage for General Relativity in
1917. The real argument might have been that nothing in physical theory at the time could
explain Mercurys perihelion advance.

73
Another thing to notice is that some of these arguments would not be very good
arguments by philosophical standards. Why would anyone think that a velocity dependent
force law for gravitation would have the same constants as a velocity dependent force law
for electro-dynamics? Why would anyone think it obviously absurd if the moons gravitational center were not at its geometrical center and that there may be an atmosphere on its
dark side as a result? Nearly every hypothesis advanced by brilliant scientists like LeVerrier
and Newcomb strike me as being as surprising and counterintuitive as the hypotheses they
refute. Nevertheless, I believe that the reasons given are good reasons for scientists to resist embracing certain hypotheses for further study or testing. The reasons scientists did not
embrace wild theories like these is that they would disrupt widely accepted value for unity
in science. The value for some unity of science principle is reflected in several of these
arguments, but especially in the argument that the constant h in a velocity-dependent gravitational law would be the same as the constant h in a velocity-dependent electro-dynamic
law.
Scientists expect that hypotheses should account for the target data. The first three
hypotheses do not account for the data in the sense that LeVerrier expected; so that if these
hypotheses were true then either they would, on certain parameterizations, fail to account
for Mercurys advance, or on other parameterizations, would be attended by additional
observational data which was absent.
The arguments against the oblate-sun hypothesis may give examples of both methods just suggested. The oblate-sun hypothesis was in conflict with the interpretation scientists gave to telescopic data involved in describing the suns shape. This is a case where a

74
hypothesis may account for a set of data central to certain lines of inquiry, but fail to reconcile with other data to which the scientific community is largely committed which can be
seen as a conflict with some version of a unity of science principle. It may also have been
that if the sun were oblate rather than prolate then the observed telescopic data would be
different than it was.
Scientists also expect scientific theories to be general. What it is for a scientific
theory to be general may be subject to some vagueness, but it is clear that scientists expected
hypotheses to be involved in theories which account for a broad range of data. While I
would not wish to offer my own theory of what it is for a theory to be general, it is possible
to point to cases where a value for general theories played a role in scientific methodology.
Newton, for example, expected the force law to explain both terrestrial and extraterrestrial
motion. Scientists working on the problem of Mercurys perihelion advance expected that
uniform laws of friction and gravitation should explain zodiacal light matter interactions
with all planets.
What this case study has developed so far is the following list of methodological
guidelines: that hypotheses account for data, that proposed hypotheses do not imply data
which is at odds with other data, that theories are sufficiently general, and that hypotheses
are proposed in accordance with a unity of science principle. In addition to these, scientists
often expect that the construction and testing of theories ought to involve the addition of
as few ad hoc hypotheses as possible. Inventive philosophers might not be swayed by the
argument that a non-inverse square law gave an odd center of gravity for the moon and
therefore ought to be rejected. Perhaps the postulates or axioms of geometry could be

75
changed. Perhaps hypotheses about the way that light travels between the moon and the
Earth or through the lenses of our telescopes could be changed in existing theories to give a
coherent story about the motion of terrestrial and extraterrestrial bodies despite Newcombs
argument. Yet for some reason, scientists reject this sort of imaginative tinkering. Scientific
theories and methodological principles are overturned by careful and methodical steps.
Bodes Law was a widely accepted principle of astronomy despite the margin of error which
appeared to be increasing as the distance of planets from the sun increased. Eventually a
hypothesis which was constructed and tested using Bodes Law (the Neptune hypothesis)
made a significant contribution to the dislodging of the principle from accepted scientific
methodology.
Finally, what is interesting to notice about the hypotheses suggested in the study
of Mercury is that they are rejected rather than accepted on the basis of certain scientific
arguments. After a discussion of the problems of induction it is tempting to think that the
fundamental puzzle of scientific methodology is how it is that hypotheses that are more
likely to be true than their competitors come to be accepted by the scientific community.
But the clever scientific arguments given by LeVerrier and Newcomb are arguments telling
us which hypotheses ought to be abandoned. None of the reasons given for the abandoning
of hypotheses obviously have to do with the relative complexity (or non-simplicity) of the
hypothesis, but rather with the methodological values suggested above: hypotheses ought
to account for data and ought not to introduce weird data, hypotheses ought to contribute to
theories which are general in some sense, the unity of science is a good thing, and ad hoc
hypotheses are a bad thing.

76
Swinburne tried to employ a reference to LeVerriers work to provide an example
of how scientists judge that simpler hypotheses are more likely to be true. I cannot see
how Swinburnes thesis might be advanced using this example. After all, LeVerrier makes
it very clear that he is more committed to his methods than he his to the number of objects that his hypotheses posit. He would rather posit planetary matter to explain the data
than overturn Newtons Laws. Some people will leap at the opportunity to point out that
LeVerrier is invoking some principle of parsimony with respect to ontological kinds in being committed to there being more of the same kind of stuff in his ontology. However, he
did not favor his particular hypotheses (Vulcan or Neptune) relative to hypotheses which
posited ghosts or effluvia. Were hypotheses positing additional kinds of substance under
consideration to explain the orbital data, we would have examples where a hypothesis is
thought to be more likely than its competitors due to some simplicity judgment about the
numbers of kinds of things posited. Instead, LeVerrier wanted to posit more of the same
kind of matter rather than to overturn a mathematical theory. To say that hypotheses were
compared for their relative simplicity in this case would be to commit a category mistake.
LeVerrier is the poster child for a scientific mantra which might go it is better to endorse
a false hypothesis than to change a good methodology.
This story shows the importance of doing scientific work to reject hypotheses and
thus to show what remains to be explained. This activity is an important part of the scientific process (of what Kuhn calls normal science). Accepting hypotheses or theories by
an application of a principle of parsimony is not part of this story. So my hope is to point
down the path towards greater philosophical focus on the part of science which demon-

77
strates problems with theories or methods rather than on the part of science concerned with
comparing which of several competing hypotheses is supposed to be the most likely. Although each of the above hypotheses are competing with the others in the sense that they
are each given initially to explain Mercurys advance, they are not compared to one another
in the arguments which refute them. So, in some sense, they are never really in competition.
Although, in one case, Newcomb mentions what he thought to be the virtue of simplicity
exemplified by one hypothesis over another, this serves no role whatsoever in the argument
which leads him ultimately to reject the view.
There may be many different ways that simplicity judgments are involved in scientific methodology. Simplicity might be involved in the construction of hypotheses. We
might wonder if it was simpler for LeVerrier to posit more of the same kind of matter
employed in other scientific explanations than to posit novel kinds of matter for his hypotheses. We should ask what kind of simplicity this is before asking any questions about
whether or not principles involving this kind of simplicity are justified.
Perhaps it is simpler, in some sense, to remain stalwart against the temptation to
tinker with extremely general and broadly accepted laws. Perhaps it is simpler to have
one rather than two or three constants in velocity dependent laws regardless of the sort of
phenomena they are posited to explain.
Simplicity might be involved in the construction of arguments which motivate certain lines of inquiry. LeVerriers argument that Mercurys perihelion advance could not
be explained by gravitational interactions with known matter in the solar system involved
methods of discrete integration which depend upon certain simplifying assumptions: roughing-

78
in the first set of calculations as if each problem is a two-bodied problem rather than a
multi-bodied problem.
Simplicity, perhaps, is involved in the expectation that our scientific theories be
general or that they unify with other theories. Even if the mathematics gets verbose or
difficult to use or to comprehend, it still may be simpler to have one law describing the
motion of all bodies, or perhaps one constant for velocity-dependent laws than it would be
to generate multiple laws.
It might even be that simplicity considerations are shot through several different
parts of the scientific methodology, since many methods implicitly involve resisting the addition of ad hoc hypotheses to the body of scientific theory. In this way, the fewest number
of ad hoc hypotheses would be a species of simplicity. However, simplicity judgments in
this case are part of the general method of theory construction, and not committed only to
a principle of parsimony.

2.7

The Simplicity of the Ancients

In this final section, I discuss the deeply entrenched features of simplicity in philosophy.
The word simplicitie probably came into use sometime in the 12th century. The
first citation in the Oxford English Dictionary is Chaucers Boethius (simplicitie) (1374).26
The suffix ity indicates an abstraction from a comparison.
26 Simplicity:

The state or quality of being simple in form, structure, etc.; absence of compositeness, complexity, or intricacy [1]. The word may have earlier Latin ancestors for all I can tell, but
etymology is not the focus of this project.

79
It is important to remember that simplicity is contrasted with complexity perhaps in
several senses. There is no upper bound on complexity, but the lower bound on simplicity
is unity. A methodological principle that promotes unity, is a principle of simplicity. To
be clear, we must ask what simplicity criterion is involved in the principle, or what kind of
unity the principle promotes. If a simplicity principle does not promote unity (the limiting
case of simplicity), then it is a principle that can only be applied in comparative judgments.
Knowing how it is that art, culture, and science reflect one another like a hall of
mirrors, it is fun to fantasize that the whole mess about the principle of parsimony was
introduced by some early scientist who infected science with some whimsical aesthetic.
Swept away by Chaucer and wine this scientist spilled his drink and observed how the
chaotic substance resolved itself into a pattern on his pillow. The next morning he returned
to his laboratory with a stained face and a new methodological principle. Western thought,
ever since, would have inherited a mistake. But I believe that the roots of simplicity in
science are far older. Roger Ariew says,
Briefly, the broad historical argument is this: the roots of what is called
Ockhams razor can be traced back to Aristotles Physica and De Caelo. One
can pick up the medieval interpretations of Aristotles principle in the Latin
translations of Averroes commentaries on the Physica and the De Caelo or the
later Medievals (Aquinas, for example). One can then see how Ockham came
about his principle of parsimony from these sources through his immediate
predecessors, Peter Auriole and John Duns Scotus.([3] pg.15)
Friedrich Nietzsche had a useful insight when he associated the names of ancient
Greek thinkers with archetypal doctrine. This is useful because it helps to organize different
kinds of simplicity principles. Additionally, if Nietzsches account of the thought of the
ancient Greeks is correct, then the place of simplicity in science is not merely that it is

80
a reflection of the values, or errors of Medieval or Early Modern thinkers, but something
which is deeply embedded in our language and thought. Nietzsche says that, what they
[the Greeks] invented was the archetypes of philosophical thought. All posterity has not
made an essential contribution to them since. ([48] pg.31)
I have wondered if Nietzsche is right on the second claim. If so, then we would not
learn anything about philosophy by reading contemporary literature in addition to reading
the Greeks. I am however, convinced that Nietzsche is right on the first claim.
If I try to fantasize about what the ancient Greek philosophers were talking about, I
imagine that they had noticed that there seemed to be order rather than chaos and the game
became one of explaining this with as few hypotheses as possible. For the pre-Socratics,
the game would have been to give just one hypothesis. The ancient Greeks were primarily
concerned with cosmogony (the logical evolution of the universe). The fundamental puzzle
appeared to be the question of how there came to be a plurality of things and how assertions
are possible.
Nietzsche says that the Greeks prior to Plato were pure types and from Plato on
philosophers have mixed the pure pre-Socratic views. This is Nietzsches useful claim. He
is saying that there is a way to think about the philosophical dialectic as several competing
hypotheses about unity. He says that it would be
correct and simple to comprehend the latter as philosophic mixed types,
and the former as pure types. Plato himself is the first mixed type on a grand
scale, expressing his nature in his philosophy no less than in his personality.
Socratic, Pythagorean, and Heraclitic elements are all combined in his doctrine
of Ideas. ([48] pg.34-35)

Nietzsche says that Thales was the notable character at the birth of Greek philoso-

81
phy. Nietzsche has Thales as a proto-empiricist. Kant seems to suspect this as well, because
he mentions that Thales may have been the one who turned geometry into an empirical science.27 It has been said that Thales turned the Greek philosophers away from explanations
of natural phenomena in terms of anthropomorphic gods and heros. However, it is interesting that Nietzsche also points out the unity hypothesis all things are one is what is
really distinctive of Thales Greek philosophy and that Western thought is shot-through
with attempts to reinterpret this doctrine.
Greek Philosophy seems to begin with an absurd notion, with the proposition that water is the primal origin and the womb of all things. Is it really
necessary for us to take serious notice of this proposition? It is, and for three
reasons. First, because it tells something about the primal origin of all things;
second, because it does so in language devoid of image or fable, and finally, because contained in it, if only embryonically, is the thought, all things are one.
The first reason still leaves Thales in the company of the religious and superstitious; the second takes him out of such company and shows him as a natural
scientist, but the third makes him the first Greek philosopher. Had he said water turns into earth, we should have but a scientific hypothesis, a wrong one
but difficult to disprove. But he went beyond scientific considerations. By presenting his unity-concept in the form of his water-hypothesis, Thales did not,
it is true, overcome the low level of empiric insight prevalent in his time. What
he did was to pass over its horizon. The sparse and unordered observations
of an empirical nature which he made regarding the occurrence and the transformations of water (more specifically, of moisture) would have allowed much
less made advisable, no such gigantic generalization. What drove him to it was
a metaphysical conviction which had its origin in a mystic intuition. We meet
with it in every philosophy, together with the ever-renewed attempts at a more
suitable expression, this proposition that all things are one.([48]pg.38-39)
We can see how other ancient philosophers cast, and recast the unity hypothesis.
Nietzsche says that, Anaximander takes two steps beyond him [Thales]. For the first he
asks himself: How is the many possible if there is there is such a thing as the eternal?.
([48] pg.49)
27 Kant

mentions this in the Preface to the Second Edition of The Critique of Pure Reason.

82
Alexander Polyhistor, in summarizing the Pythagorean doctrine says that the first
principle of all things is the One. From the One came an Indefinite Two (dyad), as matter
for the One, which is cause.28 Alexanders discussion of Pythagorean doctrine involves not
only the role of the unity in cosmology but also its causal role. Pythagoreans also thought
that there was a relationship between mathematical cosmogony and the mathematical basis
for music theory29 . The history of science is shot through with stories of contributors who
cast and recast these Pythagorean values (often in Christian theological frameworks). In
general we ought to ask of the role that simplicity has played in reasoning, why it is thought
to be related either to causation or to aesthetic values.
We can also imagine the opposing camps occupied by Parmenides and Heraclitus
giving their versions of the unity hypothesis. Parmenides is thought to have held the view
that all things are one and that change is impossible. It is also thought that Parmenides was
the primary influence in Platos division between reality and illusion. Heraclitus held the
view of the unity of opposites; that everything is changing. By modern categorical labels,
Parmenides would be the anti-realist extraordinaire and his distinction between reality and
illusion would preclude any empiricist commitments. But Heraclitus, who may be thought
of an another kind of anti-realist, offered an empirical hypothesis.
I see nothing other than becoming. Be not deceived. It is the fault of your
myopia, not of the nature of things, if you believe you see land somewhere in
the ocean coming-to-be and passing away. You use names for things as if they
rigidly, persistently endured; yet the stream into which you step a second time
28 Cornford

says that, writing in the first century B.C., Alexander would certainly have been
influenced by Platos Timaeus and therefore it is not entirely clear if this is properly a Pythagorean
view or if it is just Platos view. [19](pp.3)
29 Aristotles

(Metaphysics v. 285b, 23)

83
is not the same one you stepped into before. ([48] pg. 52)

I am not sure how the Heraclitian doctrine of the unity of opposites was robbed of
its empirical content, but if the thief left any clues he probably left them in the Timaeus. It is
Plato who tried to take the middle ground between the Pythagoreans who give explanations
in terms of what remains the same and the Heraclitian doctrine of change.
Although modern thinkers tend to see Aristotle as rebeling against Platos division
between appearance and reality, Aristotle probably saw himself as the defender of Platos
middle way. Aristotle framed this issue when he promulgated the syntax of scientific discourse in the Physics.
Now that we have made these distinctions, here is something we can grasp
from every case of coming to be, if we look at them all in the way described.
In every case there must be some subject that comes to be [something]; even if
it is one in number, it is not one in form, since being a man is not the same as
being an unmusical thing. (By in form I mean the same as in account.) One
thing [that comes to be] remains, and one does not remain.[4] (Physics Book
I)

What changes is predicated of the subject which does not change. In order to engage
the grammar of science we must establish the vocabulary of science. That is, we must
establish which terms are said to change and which are said to stay the same. We must
also show which are the law (or law-like) ways in which things come to be. That is, we
must generate statements of the form, it is a law that F x Gx. Unless the Parmenides
of Platos dialog is right that there is One, and nothing else can truly be asserted, then
some things change and some things stay the same. The scientific language of discourse
is committed to some things changing and others staying the same. If it is ever the case

84
that this is not so, then the physical world will either grind to a screeching halt, or be torn
asunder at the joints.30
The difficult questions about parsimony as a methodological principle in science
might be seen as a lasting battle between methods that generate hypotheses about sameness
and hypotheses of change. These two general versions of parsimony, in the abstract, are in
conflict and yet I see no way to avoid doing science in the basic way outlined by Aristotle.
It is as if, in the Fourteenth Century, the appearance of the word simplicity in Europe
indicated that modernity had finally given up on the Pythagorean ideal. Humes critique of
induction is the herald of the fact that the empirical hypotheses of Thales and Heraclitus had
settled securely into the foundations of scientific methodology, but without their empirical
content!
In the introduction to this chapter I promised to show how deeply principles of
parsimony are embedded in scientific discourse and I promised to show how the Ancient
Greeks can provide for us these very abstract archetypes by which to categorize this discourse. Europeans in the wake of the Copernican Revolution may have given up on the
simplicity of ancients because of the straight forward fact that the ancients gave hypotheses
about unity, and since there is no upper bound on complexity, the usefulness of the word
simplicity may indicate that people noticed that the hypotheses of the Ancient Greeks
were too simple to be employed in the scientific project.
30 Plato

appears to have the character Socrates defend his merger of the Heraclitian and
Pathagorean views by giving an interpretation to Homer saying, so long as the heavens and the
sun continue to move round, all things in heaven and earth are kept going, whereas if they were
bound down and brought to a stand, all things would be destroyed and the world, as they say, turned
upside down. (Theaetetus[49])

85
Plato mixed the pure hypotheses of the ancients and this shaped modernity. Aristotle followed Plato in mixing the Pythagorean unity of that which is insensible with the
empirical unity of Heraclitus. Some things change and some things stay the same and I
cannot see how to do science unless we think of it in this very abstract way. The scientific
project involves categorizing those things which stay the same and then giving hypotheses
about the law (or law-like) ways in which they change. Kant points out that the terms of
the science of geometry were fixed before geometry became a science. Similarly, Kant
points out that Galileo and Descartes fixed most of the terms of the physical sciences before Newton developed his laws of motion. One of the most important projects involved
in science is the project of determining which are to be the signs which stay the same (the
predicate terms) for a particular body of discourse. We can think of the Logical Behaviorists as suggesting a massive empirical research program which would result in a tome
of operationalized psychological predicates. Another way to think about this part of the
scientific project is as a set of questions about where to locate the stay-the-same-structures.
Kant locates the structures in the minds of individuals. Chomsky does something similar
locating the universal grammar in the human mind. Kuhn locates that which stays the same
(at least for a while) in society. This archetype of philosophical thought can even be seen in
Russells work where he mentions a common materialist intuition. One great reason why
it is felt that we must secure a physical object in addition to the sense-data, is that we want
the same object for different people.31
The mixed hypotheses of Plato and Aristotle are useful, but they lead down the
31 Chapter

II of The Problems of Philosophy. The emphasis on same is Russells

86
garden path to skepticism also. Hume noticed this. Hume needed only to reflect on his own
experience in order to get a Heraclitean hypothesis of change. Hume finds no thing or self
which stays the same as his thoughts glide by, posturing in this or that way. The problem
that arises due to the mixing of the Pythagorean and Heraclitean hypotheses is that the only
empirical hypothesis appears to be the Heraclitean one. But the doctrine of flux, by itself,
will not do the work of science.
Scientific methodologies will involve several different kinds of simplicity criteria
because science is fundamentally committed to a project where some things are said to
change and others to stay the same. It is also the case that scientific hypotheses make
substantive claims about what there is and about what the fundamental mechanisms of
change are. We are faced with a variety of questions about simplicity the moment that we
step into the scientific enterprise. The next order of business is to see what might be done
by way of organizing and categorizing kinds of simplicity in science.

87
CHAPTER 3
THE COMPLEXITIES OF SIMPLICITY

The previous chapter introduced several issues. Nave resistance to the philosophy
of simplicity really only serves to motivate it. Simplicity is subject to vagueness, and the
clarification of simplicity is a philosophical project which is motivated, at the very least, by
the fact that we wish to avoid slipping into the equivocal use of terms. It also turns out that
the arguments given to clarify simplicity defeat a few nave dogmatisms. On the one hand,
it is not the case that there are no philosophical problems of simplicity in science. Nor is it
the case that simplicity reflects merely pragmatic or aesthetic values, for this claim is either
false or too vague to evaluate. On the other hand, the clarification of simplicity shows that
it far from clear that Ockhams Razor is the final court of appeal in theory construction and
selection as many have believed it to be. Again, at the very least, this is because Ockhams
Razor is subject to vagueness. There are many razors, not all of which give compatible
results. But various principles of parsimony may indeed play important roles in science.
When a principle of parsimony is put to work in science, it is in conjunction with
many other considerations. It may not even be possible to imagine cases where all other
things are equal and some principle of parsimony, which is specified in a non-arbitrary
way, would be the only court of appeal for judging that one scientific theory is more likely
to be true than its competitors. The previous chapters discussion of Swinburnes thought
experiment shows how a defense of a scientific principle of parsimony with a ceteris paribus
clause can go wrong. Scientists are concerned with satisfying many diverse goals when
they construct theories. I hope that I have shown that philosophers must proceed very

88
carefully if they wish to construct thought experiments in the philosophy of science. The
general goals of this chapter are first to categorize simplicity criteria in science, and second
to elucidate the relations between these criteria and some of the other goals of science.
I hope that the previous chapter has also shown that it would be misguided to
claim that judgments of simplicity play no role whatever in science. The fact that principles of parsimony govern non-deductive inferential judgments (whether they are rationally
grounded or not) should be sufficient to show this. In this chapter I hope to explain precisely
why it is so difficult to conceive of situations where empirically adequate theories satisfy
equally the desiderata of science, and we are left only with simplicity judgments to select
one over the other for testing. Once the various features of scientific theories that might
be specified by simplicity criteria have been categorized and the other desirable features of
scientific theories have been identified, it becomes clear that scientific theory construction
involves a delicate act of balancing the various desirable features of theories.
I have also tried to show how it is that the grammar of science is fundamentally
committed to at least two very abstract notions of simplicity: the doctrines of sameness
and change.1 One simplicity doctrine is intuitively simpler than two, so in this way the
grammar of science is already committed to complexity. Complexity can be expected to
increase from this very abstract level as workable theories are constructed and the things
that stay the same and those that change are defined. Even Aristotles basic example where
a man comes to be musical from having been unmusical involves one subject term and two
1 Platos mix of the Pythagorean doctrine and the Heraclitean doctrine establishes a pattern which

is found repeated throughout Western philosophy, not just in science. See the final section of Chapter
One.

89
predicate terms. Additionally, if we read a bit further in Aristotles Physics we see that
Aristotle has at least two different conceptions of necessity that govern change.
Mario Bunge holds that science marches in the direction of complexity, not simplicity. Perhaps this is not at all surprising. It is a deeply entrenched feature of the grammar of
science that multiple predicate terms must be defined, that subject terms must be defined,
and that mechanisms of change must be posited. In this chapter I present Bunges detailed
analysis of simplicity and synthesize Bunges arguments about the relationships between
the various species of simplicity and some of the other desiderata of science into a new, and
I hope, useful form.2
This chapter also includes a discussion of the valuable work done by several other
philosophers who join Bunge in criticizing Logical Positivism. The period of the criticism of positivism is a fertile one for philosophy. In this period, we find Nelson Goodman
making important contributions to the analysis of simplicity. We also find Carl Hempel
and Ernest Nagel laying the ground work for the contemporary discourse on scientific explanation. The analysis of scientific explanation is related to simplicity on several fronts,
including the relation of simplicity judgments to the construction of explanations that are
general and that are conceptually related to other accepted explanations. This chapter introduces the terminological distinctions and arguments contributed by these philosophers.
It is useful to work through a bit of the history of the philosophy of science because these
authors contributed to analytic clarity on the issue of simplicity judgments and their rela2 The

title of this chapter is a tribute to Bunge, whose 1962 article is named The Complexity of
Simplicity

90
tions to the other desiderata of science. It is also useful to investigate these arguments to
show what reasons we have for rejecting the positivists wholesale hostility towards metaphysics. Although not all simplicity judgments in science make ontological commitments,
we find simplicity judgments embedded in a web of interrelated judgments, some of which
also make ontological commitments.
In the first section of this chapter I present Bunges categories of simplicity. Bunge
argues that simplicity is not of one kind but of many. Although judgments of simplicity are
part of scientific methodology, we should not expect an overall measure of the simplicity
of theories due to the extreme heterogeneity of simplicity. Bunge also gives arguments
which show why it is extremely difficult to give a formal basis for measures of simplicity.
I call this the the gauging problem. Bunge had several objectives in giving his analysis
of simplicity. One goal is to dispel the sorts of dogmatic views mentioned above that
simplicity is hopelessly vague or that simplicity is the final court of appeal in theory selection. Another goal is to show that it is far from clear that judgments of simplicity in
science can be reduced to judgments of one kind or to judgments of other kinds. Ockhams
Razor is mentioned in many places in philosophy, but it can be crippled by vagueness, so
philosophers can offer a bit of analytic clarity to this perennial issue. Bunge also makes
suggestions about future research programs and I hope to advance his project.
The second section of this chapter is an attempt to synthesize many of the arguments
given by Nelson Goodman in the 1940s and 50s with the arguments given by Bunge, and
Hempel from the 1960s. Truth may be one aim scientists have in constructing and testing

91
theories3 , but if it is, it is one among many aims. Other aims may include: explanatory
depth, linguistic exactness, accuracy, systematicity, unity of science, representativeness
(Bunges term), and testability. Call these the desiderata of science. The desiderata of
science are not all logically independent of one another. Depth, for example, is related
to testability, and testability is related to accuracy. As we shall see systematicity is related to testability, unity of science, and representativeness. Each of these terms and their
interrelations will be discussed in detail in the second section.
Discussions of the desiderata of science and their relations to one another are distributed throughout many articles and it is very difficult to discover what are the essential
features of each and to generate a comprehensive view of their relations to one another.
Perhaps some people can manage all of this information without the aid of tools. I suspect
that Nelson Goodman and Mario Bunge actually did so partly due to the fact that these
philosophers had a deep understanding of science. I, however, cannot manage them without the aid of tools. I developed a conceptual map to manage, organize, and analyze the
tremendous volume of interrelations between, what Bunge calls the metascientific criteria.
I present these results in the second section of this chapter. In order to engage this analysis,
we must first clarify the various species of simplicity relevant to science and the problems
involved in measuring them. Questions about the relation (or relations) between judgments
of simplicity and the truth of theories or their relations to ontology are important questions,
but I see no way to give a precise formulation of these questions without first clarifying the
roles played by various judgments of simplicity in science. We cannot begin to understand
3 This

should be appropriate to assert, regardless of ones conception of what the truth is.

92
the roles that simplicity judgments might play in science without first understanding how
simplicity judgments are related to one another and to the other aims of science. My hope is
that the conceptual map will aid in showing why this is and that it will also provide a useful
piece of the dialectic, since it is convenient to critically engage something that organizes
terms an relations in this way.
The questions of interest to philosophers about whether or not simplicity is, in any
way, a feature of the mind or of the mind-independent world may be of interest across interdisciplinary boundaries. Perhaps these questions are also of interest to physicists, cognitive
scientists, and social scientists. But because simplicity is not of one kind but many, a bit of
dialectical separation can be kept between questions about the relationship between some
judgments of simplicity and the truth and the other important questions about the roles of
simplicity in the construction and testing of scientific hypotheses and theories. Perhaps
some of the simplicity criteria at work in science are different from those of interest to
the classical epistemologist. In the philosophy of science there may also be field-specific
questions about how to classify and gauge simplicity.
There are also questions about what roles properly specified simplicity criteria
might play in scientific theory construction and testing. We do appear to have theories
that are empirically adequate in the sense that up until now they appear to account for the
relevant data. However, theories which compete in the sense that they fit existing data may
give different predictions, and competing theories are not conceptually equivalent because
they posit very different kinds of things and different fundamental mechanisms. Currently,
for example, there are many different interpretations of quantum physics which are not

93
conceptually equivalent.4 There are also various theories of General Relativity which are
not conceptually equivalent.5
The criteria which aid in theory construction, testing and selection are called metascientific criteria. Simplicity criteria are often listed among the metascientific criteria. I
follow Mario Bunge in calling the analysis of the metascientific criteria and their interrelations Metascientific analysis. Simplicity judgments may contribute to the construction or
testing of a theory by aiding in the satisfaction of the other metascientific criteria. Simplicity judgments may also conflict with the other metascientific criteria. In these cases
simplicity is usually trumped. Mario Bunge says that
only those simplifications will be admitted in science which render the theory more manageable, more coherent, or better testable: no simplification will
be accepted if it severely cuts down either those characteristics or the depth,
the explanatory power, or the predictive power of the theory.([10]pg. 123)

The problem of giving necessary or sufficient conditions for the relation between
simplicity criteria and the truth is only one of the problems for the philosophy of simplicity.
Elucidating the relations between different kinds of simplicity criteria and one another and
the other metascientific criteria presents an array of difficult and interesting problems. In
addition to these problems there are fundamental problems selecting a formal gauges of
simplicity.
The first section distinguishes general species of simplicity criteria and explains the
4I

do not know if it would be appropriate to say that the different interpretations of quantum
theory are themselves competing theories. I follow Bunge only to point out that these are not
conceptually equivalent.
5 R.H.

Dicke and A.N. Whitehead have given theories which are often cited as theories which
compete with Einsteins General Relativity. [50]

94
problems involved in gauging simplicity. The second section explains the interrelations between metascientific criteria. The concluding section explains how the analysis contributes
to the philosophical dialectic by revisiting some of the issues raised in the previous chapter.

3.1

Categorizing and Gauging Simplicity

Bunge organizes simplicity into five categories: syntactical simplicity, semantical

simplicity, epistemological simplicity, pragmatical simplicity, and ontological simplicity
[10]. This section presents Bunges arguments. Contemporary journal publications regularly feature articles which discuss issues related to simplicity in science [24] [54] [55] [22].
Bayesian curve fitting algorithms have seen a recent spike in academic popularity. Yet, I
find Bunge and Goodman too infrequently cited in these articles, because Bayesian curve
fitting algorithms depend upon syntactical gauges of simplicity which Bunge and Goodman have shown to have many deficiencies. It is worth giving some careful attention to the
articles that served a key role in developing contemporary debates. I also follow Bunge in
thinking that this analysis helps to dispel dogmatisms, to avoid unnecessary vagueness, and
to promote new research programs.

3.1.1

Ontological Simplicity

Ontological simplicity, in the sense relevant to science, is a comparative measure of

the number of kinds of entities or relations posited by a theory.6 Questions about simplicity
6I

say in the sense relevant to science, because there may be other issues in ontology where
simplicity plays a central role. Questions about whether or not propositions are to be admitted to
ontology or whether there are universals or binding relations between universals and particulars are
issues which may also involve judgments of simplicity.

95
are fundamental for ontological inquires. Recall that when we judge simplicity, there is
no upper bound on complexity and the lower limit of simplicity is one (or unity). If all is
change, or if there is only one thing and nothing else can be asserted truly, then the correct
ontological hypotheses were promulgated long ago by the Pre-Socratics. However, few are
convinced of these views. If we think that there is an assertory foundation for science (that
scientific statements are either true or false), then we move away from these views and in
the direction of complexity, because, at the very least, we are committed to Platos mixed
view: a verisimilar scientific theory would be committed to some things changing while
others stay the same.
Platos synthesis of the Heraclitean and Pythagorean hypotheses is adopted by Aristotle and henceforth forms the basis for the grammar of science. Some things change and
some things stay the same. As scientists and philosophers we do two projects (often simultaneously). We categorize which things are of the stay-the-same sort and we posit mechanisms by which things change. On the face of it, science is committed to some ontological
complexity. This is because science is a mix of the pure doctrines of change and unity.
But it is not clear what would count as ontological complexity. It is notoriously difficult to
determine how to organize experience so that we could even start to get a measure of the
simplicity of things which might correspond to reality. Consider the two sets of objects in
figure 3.1.

Figure 3.1: The Counting Problem

96
If we think about ways to organize the two sets so that we may compare their relative
complexities, it is natural to consider various counting strategies. On one way of counting,
the two sets are equally simple (or equally complex) because they each contain two objects.
On this way of counting, we can say that each set has a complexity of 2. But we also
notice that these two sets are different in certain respects. So the identification of the
ways in which they differ might be the first step in establishing some basis for comparative
judgments. On one way of counting colors, the set on the left counts 1, and the set on the
right counts 2 because there is one type of color on the left and 2 types of colors on the
right. But if we count by comparing types of shapes, then the left set counts 2, and the right
set counts 1.
It is difficult to say what, if anything, about the context of the presentation of these
objects could determine how we are to rank the counts given to various types of attributes.
Are shapes to be given a more hallowed ontological status than shades are to be? In this
case, the problem grows even more puzzling when we consider counting by the number of
sides of each shape. Now the circle all by itself, presents a problem. Does the circle have a
side-count of 1 or 0 or ?7
7 Popular

author William T. Vollmann said that, on the subject of what ought to be, lets remind
ourselves that the purpose of conceptualization is to transform realitys perceptual randomness into
patterns. A perfect circle excels in beauty, elegance and mathematical simplicity.([63] pg.31) Vollmann goes on to claim that there has been psychological research which shows that people remember things as circular even if they are not. He does not cite the source and this point, if true, would
be irrelevant to mathematical simplicity because it is not clear that a circle is not just a special case
of an ellipse. Vollmann is neither a scientist nor a philosopher. We should not expect this sort of
carelessness from scientists or philosophers. I find the quote interesting because it might reflect
just the sort of ideas which people of our epoch are conditioned to accept without question. After
all, despite Vollmanns carelessness, he is a very popular award winning author, so perhaps he does
reflect popular dogmatisms.

97
The problem presented here is not a problem judging simplicity. We know how to
compare larger numbers with smaller numbers. If we know what to count then we have
no problem counting and no problem comparing the results. The problem may be seen as
giving us a two part objection strategy: 1) in some situations it is not clear how to rank
incompatible counts, and 2) in some situations it is not clear how to get started counting in
the first place. This problem marks one trail head into the jungle of metaphysics. Richard
Fumerton gives a similar example where the difficulty is in deciding which similarities or
differences will serve as the basis upon which we organize books on a bookshelf.8 Fumertons point is that radical metaphysical relativism is not the only port in this storm.
The fact that we notice similarities and differences between objects and their attributes in a variety of ways is common dialectical ground held by all parties venturing
metaphysical inquiry. The fact that there is a problem ranking incompatible measures of
the complexity should not stack the deck for or against any metaphysical positions, because
the problem itself favors no position. Still, the suggestion that we count the sides of a circle
may appear to be symptomatic of some philosophical sleight of hand. If so, this l`egeret`e
des mains has given philosophers to mystify themselves for two and a half thousand years.
It is true that in some cases we would be forced to solve deep and enduring issues in metaphysics in order even to propose a measure of complexity. However, this is not always the
case. In some cases we can just stipulate, perhaps on pragmatic grounds, the criteria which
select the objects to be counted, and we can establish a measure of simplicity. In such cases
it is not a problem to count the things specified, and it is not a problem making judgments
8 Realism

and the Correspondence Theory of Truth (2002)

98
of simplicity. Epistemological questions remain, such as determining which, if either, of
two incompatible simplicity judgments is justified. The question of justification will be
addressed in the next chapter.
Every species of simplicity may be subjected to this same objection strategy. In
some cases it is not clear how to establish a formal measure of simplicity/complexity. In
other cases formal measures are, at least in principle, available. When formal measures are
available it is very easy to select another simplicity criterion and generate a measure which
gives incompatible results (objection strategy (2)).

3.2

The Simplicity of Systems of Signs

Theories are systems of hypotheses with interpreted terms. Systems are interrelated
hypotheses, and there are various ways of relating hypotheses. Hypotheses are constructed
out of linguistic symbols, and for this reason we can consider the linguistic features of
hypotheses which might be specified by simplicity criteria. Bunge says that,
There are basically four types of sign in the field of discourse: terms,
propositions, proposals, and theories. Hence, we must begin by studying the
formal or structural simplicity of terms (designating concepts), sentences (expressing propositions and proposals), and theories (systems of propositions).
In turn, since propositions and proposals are built out of predicates (like between), names of constants (like Argentina), and variables (like x), logical
constants (like or), logical prefixes (like all), and modal prefixes (like possibly), a methodical study of logical simplicity should begin by examining the
formal complexity of predicates. 9 ([11] pg.114)
9 I do

not know why Bunge uses the word proposition when he wishes to discuss the simplicity
of systems of signs. It might seem like statement or expression might be better since we wish
to gauge the complexity of systems of symbols and it is not obvious how to show that propositions
are structured in the same way that sentences are because different languages have different rules
of grammar. Also, I do not know if Bunge makes a distinction in the uses of words proposal,
hypothesis, and axiom. These words have different definitions and it seems like all of them
ought to be included in the analysis. Probably, the running together of these terms reflects a positivist

99
In contrast with the tightly knit systems of signs found in scientific theories, we
sometimes find loosely related groups of conjectures in non-scientific discourses. I am often puzzled by what it is that makes some people apparently more successful than others
at slinging terms like upper and lower chakra in New Age contexts far removed from
the ancient Upanishads, that is, torn from their original systematic place. Some of this
language, I think, fails to be explanatory. It appears, instead, to be some sort of magic for
social maneuvering. Part of what makes the use of these terms successful in specific social circles is probably that they are deployed in a deliberately vague way. After all, some
kinds of vagueness are favorable to predictions recall the first chapter discussion of the
inferences made by Sherlock Holmes. The contrasting case with non-scientific discourse
is useful to consider because it may draw attention to a virtue of scientific theories. Most
scientists and philosophers are in agreement about which fields of discourse are at least
candidates to fall under the heading of science. While some may disagree about whether or
not psychology (for example) is a science, most will agree that it is an important question
to ask whether or not it is. However, most scientists and philosophers will agree that it is
not worth discussing the scientific merits of New Age babblings, mystical prophesies, or
conspiracy theories. Although the so-called pseudo-sciences may have some of the characteristics of genuine scientific theories, they are lacking in far too many to be seriously
considered scientific. Often vagueness is what takes a discourse far from the path of science. The contrast shows us that we seek a kind of precision in the language of science.
residue because several years passed before the axiomatic account of theories stood sufficiently
challenged.

100
One reason to seek formal mathematical or logical expressions in theory construction is for
precision.
We can think of theories as systems of signs. Some of the signs are formal and some
of them are the signs of natural language. The formal language in which a theory is codified
can contain formal logical signs like or, not and all. They also consist of what are
often called extralogical bases, and the natural language interpretations of these symbols.
The formal language will also have rules that specify which sequences of symbols count
as well-formed sentences or formulae. They will also have rules specified which govern
the transformation of well-formed-formulae into other well-formed-formulae. These are
the transformation rules of the system. The simplicity of extralogical predicates is the first
topic of study.

3.2.0.1

The Syntactical Simplicity of Predicates

Consider the various ways in which a predicate may be simple or complex:

Predicates may be formally simple like the predicate extended. They may also be
complex like extended over a sphere.
Predicates may be one-placed or many-placed.
Predicates may be first-order or second-order.
Predicates may be dichotomic (presence/absence predicates like hole) or metrical
10

10 This

list is quoted from Bunges The Complexity of Simplicity(1962)

101
Immediately it is apparent that there are several bases upon which to judge the
simplicity of a predicate. Following the pattern suggested by the problem of establishing a
basis for judgments of ontological simplicity it is obvious that, here too, our choice of the
gauge for the logical simplicity of predicates must be non-arbitrary. This problem is even
more difficult than it might first appear to be.
The first problem is that a predicate does not necessarily reveal the complexity of
what it denotes. Equally simple signs S and C can be used to denote sequences of
ideas which are more or less complex. For example, S could stand for the class of gradeclasses of pupils at Sonora High School and C could stand for the class of grad-classes of
pupils at Cassina High School. Perhaps both schools have 100 students, but let us say that
Sonora has four grades (9-12) and Cassina has only one grade. In this case S would denote
a class of grade-classes of pupils with the more complex structure11 .
Bunge considers an even trickier example. Let us say that we wish to gauge the
complexity of a dichotomic predicate like black. Bunge walks through a proposal about
how to measure the complexity of this predicate in order to show what is unsatisfactory
about it. We might attempt to measure the complexity C of the predicate by adding the
number of its atomic constituents A to the number of number of places P and to its degree
D.

C =A+P +D
The complexity measure of the predicate black would be,
11 This

sort of example is used by Goodman (1943)

(3.1)

102

C = 1 + 1 + 1 = 3.

(3.2)

This suggests that the complexity measure may be normalized by dividing by 3.

Since simplicity is inversely proportional to complexity we could get a simplicity measure
from,

S=

3
1
=
.
C
A+P +D

(3.3)

This would give a function that ranges from infinite complexity (simplicity of 0)
to perfect simplicity (1). However, this measure is generated from a common sense usage
of the term black. In physics black has a special definition involving the extralogical
predicates electromagnetic wavelength, absorbs and absorption coefficient x. The definition is an object is black when it completely absorbs all electromagnetic wavelengths.
This might be stated more formally, black (x) =df (y) [electromagnetic wavelength y
x absorbs y (absorption coefficient x = 1)]. The atomic value of the predicate electromagnetic wavelength is 2 (because it involves a two predicates) and 1 for absorbs and for
x. Similar to the example above, the predicate black would denote a complex structure
given its technical definition in physics. The total molecularity is 2 + 1 + 1. Since absorbs
is a two place predicate the total number of places is 1 + 2 + 1 and the total order of degree
is 1 + 1 + 1. This gives,

S=

1
3
3
=
= .
C
4+4+3
11

(3.4)

103
Equations (2.1) and (2.2) give incompatible results. On one proposal about how
to measure the complexity of a predicate it is not clear how to establish a non-arbitrary
way to select the gauge. Scientific theories contain formal predicates and natural language
interpretations of those predicates. It is not clear how we can establish a non-arbitrary
formal measure of the syntactical simplicity of predicates which captures both features
of a scientific theory and which gives consistent results. The problem is perhaps even
worse because even the formal measure of simplicity of the scientific predicate black
may hide some of its structure. Some metrical predicates like wavelength have an infinite
range while others, like absorption coefficient x have a finite range (but may take an
infinity of values). We might ask what would non-arbitrarily justify either excluding or
including these features of the syntactical complexity of predicates in a gauge of simplicity.
Additionally, it is not clear that we ought to adopt this proposed measure of predicate
simplicity because it involves the linear sum of non-homogeneous predicates. We would
need some reason for thinking that non-homogeneous predicates should stand on par in
their relative complexity measures.
There may be further problems gauging predicate simplicity. The basic predicates
of a scientific theory appear immersed in a system and may be related to other symbols
by law statements and perhaps this too ought to fall under the scope of the analysis of
the complexity of predicates. Also, many symbols are governed by certain transformation
rules. Bunge points out that a symbol which is part of a system may seem to be simple on
the face of it, yet by the transformation rules allowed by the system, it may be infinitely
complex. For example, the numeral 1 may be written as,

m n
, ,
m n

i (i), (1)2n and so

104
on. These predicates are not syntactically equivalent, so just by applying the transformation
rules specified by a system to some of the predicates we might achieve an ad hoc gain in
syntactical simplicity.[11] If the transformation rules were given so that each of these symbols is supposed to express the same propositional content, then contradictory measures of
simplicity are derivable by application of the transformation rules. This is another example
of objection strategy (2).

3.2.1

The Syntactical Simplicity of Formal Expressions

The problems of predicate simplicity are bad enough and formal expressions are
strings of symbols. Nevertheless, much of the more recent philosophical literature on simplicity has been dedicated to the syntactical simplicity of formal expressions (or propositions, if we follow Bunges terminology) than it has been to the syntactical simplicity of
predicates. Scientists in many fields employ the results of this trend in analysis when they
use computer programs to fit functions to data. So the analysis of the syntactical simplicity
of formal expressions may be of interdisciplinary interest.
The analysis of syntactical complexity will involve several things. Bunge says that,
syntactical simplicity (economy of forms) depends on (1) the number and
structure (e.g., the degree) of the specific primitive concepts (basic extralogical
predicates); (2) the number and structure of independent postulates, and (3) the
rules of statement transformation. ([10] pg.121)

Early in the twentieth century Dorothy Wrinch and Harold Jeffreys proposed a measure of the syntactical complexity/simplicity of mathematical expressions in a probability
text book and a series of articles. If it is assumed that the connectives are all equally simple,
then we might attempt to gauge the simplicity of an expression by counting the number of

105
connectives or the number of independent freely adjustable parameters. This approach does
neglect the complexity of the atomic terms discussed above. Jeffreys wanted his simplicity
measure to give the prior probability, given by 2k for physical theories. On the assumption
that all physical laws are expressible in differential form, he suggests that we can analyze
families of equations with the simplicity measure k where k = order + degree + the absolute value of coefficients. The order of a differential equation is the highest derivative. The
degree of a differential equation is the power of the highest derivative term. Consider the
Newtonian equation for position under gravitational acceleration,

1
s = a + vt + gt2 .
2

(3.5)

Cleared of roots and fractions 3.3 is expressed in differential form as,

d2 s
= g.
dt2

(3.6)

In this case, Jeffreys variable k is 2 + 1 + 1 = 4. This is generalized to families of

equations. The linear family of curves (LIN) is given by,

f (x) = 0 + 1 b.

(3.7)

The parabolic family of curves (PAR) is given by,

f (x) = 0 + 1 b + 2 c2 .
The cubic family of curves (CUB) is given by,

(3.8)

106

f (x) = 0 + 1 b + 2 c2 + 3 d3 .

(3.9)

On Jeffreys proposal CUB would have simplicity k = 5 which is more complex

than PAR with simplicity k = 4 and LIN with simplicity k = 3. If LIN expressed a proposition embedded in a physical theory, then LIN would have the higher prior probability.
The objections to this approach are well known. The first objection is the subfamily objection. If we think about the curves generated by functions as fitting some data
then we could achieve syntactical simplicity in an ad hoc way. Since each of the functions
with a smaller number of adjustable parameters is a sub-family of the families with more
adjustable parameters there are infinitely many cases where CUB overlaps with PAR and
PAR overlaps with LIN. These are cases where the variable with the highest exponent takes
the value of 0. We could get an ad hoc gain in syntactical simplicity by fixing the values
of a higher dimensional curve at the values which give the best fit (where fit involves not
merely the fit with the original data but some hindsight judgment about what also would
have given the best predictions). In this way the number of adjustable parameters would
be reduced and the syntactical simplicity value would be maximized. Forster and Sober
[24] suggest that a rule can be added to exclude overlapping cases. However, this solution
depends upon an inductive inference that curves which fit data too well are too good to
be true. Since we ultimately wish to investigate whether or not any simplicity criterion
is justified, this solution may not be acceptable because it begs the question against the
inductive skeptic.
Another objection raised by Hesse [31] is that merely changing the choice of coor-

107
dinate system will give contrary simplicity measures for equations that generate identical
curves. Consider the equation for a circle in Cartesian Coordinates,

r 2 = x2 + y 2 .

(3.10)

A spurious gain in simplicity can come from selecting polar coordinates where the
equation for a circle is,

r = a.

(3.11)

If equations (3.8) and (3.9) both express the same proposition, then a contradiction
is derivable from Jeffreys gauge (another example of objection strategy (2)).
Bunge objects that Wrinch and Jeffreys did not analyze predicate complexity and
restricted their proposal to metrical predicates, and so the problems of predicate complexity
remains on this approach. Also, it is not clear why order and degree are given equal weight.
Again it is not clear what would justify the linear addition of the values of heterogenous
characteristics. Perhaps we should be suspicious about a measure which makes a 1/2-order
variable more simple than a first order variable. Finally, it is not the case that all physical
laws are expressible in the form of differential equations ([11] pg.119-120) which is a
fundamental assumption made by Jeffreys in his early work. So the applicability of this
method is limited.

108
3.2.2

The Simplicity of Theories

Theories are complex wholes constructed out of postulates, hypotheses and propositions which are themselves constructed with predicates and are embedded in a logical
system with transformation rules. At present, no formal gauge of the complexity of theories has been offered. Goodman [27] and Bunge [11] point out that it will do no good to
begin counting postulates since all postulates could be turned into one postulate by conjunction.
It is difficult to know how to approach gauging the simplicity of a theory. Bunge
says,
That not only the number and complexity of a theorys postulates but also
the number and complexity of its transformation rules are relevant to its Lcomplexity is plain: after all, we are interested in the complexity of systems
that, from a semiotic point of view, can be regarded as languages, and the
formal complexity of a language is determined by the complexity of both its
vocabulary and its grammar ([11] pg.122).12

We do not know how to gauge the simplicity of transformation rules themselves

and the application of these to a systems basic postulates may give infinitely many results.
What matters is the conceptual cohesiveness that the transformation rules (and definitions)
contribute to. But, as of yet, there is no measure of conceptual cohesion.

3.2.3

Semantical Simplicity

Semantical simplicity is at least as important as syntactical complexity. There are

various theories of meaning. Bunge appears to hold that it might be possible to count the
12 L-simplicity

stands for logical simplicity. Bunge uses this term in his general discussion of
syntactical simplicity

109
specifiers of meaning of the basic predicates of a theory. Roughly, the idea is that theories
may involve several operational definitions and that we could count the number of basic
(unoperationalized) terms. If this view about semantics is false, then the problems involved
in selecting a gauge of semantical simplicity may be more serious than the problems Bunge
outlined. Perhaps things would be easier if Bunges view about meaning were correct
because we might be able to count specifiers of meaning. Even in that case, Bunge shows
how difficult it would be to select a non-arbitrary gauge of semantical simplicity.
Bunge says that,
Semantical simplicity (economy of presuppositions) depends on the number of specifiers of meaning of the basic predicates. Semantical simplicity is
valued within limits because it facilitates both interpretation of signs and fresh
starts.([10] pg.121) 13

The terms theory and hypothesis are of equal syntactical complexity, but theory is semantically more complex than hypothesis because the latter appears in the
definiens of the former. The same can, perhaps, be said of electron and mass. However,
on Newtonian mechanics force and mass are of equal semantical complexity because
they are primitives in that system. Bunge investigates the proposal that we might measure the semantical simplicity S of a term t in a language L as the inverse of the terms
extralogical specifiers of meaning M :
13 I

do not know exactly what Bunge means by fresh starts but on the basis of some of his
other publications, it would appear that he has in mind something like the situations that lead to
the construction of revolutionary scientific theories. This is because Bunge holds that scientific
theories tend to become increasingly complex until they become untestable or disunified with the
other accepted theories in the body of science. Bunges view shares some features with Kuhns [41].
To borrow Kuhns terminology, normal science contributes to the increasing complexity of working
theories until crises arise. [12]

110

S(t, L) =

1
.
M (t, L)

Bunge points out that several objections arise. In the Peano arithmetic 1 and successor are primitives and each would have a semantical complexity measure of 0. Since
2 is the successor of 1 it would have a semantical complexity of 1. However, if we
choose 0 as primitive then 1 has a complexity value of 1 and 2 of 2 (objection strategy
(2)). Additionally, it is counterintuitive that 10,000 would be 1000 times more complex
than 10. This problem might be met by choosing a smoothing function: log M (t, L). The
problem with this proposal is that the semantical complexity of 1 would be 0. The corresponding simplicity value for 1 would be indeterminate. There are infinitely many ways
to formulate a gauge of semantical simplicity. It is sufficient for this objection to notice
2

that many alternative smoothing functions are available like: 1 eM , log(1 + M ) and
tanh(M ). Even if a formal measure of semantical simplicity may be, in principle, possible
for formalized languages, it is impossible for natural languages. The analysis may have
to encompass the whole of human culture in order to track down the meaning specifiers
for terms like idea, and culture is a moving target. Finally, if the meaning specifiers for
a formal system could be counted, we could not justify giving them all the same weight
unless it could first be determined which were conceptually primitive[11].

3.2.4

The Semantical Simplicity of Expressions

Bunge shows us that there is a similar pattern of problems for gauging the semantical complexity of expressions (or propositions). If a measure of the complexity of the

111
terms of an expression were available, we might be tempted to give a simplicity gauge as
the inverse sum of the predicates in the statement. However, in order to do this each predicate would have to be intensionally independent. And this is an unlikely situation for the
theoretically embedded statements of science. Bunge suggests that it might be feasible to
get a comparative gauge of the semantical simplicity of the expressions of a formal language by counting the number of presuppositions required by a statement. Bunge suggests
that by this standard the statements the Universe exists for all eternity would be semantically simpler than the statement the Universe has a beginning because the latter requires
suppositions about creation mechanisms. I suggest that even this example is subject to
the same pattern of criticisms which Bunge has raised for the other gauges of simplicity.
Cosmological arguments posit mechanisms (or movers) for the basic structure of the world
regardless of its age. So even this example does not provide a clear case where the counting
of suppositions will help to gauge semantical simplicity.

3.2.5

The Semantical Simplicity of Theories

Bunge claims that a gauge of the semantical simplicity of theories would be of

great use for scientists. He cites General Relativity as an example of a theory which, although it has empirically equivalent competitors, is preferred for its semantical simplicity
of postulates despite its syntactic and psychological complexities (a sub-class of pragmatical complexity). We do not yet know how to gauge the simplicity of terms or postulates,
so it is not clear what more can be said here, but Bunge has urged semanticists to work
on this project. Interestingly, modern work on semantics has gone a route which involves

112
ontological proliferation (often positing possible worlds and their inhabitants).

3.2.6

Epistemological Simplicity

Bunge defines epistemological simplicity saying that

epistemological simplicity (economy of transcendent terms) depends on
closeness to sense-data. Epistemological simplicity is not desirable in and
of itself, because it conflicts with logical simplicity and with depth. 14 ([10]
pg.121)
The reverse of epistemological simplicity is epistemological complexity or abstractness15 . Bunge raises a criticism of phenomenalist languages (to whatever extent these
are possible). This criticism is also presented in detail by Cartwright (1983). Bunge says
that phenomenalist languages achieve epistemological simplicity at the expense of syntactical complexity and a loss of insight (or at the cost of shallowness). It is not immediately
clear why phenomenalist languages would cost insight. This is discussed in more detail
in the next section in conjunction with the discussion of explanatory depth. As a matter of fact about the way that scientists actually construct theories, it is the case that they
posit ways that the world is, and they posit fundamental mechanisms of change. They do
not choose fundamental variables so that their referents (or meaning specifiers) are mere
sensations[11].
On the other hand, scientists do not seek epistemological complexity either. They
seek to limit the number of transempirical terms to which a theory is committed. Somehow
14 The
15 It

term logical simplicity covers both syntactical and semantical simplicity in this case.

might not matter much if one is a rationalist or a skeptic about sense-data. We can follow
Bunges project and then attempt to apply it to whatever it is thought might be the atoms of knowledge. Bunge himself is skeptical about sense-data. He mentions this in The Myth of Simplicity
(1963)

113
there must be a trade-off between epistemological simplicity and abstractness. Scientific
theories do posit kinds of things and mechanisms of causation and, as we shall see in the
next section, this is because such posits contribute to the cohesiveness of science, and to
the serendipic power of theories (the power to make novel predictions).
Epistemological simplifications may come about due to judgments not based on
the distance away from sense experience. If the postulates involved in a theory are not
scrutable, then we will endorse an epistemological simplification. Bunge says that
Unnecessarily complicated assumptions and theories should be avoided;
that is, hypotheses and theoretical systems employing inscrutable predicates,
such as Providence and collective unconscious, should be shaved with Occams razor. Notice, however, that Occams razor does not hang in the air, but
falls under the more general rule, Do not propose ungrounded and untestable
hypotheses([11] pg.129).
This, perhaps, suggests conceptually reducing epistemological simplicity to syntactical simplicity with certain metascientific constraints. That is, we would not add to a
theory basic predicates if they conflict with the other desiderata of science, like scrutability
and depth. The gauging problem arises due to a pattern of objections which show that,
in some cases, we do not know what to count, that incompatible measures arise for any
criterion and for any gauge by changing one fundamental assumption, that we do not know
how to rank the values of a heterogeneous set of objects (objection strategy (1)) or that the
formal gauge is arbitrarily selected from many other possible gauges, some of which give
contrary measures (objection strategy (2)). Concerning epistemological simplicity, it may
turn out to be the case that what might appear to be a simplifying application of Ockhams
Razor is better understood as a judgment based on scientific principles with epistemological relevance. This suggests that perhaps epistemological simplicity can be gauged by the

114
inverse proportionality of the count of inscrutable predicates in an expression, a hypothesis
or a theory. I do not know if any inscrutable predicates ought to be included in a scientific theory. However, scientists do posit causal mechanisms, and Humeans will view the
causal necessity involved in these as inscrutable. We might also be suspicious about the
scrutability of the posits made by possible world semanticists.

3.2.7

Pragmatical Simplicity

Bunge analyzes pragmatical simplicity into five kinds,

1. Psychological simplicity (intelligibility).
2. Notational simplicity (economy and suggestive power of symbols).
3. Algorithmic simplicity (ease of computation).
4. Experimental simplicity (feasibility of design and interpretation of empirical tests).
5. Technical simplicity (ease of application to practical problems).([10] pg.121-122 and
[11] pg.130)

3.2.7.1

Psychological Simplicity
Bunge suggests that psychological simplicity could be measured relative to individ-

uals by the amount of time it takes to learn something, master a skill, or to solve a problem.
It also could be measured as the inverse proportion of the number of unsuccessful trials.
Perhaps cognitive science will be able to add something to these proposals in the future.
Psychological simplicity is certainly a desirable feature of scientific practice although it is not clear how it could be an intrinsic feature of systems. Science involves

115
modeling and modeling involves simplifying assumptions. Recall the first chapter discussion of the orbital calculations made by LeVerrier and Newcomb. [50] Certain assumptions
are made in order to begin a systematic inquiry which may be dropped later in the development of the theory. For instance, it would not be uncommon to assume that orbital bodies
move free from friction, but the Zodiacal Light Hypothesis may have suggested that such a
simplifying assumption would inhibit a high degree of precision in the calculations. There
is no general solution to the three-bodied orbital problem in Newtonian Mechanics. Yet, by
introducing certain simplifying assumptions we can model the behavior of the solar system
with Newtonian Mechanics using the methods of discrete integration. Modeling planetary
interactions involves more steps for increasing degrees of accuracy.
Psychological simplicity is also of pedagogical interest. We explain things to others
by appealing first to ideas or concepts which are easy to understand and then we aggregate
them to bring the student to a new level of understanding. This depends, at least partly,
on social and cultural conditions. In vulgar terms, the conscientious instructor is always
looking for ways to connect with students. What counts as intuitive may shift from year to
year and from class to class. Bunge says that,
Psychological simplicity is desirable for both practical (e.g., didactic) and
heuristic purposes: if preliminary theoretical models are to be set up, details
must be brushed aside, and easily understandable notions (intuitive ideas)
must be seized upon. The obvious must be tried first, if only to dispose of
it early: this is a well-known rule of intellectual work. However, it should
be borne in mind that (1) psychological simplicity is culturally and educationally conditioned; i.e., is not an intrinsic property of sign systems; (2) the
deliberate neglect of a given factor should always be justified; (3) we must be
prepared to sacrifice psychological economy to depth and accuracy whenever
the former becomes insufficientfor, whatever science is, it certainly is not a
business whose concern is to save experience with the minimum expenditure
of work.([11] pg.130)

116
3.2.7.2

Notational Simplicity
Bunge says that

Notational simplicity, or economy and suggestive power of signs. Thus,

vector and tensor representations are notationally simpler than the corresponding analytica (expanded) mode of writing, and the symbol P (h/e) for the
probability of the hypothesis h on the evidence e is suggestive, hence easy to
retain. Needless to say, compactness facilitates interpretation and favors retention, and well-chosen symbols may be heuristically valuable.([11] pg.131)

In constructing systems, theorists try to select symbols which are suggestive and
therefore easy to remember and they seek a notational basis which is easy to manipulate.
On the other hand nominal definitions may hide semantical or syntactical complexities.
Just consider the loss of meaning involved in the expediencies of the current fad known as
text messaging.

3.2.7.3

Algorithmic Simplicity
Bunge says that,

Algorithmic simplicity, or ease of computation, is something like the reciprocal of the number of steps in logical or mathematical calculation (this number
is actually computed in the programming of machines). Algorithmic simplicity depends on logical, semantical, and psychological simplicity, though not in
a simple way. Thus the relation y < x is S-simpler than y = x in the sense
that the equality relation is defined in terms of the two inequality relations.
But the corresponding point set xy(y < x) is algorithmically more complex
than xy(y = x), because finding the frontier of the former set presupposes the
determining of the latter set. ([11] pg.131)

In some cases algorithmic simplicity is achieved by a syntactical or semantical complication. For example, some integration techniques involve substituting more complex
terms for single variables. Bunge also points out that some theories may have a very sim-

117
ple notational basis but involve algorithmic complexities. This is acceptable to science
when the theory gives an improvement in depth and accuracy.

3.2.7.4

Experimental Simplicity
Bunge says that experimental simplicity is simplicity in the design, performance,

and interpretation of observations or experiments. He says that experimentalists prefer

hypotheses having the operationally simplest consequents, however complex they may be
in all remaining aspects. ([11] pg.131-132)

3.2.7.5

Technical Simplicity
Bunge uses the term technical simplicity to refer to the relative ease of application

of theories or practices when the goals are non-cognitive. Time or financial constraints may
influence the judgment to put rough or simple theories into practice. Following Bunges
pattern of analysis, the suggestion might be made that technical simplicity be measured by
the relative amount of time or resources saved by selecting one theory or process over another. As long as science is conducted in roughly the same inertial reference frame, then the
time intervals between the introduction of a theory or process and its successful application
might provide a basis for gauging technical simplicity. It seems natural that the economic
notion of opportunity cost might give an economic gauge of technical simplicity. However,
until the final theory of economics is given, I cannot see how to gauge technical simplicity in purely economic terms. My guess is that scientists and engineers regularly make
judgments of technical simplicity on the basis of algorithmic simplicity when computing
is involved. This is because an economic value can be given to floating point operations

118
per second. Scientists may need to run very complex computational models on mainframe
computers, and so they must rent FLOPS.

3.2.8

Why Categorize Simplicity?

Bunge has given us a way to organize the simplicity in science dialectic. Many
different kinds of simplicity criteria are relevant to science. Simplicity criteria in science
are heterogeneous because gains in some kinds of simplicity come at the expense of other
kinds of simplicity. Given the extreme heterogeneity of simplicity, no overall measure is
possible. We do not even know how to gauge some species of simplicity. We have only
rough sketches of how to measure simplicity of other types.
One reason that Bunges analysis is useful is because it lays the ground work for
seeing how it is that the various species of simplicity are relevant to science. This is something that we should know better when we ask questions about the justification of specific
measures of simplicity which are already at work in science.
Clearly, a great deal of work is still needed even to get basic proposals of formal
gauges of simplicity of some kinds. Formal gauges of simplicity are of great use in science,
both in the early stages of theory development and in error analysis. We also want to know
which kinds of simplicity judgments are irrelevant to science.
At the very least Bunges analysis should put to rest some of the nave dogmatisms
mentioned in the previous chapter which threaten to stop the philosophy of simplicity before it gets started. Ockamists should be able to see that principles of parsimony involving
criteria that select ontological objects can be formulated with some clarity and that these

119
do not play the only role in scientific methodology. Additionally, the Ockhamist may loose
some of the confidence which usually accompanies dogmatism when it is noticed that there
are serious problems justifying the selection of criteria which select ontological objects and
additional problems justifying gauge selection. The nave pragmatist should be able to see
that it is not true that there are no philosophical problems of simplicity in science. It is
not the case that all of the kinds of simplicity at work in science reflect merely pragmatic
or aesthetic values. In so far as the final justification for desiring science may turn out to
be pragmatic, it may also be the case that only pragmatic reasons are available for the justification of syntactic or semantic judgments of simplicity in science. But that would not
mean that a particular species of simplicity itself is of the pragmatic kind.

3.3

Metascientific Criteria and Their Relations

Mario Bunges 1961 article The Weight of Simplicity in the Construction and Assaying of Scientific Theories yields several results. [11] The general goal is to show that it
is misguided to think that simplicity judgments are the final court of appeal in theory construction and selection. On Bunges view, there are several ways that such a notion might
be misguided: 1) Simplicity is not one concept, but many and the extreme heterogeneity of
the simplicity criteria employed in science rules out the possibility of any overall measure.
2) Simplicity is at best a very weak criterion of theory construction and selection because
in many cases where it conflicts with the other desiderata of science, simplicity is trumped.
3) Simplicity is not relevant to many of the desiderata of science and in some cases, due to
the fact that different species of simplicity generate competing measures, the relevance to

120
some of the desiderata of science is ambiguous.
It is not clear how to evaluate Bunges argument that simplicity criteria are, at best,
only weak criteria of theory construction and selection. It may be the case, as Bunge has
argued, that simplicity criteria are weak relative to the other desiderata of science. But
if any simplicity criterion turned out to be indispensable for science, then we might, by
another standard, say that such a criterion is extremely important. Alternatively, even if
simplicity criteria in science were justified on purely pragmatic grounds, then their analysis
would still be valuable at least for pragmatic reasons. What is of interest is that Bunge
has given arguments about the relations between various simplicity criteria and the other
desiderata of science. I wish to advance Bunges project by focusing on developing the
analysis of the these relations. This project is ambitious enough as it is, so I will omit
any discussion of Bunges arguments regarding the irrelevance of simplicity criteria to the
desiderata of science.
To the best of my knowledge, no one has contributed to Bunges project to give
a comprehensive analysis to the interrelations between the desiderata of science. Others
have focused on bits and pieces of these issues, but never the whole thing. It is natural to
focus on bits and pieces of the puzzle because the numbers of interrelations between the
desiderata of science get unmanageable after only a few arguments, and each of these issues
is deserving of focused discussions. Bunge gives twenty arguments about the interrelations
between the desiderata of science. As stated above, my work is not comprehensive either. I
synthesize as many of Bunges arguments as I can into the beginning of a holistic picture of
the interrelations between the desiderata of science. Because there are so many features of

121
this analysis, I have used a conceptual map to organize and present the results of Bunges
arguments.
In order to set up the conceptual map, it must be determined what are the key terms
of the discourse. This is very difficult. The reason is that Bunge uses technical terms in a
very specific way, although it is not the purpose of his analysis to advance detailed theories
about the uses of all of those terms. So, in order to elucidate the desiderata of science it
is necessary to consult a few other sources. But it is often the case that the terms are used
differently by other authors or there are arguments given that we should conceive of the
desiderata of science differently. For example, Bunge uses the term refutability, and it
is not clear in the 1961 and 1962 articles that this term ought to be used as a synonym
for falsifiability which means only that it must be, in principle, possible to refute a hypothesis.16 Also, Bunge discusses explanatory depth but does not advance a theory of it.
Theories of scientific explanation have recently involved discussions of depth and this can
be seen in the work of authors such as Michael Strevens, Christopher Hitchcock and James
Woodward[57][32]. Extensive work has been done on this topic and there are many different theories of scientific explanation. Not all of them share the same gauges of depth. Still,
this recent work builds on a tradition arising from the work done by Hempel and Nagel and
we can cull from it some instructive points which contribute to augmenting Bunges unique
project.
The conceptual map program generates a graphic rendering of the metascientific
16 It

appears that refutability is the same as falsifiability in Bunges use. See The Myth of
Simplicity (pg.91)

122
criteria and their relations. As the arguments are presented the conceptual map will be
updated accordingly. The various simplicity criteria are represented by the image in the
figure 3.2.

Ontological
Simplicity

Syntactical

Pragmatical

Semantical

Epistemological

Figure 3.2: Simplicity

3.3.1

Systematicity

Goodman and Bunge have argued that systematicity is, for various reasons, a desirable aim for science. Roughly put, systematicity is the connectedness between the basic
terms or ideas of an interpreted formal theory. We can think of systematicity as a gauge of
how comprehensively a system expresses the interrelations between its fundamental terms
or ideas. These are rough characterizations of systematicity because there are different
ways to achieve it.
Goodman opens his 1943 paper On the Simplicity of Ideas by saying that The
motives for seeking economy in the basis of a system are much the same as the motives for
constructing the system itself.([27] pg. 107)
We wish to have systems that adequately express their subject matter. Bunge included a series of diagrams in his discussion of conceptual connectedness, and it may help

123
to start with Bunges discussions about one way to achieve systematicity. However, Bunge
does not investigate theories of concepts. Bunge himself has said that no measure of conceptual connectedness is presently known and this may account for why his diagrams are
left open to interpretation. Possibly, conceptual connectedness is expressed by the interdefinability of terms, or by partial interpretations. However since these are expressible in
formal languages I will save these topics for the discussion of the relation between syntactical and semantical simplicity criteria and systematicity. A further discussion of the definitional structure of terms appears in the subsection on explanatory depth. I can speculate
a little about what might be unique about conceptual connections. On one view, concepts
can be complex wholes with primitive concepts as their proper parts. On this view, the
part-whole relation would account for conceptual connectedness. On another view, concepts are structured just in case they stand in a privileged relationship to other concepts by
way of some inferential disposition. On a third view, that concepts could be related is by
a subject knowing that two concepts have the same referent but different modes of presentation17 . On a linguistic analysis, this appears as the identification of coreferential terms. I
reproduce Bunges series of diagrams in the following.
If, for heuristic purposes, we assume an axiomatic account of theories of the sort
which Bunge appears to embrace in the 1960s, an example of an unsystematic set of mutually independent postulates C1 through C6 would be,

...C1 ..., ...C2 ..., ..., ...C6 ...

17 See

[45] for a detailed discussion of concepts.

(3.12)

124
in which none of the basic predicates C1 through C6 appear in more than one axiom. Let
the symbol stand for a conceptual connection of any preferred flavor. A more organized
system would be,

C1 C2 , C3 C4 , C5 C6 .

(3.13)

A yet more systematic set would be,

C1 C2 , C2 C3 , C3 C4 , C4 C5 , C5 C6 ,

(3.14)

which is equivalent to,

C 1 C2 C3 C4 C5 C6 ,

(3.15)

So one sufficient condition for systematicity is conceptual connectedness. Let the

symbol

depict the sufficient for relation and update the conceptual map in figure 3.3.
A common view is that concepts have a definitional structure. Perhaps then the

definitional power of a term tracks the structure of concepts. This is something which
can be analyzed by focusing on linguistic symbols. In the early 1940s Nelson Goodman
launched an inquiry into the linguistic structures that contribute to systematicity. Goodman
points out that if we can reduce the number of primitives in a system then the system more
comprehensively expresses the relationships between the elements of the subject matter.
But Goodman is also correct to point out that it will not be satisfactory to seek theoretical
simplicity by merely reducing basic predicates. A rough version of Goodmans argument is
that we could get a spurious gain in this sort of syntactical simplicity in (3.11) by defining

125

Ontological
Simplicity

Syntactical

Pragmatical

Semantical

Epistemological

Conceptual
Connectedness

Systematicity

Figure 3.3: Conceptual Connectedness

C1 &C3 as one predicate D1 . This is the sort of argument that we might expect from a
nominalist, but even Goodman did not think that we could define terms in any old way
we like. The same point can also be made in a different way. Bunge says that the gain in
systematicity from (3.10) to (3.13) has nothing to do with the number of basic postulates.
All of the sets have the same basic postulates. Conceptual connectedness is necessary
for having a system and merely reducing the number of extralogical bases of a system is
sufficient systematicity. Conceptual connectedness is sufficient for systematicity and and
so is syntactical simplicity. However, neither is necessary for systematicity.

126

Ontological
Simplicity

Syntactical

Pragmatical

Semantical

Epistemological

Conceptual
Connectedness

Systematicity

Figure 3.4: Systematicity

Although syntactical simplicity is sufficient for systematicity, what we wish to know
is whether or not the extralogical bases really are conceptually basic. Because this is problematic, it might be tempting to say that since conceptual connectedness is sufficient for
systematicity, we can avoid questions about syntactical simplicity altogether. This is not
so easily put into practice. As stated in the previous section, no measure of conceptual
connectedness is presently known. Also in the early stages of the development of a theory
it may not be clear what the fundamental concepts are, and we also have reasons to seek
the exactness of logical or mathematical statements. So pragmatic constraints may make

127
syntactical simplicity difficult to dispose of in actual practice.
The kind of simplicity that is relevant to systematicity does not depend merely upon
the simplicity of the signs used to denote properties or classes of objects. The earlier discussion of syntactical simplicity showed that equally simple signs may denote things which
are structured differently even if they pick out the same number of individuals. Scientific
theories do not consist merely of formal or mathematical expressions. The signs of the formal theory are given an interpretation. Goodman suggests that we can think about one type
of simplicity that is relevant to systematicity by thinking about how the signs of a formal
system have their denotations specified by the theorys physical interpretation. Goodman
suggests that we might think of the defining power of an idea, not the syntactical structure
of a theory, as the basis upon which to gauge the kind of simplicity that is relevant to systematicity. Since predicates might hide the structure of concepts, we can see why authors
like Goodman and Bunge, with their views about semantical simplicity, thought that the
defining power of terms would be relevant to systematicity. Goodman says that one natural
approach might be to try to gauge simplicity in the following way:
If a given idea of basis A were definable solely in terms of B and logic,
while B were not definable solely in terms of A and logic, it would be clear
that B is the more powerful idea or basis, A the weaker or simpler.([27] pg.110)

Goodman says that this approach is subject to two objections. One objection is
that the approach will not work for cases where neither A nor B are definable in terms of
one another. The other objection is that this approach gives counter intuitive results. For
example, from the system K consisting of A and B we can get C, D, and E,

128

C =df (A B), (A B)
(A B)
D =df (A B),

E =df (A B), (A B).

Then we can create system K 0 consisting of C, D, and E. The counterintuitive result is that
C, D, and E are definable in terms of A and B but we cannot define A and B in terms of
K 0 . This makes K 0 weaker than K. Since K 0 is more complex than K, weakness is not the
same as simplicity in the sense relevant to systematicity. What is relevant to systematicity
is not weakness but economy.
Goodman points out that in response to the first objection we might try to establish
the rule,
if a basis A is definable solely from logic and some basis that is in a specified sense of the same kind as B, while B is not definable solely from logic
and some basis that is in that sense of the same kind as A, then A is weaker
than B ([27] pg.110).
But Goodman explains that this solution would decide what are kinds independently of defining power, and if we can do this then we do not require this sort of approach
to measuring simplicity. After all, if we had some independent way of deciding the questions of ontology, then that would be sufficient for deciding which terms of an interpreted
formal theory are basic. Syntactical simplicity is sufficient, but not necessary for systematicity. If we had an independent way to settle the questions of ontology or of the structure
of concepts, we would not require syntactical simplicity as a guide to the early stages of
theory construction.

129
3.3.2

Linguistic Exactness

Goodmans discussion illuminates an interesting, and perhaps problematic feature

of simplicity in theory construction. When we judge simplicity by comparing the defining
power of basic postulates, weakness is inversely proportional to the kind of simplicity that
we seek. I will follow Bunge and call one of the desiderata of science linguistic exactness.
Linguistic exactness is inversely proportional to syntactical simplicity. As we accumulate
more and more data, we could always introduce ad hoc data-specific terms to our theories.
But we seek economy rather than mere fit with the data. The addition of terms to the postulate basis may involve semantical complexities as well. Goodman and Bunge sought to
find what places a constraint on syntactical complexities by analyzing the defining power
of terms. Semantical simplicity is shot through with its own problems, and perhaps semantics is not properly analyzed by way of the defining power of terms. What we know
is that linguistic exactness is inversely proportional to syntactical simplicity. We do not
yet know how, and in how many ways syntactical complexities are constrained. The conceptual map can be updated with the symbol

depicting the inverse proportionality

relation between syntactical simplicity and linguistic exactness. Let the symbol

de-

pict the would be sufficient for relation, and this is added between ontological simplicity
and syntactical simplicity since Goodman has reminded us that if we could settle matters
of ontology then this would settle independently questions of syntactical simplicity.
Goodman published his first proposal about how to gauge the relative complexity
of systems in 1943. Over the next decade Goodman published several different proposals.
Some of them involve a set-theoretical measure, others a measure of the predicate length

130

Ontological
Simplicity

Syntactical

Pragmatical

Semantical

Epistemological

Linguistic Exactness

Conceptual
Connectedness

Systematicity

Figure 3.5: Linguistic Exactness

and others involve an axiomatic measurement. Each approach, if successful, would be
sufficient for gauging syntactical simplicity in the way relevant to systematicity and since
there are many approaches, no one is necessary.
The diagram should help to illustrate one interesting puzzle which scientists are
faced with. Both linguistic exactness and systematicity are desirable features of scientific
theories, only there is a trade-off between them when systematicity is achieved by way of
syntactical simplicity. For this reason Goodman is correct to say that, it is quite wrong
to think of the search for economy as a sort of game, inspired by an abnormal love of

131
superficial neatness. ([27] pg.107)
This point, yet again, underscores the importance of rejecting half-baked resistance
to a philosophy of simplicity on the grounds that the role of simplicity in science is reducible only to aesthetic values. Pragmatic constraints clearly play a role in theory construction. However, it would be quite wrong to think syntactical simplicity criteria are
themselves pragmatic criteria. This would be to conflate the objects specified by the criteria with the justification for defining the criteria.

3.3.3

Testability and Accuracy

Both systematicity and linguistic exactness are desirable features of scientific theories. Systematicity is desirable for the reason which has already been stated: we wish to
have scientific theories which comprehensively express the interrelations comprising their
subject matter. Systematicity is also an aim of scientific theory construction because it is
related to testability and accuracy. Mario Bunge says that,
A dough of vague assumptions all standing on the same logical level, without strong logical relations of deducibility occurring in its body, cannot be
tested the way genuine theories are: since all of the propositions of the pseudotheory are loosely related to one another, every one of them will face separately the trials of logic and/or experience. How could we test the axioms of a
factual theory if we cannot spot their logical consequences? A chaotic mass of
conjectures lacking logical organization as is the case with psychoanalysis
cannot be subjected to the test of experience as a whole: experience may at
most confirm some of the loosely related conjectures of the pseudotheory, but
no evidence will ever conclusively refute the whole set of vaguely stated ad
hoc hypotheses-especially if they are mutually shielding. And a theory which
stands no matter what experience may say, is not an empirical theory.([10]
pg.125)
Recall an example in the first chapter where an insane neighbor cannot explain
broken glass and a missing sound system, but who thinks that the footprints between the

132
broken glass and the place where the electronics once were are confirmation for his alien
invasion hypothesis. This pseudotheory fails in two ways. It is not systematic in that it
does not comprehensively express the relationships between the data. It also fails because
it is not falsifiable. Conspiracy theories are like this also. They often consist of vague or
mutually shielding hypotheses so that they may be taken to explain any data that the theorist
wishes them to explain. Systematicity and linguistic exactness are necessary for testability.
Now, I suppose that mutually shielding hypotheses do have a kind of relatedness. However,
we will not allow the theories of the fortuneteller or the conspiracy theorist to be scientific.
Perhaps, then falsifiability provides a necessary constraint on conceptual connectedness.
Let the symbol

depict the necessary for relation and update the conceptual map with

these relations.
As it turns out, the set of relations depicted is not yet precise enough to capture the
ways in which the desiderata of science are related. Testability has several components.
Theories must be scrutable, they must have explanatory power, they must fit the data, and
they must make predictions.
To say that scientific theories must be scrutable means that they must be able to be
investigated by the public methods of science. This does not mean that all of the terms of
a scientific theory must be observable. Rarely is this the case. All it means is that exact
connections must be established between the terms of the theory and observable predicates.
Bunge says that for this reason terms like elan vital are not suitable for scientific theories.
([10] pg.126) What Bunge has shown is that systematicity is necessary for theories to be
scrutable. Additionally, srcrutability is, in many cases, inversely proportional to syntactical

133

Ontological
Simplicity

Syntactical

Pragmatical

Semantical

Epistemological

Linguistic Exactness

Testability
Falsifiability

Conceptual
Connectedness

Systematicity

Figure 3.6: Testability

simplicity. Bunge says that,
A theory containing a large number of scrutable predicates will be preferable to another theory containing fewer predicates but all or part of them inscrutable, if only because the former theory will be testable, unlike the latter.
The methodological status of the predicate basis is far more important than its
logical structure and number. Thus, electrically charged is both syntactically
and semantically more complex than providential, yet it is scrutable and may
consequently occur in scientific theory, whereas the latter cannot ([10] pg.126).

The conceptual map is now updated to show that the relevant feature of testability
which systematicity and linguistic exactness are necessary for is scrutability and to show
that scrutability can be, under certain circumstances, inversely proportional to syntactical

134
simplicity, that is when the predicate basis is complicated by inscrutable predicates. Let the
symbol

depict what can be an inversely proportional relation.

Ontological
Simplicity

Syntactical

Pragmatical

Semantical

Epistemological

Linguistic Exactness

Scrutability

Testability
Falsifiability

Conceptual
Connectedness

Systematicity

Figure 3.7: Scrutability

3.3.4

Accuracy

When we think about how well a theory fits data and how well it makes predictions
we must think about accuracy, which is another desideratum of science. This is because
accuracy is intimately related with fit and predictive power. Recall the first chapter discussion of Sherlock Holmes. It was convenient that Holmes omitted certain specifics about

135
the properties of the keys and coins which were hypothesized to be the causes of some
of the watch scratches. Under similar circumstances such specifics might have turned out
to be relevant to the truth or falsity of the hypothesis, and such syntactical complexities
would also have increased the chances that some of his hypotheses about the history of the
watch would turn out to be false. In other words, Holmes could have given more hypotheses or more specific hypotheses but he didnt and this worked out in his favor. The more
specific hypotheses are, the more likely they are to be falsified. This shows that accuracy
demands linguistic exactness. It also shows that there is an inverse proportion between fit
and predictive power and accuracy. Bunge says that,
The more exact a statement is, the easier it will be to dispose of it; vagueness and ambiguity the secret of the success of fortune-tellers and politicians are the best protections against refutation. Now, accuracy demands
complexity, both formal and semantic: suffice to compare the simplicity of
presystematic, ordinary, discourse with the complexity of scientific discourse;
compare small with of the order of one atomic diameter, and x > a with
x = a.([10] pg.126)

We wish to have scientific theories which are, in principle, refutable. In this respect
syntactical simplicity is can be both favorable and unfavorable to science. There is some
trade-off between syntactical simplicity and testability. How much of a trade-off is allowed
depends upon how the other metascientific criteria are satisfied.
The same logical relation holds between accuracy and the relevant features of testability: fit and predictive power. In this case the only difference between these two features
of testability is one of temporal order; either the hypothesis is given after the data are collected or more data are collected after the hypothesis is given. From one perspective we say
that a hypothesis fits the data given certain accuracy constraints or we say that it does not fit

136
the data for some accuracy constraints. From the other perspective we might call a datum
confirming if it falls within the accuracy constraints for a specific hypothesis and disconfirming if it falls outside of the accuracy constraints for that hypothesis. For this reason, fit
and predictive power are combined in the conceptual map. Additionally, linguistic exactness is necessary but not sufficient for accuracy. Other factors which contribute to accuracy
have to do with what sort of experimental equipment is used, the skill of the experimenter,
and the experimental controls involved in a particular experiment. The conceptual map is
now updated to show the inverse proportionality relation between fit/predictive power and
accuracy, and the necessary for relation between linguistic exactness and accuracy.

3.3.5

Depth

Despite the importance of elucidating explanatory depth, the discourse still occupies a very focused corner of philosophy. There is at present no Stanford Encyclopedia
of Philosophy entry on depth and the arguments involving explanatory depth are strewn
throughout the literature on scientific explanation and confirmation. Bunge claims that scientists will not make simplifications at the expense of explanatory depth, so it is important
to know a bit more about this explanatory virtue.
Depth comes in degrees. Some explanations are deeper than others. Brad Westlake
([64]) credits Hempel with having noticed that explanations may be ranked in order of the
range of phenomena that can be explained by the laws involved in them. In some kinds
of cases a law which figures in more kinds of explanations than another which explains
some of the same phenomena will be the deeper. By this standard Newtons law of gravita-

137

Ontological
Simplicity

Syntactical

Pragmatical

Accuracy

Semantical

Epistemological

Linguistic Exactness

Predictive
Scrutability

Power/Fit

Testability
Falsifiability

Conceptual
Connectedness

Systematicity

Figure 3.8: Accuracy

tion is deeper than Keplers laws of motion because Newtonian mechanics applies to both
terrestrial and celestial motions. Newtonian mechanics is also thought to be deeper than
Galileos Laws of motion. In one way, this might appear to conform to the same standard
because although Galileo was interested in unifying the terrestrial and celestial laws of motion, Newtons mechanics also explains tidal action. However, the reason that Newtonian
mechanics explains tidal action has to do with Newtons development of calculus and his
treatment of the whole mass of the Earth and the whole mass of the moon as collections
of point masses. In this case, the increase in the generality of Newtonian mechanics over

138
Galilean mechanics has to do with an ontological reduction of wholes to their constituent
parts and their relations. So, generality of some sort is a symptom of depth. However,
generality can be increased in a number of different ways and what it is that contributes to
depth, and thereby gives some increase in generality of some kind depends upon several
things including specific notions of scientific explanation and reduction. I wish here only
to distinguish several kinds of generality so that some of the symptoms of depth can be
identified.
Generality is sometimes referred to as scope. A law statement has its scope defined
by the number of kinds of properties or entities to which it can be applied. There are several
ways to achieve generality in scientific explanations.
Level Generality: a statement expressing a law at the genus level may be expected
to have a greater generality than a statement expressing a law at the species level. Laws
about the behaviors of the genus of critters Solenopsis apply to more kinds of entities than
do statements expressing laws about Southern fire ant behaviors only.
Reductive Generality: reductive generality may be achieved by explanations which
involve properties had in common by many individuals. Explanations of this kind may
also take a form where a part had by particular individuals appears in the postulate basis
of a theory because it explains the behavior of a whole class of individuals. This is due
to the fact that parts have attributes and structural properties. For example, law statements
involving the term chromosome makes appeal to a part of particular living organisms in
order to explain things about the whole organism. A class of entities have these parts and so
this sort of law applies to all entities of this class. Such laws are more general than Mendels

139
Laws involving appeal to gametes which Mendel knew to give only species-specific laws.
Unity of Science Generality: Generality of this kind might be analyzable in terms
of level generality or reductive generality. However, it should be discussed as a distinct
type of generality for three reasons. 1. The unity of science may actually be the primary
goal for scientists when they construct some theories. 2. The kind of generality which is
relevant to explanatory depth is often relative to a specific account of scientific explanation.
3. Accounts of reduction in science are various, and as we shall see in the next chapter, the
word reduced has been used in different ways to describe the relationship between two
sets of laws, theories, or concepts. The suggestion was made in the previous chapter that
LeVerrier and Newcomb were concerned with the unity of science in their investigations
of the range of application for Newtons Laws of motion. Newtons laws are expressible
in terms of momentum which is had by both terrestrial and celestial bodies indicating that
Newtons laws are deeper than the Keplerian laws of motion.
Holistic Generality: If there are laws relating multiply realizable properties then
these laws would also achieve a kind of generality. If the color red is thought to be a property of the mind-independent world then it is weakly emergent because it would be realized
in many different ways: by emission, absorption, reflection, and transmission. If there are
properties that cannot be token-token identified with their compositional properties, then
these would defy any kind of reduction. If there are laws relating such properties, then
nature would be disunified.
It was stated above that ontological simplicity of some kind may be sufficient for
syntactical simplicity. Consider how this might work. It is possible to satisfy several of

140
the metascientific criteria at once by doing a bit of ontology when we construct theories.
Positing, lets say, a causal mechanism will solve some of the problems of systematization
because it would establish necessary connections between the terms of the theory. It would
also contribute to generality if the mechanism relates several different kinds of properties
or if it may be used in many different explanations. Additionally, if the causal mechanism
posited by a law represents the way that the world really is (if it is a true posit) then it is
natural to expect that the law will exemplify other desirable features: it will have serendipic
power and it will support directional counter-factuals. Serendipic power is Bunges term
for the power of a theory to make novel predictions and what distinguishes a law from an
accidentally true generalization is that a law supports counterfactuals. A causal mechanism
posited by a theory will support directional counter-factuals linguistically even if it does
not represent a way that the world is, but unless it is true, it cannot be expected to give true
predictions.
As a desideratum of science, explanatory depth is not conceptually bound to causation. There are other kinds of explanations. Identities, for example, may be explanatory
and they may contribute to depth by making possible the formulation of laws with greater
scope. Also, the careful construction of operational definitions can give results which exemplify generality.
It would be extremely difficult to satisfy the conditions of explanatory depth without making ontological commitments. Scientific theories make substantive claims about
the way that the world is. Scientists rarely operationalize the theoretical terms of their theories, perhaps for pragmatic reasons or perhaps because they take themselves to be giving

141
a correct description of reality. The question is whether or not it is, in principle, possible to translate all of the primitive postulates of a scientific theory into phenomenological
statements. Hempels famous 1958 article The Theoreticians Dilemma presents arguments which can be employed to show why it is difficult to replace a scientific theory with
another theory using only phenomenological terms and yet preserve the features of explanatory depth, no matter how the distinction between phenomenological and theoretical
is drawn.
Hempels dilemma is:
If the terms and principles of a theory serve their purpose they are unnecessary, and if they dont serve their purpose they are surely unnecessary. But
given any theory, its terms and principles either serve their purpose or they
dont. Hence, the terms and principles of any theory are unnecessary. ([29]
pg.49-50)

It is the first sentence that Hempel has pragmatic reasons to reject. So long as we
desire depth and predictive power, and so long as we do not wish to systematize by admitting absurdities to the postulate basis, it is necessary that our theories make ontological
commitments.
Hempel illustrates this point by explaining that the empirical generalization; wood
floats on water; iron sinks is only applicable to a limited range of situations. Some kinds
of wood sink and iron formed into a battleship will float. An operational definition can
be constructed which captures the more general facts. The specific gravity of a thing can
be defined as the quotient of its weight and its volume, and then a new generalization can
be asserted: A solid body floats on a liquid if its specific gravity is less than that of the
liquid ([29] pg.43-44). Including this generalization in an explanation would improve the

142
indicator of depth over the previous generalization.
The first part of Hempels argument involves pointing out how theoretical terms
get mixed into general law statements along with observable terms. We can operationalize
specific gravity by defining it as the quotient of weight and volume, which are observable. Nevertheless, when we compare the floatability of two objects we will compare
their specific gravities. This is because it is notationally simpler (easier) to express things
this way in this sort of case. What should be recognized is that the theoretical detour (as
Hempel calls it) is conducive to giving systematic connections between observable terms
and general laws. Of course the question is, can we operationalize all of the terms of any
general law statement? Even Carnap begins with the very guarded claim that,
In the case of many words, specifically in the case of the overwhelming
majority of scientific words, it is possible to specify their meaning by reduction
to other words. ([45] pg.11)

Hempel argues that we have pragmatic reasons not to operationalize all of the terms
of our theories. Some of the terms of a theory certainly are definable in terms of the others.
However, we must take some terms as primitive. The question then becomes, can we take
all and only observational terms as primitives? A problem then arises due to the fact that
a definition must give necessary and sufficient conditions for the application of a term.
If we attempt to give operational definitions for the primitive terms of a theory by mere
conditional statements, a crippling objection arises. Consider a basic logical behaviorist
proposal. Suppose that we wished to define Sues anger by way of a standard conditional
statement defining the anger A of the subject x in terms of the observable initial conditions
C and the expected behavioral outcome conditions E.

143

Ax (Cx Ex).
The obvious problem with this definition is that the statement is true whenever the
antecedent is false. Sue would then be angry in every case where the initial conditions
are not met. The only way to preserve this sort of definition and avoid this problem is
to bind Cx to Ex nomologically by positing a causal mechanism which would make the
conditional express causal necessity. This brings us right back to Aristotles original view
of the Physics. A man comes to be musical from having been unmusical when caused to
become so.
Carnap suggested an alternative which would avoid this difficulty by giving partial
specification of meaning or by so-called reduction sentences. Carnaps suggestion would
replace the above definition with a partial specifier of meaning for A with a definition of
the form,

Cx (Ax Ex).
This says that under the test conditions C Sue is angry A if and only if she exhibits a
behavior of the kind E. Of course this is a partial specification of meaning, and to get a full
specification of meaning for the general term anger we would have to generate a series of
reduction sentences for all of the possible subjects and all of the conditions for which this
general term might be applied. This is precisely the situation which undermines predictive
force. It does so because the program tends towards both syntactical and semantical complexities which trade-off with predictive power. First, the kind of syntactical simplicities

144
that fit the existing data very precisely will tend to run against predictive power unless accuracy constraints are loosened. Second, the goal to generate partial definitions for all of
the terms of a theory will run into a dilemma. On the one hand, if partial definitions are
supposed to be analytic truths, then they do not have empirical content. On the other hand,
if the wild disjunction of partial definitions for a term are supposed to be related in some
way then it is because they get at some underlying law of nature.
We can see how wild the disjunction of reduction sentences for a general term would
be if we switch from psychology back to the examples from the first chapter involving
physics. LeVerrier and Newcomb were working with the general term mass which figures
in both terrestrial and celestial predictive statements. The reduction sentences for objects
that are small enough to shake around could be given in terms of the felt resistance to the
change in motion of the object. However, we would have to define the mass of Jupiter in
some other way. In this case it might be counter intuitive to think that mass really means
wildly different things depending on context despite the fact that one and the same general
term appears in the predictions of billiard ball movements and of planetary orbits. Finally,
it is not clear what sorts of experimental controls would have to be established in order to
give empirical interpretations for the theoretical term mass when applied to things like
electrons or planets which cannot easily be manipulated. Even Carnap was skeptical that
experimental conditions could be established to unambiguously specify the meaning of
the mass of the electron. So the impending project of giving partial definitions for all of
the uses of the term mass from electrons to bread boxes to satellites to black holes might
give us pragmatic reasons to just take mass as an un-operationalized primitive.

145
There is one final proposal to consider for the problem of giving necessary and
sufficient conditions for the application of a term. It might be suggested that we attempt
to give partial definitions in purely empirical terms by appealing to the entire body of
scientific empirical research. However, there are two ways to think of this. One way would
be to include all of the empirical research up to now on the assumption that science is more
or less on the right track and that future scientific theories will include the terms science
now has. The other approach would be to imagine a completed science and speculate about
what the terms of that discourse will be. The first proposal is problematic because one thing
that we know from the history of science is that revolutions happen, and when scientific
revolutions happen we rethink our interpretations of empirical research from the past. After
all, the benefit of hindsight shows us that the arguments given by LeVerrier and Newcomb
were some of the key arguments which helped to set the stage for a scientific revolution
which would tend to make us consider the project of giving operational definitions for
absolute time and for force to be inappropriate. The second proposal is problematic,
again, because revolutions happen and we cannot imagine what a future science will involve
as far as empirical programs are concerned. Perhaps a future science will find a middle
ground between introspective psychology and the experimentalist traditions by including
mental predicates as observation terms. It is not clear how to justify settling on one of these
two methodological presuppositions.
If we posit causal mechanisms in our scientific theories then the conditional statements of those theories will express causal necessity, and we would solve the formal problems involved in balancing systematicity with depth. Because conditional statements allow

146
us to derive other statements, systematicity is achieved by deductive logical connections.
This would serve well the desire for depth and, if the posits were true, for predictive power.
If we posit primitive terms (like mass) with instances that fall under the scope of
a universal quantifier ranging over a class of entities then we have also given necessary and
sufficient conditions for the application of terms which can be the values of the variables
bound by this quantifier. This sort of posit too serves well the desire for predictive power
and for depth, and it contributes to systematicity by establishing logical connections.
Some things change and some things stay the same, and we could have learned
from Hume that we cannot engage the project of explaining such a world without transempirical terms (which suggested to Hume that there is a non-rational foundation for human
reasoning). Hume says that,
If reason determind us, it woud proceed upon that principle, that instances, of which we have had no experience, must resemble those, of which
we have had experience, and that the course of nature continues always uniformly the same. (A Treatise of Human Nature 1.3.6.4)

But it is the uniformity of nature principle which Hume claims that we cannot justify
by experience.
Why from this experience we form any conclusion beyond those past instances, of which we have had experience? If you answer this question in
the same manner as the preceding, your answer gives still occasion to a new
question of the same kind, even in infinitum; which clearly proves, that the
foregoing reasoning had no just foundation. (A Treatise of Human Nature
1.3.6.10)

We also could have learned this just by thinking about Aristotles scientific adaptation of Platos mixed view. In the Posterior Analytics Aristotle says,

147
We suppose ourselves to possess unqualified scientific knowledge of a
thing, as opposed to knowing it in the accidental way in which the sophist
knows, when we think that we know the cause on which the fact depends, as
the cause of that fact and of no other, and, further, that the fact could not be
other than it is.(Book I, Chapter 2)
In so far as we desire our theories to have predictive power and in so far as we desire
the features of explanatory depth, we have pragmatic reasons to posit causal mechanisms
which govern change and to take as primitives general terms (which stay the same). Systematicity is necessary for testability, and in so far as we wish to have scientific theories
rather than conspiracy theories or pseudosciences, we have pragmatic reasons to posit the
sort of primitives that maximize the logical connections of a theory.
Depth is related to testability because testability is related to systematicity. A theory
constructed using the predicate basis suggested by the behaviorist appears doomed either to
be crippled by triviality or to become so relativized that it will not make predictions at all,
let alone novel predictions. To use a turn of phrase popular in contemporary philosophy of
mind debates, disjunctive predicates are not projectible. The logical behaviorists obsession
with scrutability understood in terms of epistemological simplicity undermines explanatory
depth, and this undermines predictive power.
As suggested above, positing mechanisms of mental causation (the very thing that
the behaviorist wished to avoid adding to the science of human behavior) would fix these
problems. Positing mechanisms of mental causation would support non-backtracking counterfactuals (because causation has a direction) and the possibility of generality with respect
to other possible properties of the theory. If the causal mechanism is shared by other theories, then it would contribute to this dimension of generality as well (a unity of science

148
kind of generality). If true then it would be expected to have serendipic power as well. We
could, of course, be mistaken about what counts as confirming or disconfirming evidence
for theories which posit causal mechanisms.
One final note on depth is that the symptoms of depth are relative to specific accounts of scientific explanation. For example, Woodward and Hitchcock defend what is
known as the interventionist account of explanation. On this view, there are several indicators of depth; among them is accuracy, which I have kept as a distinct desideratum
of science. Brad Westlake has also contributed to the project to taxonomize some of the
leading accounts of explanation and to show the implications for ontology when depth is
gauged in each way [64]. Westlake defends an abstractionist account of explanation and
this account differs from the Woodward and Hitchcock account mainly due to a difference
in views about which direction reductive explanations ought to go in.
Since more work needs to be done on explanatory depth, an exhaustive analysis
cannot yet be presented. However, with this minimal understanding of how to gauge depth,
we can see some of the ways to construct theories that exemplify some of the symptoms
of depth. Operational definitions might turn the trick in some cases, although there are
many problems with operational definitions. However, when it comes to the fundamental
postulates of a theory, we have pragmatic reasons to take Hempels the theoretical detour.
The conceptual map is now updated to show that a necessary condition constraining judgments of syntactical, semantical, epistemological and pragmatical simplicity in
theory construction is that depth not be sacrificed. Obviously, since the positing of causal
mechanisms would be one way to satisfy the conditions of explanatory depth, ontological

149
proliferation is not constrained by the criteria of explanatory depth.

Ontological
Simplicity

Syntactical

Pragmatical

Accuracy

Semantical

Epistemological

Linguistic Exactness
Depth
Predictive
Scrutability

Power/Fit

Testability
Falsifiability

Conceptual
Connectedness

Systematicity

Figure 3.9: Depth

3.3.6

Representativeness

One natural way to think of representativeness is as a criterion which specifies that

theoretical postulates aim to represent the way that the world really is.
More than a half century ago logical positivism was crumbling. Many philosophers
and scientists were drawn to the scent of the decaying positivist steed and the collapse
of the distinction between theoretical languages and observation languages as well as the

150
collapse of the analytic/synthetic distinction and the failures of attempts to give necessary
and sufficient conditions for meaning on the verificationist theory. In particular, the positivists hostility to metaphysics faced serious criticism. Bunge was certainly among those
circling the positivist carcass. My guess is that it is closest to Bunges intention to give the
straightforward interpretation to representativeness: that a theory represents the way that
the world really is, rather than taking the basic postulates of a theory to refer to sensations.
Nancy Cartwright says that,
Philosophers distinguish phenomenological from theoretical laws. Phenomenological laws are about appearances; theoretical ones are about the reality behind the appearances. The distinction is rooted in epistemology. Phenomenological laws are about things which we can at least in principle observe
directly, whereas theoretical laws can be known only by indirect inference.
Normally for philosophers phenomenological and theoretical mark the distinction between the observable and the unobservable.
Physicists also use the terms theoretical and phenomenological. But their
usage makes a different distinction. Physicists contrast phenomenological with
fundamental. For example, Pergamon Presss Encyclopaedic Dictionary of Physics
says, A phenomenological theory relates observed phenomena by postulating certain equations but does not enquire too deeply into their fundamental
significance.([13] pg.1)

Representative theories posit theoretical entities, and one desideratum of science is

to get these posits to represent reality. This section will explain some of the interrelations
between representativeness and systematicity, depth, testability, and simplicity.
Hempel says that,
Scientific systematization is ultimately aimed at establishing explanatory
and predictive order among the bewilderingly complex data of our experience, the phenomena that can be directly observed by us. It is a remarkable
fact, therefore, that the greatest advances in scientific systematization have not
been accomplished by means of laws referring explicitly to observables, i.e.,
to things and events which are ascertainable by direct observation, but rather
by means of laws that speak of various hypothetical, or theoretical, entities.

151
i.e., presumptive objects, events, and attributes which cannot be perceived or
otherwise directly observed by us.([29] pg.41)

Cartwright distinguishes between the philosophers uses of the words phenomenological and theoretical, but perhaps philosophers and scientists are talking about the same
thing, which is that theoretical statements involve the theoretical detour. Bunge has stated
that epistemological simplicity competes with logical simplicity and with depth. He has
also said that phenomenalist languages come at the cost of syntactical simplicity and insight. Perhaps now it is clear why he said this. We could have learned this lesson from
Berkeley, who attempted to give a radical empiricist interpretation of how Gods simplicity is reflected in the created world of sensations [6]. Even Berkeley helps himself to the
most exotic theoretical detour, positing archetypes, as he struggles to systematize an unwieldy view. On the most charitable of interpretations, Berkeleys suggestion that we operationalize the fundamental terms of science leads to considerable syntactical complexities.
Epistemological simplicity leads to syntactical and semantical complexities and it does not
reconcile with the features of explanatory depth, thereby undermining systematicity and
predictive power.
Now an interesting set of questions arises about to how to update the conceptual
map. We have seen that either the positing of fundamental causal mechanisms or the
positing of general theoretical terms would be sufficient for satisfying the conditions of
explanatory depth which is necessary for predictive power. It has also been shown how the
theoretical detour is sufficient for the formal requirements of systematicity. It is tempting,
then, to say that ontological proliferation favors representativness. Or, in other words, that

152
there might be an inverse relation between ontological simplicity and representativness.
Yet, recall that the problems (1) and (2) which arise when we try to organize sense experience in an ontologically responsible way do not arise when we just stipulate the kinds of
things to be counted. Perhaps we could stipulate simplicity criteria with respect to causal
mechanisms on the pragmatic grounds just argued for. Positing one causal mechanism
might seem to systematize a theory better than several independent causal mechanisms.
Having just one kind of thing is the clearest case of ontological simplicity, and this is what
systematizes Newtonian mechanics. Yet it is not clear that positing just one kind of causal
mechanism is the best way to achieve systematicity because causal mechanisms could be
dependent on one another.
Suppose, for example, that physicists posited one kind of causal mechanism and
psychologists posited another kind of causal mechanism which happens to be restricted by
the range of what is possible given the physicists kind of causation. After all, a complete
theory of the mental would involve certain physical limits because consciousness can be
interrupted by a sharp blow to the jaw, for example. In this kind of situation, psychology would enjoy a certain unification with physics. It could turn out that the theories of
psychology would have to be codified in modal systems like S4 or S5 which allow for
nested modalities of different sorts. If we have reasons to employ S4 or S5 modal logics
it is because they might be thought to adequately express the relationships of theories of
this sort and because they establish greater systematic connections than do modal systems
which do not allow for nested modalities. However, nested modalities may not be the kind
of relations that contribute to depth and it would be to endorse an outdated view to sug-

153
gest that only deductive relations contribute to systematicity. Ontological proliferation may
contribute to representativeness, but it is not clear what to say about the relation of causal
mechanisms to representativeness. We should accept ontological proliferation only to the
extent that such a postulate basis maximizes the other desiderata of science. After all, we
should expect some ontological complexities in science, since its very grammar is committed to greater complexities than Parmenides and Heraclitus would have allowed, but the
world may have a specific number of kinds of entities, kinds of properties and kinds of
relations. We do not know what the upper bound on the complexity of the world is.
We can now update the conceptual map to show that representativeness is sufficient
for depth. Representativeness is sufficient for systematicity and it stands in an inversely
proportional relation with ontological simplicity in the most basic sense that we could easily have learned from Aristotle who followed Plato down the path of the mixed-view. We
can also add relations showing that epistemological simplicity is inversely proportional to
syntactical and semantical simplicity and inversely proportional to depth (assuming that
perhaps losses in depth might come in degrees if we attempt partial specifications of meaning).

3.3.7

The Unity of the Sciences

Some version of the unity of science has been taken by many philosophers and
scientists to be a goal of science. Bunge has said that Simplificaton is sufficient for unification, but it is not necessary to this end and should be minimized in view of the important
goals of accuracy. ([12] pg.89)

154

Ontological
Simplicity

Syntactical

Pragmatical

Accuracy

Semantical

Epistemological

Linguistic Exactness
Depth
Predictive
Scrutability

Power/Fit

Testability
Falsifiability
Representativness

Conceptual
Connectedness

Systematicity

Figure 3.10: Representativness

In the same article Bunge goes on to say of World View Compatibility that,
[it is desired] that the theory be consistent with the prevailing world outlook
- a desideratum which, if exaggerated, leads to the banishment of originality.
Simplicity is as inconsistent with world-view compatibility as it was found to
be with external consistency.([12] pg.105-106)
By external consistency Bunge refers to whatever the features of a scientific theory are that make it compatible with other theories. Scientists had very good reasons to be
suspicious of Darwins evolutionary theory for many years. The reason is that it suggested
that the Earth was millions of years old. This result was incompatible with the results in
physics and astronomy. The mass of the sun was known and it had been used in the sorts

155
of calculations performed by astronomers like those of LeVerrier and Newcomb for many
years. The problem was that the only known process by which matter could be converted
into heat was by chemical processes. It had been calculated that if the sun converted 100%
of its chemical energy into heat it would burn out in a scant few hundred thousand years18 .
A theory of atomic energy was needed to reconcile Darwinian theory with physics.
The problem with discussing the unity of science is that it is subject to a great deal
of vagueness. Again, all that can be accomplished here is to gesture at some of the features
of this goal of science and their relations to the other desiderata of science. Jordi Cat opens
the Stanford Encyclopedia of Philosophy article on the unity of science saying that,
The topic of the unity of science includes the following questions: Is there
one privileged, most basic kind of stuff, and if not, how are the different kinds
of material in the universe related? Can the various physical sciences (physics,
astronomy, chemistry, biology) be unified into a single overarching theory, and
can theories within a single science (e.g., general relativity and quantum theory
in physics) be unified? Does the unification of these parts of science involve
only matters of fact or are matters of value involved as well? Moreover, what
kinds of unity in the sciences are there: is unification a relation between concepts or terms (i.e., a matter of semantics), or about theories they make up?
And is the relation one of reduction, translation, explanation, or logical inference? ([14])

It may already be clear how to extend the discussion of the desiderata of scientific
theory construction and testing to the discussions about the unity of science. This section
has covered some of the goals and difficulties involved in analyzing conceptual connectedness, of giving reduction sentences, of satisfying the features of scientific explanation and
of systematizing under deductive relations. It would be natural to extend these discussions
18 Bunge

makes brief mention of the cool reception given to Darwinian theory in France in his
1961 article. [10]

156
further to inter-theoretic relations. It may also become clear how much distance we can get
out of the examples raised in the first chapter. The discussion of Platos mixed view sets
the stage for thinking about the problems of induction, for thinking about the grammar of
science, and for seeing the basic reasons articulated by Hempel for the theoretical detour.
This discussion also sets the stage for thinking about the historical facts about the heritage
of proclivities for unity in Western thought. Cat says that,
The general questions should be carefully distinguished from any of the
different specific theses addressing them and should be noted as the linking
thread of a time-honored philosophical debate. The questions about unity belong to a tradition of thought that can be traced back to pre-Socratic Greek
cosmology, in particular to the preoccupation with the question of the one
and the many. In what senses are the world and, thereby, our knowledge of
it, one? A number of representations of the world in terms of a few simple
constituents considered fundamental emerged: Parmenides static substance,
Heraclitus flux of becoming, Empedocles four elements, Democritus atoms,
or Pythagoras numbers, Platos forms, and Aristotles categories.([14])
Cat goes on to say that,
With the advent and expansion of Christian monotheism, the organization
of knowledge reflected the idea of a world governed by the laws dictated by
God, creator and legislator.([14])
Of course the Christian revolution contributed significantly to the thought of the key
thinkers of the scientific revolution. Cat says that,
The emergence of a distinctive tradition of scientific thought addressed the
question of unity through sciences designation of a privileged method, set of
concepts and language. In the late 16th century Francis Bacon held that one
unity of the sciences was the result of our organization of discovered material
facts in the form of a pyramid with different levels of generalities; these would
be classified in turn according to disciplines linked to human faculties. In accordance with at least three traditions, the Pythagorean tradition, the Bibles
dictum in the Book of Wisdom and the Italian commercial tradition of bookkeeping, Galileo proclaimed at the turn of the 17th century that the Book of
Nature had been written by God in the language of mathematical symbols and

157
geometrical truths; and that in it the story of Natures laws was told in terms of
a reduced set of objective, quantitative primary qualities: extension, quantity
of matter and motion. In the 17th century, mechanical philosophy and Newtons systematization from basic concepts and first laws of mechanics became
the most promising framework for the unification of natural philosophy. After
the demise of Laplacian molecular physics in the first half of the 19th century,
this role was taken over by ether mechanics and energy physics.([14])

Various projects to classify the parts and the structure of the world or to do the same
for human knowledge have fallen under the term unity of science. It appears that the ancients held metaphysics to have logical priority over epistemology and modern thinkers
invert this relation. The ancients wondered how the world and thereby our knowledge of
the world was one. Modern thinkers may wonder how our knowledge of the world is one
and whether or not that corresponds to the way that the world is. The structure of the dialectic of metaphysics perhaps was mapped onto the dialectic of epistemology in the Early
Modern period. We can begin by asking questions about what structures must be in the
mind in order for there to be some unity of thought and then ask the question of whether
or not any of these correspond to the world to shift the discourse from epistemology to
metaphysics, unless of course, we accept Kants Copernican Revolution in metaphysics
where the study of the cognitive structures just is the study of metaphysics. For this project
I suggest taking the modern approach and asking the epistemological questions first. The
arguments for this approach have been given above. When we investigate scientific explanation (epistemology) we end up with pragmatic reasons to take general terms and causal
mechanisms as fundamental (ontology).
We have already seen that concepts may be structured definitionally, mereologically, or inferentially. In any case, conceptual unification is sufficient for systematicity. So,

158
if there is a unity of scientific knowledge then it contributes to conceptual connectedness.
Different theories may also be connected by simply sharing fundamental postulates. Also,
theories could be unified structurally, as suggested by the example where two theories share
a fundamental causal mechanism, so that the operation of one is governed by the range of
possibilities of a causal mechanism posited by the other. So, representativness would be
sufficient for the unity of science just in case nature turns out to be unified. The next chapters discussion of scientific reduction explains in more detail how it could turn out that
nature is not unified in this sense.
We seek linguistic exactness. This is why it is preferable to codify theories in logical
or mathematical systems. A loss in linguistic exactness leads to a loss in accuracy. We now
know why Bunge said that simplicity would be sufficient for the unity of science, only we
will not accept simplifications which decrease accuracy if the expense involves a loss of
depth or testability. A form of ontological simplicity would be sufficient for the unity of
science, that is if more than one theory share a fundamental (un-operationalized) postulate.
This could well amount to syntactical and semantical simplicity as well. However, it has
been shown above that systematicity and depth benefit from ontological complexity so
ontological simplifications ought to be accepted for the unity of science only on balance
with the other desiderata of science. It has already been shown that some of the desiderata
of science impose necessary constraints on several of the species of simplicity.
Depth may be indicated by generality with respect to other systems. Maximizing
this feature of depth contributes to a version of the unity of science which Bunge calls external consistency. Representativeness is sufficient for any version of the unity of science

159
sketched by Cat. An increase in the versions of the unity of science having to do with the
structure of knowledge would be sufficient for conceptual connectedness. Since the focus
of my investigation begins with the desirable features of scientific theories expressed in
symbols the relevant versions of the unity of science are sufficient for conceptual connectedness.
Before adding these relations to the conceptual map it is worth noting that different
scientists may have seen the unity of science as a desideratum which imposes necessary
conditions on various goals of scientific theory construction. I will not try to show what
necessary conditions might be imposed on the other desiderata of science by a preference
for scientific unity because this might involve a serious investigation into the history of
science. The first chapter has given good reasons to suspect that scientists like LeVerrier
and Newcomb gave a very high degree of priority to scientific unity. When we look at the
diagram generated by the conceptual map, we can imagine that some scientists may have
held the unity of science to be so important that it may extend an array of necessary conditions to the other desiderata of science. However, because depth is sufficient for scientific
unity it cannot be precisely determined if these scientists rejected certain hypotheses because they competed with scientific unity or with some of the features of depth. Perhaps it
would be a contribution to the philosophy of science to investigate further how historically
influential theories and experiments were constructed with considerations given to depth
and to scientific unity.
The conceptual map is now updated to show that depth and representativeness are
both sufficient for the unity of science and the unity of science is sufficient for conceptual

160
connectedness.

Ontological
Simplicity

Syntactical

Pragmatical

Accuracy

Semantical

Epistemological

Unity of Science
Linguistic Exactness
Depth
Predictive
Scrutability

Power/Fit

Testability
Falsifiability
Representativness

Conceptual
Connectedness

Systematicity

Figure 3.11: Unity of Science

3.3.8

Other Relations

The discussion from the first section showed that pragmatical simplicity often is
opposed to syntactical and semantical simplicity. This can occur in situations where judgments of psychological simplicity or notational simplicity lead to complexity of the predicate basis. However we will not accept these simplifications if they render a theory less
testable or deep. Pragmatical simplifications are constrained to some extent by some of the

161
other desiderata of science. It has been argued that there may be cases where pragmatical
simplifications will lead to a decrease in accuracy and this would lead to an increase in predictive power. However, what will not be acceptable is a decrease in scrutability or depth.
It is important to bear in mind that scrutability should not be interpreted in the phenomenalists sense as entailing that epistemological simplicity criteria govern selection of the
predicate basis. The conceptual map can be updated to show the inverse proportionality
relation between pragmatical simplicity and syntactic and semantic simplicity and the fact
that scrutability provides a necessary condition for pragmatical simplicity.
I have followed in the footsteps of Goodman and Bunge by contrasting scientific
theories with pseudoscience and conspiracy theories. We can see that certain types of
complexities of the predicate basis which are designed in an ad hoc manner to promote
the kinds of cohesiveness that will guard against refutation are to be avoided. However,
in science the requirements of depth, predictive power, accuracy and linguistic exactness
restrict logical complications of this sort. Now we can see that syntactical and semantical
simplicity, properly restricted, do contribute to falsifiability. We can add to the conceptual
map the necessary constraints that falsifiability places on these kinds of simplicity.

3.4

Evaluation

Several results can be gathered from the discussions of this chapter. We can revisit some of the issues raised in the previous chapter with a finer grained set of terms and
distinctions. We can now give an analysis to principles of parsimony and to some of the
key roles played by judgments of simplicity in scientific theory construction. We can also

162

Ontological
Simplicity

Syntactical

Pragmatical

Accuracy

Semantical

Epistemological

Unity of Science
Linguistic Exactness
Depth
Predictive
Scrutability

Power/Fit

Testability
Falsifiability
Representativness

Conceptual
Connectedness

Systematicity

Figure 3.12: Other Relations

establish the foundations for criticisms of the contemporary explosion in the literature dedicated to curve-fitting. I will also make a proposal about how to advance future research
programs into the roles of the desiderata of science in theory construction.

3.4.1

Ockhams Razor and Conceptual Analysis

I suggest that we stop using the term Ockhams Razor altogether. Ockhams
Razor has been shown to be subject to vagueness and contemporary uses of the term invite
historical confusions. The philosophical problem with Ockhams Razor is that the ceteris
paribus clause is impossible to satisfy for two different competing scientific hypotheses.

163
In some cases it may appear that a scientists actions would be explained if she employed a principle of parsimony in her methodology. However, a closer look at some of
these cases reveals that a different sort of reasoning may be involved. Although many have
thought that some version of a principle of parsimony served as the final court of appeal
in selecting between competing theories, we can now see that some versions of simplicity
(especially syntactical simplicity) may just be a symptom of the scientists desire to maximize depth, predictive power, or the unity of science. In many cases simplicity judgments
do play a role in theory construction, it is just that these ought to be kept distinct from
principles of parsimony which assert something about the likelihood of a theory to be true.
Some judgments of simplicity may be reduced to judgments of other sorts; however, given
the extreme heterogeneity of simplicity, no overall measure of the simplicity of theories is
possible. After all, simplicity is difficult to gauge, and in cases where a clear gauge is available, we have seen that maximizing one species of simplicity leads either to complexity for
some other species or to an unacceptable loss in one of the other desiderata of science; we
do not, for simplicitys sake, render theories less testable, less deep, or unfalsifiable.
The devastating result of our study of the interrelations between the metascientific
criteria is for the family of principles which involve a ceteris paribus clause. There are
always trade-offs between the metascientific criteria, so other things never are equal. My
objection to Swinburnes example in the first chapter is not a case of meager bickering
about example choice. First, if one is to be a Kantian about the methodological role of an
indispensability principle, then having an example is logically prior to putting the test of
indispensability to it to see how experience conforms (or fails to conform) to a concept.

164
Kant makes it quite clear that the Egyptians groped about with mathematics long before
some Greek decided to let the figure inform him. My best guess is that Swinburne wishes
to employ some version of the indispensability principle in the argument for his multifacetted obsidian razor. If I am right about this, then Swinburne must find an unproblematic
example.
Secondly, the analysis of simplicity shows us that the ceteris paribus clause will be
impossible to satisfy in a non-arbitrary way. It is tempting to suppose that scientists like
LeVerrier investigated unrestrained conceptual space as they dreamt up hypotheses to explain the phenomena they were interested in, and that here some version of the principle of
parsimony helped them to decide which hypotheses deserved further attention. The problems (1) and (2) from the first section show us that this cannot be the correct story to tell
about the role of the imagination in scientific theory construction. We require some principled way to specify gauges of simplicity in the first place. It is unlikely that LeVerriers
methodological principles were selected arbitrarily. Although I do not yet know precisely
what LeVerriers methods were, we may have a partial guide to determining what some of
them were. It is very implausible that LeVerrier could have offered infinitely many competing hypotheses for any of the data that was of interest to him. He took only a very few to
be candidates in the first place and then he ruled out each of these by way of arguments that
were widely accepted in the scientific community. We can learn something about scientific
methodology by investigating the arguments given to rule hypotheses out. But we might
also learn something by the ways that hypotheses were constructed in the first place, even
if they are constructed only to be ruled out a short while later. It seems to me quite wrong

165
to suggest that LeVerrier could have given infinitely many competing hypotheses. It seems
that he could not. To give any more (or many more) than he did probably would have
conflicted with some of his methodological principles. Swinburne appears to think that the
relevant question is, why did LeVerrier give only one hypothesis in the case of Neptune?
The better question to ask is what LeVerriers hypotheses had in common such that he only
ever offered a handful as candidates to explain some data. After all, two ad hod hypotheses
could be turned into one by conjunction unless this is constrained by a principle of syntactical simplicity as well. In such a case, the hypothesis with conjunctive ad hoc additions
would be doubly complex compared to its rival, but not all other things would be equal.
Whatever answers suggest themselves to the question of why only a few contenders
are allowed into the ring, I do not see how any of them could involve principles with ceteris
paribus clauses. Suppose that we stipulate the simplicity criterion to play the role in a principle of parsimony with a ceteris paribus clause, to select numbers of ad hoc hypotheses.
At first blush, this might seem like a clear gauge of how hypotheses are generated so that
they diverge only a little from the accepted theory. The problem is that scientific hypotheses are systems of signs and they have content. This means that there will always be another
gauge of simplicity available to give a different measure. As we have seen, it is pretty easy
to conjure incompatible measures of simplicity by selecting some feature and then trying
out different gauges for that feature.
Consider two of LeVerriers hypotheses about the behavior of Mercury: the InterMercurial planet hypothesis and the Inter-Mercurial ring hypothesis. It might be thought
that LeVerriers hypotheses vary from the accepted theory about the matter in the solar sys-

166
tem, in each case, by only one ad hoc hypothesis. However, these hypotheses are neither
conceptually nor empirically equivalent. This shows us that incompatible gauges of simplicity are immediately available. The Inter-Mercurial planet hypothesis might be said to
posit just one more thing if we count things as Swinburne does, or perhaps that it posits
0 extra kinds of things, a naught difference in complexity from the accepted ontology of
Newtonian theory. This hypothesis was also calculated to be very easy (pragmatically simple) to test for and (in an ironic Ockhamian-twist) therefore very unlikely, since the patient
eyes of astronomers had not collected data compatible with this hypothesis.
It might also be thought that the Inter-Mercurial ring hypothesis varies by only
one ad hoc hypothesis from the accepted theory about the matter in the solar system but
this hypothesis would count in the millions of things if we count as Swinburne does, or
alternatively 0 kinds of things. Finally the initial suggestion of the Inter-Mercurial ring
hypothesis was that it would be very difficult to test for (pragmatically complex) if the
rings were at the right sort of inclination to be always invisible from Earth.
It appears that LeVerrier rejected these hypotheses because they failed to unify with
the accepted body of science and because they could be calculated to give absurd empirical
predictions. Testability and unity of science may well have been the criteria which inspired
the introduction of these hypotheses in the first place rather than some vague or arbitrary
principle of parsimony. LeVerrier sought to give theories which were empirically adequate,
which had predictive force, and which unified with the body of science, and then he committed more to discussing what was unlikely rather than what was likely. My objection to
Swinburnes version of the principle of parsimony and to any other in that family is that sci-

167
entific hypotheses have content and this lays open the possibility that incompatible gauges
of simplicity are available. But without the ceteris paribus clause, a principle of parsimony
cannot be the final court of appeal in scientific theory selection because some other principles must guide the non-arbitrary selection of simplicity criteria. Objection strategies (1)
and (2) arm us to show that other things never are equal, because both arbitrary and nonarbitrary gauges of the syntactical, semantical, ontological, epistemological or pragmatical
features of any two ad hoc hypotheses will not give equivalent results.
There may be other versions of the principle of parsimony which do guide theory
construction. Ariew interprets Ockham as promulgating one principle about explanations
and another about statements or concepts. [3] Newton appears to have promulgated principles about kinds of causes and about kinds of entities. Properly formulated these principles
might actually be a guide to the desirable features of scientific theories, at least in the
early stages of theory development. We might plausibly formulate principles of parsimony
which select some of the features of scientific theories guided by considerations for depth
or systematicity.
Suppose that we follow Newtons apparent principle to limit kinds of causes in
the postulate basis. This immediately satisfies some of the features of explanatory depth
and it contributes to systematicity better than positing several independent kinds of causes.
Additionally, this kind of posit will be reflected by semantical gauges of simplicity and
perhaps by syntactical simplicity, depending on the number of laws hypothesized and the
syntactical gauge selected. Also, if we peer over the cubical to borrow this kind of causal
mechanism from a neighboring theorist, we would satisfy the desire for the unity of science.

168
The analysis of this chapter has been dedicated to the kinds of simplicity that systems of signs might have, but when we begin to think about the unity of science new kinds
of simplicity suggest themselves. Perhaps the degrees of closeness to other systems might
serve as the basis for judgments of simplicity also. The sharing of basic postulates of the
same category would certainly be a candidate for a gauge of scientific unity. However,
these sorts of simplicity gauges are only useful in the preliminary stages of theory development. As a theory marches into what Kuhn calls normal science, these principles can be
jettisoned. To seek a circulatory theory of government by embezzling the circulatory theory of blood is to seek a fiction. Darwins theory of natural selection set securely upon its
path of normal science only after it joined with the mechanisms of molecule synthesis and
replication. One example of this version of simplicity leads in one direction to unlikeliness
and in the other to complexity. With these reflections, a skeptical eyebrow is raised about
how long our most cherished unified scientific theories will remain, conceptually, as they
are today. Perhaps more things change than stay the same. Bunge has argued that scientific
theories march in the direction of the complexity of the postulate basis as they mature. The
analysis of this chapter gives us good reason to believe Bunge on this point. The more
over and underpasses that are taken by the theoretical detour, the more our theories will be
systematic.
What does this have to do with likelihood? Something, I would say, but it does not
obviously have to do with the likelihood that a highly systematic theory will be true. The
more lines that cross the face of a mature theory the more likely it is to be replaced by

169
fresh blood19 . Complexity of the postulate basis marches in the direction of testability, and
because linguistically complex theories fit better existing data, in this direction lay future
anomalies.
Let us never speak of ceteris paribus principles of parsimony again. Nothing that
we can imagine satisfies the ceteris paribus clause in a non-arbitrary way. Counterexamples
will forever be conceivable, and this suggests that as a finial court of appeal in scientific
theory selection, there is no Ockhams concept at all.

3.4.2

Avoiding Obsessions with Syntactical Simplicity

The analysis presented in this chapter allows us to give, at least partial answers to
some of the questions raised in the previous chapter. Recall the discussion of the forensic
hypotheses offered by Sherlock Holmes. Holmes does conduct some inferences similar the
inferences made by scientists. In the first chapter the following questions were introduced:
1. What, if anything, would justify hypotheses which track causation when this form of
inference cannot give to us an idea of the necessary connection between causes and
effects, but only an idea of their being correlated in the past?

2. What, if anything, would justify selecting just one hypothesis from among infinitely
many which would also account for the data?
19 Bunge uses the term fresh starts to describe this situation.

[10] It shares some common ground

with Laudans pessimistic meta-induction as well. I would not say that this view is similar to
Kuhns Gestalt Shifts partly because Bunge does not agree with this and so, could not have meant
it. [41] Also, I would not endorse the Gestalt Shift view because Laudan has correctly criticized this
notion.[43]

170

3. What, if anything, would justify a certain degree of inaccuracy in hypothesis formation and what would constrain hypotheses from becoming incorrigible due to vagueness?
We can give pragmatic answers to these questions. Hempel argued that we posit
causal mechanisms for pragmatic reasons. We do not need to know (in some Cartesian
sense) what necessary connection there is between points of data, we only need to posit
causal mechanisms to describe them if we wish to have systematic, an therefore testable
theories. Our postulate basis may be jettisoned if it fails to give predictive results, or if it
comes into extreme dissonance with the accepted body of science, or if it would be impossible to falsify its predictions. None of these problems are serious by the pragmatists
lights. Failures just guide us to select another causal mechanism for the postulate basis
of our revised theory. Negative results might lead us to reconsider the way in which we
classify the data that we tried to describe with some antiquated theory and with some antiquated causal mechanism, or they might inspire us to peer over the cubical in an attempt to
systematize by way of the unity of science.
The second question is too vague to answer as it is. It ought to be made precise
by asking, what if anything would justify selecting just one hypothesis in field X from
among infinitely many which would also account for the data? If we are talking only about
accounting for data and making predictions, then it is obvious that people do not select
just one hypothesis to do this. Engineers still use Newtonian Mechanics everyday. It is
pragmatically simpler than any other alternative involving quantum physics or some general

171
theory of gravitation. In solid state physics it is often convenient to model the properties
of a semi-conductor by treating atoms as if they conformed to the Bohr model. These
pragmatical simplifications might appear to sacrifice depth. My guess is that physicists
would scoff at this charge. They would say that in these cases the postulate basis relevant
to the focus of their projects are preserved and that calculations made in these ways are
approximations which are well within relevant accuracy constraints.
Even if we are not talking about engineering, it does not appear to be the case
that theorists themselves generate only one theory. Simon Newcomb was an exemplary
scientist and it was not because he advanced and defended a precious theory all his own.
It is because he collected many hypotheses from his predecessors and generated a few
of his own, painstakingly working out the deductive consequences of each to see if these
unified with the accepted body of science or if they gave testable results. Newcombs
contribution to science was to make is quite clear that the accepted body of science in
the Nineteenth Century could not explain certain data. This kind of work is extremely
important and I think that philosophers tend to forget about it when they get to focussing
on the weight of evidence in crucial experiments, about the puzzles of theory selection, and
about revolutionary theories themselves without thinking about what set the stage for the
revolution.
So, one answer to question (2) is that not all scientists do select just one hypothesis
from amongst a set of competitors. Another answer is given by the discussion of ceteris
paribus clauses. It is very difficult to find cases where there are genuine competitors.
Empirical adequacy is not the only desideratum of science.

172
On the other hand, perhaps there are cases where one theory is selected for testing
and further research and in these cases again, pragmatic reasons should be easy to give for
why there are no competitors in some cases. This must be evaluated on a case-by-cases
basis.
The answers to question (3) are the same as the answers to question (2). Certain
inaccuracies may be permitted given that scrutability, and falsifiability are not sacrificed.
Additionally, questions (2) and (3) are intimately related. Recall that in the previous chapter
it is pointed out that Holmes appears to have improved the chances that his retrodictions
would be true by not getting too specific about some of the details about the causes of the
watch scratches. This point gestures in the direction of the curve-fitting problem. Forster
and Sober say that,
Curve fitting is a two-step process. First one selects a family of curves
(or the form that the fitted curve must take). Then one finds the curve in that
family (or the curve of the required form) that most accurately fits the data.
These two steps are universally supposed to answer to different standards. The
second step requires some measure of goodness-of-fit. The first is the context
in which simplicity is said to play a role. Intrinsic to this two-step picture
is the idea that these different standards can come into conflict. Maximizing
simplicity usually requires sacrifice in goodness-of-fit. And perfect goodnessof-fit can usually be achieved only by selecting a complex curve. ([24] pg.1-2)

Forster and Sober discuss Akaikes curve fitting algorithm and some of its possible
applications in this article. Akaikes idea is that we can use the trade-off between accuracy
and goodness of fit and the trade-off between linguistic exactness and syntactical simplicity
to fit curves to data. We can do this partly because linguistic exactness places some constraints on accuracy. The gauge of syntactical simplicity used by Akaike is very similar to
the one proposed by Jeffreys.

173
We have already seen how many objections are available for Jeffreys gauge of simplicity. The application of this gauge is limited to hypotheses which are expressible in a
differential form. It is also the case that this must be used in cases where arbitrarily selecting a different coordinate system would be ruled out, perhaps because a mathematical
transformation would change the physical meaning of the hypothesis. It must also be the
case that some reasons are available to justify giving equal weight to the values of heterogeneous variables and that other kinds of simplicity are ignored. In short, Akaikes curvefitting algorithms cannot be thought of as general solutions for all cases of curve-fitting in
science. Forster and Sober are perfectly aware of these problems.
This view of the curve fitting problem engenders two puzzles. The first
concerns the nature and justification of simplicity. What makes one curve simpler than another and why should the simplicity of a curve have any relevance
to our opinions about which curves are true? The second concerns the relation
of simplicity and goodness-of-fit. When these two desiderata conflict, how is a
trade-off to be effected? A host of serious and inventive philosophical proposals notwithstanding both these questions remain unanswered. ([24] pg.2)

Now, I think that some of the recent excitement about Akaikes algorithms arises for
two reasons. 1) It is tempting to think that because Akaikes algorithms give a mathematical
solution to the curve-fitting problem which means that they add no presuppositions to the
scientific basis. 2) Because questions about justification remain, it is also easy to be swept
away by the hope that these algorithms do something magical and avoiding metaphysics
altogether. Both of these views would be mistaken.
It has already been shown that it might be practical to seek some kind of syntactical
simplicity in the early stages of theory development. The reasons are that if we can formulate a theory mathematically, then we satisfy the desire for linguistic exactness. If we

174
gauge the simplicity of a mathematical hypothesis by its syntactical properties then we satisfy a desire for systematicity. When it is possible to use an algorithm involving a gauge of
syntactical simplicity we would have good pragmatic reasons to do so. Nevertheless, if the
final theory (which may take generations to develop) is deep and gives predictive results,
then we will not think that it is because of the syntactical simplicity of its hypotheses, but
because of its representativeness and depth.
Secondly, it is not the case that an actual application of one of Akaikes algorithms
is completely presuppositionless. Data does not appear from nowhere. Data is collected by
skill practitioners who subscribe to a rigorous methodology. This means that their methods
are theoretically embedded. Applying Akaikes algorithms to data collected in a theoretically embedded way should only be expected to reveal things about the presuppositions of
the theories which were employed in generating the data in the first place. Perhaps curvefitting algorithms are best used in error analysis and maybe their successful employment
helps us to notice things about the relationships between background theories.
The point of all of this is that mere syntactical simplicity has not yet answered the
Humean skeptic. Why would it? Hume saw that there was something essential about science that forced us to posit transempirical terms. Of course this is the case. It was so
by Platos design. Now, Hume thought that something non-rational happened in the mind
when certain fictions were posited to solve the problem of induction and to avoid the discomfort of confronting contradictions. I do not know if Hume was right about this. In
Reduction: Ontological and Linguistic Facets ([30]) Hempel has argued that the linguistic turn or the obsession with the linguistic analysis, leaves us still with the problem

175
of identifying what are the natural kind terms. This analysis shows why this is the case
also. Hempel has shown us that we have pragmatic reasons to take the theoretical detour,
and it is this that makes our theories deep and this that makes them have predictive power.
No matter how we count the syntactical or semantical properties of hypotheses, they have
content and so the symbols out of which a theory is constructed reflect the postulate basis
and whatever it is (models or axioms) that relate the fundamental postulates. When the
analysis reaches certain levels of abstraction, it is easy to forget this, and this is what I
believe happens to those who become obsessed with syntactical simplicity gauges as a way
to avoid doing metaphysics.

3.4.3

A Modest Suggestion

The arguments of this chapter regarding depth and representativeness have been
thoroughly pragmatic. For this reason it is very easy to make another pragmatic suggestion: that we relativize the analysis of simplicity to specific scientific theories or bodies
of theories. Platos mixed view is committed to two abstract kinds of simplicity to start
with; a certain complexity on one way of counting. Given the nature of Platos way of
mixing the Pythagorean hypothesis with the Herclitian hypothesis, we are left looking for
yet another kind of simplicity to solve the epistemological problems that we have inherited. By taking a relativistic approach to scientific theory construction and evaluation, we
can let the abstractness of Platos mixed view inspire creativity in filling out the details.
We can let the extreme heterogeneity of simplicity work for us so long as we remember to
make simplicity criteria clear, so that we do not stumble into equivocal or vague uses of the

176
term simplicity. We might, then on pragmatic grounds find non-arbitrary ways to select
simplicity criteria which aid in theory construction and the testing of existing theories. I
suggest that we try out different gauges of simplicity for different theories to see what sorts
of results we get. Failures might show us as much about our theories as do the successes
by this method.
I am not sure that mere conceivability will do the work that philosophers intend for
it in the philosophy of science. In the Swinburne example I do not know how to imagine
Newtonian mechanics to be true and then conceive of the alternative hypothesis he asks
us to conjure. In the philosophy of science philosophers run the risk of conceiving of
examples which are not scientific. The present discussion should cast this in even better
light. The conceptual map program renders a very complicated diagram. It very difficult
to think of examples where we change, lets say, pragmatical simplicity by introducing a
hypothesis, and yet retain the proper relations to the other desiderata of science. My project
is supposed to contribute to conceptual analysis by creating an image which allows us to
consider many relations at once. This would be too great a task for my mind without an aid.
Bunge provided all of the parts needed for the whole represented by the diagram which the
conceptual map program renders. Synthesizing these arguments into the whole which the
conceptual map program produces does not give us something more than the whole, but we
may be in a better position to engage these arguments.
Relativizing analysis to systems might have some interesting and helpful results.
As a practical matter, scientists and philosophers review theories and articles regularly and
it would be helpful if they had a tool which allows them to refer very quickly to the inter-

177
relations between the desiderata of science. This may help them to evaluate the theories
before them. If the arguments which generated this conceptual map program are all sound,
then people in these positions may only glance at the diagram and judge whether or not
a proposed theory will satisfy the demands of depth, systematicity and testability. If any
are unsound, then the arguments given to construct the conceptual map should be criticized
and the program updated.
I would think the relativized view would be helpful to pure academic research as
well. We should explore different approaches to the development of theories. It might help
to think holistically of the desiderata of science and then try out different combinations of
simplicity gauges for systems to see both how they cohere with the desiderata of science
and to see what sorts of empirical predictions these systems give.
This suggestion is not incompatible with either scientific realism or scientific antirealism. The anti-realist should be happy to accept the relativized approach and the realist
might expect that successful theories generated by the relativized approach will, in the long
run, contribute to triangulating the fundamental structure of reality. After all, scientific
theories make substantive claims about the way that the world is. This is consistent with
the desiderata of depth and representativeness.

178
CHAPTER 4
SIMPLICITY AND JUSTIFICATION

Like a gemstone, the clearer we make simplicity the more difficult it is to handle.
The discussion of the previous chapter helps to clarify issues in the philosophy of simplicity generally. It also contains arguments which help to avoid traditional pitfalls, like the
mistake of considering a principle of parsimony to be the final court of appeal in scientific theory evaluation. On the other hand, we have acquired many terms, distinctions and
relations which can be difficult to manage. The aim of this chapter is to investigate the
question of the justification of simplicity principles in scientific methodology. I will explain how one version of the question of the justification of simplicity is simply confused.
Then I will attempt to formulate the questions about justification that may be reasonably
asked. I will then organize the types of arguments that might be offered to justify principles
of parsimony in scientific theory construction and evaluation. I will show which arguments
might bear fruit and apply the results of the analysis of simplicity in science to show which
kinds of arguments should be eliminated from this enterprise. My analysis shows first that
the problem of the justification of simplicity judgments in science is motivated by the fact
that we are still awaiting arguments (rather than mere assertions) that the problems of simplicity can be shuttled off to aesthetics or psychology. Second, I show that some argument
forms may be in the running to give this justification although many of them may depend
upon solutions to traditional problems in metaphysics and epistemology. Induction, some
kind of transcendental deduction, and inference to the best explanation may all be in the
running, although each of these, in their own ways, may involve commitments that are

179
unacceptable to some philosophers.
Bas van Fraassen says that
Studies in philosophy of science divide roughly into two sorts. The first,
which may be called foundational, concerns the content and structure of theories. The other sort of study deals with the relations of a theory on the one
hand, to the world and to the theory-user on the other. ([61] pg.2)1

This chapter investigates issues analogous to the second kind of study distinguished
by van Fraassen. This study focusses on what types of inferences might be employed to
show that there is a relation between a scientific methodology and the scientist or between a
methodology and the world. The first and second chapters have shown that it is not merely
our theories that make ontological commitments, but our methods as well. When our analysis of methodological principles is sufficiently clear, we see that they involve criteria, some
of which select ontological objects. This is unavoidable. Hempels arguments are typically
taken to show that we cannot elucidate scientific theories by reducing all of the terms of
science to those which denote only that which is empirically basic. We could have learned
a similar lesson from Bunge who has shown that empirical simplicity involves trade-offs
with the other desiderata of science trade-offs which tend toward unacceptable sacrifices
in depth and predictive power. It is difficult to see how to balance the meta-scientific criteria
1 The

quotation from van Fraassen must not mislead. Van Fraassen is perhaps best known for advancing a particular scientific antirealist view. I cannot do justice to the scientific realism/antirealism
controversy. One reason for this is that there may be as many realist and anti-realist views in the
philosophy of science as there are philosophers of science. Organizing this discourse would require
a book all its own. Another reason is that it might take yet another book to organize the various
arguments on meaning and reference which are crucial to this debate. Still, van Fraassens work
plays a key role in several of the issues presented in this chapter. This is because the issue of simplicity and issues involving types of inference have arisen in the debate between van Fraassen and
his critics. Fortunately, van Fraassen and his critics are systematic thinkers and we can easily pluck
from the debate the issues relevant to simplicity in science.

180
by an empirical criterion alone. Hempels arguments give insights beyond just the elucidation of theories. They reveal similar problems for the elucidation of scientific methodology
as well.
Why distinguish the philosophy of methodology from the philosophy of theories?
The answer is that underdetermination runs both ways from theory to methodology and
from methodology to theory. A popular twentieth century view must be put to rest, namely
that we can read off of a theory what are the methods for its construction and evaluation.
Bunges arguments show that a theory may be constructed by using many different criteria
of the same species. Consider, for instance, that we may chose infinitely many gauges of
syntactical simplicity, many of which will generate the same theories because the transformation rules of the system will allow different expressions of predicates and functional
relations. One goal that we might strive for in the elucidation of scientific methodology
would be the discovery of which methods are instrumentally reliable that is, are a reliable
guide to the construction and evaluation of the scientific theories that we accept. We might
have this aim so that the methods identified could be used to construct entirely new theories.
Because they are general in this way, we should expect that any particular methodology (a
cluster of methodological principles and criteria) underdetermines any particular theory.
Underdetermination runs from methodological principles to theories partly because even
clearly defined criteria may be satisfied in many ways and also because clusters of criteria
are involved in the construction and evaluation of theories. Even if the criterion involved
in a popular principle happened to pick out a feature of the world or the mind, that principle could be mixed in with a bunch of other unjustified principles that are unrelated to

181
the truth, that are themselves false, or that generate false theories. Just imagine the new
age religious theorist who constructs her theory using a few legitimate (and perhaps justifiable) scientific principles, but who should not be taken to have given a scientific theory
because the methodology is mixed with mistaken principles. Perhaps we can pinpoint her
error if she advances an argument that from quantum theory we can deduce that humans
have free will. No such deduction is possible, but perhaps humans do have free will and
perhaps only random causation can explain quantum tunnelling. Methodological principles
committed to free will and random causes2 may turn out to be justified. Perhaps this theory
leapt the rails of science because it was constructed using some other principles that do not
involve the satisfaction of meta-scientific criteria. I do not know what draws people into
Lucretius mistake. Perhaps it is the swerve principle: if determinism is false, then humans
have free will. Even a principle involving a criterion committed to the correct ontology is
insufficient for the construction of theories that exemplify depth, fit, predictive power, or
systematicity (let alone truth). An ontological criterion might be involved in an example
like this. It may select kinds of causes. Current interpretations of physical theory are typically committed to at least two kinds of causes: deterministic and random. It looks as if
it is true that determinism is false, but even if the swerve principle were constructed using
this criterion, it is surely false and its inclusion would pollute a scientific methodology. For
these reasons, our epistemological investigation of the methods of science in some respects
2 As

I understand it, the strong force binds together the nuclei of atoms. But the strong force
is a short range force and protons repel one another because they are positively charged. Spectral
analysis shows that the nuclei of atoms are sometimes formed in states with energies too low to
overcome repulsive electro-magnetic forces. Somehow, protons occasionally jump an energy barrier
and stick together.

182
parallels arguments about the ontology of theories, but deserves its own dialectical space
anyway.
Some types of arguments may be in the running for showing that there is a relationship between a methodology and the world, and if there is a high degree of continuity
between traditional epistemology and the philosophy of scientific methodology, then these
are the same arguments that would be in the running for solving the problem of skepticism.
Other types of arguments would be of the sort that, at best, they could only be employed to
show that there is some relationship between a methodology and its user, but taken alone
arguments of this type may not specify the nature of that relation. It may seem as if this
is well worn philosophical territory, but the analytic treatment of simplicity itself is rare
despite the obvious fact that this cuts across the perennial issues in philosophy. Now that
we have a sketch about how to clarify simplicity criteria (at least of some sorts), I wish to
investigate which argument forms might be employed to justify them.
We have seen that pragmatic justification is easy to come by for specific simplicity
principles in specific contexts. This is the case for many kinds of psychological simplicity.
For example, the clever choice of a coordinate system and its origin can contribute to algorithmic simplicity, which is useful for solving certain problems. Another kind of simplicity
judgment is involved in the methods for estimating discrete integrations where iterated dependencies are expressed by the relevant equations. LeVerrier and Newcomb used methods
of this sort to model planetary motions.3 Some problems cannot be solved without simplic3 Although

the example from the first chapter involves Newtonian Mechanics, Monte-Carlo and
Runge-Kutta are methods of this sort developed in the Twentieth Century and which are used in a
variety of scientific fields and in economics.

183
ity principles like these that aid in estimating the solutions to complex multi-dimensional
integrals (I do not use the word parsimony in this context because these judgments would
rarely be taken to track ontology). The simplicity criteria for these principles will be chosen within some practical range so that they balance with other meta-scientific criteria like
depth and accuracy. The remaining question is whether or not all simplicity criteria useable in science can be justified on pragmatic grounds, or if any might be justified in some
other way. If the answer to this question is that it is not possible to reduce the justification
of every simplicity criterion to pragmatic justifications, then something like a traditional
question about justification is motivated. My view is that questions about the relation between simplicity criteria and the truth cannot be reduced to pragmatic justifications without
residue. Even if aesthetic judgments play a significant role in scientific theory construction
and appraisal, we might wish to know why aesthetic judgments about ontology generate
theories that make good predictions.4
To show that some principle of parsimony is justified in the sense relevant to this
investigation is to show that there is a relationship between some feature of the principle
and the truth. Often the chore is taken to be that if it is claimed that some principle of
parsimony that involves simplicity criterion is truth conducive, then it must be shown
that the world is simple in respect. Put in another way, the traditional question of the
4 An

old engineering principle is, if it looks good, then it will go fast. Why do sports cars and
fighter jets look like sharks? Probably, some sharks are on to something about nature that we could
either imitate or discover by an application of the laws of physics. Perhaps we never hear about the
companies filled with people who thought that daffodils were beautiful because their fighter jets did
not perform very well. Claiming that judgments are aesthetic in nature does not bring an end to our
questions for an explanation of why certain judgments generate the results that we desire. Perhaps
there are even scientific explanations for why people have some of the tastes that they have.

184
justification of a principle of parsimony has been taken to be the question of whether the
simplicity criterion involved in the principle succeeds in picking out a feature of the world.
This formulation of the problem is nearly, but not quite appropriate. Simplicity
judgments are comparative, and therefore, it would be to commit a category mistake to ask
whether or not simplicity corresponds to the way that the world is. In some respects, simplicity is like tall or short. These predicates are comparative and therefore context
dependent. So, I would not ask whether simplicity is related to the truth, rather whether
some sub-set of the meta-scientific criteria, those which govern ontological simplicity judgments, select objects had by the world. Again, not all simplicity criteria select ontological
objects. Some select linguistic, epistemological, semantical or psychological objects for
comparison. We wish to formulate the question about the justification of simplicity judgments in science by asking it about the sub-set of the cluster of meta-scientific criteria that
select ontological objects.
This formulation at least points in the right direction because it avoids the confusion
between comparative predicates and attribute predicates by asking whether or not any of
a specific subset of the meta-scientific criteria succeed in picking out kinds of things in
the world. Still, perhaps a few too many questions are begged by formulating the problem
in the above way. For one thing, we wish to know whether or not the truth is even an
aim of science (or if it is among the meta-scientific criteria). If the truth is a desideratum
of science, then we would need to know what relations it stands in to the other metascientific criteria. Questions about what the truth is and how it is related to anything deserve
volumes of focused investigation and so are out of range here. Another problem is that this

185
formulation does not make room for the two possible aspects of our study: the relation
between a methodology and the world and the relation between a methodology and its
user. If possible, it would be nice to formulate the problem so that it makes room for views
that begin and end with the relation between a methodology and the its user.
It will be useful to adopt the terminology that Richard Boyd ([8]) employs to avoid
begging any questions in the debate between scientific realists and scientific anti-realists. I
shall follow Boyd in calling a scientific theory or hypothesis instrumentally reliable when
its empirical predictions are approximately true.5 I will call a scientific methodology instrumentally reliable when it is a reliable guide to the acceptance of instrumentally reliable
theories or hypotheses.6 This still leaves much room for further work on what kind of relation there might be between a principle of parsimony and the truth such that it is truth
conducive, how many places the relation has, what the relata of the relation are, or how to
account for verisimilitude. At least since Hume, philosophers have recognized how difficult is the problem of showing that there are necessary connections at all and for at least
one hundred years the makes probable relation has met with similar difficulties.
If a cluster of principles were committed to the correct ontology and their interrelations were also in accordance with the truth (whatever that entails), then it would be no
mere accident that these principles were instrumentally reliable. So something stronger
than the material conditional would relate true principles to instrumental reliability. Also
since something stronger than the material conditional would relate ontological criteria and
5 This

notion of truth is whatever the constructive empiricist has in mind when he says that
theories are construed as true.
6 Boyd

(1985) Lex Orandi Est Lex Credendi pp.4

186
the interrelations between the metascientific criteria to the truth, something stronger than
the material conditional relates ontological criteria and the interrelations betwen the metascientific criteria to instrumental reliability. Our very difficult problem, then, is to give an
argument that there is some stronger-than-material-conditional relation between clusters of
metascientific criteria selecting either objects of the mind or of the world and related in
specific ways to one another and their instrumental reliability. Presumably, we would have
a collection of instrumentally reliable theories that are taken to be paradigmatic of science.
Also, we would have the methodological principles used in constructing these theories,
testing them, and in preserving them in those cases where competitors were introduced. In
order to proceed, I assume that it is possible to accumulate a bunch of paradigmatic instrumentally reliable theories. Perhaps this is not possible. However, many philosophers of
science think that this is not only possible, but that they can identify some of these theories.
The problem, then, is to show that the methodological principles employed in the construction and appraisal of these theories make these (and future) theories probable. The problem
is to give an argument that justifies one of the following kinds of claim:
Principle: 4.0.1. Simplicity criteria related in way R select objects of the world principles of parsimony ,R are instrumentally reliable.
or, perhaps more modestly,
Principle: 4.0.2. Simplicity criteria related in way R select structures that require theories to take certain forms principles of parsimony ,R are instrumentally reliable.
is the set of meta-scientific criteria that make ontological commitments and R

187
is the specific set of interrelations between the meta-scientific criteria. The relata of R
include not only the criteria of but some of the other criteria as well. As we have seen
some pragmatical criteria may be indispensable for some methods. The formulation of
4.0.2 is meant to capture the view that some philosophers (like Swinburne) attempt to
argue for. The structures that require theories to take certain forms might be found in the
mind, in the world, or some combination thereof. The degree to which a philosophical view
is an Idealist view might be gauged by the degree to which a view is committed to these
structures being in the mind.
There is wide agreement among scientists and philosophers that some principle of
parsimony is indispensable for scientific practice. If we are to give a justificatory argument, what we require is an argument that will take us from the truth of the consequent
to the truth of the antecedent of either (4.0.1) or (4.0.2). If we are to deny the justification of simplicity principles in science, then arguments would need to be given that the
instrumental reliability of theories constructed by methodology,R have nothing to do with
.
I discuss five general strategies which might engage the question of the justification
of (4.0.1) or (4.0.2).
1. Offer a reductive analysis of simplicity principles and show that the reduction bases
are relevant to the question of justification.
2. Deny that the way that the world is has anything to do with the instrumental reliability
of any principle of parsimony.
3. Offer an inductive argument where the antecedent of either (4.0.1) or (4.0.2) appears

188
as the conclusion.
4. Offer an argument involving an indispensability principle for the truth of the antecedent of either (4.0.1) or (4.0.2).
5. Offer an argument that the best explanation for the consequent of either (4.0.1) or
(4.0.2) is that the antecedent is true.
Initially, it may appear that there is no further discussion to be had on the relation of
simplicity criteria to the truth because the question about the justification of simplicity principles might be misconceived from the outset. Unless we have independent grounds upon
which to settle the questions of ontology, only comparative judgments of simplicity are
possible. Since simplicity judgments are comparative, it would be a mistake to ask whether
some particular simplicity criterion is related to the truth, since we need only to apply the
same criterion to a different set of theories to get different results. The question that we
want an answer to anyway is the question of whether or not the methodological principles
at work in science have any correct ontological commitments and whether or not the metascientific criteria have been balanced in the appropriate ways. An answer to the question of
ontology would settle many other questions raised in the second chapter, like the question
of how to systematize theories or how to get predictive success. Goodman pointed out that
if only we had some independent way to settle the questions of ontology, many if not all
of the puzzles about how to specify and how to gauge simplicity may disappear as well.
Our question is whether or not a principle involving simplicity criteria can be a guide to
ontology, not whether we end up with simplicity of some form as a result of having settled
matters of ontology (supposing any of these can be settled) by some other means. Recall

189
Lavoisiers claim that a simple principle gave rise to astonishingly simple results. I do not
know what Lavoisiers methodological principle was nor do I know why he was astonished.
The only situation that would be astonishing would be if one kind of simplicity principle
used to construct a theory regularly generated theories exemplifying a completely different
kind of simplicity. Although it is surely not what Lavoisier had noticed, it would be astonishing if a principle involving a criterion of psychological simplicity gave ontologically
simple results. What we wish to know is what set of relations would make something like
that happen and what relation the ontological criteria involved in methodology have to the
instrumental reliability of the theories they are used to construct or evaluate.
The complexity of the interrelations between the metascientific criteria raises another sort of question about justification which may still involve simplicity criteria and perhaps this paves the way to formulate the present inquiry. The question is, what if anything
explains why a particular set of methods is instrumentally reliable? Clusters of simplicity
criteria are involved in theory construction and testing. For a pair of theories (an old one
and a newer version or two which compete to explain some data) it may be possible to
establish unambiguous and non-arbitrary criteria specifying the objects of comparison and
the gauges for their comparison. Simplicity criteria are related to one another and to the
other metascientific criteria sometimes in a complementary way and sometimes by inverse
proportionality, and sometimes they are not related to one another at all. Of all of the,
perhaps infinitely many, clusters of principles that could be employed in scientific theory
construction and evaluation, we wish to know if any are justified.
In order to engage this question, the analysis from the previous chapter must be

190
applied to actual cases of theory construction and testing. However, we can get a rough
notion of what sort of cluster of methods might be the subject of justificatory arguments by
using some of the results from the first chapter about the history of LeVerriers hypotheses.
Suppose that a scientist is confronted with LeVerriers puzzle: how to get the physical theory to give accurate predictions about the positions of celestial bodies. Suppose also
that this scientist embraces methodological principles much like those that Newton appears
to have promulgated: simplicity with respect to the positing of kinds of causes where systematicity is the aim, simplicity with respect to ad hoc causes, and simplicity with respect
to kinds of properties and kinds of entities (as opposed to property instances or numbers of
individuals). The theories which provide the basis for comparison are T Newtonian Mechanics and T 0 Newtonian Mechanics augmented with additional hypotheses. This means
that the construction of T 0 would be constrained from moving very far from T in certain
respects. By adding no additional kinds of causes or kinds of entities to T the goal of the
unity of science will be satisfied and systematicity and depth will be preserved. Such a
methodology contributes to pragmatical, epistemological, and semantical complexities because it is difficult to test for inter-Mercurial matter and because semantical complexities
will be involved in the introduction of terms denoting additional entities in the solar system.
So, we can see that our interest will be in clusters of interrelated principles, because justification is required for constructing theories that on the basis of some kinds of simplicity
principles when they are always traded-off with the satisfaction of other kinds of simplicity.
If we were able to justify an instrumentally reliable methodology, and simplicity criteria were part of that methodology, we would have a set of simplicity criteria which could

191
be put to work in methodological principles for the construction and evaluation of theories.
We would have another set of criteria that are not relevant to simplicity because complexities of some kinds are allowed or even to be expected. The justificatory argument would
then be expected to give some kind of complex conclusion asserting that Methodology,R
is related to the truth. The conclusion may be expected to assert things like that the ontology of the causes of Methodology,R corresponded to the kinds of causes in the world or
that the properties posited by Methodology,R correspond to the properties of the world or
the generalizations of the theories generated by Methodology,R express relations between
the entities and properties which it posits which are some of the relations of the world and
that the testing procedures used to obtain data and the theories which inform those testing
procedures are also justified.
Perhaps we should not expect to get these results from an application of a justificatory argument to the methodology of LeVerrier since one of his aims was to preserve
Newtonian mechanics. Hindsight shows us that Newtonian Mechanics could not be preserved and that some of the data that LeVerrier counted in favor of his hypothesis were
misunderstood. For this study it may be more appropriate to investigate some of historys
most secure changes (if there are any such things), like how it was that Special Relativity unified electro-magnetics and physical dynamics. Even though LeVerriers endeavors
to save Newtonian theory failed, we do have an important glimpse into how a complex
set of methodological principles can contribute to the evaluation of theories from the brief
historical study in the second chapter. Recall that the arguments given by LeVerrier and
Newcomb established that Mercurys perihelion was a phenomenon that required expla-

192
nation. In the early part of the twentieth century, General Relativitys ability to explain
this played a key role in the acceptance of GR. My project cannot triangulate precisely the
fundamental principles of science. This project must be left to historians of science or for
future inquiries. However, what should be clear is that the justificatory arguments would
involve a complex set of issues including ontological justifications for the theoretical postulates of a theory and also justification for inter-theoretical relations. For this project some
argument must track backwards from the truth of the consequent of either (4.0.1) or (4.0.2)
to the truth of the antecedent of either (4.0.1) or (4.0.2) for whatever the properly specified
set of methodological principles are given a particular set of scientific theories. To do this,
there are, as noted, five strategies: give a reductive argument, deny that it can be done, give
an inductive argument, give an argument involving an indispensability principle or give an
argument to the best explanation.

4.1

Reduction

Reduction is a term that is probably used far more frequently in the philosophy of
science than it is made clear. It is often thought that there is a relation between reduction
and ontological simplicity. It is also common to associate successful reductions with increases in explanatory depth. Increases in explanatory depth achieved by some reductions
in science are often associated with what is sometimes called fundamentality: the idea that
quantum physics gives, provisionally, the final ontology of the physical world. However,
there is no obvious reason to expect that fundamental physics will give us our only reductive bases. Perhaps multiply realizable properties (like color) are reductive bases also.

193
Woodward and Hitchcock, for instance, defend the view that temperature is basic instead
of molecule and kinetic energy. The first part of this section aims to clarify different
uses of the word reduction in the philosophy of science. I do this by discussing Marshall
Spectors view about how to elucidate reduction. This involves contrasting Spectors view
with other famous attempts to elucidate reduction and also with some very different ways
that the word reduction is used in various branches of philosophy and physics.
The second part of this section investigates an epistemological motivation for reduction. The question is whether or not reductions may be performed on methodological
principles involving simplicity criteria with the result that all concepts of simplicity could
be reduced either to one concept of simplicity or to metascientific criteria of other kinds.
I will explain why this kind of reduction would be a conceptual reduction instead of an
explanatory reduction.
If all simplicity judgments can be conceptually reduced to one type of simplicity
judgment, then our arguments about justification could begin with those reductive bases.
Alternatively, if each unique kind of simplicity judgment can be conceptually reduced to a
judgment of some sort other than a simplicity judgment then arguments about justification
could begin with the reductive bases (whatever these turn out to be). Successful reductions of simplicity might at least provide our epistemological inquiry with some welcomed
direction. Ultimately the fantasy might be to diffuse the problems of the justification of
simplicity principles in science to other problems of justification or perhaps to draw entirely to a close our discussion of the justification of simplicity criteria by completely reducing simplicity to the other metascientific criteria. The primary argument of this section

194
is really Mario Bunges argument recast with some contemporary terminology. Although
reductions of specific simplicity criteria to other metascientific criteria may be possible
in specific scientific contexts, no overall reduction of simplicity is possible. When this is
added to the results from the previous chapter, which show that simplicity criteria can be
made clear and that many do play important roles in theory construction and testing, we
are forced to take seriously the question about the justification of simplicity.

4.1.1

Clarifying Reduction in Science

Marshall Spector ([56]) has argued that reduction is best elucidated by the notion
of a term-by-term replacement involving two different vocabularies. On this view of reduction, metalinguistic functions may define the terms of a reduced vocabulary in terms
of a reducing vocabulary. This view is defended on the basis that it avoids several problems faced by the standard analysis given by Ernest Nagel. [47]

Nagels attempt to

elucidate reduction in science involved the explanatory reduction of laws to other laws that
explain overlapping sets of data. Spectors approach to reduction is more appropriate than
the standard analysis for an application to the metascientific criteria. This is because the
metascientific criteria are not any more a part of physical theories than a metalangauge is
a part of an object language, and whatever might be explained by a reduction of some set
of metascientific criteria to other metascientific criteria should not be understood as having
been explained in the same way that a theory is said to explain data or in the same way that
one theory explains another.
7 Nagels

account of theoretical reduction met with many objections. Some of the problems with
Nagels approach will be outlined. See Spectors ([56]) Chapter 3 for a detailed discussion.

195
Spector traces the genesis of the modern discourse on reduction to Russells Maxim
which was presented in the essay Logical Atomism (1924). The maxim states that whenever
possible, constructions involving known entities could be substituted for unknown entities.
Russell and Whitehead found by experience[56] that this principle could be applied to
mathematical logic and suggested its application for other areas.
It is fairly easy to understand the motivations for this kind of substitution in some
cases involving applied mathematics. Take, for example, imaginary numbers, which regularly appear in physics and engineering contexts. When a convex automobile side-view
mirror is designed on paper we find that a virtual image is projected into the complex numbered plane (inside the mirror). It might be thought that it presents a troubling paradox or
an outright contradiction to say that imaginary numbers are involved in predictions about
the real world at least in this case, about an image with real numbered dimensions. The
idea is that imaginary numbers can be defined using functions that only involve only real
numbers. This reduction carries ontological implications since the result suggests that we
need not reify imaginary numbers, at least in the case of mirrors.
The motivations for ontological reductions in the philosophy of science are usually
a bit different from the motivations in the above example though. One tempting application
of this kind of reduction is to reduce unobservables to observables. Perhaps Berkeleys
view that it is a manifest contradiction to assert that what is unobservable exists is a view
of this sort. Modern arguments in the philosophy of science have cast serious doubt on
whether the observable-unobservable dichotomy can be maintained. One of the reasons for
suspicion about the observable-unobservable dichotomy is given in the previous chapters

196
discussion of Hempels Theorticians Dilemma. Deep scientific explanations end up being
a mixed bag of predicates denoting bread-box sized objects and theoretical terms.
Having noticed problems of this sort, we might consider ontological reductions.
Perhaps we adopt a philosophy of language on which our theoretical terms refer to the
most basic entities that we know of in the web of causes and effects and seek reductions
that go in a different direction, from macro-objects to micro-objects. This kind of approach
to reduction has been investigated for possible application in several adjacent areas in the
philosophy of science, including projects to elucidate the scientific progress exemplified
by the replacement of older theories with newer ones that account for the same data. This
approach to reduction may also cast light on attempts to understand how successful theories
are viewed as deep or how theories satisfy the unity of science criterion.8 Here, again, the
early critics of the Positivist program (Bunge, Goodman, and Hempel) offer useful insights.
As Goodman has argued, the notion of simplicity relevant to science is a notion of economy
rather than one of mere linguistic simplicity, and this notion is related to systematicity
which provides a basic criterion for theory construction. It is tempting, then, to take the
fundamental particles and their properties (perhaps protons, neutrons and electrons with
parity, spin and charge, or quarks with charm) as our reductive bases. Even so, if the
special sciences are autonomous, then no such reduction is possible for every scientific
statement.
Perhaps it is now clear why the discourse about reduction in the philosophy of
8 Reduction

(term-by-term replacement), explanation, and the goal to unify scientific knowledge

come together in Philip Kitchers Explanatory Unification (1981). Kitcher says that unification is
achieved by using similar arguments in the derivation of many accepted sentences (pg.519).

197
science is as complex and diverse as it is. There are different notions of reduction that are
motivated in different ways and standing in diverse relations to other issues in philosophy
more broadly. It is often, but not always the case that the motivations for reduction are
ontological. Accounts of reduction may differ in what is taken to be the reductive basis:
micro-physical entities or properties, macro entities or properties, particulars or abstracta.
Perhaps we can also see why confusion about reduction in science is easy to come
by. First, there are many distinct ways to consider reduction in science and some of these
may be incompatible with one another, depending on what kind of account of explanation is embraced. Second, the word reduce is often used differently by philosophers and
scientists. Let us consider a few of the very different ways that this term has been used.
It is popular to say that Keplers laws of motion were reduced to Newtons laws of
motion and it is popular to say that thermodynamics was reduced to classical mechanics. It
is true that Keplers laws are derivable from Newtons laws plus simplifying assumptions
(this last part is often omitted, contributing immediately to error). It is not true that Keplers
laws are strictly derivable from Newtons laws. This is one of Kuhns important insights in
The Structure of Scientific Revolutions([41]). Keplers laws predict elliptical orbits for all
of the planets in a solar system and they predict that the planets will sweep out equal areas
in equal times. In order to get these results from Newtonian mechanics one must discount
several important relations described by Newtonian mechanics including: the gravitational
interactions between planets and other planets, and the fact that planets are not perfect
spheres with evenly distributed densities.9 When our predictions about celestial mechanics
9 Even

though Newtonians model masses as point masses, some very tricky integrals may be

198
require a high degree of precision these simplifying assumptions are no longer appropriate.
This was the situation for LeVerrier and Newcomb given the scientific climate they found
themselves in. At a high degree of precision, Newtonian orbits are not Keplerian orbits.
Additionally, the first chapter discussion of Newcombs work explains why it is the case
that celestial motions are not derivable from Newtonian mechanics. So if it is the case
that the concepts of the large scale phenomena of planetary orbits are reducible to the
fundamental laws of Newtonian mechanics then deducibility is not the relation upon which
this reduction depends.
A similar problem arises for the popular example of the reduction of thermodynamics to classical mechanics plus molecular theory (this last part is often omitted, contributing
immediately to error). The usual example is that temperature is reduced to a function of the
masses and velocities of the molecules of a system or, perhaps more precisely, that temperature is reduced to mean kinetic molecular energy. However, the laws of thermodynamics
are not derivable from the dynamical laws involving molecules. The laws of thermodynamics apply to systems composed of things smaller than the molecular theory includes
(like plasmas or electron gasses). A plasma, for example, has a temperature although it is
not composed of molecules. If molecules or atoms are taken to be in our reductive bases,
then temperature is multiply-realizable because systems composed of things much smaller
than molecules or atoms have temperatures. Again, perhaps deduction is not the relation
essential to reductions.
employed to get very accurate descriptions of some of the behavior of planets, including wobbles
(or nutations) in the precession of the planets.

199
This rough sketch of these examples helps to raise several key points. In the case
of the mathematical reduction of imaginary numbers to real numbers the motivations can
be made clear. One motivation is ontological. Another motivation is psychological (algorithmic). Hempel has stated that the philosophical motivations for scientific reductions
are ontological also. However, when we seek the precision afforded by a linguistic analysis we may say nothing about ontology, rather only something about how it is possible
to define a set of terms ([30] pg. 195-196). The discussion of Goodmans analysis of
the defining power of terms raises the same issue. Both of these authors would probably
agree that they have shown that the Positivists rejection of metaphysics does not end up
being justified. I believe that Hempels arguments against the Positivists hostility toward
metaphysics contributed to a common contemporary view that places metaphysics back at
the forefront of inquiry, especially involving the ontology of theories. Similar arguments
place metaphysics at the forefront of our investigation into scientific methodology, which
I believe deserves some dialectical separation from questions about how to elucidate scientific progress or scientific explanation. We might also ask what the motivations for a
philosophical dialog about reduction in science are. Is it the case that we are convinced
that reduction takes place in the sciences and we are searching for paradigmatic examples
of it? Is it the case that we think that there is progress in science and we wish to elucidate
the notion of progress, hoping that some reductive analysis will bear fruit? Do we have a
notion of reduction, and we are looking for uses for it in the same way that mathematicians
are currently exploring the utility of model theory? These motivations are all very different
and so should be kept clear.

200
At least the leading authors who contributed to the reduction in science dialectic
had some specific things in mind. Ernest Nagel sought to elucidate reduction when he gave
a detailed discussion of thermodynamics. Nagel held that reduction was explanatory and
endeavored to fit the analysis of reduction to the deductive-nomological model of explanation. Spector holds that heterogeneous reductions can be explanatory and he employs
the above examples to show that reductive explanations need not depend upon deduction.
Spector attempts to split the horns of Hempels linguistic-ontological dilemma by claiming that term-by-term concept replacement is the better way to elucidate reduction than by
deductive arguments. On Spectors proposal the ontological impetus for reduction may be
preserved, and by expressing the reduction in the metalanguage linguistic precision is preserved. In this way, reductions may be explanatory, but not in the same way that scientific
laws explain data.
One final example will help to clarify uses of the word reduction in the philosophy of science discourse. We must not confuse reduction with the phrase reduces to.
Physicists say that relativistic physics reduces to classical mechanics as

v
c

0. This kind

of reduction would be a homogeneous reduction. This kind of reduction is called homogeneous because the reduced theory does not contain different terms than the reducing theory,
although the reducing theory may contain different terms than the reduced theory. Deduction may play a role in this kind of reduction if a principle of the following sort is embraced
at the methodological level: If error constraint is not met or exceeded and

v
c

0, then

replace the laws of relativistic mechanics with the laws of classical mechanics. In this
sense of the word reduce a similar principle might govern the replacement of the laws

201
of quantum mechanics with the laws of classical mechanics when values for momentum
are much larger than Plancks constant h:

0. Useful points can be drawn from this

example. Physicists may use the term reduce in a way that is quite different from the
way in which philosophers use the term and this can contribute to confusion if the two uses
are not kept clear. Deduction may play a role in some kinds of reduction. However, the
homogeneous kind of reduction given in this example is neither explanatory nor ontological in nature. This kind of reduction is not explanatory because classical mechanics does
not explain relativistic physics. Classical mechanics is not merely a less accurate way to do
relativistic calculations. It is false. The first chapter discussion of Newcombs work shows
why classical mechanics cannot account for some data. This kind of reduction is a pragmatic notion which indicates that approximations are allowed in some contexts. Finally,
the key notion involved in this kind of reduction is not a notion involving term-by-term
replacement. Rather, the relevant notion involves the replacement of the laws of one theory
with the laws of another.

4.1.2

The Conceptual Reduction and Simplicity

There is another application of Russells Maxim which ought to be considered for

the goals of this chapter: the conceptual reduction. We have many examples of this kind
of reduction from applied science and engineering where different kinds of systems are
modeled using the same mathematics because the descriptions of the systems share certain
formal characteristics. It is common to design and diagnose automotive electrical systems
on the hydraulic model of current flow. Also, acoustical systems may be modeled using

202
electrical circuits and disease propagation may be modeled using the same equations that
model radioactive decay. In each of these cases two different systems share certain structural features that makes it possible to swap the terms or concepts of one discipline for the
terms or concepts of another in the very same equations. This kind of reduction is a conceptual reduction rather than an explanatory reduction because it is certainly not the case
that hydraulics explains electricity or that disease propagation explains radioactive decay.
Obviously, this kind of reduction is useful. It does not, however, provide any explanation of
the sort relevant to scientific explanation. If conceptual reduction provides any explanatory
insights they are insights about minds, either human or divine.
Let us now see if simplicity principles in science share sufficient structural similarities with other principles so that a conceptual reduction might be possible. It would be
nice if an abstract analysis of the conceptual reduction of metascientific criteria were possible. But it is not because simplicity criteria are radically heterogeneous and they stand in
unique interrelations with one another. Conceptual reductions of specific principles to principles involving other metascientific criteria may be possible. Recall that Bunge suggested
that in some cases, a principle of parsimony that appears to select ontological objects, or
at least to have ontological implications for theory construction, might actually serve an
epistemological role. Bunge said that
Unnecessarily complicated assumptions and theories should be avoided;
that is, hypotheses and theoretical systems employing inscrutable predicates,
such as Providence and collective unconscious, should be shaved with Occams razor. Notice, however, that Occams razor does not hang in the air, but
falls under the more general rule, Do not propose ungrounded and untestable
hypotheses([11] pg.129).
I will follow up on Bunges suggestion that conceptual reductions of specific prin-

203
ciples may be possible by reviewing the study of the arguments given by LeVerrier. Recall
that LeVerrier constructed several hypotheses initially aimed at explaining Mercurys perihelion. He rejected each because they could not account for Mercurys perihelion without
also suggesting other data that was incompatible with the body of collected data. An interMercurial planet and inter-Mercurial astroids with mass sufficient to account for Mercurys
perihelion would be observable. Inter-Mercurial matter either would not account for Mercurys perihelion or would be observable. It seems plausible that these hypotheses were
rejected because they did not fit the data. I assume that these hypotheses were rejected because they failed to satisfy a straightforward empirical criterion. However, these hypotheses
were constructed using additional criteria, and this tells us much more about LeVerriers
methodology. After all, LeVerrier did not posit some kind of ether pulling on Mercury, and
he did not attempt to alter Newtons law of gravitation. This suggests that LeVerrier may
have constructed hypotheses by employing some kinds of simplicity principles.
It seems quite clear that a scientist who constructs an astroid hypothesis to explain
some data does not include in his methodology a criterion of simplicity with respect to
numbers of entities. He may have held principles that involved simplicity with respect to
kinds of substances though. He may also have held a principle of syntactical simplicity
which could account for the fact that he did not wish to explain Mercurys perihelion by
adding terms to Newtons law of gravitation. However, it is also possible that he had a more
fundamental principle that committed him to the unity of science. Perhaps simplicity with
respect to kinds of causes and syntactical simplicity can be replaced with a unity of science
principle with the same results. My review of LeVerriers work makes this seem plausible

204
also. Obviously, settling which, if any of these, was the case is the job for historians of
the philosophy of science, but I do think that this shows that conceptual reductions of
some simplicity principles to principles involving other metascientific criteria are possible.
If conceptual reductions of specific methodological principles are possible, we still know
nothing about which principles are more fundamental. For this additional arguments are
required.
Let us consider a different approach to the conceptual reduction of simplicity criteria. Perhaps the most promising approach is to forget about theories for a moment and to
explore the possibility that all kinds of simplicity are constructed in the mind on the basis
of one type. It is natural to speculate that people originally get some idea of simplicity
through experience and then annex that idea to other cases, thereby acquiring the entire
stock of simplicity criteria which govern judgments in the mature mind. Perhaps as a child
I lifted two objects and felt that one offered less resistance than the other. The experience
might have provided the basic inequality from which the mind abstracts the form of the
comparative simplicity relation. This form might be applied to other kinds of cases. Perhaps some skill is more difficult to learn than another or perhaps one sentence consists of
more terms than another and in both of these cases the mind borrows the general less than
relation from previous experience and applies this relation to the new cases. On this story,
simplicity criteria would select the objects of comparison for special cases of judgments of
inequality. For all I know, this could well be the way that the mind acquires a mixed bag of
simplicity criteria.
However, in any particular case of theory construction or evaluation I will have to

205
call upon several different judgments of simplicity. Some of these criteria are complementary; syntactical simplicity gauged in a certain way may be compatible with ontological
judgments of simplicity. But some of these judgments are always incompatible with others. Syntactical simplicity is inversely related to epistemological simplicity. This shows
that when we are confronted by any particular case involving simplicity judgments there
will be gains and there will be losses in simplicity depending on which criterion is under
consideration. Perhaps it is true that simplicity concepts share structural similarities, but
when we consider clearly defined simplicity criteria in science, no conceptual reduction of
all of the kinds of simplicity criteria to just one type is possible because contradictions will
arise.
The question about how human minds come to have their ideas is a different question from whether or not the methods of science are reducible to one another in specific
cases. They are both important questions. But when it comes to the methods of science, no
conceptual reduction of simplicity criteria is possible. Simplicity, as Bunge has argued, is
radically heterogeneous. This fact, the fact that complexity of some sort is always traded
with simplicity of another sort, is cast in the very grammar of science as Aristotle advanced
it. We may lean a bit more on the Pythagorean notion of simplicity, as Aristotle may have,
at the cost of some syntactical and ontological complexities, positing many terms of the
stay-the-same-sort and several mechanisms of causation. Or we may lean on the Heraclitian notion a bit more like Kuhn and Laudan, holding that more things change than stay
the same; where simplicity is gained by positing flux, we incur syntactical and semantical
complexities when we give names to all of the things which change.

206
Some of the problems for contemporary literature on Bayesian and Akaike information criteria are laid bare by these observations. Some authors are overtaken by the
idea that traditional problems of ontology are dissolved by Akaikes equations, which relate accuracy and syntactical simplicity with predictive power (or testability). The database
rendering from the previous chapter shows us that it is not surprising at all that these are
related to one another. These curve fitting algorithms suggest that we can reduce simplicity
judgments in theory selection to syntactical simplicity and epistemological criteria. However, not all of the problems of ontological simplicity in science are eliminated in this way.
Syntactical simplicity is not the only kind of simplicity relevant to theory construction and
testing. Given the extreme heterogeneity of simplicity concepts, no overall conceptual reduction is possible. It matters not what reductive basis is chosen. There will always be
a trade-off between some kinds of simplicity and others. We require some guide to acceptable accuracy constraints in order to know what counts as empirically adequate in any
particular case. For this, we must appeal to depth and this raises the ontological question again because, as we saw in the previous chapter, must take some un-operationalized
primitives in order to satisfy the formal requirements of depth on balance with the other
metascientific criteria.
It is intriguing to consider cases where conceptual reductions of specific simplicity principles may be available so that a criterion that makes an ontic suggestion may be
stated in other terms. Speculating about this option, I would say that it would probably
require a very robust anti-realist account of science, so that the instrumental reliability of
the relevant methodological principles could be accounted for in terms of pragmatic, aes-

207
thetic, and social goals. The reason that this is difficult to imagine at this stage is because
well constructed scientific hypotheses satisfy several metascientific criteria at once. It is
important to rid ourselves of the Positivist notion that empirical adequacy is the only relevant metascientific criterion. For one thing, empirical adequacy does not appear to be a
strictly empirical criterion. There is a trade-off between accuracy and predictive power
a trade-off which is balanced by appeal to the other metascientific criteria, some of which
suggest ontology. For another thing, hypotheses balance several different criteria: accuracy
and predictive power are traded for one another when notational simplicity is adjusted this
way or that, and as certain epistemological simplicities are traded for abstractions, and all
of this must be achieved without sacrificing depth. It will also be very difficult to find cases
where some unity of science goals do not play a significant role in theory construction or
testing. It seems, then, that the first salvo in this battle must be fired by the historians of
science. In any case, I do not expect examples to arise that change Bunges insight that
simplicity is radically heterogeneous and that no overall reduction is possible.
My strategy, in this section, was to rule out the possibility that the questions about
the justification of ontological simplicity criteria can be eliminated by reductive analyses.
First, the term reduction is subject to vagueness. Second, it is used in the philosophy
of science in several ways. It is only conceptual reduction that might play a role in our
present inquiry. Bunges arguments show that no overall reduction of simplicity in science
is possible. However, it might be possible to conceptually reduce some simplicity principles
to others or to methodological principles that do not involve simplicity criteria. However, if
conceptual reductions of specific principles are possible, we know only something about the

208
structural similarities of the principles. We do not know which are more fundamental. All
of this shows that the possibility of conceptual reducing simplicity principles cannot defeat
an investigation into the justification of simplicity judgments. However, I believe that the
preceding discussions have motivated the question of justification. Hempels arguments
show that we cannot give a purely empirical elucidation of scientific theories. Ontological
criteria will serve in principles that make theories deep.

4.2

Denial

I am not sure who would deny that at least some simplicity principles are instrumentally reliable. Even if a mature theory migrates far from its early form by acquiring
many additional terms and relations, or by shifts in the semantics, it is still the case that
syntactical simplicity, in perhaps various forms, aids in the early stages of theory development by achieving systematicity and testability. It is still the case that scientists seek
to balance the trade-offs between notational or algorithmic simplicities and scrutability. It
would seem, then, that even the die-hard pragmatist would be inclined to accept that some
simplicity principles are instrumentally reliable. The question ought to be as it has been
cast here: is the truth related to any of the metascientific criteria?
But there remain two possibilities for denying that arguments can be given to justify
simplicity principles. One might take simplicity as a methodological primitive which cannot be justified, or, as suggested in the previous section, perhaps some simplicity principles
can be conceptually reduced to other criteria given a sufficiently robust anti-realist account
of science. Alan Baker explains these options in the Stanford Encyclopedia of Philosophy

209
entry on simplicity,
Some philosophers have approached the issue of justifying simplicity principles by arguing that simplicity has intrinsic value as a theoretical goal. Sober,
for example, writes:
Just as the question why be rational? may have no non-circular answer,
the same may be true of the question why should simplicity be considered in
evaluating the plausibility of hypotheses? (Sober 2001, p. 19).
Such intrinsic value may be primitive in some sense, or it may be analyzable as one aspect of some broader value. For those who favor the second
approach, a popular candidate for this broader value is aesthetic. Derkse (1992)
is a book-length development of this idea, and echoes can be found in Quines
remarks in connection with his defense of Occams Razor concerning his
taste for clear skies and desert landscapes. In general, forging a connection between aesthetic virtue and simplicity principles seems better suited to
defending methodological rather than epistemic principles.([?])
Baker goes on to say that,
Kuhn (1977) takes this line, claiming that how much weight individual
scientists give a particular theoretical virtue, such as simplicity, is solely a
matter of taste, and is not open to rational resolution. ([?])
I agree with Baker. It would seem that the reduction of simplicity to aesthetic values
would better suited to a defense of methodological principles than epistemic principles.
But our previous discussion shows how these views are a bit misguided anyway. Bunge
has successfully argued that simplicity is radically heterogeneous. No overall reduction is
possible. Even if one kind of simplicity were conceptually fundamental, we would need
to know which one this is. So, I cannot see how it would work out that simplicity is
conceptually fundamental as Sober suggests that it might be. The word it does not refer
to just one thing in the philosophy of simplicity. The same problem arises for attempts to
reduce simplicity judgments to aesthetic values. This reduction is just impossible. We have
shown already that many of the simplicity criteria that aid in the construction and appraisal
of theories are justified on pragmatic grounds.

210
Kuhns claim is not sufficient to deny the justification of all of the different kinds
of simplicity principles in science. Probably, Kuhn did not intend to deny the instrumental
reliability of some methods, at least at certain stages of history. Also, Kuhn was probably
thinking only of a kind of ontological simplicity and imagining that scientists reflect only
their tastes for certain ontological commitments in theory construction and testing. On
Kuhns account, scientists would acquire these tastes early in their careers from text books
and teachers. Kuhn is right to criticize text books that run together different theories from
different ages in a misguided attempt to give to students a holistic picture of progress in
science. My own astrophysics textbook freely switches between Keplerian laws and Newtonian laws of motion without mentioning the almost comic fact that Kepler thought that
he was unifying the laws of celestial motion with the laws of musical harmony! I am as
unsettled as Kuhn is about this sort of thing. It is not just that no reductive explanation will
account for the replacement of Keplerian laws of motion by Newtonian laws of motion,
but no conceptual reduction is possible because Keplers ancient Pythagorean notions of
causation are totally different from Newtons notion of causation. The two theories are
totally different and they do not even share structural similarities. I agree with Kuhn that it
is either nave or irresponsible of authors to subject students to this sort of confusion.
Although I share with Kuhn these concerns, his claim that simplicity is solely a
matter of taste does not account for the sort of thing that Goodman and Bunge have pointed
out, that in the early stages of theory construction, economy of form is desirable because
it contributes to systematicity and testability. This is not merely a matter of taste and it is
open to rational resolution. Really all that Kuhn has pointed out is that there is a problem

211
justifying how incompatible measures of simplicity are to be weighted. But the gauging
problem is different from the counting problem. Each require specific justifications for
specific applications. This same point is made by Sober in the article cited by Baker,
Several philosophers have asserted, without providing much of a supporting argument, that the trade-off problem [the problem of how to weight simplicity measures against the other metascientific criteria] has no objective solution. For example, Kuhn (1977) claimed that scientists differ in how much
importance they assign to one virtue of a theory as opposed to another, and
that this difference is just a matter of taste. One scientist may think that the
most important demand on a theory is that it should make accurate predictions;
another may hold that the first duty of a theory is that it be elegant and general.
Kuhn does not offer much of an argument for this claim; it merely constitutes
his impression of what would be left open by any compelling and reasonably
complete set of epistemological standards. Of course, it is not in dispute that
scientists have different temperaments. But from the point of view of normative theory, it is far from obvious that no uniquely best trade-off exists between
simplicity and the other factors that affect a theorys plausibility. Once again,
this cannot be settled in advance of examining particular proposals [about how
to measure simplicity]. ([55])
The better approach is to deny that the world being any sort of way explains the
instrumental reliability of methodological principles involving simplicity criteria. For this
project, I leave it to the defenders of scientific anti-realism to marshal their arguments. It is
important to keep in mind the serious challenge faced by those who wish to say anything
about the instrumental reliability of simplicity criteria in scientific methodology. The problem is that one method can give different results when a theory A is compared to a theory
B and when the same theory A is compared to another theory C. Simplicity judgments
are comparative. The problem of specifying a suitably abstract methodological principle
that may have its instrumental reliability either explained or explained away is faced by
both scientific realists and anti-realists. Also, there are two two distinct problems which
require justification once they are suitably analyzed: the counting problem and the gauging

212
problem. These problems have been offered many different rigorous formulations. These
are employed by scientists in a wide variety of fields every hour of every day. Typically
though, the modern scientist just points and clicks, interfacing with some software program, to classify data and fit curves. Rarely does the practicing scientist ask what method
was used (nearest-neighbor, decision trees, linear regression, neural net, BIC, AIC, or some
combination). Since the scientist rarely asks which method her software program employs,
she rarely asks whether or not different results would be obtained by switching methods.
Sometimes the results are different, and this question ought to be asked. Once we realize
that different methods can give different results the question of what justifies any particular
method is once again raised. The its always worked for me, mantra of normal science, is
typically not an inductive justification, but an appeal to pragmatic values aimed at putting to
rest the pesky philosophers inquiry about justification. However, the mantra is an inhibitor
when what Kuhn calls periods of crisis demand changes in our methods.
My hope is that attempts to deny the possibility of the justification of simplicity
principles in scientific methodology has run its course in philosophy. Perhaps it is natural
for this dogma to die a few publishing generations after the death of Positivism. Hempel
and others argued that the Positivists could not justify a wholesale hostility to metaphysics.
Additionally, the very notion that what distinguishes science from other human activities is
its ability to bring about wide-spread agreement between its skilled practitioners inspired
attempts to explain or to explain away the idea of progress in science. This idea about
science is misguided also. I suggest that the clusters of metascientific criteria that govern
activities of various sorts are different and this is sufficient to distinguish activities from

213
one another. The metascientific criteria differ from the metamusical criteria and this is
sufficient to explain the unique paths that different human intellectual endeavors take.10
Kuhn is trapped in a post-Positivist kind of view, that simplicity reflects ontology. In some
cases this is true. In some cases it is not. I wish to revive the idea that arguments are
required for our discussion of the justification of simplicity judgments in science. Perhaps
arguments can be given to justify this or that kind of simplicity in science. Perhaps, as Sober
suggests, some kind (or kinds) of simplicity are conceptually fundamental and we cannot
justify them with arguments, but we still require arguments to distinguish the conceptually
fundamental kinds of simplicity from those that are not.
Finally, I believe that the analysis of the metascientific criteria and their interrelations suggests something that, for whatever reasons, has not fallen squarely into the limelight. Although several arguments from the twentieth century reignited a discussion about
the ontology of theories, the point that we must also discuss the ontology of scientific methods seems to have slipped just below the radar of many (but not all) discussions. Perhaps
this is due to the fact that the philosophy of scientific methodology has not been given
its own dialectical space. It is clear that Paul Churchland (discussed in the next section)
thought of this and employed the idea in an argument against scientific anti-realism. How10 Although

musicians and composers clearly have their own unique styles, these styles may arise
from the various ways that each individual satisfies widely agreed upon metamusical criteria. For
example, it may be widely accepted that we prefer non-parallel harmonic motion and that this criterion is only sacrificed when special effects are desired. In other words, there may be a trade-off
between the metamusical criteria. We may find that there are syntactical simplicity criteria in music,
but it is not likely that the specific gauges of syntactical simplicity in music are the same as those
employed in science. A contrasting difference between science and music is probably that depth
is not among the metamusical criteria. Although some explanation may be, in principle, available
for what makes a half diminished chord well-placed, a chords being well-placed does not explain
anything.

214
ever, this does not show us precisely why considerations of ontology may be unavoidable
in the elucidation and justification of methodological principles. The string of reasoning
that leads us here is rather complex in its details, but we ought to be in a position to list
the relevant general topics. 1) Methodological principles involve criteria. 2) Criteria select
objects. 3) The metascientific criteria are radically heterogeneous, so no over all reduction
is possible even for a particular genus (say simplicity). 4) The metascientific criteria stand
in specific set of complex interrelations. 5) Any attempt to make one genus of the metascientific criteria conceptually fundamental faces either the problem of the irreducibility of
the metascientific criteria, or leads to methods that commit us to unacceptable sacrifices in
how the others are satisfied. This final point is exemplified in several different problems
that have been discussed like the linguistic-ontological dilemma or the problems involved
in an obsession with epistemological simplicity. In short, we do not define the metascientific criteria in terms of one another. We balance them against one another, and our
methodological choices have ontological consequences.
This is not merely a restatement of Hempels dilemma. The arguments about the
justification of methodological principles is an issue distinct from questions about the ontology of theories. Again, consider that perhaps we can elucidate a unity of science criterion epistemologically - that embracing it involves aiming to unify the encyclopedia of
scientific knowledge. Still, theories constructed with a unity of science principle may make
ontological commitments (or resist them) like not positing invisible frictionless substances
or changing the laws of nature. This ought to motivate a genuine discussion about the justification of methodological principles. The study of ontology and the questions of tradi-

215
tional epistemology do not apply to theories alone, but to methodologies as well. In some
cases the questions of traditional epistemology do line up with the questions about the
justification of methodological principles because methodological principles often make
ontological commitments.

4.3

Induction

In this section, I argue that inductive justifications of simplicity criteria in science

do not fail because of any obvious formal problems. Still, I do not favor this approach
because I think that it is doomed to fall short of the aims of this project.
Induction was discussed briefly in the first chapter, using an example from a Sherlock Holmes novel. What we might expect for the justification of some specific cluster of
methodological principles would be a collection of case studies involving those methods
and the instrumentally reliable theories that they were employed in constructing or evaluating.
1. Past Observation Set 1: Methodology,R was employed in constructing and evaluating T1 , and T1 is instrumentally reliable.
2. Past Observation Set 2: Methodology,R was employed in constructing and evaluating T2 , and T2 is instrumentally reliable.
3. Past Observation Set 3: Methodology,R was employed in constructing and evaluating T3 , and T3 is instrumentally reliable.
4. On the basis of uniform past experiences, I have come to expect that Methodology,R
gives instrumentally reliable theories.

216
Therefore: It is probable that using Methodology,R to construct and evaluate T4 will result in
T4 being instrumentally reliable. (just like in every other case that I have observed)
Now, is a set of simplicity criteria. The fourth premise involves the phrase, on
the basis of uniform past experiences and this involves a uniformity of nature criterion
which is often noted to be a simplicity criterion. This immediately raises the question of
whether or not this argument is circular. It is far from obvious that this is the problem with
this argument.

4.3.1

The Question of Circularity

Swinburne raises an objection to the circularity of inductive arguments employed

to justify simplicity criteria. Swinburne says that
the claim was only that usually the simplest theory has proved the better
predictor, and that would justify the assertion that probably a given simplest
theory will prove the better predictor on some future occasion.
However, even this modest claim about the history of science that usually the simplest theory has proved the better predictor seems very doubtful. In
many areas of inquiry the simpler laws which served well in the past have
been replaced by more complicated laws the Boyle to Van der Waals story
is hardly unique. But even if the simplest theories have usually proved better
predictors, this would not provide justification for the subsequent use of the
criterion of simplicity, for the reason that the justification itself relies on the
criterion of simplicity. There are different ways of extrapolating from the corpus of past data about the relative success which was had by actual theories and
which would have been had by possible theories of different kinds, if they had
been formulated. Usually simplest theories predict better than more complex
theories is one way. Another way is an extrapolation of the form Usually
theories formulated by the Greeks in the bath, by Englishmen who watch apples drop, or Germans who work in patent offices...etc, which initially appeal
to the scientific community predict better than other theories. An extrapolation of this kind, spelled out at great length, would yield the data of the past
history of science just as well as the extrapolation which connects predictive
success with simplicity. Of course, this kind of extrapolation is an absurd way
of extrapolating from the past data concerning the relative success of theories.

217
And why is it absurd? Because it is far less simple than the obvious way of extrapolating. We have assumed the principle in providing its justification! Any
other purported empirical justification of the use of the criterion of simplicity
will prove equally circular. ([59] pg. 52-53)

Swinburne has not yet made his case. is a set of the heterogeneous metascientific
criteria specifying a variety of objects for comparative judgments. So long as does not
include a unity of nature criterion, this argument does not employ the principle of the
uniformity of nature in the making of the inference. Perhaps consists of syntactical and
epistemological simplicity criteria. These might specify numbers of logical connectives
in a sentence and the degrees of accessability to a terms denotation. So long as all we
wish to give is an argument that past experience makes it probable that will generate
instrumentally reliable theories, then it would seem that this is what we have. Swinburne
knows that there are many different simplicity criteria. This will be made clear in my
discussion of his view in the next section. Since Swinburne knows this, it is not quite clear
why he uses the word criterion in this passage, although I have some suspicions about it
which are my next topic of discussion.

4.3.1.1

Sameness and Change

Although Plato did not wish to do philosophy of nature, probably due to his view

that sensations are merely illusory images of what is real, Plato does appear to be the
father of the middle way between Pythagoras and Heraclitus. Aristotle founded science
upon this middle way and now we are stuck with it. I cannot see how to avoid accepting
that some things change and some stay the same. I cannot see how to reconceive of the
scientific project involving attempts to give both explanations and predictions about the

218
regularities in nature. Nietzsche, with his notorious eye for irony, sees this pattern and its
problems echoed throughout philosophy. Scientists have a serious challenge giving deep
explanations about the regularities in nature. Another serious challenge is the one faced
by philosophers, and sometimes scientists acting as philosophers, to categorize the things
of the stay-the-same sort and the things of the change sort. Perhaps Swinburnes objection
reflects the many problems of induction. Swinburne says that There are different ways of
extrapolating from the corpus of past data...([59] pg.52)
Swinburne may be pointing to the fact that simplicity criteria may be involved in
categorizing the relevant terms of the stay-the-same sort. If the principle involved in this
is the principle of the uniformity of nature, then indeed this argument is circular. Since we
have ways to formally analyze many different kinds of simplicity criteria, it seems to me
that we could discover whether this is the case. In order to avoid the circularity, though, all
that is required is that none of the criteria used to sort predicates into classes or to govern
the inductive inference are members of . This does not mean that induction would justify
every principle of simplicity. For all we know, the criteria employed in the sorting of predicates and the criterion of induction could be justified by some other arguments (perhaps
some indispensability argument). If that were the case, then the inductive argument would
have no formal problems.
Swinburne probably notices that the problems about how to fill out the details of
the Pythagorean-Heraclitean mixed principles are fundamental problems. Hume noticed
the same thing. I see the problem, but it is far from obvious that an attempt to inductively
justify the highly specialized, highly formalized syntactical simplicity criteria employed in

219
curve-fitting algorithms (for example) would employ those very same criteria to govern the
inductive inference. In a sort-of unsatisfactory way, the radical heterogeneity of simplicity
comes to the rescue of induction, at least in this very specialized area of inquiry, because it
is not obvious than an inductive justification of simplicity principles in science is viciously
circular. However, induction itself still depends upon some justification of the simplicity
criteria that govern it.
This result is interesting, and perhaps it is a game-changer for some controversies
in the philosophy of science. It might be true, as Swinburne expects, that answers to the
justification of simplicity in science await the skeptic to be given a final answer. However,
we should not expect the arguments meant to justify simplicity in science to be the very
same arguments aimed at answering the skeptic. The degree to which the philosophy of
science has inferential criteria in common with classical epistemology I will call the degree
of continuity between these disciplines. I do not know how much continuity to expect, but
the radical heterogeneity of simplicity suggests that there is some discontinuity. If this is so,
then we would not answer the skeptic with the same argument that justified some scientific
methodology.

4.3.1.2

Reasons Not to Use Induction

Perhaps my reasons to be suspicious about the possibility of giving successful in-

ductive arguments to justify simplicity principles in science are related to Swinburnes.

Induction does not seem capable of telling us what we want to know in this case and I
do not see how to reconcile an inductive argument for the justification of simplicity with

220
changes in science. I do not agree with Swinburne that inductive arguments justifying simplicity in science are circular, but I do agree that they do not get at the epistemologically
fundamental issues.
First, let us suppose that the these methods have always worked for me claim is
the claim of inductive justification. An inductive argument linking methodology,R to true
theories is exactly what one would expect if there were some necessary connection between
the use of a specific set of interrelated methodological principles and the generation of true
(or nearly true) theories. It seems that giving an inductive justification for the use of simplicity principles in science cannot, by itself, stop me from wondering if there is a reason
why that was possible. Induction cannot tell us anything about necessary connections, if
there are any. Perhaps I have shown that an inductive justification of simplicity principles
in science would not fail because of any obvious circularity. I do not know if I have shown
that such an argument is possible, and it looks very unlikely that someone will actually give
it. For me to be convinced that induction is the alpha and the omega of the justification of
simplicity in science, I would have to be convinced that no other argument could do the
trick. Only in that case would I let the justification of simplicity live by virtue of its heterogeneity or die by the sins of circularity. I do not believe that all other arguments forms
have yet been ruled out.
Second, Swinburne may be on to something when he objects that there are different
ways of extrapolating data. One fundamental problem with the argument from induction
is that we are as justified in believing that things will change as we are in believing that
things will remain the same. Kuhns arguments drew attention to the important fact that

221
revolutions are an important part of science. Laudan holds that the flux of science is more
constant than even Kuhn thought that it was. Science changes and we would not wish
to load the inductive dice by picking a bunch of methodological principles that generated
theories of the same sort that we have generated in the past. Swinburne gives an example
of a law of thermal dynamics that was replaced by a more syntactically complex law. The
problem with inductive justification is that we would be exactly as justified in believing
that complexity favors science as believing that simplicity favors it. At some point, we
want to generate entirely new theories in entirely new areas of study. If we wish to justify
the methodological principles that will take us into the unknown and guide us through
the unexpected, we might wish to know something more than induction can provide. We
might wish to know something about the human mind or about the world. One original
motivation for looking into the roles of simplicity judgments in science arises directly from
the suspicion that complexity might just be the friend of science. Induction, alone, cannot
decide between the complexity and simplicity.

4.3.2

Indispensability

Kant suggested that we employ an indispensability principle to solve the problem of

induction. This would change the approach to traditional metaphysics which relied mainly
on a principle of non-contradiction. The change would be that we might discover something about the mind rather than about the mind-independent world we might discover
which concepts make possible the having of specific experiences. In its strongest form, the
indispensability principle states that having some concept is both necessary and sufficient

222
for experiencing sensations in the unique way that we do.
Unfortunately, Kant did not know that paradigmatic examples would be difficult to
come by. Kant thought that he could argue that space and time concepts are indispensable
for the kind of knowledge that we get from doing physics. As it turned out, even length and
time would be conceived of relative to an inertial frame of reference after the splash made
by Einsteins Special Relativity. Science also ended up embracing non-Euclidian geometries, so that even the axioms of extension in space and time cannot easily be shown to be
indispensable for the way that a scientist would experience experiments while embracing
any particular theory. But this kind of inference did not die along with any hopes for a
Copernican Revolution in metaphysics. Some authors still hope that an indispensability
principle will solve the traditional problem of showing that our theories are related to the
truth.
One approach is to employ the indispensability principle in an argument aimed at
showing that theories constructed and evaluated by methodology,R will be instrumentally
reliable because we have concepts that are filled out when the criteria are satisfied (however this works out in experience). On this version of the indispensability argument some
concepts are identified as being both necessary and sufficient for the having of some experiences.
Another approach involves embracing a weaker version of the indispensability principle. A weaker form of the principle might state only that some concepts are sufficient (but
not necessary) for the instrumental reliability of scientific theories. For our present inquiry,
the idea would be to show that the reason that specific theories constructed using a specific

223
methodology give descriptive and predictive results within acceptable accuracy constraints
is that they produce theories with the correct ontology. This argument form should be
expected to identify a feature of cognition that makes possible our having experiences of
a specific sort the sort had by those who embrace true theories. Perhaps, moored to a
causal theory of reference and with the sorting of some tricky details about the structure of
concepts, this argument form would also solve the traditional problem of ontology.
One version of the indispensability argument is as follows:
1. We have good reason to believe the literal truth of our most successful (instrumentally
reliable) scientific theories.
2. Methodology M,R is indispensable for scientific practice.
C: We have reason to believe in the existence of the properties and mechanisms posited
by theories constructed using M,R .
So, if M,R is a method which is indispensable for the construction or testing of
scientific hypotheses or theories, then we have reason to believe that the properties and
mechanisms posited by M,R exist 11 .
In this section I present and evaluate arguments given by two modern authors which
appear to take this form. Richard Swinburnes argument, I think, does properly take the
form of this argument. It involves the strong form of the indispensability principle. Paul
11 This

indispensability argument borrowed from Alan Bakers Mathematics, Indispensability

and Scientific Progress (2001) where it is argued that that this principle ought to be understood
broadly so that it applies not only to current scientific principles and methods, but also to any which
history has shown to have played an indispensable role in the development of science. I have altered
Bakers argument a bit by including the , R subscript. I believe that I have shown that even a
methodology committed to the correct ontology had better have principles that stand in the correct
relations to one another. I changed Bakers word abstracta to properties and mechanisms.

224
Churchlands argument appears to take the form of this argument with the weak version
of the indispensability principle, although it is possible that Churchlands argument should
be analyzed as an argument to the best explanation. Arguments to the best explanation are
discussed in the next section.
I raise objections to the arguments given by Swinburne and Churchland. I conclude
that Swinburnes approach is more promising than the weak version of the indispensability
argument.

4.3.3

Richard Swinburnes Argument

Swinburnes Simplicity as Evidence of Truth (1997) is a short and lucid defense of

the view that, other things being equal, the simplest hypothesis among competitors is most
likely to be true. Swinburne says,
I seek in this essay to show that other things being equal the simplest
hypothesis proposed as an explanation of phenomena is more likely to be true
than is any other available hypothesis, that its predictions are more likely to
be true than those of any other available hypothesis, and that it is an a priori
epistemic principle that simplicity is evidence of the truth. ([59]pg.7)

We learn a bit more about Swinburnes view when he says that

If one theory is superior to another in yielding the data to a higher degree
of inductive probability, or in yielding more data to the same degree of probability, then as such it is more likely to be true, but any greater simplicity of
a rival theory is a compensating factor which could lead to equal probability
overall or even to the greater probability of the rival. ([59]pg. 19)

Swinburnes defense of principles involving simplicity has many merits. Among

them is Swinburnes sensitivity to the fact that the metascientific criteria stand in a tight set
of interrelations. We would expect more accurate predictions and perhaps some increase

225
in the number of data (or, perhaps the kinds of data) predicted by deeper theories and it
is likely that Swinburnes arguments reflect sensitivity to the relations between depth and
the other metascientific criteria. Swinburnes view should also accommodate a variety of
views about curve-fitting that are based generally upon the gauge of syntactical simplicity
introduced by Jeffreys.
Swinburne knows that we construct and evaluate theories according to some criterion of how well a hypothesis fits the data and some criterion of how well a hypothesis
makes predictions. He also knows that some criterion of fit (in a different sense) with
our background knowledge plays a crucial role in (what Kuhn would call) normal science.
These notions deserve more elucidation than they have been given in the previous chapter,
but Swinburnes sensitivity to these features of science suggests that his view would work
nicely with the mixed bag of empirical and non-empirical criteria. Swinburne uses slightly
different terminology, but as I have discussed them, his a posteriori criteria would include
predictive power/fit, accuracy, and the unity of science. Swinburne adds two a priori criteria
of theory appraisal, one involving the trade-off between conceptual content and predictive
power and the other involving simplicity.12 Swinburnes sensitivity to the conceptual content of theories reveals his sensitivity to the kinds of issues discussed by Goodman about
economy. It is Swinburnes aim to argue only that simplicity is an indispensable a priori
criterion of theory evaluation, but I take it to be a merit of his view that he sees that several
12 Swinburne

says that there are four criteria of theory evaluation, two a posteriori and two a
priori. I do not see how to do this job with just four criteria. The suggestion that there is a criterion
of the trade-off between content and predictive power appears to conflate the metascientific criteria
with their interrelations.

226
other criteria are involved in the construction and appraisal of theories.
Another merit of Swinburnes view is that he distinguishes different kinds of simplicity criteria. I have argued that this is where all philosophies of simplicity must begin.
I will present Swinburnes view and raise one objection and another important challenge.
My objection is that his criteria depend on the ceteris paribus clause and that my arguments
have shown that no principle of parsimony in science can depend on this. My challenge
for proponents of this strategy is to show how the indispensability argument can uniquely
select any particular cluster of simplicity criteria as those that are necessary for generating
instrumentally reliable theories.

4.3.3.1

Swinburnes Kinds of Simplicity

Swinburne distinguishes five different criteria of simplicity, or as he calls them

facets of simplicity. Swinburnes simplicity criteria specify numbers of things, kinds of

things, a criterion of graspability which may involve notions of epistemological simplicity
or psychological simplicity (a kind of pragmatical simplicity), numbers of independent
laws, and a composite criterion involving syntactical simplicity and either pragmatical
simplicity or epistemological simplicity. I will let Swinburnes own words show how he
distinguishes simplicity criteria.
Swinburnes first facet is a criterion selecting the numbers of things postulated.
The first facet of simplicity is just a matter of number of things postulated.
A theory which postulates one entity (or property of an entity) rather than two,
two rather than three, is (other things being equal) simpler. ([59]pg. 29)

His second facet is a criterion selecting numbers of kinds of things postulated.

227
A theory which postulates three kinds of entities (or properties of entities)
is (other things being equal) simpler than one which postulates six, and so
on. A theory which postulates three kinds of quark is simpler than one which
postulates that quarks have just certain properties, such as spin, is simpler than
one which postulates that they have these properties and also charm as well.
([59]pg. 30)

Swinburnes third facet appears to be either a kind of semantical simplicity or a

kind of psychological simplicity.
A formulation of a theory which contains a term referring to an entity or
descriptive of a property which can only be grasped by someone who grasps
some other term will be less simple than an otherwise equally simple formulation of a theory which contains the latter term instead. Thus if grue can only
be understood by someone who understands green but not conversely, then
all emeralds are green is simpler than all emeralds are grue. ([59]pg. 30)

Swinburnes fourth facet is a criterion selecting numbers of separate laws. It should

be noted that Swinburne embraces a causal theory of explanation, so this criterion will
select causal laws and will be related to kinds of causation in some way. A formulation
of theory consisting of few separate laws is (other things being equal) simpler than one
consisting of many laws. ([59]pg. 31)
Swinburnes fifth facet appears to capture some of the things discussed by Goodman about economy of form. The fifth facet involves the fewness of variables related by
individual laws and, also, what might be thought of as the defining power of these variables.
A formulation of a theory is simpler in which individual laws relate few variables rather
than many.
Two sub-facet are involved in this facet of mathematical simplicity. One
is that fewer terms in an equation make it simpler. y = z + x is simpler than
y = z + x + x2 . Secondly, other things being equal, an equation is mathematically simpler than another in so far as it uses simpler mathematical entities or

228
relations than that other. A mathematical entity or relation is simpler than
another one if can be understood by someone who does not understand ,
but cannot be understood by anyone who does not understand .([59]pg. 32)
I believe that Swinburnes argument does take the form of an indispensability argument involving necessity very much like the example that I have given above, although
he does not state it this way explicitly. In a few places Swinburne says that a person who
does not judge the a priori probability of hypotheses by their simplicity would be said to be
irrational, and I infer that what he means is that simplicity principles are indispensable for
rational thought. Two passages which support this are also useful for the critical discussion
at the end of this section. Swinburne says that,
if there are two theories which yield the observations made so far, one predicts that all life in the northern hemisphere will be destroyed tomorrow and
the other predicts that all life in the southern hemisphere will be destroyed
tomorrow, and there is no time to test further between them, but the latter is
complicated and the former is simple, any northerner would be on the aeroplane to the south tonight and think that they were highly rational to do so.
([59]pg.50)
Perhaps the most telling piece of text evidence that Swinburnes argument takes the
form of the indispensability argument with necessity is a passage where he argues against
the possibility that inductive justification can be given for simplicity principles.
All such attempts to prove from some theorem of mathematics or logic that
the simpler theory is more probably true, are I suggest, doomed to failure. The
fact however unwelcome to many is that, if the principle of simplicity is
true, it is a fundamental a priori truth. If data ever render one theory or one prediction more probable than another, that can only be because there are a priori
criteria for extrapolating from the data in one direction rather than another. Yet
there is no truth of logic with a consequence about which direction extrapolation yields probable truth. So if any proposition which is not analytic is
synthetic it is both synthetic and a priori that (other things being equal) a
simpler theory is more probably true than a complex one. If simplicity could
be justified further, it would derive that justification from some higher a priori
criterion, and that one would be fundamental. ([59] pg.55-56)

229
I conclude that Swinburnes argument is that these five criteria of simplicity are
indispensable for the rational enterprise of science.

4.3.3.2

Criticisms of Swinburnes View

My objection to Swinburnes view is that each facet depends on a ceteris paribus

clause. No simplicity principle in science can depend on this clause. The analysis of the
interrelations between the metascientific criteria show us this. There is no way to construct hypotheses that are different from one another and have them not differ along several
dimensions of the metascientific criteria. For this reason, a simplicity principle cannot
involve simplicity criteria that depend upon ceteris paribus clauses. Appeal to a ceteris
paribus clause is an attempt to give analytic treatment to simplicity without settling the
difficult matter of weighting the metascientific criteria. We cannot get around this problem
which reigns in the freedom of conceptual analysis. If any two hypotheses account for some
data equally well, but are truly different hypotheses, then they will differ conceptually. This
means that, necessarily, other things never are equal.
It is interesting that Swinburne does not appear to notice this problem. He sees that
there is a trade-off between the various simplicity criteria. He says that,
in order to compare theories, we need to compare their simplest formulations. But it is not always clear which is the simplest formulation of a theory. For it is always possible to give a formulation of a theory which makes
it simpler in one respect at the cost of loss of simplicity in another respect.
We can for example reduce many laws to few by introducing variables with
more components (scalars such as mass having only one component, vectors
such as velocity in three-dimensional space having three components, tensors
having many more), and so compressing the information into a shorter form.
Maxwell is known for having propounded four laws of electromagnetism, but
Maxwell used only vectors and scalars. Put his theory in tensor form, and you
can express it with only two laws. But the gain in simplicity in fewness of

230
laws is balanced by the loss involved in introducing variables, that is, values of
properties, more remote from observation (you cannot grasp the concept of an
electromagnetic field tensor without grasping the concepts of electric field and
magnetic field, but not vice versa), and more complex mathematical entities.
Formulations of theories may or may not become simpler when there is a loss
in respect of one facet but a gain in respect of another. ([59]pg. 34-35)

Swinburne is right that it is always possible to formulate theories that are simpler
in one respect and more complex in another, because it is always actually the case that
when competing scientific theories are offered they are simpler in some respects and more
complex in others. Swinburne is right to make a point very similar to Goodmans, which
is that it is possible to achieve spurious gains in simplicity of a particular sort and gauged
in a particular way just by defining terms a bit differently. Goodmans point, and, I take it,
Swinburnes as well, is that we seek economy rather than mere syntactical simplicity. What
deserves to be noticed is that it is not only the case that it is always possible to reformulate
a theory, or even that it is always possible to solve the ranking or gauging problems in
different ways, but that different theories are different and so will either differ along more
than one kind of simplicity measure or else they will differ with respect to the ways in
which the other metascientific criteria are satisfied. Individual hypotheses H1 and H2 may
differ by only one term. But this difference will be reflected in the semantics, in algorithmic
simplicity, in psychological simplicity, and in epistemological simplicity. Recall that there
is a trade-off between syntactical simplicity and epistemological simplicity. Also, two
competing hypotheses should not be expected to be equal in their depth, or in the ways
that they are thought to unify with the rest of our physical theories. At the very least, it is
difficult to conjure an example where two scientific hypotheses account for the same data

231
and both satisfy the other metascientific criteria equally and in the very same ways.
Recall the earlier discussion Swinburnes example that clearly does not successfully
compare two hypotheses differing in their satisfaction of one criterion of simplicity while
holding other things equal. LeVerrier could not have posited seven hundred bodies with a
common center of mass to explain the wobbly orbit of Uranus instead of just one because
this configuration of bodies would not behave in that way without either changing other
data about the orbits of Uranus and the other planets, or without depending upon some
very strange physical laws. The seven hundred body hypothesis would change either the
expected data or it would depend upon non-Newtonian laws or both. The fact that LeVerrier and Newcomb knew this is reflected in their arguments about Mercurys perihelion
advance. LeVerrier, for example, argued that the sun could not have rings that affected
Mercurys orbit in a way consistent with the data. The rings would have to be at just the
right angle of inclination to be invisible from Earth, but at that angle they would not give
Mercury an orbit consistent with the data.
Although I cannot yet accept Swinburnes argument, because it is for a principle of
parsimony with a ceteris paribus clause, a finer analysis of the interrelations between the
metascientific criteria may aid in constructing an argument along similar lines. Swinburne
says that There is this crucial a priori element affecting the probability of [a hypothesis] h
is the claim of this paper, which affirms that it is a function of simplicity and (inversely) of
content. ([59] pg.60-61)
It is true that if we could find some acceptable way of giving an empirical analysis of
both data and auxiliary hypotheses, as Swinburne suggests, perhaps we could put all of the

232
empirical data predicted by background theories into the evidence, then there is an inverse
relation between the content of a hypothesis and its fit with the data, only if we assume
that fit involves predictive power. However, as Goodman pointed out, we are not interested
in mere simplicity, but in economy. More precisely, there is an inverse relation between
the accuracy of a hypothesis and its ability to make correct predictions, and there is an
inverse relation between the syntactical simplicity of a hypothesis and linguistic exactness,
which places a necessary constraint on accuracy. We have seen that the metascientific
criteria are related in many other ways as well. Depth provides a necessary constraint on
syntactical, semantical and epistemological simplicity criteria also. Swinburne may be a
friend of the view that explanatory depth is achieved by getting ontology correct and I
would think that this might favor an argument like his because it might aid in triangulating
what are the relevant simplicity criteria in different departments. In order to pursue this
line, some very difficult work would need to be done to establish which theories and which
associated methods were paradigmatic of scientific success and then to give to these the
kind of detailed analysis that Bunge has outlined. It seems to me that if the balance between
the metascientific criteria employed in the construction of instrumentally reliable theories is
more delicate, then the indispensability argument is more insulated from challenges. This
would be the case if it could be shown that one and only one set of interrelated criteria
are involved in our instrumentally reliable methods. It would be a very difficult case to
make, and one that I do not believe has yet been made. The first move in this direction is to
review the work of historians of science to identify specific success stories in science and
then to see which hypotheses were considered by scientists and what were the arguments by

233
which each contender was ruled out. In the effort to triangulate paradigmatic instrumentally
reliable methods, I suggest that we might have better luck by looking at the cases involving
the ruling-out of hypotheses than by looking for cases where entire theories were thought
to be somehow confirmed. The reasoning in these cases is fairly clear and confirmation is
a thorny issue.
My challenge for Swinburne, and for proponents of this kind of argument, is to
show that a specific cluster of clearly defined simplicity criteria really are indispensable for
the construction and evaluation of our instrumentally reliable scientific theories. Swinburne
has made an attempt to give this argument. He says, to summarise the claims in a nutshell:
either science is irrational (in the way it judges theories and predictions probable) or the
principle of simplicity is a fundamental synthetic a priori truth. ([59] pg.61)
It may be only a minor problem to show that some specific cluster of simplicity
criteria employed in a methodology are sufficient for the construction of instrumentally
reliable theories. The real problem is showing that some cluster is necessary for that end.
Kant notices this problem when he outlines a similar style of indispensability argument for
the concept from which all moral duties are derived.
For in such a case it is easy to decide whether the action [which accords
with duty] was done out of duty or for some self-interested goal. This distinction is far more difficult to perceive when the action accords with duty but the
agent has in addition a direct inclination to do it. For example, it is certainly
in accord with duty that a shopkeeper should not overcharge an inexperienced
customer; and, where there is much business, a prudent merchant refrains from
doing this and maintains a fixed general price for everybody, so that a child can
buy from him just as well as anyone else. People thus get honest treatment. But
this is not nearly enough to justify our believing that the shopkeeper acted in
this way out of duty or from principles of honesty; his interests required him
to act as he did. We cannot assume him to have in addition a direct inclination
towards his customers, leading him, as it were out of love, to give no one pref-

234
erential treatment over another person in the matter of price. Thus the action
was done neither out of duty nor from immediate inclination, but solely out of
self-interest. (Groundwork for the Metaphysics of Morals, Chapter One)

The problem, in Kants example, is that a shopkeepers fair prices could be established either from selfish ends or from acting on a moral duty. Just by reviewing the
datum, we cannot know which is the case. There is a similar problem for an argument that
any particular cluster of properly specified metascientific criteria are indispensable for the
construction or evaluation of instrumentally reliable theories. It is possible to conceptually reduce specific principles of simplicity to other methodological principles. Since the
reduction is a conceptual reduction, we would know only something about the structural
similarities of the reducing and reduced principles, but we would not know which is more
epistemologically fundamental. Again, we can see this in the arguments given by LeVerrier and Newcomb when they evaluated and ruled out hypotheses proposed to explain
Mercurys perihelion advance. A specific hypothesis about inter-Mercurial matter (recall
that there were several of this sort) may have been ruled out because it did not satisfy one
of the following criteria: simplicity of entities, simplicity of syntax, semantical simplicity, conformity with the data, or unity with accepted scientific laws. If limited conceptual
reductions of specific methodological principles are possible (and I suggest that they are),
then the indispensability argument alone cannot determine which specific principles are
necessary for the construction or evaluation of instrumentally reliable theories.
Now, I suggested that a close look at the work done by historians of science might
aid in triangulating indispensable principles and this is where I think that we should look.
However, this challenge is more difficult than it may at first appear. The above example

235
from Kant is employed to show that the problem of induction is unavoidable so long as our
investigation involves sensations. Kant concludes that he will never find law-like connections between sensations and therefore, he must investigate the form of the concept alone
in order to discover any universal and necessary connections (which Kant says is easy). In
our present inquiry into the justification of scientific methodologies, we have an empirical
component: the instrumental reliability of theories constructed by specific methods. The
fundamental problem remains to show that synthetic a priori judgments are possible. For
this reason, I cannot yet see how to defend this kind of argument.

4.3.3.3

Piety and Swinburnes Examples

I wish to conclude this subsection by reviewing what may appear to be a challenge

to my own project. This challenge arises from the if our lives depended upon some judgment kind of case that Swinburne has given. Swinburne gives an example where a scientist
must decide whether or not to fly to the southern hemisphere on no further information than
that a hypothesis about some northern hemisphere cataclysm is simpler, in some respect,
than a competing hypothesis about the destruction of the southern hemisphere. This kind
of example appears to confront my project with a dilemma: either accept that scientists are
irrational or accept that simplicity judgments are fundamental for rational thought. I reject
the dilemma. Although I might be inclined to accept that science is not entirely rational
(but not fundamentally irrational), it is not yet clear that these are the only two options.
It is key for this example that the individual deciding to fly to the southern hemisphere is a scientist. The reason is that we wish for the individual making this judgment

236
to be someone who understands science. What is not clear is whether the principles that
govern scientific inferences collapse into the principles that govern the inferences of classical epistemology. The scientist in this example may just give up on knowing anything
on the basis of scientific principles in this kind of case and revert to some common human
judgment. If human knowledge is possible, then I suggest that scientists are not the only
people who have it. If humans have knowledge, then they have it without understanding the
specific, and sometime highly technical, esoteric principles of science. Science may be a
rational enterprise because it shares some, but not all, of the principles of fundamental rational inference. There may not be a high degree of continuity between classical epistemology
and the philosophy of science although the word epistemology is correctly applied in both
cases. It is possible that classical epistemology and the realism versus anti-realism controversy in the philosophy of science are, to some extent, dialectically independent. I do not
see why non-scientists should be thought of as employing Swinburnes criteria when they
make rational judgments, so it is not clear that a scientist who makes a rational judgment
(supposing that is what it is in this example) is employing Swinburnes criteria. Perhaps
the metaepistemological criteria and the metascientific criteria partially overlap. Swinburne
must first argue that they are identical before he can force us to embrace this dilemma.

4.3.4

Paul Churchlands Argument

Paul Churchland employs what may be a version of the second kind of indispensability argument in The Ontological Status of Observables: In Praise of the Superempirical Virtues [18]. If this is the correct way to understand Churchlands inference, then

237
it would be an inference that some concept, or set of concepts are indispensable for the
experiences that we do have, but only as a contingent matter of fact. The superempirical
virtues, Churchland argues, are some of the fundamental cognitive features had by critters
who succeed in breeding, reasoning and doing science of the sort that we are familiar with.
These map very generally onto the metascientific criteria; including, of course, simplicity.
Churchlands arguments in this article are specifically directed against the constructive empiricism of Bas van Fraassen. Constructive empiricism is committed to the following two claims:
1. The truth of a theory is measured by nothing more than descriptive excellence at the
observational level.
2. Acceptance of the theory should not involve commitment to unobservable entities.
The thrust of these two claims is that empirical adequacy is the only theoretical
virtue in science and that in accepting theories as true, we need not reify the terms postulated by the theoretical detour. Churchland argues that there are other theoretical virtues:
simplicity, coherence, and explanatory power.
Churchlands strategy is to press the Constructive Empiricists commitment to a
distinction between what is unobservable and what is unobserved. Churchlands claim is
that the distinction between observable and unobservable seems intuitively plausible only
because of the contingent state of affairs that human beings find themselves in. Evolutionary events, random and deterministic causation tell the story about how humans came to
sense only certain wave lengths of E-M radiation, specific frequencies of sound, and things
only of certain sizes or at certain distances. Were we to have our heads fitted with artificial

238
transducers, or were evolutionary forces to have favored electron microscope eyes (so the
argument goes), we would have developed a discourse about the strands of DNA which we
would see, instead of the breadbox-sized-objects-discourse which we are accustomed to.
The constructive empiricist must say that there are unobserved entities, but Churchlands
thought experiment is supposed to show that some of the entities that the constructive empiricist wants to call unobservable might just be unobserved. If Churchlands objection
is on point, then the constructive empiricist needs to make a principled distinction between
what is supposed to be unobservable and what has merely not been observed. Churchland
reasons that if a principled distinction is not defensible, then the constructive empiricist
must either give in to skepticism or else accept the fact that we accept theories on the basis
of criteria that cannot be reduced to empirical criteria.
Churchland says that
there is no way of conceiving or representing the empirical facts that is
completely independent of speculative assumptions, and since we will occasionally confront theoretical alternatives on a scale so comprehensive that we
must also choose between competing modes of conceiving what the empirical
facts before us are, then the epistemic choice between these global alternatives
cannot be made by comparing the extent to which they are adequate to some
common touchstone, the empirical facts. ([18]pg. 41)
Churchland concludes,
values such as ontological simplicity, coherence, and explanatory power
are some of the brains most basic criteria for recognizing information, for
distinguishing information from noise. ([18]pg. 42)
Added to this is a premise arising from a rich discussion in the philosophy of science: that even everyday observations are theory laden. The conclusion is that the constructive empiricist cannot make empirical adequacy the sole theoretical virtue. Humes

239
problem is a problem at the level of medium-sized dry goods as it is at any other scale.
Churchland says,
I assert that global excellence of theory is the ultimate measure of truth and
ontology at all levels of cognition, even at the observational level. Van Fraassen
asserts that descriptive excellence at the observational level is the only genuine
measure of any theorys truth and that ones acceptance of a theory should
create no ontological commitments whatever beyond the observational level.
Against van Fraassens first claim I will maintain that observational excellence or empirical adequacy is only one epistemic virtue among others of
equal or comparable importance. And against his second claim I will maintain that the ontological commitments of any theory are wholly blind to the
idiosyncratic distinction between what is and what is not humanly observable,
and so should be our own ontological commitments. ([18] pg.35)
Perhaps Churchlands argument does take the form of the indispensability argument
without the necessity claim. I will suppose, for the purpose of analysis, that this is his
argument, or at least that someone may give a very similar one intending it to be of this
form. If this is the case, then the argument would purport to show that humans have specific
concepts that we employ in solving the problem of Humean skepticism both when we
bump around the world of sensations and when we do science. If successful, this argument
would aim to put van Fraassen in a dilemma: either give a principled distinction between
unobserved and unobservable or admit that the skeptic must be given a final answer before
the realism versus anti-realism controversy in the philosophy of science can be settled.

4.3.4.1

Criticisms of Churchlands View

Although I agree with Churchland that empirical adequacy is but one of many meta-

scientific criteria, I do not share Churchlands reasons for holding that empirical adequacy
is not (cannot be) the sole criterion of theory acceptance. My reason is that empirical adequacy is a composite notion involving both the fit of a theory with the existing data and its

240
ability to make predictions. In order to gauge how well a theory satisfies the composite criterion of empirical adequacy, we need to know what the accuracy constraints are. We can
learn some important things about how accuracy constraints are established by empirical
studies, but we cannot learn everything. Empirical studies may tell us about the physiological and psychological capabilities of experimenters or about how experimenters are
influenced by social phenomena. However, we cannot give a purely empirical elucidation
of how accuracy constraints are determined when we consider how experimental equipment
is designed or how experimental controls are established. These things are determined by
appeal to theories.
For example, the famous Cavendish experiment, which is aimed at measuring the
value of the gravitational constant, depends for its construction upon theories about springs
and theories about optics13 . Additionally, this experiment is very sensitive to vibrations
and so theories about possible sources of vibration contribute to the experimental controls
placed on any particular run. The experiment may be performed on a massive spring loaded
table or run only at non-commute hours of the day to reduce the vibrations contributed by
cars driving near by. In this way, scientific experiments are said to be theory laden. This
is because the design of this experiment and the interpretation of its results depend upon
theories about the behavior of metals, about inertia, and sources of vibration theories that
are independent of the theory of gravitation. Churchland is quite right to point out that our
experiments are theory laden. It is true that scientists who perform this experiment take into
13 Because

the deflection of a set of masses suspended by a spring, in this experiment, is quite

small, the experiment typically bounces a beam of light off of a mirror and projects that beam out
many feet (or yards) so that small movements can be detected.

241
account these factors and some of them are reflected in considerations about the accuracy
of the results.
Accuracy constraints, however, are also determined by the ways in which they are
balanced with many of the other metascientific criteria. Most importantly, depth will play
a key role. All scientific experiments involve simplifying assumptions. No simplifying
assumptions will be allowed at the expense of explanatory depth. I do not believe that a
purely empirical elucidation of depth has yet been given. For these reasons, I end up with
a conclusion much like Churchlands that empirical adequacy is not the sole criterion of
theory acceptance. I do not end up with this conclusion because Churchland has convinced
me that run-of-the-mill rational thought involves precisely the same criteria that scientific
thought does.
In any case, it was not Churchlands project to justify any specific principle of
parsimony involving a specific criterion of simplicity as it is in the brain. His point is
to make trouble for a fundamental commitment of constructive empiricism. In this he
succeeds.
Not surprisingly, my objections to this kind of inference begin with the clarification of simplicity criteria. We know that there are many ways to categorize and to make
precise the fine-grained details about simplicity criteria. There are many different kinds of
simplicity criteria and these stand in unique relations with one another and with the other
metascientific criteria. No overall reduction of simplicity is possible, so humans do not
have just one simplicity concept. In some cases, specific methodological principles may
be conceptually reduced to others but successful reductions of this sort leave us, still, with

242
a radically heterogeneous bag of criteria some empirical and some non-empirical. Conceptual reductions of specific criteria leave us in the lurch about which are supposed to be
epistemologically fundamental.
Again, consider the ways in which the Kantian indispensability argument is challenged. Kant thought that space and time concepts were indispensable for having knowledge. Only, Kant was thinking that Newtonian physics was pretty well the final word on
natural philosophy. Alas, but Euclidian geometry would go the way of the dinosaur when
Einstein gave a theory about curved space-time and a star was viewed in a way that it should
not be, if Newtonian physics were true, during a solar eclipse. The problem arises when
many concepts would appear to be sufficient, but no one necessary, for our theories to be
instrumentally reliable within specific accuracy constraints.
Similarly, Nietzsche may have Kants ethical groundwork as a target in The Genealogy of Morals14 . If Nietzsche succeeds in showing that moral judgments change in such a
radically heterogeneous way that no overall conceptual reduction is possible, then he will
have shown that no single concept of morality is indispensable for human moral judgments.
This would show us that practical anthropology is really the final study in the field of ethics
and that this shows us something about society, but nothing about the structures of the hu14 Nietzsches

primary targets are Spencer and Huxley who published popular books blurting
out precisely the sort of thing that one might expect from Englishmen bobbing in the wake of the
Christian conceptual revolution that the term good was originally used to compliment unegoistic
acts, but after humans evolved, the species forgot this, but continued to use the term this way. There
are many problems with this view. It commits the genetic fallacy. It is not based on research. It
appears to depend upon a teleological view of evolution. It would be odd to report something that
the species forgot and that is complimented for being good today. Nietzsche probably suspects that
the defenders of this view got squeamish about doing any real history because they had a notion that
the Christian conceptual revolution was a bloody and ferocious one.

243
man mind. If we give up the necessity claim of the indispensability argument, we give up
the original aim of this study which is to justify the stronger-than-material-conditional relation between the truth of our theories and the instrumental reliability of the methodologies
that generated them. This is a problem. The more serious problem is that when we show
that many different principles are sufficient, but no one is necessary, for the construction
of our instrumentally reliable theories, we notice that an indispensability argument cannot
determine anything about the mind. My objection to the indispensability argument minus
the necessity claim is that it is bad cognitive science. For every reason discussed in this
subsection, Swinburnes style of argument would be the less sinful way to go.
Finally, it may not be appropriate to characterize Churchlands actual argument as
an indispensability argument at all. Alan Baker says that the modern style of indispensability argument where we have good reason to believe X because of the success of some
body of scientific work committed to X is actually an argument to the best explanation
that X can justifiably by reified. The inference to the best explanation argument strategy is
the next topic of discussion.

4.4

Inference to the Best Explanation

In this final section, I wish to investigate the possibility that the justification of
simplicity principles in science might be based upon the inference to the best explanation
(IBE). Paul R. Thagard has given a thorough defense of IBE as a form of inference in
science. Thagard also suggests that his analysis is general enough to be applied to issues in
classical epistemology as well. Thagard says that a merit of his analysis of IBE is that
it makes possible a reunification of scientific and philosophical method,

244
since inference to the best explanation has many applications in philosophy,
especially in metaphysics. Arguments concerning the best explanation are
relevant to problems concerning scientific realism, other minds, the external
world, and the existence of God. ([60] pg.92)

I present Thagards analysis and attempt to sketch how IBE might be applied to the
justification of simplicity principles in science.
I also present Richard Fumertons argument that IBE cannot be used as an alternative to induction in answering the skeptic. Fumertons arguments raise interesting issues
about the role of simplicity judgments in IBE. I might be convinced not to use IBE to answer the skeptic; however, I do not yet see why IBE suffers from any formal problems if it
is employed to justify simplicity principles in science. If the argument from IBE is circular,
it is because the principles of simplicity that govern the inference are the same as the principles appearing in the conclusion. For reasons similar to those discussed in the previous
sections, the radical heterogeneity of simplicity, perhaps, comes to the rescue when IBE is
charged with circularity. However, if IBE needs rescuing then this rescue does rely on the
notion that if simplicity principles in science were to be justified by IBE, then the skeptic
would have received a final answer by way of some other argument.
I do not yet know how the IBE justification of simplicity in science would go in all
of its details. Part of this has to do with the same issue that has previously been discussed.
It seems that a few enduring controversies in the history of science need to be settled so that
we can see which scientific theories and which associated methodologies are paradigmatic
of scientific success. What I aim to show is that if IBE is employed to solve the problem of
the justification of simplicity in science then 1) it is far from obvious that IBE suffers from

245
any formal difficulties and 2) it is not yet clear that it conceptually reduces to induction.

4.4.1

How IBE Might Go

Thagard says that, to put it briefly, inference to the best explanation consists in
accepting a hypothesis on the grounds that it provides a better explanation of the evidence
than is provided by alternative hypotheses.([60] pg.77)
In the second chapter, I suggested that IBE would be employed in an argument of
the following form,
1. Observation set O.
2. Observation set O0 .
3. Observation set O00 .
4. If h were true then it would be better than the disjunction of all of the (perhaps
infinitely many) hypotheses which give a unified explanation for the bunch of observations which would otherwise be a totally disconnected set of coincidences.
Therefore: h is probably true.
If this argument form were applied to justify simplicity in science it might go as
follows,
1. Methodology,R was employed in constructing and evaluating T1 and T1 is instrumentally reliable.
2. Methodology,R was employed in constructing and evaluating T2 and T2 is instrumentally reliable.
3. Methodology,R was employed in constructing and evaluating T3 and T3 is instru-

246
mentally reliable.
4. If the world has the features selected by the members of and those features are
related in Methodology,R so that it generates instrumentally reliable theories, then
this would be a better explanation than the disjunction of all of the other (infinitely
many) hypotheses which give a unified explanation for the fact T1 , T2 , and T3 are
instrumentally reliable, which would otherwise be a totally disconnected set of coincidences.
Therefore: Probably the world has the features selected by the members of and those features
are related in ways that Methodology,R tracks such that it generates instrumentally
reliable theories.
This formulation still does not look much different from induction due to the fact
that each of the premises are uniform cases. The proponent of induction could claim that
to get the story right about how the human mind works we should replace (4) with the
principle of the uniformity of nature. I wish to see if it is possible to avoid this stalemate
by giving IBE a unique form in this context. Perhaps the argument form should go more
like this:
1. Methodology,R1 was employed in constructing and evaluating T1 and T1 is instrumentally reliable.
2. Methodology,R2 was employed in constructing and evaluating T2 and T2 is instrumentally reliable.
3. Methodology,R3 was employed in constructing and evaluating T3 and T3 is instru-

247
mentally reliable.
4. If analysis reveals that could be a subset of , , and and R could be a subset
of R1 , R2 , and R3 , and Methodology,R could play a role in the construction of T1 ,
T2 , and T3 , then the world has the features selected by the members of and those
features are related in Methodology,R such that it generates instrumentally reliable
theories, then this would be a better explanation than the disjunction of all of the other
(infinitely many) hypotheses which give a unified explanation for the fact T1 , T2 , and
T3 are instrumentally reliable, which would otherwise be a totally disconnected set
of coincidences.
Therefore: Probably the world has the features selected by the members of and those features
are related in ways that Methodology,R tracks such that it generates instrumentally
reliable theories.
This argument form looks distinct from induction because it attempts to give a unified explanation for a heterogeneous, rather than homogeneous set of observations. But,
the argument does attempt to draw something homogeneous out of the observations: that
Methodology,R may be seen as playing a role in each case. Perhaps what is unique about
this form of argument is that those who give it would not have to wait for historians to
sort out every detail about which scientists took which principles to be epistemologically
fundamental and then show that they had a few in common. The philosopher of science
could attempt to give a conceptual reduction of any principles appearing in the scientists
methodologies to other principles, showing that these reductions do not change the interrelations between the metascientific criteria. If it could be shown that all of the distinct

248
methodologies might share a common subset of principles, then, given the complexity of
the metascientific criteria and the tightness of their interrelations, it would be a miracle
if anything else accounted for their instrumental reliability than that methodology,R was
related to the truth.
This project would involve a mountain of analytic and historical work! It seems that
identifying a common subset of metascientific criteria and interrelations amongst several
heterogeneous methodologies would be a miracle all its own. I cannot give the argument,
nor can I spell out any specific details about which sorts of principles might be involved or
how they might be related. However, I do think that this is the best case to be made that
IBE would give us a form of argument distinct from the inductive argument.
Richard Boyd argues for a version of scientific realism by employing IBE[9]. Boyds
strategy involves first, arguing that the scientific anti-realist has a fundamental epistemological commitment to IBE because science is fundamentally committed to IBE in generating
theories from data. Second, according to Boyd, that the only competing hypotheses about
the instrumental reliability of scientific methodologies are the realists hypothesis and the
hypothesis that it is a miracle (i.e. that there is no explanation other than luck) that our
instrumentally reliable methods collect instrumentally reliable theories over time. Third,
Boyds explanation for why the scientific methods employed by the mature sciences collect
instrumentally reliable theories over time is in part that the auxiliary hypotheses needed for
conducting experiments and interpreting the data are approximately true.
Boyd points out that antirealists themselves account for the projectibility of predicates and degrees of confirmation by appealing to background theories.

249
What Kuhn and other constructivists insist (correctly, I believe) is that
judgments of projectibility and degrees of confirmation are quite profoundly
dependent upon the theories that make up the existing theoretical tradition or
paradigm. ([9] pg.57)

Boyd argues that scientific judgments about the projectibility of predicates and the
degrees of confirmation conferred upon hypotheses by data depend on the inference that
background theories are approximately true. He says,
...this conception of the enterprise of science provides the only scientifically plausible explanation for the instrumental reliability of the scientific
method. In particular, I argue that the reliability of theory-dependent judgments of projectability and degrees of confirmation can only be satisfactorily
explained on the assumption that the theoretical claims embodied in the background theories which determine those judgments are relevantly approximately
true, and that scientific methodology acts dialectically so as to produce in the
long run an increasingly accurate theoretical picture of the world. ([9] pg.59)

Boyds conclusion is that the antirealist is committed to IBE in virtue of her commitment to enumerative induction and must justify limiting IBE to observables.
The rejection of abduction or inference to the best explanation would place
quite remarkable strictures on intellectual inquiry. In particular, it is by no
means clear that students of the sciences, whether philosophers or historians,
would have any methodology left if abduction were abandoned. If the fact
that a theory provides the best available explanation for some important phenomenon it is not a justification for believing that the theory is at least approximately true, then it is hard to see how intellectual inquiry could proceed.
Of course, the antirealist might accept abductive inferences whenever their
conclusions do not postulate unobservables, while rejecting such inferences to
theoretical conclusions. In this case, however, the burden of proof would no
longer lie exclusively on the realists side: the antirealist must justify the proposed limitation on an otherwise legitimate principle of inductive inference.
([9] pg.67)

I wish to make three points about Boyds argument: 1) it accommodates the fact
that science often involves the ruling-out of hypotheses, 2) perhaps the argument could be

250
employed to justify methodology,R , and 3) that simplicity criteria of some kinds may be
involved in the construction of his argument.
1) The fact that the efforts of LeVerrier and Newcomb to preserve Newtonian dynamics led to the key arguments that helped to overturn Newtonian dynamics may seem to
present the scientific realist with a puzzle how to justify the claim that scientific theories
are approximately true, or that their terms succeed in referring, or that scientific progress
builds upon earlier theories, when exemplary scientific work sometimes succeeds only in
overturning its paradigm. Boyds view would accommodate this, because his inference
would be that the theories needed to collect data about planetary motions are roughly true.
In other words, that our theories about optics are roughly true. For all I know, Boyd may be
right to claim that some scientific methods depend upon the truth of background theories.
2) What we want is some argument that methodology,R is justified. It would appear that the style of argument that Boyd gives against the anti-realist might do this. What
would explain the instrumental reliability of our methodologies is that the background
theories that some of the methods depend upon are approximately true, and also that the
methodological principles are truth conducive, in the sense of preferentially selecting theories that are true or that progressively approach the truth. This style of argument would aim
to show that the only explanation for the instrumental reliability of the theories of mature
science is that methodologies of mature science involve principles that are related to one
another in just the sort of way that is conducive to generating approximately true theories,
and it would also aim to show that those methodological principles involving criteria that
make ontological commitments are committed to the true ontology, or something close to

251
the true ontology. This argument could be extended to justify the simplicity criteria that
are constituents in the methodology of mature science. However, this argument, alone,
does not tell us which methodological principles standing in which interrelations are truth
conducive. Identifying the specific principles of mature science requires empirical investigation.
3) Various simplicity criteria may be involved in the assignment of prior probabilities to hypotheses that purport to explain the instrumental reliability of the mature sciences.
Boyd holds that there are only two hypotheses competing to explain the instrumental reliability of scientific methodology; the miracle hypothesis and the realist hypothesis. He
concludes that the miracle hypothesis does not explain the instrumental reliability of scientific methodology. However, we might ask why Boyd considers only two hypotheses,
and if our Boyd-style argument for the justification of methodological principles should do
the same. Laudan has argued that there have been many rogue theories which enjoyed
periods of instrumental success, but are now thought to be false, and to contain theoretical
terms that, by the lights of current theories, lack reference. [42] Boyd must have some
principled way to show that these sorts of hypotheses have an insignificant prior probability relative to the realist hypothesis and the miracle hypothesis. We must get an idea
about how this kind of inference goes. In order to answer this challenge so that the Boydstyle argument can be employed to justify methodology,R , we need a principled way to
distinguish rogue theories from the theories of mature science. Perhaps arguments are
available to the effect that rogue theories were not actually more instrumentally reliable
(however this is gauged) than the theories of the mature sciences (whatever those are). If

252
this cannot be cannot be done for all rogue theories, then this style of argument to the best
explanation may still involve judgments of simplicity in the assignment of the prior probabilities. If the notion of best in IBE depends upon simplicity judgments, then we must ask
two important questions. Would an argument to the best explanation that methodology,R
is truth conducive be circular? Would the justification of the simplicity criteria involved in
IBE depend upon a track-record of giving probably true conclusions?
The remaining problem is, as suggested in the second chapter, how to understand
the word best involved in IBE. This is where Thagards defense of IBE comes in handy.

4.4.2

Thagards Analysis

Thagards view is that the word better or best involved in the crucial premise
of IBE can be analyzed as a set of criteria which have to be balanced against one another.
Those criteria are: consilience, simplicity, and analogy. Thagard says that,
by criteria I do not mean necessary or sufficient conditions. We shall see
that the complexity of scientific reasoning precludes the presentation of such
conditions of the best explanation. A criterion is rather a standard of judgment
which must be weighed against other criteria used in evaluating explanatory
hypotheses. ([60] pg.79)

The criterion of analogy would be satisfied in the above example by the reductive
analysis (supposing that one could be given) that resulted in some common set of interrelated methodological principles in every case. Simplicity and consilience require a bit
more detailed discussion. Most of Thagards paper is devoted to spelling out the criterion
of consilience. He says that,
the notion of consilience is derived from the writings of William Whewell.
Consilience is intended to serve as a measure of how much a theory explains,

253
so that we can use it to tell when one theory explains more of the evidence
than another theory. Roughly, a theory is said to be consilient if it explains at
least two classes of facts. Then one theory is more consilient than another if
it explains more classes of facts than the other does. Intuitively, we show one
theory to be more consilient than another by pointing to a class or classes of
facts which it explains but which the other theory does not. ([60] pg.79)

I have tried to make an argument form involving IBE capture this feature of consilience. The way that this is supposed to happen is, first by assuming that a conceptual
reductions of several principles from several distinct methodologies will arrive at a common set of principles and interrelations, and second, that a commitment to the ontology
suggested by this methodology would explain a diverse set of facts. Thagard also says
that, in inferring the best explanation, what matters is not the sheer number of facts explained, but the variety, and variety is not a notion for which we can expect a neat formal
characterization. ([60] pg.83)
In scientific explanations, the focus of Thagards paper, we would expect the criterion of consilience to satisfy the aims of systematicity, depth, and perhaps the unity of
science. Perhaps, we would expect a criterion of consilience to do something similar in
the inference about the justification of simplicity principles in science it would contribute
to the systematicity, unity, and generality of the hypothesis aimed at explaining the instrumental reliability of a methodology. I have suggested that scientific theories succeed in
satisfying the metascientific criteria by doing more than we can account for by a mere fit
with the data and by the systematicity to which syntactical simplicity contributes. This is
due to the fact that we do not know everything that there is to know about accuracy by appeal to empirical and linguistic criteria alone we must balance these with depth. Thagard

254
says something similar.
A consilient theory unifies and systematizes. To say that a theory is consilient is to say more than that it fits the facts: it is to say first that the theory
explains the facts, and second that the facts it explains are taken from more
than one domain. These two features differentiate consilience from a number of other notions which have been called explanatory power, systematic
power, systematicization, or unification. For example, Carl Hempel has
given a definition of systematic power which is purely syntactic, and hence
much more exact than the above definition of consilience. However, it is not
applicable to the sort of historical examples I have been considering, since
it concerns only the derivation of sentences formed by negation, disjunction,
and conjunction from atomic sentences Pa; it therefore does not represent
the way in which Huygens, Lavoisier, and Darwin systematize by explaining a
variety of facts, including those expressed by laws. ([60] pg.82)
I have attempted to formulate this version of IBE for the justification of simplicity
in science guarding against IBE collapsing into an indispensability argument. This would
happen if the crucial premise said something like,
4 If analysis reveals that is the only subset of , , and and R is the only subset
of R1 , R2 , and R3 , and Methodology,R plays an essential role in the construction
of T1 , T2 , and T3 , then the world has the features selected by the members of and
those features are related in ways such that Methodology,R generates instrumentally
reliable theories, because there are no other alternatives and otherwise T1 , T2 , and T3 ,
would be a totally disconnected set of coincidences.
Additionally, this form of argument might be pliable enough to accommodate the
fact that scientists may have in common a few mistaken principles. Analogously with
scientific explanation,
According to Wesley Salmon, variety of instances is important in that it
helps us to eliminate alternative hypotheses; according to Clark Glymour, variety is needed in order to compensate for cases where errors in one or more
hypotheses, or in evidence, may cancel each other out. ([60] pg.85)

255
Finally, we wish to guard against the objection that the maximally consilient explanation would explain any fact whatsoever. Thagard suggests that the solution to this
problem involves balancing consilience with other criteria.
The limit to these adjustments depends on the increase in consilience of
the theory being offset by a decrease in satisfaction of other criteria, such as
precision and simplicity. ([60] pg.85)
Thagards view is that simplicity places a constraint on consilience. His simplicity
criterion selects numbers of ad hoc hypotheses. Thagard says,
a simple consilient theory not only must explain a range of facts; it must
explain those facts without making a host of assumptions with narrow application.
An ad hoc hypothesis is one that serves to explain no more phenomena than
the narrow range it was introduced to explain. Hence a simple theory is one
with few ad hoc hypotheses.([60] pg.87)
This notion is included in the IBE argument for the justification of simplicity.
The idea is that we need posit nothing more to explain the instrumental reliability of
methodology,R than that it is committed to the correct ontology.

4.4.3

Circularity and IBE

For the very same reasons discussed in the section on induction, it is far from obvious that IBE employed to justify simplicity in science is committed to circularity just because it involves a criterion of simplicity. Thagards IBE in scientific explanation involves
a criterion of simplicity with respect to ad hoc hypotheses. He is probably correct to assert
that scientists embrace such a criterion. If IBE had a similar form for the justification of
simplicity principles in science, it might also involve a criterion of simplicity with respect
to ad hoc hypotheses that would place a constraint on consilience. It is not obvious that the

256
criterion is the same in both cases. Scientific explanations are often causal explanations.
The explanation of the instrumental reliability of a methodology is not obviously a causal
explanation. Therefore, IBE in scientific explanation could involve a simplicity criterion
selecting numbers of ad hoc scientific hypotheses and IBE employed to justify simplicity
in science could select numbers of ad hoc metaphysical hypotheses. So long as IBE uses
different criteria, the general argument form can be employed to justify the criteria used by
that argument form in another field of discourse. Again, the final justification of simplicity
in science might wait for the skeptic to be given a final answer. Another way of putting this
is, dont wish for science to solve the problems of ontology.
Does that mean that there is really no point in the epistemological study of simplicity in science? After all, everyone is waiting for the skeptic to receive her final answer. I
do not see why this is the case. If we learned something about how inferential criteria are
balanced by investigating the philosophy of science, then we learned something in general.
Perhaps, this lesson was not available by consulting mere imagination alone. If we learned
something about how the classical problems of epistemology are different from the problems in the philosophy of science, then we learned something by way of contrast. Anyway,
for all I know, an anti-ad hoc-hypothesis criterion of metaphysics might be justified by
some transcendental argument, not by IBE. In that case, we really would need two different analyses: one to show how the transcendental argument answers the Humean skeptic
and the other to show how simplicity criteria in science are justified.

257
4.4.4

Fumertons Challenge

Fumerton has argued that IBE will not serve as a viable alternative to induction
in answering the problem of skepticism.[25] The reason is that the inferential criteria, in
particular, simplicity would have to be justified themselves. It is not clear what could
justify constraining consilience with an anti-ad hoc-simplicity criterion except that past
experience showed us that this method uniformly generated true hypotheses. Appeal to
uniform past experiences is an appeal to induction. Probably, one of the inferences, IBE,
induction, or indispensability is fundamental if the skeptic is to receive an answer, but
only because an infinite regress would have to be avoided. However, in the philosophy
of simplicity in science, I do not see that we have to take any of these argument forms
to be fundamental. None appear to share sufficient structural similarities for conceptual
reductions to be available.

4.4.4.1

Structural Similarities with Metaepistemology

Our present inquiry shares a few structural similarities with the internalism/externalism

controversy in metaepistemology. Fumerton characterizes the form of the controversy in

the following way:
Inferential internalism maintains that someone S justifiably infers one proposition P from another E only if 1) S is justified in believing that E confirms or
makes probable P and 2) S is justified in believing E. The view is, of course,
highly controversial and all well-known versions of externalism deny at least
condition 1). ([25] pg.152)
Fumerton says that externalists will have no need of an argument to the best explanation since they have rejected 1), which is the only claim that would require an argument
of this form for its justification. In our present investigation, we have slightly different

258
claims. We make the very plausible assumption that we are justified in believing 2s) that
some of our scientific theories are instrumentally reliable and then we wish to see if there
are any arguments available to show that 1s) something about our instrumentally reliable
methodologies makes probable the production or preservation of instrumentally reliable
theories. I have shown that it is, at least, very difficult to reject 2s) and that the usual claims
aimed at rejecting 2s) are either false or too vague to get the job done. For this reason
we are left with only a few candidate argument forms: (I) induction with the assumption
that problems in the philosophy of science are independent of problems in classical epistemology, (TD) a transcendental deduction supposing that i) synthetic a priori knowledge
is possible and that ii) paradigm examples of instrumentally reliable methodologies might
be given and analyzed in such a way that a set of methodological principles worthy of an
indispensability argument could tell us something about the mind, and (IBE) inference to
the best explanation.
Fumerton says that
I suggested that one who accepts externalism would probably feel no particular need to try to subsume various fundamental belief-forming processes
under some more general pattern of argument like inductive reasoning or reasoning to the best explanation. On the other hand, I should hasten to emphasize
that if externalism were true, the philosopher qua philosopher should probably
have no very strong opinions on what processes do and do not generate justified
beliefs. ([25] pg.155)

Similarly, I say that if simplicity judgments are reducible to psychological states or

aesthetic values, then philosophers should have no strong opinions about the methods that
bring about instrumentally reliable theories. However, I have argued that an overall reduction of simplicity criteria is impossible. So, the problem of the justification of scientific

259
methodological principles is motivated, and we have only a few strategies for dealing with
this problem.
I made note of the fact that we would need to add some special premises to the
argument involving IBE in order for it to look distinct from the inductive argument. When
we do this, we introduce a bunch of inferential criteria and these must be justified somehow.
Fumerton says a similar thing.
If it were true in general that plausible reasoning to the best explanation is
enthymematic inductive reasoning, it trivially follows that reasoning to the best
explanation cannot serve as an alternative to inductive reasoning in bridging
gaps between the available evidence and our commonsense beliefs.
One can, of course, reject the suggestion that reasoning to the best explanation typically collapses into inductive reasoning by adding more to the
premises of arguments to the best explanation. Thus instead of a conditional
simply asserting that our observations would be explained by E, we could have
a conditional saying that the observations would be explained best by E where
our criteria for comparing explanations do not rely on any inductively supported connections.([25] pg.159)
Fumertons concern in this paper is in the possibility of employing IBE to solve the
problem of skepticism. He concludes that this is not possible because the justification of
the inferential criteria will depend upon induction. Fumerton says that,
If we rely on any information about the past at all in justifying our belief that most events have causes, we will need an independent solution to the
problem of justifying beliefs about the past, and thus if the preceding argument is correct we will have foreclosed the possibility of using reasoning to
the best explanation in an attempt to justify beliefs about the past. If in reaching the conclusion that most events have causes we rely on the fact that our
sensations have causes, we will again need a prior solution to skepticism about
the physical world and we will be precluded from using reasoning to the best
explanation in order to get that solution.([25] pg.162-163)
Fumerton may have successfully intimidated me into avoiding IBE to answer the
skeptic, but, in all fairness, the skeptic had me intimidated already. The extreme hetero-

260
geneity of simplicity suggests that it is not obvious that the criteria of IBE employed in
some case would be justified by induction rather than IBE with a slightly different set of
criteria. There is no circularity here. Justificatory arguments stacked in this kind of way
either terminate with an answer to the skeptic or go on infinitely (if there are infinitely
many simplicity criteria that would actually constrain consilience in the appropriate ways).
If these arguments terminate, then I agree that either IBE, induction, or indispensability is
fundamental. If justificatory arguments regress, it is not clear that this is a vicious regress.
Perhaps avoiding a vicious regress here depends upon a coherence theory of justification.

4.4.5

The Independence of IBE?

Finally, I wish to show why I am not yet convinced that IBE can be conceptually
reduced to some other argument form. However, I hasten to add that I could easily be
convinced that IBE does collapse to some other inference, were a few things to fall into
place. In my set up of the argument involving IBE, I showed how its form should be
distinguished from an indispensability argument. If philosophers could show that some
methodology,R was the one and only methodology possibly had in common by several
independent paradigmatic cases of successful theory construction and evaluation, then IBE
would collapse into indispensability. The problems for this are paramount, including sorting out paradigmatic cases and what successful amounts to and giving the relevant
metascientific analyses.
I am also not yet convinced that IBE conceptually reduces to induction. Fumerton
says,
I dont think there is a legitimate form of reasoning to the best explanation

261
that is distinct from inductive reasoning and thus I dont think reasoning to the
best explanation can circumvent the traditional problems involved with finding
an inductive bridge between available evidence and commonsense beliefs.([25]
pg.169)

I agree with Fumerton that it is difficult to see how IBE could circumvent the traditional problems of bridging the gap between evidence and common sense beliefs. I do
not see that this is because IBE is not distinct from induction. Perhaps this can be seen by
noticing that the problems of induction are a bit different from the problems of IBE.
In the first chapter I noted that if it turned out that we had some idea of the necessary
connection between one cause-effect pair we could reason from cause to effect in that
kind of case. However, we still would be unable to reason from effect to cause. It is
not clear that IBE and induction share sufficient structural similarities that one could be
conceptually reduced to the other because we criticize induction and abduction differently.
Both the problems of underdetermination by data and the problem of circularity apply
to reasoning from cause to effect. But reasoning from effect to cause is crippled by the
underdetermination of theory by data alone. In causal reasoning, there is an asymmetry
between the problems of reasoning from cause to effect and reasoning from effect to cause.
This is because the uniformity of nature principle does not sort out which things are of
the stay-the-same sort. IBE includes criteria that help to do this job. Another way to put
these criticisms would be that the inductivist must figure out how to distinguish accidental
generalizations from laws of nature, but the proponent of IBE must give an account of what
the best explanation is.
Both forms of inference face the problem of showing that what the stronger than

262
material conditional relation is between antecedent and consequent. Induction alone just
cannot show us this because we would have to rely upon past experience to establish which
things stayed the same and which change, but experience is a mix of change and sameness.
The argument involving IBE posits some necessary connection or some makes probable
connection between antecedent and consequent, and it appears to have criteria appropriate
for governing this inference. The problem with IBE is that it, alone, cannot tell us precisely
what this relation is.
What would it take to talk me into accepting that IBE reduces to induction? I would
have to be shown that the crucial premise of IBE could be replaced with the crucial premise
of induction and that the criticisms of IBE remain the same. I think that the question of
interest is not whether IBE conceptually reduces to induction (or vise versa), but whether or
not the justification of IBEs inferential criteria must depend upon induction, not upon IBE
using different criteria or upon indispensability. What about the possibility of an infinite
regress? I, for one, am not all that concerned. I was suspicious about simplicity judgments
in science to begin with. I could easily come around to the view that science is not entirely
rational. Unlike some people who are incensed by the suggestion that science is like art
in some ways. I am not. Both involve some rationality (because they can be done in very
stupid ways). However, it is not yet clear that each is entirely rational. I do not know why
this is a bad thing, nor why it would involve an impiety to think of the world as being closer
to chaos than to simplicity. I could easily be convinced that the terminus of the justificatory
arguments for simplicity in science is in the answer to the skeptic or in some non-rational
judgments. Again, I say, do not simply assert one or the other. Marshall your arguments.

263
CHAPTER 5
CONCLUSIONS

My conclusions are that induction, indispensability and IBE are the only remaining contenders for arguments meant to justify simplicity principles in science. So what?
Anyone familiar with classical epistemology could have told us that the dialectic long ago
revealed this three-way stand-off in the fight for justificatory supremacy. Hold on, I say.
Is it right to say that the preceding amounted to common philosophical knowledge? The
introduction laid out some very ambitious objectives which I believe have been met.
First I motivate giving analytic treatment to simplicity. It is tempting to lunge at the
questions of justification and thereby miss the valuable and difficult project of clarification.
Typically, lunges at justification arise in semi-casual philosophical conversations. Some
people think that there are no problems of simplicity in philosophy because they have
allowed themselves to slip into thinking that simplicity judgments reflect subjective values
and that there is nothing much more to say on the matter. Others appear to think that the
frequency of the use of terms like Ockhams Razor by legitimate authorities (like Smart)
indicates that some experts must know what is going on with simplicity. The appeal to
subjective values and the appeal to authority both miss the crucial first step. We must be
clear about what we are refuting or defending. Anyway, some philosophers actually do put
into print claims intended to slip by the questions of the justification of simplicity. Kuhns
claim that how much weight scientists give a particular theoretical virtue is a matter of taste
cannot be quite right. It could be right to say that how much weight a particular scientist
gives to a theoretical virtue on balance with the other virtues is a matter of pragmatic and

264
aesthetic judgments. My inclination towards this view will increase as problems with the
inductive, indispensability and IBE approaches are revealed. Other authors have said things
that require a bit of work to clarify. Smart appears to be one of these. It would be nice to
find out what Smart thinks that Ockhams Razor is - a principle that scientists do not violate
or a useful heuristic. In any case, I believe that I have shown that arguments are required
either to reject principles of parsimony or to justify them. We wish to avoid equivocation,
so we must begin by analyzing simplicity judgments.
A study of history can supply the basic ideas about the roles played by simplicity
judgments generally and in science. Nietzsche had a useful insight. He noticed that the
archetypes for the problems of philosophy are to be found in Platos mixed view. Nietzsche
provided a template for understanding many contemporary issues including the problems
of skepticism, and perhaps Nietzsches template also provides a guide by which various
realist and anti-realist positions may be categorized.
By way of example, consider once again that Kuhn was critical of the notion that
what was distinctive of science was its ability to bring about widespread agreement between its skilled practitioners. He was also critical of the view that scientific progress
could be accounted for by showing that older scientific laws could be deduced from newer
laws. There are probably many things wrong with these views. However, Kuhns unique
argument is that these views are incompatible with a very important feature of science - the
revolution. Applying Nietzsches insight we see that Kuhns attempt to elucidate science
aimed to balance change and sameness. Kuhns view about sameness is that somehow the
terms and methods of science are established by paradigm theories for periods of time that

265
involve normal science - the working out of the details of paradigm theories. Eventually,
changes in experimental equipment and the increasing complexity of theories will lead to a
crisis - paradigm theories will no longer explain or predict the relevant data. When a crisis
arises, the stage is set for an overturning of the old paradigm. Applying Nietzsches insight
again, Larry Laudan argues that change in science is more fluid that Kuhns view allows
for. He says that his argument is thoroughly Heraclitean (Science and Values pp. 64).
Second, I believe that the conceptual map of the metascientific criteria is valuable
in many ways. Although the arguments about the interrelations between the metascientific
criteria are not my own (they are mainly Bunge and Goodmans arguments), the development of the conceptual map may be a small contribution to the dialectic. The graphic
rendering allows us to have a holistic picture of the interrelations between the metascientific
criteria with only a glance. This may be a useful aid for both scientists and philosophers
who do not specialize in the metascientific analysis as Bunge and Goodman did. Also, this
particular format is chosen so that it will contribute to the philosophical dialectic. Although
I have said quite a lot about the interrelations between the metascientific criteria, not nearly
enough has yet been said. The open-source LaTeX program may provide a key foundation
for future criticism and contributions. Philosophers who wish to criticize the arguments
given by Bunge or who wish to contribute arguments about some of the other relations
need only to locate the relevant sections of the LaTeX code and make the adjustments that
they argue for. We could do the usual thing in philosophy and present a whole bunch of
principles with associated arguments. After all, the conceptual map just shows us a bunch
of logical relations. The advantage of the conceptual map is that both its rendering and its

266
code are compact.
Third, I believe that the arguments about Ockhams Razor are decisive. This
term is subject both to vagueness and to historical confusions. It is not that philosophers
have nothing to say about simplicity or about principles of parsimony. It is quite clear that
many kinds of simplicity play roles in science. Some of those roles may be in principles
of parsimony. I offer a revisionary thesis. We should not use the words Ockhams Razor
or Occams Razor ever again. In every case, we should formulate the principle and
specify the criteria that govern the judgment, then discuss only the principle that we mean
to discuss. Revise Ockhams Razor, I say, for clarity and posterity.
There is an even more devastating result for Ockhams Razor. The popular addition of the ceteris paribus clause to principles of parsimony cannot be justified in the
philosophy of science. Scientific theories must satisfy many different criteria. The metascientific analysis shows us that there is always a trade-off between some kinds of simplicity and others. When we construct competing scientific theories we cannot hold all other
parameters constant while altering just one. Even if we did have two different theories that
accounted for the same data, and even if they did give the same predictions (which would
be rare if even possible), they would not differ by only one kind of simplicity. Suppose that
two hypotheses differed in the kinds of causal mechanisms that they posited. This would be
reflected in the syntax of the theories and in their semantics as well. If it were possible to
give a fully empirical analysis of the other metascientific criteria, then the two theories, so
long as they really are different would no longer account for the same data, and hence, they
would not really be competitors. In any case, I am doubtful that a fully empirical analysis

267
of the metascientific criteria is possible. The sign suggesting that this path is a dead end
reads, Hempels Theoreticians Dilemma - Turn Back Now.
Fourth, this project opens the door for future research in the philosophy of science.
I would like to apply these results to a critique of confirmation theory. Our brief historical account of the arguments given by LeVerrier and Newcomb show us that simplicity
judgments of various sorts may be involved in the construction of hypotheses which are
ruled out for various reasons. Now, some of these simplicity criteria are justified on pragmatic grounds, like those involved in the methods of discrete integration. It is likely that
these scientists aimed primarily for another kind of simplicity - a simplicity with respect
to concepts. It is plausible that LeVerrier and Newcomb valued a kind of unity of science
principle. One version of a unity of science principle is that it aims to unify the encyclopedia of human knowledge. Why would we ask what justified their judgments when we now
believe that they were endeavoring to preserve a theory that has been replaced? The reason is historical. When General Relativity was published, scientists were prevented from
viewing the first possible solar eclipse to see if the light from distant stars was bent near
the sun. They were prevented from performing this experiment because of World War I.
Einsteins General Relativity was not without competitors. Dicke and Whitehead also gave
general theories of gravitation. Einsteins GR was accepted because it could account for
Mercurys perihelion advance. The only reason that scientists had for taking this to be a
virtue of the theory was that LeVerrier and Newcomb had argued that Mercurys perihelion
advance was a genuine anomaly for Newtonian theory. This, I should think, would lead us
to ask if the methods employed by LeVerrier and Newcomb were justified. This suggests

268
that scientific confirmation and disconfirmation are not as cleanly based on evidence as
some authors have suggested (see Achinstein, The Book of Evidence 2001, for a detailed
discussion).
Finally, the metascientific analysis lays the groundwork for epistemological studies
and perhaps cues metaepistemological inquiry. It is true that one result is not surprising.
Induction, indispensability and IBE are in a stand-off when it comes to the justification of
simplicity. The arguments of the third chapter contribute one small thing to this discussion.
They show how, and with what care, these arguments must be constructed by those who
wish to defend them.
The more subtle insight that we might gain from the discussion of the justification
of simplicity in science has to do with the order of intellectual divisions of labor. I do
not yet have an idea about how cleanly the problems of classical epistemology map onto
the specific problems of the philosophy of science. It is at least worth asking a question
about how much continuity there is between classical epistemology and the philosophy of
science. We might put this question another way. How many of the metascientific criteria
are also metaepistemological criteria? As a first guess, I would say, not many. It seems
unlikely that epistemologists actually aim for syntactical simplicity in their theorizing. I
seems unlikely that the kind of generality that we would expect from epistemological theories would be gauged in the same way that we would gauge the generality of scientific
theories. Epistemologists have been known to make an appeal to causation in their arguments. The positing of causal mechanisms is one way to systematize a theory and achieve
explanatory depth - if, of course, the posits are true. However, the epistemologist should be

269
wary about borrowing causal mechanisms from science to construct her theories. The final
justification for the judgments of simplicity that lead scientists to posit the causal mechanisms that they posit may just depend upon answers to questions in classical epistemology.
At least there is good reason to be suspicious about the degree of continuity between
classical epistemology and the philosophy of science. Aristotle did make some important
changes to Platos mixed view. Ever since, the project of science has been committed to a
delicate balancing act involving constructing theories about the laws that govern how some
physical things change while others remain the same. Aristotles notions of causation must
have been different from Platos. We should ask if the notions of causation in science are
still different from those in classical epistemology. I take the most interesting result of the
analysis of simplicity in science to be the inquiry that it suggests. This new investigation
would be into the degrees of continuity between classical epistemology and the philosophy
of science. This investigation makes it look plausible that the ways that meta-criteria are
specified and balanced against one another distinguish activities of different sorts and reveal
how different activities are related to one another.

270
APPENDIX A
SHERLOCK HOLMES AND THE SIGN OF THE FOUR

[Watson is the first-person voice, and he addresses Holmes in this spot to put his
skills to the test, right after Holmes has performed one of his deductions.] In this case
it certainly is so I replied after a little thought. The thing, however, is, as you say, of the
simplest. Would you think me impertinent if I were to put your theories to a more severe
test?
On the contrary, he answered, it would prevent me from taking a second dose of
cocaine. I should be delighted to look into any problem which you might submit to me.
I have heard you say it is difficult for a man to have any object in daily use without
leaving the impress of his individuality upon it in such a way that a trained observer might
read it. Now, I have here a watch which has recently come into my possession. Would you
have the kindness to let me have an opinion upon the character or habits of the late owner?
I handed him over the watch with some slight feeling of amusement in my heart, for
the test was, as I thought, an impossible one, and I intended it as a lesson against the somewhat dogmatic tone which he occasionally assumed. He balanced the watch in his hand,
gazed hard at the dial, opened the back, and examined the works, first with his naked eyes
and then with a powerful convex lens. I could hardly keep from smiling at his crestfallen
face when he finally snapped the case to and handed it back.
There are hardly any data, he remarked. The watch has been recently cleaned,
which robs me of my most suggestive facts.
You are right, I answered. It was cleaned before being sent to me.

271
In my heart I accused my companion of putting forward a most lame and impotent
excuse to cover his failure. What data could he expect from an uncleaned watch?
Though unsatisfactory, my research has not been entirely barren, he observed,
staring up at the ceiling with dreamy, lack-lustre eyes. Subject to your correction, I should
judge that the watch belonged to your elder brother, who inherited it from your father.
That you gather, no doubt, from the H. W. upon the back?
Quite so. The W. suggests your own name. The date of the watch is nearly fifty
years back, and the initials are as old as the watch: so it was made for the last generation.
Jewellery usually descends to the eldest son, and he is most likely to have the same name
as the father. Your father has, if I remember right, been dead many years. It has, therefore,
been in the hands of your eldest brother.
Right, so far, said I. Anything else?
He was a man of untidy habits very untidy and careless. He was left with good
prospects, but he threw away his chances, lived for some time in poverty with occasional
short intervals of prosperity, and finally, taking to drink, he died. That is all I can gather.
I sprang from my chair and limped impatiently about the room with considerable
bitterness in my heart.
This is unworthy of you, Holmes, I said. I could not have believed that you
would have descended to this. You have made inquiries into the history of my unhappy
brother, and you now pretend to deduce this knowledge in some fanciful way. You cannot
expect me to believe that you have read all this from his old watch! It is unkind and, to
speak plainly, has a touch of charlatanism in it.

272
My dear doctor, said he kindly, pray accept my apologies. Viewing the matter
as an abstract problem, I had forgotten how personal and painful a thing it might be to you.
I assure you, however, that I never even knew that you had a brother until you handed me
the watch.
Then how in the name of all that is wonderful did you get these facts? They are
absolutely correct in every particular.
Ah, that is good luck. I could only say what was the balance of probability. I did
not at all expect to be so accurate.
But it was not mere guesswork?
No, no: I never guess. It is a shocking habit destructive to the logical faculty.
What seems strange to you is only so because you do not follow my train of thought or
observe the small facts upon which large inferences may depend. For example, I began by
stating that your brother was careless. When you observe the lower part of that watch-case
you notice that it is not only dinted in two places but it is cut and marked all over from the
habit of keeping other hard objects, such as coins or keys, in the same pocket. Surely it is
no great feat to assume that a man who treats a fifty-guinea watch so cavalierly must be a
careless man. Neither is it a very far-fetched inference that a man who inherits one article
of such value is pretty well provided for in other respects.
I nodded to show that I followed his reasoning.
It is very customary for pawnbrokers in England, when they take a watch, to
scratch the numbers of the ticket with a pin- point upon the inside of the case. It is more
handy than a label as there is no risk of the number being lost or transposed. There are no

273
less than four such numbers visible to my lens on the inside of this case. Inference that
your brother was often at low water. Secondary inference that he had occasional bursts
of prosperity, or he could not have redeemed the pledge. Finally, I ask you to look at the
inner plate, which contains the keyhole. Look at the thousands of scratches all round the
hole marks where the key has slipped. What sober mans key could have scored those
grooves? But you will never see a drunkards watch without them. He winds it at night,
and he leaves these traces of his unsteady hand. Where is the mystery in all this?
It is as clear as daylight, I answered. I regret the injustice which I did you. I
should have had more faith in your marvellous faculty. May I ask whether you have any
professional inquiry on foot at present?
None. Hence the cocaine. I cannot live without brainwork. What else is there to
live for? Stand at the window here. Was ever such a dreary, dismal, unprofitable world? See
how the yellow fog swirls down the street and drifts across the dun- coloured houses. What
could be more hopelessly prosaic and material? What is the use of having powers, Doctor,
when one has no field upon which to exert them? Crime is commonplace, existence is
commonplace, and no qualities save those which are commonplace have any function upon
earth.

274
APPENDIX B
SIMPLICITY AND PARSIMONY

I suggest that philosophers attempt to reserve distinct uses for the words simplicity and parsimony. Alan Baker points out that the words are often used interchangeably1 .
It is true that the words get used interchangeably by news reporters, magazine writers, writers of fiction and many others who employ the word in colorful ways. But as philosophers
we may wish to be more careful. I have three reasons for using the words as I do.
1. The definitions of the two words are different in ways which suggest that they refer
to different kinds of things.
2. There are two different things which philosophers might like to refer to, one is a
methodological principle and the other is a state of affairs or attribute.
3. Some scientists appear to keep the words distinct in the very way that I suggest we
do.
The Oxford English Dictionary gives the definition of simplicity

Definition B.1. The state or quality of being simple in form, structure, etc.;
absence of compositeness, complexity, or intricacy.

and of parsimony
Definition B.2. Economy in the use of assumptions in reasoning or explaining; esp. in law of parsimony n. (also principle of parsimony) the prin1 [5]

275
ciple that no more entities, causes, or forces than necessary should be invoked
in explaining a set of facts or observations (cf. Ockhams razor n.).
Both are nouns and both have adjective forms like simplex and parsimonious. However
simplicity is a state of affairs or an attribute and parsimony is an economy in use.
These appear to be words which refer to different types of things. I suggest that we use
simplicity in reference to states of affairs or attributes and that we use parsimony in
reference to a methodological principle. These are two different things which philosophers
wish to discuss and these uses are consistent with the definitions.
I am no lexicographer and probably we would do well to consult one on why the
definitions of parsimony and simplicity are different in the ways that they are. I do
not know what sort of arguments might be marshalled for the view that the two words actually mean the same thing. For all I know, the word parsimony may be like the word
podium which has morphed due to pervasive misuse. I consulted the Corpus of Contemporary American English and found that the uses of parsimony as a synonym for
simplicity in spoken word, fiction, magazine articles, newspapers and academic journals was considerable. However, one group uses the words in consistently distinct ways:
evolutionary biologists.
In the 2008 Bioscience article The Green Algal Underground: Evolutionary Secrets of Desert Cells Cardon, Gray, and Lewis use parsimony to refer to a methodological principle,
Two methods were used to estimate the number of independent transitions
to the desert habitat in green algae: with phylogenetic trees obtained from
Bayesian phylogenetic analyses, parsimony reconstruction (under optimizations favoring either reversals or parallel changes) and Bayesian mapping led

276
to a conservative estimate of 14 to 17 independent transitions from aquatic
ancestors to the desert habitat.

David Irwin uses the word in the same way in the 2006 Bioscience article Evolution of Hormone Function: Proglucagon-derived Peptides and Their Receptors
Phylogenies were generated by both parsimony and distance methods. For
parsimony methods, the preferred phylogenetic tree would be one that minimizes the number of mutational steps necessary to account for the diversity of
amino acid sequences.
Elliott Sober works on the philosophy of evolutionary biology, and he uses the
words this way in Reconstructing the Past, Parsimony, Evolution, and Inference [54]. Elliott Sober is one of the worlds leading experts on simplicity and parsimony. I suggest that
we follow these careful scientists and philosophers in the use of these terms.

277
APPENDIX C
METASCIENTIFIC CRITERIA CONCEPTUAL MAP

%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Author

: Dan Schulz (Jan. 2011)

%
% Title

: Map of the metascientific criteria

% and their interrelations

%

Affiliated with the University of Iowa 2011

% Notes

:In LaTeX 2e with tikz graphics package.

%
% Tags
%

: simplicity, metascientific criteria, mindmap,

layers,

\documentclass{article}
\usepackage{tikz,times}
\usetikzlibrary{arrows,positioning}
\usetikzlibrary{mindmap,backgrounds}

\begin{document}
\tikzstyle{root concept}+=[concept color=gray!80, text=black]

278
\tikzstyle{level 1 concept}+=[set style={{every child}=[
concept color=blue!50, inner sep=8pt, minimum size=9mm]}]
\begin{figure}[h]

\centering\scalebox{.5}{
\begin{tikzpicture}[mindmap,level 1 concept/.append
style={level distance=130,sibling angle=40},
extra concept/.append style={color=blue!50,text=black}]
\begin{scope}[mindmap]
\node [concept] at (0,0) {\small Simplicity}
[clockwise from=-30]
child [concept]{node [concept] (prag)
{\footnotesize Pragmatical}}
child {node [concept] (epi)
{\footnotesize Epistemological}}
child {node [concept] (sem) {\footnotesize Semantical}}
child {node [concept] (syn) {\footnotesize Syntactical}}
child {node [concept] (ont) {\footnotesize Ontological}};
\end{scope}

% Accuracy
\begin{scope}[mindmap]

279
\node [concept, concept color=yellow, text=black](acc)
at (-7.2,-4)
{\small Accuracy};
\end{scope}

% Linguistic Exactness
\node [concept, concept color=orange, text=white](lin)
at (-8.6,-9)
{\small Linguistic Exactness};

% Testability

\begin{scope}[mindmap, concept color=red, text=white]

\node [concept](tes) at (-8.3,-14) {\small Testability}
child [grow=35]
{node [concept] (scr) {\footnotesize Scrutability}}
child [grow=-5]
{node [concept] (fal) {\footnotesize Falsifiability}}
child [grow=-222]
{node [concept] (fit)
{\footnotesize Predictive Power/Fit}}

280
;
\end{scope}

% Sytematicity

\begin{scope}[mindmap, concept color=violet,text=white]

\node [concept](sys) at (-7.2,-20) {\small Systematicity};
\end{scope}

Scientific Unity

\begin{scope}[mindmap, concept color=blue!50!black,text=white]

\node [concept, text=white](uni) at (-2,-8.5)
{\small Unity of Science};
\end{scope}

Conceptual Connectedness

\begin{scope}[mindmap,
concept color=blue!80!black,text=white]
\node [concept, text=white](ccd) at (-2.,-18)
{\small Conceptual Connectedness};
\end{scope}

281
% Representativness
\begin{scope}
\node [concept, text=white](rep) at (2.4,-15)
{\small Representativness};
\end{scope}

% Depth

\begin{scope}[mindmap,
concept color=green!50!black,text=white]
\node [concept](dep) at (4,-9.5) {\small Depth};
\end{scope}

% Relations between metascientific criteria

\begin{pgfonlayer}{background}
\draw [>=stealth, ->, line width=4pt, gray]
(ccd) edge (sys)
(syn) edge (sys)
(rep) edge[bend right=28] (sys)
(rep) edge (dep)
(dep) edge (uni)

282
(rep) edge (uni)
(uni) edge (ccd)
;
\draw [>=stealth, ->,
line width=4pt, dashed, gray]
(ont) edge[bend right=60] (syn.west);

\draw [>=stealth, <->, thick,

shorten <=2pt, shorten >=2pt]
(lin) edge[bend left=45] (syn.west)
(fit) edge[bend left=45] (acc.west)
(rep) edge (ont)
(epi) edge (syn)
(epi.north) edge[bend right=45] (sem.north)
(epi.east) edge[bend left=45] (dep.east)
(scr) edge[bend right=35] (prag)
;
\draw [>=stealth, <->, thick, dashed,
shorten <=2pt,
shorten >=2pt]
(scr) edge[bend right=25] (syn.south)

283
;
\draw [>=stealth, ->>, thick, shorten <=1pt,
shorten >=1pt]
(sys) edge (tes)
(lin) edge (tes)
(lin) edge (acc)
(fal) edge (ccd)
(dep) edge (fit)
(dep) edge (prag)
(dep) edge (epi)
(dep) edge (sem)
(dep) edge (syn.south)
(fal) edge[bend left=25] (syn)
(fal) edge[bend right=25] (sem)
;
\end{pgfonlayer}
\end{tikzpicture}
}
\caption{}
\end{figure}
\end{document}

284
REFERENCES

[1] Oxford English Dictionary. Oxford University Press, online edition, 2012.
[2] Peter Achinstein. The Book of Evidence. Oxford University Press, Oxford, 2001.
[3] Roger Ariew. Ockhams Razor: A Historical and Philosophical Analysis of Ockhams
Principle of Parsimony. PhD thesis, University of Illinois, Champane-Urbana, 1976.
[4] Aristotle. Physics. In Steven M. Cahn, editor, Classics of Western Philosophy, Seventh
edition, page 197. University of Chicago Press, Indianapolis, 2006.
[5] Alan Baker. Mathmatics, Indispensability and Scientific Progress. Erkenntnis, 55(1),
July 2001.
[6] George Berkeley. A Treatise Concerning The Principles of Human Knowledge. The
Open Court Publishing Company, Chicago, 1904.
[7] Max Born. Physics in My Generation. Springer-Verlag, New York, 1969.
[8] Richard Boyd. Lex Orandi est Lex Credendi. In Paul M. Churchland, editor, Images
of Science; Essays on Realism and Empiricism, pages 334. University of Chicago
Press, Chicago, 1985.
[9] Richard N. Boyd. The Current Status of Scientific Realism. In Jarret T. Leplin, editor,
Scientific Realism, pages 4380. University of California Press, Berkeley, 1984.
[10] Mario Bunge. The Weight of Simplicity in the Construction and Assaying of Scientific Theories. Philosophy of Science,, 28(2):120149, 1961.
[11] Mario Bunge. The Complexity of Simplicity. The Journal of Philosophy, 59(5):113
135, 1962.
[12] Mario Bunge. The Myth of Simplicity. Prentice-Hall Publishing Company, Englewood
Cliffs, N.J., 1963.
[13] Nancy Cartwright. How the Laws of Physics Lie. Oxford University Press, Oxford,
1983.
[14] Jordi Cat. The Unit of Science. The Stanford Encyclopedia of Philosophy, 2007.
[15] Stanley Cavel. The Claim of Reason. Oxford University Press, Oxford, 1979.

285
[16] Patricia Churchland. Epistemology in the Age of Neuroscience. The Journal of Philosophy, 84(10):544553, 1987.
[17] Paul Churchland. Matter and Consciousness. MIT Press, Cambridge Massachusetts,
1984.
[18] Paul Churchland. The Ontological Status of Observables: In Praise of the Superempirical Virtues. In Paul M. Churchland, editor, Images of Science; Essays on Realism
and Empiricism, pages 3547. University of Chicago Press, Chicago, 1985.
[19] F.M. Cornford. Plato and Parmenides; Parmenides Way of Truth and Platos Parmenides. Routledge, Great Britain, 1939.
[20] Rene Descartes. Meditations on first philosophy. In John Cottingham, editor, The
Philosophical Writings of Descartes. Cambridge University Press, New York, 2004.
[21] J. M. Dieterle. Ockhams Razor, Encounterability, and Ontological Naturalism. Erkenntnis, 55(1), July 2001.
[22] David L. Dowe, Steve Gardner, and Graham Oppy. Bayes not Bust! Why Simplicity is
no Problem for Bayesians. British Journal for the Philosophy of Science, 58:709754,
July 2007.
[23] Richard Foley. Whats to be said for simplicity? Philosophical Issues, 3:209224,
1993.
[24] Malcolm Forster and Elliott Sober. How to Tell when Simpler More Unified, or Less
Ad Hoc Theories will Provide More Accurate Predictions. The British Journal for the
Philosophy of Science, 45(1):135, 1994.
[25] Richard Fumerton. Skepticism and Reasoning to the Best Explanation. Philosophical
Issues, (2):149169, 1992.
[26] Julia Gohner, Marie I. Kaiser, and Cristian Suhm. Is Simplicity an Adequate Criterion
of Theory Choice? In Richard Swinburne; Christian Philosophy in a Modern World,
pages 3245. ontos Verlag, Frankfurt, 2007.
[27] Nelson Goodman. On the Simplicity of Ideas. The Journal of Symbolic Logic,
8(4):107121, 1943.
[28] Nelson Goodman. Fact, Fiction, and Forecast. Harvard Universtiy Press, Cambridge,
1955.
[29] Carl G. Hempel. The Theoreticians Dilemma. Minnesota Studies in the Philosophy
of Science, II(2):3798, 1958.

286
[30] Carl G. Hempel. The Philosophy Of Carl G. Hempel. In James H. Fetzer, editor,
Studies in Science Explanation and Rationality. Oxford University Press, New York,
2001.
[31] Marry Hesse. The Structure of Sientific Inference. University of California Press,
Berkeley, 1974.
[32] Christopher Hitchcock and James Woodward. Explanatory Generalizations, Part II:
Plumbing Explanatory Depth. NOUS, 2003.
[33] David Hume. An Enquiry Concerning Human Understanding. Oxford of University
Press, Clarendon Street Oxford, 1748.
[34] Immanuel Kant. Groundwork for the Metaphysics of Morals. In Steven Cahn and
Peter Markie, editors, Ethics; History, Theory, and Contemporary Issues, pages 313
352. Oxford University Press, New York, 2012.
[35] Sren Kierkegaard. Concluding Unscientific Postscript. In Steven M. Cahn, editor, Classics of Western Philosophy, Seventh edition, pages 10501051. University of
Chicago Press, Indianapolis, 2006.
[36] Jeffrey King. Structured Propositions. Stanford Encyclopedia of Philosophy, 1997.
[37] Philip Kitcher. Explanatory Unification. Philosophy of Science, 48(4):507531, 1981.
[38] Philip Kitcher. Four Decades of Scientific Explanation. In Salmon Wesley Kitcher,
Philip, editor, Scientific Explanation. University of Minnesota Press, Minneapolis,
1989.
[39] Johannes Korbmacher, Sebastian Schmoranzer, and Ansgar Seide. Simply False?
Swinburne on Simplicity as Evidence of Truth. In Richard Swinburne; Christian
Philosophy in a Modern World, pages 4650. ontos Verlag, Frankfurt, 2007.
[40] Saul Kripke. Naming and Necessity. Harvard University Press, Cambridge, MA.,
1972.
[41] Thomas S. Kuhn. The Copernican Revolution: Planetary Astronomy in the Development of Western Thought. President and Fellows of Harvard College, 1957.
[42] Larry Laudan. A Confutation of Convergent Realism.
48(1):1949, 1981.

Philosophy of Science,

[43] Larry Laudan. Science and Values. Univerisity of California Press, Berkeley and Los
Angeles, 1984.

287
[44] Gottfried Leibniz. Discourse on Metaphysics. In Garber Daniel Ariew, Roger, editor,
Philosophical Essays. Hackett Publishing Company, Indianapolis, 1989.
[45] Eric Margolis and Stephan Laurence. Concepts; Core Readings. MIT Press, Cambridge Massachusetts, 1999.
[46] Margaret Morrison. Models, Measurement and Computer Simulation: The Changing
Face of Experimentation. Philosophical Studies, 143:3357, 2009.
[47] Ernest Nagel. The Structure fo Science. Harcourt, Brace and World Inc, New York,
1961.
[48] Friedrich Nietzsche. Philosophy in the Tragic Age of the Greeks. Regenery Publishing
Inc., Washington D.C., 1962.
[49] Plato. Theaetetus. In Edith Hamilton, editor, The Collected Dialogues of Plato, page
858. Princeton University Press, Chicago, 1989.
[50] N.T. Roseveare. Mercurys Perihelion: From Le Verrier to Einstein. Oxford University Press, Oxford, 2001.
[51] W.C. Salmon. The Foundations of Scientific Inference. The University of Pittsburg
Press, Pittsburg, 1967.
[52] Brian M. Scott. Technical Notes on a Theory of Simplicity. Synthese, 109(2):281
289, 1996.
[53] J.J.C. Smart. Sensations, and Brain, Processes. The Philosophical Review, pages
141156, 1959.
[54] Elliot Sober. Instrumentalism, Parsimony, and the Akaike Framework. Philosophy of
Science, pages 112123, 2002.
[55] Elliot Sober. What is the Problem of Simplicity? In H. Keuzenkamp, M. McAleer,
and A. Zellner, editors, Simplicity, Inference, and Econometric Modelling, pages 13
32. Cambridge University Press, Cambridge Massachusetts, 2002.
[56] Marshall Spector. Concepts of Reduction in Physical Science. Temple University
Press, Philadelphia, 1978.
[57] Michael Stevens. Depth. Harvard University Press, Cambridge, MA., 2006.
[58] Lloyd Strickland. Leibniz Re-Interpreted. Continuum International Publishing, London, GBR, 2006.

288
[59] Richard Swinburne. Simplicity as Evidence of Truth. Marquette University Press,
Milwaukee, Wisconsin, 1997.
[60] Paul R. Thagard. The Best Explanation: Criteria for Theory Choice. The Journal of
Philosophy, 75(2):7692, 1978.
[61] Bas van Fraassen. The Scientific Image. Clarendon Press, Oxford, 1980.
[62] Bas vanFraassen. Empiricism in the Philosophy of Science. In Paul M. Churchland,
editor, Images of Science; Essays on Realism and Empiricism, pages 245305. University of Chicago Press, Chicago, 1985.
[63] William Vollmann. Uncentering the Earth. W.W. Norton and Company, New York,
2006.
[64] Brad Westlake. Explanatory Depth. Philosophy of Science, pages 273294, 2010.