The Anti-Realist Side of the Debate

A Theory's Predictive Success does not Warrant Belief in the
Unobservable Entities it Postulates

Andre Kukla and Joel Walmsley

6.1 Introduction

One problem facing the epistemology of science concerns the fact that many scientific theories consist of claims
about unobservable entities such as viruses, radio waves, and quarks. If these entities are unobservable, how are
we to know that the theories that make reference to them are true? The question we ask is "Does the success of a
scientific theory ever give us reason to believe in the unobservable entities it postulates?" The existence of the
unobservable entities postulated by a theory is, of course, entailed by the truth of that theory. If a theory is true,
then the entities it postulates exist (if they didn't, the theory wouldn't be true). Accordingly, where we talk of the
truth of a theory, we mean that to include the existence of the unobservable entities it postulates, as will be
noted.

The fact is that scientists do hold theories about unobservable entities, and for the most part, they hold them to
be true. At the risk of overgeneralization, then, we might say that most practicing scientists are realists, where
realism is the thesis that scientific theories are true (with all that that entails). It must be said that this view is a
particularly strong (and therefore fragile) form of realism. This chapter aims more ambitiously to unseat the
argument from scientific success for the much weaker realist thesis that there are nomologically possible
circumstances (circumstances which are possible given the laws of physics in this world) wherein we would be
justified in believing in the unobservable entities postulated by a theory. Henceforth, we use the term "realism" to
refer to this view - which Leplin has called "minimal epistemic realism" - that it is logically and nomologically
possible to attain a state that warrants belief in a theory. It is worthy of note that there may be other good
reasons for being a (minimal epistemic) realist - the aim of the present chapter is to show that the success of
science isn't one of them: the success of science does not warrant even the weakest realist thesis.
 

We should start by saying what we mean by the "success of science." Let us say
that by this phrase, we mean that our scientific theories allow us to make signifi
cantly more correct predictions than we could without them. A less circumscribed,
but nonetheless consistent, version is Laudan's broadly pragmatic definition:
A theory is successful provided it makes substantially more correct predictions, that it
E leads to efficacious interventions in the natural order, or that it passes a suitable battery
of tests. (1984, p. 109) Nothing substantial in our argument will hinge on the difference
between these two formulations.

The argument for realism from the success of science is often referred to as the "no miracles" argument,
following Putnam's (1979) contentious formulation. In his view, "[t]he positive argument for realism is that it is
the only philosophy that doesn't make the success of science a miracle" (p. 73). In order to block Putnam's
conclusion, then, we need to show that there is a coherent explanation for science's success, which neither
requires nor entails realism. This is to say that even if we grant that realism does explain the success of science,
Putnam's claim can be undermined by showing that there are comparably good anti-realist explanations.
 

6.2 Anti-Realist Explanations for the Success of Science

Leplin (1987, 1997) has conducted a thorough investigation of this topic and concludes that there are no viable
anti-realist arguments for the success of science. He considers two categories of putative anti-realist explanations
for the success of science: those that allude to evolutionary-type mechanisms of theory selection, and those that
regard a theory's empirical adequacy (i.e., the extent to which the theory correctly describes observations and
experimental results) to be an explanation of its success. According to Leplin, both strategies fail, so the only
remaining explanation of scientific success is the truth of theories. His critique of the evolutionary alterna
tives is persuasive - more so than his critique of the empirical adequacy alternative - so let us consider them in
that order.

The archetypal evolutionary explanation for the success of science can be traced to van Fraassen (1980),
according to whom the success of our theories is due to their having been chosen by a mechanism of Darwinian
selection:

Species which did not cope with their enemies no longer exist. That is why there are
only ones who do. In just the same way, I claim that the success of current scientific
theories is no miracle. It is not even surprising to the scientific (Darwinist) mind. For any
scientific theory is born into a life of fierce competition, a jungle red in tooth and claw.
Only the successful theories survive - the ones which in fact latched on to actual regu
larities in nature. (p. 39)

This thesis often goes by the name of "evolutionary epistemology," and amounts to the view that theories are
selected by the scientific community for predictive success and that this practice suffices to account for the
success. It has to be defended against the charge that the competition among theories fails to satisfy some of the
conditions necessary for a process of Darwinian selection to take place. Leplin's (1997) analysis, which we now
summarize, shows that van Fraassen's account fails to provide an explanatory alternative to theoretical truth.

Leplin draws an analogy with the tennis-playing abilities of Wimbledon finalists:
How do we explain the fact that Wimbledon finalists are so good at tennis? The
question admits of two different interpretations. If we want to know why it is that
Wimbledon finalists are in general good tennis players, it is appropriate to cite the
stringency of the selection procedures for entry into the tournament. This answer,
however, does not tell us why particular individuals - Sampras and Agassi, for
example - are good at tennis. To answer this question, we would have to cite the rel-
evant properties of the individuals, such as their training and genetic endowment. To
make the analogy explicit, the problem with evolutionary epistemology is that the
question it answers is not the same as the question answered by theoretical truth. The
Darwinian account explains how we come to be in possession of successful theories,
but it doesn't explain what it is about these theories, which we happen to possess,
that makes them successful. Theoretical truth does not explain how we come to possess
these particular theories. But it does answer the second type of question with regard
to predictive success: it is an attribute of these particular theories that accounts for
that success. We must conclude that evolutionary epistemology and theoretical truth
are not explanatory rivals - van Fraassen's account does not give us an anti-realist
explanation for the success of science and the "no miracles" argument remains
unscathed.

Leplin notes that the same can be said for Laudan's (1984) elaboration of van Fraassen's Darwinian account,
which specifies the mechanism by which theories are selected. According to Laudan, success is to be explained
by the fact that scientists' use appropriate experimental controls and methodologies in testing their theories.

Indeed, the fact that our theories have been tested by such-and-such a methodology may explain why we
possess successful theories, but it does not explain why quantum mechanics and not some other theory has
survived all those tests. The hypothesis that quantum mechanics is true, by contrast, provides at least an
explanation for its success.

Before we turn to his critique of the empirical adequacy alternative (i.e., the part with which we
disagree), let us briefly set a few ground rules for the discussion. We will suppose, for now, that no other
potential problems with the "no miracles" argument need to be worried about. For the sake of argument, let us
grant the realist an awful lot, and assume that science is successful, that the realist argument from the success of
science is not question-begging or invalid, that the notion of (mere) approximate truth is both meaningful and
measurable, and that approximate truth does explain the success of science. Given these assumptions, it will
make no difference whether we talk of a theory's truth or its approximate truth. We can, therefore, play fast and
loose with the distinction as well as with the distinction between empirical adequacy and approximate empirical
adequacy. For the sake of expository convenience, we speak of the truth of theories where we mean their truth
or approximate truth, and of empirical adequacy where we mean empirical adequacy or approximate empirical
adequacy. Provisionally, let us also define "T is (approximately) empirically adequate" as "T is (approximately)
empirically equivalent to a true theory." (Note that this definition of "empirical adequacy" differs from the
conventional one whereby a theory is empirically adequate if it gives a correct description of observations and
experimental results.)

Leplin (1987) defines surrealism as the thesis that the explanation for the success of our theories is that
they are empirically adequate in the sense just defined - that is, our theories are empirically equivalent to true
theories. Surrealism holds that the world behaves as if our theories of the world were true, but it "makes no
commitment as to the actual deep structure of the world." Surrealism "allows that the world has a deep
structure," which is to say that it allows that there is a true theory of the world whose theoretical terms denote
real entities, but it "declines to represent" these structures (Leplin, 1987, p. 520). Leplin attributes the
development of this view to Fine (1986) but, as Leplin rightly notes, Fine does not actually endorse surrealism,
he simply cites its coherence to discredit Putnam's "no miracles" argument.

Leplin presents two arguments designed to show that surrealism collapses into realism. The first is that
the surrealist explanation for the success of science presupposes the realist explanation: To suppose that
Surrealism explains success is to suppose that for the world to behave as if the theory is true is for the theory's
predictions to be borne out. But this is to suppose that theoretical truth is normally manifested at the
observational level, and thus to presuppose the original, realist explanation of success rather than to provide a
weaker alternative. (1987, p. 523) If this is right, then surrealism does not qualify as an alternative to the realist
explanation. But this claim merely equivocates between two senses of "explanation." Let us explain. We need to
disambiguate the phrase "A explains B," and do so without becoming entangled in the various philosophical
questions concerning explanation that are peripheral to this chapter. To say that A is an explanation for B is to
make a claim which is consistent with A being false and the possibility of there being some C, incompatible with
A, which is also an explanation for B. To say that A is the explanation for B is clearly to imply that A is true, and
that the truth of A is the reason why B is true. "A explains B" is ambiguous between these two options.
Returning to the issue at hand, to suppose that the empirical adequacy of a theory T explains its success (in either
sense of "explains") is indeed to presuppose that T's truth must also qualify as an explanation of its success. But
it is not to presuppose that T's truth is the (i.e., the correct) explanation of its success, or even that T is true at
all. Both realists and anti-realists acknowledge that scientists frequently make use of theories that are known to
be false in predicting observations. Scientists sometimes represent gas molecules as minute billiard balls. The
explanation for why these patently false representations succeed is presumably surrealistic: they are approxi-
mately empirically equivalent to other theories about gases that are true. If we accepted Leplin's claim about
surrealism's presupposition of realism, then we would have to conclude that gases are made up of tiny billiard
balls.]

One might argue that, whilst empirical adequacy does not presuppose truth simpliciter, it does
presuppose approximate truth: the success of the billiard ball theory of gases does not entail that gas molecules
are billiard balls, but it does entail that gas molecules are relevantly like billiard balls. Most realists would be
content with the verdict that our best theories are approximately true, so the surrealist explanation of the
success of science may presuppose the realist explanation after all. The temptation to think that approximate
empirical adequacy entails approximate truth stems from the following considerations. For a theory to be
approximately true means, let's say, that it gets most or many of its claims right, even though it may also get
a some things wrong. Similarly, for a theory to be approximately empirically adequate, it must get a lot of
its observational claims right, even though it may also get some of them wrong. But to get a lot of
observational claims right is to get a lot of claims right, tout court. Therefore, approximate empirical
adequacy entails approximate truth.

The problem for realists, though, is that this notion of approximate truth has no connection with scientific
realism. If getting any sizeable part of the story right is enough to count as approximate truth, then it is
possible for an approximately true theory to get the theoretical part of the story entirely wrong despite getting
the observational part right. Getting a lot right is certainly necessary for approximate truth, but if approximate
truth is to have any bearing on the issue of realism, it should also a be stipulated that approximately true theories
get at least some things about theoretical entities right, not just observational statements. There is no need here
to specify "just what part or how much of the theoretical story an "approximately true" theory has to get right.
But however slight that theoretical requirement may be, the result is a notion of approximate truth that is no
longer presupposed by approximate empirical adequacy. The surrealist explanation does not presuppose the
realist explanation because getting an arbitrarily large part of the observational story correct is logically
compatible with getting the theoretical story totally wrong.

Let's call the hypothesis that every theory has empirically equivalent rivals EE. Leplin's second argument
is that surrealism threatens to collapse into realism unless EE is true. For suppose that we have a successful
theory T which has no empirically equivalent rival. Then the surrealist explanation for T's success - namely that T
is empirically equivalent to a true theory - entails that T itself is true. This is to say that the surrealist explanation
entails the realist explanation. To maintain its distinctiveness, surrealism must be committed to the truth of EE.
Thus Leplin regards it as unproved that surrealism provides a genuine explanatory alternative, since it is an open
question whether EE is true. The first response to make is that it is not an open question as to the universal
availability of empirically equivalent rivals. Kukla (1998) argues for several algorithms for transforming any
theory into an empirically equivalent rival theory (see chapter 5 in particular). We shan't insist on that point here -
rather, let us concede, for the sake of argument, that the status of EE is uncertain, and hence that the
distinctiveness of the surrealist explanation is also uncertain. We can nonetheless weaken surrealism in such a
manner as to block the derivation of realism from the negation of EE whilst still leaving an available explanation
for the success of science.

Let us here revert to the more traditional definition of empirical adequacy: T is empirically adequate if all
of its empirical consequences are true. As before, the empirical adequacy of T entails that the observable world
behaves as if the "deep structures" postulated by T do exist, but it is no longer required that the observable
phenomena be consequences of any real "deep structures." This is to say that, in virtue of the fact that it drops
the requirement that T's empirical consequences also be the consequences of a true theory, this notion is logically
weaker than the previous one. We might give the name weak surrealism to the thesis that our best theories are
empirically adequate in the new diminished sense, and strong surrealism to the thesis that our best theories are
empirically equivalent to true theories.

Now consider the hypothesis that weak surrealism explains the success of science.° In fact, such a hypothesis is
immune to the second difficulty that Leplin makes for the strong surrealist explanation. Leplin claims, however,
that such an escape from the collapse of strong surrealism merely lands us in another dilemma:

The statement that things happen as they would happen were our theories true invites two
directions of analysis. On one analysis, Surrealism simply tells us what goes on at the observational level.
Then it is not simply metaphysically weaker than realism; it is not metaphysical at all. But neither is it
explanatory. Rather, it restates the explanandum. (1987, p. 522)

The "direction of analysis" alluded to in this passage is the direction of weak surrealism. The other
direction is strong surrealism. According to Leplin, in their choice between strong and weak surrealism,
anti-realists are caught between the Scylla of realism (if they choose strong surrealism, their explanation threatens
to be indistinguishable from that of the realists) and the Charybdis of vacuity (if they opt for weak surrealism,
then their explanatory hypothesis amounts to no more than a recapitulation of the explanandum).

But there is a position between strong surrealism and the explanation that Leplin regards as vacuous, and
so it seems his argument conflates three putative explanations of the success if science into two. Let's call T* the
set of all empirical consequences of T. Weak surrealism is the view that T* is true. But one can hold that T" is
true in two different ways. On the one hand, one may believe that the statement "The observable world behaves
as if T is true" is itself a nomological truth about the world - a law of nature and part of T just like any other
law-like statement about the world contained in T. On the other hand, one may believe in the truth of a set, X, of
lawlike empirical generalizations, along with the (nonnomological) hypothesis that, as it happens, X is
coextensive with T", perhaps by accident. In the latter case, one believes two things: (1) for each statement in X,
that statement is a true law-like generalization; and (2) these generalizations exhaust T". The conjunction of (1)
and (2) amounts to saying that one does not believe in a further law that X is coextensive with T*; rather, one
accepts this coextensiveness as a contingent accident.

The difference between the two can be brought out by considering what happens if some previously
unconsidered empirical hypothesis E is found to be a consequence of T. If you believe that it is a law of nature
that the world behaves as if T is true, then you will add E to your stock of beliefs. But if you think that the
coextensiveness of T* and your empirical beliefs is merely accidental, then the same discovery will leave the
status of E open; its derivation from T will not be a rationally compelling reason for its adoption into T". Indeed,
the discovery that E is a consequence of T might be a good reason for giving up the belief that the true empirical
generalizations are coextensive with T*. We can, therefore, reserve the name "weak surrealism" for the first
view, and call the second view fragmentalism. Fragmentalism is logically weaker than weak surrealism.

We thus have a series of progressively weakened views about theories: (1) realism (our theories are true);
(2) strong surrealism (our theories are empirically equivalent to true theories); (3) weak surrealism (the
observable world behaves as if our theories were true); and (4) fragmentalism (the true empirical generalizations
happen to be coextensive with the empirical consequences of our theories). The further down the list we go, the
less danger there is of a collapse into realism, but also the fewer resources we have for explaining the success of
science. Leplin allows that strong surrealism explains the success of science but notes the danger of its collapse
into realism.

On the other hand, a fragmentalist account of success falls prey to the charge of vacuity, since according
to that view, the ultimate truth about the world - or as much of it as we can ascertain - is given by a collection of
empirical generalizations, and the truth of our empirical generalizations is exactly the explanandum in the success
of science argument. But Leplin doesn't distinguish vacuous (4) from (3) - and (3) seems to avoid both Scylla
and Charybdis. According to weak surrealism, the truth of our empirical generalizations is explained by the fact
that (i) these generalizations are consequences of T and (ii) the observable world behaves as if T were true. Since
this claim goes beyond what is contained in the explanandum, weak surrealism avoids the charge of vacuity. But
there is no danger of it collapsing into realism, even if it turns out that some theories have no empirically
equivalent rivals. Thus, weak surrealism is an anti-realist rival to the realist explanation for the success of science.
Of course, the existence of coherent and nonvacuous rivals to realism does not show that these explanations are
as good as theoretical truth. The realist may abandon Putnam's own formulation of the "no miracles" argument
and weaken her thesis to the claim that the realist explanation is better than the anti-realist explanation. The
realist intuition is that the weak surrealist claim that the empirical adequacy of T is a fundamental fact about the
world is unintelligible. We have no idea what could keep all the disparate empirical consequences of T in line, if
not the causal mechanisms postulated by T itself, or at least, as strong surrealism asserts, the mechanisms
posited by some other theory that is empirically equivalent to T: simply saying that they are kept in line doesn't
explain anything. But what is the epistemic import of intelligibility?

It used to be thought that teleological explanations gave us understanding, but now we want to know the
mechanism behind an appearance of teleology. Explanations that posit action at a distance have been considered
intelligible and unintelligible by successive generations. The point is that the notion of "intelligibility" is at least
relative if not vague. It would seem that the realists have in mind an archetypal explanation as one that satisfies
our explanatory hunger, likely mechanistic in flavor.

But it doesn't seem that the "intelligibility" of this kind of explanation signifies anything more than a
psychological willingness to stop inquiry at that point. It would be, of course, premature to claim that there
cannot be an epistemically relevant notion of intelligibility that brands weak surrealism as unintelligible, but at
present there does not seem to be such a notion formulated. In fact, the only relevant discussion, to our
knowledge, concludes that the intelligibility of scientific theories is a matter of "psychological acclimation"
(Gushing, 1991, p. 346).

Does this critique break one of the ground rules? Recall that we granted the valid ity of the "no miracles"
argument, and then went on to argue that weak surrealism is a coherent, nonvacuous, and distinct explanation of
the success of science. If we are right, then realists need to show that theirs is a better explanation than that of
the weak surrealists. They do not have to show that explanatory goodness has epistemic import, since that much
was assumed when the validity of the "no miracles" argument was granted - if explanatory goodness were not
epistemically relevant, then the conclusion of the "no miracles" argument wouldn't follow from its premises. It is
precisely this principle that we call into question, but to grant that explanatory goodness is epistemically relevant
is not to give the realists license to use any features of explanations they like in making a comparative evaluation.
It would be cheating, for example, for the realist to claim that their theory is superior on the grounds that the
positing of theoretical entities is an indicator of explanatory goodness.

The problem is that the assumption of the epistemic relevance of explanatory goodness doesn't tell us
which theoretical properties are indicative of this epistemically relevant explanatory virtue. We can no longer deal
with Leplin's problem in isolation from other problems faced by the "no miracles" argument. The realists' task is
to find a property, p, of explanations which simultaneously satisfies this pair of criteria: (1) theoretical truth has
more p than empirical adequacy; and (2) the possession of p is relevant to beliefworthiness. The realist must at
the very least show that there exists some p such that theoretical truth has more of it than empirical adequacy,
and that p is not demonstrably irrelevant to beliefworthiness. It is easy to satisfy either requirement by itself. On
the one hand, theoretical truth is undoubtedly a more intelligible explanation of success than empirical adequacy.
But if, indeed, intelligibility is relative, then it cannot play the role that realists need it to. On the other hand, an
example of a property of explanations whose epistemic credentials cannot be denied is the probability that it is
true. But since theoretical truth is logically stronger than empirical adequacy, it cannot be maintained that the
realist explanation has more of this property than the surrealist explanation. The trick is to satisfy both
requirements at once, and realists have not yet managed to perform this feat.