On structural realism

Another approach related to this topic is the so-called structural realism of John Worrall (1989). Worrall’s motivation is to do justice to the pessimistic meta-inductionargument – the argument that since the observable posited by past scientific theories turned out nonexistent, it is likely that the unobserv-ables of today’s science are also nonexistent. He agrees that the pessimistic meta-induction is correct in pointing out that there is a radical discontinuity at the level of unobservables. However, Worrall argues for a level of continuity and progress in the dynamic of theories. According to him, the mathematical content of a superseded theory is typically retained and embedded in the for-malism of the successor theory. In other words, Worrall advocates a continuity at the level of theories’ mathematical structures, irrespective of the content of the concepts filling out these structures. Presupposed is a distinction which Worrall draws between thenature and thestructure of an entity. The latter is given by the mathematical equations defining the entity, whereas the former just cannot be quantitatively described. In an important sense, as structural realism seems to claim, the only firm knowledge we have is about structures:

in trying to find out more about structures, all we can get is more structure.

I am rather expedite in rejecting structural realism. For one thing, I do not share its uneasiness either about the pessimistic meta-induction argument, or about Kuhn’s (1962) conception of the incommensurability of scientific paradigms, which also inspired Worrall. Second, and more importantly, our argumentation showed in sufficient measure that we can believe in unobserv-able entities on the grounds of their causal properties. Therefore, I believe that one of structural realism’s main motivations (skepticism about unobservable entities) is misconceived.

Worrall wanted to offer us ‘the best of two worlds’: the Kuhnian world of radical discontinuities at the theoretical level, and continuity at a level in-between empirical laws and theoretical accounts of mechanisms and causes (cf.

Worrall 1989: 111). Yet, the former world does not cause us any anxiety, since we reject Kuhn’s conception of incommensurability. As to the latter one, we already have its best: scientific realism.

Chapter 4

The Underdetermination Argument:

The Theoretical/Observational Distinction

In the introductory chapter, scientific realism was defined as the doctrine claiming that most of the essential unobservable entities posited by the well-established scientific theories do exist independently of our minds, language, and theories. Moreover, the theoretical descriptions of those entities’ properties and of the relations among them are approximately true. Typically, scientific realists are also committed to the view that we can ascertain that science’s unobservables exist and that the theories about them are approximately true.

This view belongs, as we have noticed when discussing the horizontal varieties of realism (see Introduction), to theepistemic dimension of scientific realism.

To reiterate from earlier discussion, there are several degrees of epistemic commitment that a realist may hold. What we deem to be the typical scientific realist position is in fact situated in the middle of a spectrum ranging from the view that our best theories are rigorously true, to the view that we are only rationally warranted to assert that our theories are approximately true.

As seen in chapter 2, the belief in the strict truth of scientific theories is unwarranted: science essentially relies on idealizations, simplifications, and approximations, facts which make its sentences, strictly speaking, false. Be-sides, past theories, even when empirically successful, turned out to include idle parts, erroneous calculations, as well as clearly false ontological assump-tions. Recognition that knowledge is fallible and acceptance of a concept of closeness to truth is a lesson learnt by most realists.

Yet, one may be rationally warranted in believing that the best scien-tific theories are approximately true even if, in fact, these theories are not approximately true. Leplin (1997) calls this position minimal epistemic re-alism. Minimal epistemic realism circumvents a significant objection levelled against ‘standard’ scientific realism, namely the ‘pessimistic meta-induction’ –

the view that, since past theories turned out to be false, it is highly probable that, in the future, we shall also come to regard our theories as false and not approximately true. However, the cost for easily coping with pessimistic meta-induction is rather high for minimal epistemic realism: it cannot account for the methodological success of science. In other terms, it cannot answer prop-erly the question, why are the methods of science so successful at generating empirically successful theories. We saw in section 2.4 that the best explanation of this success relies on those theories’ being approximately true.

Taking one step further towards scepticism, we reach a situation in which any possible ground for rational belief in a scientific theory is denied. While it may be that the theorycan well be true from God’s perspective, even if it was, we as mortals could never acquaint the grounds to believe it. Obviously, this step takes us beyond the borderline between realism and antirealism. To claim that we can never have grounds to believe in any theory is the very definition of epistemic antirealism. The rejection of the latter comprises the subject of the current and the next chapter. Epistemic antirealism is famously exemplified by van Fraassen’sconstructive empiricism, which we labelled as a form of agnostic instrumentalism – i.e., instrumentalism agnostic with respect to the referentiality of theoretical entities. Accordingly, a substantial part of the current chapter will be devoted to a discussion of various aspects of constructive empiricism. But there are also agnostic forms of phenomenalism and of the philosophy of as-if which, while admitting the possibility that the unobservables posited by science may exist, decline any reason to believe in them.

What these doctrines share is the empiricist credo that all knowledge is ultimately based onsense experience. Empiricists typically assume that beliefs about sense experience are justified by directly ‘reading off’ the contents of sense experience. Thus, if an object is visually presented to me as a green car, I merely read off this appearance in forming the belief that there is a green car in front of me. The scope and accessibility of these sense experience-contents is a contentious matter. Phenomenalism, for example, claims that one’s perceptual beliefs receive support solely from one’s own sense experience.

Constructive empiricism by contrast, appears to claim that sense experience-contents are public, shared by all those perceiving the sensory stimuli from a given environment. It is beyond my purpose to go any further into this topic.

What is relevant for the discussion of epistemic antirealism is the common empiricist conviction that the leap from beliefs about sense experience to beliefs in unobservables is epistemically illegitimate.

What is the motivation for such a scepticism? Why do empiricists take be-lief in unobservables to be so shaky? How can empirical evidence be deemed incapable of guiding theory choice? The answer can be devised in terms of

the doctrine of underdetermination of theory by all possible evidence. We’ll immediately see that there are different formulations of underdetermination, increasingly threatening to the realist theses. All these formulations have the claim in common that indefinitely (possibly infinitely) many incompatible the-ories entail any given set of evidence. That corresponds to saying that no theory can simply bededuced from evidence. There is no unique upward path, from evidence tothe theory entailing it; there are, as the arguments goes, indef-initely many such upward paths. Before entering into technical details, some terminological preliminaries are required. We label, respectively, E the given evidence,T1, . . . , Tna set of rival theories, andR the rules employed to choose between these theories. Let EC(T) stand for the empirical content of theory T, i.e. the set of observational sentences entailed by T. Thus, we say that theories T and T’ are empirically equivalent if and only if EC(T) =EC(T⁰).

The latter phrase can be spelled out as follows: given a neutral observation language L_O,T and T’ are empirically equivalent in case they have the same deductive connections to the evidential basis, E, formulated in LO.

Thus equipped, the core of the underdetermination thesis may be expressed as follows: if both T and T’ deductively entail E, then the evidence E and the rule of choice R cannot provide sufficient rational grounds for preferring any of the two theories to its rival. As the incoming analysis will show, several clarifications are in order.

First, is one to take E as the existent body of evidence, or as all possible evidence? If only the existent body, then shouldn’t it be expected that some future empirical discoveries may drastically change one’s preference for one theory in favor of its empirical equivalents? And if all (logically and nomologi-cally) possible evidence, then possible for whom? Only for humans, with their sensory perceptual limitations, or for any imaginable rational being?

Second, we have defined the empirical content of a theory as its set of observable consequences. It is known that such derivations require without exception the presence of auxiliary theoretical statements (i.e., information about the initial conditions, about the instruments in the laboratory, about the mathematical and logical methods, etc.). But, as Laudan and Leplin (1991) argue, these auxiliary assumptions vary over time in two respects: they are defeasible andaugmentable. Accordingly, the set of observational consequences of a theory varies over time. This fact urges a diachronic treatment of the relation between theory and empirical evidence.

Third, it is relevant to know whether a given empirically successful theory has one empirically equivalent rival, a finite number of rivals, or an infinite number of rivals. Assume, for example, that the upward path from evidence to theory proves indeed to be blocked. It is one thing to have only two empirically equivalent rivals T and T’, given that our subjective probabilities for each of

them arep(T) =p(T⁰) =.5. It is another thing to have a finite (no matter how large) number of empirically equivalent rivals,T₁, . . . , T_n, since our subjective probabilities for each of them are p(T₁) = . . . = p(T_n) = 1/n. Finally, it is another thing altogether to have an infinite number of rivals, being that the respective subjective probabilities are zero.

Naturally, each of these eventualities leads to a different formulation of the empirical equivalence thesis. Thus, depending on whether we take the notion of empirical evidence in a restrained sense (as given evidence) or in a broad sense (as possible evidence), we obtain two extreme versions of the underde-termination thesis (henceforth UD): first, a weak version (W U D), which is no threat to scientific realism:

WUD: For any body of evidence, there are indefinitely many mu-tually contrary theories, each of them logically entailing that evi-dence.

WUD has been quite similarly formulated by Laudan (1996), Newton-Smith (2000), and Devitt (2003). Second, there is a strong version of the UD thesis (SUD), which the realist has to take pains to block:

SU D: Any theory T is radically underdetermined in that all pos-sible evidence could not justify it over its empirically equivalent rivals. (Devitt 2003)

It is an important task of this chapter to argue that there are no reasons to believe SUD.

Let us now systematize the debate over UD. Formally, the UD argument runs in the form of the followingmodus ponens:

(i) The thesis of empirical equivalence (EE): For any given theory, there are indefinitely many empirically equivalent rivals;

(ii) The entailment thesis (EE→ UD): Empirical equivalence entails under-determination.

Accordingly, the realist can adopt one of the following lines of counterattack:

• She can show (i) to be incoherent or unsustainable.

• She can accept that (i) is coherent and instead try to show that it is false.

• She can admit (i) but, nonetheless, deny (ii).

These realist strategies structure our agenda in this chapter. Throughout the current chapter, we argue in detail against the possibility of a principled distinction between theory and observation, on which epistemic antirealism essentially relies. If correct, the outcome of this chapter undergirds the first realist counterattack strategy. Chapter 5 will be dedicated to the analysis of the other two strategies.

Some realist philosophers maintain that UD is not intelligible since it relies on the thesis of empirical equivalence (EE) which, to be formulated, presup-poses a neat dichotomy between the observable and the theoretical. The ob-jection is that such a dichotomy is untenable. Let us spell this out. EE claims that any theory T has empirically equivalent rivals. As already seen, two the-ories, T andT’, are empirically equivalent if and only ifEC(T) =EC(T⁰), i.e.

if bothT andT’ entail the same body of evidence,E. But to specify a body of evidence means to formulate a set of observational sentences. To see that both T and T’ imply the same E, we need to specify E in a neutral observation language, L_O.

Now, the neutrality ofLOcan be taken to be relative to a specific epistemic context. Thus, what counts as an observation depends on the specific cognitive interests of the scientists. Depending on which issues an experiment is designed to settle, different empirical assumptions are taken for granted. The point has been made, among others, by Fodor (1984):

We can’t test all our beliefs at once. It is perfectly reasonable of working scientists to want to mark the distinction between what’s foreground in an experiment and what is merely taken for granted, and it’s again perfectly reasonable of them to do so by relativizing the notion of an observation to whatever experimental assumptions are operative. (Fodor 1984: 25–6) However natural in scientific life, this background-relative manner of distin-guishing observation from theory is not what the empiricist antirealist needs.

Recall that her scepticism is grounded on the alleged irrationality of the leap from beliefs about sense experience to beliefs in unobservables. These epis-temic scruples are conspicuously offended by such a relativization, because to admit that observation hinges on scientists’ theoretical background and epis-temic interests is already to admit that the leap to theoretical beliefs has been committed. Thus, what the epistemic antirealist needs is a narrow construal of the notion of observation, as to include only sense experience.

As we shall see, the antirealist has serious trouble in doing justice (as she must) both to the fact that observation is thoroughly theory-laden, and to the idea that there is an observational language neutral with respect to all theories. But before taking this line of argumentation, let us take a detour to

the so-calledreceived view of theories and the support it offers to instrumen-talism. This provides a good setting for the subsequent examination of the theory/observation distinction.

4.1 The theoretician’s dilemma

In subsection 2.1.3 we have confronted realism with instrumentalism on the ex-planatory dimension, and argued that the former is preferable because, unlike the latter, it can offer causal explanations. The focus now shifts to instrumen-talism’s deeper epistemic roots and to the reasons why it has been such an appealing philosophy of science.

The instrumentalist impetus came from the conviction that science can well get along without appeal to theoretical unobservable entities in achieving its aims of making predictions and establishing connections among observables.

These purposes have been thought to be attainable only by means of laws and mechanisms couched exclusively in an observable vocabulary. Indeed, many empirical generalizations can be formulated in observable terms. Here is an example offered by Hempel (1958: 43):

(1) Wood floats on water; iron sinks in it.

For many practical situations, that is a correct phenomenological description.

Nonetheless, it suffers from evident deficiencies: “it refers only to wooden and iron objects and concerns their floating behavior only in regard to water.

And, what is even more important, it has exceptions: certain kinds of wood will sink in water, and a hollow iron sphere of suitable dimensions will float on it.” (Hempel 1958: 43). These shortcomings can, for this specific instance, be easily remedied by introducing the concept ofspecific gravity of a body x, defined as s(x) =w(x)/v(x), wherew and v are, respectively, x’s weight and volume. A generalization of (1) is thus obtained as follows:

(2) A solid body floats on a liquid if its specific gravity is less than that of the liquid. (1958: 44)

This statement correctly predicts the floating behavior of any solid body upon any liquid. Yet, the key concept of the generalization, s, though defined in terms of observable characteristics of x, is not itself observable. Why use s? Why not formulate empirical generalizations directly and exclusively in observable terms? No doubt, (2) is equivalent to (2’):

(2’) A solid body floats on a liquid if the quotient of its weight and its volume is less than the corresponding quotient of the liquid.

(1958: 46)

So theoretical terms seem to be superfluous. They can be used as economi-cal abbreviations, relating several observational characteristics. It would have been uncomfortable to use in all applications of Newtonian gravitation theory the formula ‘the product of the masses of two bodies divided by the square of the distance between them’ instead of using the concept of ‘gravitational force’; and it would have plainly been impracticable to do so with the quantum-mechanical concept of ‘wave function’ (granted that it can be defined in ob-servational terms). But are theoretical terms nothing but economical devices?

And assuming a computation power high enough to allow their substitution with observables to be carried out in all situations, could theoretical terms be entirely avoided? The force carried by this question is expressed in Hempel’s famoustheoretician’s dilemma:

If the terms and principles of a theory serve their purpose [i.e., they establish definite connections among observable phenomena] they are un-necessary, ...and if they don’t serve their purpose they are surely unnec-essary. But given any theory, its terms and principles either serve their purpose or they don’t. Hence, the terms and principles of any theory are unnecessary. (Hempel 1958: 49–50)

The conclusion is in apparent harmony with the instrumentalist philosophy. In order to approach it, a closer look at the concept of theory that has supported this view is useful.

Logical positivism set out to develop an observational language for science, L_O, which contains, apart from the logical vocabulary, only observational terms – individual constants and predicates. The observational terms directly des-ignate observable entities and properties, while all other terms are explicitly definable in L_O. Thus, the full language of science, L, which also includes theoretical terms such as ‘electron’, ‘energy’, ‘gene’, etc., is reducible to or translatable into LO.¹

1The precise nature ofLOwas a disputed matter among logical positivists. The divergence emerged as a debate overprotocol sentences – the elementary form in which the results of scientific experimentation are recorded – in the 1930s within the Vienna Circle. Carnap had expressed in his Aufbau (1928) the phenomenalist view that such propositions must express private complexes of sensations. For the ‘left-wing’ oriented Neurath, this view was incompatible with the public and intersubjective character of science. Consequently, he advocated the view that protocols are accepted by the scientific community as reporting the results of publicly accessible observations. Protocol sentences, as Neurath maintained, must be expressible in thephysicalisticlanguage of unified science, which makes them, like all other sentences, revisable in principle. This view was strongly opposed by the ‘right-wing’ oriented

1The precise nature ofLOwas a disputed matter among logical positivists. The divergence emerged as a debate overprotocol sentences – the elementary form in which the results of scientific experimentation are recorded – in the 1930s within the Vienna Circle. Carnap had expressed in his Aufbau (1928) the phenomenalist view that such propositions must express private complexes of sensations. For the ‘left-wing’ oriented Neurath, this view was incompatible with the public and intersubjective character of science. Consequently, he advocated the view that protocols are accepted by the scientific community as reporting the results of publicly accessible observations. Protocol sentences, as Neurath maintained, must be expressible in thephysicalisticlanguage of unified science, which makes them, like all other sentences, revisable in principle. This view was strongly opposed by the ‘right-wing’ oriented