Prospective realism

HAROLD

I. BROWN*

PROSPECTIVE REALISM IN THIS PAPER I will develop and defend a view that I call ‘prospective realism’.

Devitt uses ‘weak realism’ to characterize the minimal view that ‘Something objectively exists independently of the mental’.’ Weak realism will serve as my starting point except that I will drop Devitt’s reference to ‘the mental’ and rephrase weak realism as the view that something exists independently of our attempts to know it; I will explain the reasons for this modification shortly. Prospective realism adds the following two theses to weak realism: (1) The attempt to learn the nature of these independent items is a reasonable aim for science. (2) Our ability to pursue this aim has been increasing through the development of progressively more sophisticated instrumentation. I have offered my own defence of the weak realist thesis elsewhere? thus I will take weak realism as given for the purposes of this paper and confine discussion to the additional claims of prospective realism. Let me emphasize at the outset that I do not maintain that realism provides the only end for science, or even the primary end. Science is a complex community endeavour in which different ends are pursued both simultaneously and seriatim.3 I am concerned only to defend the legitimacy of realism for those who choose to pursue it. I will confine discussion here to physical science. This will ease the exposition considerably, for we must be extremely careful in formulating even weak realism when we turn to biology or psychology. The thesis that there is a physical world that exists apart from human knowers is reasonably clear. Obviously, human biology does not exist apart from human beings, and psychological phenomena do not exist independently of ‘the mental’ although biology and psychology are legitimate subjects of scientific study. The attempt to extend prospective realism to these cases does not pose an intractable problem, but I will not pursue the issue here. The paper will be organized into three main parts: In Part I I will argue that the realist aim is prima facie reasonable and that those who oppose realism as an aim for science must bear the burden of showing that this aim is miscon*Department of Philosophy, Northern Illinois University, DcKalb, IL 601 IS, U.S.A. Received 28 January 1989. ‘M. Devitt, Realism and Truth (Princeton: Princeton University Press, 1984). pp. 15, 22. 2H. Brown, Observation and Objecriviry (New York: Oxford University Press, 1987). ch. 6.

‘For discussion see H. Brown, ‘Normative Epistemology and Naturalized Epistemology’, Inquiry 31 (1988). 69-74. Stud. Hiss. Phil. Sci., Vol. 21, No. 2, pp. 211-242, 1990.

Printed in Great Britain 211

0039-3681/90 s3.00 + 0.00 Q 1990. Pcrgamon Press plc.

212

Studies in History and Philosophy

qf Science

ceived. There are many who have happily taken up this burden; in Part II I will respond to the major current attempts to argue that the realist aim is not a reasonable one to pursue. In the course of our response we will find that these antirealist arguments do make several points that realists must face. Extracting these points will allow us to see more clearly what forms of realism are defensible and thus to sharpen the formulation of prospective realism. In Part III I will develop and defend thesis (2). I. Prima Facie Realism While my main aim in this part of the paper is to argue that realism provides a prima facie reasonable end for science, I want to begin with four preliminary considerations that will narrow the range of our discussion. First, since I will be arguing only that realism provides a reasonable scientific end for those who choose to pursue it, I will not consider arguments that seek to defend the cognitive legitimacy of nonscientific cultures. Note, however, that if realism is successfully pursued, its results are ignored only at the peril of those who choose to do so. Second, I will make no assumptions about the ultimate nature of the physical world or about the final shape of physical science. Instead, I will adopt Shapere’s thesis that successful science has proceeded in terms of distinct domains that are, to a significant degree, isolated from other domains.4 Questions such as whether any two distinct domains will eventually be merged, or whether all scientific subjects will ultimately be unified into a single domain, will be answered only by scientific research. I will seek to avoid importing unstated assumptions as to what we will find in a given domain by using ‘item’ as a neutral term that involves none of the connotations connected with ‘events’, ‘processes’, ‘things’, ‘ objects’ and so forth. In this terminology, realists seek to understand the nature of the items in their research domains. Third, Hacking has distinguished realism with respect to theories from realism with respect to entities ;5 for present purposes I will ignore this distinction. That is, I will take realism to be the thesis that science seeks true propositions, where some of these propositions assert that a specific item exists (e.g. a trans-Uranian planet) or that items of a particular type (e.g. electrons) exist. Such claims do not occur in science as isolated assertions. They have no scientific interest unless they are embedded in a body of theory that is sufficiently rich to permit the derivation of testable consequences of the existence of these items. I will, however, assume that scientists can agree that an item exists but disagree about its properties to some significant degree. ‘D. Shapere, ‘ScientificTheories and Their Domains’, The Structure of Scientific Theories, F. Suppe (ed.) (Urbana: University of Illinois Press, 1974), pp. 518-565. ‘I. Hacking, Representing and Intervening (Cambridge: Cambridge University Press, 1983). pp. 21-28.

Prospective Realism

213

Fourth, some comment is in order about my use of ‘truth’ although space does not permit a detailed discussion of this notion here. I will assume, without further discussion, a correspondence view of truth. That ‘is, I will assume that realists seek descriptions of physical items that exist apart from any of our cognitive relations to them. We must, of course, distinguish what it means to say that a proposition is true from the procedures we use to establish that a proposition is true. Discussion in this paper will be limited to the latter question. I have two arguments to offer on behalf of the claim that realism provides a prima facie reasonable cognitive end. The first argument consists of noting that, as a matter of simple curiosity, the desire to understand the world needs no defence - unless grounds can be presented to show that this end cannot be successfully pursued. In other words, those who maintain that the attempt to satisfy such curiosity is pointless or even harmful must bear the burden of proof. The second argument derives from practical considerations. Many antirealists hold that the only legitimate aim of science is some practical outcome such as the ability to make predictions or control nature.6 Now I have already noted that science need not be viewed as pursuing a single end. As a result, the common tendency to treat realist and antirealist ends for science as if they were incompatible alternatives is a mistake. In one respect these goals can be pursued simultaneously since a true account of the items in a domain provides the only fully reliable means of prediction and control. To see why, note that when we are engaged in a practical activity such as treating a disease or designing an airplane we are often concerned to make predictions about the outcomes of specific policies, and we want to make predictions that will turn out to be true. We want, for example, to predict correctly the outcome of a therapy or the ability of the joint between wing and fuselage to withstand the stress of repeated take-offs. Such predictions go beyond a description of our present circumstances and thus require inference. But the conclusion of an inference is fully reliable only if the premises are true. Whether an inference is deductive, inductive, analogical, or what have you, one way of casting doubt on its conclusion is by challenging the truth of its premises. We often do infer true conclusions from false premises and, in the absence of true premises, we regularly make predictions on the basis of correlations whose foundation and limits we do not understand. But while we often proceed in this way because it provides the only available approach, our conclusions would be more reliable if they were inferred from a true account of the items we are dealing with. A 61nstrumentalism was once the most common form of antirealism, where traditional formulations of instrumentalism held that certain types of scientific proposition have no truth-value. Many contemporary antirealists agree that the propositions in question have a truth-value, but argue that there are insuperable difficulties in determining that truth-value.

214

Studies in History and Philosophyof Science

true account may be beyond us in many domains, but this must be shown. Prima facie a true account is desirable, and those who maintain that this end is not worth pursuing must bear the burden of proof. II. Antirealism

Currently, many of the most influential arguments against realism derive from a variety of forms of relativism. Relativists argue that which scientific claims are put forward and the grounds we develop for accepting those claims are, in some deep sense, dependent on historical or cultural circumstances. This, in turn, is supposed to provide grounds for denying that science can achieve the trans-cultural, trans-historical knowledge that realists seek. Much of the challenge of relativism derives from such undisputed facts as that science came into existence in specific cultures at specific points in their history and that some cultures are better able to sustain scientific institutions than are other cultures -just as some cultures are better able than others to sustain specific religious or social structures. Thus scientific realists face the task of providing specific reasons for holding that scientific procedures allow us to establish results in a way that transcends the culture in which those results were put forward and accepted. As a result, relativism will provide a central theme of this part of our discussion - although I will also consider some arguments that seek to establish antirealism without relativism. I will discuss three types of antirealist argument: the historical induction, arguments from underdetermination, and arguments from incommensurability. 1. The historical induction The historical induction attempts to use evidence from the history of science as a basis for concluding that we have no adequate grounds for asserting the truth of any current scientific claims. Proponents of this argument hold that the history of science provides a continuing saga of theories that were once widely accepted on the basis of sound scientific evidence and were therefore taken to provide a true account of the items in their domains. Nevertheless, we now reject these theories as utterly false. Yet, if we look at the most successful current scientific theories we find that the evidence is of the same general type as the evidence that was invoked on behalf of those earlier theories. As long as we confine our attention to a current theory and the evidence in its behalf that evidence may seem impressive, and it may be difficult to imagine how a radically false theory could have provided such a striking array of correct predictions. But our predecessors drew the same conclusion on the basis of the evidence available to them, and we agree that their theories were false. This, it is argued, provides powerful inductive grounds for believing that currently favoured theories are also false.

Prospective Realism

215

The above argument may be deployed against a relatively strong form of realism or against a weaker form. Stronger forms of realism hold that contemporary theories either have achieved truth in their domain or have done so approximately; Laudan has developed the above argument in considerable detail against the stronger forms of realism.’ Weaker forms of realism hold only that scientific procedures are capable of leading us to truth but can remain agnostic on the status of current theories. Although I am concerned to defend only the weaker view in this paper, I will argue that the historical induction fails as an argument against stronger forms of realism, and thus a fortiori against prospective realism. One realist response to the historical induction is to challenge its historical basis;8 I will pursue a different strategy here. I will argue that the historical induction is an extremely weak inductive argument. Whatever account of induction we may ultimately accept, it will presumably include the following two necessary conditions for a successful inductive argument: the premises of the argument must provide (1) a sufficient number of cases that are (2) sufficiently similar to the situation described in the conclusion. Thus, to support the conclusion that we should not accept current scientific claims as true, proponents of the historical induction must provide a substantial history of cases with epistemic features that are sufficiently similar to the epistemic features of the current theories they would challenge. But if we look at the cases that are usually supplied, these two criteria work against each other. If we focus on cases that are closely similar in the relevant epistemic respects to current science, there are very few - not enough to make for an impressive inductive argument. When we attempt to widen our set of cases, we find proponents of the argument relaxing the similarity requirement. Let me develop this claim. The single most impressive example for proponents of the historical induction is provided by the fate of classical mechanics. Here we have a theory that was undoubtedly successful on every accessible measure of scientific success and that for generations provided the very paradigm of science. Nevertheless, this theory failed and failed radically. Let me emphasize that I am not going to pursue the common realist strategy of seeking some sense in which Newtonian mechanics was correct or partially correct. Rather, I agree that this theory embodies a radically incorrect image of the physical world. Moreover, even on an observational level classical mechanics yields wildly incorrect predictions for major portions of its domain, a result that was long hidden because of the limited range of available instrumentation. This point cannot be too strongly stressed: Classical mechanics as classically understood included all velocities in ‘L. Laudan, ‘A Confutation of Convergent Realism’, Phifosophy of Science 48 (1981), iW9. ‘See, for example, E. McMullin ‘A Case for Scientific Realism’, Scienrifc Realism, J. Leplin (ed.) (Berkeley: University of California Press, 1984). pp. 17-18.

216

Studies in History and Philosophy of Science

its domain with no restrictions on achievable velocities, and it included all material entities in its domain irrespective of size or mass. In fact, classical mechanics gives incorrect results for all dynamical situations and even for some static situations (e.g. determinations of total energy). Classical mechanics does give ‘sufficiently correct results’ for some purposes, but which results count as ‘sufficiently correct’ changes as our aims and our ability to measure change. However, impressive as this example is, one example does not provide much inductive basis for the conclusion that currently successful scientific theories are also false. A convincing inductive argument should display a series of such cases - and proponents of the historical induction have been well aware of this point. Much of the strength of the historical induction derives from the work of historians who reject the view that science came into existence in the seventeenth century. Instead, they argue, science as we understand that notion existed throughout the ancient and medieval periods, as well as in some nonwestern cultures. Antirealists such as Kuhn and Laudan regularly include these cases in their accounts of science. We can, however, accept these earlier views as genuine science and still hold that they are examples of science that are different from current science in ways that block their relevance for the historical induction. I want to discuss two important differences. First, consider the relation between classical physics and twentieth-century physics. Classical physics plays a variety of useful roles within contemporary physics, as well as in physics education. Even if we do not take this fact as an argument on behalf of the approximate truth of classical physics, we can still note that the older sciences that are supposed to widen our inductive base did not have this relation to their successors. For example, once Newtonian mechanics had been established there was no longer any role in physics for epicycles or the distinction between natural and violent motion. It was not necessary that physics students know anything about these older views, nor did those views provide a useful approximation even for a small range of cases. In a similar way, after Lavoisier phlogiston chemistry vanished, and a similar fate awaited the doctrines of crystalline spheres, humours, caloric, and many other older scientific views.’ Thus the relation between classical physics and its successors is significantly different from what occurred in the other cases of theory succession that are typically cited on behalf of the historical induction. It does seem that science underwent a particularly deep transformation in the ‘Kuhn cites one apparent exception: Surveyors still find it useful to treat the earth as stationary. T. Kuhn, The Structure of Scientific Revolutions, 2nd edn (Chicago: University of Chicago Press, 1970). p. 102. But there is a great difference between using a single hypothesis that is found in an earlier view and making detailed use of that view. Treating the earth as stationary requires only the optical relativity of motion, which can be established independently of pre-Copernican astronomy. There is nothing here analogous to, say, the procedure of writing down classical equations and systematically transforming them into quantum-mechanical equations.

Prospective Realism

217

seventeenth century, and this casts doubt on the legitimacy of using the fate of earlier theories as grounds for predicting the fate of post-seventeenth-century theories. Second, beginning with Galileo’s use of the telescope in 1609, scientists have been designing and deploying instruments that increase our ability to interact with the physical world and gather evidence that is relevant to theory evaluation. I will have much to say about such instrumentation in Part III. For the moment I want to suggest only that the introduction of such instrumentation brought about a qualitative change in the way science is pursued. Earlier views were still science in the sense that researchers put forth testable theories, evaluated these theories against the best available evidence, sought to solve research problems in their domains, and so forth. But all of this is compatible with the claim that modem instrumentation has changed the detailed pursuit of science in a way that should make us wary of drawing inferences from the fate of previous science to the fate of current science. I submit that the above two examples are sufficient to undercut the historical induction as a general argument against realism. Proponents of this argument just do not have enough sufficiently similar cases to make for a plausible inductive argument. There is another reason for rejecting the historical induction that will hold even if we were to accept the earlier cases into our inductive base. When we look at a specific current theory we often find that there is a substantial body of favourable evidence from the empirical tests of that theory. This evidence provides inductive support for the claim that the theory is true. In effect, proponents of the historical induction maintain that the historical evidence on behalf of the claim that successful scientific theories eventually fail overwhelms the observational and experimental evidence on behalf of, say, special relativity, or elementary quantum theory, or plate tectonics. Yet proponents of the historical induction have never carried out a comparative evaluation of the inductive support for the historically based claim that scientific theories are false, and the evidence that supports specific contemporary theories. In Part III I will argue that contemporary ‘theories have been tested much harder than their predecessors. If this is correct, then the evidence supporting some current scientific theories may well be much greater than the historical evidence against realism in general. ‘OAgain, the only point that I am after here is that the historical induction fails to provide inductive grounds for maintaining that all successful scientific theories are false, and thus fails to provide a reason for holding that the realist end cannot be successfully pursued.

“To my knowledge, the above argument first occurs in M. Levin, ‘On Theory-Change and Meaning-Change’, Philosophy of Science 46 (1979), 420-421. !3ee Brown, op. cit., note 2, pp. 217-2 18 for further discussion.

218

Studies in History and Philosophy of Science

2. Underdetermination

All viable forms of realism hold that scientific theories are ultimately to be evaluated on the basis of observational data. But, it is argued, the available data is never sufficient to dictate a unique decision on the status of a theory, while realism requires that we be able to arrive at a unique theory in each domain. I want to note three areas in which underdetermination occurs before offering a realist response.” Underdetermination appears, first of all, when we consider the grounds for accepting any universal generalization. The classical problem of induction turns on an underdetermination argument: Observation is supposed to provide the evidence that will justify accepting universal propositions, but the content of a universal proposition always goes beyond the evidence and it is not clear what constitutes sufficient grounds for accepting such a proposition. The defender of realism is thus placed under the obligation of providing and justifying some grounds for assessing when we have sufficient evidence for concluding that a universal proposition is true. The antirealist thrust of the problem of induction becomes even more pressing with Goodman’s ‘new riddle of induction’.‘* Goodman noted that we can always find alternative universal generalizations that yield different predictions for unexamined cases but that are equally well supported by the available data. Thus even if we could find the long-sought criteria for inductive evaluation of universal propositions, we would still lack suflicient grounds for the unique choice between incompatible alternatives that realism requires.” Underdetermination reappears when we invoke observational evidence to refute a universal proposition. As Popper noted long ago, refutation is logically cleaner than confirmation. But Popper’s attempt to use this point as the basis for a logical analysis of scientific method based on the refutation of individual hypotheses failed because - as Duhem, Quine and many others have pointed out - scientific propositions do not face observational data one by one, but only as members of a corporate body of propositions.14 Thus while refutations do show that something is wrong somewhere, there are always alternative points at which the thrust of modus tollens can be focused. Two consequences of this form of underdetermination are particularly troubling

“We will encounter a fourth form of underdetermination in Section 3. ‘*N. Goodman, Fucf, Fiction and Forecusr, 2nd edn (Indianapolis: Bobbs-Merrill, 1965). “Much of the debate over Goodman’s argument has turned on peculiarities of the artificial example he constructed. See H. Brown, Rationufity (London: Routledge, 1988), pp. 23-29, for examples from real science that make the same point. “We need not hold, with Quine, that all of science is at stake in every such case. It is sufficient that each purported refutation involve a substantial set of propositions - perhaps all of the propositions currently accepted in a given domain.

Prospective Realism

219

from a realist perspective. First, it seems to allow us to choose some propositions that we will permanently protect from refutation. These propositions will be treated as if their truth has been established, but they will achieve this status by convention, rather than because we have proven that they are true. Second, different scientists can presumably confer this status on different propositions, and observational data will be impotent to choose between these alternatives. Once again, the realist aim of selecting a single theory in each domain seems to be blocked. A third form of underdetermination arises when we consider our grounds for believing in the existence of nonobservables. Strictly speaking, accepting the existence of a nonobservable is a special case of accepting the truth of a proposition. But the case deserves special note because of the pervasive role that nonobservables play in modem science and in philosophical discussions of scientific realism. Ultimately our grounds for accepting scientific claims rest on observation. Thus theories that postulate nonobservables must have observable consequences and our grounds for accepting these theories derive from those observable consequences. But if this is the case, why should we believe any more than the observable consequences of these theories? The point becomes especially pressing when we note the possibility that different theories that postulate different nonobservables may yield the same observable consequences. Again, the present point. is that our belief in nonobservables is underdetermined by the relevant evidence. What are we to make of these underdetermination arguments from a realist point of view? I want to separate two different interpretations of the significance of these arguments. First, they can be read as arguments on behalf of relativism; I will reserve discussion of this interpretation for the next section. However, underdetermination provides an important challenge to realism even for those who reject relativism. I will devote the remainder of the present section to this challenge. Underdetermination is a genuine feature of scientific decision-making, and a viable form of realism must come to grips with it. One effect of underdetermination is to raise a question as to the degree of certainty we should expect from a scientifically adequate theory. Underdetermination arguments are sometimes used to show that alternative views are always logically possible given a finite body of observational evidence. This is undoubtedly correct, and it underlines an epistemic constraint on a viable realism. Realists must reject the demand that all logically possible alternatives have to be defeated before we have an epistemic right to believe a theory. As a result, realists must acknowledge that attempts to achieve the realist end are fallible and intrinsically risky. The two positive arguments for realism offered above provide some grounds for holding that these risks may be worth taking, and I will shortly offer reasons for thinking that these risks are unavoidable in any case.

220

Studies

in History

and Philosoph_v of Science

Unfortunately, accepting fallibilism does not solve the problem of underdetermination. Instead, it complicates our epistemic situation immensely. In effect, I have just conceded that logic and observation are not sufficient to decide the epistemic fate of scientific theories; we need some additional criteria. In fact, we must supplement logic and observation with methodological rules - but the problems we have been considering reappear with regard to methodological rules. As a number of philosophers have argued in recent years, the proper methodology for pursuing science depends on the nature of the world and on what sort of cognitive agent we are.15 But these must be discovered by the same procedures that lead us to accept scientific theories. In other words, the methodology of science is integral to science and our methodology will be underdetermined in the same ways that our science is underdetermined. Indeed, the two most troubling antirealist consequences of underdetermination the possibility of permanently protecting some favoured propositions and our ability to find significantly different theories that are equally well supported by the observational data - are increased if our manoeuvring room includes theability to alter our methodology. Thus an adequate defence of realism will have to provide a realist account of methodology. In addition to accumulating enough evidence to select a single theory in a domain, we will also have to accumulate enough evidence overall to select a methodology - and we will have to do this in a way that does not amount to a self-serving choice of methodology on the part of those who prefer a particular theory. Further discussion of the status of methodology is best deferred to the next section since the relevant points will be sharpened by our discussion of incommensurability. In the remainder of this section I want to consider two antirealist positions that are deeply dependent on underdetermination arguments: those of van Fraassen and Laudan. Van Fraassen’s viewsi have been the subject of extensive discussion and it will not be possible to survey that debate here. I want only to underline the role of underdetermination arguments in his attack on realism. Van Fraassen’s antirealism is primarily aimed at nonobservables. He agrees that scientists make use of theories that postulate nonobservables and that some such theory might be true, but he contends that we are justified in believing only the observable consequences of these theories. Much of van Fraassen’s argument against realism seeks to make two points: (1) available realist attempts to

ISFor example C. Hooker. A Realistic Theory of Science (Albany: SUNY Press, 1987); L. Laudan. Science and Values (Berkeley: University of California Press, 1984) and ‘Progress or Rationality? The Prospects for Normative Naturalism’, American Philosophical Quarterly 24 (1987). 19-31; D. Shapere, Reason and rhe Search for Know/edge (Dordrecht: D. Reidel, 1984) and ‘Method in the Philosophy of Science and Epistemology’, The Process of Science, N. Nersessian (ed.) (Dordrecht: Martinus Nijhoff, 1987), pp. l-39. lbB. van Fraassen. The Scienfifc Image (Oxford: Oxford University Press, 1980).

221

Prospective Realism

justify belief in nonobservables do not succeed; (2) the uses scientists make of theories that postulate nonobservables require belief only in a theory’s observable consequences. If successful, these arguments would leave the status of nonobservables unresolved. Van Fraassen’s major direct argument for withholding belief in nonobservables is an underdetermination argument: Theories that postulate nonobservables have excess content that goes beyond their observable consequences, but the scientific justification of those theories is based solely on observable consequences. Thus we have no justification for believing any more than these observable consequences. Hooker has argued that van Fraassen’s position is based on a strategy of avoiding epistemic risk.a” Belief in nonobservables involves greater epistemic risk than is involved in restricting belief to observable consequences. But, van Fraassen argues, science can proceed without taking these risks and they should thus be avoided. Realists can immediately respond that these risks may be well worth taking. It is highly likely that a significant portion of the universe is beyond the range of our senses - a point with which van Fraassen agrees and some of us are curious about that realm. Van Fraassen has not shown (nor attempted to show) that this curiosity cannot be satisfied, and those who share this curiosity may consider the additional risk of error to be worthwhile. To be sure, any claims to have discovered the truth about some portion of this realm will be fallible, but we have already seen that realists must be fallibilists. In addition, it is not clear that van Fraassen’s epistemic programme is less risky than the realist programme. Van Fraassen acknowledges that theories including theories that postulate nonobservables - play a number of central roles in science. Scientists rely on these theories for, among other things, designing instruments and developing programmes of research. Hooker has argued that this leaves us with an approach that is no less risky than the realist approach,‘8 and Melchert has argued that van Fraassen’s antirealism collapses into realism.” The pragmatic argument on behalf of realism that I offered in Part I strengthens the present point. For any inferences drawn from claims that are not in fact true are ipso facto risky. Thus we have a vital pragmatic interest in discovering true theories and replacing any false theories that we hold with true theories. It may turn out that we cannot achieve this end, but van Fraassen has offered no argument for this conclusion and thus no argument to show that the pursuit of the realist end is pointless.

“C. Hooker, ‘Surface Dazzle, Ghostly Depths’, Images o/ Science, P. Churchland and C. Hooker (eds) (Chicago: University of Chicago Press, 1985), pp. 153-196. “l&f.. pp. 169-170. lPN Melchert ‘Why Constructive Empiricism Collapses into Scientific Realism’, Ausfralasian Joumbl of Philoiophy 63 ( 1985), 2 13-2

15.

222

Studies in History and Philosophyof Science

Finally, it is not clear just how great any additional risks entailed by realism must be - in the long run. It is presumably riskier to believe on the basis of skimpy evidence than on the basis of rich and varied evidence. But if this is the case, the magnitude of the risk involved in believing a proposition depends on the range of observational evidence we can accumulate. I want to develop the significance of this point by considering an additional argument that van Fraassen sometimes uses against realism. XJVan Fraassen advocates a probabilistic view of observational support. On this view, observation statements are deduced from premises that essentially include the hypothesis in question; confirmation of these observation statements increases the probability of the hypothesis. But the theory of probability now guarantees that the hypothesis can never be more probable than the evidence. Moreover, since scientifically interesting claims about nonobservables always have greater content than the observations that support them, these hypotheses will always be less probable than the supporting observations. ‘Credibility varies inversely with informativeness” and believing a theory’s claims about nonobservables is a gratuitous move that has no epistemic legitimation. Note, however, that while van Fraassen is undoubtedly correct in holding that it is always riskier to believe a hypothesis than to believe its observable consequences, this tells us nothing about how risky it is to believe that hypothesis. Presumably, in an adequate probabilistic confirmation theory the probability of a hypothesis will be a function of the amount and variety of available evidence. But then the magnitude of the additional risk will depend on our ability to gather evidence. In Part III I will argue that our ability to gather evidence has been increasing dramatically. If that argument is correct, then we may be facing a situation of declining epistemic risk in accepting some claims about nonobservables. Note also that just how high a probability is sufficient for belief is a matter to be decided in establishing our methodological rules.22 But this methodological decision is under at least some degree of empirical control. For example, if we judge that we have been believing too many theories that we later discover to be false, we might raise the probability value required for belief. Or, if we find that our present methodology leads us to withhold belief from a number of theories that have continued to pass stringent observational tests, we might decide to lower the probability threshold for belief. Moreover, our empirical data-base for evaluating methodological rules grows as our scientific experi*OForexample B. van Fraassen, ‘Empiricism in the Philosophy of Science’, Images of Science, P. Churchland and C. Hooker (eds) (Chicago: University of Chicago Press, 1985) p. 247. “Ibid., p. 280. “R. Jeffrey points out that we may accept different probability values as adequate for belief in different domains. See ‘Probability and the Art of Judgment’, Observarion. Experiment, and Hypothesis in Modern Physical Science, P. Achinstein and 0. Hannaway (eds) (Cambridge: MIT Press, 1985), p. 103.

Prospective Realism

223

ence increases. The present size and vigour of the scientific community suggests that this data-base is growing faster than ever before. Thus the epistemic risks involved in accepting methodological rules may also be declining. Laudan’s antirealism is not just aimed at nonobservables. Laudan holds that the attempt to discover any universal truths through science is pointless because we have no way of determining whether we have achieved that aim.23 Again the argument rests on the underdetermination of scientifically interesting truths by the data, and again the possibility of a realist response turns on how high a standard of justification is required and on how high a standard of justification we can achieve. 3. Incommensurability The incommensurability thesis holds that the history of a science displays a sequence of radically different fundamental theories with no objective basis for a comparative evaluation of these theories. Traditionally it was held that observational data and methodological rules provide the grounds for theory choice, but in recent years it has been argued that neither observation nor methodology can do the job since these are dependent on the theories under evaluation. I want to develop this claim and consider its consequences for realism. One preliminary point will set the stage for this discussion. Every scientific theory embodies a system of concepts and different fundamental theories embody different concepts. These concepts, like the theories themselves, are human creations; nothing in our experience imposes a unique set of concepts on US.*~Thus when we choose a fundamental theory, we are also choosing a conceptual system. Now realists hold that one aim of science is to find the set of concepts that correctly describes the items in each domain. But the incommensurability thesis entails that there is no way of choosing between conceptual systems that will provide grounds for holding that the selected system of concepts describes items that exist independently of that theory. I want to develop the main arguments for this conclusion. The central argument to show that observation is theory-dependent turns on the point that our senses do not, by themselves, provide the evidence required to evaluate a theory because perceptual data has no cognitive significance until uFor example L. Laudan. Progress and its Problems (Berkeley: University of California Press, 1977). pp. 125-126; Science and Values (Berkeley: University of California Press, 1984). pp. 53, 137. Tontemporary relativists are particularly fond of this point. See, for example, B. Barnes, T. S. Kuhn and Social Science (New York: Columbia University Press, 1982), pp. 22-24; D. Bloor, ‘Durkheim and Mauss Revisited: Classification and the Sociology of Knowledp’, &dies in History and Philosophy of Science 13 (1982), 267-297, and Wittgensrein: A Social Theory of Knowledge (New York: Columbia University Press, 1983). pp. 38-41; A. Pickering, Constructing Quarks (Chicago: University of Chicago Press, 1984) pp. 405-407.

224

Studies in Histor_v and Philosophy of Science

it has been conceptualized. We can, of course, look at the dial of a laboratory instrument and note that it is pointing to the five. This is, in an important sense, an observation, and it requires no significant scientific background. But this observation cannot confirm or disconfirm any scientific theory until a host of questions have been answered. These questions include, among others: What units are to be attached to the number read off the dial? How accurate is this instrument? Is this accuracy sufficient for the test in question? Why is this an appropriate instrument for testing this particular theory? The answers to all of these questions depend on the currently accepted body of science.‘5 Clearly. this situation raises the specter of underdetermination once again: If an observation appears to contradict the theory under test, we can protect that theory by questioning other aspects of the scientific background of the experiment. But an additional problem arises in this case. For in order to answer the above questions we must, first of all, describe the reading from the instrument in terms of the concepts embodied in the theory we are testing. The point is particularly clear if confirmation and disconfirmation derive from logical relations between ‘observation sentences’ and ‘theoretical sentences’. No such logical relations will obtain unless the observation sentences are couched in the language of the theory under test. Suppose, however, that we are attempting to choose between two competing theories that embody different conceptual systems. Proponents of each theory will conceptualize the data using the conceptual system of their preferred theory. But we now have two different bodies of data - one for each theory - rather than a single body of data against which the two theories can be compared. Furthermore, even if we could isolate a relevant body of observational data that is independent of our theories, we would still not have enough for objective theory choice. Given a theory and a body of data, we need criteria for deciding whether the data supports the theory, contradicts it, or whatever. These criteria are supplied by methodological rules, and we have already noted that the methodology of science is discovered as science develops. This will not cause a problem as long as the methodological rules used to evaluate a theory are independent of that theory, but new fundamental theories provide a major source of methodological innovation. For example, before the advent of quantum theory, the demand for causality served as a central methodological rule for science. Quantum theory is noncausal and the acceptance of quantum theory thus requires that this rule be relaxed. Some physicists, such as Einstein and contemporary hidden variable theorists, have taken this noncausality as grounds for holding that there is something wrong with quantum theory.

“The classic statement of this point is P. Duhem. The Aim and Structure q/Physical Theorv, P. Wiener (trans) (New York: Atheneum, 1962), p. 145. See C. Hooker, op. cit., note IS. ch. 4 for a particularly detailed analysis.

Prospective Realism

225

Other physicists take the success of quantum theory as grounds for rejecting the demand science always seek causal theories. The arguments on each side of this debate certainly appear to be circular. Rather than evaluating two competing scientific theories by appeal to methodological rules that are independent of those theories, each side evaluates a key methodological rule on the basis of its preferred theory. Proponents of the incommensurability thesis argue that this is what typically occurs when a choice must be made between fundamental theories, and thus that methodological rules do not provide a theory-free basis for choosing between such theories. Thus the fact that a theory fares well given its interpretation of the data and its preferred set of methodological rules provides no grounds for holding that the theory is true in the sense that realism requires. Kuhn, for example, concludes: ‘There is, I think, no theory-independent way to reconstruct phrases like “really there”; the notion of a match between the ontology of a theory and its “real” counterpart in nature now seems to me illusive in principle’.26 One more wrinkle, deriving from recent work on the sociology of science, will complete this relativistic vision. Scientists do choose between fundamental theories. If the above arguments are correct, observation and methodology cannot account for these choices; some other factors must be invoked. The determining factors, it is argued, are social: Scientific theory choice is ultimately determined by features of the historical period, culture or subculture in which the scientists are working, class considerations, and such. But if this is the case, then scientific theories are cultural products to be set alongside, say, religious beliefs or artistic styles rather than the result of some contact with a reality that transcends all cultures. We may accept a scientific theory, but we cannot bring forth any reasoned argument to show that one theory has a better grasp on truth than another. I want to note an historical point before I tackle these antirealist arguments directly. During the first half of this century logical empiricism provided a detailed development of the view that observation and methodology provide the grounds for scientific theory choice. Observations, it was held, simply occur to us independently of any of our theories or beliefs, while methodological rules were taken to be logical rules and thus known a priori. The incommensurability argument was posed in the course of a reaction against logical empiricism and is aimed at undercutting just these two presumed foundations of scientific evaluation. But empiricism is a notoriously antirealist position. Thus, even if the above arguments succeed in undercutting empiricism, we should not accept them as a refutation of realism without further consideration. Indeed, before attempting a realist response to these arguments I will concede their *60p. cit.. note 9, p. 206. This passage is from Kuhn’s ‘Postscript’. which was written after due consideration of his critics.

226

Studies in History and Philosophyof Science

central points. That is, I agree that epistemically relevant observation takes place only in terms of some system of concepts, and that these concepts are human products generated in specific historical and social contexts. Similarly, I agree that our standards of epistemic evaluation develop as science develops, and that major theoretical changes often bring about changes in methodology. But, I will argue, these concessions do not settle the question of whether we are able, as science proceeds, to formulate and justify theories and cognitive standards that transcend the culture in which they appeared. To be sure, if realism is defensible we must find some grounds for evaluating scientific theories that are independent of our current scientific beliefs; ideally, these grounds should be independent of the content of any specific scientific theory. But, I will argue, this demand can be met if we properly understand the nature of observation and the epistemic status of evaluative criteria. Let me begin working my way towards this conclusion by noting that if the realist project is to have any chance of success, we must be able to achieve two kinds of access to items that exist apart from any of our theories. We require epistemic access: we must have some means of assessing whether a theory succeeds in describing the items in its domain. But before this issue can arise, we require conceptual access: we must be able to form concepts that can serve as descriptions of items that exist apart from our cognitive relations to those items. Since the incommensurability thesis denies both kinds of access, we can begin with the more fundamental issue of conceptual access. There is a long history of arguments that seek to show that such conceptual access is impossible. A full discussion is beyond the scope of this paper, but I want to indicate briefly why the problem is far from insurmountable. Our problem is to find an account of the way in which concepts are specified that is compatible with realism. Discussions of concept formation have largely turned on two positions: the empiricist view that concepts are ultimately specified by correlation with experience and the thesis that concepts are implicitly defined through the internal structure of the theories in which they occur. Concept empiricism is clearly antirealist. Realism must allow for the possibility that the physical world includes items that are utterly different from anything we have experienced; if our concepts are limited by the range of our experience, then we cannot form concepts that will allow us to describe such items. Definite descriptions will allow us to refer to such items as, for example, the cause of a particular sensation, but these descriptions do not introduce any new conceptual content. Our concepts of trans-empirical items remain parasitic on those concepts that refer to items we experience. As a result, contemporary realists generally reject the empiricist view of concept formation - but so do contemporary relativists. The latter typically hold that concepts are specified by implicit definition. Now Newton-Smith has argued that implicit definition entails that each theory contains distinct concepts and that this is

Prospective Realism

227

suthcient to yield incommensurability and block realism.*’ One realist response (which Newton-Smith adopts) is to opt for the Kripke-Putnam view of reference and to hold that realism requires only a theory of reference, not a theory of concept formation. Space does not permit discussion of this option here, but I want to suggest that we should not concede the implicit-definition view of concept formation to the antirealists. To see why, recall that I am currently concerned only with conceptual access, that is, with our ability to form concepts that describe a trans-empirical realm. But if concepts are determined by their internal relations - i.e. if experience plays no role in specifying our concepts - then there is no reason to hold that the range of application of these concepts is limited by what we experience, and thus no reason for denying conceptual access to items that transcend experience. In a similar way, we can agree that concepts are formed by individuals living in a particular culture. But it does not follow that these concepts cannot also describe items that are independent of that culture. Thus realists have available at least two approaches to the problem of conceptual access.2s The crucial issue for scientific realists is epistemic access - can realists provide grounds for holding that some conceptual system does in fact describe the items in some domain? The remainder of this paper will be concerned with the problem of epistemic access. It will be helpful to distinguish two different questions that arise at this point: (1) Are there any grounds for a rational choice between competing fundamental theories? and (2) Assuming an affirmative answer to question (1) do these principles of choice provide grounds for holding that the preferred theory is true? A rational choice between two theories must be mediated by some beliefs that are held in common by proponents of both theories. The incommensurability thesis denies that there are any such mediators in cases of fundamental debate, but I have argued elsewhere that when we examine actual

2W. Newton-Smith, The Roriondity of Science (London: Routledge & Kegan Paul, 1981). pp. 9-13. *‘A third approach is provided by the work of Wilfrid Sellars. On the empiricist view, only sensory experience can specify concepts. On the implicit-definition view, concepts are who//J specified by relations to other concepts. On Sellars’ approach, scientific concepts are determined by both implicit definitions and ties to the extra-conceptual world - but not to sensations. Sellars has also provided a powerful account of how new concepts can be introduced by analogy. In H. Brown, ‘Sellars, Concepts and Conceptual Change’, Synfhese 68 (1986). 275-307, I argue that Cellars’ account providestools for systematically comparing conceptual systems that are similar but not identical. The ability to compare similar conceptual systems undercuts Newton-Smith’s argument that implicit definition entails incommensurability. Note also that Mars has developed his views on concepts while defending scientific realism.

228

Studies

in History

and Philosoph?,

of Science

cases we find that this is false. 29I will not repeat those arguments here. and I will take it as established that question (1) receives an affirmative answer. The story with respect to question (2) is more difficult. As Boyd notes, a debate may be capable of a reasonable resolution on the basis of common principles without this providing grounds for a realist interpretation of the outcome.30 For example, on a Kantian view universally shared concepts and principles provide the common ground for resolving all scientific debates, but do so in a way that explicitly blocks a realist interpretation of the conclusion. Turning to a more recent example, one of Laudan’s goals in developing his ‘reticulated model’ of theory choice was to argue that there is always a sufficient common basis to assure a rational resolution of scientific debates, but this argument is developed in the context of an antirealist account of science.” Realism requires more than just reasonable grounds for preferring one theory to another. Realism requires specific grounds for holding that a preferred theory correctly describes the items in its domain. The key step in meeting this realist demand is to note that physical theories do not make claims about themselves. Rather, physical theories make assertions about items that are presumed to exist independently of our theorizing. We can pursue the realist end by testing theories against those items and allowing those items, so to speak, to enter into the discussion. This proposal has a number of virtues that are well worth exploring. Note, first, that on this proposal the items that limit theorizing are indeed independent of our beliefs. Moreover, the proposal provides an immediate response to the worry that theory testing must be biased whenever we use the theory under evaluation as a basis for conceptualizing our data. For no matter how deeply a theory may be involved in the design and interpretation of an observational procedure, as long as that procedure brings us into contact with items that exist independently of the theory, those items need not behave in the expected manner. Antirealists sometimes make extreme remarks about our inability even to notice outcomes that violate our expectations, but this is far *‘See H. Brown. ‘For a Modest Historicism’, The Monist 60 (1977). 54&555; ‘Incommensurability’. Inquiry’ 26 (1983), 3-29; Rationaliry (London: Routledge. 1988) pp. 207-224. L. Laudan. Science and Values (Berkeley: University of California Press, 1984). develops a similar approach to rational theory choice. Kuhn insists that he never intended ‘incommensurability’ to be interpreted as the claim that there are no common elements mediating scientific revolutions but only as the claim that the common elements are not sujicient to dictate theory choice: see, e.g.. ‘Theory Change as Structure Change’, Historical and Philosophical Dimensions of Logic, Methodology and Philosophy of Science, R. Butts and J. Hintikka (eds) (Dordrecht: D. Reidel. 1977). pp. 289-309. Kuhn has also urged that the proper conclusion to draw at this point is that there IS something wrong with the demand that rational choice must be dictated by such common elements: see, e.g.. ‘Reflections on My Critics’, Criticism and rhe Growth of Knowledge, 1. Lakatos and A. Musgrave (eds) (Cambridge: Cambridge University Press, 1970), pp. 234-235, 263-264. ‘OR Boyd ‘The Current Status of Scientific Realism’, ScienriJic Realism, J. Lephn (ed.) Berkeley: University df California Press, 1984), pp. 53-54. “L. Laudan. op. cit.. note 29.

229

Prospective Realism

from the scientific norm. Recall that many antirealists use underdetermination arguments to show that theories need not be abandoned in the face of anomalies - but this surely requires that anomalies occur and are recognized as problematic. If comprehensive scientific frameworks literally determined what scientists observe, there would be no anomalies.32 On the other hand, if observation involves interaction with items that exist independently of any theory, then we can understand why anomalies occur. Theory may guide observation in a number of ways - it may tell scientists where to look, what instruments to use, how to interpret the resulting data, and much more - but theory cannot predetermine the outcome of an observational procedure. In other words, the fact that observation sometimes yields anomalies is sufficient to show that something beyond our theories can provide an epistemic challenge to those theories. And this is enough to show that science does have epistemic access to items that are independent of our theories, and that these items provide some constraints on what theories are acceptable. We have here the first step towards a defensible realism, and while it is far from the entire story, it does provide a basis for a clearer statement of the task that remains. Realism may not be viable in every domain, but ifrealism is to be viable at all,

there must be some domains in which observation can provide enough constraints to select a single theory. Before we consider whether there are any reasons for

thinking observation can do the job, there is one remaining challenge from the incommensurability thesis that we must discuss. Even if we can form the required concepts and develop a large body of observational constraints on our theories, the problem of standards of evaluation remains. If those operating from different theories may arrive at incompatible conclusions - even on the basis of the same evidence - because they evaluate the bearing of that evidence on theory differently, then realism cannot be sustained. Now this objection turns on the thesis that we have no a priori grounds for accepting a single set of epistemic standards that must be applied in all cases of theory evaluation. Instead, every feature of science, even evaluative criteria, ultimately derives from decisions made in some intellectual tradition. I think this thesis is correct, but if we take it seriously, we can blunt the force of the objection. For if evaluative standards are part of science and are developed as science develops, then the line between methodology and substantive scientific claims is not as sharp as had traditionally been assumed. As a result, evaluative standards are capable of observational evaluation. But then the possibility that theory change may bring about a change of methodology is a virtue not a problem - for learning the appropriate methodologies ‘*Cf. R. Boyd, op. cir.. note 30, p. 60,

and ‘Observations, Explanatory Power, and Simplicity’,

Observation,Experimentand Hypothesis in Modern Physical Science, P. Achinstein and 0. Hannaway (eds) (Cambridge: MIT Press, 1985). p. 49.

230

Studies in History and Philosophy of Science

is part of the process of learning about the world. I want to cite just three examples to underscore these points. First, conservation of parity once provided a criterion that was used to evaluate the acceptability of physical theories but the scope of this criterion has been limited as a result of empirical developments within physics.j3 Second, the demand that scientific claims be evaluated against observation has been a constant feature of science throughout its history, but what counts as an observation has changed substantially as science has developed. To take but one example, the rejection of simultaneity at a distance in special relativity places constraints on observation that were not invoked by earlier science. We cannot, for example, determine the shape of a moving item by looking at it through binoculars. Third, we have already noted that the demand that scientific theories be causal theories has been challenged by quantum theory. Moreover, it is simply misleading to suggest that some physicists rejected the demand for causality because they wanted to accept quantum theory, and rejecting causality allowed them to do this. Rather, we have here a theory that has been amazingly successful in meeting observational tests and the rejection of causality is a central feature of this theory as it is currently understood. As a result, the evidence for quantum theory as a whole is also evidence against universal causality in the physical world. Thus the relaxation of the demand for universal causality is not just a self-serving methodological decision on the part of those who wish to accept a particular theory. Rather, it is part of a comprehensive theoretical response to an impressive body of evidence. Of course, the issue has not been eternally closed. New observational or theoretical results may lead to further reconsideration of the causal principle -just as they may lead to a reexamination of quantized energy levels. The upshot of this discussion is that the inclusion of evaluative standards in the body of science brings those standards under observational control and thereby eliminates any special problem generated by acknowledging that scientific standards are subject to change as science develops. To be sure, this view of methodology provides a wider range of options to those who would defend a favoured thesis in the face of unwelcome observations than would be the case if methodology were determined a priori. But if the evaluation of methodology is part of the process of evaluating scientific theories, then discovering the correct methodology in a domain is part of the problem of discovering the true theory of that domain. This returns us to the key point of the present discussion: Our ability to achieve the realist end depends ultimately

‘Cf. A. Franklin, The Neglect of Experiment p. 36.

(Cambridge: Cambridge University Press, 1986).

231

Prospective Realism

on how strongly theory choice is constrained by observation. I turn, finally, to that question.34 III. Prospective Realism 1. Perception At some level, our ability to gather evidence about the physical world depends on our perceptual abilities; our next project is to consider how perception provides evidence for physical theories. Historically, there have been three types of philosophical theories of perception: phenomenalism, direct realism, and indirect realism. Phenomenalism had a long run as the dominant theory and its failings are well known. For our purposes, the most important of these failings is that phenomenalism fails to account for either conceptual or epistemic access to items that exist apart from perceptual experience - items that may have properties quite different from anything that appears in our experience. Phenomenalism has few current defenders; direct realism is now the dominant view among epistemologists. Direct realists hold at least the following two theses: (1) There is a world that exists apart from our awareness of it; (2) Perception involves a ‘direct’ relation between the perceiver and items in that world. The first thesis is common to direct and indirect realism, but the story is more complex with respect to the second thesis. Traditionally, indirect realists held that perception is a triadic relation between a physical item, a perceiver, and some third item - variously characterized as an idea, sensation, sensum, or sense-datum. Perception makes us directly aware only of this third item, which was held to be internal to the perceiver’s mind and private. But this third item was also supposed to stand in some relation to items in the physical world, and thus give us indirect epistemic access to the physical world. In effect, indirect realists attempted to combine the central theses of phenomenalism and direct realism, agreeing with phenomenalists that we are directly aware only of private mental items, but holding that these mental items somehow serve as a source of knowledge of public physical items. Not surprisingly, this attempt to combine physical and nonphysical entities is unstable; since Berkeley’s attack on Locke there have been relatively few indirect realists. Phenomenalists and direct realists agree that if perception is to serve as a source of knowledge, perception must be a relation between a perceiver and one other item; they disagree on which of the remaining items from the indirect realist account should be rejected. There is much more to be said about our choice of methodology. For example, similar methodologies can be often used in different domains, and the evidence for evaluating a methodology can thus be considerably broader than the evidence for evaluating a specific theory. N. Rescher develops this last point in some detail: see MethodologiculPragmatism(Oxford: Basil Blackwell, 1977). Further discussion of this issue would take us too far from the main topic of the present paper. 21:2-o

232

Studies in History and Philosophyof Science

At this point it becomes crucial that we clearly distinguish ontological issues from epistemological issues. Contemporary direct realists are, by and large, materialist-naturalists who deny that perception involves any nonphysical items. I will accept this ontological thesis here without further discussion.35 But even if we deny that there are ontological intermediaries between the perceiver and the physical world, it does not follow that there are no epistemological intermediaries. That is, even if no third entity stands between the perceiver and the item perceived, it does not follow that perception provides a direct revelation of the nature of that perceived item. Moreover, there are substantive reasons for maintaining that perception does not directly reveal the physical world. At the very least, the old arguments from nonveridical perception show that physical items are not always as they appear to be. More importantly, if the physical world is anything like the world described by contemporary physics, then perception does not show us what physical objects are like. Thus even if perception brings us into direct contact with items in the physical world, and even if perception provides our only source of information about those items, perception provides only indirect epistemic access to items in the physical world. There is still a great deal of work to be done in moving from physical items as we perceive them to an understanding of the properties of those items. It will be useful to restate this point in Dretske’s terminology.36 Granting that perception involves a flow of information from the physical world to the perceiver, and that this information is independent of our theories or beliefs, it does not follow that this information is available in a form that allows us simply to read it off from perceptual experience. There is an obvious analogy here to a coded message or to a text in a language that we do not understand. Both of these carry information in Dretske’s objective sense, but if we wish to understand that message we must learn how to extract this information. Thus direct realism leaves the central epistemologicul problem of perception unanswered: How do physical items, as they appear to us, allow us to evaluate claims about the actual properties of those physical items? We can move towards an answer to this question - and also strengthen our grounds for holding that our epistemological relation to the physical world is indirect - by raising another question: What is the relation that holds between the perceiver and a physical item when perception occurs? From a naturalist point of view there is only one acceptable answer - it must be a causal relation. Let us, then, examine some central features of causal relations, without attempting a complete account. I’The adverbial theory provides one way of dealing with the ontology of perception that is compatible with the following discussion-of the epistemology of perception. For discussion see Brown, op. cit., note 2, pp. 122-129, 153-157. j6F, Dretske, Knowledge and the Flow of Inform&on (Cambridge: MIT Press. 1981).

Prospective Realism

233

2. Causality One central thesis of this section is that the Humean view of causal relations as holding fundamentally between pairs of events does not provide the basis for a scientifically adequate account of causation. Instead, I will take the caSe of two interacting physical items as the paradigm of a causal interaction. When two physical items interact, some properties of each of these items will, in general, be changed and a scientifically interesting account will be concerned with these changes. Let me develop the point of this kind of account, beginning with Hume’s favourite example: A moving billiard ball, A, hits a stationary billard ball, B, and B begins to move. Now the mere fact that B moves is of little scientific interest. A physicist will want, at a minimum, to determine B’s velocity. Ignoring friction and B’s brief acceleration, this velocity is strictly determined, given conservation of energy and momentum, by B’s mass along with A’s mass and velocity and the elasticity of the two balls. This rough initial example will serve to illustrate a number of important points. First, A’s velocity will also be changed as a result of the interaction; a complete description of the outcome of this interaction will include changes in both A and B. In a given case, we may pay attention to only a part of this total outcome because that is what interests us. When we focus our attention on B’s new velocity we describe A as the cause of that velocity. If we were interested in A, we could have described B as causing a change in A’s velocity. Strictly speaking, two physical items interacted, and some properties of each (e.g. velocity, momentum, for a time acceleration) were altered. The characteristic feature of a causal interaction is that all these changes are rigidly determined by properties of the items that interact in conjunction with some laws of nature. Second, only a subset of an item’s properties will typically be involved in a particular causal interaction. For example, each of our billiard balls has a certain reflectance and magnetic permeability, but these do not enter into the causal interaction of the above example. These properties would certainly enter into other causal interactions. Third, given that the final states of the two balls are determined by the initial states of these balls, the final states carry information about the initial states.37 Given a sufficient body of background knowledge, we can sometimes extract that information from a study of the final states. In this rather simple case, two kinds of background knowledge are required: knowledge of the relevant laws of nature38 and knowledge of the final states of the items involved in the interaction. In our example, the relevant laws are conservation of momentum

“I am only maintaining that causality provides a sufficient condition for information mission, not that it provides a necessary condition. UFor present purposes I will make no distinction between laws and theories.

trans-

234

Studies in History and Philosophy of Science

and energy. If I was unaware of these laws I could not begin to back-figure the initial state from the final state, while if I accepted a different set of laws I would (in general) arrive at different results for that initial state given a specific final state. (Consider, for example, the velocities and distances we would attribute to quasars had they been detected when astronomers still accepted the pre-relativistic formula for the Doppler effect.) The second kind of background knowledge we require comes from our examination of the specific case before us. A slightly more complex version of our example will help bring out the relevant points. In this new version we will drop the requirement that B is initially stationary, and we will not make any assumptions about the direction of motion of the balls other than that they collide. However, we will continue to simplify our problem by assuming that the collision is completely elastic, that masses do not change with velocity, and that the billiard balls remain on the surface of the table, although we will ignore friction. Assume, then, that we want to know A’s mass and initial velocity, and that these were not determined before the collision. Whether we can determine A’s mass and initial velocity depends on how many of the remaining parameters are available. Suppose, to begin with, that A is now inaccessible but that we can examine B both before and after the collision. (Perhaps we intentionally interposed B in A’s path in order to study A.) Our two laws of nature provide three equations: two for the components of momentum and one for energy. We have ten unknowns in these equations: two components of the initial velocity of each ball (giving four unknowns), two components of the final velocity of each ball (another four unknowns), and the mass of each ball. Our examination of B allows us to determine five of those unknowns: two components each for B’s initial and final velocities, and B’s mass. We do not have enough data to determine A’s initial velocity or mass, but we could improve our situation if we could increase the range of data available. In this simple case, if we could find a way of detecting A’s final velocity, we would have enough data available to compute A’s initial velocity and mass. Contemporary scientific cases are rarely this simple, but we do have here a first glimpse of the process by which scientists learn properties of an item, I, that is not directly accessible by studying properties of some other item that has interacted with Z and that is accessible. Sometimes we can study outcomes of causal interactions and use the data derived from this study, in conjunction with accepted laws, to back-figure properties of the items involved in causing those outcomes. In a typical scientific case the procedure will look rather different since scientists will rarely be able to determine the causes that concern them by plugging data into a set of equations and solving those equations. Typically, the argument will be cast in hypothetico-deductive form: a cause is postulated and it is then argued that this postulate provides the missing

Prospective Realism

235

premises needed to.account for the data. But this is somewhat misleading since scientists will not conclude that they have postulated the, correct cause until they can show that the account in question is the only acceptable account. Once this has been done, the full argument can be put into deductive form with the claim that a particular item was involved in causing the outcome before us as the conclusion. Dispute about the conclusion need not be closed off, but the dispute will focus on the acceptability of specific premises.39 Fourth, a causal interaction will often produce effects that do not copy any of the properties of the items that were involved in that interaction - even though the effect will be completely determined by those items. In this regard our billiard ball example is a bit misleading. It is not completely misleading because the final velocities of the balls need not be equal to any of their initial velocities. We can extract information about some of the initial velocities from the final velocities and other information, but we do not do this by simply reading, say, A’s initial velocity off B’s final velocity. Still, we are using a velocity as a source of information about another velocity. A different example will illustrate a more radical situation. Suppose I spend too much time in the sun without a shirt. Being rather fairskinned, after a while I will be the subject of two new items: my skin will be blistered and I will be feeling pain. The sun - or, more precisely, photons from the sun - caused both of these new items, and the causal account follows the essential features of our billiard ball example.“’ The new items that appear are determined by (among other things) properties of the sun and of my skin. If my skin were darker or if the sun’s spectrum were appropriately different, my skin would not be blistered and I would not be feeling pain. The sun was not affected by this particular interaction, although the photons that actually interacted with my body were affected; and so forth. Note, however, that neither my blisters nor my pain copy any properties of the photons or of the sun or of my skin prior to my encounter with the sun. In other words, a causal interaction can bring about a property that is not found in any of the items that entered into that interaction. This, however, does not prevent a causal outcome from carrying information about its causal antecedents. For example, the present condition of my skin carries information about the sun and, given a lt is now common scientific practice to describe cases in which these conditions can be met as observationsof the cause in question. For defencc of this usage and some detailed examples see H. Brown, ‘Naturalizing Observation’, The Process of Science. N. Nersessian (ed.) (The Hague: Martinus Nijhoff, 1987), pp. 179-193, and op. cir., note 2; D. Shapere, ‘The Concept of Observation in Science and Philosophy’, Phtiosoph~ of&ience 49 (1982), 485-525. However, because this usage raises issues that are not germane to the points I wish to make, I will avoid the term ‘observation’ completely in the remainder of this paper. ‘Otis reference to photons, along with other references to postulated entities in the discussion to follow, may seem to beg the question with respect to realism. 1 will respond to this objection at the end of the paper. For the time being, I will write as if there are no substantial questions about whether photons, electrons, and such exist since this will simplify the discussion.

236

Studies in History and Philosophy of Science

sufficient body of background knowledge, we might be able to extract that information. Fifth, and finally, although there will surely be cases in which we cannot figure-back to properties that we wish to discover, our abilities in this regard are not as limited as is sometimes assumed. In many scientific cases we are interested in items that exist through significant periods of time, or in members of a class of identical items (e.g. electrons). In such cases we are not limited to one quick glimpse; we can return to these items again and again for further study. Our discussion has indicated two ways in which such continued study might increase our ability to determine specific causes from a study of their effects: we can discover new laws, and we can develop new ways of interacting with the item that concerns us and thereby widen the variety of data at our disposal. Now, perception of an item in the physical world results from a causal interaction between a perceiver and that item. How the item is perceived is dependent on properties of that item, properties of any intervening medium, and properties of the perceiver’s perceptual system - that is, her sense organs, nervous system and brain. As a result, the way an item appears carries information about all of these, but we cannot assume that the properties of a physical item are directly revealed by the way that item is perceived. This may occur in some cases. (High-fidelity sound systems are designed so that the speakers will produce sound waves that closely approximate the sound waves that impinged on the microphones.) But whether any properties of a physical item can be read directly off its perceptual appearance must be established in individual cases. Given the complexity of the causal processes involved, it would be surprising to find that such copying occurs often - and the available scientific accounts of the physical world suggest that this is rare indeed. Still, even when perception does not directly reveal the properties of physical items, perception provides a source of information about those items, and in this sense perception is an indirect source of knowledge about the physical world. It may be extremely difficult to extract the information carried by perception in a specific case, but our discussion suggests that one way in which we may be able to improve the situation is by increasing the range of available data. This is the point at which our causal account of perception will pay epistemic dividends for realism. For on this account of perception, the use of instruments to increase the range of data available is completely demystified. Instrumentation extends our senses by allowing us to enter into new causal interactions with items in the physical world and thereby increase the amount of available information about that world. I want to develop this point next. 3. Instrumentation The familiar fact that each of our senses detects some items that are not detected by other senses should be enough to make intelligible the suggestion

Prospective Realism

237

that the world might also include items that none of our senses detect. As science has developed, scientists have been led, often reluctantly, to postulate such items and they have also constructed instruments that allow them to study these items. Scientists use several types of instruments that serve different functions, but I will focus attention here on instruments that operate by entering into a causal interaction with the items under study and producing outputs that we can perceive. Perhaps the oldest instrument of this type is the magnetic compass which lets us see the direction of the earth’s magnetic field, even though we have no sense that directly responds to this field. Galileo’s use of the telescope provides the first case of the systematic use of this kind of instrument for scientific purposes. Twentieth-century scientists have been especially prolific in postulating items that we cannot perceive; their increasing tendency to postulate such items has gone hand-in-hand with an increasing ability to develop instrumentation that allows them to interact with, and thus study, the postulated items. This is exactly what we should expect given a suthciently broad-minded version of empiricism. The central empiricist insight is that we cannot learn about the world just by thinking. Rather, we must interact with the world and gather information that will guide our thinking and allow us to test our hypotheses. We must be particularly careful not to confuse this central empiricist thesis with detailed elaborations that offer accounts of how this aim of learning about the world is to be, pursued in practice. There have been many such elaborations. Historically important examples include the Aristotelian thesis that we should rely on our unaided senses since these are appropriate for showing us what the world is like; the early modem view of Galileo, Boyle, Locke and others that we cannot take the dictates of our senses at face-value because there is a larger set of qualities in sense perception than in the physical world; the classical empiricist view that the qualities available in sense perception provide the only material for constructing a knowable world; and even the Kantian view that the pursuit of empirical knowledge can only take place within a framework of synthetic a priori propositions that are internal to a knowing mind. To these we can now add the thesis that much of the world cannot be detected by our senses but that we can still study these items given appropriate instruments. Each of these elaborations is, in part, an offshoot of the state of science at a particular point in its development. Indeed, one powerful reason for taking the present proposal especially seriously is the fact that it has emerged - often against considerable resistance - from persistent attempts by scientists to carry out the empiricist programme of learning about the world by interacting with it and by testing hypotheses in such interactions. The next logical step in this discussion would be to display examples of such instrumentation. I will not take that step here because of space limitations and because there is a current and growing body of literature in which such studies

238

Studies in History and Philosophy of Science

are offered.41 My immediate concern is to argue that the use of such instrumentation generates no special epistemological problems once we view it from the causal perspective sketched above. The key point should by now be clear: Given that our senses operate by establishing a causal interaction with items in the physical world, nothing categorically new is introduced by the kind of instrumentation we are considering. These instruments also work by establishing a causal interaction between us and the items we would study. Because of this causal interaction, the output of our instruments carries information about the items under study, but this information rarely occurs in the form of a clear text. We must interpret the output of our instruments in order to extract the information we desire - but this is the case with respect to our unaided senses too. Our ability to extract information about causes from the output of our instruments is highly dependent on our current understanding of natural law.42 But our account of perception yields the same result for our senses. Even the apparently atheoretical process of simply taking what we perceive as a direct revelation of properties of physical items is based on a set of assumptions - and we have powerful reasons for believing that those assumptions are false. In other words, once we accept a naturalistic account of how our senses work, we should conclude that scientific instrumentation increases the range of available information about the physical world, and thus improves our ability to learn about that world. There is one argument for the claim that our unaided senses play a special epistemic role that requires a response. This argument takes off from the point that in the final analysis (leaving science fantasy examples aside) we can only pick up information about our environment through the use of our senses. Even the most sophisticated instruments must provide an output that we can sense. This, in turn, suggests that the uncertainties involved in our use of instruments will always be added to any uncertainties involved in the use of our senses. Thus a long, complex instrumental chain always involves greater uncertainty than a shorter chain, and unaided sense perception will provide the upper limit of reliability.43 This argument is seductive, but it is also incorrect. There are just too many cases available in which we get better information from a more complex causal chain than from a simpler one - as anyone who wears eye glasses or who has “See H. Brown, op. cit., note 2 and note 39; A. Franklin, op. cit.. note, 33; P. Galison, How Experimenrs End (Chicago: University of Chicago Press, 1987); I. Hacking, op. cit., note 5; D. Shapere, op. cit.. note 39. For antirealist interpretations see R. Ackermann, Dora, Insrrumenrs, and Theory (Princeton: Princeton University Press, 1985); A. Pickering, op. cit., note 24. 4zThe noncausal nature of quantum theory places limits on our ability to carry out this programme. For discussion see H. Brown, op. cir., note 2, pp. 144-149. “For a recent statement of this argument see J. Fodor, ‘Observation Reconsidered’, Philosophy of Science 51 (1984). 24.

Prospective Realism

239

ever used a simple magnifying glass should recognize. To see where the argument goes wrong, we must distinguish the simple conjunction of an instrument and an unaided sense from the case in which we use an instrument that compensates for a limitation of that sense. Let me take a different sort of example to underline my point. A person with an injured leg may have a limited ability to stand erect for several minutes. A cane, by itself, has virtually no probability of staying erect. But it would be absurd to conclude that the probability of the person using the cane being able to remain erect must be less than the probability of the cane remaining erect. The multiplication theorem from probability theory does not apply here because a person using a cane is not simply a case of the conjunction of a person and a cane. In a similar way, Galileo’s earliest telescopic studies led him to the conclusion that telescopic astronomy is more reliable than naked-eye astronomy because the telescope corrects specific errors caused by features of our eyes. Galileo attempted to identify those errors and explain how the telescope eliminated them. We still take Galileo’s point to be correct, although the detailed mechanisms are more complex than Galileo imagined.” In a similar way, eye glasses are designed to correct specific refractive errors, computers improve on our rather limited calculating abilities, and so forth. Note also that scientific instruments often increase the quantity and variety of data available to us. It should not surprise us to find that we can arrive at more reliable conclusions from large amounts of varied data than from more limited data. And even if the above argument were correct, it would not negate another crucial point: There are too many items in the world that our senses do not detect at all and we thus have no choice other than to rely on instruments if we would study those items. In addition, there are no a priori grounds for holding that the items we can detect with our senses are more significant for understanding the physical world than the items that our senses do not detect. Thus the fact that all instruments must yield outputs that we can sense provides a pragmatic constraint on the design of instrumentation, but it does not provide a limit on our ability to study the world. I want to end this section by returning for a moment to van Fraassen. Given the relation between instruments and senses we have been discussing, it is difficult to see what reasons van Fraassen could bring forth for giving data provided by the use of our unaided senses an epistemic significance that is to be withheld from data derived with the help of instruments. Van Fraassen does not appear to identify perception with the awareness of sense-data or other such private items; rather, he seems to hold that perception provides information about items in the world. He also insists that science tells us what we can “For discussion see H. Brown, ‘Galileo on the Telescope and the Eye’, Journal of the Hi.srory of Ideas 46 (1985), 487-501.

240

Studies in History and Philosophy qf’ Science

perceive. But once science has taught us that the relevant interaction between our senses and physicalsitems is a causal interaction, there remains no reason for refusing to treat data derived from our instruments on a par with data derived from our senses - particularly when we note that in many cases we understand the operation of our instruments better than we understand the operation of our own sensory systems. Note also that in responding to Gutting’s question as to why we should confer special status on what is perceivable but not in fact perceived,45 van Fraassen presses the point that we should limit belief to cases in which there is ‘accessible evidence’.& But the point that I have been arguing here is that in so far as we are concerned with beliefs about items that exist independently of our perception, there is no epistemically significant sense in which unaided perception provides evidence but perception aided by instruments does not. There seems to be no more reason for limiting accessible evidence to the evidence we can acquire without instrumental aids than there is for limiting diggable holes to the holes we can dig without instrumental aids. 4. Realism We can now see how the development of new instrumentation improves realism’s prospects. From an empiricist perspective, our ability to learn about the world depends on our ability to do more than just speculate. We must, at a minimum, be able to test hypotheses about items in the world by interacting with those items. Earlier in our cognitive history sense perception provided our only means of carrying out such tests and of gathering information about the world. But since the invention of the magnetic compass we have been gathering some information beyond the limits of our unaided senses; and since Galileo turned his telescope on the heavens scientists have been gathering such information in a systematic manner. As a result, we can now marshal1 a wider body of data about physical items than we could without the use of instruments and we can thus place our theories under more stringent empirical constraints than we could without these instruments. I argued above that the central question for realism is whether we can eventually develop a sufficient set of constraints to select one theory-cummethodology in some domains. This question will have to remain open for the present and the forseeable future, but modern instrumentation provides grounds for being optimistic. Because of the increasing range and sophistication of our instruments - especially in this century - currently accepted theories have been tested harder than any previously accepted theories. For “G Gutting, ‘Scientific Realism versus Constructive Empiricism’, Images of Science.P. Churchland and C. Hooker (eds) (Chicago: University of Chicago Press, 1985), pp: 127-130. “Op. cit.. note 20, p. 254.

Prospective Realism

241

example, theories of energy generation in stars must now account not only for the energy carried by the visible portion of the electromagnetic spectrum, but for the full range of radiation and particles that stars produce. Instruments that allow us to study a wider range of these effects open up the possibility of significant new confirmations of existing theory - along, of course, with the possibility of new refutations. Similarly, special relativity has passed significantly tougher experimental tests than its classical predecessor. The MichelsonMorley experiment (whatever its original purpose) supports special relativity and undermines classical mechanics. The experiment was based on a new instrument that allowed for a degree of precision never before attainable. Further tests have been provided by the Kennedy-Thorndike experiment, the muon experiment on time-dilation, and accelerators that permit physicists to measure the response of particles to increasing accelerations. At the same time, special relativity has been used in conjunction with other theories, particularly quantum theory, to predict previously unexpected phenomena as well as physical parameters that have been measured to previously unheard of degrees of precision. Given that a failure of any of these predictions could provide evidence against special relativity, their success must be counted as providing some evidence on behalf of this theory. I hasten to add that I am not claiming that we now have sufficient evidence to declare that special relativity is true, full stop. I am defending only the more modest thesis that we have better reasons for believing special relativity than we had for believing its predecessors because the empirical constraints on special relativity are much greater. We also have every reason to expect that scientists will continue to design increasingly more powerful instrumentation and thus to place scientific theories under increasingly greater empirical constraints. This brings us back to the thesis of prospective realism: There is no guarantee that realism will succeed, but the realist goal is pursuable and our ability to pursue this goal has been increasing because of the increasing power of our instruments. I turn, finally, to the objection that I have begged the question on behalf of realism in writing of such items as photons, electrons and the electromagnetic spectrum. This was done for ease of exposition, but to make the points that concern me here, all we need assume is that there is a world that exists apart from us and that we are interacting causally with that world. For, even if currently accepted theories fail radically, so that we are left with no basis for understanding our instruments or for interpreting their outputs, our instruments still provide a wide range of causal interactions with the world, and thus generate constraints that scientific theorizing must meet. In the absence of an adequate theory of the instrument, this would leave us with data in search of an interpretation, but it would still leave us with a rich body of constraints on our theorizing.

242 Acknowledgements -

Studies in History and Philosophy of Science

This material is based upon work supported by the National Science Foundation under Grant No. DIR-8807867. The Government has certain rights in this material. I want to thank Mark Heller, C. A. Hooker, David Hull, Andrew Lugg and Harvey Siegel for comments on an earlier version.