From owner-bhaskar@jefferson.village.Virginia.EDU Sat Nov 22 22:04:43 1997 Received: (from domo@localhost) by jefferson.village.Virginia.EDU (8.8.5/8.6.6) id WAA29541 for bhaskar-outgoing; Sat, 22 Nov 1997 22:04:37 -0500 X-Authentication-Warning: jefferson.village.Virginia.EDU: domo set sender to owner-bhaskar@localhost using -f Received: from ( []) by jefferson.village.Virginia.EDU (8.8.5/8.6.6) with SMTP id WAA36448 for <>; Sat, 22 Nov 1997 22:04:33 -0500 Received: from by (5.x/SMI-SVR4) id AA29463; Sat, 22 Nov 1997 20:04:11 -0700 Received: by (SMI-8.6/SMI-SVR4) id TAA14686; Sat, 22 Nov 1997 19:54:17 -0700 Date: Sat, 22 Nov 1997 19:54:17 -0700 Message-Id: <> From: Hans Ehrbar <> To: bhaskar@jefferson.village.Virginia.EDU Subject: BHA: rts2-32 Sender: owner-bhaskar@jefferson.village.Virginia.EDU Precedence: bulk Reply-To: bhaskar@jefferson.village.Virginia.EDU 148 A Realist Theory of Science 2. THE SURPLUS-ELEMENT IN THE ANALYSIS OF LAW-LIKE STATEMENTS: A CRITIQUE OF THE THEORY OF MODELS It has often been held that a constant conjunction of events is not a sufficient condition for a causal law. This may be because it is regarded as incapable of sustaining the intuitively obvious The Logic of Scientific Discovery 149 and important difference between necessary and accidental sequences or in Johnson's time-honoured terminology between universals of law' and `universals of fast'.4 Or it may be because it is regarded as incapable of licensing what it is intuitively felt causal laws do licence, namely counter-factual conditionals.5 It is never seriously denied that we feel, and scientists act as if, some but not other sequences of events are `necessarily connected'; so that we must possess the concept. What the radical empiricist, in the form of Hume, denies is: (a) that there is any objective basis for this distinction, i.e. that it corresponds to any real difference between the two sequences of events; and (b) that there is any justification, apart from habit or custom, for our ascriptions of natural necessity and accident.6 Most philosophers since Hume have attempted to show how he was wrong in (b) without objecting to (a). I want to argue that Hume was wrong in (a); and that it is only if we can establish this that we can show why he was wrong in (b) also. The radical empiricist challenge to philosophers then is to provide an alternative account of the `surplus-element'7 in the analysis of law-like statements; that is, that element over and above the (presumed) constant conjunction that explains our ascriptions of necessity; and which will show how, and the conditions under which, a distinction between necessary and accidental sequences and the assertion of counter-factuals can be rationally justified. The usual response to this challenge consists in the attempt to locate the surplus-element in the statement's `explanation', and more particularly in the `theory' which explains it. However the terms `explanation' and `theory' cover a gamut of philosophical positions, which must now be considered. The deducibility of a law-like statement from a set of higher order statements is often regarded as a criterion of `explanation'.8 4 W. E. Johnson, Logic, Vol. III, Chap. 1. 5 R. Chisholm, `The Contrary to Fact Conditional', Mind 55 (1946), reprinted in Readings in Philosophical Analysis, eds. H. Feigl and G. Maxwell, pp. 482-97. 6 (a) and (b) correspond of course to Hume's two definitions of `cause'. See D. Hume, Treatise, p. 172 and Inquiry, pp. 76-7. 7 I owe this term of G. Buchdahl, op. cit., p. 27 and passim. 8 See e.g. R. B. Braithwaite, op. Cit., Chap. 8; C. G. Hempel, op. cit., Chap. 12; and E. Nagel, op. Cit., Chap. 4. 150 A Realist Theory of Science However if deducibility is the only criterion for explanation and the source of the surplus-element is its explanation there will be an infinite number of surplus-elements for *any* statement. Hence any statement can be said to be law-like on an infinite number of grounds!9 Deducibility alone cannot explicate the distinction between necessary and accidental or nomic and non-nomic universals. Moreover additional criteria such as simplicity can only reduce the number of possible explanations for a statement which has already been identified as law-like. But they cannot be used to say which statements are law-like and so possess the surplus-element. For even if there were a simplest explanation for every statement, there are no absolutely simple explanations. Thus such criteria can at best be used to explain why we choose one explanation rather than another, but not why one statement rather than another is regarded as law-like.10 Of course it might be objected that when everything is explained all factual statements will be law-like. But what would count as an explanation then? Could it be anything other than an inexplicable constant conjunction of events, as in the case of 9 The Jesuit mathematicism Clavius demonstrated this fallacy in Osiander's apologetic preface to Copernicus' De Revolutionibus. Osiander had argued, as Galileo was later invited to before the Inquisition, that the helio-centric theory was merely a mathematically adequate representation of the facts of planetary motion that made no claim to be true. Clavius pointed out that it was never a good argument in favour of a theory that it `saved the appearances', as a true result could be derived from any number of absurd or false premises. (Cf. J. Losee, An Historical Introduction to the Philosophy of Science, pp. 44-5.) Indeed even if we exclude all premises which we know to be false or which are not explicitly defined there will still be an infinite number of sets of premises from which the facts can be deduced, provided we allow for the introduction of artificial predicates such as Hesse's `tove' (M. B. Hesse, Models and Analogies in Science, p. 30), of which place- and time-dependent predicates such as Goodman's `grue' (N. Goodman, Fact, Fiction and Forecast, p. 74) merely form a special class. Hence deducibility cannot provide a sufficient criterion for choosing one set of premises rather than another (the source of Goodman's paradox) or for justifying one statement rather than another as law-like. 10 This is of course a very poor best. For (i) the simplest of any small number of explanations is not necessarily the best (cf. M. Bunge, The Myth of Simplicity, pp. 51-134); (ii) there will still be an in principle infinite number of equally simple explanations, if we restrict ourselves to formal or syntactical criteria alone (cf. J. J. Katz, The Problem of Induction and its Solution, Chaps. 4 and 5). The Logic of Scientific Discovery 151 Mill's unconditional laws?11 If it could not, we are back with Hume, and have done nothing to allay the sting of the radical empiricist challenge. If it could, some alternative non-Humean analysis of the ultimate or highest-order laws must be given which will show how they, as uniquely qualified `explainers', do possess a genuine surplus-element. We are thus faced with the following dilemma: either explanation is achieved by subsumption under higher-order laws in which case the problem is merely shifted, for a surplus-element must be found for them if they are to qualify as `laws'; or an alternative analysis of `explanation' must be given, which does not identify the explanans with a further set of laws, and so provides room for the location of a surplus-element in the analysis of laws, within the context of their explanation, at any one level. It might be thought that it is in the capacity of the law-like statement to yield successful predictions that the source of the surplus-element lies. But this will not do without an analysis of the `capacity' or `power'. For the Humean it is the past and actual successes of the statement that count, not its potential ones. And these can at best explain, not justify, the surplus-element. It is the surplus-element that must provide our inductive warrant, if we have one: rather than the other way round. Moreover even an accidental generalization is capable of yielding correct predictions, viz. as long as the conditions that account for it persist. This suggests that, even if we were to possess some general inductive warrant, predictive success alone could not differentiate necessary from accidental sequences or license the assertion of counterfactuals. It seems clear that if we are to get any further in our search for the surplus-element the idea of purely formal differentiae must be abandoned. Inductive considerations prove no better than deductive ones. For accidental generalizations may be inductively confirmed, just as they may be deductively explained. In practice then the non-radical empiricist, if he is not to concede the game, is forced to re-examine the account of science that seems to render any non-Humean conclusion impossible. The fundamental fact about science that has been missing from the discussion so far is the existence at any moment of time of an antecedently established body of theory. And it is here that the 11 J S. Mill, A System of Logic, Vol. I, p. 378. 152 A Realist Theory of Science non-radical empiricist attempts to locate the surplus-element. But can `theory' do what experience and deducibility fail to do, i.e. provide a rational ground for our ascriptions of natural necessity? The answer clearly depends upon the extent to which the former contains components irreducible to the latter. And the onus is on the philosopher who attempts to locate the surplus-element in the systematic organization of our knowledge or the capacity of a theory to explain many different laws12 or to predict novel kinds of facts13 to show how their concept of theory escapes Humean analysis. Goodman's notion of entrenchment,14 for example, functions in exactly the same way as Hume's notion of custom and can no more justify our attributions of necessity than the latter could. In short, unless theory contains elements irreducible to experience and truth-functional operations on it there is no basis for a non-Humean theory of natural necessity.15 Thus the possibility of the latter depends upon some terms of the theory not being explicitly defined in terms of experience and/or some statements of the theory not being deductively connected and/or some ideas of the theory being non-propositional in logical (or non-sentential in linguistic) form. These establish the possibilities of intensional relationships between predicates, non-deductive (e.g. analogical) relationships between ideas and non-propositional (e.g. iconic) ideas respectively as potential sources of necessity. It is the second of these that has been most thoroughly explored; and it is to Campbell's initial formulation of the theory of models that I now turn. On Campbell's view a theory must contain not only a `dictionary' correlating some, but not all, of the theoretical concepts with empirical terms but a `model' for the hypotheses or theoretical statements of a theory, by means of which its hypothetical subject matter may be imagined to be like in some, but not all, respects the real empirical subject matter of some field which is already known.16 On this view the surplus- 12 E. Nagel, op. cit., pp. 64-5. 13 I. Lakatos, op. cit., p. 116. 14 N. Goodman, op. cit., pp. 92-122. 15 For, as Craig's theorem shows, if it does not the theoretical component is then completely eliminable. See W. Craig, `The Replacement of Auxiliary Expressions', Philosophical Review 65 (1956), pp. 35-55. 16 N. R. Campbell, The Foundations of Science, esp. Chap. 6. The Logic of Scientific Discovery 153 element just is the model. Thus what distinguishes Boyle's law from a merely accidental generalization is, according to Campbell, the corpuscularian model informing the kinetic theory of gases. By means of this model gas molecules are imagined to be, in certain respects, like billiard balls bouncing off each other and exchanging their momentum by impact. And it is in our prior understanding of this that the necessity of the gas laws ultimately lies. Notice that for Campbell it is not the mere availability of a theory or even the organization that the theory makes possible (e.g. the fact that Boyle's law, Charles' law and Graham's law are all deductive consequences of the kinetic theory) but the interpretation theory explains.17 As a critique of the deductivist view of the structure of scientific theories, as typified by Mill, Duhem and Hempel, Campbell's case is a strong one. The deductivist, he says, merely exhibits `the dry bones of science from which all the spirit has departed".18 His project is to revitalize it. He sees the driving force of science as the exploitation of analogies in the conquests of new fields, without which neither theory nor the range of facts could grow or the language in which to state them develop.19 But is his case unanswerable? How does it fare when faced with the challenge of radical empiricism? Is it capable of providing an adequate account of the surplus-element in the analysis of law-like statements? To answer these questions we must look more closely at the terms of the modelling relationship which is intended to provide the basis for a non-Humean theory of natural necessity. Now essential to Campbell's correction of the deductivist view of explanation is the idea that for the explanation of a range of phenomena say E_a to have occurred the relationship between the theory T_a which explains the phenomena and from which the latter is deducible must be supplemented and informed by another relationship. This is a relationship of analogy not deduction; and it is by means of it that we render T_a intelligible to ourselves. See Diagram 3.2 below. According to Campbell the entities and processes postulated at T_a are unknowable; i.e. 17 Ibid. pp. 126-40. 18 N. R. Campbell, What is Science?, p. 99. 19 N. R. Campbell, Foundations, pp. 132-7; and M. B. Hesse, op. cit., pp. 35-43. 154 A Realist Theory of Science they do not constitute part of the phenomenal world described by science. Although we cannot know what produces E_a we can imagine it to be like something we do know. Such an act of imaginative daring need not be totally arbitrary. For it is possible to conceive of principles of analogical, just as there are principles of deductive or inductive reasoning. Only when we have constructed a model can we be said to have achieved T_a |\ | \ | \ Analogy | \ | \_ Deduction V /\ | \ | \ E_a E_b Diagram 3.2 Campbell's `Theory' scientific understanding. That is, not just saved the facts, preferably with elegance and economy, but explained them. Using the analogy provided by E_b a real or empirical phenomena can thus `enliven' the abstract theoretical relationships from which E_a is deduced. And E_b does this by standing in for or representing (in the sense of the German `darstellung') the unknown causes of E_a. Explanation thus involves, centrally, the substitution in our imagination of a real or empirical relationship for an unreal or theoretical one.20 This is Campbell's debt to empiricist ontology: a debt that it summed up by his tacit acceptance of the concept of the empirical world. For on his theory T_a cannot be, or come to be known as, real; though it is at any moment of time, and perhaps forever, unperceivable to us. For him theoretical entities, such as molecules, can only be said to be `real' by analogy with material objects. Campbell does not deny that deducibility is necessary for explanation, merely that it is sufficient. His theory may thus be regarded as providing an alternative shave to Occam's razor. Tyndall formulated the criterion for the selection of explanations 20 N. R. Campbell, op. cit., pp. 243-56. The Logic of Scientific Discovery 155 implicit in Campbell's theory as follows: `ask yourself whether your imagination will accept it'.21 Now such a criterion is clearly capable of selecting a theory within a given metaphysical schema, such as that provided by the classical mechanical world-view. But it is not capable of judging between different schemas, when it is precisely the nature or the limits of the imagination that is in question. To take an obvious example: Aristotelian and Galilean dynamics are in conflict over whether when a stone falls to the earth, the earth should be conceived as fixed (Aristotle, Ptolemy and Tycho Brahe) or as moving (Copernicus, Giordano Bruno, Kepler and Galileo). Now, try as you may, there is no neutral way of conceiving the falling of the stone.22 Our imagination, although not fixed, is either Aristotelian or Galilean. Tyndall's criterion cannot help us to decide between the competing frameworks, because what it in question is the nature of the concept in terms of which any motion has to be understood. There is a similar break involved in the transition from Newtonian to Einsteinian dynamics. Part of the trouble with current micro-physics is that our imagination cannot accept it, and yet we have every reason to believe it to be true. If Tyndall's criterion were acted upon it could have effects on scientific practice as conservative and dogmatic as the consistency and meaning-invariance conditions of classical empiricism. A new scientific ontology or a fundamental change in scientific concepts may transform our conception of what is plausible. At such times in the history of science it becomes necessary for the scientist to stand Tyndall's criterion on its head, and dizzily ask himself whether he can continue to accept his imagination. Although its inadequacy to deal with fundamental scientific change is most evident, Tyndall's criterion is no less inadequate to deal with the continuing processes of conceptual micro-adjustment, in which our imagination is continually modified and extended, that are a part and parcel of the process of `normal science'. More generally, it is always legitimate for 21 J. Tyndall, `Scientific Uses of Imagination', Fragments of Science for Unscientific People, p. 131. 22 P. K. Feyerabend, `Problems of Empiricism, Part II', The Nature and Function of Scientific Theory, ed. R. G. Colodny, p. 317. 156 A Realist Theory of Science scientists to ask and sometimes possible for them to answer, questions about whether gases are really composed of molecules or whether the earth really moves. Such questions cannot be rephrased as questions about the plausibility of our conceptions. This would be, in terms of Diagram 3.1, to reduce phase (1) to phase (2). Rather the normal procedure in science is if we have a plausible conception to go on to ask whether it is true, which is to ask whether the entities and processes it postulates are real, or only fictional. Plausibility is a prima facie criterion for a theoretical explanation. But is is neither sufficient, nor in the last instance necessary. How does Campbell's theory fare as a response to the challenge of radical empiricism? According to it, the surplus-element in the analysis of law-like statements is the model at the heart of the theory that explains it. But for Campbell the model cannot prompt questions about the reality of the abstract entities and processes postulated in the theory. For theoretical entities are by definition unperceivable and hence, given the fundamental equation of empiricist ontology, viz. real = empirical, cannot exist. Models function then not as knowledge-extending but as essentially pragmatic devices, servicing the needs of the understanding. Theory involves a journey from one set of experiences E_b to another E_a. Because of this it is always possible for the radical empiricist to ask whether the journey is really necessary. Moreover, even if a way could be found of showing that some model is necessary, there would seem to be no way of justifying the choice of any particular one (given that the idea that its necessity could be demonstrated a priori is rejected as being inconsistent with the fact of scientific change). To this it may be contended that models are necessary not only as conceptual crutches for the tender-minded and as heuristic devices for the young (which the radical empiricist may graciously concede) but for a theory's growth and development, and in particular (so as not to beg the issue by positing non-empiricist criteria of development) for the generation of facts empirically relevant for the theory but which would not have been forthcoming without it.23 But this only pushes the argument back a stage further. In a completed science models would 23 See e.g. M. B. Hesse, op. cit., pp. 35ff. Cf. also P. K. Feyerabend, `Problems of Empiricism', op. cit., pp. 173ff. The Logic of Scientific Discovery 157 be dispensable. For, as Duhem has put it, `to explain is to strip reality of the appearances in which it is wrapped as in veils in order to see this reality naked and face to face'.24 When we have done this, what more can there be to do? The objection that `explanations are practical context-bound affairs'25 either is covered by the heuristic role allowed to models or depends upon the incompleteness of science, in which case their nemesis is merely (if perhaps indefinitely) postponed. We are thus forced inexorably back to a particular conception of reality, the only `world' that Campbell's account of science contains: the world of Mach and Hume. In such a world causality is bare and invariant conjunction; and scientific knowledge consists, for its part, in `description, that is the mimetic reproduction of facts in thought, the object of which is to replace and save the trouble of new experience'.26 Suppose now that arguments are advanced to show that no science can ever be complete in the requisite sense. Science still remains, on the Campbellian conception, a purely internal process, locked in a dosed circle of thought. Science is still a creature of custom and habit, the only difference being that the habit is now one of the imagination, rather than sensation. In virtue of their shared ontology Campbell is closer to Mach and Tyndall to Occam than one might think.27 In neither case can the possibility of major conceptual revisions be accommodated or the mechanism of scientific discovery be displayed. Let us apply to Campbell's theory the litmus test for the adequacy of an account of science developed in Chapter 1. Can it sustain the idea of the applicability of the concept in question, viz. that of necessary connection, in a world without men? The answer is obvious. In the case of Campbell, as of Hume, there is still no difference, independent of men, between a necessary and an accidental sequence of events. The Campbellian can at best talk of a nomically necessary statement; he cannot talk of a nomically necessary sequence. The attempt to locate 24 P. Duhem, op. cit., p. 7. 25 M. Scriven, `Truisms', p. 450. 26 `This', says Mach, `is all that natural laws are', op. cit., p. 192. 27 Indeed one might be tempted to see the difference as merely one of taste or temperament as when Duhem compared the `rolling drums', `pearl beads' and `toothed wheels' of the mechanical models of English physicists such as Maxwell, Kelvin and Lodge with his own Cartesian conception of an axiomatic electricity. See op. cit., pp. 70-l . 158 A Realist Theory of Science the surplus-element in the analysis of law-like statements in the imagination of men is a failure. For transcendental realism the surplus-element distinguishing a law-like from a non law-like statement is the concept of the generative mechanism at work producing the effect in question. Such mechanisms exist and act independently of men; so that the necessity can be properly ascribed to the sequence. Moreover as the world is open not all events will be connected by a generative mechanism; so that the transcendental realist can sustain a concept of natural accident. Only a real difference between necessary and accidental sequences can justify our distinguishing law-like from non-law-like statements. Hence one cannot deny Humean conclusion (b) (on page 149 above) without objecting to Humean conclusion (a), and thus to the ontology that implies it. Nowhere is the anthropocentricity of post-Humean philosophy more evident than in the notion that natural necessity must be sought in the behaviour or nature of men. And nowhere is the displacement of rational intuitions more obvious than the attempt to locate structure in the imagination of men. `Connection' is, as Chisholm has remarked, an `ontological category and a source of embarrassment to empiricism'.28 But it is not an irreducible one. For its basis lies in the generative mechanisms of nature which connect events as cause and effect and which exist as the powers of things. Thus to assert a counterfactual is not to make a meta-statement29 (which would be to make a statement about its grounds), but to make a statement about the way some thing would have behaved (exercised its tendencies, liabilities or powers) had the conditions in fact been different. Theory is not an elliptical way of referring to experience,30 but a way of referring to hypothesized inner structures of the world, which experience can (in ways to be explored in paragrpah 3 below) confirm or falsify. We are not locked in a dosed circle of thought; because there are activities, viz. perception and experimentations by means of which under conditions which are deliberately generated and carefully controlled, relatively independent cross-bearings on the intransitive objects of thought 28 R Chisholm, op. cit., p. 496. 29 See e.g. E. Nagel, op. cit., p. 75; or S. Toulmin, op. cit., p. 185. 30 Ibid., p. 185. The Logic of Scientific Discovery 159 can be obtained. Such activities are not independent of thought, but their results are not implied by them either. Campbell's achievement is to have seen that scientific theory cannot be identified with a deductive system erected on the basis of a single set of experiences. But he made two mistakes. He too, like the empiricists, missed the essential point that science is essentially developing; so that the hypothetical mechanisms of yesterday may become today's candidates for reality and tomorrow's phenomena. But behind this failure also lay an inadequate intransitive dimension, and in particular the absence of the concept of objects apart from our changing knowledge and possibilities of perception of them. Campbell's theory has been extended in two ways. Some have rectified his first mistake but not his second, and viewed science as a sequence of models, an unfolding process of shifts in intellectual fashion. Others have developed his theory in a realist way. Harre, for example, has drawn attention to the role of the existential questions prompted by the creative use of analogies in the development of science.31 By way of concluding my discussion of Campbell's theory I want to sketch out such a dynamized realist version of it. In Diagram 3.3 below the dotted lines now stand for relationships of deduction and the continuous lines for relationships of T_alpha /|\ (vertical lines / | \ dotted, all others / | \ continuous) / | \ / | \ / | \ / | \ / | \ E_x / |E_alpha \E_ta = T_a ______________________________________________ | |\ | | \ | | \ | | \ | | \ | | \ | | \ | | \ | | \ E'_a <-- E_a E_b Diagram 3.3 A Dynamic Realist Development of Campbell's Theory 31 R. Harre, op. Cit., esp. Chaps. 2-3. 160 A Realist Theory of Science analogy (to indicate their reversed relative importance). T_a has come to be established as real, and in this case also is perceivable. In the course of this process facts E_a have been corrected and now become facts E'_a. T_a now provides one of the sources for a new model designed to explain phenomena E_a. And the process of checking its reality (which will almost certainly modify our conception of it) has begun. Needless to say there will in general be more than one model for E_a. The state of chemistry c.1930 provides an illustration of the model. T_a is Prout's hypothesis and T_alpha the theory of sub-atomic structure. E_a-->E'_a, consists in the elimination of the impurities that dogged the verification of Prout's hypothesis for over a century. And the new model might be the Bohr-Rutherford model of atomic structure; which conceived as a hypothesis about the internal structure of atoms is, we now know, false. The source of such models may lie either in some general conceptual scheme (such as atomism in chemistry) or some other science or proto-science (such as the wave models of light and sound in particle physics). The subject of such models is the unknown but knowable intransitive structure of the world. It is by means of the experimental testing of the hypotheses suggested by already existing knowledge that new knowledge comes to be produced. The problem of the surplus-element, and Hume's challenge, has another aspect. This turns on the question of what warrant we have for distinguishing between cases of genuine and pseudofalsification, and hence for invoking the CP clause in defence of generalizations in the former case. This calls into question the necessity of deducibility, not just its sufficiency in the explanation of laws. Science needs a concept of pseudo-falsification for three reasons, two of which are epistemic and one of which is ontological. Firstly, because a theory may not be at present sufficiently refined or developed to cope with anomalous counter-instances; that is to say, every theory needs a `protective-belt' for its development.31 Secondly, because the `facts' may be wrong: either in the simple sense that they are misrepresentations of the phenomena or more profoundly because they depend upon 32 I. Lakatos, op. cit., pp. 134-8. The Logic of Scientific Discovery 161 false or inadequate observational theories.33 As is well known, every new theory is faced with innumerable anomalies and counter-instances of these kinds. They form in a sense the staple diet of normal science. A successful theory is one which, like Newton's, though it never resolves them all and generates new ones in the process of their resolution, `turns each new difficulty into a victory for its programme'.34 Thirdly, science needs a concept of pseudo-falsification because a countervailing cause or interfering agent may be at work generating the `counter-instance'. It is only under closed conditions, as we have seen, that a theory can be given a fair test or that a crucial experiment - Bacon's `instance of the fingerpost'35 - becomes possible. The problems of the necessity and universality of law are indeed inextricably linked, but not in the way Hume thought. For if the surplus element in the analysis of law-like statements is the concept of a generative mechanism at work and this concept is irreducible to that of a sequence of events then it is quite rational to uphold an ontological distinction between cases of genuine and pseudo-falsification (in which, as exemplified by the case of Prout's hypothesis referred to above, our epistemic distinctions too may be grounded). For we may readily allow that the generative mechanism in virtue of which natural necessity is ascribed is not undermined by the instability of the conditions under which it operates. So that if a law has been confirmed under closed conditions and there is no reason to suppose that the generative mechanism at work in those instances has ceased to operate, the law that the concept of the mechanism grounds may be supposed to continue to apply outside the conditions under which it was confirmed, whether or not the consequent of the statement happens to be realized. By now it would, I think, be generally agreed that models play some cognitive role in science and that there is a feature about such models which renders them irreducible to the experiences that they are in some way intended to embroider or explain. (This feature is, I have argued, typically an idea of a 33 As Feyerabend has put it: a theory may be in trouble only because of `the backwardness of the observational ideology'. See `Problems of Empiricism, Part II', pp. 292ff. 34 P. S. de Laplace, The System of the World, Bk. III, Chap. II. 35 F. Bacon, op. cit., Bk. II, Aphorism XXXVI. 162 A Realist Theory of Science mechanism which would, if it were real, generate the phenomena in question.) But the representatives of the three traditions in the philosophy of science differ radically in their interpretations of the status and role of such models, and of the irreducible concept that constitutes its essential core. The classical positivist view is that it is merely a heuristic device (Duhem, Hempel and Brodbeck). This is liable to encourage the view that the rationale for distinguishing necessary from accidental sequences is solely pragmatic; that it is, as it were, a question of our greater attachment to the former (Quine),36 or of the deeper entrenchment of their predicates in our conceptual system (Goodman). Similarly it encourages the idea of the CP clause as a device that can be relaxed or invoked, switched off or on, according to whether or not we are prepared to forego the falsified law-like statement. This view carries the implication of course that the use of the CP clause is bound to be more or less arbitrary or dogmatic.37 And this in turn creates the Kuhn-Popper problem of the functions of dogma. The concept of the generative mechanism may be given a firmer status, and the distinctions it grounds a better rationale, by seeing its function as concerned essentially with the development of science. Protection from pseudo-falsification then becomes protection from too easy or too early falsification; that is, before the full potentialities of the theory have been developed (Lakatos and Feyerabend). This view allows that our knowledge is structured - that it contains, as it were, layers of different age. The conditions of knowing are here explicitly distinguished from the conditions of being. But positivism still provides the underlying account of the world. And because of this the rationale of the concept of the generative mechanism, which forms the heart or essential core of the theory, is still more or less pragmatic, still science- or knowledge- or man-dependent. The third position consists in coming to see not just our 36 `Any statement may be held true come what may, if we make drastic enough changes elsewhere in the system', W. V. O. Quine, From a Logical Point of View, p. 43. 37 See e.g. K. R. Popper, Logic of Scientific Discovery, p. 42 and pp. 80-2; and T. W. Hutchison, The Significance and Basic Postulates of Economic Theory, pp. 40-6. I have of course argued (in 2.4 above) on quite distinct realist grounds that once the irrationality of pseudo-falsification is granted the CP clause becomes superfluous. The Logic of Scientific Discovery 163 knowledge but the world itself as structured and differentiated. According to this conception, which is that of transcendental realism, science is concerned neither with the incessant accumulation of confirming facts (or the incessant search for falsifying ones), nor even with its own growth and development, but rather with the understanding of the different mechanisms of the production of phenomena in nature. Thus it allows that under certain conditions the concept of the generative mechanism at work may be given a realist interpretation as a representation in thought of the transfactually active causal structures of the world. The possibility of such an interpretation supplements internal consistency and contextual plausibility as a constraint on the possible forms of theoretical advance; and it constitutes the ultimate goal of all theory construction. Now empirical realism generates the following dilemma: Either theoretical entities refer ultimately to experience, in which case they can be eliminated. Or theoretical entities constitute experience (in whole or in part), in which case they cannot be eliminated, but must, given the equation of empirical realism, constitute the world (in whole or in part). Now as long as an ontology based on the category of experience is retained there can be no grounds independent of man for ascribing necessity to some but not other statements. On the first horn this generates the problem of what justifies our belief that the future will resemble the past, or the unobserved the observed, i.e. the problem of the induction. But on the second horn it generates the problem of what justifies the assumption of intellectual conformity. And, on this horn, scientific change, or even dissent, actually constitutes (in whole or in part) a breakdown in the uniformity of nature! --- from list --- .