REVIEWS

Analyses of Theories and Methods of Physics and Psychology, Minnesota Studies in the Philosophy of Science, Vol. IV (ed. by Michael Radner and Stephen Winokur), Minneapolis, University of Minnesota Press, 1970, vi+441 pp., $12.50. This fourth volume in the Minnesota Studies in the Philosophy of Science is divided into two parts. The first part consists of papers and discussions from a conference on the problem of correspondence rules held in 1966 at the Minnesota Center for the Philosophy of Science. However, some of the papers and much of the discussion have little to do with correspondence rules. The second part consists of papers by various members of the staff of the Minnesota Center and visitors to the Center. These papers, according to the editors, "reflect work going on in 1969-70" (p. vi). The papers in this part are on a variety of topics ranging from a critique of Griinbaum's philosophy of geometry to methodological difficulties of psychoanalytic research. Herbert Feigl in his 'The Orthodox View of Theories: Remarks in Defense as Well as Critique' sets the stage for the conference. Feigl outlines the orthodox positivist view of scientific theories: a pure calculus connected to some empirical subject matter by rules of correspondence. Feigl points out that this view has come under strong attack in recent years. He expresses his disagreement with critics such as Feyerabend and makes some attempt to answer them. However, his paper is short and his defense is not fully developed. In a long and provocative essay 'Against Method: Outline of an Anarchistic Theory of Knowledge' Paul K. Feyerabend maintains that "anarchism, while perhaps not the most attractive political philosophy, is certainly an excellent foundation for epistemology, and for the philosophy of science" (p. 17). At times Feyerabend seems to be saying that in such an epistemology there are no standards, no rules, no method and "anything goes" (p. 26). At other times, however, he seems to be holding a

Syntheae 25 (1972) 219-228. All Righta Reserved Copyright © 1972 by D. Reidel Publishing Company, Dordrecht-FIolland

220

REVIEWS

more moderate position. Instead of maintaining that there are no rules he seems rather to be maintaining that there are no rules that should not on occasion be broken: "Considering any rule, however 'fundamental', there are always circumstances when it is advisable not only to ignore the rule, but to adopt its opposite" (p. 22). Yet it is not completely clear that even this is his view. For he does advocate one rule that he seems to regard as absolute: the rule of the proliferation of theories. He says that we should "invent and elaborate theories which are inconsistent with the accepted point of view" (p. 26). In any case, Feyerabend argues that the distinction between the context of justification and the context of discovery should be eliminated in the actual ongoing practice of science. His critique of Popper and Lakatos makes clear that he believes that subjectivity and irrationality are essential parts of the testing of theories. Choice between theories which "are sufficiently general to yield a comprehensive world view and which are empirically disconnected" (p. 91) is a matter of taste. Why Feyerabend stops with these theories is unclear. Given his anarchism why is not the choice of every theory, every method, everything in science "a matter of taste"? Indeed, why is not the choice between metascientifie theories (including Feyerabend's) "a matter of taste" ? One is scarcely comforted by being told that matters of taste are "not completely beyond the reach of argument" (p. 91). For we have been told earlier, "It follows that appeal to argument either has no content at all, and can be made to agree with any procedure, or else will often have a conservative function: it will set limits to what is about to become a natural way of behavior. In the latter case, however, the appeal is nothing but a concealed political maneuver" (p. 24). Such a view of argument would seem paradoxically to cast doubt on the very arguments in Feyerabend's own paper and, in particular, on his argument for his view of arguments. However, what Feyerabend's paper may lack in logical coherence it makes up for in sheer bulk; it is 112 pages long with 35 pages of notes consisting of 221 footnotes. The late N. R. Hanson in 'A Picture Theory of Theory Meaning' argues against the view that a general account of the relation between observation and theory can be given. He suggests that "philosophers of science take up ... the analyses of scientific theory one at a time, and not try to settle the issue all at once" (p. 135). In particular, he maintains that sometimes

REVIEWS

221

there is no well articulated theory with which to coordinate observation and measurement and no one simple relation between theory and evidence. What is unclear in Hanson's argument is whether people who held the standard view of scientific theories, e.g. Feigl and the early Hempel - the view that Hanson seems to be talking about - would deny what Hanson says. Hanson affirms several times in the paper that he is concerned with the "context of strategies, the sort of reasonable considerations in virtue of which a man's argument might be designed one way rather than the other" (p. 134). Now whatever the status of the context of strategies it is not the context of justification and this was the concern of advocates of the standard view. Carl G. Hempel in his lucid 'On the Standard Concept of Scientific Theories' formulates the positivist view of theories outlined by Feigl. Hempel, although at one time an advocate of this view, now finds it "misleading in certain philosophically significant respects" (p. 147) and offers criticism of it. He substitutes in its place a different view of scientific theories. A scientific theory is, according to Hempel, a set of the logical consequences of the set of internal principles and bridge principles. The internal principles specify "the theoretical scenario", (p. 142) i.e. the basic entities and processes postulated by the theory, as well as the theoretical laws that are assumed to govern them. The bridge principles indicate the way the scenario is connected with previously explained phenomena. Hempel admits that the distinction between internal principles and bridge principles is vague. This view does not assume a theory free observational language or an uninterpreted calculus as does the standard positivist view. Furthermore, on such a view analogical models are sometimes not just useful as if devices but give tentative descriptions of the actual microstructure of the objects postulated by the theory. These descriptions form an integral part of the theory. In the last section of his paper Hempel considers the question of how the meaning of new terms is to be specified. Hempel now believes that the problem is misconceived. He used to think that unless the meaning of new terms was specified in terms of an antecedently available vocabulary the new terms could not be meaningful. But there are various ways of understanding terms, Hempel now believes, besides defining them by relating them in some way to antecendently understood terms, e.g. learning how to use them in context.

222

REVIEWS

Hempel's critique of the standard view of scientific theories is in fact Hempel criticizing art earlier Hempel; few philosophers could bring themselves to do this let alone do it with so much grace. In 'An Inductive Logic of Theories' Mary Hesse considers the problem of inductive inference in relation to theories by considering a problem first raised by Hempel. When one combines the special consequence condition (If some evidence e confirms a hypothesis t, then e confirms every logical consequence of t) and the converse consequence condition (If t logically implies e and e is confirmed, then e confirms t) one gets counterintuitive results. Indeed, Hempel shows that any statement e 1 confirms any other statement e2. Yet as Hesse points out certain inferences which turn on these conditions seem intuitively justified. Consider an example of Putnam's cited by Hesse. Suppose in, say, 1940 evidence statement et supports a nuclear theory t. (We suppose that t entails el and by the converse consequence condition el confirms t.) But t entails what would happen in "experimental situations so far wholly unrealized in any previous experience" (p. 165), viz. an atomic explosion. Let e2 be a sentence describing an atomic explosion. Then e2 is confirmed by the special consequence condition. This inference seems justified. Hesse suggests that the way to avoid the counter-intuitive results and preserve the inferences one wants to preserve is to make el relevant to ez, the relevance being in terms of analogy. She does not give a general account of analogy but considers cases which are intuitively justified and works out the details of the analogy for such cases. Unfortunately, however, there seem to be cases where the phenomena described by the evidence sentences are not in any obvious sense analogous yet where inductive theoretical inferences are justified. (Indeed the example cited above seems to be a counterexample to Hesse's own view. It is significant that she does not try to show the analogy between the body of chemical and physical evidence existing in 1940 and the atomic explosion.) Hesse could have in mind a very weak sense of analogy. In this case there would be an analogy between the evidence and the explosion. However, if a very weak analogy is allowed, there is a danger that the counterintuitive results will reappear for in a weak sense of analogy any two things are analogous to one another. Until one gets a more adequate account of analogy Hesse's approach does not seem too promising. In any case, more promising alternatives are available. 1

REVIEWS

223

Grover Maxwell in his thoughtful 'Structural Realism and the Meaning of Theoretical Terms' combines Russell's principle of acquaintance with Ramsey sentences to give an account of the meaning of theoretical terms. Russell's principle, as formulated by Maxwell, is: All the descriptive terms in any meaningful sentence refer to items with which we are acquainted. This principle does not imply that one cannot refer to theoretical entities in science. Despite the fact that we are not acquainted with these entities we refer to them by Ramsey sentences. Consider the following sentence containing theoretical term T and observational term O: (1)

(x) (Tx = (3z) (Oy)).

Now instead of using (1) we use

(2)

(34) (x) (~x = 0z) (oy)).

This technique enables us to eliminate theoretical terms thereby solving the problem of their meaning. In (2) the only descriptive term that appears is a term which refers to an item with which we are acquainted. Nevertheless, we still refer in a way to a theoretical entity. To be sure, the use of Ramsey sentences makes it impossible to know the intrinsic properties of theoretical entities, but it makes it possible to know the structure of such entities. Maxwell argues that his approach is applicable no matter how the division between theoretical and observational terms is made, as long as the meaning of terms in the one class (observational terms) is relatively unproblemafic and the meaning of terms in the other class (theoretical terms) is problematic. However, is there anything problematic about the meaning of theoretical terms in science? It is by no means obvious that there is. As Hempel points out in his paper in this volume, the problem of the meaning of theoretical terms may have been misconceived. There is no reason why the meaning of theoretical terms cannot be grasped by seeing how they are used in context. But even if the meaning of theoretical terms is problematic, Maxwell's method of elimination seems to involve a cost that many would not want to pay. We are forbidden on his view from knowing the intrinsic properties of unobservable entities. Other ways of proceeding are surely preferable if any are available but Maxwell does not consider any possible alternatives. In his pithy 'Notes on Feyerabend and Hanson' Joseph Margolis raises some important questions concerning Feyerabend's thesis about

224

REVIEWS

theory laden observation. Feyerabend argues that all observational terms are theory laden and he bases his thesis on a consideration of certain historical examples. Margolis argues that other historical examples given by Hanson suggest that in these cases at least observational terms are theory free. In these cases, according to Margolis, there were "no relevant theories available that might have borne a relationship to observation sentences at all like that obtaining in Feyerabend's example" (p. 194). Thus Feyerabend's general thesis on theory laden observational terms is dubious. Margolis maintains further that since Feyerabend admits that observables themselves (in contrast to the meaning of observational terms) do not change despite changes in the observational language there must be some "residual descriptive language" (p. 195) that is theory free in the relevant sense in which these observables may be identified. In the most technical paper in the volume, 'The Crisis in Philosophical Semantics', William W. Rozeboom develops a novel approach to "the meaning of theoretical terms". He constructs a semantic apparatus that enables one to state the truth conditions of theoretical expressions in purely observational language. Nevertheless a theoretical term in the theoretical expression designates some state of affairs that is not designated by any observational term. One interesting feature of his approach is that theoretical terms do not in general have unique referents. Actually Rozeboom does not give an account of the meaning (connotation) of theoretical terms. Rather, he gives an account of the denotation of such terms. Indeed, Rozeboom argues that "meanings are what goes on inside people's heads" (p. 204f), and the job of determining meaning is ultimately a job for psycholinguistics and psychology. Rozeboom is justified in being skeptical that philosophers would 'appreciate' his psychological account of meaning. For although it might be necessary to know what goes on in someone's head in order to know what he means, it does not follow - nor does it seem correct- that meanings a r e what goes on inside someone's head. In any case Rozeboom, like Maxwell, bases his semantic reconstruction on the supposition that the meaning of theoretical terms is problematic in a way that the meaning of observational terms is not. This assumption is by no means obviously true. In the lead off paper in the second part of the volume, 'Topological and Metrical Structure', William Demopoulos argues that Griinbaum has not shown that the choice among metrics in physical geometry is a

REVIEWS

225

matter of "descriptive convenience". The editors in the preface refer the reader to Griinbaum's rebuttal to Demopoulos' argument. 2 Here Griinbaum accuses Demopoulos of misunderstanding his position and challenges Demopoulos' view on the non-conventionality of the relation of distance which plays a crucial role in his argument against Grtinbaum. Keith Gunderson in his imaginative 'Asymmetries and Mind-Body Perplexities' attempts to dissolve one important obstacle that he believes has stood in the way of accepting physicalism, viz. the persistent inability of people to view themselves (in contrast to other people) as objects of scientific investigation. This asymmetry, according to Gunderson, has led people to suppose that the subject, self, ego, etc. is not subject to physicalistic reduction. Gunderson attempts to dissolve the problem by constructing examples in which an analogous assymmetry holds but which are 'metaphysically neutral'. For example, the fact that a submarine's periscope cannot locate itself in its own crossbeam does not entail any ontological difference between "the nature of the periscope doing the sighting and the thing it can sight" (p. 305). Similarly just because the Cartesian ego "is safe from spotting itself in the physical world" (p. 305) it does not mean that this ego is not a member of the physical world. Now whether Gunderson's arguments will have the dissolving power he hopes for is unclear. For some of us it has never seemed hard "to view the mind as an item wholly in the physical world"; it never seemed hard even before Gunderson's examples were pointed out to us. On the other hand it is unclear that keeping Gunderson's examples clearly in mind will make much difference for others; it will still seem hard to view one's mind as a physical object. In short, despite Gunderson's ingenious examples, it has yet to be shown that the intuitions people have or lack about the differences between self and physical object are the result of not seeing the points that Gunderson's examples reveal. Paul E. Meehl in 'Psychological Determinism and Human Rationality: A Psychologist's Reaction to Professor Karl Popper's 'Of Clouds and Clocks" criticizes Popper's thesis that human rationality is incompatible with psychological determinism. Part of the reason for not thinking that rationality is compatible with determinism, according to Meehl, is the mistaken identification of determining factors with the non-rational. But although reasons and causes are not identical and are distinct categories, having a reason and giving a reason can be causes of human behavior.

226

RBVInWS

Indeed, one could give a complete physical micro-causal account of human decision making without any explicit mention of reasons. But the truth of such an account would be compatible with a molar account in which having reasons and giving reasons decisively influenced behavior. Furthermore, abstract syllogistic arguments might be physically modeled in certain sub-systems of the human brain. Meehl's arguments are certainly convincing. One might only wish that he could have made them in less than sixty two pages. In perhaps the most disturbing paper in the volume 'Nuisance Variables and Ex Post Facto Design' Paul Meehl concludes that "a large portion of current research in the behavioral sciences, while meeting the conventionally accepted standards of adequate design, must be viewed as methodologically unsound" (pp. 401-402). Meehl's major target for criticism is the common non-laboratory experimental design in which control and experimental groups are matched with respect to some "nuisance variables", variables which "need to be controlled". Meehi persuasively argues that such matching may have the effect of either making the sample unrepresentative or of bringing about systematic unmatching of other variables or of making certain variables (without justification) input variables. Consider the first problem. Suppose one wishes to compare schizophrenics and manic depressives on some psychological variables. Knowing the relation between these mental diseases and socioeconomic class experimentors would normally match their sample of schizophrenics and their sample of manic depressives with respect to socioeconomic class. But in so doing, Meehl argues, "the schizophrenic group sampled is pulled upward from their population social class value and the manic depressive group is pulled downward from their population social class value" (p. 380). This unfortunately makes the sample unrepresentative. The problem posed by Meehl can be construed as a problem in making corrections for nuisance variables. However, in making such corrections counter-factual inferences are involved. Thus Meehl relates the problem of correction for nuisance variables to the philosophical problem of interpreting counter-factuals. In the social sciences, Meehl argues, by and large the meaningfulness and truth of such counter-factual inferences are difficult to ascertain. Thus the problem of making corrections for nuisance variables is connected closely with deep philosophical problems.

REVIEWS

227

Paul Meehl in a third paper entitled 'Some Methodological Reflections on the Difficulties of Psychoanalytic Research' considers some of the methodological problems involved in investigating psychoanalytic theory. Meehl rejects the operationist point of view in evaluating psychoanalysis as well as any requirement that psychoanalysis should at present be reduced to neurophysiology. He adopts instead what he calls a 'neoPopperian' position which is a 'softening' of the orthodox Popperian falsifiability approach. A theory is admissable not only if we know how to test it but if "we know what else we would need to know in order to test it" (p. 412); a theory is not to be given up in the light of adverse evidence "if there are fairly strong results corroborating it" (p. 412) for these results suggest that perhaps the auxiliary hypotheses are false or that the theory, although false, is a close approximation to the truth. Meehl warns that such softening of Popperianism must not lead to clinging to a theory in spite of all adverse evidence. Furthermore, Meehl notes that when one should stick to a theory and for how long are difficult questions. According to Meehl the problem with applying the neo-Popperian view to psychoanalytic research is the difficulty in formulating psychoanalytic theory in such a way that 'high risk' predictions, i.e. precise predictions, can be made from it. Furthermore, use of the psychoanalytic clinical session as a source of data has two problems: (1) the analyst, because of his theoretical biases, may intervene and influence the discourse of the patient to conform to his biases; (2) it is extremely difficult to 'objectify' the classification of the patient's verbal output. The relation between the neo-Popperian position that Meehl adopts and the particular problems in psychoanalytic research that Meehl points out is none too clear. However, one relation seems to be this: Strong corroboration requires precise predictions and, according to Meehl, it is presently impossible to have precise predictions in psychoanalysis. What is unclear, however, is why on the neo-Popperian view psychoanalysis is not therefore rejected. From Meehl's formulation of neo-Popperianism one supposed that if there was adverse evidence and no strong corroboration a theory should be rejected. Meehl seems to admit that there is no strong corroboration in the case of psychoanalysis. There also seems to be negative evidence, given a suitable interpretation of psychoanalysis.8 Yet Meeh/does not reject psychoanalysis.

228

REVIEWS

In the final paper of the volume, 'Popper and Laplace', Michael Radner considers Popper's indeterminism as presented in 'Clouds and Clocks' as well as in 'Indeterminism in Quantum Physics and Classical Physics '4 and contrasts Popper's position with Laplace's and Norbert Wiener's. Radner argues that Popper's critique of Laplacian determinism is misdirected. At most Popper has shown that a version of determinism based on the idea of a prediction machine is false, but not that Laplace's version of determinism, based on the idea of a super intelligence, is false. He draws some interesting analogies between Wiener's indeterminism and Popper's indeterminism and concludes that Lap/ace's paradigm has been rejected by people like Wiener but "no clear and compelling account" (p. 427) has replaced it. Several papers in previous volumes of the Minnesota Studies have for good reason become minor classics in the field: for example Scriven's 'A Study of Radical Behaviorism' in Volume I, Hempel's 'The Theoretician's Dilemma' in Volume II, Feyerabend's 'Explanation, Reduction and Empiricism' in Volume III. Whether any paper in the present volume will achieve that status is very uncertain. Nevertheless, the volume contains many papers that are insightful and provocative and which further recent discussion in several areas of philosophy of science. MICHAEL MARTIN

Boston University

NOTES 1 See B. A. Brody, 'Confirmation and Explanation', JP 65 (1968), 282-299. 2 See Adolf Grtinbaum, 'Space, Time and Falsifiability',PS 37 (1970), 469-588. See esp. pp. 519-520. 8 Michael Martin, 'Mr. Farrell and the Refutability of Psychoanalysis', Inquiry 7 (1964), 80-98. 4 K. R. Popper, 'Indeterminism in Quantum Physicsand in ClassicalPhysics, Part II', BJPS 1 (1950), 173-195.

Wolfgang Stegmiiller, Probleme und Resultate der Wissenschaftstheorie und Analytischen Philosophie, Band 11: Theorie und Erfahrung. SpringerVerlag, Berlin-Heidelberg-New York, 1970, XV +485 pp., DM 7 6 , US $22.00. In two rather long volumes Stegmiiller presents comprehensive discussions of most of the important general problems in the philosophy of science. Volume I, Wissenschaftliche Erkliirung und Begriindung, is concerned with explanation, laws, causality, statistical systematizations, teleology, and related topics. Joseph J. Kockelmans' informative review of this first volume can be found in Philosophy of Science 38 (1971), 126-132. Volume II, Theorie und Erfahrung, is concerned with concept formation in the sciences and with certain aspects of the structure of theories. This volume can be read independently of the first. Theorie undErfahrungis divided into three main parts. Part A, consisting of Chapters I-II, is concerned with conventional and empirical factors in concept formation and the construction of theories. Part B (Chapters III-IV) discusses problems about the empirical significance of scientific languages and investigates the motivations for the use of partially interpreted theoretical terms. Part C (Chapters V-VII) discusses the partial interpretation of theoretical terms, as well as the elimination of such terms by means of Craig's theorem and Ramsey sentences. In addition there is an appendix concerned with the problem of the logical consistency of the use of classical probability calculus in quantum theory. There is also an epilogue entitled, 'What is Scientific Progress?'. The book contains a useful bibliography and index. Stegmiiller's book consists largely of a critical exposition of investigations of the above topics by members of the logical empiricist, and related, schools. Thus he relies heavily on the work of such philosophers as Carnap, Reichenbach, Braithwaite, Hempel and Nagel. However, the book is not a mere historical exposition; it is a well organized, systematic investigation which has considerable scope and great depth. Thus it will be most practical to review briefly the most important general characteristics of the author's investigations. Chapter I contains an excellent discussion of the techniques of construction of classificatory, comparative, and quantitative concepts. Mathematical technicalities are kept to a minimum, for the author is not in-

Synthese 25 (1972) 229-233. All Rights Reserved Copyright © 1972by D. ReidelPublishing Company, Dordrecht-Itolland

230

REVIEWS

terested in presenting detailed axiomatic formulations of many types of measurement systems. The emphasis is rather on basic semantical and epistemological aspects of concept formation. Stegmfiller is especially concerned with the interplay of five factors in concept formation. These factors are conventional stipulation, empirical findings (e.g., experimental data), hypotheses (e.g., hypothesized empirical generalizations), simplicity considerations, and fruitfulness (e.g., in the formulation of laws). Stegmiiller illustrates these factors in several examples of concept formation, including discussions of the measurement of length and time. The last two of the five factors can help to determine the choice of a particular convention from a set of many possible conventional choices. Chapter II, which discusses theory construction, also is concerned with these five factors. This chapter contains an analysis of Newtonian mechanics which describes various possible interpretations of particular Newtonian laws as either conventions or as empirical hypotheses. In addition, Chapter II contains very clear expositions of Poincart's view of the relative simplicity of Euclidean geometry, and of Reichenbach's interpretation of the relative simplicity of Einstein's general relativity theory. Parts B and C together (which comprise about 250 pages) contain the most clear and detailed discussion I have ever seen of the problems surrounding empirical significance and theoretical terms. Unfortunately, because of the great amount of detail presented, this discussion is in some places tedious. However, the presentation is well organized and the author's judgments are generally quite reasonable and well-defended. Chapter III begins with a statement of the basic thesis of empiricism, which Stegmiiller says is composed of two subtheses. The first of these asserts that any meaningful statement is either synthetic or has its truthvalue analytically determined, and also asserts that the truth-values of the latter statements are determined only by linguistic conventions and are immune to change by experience. The second subthesis, which is Stegmiillefts concern, asserts that the truth-value of any synthetic statement can only b e determined by empirical means. According to Stegmiiller, the problem of formulating a criterion of empirical significance is the problem of developing a precise formulation of this second subthesis. He then proceeds with a critical exposition of some of the early attempts to develop criteria based on verifiability, falsifiability, and confirmability. Concluding that these have all failed, and that similar attempts are likely to fail, he

REVIEWS

231

then discusses criteria based on translatability into a so-called empirical language. He attributes this idea (at least in part) to Carnap's famous paper, 'Testability and Meaning'. On the basis of an observation of Scheffier about the concept of translation, Stegmfiller revises this criterion to one of membership in an empirical language. This still leaves the problem of describing an empirical language. Originally it was supposed that this language would have observation terms and in addition only other nonlogical terms which are definable in terms of the observation terms. In Chapter IV Stegmiiller reviews several of the problems, most of which are by now well-known, involved in trying to define certain kinds of important scientific concepts in an observation language. In addition, there is a discussion of what Stegmtiller calls the Braithwaite-Ramsey Conjecture. This, very roughly, is the conjecture that a theory containing undefinable theoretical terms will in general have greater predictive and explanatory power than one whose theoretical terms are all definable in the given observation language. Although this conjecture (in one form or another) is not uncommon nor implausible, it is rather troublesome to state clearly and prove rigorously. In his Scientific Explanation Braithwaite presents in a complicated way an example to illustrate the conjecture. Stegmiiller has a very elegant presentation and discussion of this example. He also formulates the conjecture in a fairly precise manner which is illuminating and thought-provoking. On the basis of Chapter IV, Chapter V begins with the premise that there are compelling reasons for the use of partially interpreted theoretical terms. Thus the total scientific language is supposed to consist of an observation language plus a theoretical language. The terms of the latter are to be partially interpreted with the help of correspondence rules. According to Carnap, in 'The Methodological Character of Theoretical Concepts' (Minnesota Studies in the Philosophy of Science, Vol. I, (1956)), this leads to the problem of formulating a criterion of empirical significance for theoretical terms. Stegmiiller presents a lengthy critical exposition of Carnap's criterion, including a discussion of previous critiques of it by P. Achinstein and D. Kaplan. After all is said and done, he concludes, for a variety of reasons, that the search for a single criterion of empirical significance is doomed to fail. Indeed, the author suggests on the basis of several plausibility arguments that there is no single explicandum which such a criterion is to explicate. On this view the logical positivists' attempt

232

REVIEWS

to eliminate metaphysics by a significance criterion was misguided. This view also suggests that one should not identify the search for a criterion of empirical significance with the problem of precisifying the empiricist's second subthesis. I wish that the author had included more direct discussion of the consequences, for empiricism, of this latter suggestion. He does, however, make some interesting comments about empty (and thus 'meaningless') questions which are empty because they presuppoCe empty theories. Chapter VI discusses Craig's theorem on the elimination of theoretical terms by axiomatization of the set of observation sentences of the theory. As usual, Stegmiiller's exposition is thorough. He carefully states the theorem and outlines the proof. This is followed by a discussion, which relies partly on previous work of Hempel, of the epistemological significance of the theorem. Chapter VII contains what is probably the most extensive treatment of Ramsey sentences available. Among the several interesting issues discussed, there is a consideration of the effectiveness of Ramsey sentences in producing inductive connections. StegmiJller reviews the two papers on Ramsey sentences by I. Scheffler and H. Bohnert which appeared in The Journal of Philosophy 65 (1968), No. 10. Scheffler claims that the Ramsey sentence of a nonanalytic theory which allegedly establishes certain inductive connections can in some cases be analytic and not yield these inductive connections. This is illustrated by a simple example. Bohnert replies by citing several reasons which allegedly show that Scheffter's example does not represent a serious shortcoming of Ramsey sentences. The issues are complex, and StegmiiUer reviews them well, but it seems to this reviewer that he too quickly sides with Bohnert in an argument which is not conclusive. Stegmiiller admits the inconclusiveness and attributes it in part to the need for better understanding of inductive arguments. No doubt this is correct, but it would have been nice to see some further discussion. Similarly, in another connection Stegmiiller seems too quick to conclude that a nonanalytic theory is empirically trivial if it has an analytic Ramsey sentence. This conclusion is based on the premise that a theory with no (nonanalytic) observational consequences is empirically trivial (p. 423). However, this premise should not be used without qualification. It is especially dangerous to apply it to single hypotheses. For instance, when

REVIEWS

233

it was first proposed, Avogadro's hypothesis by itself probably had no nonanalytic observational consequences. Yet it was nontrivial, in an important sense of 'nontrivial', because it led to observational consequences when used in conjunction with suitable auxiliary hypotheses (some of which were supplied by Avogadro). Thus, Stegmiiller's premise should at least be qualified as applying only to 'well-developed' theories, whatever those are. But when it is so qualified, the premise becomes less interesting because it has less practical applicability. It would also appear that some of Stegmiiller's conclusions about Ramsey sentences become less interesting, when these conclusions are suitably and similarly qualified. It is not possible to mention here all of the many interesting issues Stegmiiller discusses. In the above, I have only indicated some of the general topics covered in his book. An overall evaluation is now in order. Theorie und Erfahrung is primarily an expository book. One who is familiar with modern writings in the philosophy of science will not find many surprises in it. On the other hand, Theorie und Erfahrung is a beautiful book, for it is written in a clear and simple style, and it is a paragon of sincere, careful, philosophical investigation. Stegmiiller exhibits a good sense for what is philosophically significant. Although there are occasional digressions, on the whole the arguments are presented in a straightforward, elegant manner. In addition, examples from the sciences are described with great clarity. My only criticism is concerned with emphasis. I do not believe that the problem of empirical significance demands the very lengthy, and sometimes tedious, discussion which is devoted to it. In spite of this, the book will probably become an effective text, and it certainly will be a valuable reference work for any philosopher interested in the issues reviewed above.

University of Texas at Austin

ROBERT L. CAUSEY

Jaakko Hintikka and Patrick Suppes (eds.), Information and Inference, Synthese Library, D. Reidel Publ. Co., Dordrecht 1970, vii + 336 pp. This is a collection of nine original papers by the editors of this volume and their associates, past and present, at Helsinki and Stanford Universities. The most prominent common theme is the use of probabilities, but they are otherwise widely different in subject matter and aims. Four of the papers concentrate largely on the uses of probabilistically defined measures of information in the formulation of methodological principles in science, and since this is the subject treated which will probably be of most general interest to philosophers of science these will be discussed first, beginning with a group of three - 'On Semantic Information' by Hintikka, 'On the Information Provided by Observations' by Risto Hilpinen, and 'Quantitative Tools for Evaluating Scientific Systematizations' by Juhani Pietarinen- which share much the same aims and assumptions, and which will be discussed together. The three authors propose a variety of measures either of information (absolute, or relative, as when we speak of the information provided by a theory about certain data), or of the systematic power of such and such a theory or hypothesis to, roughly, 'reduce our uncertainty' concerning a given body of data. What the different measures have in common is the fact that, where h is the hypothesis and d is the 'data set', all measures are defined to be functions of suitably defined probabilities P(h), P(h/d) (the probability of h given d) which are assumed to satisfy the usual axioms of probability but are otherwise left unspecified (except in Hintikka's case, since he identifies them with logical probabilities). For instance, Hintikka introduces two possible measures of the information provided by h about d: namely Cout (h/d) = P(d &,~ h)/P & d; and Inf(h/d) = - logP(h/d). Hilpinen considers still another relative information measure and Pietarinen considers four possible measures of systematic power, all essentially variants on the above theme. A number of methodological claims about information, systematic power and so on are discussed in the light of the proposed information measures, among which the following may be mentioned. (1) Hintikka comments briefly on a Popperian thesis to the effect that scientists should aim for informative theories, rather than probable ones, noting that informativeness and probability must conflict if Cont (h/d) measures inforSynthese 25 (1972) 234-240. All Rights Reserved Copyright © 1972 by D. Reidel Publishing Company, Dordreeht-Ilolland

REVIEWS

235

mation (pp. 8, 9). (2) Pietarinen (pp. 133-4) gives an illuminating discussion tending to counter claims of Popper's that high informative content, testability, logical strength and systematic power all go together. (3) Pietarinen (pp. 139-43) formulates five rules for accepting hypotheses, some of the flavor of which can be gathered from the following rough version of his Rule 1 : accept a hypothesis if and only if it is sufficiently probable and sufficiently informative (about some given body of data d). (4) Hilpinen uses his information measures to clarify claims (such as Carnap's Principle of Total Evidence) according to which it is desirable to gather more, or more weighty, evidence in relation to a hypothesis, irrespective of the probability of the hypothesis on the evidence. I will shortly question the appropriateness of the above information measures, or any measures of this type, to do the kinds of methodological jobs which these authors propose for them, but for the moment I would like to query the basic undertaking here of proposing 'precise' information measures in terms of which to formulate methodological claims, where what is unclear about such claims is not so much what is meant by 'information' but just what it is they recommend - and why. One can cook plausible information measures ad infinitum, but unless he Can give coherent sense to such claims as that science should 'aim for' theories which are informative but not necessarily probable, these measures must remain idle tools. What does it mean to say that a theory is probable? The sense of probability which seems most appropriate to these endeavors is either logical or subjective: i.e., a theory is probable if it either is or ought to be believed. But it verges on contradiction to claim that science should aim for theories which are informative but which either are not or ought not to be believed. It is similarly paradoxical to formulate rules of acceptance according to which theories which are or ought to be believed ought not necessarily to be accepted. The present writers are not the originators of this sort of incoherent talk about what scientists ought to do or think or strive for, but they do, unfortunately, 'preserve the tradition'. Roger Rosenkrantz' paper 'Experimentation as Communication with Nature' sketches a framework (derived from a theory of experimentselection due to Raiffa and Schlaiffer in their book Applied Statistical Decision Theory, Harvard Business School, 1961) within which it should be possible under certain circumstances to rationalize claims to the effect that one should seek 'informative' data, by showing what the utilitarian

236

REVIEWS

value of such data-acquisition might be. Leaving aside what is inessential to our present concerns, the four basic elements of the framework are: (1) possible experiments, e, which might be made (which could just as well be called observations), (2) possible results, z, of the experiments, (3) possible actions, a, which could be taken subsequent to experimenting, and (4) possible states of nature, s. The significant outcomes of the action, a, taken are assumed to be determined by the act and the state of nature, but the experimentor or observer is supposed to decide which action to take on the basis of maximizing expected utilities. These expected utilities are determined by the probabilities which the observer estimates for the different states of nature, and these probabilities are determined in part by what the observer observes the result of his experiment to be. According to the theory these are posterior probabilities determined by Bayesian inference from prior probabilities for the states of nature, plus 'given" conditional probabilities that experimental result z will occur if experiment e is performed and the true state of nature is s. The foregoing framework permits the definition of both a 'utilitarian value', v(e, z) of performing experiment e and observing result z, and an 'informational' or 'decrease in uncertainty as to the true state of nature' value, I(e, z) of performing e and observing z. v(e, z) is the difference between the expected utility of the action which would be taken if e were performed and z observed and that of the action which would be taken 'blind' if no experiment were performed and the experimentor acted on prior probabilities. The uncertainties entering into the computation of I(e, z) are measured as the entropies of the posterior and prior probability distributions over the states of nature. Taking expected values with respect to z (holding e fixed) gives measures v(e) and I(e) of the expected utilitarian value of the information derived from performing e, and of the expected 'intrinsic' information increase resulting. Both of these values are always non-negative, and this can be interpreted to mean that as long as experimentation or observation costs nothing it never hurts to perform it (it is desirable in a utilitarian sense to do it), and furthermore the probabilities are that 'intrinsic information' as to the true state of affairs will be gained in this way. Hilpinen (p. 103) accepts this interpretation as justifying a version of Carnap's Principle of Total Evidence, though we shall question this in a moment. The foregoing suggests how one might go about justifying methodolo-

REVIEWS

237

gical claims as to the desirability of doing something like 'maximizing information-acquisition', for if it could be shown that increases in intrinsic information l(e, z) always went with increases in the utilitarian value v(e, z) of the information, this would show the utilitarian value of acquiring information. Unfortunately, Rosenkrantz, drawing on an example originally presented by Joseph Sneed ('Information, Entropy and Decision', Synthese 17, 392-407), shows that these two measures are not always monotonically related, so the utilitarian justification for information-acquisition is not carried through. What one would like to see (perhaps it already exists in the literature) is a more detailed study of the intereonneetions between the two measures, where the information measure might be expected to appear, as does the entropy measure in the fundamental coding theorems of information theory, in inequalities bounding the utilities v(e, z). Rosenkrantz does not discuss this, however, and his paper concludes with some highly suggestive remarks as to how considerations applicable to informativeness of experiments and their results might be extended to theories. A criticism applicable to all four of the papers so far discussed has to do with the fact that the information measures they employ measure what might be called the quantity of the information, but not its quality. To illustrate suppose that you and I both begin by thinking there is a 50-50 chance of the sun's rising tomorrow, then you get information leading you to think the chance of its rising is 0.999 and of its not rising is 0.001, while I get information leading me to the opposite conclusion: that the chance of the sun's rising is 0.001 and of its not rising is 0.999. Measuring information by reduction in entropy, we have both gotten the same amount of information. But I think we would say intuitively that we have not gotten equally good information. One way in which this distinction is important has to do with the possible use of information measures in the formulation of methodological recommendations. For it seems to me to make little sense to recommend the search for informative theories in science, if the measures used to determine the informativeness of the theories are blind to the quality of the information they provide. 1 1 Joseph Hanna has considered the problem of devising measures of the quality of information contained in a set of probability estimates in his paper 'A New Approach to the Formulation and Testing of Learning Models', Synthese 16, 344--380. His measure turns out to be a generalizationof the entropy measure.

238

I~EVlEWS

Taking into account the fact that information and the probability estimates on which it is based may be better or worse, and not just more or less, also brings into question the interpretation of the non-negativeness of v(e) as meaning that cost-free experimentation is never undesirable. It can be undesirable if the experimentor misjudges the prior conditional probabilities associating experimental outcomes or results of observation with possible states of nature (the possibility of this sort of mistake is left ambiguous in describing these probabilities as 'givens'). For example: a pedestrian who thinks the green light means 'stop' and the red light means 'go' would be better off not to look at the signal. I must be unjustly brief in commenting on the contents of the remaining five papers of this volume. Dean Jamison's paper 'Bayesian Information Usage' concerns the use of second and higher order 'probability of probabilities' theories as a way of accounting for or 'rationalizing' first order theories of probability-change in the light of evidence. Readers will call to mind Laplace's theory leading to his rule of succession as an instance of this approach (supposing that we initially suppose all first order probabilities that an instance of the series will have the given property to be equally probable, and use Bayesian inference to update the probabilities of first order probabilities). The advantage of the approach is to make possible a plausible justification of what seem to be very ad hoc rules of first order probable inference (like Laplace's rule). Jamison shows how both Carnap's-continuum and Hintikka's two-dimensional continuum of inductive methods (both formulated originally in terms of first order probabilities) can be represented as instances of second and third order probability models, respectively, and the former can be generalized thereby in a natural way. The most useful contribution of this paper, though, will probably be its sketching of important mathematical techniques for working with higher order probability models (using Dirichlet integrals), since undoubtedly the most important bar to using such models is the mathematical problems to which they give rise. Zoltan Domotor's paper 'Qualitative Information and Entropy Structures' takes the fundamental measurement theory approach to the measurement of entropy and information. That is, he supposes that we are initially provided with qualitative relations of 'greater or lesser informativeness' and the like holding either between propositions or 'experiments' with mutually exclusive possible results, and we ask what conditions these

REVIEWS

239

qualitative relations must satisfy if there are to exist standard (entropic) measures of information which 'agree' with the relations (and how uniquely the relations 'determine' the measures). The mathematical problems encountered are considerably more difficult than those dealt with in the more elementary parts of fundamental measurement theory, and most of the attention is centered on these rather than the substantive question as to what intuitions are as to whether one proposition is more informative than another, or whether particular axioms formulated about that are true. Nonetheless, some interesting representation theorems are proved, and novel methods are used in their proofs. It is impossible to do more than describe the main themes of the two papers 'Surface Information and Depth Information' by Hintikka and 'Towards a General Theory of Auxiliary Concepts and Definability in First-Order Theories' by Hintikka and Raimo Tuomela, since both deal with issues in the philosophy of science through application of Hintikka's theory of distributive normal forms, and their main contentions are scarcely intelligible without a background in that theory. The first of these papers utilizes the distributive form theory, with its concept of a constituent generalizing and relativizing to depth the Carnapian idea of a Qpredicate (maximally consistent predicate expression in a monadic calculus with a finite number of predicates), to suggest ways for generalizing to languages with n-ary predicates familiar symmetry principals in the foundations of probability formulated in terms of constituents. Because constituents are now relativized to depth (defined by the number of nested quantifiers involved), we get a series of symmetric probability assignments, one for each depth, d. Because it is possible for a constituent which is not trivially inconsistent at depth d to nevertheless be inconsistent (which must ultimately show up when it is 'analyzed' to greater depth), it is possible to assign non-zero probability at depth d to formulas which are inconsistent, but not trivially so at depth d. Hintikka exploits this possibility as a way of capturing the idea of the information yielded by the discovery of inconsistency (or logical consequence), which goes along with the change in probability from the level at which the formula has non-zero probability to that at greater depth where it is trivially inconsistent and so has zero probability. Not all details of Hintikka's way of developing this idea seem to me very plausible (for example, change in probabilities from one level to a deeper one does not have to be Bayesian - so that discovery of con-

240

REVIEWS

tradiction would appear to be fundamentally different from discovery of mere falsehood, and the questionable Cont measure of information is still used), but the guiding intuitions seem most interesting and are easily separable from the more doubtful details of their application. The Hintikka-Tuomela paper investigates certain advantages to be gained by adjoining auxiliary concepts (possibly by explicit definition) to a theory involving 'primary concepts' in which we are interested. Their principal technical tool is a method of'projecting out' those consequences expressed in terms of the primary concepts alone of a theory involving both primary and auxiliary concepts, which are in a sense 'obvious' when the theory is analyzed into its distributive normal form at depth d. Any 'primary statement' which is a consequence of the theory must ultimately be projected out in this way, but only if the set of these consequences is finitely axiomatizable will all of the 'primary content' of the theory be arrived at after a finite number of steps. This, of course, is one of the possible advantages of axiomatizing with auxiliary concepts since it is well known that theories with auxiliary concepts may be finitely axiomatizable while their 'projections' (set of consequences in the language of primary concepts alone) may not be. Other advantages which can be brought out by the projecting technique (e.g., of introducing explicit definitions) are discussed. The last paper 'Learning and the Structure of Information' by Dean Jamison, Deborah Lhamon, and Patrick Suppes presents a detailed exposition and comparison of probabilistic models of paired-associate learning. All that an outsider can say on being confronted with this wealth of complicated models, each with its ponderous superstructure of theorems, is "God help us when they get around to working with triples". That may be the psychological version of the three-body problem. Anyway, the focus of the paper is on empirical models and not methodological issues, and since this review is concerned with the latter I shall not attempt further summary.

University of California, Berkeley

ERNESTW. ADAMS

Norwood Russell Hanson, Perception and Discovery, Freeman, Cooper & Co., 1969, 435 pp. Before he died in the crash of his airplane, Norwood Russell Hanson had been writing an elementary textbook in the philosophy of science. In this volume his former student, Willard C. Humphreys, has brought Hanson's goal to fruition by carefully editing substantial portions of Hanson's lecture notes, which were to be used as a basis for the text, along with some material that Hanson had completed or nearly so. The result captures Hanson's dynamic style with remarkable fidelity, and will be welcomed by students as well as teachers who complain of lifeless presentations of so much material in the philosophy of science. Hanson was a van Gogh rather than a Vermeer among philosophers. He lacked patience for fine detail. His penchant was for rough brush strokes that would suggest the important features and that might stimulate others to pursue details. But these brush strokes also were intended to make one see old things in novel ways. This procedure has many advantages that students should find valuable, although it does make the work harder for those who want detail, especially detailed argumentation, before they can accept a philosophical view. For these students the book will be hard-going indeed! A large portion of the text is devoted to the concept of seeing. Do the 13th century and 20th century astronomers see the same thing when they look to the east on a clear morn? A Yes answer, says Hanson, is offered by those who define 'seeing X' as 'having a normal retinal reaction to X' (p. 66), a No answer by those who define it as having an X-like visual sense-datum (p. 70). But both definitions are inadequate, Hanson argues. The right view is to realize that seeing, other than the sort of 'phenomenal' seeing we do when the occulist gives us an examination, involves 'seeing as'. When we see an object X we see it as something Y, which means that we "see that if certain things were done to the objects before our eyes, certain other things would probably follow" (19. 112). Seeing X, therefore, involves knowing things about X. So the 13th and 20th century astronomers don't see the same thing, but not because their retinal reactions are different and not because their sense-data are 'private', but because the 13th century astronomer sees what he does as a body revolving about the earth, whereas the 20th century astronomer does not. Synthese 25 (1972) 241-247. All 2~ightsl~eserved Copyright © 1972 by D. Reidel Publishing Company, Dordrecht-Holland

242

REVIEWS

Hanson's attempt to develop Wittgenstein's ideas about 'seeing as' and to apply them to problems in the philosophy of science is lively and provocative, but is it right? To begin with, why does one need to assume, as Hanson seems to do, that if we say that the two astronomers Hanson describes see the same thing, then we must be adopting the 'retinal' definition of 'seeing'? Many philosophers have wanted to say that these astronomers do see the same object, viz. the sun, but that they simply have different beliefs about its motion, and such philosophers have not adopted a 'retinal' definition of 'seeing'. Recently this latter point of view has been defended very persuasively by Fred I. Dretske, in his excellent book Seeing and Knowing, on the basis of a definition of 'seeing' that is non-retinal. According to Dretske there is a non-epistemic sense of 'see', a sense in which if a person A sees X this does not imply that A has any particular beliefs (and hence any particular knowledge) about X or about his seeing of X. Briefly, on his view, if A sees X then X looks some way to A and this is such as to allow A to visually differentiate X from its immediate environment. In accordance with this definition, our two astronomers could both see the s u n - the same sun - if the sun looked some way to each, a way that would allow each to visually differentiate it from its immediate environment. Such astronomers could hold quite different beliefs about the sun, and indeed on this definition the sun could look different to each, yet we would not have to say that each sees something different. Let me now comment on some of Hanson's specific claims about seeing. (1) Seeing X involves ~eeing X as IT, and this involves knowing something about X (p. 107). Such a claim is incorrect if there is, as Dretske claims, a non-epistemic sense of 'see'. (2) Cases of seeingin which 'seeing X' doesnot entail "knowing what X is' are cases of "phenomenal" seeing, in which what we see are "shapes, color patches, oscillations, andpointer readings" (pp. 108-110). Again, if Dretske is correct in claiming that there is a non-epistemic sense of'see', then when, say, the native sees the airplane for the first time and does not know what it is, he does see the airplane and not simply shapes and color patches. (3) Seeing X as Y entails seeing that X may be expected to behave in all the ways that Y's do (p. 116). This seems incorrect for certain fairly standard uses of 'seeing X as Y'. To modify an example used by Austin, I

REVIEWS

243

may see yonder cloud as a horse one moment and as an antelope the next, but I don't expect it to behave in all the ways that horses or antelopes do. The point is that Hanson wants to claim that seeing as involves seeing that, and this just isn't so. Our 13th century astronomer may see the sun (or 'that object in the sky') as a body that revolves around the earth. But it doesn't follow from this that he sees that the sun is a body that revolves around the earth. For that would imply that the sun is such a body. One can see X as a Y without seeing that X is a Y, without seeing that X may be expected to behave in all the ways that Y's do. (4) Seeing objects involves having visual impressions which we see as objects (p. 111). This is at best very dubious. If I see a chair I don't see my visual impressions as a chair, for I don't see them at all (a point Hanson seems to grant earlier, on p. 82, in his critique of the sense-datum theory of seeing). Hanson wants us to note ai~inities between seeing ordinary objects and seeing reversible and shifting-aspect figures in drawings, e.g., the 'duck-rabbit' figure. In the latter case it is appropriate to speak of seeing the assemblage of fines as a duck or as a rabbit. That's because in such eases we have an ambiguous figure. But what reason is there for supposing that all figures, and indeed all objects, are ambiguous in all contexts. When there is an ambiguity or the likelihood of one in a given context, or when X looks to one like something which it isn't, there is a point to the 'seeing as' locution, but otherwise to say that I see the object in front of me (the chair) as a chair is to say something inappropriate, because (as Carnap would have put it) a 'pragmatic' convention associated with the use of 'seeing as' is being ignored. (I might say that I see it simply as a chair in rebutting the suggestion that it is ambiguous or that it looks like a non-chair.) More importantly, if Dretske is right, to say that I see the chair as a chair is to say something quite possibly false. I may see the chair without seeing it as anything at all (without having any beliefs about it or about the way it looks). Perhaps Hanson's claim that seeing X involves seeing something as an X means that seeing X involves seeing something that looks like an X. But that could not be true, since one might see X in a situation in which it didn't look like an X at all. I think the most that Hanson could say here is that if a person sees X t h e n Xlooks some way to him. This needn't suggest any ambiguities or X's looking like something which it isn't, and if Dretske is right, it is indeed part of what is meant by a claim that someone

244

REVIEWS

sees X. However, I would hesitate to speak of this as seeing something as an X, since (I believe) the latter expression is used in such a way that I can see something as an X without there being any X to see, whereas I cannot see an X unless there is an X to be seen. (5) The 13th century astronomer sees the sun as moving around the earth, the 20th century astronomer sees it as something different (p. 119). But if, as Hanson claims, seeing as entails seeing that, and if, as I claim, seeing thatp is true entails thatp is true, then the 13th century astronomer cannot see the sun at all, in Hanson's sense. So much for the concept of seeing. Now let me comment briefly on two other topics discussed. (A) Facts. Hanson attacks the view that facts are independent of language and of our beliefs about the world. We say that grass is green, whereas the Arab says that the grass greens, these are both facts, and they may be mutually exclusive (p. 178). But if they are mutually exclusive, then it can't both be true that grass is green and that grass greens. So how can both of them be facts? (I find it very difficult to abandon the view that i f p is true then that p is true is a fact.) Hanson seems to be making an inference from (1) language, and therefore our views embedded in it, affects what we take the facts to be, to (2) language affects the facts. (1) seems reasonable enough, but it doesn't entail the relativism espoused in (2). Sometimes Hanson is content just with asserting (1): "The traces of the language in which we state the facts are never absent from our conceptions as to what the character of the facts actually is" (p. 187). But often he suggests (2): "The shape of the facts is determined largely by the language in which we state those facts" (p. 187). (B) Causation and Theory-Laden Terms. As he did in his earlier work Patterns of Discovery, Hanson here emphasizes the 'theory-laden' character of causal talk. "When we say that A causes B, we say that because 'A' is a word that is ... impregnated with logical considerations. It is ... theory-laden, rule-governed, and anything but neutral" (p. 313). There are, Hanson urges, certain words like 'wound' and 'poison' that are 'causewords' - theories, diagnoses, analyses are built into them. 'Effect-words' are "much less charged with theory and hence far less able to explain their causes" (p. 303). In a completely 'phenomenal' language - in a language containing only words like 'red patch', 'smooth', 'sour' - "causal connections could not even be expressed" (p. 302).

REVIEWS

245

Hanson's view that there are cause-words that explain and effect-words that don't is, at best, an exaggeration. It isn't words that explain: the word 'poison' doesn't explain anything (a point Hanson grants later). Moreover, even to say that only certain words can designate causes, or only certain words can appear in explanations, is a gross exaggeration. True, I can use Hanson's cause-word 'wound' in an explanation of my scar ('scar' being an effect-word for Hanson). But equally I can use the word 'scar' in explaining what is causing my itch. Indeed, even Hanson's 'phenomenal' terms can be used in causal explanations. I might say that the appearance of that yellow patch caused a sour taste in my mouth. Perhaps, then, we should relativize the notion of theory-dependence by speaking of one word as being theory-laden by contrast with another. Hanson's thesis would then be that ifA causes B then the term 'A' is theory-laden by contrast with the term 'B' (or 'A' is more theory-laden than 'B'). What might such a claim mean? (1) It might mean that the fact that A occurs can be used in an explanation of why B occurs. (2) It might mean that there is some theory (to which the term 'A' is tied) which together with the fact that A occurs explains why B occurs. If (1) is what Hanson has in mind by 'theory-dependence', then his claim that A causes B only if 'A' is theory-laden by contrast with 'B' is true but philosophically not very illuminating. It is just the claim that if A causes B then you can explain why B occurs by appeal to the fact that A occurs. In short, it is just a claim that there is an intimate relationship between causation and explanation. If (2) is what Hanson has in mind, then his claim that A causes B only if 'A' is theory-laden by contrast with 'B' is much more exciting, but, I think, false, given any fairly standard interpretation of the notion of theory. If I claim that wounds cause scars I am, to be sure, committed to saying that scars can be explained by appeal to wounds. That's point (1) above. But need I be committed to saying that there is some theory to which the term 'wound' is tied which, together with 'Jones was wounded' explains Jones' scar? No doubt I believe that some explanation of how wounds cause sears can be found, and this may even be a theory. But is my use of'wound' tied to that theory? Not in any sense that I know. I may certainly use the term 'wound' properly without knowing any such theory. In short, I may know that A causes B without knowing how A causes B or without knowing any theory which explains how. Of course, I do know that wounds cause scars, and if that's a 'theory' then I know a theory about

246

REVIEWS

wounds. But if that is all Hanson has in mind, then his position would be even less illuminating than (1) above. It would amount to saying that if I claim that wounds cause scars then I am committed to the 'theory' that wounds ¢ause scars. Towards the end of his chapter entitled 'Theory-Laden Language' Hanson claims that "cause and effect talk ... is always talk within a definite context. It ought not to be considered just in general" (p. 311). When we say that fire causes heat, it's not that the word 'fire' explains the word 'heat' but rather that in a given context these words are short for propositions, say, 'there is a fire in the building' and 'there is heat in the building'. The problem with this is that we do make general causal claims like 'fire causes heat' or 'smoking causes ill health' where no particular context seems to be envisaged. Is Hanson's position merely that the words 'fire' and 'heat' can be used in certain contexts to make a causal statement? Or is he saying that any causal claim, even a general one, must in some way be relativized to a context? If the former is his view, then, as I have urged, there are pairs of terms from his 'phenomenal' list for which the same claim can readily be made. (And anyway if a word can function as a 'cause-word' in one context and not in another, why draw up lists of 'cause-words' and 'effect-words' in the first place?) It the latter is his position, then he owes us an explanation of how context is supposed to operate to produce a causal claim. Why and how must causal statements like 'fire causes heat' be relativized to a context? There are numerous other topics Hanson discusses in the volume, e.g., definitions, laws, and probability, and he has stimulating things to say. He makes one think. I have chosen to comment on his discussion of seeing, facts, and causation because there is a unifying Hansonian theme here: theory-dependence. Scientists confront nature with theories when they see, when they report the facts, and when they make causal judgments. Seeing is not simply opening one's eyes and looking but a rich, theory-infused activity involving seeing that objects would behave in certain ways under appropriate conditions. Facts are not entities which exist completely independently of us, they have human ties, they are affected by our theoryladen language. And to say that two events are causally related is not simply to speak of a spatial or temporal connection between them but of a relationship that exists in virtue of our theories which connect them, theories to which causal words are tied.

REVIEWS

247

Now I think that in each of these cases there is a truth that Hanson is getting at, though in each instance he overshoots the mark. Seeing that can indeed be theory-laden. The physicist may see that alpha particles are passing through the cloud chamber, and he would not be able to do so if he lacked a theory or at least some information about alpha particles and cloud chambers. But not all seeing, even scientific seeing, is seeing that, or at least Hanson has not demonstrated that it is. There is a plausible case to be made (and Dretske has made it) for distinguishing seeing that is theory-laden from seeing that is not. Second, how we express what we take the facts to be does depend on our language and on our views (our 'theories') about the world. But it doesn't follow from this that the facts themselves are determined by our language and our theories. Third, if we say that A causes B we do suppose that B can be explained by appeal to A. We may even suppose that there is some theory of A that will explain how A causes B. But we need not suppose that the use of 'A' is tied to any such theory in order to claim with justification that A causes B. We need not know of any such theory in order to make this claim. To be sure, there are cases when we make a causal claim that we would defend by appeal to a theory showing how A does cause B. But there are cases in which we make a causal claim without knowing any such theory. Characteristically, Hanson loves to produce extreme cases. That's one of the things that makes him exciting and provocative. Either seeing is simply what meets the eyeballs or else it is swimming in theory. Either facts are like planets or else they are like poems. Either causation is 'one damn thing after another' or else it is theoretically supercharged. Happily (I think) one can accept the importance of theories in our talk about seeing, facts, and causation without being committed to any of these extreme positions.

The Johns Hopkins University PETER ACHINSTEIN

R. Hart6, The Principles of Scientific Thinking, London, Macmillan, 1970, 324 + x pp., ~ 4.50. In the early seventeenth century Francis Bacon revived the organon of the natural sciences, recommending that phenomena should be classified and analyzed according to the Natures and Powers of the bodies in which they manifested themselves. About a hundred years later David Hume raised a series of fundamental and perplexing questions about all such qualities and efficacities. What grounds do we have for attributing to them inherence or potency, other than appearances and sucessions of appearances? If we have no other grounds, may not appearances be deceptive and apparent connections illusory? If our fundamental beliefs about the workings of nature may be riddled with deceit and illusion, might not events surprise even our most confident expectations? And if this is so would it not be the part of modesty, not to say wisdom, to restrict our confidence to what is less precarious, such as descriptions of apparent fact and manipulations of language that are under our own control? Nobody has ever given answers to these questions that can offer the slightest reassurance to those who find Hume's brand of scepticism disturbing. Not that Hume himself ceased to believe, or expected anyone else to cease to believe, in the efficacy of natural causes as a matter of empirical confidence; he was a philosophical and not a scientific sceptic. He anticipated charges of inconsistency on this score: "My practice, you say, refutes my doubts. But you mistake the purport of my question" (emphasis added). The force of his work, and of the work of those who followed him in a tradition whose most recent avatar is logical empiricism, was to ask what limits are imposed on our knowledge by this restriction to appearances as its source and warrant, how within such limits it can best be organized, and what risks are incurred when we are forced to venture outside them, as for practical purposes we often are. One line of enquiry these questions have naturally led into has been the investigation of claims to knowledge (such as religious or metaphysical claims) that can be shown to lie wholly outside the limits; another has been the investigation of the structure of science, because as Bacon correctly saw science is the most perfected species of human knowledge, and hence poses epistemological questions in their clearest and most acute form. All this is commonplace enough, and it would hardly be necessary to Synthese 25 (1972) 248-253. All Rights Reserved Copyright © 1972 by D. ReidelPublishing Company, Dordrecht-Holland

REVIEWS

249

repeat it if it were not for the fact that a good many philosophers of science seem, these days, to have forgotten it. Harr6, in The Principles of Scientific Thinking, is a case in point: from the conclusions he draws, Hume might just as well never have lived, and nothing philosophical, except dreadful mistakes, have happened since, at any rate among those (apart from Kant) who were influenced by him. Not that Harr6 ignores Hume, indeed he is at pains to neutralize what he takes to have been Hume's arguments. But he has mistaken the purport of Hume's question, and for that matter the purport of a great deal of what has happened in the philosophy of science in this century. He is back to Natures and Powers again in a spectacular regression that succeeds in annulling 350 years of philosophical history. How has this feat been managed? And why should it have been thought desirable ? To answer the second question first: The trouble seems to be that the logical-reconstructionist version of the philosophy of science, as practiced by empiricists in the last generation or so, seems too remote from the actual business of the scientist to be plausible as an account of the nature and operation of science. Harr6 "tries to present a theory of science with some resemblance to scientific theory and practice": Now this is a wellworn objection that echoes, among others, Toulmin, Kuhn, and Bunge, and it seems to me to rest on a profound misunderstanding. Why should a theory of science 'resemble' scientific theory and practice? Theories of genetic inheritance do not resemble fathers and sons, nor do theories of disease resemble leprosy or chicken-pox. The relation between a theory and what it is a theory of is obviously a central problem for the philosophy of science, but the fact that something is a theory, rather than, say, a descriptive narrative, suggests that this relation should be one of explication, perhaps, rather than resemblance. The point is less to tell the scientist what he thinks he is doing or trying to do ("preserving his major intuitions", as Harr6 puts it) than to find a philosophically illuminating way of representing what he is able and not able to do. As far as the practice of his trade is concerned the scientist can do all sorts of exciting things, but we have to ask whether he is truly giving us knowledge of the ultimate constitution of the world, for example, as opposed to devices for anticipating its behaviour, and it is on such points that Humean scepticism comes to bear. We hope for knowledge of the ultimate constitution of the world, but we can't know whether or not we have got it.

250

REVIEWS

This is the sort of thing that makes Harr6 lose patience. Of course we know the ultimate constitution of the world, he says, not to be sure in as much detail as we would like, but enough to scotch all this positivistic pessimism. What made us think we couldn't know it can only have been a misunderstanding of the nature of scientific knowledge, engendered by a foolish craving for logical rigour. The ultimate constitution of the world is point-centers of influence, having powers. Along the way to this triumphant conclusion Harr6 achieves, or at any rate announces, a Copernican revolution in the philosophy of science. This consists in shifting the perspective of theoretical understanding from linguistic constructions (sets of deductively-related sentences, and so on) with heuristic models attached, to "pictures of the inner structure and constitution of things" with heuristic sentences attached. And in fact there is a lot of sense in what he says about models; the value of his book lies in the original analysis he suggests for various types of model, and this part of it constitutes a welcome addition to the literature of the subject. The trouble is that he claims so much more, and with such unnecessary polemics. The Copernican revolution in this case fails for lack of a Ptolemy. Harr6's candidate for the job is the set of gullible enthusiasts, worse than Simplicio even, who have swallowed the positivist dogma, celebrated its myths, received its logic (the phrase 'received logic' occurs several times). The only trouble is that these seem to correspond to no identifiable figures in the recent history of the philosophy of science. There are some names, to be sure - Carnap, Hempel, Goodman - but they are for the most part attached to misrepresentations of the views held by Carnap, Hempel, and Goodman, to name some actual philosophers. Hart6 accuses the tradition these philosophers represent (in their very different: ways) of having spawned "an enormous scholastic literature of an essentially trivial nature", and I must admit to some sympathy with him on this point, but that cannot count against the view he dislikes, since it is a universal phenomenon in intellectual life - such a literature always accompanies any movement of importance. Also "essentially trivial" is a dangerous phrase for Harr6 to be bandying about. The misrepresentation I refer to consists less in making mistakes about what the philosophers in question actually said, than in misunderstanding the problematic context in which they said it. For example he starts off with a little joke, a cut of

REVIEWS

251

"'Cygnus Melancoryphus, the Black and White Swan... An awful warning to those who suppose that 'All swans are white' and its confirmation or falsification by instances exhausts the logic of the laws of nature". Now nobody I have ever met or heard of has been naive enough to think that 'All swans are white' and its confirmation or falsification by instances exhausts the logic of the laws of nature, but some people for whom I have great respect have noticed that if profound philosophical difculfies are unchained by the analysis of even such elementary and straightforward general statements, more complicated ones are afortiori likely to resist easy treatment. It is simply no argument against this point to bluster about our really knowing so obviously how it all is. "In fact the future is not open, causality is generation, prediction can be rational, and our confidence in our knowledge of the future well-founded", says Harrr. "We know that there are generative powers". These and many other examples are abuses of the assertoric mode, offences against the restraint that the whole history of philosophy has painfully demonstrated to be the first demand of philosophical responsibility. Harr6 offers no new evidence or proof for these claims. It is of course quite correct to represent such attitudes as habitual, but it is not correct to represent them as well-founded. Consider for example Harrr's analysis of the notion of a physical law. Laws, he says, are not properly so called until a generative mechanism has been identified. (This seems arbitrary - were Kepler's Laws not laws until Newtonian theory came along, and what kind of generative mechanism did Newtonian theory really offer, apart from mathematical deduction?) At all events this requirement is said to eliminate the perplexing problem of how to tell laws from accidental regularities, since in the latter case no such mechanism can even be imagined. But suppose some law really is accidental - that by a staggering coincidence all the observed cases have come out right, and continue to do so, for the time being at any rate. Suppose further that a plausible generative mechanism is ready to hand, which happens in fact to have nothing to do with the phenomenon in question even though it seems quite obvious to us that it does. Hart6 has made it impossible even to entertain the possibility; he will ascribe a power to the supposed mechanism, thus conferring on it the status of causal agent, and he will say that we just know that this power and this agency exist. Now the intuition that leads him to do this is precisely the

252

REVIEWS

intuition that scientists work with and that he wishes to defend, and it is

practically justified in most cases. The question is whether it is philosophically justifiable. This is the question that Hume and his successors answered in the negative. In a way it is an obtuse question, but then philosophy is an obtuse discipline; it persists in seeing trouble where common sense sees none, hut reiterating the assertions of common sense does not eliminate the trouble. Now of course it is a possible strategy to maintain that the whole set of topics raised in some line of enquiry is somehow off the point, or that some analytic misunderstanding or carelessness vitiates the way they are dealt with. This might be convincingly shown if another set of topics could be put forward, that had not previously been thought of, and dealt with in a careful and rigorous way. Harrt's topics have the disadvantage that they have been thought of, and dismissed as unfruitful, by a great many people, and to this must be added the further disadvantage that his treatment of them is often surprisingly loose. His polemical stand is so aggressive that one might expect all arguments to be marshalled with the utmost care, yet both in criticism and in speculation he is often merely rhetorical or programmatic. Some examples: to continue for the moment with the topic of scientific law, he supposes the point of the argument about subjunctive conditionals to be the representation of the law itself as a great subjunctive conditional, i.e. ruling out cases in which it has already been shown to hold, whereas the problem is whether the law will sustain particular subjunctive conditionals. He puts down reduction sentences with the remark that A scientificlanguage does not grow by addition of predicates which are related in this exhaustive way to observables, in the first instance. For example, the predicate 'magnetic' is used to ascribe a definite and independent property or power to substances, and furthermore a property which, because of the accidental constitution of our sense organs, is never among the observableproperties of things; yet before Gilbert elevated it to metaphysical status magnetism was precisely an observable (although dispositional) property, related to other observables (lodestones, geographical orientation) in just the way required by reduction sentences (if freely suspended, in the vicinity of a lodestone, etc., then if a piece of a given substance orients itself in a given way, etc., then it is magnetic). Of course if by 'observable' is meant 'visible' this won't work, but then mass, for example, isn't an observable either. He is

REVIEWS

253

highly critical of the logical empiricists for their unexamined use of ostensive definitions, predicates of quality, and the like, but uses terms like 'unit', 'entity', 'structure', 'essence' almost without any definition at all. He invents some dubious operations in physics: "Planck's Law, which is Rayleigh-like at low temperatures and Stefan-like at high temperatures, can be shown to be something very like a logical conjunction of the two more restricted laws". And so on. There is a lot more to be said, but perhaps this is enough. My dissatisfaction with the book can be summed up under three heads: (1) it promises more than it delivers ("I shall give an account of the true facts of the matter", says Harrr). (2) It misrepresents the views it attacks. (3) It reintroduces an outmoded and essentially uninformative way of characterizing the composition of the physical world under the guise of something novel and enlightening. Hart6 claims to set out "a new view of theories in which they are to be seen as essentially concerned with the mechanisms of nature, and only derivatively with the patterns of phenomena". But that is the old view; it is what we would all have liked to stick to all along; only it has not proved possible to substantiate it. So theories remain hypothetical, and hypotheticals do not turn into categoricals just because one shouts them at the top of one's lungs. And as to the attribution of powers, I cannot for the life of me see what it adds to the attribution of theoretical predicates in the usual way. Lime juice, to use an example of Harrr's own, prevents scurvy; what more do we learn about it when told that it has the power to do so? Let me reiterate in closing that the book has great merits; not only the chapter on models but also a rather unexpected detour through the mind-body problem in the chapter on principles, while not as definitive as Hart6 seems to think, offer suggestions that are worth considered attention. Even the Copernican revolution, megalomania apart, provokes a certain reflection. All the more regrettable, then, that the book should also contain so much old metaphysical wine in such worn dialectical bottles.

Hunter College, City University of New York

PETER CAWS