Salmon on Explanation Bas C. Van Fraassen The Journal of Philosophy, Vol. 82, No. 11, Eighty-Second Annual Meeting Ameri...

Author:
van Fraasen Bas

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!

Salmon on Explanation Bas C. Van Fraassen The Journal of Philosophy, Vol. 82, No. 11, Eighty-Second Annual Meeting American Philosophical Association, Eastern Division. (Nov., 1985), pp. 639-651. Stable URL: http://links.jstor.org/sici?sici=0022-362X%28198511%2982%3A11%3C639%3ASOE%3E2.0.CO%3B2-D The Journal of Philosophy is currently published by Journal of Philosophy, Inc..

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/jphil.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact [email protected]

http://www.jstor.org Tue Jun 19 06:29:38 2007

SALMON O N EXPLANATION AND CAUSALITY

639

properties and explanatory dependence emerge from the project of unifying the regularities we discover in nature. On this approach, theoretical explanation is primary. Causal concepts are derivative from explanatory concepts. In explaining particular events we answer as many questions as we can, drawing on our view of the order of natural phenomena. In some cases, our ideal of understanding may not be completely realizable. The two approaches depart from Hempel in different directions. Nobody has yet developed a radical "top down" approach with the clarity and thoroughness that Salmon has brought to its "bottom up" rival. My aim has been to suggest that, for all its thoroughness, Salmon's program still faces some research problems and that the case against an alternative view is not entirely closed. PHILIP KITCHER

University of Minnesota

SALMON ON EXPLANATION"

A

CCORDING to Wesley Salmon's significant and seminal study, science provides us with understanding of a phenomenon exactly to the extent that it provides an explanation, and provides explanation to the extent that it fits the phenomenon in a discernible pattern of causal relations. I suppose I disagree on both points; but this statement is misleading, for in fact my agreement with Salmon is much more extensive than our disagreement. The causal order as described by Salmon I can understand without sacrifice of philosophical scruples (though I am less certain that the notion of causality is entirely captured), and the kind of explanation he characterizes is, I think, of central importance, though not the only kind. T o bring out our agreements and disagreements more concretely, I shall discuss two topics: the logic or structure of explanation, and the limits of causal modeling. *Contribution to a n APA symposium o n Wesley Salmon, Scientific E x p l a n a t i o n a n d t h e Causal Structure o f t h e W o r l d (Princeton, N.J.: University Press, 1984), to be held o n December 30, 1985. Philip Kitcher will be co-symposiast, and Wesley Salmon will comment; see this JOL'RNAL, this issue, 632-639 and 651-654, respectively, for their contributions. Parenthetical page references i n the text of this paper are to Salmon's book. T h e author wishes to acknowledge the support of the National Science Foundation. 0 1985 T h e Journal of Philosophy, Inc. 0022-362X/85/8211~0639$01.20

640

THE JOURNAL OF PHILOSOPHY I

Salmon and I agree that an explanation must provide a bit of relevant, missing information. He illustrates this with the question whether Laplace's demon would have an explanation for everything that happens or only a description of it. Salmon points out that the demon, who knows one particular state of the world in detail and can calculate in accordance with Newton's mechanics, may still lack "several items of knowledge, essential to explanation but not indispensable to prediction" (132). These include: which are the causal laws; which are real and which pseudo processes; which among all events are causal interactions. This shows graphically that, i n Salmon's ontic conception, the information involved in explanation is information about a particular subject: causal relations. I would disagree; I think almost any bit of information could play the crucial role of allowing a questioner to complete his understanding in the particular respect in which he wants it to be complete. T h a t is the pragmatic conception; the difference lies of course in the construal of 'relevant'. Salmon objects with the example of a congressman who is satisfied with the explanation of a plane crash (ice build-up on the wings). His objection is that the further information that the plane derives its lift, due to the Bernoulli principle, from wings that have the shape of air foils, still constitutes a better or fuller explanation, regardless of what the congressman asked for. Now it is true that if the congressman were to learn enough physics to understand this point he would then be in a position to answer more why questions about the incident than if he did not. But this is, I think, true of any information-that is, it is potentially just the information needed to furnish an answer to a why question (and to other sorts of question as well, of course). Our different reactions to the example clearly show that, on Salmon's conception, what the relevant information is depends on the event to be explained, and not on facts about the questioner. T h e relevance is that it is information about a causal mechanism. Now I wonder whether Salmon also takes this attitude about the point that the explanation provides missing information. Suppose the congressman, unlike Laplace's demon, had all the information in question-that he had learned (and not forgotten) about the Bernoulli effect. Would the information about the ice on the wings then constitute the explanation? I feel we should say yes, but this may be because I automatically equate explanation with answer to a why question, and so the questioner's background information plays an obvious role. If Salmon says no, it is a little difficult to see how he can stop short of identifying every explanation with a very large

SALMON ON EXPLANATION A N D CAUSALITY

641

chunk of the sciences, plus a good deal of information about initial conditions, typically to a large extent unknown to all participants in the discussion. The human phenomena we usually classify as explanations must then be reclassified as pointers to or sketches of explanations or, more precisely, of conjectures concerning the existence of explanations. One virtue I would claim for a pragmatic account is that it succeeds more readily and simply in "saving" these human phenomena. A more important point, to Salmon, concerns the relation between explanation and understanding. Discussions with Karel Lambert over the past few years have brought it home to me that I definitely cannot equate the two. (This is perhaps also a reason why I cannot accept what Philip Kitcher calls "global accounts" of explanation.) Certainly, when someone asks for an explanation, he is asking for a way to complete his understanding in certain respects. But he may have a mistaken presupposition: it may be that his information and even his understanding are already complete, or as complete as science can make them, at least in that particular respect. Salmon says that, with my second criterion for evaluating explanations-that the answer should describe factors that favor the topic with respect to other members of the contrast class-we come to a fundamental parting of the ways (108). I suspect that this parting has to do with that question about understanding and explanation. As an example Salmon considers a Mendelian experiment on the color of pea blossoms; about 3/4 of the offspring are red and 1/4 white. The explanation in terms of the genetic character of the population does not refine these probabilities. Salmon writes: O n van Fraassen's theory, then, the explanation of the color of the red blossom (which favors the topic of its question) is clearly superior to the explanation of the color of the white blossom (which does not favor the topic of its question). Nevertheless, as I remarked previously, I am firmly convinced that, in such cases, we understand the unfavored ones just as well (or as poorly) as we understand the favored ones (109).

I am equally firmly convinced of this, and agree completely: both before and after receiving information about the genetic character of the population, our understanding of the occurrence of red flowers may be neither better nor worse than our understanding of the occurrence of white ones. But I do not agree with the word 'nevertheless', which signals another conviction about the relationship between understanding and explanation. We can imagine the

642

THE JOURNAL OF PHILOSOPHY

following dialogue, between people who have studied Mendelian genetics, but who do not have equal knowledge of the origin of certain flowers on the table: (1) Why is this flower red? It grew from a seed obtained from the population with genetic constitution X (described in Salmon's example). (2) Yes, but this second flower grew from a seed taken from the same population as the first, and is white-so why was that first flower red? Sorry, there is no further explanation to be had.

For the first question to receive a good answer, the answer should contain information which the questioner did not already have, which should redistribute his prior probabilities for the flower colors in a certain way. Note that the same response could have been a good answer to another question: (3) Why is this flower white? Because it grew from a seed obtained from the population with genetic constitution X (described in Salmon's example).

for example, if relative to the questioner's background information the flower could have come equally well from any one of ten populations, nine of which yield hardly any white flowers at all. The place where we n o longer have information that is new for the questioner, and could favor the topic-given his background information-is (2), and there again the situation is symmetric with respect to red and white. Here I would say the explanation request is rejected, because there is n o further explanation to be hadthe questioner already has all the information he could have. If he does not realize that, his understanding might be increased when that is pointed out to him. (Perhaps not in the case of Mendelian genetics, where we suspect there are undiscovered, predictively valuable, hidden variables, but at least in the case where he is told that the probabilities are irreducible.) But I still would not say he had received an explanation-his question was not answered but rejected. So I think the real focus of disagreement may be the relation between explanation and understanding. I have not previously tried to say anything about the latter. All I have to say here is that our understanding can sometimes be increased otherwise than by receiving explanations and that we may sometimes be in a position (in view of our interests, background information, and beliefs) to receive an explanation of one case and not of another even though the total scientific information, by itself, harbors no such asymmetry.

SALMON ON EXPLANATION AND CAUSALITY

643

Salmon has very correctly emphasized the connections between my account of explanation and other accounts, such as the recurrence of the problem of the reference class under another name. He has also posed a problem (110) which I want to take seriously, but have at present no answer for. It is this: I take it that a why question is always (explicitly or implicitly) contrastive-we ask why this flower should be, e.g., white rather than red like the others, and there may or may not be an answer. If there is an answer, then, I think, it describes a factor that favors the topic-in this case, white color-as against the contrast cases-in this case, red color. But, as Salmon points out, there is a difference between being told why P and why not Q and being told why P and that not (2. This suggests a stronger criterion of evaluation than I have utilized, and that suggestion may be correct. One thing I may perhaps be allowed to emphasize again: I do not pretend that the criteria for evaluating answers to why questions are either singly or jointly complete, or even that they are all the criteria applicable in common to all cases (i.e., to all why questions regardless of relevance relation, the other crucial ingredient). I was less than intensely concerned with this because I felt that the main traditional puzzles-asymmetries and rejections-could be understood before the evaluation of answers. In the last few paragraphs I have been at pains to defend my own account against Salmon's criticisms. I hope I have not abused the symposium's hospitality in this way, and hasten back to other aspects of Salmon's book. I1

On Salmon's ontic conception of explanation, science provides us with an explanation of a phenomenon to the extent that it fits the phenomenon into a discernible pattern of causal relations (121). This causal pattern is constituted by a net of spatiotemporally continuous causal processes, which intersect in events (common causes) classifiable as either conjunctive or interactive forks (see page 228, for summary). As Salmon realized very well, the place where this conception is prima facie called into severe doubt is quantum mechanics. He has responded with a lucid and insightful discussion of this subject and its difficulties, including those posed by the Einstein-Podolsky-Rosen (EPR) thought experiment and Bell's inequalities (242-259). In his overview of the situation Salmon acknowledges that quantum mechanics does not always provide a causal account of the sort he had described as needed for explanation. O n the other hand he raises implicit objections against the possible assertion that quantum theory does provide an account of a sort that gives us scientific understanding of those phenomena.

644

THE JOURNAL OF PHILOSOPHY

He points to the continuing question whether the so-called "reduction of the wave packet" is (supposed to be) a real process or only a mathematical calculation stage. Any interpretation of the theory must make this clear, and if we don't know the answer then a fortiori we have not found any interpretation satisfying. I would not take quite the same view. Suppose that we knew of several interpretations, which disagreed with each other on this point, but were each in themselves entirely coherent and feasible. Then we would have n o answer to that question, but we would have a completely satisfactory understanding of the theory. (And indeed I think we may be in roughly this situation-was there ever an accepted basic theory of physics totally devoid of remaining philosophical puzzles?) Our different attitudes here, however, may relate to the realist/anti-realist opposition. T h e different interpretations do not-if they are interpretations rather than revisionsaffect the question of empirical adequacy. Therefore I would say that even unqualified acceptance of the theory does not involve choosing among them. A realist has, I think, higher criteria for acceptability. But he must face the fact that quantum theory is today all but universally accepted in the scientific community, with n o more qualification than such acceptance ever receives in practice, and without such a choice between interpretations, and without anxiety about when the interpretations being developed will become complete and trouble-free. T o me this looks normal, but to a realist it may be deplorable double-think. T o return to causal modeling: Salmon plays with two ideas about how the confrontation with quantum mechanics may eventually force a new characterization of satisfactory scientific explanation. T h e first comes after a presentation of Bernard d'Espagnat's ideas: "although the argument is not rigorous, the strong suggestion is that nature seems to exhibit action-at-a-distance in the quantum domain" (251). T h e second idea is a sort of physical holism: "The basic philosophical question seems to be this: Is local conservation of a physical quantity any less mysterious than remote conservation; is remote conservation any more miraculous than local conservation?" (257). In other literature I have seen this second idea usually paired either with the idea that not particles but fields are the real individuals (and that nonlocal conservation is not mysterious if it happens to a single individual existing throughout space-time) or with the idea that quantum logic describes the structure of events and possibilities in some revolutionary new way. (Good examples of these two are the papers by Allen

SALMON ON EXPLANATION AND CAUSALITY

645

Stairs and Allen Ginsberg in Synthese LXI'.) Without denying the potential value of either idea (for the construction of incompatible but severally coherent interpretations of the theory) I am still surprised at the omission of a third. What about the Copenhagen line that the probabilities entailed by quantum theory are all conditional on measurement, coupled with a rejection of the search for mechanisms that bring about the expected statistical correlations? Perhaps the omission is not surprising, because this line comes right out and rejects the demand for explanation in terms of mechanisms (at least, at all levels). Yet the two ideas Salmon does play with are, I think, equally far from this original wish. Indeed, when he spoke of remote conservation I began to think he was veering toward Copenhagen, until he asked "Or is this one of the fundamental mechanisms by which nature operates?" (258). With this, I think, Salmon implicitly denied that a good theory could entail the statistical correlations in question without entailing the existence of a mechanism that "brings them about." It would just have to be a mechanism quite different from the sort envisaged in terms of common causes-but this negative characterization is all we have. Action at a distance is a concept with a persuasive mechanical name-action explains the remote correlation. But such action! Not action conveyed by the outward travel of matter or energy. Or is it instantaneous travel? But that makes no sense. If I traveled instantaneously, I would be at start, and finish, and all points in between at the same time. How does that differ from being a n extended entity that occupies the whole track? Only, I think, by a verbal distinction. In any case, instantaneous travel in one frame of reference is noninstantaneous in another. Travel faster than light, then? An envisaged revision of relativity theory? T h e sole motive, to save some traditional philosophical principle of interpretation, is not the sort that moves science. I have two reasons for preferring something closer to the Copenhagen view. T h e first is that the relevant phenomena do not exhibit, and the quantum theory does not imply, empirically verifiable action or signals faster than light. In the strikingly macroscopic effects, meters apart, of the Aspect experiments, a person might be tempted to say that it "looks as if there were" faster-than-light communication. But closer attention reveals that there is nothing a person could do at one arm which would affect the concurrent freI "Sailing into the Charybdis: Van Fraassen o n bell'^ Theorem," Synthese, LXI, 3 (December 1984): 351-359; and "On a Paradox in Q u a n t u m Mechanics," ibid., 325-349, respectively.

646

THE JOURNAL OF PHILOSOPHY

quency count at the other arm. It is true that Erwin Schrodinger, in his 1935-1936 papers reflecting on the Einstein-Podolsky-Rosen paradox, and assuming the reality of the reduction of the wave packet, deduced the existence of "spooky action at a distance." (This is what Albert Einstein called it in a 1948 letter to Max Born.) For he said of the two systems X and Y, separated after previous interaction, that the experimenter could steer Y into almost any state, with positive probability, by carrying out suitable experiments on X. This deduction was correct in itself, but faulty in the causal terminology of the conclusion. For he neglected to ask: What is the probability that a measurement testing for that state on Y, will have a positive outcome, regardless of whether any measurements are carried out on X? The answer is just the same as the probability conditional on those measurements being performed on X. Thus those measurement operations on X are only like pricking a voodoo doll: it is true that the derived measurement outcome for Y has a positive probability, and also that the interpretation makes the operation responsible for that, but the probability is no higher than without the operation. The confusion that sustained the conclusion was with another conditional probability: of the desired measurement outcome on Y, given a certain outcome of those measurements on X. But the experimenter can decide only on what measurements to carry out on X, and not on what their outcomes will be. So this other conditional probability is not relevant to the question whether the experimenter on X has influence on Y. There is, in the EPR-type experiments, no sign of empirically verifiable signaling faster than light. There may be nevertheless, behind the phenomena, so to say, on certain interpretations (interpretations that attribute physical reality to the collapse of the wave packet). But the phenomena give us no compelling basis for such a conclusion. It cannot be emphasized enough that the experimental violations of Bell's inequalities are on the macroscopic level. They are there, equally verifiable in frequency counts, even for people who reject quantum theory as utter nonsense. And the deduction of the inequalities from the assumption that phenomena admit of a common-cause model makes no appeal at all to quantum t h e ~ r y . ~ The relation between causality and Bell's inequalities is an issue apart from quantum theory. It had seemed to me for a little while that, once Salmon introduced the interactive fork in his presidential address "Why Ask 2 ~ e my e "Charybdis of Realism," Synthese,

LII,

1 (July 1982): 25-28.

SALMON O N EXPLANATION A N D CAUSALITY

647

' ~ h y ? ' ? "things ,~ looked a little better for causal modelings. The events at the source in those particle-pair experiments constitute an interactive fork, as does the initial interaction in the EPR thought experiment. Unfortunately quantum theory does not entail that EPR-type correlations exist only in composite systems whose parts have been in previous interaction. Indeed, in the treatment of assemblies of "identical" particles, the permutation-invariance requirement on the state implies that any such assembly is always in a state that entails such nonclassical correlations. Now it is true that even physicists sometimes query the correctness of this. What about a system consisting of two electrons, one in our galaxy and one in another? The suggestion is, I think, that our evidential support for the theory won't be any worse if we drop the permutationinvariance requirement for such a case. Fine; but the point is empty. For if we put this information about locations into the state, the probabilities for measurement outcomes become negligibly different whether the invariance requirement is imposed on the state or not. Anyway, all we need for the present point (about the impossibility of explaining correlations through previous interaction) are mundane cases of particles, newly created or having come together from separate sources, but without previous interaction. I think Salmon must have realized this from the beginning, for he did not attempt any such defense; but it took me a little while to appreciate it. I said I had two reasons for preferring something closer to the Copenhagen view. The second concerns the idea of remote (nonlocal) conservation laws. As I said, the sort of holism considered here has appeared especially in connection with two other ideas: that primacy should go to (a) quantum field theory, or (b) quantum logic, interpreted ontologically. But of course it had also appeared much earlier, in another way, in Niels Bohr's reaction to the EPR article-coupled with Bohr's notorious agnosticism about the transempirical or transphenomenal. Now I have absolutely no quarrel with primacy for fields-Ginsberg and W. M. DeMuynck among others have done a great deal, in my opinion, to make this attractive to philosophers. That is, it gives us an attractive way to get used to giving u p the demand for causal explanation of nonlocal correlations, in the traditional form. We should not pretend that it gives us a way of hanging on to what we had found it so painful to Pacific Division, March 24, 1978; Proceedings and Addresses of the American Philosophical Association, Lr, 6 (August 1978): 683-705.

648

THE JOURNAL OF PHILOSOPHY

lose; and I don't think anyone pretends that. It shares with Bohr's acceptance of a certain kind of holism, that it seeks peace through surrender-and I wholeheartedly agree that, for the philosophical fly, surrender of old principles may be the only way out of the fly bottle. The quantum-logical approach is a different matter, holding out much greater promise, at a much greater price. The argument that leads from necessary conditions for a causal model to Bell's inequalities proceeds via Boolean logic and its concomitant Kolmogorovian probability theory. What all may we not be able to retain, in the way of traditional philosophical premises, if we opt for a new logic and its concomitant new probability calculus! Those new and unsuspected phenomena that have come to light, do they not show us a new structure in the space of possibilities? And now nonclassical logic has become macroscopically manifest. I want to explain exactly how it may look as if the phenomena lead us forcibly away from Boole and Kolmogorov, and why in fact they do not. Of course, in this I am not arguing against Salmon, but against a common rival point of view. This rival envisages a much more radical rejection of the reasoning espoused in Salmon's work than I am urging, and I am anxious that it should not find a fulcrum in those difficulties with quantum theory. Consider conditional probabilities for two-valued variables A , B, C, . . . of the form P (a measurement of A yields value 1, given that a measurement of B has yielded value 1) = PAB.Given a finite set C of such equations, PAB,PAC,PBC,PCB,. . . , can we find a single probability function from which all of these derive? That question needs to be made precise. Let us define a simple Kolmogorov model for 2 to be a finite Boolean algebra of sets, some of its members also denoted as A , B, C, . . . , and an ordinary probability function P defined on this algebra, such that for each equation Pxr = r in C, we have P(X n Y ) = rP(Y). The question we asked may be given the precise form: does C admit of a simple Kolmogorov model? And the answer is yes, if and only if C implies no violation of Bell's inequalities. It is not difficult to see how to transpose the deduction that leads from the assumption of a common cause (conjunctive fork) to Bell's inequalities, to the present case. For note that, if x is any atom of the Boolean algebra in question, then P(X n Y 1 x ) = P(X I x) P(Y I x), no matter how strong the correlation between X and Y. The conclusion that phenomena involving violation of Bell's inequalities admit no common-cause model may have been an interesting if painful surprise. But the idea that they admit no descrip-

SALMON ON EXPLANATION AND CAUSALITY

649

tion satisfying Boolean logic and Kolmogorov's probability theory is a surprise of another order of magnitude. Remember after all that we can witness these phenomena on the macroscopic level, the relevant parts being frequency counts. Please reserve judgment on this for a moment while I continue the story begun in the preceding paragraph.4 Define a Hilbert-space model for that finite set C of equations to be a Hilbert space, with some vectors (PA, (PB,(PC,. . . in the space such that, for any equation PXY= r in C, r = I ( (PX,(PY ) 1 2 . We can now make the original question precise in a different way: does 2 admit a Hilbert-space model? The answer is yes, if and only if C satisfies certain numerical conditions, not so restrictive as Bell's inequalities (more restrictive if the Hilbert space is to be on the real numbers, and less restrictive if on the complex numbers). These conditions are certainly not tautological, and could conceivably be violated in nature too, though no such violations have been encountered in the experiments. This positive result is perhaps even more disquieting than the negative results discussed above, because the Hilbert space models could be called non-Boolean, non-Kolmogorovian probability models. They give real substance to the ideas that classical logic and probability theory have had their day, for they provide a precise formulation of an alternative. So far I have not stinted on the side of my opponents. But now I should like to point out that the deduction of Bell's inequalities from the assumption of a common-cause model had a much weaker premise than that of the existence of a simple Kolmogorov model. The word 'simple' is not redundant. Here is an example of a set C that does not admit a simple Kolmogorov model: PAB,PAC, and PBCall equal to zero. That means that if you measure one of A and B say, and find value 1 for it, you definitely will find another value for the other. But I also had said that A, B, C are two-valued. How could three variables each be two-valued and yet never have the same value? That is impossible. Yes, that is impossible. But the original description of C gave its equations as: P (a measurement of A yields 1 given that a measurement of B yielded 1) = PA& The situation described in the preceding paragraph is therefore not impossible, if the variables are not all three jointly measurable. A less simple Kolmogorov model now associates two sets with each variable: say A + and MA-read 4 ~ elegant n exposition of the above was given by Arthur Fine, Physical Reuiew Letters, xLvrr, (1982): 291-295; for the continuation below, see L. Accardi and A. Fedullo, Lettere a1 Nuovo Cimento, x x x ~ v(,1982): 161-172.

650

THE

JOURNAI. OF

PHILOSOPHY

"MA" as " A is measured." And then it could model 2 as follows: P x y = P ( X +n Y' I M X n M Y ) . We add that although P ( M A n M B ) , P ( M A n M C ) , a n d P ( M B n M C ) a r e each positive, P ( M A n M B n M C ) equals zero. It is clear, however, that we will be in difficulty again if we now add: but A, B , C each must have one of its two possible values at all times, whether measured or not. The only remedy here is to say, in Copenhagen style: Who says that physical magnitudes are like that? Why should they always have definite values? And what empirical results could make us think they do? So suppose that certain phenomena, described by such a set 2, admit no simple Kolmogorov model but d o admit a Hilbert-space model. They still admit a less than simple Kolmogorov model; so there is no reason to think that classical logic and probability theory have come a cropper. That model includes probabilities for the acts of measurement-information not expected, or required, from science. So we shall be happy if science gives us the set C of important conditional probabilities, and even happier if it gives us that set in the calculationally manageable form of a Hilbert-space model. In fact, all we need, to feel happy and at peace here is the Copenhagen agnosticism with respect to what happens to measurable physical magnitudes when they are not being measured. Perhaps they have n o values; or have values different from what measurements would have shown, had those been carried out; or whateverlet us hr interested in, but not unduly anxious about, the logically possible interpretations along all such lines. I11

Let me sum u p the second part of my paper as follows. Salmon explicitly rejected apriorism about what scientific explanation must be or do ( 2 4 2 ) . T h e very demand of explanation through causal mechanisms which he takes as paradigm was strongly reinforced by the victories of relativity theory, and could not have been met by Newton's celestial mechanics, as he himself points out. In turn, it is challenged by the successes of quantum theory, and this challenge to his conception was presented nowhere as incisively and precisely as by Salmon himself. T h e question is now, How shall we react to this challenge? When it comes to the form of explanation in general, I think I am much more liberal than Salmon; that is inherent in the difference between the pragmatic and the ontic conceptions. But when it comes to the evaluation of proffered explanations, as good, better, or best, I am no more inclined than he is to think there are no standards. We agree in our desire for a concrete and informative, yet comprehensive, account of what physical theory is and does. But I respectfully submit the following points

SALMON O N EXPLANATION AND CAUSALITY

65 1

concerning the challenge of quantum theory: (1) the difficulties for the ideal of the causal order arise at the level of the observable phenomena, and are independent of the intelligibility of the theory; (2) neither the phenomena nor the theory imply the possibility of empirically verifiable signals or action faster than light; (3) the nonclassical correlations implied by quantum states may be described in terms of remote (nonlocal) conservation laws, but there is no sign of remote conservation mechanisms; (4) the temptation to see the phenomena as violating the classical logic and probability framework delineated by Boole and Kolmogorov is understandable enough, but also finds no warrant in the phenomena. This list of four points is clearly not a list of disagreements with Salmon, for some of this he points out himself. But I wonder if, taken all together, they do not provide some good reasons to reduce the distance between Pittsburgh and Copenhagen. BAS C. VAN FRAASSEN

Princeton University CONFLICTING CONCEPTIONS OF

SCIENTIFIC EXPLANATION*

I

N his brief but penetrating paper, Philip Kitcher has raised a number of important issues. The most fundamental, I think, involves the conflict between two venerable intuitions: explanation as derivation and explanation as identification of mechanisms. A primary virtue of the former, he maintains, is the unification it achieves by appeal to general principles that enables us to use "the same pattern of derivation in case after case." I agree that unification has enormous explanatory power but would argue that it plays an equally central role for the mechanist. Just as the derivationist sees the same pattern of derivation in case after case, the mechanist sees the same kinds of basic mechanisms working in widely differing circumstances. Just as the derivationist appeals to general laws to carry out derivations, the mechanist recognizes that the basic mechanisms conform to general laws. The two viewpoints are equally committed to the covering-law principle and to the notion that our understanding of the world is improved through unification of diverse phenomena under general principles. Are there, then, any really important differences between the *Abstract of a paper to be delivered in an APA symposium o n Wesley Salmon's Scientific Explanation and the Causal Structure of the World, December 30, 1985, responding to papers by Philip Kitcher and Bas van Fraassen; see this JOURNAL, this issue, 632-639 and 639-651, respectively.

0022-362X/85/8211/065lf00.50

@

1985 T h e Journal of Philosophy, Inc.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/jphil.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

JSTOR is an independent not-for-profit organization dedicated to and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact [email protected]

http://www.jstor.org Tue Jun 19 06:29:38 2007

SALMON O N EXPLANATION AND CAUSALITY

639

properties and explanatory dependence emerge from the project of unifying the regularities we discover in nature. On this approach, theoretical explanation is primary. Causal concepts are derivative from explanatory concepts. In explaining particular events we answer as many questions as we can, drawing on our view of the order of natural phenomena. In some cases, our ideal of understanding may not be completely realizable. The two approaches depart from Hempel in different directions. Nobody has yet developed a radical "top down" approach with the clarity and thoroughness that Salmon has brought to its "bottom up" rival. My aim has been to suggest that, for all its thoroughness, Salmon's program still faces some research problems and that the case against an alternative view is not entirely closed. PHILIP KITCHER

University of Minnesota

SALMON ON EXPLANATION"

A

CCORDING to Wesley Salmon's significant and seminal study, science provides us with understanding of a phenomenon exactly to the extent that it provides an explanation, and provides explanation to the extent that it fits the phenomenon in a discernible pattern of causal relations. I suppose I disagree on both points; but this statement is misleading, for in fact my agreement with Salmon is much more extensive than our disagreement. The causal order as described by Salmon I can understand without sacrifice of philosophical scruples (though I am less certain that the notion of causality is entirely captured), and the kind of explanation he characterizes is, I think, of central importance, though not the only kind. T o bring out our agreements and disagreements more concretely, I shall discuss two topics: the logic or structure of explanation, and the limits of causal modeling. *Contribution to a n APA symposium o n Wesley Salmon, Scientific E x p l a n a t i o n a n d t h e Causal Structure o f t h e W o r l d (Princeton, N.J.: University Press, 1984), to be held o n December 30, 1985. Philip Kitcher will be co-symposiast, and Wesley Salmon will comment; see this JOL'RNAL, this issue, 632-639 and 651-654, respectively, for their contributions. Parenthetical page references i n the text of this paper are to Salmon's book. T h e author wishes to acknowledge the support of the National Science Foundation. 0 1985 T h e Journal of Philosophy, Inc. 0022-362X/85/8211~0639$01.20

640

THE JOURNAL OF PHILOSOPHY I

Salmon and I agree that an explanation must provide a bit of relevant, missing information. He illustrates this with the question whether Laplace's demon would have an explanation for everything that happens or only a description of it. Salmon points out that the demon, who knows one particular state of the world in detail and can calculate in accordance with Newton's mechanics, may still lack "several items of knowledge, essential to explanation but not indispensable to prediction" (132). These include: which are the causal laws; which are real and which pseudo processes; which among all events are causal interactions. This shows graphically that, i n Salmon's ontic conception, the information involved in explanation is information about a particular subject: causal relations. I would disagree; I think almost any bit of information could play the crucial role of allowing a questioner to complete his understanding in the particular respect in which he wants it to be complete. T h a t is the pragmatic conception; the difference lies of course in the construal of 'relevant'. Salmon objects with the example of a congressman who is satisfied with the explanation of a plane crash (ice build-up on the wings). His objection is that the further information that the plane derives its lift, due to the Bernoulli principle, from wings that have the shape of air foils, still constitutes a better or fuller explanation, regardless of what the congressman asked for. Now it is true that if the congressman were to learn enough physics to understand this point he would then be in a position to answer more why questions about the incident than if he did not. But this is, I think, true of any information-that is, it is potentially just the information needed to furnish an answer to a why question (and to other sorts of question as well, of course). Our different reactions to the example clearly show that, on Salmon's conception, what the relevant information is depends on the event to be explained, and not on facts about the questioner. T h e relevance is that it is information about a causal mechanism. Now I wonder whether Salmon also takes this attitude about the point that the explanation provides missing information. Suppose the congressman, unlike Laplace's demon, had all the information in question-that he had learned (and not forgotten) about the Bernoulli effect. Would the information about the ice on the wings then constitute the explanation? I feel we should say yes, but this may be because I automatically equate explanation with answer to a why question, and so the questioner's background information plays an obvious role. If Salmon says no, it is a little difficult to see how he can stop short of identifying every explanation with a very large

SALMON ON EXPLANATION A N D CAUSALITY

641

chunk of the sciences, plus a good deal of information about initial conditions, typically to a large extent unknown to all participants in the discussion. The human phenomena we usually classify as explanations must then be reclassified as pointers to or sketches of explanations or, more precisely, of conjectures concerning the existence of explanations. One virtue I would claim for a pragmatic account is that it succeeds more readily and simply in "saving" these human phenomena. A more important point, to Salmon, concerns the relation between explanation and understanding. Discussions with Karel Lambert over the past few years have brought it home to me that I definitely cannot equate the two. (This is perhaps also a reason why I cannot accept what Philip Kitcher calls "global accounts" of explanation.) Certainly, when someone asks for an explanation, he is asking for a way to complete his understanding in certain respects. But he may have a mistaken presupposition: it may be that his information and even his understanding are already complete, or as complete as science can make them, at least in that particular respect. Salmon says that, with my second criterion for evaluating explanations-that the answer should describe factors that favor the topic with respect to other members of the contrast class-we come to a fundamental parting of the ways (108). I suspect that this parting has to do with that question about understanding and explanation. As an example Salmon considers a Mendelian experiment on the color of pea blossoms; about 3/4 of the offspring are red and 1/4 white. The explanation in terms of the genetic character of the population does not refine these probabilities. Salmon writes: O n van Fraassen's theory, then, the explanation of the color of the red blossom (which favors the topic of its question) is clearly superior to the explanation of the color of the white blossom (which does not favor the topic of its question). Nevertheless, as I remarked previously, I am firmly convinced that, in such cases, we understand the unfavored ones just as well (or as poorly) as we understand the favored ones (109).

I am equally firmly convinced of this, and agree completely: both before and after receiving information about the genetic character of the population, our understanding of the occurrence of red flowers may be neither better nor worse than our understanding of the occurrence of white ones. But I do not agree with the word 'nevertheless', which signals another conviction about the relationship between understanding and explanation. We can imagine the

642

THE JOURNAL OF PHILOSOPHY

following dialogue, between people who have studied Mendelian genetics, but who do not have equal knowledge of the origin of certain flowers on the table: (1) Why is this flower red? It grew from a seed obtained from the population with genetic constitution X (described in Salmon's example). (2) Yes, but this second flower grew from a seed taken from the same population as the first, and is white-so why was that first flower red? Sorry, there is no further explanation to be had.

For the first question to receive a good answer, the answer should contain information which the questioner did not already have, which should redistribute his prior probabilities for the flower colors in a certain way. Note that the same response could have been a good answer to another question: (3) Why is this flower white? Because it grew from a seed obtained from the population with genetic constitution X (described in Salmon's example).

for example, if relative to the questioner's background information the flower could have come equally well from any one of ten populations, nine of which yield hardly any white flowers at all. The place where we n o longer have information that is new for the questioner, and could favor the topic-given his background information-is (2), and there again the situation is symmetric with respect to red and white. Here I would say the explanation request is rejected, because there is n o further explanation to be hadthe questioner already has all the information he could have. If he does not realize that, his understanding might be increased when that is pointed out to him. (Perhaps not in the case of Mendelian genetics, where we suspect there are undiscovered, predictively valuable, hidden variables, but at least in the case where he is told that the probabilities are irreducible.) But I still would not say he had received an explanation-his question was not answered but rejected. So I think the real focus of disagreement may be the relation between explanation and understanding. I have not previously tried to say anything about the latter. All I have to say here is that our understanding can sometimes be increased otherwise than by receiving explanations and that we may sometimes be in a position (in view of our interests, background information, and beliefs) to receive an explanation of one case and not of another even though the total scientific information, by itself, harbors no such asymmetry.

SALMON ON EXPLANATION AND CAUSALITY

643

Salmon has very correctly emphasized the connections between my account of explanation and other accounts, such as the recurrence of the problem of the reference class under another name. He has also posed a problem (110) which I want to take seriously, but have at present no answer for. It is this: I take it that a why question is always (explicitly or implicitly) contrastive-we ask why this flower should be, e.g., white rather than red like the others, and there may or may not be an answer. If there is an answer, then, I think, it describes a factor that favors the topic-in this case, white color-as against the contrast cases-in this case, red color. But, as Salmon points out, there is a difference between being told why P and why not Q and being told why P and that not (2. This suggests a stronger criterion of evaluation than I have utilized, and that suggestion may be correct. One thing I may perhaps be allowed to emphasize again: I do not pretend that the criteria for evaluating answers to why questions are either singly or jointly complete, or even that they are all the criteria applicable in common to all cases (i.e., to all why questions regardless of relevance relation, the other crucial ingredient). I was less than intensely concerned with this because I felt that the main traditional puzzles-asymmetries and rejections-could be understood before the evaluation of answers. In the last few paragraphs I have been at pains to defend my own account against Salmon's criticisms. I hope I have not abused the symposium's hospitality in this way, and hasten back to other aspects of Salmon's book. I1

On Salmon's ontic conception of explanation, science provides us with an explanation of a phenomenon to the extent that it fits the phenomenon into a discernible pattern of causal relations (121). This causal pattern is constituted by a net of spatiotemporally continuous causal processes, which intersect in events (common causes) classifiable as either conjunctive or interactive forks (see page 228, for summary). As Salmon realized very well, the place where this conception is prima facie called into severe doubt is quantum mechanics. He has responded with a lucid and insightful discussion of this subject and its difficulties, including those posed by the Einstein-Podolsky-Rosen (EPR) thought experiment and Bell's inequalities (242-259). In his overview of the situation Salmon acknowledges that quantum mechanics does not always provide a causal account of the sort he had described as needed for explanation. O n the other hand he raises implicit objections against the possible assertion that quantum theory does provide an account of a sort that gives us scientific understanding of those phenomena.

644

THE JOURNAL OF PHILOSOPHY

He points to the continuing question whether the so-called "reduction of the wave packet" is (supposed to be) a real process or only a mathematical calculation stage. Any interpretation of the theory must make this clear, and if we don't know the answer then a fortiori we have not found any interpretation satisfying. I would not take quite the same view. Suppose that we knew of several interpretations, which disagreed with each other on this point, but were each in themselves entirely coherent and feasible. Then we would have n o answer to that question, but we would have a completely satisfactory understanding of the theory. (And indeed I think we may be in roughly this situation-was there ever an accepted basic theory of physics totally devoid of remaining philosophical puzzles?) Our different attitudes here, however, may relate to the realist/anti-realist opposition. T h e different interpretations do not-if they are interpretations rather than revisionsaffect the question of empirical adequacy. Therefore I would say that even unqualified acceptance of the theory does not involve choosing among them. A realist has, I think, higher criteria for acceptability. But he must face the fact that quantum theory is today all but universally accepted in the scientific community, with n o more qualification than such acceptance ever receives in practice, and without such a choice between interpretations, and without anxiety about when the interpretations being developed will become complete and trouble-free. T o me this looks normal, but to a realist it may be deplorable double-think. T o return to causal modeling: Salmon plays with two ideas about how the confrontation with quantum mechanics may eventually force a new characterization of satisfactory scientific explanation. T h e first comes after a presentation of Bernard d'Espagnat's ideas: "although the argument is not rigorous, the strong suggestion is that nature seems to exhibit action-at-a-distance in the quantum domain" (251). T h e second idea is a sort of physical holism: "The basic philosophical question seems to be this: Is local conservation of a physical quantity any less mysterious than remote conservation; is remote conservation any more miraculous than local conservation?" (257). In other literature I have seen this second idea usually paired either with the idea that not particles but fields are the real individuals (and that nonlocal conservation is not mysterious if it happens to a single individual existing throughout space-time) or with the idea that quantum logic describes the structure of events and possibilities in some revolutionary new way. (Good examples of these two are the papers by Allen

SALMON ON EXPLANATION AND CAUSALITY

645

Stairs and Allen Ginsberg in Synthese LXI'.) Without denying the potential value of either idea (for the construction of incompatible but severally coherent interpretations of the theory) I am still surprised at the omission of a third. What about the Copenhagen line that the probabilities entailed by quantum theory are all conditional on measurement, coupled with a rejection of the search for mechanisms that bring about the expected statistical correlations? Perhaps the omission is not surprising, because this line comes right out and rejects the demand for explanation in terms of mechanisms (at least, at all levels). Yet the two ideas Salmon does play with are, I think, equally far from this original wish. Indeed, when he spoke of remote conservation I began to think he was veering toward Copenhagen, until he asked "Or is this one of the fundamental mechanisms by which nature operates?" (258). With this, I think, Salmon implicitly denied that a good theory could entail the statistical correlations in question without entailing the existence of a mechanism that "brings them about." It would just have to be a mechanism quite different from the sort envisaged in terms of common causes-but this negative characterization is all we have. Action at a distance is a concept with a persuasive mechanical name-action explains the remote correlation. But such action! Not action conveyed by the outward travel of matter or energy. Or is it instantaneous travel? But that makes no sense. If I traveled instantaneously, I would be at start, and finish, and all points in between at the same time. How does that differ from being a n extended entity that occupies the whole track? Only, I think, by a verbal distinction. In any case, instantaneous travel in one frame of reference is noninstantaneous in another. Travel faster than light, then? An envisaged revision of relativity theory? T h e sole motive, to save some traditional philosophical principle of interpretation, is not the sort that moves science. I have two reasons for preferring something closer to the Copenhagen view. T h e first is that the relevant phenomena do not exhibit, and the quantum theory does not imply, empirically verifiable action or signals faster than light. In the strikingly macroscopic effects, meters apart, of the Aspect experiments, a person might be tempted to say that it "looks as if there were" faster-than-light communication. But closer attention reveals that there is nothing a person could do at one arm which would affect the concurrent freI "Sailing into the Charybdis: Van Fraassen o n bell'^ Theorem," Synthese, LXI, 3 (December 1984): 351-359; and "On a Paradox in Q u a n t u m Mechanics," ibid., 325-349, respectively.

646

THE JOURNAL OF PHILOSOPHY

quency count at the other arm. It is true that Erwin Schrodinger, in his 1935-1936 papers reflecting on the Einstein-Podolsky-Rosen paradox, and assuming the reality of the reduction of the wave packet, deduced the existence of "spooky action at a distance." (This is what Albert Einstein called it in a 1948 letter to Max Born.) For he said of the two systems X and Y, separated after previous interaction, that the experimenter could steer Y into almost any state, with positive probability, by carrying out suitable experiments on X. This deduction was correct in itself, but faulty in the causal terminology of the conclusion. For he neglected to ask: What is the probability that a measurement testing for that state on Y, will have a positive outcome, regardless of whether any measurements are carried out on X? The answer is just the same as the probability conditional on those measurements being performed on X. Thus those measurement operations on X are only like pricking a voodoo doll: it is true that the derived measurement outcome for Y has a positive probability, and also that the interpretation makes the operation responsible for that, but the probability is no higher than without the operation. The confusion that sustained the conclusion was with another conditional probability: of the desired measurement outcome on Y, given a certain outcome of those measurements on X. But the experimenter can decide only on what measurements to carry out on X, and not on what their outcomes will be. So this other conditional probability is not relevant to the question whether the experimenter on X has influence on Y. There is, in the EPR-type experiments, no sign of empirically verifiable signaling faster than light. There may be nevertheless, behind the phenomena, so to say, on certain interpretations (interpretations that attribute physical reality to the collapse of the wave packet). But the phenomena give us no compelling basis for such a conclusion. It cannot be emphasized enough that the experimental violations of Bell's inequalities are on the macroscopic level. They are there, equally verifiable in frequency counts, even for people who reject quantum theory as utter nonsense. And the deduction of the inequalities from the assumption that phenomena admit of a common-cause model makes no appeal at all to quantum t h e ~ r y . ~ The relation between causality and Bell's inequalities is an issue apart from quantum theory. It had seemed to me for a little while that, once Salmon introduced the interactive fork in his presidential address "Why Ask 2 ~ e my e "Charybdis of Realism," Synthese,

LII,

1 (July 1982): 25-28.

SALMON O N EXPLANATION A N D CAUSALITY

647

' ~ h y ? ' ? "things ,~ looked a little better for causal modelings. The events at the source in those particle-pair experiments constitute an interactive fork, as does the initial interaction in the EPR thought experiment. Unfortunately quantum theory does not entail that EPR-type correlations exist only in composite systems whose parts have been in previous interaction. Indeed, in the treatment of assemblies of "identical" particles, the permutation-invariance requirement on the state implies that any such assembly is always in a state that entails such nonclassical correlations. Now it is true that even physicists sometimes query the correctness of this. What about a system consisting of two electrons, one in our galaxy and one in another? The suggestion is, I think, that our evidential support for the theory won't be any worse if we drop the permutationinvariance requirement for such a case. Fine; but the point is empty. For if we put this information about locations into the state, the probabilities for measurement outcomes become negligibly different whether the invariance requirement is imposed on the state or not. Anyway, all we need for the present point (about the impossibility of explaining correlations through previous interaction) are mundane cases of particles, newly created or having come together from separate sources, but without previous interaction. I think Salmon must have realized this from the beginning, for he did not attempt any such defense; but it took me a little while to appreciate it. I said I had two reasons for preferring something closer to the Copenhagen view. The second concerns the idea of remote (nonlocal) conservation laws. As I said, the sort of holism considered here has appeared especially in connection with two other ideas: that primacy should go to (a) quantum field theory, or (b) quantum logic, interpreted ontologically. But of course it had also appeared much earlier, in another way, in Niels Bohr's reaction to the EPR article-coupled with Bohr's notorious agnosticism about the transempirical or transphenomenal. Now I have absolutely no quarrel with primacy for fields-Ginsberg and W. M. DeMuynck among others have done a great deal, in my opinion, to make this attractive to philosophers. That is, it gives us an attractive way to get used to giving u p the demand for causal explanation of nonlocal correlations, in the traditional form. We should not pretend that it gives us a way of hanging on to what we had found it so painful to Pacific Division, March 24, 1978; Proceedings and Addresses of the American Philosophical Association, Lr, 6 (August 1978): 683-705.

648

THE JOURNAL OF PHILOSOPHY

lose; and I don't think anyone pretends that. It shares with Bohr's acceptance of a certain kind of holism, that it seeks peace through surrender-and I wholeheartedly agree that, for the philosophical fly, surrender of old principles may be the only way out of the fly bottle. The quantum-logical approach is a different matter, holding out much greater promise, at a much greater price. The argument that leads from necessary conditions for a causal model to Bell's inequalities proceeds via Boolean logic and its concomitant Kolmogorovian probability theory. What all may we not be able to retain, in the way of traditional philosophical premises, if we opt for a new logic and its concomitant new probability calculus! Those new and unsuspected phenomena that have come to light, do they not show us a new structure in the space of possibilities? And now nonclassical logic has become macroscopically manifest. I want to explain exactly how it may look as if the phenomena lead us forcibly away from Boole and Kolmogorov, and why in fact they do not. Of course, in this I am not arguing against Salmon, but against a common rival point of view. This rival envisages a much more radical rejection of the reasoning espoused in Salmon's work than I am urging, and I am anxious that it should not find a fulcrum in those difficulties with quantum theory. Consider conditional probabilities for two-valued variables A , B, C, . . . of the form P (a measurement of A yields value 1, given that a measurement of B has yielded value 1) = PAB.Given a finite set C of such equations, PAB,PAC,PBC,PCB,. . . , can we find a single probability function from which all of these derive? That question needs to be made precise. Let us define a simple Kolmogorov model for 2 to be a finite Boolean algebra of sets, some of its members also denoted as A , B, C, . . . , and an ordinary probability function P defined on this algebra, such that for each equation Pxr = r in C, we have P(X n Y ) = rP(Y). The question we asked may be given the precise form: does C admit of a simple Kolmogorov model? And the answer is yes, if and only if C implies no violation of Bell's inequalities. It is not difficult to see how to transpose the deduction that leads from the assumption of a common cause (conjunctive fork) to Bell's inequalities, to the present case. For note that, if x is any atom of the Boolean algebra in question, then P(X n Y 1 x ) = P(X I x) P(Y I x), no matter how strong the correlation between X and Y. The conclusion that phenomena involving violation of Bell's inequalities admit no common-cause model may have been an interesting if painful surprise. But the idea that they admit no descrip-

SALMON ON EXPLANATION AND CAUSALITY

649

tion satisfying Boolean logic and Kolmogorov's probability theory is a surprise of another order of magnitude. Remember after all that we can witness these phenomena on the macroscopic level, the relevant parts being frequency counts. Please reserve judgment on this for a moment while I continue the story begun in the preceding paragraph.4 Define a Hilbert-space model for that finite set C of equations to be a Hilbert space, with some vectors (PA, (PB,(PC,. . . in the space such that, for any equation PXY= r in C, r = I ( (PX,(PY ) 1 2 . We can now make the original question precise in a different way: does 2 admit a Hilbert-space model? The answer is yes, if and only if C satisfies certain numerical conditions, not so restrictive as Bell's inequalities (more restrictive if the Hilbert space is to be on the real numbers, and less restrictive if on the complex numbers). These conditions are certainly not tautological, and could conceivably be violated in nature too, though no such violations have been encountered in the experiments. This positive result is perhaps even more disquieting than the negative results discussed above, because the Hilbert space models could be called non-Boolean, non-Kolmogorovian probability models. They give real substance to the ideas that classical logic and probability theory have had their day, for they provide a precise formulation of an alternative. So far I have not stinted on the side of my opponents. But now I should like to point out that the deduction of Bell's inequalities from the assumption of a common-cause model had a much weaker premise than that of the existence of a simple Kolmogorov model. The word 'simple' is not redundant. Here is an example of a set C that does not admit a simple Kolmogorov model: PAB,PAC, and PBCall equal to zero. That means that if you measure one of A and B say, and find value 1 for it, you definitely will find another value for the other. But I also had said that A, B, C are two-valued. How could three variables each be two-valued and yet never have the same value? That is impossible. Yes, that is impossible. But the original description of C gave its equations as: P (a measurement of A yields 1 given that a measurement of B yielded 1) = PA& The situation described in the preceding paragraph is therefore not impossible, if the variables are not all three jointly measurable. A less simple Kolmogorov model now associates two sets with each variable: say A + and MA-read 4 ~ elegant n exposition of the above was given by Arthur Fine, Physical Reuiew Letters, xLvrr, (1982): 291-295; for the continuation below, see L. Accardi and A. Fedullo, Lettere a1 Nuovo Cimento, x x x ~ v(,1982): 161-172.

650

THE

JOURNAI. OF

PHILOSOPHY

"MA" as " A is measured." And then it could model 2 as follows: P x y = P ( X +n Y' I M X n M Y ) . We add that although P ( M A n M B ) , P ( M A n M C ) , a n d P ( M B n M C ) a r e each positive, P ( M A n M B n M C ) equals zero. It is clear, however, that we will be in difficulty again if we now add: but A, B , C each must have one of its two possible values at all times, whether measured or not. The only remedy here is to say, in Copenhagen style: Who says that physical magnitudes are like that? Why should they always have definite values? And what empirical results could make us think they do? So suppose that certain phenomena, described by such a set 2, admit no simple Kolmogorov model but d o admit a Hilbert-space model. They still admit a less than simple Kolmogorov model; so there is no reason to think that classical logic and probability theory have come a cropper. That model includes probabilities for the acts of measurement-information not expected, or required, from science. So we shall be happy if science gives us the set C of important conditional probabilities, and even happier if it gives us that set in the calculationally manageable form of a Hilbert-space model. In fact, all we need, to feel happy and at peace here is the Copenhagen agnosticism with respect to what happens to measurable physical magnitudes when they are not being measured. Perhaps they have n o values; or have values different from what measurements would have shown, had those been carried out; or whateverlet us hr interested in, but not unduly anxious about, the logically possible interpretations along all such lines. I11

Let me sum u p the second part of my paper as follows. Salmon explicitly rejected apriorism about what scientific explanation must be or do ( 2 4 2 ) . T h e very demand of explanation through causal mechanisms which he takes as paradigm was strongly reinforced by the victories of relativity theory, and could not have been met by Newton's celestial mechanics, as he himself points out. In turn, it is challenged by the successes of quantum theory, and this challenge to his conception was presented nowhere as incisively and precisely as by Salmon himself. T h e question is now, How shall we react to this challenge? When it comes to the form of explanation in general, I think I am much more liberal than Salmon; that is inherent in the difference between the pragmatic and the ontic conceptions. But when it comes to the evaluation of proffered explanations, as good, better, or best, I am no more inclined than he is to think there are no standards. We agree in our desire for a concrete and informative, yet comprehensive, account of what physical theory is and does. But I respectfully submit the following points

SALMON O N EXPLANATION AND CAUSALITY

65 1

concerning the challenge of quantum theory: (1) the difficulties for the ideal of the causal order arise at the level of the observable phenomena, and are independent of the intelligibility of the theory; (2) neither the phenomena nor the theory imply the possibility of empirically verifiable signals or action faster than light; (3) the nonclassical correlations implied by quantum states may be described in terms of remote (nonlocal) conservation laws, but there is no sign of remote conservation mechanisms; (4) the temptation to see the phenomena as violating the classical logic and probability framework delineated by Boole and Kolmogorov is understandable enough, but also finds no warrant in the phenomena. This list of four points is clearly not a list of disagreements with Salmon, for some of this he points out himself. But I wonder if, taken all together, they do not provide some good reasons to reduce the distance between Pittsburgh and Copenhagen. BAS C. VAN FRAASSEN

Princeton University CONFLICTING CONCEPTIONS OF

SCIENTIFIC EXPLANATION*

I

N his brief but penetrating paper, Philip Kitcher has raised a number of important issues. The most fundamental, I think, involves the conflict between two venerable intuitions: explanation as derivation and explanation as identification of mechanisms. A primary virtue of the former, he maintains, is the unification it achieves by appeal to general principles that enables us to use "the same pattern of derivation in case after case." I agree that unification has enormous explanatory power but would argue that it plays an equally central role for the mechanist. Just as the derivationist sees the same pattern of derivation in case after case, the mechanist sees the same kinds of basic mechanisms working in widely differing circumstances. Just as the derivationist appeals to general laws to carry out derivations, the mechanist recognizes that the basic mechanisms conform to general laws. The two viewpoints are equally committed to the covering-law principle and to the notion that our understanding of the world is improved through unification of diverse phenomena under general principles. Are there, then, any really important differences between the *Abstract of a paper to be delivered in an APA symposium o n Wesley Salmon's Scientific Explanation and the Causal Structure of the World, December 30, 1985, responding to papers by Philip Kitcher and Bas van Fraassen; see this JOURNAL, this issue, 632-639 and 639-651, respectively.

0022-362X/85/8211/065lf00.50

@

1985 T h e Journal of Philosophy, Inc.

Our partners will collect data and use cookies for ad personalization and measurement. Learn how we and our ad partner Google, collect and use data. Agree & close