Van Brakel (1993), Grush (1993), and Morris (1993) dispute the analysis of the frame problem which is advanced in Fetzer (1991a) and (1991b). I suggest that the frame problem can best be understood as a special case of the problem of induction. Those who persist in maintaining that the frame problem is exclusively a problem of representation miss the point that without a solution to the problem of change even the availability of a representation scheme serves no suitable purpose because there is no solution to the problem of change to represent. And those who persist in maintaining that the frame problem is one of "common sense" rather than one of scientific knowledge need to understand that changes in the world come about as a causal consequence of the operation of natural laws, the objects of scientific discovery. The resources "common sense" provides are not adequate to the problem.
2. Somewhat to my surprise, this suggestion of mine, which I considered to be sufficiently well-founded as to be almost beyond serious debate, provoked a negative response, first from the editors who were reviewing the proceedings of the Pensacola Conference for possible inclusion in a book -- I have in mind Patrick Hayes especially, but also another referee, who has remained anonymous, a disagreement mediated by Ken Ford that culminated in the publication of the exchange between us -- and now van Brakel and Grush, both of whom reply to my prior remarks on van Brakel, and Morris, who like Hayes, presumes that the frame problem is a matter of representation. What I would like to do here, therefore, is to explain the reasons why the frame problem may be viewed as a special case of the problem of induction and to respond to the criticism of these commentators.
3. Let us assume that at least three interrelated issues are involved here; these might be thought of as the problem of change, the problem of representation, and the problem of implementation, respectively. The problem of implementation can be solved if a suitable representation can be designed and implemented in the form of a computer program. If there were no suitable representation to implement in the form of a computer program, however, it would be impossible to solve the problem of implementation. Similarly, the problem of representation can be solved if there is a suitable language (or representation scheme) for representing solutions to the problem of change. If there were no suitable solution to the problem of change, however, then (even given a representation scheme) there would be nothing to represent.
4. A much simpler response to van Brakel (1992), therefore, might have been to observe (as Hadley [1988], among others, has previously observed) that while some computer scientists, such as Patrick Hayes, want a narrow, technical definition of the frame problem, philosophers tend to "include any problem whose solution is PRESUPPOSED by a solution to the narrow problem" within its scope (Hadley 1988, p. 34; original emphasis). Thus, if the problem of implementation cannot be solved without solving the problem of representation, and if the problem of representation cannot be solved without solving the problem of change, then (as I see it) the frame problem has three dimensions in the absence of solutions to which it cannot be solved, where the most basic of the three (from a logical point of view) turns out to be the problem of change.
5. In retrospect, I now believe that 4.0 above is most of what I wanted to say about van Brakel (1992) and that I came across as excessively provocative in Fetzer (1993a). The electronic media have their distinctive strengths and weaknesses, as I observed in Fetzer (1992a), and I appear to have succumbed to the temptation to say what I had to say more forcefully than I ought to have said it. Nevertheless, while I concede to excesses in the form of my presentation, I maintain the content of my position. If there is a "family of frame problems" that van Brakel (1992) identifies with [A] "Which things change and which don't?", [B] "How can (solutions to) [A] be represented?", and [C] "How can/do we reason about [A]?", this "family" seems to be close enough to my "dimensions" for me to again affirm in 4.0 the views of Fetzer (l993a).
6. Yet the position advanced by Hayes (1991), for example, might possibly be based on the perception that the problem of representation can be solved if we can design a suitable language (or representation scheme) for representing solutions to the problem of change, whether or not the problem of change itself is amenable to solution. A distinction of this kind, I suppose, would enable Hayes to contend that the problem of induction -- which I take to be the genus (or general problem) of which the problem of change is a species (special case) -- is "logically irrelevant" to the frame problem, when it is strictly defined as a problem of representation in this fashion. This might make sense of Hayes (1991), but it appears to be a "Pickwickian" pose to adopt if there is no solution to the problem of change to represent.
7. I do not take van Brakel to be assuming a "Pickwickian" stance, and he does acknowledge that the problem of induction has some relevance here. He seems to think that by describing his [A], [B], and [C] as "three dimensions" of the frame problem, I am thereby committing a mistake, but apparently because NOTHING, including the frame problem, can be precisely defined (van Brakel 1993). Insofar as I am maintaining that the frame problem is SUBSUMED by the problem of induction, he suggests, I have misrepresented their relations, because they are more loosely connected. I would grant that the frame problem cannot be COMPLETELY subsumed by the problem of induction, because it has the representational and implementational aspects that are not elements of the problem of induction, as that problem is traditionally understood.
8. The point of describing "the frame problem" as "a special case" of the problem of induction, therefore, was to emphasize that the frame problem cannot be solved without a solution to the problem of change, which is the DIMENSION of the frame problem that is "a special case" of the problem of induction (except when the frame problem is given the "Pickwickian" twist described in 6.0). To this extent, I do not want to contest van Brakel's denial that the problems are identical. But I do want to insist that a COMPLETE SOLUTION to the frame problem (except in its "Pickwickian" guise) presupposes an adequate solution to the problem of induction. Thus, a solution to the problem of induction is necessary (essential) to resolve the frame problem.
9. Nothing here, however, mitigates my skepticism about the possibility of complete descriptions for concrete events as features of the world's history. I take it that van Brakel and I agree on this matter, which Hempel (1965), especially, has patiently explained, but that we disagree on its significance. Thus, as Hempel emphasizes, we can never completely describe, much less completely explain the occurrence of any concrete event, which means that at best an event can only be explained as an event under a description or perhaps as an event of A SPECIFIC KIND (Hempel 1965, p. 233, for elaboration). I would readily acknowledge that, if the explanation of any event required an explanation for every property of that event, then explanation would be impossible, since it is impossible to even describe every property of any such occurrence.
10. At this juncture, we appear to part ways. Van Brakel construes me as thinking that the impossibility of complete descriptions can be overcome by appealing to "a panacea like relevance or salience." Insofar as I did not specifically contest this description in Fetzer (1993a), van Brakel infers that I agree that appeals to MAXIMAL SPECIFICITY and the like are "useless as a contribution to solving the frame problem. The problem is just pushed ahead" (van Brakel 1993, Sec. 4). Strictly speaking, of course, I do not think that anyone can "break out" of the problem of complete descriptions, which appears to be impossible. What I think is that explanations for events as events of specific kinds or under certain descriptions can be provided without having to solve the unsolvable problem van Brakel cites.
11. As for relevance as a "panacea," I view the notion of RELEVANCE as among the most basic of all concepts, but it is not therefore unambiguous. In fact, a number of different conceptions must be distinguished, which range from logical relevance to evidential relevance to explanatory relevance, and a subset of explanatory relevance conceptions that range from statistical to causal to nomic relevance. The notion underlying these conceptions of explanatory relevance, for example, is that a property F is explanatorily relevant to the occurrence of an outcome E when the presence or absence of F makes a difference to the occurrence of an outcome of that kind, where making this precise is one of the objectives of the theory of explanation as it has been pursued by Hempel (1965), Salmon (1971), (1984), and Fetzer (1981), (1993b).
12. The theory of explanation that I propose depends on a conception of laws of nature of which Hume would have disapproved. I maintain that a distinction ought to be drawn between "permanent" and "transient" properties, where permanent properties are contingent dispositions something cannot lose without also losing a corresponding reference property. Dispositions are envisioned, in turn, as single-case causal tendencies that may be of universal or of probabilistic strength. These conceptions provide an ontological foundation for fixing the truth conditions of logically contingent subjunctive conditionals. Lawlike sentences are true only if they are maximally specific and are empirically testable by attempts to establish that they are false. I have elaborated this account in some detail, most recently in Fetzer (1993b).
13. Van Brakel concedes the existence of laws of nature but believes they "cannot be applied to concrete events without the addition of unspecified ceteris paribus conditions." On this point, we are in substantial disagreement. There are several approaches toward understanding the nature of laws of nature and the structure of scientific theories, some of which bear similarities to the position that van Brakel endorses. It is my position that the success of empirical science in discovering laws of nature depends upon there being no more than a finite number of kinds of factors that make a difference to the occurrence of any specific event. When there are more than a finite number of factors or we cannot discover them, science cannot succeed.
14. If van Brakel would like to address the problem of the nature of laws of nature and the logic of scientific explanations, I would welcome a debate within a more appropriate forum. For present purposes, however, it would probably be more suitable for me to address briefly the questions he poses about necessary and sufficient conditions for events and concepts. There are various positions like van Brakel's, some related to Wittgenstein, that hold that necessary and sufficient conditions are never available for causes (because there are infinitely many factors that influence the occurrence of an event, because ceteris paribus conditions are always required, and so forth) or for concepts (because their meaning is to be found in their use, because they are only [at most] connected by family resemblance relations, and so on).
15. Van Brakel claims that I appeal to "some worn out positivistic ideas if the 1930s," but I do not imagine that he has much familiarity at all with my views on the problem of meaning (Fetzer 1990, esp. Part I, 1991c, 1991d, 1991e or 1992b, for example). As for necessary and sufficient conditions in the case of concepts, I presume that nominal definitions, meaning analyses, and empirical analyses are amenable to being understood in terms of necessary and sufficient conditions, as those are ordinarily understood, but that ostensive definitions, explications, and contextual definitions are not. Insofar as the situation with respect to events is closely related to Hayes's position, I shall consider this next. But I would concede that the matter of meaning is vastly more complex than was implied by my recent comments on van Brakel.
16. Hayes wants to maintain a distinction between "common sense" and "scientific knowledge" where the frame problem is a problem for common sense that does not require scientific knowledge. If a strong distinction were drawn between "common sense" and "scientific knowledge," the frame problem could be be said to be the common-sense version of the problem of induction and that would still be distinct, which would trivially distinguish between them. I maintain that the existence of "common sense" should not be taken for granted (Fetzer 1990, pp. 141-145), though I am less skeptical that there is something called "ordinary knowledge" that differs from scientific knowledge more in degree than in kind (Fetzer 1981, pp. xii-xiii). Here I simply want to advance some examples that seem to illustrate some differences which are at stake here:
17. It appears to be a matter of common sense to suppose that someone who is stabbed through the heart will die. Yet there are conditions under which people is stabbed through the heart will not die, for example, when they are connected to a heart-lung machine while undergoing heart surgery. It also appears to be a matter of common sense to suppose that all people who take a large dose of a potent poison will probably kill themselves and that people who take twice as much will almost certainly kill themselves. Yet if one were to take a large dose of an alkaline poison and another large dose of an acidic poison, one would actually reduce rather than increase the probability that of killing oneself. Other cases are not difficult to generate by including or by excluding various causally relevant conditions.
18. It might be "common sense" to infer that one will die if stabbed through the heart, but that is by no means always the case. And it might be "common sense" to think that those who take twice as much poison will increase their chances of killing themselves, but that is not necessarily what happens. If we happen to be surgeons, we may draw inferences that nonsurgeons would not draw. And if we happen to be chemists, we may draw other inferences that nonchemists might not draw. What qualifies as "common sense" can clearly differ from person to person and from time to time as a function of factors such as age, education, and general experiences in life. There is no specific set of beliefs that qualifies as "common sense" that is common to the old and the young, the educated and the ignorant, rich and poor alike.
19. If we are considering an appeal to "common sense" as a foundation for solving the frame problem, we have to ask, "Common sense for whom?" As soon as we recognize that the presence or the absence of causally relevant factors makes a difference in each specific case, we ought to consider the prospect of discovering the complete range of relevant factors in the kind of cases that make a difference to us. If empirical science is an attempt systematically to discover complete sets of relevant conditions that make a difference in various domains (such as physics, chemistry, biology, and so forth) then it should be obvious that scientific knowledge might provide a more reliable foundation for anticipating the course of experience than the highly variable and incomplete knowledge that passes for "common sense" can possibly provide.
20. The example of a match's lighting when it is struck in the presence of oxygen, provided that it is not wet, is of appropriate chemical composition, and struck with suitable force, was intended as an illustration of contingently necessary and sufficient conditions, where those conditions are individually necessary and jointly sufficient in the case of specific events (tokens) as instances of laws. Other illustrations are at hand of conditions that are individually necessary or jointly sufficient in the case of generic events (types) as instances of other laws. Animal life, for example, cannot endure without food and water, which are necessary to those forms of life. And for any system (including human beings) that wants to lose weight (or mass), it is sufficient for that system to expend more energy than it consumes.
21. Appeals to "common sense" may make no difference when the consequences of making mistakes are not severe. But when the effects would make a difference in our quality of life or its continuance (whether in an individual case or in relation to the species), the risks of relying upon "common sense" appear to outweigh the benefits. Scientific knowledge, after all, promises to provide knowledge of the kind that is required to enhance our prospects of success. In my response to Hayes (1991), I accordingly provided a diagram representing the relative preferability of programs based on scientific knowledge as opposed to those based upon expert opinion as opposed to those based on "common sense" (Fetzer 1991b). (Please note, however, that as published in Ford and Hayes, eds. (1991), Figures 1 and 2 are reversed.)
22. When Grush (1993) suggests that there may be "less disagreement between van Brakel and Fetzer than Fetzer makes out, at least on these issues," I am accordingly prepared to agree with him. I believe I was overly confrontational and consequently exaggerated some differences between us. Other differences remain, however, especially concerning the nature of laws, the requirement of maximal specificity, and the conditions that definitions must satisfy. Morris (1993), of course, provides a review of the entire book, including a commentary on my exchange with Hayes. In his view, my response to Hayes does little more than restate the main themes of my original piece. I think that Morris has not appreciated what I have said there, but one hopes that he will find something of substance that merits further consideration here. [Minor technical point: in Sec. 4.8, Morris mistakenly claims that inferences involving "closed systems" (which satisfy maximally specific descriptions) are "deductively valid," but that is the case only when the laws of systems of those kinds happen to be universal rather than probabilistic.]
23. Those who persist in maintaining that the frame problem is exclusively a problem of representation (or a problem of implementation), therefore, appear to miss the point, namely, that without a solution to the problem of change, even the availability of a representation scheme serves no suitable purpose, because there is no solution to the problem of change to represent. And those who persist in maintaining that the frame problem is one of "common sense" rather than one of scientific knowledge need to understand that changes in the world come about as a causal consequence of the operation of natural laws, which are the objects of scientific discovery. Surely no useful purpose can be served by restricting attention to "common sense" when the resources that it provides are not adequate to the problem.
REFERENCES:
Fetzer, J. H. (1981) Scientific Knowledge: Causation, Explanation, and Corroboration (Dordrecht, The Netherlands: D. Reidel, 1981).
Fetzer, J. H. (1990) Artificial Intelligence: Its Scope and Limits (Dordrecht, The Netherlands: Kluwer Academic Publishers, 1990).
Fetzer, J. H. (1991a) The Frame Problem: Artificial Intelligence Meets David Hume. In: Ford and Hayes (1991).
Fetzer, J. H. (1991b) Artificial Intelligence Meets David Hume: A Response to Pat Hayes. In: Ford and Hayes(1991).
Fetzer, J. H. (1991c) Philosophy and Cognitive Science (New York, NY: Paragon House Publishers, 1991).
Fetzer, J. H. (1991d) Aspects of the Theory of Definition. In: Fetzer (1991f).
Fetzer, J. H. (1991e) Primitive Concepts: Habits, Dispositions, and Laws. In: Fetzer (1991f).
Fetzer, J. H., ed. (1991f) Definitions and Definability: Philosophical Perspectives (Dordrecht, The Netherlands: Kluwer Academic Publishers, 1991).
Fetzer, J. H. (1992a) Editor's Preface. Minds and Machines 2 (1992) p. iii.
Fetzer, J. H. (1992b) Connectionism and Cognition: Why Fodor and Pylyshyn are Wrong. In A. Clark and R. Lutz, eds., Connectionism in Context (Heidelberg, FRG: Springer-Verlag, 1992) pp. 37-56.
Fetzer, J. H. (1993a) Van Brakel's Position Appears to be Incoherent. PSYCOLOQUY 4(14) frame-problem.4.
Fetzer, J. H. (1993b) Philosophy of Science (New York, NY: Paragon House Publishers, 1993).
Ford, K. and P. J. Hayes, eds. (1991) Reasoning Agents in a Dynamic World: The Frame Problem (Greenwich, CT: The JAI Press, 1991).
Grush, R. (1993) Van Brakel's Position is Perfectly Coherent. PSYCOLOQUY 4(24) frame-problem.8.
Hadley, R. (1988) Review of Z. Pylyshyn, ed., The Robot's Dilemma: The Frame Problem in Artificial Intelligence, Canadian Philosophical Reviews VIII (1988) pp. 33-36.
Hayes, P. J. (1991) "Commentary on 'The Frame Problem: Artificial Intelligence Meets David Hume'", in Ford and Hayes, eds. (1991).
Hempel, C. G. (1965) Aspects of Scientific Explanation (New York, NY: The Free Press, 1965).
Morris, R. (1993) The Changing Scene. PSYCOLOQUY 4(25) frame-problem.9.
Salmon, W. C. (1971) Statistical Explanation and Statistical Relevance (Pittsburgh: University of Pittsburgh Press, 1971).
Salmon, W. C. (1984) Scientific Explanation and the Causal Structure of the World (Princeton: Princeton University Press, 1984).
Van Brakel, J. (1992) The Complete Description of the Frame Problem. PSYCOLOQUY 3(60) frame-problem.2.
Van Brakel, J. (1993) Unjustified Coherence. PSYCOLOQUY 4(35) frame-problem.7.