It is argued that the intuitively sanctioned distinction between beliefs and non-belief states that play a role in the proximate causal history of beliefs is a distinction worth preserving in cognitive psychology. The intuitive distinction is argued to rest on a pair of features exhibited by beliefs but not by subdoxastic states. These are access to consciousness and inferential integration. Harman's view, which denies the distinction between beliefs and subdoxastic states, is discussed and criticized.

Hintikka has criticized psychologists for “hasty epistemologizing,” which he takes to be an unwarranted transfer of ideas from psychology (a discipline dealing with questions of fact) into epistemology (a discipline dealing with questions of method and theory). Hamlyn argues, following Hintikka, that Gibson's theory of perception is an example of such an inappropriate transfer, especially insofar as Hamlyn feels Gibson does not answer several important questions. However, Gibson's theory does answer the relevant questions, albeit in a new and radical way, which suggests that the alleged distinction between psychology and epistemology is suspect. In fact, contrary to Hintikka and Hamlyn's claims, Gibson's theory of perception appears to be a valuable source of epistemological as well as psychological ideas.

This paper contrasts two information-theoretic approaches to statistical explanation: namely, (1) an analysis, which originated in my earlier research on problems of testing stochastic models of learning, based on an entropy-like measure of expected transmitted-information (and here referred to as the Expected-Information Model), and (2) the analysis, which was proposed by James Greeno (and which is closely related to Wesley Salmon's Statistical Relevance Model), based on the information-transmitted-by-a-system. The substantial differences between these analyses can be traced to the following basic difference. On Greeno's view, the essence of explanation lies in the relevance relations expressed by the conditional probabilities that relate the explanans variables to the explanandum variables; on my view, in contrast, the essence of explanation lies in theories viewed as hypothetical structures which deductively entail conditional probability distributions linking the explanans variables and the explanandum variables. The explanatory power of a stochastic theory is identified with information (regarding the values of explanandum variables) which is “absorbed from” the explanans variables. While other information which is “absorbed from” the explanandum variables (through the process of parameter estimation, for example) reflects descriptive power of the theory. I prove that Greeno's measure of transmitted information is a limiting special case of the E-I model, but that the former, unlike the latter, makes no distinction between explanatory power and descriptive power.

This paper presents a theory of conclusions based upon the suggestions of Tukey [21]. The logic offered here is based upon two rules of detachment that occur naturally in probabilistic inference, a traditional rule of acceptance, and a rule of rejection. The rules of detachment provide flexibility: the theory of conclusions can account for both statistical and deductive arguments. The rule of acceptance governs the acceptance of new conclusions, is a variant of the rule of high probability, and is a limiting case of a decision-theoretic rule of acceptance. The rule of rejection governs the removal of previously accepted conclusions on the basis of new evidence. The resulting theory of conclusions is not a decision-theoretic logic but does, through the aforementioned limiting property, provide a line of demarcation between decision and conclusion (i.e., nondecision) logics of acceptance. The theory of conclusions therefore complements decision-theoretic inference.

The theory of conclusions presented here satisfies Tukey's desiderata, specifically:

1. (1) conclusions are statements which are accepted on the basis of unusually strong evidence;

2. (2) conclusions are to remain accepted unless and until unusually strong evidence to the contrary arises;

3. (3) conclusions are subject to future rejection, when and if the evidence against them becomes strong enough.

Finally, the preferred theory of conclusions has a strong conservative bias, reflecting Tukey's aims.

This article is primarily a study of the group selection controversy, with special emphasis on the period from 1962 to the present, and the rise of inclusive fitness theory. Interest is focused on the relations between individual fitness theory and other fitness theories and on the methodological imperatives used in the controversy over the status of these theories. An appendix formalizes the notion of “assertive part” which is used in the informal discussion of the methodological imperatives elicited from the controversy.

While the use of so-called idealizations in science has been widely recognized for many years, the philosophical problems that arise from this use have received relatively little attention. Even a cursory reading of the philosophical literature devoted to these problems ([1]; [2]; [4], pp. 160–171; [5]; [7], pp. 54–63) reveals that the following questions remain unanswered: In general, what, if any, are the distinguishing characteristics of idealizations? More specifically, do idealizations have any distinguishing syntactic or semantic characteristics? In addition to these questions there exist the following pragmatic questions, questions relating to the ways in which idealizations are used in science: How are idealizations used in explanations? (see [2]). Do these explanations have any peculiar characteristics—characteristics not shared by deductive-nomological (D-N) explanations? If we assume that (at least some types of) idealizations are false (or contain false components) or “do not obtain,” how is it that they can have any explanatory power? (see [7], p. 58). Further, there are questions of more general philosophic concern. How do the problems of idealizations relate to those of simplicity, for an idealization seems to be, in some sense, a type of simplification? How do the problems of ideal laws and theories relate to the general problems of scientific laws and theories?

Germund Hesslow has argued recently [2] that a probabilistic theory of causality as advocated by Patrick Suppes [4] has two problems that a deterministic theory avoids. In this paper, I argue that Suppes' probabilistic causal calculus is free of each of these problems and, moreover, that several broader issues raised by Hesslow's discussion tend to support a probabilistic conception of causes.

In a recent paper, Michael Levin argues that both statements reporting causal relations and causal explanatory statements are extensional. We show that his argument for the extensionality of causal explanatory statements fails to establish that conclusion. His claim that certain ‘because’ statements are elliptical for statements of what he terms the ‘causal explanatory’ form is unsubstantiated. The argument for the referential transparency of the allegedly explanatory form, regardless of whether it is a distinct explanatory form, fails because of scope problems. Finally, we argue that the apparent referential opacity of explanations, the attribution of certain kinds of properties to explanations, and our satisfaction with explanations in what appear to be disparate forms are best accounted for, not by the assumption of ellipticality of statements in various forms for statements in some inherently explanatory form, but rather by an account of explanation as a speech act, the success or failure of which may be affected by which of several co-referring expressions is employed.

Limitations of space dictate that we confine ourselves to Miss Stern's most salient comments. First, a preliminary point. Miss Stern says “Levin offers no argument” for why “e happened because of c” implicitly contains an explanatory description, while “c caused e” does not. But surely the remark ([1], 273) that we often know that c caused e without knowing why c caused e is just such an argument. Our linguistic intuition suggests that we use the first locution in this case; Miss Stern's evidently does not. Cases of non-diagnostic causal ascriptions do exist, and even ordinary language must provide for them; how it does so is a verbal dispute about English.

In a recent article [1], Kenneth Sayre presents what he takes to be an alternative to Salmon and Reichenbach's screening-off model of causal relations. His statistical model, based on communication theory, supposedly “has the added advantage of distinguishing cause and effect without reference to temporal distinction” (p. 203 abstract, cf. 212–213). Unfortunately his model falls far short of this grand claim, faltering before the same sort of examples I had previously used to assail the screening-off model [2].

Having noted my intention to improve upon other statistical analyses of causal relatedness by providing theoretical means to distinguish cause and effect without reference to temporality [1], Professor Shrader argues [3] that the alternative model I offer falters “before the same sort of examples he had previously used to assail” these other analyses. Although the assailant seems both to have misunderstood the model in question and to have overestimated the effect of his counterexamples, his objections provide a welcome opportunity for clarification.

In his article “On Mechanical Recognition” ([6]) R. J. Nelson brings to bear a branch of mathematical logic called automata theory on problems of artificial intelligence. Specifically he attacks the anti-mechanist claim that “[i]nasmuch as human recognition to a very great extent relies on context and on the ability to grasp wholes with some independence of the quality of the parts, even to fill in the missing parts on the basis of expectations, it follows that computers cannot in principle be programmed to recognize or learn to recognize all patterns” ([6]), p. 24). Nelson proposes, contrary to this claim, that “gestalt recognition is not beyond digital automata” ([6], p. 24). in particular, he claims, he will establish by what he calls “Turing machine arguments” ([6], p. 25) that the following four theses are true:

1. (1) automata can recognize different pattern types in one and the same set of instances;

2. (2) automata can recognize the “same” pattern in different (even completely disjoint) sets of instances;

3. (3) automata can recognize incomplete, degraded patterns having missing or indeterminate parts;

4. (4) automata can recognize family resemblances. ([6], pp. 24–25)