Herbert Feigl, The "Mental" and the "Physical": The Essay and a Postscript (1967).

B. The Inference to Other Minds.

Behaviorism and phenomenalism display interesting similarities as well as fundamental differences. According to logical behaviorism, the concepts of mental states, dispositions, and events are logical constructions based on (physical1) characterizations of behavior. According to the more recent formulations of physicalism (Feigl, 113, 116; Carnap, 73; Sellars, 315) the "logical construction" thesis is inadequate and has to be replaced by an analysis in terms of postulates and correspondence rules. Very simply and very roughly, this means -- in the material mode of speech -- that for physicalism mental states are inferential ("illata," cf. Reichenbach, 273). Contrariwise, modern phenomenalism (Carnap, 60; Ayer, 12; Goodman, 135) had maintained that the concepts of physical things, states, dispositions, and events are logical constructions based on concepts designating the phenomena of immediate experience. And in the "revised" version of phenomenalism, i.e., a genuinely realistic epistemology based on phenomenal data, a doctrine which should not be called "phenomenalism" at all, the concepts of physical objects are inferential ("hypothetical constructs," "illata"). But this doctrine is in many of its tenets consonant with classical critical realism (von Hartmann, Külpe, Schlick, R. W. Sellars, D. Drake, C. A. Strong, J. B. Pratt, A. O. Lovejoy, G. Santayana). In contradistinction to critical realism, there is the earlier doctrine of neutral monism developed by the neorealists, especially E. B. Holt and Bertrand Russell (before his later critical realism), and historically rooted in the positivism and empiriocriticism of Hume, Mill, Mach, and Avenarius. Russell (284, 287) was the primary influence in Carnap's early epistemology (60, 61); and this sort of neutral monism was also adopted in prefatory philosophical remarks of some psychologists like E. C. Tolman (336), C. C. Pratt (260), and others.

The distinctive mark of neutral monism is a conception of the "given" which (1) is subjectless, i.e., it does not allow for the use of the personal pronoun "I"; and (2) is "neutral" in the sense that the given is characterizable as neither "mental" nor "physical." It maintains that both mentalistic concepts (the concepts of psychology) and physical concepts (those of physics) are logically constituted out of the more basic concepts designating neutral data. Psychology and physics are here understood as more or less systematic knowledge both on the level of common life, and on the more advanced level of science. Disregarding some technical logical questions, the data upon which the construction is based turn out to be items of immediate experience (sentience) and are thus "mental" after all, in one of the two senses of "mental" which we have been at pains to explicate.

This is not the place to review the many arguments1 which have been advanced in the refutation of phenomenalism. If an epistemology with a phenomenal basis can at all be worked out satisfactorily, then these data have to be conceived as lawfully related to the physical objects of everyday life. This means that the doctrine of logical constructionism or reductionism, i.e., of the explicit definability of physical concepts in terms of phenomenal concepts, has to be abandoned. The logical relations involved here are synthetic, and the translatability thesis is not just utopian (owing to the always admitted complexities), but completely inadequate, if not quixotic. I remain unimpressed with the significance of Craig's theorem (cf. Hempel's essay in Minnesota Studies in the Philosophy of Science, Vol. II) in this connection. An infinite set of postulates is not what phenomenalists ever had in mind. And I believe there are other grave objections to that sort of a translatability doctrine. The kind of translatability which Craig's theorem allows for concerns only the empirical content of theories in the sense of all conceivable evidential (confirming) statements, but not in the sense of the factual reference of the postulates (and, hence, of the theorems).

Mutatis mutandis, it is now realized in many philosophical and psychological quarters2 that the thesis of the translatability of statements about mental states (in phenomenal language) into statements about peripheral behavior (in descriptive, not theoretical physical1 language) must also be repudiated.

With this firmly established orientation, the inference of sentience (raw feels) in other organisms seems prima facie restored to its original form as an argument from analogy. I have no doubt that analogy is the essential criterion for the ascription of sentience. But a closer look at the logic of the inference will prove worthwhile. The inference from peripheral behavior to central processes, very much like the inference from skulls to brains contained in them, is intersubjectively confirmable, and this in the sense that independent intersubjeetive evidence for the truth of these conclusions is in principle available. Just this is, of course, not the case for the conclusions regarding mental states, if by mental states (sentience, raw feels) one means something that is not identifiable (i.e., not explicitly definable in physical terms) with either overt-behavioral or central-neural states or processes.3 If, contrary to the suggested orientation, such identifications could be made, i.e., if explicit definition could plausibly be given as an analysis of the meaning of phenomenal terms, then indeed no analogical inferences would be required. Nevertheless, considerations of analogy would be suggestive, though never decisive, for the terminological conventions according to which we apply or refrain from applying phenomenal terms to the behavior of animals and plants (let alone lifeless things).

If, however, phenomenal terms are logically irreducible to physicalistic terms, then parallelistic (epiphenomenalist) dualism is the most plausible alternative view. But inreractionistic dualism is empirically much less defensible, and its methodological orientation too defeatist, to be acceptable to the current scientific outlook (cf. section II, above). And epiphenomenalism also has generally been considered objectionable because it denies the causal efficacy of raw feels; and because it introduces peculiar lawlike relations between cerebral events and mental events. These correlation laws are utterly different from any other laws of (physical2) science in that, first, they are nomological "danglers," i.e., relations which connect intersubjectively confirmable events with events which ex hypothesi are in principle not intersubjectively ajid independently confirmable. Hence, the presence or absence of phenomenal data is not a difference that could conceivably make a difference in the confirmatory physical1-observatfonal evidence, i.e., in the publicly observable behavior, or for that matter in the neural processes observed or inferred by the neurophysiologists. And second, these correlation laws would, unlike other correlation laws in the natural sciences, be (again ex hypothesi) absolutely underivable from the premises of even the most inclusive and enriched set of postulates of any future theoretical physics or biology.

No wonder then that after a period of acquiescence with epiphenomenalism during the last century (T. H. Huxley, et al.), the behaviorist movement in psychology took hold, and exercised an unprecedented influence in so many quarters. Behaviorists, in their way, repressed the problem either in that they denied the existence of raw feels (materialism); or in that they defined them in physical-observation terms (logical behaviorism); or they maintained that the subject matter of scientific and experimental psychology can be nothing but behavior (methodological behaviorism), which leaves the existence of raw feels an open question, but as of no relevance to science. Our previous discussions have, I trust, clearly indicated that behaviorism in the first sense is absurdly false; in the second sense it is inadequate as a logical analysis of the meaning of phenomenal terms; and in the third sense, it is an admittedly fruitful but limited program of research, but it entails no conclusion directly relevant to the central philosophical issue.

The repudiation of radical behaviorism and of logical behaviorism entails the acceptance of some sort of parallelistic doctrine. Recent arguments for this position4 are prima facie highly persuasive. The basic point is simply that each of us knows his own states of immediate experience by acquaintance, and that by analogical reasoning we can infer similar, though never directly inspectable, states of experience in others. Direct inspection of the mental states of others is now generally considered a logical impossibility. For example, the subjunctive conditional, "If I were you, I would experience your pain,'' is not merely counterfactual, but counterlogical in that the antecedent of the conditional involves an outright inconsistency. The air of plausibility of the mentioned subjunctive conditional derives from entirely other, quite legitimate types of subjunctive conditionals, such as "If I had a broken leg (as you do), I should feel pain"; or "If I had (some traits of) your personality, flattery would please me." The logical grammar of personal proper names (or pronouns) however is such that it is downright self-contradictory to say (in a reasonably constructed and interpreted language) that Smith is Jones, or that I am you. The Mont Blanc cannot conceivably be identical with Mt. Everest!

Indirect verification or confirmation of statements regarding the mental states of other persons is however clearly possible once we have established laws regarding the correlation of the φ's with the ψ's for our own case. And as we have pointed out, these laws could in principle be most directly established with the help of an autocerebroscope. On the level of common life, of course, the correlations between neural and mental states are totally unknown. But a great many behavioral indicators are constantly being used in the (probabilistic) ascription of mental states. Logical analysis (Carnap, 73; Scriven, 306; Feyerabend, 119; Watling, 342; Feigl, 110, 111, 112, 114) has, I think, quite convincingly demonstrated the need for distinguishing the evidential bases from the factual reference of concepts and statements. The behavioral indicators serve as evidential bases for the ascription of mental states. Only the person who experiences the mental state can directly verify its occurrence. But there is no reason whatever to assume that when A reports his mental state, and B talks about it on the basis of behavioral evidence (or, if this is feasible, on the basis of neurophysiological evidence), that what they are talking about is not the very same mental state. This is indeed the way in which ordinary communication is understood. For example, if the doctor tells me a moment before lancing my abscess, "This will hurt," it is I who can directly verify this prediction. Moreover, most of us have learned from childhood on how to conceal our thoughts, feelings, sentiments, how to dissimulate, playact, etc. And so we can justifiably say that behavioral symptoms do not reliably indicate mental states. In the light of the basic principles of normal induction and analogy, involving symmetry considerations, solipsism (with its arbitrary asymmetries) must be regarded as an absurdly false, rather than as a meaningless doctrine.

If we had completely adequate and detailed knowledge of the neural processes in human brains, and the knowledge of the one-one, or at least one-many ψ-φ correlation laws, then a description of a neural state would be completely reliable evidence (or a genuine criterion) for the occurrence of the corresponding mental state. If these central neural events are essential intermediate links in the causal chain which connects stimuli with responses, then these central states are (probabilistically) inferable from stimulus-response situations. In this respect they have a logical status similar to the mental states as they are inferred from behavior in everyday life, or as the basis of psychological test situations. One may therefore wonder whether two steps of inference are really needed for a full logical reconstruction of the scientific ascription of mental states to other persons; the first step being the one from overt behavior to central neural events, and the second step being the one from neural events to mental states. I shall return to this question in subsection E, where I shall discuss the arguments for and against the identification of raw feels with the denotata of certain theoretical physical1 (or physical2) concepts.


Notes

1. Cf. Freytag (128); Külpe (191); Broad (50, 51); Schlick (298); Reichenbach (273); Pap (248); Lovejoy (204); R. W. Sellars (307); W. S. Sellars (308); B. Russell (288); Kneale (179); Beck (24); Feigl (110, 111); Berlin (35); Watling (342); Braithwaite (48); E. J. Nelson (237, 238); and now, after a drastic change in outlook, even Ayer (18) is close to a critical realist position.

2. Cf. Hempel (146); Camap (64, 67, 73); Kaufmann (175); Jacobs (163); Pap (242, 243, 245, 248); Ayer (18); Feigl (113, 116); Cronbach and Meehl (79); Scriven (306).

3. This is my way of stating succinctly the puzzle of "Other Minds" as it is understood in the long (unfinished) sequence of agonizing articles by John Wisdom (354), and in many other authors' publications, notably: Carnap (61, 62); Schlick (299); Ayer (15, 18); Austin (10); Pap (243, 248); Hampshire (141); Watling (341); Mellor (223).

4. By Pap (243); Hampshire (141); Watling (341); Ayer (18).