Herbert Feigl, The "Mental" and the "Physical": The Essay and a Postscript (1967).
D. Reduction and Identification in Scientific Theories.
In order to decide whether the mental and the physical can in some sense be identified, it is indispensable to cast at least a brief glance at the logic of reduction and identification in the sciences, especially in physics, biology, and psychology. Although these reflections will not provide us with the complete solution of the problem, they will be helpful and suggestive.
It was pointed out and briefly discussed in section II that the advance of scientific theories consists essentially in the reduction of a variety of originally heterogeneous observable facts and regularities to a unitary set of explanatory concepts and postulates. Customarily it is said, for example, that visible light is electromagnetic radiation (within a certain interval of wave lengths); that table salt is NaCl; that magnetized iron is an aggregate of iron atoms with a characteristic spin of certain of their electrons; that the transmitters of hereditary traits are the genes in the chromosomes of the germ cells; that (at least) short range memory traces are reverberating circuits in cerebral cell assemblies, etc. The "is" and the "are" in these sentences represent identities. But these identities differ in their mode of certification from the analytic identities of pure logic and mathematics. For extremely simple illustrations consider the general theorem of set theory "[S v T] = - [ - S · - T]" and the specific arithmetical identity "√64= 23" which hold by virtue of presupposed definitions and the principles of logic or arithmetic.
But the identities established in the factual sciences are confirmed on the basis of empirical evidence. This is very like the empirically ascertainable identity of Shakespeare (or could it be Marlowe?) with the author of Hamlet, or the identity of the author of Hamlet with the author of King Lear. Of course there are also such empirically ascertainable identities as those of Tully and Cicero, of William Thompson and Lord Kelvin, or of the evening star and the morning star. In the examples just given we have (extensional) identities of individuals labeled or uniquely described in two or more ways. When it comes to properties (universals), the identity may be either intensional or extensional. An illustration of the first is, e.g., the identity of d# and eb in the well tempered scale of music. An illustration of the second is the identity of the chemical element with atomic number or nuclear charge 20 with calcium characterized as a constituent of limestone, of atomic weight 40, having a melting point of 810° C, a specific heat of 0.169 at 20° C, etc.
In the case of analytic identities of individuals or of properties we may speak of the synonymy of names or predicates, respectively. (This applies, of course, also to two-place, three-place, etc. predicates, i.e., to dyadic, triadic, etc. relations. Thus, e.g., "earlier than" is logically synonymous with "temporally precedent to" or with the converse of the relation "later than"). The identity of the class of rational animals with the class of featherless bipeds (disregarding plucked birds), or with the class of laughing animals (disregarding hyenas), is extensional and empirical. Of course, extensional identity, be it logically necessary or empirical, is implied by intensional identity, but not vice versa. There is no longer any reason to be puzzled about identity being a relation. The proper explication of identity consists simply in the recognition that one and the same individual (or universal) may be designated by different labels or described by different characterizations. This could (but need not) be formulated by saying that the relation of identity fully explicated amounts to a triadic relation between labels (L), or descriptions (D) and a referent (R). The following diagrams represent the simplest paradigmatic situations.
L1 → R ← L2 or L → R ← D or D1 → R ← D2Since I am not a nominalist, having remained unconvinced by the arguments of Quine, Goodman, and White (269, 242), I see no objection to introducing universals as referents of predicates or relations. And since I am not a Platonic realist either. I am quite willing to consider talk about universals as a convenient facon de parler, rather than as a matter of profound "ontological" significance. In my previous example I regarded "d#" and "eb" as different labels for the same kind of musical tone-as-heard. Similarly I see no reason whatever to deny that "calcium" and "element of atomic number 20" designate the same kind of substance. This amounts to saying that the identity of universals, if it is not based on the logical synonymy of intensions, can amount only to an extensional (in this case, empirical) equivalence of two classes.
Prima facie the identifications achieved by scientific laws and theories appear to be cases of co-extensiveness, i.e., of extensional equivalence. This is certainly the case with identifications based on empirical laws. A metal characterized in terms of its thermal conductivity may be identical with the metal characterized by its electric conductivity. The ascertainment of the identity, in this case, depends upon the validity of the Wiedemann-Franz law according to which there is a linear relationship between the two kinds of conductivity. Now, while I grant that the word "identity" has only one meaning, and this is the meaning defined by the (properly understood) Leibniz principle of identitas indiscenibilium, the modes of ascertainment of identity are for our purposes the essential consideration. I shall therefore take the terminological liberty of speaking of different kinds of identity, viz., (1) logical, (2) empirical; and under (2) I shall distinguish (a) accidental, (b) nomological, (c) theoretical identities. In more precise but also more cumbersome language this would amount to distinguishing the various modes of ascertainment of identity, or the types of validity that assertions of identity may have.
The identity of the class of rational animals with the class of featherless bipeds may be considered not only as logically contingent, but as empirically accidental; in the same sense as we consider it empirically accidental that the city which is the seat of the United States Government is identical with the city in which on January 17, 1956, at 11:00 a.m. the temperature was (say) 43° F., the barometric pressure 30 inches, and the relative humidity 89 per cent. The referent of these descriptions is the one city of Washington, D.C. This is identity of individuals. Nomological identities rest on empirical laws; theoretical identities depend upon the postulates and definitions of a scientific theory. Since all types of identity, except the logical, are established on the basis of empirical evidence, they must therefore be formulated in synthetic statements.
There is, however, the temptation to regard certain well established theoretical identities as analytic. For example, if "gas pressure" is defined in terms of the sum of the momenta delivered by the molecules of a gas to the walls of its container, then of course within the context of the kinetic theory of gases, the identification of pressure with the sum of the molecular momenta is analytic. But, as Ernest Nagel (230) has made clear, if we mean by "the pressure of a gas" that property of it which is measurable by manometers, and which has a variety of well-known lawful connections with the volume, the temperature, etc. of the gas, and thus "manifests" itself in a variety of ways, then clearly it was a discovery, yielding new information, that revealed to us the relation of gas pressure (the "macro"-concept) to certain aspects of molecular motion. This is clearly synthetic. The interesting point which makes it so tempting to view the relation as analytic is, however, worth a little discussion. It is not simply the much vaunted arbitrariness of definitions.1 It is rather that the macro-properties and macro-regularities of gases can be derived2 from the assumptions of the molecular-kinetic theory. A full fledged micro-theory of thermal conduction, convection, diffusion, etc. thus enables us, among other things, to derive the regularities of such indicating instruments as the manometers, thermometers, etc. The expansion of the volume of the gas in the gas thermometer is an immediate logical consequence of the (assumed) increase in the average velocities of the molecules making up the gas, and the initial and boundary conditions which characterize the micro-state of the instrument. Quite analogous considerations apply to the electron theory of electric currents and the measurements of electromotive force and current intensity with the help of such indicating instruments as the voltmeter and the ammeter.
The explanation of the macro-behavior of organisms is sought along methodologically similar lines. Neurophysiological laws and neural-endocrine-muscular, etc. states will presumably suffice for the explanation of even as complex and intricate behavior as that of human beings. Disregarding the ultimately (possibly inevitable) statistical aspects of some of the laws or of the assumptions about initial and boundary conditions, the neurophysiology of the future (3000 A.D.?) should provide complete deductive derivations of the behavior symptoms of various central states whose ψ-correlates are the familiar sensations, perceptions, thoughts, beliefs, desires, volitions, emotions, and sentiments (known by acquaintance and described in phenomenal language). Perhaps I should make clear that I am here trying not so much to convince my readers of the feasibility of what he may consider an entirely Utopian and quixotic program for science. I am rather concerned to argue conditionally, i.e., if this physicalistic program can be carried out, then there would be something like an empirical identification of the referents of molar behavior theory concepts with the referents of some neurophysiological concepts. In its logical and methodological aspects this would be quite analogous to,the identification of, e.g., the property of magnetism (as conceived in the macro-theories of physics) with certain micro-structures and processes involving electron spins, etc., ascribed to the atom and quantum dynamics of ferromagnetic substances. These identifications, like all others of a similar kind3 appear as analytic only because of the mentioned relations of deducibility which we know (or believe) to hold between the micro-theoretical and macro-nomological or macro-descriptive propositions.
But a more accurate analysis reveals invariably a synthetic-empirical feature somewhere in the context of such scientific explanations. Just where this feature is located depends largely on the nature of the logical reconstruction by means of which we analyze those explanations. In the case of the length of the mercury column in a thermometer, or the volume of the gas in a gas thermometer, the derivation of their (respective) expansions under the condition of increasing heat intensity is so direct that the "identity" appears deceptively as a logical one. But even here, empirical regularities enter in. In addition to considerations of the respective thermal expansion coefficients of gases or mercury as compared with those of the glass of the instruments, there are the laws of geometrical optics regarding the paths of the light rays, and the laws of psychophysics and of psj'chophysiology concerning the visual perception of the mercury column or of the indicator (e.g., a drop of ink) of the gas thermometer.
Just where we decide to put the boundary (or "partition") between the data of observation and the inferred state of affairs is thus a matter of convenience in epistemological reconstruction. But somewhere we must put it, if we are not to lose sight of the empirical character of the relation between the data and the illta. In one reconstruction the data statements concern the observables of common life. This is the epistemology favored by thinkers like Popper, Carnap, Reichenbach, Hempel, Ryle, Black, Skinner, and W. Sellars. They all agree in this respect even if they differ sharply in others. They all accept in one way or another an intersubjectvie (physicalistic) thing-language as the basis of epistemological reconstruction. Bertrand Russell, in his later works, is about the only thinker who has made a valiant attempt to combine acceptance of a phenomenal basis with a realistic (non-phenomenalistic) reconstruction. This means that, as a realist, he has long ago abandoned the earlier phenomenalistic translatability doctrine, and has ever since regarded the relation between physical object statements and phenomenal data statements as one of probabilistic inference. I believe this position still needs considerable logical clarification, but I also believe that it is basically sound, in that it pursues the epistemological analysis down to data which involve only that minimum of inference which knowledge by acquaintance requires. (This was more fully discussed in the preceding subsection.)
No matter where the line is drawn between observables and inferred entities, the most adequate reconstruction, it seems to me, has to be rendered in any case in terms of nomological nets. To return to the temperature example, we may say that the intensity of heat in an oven is indicated by various observable effects, but is not identical with any single one of them, nor is it identifiable with a disjunction (or other logical function) of the observable indications. The intensity of heat is nomologically, and hence synthetically, related to the indications of indicators. This is not to be confused with the quite obviously synthetic character of the functional or statistical relations between the indications themselves. Empiricists, positivists, and operationists have of course always stressed the empirical character of these correlations.
But even when theories (spelling out nomological networks) are adumbrated only in the form of extremely vague "promissory notes," the practice of scientific thinking clearly demonstrates that theoretical concepts (hypothetical entities) are never reducible to, or identifiable with, observable data (or logical constructions thereof). When, e.g., the spirochaete treponema pallida was still undiscovered, the "disease entity" general paresis was conceived as the causative factor which "produces" the various symptoms of that disease. Examples of this sort could be multiplied indefinitely from all the sciences. Theoretical concepts are "anchored" in the observables, but are not logically (explicitly) definable in terms of the observables. To be sure, it is the "congruence," "consilience," "convergence," or whatever one wishes to call the testable correlations between the observables that allows for the introduction of fruitful theoretical concepts. It is indeed this consilience which provides the empirical basis for the specification of the meaning of theoretical concepts. Abstract postulates alone determine only their logical or mathematical structure, but never their empirical significance.
New evidential bases, such as the microscopic bacteriological findings, provide additional, and usually crucially important, "fixes" upon the theoretical concepts. Nevertheless they amount essentially to enrichments of the nomological net, and thus to a revision of the "weights" of the various other indicators. Thus, in present day pathology, the presence of the spirochaete is a criterion of general paresis, and even if many of the usual symptoms were absent, the disease would be ascribed to a patient if a sufficient concentration of the spirochaetes in the nerve tissues were verified. The fact that the bacteriological evidence is correlated with the (more "superficial") symptoms is of course something that only observations could have confirmed. But this need not prevent us from saying that the disease entity general paresis as construed before, or independently of, the evidence for the presence of the spirochaete, can be rightfully identified with the disease characterized with the help of the bacteriological evidence.
I conclude that it is proper to speak of "identification," not only in the purely formal sciences where identity consists in the logical synonymy of two or more expressions, but also in those cases in which the mode of ascertainment is empirical. The important consequence for our problem is then this: Concepts of molar behavior theory like habit strength, expectancy, drive, instinct, memory trace, repression, superego, etc., may yet be identified in a future psychophysiology with specific types of neural-structure-and-process-patterns. The identification, involving as it will, factual discoveries, is empirical in its mode of certification, but it is an identification nonetheless.
Notes 1. What is arbitrary in definitions is usually very uninteresting and inconsequential, in contrast to what is not arbitrary.
2. It was customary to assume that these derivations are deductive. But some of the premises in this case are statistical laws; hence some of the derivations of descriptive-observational or empirical-regularity conclusions are probabilistic. Strict deductions, however, can be found in classical thermodynamics, classical electrodynamics, in the theory of relativity and other examples of "classical" scientific theories. Even in statistical mechanics some derivations are strictly deductive, others so highly probable that for practical purposes they can be considered as ("nearly") deductive.
3. E.g., table salt = NaCl; Units of heredity = Genes; Light = electromagnetic waves; the chemical bond = electromagnetic forces playing between the atoms within a molecule; memory traces = reverberating neural circuits; etc., etc.