(Cont'd) Naturalism and Instrumentalism in the Metaphysics of W. V. Quine

Part 2 of 2

Mar 20, 2025

File:Willard Van Orman Quine on Bluenose II in Halifax NS harbor 1980.jpg — Image credit: Mary Claire Quine, via Wikimedia

[This post concludes the essay begun in the previous post.]

III. Resolving the conflict

Assuming that I am correct about the conflict in Quine's work, I propose that this conflict can best be resolved, and the most tenable position achieved, by returning to the instrumentalism of "Two Dogmas". I propose that this be done by discarding the anti-mentalist and extensionalist elements of Quine's theories which generate the conflict.

Quine's anti-mentalism has been effectively attacked by numerous critics (notably by Searle, in Martinich, 476ff). To this criticism I will add that Quine's proposed solution to the problem of induction, which is premised upon his anti-mentalism, has numerous problems.

Firstly, Quine's anti-mentalist account of induction and its relation to habit makes the conditioned responses of a flatworm type-identical to a medical researcher's judgement that cigarettes cause cancer. Given Quine's behaviorist/ verificationist theory of language and meaning, he might be inclined to simply dismiss such an objection, but if he did, it would no longer be clear that Quine was even addressing the problem of induction as it is usually understood. For the problem of induction is usually construed as a problem concerning judgments, not simply behavior, and an account of the reliability of judgments as such cannot ignore intensional considerations, for the truth of a judgement is not identical to the conduciveness to survival of a habitual behavior.

Quine could perhaps escape this objection by claiming that he does not accept ordinary notions about truth as correspondence. But there are more serious problems with his account.

Quine explains induction ultimately in synchronic terms: the success of induction is based on the accordance of the quality spacing with the "functionally relevant groupings of nature". But this sort of accordance is not sufficient for successful induction. The sort of similarity that justifies induction is diachronic: induction is justified when the assumption that the future will be like the past is true. Even if our quality spacing matches nature perfectly at a given time, who is to say that either nature or our quality spacing cannot change? In fact, science indicates that our innate quality spacings do change as we mature: adolescents view members of the opposite sex differently than do children, for example.

In any case, applying a sense of similarity requires drawing a comparison, and comparisons can only be drawn between existents that the one who draws the comparison has had experiential acquaintance of. Since no one not gifted with precognition has had experiential acquaintance of future events, future events cannot be compared to prior events (except by oracles). So the judgement that the future will be like the past, which if true is what renders induction effective, cannot be based solely on an application of the similarity sense. This is of course even more clearly the case if one happens to regard future events as nonexistent.

As for Quine's nominalist/extensionalist account of similarity and kinds, in addition to its conflicts with Quine's other work, the position Quine takes in "Natural Kinds" is fraught with internal conflicts.

Quine begins his examination of similarity by rejecting the definition of similarity in terms of properties, but ends by doing just that. For the explication of water-solubility that was supposed to provide a model for the legitimation of similarity depends on the existence of chemical properties like "being an atom".

It is perhaps no surprise that Quine should find himself forced into a contradiction at the end of his analysis. For a close reading of Quine's essay reveals a persistent tension between two competing and incompatible conceptions of similarity.

First, there is the bottom-up, data-driven, extensional view of similarity revealed in the passages I have already quoted. Empiricists from Francis Bacon on have been most comfortable with this sort of perspective, and this is the view that would best serve the naturalist project.

But, perhaps in spite of himself, Quine also allows a top-down, theory-driven, intensional view of kinds to creep into the discourse, as is revealed in the following passages:

If this definition (of the water-soluble kind) covers just the desired things, it owes its success to a circumstance that could be otherwise. . . . If the trend of events had been otherwise, perhaps the solubility concept would not have been wanted. . . . As between theoretical and intuitive kinds, certainly the theoretical ones are the ones wanted for purposes of defining solubility and other dispositions of scientific concern. (131, emphasis mine)

‘Wanted” implies that there is some desired outcome, prior to the experience in question.

It is also significant that Quine repeatedly mentions trial-and-error (121, 128), as opposed to serendipity or accident. Trial-and-error also implies a desired outcome. Top-down considerations are required in order to make sense of the following passages:

"Cosmically, colors would not qualify as kinds" (127).

"By primitive standards the marsupial mouse is more similar to the ordinary mouse than to the kangaroo; by theoretical standards the reverse is true" (128).

It is easy to see how statements like these might follow from presuppositions regarding, say, the unity of science. But what purely observational considerations would warrant such statements?

The contrast is sharpest when Quine discusses the development of the similarity sense:

By the trial-and-error process of theorizing (man) has regrouped things into new kinds which prove to lend themselves to many inductions better than the old" (128).

This suggests that the process of development is an active process, driven by human purposes. But the following passage suggests the opposite:

. . . one's sense of similarity or one's system of kinds develops and changes and even turns multiple as one matures, making perhaps for increasingly dependable prediction. And at length standards of similarity set in which are geared to theoretical science" (133).

This suggests that the development of the similarity sense is a passive process.

Perhaps I am making too much out of what may be merely stylistic considerations. Perhaps Quine could have phrased his essay in a way that unambiguously endorsed the bottom-up view. But I believe that would have been a mistake, even if it were possible, which I doubt. I base this belief and this doubt on the nature of the relationship between similarity and kind.

Thus far I have not questioned Quine's initial linkage of the notions of similarity and kind. In view of the problems with Quine's account, I believe such questioning is now warranted.

The notion of a kind and the notion of similarity or resemblance seem to be variants or adaptations of a single notion. Similarity is immediately definable in terms of kind; for, things are similar when they are two of a kind (117).

But Quine is unable to come up with a definition of kind in terms of similarity and acknowledges that "the relation between similarity and kind, then, is less clear and neat than could be wished" (121). In light of this I will now present an alternative proposal, one which does not assume anti-mentalism, extensionalism, or nominalism.

I suggest that similarity and kind are indeed related, but they are not the same notion. They "vary together", as Quine puts it, because both are definable in terms of properties (including polyadic properties). But they are not the same, for there are grounds for viewing similarity as an objective empirical relation, in at least some of its applications, but not so for kinds.

Quine tells us that kinds are sets, but does not tell us what distinguishes kinds from other sets. What differentiates kinds is that they are sets of properties--this and nothing more. This means that any conceivable combination of properties can qualify as a kind.

Things are similar to the extent that they have properties in common. Since kinds can be constructed out of any groupings of properties, it will follow that things that are similar will belong to some constructible kind. But this does not imply that sameness of kind is constitutive of similarity.

If there are mind-independent natural properties, and if there are natural objects that do or do not instantiate those properties, then for any two objects, it will be objectively true or false that those objects are similar in some respect, and humans who perceive this are therefore perceiving an objective empirical relation.

But any two objects that have any properties at all can be viewed as belonging to a kind. This much is objectively true, but it is a truth of set theory, that is to say it is a purely formal truth and not a truth of empirical science. Hence when humans view things as belonging to the same kind, they are not simply perceiving an objective empirical relation.

Quine is right when he says that learning requires the ability to draw distinctions. So we must have some innate ability to discern whether two objects or situations are the same or different in some respect. But this notion of respect merits more attention than Quine gives it.

"Some people have thought . . . that (the sense of similarity) depends irreducibly on respects . . . . Still one would like to be able to show that a single general standard of similarity . . . is all we need, and that respects can be abstracted afterward" (122).

Quine supposes that one can see that a yellow cloth is more similar to a yellow ball and a yellow block than to a red ball and a red block without any prior knowledge of respects (Quine does not say whether a yellow cloth is more similar to a yellow ball or a red cloth, though he does allude to Goodman's "difficulty of imperfect community").

But Quine ends up making respects essential to the scientific definition of similarity:

general similarity is as yet undefinable (though it might be definable if all science can be reduced to the science of elementary particles), but chemical similarity can be defined in chemical terms, and similarly for the similarity that is relevant for other "mature" sciences.

The upshot of this is that discriminations of theoretical kinds depend on discriminations of respects. This is as it should be. The notion of respects is a purposive one that unites the formal notion of kinds, the empirical notion of similarity, and the phenomenological notion of comparison. The notion of respects is necessary to account for our ability to spotlight certain properties and apply our similarity sense to those while ignoring all the others. Without some such ability, we would have a single unchanging similarity sense to apply in all situations.

To say that two things are similar in respect to color, or chemical properties, or whatever, is to perform two constructive operations, in addition to the operation of comparison. Firstly, a kind is constructed, singling out some subset of all the properties that animal awareness can be cognizant of. Secondly, that kind is itself singled out as the basis for a comparison. Now, are these three operations data-driven or theory-driven? Active or passive? Innate or acquired? Quine's account conflates the three, and so gives conflicting answers to these questions, though he comes close to grasping the answer when he says, ". . . we retain different similarity standards, different systems of kinds, for use in different contexts" (129).

Having distinguished the three, I will propose the following: the operation of comparison is innate, passive, and data-driven; once a basis for comparison is selected, the degree of similarity is simply perceived in some broad sense. But the constructive operations that make the operation of comparison possible are theory-driven (in the sense that the theoretical constructs of respect and kind constrain the resultant observations and not vice versa, although these constructs may be themselves constrained by the availability of properties to animal awareness), and variable with respect to the innate/acquired and active/passive dimensions.

I say variable because it would appear that certain kinds are recognized by everyone, giving grounds to the speculation that these kinds are innately salient to all or most humans, who are passively driven to construct these kinds and make comparisons on the basis of them.

Quine's evolutionary account is a plausible explanation for this. But then there are kinds that are not universally recognized. These kinds are contingently constructed in response to contingent human projects and purposes. In these cases the constructive operations are active and acquired.

To see how this happens, let us take the shopworn example of the Inuit classification of snow. Why do the Inuit who live in the tundra have multiple concepts corresponding to the city-dweller's single concept of "snow"? Presumably, because in the course of their history the Inuit noticed that variations in the properties of their environment correlated with variations in the success of certain endeavors, such as building an igloo or driving a sled. Through active theorizing (the trial-and-error of Quine's account) they were able to isolate the pertinent kinds and make the relevant comparisons that allowed them to categorize the phenomena into "functionally relevant groupings".

[Aside (2025): This, by the way, is why I never submitted this paper for publication. I wanted to come up with a better example to illustrate this point. Then I started working on other projects and never got back to this one.]

But what is functionally relevant for the inhabitant of the tundra is not necessarily functionally relevant for the resident of Toronto. The process of trial-and-error and the operations involving similarity, kinds, and respects do not reveal the objective "joints of reality", they instead reveal possible solutions to the particular problems of particular people living in particular circumstances. As such they are constrained, but not determined, by what is "really out there". Thus, the counterintuitivity of predicates like "grue" may reflect natural features, but it also reflects the fact that, regardless of whether "grue" is a real predicate, humans have no use for it at present.

In short, even if there is an innate similarity sense, kinds are instrumental groupings, not natural ones. This brings us back to the instrumentalism of "Two Dogmas".

However, one factor that might detract from the tenability of the earlier instrumentalism is the verification theory of meaning. Verificationism is generally considered to be a dead issue (cf. Hempel, in Boyd et al. 1991, and Putnam, 1975, for criticisms of verificationism).

But need instrumentalism be based on verificationism? I do not believe so. I believe we can accept Quine's holism about meaning and instrumentalism regarding theoretical entities without being committed to the reduction of theoretical discourse to phenomenal terms. For it seems to me that an instrumentalist could consistently expand the basis of reduction to include not only phenomenal terms but also purposive terms; that is, the theoretical entities of science are to be understood not only in terms of their relation to human phenomenal experience, but also in terms of their relation to human goals. A moral cognitivist could go even further, and expand the basis to include their relation to human obligations.

I discuss this in somewhat further detail in Robertson, 1996, and there are hints along these lines in Sartre, 1966, but I cannot claim that it has been proven that such a theory could avoid all the problems that have led to the demise of verificationism. But even if it does not, the instrumentalist has the option of simply returning to an old-fashioned, "internalist" theory of meaning that places meanings "in the head". Quine's arguments against such a view presupposed the truth of verificationism, and more recent criticisms of internalist semantics from proponents of the "new theory of reference" have not been successful (cf. Putnam, 1975; I respond to Putnam's attacks in Robertson, 1996).

Such a version of instrumentalism will not be compatible with behaviorism, eliminativism, or any other version of anti-mentalism, nor will it be compatible with the rejection of the first-person point of view, which means that it will not be compatible with certain versions of naturalism.

More seriously, an instrumentalism that allows human goals in its basis of reduction will not be compatible with any version of naturalism. For naturalism is based on the methods of science as science has been understood since the time of Galileo. This means that it does not countenance final causality as an irreducible element of knowledge or ontology (cf. Descartes; also Brenner, 1989).

IV. Conclusion

In the early part of this century, instrumentalists like Dewey and positivists like Ayer considered themselves naturalists. But, probably in response to concerns about verificationism, the more recent work of naturalists such as H. Putnam and R. Boyd has increasingly moved towards realism and away from the problematic idealism of instrumentalism and positivism. The work of W. V. Quine illustrates this trend.

In "Natural Kinds", Quine's ambitious attempt to explain similarity, kinds, induction, dispositions, subjunctive conditionals, and singular causal statements in a single naturalistic synthesis ultimately fails. Even if naturalism is taken for granted, Quine does not address the most important aspects of induction, wrongly identifies the notions of similarity and kind, and explicates the connection between similarity and science in a self-contradictory manner. In addition to these problems, the realism of "Natural Kinds" is not compatible with the instrumentalism of Quine's earlier work, "Two Dogmas of Empiricism".

A non-naturalistic account of similarity need have none of these flaws, and can take proper account of intensional considerations that are essential to a complete understanding of judgments. Such an account will not yield the hoped-for scientific explanation of induction and similarity, nor will it be compatible with naturalism, but this is not a problem for the non-naturalist. Such an account will, however, be consistent with the instrumentalism of "Two Dogmas of Empiricism".

It may be that naturalism and instrumentalism are irreconcilable. If this is the case, then the correct choice is instrumentalism. To put it another way, the proper response to the verificationist failure to unite naturalism and idealism is not to adopt realism, but rather to discard naturalism.

Logos & Liberty

Discussion about this post