Suppose that itâ€™s true that (1) a is F and that (2) b is not F. What would prevent a subject from judging that (3) a and b are distinct? There are several options.
The No-Relevant-Information option: The subject believes neither (1) nor (2)
The Incomplete-Information option: The subject believes either (1) or (2) but not both
The Inferential-Failure option: The subject believes both (1) and (2) but nonetheless fails to infer (3)
Not all of these options are equally appealing.
The Information Deficiency Options
Note that both the No-Relevant-Information option and the Incomplete-Information option are consistent with there having been some prior time in which the subject had the beliefs which, according to the options, the subject currently lacks. So, for example, it would be consistent with the Incomplete-Information option to say of the subject that at time t, the subject perceived and thus believed that (1) a is F; at time t + 1 perceived and thus believed that (2) b is not F; and at time t + 2, when queried about whether (3) a and b are distinct, doesnâ€™t know because the subject has already forgotten either that a is F or that b is not F. It would likewise be consistent with the No-Relevant-Information option that the subject has forgotten at time t + 2 both (1) and (2).
However, insofar as the No-Relevant-Information option and the Incomplete-Information option are cashed out in terms of memory failure, the threat looms that the re-identifiability criterion for concept possession is unsatisfied. Of course, this sort of threat looms only if it is assumed that, for example, time t+1 was the first and only time the subject has a mental representation of b and the representation in question is atomic. If the alleged belief attributed in (2) is a first and fleeting atomic representation of b, then it is a poor candidate for a concept of b. If so-called demonstrative concepts are supposed to be atomic one-shot representations, then the memory-failure versions of the No-Relevant-Information and Incomplete-Information options point out serious problems for the demonstrative-concepts defense of conceptualism.
The threat to conceptualism posed here by memory failure can be headed off, however, if instead of construing the representations of a and b as demonstratives, we construe them instead as descriptions (definite or otherwise). Thus would the representations be non-atomic and even if b is being represented for the first and last time, its representation is composed of parts each of which may have a life history satisfying the demands of re-identifiability.
Another way of cashing out the No-Relevant-Information and Incomplete-Information options would be in terms of a failure of perception instead of memory. So, even though the subject might be presented with stimulatory conditions potentially conducive to perceiving that a is F and b is not F, the subject nonetheless fails to actually perceive that either that a is F or that b is not F. (This may be precisely the sort of thing going on in change-blindness.) Insofar as we can cash out the No-Relevant-Information and Incomplete-Information options in terms of perceptual failures, the kind of worry the non-conceptual content proponents want to raise about the representation of a and b gets blocked. This is because it no longer looks like we have a representation of, e.g., b that fails to be a conceptual representation. Insofar as we are relying on the sort of perceptual failure described above, thereâ€™s no need to attribute a representation of, e.g., b at all.
The Inferential-Failure Option
Turning now to the Inferential-Failure option, the question arises of how there can be such a failure of inference without raising doubts about whether the subject actually believes (1) and (2) in the first place. There are various ways this might get cashed out, none of which make the Inferential-Failure option a particularly plausible model for Kelly cases.
Way One: Either (1) or (2) are believed non-occurently. Iâ€™m sure that itâ€™s common to have beliefs but not draw the simple logical conclusions of those beliefs precisely because the beliefs are not currently contemplated. Problem: Kelly cases involve occurrent mental states, so standing or abeyant beliefs are poor models.
Way Two: Either (1) or (2) are really complicated, like some proposition from string-theory, and thus their deductive consequences are not immediately apparent. Problem: Kelly cases involve experiences of colored paint chips. That ainâ€™t rocket science. Another problem: Why would complication block inference? If itâ€™s due to a load on memory, then this threatens to collapse this response into memory-failure versions of the No-Relevant-Information and Incomplete-Information options.
Way Three: Either (1) or (2) are believed occurrently but non-consciously. Problem: If the topic is non-conscious mental processing, then Iâ€™m not interested in the topic anymore. Iâ€™m interested in versions of Kelly cases that involve conscious experience. Insofar as Kelly cases involve conscious states, then un-conscious states are poor models.
I currently canâ€™t think of any other Ways, so I currently canâ€™t get excited about the Inferential-Failure Option
The Bottom Line
The conceptualist has some promising options for providing intellectual models of Kelly cases. Kelly cases can be modeled either in terms of perceptual failure or memory failure. If they are modeled in terms of memory failure, then a demonstrative-concepts defense is no longer available. But a conceptualist response would be available nonetheless.