According to Dennett, the assumption—or ‘‘myth’’—of rationality lies at the core of folk psychology:
However rational we are, it is the myth of our rational agenthood that structures and organizes our attributions of belief and desire to others and that regulates our own deliberations and investigations. We aspire to rationality, and without the myth of our rationality the concepts of belief and desire would be uprooted. Folk psychology, then, is idealized in that it produces its predictions and explanations by calculating in a normative system; it predicts what we will believe, desire, and do, by determining what we ought to believe, desire, and do. (1987: 52)
This points to a strong version of RT, which might be formulated as follows:
(SRT) Attributors always interpret targets, insofar as possible, as con-forming with all normative principles of reasoning and choice.
In his early writing on intentional systems, Dennett applied something like (SRT) to belief and logical deduction. Proceeding on the widely held as-sumption that it is a normative principle of reasoning that you should be-lieve whatever follows by rules of logic from anything else you bebe-lieve, he wrote:
The assumption that something is an intentional system is the assumption that it is rational; that is, one gets nowhere with the assumption that entity x has beliefs p, q, r. . . unless one also supposes that x believes what follows from p, q, r. . . . So whether or not the animal [who is interpreted as an intentional system] is said to believe the truths of logic, it must be supposed to follow the rules of logic.
(1978b: 10–11)
Is it true that whenever attributors assign some belief P to a target, they also assign to her beliefs in all the logical consequences of P? This is palpably false. Attributors clearly allow for the possibility that people forget or ignore many of their prior beliefs and fail to draw all the logical consequences that might be warranted (Stich, 1981). In any case, the logical consequences of any prior set of beliefs are infinite, and it is doubtful that naı¨ve attributors impute infinitely many beliefs to their targets.
The problem isn’t confined to deductive closure; it applies equally to in-consistency avoidance. A putative norm of rationality is to avoid believing all members of any inconsistent set of propositions. The question for (SRT) is whether attributors try to interpret their targets’ beliefs so as to ‘‘protect’’
them from violating this norm. Do attributors really proceed in this protec-tionist spirit? Absent relevant empirical work,2I appeal to a thought exper-iment (the same methodology used by Dennett).
Here is a variant of the paradox of the preface. George publishes a book in which he modestly concedes (asserts) in the preface that the book surely contains some falsehood, although he has been very careful and believes each thing asserted in the text. The resulting set of assertions by George is in-consistent. (If all propositions in the text are true, then the proposition in the preface is false, etc.) If attributors conformed to (SRT), no attributor would attribute to George belief in all these propositions. Instead, an attributor would find some different pattern of beliefs to assign to George. But—here comes the thought experiment—as one everyday interpreter, I would cer-tainly not desist from attributing to George belief in all the asserted propo-sitions. On the contrary, I would assume that George simply didn’t notice the inconsistency and went ahead and believed all the propositions. So the norm of inconsistency avoidance does not guide belief attribution as (SRT) claims (Goldman, 1989).
Responding to worries of this type lodged by Stich (1981), Dennett grants that he has ‘‘hedged and hinted and entertained claims that [he has] later
qualified or retracted’’ (1987: 94). So he proceeds once again to issue some retractions:
[A] few words on what rationality is not. It is not deductive closure.. . . Nor is rationality perfect logical consistency. (1987: 94–95)
If rationality of belief isn’t determined by these logical relations, what does determine it?
The problem is more acute than Dennett realizes. In denying that rationality is governed by deductive closure, he means that rationality doesn’t require agents to infer all logical consequences of their prior belief sets. But what about inferring a single proposition that logically follows from one’s prior belief set? Is that always rationally permissible? Not so. Sometimes it is more rational to retract prior beliefs than to add new consequences of those beliefs.
If you suddenly notice that one logical implication of your prior beliefs is a bizarre proposition, it is more rational to retract some prior beliefs than to accept that bizarre proposition. Unfortunately, unambiguous advice about rational belief formation cannot be obtained from principles of logic (Harman, 1973; Goldman, 1986). But if logic doesn’t generate rules of rational belief, what does? Epistemological theorists, I fear, have no satisfactory answer.
Although epistemologists lack such a theory, ordinary belief attributors might somehow manage to attribute beliefs in accord with the right rules of some still unknown theory of rationality. But what reason is there to suppose this?
Even if epistemology (by my lights) has no satisfactory theory of rationality, plenty of workers in related disciplines think they know what rationality consists in. Many researchers think that Bayesianism supplies principles of rational probabilistic judgment; others think that the transitivity of preference and the sure-thing principle constitute principles of rational preference or choice. Sup-pose some of these theories are right. How does the rationality approach to attitude attribution fare if these are the right principles? Do folk attributors use them in imputing attitudes to others? When it comes to the probability judg-ments, preferences, and choices of naı¨ve agents, a large body of empirical lit-erature suggests that people don’t themselves conform to these precepts. They could, of course, violate these precepts in their own cognitive life but assume that others conform to them. But what reason is there to suppose this? On the con-trary, the empirical literature strongly suggests that naı¨ve agents have no clear grasp, even a tacit grasp, of these principles (Kahneman, Slovic, and Tversky, 1982; Kahneman and Tversky, 2000; Gilovich, Griffin, and Kahneman, 2002). If so, how could it be that they employ them in attributing attitudes to others?
Finally, attributors almost certainly impute to targets some clear cases of irrational belief-forming transitions. Leaping to conclusions and engaging in wishful thinking are foibles most people exemplify at least occasionally.
Presumably, they are examples of epistemic irrationality. Are we to suppose that attributors never appreciate these foibles and never impute beliefs to
others arising from them? That is precisely what (SRT) implies. It says that attributors always interpret targets as conforming to all normative principles of reasoning, so they must not interpret them as making mistakes like leaping to conclusions or engaging in wishful thinking. Surely, such a strong version of the rationality approach isn’t right.
Weaker formulations of the rationality approach are obviously possible.
Dennett sometimes seems attracted to this line.
Of course we don’t all sit in the dark in our studies like mad Leibnizians ra-tionalistically excogitating behavioral predictions from pure, idealized concepts of our neighbors, nor do we derive all our readiness to attribute desires to a careful generation of them from the ultimate goal of survival. We may observe that some folks seem to desire cigarettes, or pain, or notoriety (we observe this by hearing them tell us, seeing what they choose, etc.) and without any conviction that these people, given their circumstances, ought to have these desires, we attribute them anyway. So rationalistic generation of attributions is augmented and even corrected on occasion by empirical generalizations about belief and desire that guide our attributions and are learned more or less inductively.. . . I would insist, however, that all this empirically obtained lore is laid over a fun-damental generative and normative framework that has the features I have de-scribed. (Dennett, 1987: 53–54)
Here Dennett substantially qualifies his rationality theory by introducing an admixture of theory-theory. Suddenly we are told that people don’t make attributions based solely on the ‘‘oughts’’ of rationality theory but (to a large extent) on empirically derived generalizations. Although this may be a sen-sible move toward greater accuracy, it also means the devolution of a once clear theory into a highly inchoate one. How, exactly, are rational oughts mixed with empirical iss? It’s anybody’s guess.
Elsewhere, Dennett proposes yet another variation on RT (1987: 98–101).
First, he says, one may decline to identify rationality with the features of any formal system. But if formal systems don’t disclose the content of rationality, where should we look? He answers:
When considering what we ought to do, our reflections lead us eventually to a consideration of what we in fact do. (1987: 98)
This makes it appear, as Dennett concedes, that he is turning to (what is now called) simulation theory, or something rather similar to it.
Now it will appear that I am backing into. . . the view that when we attribute beliefs and other intentional states to others, we do this by comparing them to ourselves, by projecting ourselves into their states of mind. (1987: 98–99) In terms of our classification of theories, of course, this would be an aban-donment of RT and a substitution of ST instead. Although Dennett doesn’t adopt this maneuver, his constant shifting makes for a blurry target.
Dennett aside, other theorists attracted to RT have suggested different weakenings of the position. Edward Stein (1996: 133–134) proposes weaker formulations along the following lines:
(WRT¢) Attributors interpret targets as always conforming to some nor-mative principles of reasoning.
(WRT¢¢) Attributors interpret targets as sometimes conforming to some normative principles of reasoning.
These proposals follow Christopher Cherniak’s (1986) conception of ‘‘mini-mal rationality.’’ According to Cherniak, a mini‘‘mini-mally rational agent would satisfy the condition that if inconsistencies arise in his belief set, he will sometimes eliminate some of them (Cherniak 1986: 16). Adapting this idea to the practice of attribution, a proponent of RT might say that attributors in-terpret their targets as minimally rational, that is, as sometimes conforming to some normative principles of reasoning. Wouldn’t the plausibility of RT be enhanced if we replaced the claim that attributors impute ideal rationality with the claim that they impute minimal rationality?
Granted, this would improve RT’s plausibility, but at the cost of extreme vagueness or underspecification. The new theory leaves vast areas of inde-terminacy as to how attributors proceed. To say that attributors expect targets sometimes to conform to some normative principles of belief formation leaves it entirely open which beliefs an attributor will attribute on specific occasions.
If a theory of attribution seeks to identify the principles that generate the specific ascriptions that attributors make, this weakened theory hardly gets to first base.