Cognitive Architecture and the Limits of Interpretationism

To interpret someone's behavior we treat it as evidence of beliefs that take their place as part of a network constructed according to norms of rationality (formalized into principles of procedural rationality: logic probability and decision theory). Call someone who can be understood this way intelligible. It follows that failures of interpretability are the result of inability to find a pattern in behavior that conforms to those norms or departs from them in systematic ways. Such people cannot be credited with beliefs or, because words express beliefs, meaningful utterances. Donald Davidson's version of interpretationism has two crucial aspects

1. only those who can be thus interpreted are intelligible.
2. interpretation is confined to the explanation of belief in terms of other beliefs and the principles of procedural rationality. One cannot advert to cognitive or neurobiological facts to explain beliefs.

This second aspect depends crucially on the idea that there cannot be a cognitive (that is, an information processing) theory of procedural rationality. The codification of those norms by formal theories such as logic, probability and decision theory is an idealization and abstraction from folk psychological principles, but not a theory of human cognition. The reason is that procedural rationality contains no rules for its application in particular contexts (more formally, the problem of finding an algorithm for inductive inference). Procedural rationality cannot tell a subject whether to apply modus tollens or modus ponens to a belief, or whether accept a new belief and revise existing ones accordingly or reject the candidate belief and keep the existing network intact (Davies 1990; Fodor 2000; Gerrans 2001). To take an example familiar from recent literature, Jean the subject of the Cotard delusion believes that she is dead and that she can feel her body. When Jean removes the inconsistency in network of beliefs by adding the additional one that this experience is unique to her, rather than abandoning the belief that she is dead, she has not violated procedural rationality but she has become less intelligible (Young and Leafhead 1996). Because procedural rationality contains no rules for cases such as this, no algorithm can be derived and hence the project of cognitive explanation, which shows how human neurobiology implements algorithms proprietary to aspects of intelligent behavior, cannot proceed.

In their different ways the papers under discussion reject (i), and, I shall suggest, provide good reason to reject (ii) as well. This would lead to the adoption of a version of interpretationism more like that of Daniel Dennett, in which intelligibility is properly regarded not solely as an artifact of interpretation but an emergent property of complex cognitive systems whose systematic behavior can be predicted using the canons of procedural rationality. Dennett and Davidson share the view that interpretability is a property of human agents engaged in purposive behavior (persons), but Davidson's version has the unwelcome consequence that the subpersonal, that is the cognitive or neurobiological aspects of behavior, cannot be invoked to explain the personal.

In contrast, Dennett's version invites the explanation of personal-level behavior in terms of the subpersonal because it is an inevitable consequence of his view that intelligibility is to be explained in terms of the subsystemic properties of the agent, which give rise to the systemic properties that make it interpretable (Dennett 1991a,b). Only this way of explaining beliefs provides the explanatory power we need in the case of apparent irrationality, because what is essentially different about deluded or irrational subjects is their cognitive or neurobiological functioning, not their inscrutability to an interpreter. In effect, Davidsonian interpretationism reverses the direction of explanation here as a consequence of its refusal to allow subpersonal facts to enter into the explanation of beliefs. This is pointed out very nicely in Bayne and Pacherie's (2004) discussion of Campbell's rationalist proposal (Campbell 2001).

At the limit of interpretation we are faced with people who cannot be made intelligible under any interpretative scheme. Alzheimer patients and epileptics undergoing seizure are examples...