Cognitive Mechanisms and Computational Models: Explanation in Cognitive Neuroscience

摘要

Cognitive Neuroscience seeks to integrate cognitive psychology and neuroscience. I critique existing analyses of this integration project, and offer my own account of how it ought to be understood given the practices of researchers in these fields. ududA recent proposal suggests that integration between cognitive psychology and neuroscience can be achieved `seamlessly' via mechanistic explanation. Cognitive models are elliptical mechanism sketches, according to this proposal. This proposal glosses over several difficulties concerning the practice of cognitive psychology and the nature of cognitive models, however. Although psychology's information-processing models superficially resemble mechanism sketches, they in fact systematically include and exclude different kinds of information. I distinguish two kinds of information-processing model, neither of which specifies the entities and activities characteristic of mechanistic models, even sketchily. Furthermore, theory development in psychology does not involve the filling in of these missing details, but rather refinement of the sorts of models they start out as. I contrast the development of psychology's attention filter models with the development of neurobiology's models of sodium channel filtering.ududI argue that extending the account of mechanisms to include what I define as generic mechanisms provides a more promising route towards integration. Generic mechanisms are the in-the-world counterparts to abstract types. They thus have causal-explanatory powers which are shared by all the tokens that instantiate that type. This not only provides a way for generalizations to factor into mechanistic explanations, which allows for the `upward-looking' explanations needed for integrating cognitive models, but also solves some internal problems in the mechanism literature concerning schemas and explanatory relevance.ududI illustrate how generic mechanisms are discovered and used with examples from computational cognitive neuroscience. I argue that connectionist models can be understood as approximations to generic brain mechanisms, which resolves a longstanding philosophical puzzle as to their role. Furthermore, I argue that understanding scientific models in general in terms of generic mechanisms allows for a unified account of the types of inferences made in modeling and in experiment.