ABSTRACT
The types of representation and the styles of computation in the brain appear to be very different from the symbolic expressions and logical inferences that are used in sentence-logic models of cognition. In this paper we explore the consequences that brain-style processing may have on theories of cognition. Connectionist models are used as examples to illustrate neural representation and computation in the pronouncing of English text and in the extracting of shape parameters from shaded images. Levels of analysis are not independent in connectionist models, and the dependencies between levels provide an opportunity to co-evolve theories at all levels. This is a radical departure from the a priori, introspection-based strategy that has characterized most previous work in epistemology.
