Abstract

Because computationalism and representationalism lay among the foundations of cognitive science, most cognitive scientists are convinced that without internal representations over which to operate, no intelligent system could do what it does. This view has recently come under attack by anticomputationalists of various stripes. Consequently, with the aim of understanding how brains work, my purpose for this dissertation is to determine whether it is appropriate, on computational grounds, to posit internal representations when explaining how intelligent systems work. ;As it is the ontological status of internal representations that is at issue here, I begin by defending the view that something is a representation just in case it is an "entity" in a stands-for relation to something else. In short, every representation is a content-bearer. Adopting the vocabulary found in the literature on intentionality, I draw a distinction between extrinsic and intrinsic representations. Whereas the former bear content just in virtue of our interpretations, the latter do so because they were produced by a mechanistic process of the "right sort," namely, one whose designed or evolved function is to produce content-bearers. The task then becomes determining whether computational systems operate over intrinsic representations. ;After exploring both the nature of computation and the interest-relative way by which something receives a computational interpretation, I argue that there are at least two types of actual computational processing--symbolic-digital processing and nonsymbolic-analog processing. As each of these types of processing corresponds to a different class of computer, the task then becomes exploring the role and status of internal representations in both digital computers and analog computers. Whereas symbolic-digital processing is mediated by internal representations, nonsymbolic-analog processing is not. Or so I argue. ;By focusing on vision, the final task is to show that brains implement nonsymbolic-analog processing. Yet because brains do produce representations , such processing is not always representation-free. Thus, given the analog nature of biological computational processing, mechanistic explanations of how brains work requires only a minimal commitment to internal representations