Overview

This chapter looks at a model of information processing very different from the physical symbol system hypothesis. Whereas the physical symbol system hypothesis is derived from the workings of digital computers, this new model of information processing draws on an idealized model of how neurons work. Information processing in artificial neural networks is very different from information processing in physical symbol systems, particularly as envisaged in the language of thought hypothesis. In order to understand what is distinctive about it we will need to go into some detail about how neural networks actually function. I will keep technicality to a minimum, but it may be helpful to begin by turning back to section 3.3, which contains a brief overview of the main features of artificial neural networks. As we work through the much simpler networks discussed in the first few sections of this current chapter, it will be helpful to keep this overview in mind.

The chapter begins in section 8.1 by reviewing some of the motivations for neurally inspired models of information processing. These models fill a crucial gap in the techniques that we have for studying the brain. They help cognitive scientists span the gap between individual neurons (that can be directly studied using a number of specialized techniques such as microelectrode recording) and relatively large-scale brain areas (that can be directly studied using functional neuroimaging, for example).