An intimate affair: Cognition and its physical basis

Functionalism is currently the dominant philosophical theory of mind. According to this theory, mental states are individuated according to the causal role they play in a system of mental processes. As a result, what it is to be a mental state is ‘disconnected’ from any particular kind of physical realization of that state, since the same function may be performed by different kinds of physical mechanism. For this reason, some philosophers argue that data about the human nervous system are irrelevant for understanding cognition. I argue that this view is mistaken: Facts about the physical systems that realize cognitive processes are relevant for understanding cognition.; The most prominent functionalist account of cognition is ‘classicism’. According to this thesis, cognitive processes are computational: Thought consists in the rule-based manipulation of mental representations. If cognitive processes are to be understood independently of the physical systems that implement them, then there must be an appropriate kind of relation holding between those processes and the substrates that realize them, one that guarantees irrelevance. A candidate relation is identified, based on the computational nature of classical systems. However, I argue that it is unlikely that the relation obtains in the case of human neural mechanisms. In the absence of a suitable relation, the likelihood of irrelevance is reduced.; Furthermore, by examining the computational features of classical systems, I show how they place specific constraints on the configuration of those physical mechanisms hypothesized to realize them. I develop a non-classical account based on recent research in artificial neural networks, and show how networks differ from classical systems with respect to those constraints. Finally, I show how networks are not implementations of classical systems. As a result, the choice between these alternatives depends on how mental processes are realized in the physical mechanisms of that system. Thus neurobiological data can be relevant to our understanding of cognition.; In conclusion, the relation of these results to problems of multiple realizability and psychophysical reduction is discussed, as are methods for deciding between classical or non-classical explanations of behavior.