On the physical limitations of pattern matching

Abstract

The powerful functional capacities (Turing computability) of digital computers are, in part, responsible for fostering the notion that understanding mind is simply a matter of determining the right algorithm—the so-called ‘strong AI’ position that cognition is computation which holds that computers can have mental states and that behavioural equivalence is a sufficient test for the existence of mind (i.e. the Turing Test). John Searle's Chinese Room thought experiment (Searle 1980), however, raises the possibility that pure symbol manipulation does not capture the essence of mind (e.g. intentionaiity) because it lacks certain features—so-called ‘causal properties’. The current status of this debate borders on being a stalemate because, on the one hand, empirical verification based on behaviour is, at the very least, a long way from being decisive while, on the other hand, intuitive arguments about mind are subjective and, as such, untestable. This paper, in contrast to arguing intuitively or relying on behaviour, challenges the computational theory of mind on purely physical grounds, thereby avoiding such an impasse. In particular it is shown that digital computers are physically limited by the very process that underlies their powerful functional capacities, pattern matching. This physical limitation, it is claimed, is fundamental to deciding whether computation is sufficient for understanding mind. Pattern matching is the physical process by which symbols in computers are causal and, therefore, serves as the physical basis for information processing in computers. To demonstrate the physical limitations of pattern matching, a general, causal framework based on how physical change is brought about is introduced, enabling an analysis of the ways in which objects physically embody and transmit information. Physical interactions are causally analysed into ‘informing’ categories which are hypothesized to be Searle's causal properties. It is claimed that differences in these properties explain the cognitively-relevant physical differences between computers and brains