Behaviour: perception, action and intelligence — the view from situated robotics

We relate the problems afflicting implementations of classical knowledge-based symbolic systems to theoretical criticisms of the paradigm, and explain why many of those pursuing research programmes designed to avoid these problems and underpinned by models of mind variously described as ‘behaviour-based’, ‘reactive’, ‘enactive’, ‘situated’, ‘embedded’ are using robot rather than computer systems as their experimental domain. We argue that mentalistic terms are only applicable to contingent historical agents embedded in the local world with which they interact, and therefore (for example) attempting to implement intelligence, semantics, etc., in a computer system is a doomed enterprise.