ABSTRACT
Throughout its brief history, the field of artificial intelligence (Ai) has been the arena of jousts between two freres ennemis, symbolicism and connectionism. No sooner had connectionism recovered from Minsky and Papert’s (1988) devastating blows than Fodor and Pylyshyn (1988) charged to the fore in the name of symbolic AI. They argued that connectionism cannot be a valid theory of cognition since it fails to account for the combinatorial syntactic and semantic structure of mental representations; at best, connectionism is just another implementation technology, an alternative means of implementing classical symbolic structures and processes. This implementationalist viewpoint has since been the traditional defense of symbolic Ai against connectionism’s cognitive claims. At the other extreme, according to Pinker and Prince’s (1988) classification, eliminativism rejects the symbol level as a valid level of description of cognitive phenomena; symbolic theories are no more than crude approximations of what really takes place in the brain and must give way to connectionist or neural theories.
