CONCEPT
The Rationalistic Tradition (Winograd)
Terry Winograd's name for the dominant assumption in AI—that intelligence is the manipulation of formal representations of the world according to explicit rules—whose failure he diagnosed from inside, and whose ghost returns in every era when machines produce impressively fluent outputs and the temptation rises to mistake the output for the understanding.
The rationalistic tradition is Terry Winograd's term for the picture of mind that organized most of classical artificial intelligence and that he argued, in
Understanding Computers and Cognition (1986), rests on a mistaken
foundation. The picture is simple and intuitive: to know about the world is to have a representation of it—a model, a set of facts and rules, an internal mirror of external reality—and thinking is the computation performed over that representation. Intelligence, on this view, is symbol manipulation according to explicit rules; language understanding is parsing syntax and matching it to a semantic representation; perception is building and maintaining a world-model from sensory input. The approach had prestige, funding, and genuine accomplishments, and its proponents built systems of considerable sophistication within carefully bounded domains. Its failure was specific: it could not handle the vast, messy, statistical character of