Integration of Knowledge and Neural Heuristics

In his keynote speech, B. Chandrasekaran (The Ohio State University) argued that the debate about the right approach to AI could be clarified by removing many confusing notions with regard to what made something a representation. In one case, content is more important than form, whereas in the other, the reverse is true. He pointed out that proper understanding of Alan Newell’s knowledge level versus symbol level distinction could illuminate many phenomena related to representation. As to the issue of integration, the first question is always, “Integrate what and what?” Many integration alternatives could exist, and not all of them make sense in a given context. After all, one cannot be so naive as to overlook the potential that a hybrid inherits the weaknesses, rather than the strengths, of its parents. The integration or synergism of knowledge-based components and neural networks in a system can be explored from their functional and structural relationships in the system. Five integration architectures can be identified: First is completely overlapped: In this architecture, the system is both a knowledge-based system and a ■ This article discusses the First International Symposium on Integrating Knowledge and Neural Heuristics, held on 9 to 10 May 1994 in Pensacola, Florida. The highlights of the event are summarized, organized according to the five areas of concentration at the conference: (1) integration methodologies; (2) language, psychology, and cognitive science; (3) fuzzy logic; (4) learning; and (5) applications.