The binding roots of symbolic AI : a brief review of the Cyc projectDeniz

Cyc is a monumental but controversial research eeort for codifying the human consensus knowledge initiated by Doug Lenat in 1984. The human consensus knowledge is meant to be the background knowledge that a human is assumed to possess in order to understand, for example, newspaper articles or encyclopedia entries. The methodology chosen is to enter this knowledge explicitly in a large knowledge base. A team of knowledge enterers have been actively engaged in this process since the onset of the project. It is a multi-million dollar, decade-long, two person-century eeort. The engineering goal of Cyc is to overcome the brittleness of conventional application programs by letting them fall back on background knowledge. The scientiic goal of Cyc is to build a system that would exhibit human level common sense and understanding. I believe that Cyc has a respectable goal but inadequate methodology. I further believe that this inadequacy comes from insisting on three limiting principles: (1) representing knowledge explicitly, (2) representing knowledge in a single uniform framework, and (3) insisting on deduction as the main inference engine.