Knoesphere: Building Expert Systems With Encyclopedic Knowledge

Abstract

The Knoesphcre project is an atte mpt to build an expert system that is encyclopedic, in the breadth of coverage of its knowledge base, and in the degree of integration of that knowledge. The primary is sue is how to ai d users in searching complex bodies of knowledge. Our approach is to frame the system more as a museum than a set of tomes, and to have the u se r take more or less guided tours of the exhibits therein. The impact of such a system on everyday life -- entertainment and, eventually, education -- is clear. We discu ss its potential for progress in AI as well: a testbed for representation, speech understanding, natural language understanding and generation, automatic story generation and animation, learning, user modelling, and planning. Having an immensely broad and moderately deep knowledge base, the system may also serve as a useful testbed for exploiting analogy and metaphor as a source of power. The work is in its early stages, hence much of what we present is the design for this system, not finished results. We do calculate the magnitude of the tasks involved in such an ambi tious endeavor, and give scenarios of its use.