CPSC 679 Project -- Knowledge Elicitation Challenges

Development of an Expert Knowledge Base for the Control of an Intelligent Robot Sumo Wrestler

Situation Assessment Module

Course Project for CPSC 679 -- Cognitive Processes in Artificial Intelligence

"Knowledge engineers acquire, store, and process the bivalent rules (of an expert knowledge base) as symbols, not as numerical entities. This often allows knowledge engineers to rapidly acquire structured knowledge from experts and to efficiently process it. But it forces experts to articulate the prepositional rules that approximate their expert behavior, and this they can rarely do." -- Kosko, 1992

"The essence of becoming an athlete is to make more and more of what you know unconscious and therefore out of reach of description, verbalization, explanation." -- Vickers, 1990

In addition to the barrier posed by the unconscious storage of motor skill, terminologies and conceptual systems of the experts rarely coincide with those of the knowledge engineer or even other experts in the same field. But it is these differences in the concept structures that is fundamental to the way that individuals perceive the world and format their experiences as knowledge. It is often the case that expertise is distributed over many experts and only partial answers to the control problem can be derived from a single source. This means that the knowledge elicitation for a rule based control system may need to be from multiple individuals.

In areas such as mathematics or science a highly developed objective language or consensual form of communication is available. This greatly eases the difficulties in multiple expert knowledge elicitation techniques. Yet, even these areas fall prey to the diversity of the concept structures of the experts. In the case of the Robotic Sumo Wrestling application, even the minimal respite of an objective language is not available.