Sharman, Duane2008-05-202008-05-201986-03-01http://hdl.handle.net/1880/46498Commercial exploitation of artificial intelligence has motivated development of expert systems based largely on heuristic principles. Recently, attention has been given to the use of embedded domain models to augment the reasoning abilities of heuristics. One reason for incorporating an explicit domain model is the belief that the human expert's heuristics are founded in a deeper understanding of the domain: that an expert's base of knowledge is more than simply a large collection of empirical observations. Secondly, acquisition of knowledge for heuristic expert systems has been recognized as the bottleneck preventing wider-scale adoption of expert system technology. It is hoped that domain models based on established scientific knowledge can expedite the transfer of knowledge from traditional scientific activities into symbolic form. The third rationale behind the demand for embedded domain models is the apparently arbitrary character of heuristics. The term "deep knowledge" has been coined to describe knowledge that derives from an explicit domain model. This usage parallels the linguistic concepts of "deep" and "shallow" structure, distinguishing the assumed internal representation of utterances and their verbal expression. This paper presents an overview of some recent literature about Deep Knowledge. The subject matter of expert systems is divided into three major area: representation, inference, and knowledge acquisition. After discussing the expected benefits of deep knowledge, a sampling of papers relevant to each area of interest is reviewed.EngComputer Scienceunknown1986-236-1010.11575/PRISM/31278