Expert systems for configuration at Digital: XCON and beyond

Reviewer: Jaak Tepandi

The proponents of artificial intelligence often refer to configuration systems at Digital (R1, XCON, XSEL, and others) as the most successful applications of knowledge-based methods. While this is in some sense true—this paper estimates the overall net income due to these systems to be in excess of $40 million per year—it is only a part of the story. In the first part of this paper, Barker and O'Connor take a holistic approach, discussing the following key factors in developing successful expert system applications: strategic/business, technical, and human resources/organizational. The discussion reads well and is well structured, although the layout of headings and subheadings seems insufficient to reflect this structure adequately. The accompanying material by Bachant and Soloway focuses on RIME, a software engineering methodology for expert systems developed at Digital's Configuration Systems Development Group. It discusses new control techniques for XCON (algorithmic control and explicitly choosing among alternatives) as well as guidelines for organizing the rule base. It will be interesting to see whether the implementation of this methodology results in something close to a hybrid of the rule- and object-oriented approaches. As a whole, familiarity with the ideas expressed in this paper is desirable for anybody trying to develop operational, integrated, successful expert system applications.