A Delphi Algorithm that Integrates Knowledge for Expert System Development

Abstract

A rapid growth of expert system development in various fields of study will likely occur in this decade. Two prerequisites are needed in order for this to happen: strong social need and technical feasibility. Given that both factors presently exist, a few areas where expert systems can help significantly include: (1) providing an interactively accessible source of updated and well-organized knowledge, and (2) assisting a user in decision making. The current research reviews areas of Artificial Intelligence that relate to the process of knowledge acquisition for expert systems. Until very recently, the primary technique for knowledge acquisition has been the time-consuming process of interviews. Typical techniques include: structured and unstructured interviews, questionnaires, and verbal reporting which incorporates protocol analysis. The functions involved in one or more of the techniques encompass extraction of meaning, data inference, and rule induction coupled with retrospective comment analysis, and behavioral observations. The purpose of the current research is to explore different avenues for data acquisition when dealing with multiple knowledge sources with the objective to develop an automated technique for knowledge acquisition. The Delphi Technique is the primary technique investigated in this study, and the result is the Delphi Manager algorithm which is based on the original version of the Delphi Exercise modified to benefit the expert system development process. Other users of the algorithm include: (1) model verification and validation, (2) forecasting, and (3) opinion polls for policy decision making. Although there are additional uses, the Delphi Manager is primarily formulated for the expert system development process. The Delphi Manager was validated by using an existing knowledge base (KB) that was compiled by a paper and pencil version of the Delphi Technique. This existing KB was part of a dissertation by Randall F. Scott entitled “A Computer Programmer Productivity Prediction Model.” The Delphi Manager has the potential to reduce significantly the time needed to collect and analyze new data. In addition, its user-friendly interface reduces the need for an advanced computer user either to build a questionnaire or to install a help facility. The program provides context sensitive help which is input by the developer through a series of templates. The Delphi Manager is also flexible enough to accommodate anyone from a novice to an advanced programmer. Improvements are suggested that are designed to provide additional program functionality and applications.