An evaluation of commercial expert system building tools

摘要

The Force Requirements Expert System (FRESH) is the first phase of the Defense Advanced Research Projects Agency's (DARPA) Navy Battle Management Program, part of the Strategic Computing Initiative. To encourage long term maintainability of the delivered system, DARPA required the use of a commercial Expert System Building Tool. The selection of a tool required the establishment of an evaluation methodology: Analyze the problem domain in terms of AI theory, examine the previous research solutions to the problem and determine their requirements, solve the problem with each tool, compare the results and make the selection. This methodology was followed for three tools that run on Lisp Machines: KEE™ from the IntelliCorp, ART™ from Inference Corporation, and Knowledge Craft™ from Carnegie Group Incorporated. The determining factors in our choice were the size and complexity of the domain and the fact that long term maintenance will be performed by people other than the original knowledge engineers. For these reasons the selection of expert system building tool for FRESH was Knowledge Craft™. This paper presents the application of this evaluation methodology, the results obtained from the tests, and selection decision given the results. Use of this methodology produces comprehensive and functionally accurate information with which to make a selection.