Validating Expert System Prototypes Using the Turing Testby Gary S. Spring, North Carolina A&T State Univ, Greensboro, United States,
Abstract: This paper describes the Turing Test technique to formalize the expert system validation process using a mature microcomputer based prototype expert system that has been developed to perform accident location analysis. Turing tests are particularly useful as a validation method where it is difficult for the developer to assess output on a case study as correct, or otherwise, or make judgements about how it differs from the human expert's opinion, or whereit is desired to validate the system using multiple experts and there is variation among the performance of the experts. The paper also discussed the relative advantages of this technique over the case study method that has been used elsewhere. The prototype expert system modified based on previous studies and a traffic each analyzed seven high accident locations in the City of Greensboro, North Carolina. Their conclusions were summarized sent to four traffic engineers outside of Greensboro to rate the cases. Their ratings are summarized and a kappa statistics, which was used to assess the consistency of the experts, is described. It was concluded that the prototype is not yet ready for full implementation.
Subject Headings: Expert systems | Validation | Case studies | Traffic accidents | Traffic analysis | Ratings | Computer analysis | Systems management | North Carolina
Services: Buy this book/Buy this article
Return to search