Competence Assessment in Configuration Tasks

Article type: Research Article

Authors: Voß, Angi^; | Karbach, Werner | Drouven, Uwe | Lorek, Darius

Affiliations: German National Research Institute for Computer-Science (GMD), Expert System Research Group, Postfach 1240, D-5205 Sankt Augustin, e-mail: [email protected]; [email protected]

Note: [] This paper has been awarded the DIGITAL AI Research Paper Award for 1990. This prize is awarded by DIGITAL from a list suggested by the Program Committee of ECAI'90

Abstract: Configuration problems are usually overspecified so that the real art lies in finding the right compromises. A competent configuration system can do that. It assesses a problem before jumping into the solution process. It identifies unsolvable subproblems for which compromises must be negotiated and locates critical ones that must be handled with priority. While problem solving it detects the real causes of deadends as early as possible and proposes relaxations for recovery. Our approach to competence is motivated by a configuration system for office room allocation. We characterize some manifestations of competent behavior and illustrate them from our application. We derive a functional architecture with modules for feasibility tests, cost assessment, assessment of missing or unreliable knowledge, problem relaxation, etc. Basic to many of these modules is the ability to abstract a concrete problem, either by dropping requirements, by weakening the conditions of the requirements, or by suitably changing the world. We sketch a problem solving strategy which solves a concrete problem by decomposing it into abstractions and merging their abstract solutions. This document is part of a research project partially funded by the Esprit Basic Research Programme of the Commission of the European Communities as project number 3178. The partners in this project are the University of Amsterdam (NL), the German National Research Institute for Computer-Science GMD (D), the Netherlands Energy Research Foundation ECN (NL), and BSR-Consulting (D)

DOI: 10.3233/AIC-1990-3301

Journal: AI Communications, vol. 3, no. 3, pp. 107-114, 1990