In practice it is important to provide assurances of performance for a developed system. This can only be done if its quality can be guaranteed. Validity is one fundamental requirement to assure a system's quality.
Considering the problems of Knowledge Acquisition as the first AI-Bottleneck the difficulties arising during the validation process can be considered as a second one in the commercialisation of expert systems (XPS). A complete system validation would imply the examination of all imaginable situations (cases). Such an exhaustive check would be a complete and real, but unfortunately unfeasible proof of a system's quality. Even the use of a quasi-exhaustive set of test cases (QUEST) is impracticable due the huge cardinality of QUEST.
The focus of this paper will be set on the objective ofa criteria-driven reduction of a given set of test cases being a quasi-exhaustive description of the XPS's domain. This reduction has to be carried out properly and without risking a significant decrease in the validity statement's credibility. The result of this reduction will be a non-exhaustive, but necessary, sufficient and reasonable set of test cases (RST) allowing a validation with a "sufficient credibility".
Expert System Validation, Validation Criteria, Test Case Based Validation