Publikationen

Turing Test technologies are promising ways to validate AI systems that may have no alternative way to indicate validity. Unfortunately, human experts are sometimes too expensive to be heavily involved in the Turing Test process. Furthermore, they often have different opinions from each other as well as from themselves over time. One way out of this situation is to employ a Validation Knowledge Base (VKB) that is the collective experience of human expert panels. VKB is constructed and maintained across various validation sessions. Furthermore, Validation Experts Software Agents (VESA) are introduced as a model of the individual experience of a particular human expert, which can serve as his/her substitute in case of the expert's unavailability. A Turing Test experiment with a small prototype system indicates the usefulness of these concepts to model the collective (VKB) and individual (VESA) validation expertise.

Current rule base maintenance is wasting refinement and inference performance. There are only few maintenance concepts, which enjoy both (1) formal rule refinement and (2) utilizing topical knowledge provided by experts within the refinement process. The current state of the art in rule base validation and refinement reveals that there is no generic validation interface and no optimal rule trace refinement. This paper characterizes two different retranslation approaches for reduced rule bases and proposes a two-step validation process, which combines a case-based approach with a rule trace validation approach.