One of the newer trends in building knowledge base systems (KBSs) is maintaining descriptions of solving processes at higher levels of abstraction, nearer to the level at which an expert deals with the domain. An approach that successively refines layers of abstraction, going from a conceptual model, which describes the desired function of the KBS in terms of generic types, to its implementation in terms of rules and objects, often raises obstacles in explaining the behavior of the KBS.
The AIDE shell suggested by Greboval and Kassel not only builds representations for a higher and a lower level of expertise simultaneously, it also automatically translates the higher level into a directly executable lower-level model. Their high-level architecture also distinguishes between domain knowledge (generic and individual concepts, properties, and relations) and the solution method (tools and operators).
Besides providing a fairly detailed overview of their system, Greboval and Kassel describe their experience in using the AIDE shell for the development of SATIN, a system that establishes a first diagnosis for children entering a pediatric service.
In comparing their work to other existing approaches, the authors sometimes blur the difference between mixing the ingredients found in other people’s work and their own contribution. The paper would also have gained from a more rigorous definition of terms and concepts.