The SPHINX system environment

Abstract

When an expert system is conceived two developmental problems immediately present: one involves development of knowledge which should come about without difficulty; the second crops up because of problems posed by validation of the knowledge base. It is therefore extremely important to concentrate on a program which, besides an extensive capacity for modifying the data base, provides help to the expert to avoid errors. This second problem may for that matter be considered as an extension of the first; in other words, to enable the storage of expert knowledge as correctly and completely as possible. The sheer mass of data is so great that it is very difficult, even for an expert, to express it completely without errors and contradictions. This is the reason why many systems are committed to programs called ‘knowledge supervisors’, which look for inconsistencies in the data, detect circular arguments, which put reverse chaining control procedures in check, list all or some of the knowledge according to the user-defined criteria, map out the ways of goal attainment, and which structure and organize in an internal coding a bulk knowledge as Teiresias has done (Davis, 1980a).