Predicting the Behavior of a Knowledge Base

This paper presents the underlying ideas of a knowledge-based system, called Shal, which, given a knowledge base, attempts to capture the key features of its behavior. The idea is that a priori understanding the behavior of a knowledge base can help to achieve the multiple tasks involved in knowledge engineering, in particular interpreting and/or compiling, contradiction discovery, and acquisition of new knowledge. Moreover, the concepts Shal uses, e.g. reasoning about sets, symbolic constraint propagation, and simple theorem proving mechanisms, can in turn be incorporated into the basic language of the knowledge-based system, thus extending its scope. As Shal syntactically is a normal knowledge base, this positive feedback can be injected into the Shal design process itself: as soon as a new concept or reasoning mode is defined, it can be used to improve the behavior of Shal and the acquisition of new knowledge to be embodied into Shal's knowledge base. This is the conventional bootstrapping process, we apply here to knowledge acquisition. This key feature makes Shal an indefinitely extendable system. We already run four successive versions of Shal, each one using the previous one, and we describe here the improvements Shal brought to the initial inference engine.