Self learning or how to make a knowledge base curious about itself

Abstract

The inference process in a probabilistic and conditional environment under minimum relative entropy, is briefly repeated following the steps knowledge acquisition, query and response. In general, acquired knowledge suffers from first and second order uncertainty. First order uncertainty is missing information in the knowledge base; second order uncertainty is the vagueness or non-reliability of the system's response to a query. Both, first and second order uncertainty can be reduced by adequate additional information. In the present paper we develop the idea of a self learning knowledge base. Once the system detects a not justifiable vagueness in a recent answer it informs the user about the second order uncertainty and requires additional information in an intelligible syntactical form. This communication reduces both, first and second order uncertainty in general. Suitable examples accompany the theoretical considerations; they are modelled and calculated by means of the expert system shell SPIRIT.