Mechanizing the Common-Sense Inference of Rules which Direct Behaviour

Abstract

We shall present a system that augments its a priori general understanding of human behavior in a spatio-temporal domain of physical activity, goals, competition, and gaming. Observing seemingly novel situations within that domain, it uses common-sense reasoning to construct an interpretation of the observed actions. This understanding is expressed in terms of the goals of the actors executing the actions. The system has the ability to focus its attention, as well as generation, generalize, and verify hypotheses. The desired result is a consistent structure of generalized hypotheses which represent both the regularities and an understanding of the observed domain.