Rational Engines for BDI Architectures

A survey of AI systems would reveal that it is somewhat awkward to do acting in reasoning (or logic-based) systems (but it is convenient to talk about representational and reasoning issues using them), and it is awkward to study reasoning and representational issues in systems designed for acting/planning. Thus, most "good" planning/acting systems are "bad" KRR systems and vice versa. For example, in a recent symposium on "Implemented KRR Systems" [Rich 1991b] out of a total of 22 KRR systems presented only 4 systems had capabilities for representation and reasoning about actions/plans (RHET [Allen 1991], CYC [Lenat and Guha 1991], CAKE [Rich 1991a] and SNePS [Shapiro 1991]). The work presented in this paper presents an approach that bridges this "representational/ behavioral gap." We extend the ontology of an intensional KRR system to facilitate representation and reasoning about acts and plans. I.e. a computational cognitive agent modeled using the extended ontology has representations for beliefs, acts, and plans, and is able to reason about them. I.e., the agent is able to represent its beliefs and desires (the ’B’ and the ’D’ of ’BDI’).