From beliefs and goals to intentions and actions: an amalgamated model of inference and acting
Deepak Kumar · 1993
The behavior of computational rational agents should be driven by their beliefs, desires, and intentions. A survey of AI systems reveals that it is somewhat awkward to do planning and acting in systems designed for reasoning about beliefs and that it is similarly awkward to study representational and reasoning issues in planning/acting systems. This is an investigation of the relationships among beliefs, plans, effective acts, and between reasoning and acting, in the context of a rational cognitive agent. It turns out that one can establish a closer relationship between inference and acting if one views inference as a kind of mental acting; i.e., a deduction rule is a propositional connective that, when applied, constitutes the mental act of believing the rule's consequents. This suggests that one only needs a single module for reasoning and acting. Once integrated in such a way, it is also possible to visualize connectives that transform beliefs into an intention to do something in reaction to some new beliefs or in the quest to answer some query. These ideas have been formalized by defining the notion of a transformer and results in a model of a cognitive agent that is able to achieve the posted goals in a unified fashion. The investigation leads to the design of an architecture called the Object-based Knowledge Belief-Desire-Intention (OK BDI) architecture. It is comprised of a representational formalism called the OK Formalism and a module called the OK Rational Engine. The OK Formalism is designed using object-oriented principles. As an implementation, we have extended the SNePS semantic network processing system into what we call the SNePS BDI architecture, which includes representations for various transformers and a rational engine, called SNePS Rational Engine (SNeRE), that is responsible for integrated reasoning and acting.