The SNePS BDI architecture

Abstract

Most AI systems are effective either for inference or for acting/planning but not for both. The SNePS BDI architecture uses propositional semantic network representations of beliefs, goals, acts, plans, and a representational concept called transformers. Transformers help capture a unified approach to acting and inference. They can be used to represent reasoning rules, reactive desires, desires for acquiring knowledge, preconditions and effects of actions as well as plan decompositions. A rational engine module operates on these representations and is responsible for the agent's reasoning and acting behavior. SNeRE, the SNePS rational engine, employs a quasi-concurrent message passing scheme, implements the underlying logic as well as the action theory, has forward, backward and bidirectional inference and acting. It also incorporates the notions of truth maintenance and spreading activation.