Design of rational agents for performing speech acts intentionally in multiagent environment

摘要

This paper presents the basic design for rational agents who intentionally perform speech acts in multiagent environments for the purpose of problem-solving in collaboration with humans or other agents. The authors aim at construction of a rational agent with the BDI (belief, desire, intention) architecture. The rational agent aims at achieving her given goal and selects a plan carefully on the basis of her belief. She then forms the selected plan as her intention, as her mental state to control future deliberation. If the precondition for the intention is satisfied, a speech act composing that intention is intentionally performed. In this paper, the FIPA (Foundation of Intelligent Physical Agents)-conformant primitive communicative acts and protocols are first defined as speech acts of the agent. The extended BDI architecture is proposed in which conditional branching, synchronized waiting for a message, and so on, generated in the communication between agents by using speech acts, are directly represented in the body part of the plan. Then, information filtering by the researcher secretary agent in a laboratory tour guide system is considered, and the usefulness of the proposed method is demonstrated. Lastly, a problem is posed regarding a courteous rejection of a request from the user by the rational agent, which will be an important issue in the future.