Toward an AgentSpeak(L) Theory of Commitment and Intentional Learning

This work is about the commitment strategies used by rational agents programmed in AgentSpeak(L) and the relationship between single-minded commitment and intentional learning. Although agent oriented languages were proposed to reduce the gap between theory and practice of Multi-Agent Systems, it has been difficult to prove BDI properties of the agents programmed in such languages. For this reason, we introduce some ideas to reason temporally about the intentional state of rational agents in order to prove what kind of commitment strategy is used by AgentSpeak(L) agents, based on the operational semantics of the programming language. This enables us to prove that any agent programmed in this language follows by default a limited form of single-minded commitment. Then we analyze how intentional learning can enhance this commitment strategy allowing preemptive abandon of intentions.