Metareasoning and Social Evaluations in Cognitive Agents

Reputation mechanisms have been recognized one of the key technologies when designing multi-agent systems. They are specially relevant in complex open environments, becoming a non-centralized mechanism to control interactions among agents. Cognitive agents tackling such complex societies must use reputation information not only for selecting partners to interact with, but also in metareasoning processes to change reasoning rules. This is the focus of this paper. We argue about the necessity to allow, as a cognitive systems designers, certain degree of freedom in the reasoning rules of the agents. We also describes cognitive approaches of agency that support this idea. Furthermore, taking as a base the computational reputation model Repage, and its integration in a BDI architecture, we use the previous ideas to specify metarules and processes to modify at run-time the reasoning paths of the agent. In concrete we propose a metarule to update the link between Repage and the belief base, and a metarule and a process to update an axiom incorporated in the belief logic of the agent. Regarding this last issue we also provide empirical results that show the evolution of agents that use it.