On the convergence of autonomous agent communities

Abstract

Community is a common phenomenon in natural ecosystems, human societies as well as artificial multi-agent systems such as those in web and Internet based applications. In many self-organizing systems, communities are formed evolutionarily in a decentralized way through agents' autonomous behavior. This paper systematically investigates the properties of a variety of the self-organizing agent community systems by a formal qualitative approach and a quantitative experimental approach. The qualitative formal study by applying formal specification in SLABS and Scenario Calculus has proven that mature and optimal communities always form and become stable when agents behave based on the collective knowledge of the communities, whereas community formation does not always reach maturity and optimality if agents behave solely based on individual knowledge, and the communities are not always stable even if such a formation is achieved. The quantitative experimental study by simulation has shown that the convergence time of agent communities depends on several parameters of the system in certain complicated patterns, including the number of agents, the number of community organizers, the number of knowledge categories, and the size of the knowledge in each category.