Emergent Behaviors in Social Networks of Adaptive Agents

Designing multiagent systems that can exhibit coherent group behavior based on a small number of simple rules is a very challenging problem. The evolution of mobile computing environments has created a need for adaptive, robust systems, whose components should be able to cooperate in order to solve the tasks continuously received from users or other software agents. In this paper, an interaction protocol for a task allocation system is proposed, which can reveal the formation of social networks as an emergent property. The agents can improve their solving ability by learning and can collaborate with their peers to deal with more difficult tasks. The experiments show that the evolution of the social networks is similar under a great variety of settings and depends only on the dynamism of the environment. The average number of connections and resources of the agents follows a power law distribution. Different configurations are studied in order to find the optimal set of parameters that leads to the maximum overall efficiency of the multiagent system.