Connected replicator dynamics and their control in a learning multi-agent system

This paper analyzes complex behaviors of a multi-agent system, which consists of interacting agents with evolutionally learning capabilities. The interaction and learning of agents are modeled using Connected Replicator Dynamics expanded from the evolutional game theory. The dynamic systems show various behavioral and decision changes the including bifurcation of chaos in physics. The main contributions of this paper are as follows: (1) In the multi-agent system, the emergence of chaotic behaviors is general and essential, although each agent does not have chaotic properties; (2) However, simple controlling agent with the KISS (Keep-It-Simple-Stupid) principle or a sheepdog agent domesticates the complex behavior.