Controlling chaos in distributed computational systems

Autonomous agents in distributed computational systems make decisions based on imperfect and delayed information. These systems are analogous to immune systems, ecologies, market economies and large human organizations. The nonlinear interactions among agents lead to a wide range of system-level dynamical behaviors, including chaos which can reduce overall performance. This paper uses a general dynamical model of autonomous agents to illustrate the stabilizing effect of a simple reward mechanism. The implications of these theoretical results are described for two examples: interacting agents on the World Wide Web and distributed controls for smart matter, i.e., materials with embedded sensors, actuators and controllers.