Abstract

The paradigm of agent based computing is becoming increasingly popular both in distributed artificial intelligence and as a general software engineering technique. The difficulty with agent based computing is that success depends not on the correctness of any one agent, but on the emergent behaviour arising from the interaction of a society of agents. As a consequence, the problem of programming agents is non trivial and poorly understood. In this paper we show that genetic programming can be used to automatically program agents which communicate and interact to solve problems. The programs evolved simulataneously define when and what to communicate, and how to use the communicated information to solve the given problem.

Links and resources

Tags