The use of decentralised multiagent systems has proved to be a profitable approach for solving a variety of complex dynamic optimisation and coordination and control problems. A common problem, however, in such multiagent systems is that agent strategies which are locally individually optimal result in suboptimal global performance of the system, for example as a result of "selfish routing" problems in data networks. This problem mirrors that seen in some biological systems where "social cheaters" may gain an individual advantage at a cost to the whole system. Here I will present an outline of an upcoming project in which I would like to try to solve the selfish routing problem in multiagent systems through the use of social evolution theory and genetic algorithms. The project should allow one to calculate agent strategies which are efficient for the global, overall system, and which are immune to selfish routing problems.
evolionary MAS, social evolution theory, multilevel selection