Evolving Multi-agent Networks in Structured Environments

A crucial feature of evolving natural systems is parallelism. The simultaneous and distributed application of rules (governed by e.g. biochemistry) is generally considered as the preposition to build up complex structures. In this article, the potential of agent-based modelling equipped with a concurrent rewriting rule system for artificial evolution is investigated. The task given to the system (pattern construction drawing from a small pool of symbols) is sequential in character, but has to be solved by a strictly parallel rule system. This requires special care in setting up the environment, and it turns out that this is accomplished only by virtue of a hierarchy of levels and modularisation. The proposed three level hierarchy resembles stages of natural evolution in which the emergence of stabilizing mechanisms and cooperative behaviour can be studied. A few preliminary simlation runs are shown and discussed.