Complexity and emergence in societies of agents

Abstract

Throughout the last decades, Darwin’s theory of natural selection has fuelled a vast amount of research in the field of computer science, and more specifically in artificial intelligence. The majority of this work has focussed on artificial selection, rather than on natural selection. In parallel, a growing interest in complexity science brought new modelling paradigms into the scene, with a focus on bottom-up approaches. By combining ideas from complex systems research and artificial life, we present a multi-agent simulation model for open-ended evolution, and a software framework (BitBang) that implements it. We also present a rule list based algorithm implemented for the brain component of the agents. Genetic variation operators were created to drive the evolution of the rule list brains. Several simulation environments were created using the BitBang framework. Experimental results are presented and analysed, validating our model. The results presented show that the model is capable of evolving agents’ controllers in an open-ended evolution simulation. We see that populations evolve sustainable reproduction behaviours, without hard-coding the reproduction conditions into the simulations. By providing evolutionary pressure through the modelling of the environment, we see that on increasingly complex environments, agents evolve increasingly complex behaviours. The rule list brain is shown to provide an important analysis advantage by having readability into the agents’ evolved behaviours. This feature proved to be especially important when unexpected behaviours emerged.