Social Simulation Using a Multi-agent Model Based on Classifier Systems: The Emergence of Vacillating Behaviour in the “El Farol” Bar Problem

In this paper, MAXCS - a Multi-agent system that learns using XCS - is used for social modelling on the “El Farol” Bar problem. A cooperative reward distribution technique is used and compared with the original selfish “El Farol” Bar problem reward distribution technique. When using selfish reward distribution a vacillating agent emerges which, although obtaining no reward itself, enables the other agents to benefit in the best way possible from the system.Experiments with 10 agents and different parameter settings for the problem show that MAXCS is always able to solve it. Furthermore, emergent behaviour can be observed by analysing the actions of the agents and explained by analysing the rules utilised by the agents. The use of a learning classifier system has been essential for the detailed analysis of each agent’s decision, as well as for the detection of the emergent behaviour in the system.The results are divided into three categories: those obtained using cooperative reward, those obtained using selfish reward and those which show emergent behaviour.Analysis of the values of the rules’ performance show that it is the amount of reward received by each XCS combined with its reinforcement mechanism which cause the emergent behaviour.MAXCS has proved to be a good modelling tool for social simulation, both because of its performance and providing the explanation for the actions.