Genetic Action Trees A New Concept for Social and Economic Simulation

Multi-Agent Based Simulation is a branch of Distributed Artificial Intelligence that builds the base for computer simulations which connect the micro and macro level of social and economic scenarios. This paper presents a new method of modelling the formation and change of patterns of action in social systems with the help of Multi-Agent Simulations. The approach is based on two scientific concepts: Genetic Algorithms [Goldberg 1989, Holland 1975] and the theory of Action Trees [Goldman 1971]. Genetic Algorithms were developed following the biological mechanisms of evolution. Action Trees are used in analytic philosophy for the structural description of actions. The theory of Action Trees makes use of the observation of linguistic analysis that through the preposition by a semi-order is induced on a set of actions. Through the application of Genetic Algorithms on the attributes of the actions of an Action Tree an intuitively simple algorithm can be developed with which one can describe the learning behaviour of agents and the changes in action spaces. Using the extremely simplified economic action space, in this paper called “SMALLWORLDâ€, it is shown with the aid of this method how simulated agents react to the qualities and changes of their environment. Thus, one manages to endogenously evoke intuitively comprehensible changes in the agents‘ actions. This way, one can observe in these simulations that the agents move from a barter to a monetary economy because of the higher effectiveness or that they change their behaviour towards actions of fraud.