Abstract
Agent-based modeling shows a fruitful approach to understanding self-organizing behavior in complex social systems. The problems on self-organization have been the primary concern of systems theory. Although there have been a lot of researches studying self-organization, we would need some new way to describe a model for emergent properties and learning process. Agent-based modeling can be expected to provide some novel concept and method on learning process in complex social systems. The essential features of agent-based modeling include a learning process of the rules based on which each agent makes a decision for his next action. Genetic algorithm is a powerful tool to implement the learning process in an evolutionary manner. This paper first provides the meaning and role of system models to describe the learning process in self-organization layer. Then the essential difficulties to develop such system models are discussed. The core step in the learning process in agent-based modeling is to produce a new rule set from an old one in the previous learning phase. This correspondence is very hard to describe as some "function" between the rule sets. We should develop more effective mathematical device and concept to build system models of the learning process.
