A Tool for Simulated Social Experiments

Abstract

A new agent-based model is presented of the investiga tion of collective behavior with a large number of deci sion makers operating in a stochastic environment. (An agent is an entity that interacts with and contributes to its environment. A set of agents can be a simplified representation of a society.) The model has three distinc tive new features: the number of agents in the model is theoretically unlimited; the agents have various distinct user-defined "personalities;" and the agents are described as combinations of cellular and stochastic learning au tomata. The combination of different personalities with stochastic learning makes it possible to simulate human- like behavior in social situations when each group mem ber must choose between maximizing selfish interests or collective interests. Our model is a framework to per form various simulated social experiments and assess the propagation of information and human influence in large-scale conflicting environments, e.g., to simulate realistic multi-person social dilemmas. We have devel oped a computational tool to implement this model. This is a powerful tool for investigating group dynamics that is also an advance in nonlinear dynamic system simu lation. It may lead to the discovery of a number of fac tors influencing human collective behavior.