Multi-Agent Modeling

Computerized Social Simulation of Agency

The notions of “agent” and “agency” have had a major role in framing research in the social and behavioral sciences. Influenced by computer science (including distributed artificial intelligence, computer networking, etc.), multi-agent modeling or social simulation, that is, simulation of social processes and phenomena on the basis of models of multiple autonomous individual agents, has become a significant aspect of the social sciences.

In this context, agents are computational entities each of which presumably represents an individual person. From their interactions, complex patterns may emerge, leading to various representations of social phenomena. Thus, the interactions among multiple computational agents provide potential explanations for the corresponding social phenomena.

Multi-Agent Social Simulation

Multi-agent social simulation has seen tremendous growth in recent decades. Researchers hoping to go beyond the limitations of traditional approaches to the social sciences have increasingly turned to agent-based social simulation for studying a wide range of theoretical and practical issues.

Traditionally, two approaches dominate the social sciences. One approach centers on the construction of mathematical models of social phenomena, usually expressed as a set of closed-form mathematical equations. Such models may be mathematically elegant but with limited expressive power. Deduction may be used to find the consequences of assumptions. The other approach may be more qualitative and conceptual, with which insights are obtained [Page 636]by generalizations from observations in an informal (nonmathematical) way.

The new approach of multi-agent modeling (social simulation) has emerged relatively recently. It involves computational modeling and simulation of social phenomena based on multiple interacting agents. It starts with a set of detailed assumptions (in the form of rules, mechanisms, or processes) regarding agents and their interactions. Simulations lead to data that can be analyzed to come up with useful generalizations. Thus, simulations are useful as an aid to developing theories, or even as theories themselves.

One of the first uses of multi-agent models in the social sciences was by Robert Axelrod, in his study of evolution of cooperation. In this early work, computational simulations were used to study strategic behavior in the iterated Prisoner's Dilemma game. Even today, this work is still influencing research in various fields. In the mid-1980s, artificial-life modeling emerged, the idea of which was to simulate life to understand the basic principles of life. This led to the application in social simulation of many interesting ideas, such as complexity, evolution, self-organization, and emergence. These ideas influenced social scientists in developing and conducting social simulations. Recently, another topic area appeared dealing with the formation and the dynamics of social networks (i.e., social structures connected through social familiarities ranging from casual acquaintance to close familial bonds).

...