In class we alluded to the fact that a computer simulation is one method to show that the results of Network Exchange Theory could be replicated from probabilistic behavior rules enacted by computer programs. Not surprisingly there this method of computer simulation can be extended to explore a variety of other social phenomena. In fact, there is an emerging field of multi-agent simulation that is allowing for interdisciplinary work between social, physical, and computer scientists to exploit the growing speed of modern computers in the pursuit of understanding through the modeling of artificially connected agents (people, animals, physical particles) and their dynamical interactions.
The basic idea behind agent based simulation is using computer code to represent the simple behavioral rules of social agents. Then exploiting the significant availability of computing power, these computer “agents” are allowed to “interact” in modeler specified ways. The goal is then to to obtain a visual or statistical representation of how these interactions between agents evolve over time and to examine the macro-scale trends that can be built-up from the micro-level interactions. These notions are extended directly from those expressed by Thomas Schelling in “Micromotives and Marcobehaviors.” One benefit of this type of modeling is that it allows for the modeler to experiment with a large number of agents and to collect data on their behaviors on a much smaller time scale than any social experiment involving humans would. Also, unlike classical economic and sociological experimentation, this form of study allows for a focal shift from the equilibrium setting to a more dynamic point of view. With the ability to perform literally hundreds of agent interactions in a fraction of a second, the total number can be huge over any extended simulation run. With such a large data set and a short run time, the modeler can extract information noting the changing behavioral norms as an experiment approaches its equilibrium.
This method of social experimentation has already provided some interesting very results. Many examples of such are presented by on of the fields pioneers Dr. Robert Axtell; a few include models that attempt to recreate an ancient civilization, explore the formation of agent coalitions based on game theoretic agent interactions, and test the evolution of behaviors when agents interact according to the Nash Bargaining game. The Nash Bargaining model, in particular, relates to our discussion of Network Exchange theory; here agents are coded to negotiate the division of a good according to the payoffs listed in a 3×3 matrix and allowing for adaptive behavior by giving each player a memory for their previous interactions. The ease with which heterogeneity is built into each agent makes each simulation run slightly different, but the hope is to identify some long term social equilibriums and to note the different dynamics by which these rest states are achieved. Analogous to our model in class, this computer model has one results in the emergence of social classes based on an unequal (but still efficient) distribution of property. Another interesting twist to this model is the importance of social “tags” that help identify the various type of agents playing the bargaining game. The role of the tags here is that it aids in the adaptive measures that an agent uses by recalling the previous plays of other similarly tagged agents. (the summary above is taken from an article “The Emergence of Classes in a Multi-Agent Bargaining Model” by Robert Axtell, Joshua Epstein and H. Peyton Young. Unfortunately I was not able to find a link for this paper, and used a paper copy received from Dr. Axtell as my source.)
The following link takes you to the book by Axtell and Epstein describing their modeling efforts in the creation of artificial agent civilizations:
The overall relevance of this post to the course is the usefulness of game theoretic ideas used to interact agents that may live on some (social or physical) network. The key feature of this method is that it allows for repeated play between agents (not necessarily the same pair of agents, however) and therefore for an evolutionary perspective on the dynamics by which marco-level equilibriums are achieved through micro-level behaviors. This notion makes available the idea of agents living on a grid (in which case they interact with their physical neighbors) or a more abstract social network (where interactions are only possible between socially connected individuals). Here the variety of topologies of the networks and the rules of the games that the agents engage in make it possible to model a large variety of social interactions and hopefully some very useful results.
* You can follow any responses to this entry through the RSS 2.0 feed.