Afzal Upal's Social Simulation Research

Evolutionary Social Simulation

Project Summary
- This project involved using the architectures designed by the agent-based research community to create simulations that model aspects of animal and human societies and using these models to answer questions about evolution and spread of various traits among these organisms. Traditionally, evolutionary social scientists have relied on static game-theoretic /general equilibrium-theoretic models of social interactions forcing them to make unrealistic assumptions such as homogeneity among agents, and limiting their analysis to small populations consisting of a few agents. As economist Scott Moss recently lamented, "in more than half a century since the publication of von Neumann-Morgenstern (194x), no significant progress has been made in the development of models that capture the process of interaction among more than two or three agents." He goes on to propose agent-based simulation is such a model. I believe that he is right and that agent-based simulation has the potential to have as much impact on social sciences as computer aided design has had on engineering.
  Recently, I created a simulation to test the hypothesis posed by some evolutionary psychologists that various male traits such as co-operating with females are enhanced by a preference by females to mate with those males who possess that trait . The society consisted of 100 agents. Each agent had two main behaviors; playing Prisoner's Dilemma game, and mating with a player of the opposite gender to produce offspring . Each agent was modeled as a Java object with the attributes of gender, reproductive cost, and four 21-bit binary strings encoding the game playing strategy of the player. The results show that even in such a simple environment there are a large number of interacting variables, which complicate the relationship between the sexual selection of co-operative males by females and the proportion of males actually co-operating with females. In fact, in most situations I modeled, sexual selection of co-operative males by females ended up causing the proportion of females that co-operate with males to increase while the proportion of males co-operating with females showed no significant increase over the random selection experiments. I am currently working on extending these experiments to study if these results generalize to other male traits and fully understand the implications of all this for the sexual selection theory.
  I am also interested in extending the existing multiagent design and simulation architectures to enable simulation researchers to better model evolution of traits in animal and human societies. The factors hindering more widespread acceptance of simulation as a valid methodology by the social scientists include (a) the overly simplistic model of agents as objects with a few attributes, and (b) the assumption that only one agent can act at any given instant of time. I would like to design tools that allow their users to build multiple agents that can act at the same time. I would also like to use the artificial intelligence knowledge representation and problem solving techniques to design populations of knowledge-based, problem solving, and interacting agents. Studying populations of such agents can tell us more about how information gets created, transmitted, absorbed, and transformed in human societies. I believe that design and use of such architectures can lead to fundamental advances in both multiagent systems research as well as evolutionary social sciences.
Further Reading
- M. A. Upal. Sexual Selection of Co-operation presented at the 2nd Lake Arrowhead Conference on Agent-based Simulation in Social Sciences held at the UCLA Lake Arrowhead Conference Center May 9-12, 2002
- M. A. Upal. Sexual Selection for Co-operation, in Proceedings of the International Conference on Genetic and Evolutionary Computation (GECCO), pages 98-109, Springer Verlag, New York, 2003.