Agent-based Models in the Social Sciences: Literature Review

The literature is older than I imagined. The first agent-based examination of a social phenomenon, Thomas Schelling’s “Models of Segregation,” was published in 1969. Though the limitations of computing power in that era inhibited the growth of the field, a steady stream of work continued to emerge. The field rose to greater prominence with Robert Axtell’s famous programming contests of the early 80s, which pitted software agents against one another in the classic iterated Prisoner’s Dilemma. This rise in stature was assisted by a publicity-friendly winner: the simple Tit-for-Tat agent, whose complete algorithm consisted of “Cooperate the first round, then choose whichever move the competitor chose the previous round,” defeated competitors of much greater complexity, and was interpreted by some journalists as a guideline to basic morality.

Joshua M. Epstein and Robert Axtell’s 1996 milestone “Growing Artificial Societies” was the first work to attempt a generalized model of society through agent-based modeling. Proceeding from their argument that the social sciences were mired in isolated intellectual silos, each giving lip-service to the notion that no one field could capture the full complexity of human society, and yet unable to find any means to combine theories, Epstein and Axtell argued that Agent-Based Modeling was uniquely positioned to examine these fields in an integrated model. So their “Artificial Society” placed software agents on a field of variable resources; it gave the agents metabolism, vision and mobility, and layered on rules for trade, reproduction with genetic variation, combat, culture, credit and disease transmission. With these tools they were able to examine social phenomena from ‘the ground up’ – identifying where the massively parallel interactions of limited agents matched, or did not match, the relevant social theory. For example, it was observed that commodities traded in the artificial society never reached an equilibrium, ‘market-clearing’ price, though it came closest when agents had fixed preferences, were immortal, and traded for a long time – in other words, when they most resembled the homo economicusof classical economic theory.

In fact, “Agent-Based Models in the Social Sciences” is not a field; there are no strict boundaries to delineate which models or techniques are useful to the social sciences and which are not. Epstein and Axtell themselves acknowledge the influence on their work of numerous related fields – cybernetics, distributed artificial intelligence, cellular automata, genetic algorithms, genetic programming, artificial life, and individual-based modeling in biology. All of these studies can really be subsumed into the overarching category of complex systems analysis. This makes it difficult to identify the ‘key literature’ for the social sciences in particular.

OVERVIEWS