How does the behaviour of individuals generate the regularities of a society? This is one of the main questions that this monograph attempts to answer. By using agent-based computer simulation, Joshua Epstein and Robert Axtell try to develop a methodology that will enable social scientists to conduct the kind of experiments that are very difficult in the social sciences. For example, it is difficult to test hypotheses concerning the relationship of individual behaviour to the social patterns of a society as a whole. Futhermore, when highly aggregated models are used to represent social processes, they tend to leave out all consequences of heterogeneous (non-average) behaviour. Social science, especially game theory and general equilibrium theory, has tended to focus on static equilibria and has ignored time dynamics. This is partly because there has been no adequate methodology for studying non-equilibrium dynamics in social systems. The work described in this monograph is a methodology that purports to overcome these problems. The approach departs from traditional disciplines in two ways: first, in the way that specific spheres of social behaviour - such as combat, trade, and cultural transmission - are treated, and second, in the way those spheres are combined.
This monograph of the 2050 Project, which is a collaborative effort of the Brookings Institution, the Santa Fe Institute and the World Resources Institute, comes with the support of some major players in the field of complexity. Its authors' ambitions are high but tempered with realism. They appreciate and admit the limitations of the methodology they develop, but nevertheless offer a major first step in studying social phenomena by simulating the growth of artificial societies.
One of the consequences of the growing interest in the application of complexity to social systems is the proliferation of academic model-builders entering the lucrative world of consultancy. This is particularly evident around Santa Fe, New Mexico, which is the home of the Santa Fe Institute and the United States centre for the study of the sciences of complexity. There is, however, a slight problem. Most are natural or computer scientists with little or no understanding of how human organisations and social systems work.
Epstein and Axtell do seem to understand such systems, and one of their major contributions is the eloquent demonstration of the necessary synthesis of social-science disciplines. They have developed a unified framework for the dynamic interaction of such diverse activities as trade, combat, mating, culture and disease. For instance, they demonstrate how intimately related economic policy and population policies can be. Although such an example supports intuitive expectations, other examples throw up totally counter-intuitive results. At the very least, artificial societies raise important policy questions and may answer some of them. In complex systems, such as human societies and organisations, there may be highly indirect and counter-intuitive ways to induce social outcomes from the bottom up. Combinations of small local reforms, exploiting precisely the non-linear interconnectedness of things - may result in desirable outcomes at the macro level. Complexity encourages this way of thinking. The monograph should therefore be of interest not just to academics working on computer simulation, but to social scientists in general.
From an epistemological standpoint, agent-based social science is neither deductive nor inductive. When building artificial societies the activity and study become "generative" - in the sense that the method aims to provide initial micro-specifications (initial agents, environments and rules) that are sufficient to generate the macro-structures of interest. By growing the artificial society from the bottom up, from simple local rules, the method allows for the emergence of patterns at the societal level. This is in contra-distinction to the more usual approach of taking highly aggregated - top down - mathematical models of national economies, political systems and so on, and "linking" them, thus yielding mega-models that have been criticised and tend to discredit interdisciplinary inquiry as such. Furthermore, some economic models assume homogeneity of agent behaviour, which is unrealistic. By introducing autonomy and heterogeneity of agent behaviour and the non-linearity of multiple, simultaneous interactions, the approach is closer to the real world. In this case it couples demography, economics, cultural change, conflict and public health. These spheres of social life emerge - and merge - naturally and without top-down specification, from the local interactions of the individual agents.
The authors demonstrate another fundamental principle of complex systems: that seemingly "social" behaviour is not driven by any social impulse but is solely a product of the agent-environment coupling. A notable property of social organisations - from ant colonies to the Supreme Court - is that their membership changes while elements of their structure remain constant.
The most significant aspect of agent-environment coupling is co-evolution. The action of agents changes the environment, and in turn the changed environment alters the behaviour of the agents. "Co-evolution with", as distinct from "adaptation to" a changing environment, is one of the key characteristics of complex systems.
The ambitious aim of the authors is to propose "a generative programme for the social sciences and they see the artificial society as its principal scientific instrument". This may or not become the case, but what they have achieved is to develop a method that reflects and is appropriate to the evolutionary process it studies.
Eve Mitleton-Kelly is director, complexity and organisational learning research programme, London School of Economics.
Growing Artificial Societies: Social Science from the Bottom Up
Author - Joshua M. Epstein and Robert Axtell
ISBN - 0 262 05053 6 and 55025 3
Publisher - MIT Press
Price - £28.50 and £13.95
Pages - 208
