Complex Adaptive Systems: An Introduction to Computational Models of Social Life

  • John H. Miller &
  • Scott E. Page
Princeton University Press: 2007. 284 pp. $24.95, £14.95 (pbk); $65.00, £38.95 (hbk) 0691127026 | ISBN: 0-691-12702-6

Generative Social Science: Studies in Agent Based Computational Modeling

  • Joshua M. Epstein
Princeton University Press: 2007. 352 pp. $49.50, £29.95 0691125473 | ISBN: 0-691-12547-3

The idea that the social sciences have anything to learn from the physical sciences has raised many hackles. Some social scientists suggest that to use particle-like models of 'agents' that interact via simple rules to explore the emergence of complex collective behaviour is to neglect the sociologist's obligation to explain why individuals behave the way they do. Ironically, this position displays a curious indifference to the 'social' aspects of life. In his recent book The Flight from Reality in the Human Sciences (Princeton Univ. Press, 2005), political scientist Ian Shapiro lamented what might be called the 'physicization' of social science. He claimed that mathematical models that mimic physics fail to engage with the political landscape of the real world and instead disgorge “stylized facts that turn out on close inspection not to bear much relationship to any political reality”.

At face value, Complex Adaptive Systems by John Miller and Scott Page and Generative Social Science by Joshua Epstein seem to encapsulate all that Shapiro deplores, but in reality they are part of the solution, not the problem. Shapiro's complaint hinges on the way social scientists have embraced models taken from economics, which themselves emulate physics. This qualifier is in fact the central issue, which Miller and Page examine in illuminating detail. As Shapiro puts it, economic theory has developed “a perverse sense of rigor, where the dread of being thought insufficiently scientific spawns a fear of not flying among young scholars”. The result is that the models take no account of real human behaviour, which is far too messy to permit any theorems that can be proved rigorously. So economic models become citadels of crystalline mathematical perfection that would shatter if touched by the harsh rays of reality.

Computer simulations probe how the Anasazi culture spread in the American southwest. Credit: I. BLOCK/GETTY IMAGES

It would be grossly unfair to suggest that this describes everything that happens in economics, let alone in all social sciences. But it is widespread, and is reflected in Miller and Page's comment that economists are scandalized to discover how cavalier physicists are in making conjectures that lack any fundamental justification. The irony is that some of the foundational aspects of statistical physics, which provided economists with the early conceptual framework for the neoclassical theory of market equilibrium, remain unproven in any rigorous mathematical sense.

It is absurd that a science as complicated and ill-posed as economics should demand a degree of rigour that not even physics enjoys. That's why these two books are part of an important trend in the social sciences. Both argue for the value of agent-based modelling (ABM) in social science. This approach involves “growing societies from the bottom up”, as Epstein has put it, rather than devising analytically airtight theorems from first principles that are tractable but transparently wrong in what they assume and imply about human behaviour.

The aim of ABM is to study whether the macroscopic patterns or regularities that we observe in society, such as price equilibria or the appearance of behavioural norms, can be generated from decentralized, local interactions between collections of agents.

Epstein's book is a collection of papers that use this approach to explore phenomena as diverse as civil violence, retirement, the emergence of classes and the spread of epidemics. His classic example is the modelling of the demographics of the Anasazi culture of the American southwest between AD 800 and 1300 on the basis of archaeological evidence. Miller and Page, meanwhile, aim to outline a general programme of what ABM is and how it might be conducted. Both books show that computational modelling is slowly beginning to take root in the social sciences. Economics, however, continues largely to resist the idea, as it is incompatible with the standard assumption that the economic system is at equilibrium.

ABM gives access to virtual worlds that rigorous theory cannot touch. In these worlds the actors may differ; they have access to limited, mostly local, information and are limited in their ability to use it; they learn from experience, make mistakes, switch allegiances and copy others. No one should mistake these realms for our own, but they certainly sound closer to it than the neoclassical model of identical rationalizing agents, in which there often seems to be no populations with sizes between two and infinity.

Newcomers to this field might nevertheless find the degree of abstraction unnerving. They might ask, for example, whether Epstein's ring of agents making binary choices based on majority polling of their neighbours, or the forest-fire models presented by Miller and Page, could possibly map onto a real social situation. Aren't these just offering vague metaphors of untested relevance? Indeed, if I have a criticism of Miller and Page, it is that they don't sufficiently address the fearsome question of how such testing might be done. This is discussed in some detail in Scott de Marchi's Computational and Mathematical Modeling in the Social Sciences (Cambridge Univ. Press, 2005). But in any event, that isn't really the point. Shapiro suggests that the role of political theorists is to rove the political landscape “debunking myths and misunderstandings that shape political practice”. Properly applied, ABM might do just that.

The famous segregation model of economist Thomas Schelling, who pioneered the ABM approach in the 1970s, showed that a high degree of social segregation does not, as one might assume, imply extreme intolerance. Conversely, and relevant to today's political climate, it showed that a combination of mobility and choice may amplify marginal preferences or imbalances into major social divisions. Agent-based models may not describe reality, but they can show how interaction and nonlinearity produce social outcomes that could not be predicted simply by inspecting the behavioural rules. They undermine the common political presumption that group behaviour is a multiplied version of individual behaviour. They expose how ideas such as market efficiency may mutate from predictions of simplistic theories into dogmas that are applied insistently to the real economy. They might not tell us why certain social phenomena happen, but they offer mechanisms for how they might.

The challenge, which cannot be over-emphasized, is to ensure that ABM does not get above its station. It is a tool, not just another method for imprinting belief and prejudice with the false authority of 'theory'. As such, these models could form part of a toolbox that helps social scientists to re-engage with reality rather than trying to reinvent it.