Reviewed by
Scott E. Page
Center for the Study of Complex Systems, University of Michigan
Complex adaptive systems models have the potential to transform social science. With them scholars can construct and analyze models that include real world features such as networks, adaptation, heterogeneity, and interactions. Advocates of social simulation models claim that these models can provide insights into the likely trajectories of economic, political, and social systems that cannot be achieved with a repertoire of mathematical and verbal models and that they also can be used for high fidelity policy testing. The notion that social simulation models can improve social science should be non controversial, thus, my puzzlement over the current state of affairs. While social simulation has made inroads, particularly in professional schools, it has yet to become a core methodology of social science. Most mathematical and verbal theorists in leading economics, political science, and sociology departments seem to have at most passing interest in computational methods. Few graduate programs offer courses in computational modeling let alone require them.
The lack of penetration from a new cohort and the lack of engagement from the old guard are all the more confusing given the widespread acceptance of computational models in the physical and biological sciences and the fact that social simulation may be the killer app (to borrow a phrase from this book's authors) of computational models.
To rail against the political and structural forces that keep social simulation methods outside the mainstream would be to follow a path to cynicism if not outright disgust. A more productive approach is to devote time and energy to widening and improving the computational road not taken. If that road can be paved with interesting models, large r-squareds, and unexpected policy insights, then traffic can only but increase. This book attempts exactly that - to show the power and potential of social simulation models, to build the road not taken.
The editors aimed to present a flyover of the current state of the art. They deserve credit for an elegant organizational structure and an interesting mix of papers and techniques. They divide the XXIV papers into three parts: model oriented, empirically oriented, and experimentally oriented. The chapters are short. None is the length of a full paper. They're meant more as tapas than as complete meals.
The brevity and diversity of the chapters represent the volume's distinguishing feature. Where else but a social simulation book could you find chapters on altruism in bats (Ch 10), taxation (Ch 2), the theory of structural balance (Ch 9), land use and water management (Chs 7 & 11), emissions trading (Ch 15), the emergence of money (Ch 19), and value chains (Ch 24)? The book has something for everyone, including chapters covering the minority game (Ch 8), a tagging game (Ch 6), and an opinion formation model (Ch 4), and chapters on logical structures (Ch 5) and group formation (Ch 21). Yes, it even has a Prisoners' Dilemma model (Ch 1) as well as something called the Buffalo Game (Ch 20).
Was this the right approach to take or would the movement be better served with fewer, deeper chapters? I think the answer to this question depends on the intended audience. These shorter, more provocative investigations provide proof of concept and hint at potential but they won't convince hardcore, mainstream social scientists to complement mathematics with computation. This approach will, I think, prove enticing for generation next.
Owing to the variety of the chapters and space constraints, I cannot possibly supply coherent reviews of each and every contribution, so I will instead touch on some general themes and hit some highlights. Several themes resonate throughout the book. First, the novelty of social simulation models is on full display. Chapter 22 introduces a model of knowledge production that is both intuitive and unexpected. And Chapter 12 addresses an empirical question - the formation and evolution of social groups - that probably wouldn't even be posed were it not for simulation methods. Not only does the question prove tractable. The authors prove able to fit their model to data.
Second, several chapters reveal the potential for constructing high fidelity models. This is especially true in the applied regional planning chapters, but we get hints of the fineness these models can achieve in the chapters on group formation and financial markets. The inclusion of explicit detail flies in the face of the modeling maxim: keep it simple stupid. Yet, when facts are known in detail, including them may improve a model. Assuming that interactions take place over a social network usually produces different results than assuming interactions take place through random mixing on the head of a pin.
Third, the chapters demonstrate to great effect how social simulation models can inform policy choice. We cannot rely entirely on past data to infer the effects of new policies. Social simulation models enable the construction of empirically rich thought experiments that would be impossible with just pencil and paper. Chapter 18, which explores the efficacy of information campaigns shows with great effect how tightly social simulation models can be linked to real policy choices.
Finally, various chapters aim to show how far social simulations have come. Specific assumptions about learning rules, network structure, payoff functions, etc ... are no longer made in ad hoc fashion but build on existing models and empirical regularities. What could once be dismissed as ill-defined buzzwords have attained full status as academic jargon. Social simulation models have come a long way from the days of "look at the cool patterns that my model generates" not to mention bloviating comments like "notice how the small world network produces an emergent punctuated equilibrium at the edge of chaos". By becoming more scientific, social simulation modelers have begun to force others to pay attention. After all, sometimes a power law really is a power law (but we have to have some way of proving that to make is so).
My appraisal of the book as solid but not spectacular stems from the unevenness of the chapters: some chapters lacked innovation or inspiration, a couple of others were incomplete. And, though far be it from me to decry diversity, the choice of topics might have been better chosen so as to produce a whole that was much more the sum of its parts. That said, few volumes that I know of reveal the breadth and potential of social simulation better than this one.
Return to Contents
of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2008