Order this book
Volterra Consulting, London
Inherently, the potential readers comprise a highly multi-disciplinary group. As the authors say, the central problems and approaches in fields such as biology, chemistry, computer science and physics make it clear that there is a common set of questions that transcends the usual disciplinary boundaries which is amenable to investigation using common techniques.
There are four main sections in the book, plus two substantial appendices, which are perhaps the most useful and interesting sections of the book for those who already have experience of agent based modelling.
The first section sets out some general themes in complexity in social worlds. They are illustrated very nicely with a simple example of the Tiebout model, already familiar in the social sciences in the context of the allocation of public goods and services. The short second section introduces the idea of models as maps, a concept with which readers of this journal will be familiar, but one which it is important to state.
In the third section, some more substantial general concepts of agent based modelling are discussed. Miller and Page tackle directly some of the key objections which are made to this approach, for example that computations build in their results, that they are hard to test, that they are hard to understand.
The core of the book is the fourth section, which discusses in detail some actual models such as the forest fire model, evolving automata, the Schelling model, a version of the minority game. The authors provide a thorough analysis both of the models themselves and how each of them illustrates aspects of the more general approach of complex systems analysis.
The section is very good. But it would have been very helpful to provide web references to enable readers to access the code of the various models and use them themselves. To someone coming new to the discipline - a curious young mainstream economist for example - it may not be at all obvious how to actually program models of adaptive complex systems. By facilitating access to the use of such models, the authors could have made it easier to gain potential converts across the social sciences. Mainstream economics textbooks, for example, go to enormous effort to provide web-based material linked to the book for both students and teachers. This example could easily have been copied - and copying, as complex systems scholars know, is often a very sensible form of behaviour in such systems.
As the authors point out in the last paragraph of the main text, the purpose of the book is to set out what has been learned about modelling complex systems during the 1995-2005 decade (only three of the book's references are to 2006 papers), rather than straying into what they describe as 'uncharted waters'.
The first appendix is where Miller and Page roam a little more freely and explore in a series of challenging notes many of the interesting issues which are outstanding in complex systems analysis.
The second is a valuable check-list of best-practice methodology. Importantly - and this is a key theme of the book - the first one is 'keep the model simple'. It is scarcely possible to over-emphasise this point, yet it is one which is frequently ignored even by experienced researchers.
As the examples in the book show, and there are many more, complex systems models containing only two or three parameters can nevertheless pose considerable challenges in understanding all their features. In general, such models are containable in that their properties can be explored thoroughly, but the difficulty scales extremely rapidly with each additional parameter. Complex systems modellers often fail to keep in mind the fact that conventional economics has had reasonable success in explaining many aspects of the socio-economic world with models which, the complicated calculus associated with them aside, are often extremely simple. Think, for example, of Solow's model of economic growth. Its ability to explain the world is definitely imperfect, but it is not completely useless and does offer some insight. Yet it is in essence a very simple model indeed.
There are two principal omissions from the book. Rather, they are not omitted but appear only in passing. The first is the importance of networks, and in particular the sensitivity of results to the topology of the network. Empirical research has revealed the existence of three 'stylised' networks in socio-economic situations: random, small world and scale free, and modellers need to explore intensively which is likely to be most relevant to the problem they are addressing.
The second relates to the calibration of models to real-life problems. Many of the examples used are consciously 'toy' models, as it were. The forest fire model is a case in point. Even in its basic form, it is a powerful model with many potential applications. But no example of an attempt to calibrate the model to a real problem is given. Complex systems models have evolved to the point where they can now address specific empirical problems and be calibrated against them, proving themselves scientifically superior to more conventional approaches. The book is too tentative, too hesitant in making such claims. But this presumably reflects the fact that it is essentially covering, as noted above, the work in the field in the late 1990s and early 2000s. These points are not made in a critical spirit, but an encouraging one. Scott and Page document what has been achieved with more abstract models, and the next phase for the complex systems modelling community is to populate the space with scientifically validated models of specific, empirical problems. Overall, the book is excellent.
Return to Contents of this issue
© Copyright Journal of Artificial Societies and Social Simulation, 2008