Complex Adaptive Systems: An Introduction to Computational Models of Social Life (Princeton Studies in Complexity) [Paperback]

At the time of writing this review, this book isn't searchable through Amazon, that's too bad because if you're reading the reviews wondering if it's worth buying, just browsing through any page from the intro or appendix B would clearly resolve any remnant hesitation. This book is a must have for anyone even remotely interested in complex adaptive systems. Scott Page and John Miller dress the landscape and state of the art of computational social science, the issues are motivated from the ground up and the existing approaches to resolve them explicitly detailed, yet using clear and jargon free language. For example, descriptions of the many concepts repeatedly used in the scientific method (of CAS et al) such as ergodicity or optimization theory are refreshing and insightful, simply stuff you don't get from textbooks, but rather that one would learn over years of experience doing.

In summary, the authors are handing us an expert summary of literature and developments of a complex field in a concise, fun and delightful read, it would be a shame to miss it.

Complexity is a hot subject. Unfortunately, the language of dynamical systems theory is advanced mathematics, which means that most of the available literature is not readily accessible to lay readers. Educated nonspecialists are left with few options aside from the occasional overview which, typically, does not delve too deeply into the subject matter. Given this state of affairs, Miller and Page's book would seem to be a godsend.

A stated aim of the book is that of providing a "clear, comprehensive, and accessible account of complex adaptive social systems" for "both academics and the sophisticated lay reader." Insofar as comprehensiveness, the authors deliver. Readers are first offered preliminary discussions on complexity in social worlds, modeling, and emergence, followed by a more detailed treatment of computational modeling as a tool for theory development and of agent-based objects as the recommended means to explore complex adaptive social systems. Then a basic framework of agent-based systems is presented, followed by discussions of unidimensional complexity models and the edge of chaos, social dynamics, evolving automata, and organizational decision making. These topics are largely illustrated with the authors' previously published models. Finally, conclusions are derived regarding the book's central theme: the "interest in between" as it pertains to complex social systems (which tend to fall in between the usual scientific boundaries). Two appendices bring up the rear: an agenda for future research in complex systems and an outline of best practices for computational modeling. The thematic coverage is ample and varied, excellent for a general introductory work on social complexity.

Insofar as clarity and accessibility are concerned, however, I find myself in disagreement with the book's blurbs. Much of the mathematical formalism has been expunged from the discussions, yes, but that by itself does not guarantee enhanced communicability. The logic of the arguments, which in this field is considerable, must now be conveyed by other means, either verbal or visual. The authors do make an effort to explain in words the basic concepts when they begin a new topic. But when they proceed to discuss an actual model, they shift gears. Instead of explaining or illustrating in detail the model's functional intricacies, they switch to summarizing their findings and present a table or figure that encapsulates the model's results. Repeated readings of the text are almost always required, but understanding does not necessarily ensue. This approach does not appear to contribute to the goal of making the models "as simple and accessible as possible."

This situation is not due to writer's oversight but to a deliberate choice. Prior to discussing their first example model (a computational version of Tiebout's model), the authors state: "Rather than fully pursuing the detailed version of the model we just outlined ... here we provide just an overview." Fateful words which amount to an announcement of their modus operandi, as the subsequent instances demonstrate. Caveat lector. The reader is also assumed to possess a working knowledge of such things as game theory, elementary combinatorics, and statistics, among others. So brush up on the basics and stay close to a search engine.

Reading this book takes time and some effort; it is not a breezy read. One never gets to see an actual piece of code or even pseudocode, which one would normally expect in an introductory book on computational modeling. The reader is left in a vacuum as to the mechanics of implementation. Still, it is a good book in terms of its conceptual content. But the inconsistency between the stated aim of providing clarity of exposition at an introductory level and the actual product the reader interacts with detracts from the book's overall quality. It seems that we are still waiting for the canonical introductory text on complex adaptive social systems.

Note: If you are looking for a general overview of complexity theory intended for a lay audience, I would suggest Melanie Mitchell's Complexity: A Guided Tour. It is excellent. At the other end of the spectrum, if you're heavily into power math, consider Complex and Adaptive Dynamical Systems: A Primer (Springer Complexity) by Claudius Gros. It is rigorous.

Living systems are generally complex, dynamic adaptive systems with emergent properties that analytical models attending only to the local interactions of the system fail to capture. We must complement the standard analytical methods of physics, biology, and economics by additional mathematical tools, such as agent-based simulation and network theory.

A complex system consists of a large population of similar entities (e.g., human individuals) who interact through regularized channels (e.g., networks, markets, social institutions) with significant stochastic elements, without a system of centralized organization and control (i.e., if there is a state, it controls only a fraction of all social interactions, and itself is a complex system). A complex system is adaptive if it evolves through some evolutionary (genetic, cultural, agent-based silicon, or other) process of hereditary reproduction, mutation, and selection.. Characterizing a system as complex adaptive does not explain its operation, and does not solve any problems. However, it suggests that certain modeling tools are likely to be effective that have little use in a non-complex system.

Such novel research tools are needed because a complex adaptive system generally has emergent properties that cannot be analytically derived from its component parts. The stunning success of modern physics and chemistry lies in their ability to avoid or strictly limit emergence. Indeed, the experimental method in natural science is to create highly simplified laboratory conditions, under which modeling becomes analytically tractable. Physics is no more effective than economics or biology in analyzing complex real-world phenomena in situ.. The various branches of engineering (electrical, chemical, mechanical) are effective because they recreate in everyday life artificially controlled, non-complex, non-adaptive, environments that can directly apply the discoveries of physics and chemistry. This option is generally not open to most behavioral scientists, who rarely have the opportunity of ``engineering'' social institutions and cultures.

Miller and Page stress that complex systems cannot be properly modeled using the statistical and mathematical tools associated with differentiable manifolds and normal statistical distributions. Rather, complex phenomena exhibit power law behavior in which statistical distributions have "fat tails" that lead to considerable activity far from the distributions central tendency. A rather stunning example, discussed in Chapter 9, is the size distribution of wars in the world occurring between 1820 and 1943. When the number of deaths in a war (a good measure of the size of the war) is 10 to the power n, the number of wars with this size is about 2 x 3 to the power 7-n.

Miller and Page do a find job of making complexity analysis accessible to the non-expert, without overwhelming the reader with specialized aspects of agent-based modeling or dynamical systems. They provide an exciting stepping-off point for detailed studies in particular disciplines.