According to the author, self-organizing systems can be divided into two general areas: distributed management and control, and self-organizing computation. This book seeks “to present a state-of-the-practice view of self-organizing systems.” The structure of the book reflects the editor’s view. As a result, the book is organized into four parts: Part 1, an introduction; Part 2, on distributed management and control; Part 3, on self organization computations; and Part 4, a concluding chapter that addresses whether algorithmic systems are truly capable of emergent behavior.

Rather than summarizing each of the 15 chapters, I will try to convey the flavor of the material; some chapters will receive more attention than others.

Part 1, an introduction, addresses the basic issue of the balance between design and self organization. In particular, it poses the question: “Can large-scale systems exhibit emergent behavior without a design that primes them for the behavior?” A powerful analogy that motivates several of the example applications is phase transition: small-scale local organization leads to large-scale organization. Part 1 concludes with a chapter laying out a theoretical foundation for the construction of self-organizing systems.

Part 2 contains the chapters on distributed management that describe applications, including self-organizing traffic lights, the health monitoring of space vehicles, decision making in multiagent systems, learning locomotion for a snakebot (showing that side winding emerges as a winning method of locomotion), and decentralized robot control. In all of these applications, it is essential that information is exchanged only with near neighbors. Note that this does not necessarily mean that the agents have no information about the states of distant neighbors, only that such data will have had to travel through intermediaries. Each of these applications shows that distributed control is possible, and describes preconditions under which convergent behavior results. The various applications use differing approaches for processing the data at each agent. It is noteworthy that the traffic light system does not require exchanges of data between neighboring lights--except for those at the same intersection.

The goals of the systems described in Part 3 are rather different from those of the systems in Part 2. Here, rather than an exchange of data for the purposes of management, as an example in the linking of the segments of a snake, the data is exchanged with the purpose of finding an algorithm that will solve some problem of interest. Thus, two chapters are concerned with a relatively recent paradigm for self-organization, based on an analogy with the biological notion of the immune system. Systems based on this analogy learn useful pieces of computation that can be slotted into some larger problem, for example a scheduling system. As much as a human expert might store useful subschedules for future reference, a computer virus detection system stores code fragments that have an affinity for known viruses.

In a different and perhaps more ambitious direction, one chapter describes a system that uses a two-dimensional (2D) or even three-dimensional (3D) array of simple digital devices to construct a machine capable of self configuration, with external control at the edges (on the surface in the 3D case). Such a machine would be capable of implementing a virtual hardware system, which can be reconfigured into machines customized for specific calculations. The geometry of the machine also implies a speed-up of certain kinds of computation (compared to a linear array of devices). The authors of the chapter provide a link where a simulator for the system can be downloaded. Other approaches to self-organized computation include a grid system that can be designed to permit clients to either migrate computations to servers or to migrate themselves, based on information about the current status of the grid; the trade off in this case is that sometimes moving the agent can be cheaper than the cost of sending and receiving messages. Another chapter contains a study of the potential for self-organizing data visualization systems. Here, again, biology provides models: bird flocks and ant foraging, for example. As was the case with Part 2, the applications display a wide variety of techniques.

Each chapter is self contained. Most require little background beyond a modicum of mathematics and straightforward computer science. Because of this, the book provides a good entry-level overview of subjects, with considerable potential for further research. It succeeds admirably in its stated goal: “to present a state-of-the-practice view of self-organizing systems.” I recommend it to anybody interested in learning about this area.