From the earliest days of the computer age, we have been fascinated with the notion that a digital machine could mimic human thought. There have been many diverse efforts to realize this vision. From today’s vantage point, they fall into two broad categories. Some are based initially on introspective analysis of cognitive processes and, more recently, on probabilistic paradigms; these are broadly termed “artificial intelligence” (AI). Others, inspired mostly by lower-level biological processes, have recently been unified under the title of “computational intelligence” (CI). This volume was written by two of the inventors of an important technique in CI, and offers an integrated survey of this latter class of techniques. The book aims to provide a high-level survey of several methods, with historical context, and concrete examples of implementations. It does not give a theoretical development of any of the methods, but features numerous references to other sources that do provide such a foundation.

Eberhart and Shi define CI as self-organization plus adaptation. This emphasis reflects the focus of the field on underlying biological mechanisms rather than high-level cognitive constructs (as in classic AI) or statistical formalism (as in much recent machine learning research). The book recognizes three families of techniques that fall under the umbrella of CI: evolutionary computation, neural networks, and fuzzy logic. The first two clearly fit the model of self-organization plus adaptation. Fuzzy logic does not fit this characterization as well, and, in addition, its roots are closer to the cognitive introspection that gave rise to AI than to the biological mechanisms that drive the other two paradigms. Its inclusion can be justified sociologically: fuzzy researchers frequently associate with workers in evolutionary and neural systems. This interaction may be driven by the smooth transitions between conclusions supported by fuzzy logic that are similar to the adaptive transitions that emerge in evolutionary and neural systems, and contrast with the crisp boundaries imposed by standard logics in conventional AI. Sometimes, a set of rules is the most straightforward way to capture the structure of a problem. In this case, the smooth boundaries of fuzzy logic are much more compatible with the general flavor of CI than the black-and-white results of theorem proving in two-valued logic. In addition, fuzzy membership functions offer a set of parameters that need to be tuned to solve a problem, and evolution and neural networks are natural technologies for this task.

The book has three broad parts. The first two chapters introduce the field of CI by offering the biological foundations of the three paradigms, describing the importance of adaptation and self-organization, and contrasting the CI enterprise with AI.

Chapters 3 to 8 discuss the three component paradigms: evolutionary computation, neural networks, and fuzzy systems. In each case, one chapter traces the history of the paradigm, introduces its concepts, and distinguishes its main subtypes, while a second chapter illustrates the approach with an implementation available on the book’s Web site. Portions of the C and C++ code are included in the book for detailed discussion. The authors include their own technique, particle swarm optimization, in the evolutionary computation group, even though its biological roots are in the ethology of mature organisms and not in ontogeny, as are the other members of this paradigm. One justification for this grouping is that particle swarm is a population-based search mechanism, as the other evolutionary approaches are.

The last three chapters discuss system-level topics. Chapter 9 addresses the notion of a CI system that incorporates more than one of the underlying paradigms. Chapter 10 is an excellent introduction to a range of performance metrics that can be applied not only to CI systems, but to AI systems as well. One weakness of CI systems, particularly neural networks and evolutionary systems, is that their decisions can be hard to explain to users; chapter 11 discusses approaches to explanation.

The book is an outstanding integrated overview of this important set of techniques. Its historical sketches of the development of each of the paradigms aid the reader in seeing how the present array of methods arose. The account of how Minsky and Papert put neural network research into hibernation for more than a decade is a classic of the sociology of science, and perhaps explains why AI and CI are considered distinct disciplines. But the book’s advocacy of integration invites a broader question. If these three paradigms can support one another so successfully, what might be accomplished by bringing CI and AI into an even broader synthesis? There has been some recent research in this direction; perhaps another decade will bring us a volume that expounds this higher-level vision.

This book is ideally suited to be used as a classroom or self-study text; each chapter is accompanied by exercises, and example code is available from a companion Web site, to serve as a basis for student projects. The book includes a detailed index and an integrated bibliography of nearly 300 items, extending to 2006.