Browse Books

Reviewer: Liz Sonenberg

The authors have positioned this book as presenting a "new computing paradigm, called autonomy-oriented computing (AOC)." AOC is based on the notions of autonomy and self-organization in entities, and seeks to provide a bottom-up approach to computation that enables the solution of "hard" computational problems and the modeling of complex systems behavior. The work results from research undertaken over the past five years or so, in the first author's research laboratory at Hong Kong Baptist University. The book addresses theoretical and practical issues in AOC, with both methodological discussions and case studies. It is organized into eight chapters, separated in two parts: Part 1 includes basic concepts and an overview of the methodology, and Part 2 contains detailed case studies on how to implement and evaluate AOC. The final chapter discusses lessons learned and future research directions. In AOC, as introduced in Part 1, the basic element is "an autonomous entity." Entities locally determine their behavior-that is, there is no global control mechanism. Interaction between entities and their environment allows evolution toward certain desired states. Self-organization is the key designated computational mechanism. So far, to me, AOC seems consistent with classical distributed artificial intelligence (DAI), and it is not obvious that it can be regarded as a new paradigm. The first three chapters of Part 2 present examples of the application of AOC to a variety of problem types. Chapter 5 demonstrates applicability to n -queens and SAT-both well-known benchmark problems in computational complexity. Extensive simulation results are provided showing performance outcomes comparable to, but not significantly better than, prior distributed techniques. Chapter 6 also presents a body of empirical data, this time from an AOC-inspired model of foraging that is tested against logs of Web-surfing behavior. Chapter 7 introduces a population-based stochastic search algorithm designed to support optimization, and again presents a substantial body of empirical work demonstrating the viability of the approach on standard benchmark problems. Up to this point in the book, it is clear that the authors have a firm command of algorithmic techniques for solving difficult problems. Also, they write well and put the work into a broader algorithmic context (examples are the discussions of distributed constraint satisfaction in chapter 5, and the brief introduction to evolutionary algorithms in chapter 7). Part 2 provides an opening salvo in support of the claim of "a new paradigm." But does it go further than the observation that these classes of problems have interesting distributed solutions__?__ Is AOC a unifying theme, or a new paradigm__?__ In that the authors include some accompanying discussion of the design challenges for the different proposed solutions, the book is a step in both directions. And the discussions in chapter 3 on design and engineering issues raise questions common to a number of approaches about design choices and performance evaluation. But the conventional challenge of creating a generative (as opposed to analytical) methodology to enable the design of emergent behavior still seems out of reach. For me, at least, a convincing case has not been made that AOC is, indeed, a new computing paradigm. The choice of content is somewhat idiosyncratic, rather than comprehensive, because of its origins. Accordingly, a considerable body of what, in my mind, is at least complementary, if not overlapping work in DAI, deriving from the autonomous agents and multiagent systems community, is overlooked. For example, the comparison with agent-oriented programming offered in chapter 1 seems to rely on an influential, but somewhat dated account of the field. This means that, for advanced graduate students, the coverage of the field is narrow. The book contains numerous illustrative examples and experimental case studies, and there is a rich collection of online resources at http://www.comp.hkbu.edu.hk/~jiming, including software (with source code), slides, and links. In this sense, the book has potential as a resource for a graduate course in evolutionary computation. Because the coverage is focused on a particular body of work, supplementing this core with additional material from other perspectives is highly desirable. In summary, as a coherent treatment of a substantial body of work, with implications and challenges for the broader field of emergent behavior, the book contains a fair amount of interesting and thoughtfully presented material, and will be of interest to many. In addition, the care taken by the authors to package the book with commentary on the approach and the place of the work in the field, and the impressive collection of additional resources, is a model that others should be encouraged to follow. Online Computing Reviews Service