Browse Books

Reviewer: H. Van Dyke Parunak

The power and rapid progress of agent technology are attracting the attention of many new users and generating the need for surveys that can efficiently introduce them to a field whose roots go back several decades. These three collections of papers by leaders in the agent R&D community provide accessible introductions to the technology for newcomers, and will interest experienced users who wish to review applications in domains other than their own. Each volume provides an overview of agent technologies and of the papers included in the book, written by the editors. The books differ in length and originality, and focus on different technical levels and application domains. Huhns and Singh is the longest of the three volumes, comprising 52 papers (including the editors' introduction). It is a compendium of previously published papers, which were not re-typeset, but have been reproduced photographically on 8.5-by-11-inch pages (papers originally published in a smaller format appear two-up, requiring the reader to hold the book sideways). The papers originally appeared between 1991 and 1997, inclusive. Thus the volume is a logical successor to Gasser and Bond's collection Readings in distributed artificial intelligence [1], with no overlap in the contents. Jennings and Wooldridge, and Bradshaw, are much shorter, offering 16 and 19 papers, respectively. Six of Bradshaw's chapters are reprints in whole or in part from earlier publications (including four papers from the July 1994 issue of Communications of the ACM ), but have been reset to present a uniform appearance with the original contributions. In addition, James White's chapter on “Mobile Agents” presents an overview of Telescript that is widely available in other collections. None of Jennings and Wooldridge's papers is identified as a reprint, although several concern projects that have been discussed elsewhere. Each volume groups its papers into several parts or chapters. Similarities and differences among these groupings reflect the editors' different technical and application interests. Each volume has an introductory chapter by the editors, introducing the subject and setting the various contributions in perspective. The introductory chapter in Huhns and Singh, “Agents and Multiagent Systems: Themes, Approaches, and Challenges,” offers a detailed taxonomy of agents; a brief survey of applications; and a review of technology under the headings of architectures and infrastructure, models of agency, and future directions and challenges. Bradshaw's introductory chapter, “An Introduction to Software Agents,” summarizes several taxonomies of agents by different authors and focuses much more attention than do Huhns and Singh on motives for deploying agents. Jennings and Wooldridge's introduction, true to its title of “Applications of Intelligent Agents,” focuses on application issues. Its taxonomy is of application domains rather than of agents in the abstract, and it discusses “the agent development bottleneck” and reviews emerging approaches to the hands-on development of agents. Jennings and Wooldridge supplement their own introduction with a survey of agent technology, by Nwana and Ndumu, and a review of different application perspectives, by Laufmann. After the introductory sections, the books by Bradshaw and by Jennings and Wooldridge offer a group of papers describing the setting of agents in the real world. Bradshaw's section focuses on the relative roles of people and agents, how they interact, how agent design should take the human user into account, and what sorts of interfaces to agents are most useful. To help readers reach a balanced assessment of claims for agents, Bradshaw includes a chapter by Ben Shneiderman, a noted critic of agent technologies, arguing that appropriate direct manipulation mechanisms are superior to interface agents, and urging empirical testing to settle the question. In Jennings and Wooldridge, this group of papers has a decidedly commercial orientation, with one chapter providing an overview of vendors of agent technology, and another reviewing various business models appropriate to doing business over the Internet. Two other papers in this part of Jennings and Wooldridge discuss the business case for constructing agent models (misleadingly titled “Practical Design of Intelligent Agent Systems”) and the personal agent paradigm. All three volumes present papers on application systems that embody agents. Bradshaw's six application papers focus on agents that help people learn, access information, and communicate with computers. The nine applications reported in Jennings and Wooldridge include some information management systems similar to those in Bradshaw, dealing with workflow management (one paper) and financial systems (three papers). In addition, they offer examples of control applications such as manufacturing, intelligent highway systems, and air traffic management. The 18 application papers in Huhns and Singh cover the broadest scope of problems, including enterprise integration, information access over the Internet, personal assistants, collaboration support, scheduling, and pedagogy. Jennings and Wooldridge do not devote a section specifically to agent technology, although one of their introductory chapters does provide an overview, and two of their application papers (those by Georgeff and Rao and by Haugeneder and Steiner) expound in some detail the abstract ideas and architectures that support the applications they discuss. Bradshaw follows the application papers with seven that focus more specifically on aspects of agent technology, including Shoham's agent-oriented programming paradigm, a report by Finin and colleagues on KQML (and an important critique by Cohen), Bradshaw's own KaOS architecture, and White on the Telescript model of open agents. Technology papers are the major focus of Huhns and Singh, with 14 papers on architectures and infrastructure and 19 on various models of agency. Thus, the volumes can be arrayed along a spectrum from a technology focus (Huhns and Singh, with no explicit discussion of the setting for agents or of market issues), through Bradshaw, to Jennings and Wooldridge, with the most thorough discussion of the business justification for agents, but less exposition of fundamental technologies. All three books include papers on applications. Bradshaw focuses on human interface and information retrieval agents, while Huhns and Singh, and Jennings and Wooldridge, offer a wider array of application domains, including control applications in heavy industry as well as purely digital applications. Business managers seeking to understand the commercial impact of agents in a wide variety of fields will find Jennings and Wooldridge most accessible. Strategists and developers in information systems will appreciate Bradshaw's focus on their problem domain. Huhns and Singh will appeal to two groups that will welcome the availability of a range of classical papers between one set of covers: established agent practitioners seeking to clean out their files, and professors who would otherwise have to assemble many of these papers in course packs for their students. All three volumes are well produced. In all three, references appear at the end of individual papers, and there is no cumulative bibliography. Huhns and Singh, and Bradshaw, but not Jennings and Wooldridge, offer cumulative indexes, each averaging about one entry per page of text.