The intended audience of this book consists of managers who are inexperienced with AI but have the job of directing an AI project. The book is also good for the AI person who is not a manager but finds he or she must manage an AI project. It is well written; each chapter begins with a good summary and ends with a useful summing-up section.

The book is divided into three parts: “Concepts,” “Management,” and “Development Model/Risks and Safety Nets.” I recommend that Part 2 be read last, however. Parts 1 and 3 fit together well. After they are digested, then the question is how best to manage the enterprise.

Part 1, “Concepts,” covers many of the ideas and issues of AI. Chapter 1, “What Knowledge-based Systems Do and How They Work,” gives the reader an idea of the breadth of AI, the categories of expert systems, the difference between knowledge and data, and how AI differs from conventional software and databases. Chapter 2, “Human Knowledge,” relates the way human beings think to the way machines reason. This material is weak and overly general. Different kinds of knowledge representation are described, but the chapter falls short after the section on rule-based methods. Chapter 3, “Conventional vs. Knowledge-based Application Development,” does a good job of contrasting the two software fields in terms of tools used, hardware, and levels of complexity of expert systems.

Part 2, “Management,” espouses the management philosophy of Richard H. Thayer [1]. This whole section is intentionally generic, applying to software engineering as a whole. Chapter 4, “Developing the Plan,” covers just that. The authors relate the topic to some AI issues, such as the availability of the expert consultant whose knowledge is to be extracted. They then discuss project requirements, scoping the job, and writing the design specification. The authors list good questions that the manager should ask and provide tables of general technical and management issues. Chapter 5, “Organizing the Application,” is concerned with organization of the project, personnel qualifications, and responsibilities. The knowledge engineer is defined especially well. Chapter 6, “Staffing the Team,” continues chapter 5, discussing the need for good communication and ways of achieving it. It includes more information on personnel qualifications and hiring. Team productivity is also a concern. Chapter 7, “Directing the Team,” has more on communication issues, technology transfer (from the expert to the software), motivating commitment, and productivity--these are good topics anywhere. Chapter 8, “Controlling the Effort,” discusses accountability, project tracking, situational analysis, and contingency planning. Reporting forms and worksheets are shown and described. The authors emphasize the need for project standards, procedures, and quality assurance.

Part 3, “Development Model/Risks and Safety Nets,” describes the development phases of an AI project, points out many pitfalls, and gives suggestions about how to avoid them. Chapter 9, “Development Model,” presents a model of the AI software development process in six phases. Phase 1, preliminary exploration, is concerned with defining the problem scope, looking at alternative approaches (including non-AI techniques), tools and expertise needed, budgeting, and planning. In phase 2, the requirements are detailed, knowledge representation is defined, and prototypes are developed; this stage ends with a detailed design specification. In phase 3, system development, the manager decides on the tools to be used and the personnel needed; the initial knowledge base to be developed and any rapid prototypes to be built are included here. The software itself is developed in this phase. The discussion of phase 4, validation, emphasizes the right things. Phase 5, implementation, seems mislabeled; “support” would be a better title. This section discusses user training, documentation, and customer support. Phase 6 is maintenance. The authors talk about types of software maintenance, including the need to keep the knowledge base current. Chapter 10, “Risks and Safety Nets,” is a good checklist of things to look out for and head off if at all possible. Little is said about what to do if they happen.

Appendix A, “Human Knowledge, Thinking, Problem Solving, and Reasoning: Examples and Descriptions,” is a continuation of Part 1 and could have been integrated with it. Appendix B, “Case Study,” is good as far as it goes but stops at staffing considerations, as if the authors had run into a page count limitation. The glossary is a good start. The bibliography has a preponderance of education, management, and software engineering references and relatively few on AI or expert system development.

Technically, Part 1 leaves something to be desired. It omits some major issues that the project manager should be aware of, such as how he or she knows when the development is completed. The authors barely seem aware of knowledge representations other than rule-based ones. They never mention problems of truth maintenance in the knowledge base. They include little material on the importance of validation tests, benchmarks, and acceptance criteria as part of the formal contract. The authors seem to assume that an AI product can be specified completely, like conventional software, prior to the beginning of actual coding. They say nothing about software configuration management or change control. Not enough emphasis is placed on documentation, system design, system operation, user manuals, or system maintenance procedures. More could be said about how to estimate costs and time and about the pitfalls related to this estimation.

The aspiring AI project manager will need additional good introduction-to-expert-systems books, of which many are available. The AI person aspiring to be a project manager will probably need additional books on software engineering management; a good list of these appears in the bibliography.