In most books, the title is general or even cute, but the subtitle really says what the book is about. The reverse is true here; the title is more accurate than the subtitle. What Rasmus has written is much more about thinking than it is about building. The content gives little guidance on actually constructing artificial intelligence software, but instead gives project managers some ideas on how to think about software in general and on what tools are best used to implement leading-edge applications.

This book is mostly a compilation of articles that Rasmus has published in the past few years on this subject. He has added both an introduction and an afterword to hold the chapters together. He also provides a good bibliography and an “Intelligent Object Buyers Guide,” which lists about 15 vendors that offer software in this marketplace.

Rasmus believes that artificial intelligence methodologies form the programming paradigm that is necessary for us to deal with the complexities of the information systems of the 21st century. He talks about the various flavors of AI (including expert systems, inference engines, neural nets, and genetic algorithms) and admits that none of them has become mainstream. Several times, for example, he refers to Doug Lenat’s CYC project, begun at the Microelectronics Computer Corporation, to build a “commonsense database” as the foundation of all AI efforts.

Chapter 1 is a fine discussion of agent technology. Rasmus discusses agents first because he believes that “they are the next great hope for AI, and perhaps the first realistic hope, given what we now know about how the human brain functions.” The chapter reads almost like science fiction, with descriptions of what we can do today mixed with scenarios of what might happen tomorrow. While he does not say much specifically about the impact of a pervasive Internet, it is easily inferred from his narrative. I visualized an ecosystem of agents doing our bidding, but always learning and, therefore, evolving minds and motives of their own. We may need Asimov’s Laws of Robotics not to allow us to live with androids but to govern an infinite and invisible web of electronic servants.

While chapter 1 is futuristic, chapter 2 has more immediate value. Since most of us do not (and probably will not) work on AI projects, Rasmus gleans lessons from AI methodologies and applies them to more mundane information technology work. He describes both frames and objects (he thinks objects are far more versatile) and shows how the concepts associated with them can be applied to a variety of software tasks. He discusses how we can use knowledge acquisition to improve application programs. He talks about “intelligent information engineering” as a way of building architectures rather than only as a methodology.

One of the best parts of this chapter is a section on the testing of intelligent systems. All software testing is hard, but testing software that is designed to change itself is particularly tricky. Considering a worst-case situation is often helpful in dealing with more straightforward problems, however. For example, the author urges us to look carefully at unexpected outputs, not just to find errors, but to infer logic paths that were not part of the design specification. We are truly talking about software that takes on a life of its own.

Rasmus follows this idea into a discussion of legacy systems and “legacy knowledge.” Humans forget; computer systems do not. Expert legacy systems can become as rigid as the most arcane organizational bureaucracy. Systems built with modular objects are easier to maintain than procedural monoliths, but humans must have the determination and apply the resources to keep both the rules and the data up to date.

A long chapter is devoted to examples of applications of AI methods. Readers who are just exploring how AI can be used may find this useful, but Rasmus does not offer much that is novel here. I found his more philosophical work much more valuable.

In the afterword, Rasmus allows himself the luxury of reflection. He questions the very premises of AI: “Computers don’t have the same sense of sense that we have. We should not expect software to relate to our world or our sensibilities if it does not exist in our world.”

He says later, “The idea that logic alone informs our consciousness is an absurd conjecture.…Wedding artificial intelligence to logic straitjackets software into deterministic behavior and therefore denies it any chance of evolution that would lead to a unique software-hardware consciousness.”

The phrase “software-hardware consciousness” reminds me of the Star Trek: The Next Generation episode in which Wesley Crusher accidentally lets loose “nanites,” simple electronic molecules that self-organize and form a viable intelligence that ends up threatening the Enterprise. If we take Rasmus’s agents, add AI methodology, and let it all evolve on the ubiquitous Internet, we may get more than we bargained for. As with all technologies, the benefits and dangers are sides of the same coin.