“My subject is knowledge science. It is the study of what communities know and the ways in which they know it.… Knowledge science looks at how knowledge is made, maintained, disputed, transformed, and transferred.” Collins provides a sociological perspective on the kinds of knowledge that expert systems can and cannot capture, and on the domains that are and are not likely to be successful.

The author argues that intelligent computers cannot be treated as isolated brains. Rather, computers must be thought of as social prostheses--replacements for humans in communities. This book studies artificial intelligence from that point of view.

Part 1 examines what computers can do and what we mean by machine-like action; this material is somewhat obvious to practitioners in the field. Part 2 deals with the idea of expert systems both to explore their potential and limitations, and as a foil for more extended analysis of knowledge in general. The book gets more interesting here, though I think the author could have used a better example than PICKUP, an expert program that detects when a person met at a party is likely to satisfy the predicate “will-come-back-to-my-place.”

Part 3 is an interesting case study of the author’s experience in trying to develop an expert system for crystal growing. The dialogues between the technical expert and the knowledge engineer show the difficulties of knowledge elicitation. Several chapters on the nuts and bolts of crystal growing are intended to demonstrate the difference between abstracting rules of practice for use in a text or an expert system and learning the practice of the craft.

In Part 4 the author has some interesting things to say about the sociology of the Turing Test, particularly the use of the fact that the machine is to mimic a man who is imitating a woman, not a man being himself. Most AI philosophers tend to overlook this point. He defines the proper protocol for the Turing Test, including items such as the abilities, training, and attitude of the interrogator; the subject matter to be discussed in the test; and the length of the test.

Finally, the book explores the outer limits of what we say and do by virtue of being told things and compares this with what we can say and do by virtue of our socialization. The author concludes, “So far, the experiment in knowledge science that is artificial intelligence points unerringly to the irreducibly social nature of human beings and their knowledge.” This book provides some interesting sociological insights into the field of AI.