The Brain Makers

Rating:
Summary: Fascinating Facts, Questionable Interpretation
Review: I found this book both interesting and limited. Having encountered AI technologies and personalities at various points in my career, it was interesting to get from the author a more comprehensive view of what was going at the personal and organizational levels during the period covered by the book. The author's view of these matters was culled primarily from his experience as editor of a newsletter called AI Trends during that period. I visualize the author writing this book by pulling from stacks of old newsletters, article clippings, corporate brochures, and notes from interviews and discussions he had as a journalist on the beat. I see this book's value mainly in that it summarizes a lot of information about AI people and organizations in one place, organizing it into thematic chapters.
The author inserts his own perspective throughout the book, with mixed results. He is attracted to the dirt, the scandal, the quirky personality, and this leads to some interesting reading, interesting in the way you might listen to the town gossip, in spite of yourself. I had to take his gossip with a grain of salt, because some of it was based on questionable interpretations of the author, but enough was substantiated to be interesting. For example, the rise and fall of AI companies is an interesting story that parallels that the recent dot com cycle, and the AI era has lessons to teach us about the business and management of technology. However the author's bias toward airing dirty laundry sometimes comes across as a sneering attitude, or at least over-dramatization, and some of the ugly pictures he paints seem ugly because of his paint, not the events he reports. For example, he presumes to classify management talent as "A-teamers" (capable) or "B-teamers" (less capable), then identifies hiring B-teamers as evidence of poor management in some companies.
The author clearly does not have a deep understanding of AI technology, and this limits his ability to achieve two things he tries to do in the book: (1) explain AI in laymen's terms, and (2) interpret the technical significance, shortfalls, and potential of AI technology. He is on target some of the time, and sometimes misleading, or even wrong. For example, as the author correctly points out, the publication of the book Perceptrons by Minsky and Papert was an intriguing chapter in AI, since it effectively shut off research in neural networks for a long time. However, his discussion of the essence of Perceptron's criticism of neural networks is misleading: he says it was that neural networks cannot ".. learn new things from past experience..", when actually the main criticism was that certain kinds of problems can never be solved by neural networks. His discussion of how researchers eventually countered Perceptron's arguments is also misleading: he cites Hopfields's showing that recurrent neural networks can do things the brain does (an important contribution), when the more relevant direct answer to the Perceptron dilema was the development of good training algorithms for multi-layer nets that could solve the "impossible" problems.
I enjoyed reading this rather long (488 pages) book. It moved along quickly, and it was interesting to find answers to a lot of "whatever happened to ..... ? " questions. The AI era covered by this book was filled with fascinating stories and people. I would have preferred a more penetrating and knowing analysis of AI technology itself, that would make it easier to separate good ideas from business blunders, circumstances from fundamental flaws. And I wanted to have less of a feeling that the author was just guessing at some of his insider insights.