Artificial intelligence, in the view of many, has so far offered more promise than performance. Chess-playing, however, is undeniably one of AI’s greatest and most easily quantifiable success stories. Computer play has improved significantly over the last 25 years: specifically, the ratings of the best chess programs have increased, on average, about 49 Elo points per year (the Elo system is a rating scale developed by Arpad Elo of Marquette University in the late 1950s); current programs are playing at the grandmaster level and, using a time series, the best performance for 1994 is projected to be about 2911, a rating that should be sufficient to accomplish the long-sought goal of chess programmers, the defeat of the world champion.
Levy and Newborn’s eminently readable book, a celebration of the programmers’ success, is at once a history of computer chess and an explanation of how chess programs work. The authors have managed to pack much useful and fascinating detail about both areas into 246 pages. My only significant complaint is that the presentation is too brief; several times I felt frustrated when the discussion was cut off without my finding out what I wanted to learn. Fortunately, a five-page bibliography is provided for those who want to pursue issues further.
For most of the historically significant programs, the authors offer not only a general description of the move-selection algorithms, but also well-annotated representative games against both human and computer opponents. These programs range in sophistication from early hand simulations to the current computer champion, DEEP THOUGHT. To fully appreciate the underlying algorithms, one has to replay these games, move by move, on a chessboard.
Those readers who have harbored a desire to create their own chess programs will find the chapter on search techniques--the essence of all computer programs--and the alpha-beta algorithm invaluable. Also useful is the chapter on endgame play, a part of the game in which computers have traditionally performed poorly. Before committing two or three months of one’s time (the authors’ estimate for writing a computer chess program), however, one should note that leading chess programs have generally been written by teams, minimally composed of an exceptional programmer who is also a good player (median Elo rating 1700) and a chess expert (median Elo rating 2200).