Browse Books

Reviewer: T. Anthony Marsland

This ACM Distinguished Dissertation not only represents a demonstration of the power of VLSI design techniques but also provides an insight into the development of a powerful chess-playing machine. The book is well written and, although of greatest interest to those working on hardware support for artificial intelligence applications, it can be appreciated by anyone with a broad knowledge of computer science methods. The opening briefly covers some of the theory behind the alpha-beta algorithm, the principal means of searching two-person game trees, and reviews the hardware move generators in both Ken Thompson's highly successful Belle chess program and the MIT CHEOPS project. This introduction is followed by a good description of the move generation problem in computer chess. Ebeling's main contribution is to provide efficient hardware support for selecting from all the “ever possible” moves to a given square, those moves that are actually legal by the pieces on the board. The major advantages of this approach are that moves may be generated both incrementally and in a predetermined desirable order. In particular, checking moves may be determined easily (normally this is an expensive software test involving a search), and captures may be partitioned into safe and sacrificial moves. Finally, the hardware can give some priority to pass-pawn moves. The resulting move-ordering quality and generation speed provides Hitech, Ebeling's demonstration piece, with a significant advantage over its competitors. Even with the advantage of speed to increase the depth of search and ordering to improve the pruning rate, the Hitech program still requires a solid base of knowledge about chess. Some of this knowledge is provided in chapter 4, and although much of this information also appears in a variety of places, the presentation here is clear, unambiguous, and is a good supplement to the other best single existing source [1]. A second contribution of the book is the hardware support for position evaluation, support that contributes heavily to the system's ability to search 200,000 positions per second. Chapter 4 is followed by a short one on the Hitech machine itself. Here the presentation is terse to such an extreme degree that only the most serious reader will find in useful. To compensate, the penultimate chapter, on testing, is almost exemplary and shows clearly how Hitech evolved and improved rapidly from June 1985 to June 1986. Perhaps one disappointment is the omission of the test suite used to show Hitech's average performance. Others might like to see if their program can match Hitech's search of only 50 percent more nodes than the minimal game tree. Of course those results also depend on the efficient use of a transposition table, and the chapter provides a good discussion on the effect of table size and the details about the replacement of entries. The tests carried out are broadly based and complement those summarized elsewhere [2]. Unlike most Ph.D. dissertations, this one is not laden with abstract theory and complex notation, but it is still unique and original in many ways. It provides a valuable contribution to the scientific literature and is interesting reading throughout. One surprise, though, is the rather short, introverted bibliography. Only three or four references to hardware aspects are included, and many of the others are to the works of Carnegie-Mellon researchers. This paucity is more an indication that the book is well focused on its topic than that the existing body of knowledge is small [2].