Does Deep Blue use AI?

When Deep Blue played Garry Kasparov in February 1996 and May 1997, the extensive IBM web pages devoted to the site claimed that Deep Blue did not use artificial intelligence. We argue that this claim is inaccurate, is representative of a wide-spread phenomenon in the field, and is ultimately harmful to AI. Does Deep-Blue use AI? In February 1996 and May 1997, the reigning world chess champion, Garry Kasparov, played Deep Blue, a chess computer built by IBM and one of the strongest in the world (Keene, Jacobs, and Buzan, 1996). Kasparov won the first match 4-2. IBM sponsored extensive coverage of both matches including a web site with up-to-the-minute results, commentary, and background information. Surprisingly, there was almost no mention of artificial intelligence in any of the IBM web pages. The one exception was in a list of frequently asked questions, one of which was, "Does Deep Blue use artificial intelligence?" Even more surprisingly, IBM’s answer to this question was "no"! (IBM, 1997) To be fair, a careful reading of their answer to this question seems to equate artificial intelligence with the simulation of human intelligence. By that measure, Deep Blue doesn’t use AI, since it plays chess very differently than a human does. For example, Deep Blue generates and evaluates about 200 million chess positions per second, something no human can do. On the other hand, it is well understood that AI is much broader than simply the simulation of human thought processes, and encompasses a great deal of techniques that are clearly not psychologically plausible. By this broader definition, does Deep Blue use artificial intelligence? While there is relatively little published on Deep Blue, it is well-known that its main algorithm, at least for mid-game play, is alpha-beta minimax search with a heuristic static evaluation function. We argue that if any technique deserves to be called AI, this one does. Chess was one of the original AI problems, and remains a canonical AItask. In fact, computer chess predates the term "artificial ntelligence". The first paper on the subject, written by Claude Shannon, and entitled "Programming a Computer To Play Chess", was published in 1950 (Shannon, 1950). In all the earliest books on AI, such as Computers and Thought (Feigenbaum and Feldman, 1963), chess and Samuel’s checkers program (Samuel, 1963) were featured prominently. One of the earliest AI techniques, heuristic search, was developed primarily to deal with problems such as computer chess. In particular, AI added to the minimax algorithm of (Von Neumann and Morgenstern, 1944) the ideas of heuristic static evaluation and alphabeta pruning, all of which were developed in the 1950s. These ideas form the core of all chess programs even today, including Deep Blue. Thus, clearly Deep Blue is a product of artificial intelligence, even if it isn’t of the cognitive simulation variety. The fact that it won a game against Kasparov in 1996 should be viewed as an achievement for AI, and if it wins their rematch in May of 1997, it should be heralded as a glorious success for AI, tackling a problem that has been worked on hard and continuously for almost fifty years.