On its face, the fate of a chess match seems trivial, so it seems peculiar that a company like IBM would devote money and work-hours to designing a computer program to play the game effectively. But chess is more than a game, it is “a test bed for Artificial Intelligence” (Sinclair, 1998). A computer skilled at chess is a precursor to computers which “think” more like humans. Machines that operate more like humans, i.e., are able to adapt to changing scenarios could be used in applications including entertainment, robotics, and military simulations. Artificial intelligence is the way to achieve this result, and computer chess is a testing ground for new software, programming techniques, and chip designs.

The most famous chess-playing computer was IBM’s Deep Blue. In 1997, Deep Blue defeated Garry Kasparov, the world’s top-ranked chess player, marking the first time a human champion was defeated by a computer (Campbell et al., 2002). Scientists are still studying the reasons for Deep Blue’s success in a study published in a recent issue of Artificial Intelligence. IBM’s dismantling of the computer following the match largely caused the lengthy delay in thoroughly analyzing the match by obfuscating efforts to study it (Kasparov, 2003)

One of the major factors leading to Deep Blue’s victory was its system of “game tree searches” (Campbell et al., 2002). A game tree search is a method of analysis that allows a “thinker” to narrow its focus on a few key issues by having a central node distribute commands to secondary chips (Campbell et al., 2002). A chess game begins with 32 pieces, many of which can make a wide variety of moves, moving in multiple directions and at varying distances. Imagine trying to decide on a move, analyzing what would happen for each move by an opponent, and your response to this. It is easy to see that the number of potential moves is enormous. This is because it is the result of an exponential function. For example, if each player had 8 pieces that could move an average of 5 different ways, then after 5 moves by each player, the number of potential combinations of moves numbers in the quadrillions (4010), too many for even the fastest computer to evaluate within time limits. Yet most chess players search deeper than five moves into matches. To alleviate this problem, Deep Blue “decided” early in the search which moves were unlikely to lead to any productive position (Campbell et al., 2002), in other words, moves that would not result in jeopardizing the opposition’s pieces or would cause danger to the computer’s own pieces. This saved time, a very important advantage in this match because the first forty moves of each player were required to be played in two hours (Campbell et al., 2002). This also would be critical in other machines employing similar methods, such as a video game, which also are limited by time constraints.

The programming further augmented Deep Blue’s strength. The program gave “credit” for the advantages of certain positions. For example, a move that is significantly better than every alternative, i.e., one that saved one’s own pieces while endangering those of the opposition, received a large credit, marking it as a superior move and more likely to be played (Campbell et al., 2002). Another vital element was the determination that if a move is good, it is better played sooner rather than later (Campbell et al., 2002). This allowed the program to search deeper into the match because it wasted less time contemplating moves already evaluated and deemed insufficient by the computer.

Although Deep Blue’s programming is what ultimately gave it victory, its chip greatly added to its strength. Part of the chip was the “move generator,” which did exactly what its name implied. It was “implemented as an 8 x 8 array,” (Campbell et al., 2002) dimensions not coincidentally those of a chessboard; in fact, Campbell and colleagues term the move generator a “silicon chessboard.” The array, was able to help the computer simulate moves digitally. Another part of the chip was the “evaluation function,” which slashed the time required per move by approximating the value of a position when sufficient (Campbell et al., 2002). This was another important method of staying within the time limits, because calculating an exact value, based on the positions and values of each individual game piece, was often unnecessarily exact and time consuming. The same principle applies to robotics, because a robot may not be required to be perfect in all situations, and speed may be desired more than complete accuracy.
The success of Deep Blue was a confluence of several factors. The technical factors—such as the tree system and the evaluation function—discussed here are the most applicable to artificial intelligence in general. Despite victory, the programmers admit that there were significant flaws, especially in efficiency (Campbell et al., 2002), showing that computers have a long way to go before they can replace humans at even menial tasks. Eventually, though, the machines will most likely be able to defeat any person consistently at chess, but the time has not yet arrived. It seems likely that Deep Blue was one in a long line of steps toward a more “human-thinking” machine.