The difference is actually quite subtle and would probably be detected only by persons experienced with computer play. A computer engaged in an electronic dialogue is said to have passed the Turing test if the computer's conversation is indistinguishable from that of a human being. At the present time, no computer has ever passed the Turing test. HAL, by comparison, would pass with flying colors -- and later turn around and try to kill the person administering the test!

Drawing on this analogy, one could devise a Turing test for computer chess programs. That is, a chess machine would pass the chess-restricted Turing test if the person playing the machine could not determine whether or not he or she was playing against another person or a machine. Most players would find it difficult to discern whether or not a Deep Blue game was played by a human or a computer. This was proven in an informal experiment conducted by Frederic Friedel, Kasparov's computer adviser. Friedel showed Kasparov a series of games in a tournament played by Deep Thought and several grandmasters. Without identifying the players, Friedel asked Kasparov to pick out the moves made by the computer. In a number of cases Kasparov mistook the computer's moves for those of a human grandmaster, or vice versa. In general, only chess players who have considerable experience playing against computers can identify computer moves.

A specific example demonstrates the difference between the human style of play and the computer style of play: the fact that chess programs exhibit a lack of understanding of the role of timing in chess. Concepts involving never, eventually, or any time can be very difficult for computer programs. For example, a weapon in the arsenal of most strong human players is the idea of a fortress -- a position where a player who has fewer or less-powerful pieces, can create an impenetrable position in which the opponent can never make progress (see figure 5.4). In the 1996 Kasparov-Deep Blue match, Kasparov was able to clinch a draw in the fourth game by means of a sacrifice that created a fortress (see figure 5.5). Although Deep Blue can be programmed to identify many different specific fortresses, detecting the general case of a fortress is still beyond its capabilities and presents us with a complicated pattern-recognition problem worthy of further research.

Another difference between human and computer styles of play can be seen by examining a position involving the ability to reason. At the conclusion of the historic match, Kasparov visited our research lab and showed us a position from which he was absolutely certain that Black would eventually checkmate. Kasparov could not say precisely how many moves it would take, and he was curious to see how Deep Blue would analyze the position. Even after several minutes of search, however, Deep Blue did not see the checkmate. Sometimes search is a very poor substitute for reasoning.

There is, of course, another obvious difference between the human style (HAL) and the computer style (Deep Blue) of play: Humans have emotion. One of the supposed advantages of computers over humans in a game like chess is that computers lack emotion. They are not embarrassed by previous mistakes, they don't slump dejectedly in their chairs when they get into a bad position. One wonders, then, whether HAL's emotional side possibly influenced his style of play (see chapter 13).

When HAL thanks Frank for "an enjoyable game," this is more than simply a pleasing platitude entered into HAL's system by his programmers. Because he possesses both emotion and general intelligence, HAL has the ability to enjoy a good game of chess. Alas, while Deep Blue is sometimes capable of playing magnificent, world-class chess, it is unable to appreciate its own moves.

How, one might speculate, would Deep Blue fare in a match against HAL? Deep Blue could find all the moves HAL plays to finish off the game with Frank in a fraction of a second. Clearly, both machines are tactically very strong. However, given HAL's general intelligence, one suspects it would be able to avoid most of the typical computer mistakes to which brute-force machines like Deep Blue are susceptible. On the other hand, Deep Blue's search strategy could be a strength; it might find counterintuitive moves that would probably be dismissed by a humanlike search. I suspect it would be a very interesting match, in which each computer would gain its fair share of wins.

The idea of HAL losing a game, however, brings up an interesting point. Throughout the film, HAL consistently asserts that he is "incapable of error." Given the overwhelming complexity of the game, it is not plausible for HAL to play perfect chess, as this would require HAL to have solved all possible chess problems. So, if HAL does not play perfect chess, there must be some winning positions in which HAL fails to play a winning move -- or drawn positions in which he doesn't find the drawing move. In the normal sense of the word, these would constitute errors. HAL's own interpretation of the word errorr remains mysterious.