The defeat at the weekend of the world's strongest chess player, Garry Kasparov, by a computer carries implications far beyond the 64 squares of the chessboard. The game of chess, with its clear rules and scoring system, its multitude of possible moves and the sophistication of its tactics, is one of the most testing created. It is thus an ideal testing ground for the claim that we can create machines with genuine intelligence, so-called artificial intelligence.
If computers can play chess better than humans, then the implications could be far reaching, for it could mean they would be able to solve other problems beyond our ingenuity.
Ever since the first digital computer - a room-sized collection of 18,000 vacuum tubes called ENIAC - was switched on in 1945, computer programmers have dreamed of building a machine that could defeat the world chess champion. In 1950, Claude Shannon, a pioneer in the field of artificial intelligence, was the first to explain how a computer might be programmed to play the game. There are far too many possible games of chess to have any hope of exhausting the possibilities. One estimate gives the number of possible chess games as around 10120 (10 to the 120th), compared with 1075 (10 to the 75th) atoms in the entire universe. Shannon's idea was to limit the computer's attention in any chess position to a short list of seven likely-looking moves that might flow from it. For each of those moves the machine would then consider seven of the opponent's replies, then seven replies to each of those, building a tree of analysis and continuing until the computer ran out of memory or time.
That model has formed the basis for all computer chess programs ever since. The chess computer has two basic components. The first is its ability to work out the moves that might flow from a single position. Computers have become able to calculate markedly more moves than they used to. The second component is the so-called evaluation function by which the computer decides which is the best move to make. Deep Blue, the computer that beat Kasparov, is astoundingly powerful at calculating possible moves.
This brute, with hardware specifically designed to suit its chess-playing program, can consider a billion chess moves every second. That is approximately equal to one move for every letter of the alphabet in 2,000 large novels - or one for every letter of the alphabet that has ever been printed in the pages of the Independent since it was published in 1986.
In the average chess position, there are about 30 possible moves. So a white move followed by a black move gives 900 possibilities. If a machine is required to look at every sequence of moves that could be made from a single position, it will have reached a billion possibilities by White's fourth move.
A human player, by comparison, will have considered only about 500 different sequences of moves before making a decision. The difference is so vast. But human players make up for the relative weakness of their ability to calculate with the quality of their judgement. Grandmasters, employing their characteristically human skills of concept formation and pattern recognition, use their experience of past games to develop an almost intuitive sense of what a good move smells like. Garry Kasparov may only look at 500 move-sequences before deciding what to play, but they will be the 500 most relevant to determining the future course of the battle. The machine's programmers can only hope that those 500 are among the 100 billion or so that passed through the machine's cogitations. The computer uses its "evaluation function" to decide which moves to concentrate on and which to choose. It is this part of its calculations that is less sophisticated that the brain of a great grandmaster.
So Kasparov's defeat does not imply that computers are intelligent. Even Deep Blue's programmers would not claim any intelligence for their vast number-cruncher of a machine. Vast numbers of calculations may end up providing a better result than human intelligence, but the process is a long way from being intelligent itself. What it does show, however, is that in an area as complex as chess, huge calculating ability may be enough to overcome a basic lack of understanding.
When the great computer pioneer Alan Turing in 1953 raised the question of whether machines could be intelligent, he suggested a test. In its simplest form, an experimenter poses questions to two respondents; one is a machine, the other human, but he does not know which is which. According to Turing, if the machine can be programmed to fool any such experimenter, then one would have to admit that it is intelligent.
Even in chess, however, machines fail the Turing Test, for there are certain types of position that they play rather poorly. Where long-term judgement is more important than short-term tactics, the machine's huge calculations are of little use. Kasparov kept his tactics nice and simple in the second game, pursuing long-term goals beyond the horizon of even Deep Blue's calculations. Once or twice the machine gave a glimpse of the feet of clay beneath its huge silicon brain. Just once or twice, it created weaknesses in its position that any strong human player would have avoided. An experienced human may sense that a certain move will lead to trouble sooner or later, but if it is later than the machine's calculations extend, then even Deep Blue may go wrong.
Even if Kasparov wins this six-game match which ends next week, however, Feng-Hsiung Hsu, the American Chinese chief programmer of Deep Blue, (his name means Crazy Bird) will be able to echo Arnold Schwarzenegger's "I'll be back." If 1 billion moves a second cannot beat Kasparov, then Deep Blue 2, with perhaps 10 billion, will surely terminate him. And if not 10 billion, then 100 billion.
This game may not have shown that computers have intelligence, yet it is only the latest, and most dramatic, in a long series of improving results by machines. They have shown that machine calculation can equal human intelligence and intuition for producing good chess. More importantly, it has shown that computers may be able to cope with decision-making problems in theoretically incalculable areas.
Computers already play a large role in trading rooms in the City and in air traffic control systems where they combine calculation with elements of decision making. Medicine is another area where computer decision making may expand.
Computers have played only a limited role in the life-and-death decisions of medical diagnosis. Yet a computer diagnostics program in one British hospital once had to be abandoned even though it had shown itself to be more consistently accurate than the doctors. Its use would, undoubtedly, have resulted in more lives saved, but it was simply not considered politically acceptable for a machine to be taking life-and-death decisions. So even though, theoretically, computers may make better decisions, patients still prefer the authority of the human touch. GRAPHIC OMITTED
Join our commenting forum
Join thought-provoking conversations, follow other Independent readers and see their replies