Chess-playing computers have come far since 1968, when an international master named David Levy placed a dollars 10,000 bet that no mechanical opponent would be able to beat him within a decade. Although Levy won the bet, the performance of chess software began to improve as soon as he collected his money. By the end of the Eighties only 1,000 or so of the world's players could claim superiority over the best computer programs; earlier this year, five of the top 10 grandmasters were beaten at five-minute chess by a program using the same processor as that which vanquished Kasparov in yesterday's 25-minute game.
It is not only professional players who are disquieted by the prospect of machines winning at chess against humans. Voltaire described chess as 'the game which reflects most honour on human wit'. Learnt in 10 minutes, studied for a lifetime, chess has long seemed to distil human intelligence. So seductive was the idea of a chess machine that hundreds of players were taken in over the years by hoaxes - among them Napoleon, who lost to a machine that subsequently turned out to have a highly intelligent midget hidden behind its cogs and pistons.
Yet the beating of grandmasters by computers should not be seen as threatening. It does not, in fact, show that machines can 'think', neither does it provide any grounds for worry about the future of human civilisation. Chess is in principle no different from noughts and crosses. Both have a finite number of possible positions; both can be resolved by the brute method of trying every move from the starting position and considering all the possible responses and counter-responses. The only difference between the two is the size of the task.
Programming a personal computer to force a draw at noughts and crosses would take no competent programmer any longer than an afternoon. Chess, on the other hand, has mind-numbingly immense permutations of possible positions. Computers have to seek short-cuts. They do this by avoiding obvious mistakes, such as sacrificing a queen unnecessarily, and by using rules chosen by programmers to evaluate which positions are better than others. As a result, they prune the 'tree' of possible moves and counter-moves down to manageable proportions.
The improving performance of chess computers in recent years has been due in part to smarter programming, but more to the astonishing increases in computing speed. The Pentium chip powering the computer that beat Kasparov yesterday, made by the US company Intel, runs fast enough to allow 100,000 positions to be analysed every second. Even at such blistering speeds, though, computers can still examine only a tiny fraction of the possible outcomes of a game.
More striking still is the disparity between human and mechanical methods of play. William Hartston, the Independent's chess correspondent, recently devised a test to distinguish the two, consisting of eight problems, four of them particularly amenable to brute-force searching and four to less formal methods, with a score of minus 1 for getting problems in the first group right, and plus 1 for those in the second. The test results divided humans from machines into two distinct groups. All the machines scored minus 6 or below, all the people minus 5 or above.
This demonstrated difference in methods illustrates why computer success at chess is so hard to replicate in any other field. Chess is not the only human activity that is formal and based on rules. But superficially comparable problems, such as diagnosis of diseases from symptoms or recognition of human speech, have proven hard to automate. And while humans can apply to one endeavour what they have learnt in another, computers and their programmers have to address each problem anew.
More important, there are great swathes of human activity in which simple rules of the kind that govern the game of chess have no place. Anything that requires natural language, for instance, still offers an insuperable obstacle to today's computers. Attempts to translate from one language to another have achieved little more than glorified dictionaries. Real understanding, in the normal sense of the word, is a giant step away.
One likely consequence of yesterday's victory for the computer will be that chess will gradually lose its allure. It will be less exciting to watch, since spectators will eventually be able to carry notebook computers capable of telling them what the world champion ought to do next. There may also be greater opportunities to cheat, even though the incentive to do so will diminish as the prizes do.
What will not happen is what Garry Kasparov suggested could happen when he said that 'if a computer can beat the world champion, a computer can read the best books in the world, can write the best plays, and can know everything about history and literature and people'. Such a mistake is evidence either of an inflated idea of chess or a misunderstanding about the rest of the world. Or both.
(Photograph omitted)Reuse content