Go for its plug, Garry!

Today in New York, the world chess champion faces a new challenge from the most powerful chess computer. William Hartston assesses the prospects
In 1770, at the court of Maria Theresa of Austria, Baron Otto von Kempelen caused a sensation by demonstrating the world's chess-playing automaton. "The Turk" was a man-sized turbanned figure seated at a chess table. When a human opponent sat at the other side of the board, it would reply to his moves with a cranking of joints as its hand moved to a piece, picked it up, and deposited it on the chosen square. And it won the vast majority of its games.

Doubters, naturally enough, suspected that a strong human player was hiding in its innards, but von Kempelen delighted in opening a variety of doors to the cabinets beneath the board to reveal a fine collection of cogs and levers, but no sign of any person operating them.

It was, of course, all an illusion. There was a man inside, but his location changed as doors opened and closed, thanks to an ingenious system of sliding platforms and pulleys. Whichever door you opened, he was never there.

After a successful tour of the courts of Europe - there is even one tale of its sweeping the pieces from the board in disgust when the Emperor Napoleon tried to test it by making a series of illegal moves - The Turk lost its novelty value and went into retirement. However, when von Kempelen died, the machine was sold to Johann Maelzel, inventor of the metronome, who added music to its charms and launched it on a highly acclaimed tour of America.

The Turk epitomised our fascination with the idea of an intellectual struggle between men and machines. However, for a machine to play respectable chess without having a man hidden inside it we had to wait for the development of electronic computers.

Since the earliest days of the computer, the game of chess, with its high premium on accurate calculation, has played an important part. The early development of artificial intelligence was haunted by this thought: if computers can't even beat humans at chess, what hope do they have for making really complex decisions? Chess was the touchstone by which machine intelligence would be measured.

In 1864, Charles Babbage had speculated about the possibilities of his "analytical engine" playing chess; in 1948, Norbert Wiener, pioneer of cybernetics, discussed the idea of a chess-playing computer; in 1950, Claude Shannon proposed a theoretical strategy for such a machine; in 1951, Alan Turing wrote out a chess algorithm that could be calculated by hand. Finally, around 1960, a machine first defeated its programmer.

All the same, the top human players could afford to be scornful of computer chess for another 30 years. While the artificial intelligentsia repeatedly predicted that a machine would be world champion "within 10 years", the decades rolled past without computers attaining professional standards. They became very good at avoiding tactical errors (which decide the vast majority of amateur games), but concepts of deep strategy continued to elude them. With increased processing speed, however, that began to change.

Chess is a game at which humans ought, by most rational criteria, to be rather bad. With up to 32 pieces scattered over 64 squares, a typical position contains too much information for us to take in. Experiments have repeatedly shown that our brains can only juggle about seven pieces of information at any one time Yet we cope very well at chess by using our highest skills of pattern recognition and concept formation to turn chess into a game of high-level judgement rather than pure calculation. The question for computers is how much calculation is necessary to render such judgement and strategy redundant.

Last year, we came close to an asnwer. Computers will never be able to analyse all the possibilities in a chess game - there are more possibilities for the first 25 moves than there are atoms in the universe - but when Deep Blue, the IBM supercomputer, started thinking at the rate of two million chess positions every second, it proved it could give humanity a fair run for its money. Indeed, in the first game of its challenge match in 1996 against Garry Kasparov, the strongest human player, Deep Blue registered an astonishing victory. In later games, however, Kasparov treated the beast with more respect and won the match by three wins to one with two draws.

Today, the return match starts. Deeper Blue thinks twice as fast as last year's model and, we are told, has considerably more chess knowledge incorporated into its program. Kasparov is still a clear favourite, but he is almost alone among top players in believing that the tide of silicon can be checked for much longer. The playing strength of computers has increased with their processing speed at a constant rate. And processing speed doubles every 18 months. At that rate, Kasparov has no more than a couple of years left before ceding his crown to a heap of metal.

Comments