New bridge computer will deliver ultimate grand slam

First draughts, chess shortly - and next bridge? Computers could soon be world-beaters in yet another popular game, according to scientists devising a new generation of programs intended to vanquish humans.

However the news, which might at first cause the game's estimated 3 million players in Britain a deal of fear, is not that bad. It will probably be 20 or 30 years before a bridge- playing computer is world champion, according to Professor Alan Bundy, of the artificial intelligence department at Edinburgh University.

"Present commercial bridge programs play pretty amateurish games," Professor Bundy said, "especially compared to chess programs, which in a few years will probably be the best players in the world." He thinks bridge programs now are at the point that chess programs were in the early 1970s.

Whereas most club players can easily beat the best commercial bridge programs, last February the chess world champion, Garry Kasparov, lost a game in standard competition to the IBM computer Deep Blue. He won the six-game match but faces a rematch next May. A computer became world draughts champion in 1994. A backgammon program beat the then world champion in a game as long ago as 1970, though analysis showed it was lucky in its dice throws.

The problem computers have with bridge is that there is "incomplete information" about the game, unlike chess where all the pieces are in view. Although there are far fewer possible combinations for the 52 cards - 1,044, compared with 10,120 moves in a chess game - the fact that most of the cards are hidden at the start complicates the process hugely.

Even after bidding, through which humans deduce much about the cards, a player only knows the position of half the cards. "One way to tackle it is to use abstraction," said Professor Bundy, a keen player. "You throw away the detail about the cards, and plan on an abstract level. Plays like squeezes, endplays and finesses are collections of moves." The Edinburgh program devises a strategy to win tricks with each suit and forms an overall plan.

Other bridge programs, being developed separately in Maryland and Oregon in the US, also work less like number-crunching computers and more like humans. Matthew Ginsberg, of the University of Oregon, has devised one that guesses where cards are and plans accordingly. The Maryland one aims to lose the fewest tricks given the cards visible.

Alan Williams, general manager of the English Bridge Union, has no fears that interest in the game will diminish, even if a world-beater does eventually turn up. "There would be a lot of resistance to having them competing," he said. "A lot of people like playing bridge for the feel of the cards - and to meet people."

Leading article, page 17