No go for computers
Ben Macintyre finds there’s one game at which artificial intelligence can’t beat humans
TEN years ago in May, to the dismay of many chess enthusiasts, the IBM supercomputer program Deep Blue beat world chess champion Garry Kasparov: the greatest chess mind alive was elbowed aside by raw computing muscle. The quality of Deep Blue’s victory is still debated, but the moment marked a turning point in the relationship between man and machine.
The computer is now dominant in almost every board and card game devised by humankind. Computers can beat us not only at chess but also draughts, Othello ( also known as Reversi), Scrabble, three- dimensional noughts and crosses, Monopoly, and even bridge and poker ( most of the time).
In these games, the computer has a blueprint for ‘‘ perfect play’’: it simply runs the board position through a databank and chooses the best next move, every time.
The unstoppable march of computer power has long been a staple of science fiction, the nightmare evoked in 2001: A Space Odyssey and The Matrix , in which artificial intelligence seeks to control and supersede humankind’s. If computers can win at the intellectual challenge of world- class chess, it is assumed, then the computerised brain, for good or ill, must be inevitable.
Yet there is one game in which the computer is still no match for humans, a game in which a competent teenager can beat the world’s most sophisticated computer program with ease, and that is the ancient Chinese board game Go, the oldest game in the world and the only one at which man remains the undisputed champion.
Go originated in ancient China long before there was writing to record it ( the invention, so legend has it, of an emperor keen to teach his foolish son the virtues of balance and patience). The game involves a simple grid board of 19 lines and two players, one with white stones and the other with black.
The object, simply put, is to stake out a larger territory by tactically placing the stones and surrounding the opponent’s forces. Go is a beautiful game to play and watch: intricate, complex and, for a computer, distinctly baffling.
The qualities that mark out the master Go player are precisely those a computer lacks: intuition, planning, character and pattern reading. Go is not merely a matter of probabilities leading to certainties, which is how Deep Blue and similar programs work.
At its best, the game reflects the defining characteristics of human intelligence. Despite the expenditure of millions of dollars, the promise of prize money and several decades of research, attempts to create a program to challenge professional Go players have made little headway. Even the most sophisticated programs can compete with experienced players only on a reduced grid.
Go is seen as a key to unlocking the secret of artificial intelligence. If computers can learn the game, some scientists believe, mankind would be a big step closer to replicating human thought processes, with great scientific benefits. Con- versely, so long as we can still beat computers at this most human game, then the spectre of a machine that can comprehend rather than merely compute will remain the stuff of fiction.
From China, Go spread throughout Asia, and it is increasingly played in the West. Like chess, despite its complexity at the highest level, Go does not require special mental equipment: Albert Einstein played it, but so does Rod Stewart and so do I ( very badly).
Humans are rather good at Go, but computers are not. Deep Blue won by brute force computing power, working out the possible moves and counter- moves far ahead of play, at a rate of 200 million positions a second. Compared with the possible permutations of Go, however, that was a doddle.
In chess, each player has on average 25 to 35 possible moves; the average number of moves in Go is more than 250. It has been calculated that there are more distinct games of Go than atoms in the known universe.
In Go, a move early in the game can affect the passage of play hundreds of moves later. The vastness of the possibilities offers wide scope for individuality, strategy, personality and intuitive spatial awareness, all of which elude computers.
If chess is a medieval battle, then Go is comparable to a world war, with unpredictable conflicts and territorial shifts across a wide territory. The hardest task is to tell who is winning at any given moment.
A Taiwanese organisation has offered $ 1 million for the first computer program to defeat a junior Go champion, yet despite some recent advances, none has yet reached that skill level.
Computers crunch numbers, but humans recognise patterns and attach values to them, sometimes subconsciously, in the same way that we recognise faces.
To win at Go, a computer would have to be able to spot subtle, complex designs and draw the correct inferences. The experienced Go player can tell whether a configuration is alive, meaning it has been developed in a way that leaves it immune to capture, or dead, in the sense that no amount of counter- moves will save it: precisely the sort of fuzzy concept that sends computers into meltdown.
These are the hallmarks of human intelligence — adaptation to uncertainty, intuition, wisdom, the ability to understand the thoughts and feelings of others, and a sense of mortality — that computers cannot replicate yet.
Alan Turing, the great pioneer of computer science, once said a machine could be considered intelligent if its responses were indistinguishable from those of a human.
Viewed from the level of an ancient Chinese board game, the prospect of a machine that can reproduce the subtleties of the human mind still seems as distant as ever. Only when machines surpass Go ( and collect $ 1 million) will artificial intelligence truly be worthy to compete with the human kind.