It doesn't matter if Kasparov loses swing in its step

Ignore Deep Blue: humans always undervalue the really intelligent things we do, says William Hartston
Click to follow
Tonight in New York, the world chess champion, Garry Kasparov, will play the deciding game of his match against the IBM supercomputer Deep Blue. After four games, the match was level at one win each with two games drawn.

With its purpose-built hardware capable of analysing up to 300 million moves a second, Deep Blue had shown its capacity to come up with some of astonishing subtlety. By sheer brute force of calculation, it arrives at answers that a human could reach only by calling on such a cloudy mental process as experience and intuition. When Deep Blue won the second game, its play received the ultimate endorsement from grandmasters among the spectators: "almost human" was how they described it.

Such a comment shows, however, that they have fallen for the great confidence trick of artificial intelligence. The insidious argument goes like this: the game of chess has shown itself to be just beyond the boundaries of human intelligence. There are some things that humans can do very well - such as playing noughts-and-crosses or solving crossword puzzles - and others that we are fairly hopeless at - such as finding our way through mazes (which rats can do just as well as us); but chess is just tantalisingly over the boundary of our capabilities: nobody can quite play it perfectly. So computer programmers were naturally attracted to chess as a test of their machines. There are plenty of people making considerable efforts to play chess well, so if computers can be programmed to play better than any human, then they will be able to outperform us at anything else, the argument goes.

But it is a specious argument, based on a misunderstanding of the nature of intelligence. For intelligence, if it is to mean anything, must reflect the way we tackle problems, not just the quality of the answers we produce. You can adopt an intelligent approach and still get a wrong answer; or you can get a right answer through a purely mechanistic, intelligence- free method. If I want to multiply 99 by 99, I may make things easier by noticing that 99 is one less than 100, and realising that the answer must be equal to 10,000 minus 200 plus one. I may still get the wrong answer, but at least my approach will have been an intelligent one.

On the other hand, I could routinely do a long multiplication with four nine-times-nines arranged in columns in my head and add them up. No intelligence needed there, which may explain why the calculating genius of an idiot savant is held in lower esteem than the expertise of a top chessplayer. Yet now we find ourselves in awe of a chess-playing computer, which is no more than a mechanical idiot savant itself.

Being fooled into thinking that a machine understands what it is doing is only one aspect of a general misunderstanding of intelligence. For the intellectual accomplishments we are most proud of are, on the whole, things we're really not doing very well at all, while we take for granted all the really clever things our brains can do.

Take, for example, two basic human abilities: recognising faces and talking. Have you ever wondered what must be going on in your brain when you do either of these? An old friend you haven't seen for years passes you in the street. Despite the new haircut, new clothes, and ravages of ageing, you recognise her immediately. But how?

Somewhere inside your head, all your experiences of human faces is constantly being processed. You know - though you could not possibly explain it - what are the essential qualities that make one face different from another.

The mental image you have of a friend's face can somehow survive the distortions of ageing, make-up or rotating through an unfamiliar angle, and it's still the same face. And when you greet that face with, "My goodness, Margaret, is it really you? I haven't seen you since you moved out of Downing Street", the words gush out of your mouth in a steady stream, with little conscious thought given to their selection. Your mind just conjures up a thought and turns it into words, maintaining an intelligible grammatical structure while simultaneously thinking of what to say next in order to keep the narrative flowing.

It's an astonishing ability which we all undervalue partly because it seems so natural, but perhaps even more because we don't know how we're doing it. That is the real problem. All our conscious thinking is very slow. Yet the intellectual achievements that impress us most are the long, meticulously constructed, logical sequences of thought such as Einstein's Theory of Relativity, or the mathematician Andrew Wiles's feat in proving Fermat's Last Theorem.

We need to see the complexity and originality of the process in order to boggle at it, yet we fail to appreciate the true brilliance of underlying mental mechanisms because they proceed so much faster than boggling speed.

There is a theory that human reasoning and decision-making have an inescapable emotional component. Perception, comprehension and reaction are all inextricably linked in a mental loop of sensing and thinking. A face is linked to the love or hostility you feel for the person behind it; a chess move to your fear of defeat or joy in victory. You cannot even cross a road without being concerned about road traffic accident statistics.

When we play chess, we use our pattern recognition skills to arrange our pieces in effective formations; we use something close to language skills to understand the meaning of an entire position on the board; and then we calculate, ponderously, one move at a time, to try to work out what might happen next, and our final decision is dictated as much by emotion as logic.

Even Garry Kasparov can make conscious calculations no faster than two moves a second. But once he has decided on his move, just look at the way he can reach out, put his fingers round the piece he wants to move, and unerringly transfer it to the intended square without knocking any of the other pieces over. And he does it quite effortlessly and very fast. There's not a computer in the world that can match him.

Even if Kasparov loses to Deep Blue at chess, it will be no cause for humanity to hang its head in shame. On the contrary. The fact that it has taken a huge, purpose-built, mainframe computer to play chess better than us finally confirms something we could never have been certain about before: humans are actually rather good at chess. But there are plenty of things we are all much better at.

Comments