Smartest? 4 January 1981

Trevor Leggett was head of the Japanese Department of the BBC.

This is one of his broadcasts to Japan

ZUBARI for 4 January 1981

Smartest?           

Hello listeners! This is the first Zubari programme of the New Year so I should wish you all a happy new year, but will it be a happy new year? Thirty years ago, it was predicted that by 1982 computers would have been developed which would be taking the decisions of government for us.  It was thought that the development of a crude chess-playing computer (as had just been done in America by Dr. Shannon, and in England by the mathematician Alan Turing) showed that computers would be able to take in more and more information, and in the end come to more logical conclusions than humans.

Everybody knew that the computers, of thirty years ago, played chess very very badly indeed.   But everyone expected that they would improve more and more, and finally surpass human beings.  And if they could play chess better than humans, they could think better than humans. So, finally, all the states would have to set up a computer to govern themselves, because other competing states would have computers, which would think better than any human beings.

That was the fear – and perhaps the warning.  A celebrated book called Colossus, which was made into a film and a TV play, dramatized the taking over of the world by the computer, Colossus.

Well, I can wish you a happy new year with confidence. I recently borrowed the best of the chess-playing computers on the market, and played several games against it.  My report is this: the machine can calculate, but it cannot plan. So, there is no danger that it will ever replace human judgement. As computers develop they will become better and better at calculations: but they will not become better at judgement. They cannot be programmed to evaluate.

The chess computer excels at exact calculation. For instance, it is very good at solving chess problems these are the Western chess version of the ‘tsumé-shōgi’ problems which appear in so many Japanese magazines. The position is given, the ‘mochi-goma’ are shown, and you have to find the win. Every move is an ō-te move. There are of course a tremendous number of possible variations; but it is just here that the computer excels.  In a short time, it can run through them all, and find the one variation which will force the checkmate or ‘tsumé’.  But what the machine cannot do is to form a plan in the ordinary game of chess. Of course, the programmers have fed into it a number of the jō-seki (opening sequences), and it chooses one of them and makes the moves. It will go on doing this mechanically unless one of its pieces or pawns is directly threatened by the opponent. In that case, it interrupts the sequence to defend its piece.

But it cannot form a plan.  So it will go on with a jōseki line which is quite inappropriate against what the opponent is doing on his side of the board.  It cannot see that an attack is building up. Until there is a direct threat to one of its pieces, it takes no preventive action.

These machines do quite well against weak or medium players, because chess (like shōgi) is so complicated that an ordinary human being tends to overlook some simple thing – like an Ōte-bisha for example. The human being guards against such things for an hour, but then becomes tired, and his attention momentarily flags.  He overlooks the chance.  The machine does not become tired, It never overlooks such things, and it seizes the chance when its human opponent overlooks one.  So it often wins.  Or rather, the human opponent makes a mistake, and loses. But an experienced player does not waver in his concentration.  Against him, the machine finds no chance for an Ōte-bisha. It keeps on moving its pieces into safe positions, but it has no plan.

I put the machine, when I played it, on to its highest level of skill.  On that level of skill, it could take up to 20 minutes to make a move. I used to begin the game in the morning; I would make my move, and the machine would reply with one of the standard jōseki.  We played on for five or six moves each, and then it would come to the end of its programme.  It would begin to ‘think’.  I would begin my translation work, and wait for the machine to announce its move.  This model calls out its move – in an American accent – and then lights up while it waits for the opponent’s move.  I could generally make my reply in a few seconds, and then go on working for 15 or 20 minutes.  I would gradually build up a strong attacking position, without making any direct threats.  The machine, oblivious of this, would move cautiously.

Then I found that I could overwhelm it with a direct attack, sacrificing several pieces.  Finally, the machine would say, “I LOSE”.

I found it rather uninteresting to keep winning in this way, so I varied my strategy. In one game, I attacked from the beginning, giving up a rook (hisha) in the attack.  NOW the machine showed its strength in calculation. It did not have to plan. What it tried to do was, to exchange off all the other pieces, so that at the end it would be a rook ahead. It nearly succeeded in winning, but I managed to lay a deep trap which was too far for it to calculate. I escaped and finally even forced it to say again, “I LOSE”.

But I must admit that when I tried this again, the machine did succeed in getting an overwhelming winning position. I struggled, but I had sacrificed too many pieces in my direct attack. I felt that I was getting to the situation where the machine would say – in its American accent – “YOU LOSE”.

Yet …… it did not.  Why not? I pulled the plug out.

© Trevor Leggett