computer chess

According to Emanuel Lasker, a former world champion, chess is a fierce intellectual struggle devoid of any scope for verbal tricks. It is the supreme battle between two minds. As such it is natural that one of the first task domains for machine intelligence studies (see MANIFOLD-9) to be chosen should be chess.

The three best chess-playing programs are those of Greenblatt, Gillogly, and Atkins and Slate; yet their rating on the U.S. Chess Federation scale is between 1400 and 1500. This is just better than most amateurs, but worse than strong amateurs. It is far below international masters (2300-2600) and grandmasters (2600-2800); Bobby Fischer's last USCF rating was 2824. the highest ever award. So there is a long, long way to go before a computer will be world champion.

Professor Donald Michie, director of the machine intelligence unit at Edinburgh, reported on some of the difficulticel in an article in the New Scientist(17/8/72). The basic problem is one of description - what are the salient and significant features of a particular,chess position? How does one develop in computers a 'knowledge' and an 'understanding' of chess? Some work is being done, applying relational structure techniques, at Edinburgh, with little success so far.

Meanwhile David Levy, now an international master, is waiting patiently to settle a bet he made with John McCarthy, Donald Michie and Seymour Papert in 1968. They wagered £1000 that Levy would be beaten by a chess program by the year 1978.

Psychologists have been attacking this problem from another angle. They have been trying to find out how humans play chess and why computers cannot match up to the game's masters. In 1966, two Russian psychologists, Tichomirov and Poznyanskaya, put forward a working hypothesis. They postulated that the way we analyse a chess problem or position perceptually requires systems fundamentally different from those which have been postulated for explaining heuristic search behaviour. The importance of the perceptual visualization is supported by considerable evidence, from their own observations of eye movements of players studying game positions.

This is substantiated by some interesting research done by the Dutch master De Groot. He presented a complex game position for only five seconds and asked the subjects to reconstruct the position. Chess masters got over 90% of the pieces on the right squares but weaker players managed only 40%. Yet, significantly there was no difference when the pieces were placed on the board randomly.

Thus it seems that perception in vitally important, especially discriminating perception, to play chess at a high standard. It would be interesting to see whether blind people can be taught to play chess well. This research also tends to indicate that the use of relational structure techniques is the most promising way of improving the chess ability of computers. Yet one still has doubts. For a two-dimensional process has to be encoded in one dimension. Which seems to be an impossible task: it is highly likely that Levy will win his bet.