Like athletes in other sports, chess players play the game with a certain style. Some for example have a positional strategy, others a tactical approach and these styles influence their shot selection at each move.
Given the way that computers are beginning to dominate even the world’s best players, is it possible to teach a computer to play in the style of Vladimir Kramnick or Gary Kasparov? And if so, would it then be possible for that computer to recognise players by their moves alone?
Those are the questions that Mark Levene and Trevor Fenner at Birkbeck College in London set out to answer.
They used a traditional learning method to teach a computer the styles of Kramnik and Kasparov. Kramnik, it turns out, has a propensity for the bishop pair and tends to prefer king-side manoeuvres. Kasparov, on the other hand, is a sucker for saddling his opponents with doubled pawns and when playing black tends to opt for an attack on white’s king when the players castle on opposite wings.
Levene and Fenner then tested the computer using moves 23-35 from a set of 123 games and got pretty good accuracy rates. But they were unable to repeat their success in games between Kramnik and another former world champion, Veselin Topalov.
Clearly, the method needs a little more work.
But it clearly has potential in ensuring that you can spot when your postal chess buddy has turned to Deep Blue for advice. And what a fantastic biometric standard too. Imagine proving your identity at customs by sitting down as white.
Ref: arxiv.org/abs/0904.2595: A Methodology for Learning Players’ Styles from Game Records