John Chamberlain
Developer Diary
 Developer Diary · You Heard It Here First · Sunday 22 February 2004
Are Chess Computers About to Overtake Humans?
Over the last few years the rated strength of chess computers has been creeping upwards and is at the verge of overtaking Garry Kasparov, the top-rated human. Jeff Sonas has written an article for ChessBase News detailing these statistical results from the SSDF. They show computer program Shredder at 2812, a mere 19 points away from Kasparov's 2831 FIDE rating. By linear extrapolation the programs seem to be getting stronger at a rate of about 50 points a year so if the trend holds computers will pass Kasparov sometime this year.

Sonas is not convinced that computers will overtake humans and points out that humans are improving too. Sonas thinks that humans have a chance of holding their own as evidenced by past downturns in computer strength which show that humans can gain relative to computers. He thinks it unlikely that computers can drift higher in strength indefinitely. Joel Lautier, a top French grandmaster, has weighed in with some similar thoughts and an analysis of the Kasparov-X3D Fritz match last year.

My Opinion - The Writing is on the Wall for Humans

I remember when top-level computer chess meant Chessmaster or Zharkov running on a home PC or mac. In those days the 2200 USCF (master) level was the glass ceiling for computer programs. Nowadays such programs would be rated below 2000 USCF. Those days are long gone. Today anybody can buy an off-the-shelf version of Shredder that has grandmaster (2600+) strength even on ordinary PCs. This is a result of computers becoming more powerful and the advance of algorithms in the programs.

One might argue that at the highest levels chess involves recognizing long-term patterns that cannot be calculated. In this argument the idea is that every additional move of calculation requires an order of magnitude stronger computer, but being able to calculate 14 moves ahead will not really be much better than calculating 13 moves ahead so there comes a point when faster calculation does not really help. I think that this argument has some merit but is incomplete because the reality is that a large part of the strength of recent computer advances is due to better heuristics. Due to heuristical programming computers can combine enough ideas with calculation that they will continue to gain against humans to the point where they are substantially stronger.

Heuristical Programming Drives Cutting Edge

In the last few years chess programmers have been putting much emphasis on improving their positional play. In the old days positions were evaluated using the crude measure of material balance, but now computers factor in king exposure, pawn weaknesses, central control and other positional considerations that a human normally uses. This costs much more processing time in the evaluation phase but overall it makes the computer's analysis much better. The end result is that the computer does not search as deeply, but the moves it does search are better quality so overall it is stronger.

There is a funny story told about Emmanuel Lasker, perhaps the greatest all-around chess player ever. An amateur asked him how many moves in advance he analyzed. He replied, "Only one--but it is always the best move." The shrewdness of his answer is that it captures the idea that the quality of move selection is more important than search depth. In computer programming this translates to having better heuristics which is exactly what the current crop of programmers are focusing on and why I consider the eventual supremacy of the programs to be inevitable.

return to John Chamberlain's home · diary index
Developer Diary · about · · bio · Revised 22 February 2004 · Pure Content