Computer chess

Yesterday morning, I happened across the third match between Kasparov and X3D Fritz and got to watch some of it live (although they cut away for women's basketball). That ended up being fascinating to watch, so I've since gone through the web site and read through the previous games with commentary.

Apart from making me wish I knew more about chess, I also find it really interesting how computers have developed into excellent chess players, but of a completely different kind than human players. The fundamental chess playing algorithm is still the same as it was back when I studied computer chess programs some ten years ago -- develop an algorithm to rank board positions and then do a search to find the move that stands the best chance of achieving the top-rated board position that you can see. This makes computer players absolute tactical wizards, and masters at playing the endgame, but it means that they can only plan as far as they can analyze. They really play nothing like a human (and chess really isn't that good of a proof of "intelligence" because of its partial susceptibility to pure brute-force search); they're excellent at springing on any mistake and pursuing it relentlessly, but long-term planning escapes them.

The third game of the match is a masterpiece of a human grand master playing against a computer and completely trapping it in its flaws. The computer had no idea the long term plan that Kasparov was developing until far too late, as Kasparov patiently played into a closed, restricted position and then set up for a long-term strategic push. The second game of the match was also an excellent example, as Kasparov was in a good position until he made a human blunder and the computer pounced mercilessly.

I expect that as our sheer processing power becomes stronger, eventually chess computers will outplay the best humans because their planning horizon will slowly get far enough ahead. It may take years, since I think an exponential increase in processing power is required, but I'm sure we'll see that.

It's sort of disappointing in a way, though, that the basic algorithm remains so simple and we keep just throwing more processing power at it. The small computer science researcher in me would be far more interested in trying to understand better what a human player does and teach the computer more about how to do that, rather than just relying on its obvious strengths.

Anyway, I recommend taking a look at the games, and the fourth game is on ESPN2 tomorrow.

Posted: 2003-11-17 21:09 — Why no comments?

Last spun 2013-07-01 from thread modified 2013-01-04