Saturday, 8 November 2008

Karpov, Anand & Computers



Karpov has recently made comments about Anand's current playing style being more computer-like (allegedly). This has been discussed at the excellent chessblog by Susan Polgar (and specific article link). The most useful comment for me was that by KWRegan which I enclose below.
KWRegan said...

"Judging creativity is hard to do without being subjective. The closest I see how to do objectively is to measure setting problems for one's opponent.

Sullivan's study (comment #4 by alchessimist) includes a measure of "complexity", and scales the raw deviations from Rybka+Crafty's best move down in complex positions. I have been developing a more-detailed model that (when finally tuned) will compute expected deviations for both the featured players and their opponents. The higher the latter, the more problems set for the opponent. My model requires taking data for (at least) the top ten moves in every position, so my engine---a special private version of Toga II which I know has not been regressed against GM games---needs to run far longer per move to reach depth 18 as in Sullivan's study.

Thus a "dynamic" player with a higher error rate may benefit if he/she also created a higher expected rate for the opponent. It will be interesting to see whether this work restores the likes of Tal and Alekhine (at their peaks) to the parity their over-the-board reputations would suggest. Alternatively, it may reveal that "positional" players are also the cagiest."
I have previously blogged (here) about Sullivan's excellent research (link) and very much look forward to seeing the results of Regan's analysis. The idea of analysing the effects of dynamic play, as measured by problems set for an opponent, is an interesting one. Dynamic play is likely to increase the effect of performance psychology, time pressure, and possibly the Einstellung (set) effect. To me, dynamic play is what makes chess a sport.

Tags: Anand - Computers - Creativity - Karpov - Sullivan
Posted by ALCHEssMIST - Alchemipedia alliance.

No comments: