Author: Tim Mirabile
Date: 14:53:49 05/11/99
Go up one level in this thread
It's important to understand that a chess rating does not correspond to some absolute level of playing strength. Humans play weaker chess (i.e. weaker moves in an absolute sense) when there is less time to think, and so do computers. When humans play in action chess events, they usually play other humans, and on average, the various playing strengths of human players drop at roughly the same rate as the time control gets shorter. So two humans who are 2300 at slow time controls will probably be roughly equal (+/- 50 points) at action chess. Since chess rating scales are arbitrary, an effort has been made to make human rapid ratings conform roughly to human slow time control ratings. Computers, on the other hand, do not drop in strength nearly as rapidly as humans do when the time control is shortened. So for a human and computer who are about equal at 40/2, playing at G/30 the computer will seem much stronger relatively, and calculating a rating for the computer at this time control using the humans slow time control rating will result in a very inflated rating relative to the human's slow time control rating. The amount is surely dependent on the hardware speed of the machine, with faster hardware machines decreasing less in strength with faster time controls than slower machines. This is due to the exponentially increasing time needed to search an additional ply as you get deeper in the search, along with the diminishing returns of searching that additional ply. This is why it is misleading to use G/30 results to rate chess computers (common with dedicated chess computers). True, most people probably do play against these computers setting them at a rate of play roughly equal to action chess, but they don't force themselves to play under the same time limit (or under any time limit at all). So what you get is a computer rating which is very inflated compared to the human's slow time control rating, plus the fact that the machine is essentially giving time odds. The result is that machines played under these conditions are not anywhere near as challenging as the G/30 rating would make it seem. On May 11, 1999 at 07:26:52, odell hall wrote: >Hi CCC > > I have heard many people say that computers would lose if the time control >were slower, for instance this match with Gelfand. I know that among humans a >person rated 2300 at action chess, will be around the same, or within 50pts at >40/2. >What evidence is there to suggest that a computer that plays at 2800 level at >action will play 300 or more pts weaker at 40/2 ? I believe I rememberLarry >kaufman saying that the actual difference between action and 40/2 chess was >about 85 pts for computers, although I do not remember what his evidence to >support this was.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.