Author: Dann Corbit
Date: 11:26:01 06/21/99
Go up one level in this thread
On June 21, 1999 at 14:20:57, James Robertson wrote: >On June 21, 1999 at 13:48:03, Dann Corbit wrote: > >>On June 21, 1999 at 13:44:38, James Robertson wrote: >> >>>Could someone please calculate the odds that the 4 non-supercomputers would >>>score an even result with these 4 GMs? If the computers are just barely GM >>>strength (2500) and the humans are fairly strong GMs (2600), the humans should >>>score 64% on average. If the computers are rated 2450, the GMs should score 70%, >>>and the humans should score 75% against 2400 players. >>> >>>But the computers had white in 3 games, and white scores 60% against black. If >>>the odds were calculated, I'll bet it would be something like the computers have >>>a 35% chance of scoring the result we saw. >>With the number of games played, the uncertainty is so high that I do not think >>you can provide a reasonable calculation. Look at how many games the SSDF plays >>before they even post a result. There is a *reason* for that. > >I know, if we are trying to do some sort of confidence interval. But these >percentages are _established_. IE if we flip a previously unflipped coin, we >still are able to predict the odds that we will get heads. Anand has never >played Morozevich, but because of their ratings we can predict Anand's >probability of success. > >It should be possible to do that here. Sure, but with Anand and Morozevich there are *thousands* of games against rated opponents. Here we have 7 games per program. Quite frankly, we can't say anything with certainty about how strong any of them are. Perhaps you could say that the ELO of the programs is 2600 +/- 2000. How's that?
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.