Author: Dann Corbit
Date: 14:07:51 06/21/99
Go up one level in this thread
On June 21, 1999 at 16:39:30, James Robertson wrote: >On June 21, 1999 at 15:57:13, Dann Corbit wrote: > >>On June 21, 1999 at 15:42:24, James Robertson wrote: >>[snip] >>>I didn't make myself clear. If the programs are GM strength then they must be at >>>least 2500. I wanted to see what the odds are of a 2500 group of players againt >>>a 2600 group of players, or if they are just IM strength, what are the odds >>>then? >>I am quite sure that I do not understand your question. But the point scoring >>expectancy for 100 ELO difference is 36% -- does that help? > >In the first post I mentioned that the 2600s should score 64%. :) > >> >>If you hope to prove that the programs are GM strength from this I guarantee you >>that it is completely hopeless. > >Exactly the opposite... many people are using this as 'proof' that comps are GM >strength. I wanted to see what the odds are that they, if they were rated 2400 >or 2500 could make this result against four 2600 GMs. There are exactly 4 games where the strength of one opponent was known. In each case, the strength of the computer program was unknown. If anyone thinks you can draw a sensible conclusion from a *single* data point, then I submit that they are mistaken. If, on the other hand, the programs had gotten crushed 0-4 (which has a 2% chance against _actual_ 2600 foes) then people would be suggesting that "The programs just are not as strong as GM's yet." In either case, both assumptions are simply wild and unfounded extrapolation. What the math says, when you examine this experiment is "Not enough data." If we have a lot of experiments like this (and, importantly, where the Computer setup does not change) then we can begin to draw valid conclusions. I wish that all the computer chess vendors would do like the Rebel team and play GM's at tournament time controls so we would actually have valid data upon which we could draw sensible conclusions.
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.