Author: James T. Walker
Date: 13:48:58 10/25/99
Go up one level in this thread
On October 25, 1999 at 12:15:28, Enrique Irazoqui wrote: >On October 25, 1999 at 10:32:34, Charles Unruh wrote: > >>On October 25, 1999 at 10:24:57, Enrique Irazoqui wrote: >> >>>On October 25, 1999 at 09:40:16, Charles Unruh wrote: >>> <snip> >"Che será, será". :) > >You might be right, and at the same time it is also possible that these 5 top >programs you mention wouldn't get a 2200 or 2300 rating in tournaments with FIDE >masters that know chess programs well. > >Maybe what happens is that we are all making the mistake of applying the Elo >system to programs. The Elo system predicts the performance of a human player in >a tournament against known human opposition and this simply does not work for >programs. The strength of programs is not transitive and their performance >against people and other programs is Elo-unpredictable. < ********************************** The ELO system cannot predict the performance of either humans or computers in a tournament. It can give you an expected result only. It is impossible to accurately predict the results in tournaments for Humans/Computers. For every example you give where the results did not match the expected results for computers I can give you 10 where the predicted results did not match the actual results for humans. There are too many variables to "Predict" the results. Over a period of time with many results you can determine the probability of certain results but the probability of error in short events like 4-6 games make the prediction a waste of time. You can come up with a "Rating" which is a measure of strength overall. It will not tell you that if you are 2550 that you will definitely beat a 2450 player on any given day or in any given event. There is no reason to believe that the prediction for humans is any more valid than for computers. You just have to recognize the limitations of the ELO system in rating people/computers. The only way to deternime if computers are capable of obtaining a GM title is to give them the same opportunity to achieve a GM Norm as any human has. Jim Walker ************************* >- Aegon 96: Dietrich Steinwender, rated 1850, scored 50%. >- Aegon 97: Jeroen Noomen, rated 2180, scored 50%. >- In the current GM challenge, a group of GMs and IMs with an average rating of >2531 scored 56% against Rebel. >- In these games, Rebel scored the same against Anand (2795), Rohde (2585) and >Sorin-Hoffman (2512). > >Etc. etc. etc. Nothing of this corresponds to what would have been >expected/predicted after the Elo ratings. > >I suspect that we are measuring the strength of programs using a tool that was >not meant for them and doesn't seem to work. > >Enrique > <snip>
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.