Author: Ed Schröder
Date: 09:23:57 12/11/99
Go up one level in this thread
Changed the subject, it's my favorite one these days...
>Posted by Jeff Lischer on December 11, 1999 at 11:09:56:
>
>In Reply to: Re: New rating list based upon Human games /SSDF brought back
>into line posted by Stephen A. Boak on December 11, 1999 at 04:21:56:
>
>On December 11, 1999 at 04:21:56, Stephen A. Boak wrote:
>
>>
>>3. It is not obvious that the relative rating of computer programs in comp-comp
>>play will hold for the same programs when they play humans at 40/2 time
>>controls.
Regarding this a few answers and questions from the CHESS2010 poll
from my pages (so far 329 entries):
Question 3: We have tried to explain that being the strongest chess program
against other chess programs doesn't necessarily mean that the program in
question is also the strongest chess program against humans. Do you agree
with this proposition?
I agree with the proposition 276
I do not agree with the proposition 31
ABSTAIN 20
The vast majority simply agrees that comp-comp is something different than
human-comp.
Question 8: If chess programs would have a FIDE rating would the order of
chess programs in comparison with computer-computer lists (such as the SSDF)
be different and to what extend?
SSDF vs FIDE would differ +/- 100 elo (or more) 111
SSDF vs FIDE would differ not much (+/- 50 elo) 112
SSDF vs FIDE would be in balance (+/- 20 elo) 23
ABSTAIN 80
I am quite surprised that only 23 people (that's < 10%) think that a comp-comp
list (SSDF taken as an example) would reflect a human-comp list. I agree too
with the proposition but there is no evidence to support it as it remains a
feeling.
>In fact, isn't Rebel Century a perfect example of this? Last year, after the
>release of Rebel 10C, Ed Schroder was involved in discussions on Rebelboard
>(and here, I think) regarding the differences in programming a computer for play
>against humans as opposed to play against other computers. At that time, Ed
>decided to optimize Rebel for play against humans, whereas Tiger would be more
>optimized for play against computers. In developing Rebel Century, he returned
>to Rebel 10B as his starting point.
>
>In doing this, Ed realized he would likely be sacrificing performance versus
>other computer programs (e.g. Fritz, Hiarcs) for the sake of improved
>performance versus strong humans. It appears that Ed has been successful --
>Rebel Century is doing well in the GM Series and may show a FIDE rating ~2500.
>It doesn't seem fair, however, to then turn around and say other programs would
>have even higher FIDE ratings based on the computer-computer games of the
>SSDF.
You are quite right. Another poll question as an answer:
Question 9: We have tried to explain that adding new chess knowledge which
makes a chess program a better positional player could lower the playing
strength in the comp-comp area and that removing chess knowledge which
makes a chess program a lower positional player on the other hand could
improve its performance in the comp-comp area, then how would you like
chess companies to handle this phenomenon in the future?
Do not compromize to the program's positional understanding 251
Do not compromize to the program's comp-comp performance 37
ABSTAIN 39
After seeing the overwhelming majority "Do not compromize to the program's
positional understanding" I felt safe enough to focus on human-comp for the
future. Sometimes Internet is just great :-)
Ed
PS, the full poll results can be found at: http://www.rebel.nl/resu2010.htm
Question 5 and 6 are pretty outdated, it would be nice to have them
recalculated now that Tiger tops the SSDF list and the fact Rebel has
played its games in the GM challenge.
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.