Author: Roy Eassa
Date: 13:04:28 10/11/02
Go up one level in this thread
On October 11, 2002 at 15:45:23, Otello Gnaramori wrote: >On October 11, 2002 at 15:34:54, Roy Eassa wrote: > >>On October 11, 2002 at 15:26:27, Otello Gnaramori wrote: >> >>>On October 11, 2002 at 15:00:05, Roy Eassa wrote: >>> >>>>I have very gradually come around to the idea that what makes a chess computer >>>>good against other chess computers may be quite different from what makes it >>>>good against strong human chessplayers. >>>> >>>>Some years ago, PCs were slow enough that the chess author had no choice but to >>>>write the program to maximize the search, or else even moderately strong humans >>>>could win simply by tactics. But I think now, with PCs over 2 GHz, just 25% of >>>>the computer's power is more than sufficient tactically against humans. Against >>>>other computers, every ounce of speed must be used to search deeper, as in Fritz >>>>or Ruffian. But against humans perhaps the great majority of the power of the >>>>CPU needs to be used exclusively to play anti-human chess: avoid locked >>>>positions, avoid allowing certain types of attacking formations, "understand" >>>>many, many types of positions better, etc. Such a program would likely perform >>>>very poorly against the likes of Fritz but could perform much better than Fritz >>>>does against top humans. >>>> >>>>My thought: there should be two totally different classes of chess programs: >>>>those that are designed to win against other programs and those that are >>>>designed to win against humans. And if you want to create a program that claims >>>>to do both, you should have it swap in a completely different set of algorithms >>>>-- and not just change a few settings -- depending upon the opponent (human or >>>>computer). >>> >>>Hi Roy, >>>I think that this idea isn't completely new to CC, in fact e.g. Rebel by Ed >>>Schroeder works in anti-gm mode without decreasing the score against other >>>machines. >> >> >>But my thinking is not a simple checkbox or small change to the program, but >>rather a completely different set of algorithms. Basically, a completely >>different program with entirely different logic. > >I see your point, but I guess that too much eval complexity (read understanding) >is probably negatively impacting on the search speed and depth. >This is a topic already debated in this forum. Yes, it would certainly negatively affect the search. But additional search is not what makes the computer succeed AGAINST GM HUMANS! The search makes it succeed against other computers. The search has been "fast enough" for humans for several years now. It may (*may*) be time to stop wasting the CPU power on the search (except for as much of the CPU power as PCs had maybe 7 or 8 years ago) and devote it to something _completely_ different. For an anti-human program on a 100 MHz computer, you need to devote all the power to the search. But for the same program on a 2000 MHz computer, maybe you should devote 100 MHz to the search and 1900 MHz to beating humans. That's what I'm proposing. (But if the opponent is a computer, then do things as they are done today -- search, search, search!)
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.