Author: Robert Hyatt
Date: 20:04:17 12/19/02
Go up one level in this thread
On December 19, 2002 at 13:18:23, Bob Durrett wrote: >On December 19, 2002 at 11:59:12, Robert Hyatt wrote: > >>On December 19, 2002 at 11:38:57, Bob Durrett wrote: >> >>> >>>The big challenge to all "serious" chess programmers seems to be to thoroughly >>>trounce the top grandmasters. That seems like a worthy goal, doesn't it? >>> >>>Barring some unforseen "miracle breakthrough" in chess engine design, will we >>>have to wait for the computers to get fast enough? Waiting for the silicon >>>monsters to solve our problems for us is the "chicken" way out. >> >>Why? That is one of the benefits of computers in general. They naturally get >>faster each year. Problems that couldn't be solved in real-time 50 years >>ago can be solved in real-time today. Traversing a game tree is a big task. > >Better pruning. > That is a way but it is significantly harder. I'm a cubic inch guy. I'd rather try to win a race with a 500ci stroker than I would with a 200ci blown, ported, polished, tweaked, 10KRPM engine. It is easier. >>You can either traverse the tree serially as a chess engine does, or do it >>mystically as the human mind does. But since we don't know how the human mind >>plays chess, emulation is impossible. > >The recent articles in Nature Magazine really make your point! The much touted >"rational thought processes" may have very little to do with chess. > >>And brute-force methods are still making >>inroads... > >Well, that seems terribly mundane to me. : ) > >> >>> >>>Rolf seems to suggest that the problem is that the top GMs can think better than >>>the chess engines. >> >>They are _different_. They learn in complex ways. Computers do not, yet. >>They match complex patterns quickly, computers do not, yet. They can >>generalize from a specific issue to similar positions, computers do not, >>yet. > >I like the sound of your "yets." Would you care to elaborate on them? Not much to say. Some would say "computers will _never_ learn." I have learned that never is a _long_ time. And I fully believe that as time goes on, computer algorithms will be developed that will go a long way toward allowing a program to reach the "self-modification" stage so that it can fix holes itself as they become apparent. But as to how long, I wouldn't venture a guess. Computers are about 60 years old. And they have come a _remarkable_ distance. Who knows what will transpire over the next 60 years... > >Also, could there be NEW alternative ways for the chess engines to match the >mental prowess of the humans? Maybe it's not necessary to "learn in complex >ways," "match complex patterns quickly," and "generalize from a specific issue >to similar positions." Those things sound too much like emulation. You have >already said, "Emulation is impossible." Maybe there are better ways for chess >engines to think. It's results that count. [True?] But don't misunderstand. >I'm not saying those things are bad. I just wonder whether or not ANYBODY fully >understands those silicon monsters. Trying to program them to emulate human >thought may be misguided, if not impossible. [Forgive me, all AI people!] It >seems much better to utilize their "natural talents" directly. Anyway, >emulators are mundane too. : ) > I think the basic problem is in understanding how _we_ do something. For specific applications, no matter how complex they are, we can develop software to accomplish that particular task. But only _if_ we know how we solve the problem. That is the missing link so far. We simply don't know how we do what we do, we just know we do it. > >Bob D. > >> >> >>> >>>Assuming he's right, what can be done to make him wrong again? >>> >>>Maybe the unsolved problems associated with planning, "positional positions," >>>and intelligent maneuvering in quiet positions are worthy problems? >>> >>>Maybe they, or similar problems, should be the focus of future efforts at >>>innovation? >>> >>>On the other hand, there is still the irritating fact that pruning schemes are >>>still not nearly as good as one might wish. If there are any REAL innovators >>>out there, maybe they can make some sort of "pruning breakthrough." Hopefully, >>>that would solve all of the problems of the chess programming world. >>> >>>Bob D.
This page took 0.02 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.