Author: Chris Carson
Date: 07:13:53 06/16/99
Go up one level in this thread
On June 16, 1999 at 09:58:45, William H Rogers wrote: >On June 16, 1999 at 07:59:57, Chris Carson wrote: > >>On June 16, 1999 at 05:36:21, alfred palang wrote: >> >>>Planning...this is where the big hole is with our programs. Tactics: programs >>>execute them with astonishing accuracy. Positionally: we need more >>>improvements! >> >>I agree with your assessment on "Tactics" and "Position". >>My opinion (+ $2.00 will get you a cheap cup of coffee, notice >>it will cost more with my opinion) is that "Positional" advantages >>create tactics, and many times the tactics are beyond the horizon >>of the computer (or the person these days with deep searches by >>many programs ie: crafty, fritz, junior on 4 processor machines). >> >>I consider planning by a computer to be partially the depth of the >>pv and planning improvements to be the continued searching for better >>moves (yes this is tactically based and not all inclusive), but with >>faster machines and broader searches, like Deep Blue, computers will >>become better planners and "Positional" skills will also improve. Just >>my view and we are a long way from the type of speed needed to make >>this happen unless you use Application Specific Integrated Circuits >>(ASICS) like the Deep Blue team did. >> >>I wonder if any research has been done to determine at what point >>programmed knowledge can be removed because of deeper searching (if at all)? > >Some work was done on this years ago. A well know program just went to great >depths but it was later decided that without chess knowledge, the depths were >not very useful as the program did not play very strong so more knowledge was >added. >In theory, if you have enough knowledge, then you don't have to search deep at >all. Knowledge implies that your initial move selection will be the best! >Trying to implement that is what chess is all about. > >>I am thinking of simple knowledge like avoid Bxh2 (if B becomes trapped). >>Just a thought, maybe knowledge should never be removed, but my opinion >>is that a "static" evaluation will be all that is necessary for deep >>searching, I just do not know what ply level is deep enough to remove >>knowledge (might be a futile quest). My program has only a "static" >>eval and is very fast, but it gets killed by Crafty (which has a very >>nice balance of great eval and speed) and CM6K. My programs end game >>is dependant on search depth + tablebases (a form of knowledge) and >>learning (another form of knowledge). Oh well, I get a kick out of watching >>it speed by, guess I also love drag racing. :) >> >>Best Regards, >>Chris Carson You make a very good point. I think I know the program you are refering to, I have the book at home (by Newborn), but the name escapes me. Darkthought and Crafty have both done some deep thinking and noted that move selction continues to change with depth. I agree that knowledge is important, move selection is a good observation, I still wonder if all knowledge is necessary at 25 ply (some still is, just not sure what type, different in middle vs end?). It seems that tactical knowledge might be the easist (not sure though) to replace with fast hardware? Your comment reminds me of a quote (author unknown). When Capablanca was world champion he lost an exhibition game to a local club master. The story goes, a reporter asked Capa how far ahead he looked during the game, Capa replied "10 moves". The reporter asked the club master the same question, the club master replied "one, but it was the right move". :) Just remembered, was the program named Belle? Best Regards, Chris Carson
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.