Author: Mike S.
Date: 10:37:09 11/25/01
Go up one level in this thread
On November 25, 2001 at 12:27:10, Victor Fernandez wrote: >On November 25, 2001 at 11:17:26, Mike S. wrote: > >>I wonder what sophisticated definition of "playing chess" it is, today's >>programs aren't capable of in your opinion? >Hi Mike ! > >Of course. I am not a programmer, but I believe that in chess it is >continued programming like 50 years ago. New techniques have been >introduced but the chessprograms continue with the rules A and B of >Shannon. The chessprograms continue being blind, they specify to analyze >of 40/40 moves and they don't plan, they continue smashing with the >"horizon effect" the chessprograms have that to begin to break the >"horizon effect", using "chunks", planning to long term and to use >algorithms to examine only 1 or 2 positions in each move discarding >the other ones very faster, in another case will never end up playing >chess well. The number of types of positions and ideas chess programs don't handle well, is shrinking rapidly IMO. Do you have examples? Furthermore, I doubt that there are that many plans ins human chess, which last longer than, say, 7 or 8 moves. I don't mean positional or strategical ideas, I mean plans involving certain maneuvres across a limited number of moves. Programs can see ~14 plies deep today in 40/2h, the extensions much deeper. Did you ever have I plan longer than 6 or 7 moves? Computer chess has long since overcome the times when many simple things were not seen by the programs. Now, there's more knowledge for special cases, bigger depths, better selectivity, better evaluation... that's why they can perform 2600+. I hear always, program X has saced a pawn for the initiative, program Y has saced a piece for a promising attack, etc. - I've seen examples of tremendously brilliant endgame technique Capablanca would have been proud of... (I don't mean tablebases but calculated moves). OTOH, I know what you mean by long term planning. Sometimes ago, I used a (very simple) example to illustrate what type of "chess thinking" so to speak, doesn't fit to the way chess programs do it: Yates,F - Nimzowitsch,A [A07] Karlsbad Karlsbad, 1923 [D]r4r1k/p5pp/b1p5/3pNp2/3Q1P1P/q2P2P1/P5B1/1R4K1 w - - 0 27 Here, an idea is to bring the white Q to h5, thus creating the threat Ng6+ (the other idea h4-h5-h6 played by Yates wasn't successfull). The Qh5 idea involves giving up the pawn d3 later on. I've tested this in 1999, on PII/333 and none of the top programs tested went for Qh5; I didn't expect them to do so. Maybe now a program chooses it (I'd be surprised). I think going for Qh5 is a better try than with the h-pawn, but it is refutable too. Maybe someone else has a better example of this type of idea. It should involve some maneuvring moves before the "target" position can be reached. Regards, M.Scheidl
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.