Author: Robert Henry Durrett
Date: 14:49:35 05/14/02
Go up one level in this thread
On May 14, 2002 at 15:44:37, Vincent Lejeune wrote: <snip> >Post-match interview with GM Ilya Smirin > >"I also see room for improvement in the computer’s way of thinking in closed >position, which is quite primitive and naive!" > > <snip> >Some aspects of chess program’s >play are still calling for serious improvement; such as it’s incapability to >carry out a long-term plan, even if this plan is simple enough (ed - see >Smirin-Hiarcs game). I also see room for improvement in the computer’s way of >thinking in closed position, which is quite primitive and naive <snip> _________________________________________________ This makes me want to ask a question of the more capable chess programmers here: "What are the ways which chess programmers might approach the problem of correcting the deficiencies Smirin has discussed above?" i.e. "How to fix it?" I am not a chess programmer. However, it seems to me to be more a matter of general thinking here rather than algorithm development or coding. First come up with the "ideas" and then try to modify the programs. Right? For example, when I play chess [USCF 1856], I look at a position and try to figure out what sort of strategy would make sense. Take an elementary case: Suppose I decide [right or wrong, it's irrelevant] that a king-attack is called for. Then I will spend most of my thinking time looking at ways to attack my opponent's king. In other words, I FOCUS my analysis down to a limited range of lines to consider. I don't see why the same thing could not happen in the "mind" of a chess engine. Nowadays, with HIARCS and other chess engines supposedly using a lot of "chess knowledge" to evaluate positions, why cannot these engines reach conclusions such as "a king-attack is called for"? If this were to be allowed by the engine's programmer, then maybe it would be possible to program the computer to FOCUS it's search to a search for a king-attack. Intuitively, it seems to me that the computer would come up with a long-range king-attack plan in spite of itself! This could be generalized to other kinds of plans based on knowledge-based evaluation of positions. If several plans seem viable, the engine could look at all of them. Perhaps some of this is already being done in the more successful engines. I have no way of knowing. But Smirin's observations must be taken seriously. It must be assumed that the deficiency he thinks he sees is really there. It seems logical to infer that some method, such as the idea suggested above, is not being used, or not enough, in the modern chess engines. So, what is the way to fix these engines so that they can do a better job of long-range planning?
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.