Author: David Dory
Date: 15:21:23 05/19/02
Go up one level in this thread
On May 18, 2002 at 07:48:54, Robert Henry Durrett wrote: >On May 17, 2002 at 21:08:20, martin fierz wrote: > >>On May 17, 2002 at 18:51:47, Robert Henry Durrett wrote: >> >>>I hope you will forgive me if I lack diplomacy. I never was good at that. >>> >>>It was not I who reopened an old can of worms. It was GM Ilya Smirin. >>> >>>The essence of my recent posts is to say that I was hoping that his observations >>>about how the chess engines performed would be discussed here. So far, nothing. >>> I was merely trying to prompt someone to address the GM's concerns. >> >>there is a good reason that there is no response to smirin's statements: what he >>said is just common knowledge here. if i remember right, it was about this: >>- computers are great in open positions >>- computers are weak in closed positions >>- computers have no long-range planning capabilities. >>this is *really old* stuff! it has been this way all the time, and will probably >>be this way for a long time to come... >> >>aloha >> martin > >Thanks, Martin, for a courteous, straightforward and honest answer to my >question. Your comments that "this is common knowledge here" and "this is >*really old* stuff" have a ring of truth about them. They certainly explain why >noone addressed Smirin's observations about the chess engine he was playing >against. > >I have been going thru the archived files in search of earlier discussions of >this topic and have not yet found them. I am still looking. If you remember >any of that, I would appreciate your help in locating them. > >If the files I'm looking for are REALLY old, then one must wonder whether or not >there has been any recent progress made in finding the solutions to these >"really old" technical problems. But perhaps people will not present their new >stuff here because they are already "burned out" on this topic. Robert, This has been a topic since computer chess was created. Search for "long range planning" "how to beat a computer", "anti-computer chess", "horizon effect", "stonewall attack", or consult any reputable book on computer chess. Search extensions which add a semi-intelligent touch to the ends of the tree search if the position warrants it (and what warrants it varies, but certainly a check, a new capture, a sudden drop in the score of the resulting position, etc., should trigger this extra search) have helped solve this vexing problem of the horizon effect. To say that the computers have no long range planning is not quite true. Their search with modern computers and closed positions is not as good as some humans' long range planning. To be honest though, the best humans' get beat about half the time when they try their * superior *, long range plan, so just how good is the human long range planning? "Great" plans which lose, just really aren't that great, no matter how glowingly the losing GM/IM describe them. The fact that the computer does it better in open positions can be no surprise, since it's search is most effective there. There is no "fix" for this. The problem will be lessened by better hardware, and better search & evaluation knowledge being painstakingly added to the search. This knowledge can't be added too quickly, because it's effect is to slow down the search, which in turn makes the program weaker. For a much better understanding of the problem, see if you can get your hands on "Chess Skill in Man and Machine", by Peter Frey (Springer-Verlag), 2nd edition. The chapter on "Paradise" (a long range planning chess program), is unique, and the whole book is invaluable, IMHO. David
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.