Author: Stephen Ham
Date: 07:33:02 04/22/05
Go up one level in this thread
On April 22, 2005 at 06:01:31, Tony Nichols wrote: >On April 22, 2005 at 05:51:29, Peter Berger wrote: > >>On April 22, 2005 at 03:39:06, Tony Nichols wrote: >> >>>The programs perform as well as they do because they are very good at tactics >>>and most importantly they have huge opening books. I know this is a >>>controversial topic but if we really want to test the strenght of programs, then >>>have them play against strong humans without opening books. >> >>I consider the opening book of current top programs as one of their main >>weaknesses, and I am surprised that this argument is so popular. >> >>Even a player of about 1800-2000 will often know (important) things about his >>pet line he won't find in any computer opening book, not to talk about titled >>players or grandmasters. And the books also contain blunders the program would >>never play on its own. >> >>Opening books are still useful, but the opening book of an Anand, Kasparov or >>Kramnik is something completely different. > >Hi, Peter >This is a complex subject. On one hand opening books are a huge benefit to >computers when playing against humans. Most of the time they keep the engines >from getting a very bad middlegame. On the other hand bad book moves can kill >the best engine. I personally can attest to not knowing anything about my pet >lines:) Seriously, In my experience many masters play poor openings. So overall >a book is definitely a benfit for computers. >Regards >Tony Hello Tony, I think Peter is totally correct here, while your claims are illogical (you even contradict yourself). Your claims are: 1) "...they keep the engines from getting a very bad middlegame." Not always, Tony. Engines still reach bad middlegames with their books. Or, as is more common, the engine reaches a middlegame that's perfectly fine, but it doesn't understand it. For example, the position may involve an IQP. But unless the engine has been programmed with the heuristics for how to play such positions, then it won't know whether the IQP is an asset or a liability. It also won't know that in general, the player with the IQP should use it to generate an attack now, while the other player should think long-term, exchange pieces and play for the endgame. But had the engine been allowed to calculate its own moves without a book, then it would more likely reach a position that suits the engine. The offset is time. I remember when ChessMaster 6 went one-on-one with other engines of its day. ChessMaster's book was extremely shallow and it was often on its own after 5-6 moves. Instead its opposition had opening books that went for another 15-20 moves. Yet ChessMaster experienced no clear impairment in having to take the time to calculate its moves versus engines that played theirs instantaneously. 2) "On the other hand bad book moves can kill the best engine." Tony, you've just contradicted your #1 statement. Indeed, some opening books have huge holes in them. Hiarcs has this problem. And when its book doesn't have holes, then the moves may not be especially good anyway. Tony, opening books for engines are merely the compiled moves of human players. GM's generally already have memorized and comprehended these moves, and often have improvements prepared. Engines understand nothing and instead play what they are compelled to by their books. They may not have the heuristics to comprehend the positions once they arrive at them. They haven't prepared any improvements. They don't keep up with the changes in opening theory like strong humans do. In short, the engine with a human opening book is at a disadvantage here versus a strong human. So please re-read what Peter wrote, since he's correct on this issue. An engine's book is probably more of a liability than an asset versus a strong human. All the best, Steve
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.