Author: Mike Hood
Date: 07:42:43 11/21/05
Go up one level in this thread
On November 21, 2005 at 07:52:35, Tord Romstad wrote: >On November 21, 2005 at 04:28:47, Martin Baumung wrote: > >>Hello Uri, >> >>>How much time does your program need to find Kxb3 that is the only practical >>>chance to save the game >> >>please enlighten us on how the game can be saved after Kxb3 - especially when >>using the EGTBs. > >Uri did not claim that Kxb3 was a draw, only that it was the best *practical >chance* to win the game. Big difference. After Kxb3, white has to find a >very hard move in order to win. Many opponents will not be able to find >the right move (I almost certainly wouldn't). After all other moves than >Kxb3, the win is trivial. Even I would be able to win. > >Another, less easily noticable problem of EGTBs is the general slowdown >of the program. It is very hard to guess how many wins a program misses >because it wastes too much time probing tablebases when a slightly deeper >search would have found an easy win. > >It is just not possible to prove the usefulness of EGTBs in practical >play with a handful of positions. The only way to measure the usefulness >is to play a big number of games with and without EGTBs and compare the >results. So far, the overwhelming evidence points toward the conclusion >that EGTB use is utterly insignificant in practical play (consider, for >instance, the difference of 2 Elo points between Fruit 2.2 and Fruit 2.2.1 >on the CEGT list). > >I am not a non-believer in EGTBs, by the way (in fact, I doubt that such >"non-believers" really exist). The existence of 5-piece and 6-piece >EGTBs is a tremendous contribution to the body of chess knowledge, >and arguably one of the most exciting advances in chess theory over the >last few decades. I am just very disappointed by the immense lack of >imagination current chess programmers (myself included) display when >trying to use this new body of knowledge. > >Tord One side-effect of EGTB's is that computers have a tendency to make "non-human" moves. This is especially apparent when using a full set of 4-piece EGTB's, with no 5-piece EGTB's. I have witnessed KRNB-KP endgames where White exchanges its rook for the pawn in order to get into a comfortable KNB-K tablebase with a mate in 30 or more. Very few humans would take a risk like that. Even a very good human player would manoeuvre his pieces carefully to capture the pawn without an exchange, then use his rook as the main mating piece. Neither approach is wrong. The computer's choice might even be a longer mate, but it's easier (faster) to find than searching the tree for the shorter mate with the rook.
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.