Author: Uri Blass
Date: 10:09:56 03/07/01
Go up one level in this thread
On March 07, 2001 at 11:56:35, Ed Panek wrote: > Just a question to everyone: > What is the highest positive evaluation you have seen a program display and not >win ...and what is the lowest eval you have seen a program display and not >lose? The lowest evaluation when no lose that I remeber was mate against itself. The program of course had a bug. This evaluation was not in a comp-comp game but in an analysis of a position. Junior can show this wrong evaluation and play the wrong move(I posted positions when it happened) In the case of Genius3 it can even can show mate in 1 against itself when it is winning. I discovered the bug in Genius3 some years ago simply by finding a position when Genius3 show evaluation of mate in 2 when there is a mate in 3. I decided to change the position and I found that I can cause Genius3 to believe that there is a mate in 1 against itself when there is no mate against itself. I remember that the idea was simply to generate a position when Genius has an illusion that the bishop in the long diagnol protect a rook at h8. Genius3 probably analyzes some line in the tree like 1...Bg7 2.Rh1-h8+ Bg7xh8 3.Bxh8 and learns from this line that h8 is protected by the bishop. It finds that Bxh8 is illegal because the bishop is pinned by a rook at g1. It learns from this line that the king cannot go into h8 because h8 is protected by the bishop and it is not truth(the bishop at g7 is in the middle but Genius probably does not count it because it learned that this bishop cannot move). Genius3 has no better moves then Bg7 so it plays Bg7 with evaluation of mate in 1 against itself(the wrong reason is 1...Bg7 2.Rh8#). Only after playing Bg7 Genius3 understands that white has no mate in 1 and say mate against itself. Unfortunately I do not remember the position that I composed to prove the problem but it was published in the chess newspaper in Israel and people with the right newspaper can post the position(I am not going to look for it) Uri
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.