Chess positions too complex for engines


Back to forum

Thibault de Vassal    (2010-12-13)
Chess positions too complex for engines

This topic may last a while but it could be interesting after all.

Why not trying to gather (and discuss) as many chess positions as possible that computers are still unable to solve with the best engines even working during days... It may help us to detect more weak points in the most recent engines.

Okay, I start with these ones:

Schuster-de Vassal : 1-0

Peter played very well in this one but as far as I can remember this position was an easy draw... The plan was to keep the rook on the e column and White cannot progress. I knew that but at this moment, I had to play this move quickly and I made a stupid mistake: I trusted the engine... Rb8+?? 1-0 .. my comp was really a prehistoric one there but I wouldn't be surprised if the recent ones were still not able to avoid to play this Rb8+

The next position is less interesting but quite funny, and it looks like that the newest engines STILL do not understand it... Actually it shows +5.00 or even more while it is an EASY draw for any human player...

Utesch-de Vassal : 1/2-1/2

Amazing, what do you think? :)

Josef Riha    (2010-12-14 09:21:40)
Chess positions too complex for engines

Hello Thibault, it should be Rb1+. Right?

Thibault de Vassal    (2010-12-14 20:55:29)
Chess positions too complex for engines

Haha, I'm really lost in this game :) Yes, Rb1+ of course. Thanks!

Hannes Rada    (2010-12-15 21:56:00)
Chess positions too complex for engines

Hi Thibault,
Thank you for the examples.
I do not agree with your position against Peter Schuster. Almost every engine is now playing Qe2 i.o. Tb1 ?
So the engines now 'understand' this position a little bit.
Your position against Wolfgang Utesch is a classical fortress and indeed a bit funny. Here all engines show +5 and Rybka even +6 .... :-)
The 'tragedy' here is not the final position, but the fact the engines would go for such a position in their analysis, as it seems to be a clear win for white. I would file this under typical missing endgame understanding.
I.e. Rybka still does not understand wrong bishop endgames....

Thibault de Vassal    (2010-12-15 23:06:54)
Chess positions too complex for engines

Hi Hannes, I'm not sure if Qe2 is good or not, I'll look at that... Well, I posted these 2 positions quickly after your message in the other discussion, but I'm quite sure that I can find tens of positions like this if I have a deeper look at my games... To be continued.

Scott Nichols    (2010-12-15 23:20:40)
Chess positions too complex for engines

Hi Thib. This is one of the main games I had in mind when I responded to your quote.

Or Game #40749

After 50. ...Be2 Rybka could only think of lines keeping the King close to the g4 pawn. Try it yourself anybody, the engine just didn't get it. It was a blitz game, but even at that time control I could see that my black Bishop could guard against the pawn advance from afar. So the winning strategy was to march the King to the other side of the board and escort the a-pawn to the queening square. This idea obviously was far beyond the engines horizon. After that game, my respect for Rybka's endgame play went down considerably.

Philip Roe    (2010-12-16 03:31:35)
Chess positions too complex for engines

This ending is taken from Nunn's Chess Endings (Om. Garcia-Otero,Cuba 2002).
ChessPosition (see diagram)

White, to play, can force a Queen and Pawns ending that would be a long slow win, but he can also make a very profound triangulation that leads much more forcefully to the win. Nunn was very impressed that White found this at the board.

I would be interested to know how the engines do on this position.

a b c d e f g h

Thibault de Vassal    (2010-12-16 09:30:34)
Chess positions too complex for engines

Pawn's endgame?! I would be surprised if recent engines don't get it yet.

Josef Riha    (2010-12-16 09:49:07)
Chess positions too complex for engines

Exactly, don't forget the engines got informations about any given position in the endgame and at some point in the middlegame from the endgame tablebases whether it's a win, draw or lose.

Thibault de Vassal    (2010-12-16 09:54:53)
Preparation in sicilian

Another game I remember, the typical win by preparation (I was absolutely not prepared in this variant yet) in a blitz game - Eros & Alberto made it well with this line during the 1st Freestyle tournament.

I don't remember what move is the point exactly at a first sight, somewhere around Qxe5 but Black did not expect what kind of endgame will happen after the next 20 moves, btw many are forced but the line may be too long for most engines. The game was lost already, and I'm not sure if engines choose the right way (wouldn't be a trap anymore)...