Forums

A study computers can't solve

Sort:
JoachimJo

Hi everyone.
As you all know, computers have gotten far better than humans at chess. But there are still some positions thay can't understand but we do.
Here is a study of mine that our silicon friends can't solve, but maybe you can ?

White to play and win.
barbarachambers1611

good player is win in ther game but if are new player then you are not win but you learn both gane rule and life relu

JoachimJo

I'll give the solution as I understand it's not easy to solve.



Arisktotle
JoachimJo wrote:

I'll give the solution as I understand it's not easy to solve.

Wonderful! I had in mind to give it a shot but I am bit in overload at the moment.

sameez1

@ JoachimJo  If I had a chess engine I would be checking that out.I am assuming that to fool a computer it has to be a move that would look silly at first. I avoided the answer i just have to find it.Thanks 

JubilationTCornpone

Interesting.  This position has 35 visits at playchess.com.  Not sure if you posted it there too but people are looking at it.

Consensus appears to be 1.Nde3.  My Komodo 10.2 scores it as +1.88 which is enough that it should be decisive but doesn't see to the end.  Someone else's Stockfish 8 calls it mate in 24, but it looks like it was allowed to run for hours.  Maybe I'll let Komodo run a few hours and see if it finds it...

JoachimJo

Actually Stockfisch 8 finally found it, after reaching depth 77... but less say any 2000+ player can solve it faster than the computer.

JubilationTCornpone
JoachimJo wrote:

Actually Stockfisch 8 finally found it, after reaching depth 77... but less say any 2000+ player can solve it faster than the computer.

Well, in fairness it found it was a forced win in 24.  But it had been recommending the same key move for hours (and all the engines I've tried recommend that key move).

 

Are you an engine expert?  Why does it take a depth 77 search to know that something wins in 24?

 

Oh and ... how long does a depth 77 search take anyway?

JoachimJo

Finding the 1st move, which is forced in order not to lose in one move, is not the same as finding the whole solution.... When Komodo says +1.88 it only tells it doesn't know how to break the forteress and plays useless waiting moves...
I'm no engine expert but a FM and it's my study... I don't know why it had to reach depth 77 to find the solution, but apparently it had to... And yeah, on a normal comp, it takes a long time.

Arisktotle

I'm not an engine expert and leave the ply question to others. It is however not strange that the engines would recommend the same (and correct) key move even after superficial analysis. The alternatives are clearly inferior and will be rejected quickly, leaving us with only one viable candidate. That does not imply the engines have a clue where the solution is going! Oops! Duplicate answer.

JubilationTCornpone
JoachimJo wrote:

Finding the 1st move, which is forced in order not to lose in one move, is not the same as finding the whole solution.... When Komodo says +1.88 it only tells it doesn't know how to break the forteress and plays useless waiting moves...
I'm no engine expert but a FM and it's my study... I don't know why it had to reach depth 77 to find the solution, but apparently it had to... And yeah, on a normal comp, it takes a long time.

Hmm.  OK.  I don't really do studies very much but now I'm curious.

If a human says "AFter 1.Nde3 Black has no good defense" ... is that sufficient?  Do you want them to say *why* Black has no defense?  Or do you expect a line all the way through to mate?  Or is there some key move down the line that needs to be found (since the first one is forced)?  Just wondering what's considered to be sufficient to say that a person (or engine) has solved a position?

JoachimJo

I've shown this position to many IM and GM including MVL, they all went like 
"Ok Ne3 dxe3 Qf5 then black waits... what's the idea ? Oh ok, Put K on a1 then Qb1-e1 and K back to d1...Now what ? Nxe3 Rxe1 Kxe1 Kg1... and now hmmm... ok I see the mate".
This is solving for a human being. For a computer, it's when they switch from something like +1 playing endlessy stupid waiting moves after 2.Qf5 to showing the mate.

JoachimJo

One consider a study solved when he doesn't have anymore to find only (forced) moves to reach the goal (win or draw). Here it stops only when black is mated, because otherwise it's a draw.