i also once wasted some life time finding out the legal drawn position with the biggest advantage according to computers, and i figured out this one:
Stockfish gives +48.2, aka white is up more than five queens, but after 1...Bg4! it is obviously a draw. However, Stockfish sees the right moves for black here, so it is just an example for a misleading engine evaluation.
Stockfish gives +48.2, aka white is up more than five queens, but after 1...Bg4! it is obviously a draw. However, Stockfish sees the right moves for black here, so it is just an example for a misleading engine evaluation.
@Linnemann Wow that's a nice position..... how long did it take you to make that? I agree, this doesn't show a glaring fault in engines (they play the best moves), therefore I struggle to understand why people see these sorts of positions as proof that the human brain must be some hyper-advanced mega-super-extreme-multidimensional-quantum-physics-defying-ultracomputer (delete where appropriate :D ) in order for most people to immediately see that such positions are drawn.
@mCoombes314 lets add some bishops and you see that the computer suddenly has big problems finding the right move:
1...Bg4 is still the only move that draws, but the computer wont find it.
If this knowledge, which lets us understand fast that this position is draw and how it is draw, would be trivial, it would be easy to code it into a computer - by a human of course, computers cant do that (which has been proven). But it is not trivial at all.
Just try to give a precise verbal definition of why the position in #12 is a draw. It would start like:
"Some white pieces can not move because a black bishop on g4 (a white square) blocks them. The other white pieces can never attack that black bishop. They also can not mate the black king. So white can not make progress. On the other hand black has not enough pieces to force a mate without white allowing to free his pieces. So no one can make progress, therefore it is a draw."
Now in order to develop an algorithm who spits out '0.00' at the end, one must define what it means 'to be blocked', what it means to 'not be able to attack a square', what it means that 'the opponents king can not be mated', what it means 'to make progress', how much pieces are enough 'to force a mate', and why all these combinations result in draw.
It is very complex to develop such an algorithm, especially if it should be generic. Probably it is not possible at all.
But Humans do exactly that. Fast. No one knows why and how, but they do. Thats a miracle.
1...Bg4 is still the only move that draws, but the computer wont find it.
If this knowledge, which lets us understand fast that this position is draw and how it is draw, would be trivial, it would be easy to code it into a computer - by a human of course, computers cant do that (which has been proven). But it is not trivial at all.
Just try to give a precise verbal definition of why the position in #12 is a draw. It would start like:
"Some white pieces can not move because a black bishop on g4 (a white square) blocks them. The other white pieces can never attack that black bishop. They also can not mate the black king. So white can not make progress. On the other hand black has not enough pieces to force a mate without white allowing to free his pieces. So no one can make progress, therefore it is a draw."
Now in order to develop an algorithm who spits out '0.00' at the end, one must define what it means 'to be blocked', what it means to 'not be able to attack a square', what it means that 'the opponents king can not be mated', what it means 'to make progress', how much pieces are enough 'to force a mate', and why all these combinations result in draw.
It is very complex to develop such an algorithm, especially if it should be generic. Probably it is not possible at all.
But Humans do exactly that. Fast. No one knows why and how, but they do. Thats a miracle.
lets see, what I've got:
de.lichess.org/analysis/fromPosition/N2b4/7B/6B1/6p1/5pP1/4pP1B/1k2PRBN/3KRQQQ_w_-_-_0_1
very interesting^^
de.lichess.org/analysis/fromPosition/N2b4/7B/6B1/6p1/5pP1/4pP1B/1k2PRBN/3KRQQQ_w_-_-_0_1
very interesting^^
In these positions though, the evaluation number doesn't affect the computer's play in any way. Playing Stockfish against itself in this position results in a draw, and a few moves after Bg4, the evaluation drops to 0, even with White playing the best moves according to SF. So, while the computer doesn't see the draw straight away, it will do after a certain number of moves. Looks like the horizon effect to me.
I suppose the fact that the human brain can instantly prune lots of useless lines and therefore tell that the position is a draw is interesting, but saying that "computers can't find the draw" is wrong - they can find it, it just takes them a while to realise it's a draw, probably due to depth limitations, and/or the idea that all lines must be treated as feasible until a certain change in evaluation shows a line to be bad.
I suppose the fact that the human brain can instantly prune lots of useless lines and therefore tell that the position is a draw is interesting, but saying that "computers can't find the draw" is wrong - they can find it, it just takes them a while to realise it's a draw, probably due to depth limitations, and/or the idea that all lines must be treated as feasible until a certain change in evaluation shows a line to be bad.
@Eldrail
Nice, you beat my record by ten pawns. I have pressed some more pawns out of this idea, now at +66 (seven queens and a knight) eval in a legal drawn position:
de.lichess.org/rDWuWuFD
Who can beat this?
@mCoombes314 of course it affects the computers play. See my example #14. Also, you can find the best move by pondering random moves, it is just a question of time. That is not the problem. The problem is to get the correct eval as soon as possible, aka 0.00, because only then the algorithm 'scales'. And the human algorithm does scale.
In other words, the computer must understand whats going on. More Knowledge will increase his skill much more than just more brute force.
Nice, you beat my record by ten pawns. I have pressed some more pawns out of this idea, now at +66 (seven queens and a knight) eval in a legal drawn position:
de.lichess.org/rDWuWuFD
Who can beat this?
@mCoombes314 of course it affects the computers play. See my example #14. Also, you can find the best move by pondering random moves, it is just a question of time. That is not the problem. The problem is to get the correct eval as soon as possible, aka 0.00, because only then the algorithm 'scales'. And the human algorithm does scale.
In other words, the computer must understand whats going on. More Knowledge will increase his skill much more than just more brute force.
@Linnemann I don't feel like your solution works for #6. Black could just run the king out after the first check, no? Then get the rook into the action if pawn ever takes.
@sedore, you are right. Kc7 busts my study. Im so sorry.
Here is a fix attempt, the solution is the same.
Here is a fix attempt, the solution is the same.
@Linnemann Ok, nice attempt! I don't really have time to try and beat this and I'm not even sure, its possible.
But I'm sure the #6 is actually working. After 1.Ba7+ you think Kc7 busts you? But can't you White just play b6? My variation goes like this:
1.Ba7+ Kc7 2.b6+ Kd7 (Kd8 3.Kc3 Bxg4 4.Kd2 Ke7 transposes into the main line) 3.Kc3 Ke7 4.Kd2 Bxg4 5.Ke1 Rg8 6. Kf1 Bf5(or any other move with the bishop) 7.Bb8!! and now White manages Kg1 if Black takes the bishop, but Black cannot come in as the bishop is protecting the pawn on h2 and can move freely on the h2-b8 diagonal.
Actually this makes the position even greater in my opinion :)
But I'm sure the #6 is actually working. After 1.Ba7+ you think Kc7 busts you? But can't you White just play b6? My variation goes like this:
1.Ba7+ Kc7 2.b6+ Kd7 (Kd8 3.Kc3 Bxg4 4.Kd2 Ke7 transposes into the main line) 3.Kc3 Ke7 4.Kd2 Bxg4 5.Ke1 Rg8 6. Kf1 Bf5(or any other move with the bishop) 7.Bb8!! and now White manages Kg1 if Black takes the bishop, but Black cannot come in as the bishop is protecting the pawn on h2 and can move freely on the h2-b8 diagonal.
Actually this makes the position even greater in my opinion :)
This topic has been archived and can no longer be replied to.