AI & Igo Hatsuyoron 120
-
Tryss
- Lives in gote
- Posts: 502
- Joined: Tue May 24, 2011 1:07 pm
- Rank: KGS 2k
- GD Posts: 100
- KGS: Tryss
- Has thanked: 1 time
- Been thanked: 153 times
Re: AI & Igo Hatsuyoron 120
One thing to understand, is that the programs don't know the difference between a problem and a game. And that has consequences in how you approach the position
-
Cassandra
- Lives in sente
- Posts: 1326
- Joined: Wed Apr 28, 2010 11:33 am
- Rank: German 1 Kyu
- GD Posts: 0
- Has thanked: 14 times
- Been thanked: 153 times
Re: AI & Igo Hatsuyoron 120
If this were true, I would assume the programs to perform better.Tryss wrote:One thing to understand, is that the programs don't know the difference between a problem and a game. And that has consequences in how you approach the position
Let me give an example, showing the two types of White replies during the growth of the hanezeki's tail:
= = = = = = = = = = = = = = = = = = = =
Intro: What both programs already know
During the sequences following mistaken White moves during the growth of the hanezeki's tail, both programs are very well aware that ...
... capturing White's three stones in the lower right is Black's privilege, and ...
...that this forces White to prevent Black from immediately getting two eyes.
To ease the following presentation, we can savely assume that these two moves are played first.
= = = = = = = = = = = = = = = = = = = =
TYPE 1: Life for or escape of the tail (NOT CHOOSEN !!!)
= = = = = = = = = = = = = = = = = = = =
TYPE 2: Confinement of the tail (MISTAKENLY CHOOSEN !!!)
Now, there is a semeai between White's stones on the lower edge, and Black's tail group.
White should only play this way if she knew that she can win this semeai. What the programs usually do NOT during their analysis.
Now, there is a semeai between White's stones on the lower edge, and Black's tail group.
White should only play this way if she "knew" that she can win this semeai. What the programs usually do NOT during their analysis (neither do the programs "know" that capturing the tail is losing the game for White).
= = = = = = = = = = = = = = = = = = = =
I would like to reason that AI does understand (short) future sequences that capture some opponent's stones (and so either make two eyes or connect to the outside), but are unable to handle a (longer) semeai correct.
Last edited by Cassandra on Wed Sep 18, 2019 4:17 am, edited 1 time in total.
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)
Igo Hatsuyōron #120 (really solved by KataGo)
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: AI & Igo Hatsuyoron 120
A couple of things. Many classical problems were, in fact, whole board problems, considering that a fight over the life or death of a corner could spill out into the rest of the board. However, if the purpose of the problem was to win the game, it might well be best to play somewhere else and possibly lose the fight in the corner. But this is not one of those problems, for two reasons. 1) The fight for life and death is huge, dwarfing anything else, at least at first. 2) Regular yose are part of the problem, not just life and death. So the purpose of the problem is to win the game.Cassandra wrote:If this were true, I would assume the programs to perform better.Tryss wrote:One thing to understand, is that the programs don't know the difference between a problem and a game. And that has consequences in how you approach the position
Given that the purpose is to win the game, komi matters. In fact, to the best of human analysis, if Black gives 7½ pts. komi, she cannot win the game. So far in this discussion, the bots are assumping a 7½ pt. komi. In this note, Cassandra has shown a number of AI mistakes by White that allow Black to win the game, even giving komi. To wit:
This is a well known defect of today's top bots. Humans do better at large semeai. In fact, not too long ago I posted a mistaken review by Elf where the semeai was not that big, but there was a tesuji at the end. Elf failed to see the human win until late in the actual play. (Or maybe the human loser resigned first, I don't recall exactly.)Cassandra wrote:Let me give an example, showing the two types of White replies during the growth of the hanezeki's tail:
{snip}
White should only play this way if she knew that she can win this semeai. What the programs usually do NOT during their analysis.
{snip}
White should only play this way if she "knew" that she can win this semeai. What the programs usually do NOT during their analysis (neither do the programs "know" that capturing the tail is losing the game for White).
= = = = = = = = = = = = = = = = = = = =
I would like to reason that AI does understand (short) future sequences that capture some opponent's stones (and so either make two eyes or connect to the outside), but are unable to handle a (longer) semeai correct.
My working hypothesis for why this happens is twofold. First, there is a horizon effect. You have to calculate the tree deeply enough to find the correct play. Humans excel at depth first search, at least consciously. Second, searching the whole board is inefficient for finding locally deep plays. Usually searching the whole board is better than local depth first search, which is one reason that today's bots play at super human levels. Another is that they are better at whole board evaluation than humans, but that is something that humans can learn from them.
Anyway, most go problems are designed for human solution, that is, for depth first local search, even if the search takes the whole board. If bots were trained on such problems, they could very likely do better than humans at them. But such training would take many thousands, maybe millions of such problems, and they do not exist, nor do we know how to have computers construct them.
I think that in the future, go playing programs will, like humans, utilize local search as well as whole board search. We know mathematically that it is more efficient to search independent regions of the board locally and combine the results of those searches into a whole board search. That is how humans do well at the endgame.
An unsolved problem is how to get computers to recognize independent regions of the board. Right now, programs are continuing to improve rapidly by improving their whole board evaluation. But eventually diminishig returns will set in, and my guess is that new programs will start to incorporate local searches in order to improve. Not using local search is a known weakness, after all.The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
-
Cassandra
- Lives in sente
- Posts: 1326
- Joined: Wed Apr 28, 2010 11:33 am
- Rank: German 1 Kyu
- GD Posts: 0
- Has thanked: 14 times
- Been thanked: 153 times
Re: AI & Igo Hatsuyoron 120
The handling of the hanezeki's tail might be difficult to "understand".
Connecting the tail while capturing three White stones is a fatal mistake, but ...
... if White ever has to capture Black's tail group, ...
...
... Black can play elsewhere four times.
= = = = = = = = = = = = = = = = = = = =
On the other hand, if White mistakenly captures the tail, ...
...
...
...
... there are also two tenuki.
= = = = = = = = = = = = = = = = = = = =
If Black is forced to capture White's group in the corner, and White does not mind losing points, ...
... she can play five moves in a row elsewhere.
Connecting the tail while capturing three White stones is a fatal mistake, but ...
... if White ever has to capture Black's tail group, ...
...
... Black can play elsewhere four times.
= = = = = = = = = = = = = = = = = = = =
On the other hand, if White mistakenly captures the tail, ...
...
...
...
... there are also two tenuki.
= = = = = = = = = = = = = = = = = = = =
If Black is forced to capture White's group in the corner, and White does not mind losing points, ...
... she can play five moves in a row elsewhere.
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)
Igo Hatsuyōron #120 (really solved by KataGo)
-
Mike Novack
- Lives in sente
- Posts: 1045
- Joined: Mon Aug 09, 2010 9:36 am
- GD Posts: 0
- Been thanked: 182 times
Re: AI & Igo Hatsuyoron 120
Not exactly. In "varvar" if white plays 3 tenuki, black plays 4 at "R3" (the letters are shifted) and then that group is unconditionally alive. Meaning that there is no hurry to capture the corner because the "O6" dragon has LOTS of liberties.
So play continues elsewhere until just dame left or the dragon's liberties have been reduced to the danger point. So white does not get all those tenukis you described, just the tenuki of 3 and sente. By the time black has to use moves to capture the corner, what you have as 9, 11, and 13 become valueless dame plays (and 5 was just white having sente after black 4)
So play continues elsewhere until just dame left or the dragon's liberties have been reduced to the danger point. So white does not get all those tenukis you described, just the tenuki of 3 and sente. By the time black has to use moves to capture the corner, what you have as 9, 11, and 13 become valueless dame plays (and 5 was just white having sente after black 4)
-
Bill Spight
- Honinbo
- Posts: 10905
- Joined: Wed Apr 21, 2010 1:24 pm
- Has thanked: 3651 times
- Been thanked: 3373 times
Re: AI & Igo Hatsuyoron 120
Whole board tsumego.
Above I mentioned that traditional tsumego sometimes spilled out onto the whole board. OC, in such cases the solution to the problem may not have been the best play on the board.
Here is a nice example, from Life and Death Research by Suzuki Tamejiro, published in 1920. Many people will be familiar with the modern version of the problem, which is more enclosed.
Above I mentioned that traditional tsumego sometimes spilled out onto the whole board. OC, in such cases the solution to the problem may not have been the best play on the board.
Here is a nice example, from Life and Death Research by Suzuki Tamejiro, published in 1920. Many people will be familiar with the modern version of the problem, which is more enclosed.
The Adkins Principle:
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
At some point, doesn't thinking have to go on?
— Winona Adkins
Visualize whirled peas.
Everything with love. Stay safe.
-
Cassandra
- Lives in sente
- Posts: 1326
- Joined: Wed Apr 28, 2010 11:33 am
- Rank: German 1 Kyu
- GD Posts: 0
- Has thanked: 14 times
- Been thanked: 153 times
Re: AI & Igo Hatsuyoron 120
One of the problems with the problem is that you are right, IF AND ONLY IF Black plays a perfect game.Mike Novack wrote:Not exactly. In "varvar" if white plays 3 tenuki, black plays 4 at "R3" (the letters are shifted) and then that group is unconditionally alive. Meaning that there is no hurry to capture the corner because the "O6" dragon has LOTS of liberties.
So play continues elsewhere until just dame left or the dragon's liberties have been reduced to the danger point. So white does not get all those tenukis you described, just the tenuki of 3 and sente. By the time black has to use moves to capture the corner, what you have as 9, 11, and 13 become valueless dame plays (and 5 was just white having sente after black 4)
However, if Black ever loses a move in the main semeai between his large group in the upper right and White's group at the left edge, he will lose the game by a large margin.
Please note that in this very special case of the main semeai getting two eyes with the move at "R3" also loses a move unnecessarily.
= = = = = = = = = = = = = = = = = = = =
Joachim Meinhardt's oki at "R3" (aka O3) -- that destroyed the classical solution to the problem -- has a similar characteristic.
This move is of NO use during several potential semeai between Black's large group in the upper right and White's group at the LOWER edge (then with only one eye) in the first half of the game. It also loses a move during these subvariations, and so should not be played.
But with the correct solution sequence, the situation towards the end of the main semeai will change, and this oki becomes the game winner for White.
Please also note that none of the seemingly "dame" plays is "valueless" then.
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)
Igo Hatsuyōron #120 (really solved by KataGo)
-
Cassandra
- Lives in sente
- Posts: 1326
- Joined: Wed Apr 28, 2010 11:33 am
- Rank: German 1 Kyu
- GD Posts: 0
- Has thanked: 14 times
- Been thanked: 153 times
Re: AI & Igo Hatsuyoron 120
This implies that one -- human being and machine alike -- cannot really learn from even the PROBLEM's move sequences INSIDE the hanezeki, unless the liberty relationship between the two OUTSIDE groups that are caught in a main semeai is also taken into account.
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)
Igo Hatsuyōron #120 (really solved by KataGo)
-
Cassandra
- Lives in sente
- Posts: 1326
- Joined: Wed Apr 28, 2010 11:33 am
- Rank: German 1 Kyu
- GD Posts: 0
- Has thanked: 14 times
- Been thanked: 153 times
Re: AI & Igo Hatsuyoron 120
Crosscut? Which Crosscut?
Both programs do not realize Black's crosscut in the upper left.
LeelaZero follows her fatal inclination to moves inside the hanezeki, assuming win rate > 65 % for Black.
KataGo, however, gives atari from the wrong side, assuming a win rate about 15 % for Black.
Probably, KataGo judges Black's large upper right group to be dead, so the best policy might be capturing White's left side.
+ + + + +
It is obvious that neither program realizes the main semeai between Black's upper right, and White's left, groups.
Both programs do not realize Black's crosscut in the upper left.
LeelaZero follows her fatal inclination to moves inside the hanezeki, assuming win rate > 65 % for Black.
KataGo, however, gives atari from the wrong side, assuming a win rate about 15 % for Black.
Probably, KataGo judges Black's large upper right group to be dead, so the best policy might be capturing White's left side.
+ + + + +
It is obvious that neither program realizes the main semeai between Black's upper right, and White's left, groups.
The really most difficult Go problem ever: https://igohatsuyoron120.de/index.htm
Igo Hatsuyōron #120 (really solved by KataGo)
Igo Hatsuyōron #120 (really solved by KataGo)