Borrowing and modifying a recent example from another topic:

Click Here To Show Diagram Code

[go]$$B
$$ +---------------
$$ | X X . O X X . O O X X . |
$$ | X O O O X X O O O X O O |
$$ | . O X X . X O . O X X O |
$$ | O O X X X X O O O X . . |
$$ --------------------
$$[/go]

Click Here To Show Diagram Code

[go]$$B
$$ +--------------------
$$ | X X . O X X X O O X X . |
$$ | X O O O X X O . O X O O |
$$ | . O X X . X O O O X X O |
$$ | O O X . X X O . O X . . |
$$ ---------------------
$$[/go]

Let's forget rules pecularities for a moment. Just look at these positions with a go player's eye. How do you see them individually, and how do you see the differences - if any - between them?

Never seen such positions in real games.

Assumed probability of the first diagram 1 : 10 billion, that of the second 1 : 1 million.

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

In the first one white is ultimately dead because of the bent four, there are three ways to start a ko that may be viable in an isolated position like this but usually the number of liberties, approach moves and ko threats needed to succeed would reduce the value of any ko (even to a negative) so that is is usually not the best way to approach it.

In the second one there are two separate areas, one depending on a bent four which is usually just dead for white and another with a ten-thousand year ko that is usually alive for black. Again, in isolated positions like this you can contemplate fighting these kos but the number of moves needed to succeed usually makes it unviable.

I am not familiar with the details of Korean rules but I think it has something to do with defining local regions and shapes. Can someone explain?


Edit =======
That wasn't very accurate, the right side shape is different from ten-thousand year ko in that the ko can't be removed to create a seki. A different conclusion must follow in the first shape when taking this into account.

If I forget the details of a known japanese rules I follow only the idea that no ko fight take place in confirmation phase and ko are considered independant.
Position 1 : all white stones are dead
Position 2 : white stones at the leflt are dead and position on the right is seki.
I am not aware of korean rule.

I see things with a chinese rules eye.
The position is difficult and many sequences must be read.
I think that if Black kills White on the left, White can kill Black on the right.
The result is the same if Black does nothing (and the left position remains as it is).

The only difference between the two positions is the extra Black stone at the top center, which is one more point for Black in the second position.

With komi, I think that Black looses the game anyway.

Korean rules can be seen here. Yes their point is to consider L/D in local regions and with normal ko rule. In some versions the text explicitly says "each position considered locally without regard to the whole board position", but the exact meaning of "locally" is not given.

So, my impression is the two examples are nearly identical strategically. In normal go W can do nothing. B can start a ko then take left corner and lose right, or solidify right while W solidifies the left. Both lines gain a point or so. The shared liberty at the center does not play a role, and all central groups are safe in all lines of play. This also matches the Chinese view.

In Japanese view the 2nd position is the known "bent4 is dead even with unremovable threats", so left corner is B's. In 1st case this unfair ruling is leveraged to see all W stones dead. The two examples are distinguished, on the assumption that B can take left corner while the rest remains unchanged (which is nonsense).

In Korean view the 1st example likely considered in whole. Two interpretations seem possible for "locally" (the other would be to split the board using left 7 W stones as border), but their example #19 seem to hint the border would be at central B group in the 2nd case - thus no border in 1st case. The end result seems everything is alive in 1st case and W left is dead in the 2nd (like in LJRG or Gérard's approach). The two examples are distinguished again, now on the above bordering principle.

The problem is all four rulings are incorrect.

In these examples B can also play the exchange (equal B/W movecount) in main game (thus Korean ruling for #1 at least leads to correct result). But imagine W also has large normal/removable ko threats elsewhere...

With a probability of 1 : 100.000.000.000.000, this is nothing to write home about.

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

As a go player (preferring territory scoring), when both players pass and I would be called to help the players scoring the game:

2nd diagram is easy. I would declare white dead on the left and I would declare the right side a seki.


1st diagram is more difficult. But white is really powerless to do anything during normal play. And also during hypothetical play, white cannot do anything as they cannot play ko threats to fight a ko on the left or the right (which I consider to be the intention behind Japanese rules for hypothetical play). So during hypothetical play, black can just remove all white's stones on the left and then on the right (the right side is only a temporary seki).
So I would be inclined to declare white dead everywhere. I don't feel that this is unfair or nonsense.

Yes, that is the Japanese view. It has its logic, but is very different from what could actually happen or achievable in real go (B could never capture left and then attack an UNCHANGED right afterwards). Btw in the other topic kvasir raised the question of confirmation with more relaxed pass-for-ko rules (or normal ko) in these examples. Let me show why that still doesn't work 100% here.

Click Here To Show Diagram Code

[go]$$B
$$ +---------------
$$ | X X . O O X . O O X X . |
$$ | X O O O X X O O O X O O |
$$ | . O X X . X O . O X X O |
$$ | O O X X X X O O O X . . |
$$ --------------------
$$[/go]

Click Here To Show Diagram Code

[go]$$B
$$ +--------------------
$$ | X X . O O X X O O X X . |
$$ | X O O O X X O . O X O O |
$$ | . O X X . X O O O X X O |
$$ | O O X . X X O . O X . . |
$$ ---------------------
$$[/go]

I flipped a stone on the left corner, enlarging W there. This makes it clearer what B wants to do in real go: take the left and give up the right in exchange, gaining a few points.

With normal ko everything would be alive, which may be more fair than the above rulings. But then B would need to make the exchange in main game. This is ok in Chinese, but in territory scoring if W also has a few large removable threats elsewhere, B is robbed on threat removal costs.

Bent4 + unremovable threat is a tough nut to crack because most approaches you can choose will be unfair in some cases. Still fair scoring doesn't seem impossible. For example, here the trade is valid, so if we would score W left dead and B right dead as well, the score would turn out correct (without B losing points in territory scoring on threat removal elsewhere).

No AI on this planet is able to continue the above game of two very young kids, who barely had a rudimentary understanding of the rules, successfully!

Would you judge all these programs as "flawed"?
Or would you not ultimately come to the conclusion that it is a system-inherent feature?

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

Good questions.

Yes. The players played according to the rules. Therefore the game should have a result.
Ah, this is the eternal question "is it a bug or a feature" ?

For me it's a bug. The chinese-style rules give an easy answer : there are 70 black stones on the board and 71 white stones, and no intersection is surrounded by stones of only one colour.
With a komi of 0, White wins by 1 point.
The rule also give an easy explanation for the children about why they should continue : if they put more stones on the board, they will get more points.

Sorry, Pio2001,


is about AI CONTINUING the game SUCCESSFULLY.

It's NOT about scoring an UNfinished position

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

In this case I didn't understand your point.

What is "continuing" the game ? Play another move ? I don't see why an AI couldn't propose a move.
What is "successfully" continuing ? Find a winning move ? Winning against whom ?

Just find the best continuation for both sides.

Will be a lot of effort, and in the very end, the final result will be the same as your temporary one.

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

@Pio2001: That position is a tsumego problem, perhaps the most difficult one that was ever created. And it is not only extremely difficult for humans, it also is for AI.

Cassandra's signature refers to it. I suppose Cassandra contributed personally to the solution that may be correct?

Bots not playing perfectly is a (temporary) limitation. These rules problems otoh are not about computational complexity. There are only a few lines to read, well within the range of human analysis. What's missing is a logical system to draw correct conclusions. With two more variants where the unremovable threat is less useful:

[W also has few large removable threats elsewhere]

Click Here To Show Diagram Code

[go]$$B
$$ +---------------
$$ | X X . P P X . O O Z Z . |
$$ | X P P P X X O O O Z O O |
$$ | . P X X . X O . O Z Z O |
$$ | P P X X X X O O O Z . . |
$$ --------------------
$$[/go]

Click Here To Show Diagram Code

[go]$$B
$$ +--------------------
$$ | X X . P P X X O O Z Z . |
$$ | X P P P X X O . O Z O O |
$$ | . P X X . X O O O Z Z O |
$$ | P P X . X X O . O Z . . |
$$ ---------------------
$$[/go]

Click Here To Show Diagram Code

[go]$$B
$$ +---------------
$$ | Z Z . O O X . O O X X . P P P |
$$ | Z O O O X X O O O X P P P P P |
$$ | . O X X . X O . O X X X P P P |
$$ | O O X X X X O O O X . P P P P |
$$ --------------------
$$[/go]

Click Here To Show Diagram Code

[go]$$B
$$ +--------------------
$$ | X X . P P X X O O X X . O O O |
$$ | X P P P X X O . O X O O O O O |
$$ | . P X X . X O O O X X X O O O |
$$ | P P X . X X O . O X . O O O O |
$$ ---------------------
$$[/go]

In the 4th case W don't want to use the right (too big to lose), so bent4 just dies normally. The 3rd case is the only one where the central shared liberty comes into play. There W have to use (and lose) the right to keep the bent4 alive, otherwise he loses everything (including the right anyway). Central stones still safe in all cases.

Chinese rules can play out and score all these examples correctly. Also under informal territory rules, normal play cannot continue but hypothetical play can be discussed with honest opponent and score all four accurately as shown above (without B losing points on threat removal elsewhere). It is only current territory rules that can't do this formally.

Why do you assess Igo Hatsuyôron 120 irrelevant for judging the performance of AI?

Why do you assess your artificially generated extremely impropable examplary positions relevant for judging the performance of rulesets?

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

I haven't said anything about Igo Hatsuyôron, and don't see what else can a ruleset judged on if not mishandled examples. And bent4+seki is not that rare.

Rarity is not really good argument anyway since disputes themselves are already very rare. So in practice territory by agreement is enough, without any confirmation or L/D rule. Many players don't even know how confirmation works in Japanese.

You wrote something about "bots". Positions like Igo Hatsuyôron 120 are very rare, as are your artificially created positions.

Your examplary position is NOT bent-four & seki. What would be very rare in itself. But it is bent-four & sandwich-seki & seki. What is even much more implausible.

The many players who do not have any idea about J89's L&D confirmation will never ever realise that they have any "problematical" position on their board. Not to mention your artificially created positions with many unlikely components. Where is the problem then?

Therefore, why do you not discuss about realistic positions that could arise in professional play only? I.e. positions that J89 had been designed for?

_________________
The really most difficult Go problem ever: http://igohatsuyoron120.de/index.htm
Igo Hatsuyoron #120 (still unresolved by professionals, maybe solved by four amateurs)

Maybe because it is not the topic of the discussion.

Maybe because a sponsor who invests hundred of thousands in a tournament would prefer having secure rules.

Maybe because go is the only abstract combinatorial game that doesn't have a completely defined rule.

In one of my games, I had the choice between creating a common seki, or creating a quintuple sandwitched seki (black group vs white group vs black group vs white group vs black group). Unfortunately, I chose the former.