Superko rules and ko-cycles rules are BAD board game design

[RobertJasiek]

the game may freely move between phases. So there is no issue with passes not being allowed during the game.

Once more, J1989 application slowly for you:

$$
$$ -----------------
$$ | O O X X X O O |
$$ | X O . X . O X |
$$ | . O X X X O . |
$$ -----------------

Click Here To Show Diagram Code

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

As long as this position persists during each occurrance of a confirmation phase, each large string is alive and the two black single stones are dead. The two middle empty intersections are dame. The large strings are in-seki. Only stones not in-seki can surround territory. Since dead stones may only be removed from territory, the dead stones do not exist on territory and may not be removed.

Therefore, after n confirmation phases and resumptions, in resumed alternation eventually White must use plays to approach and remove the single black stones as follows:

...
White plays to remove one single black stone,
Black passes,
White plays to remove the other single black stone,
Black passes,
White passes,

$$ 2 black prisoner stones
$$ -----------------
$$ | O O X X X O O |
$$ | . O . X . O . |
$$ | O O X X X O O |
$$ -----------------

Click Here To Show Diagram Code

[go]$$ 2 black prisoner stones
$$ -----------------
$$ | O O X X X O O |
$$ | . O . X . O . |
$$ | O O X X X O O |
$$ -----------------[/go]

the players agree that all stones on the board are alive in the next confirmation phase,
this agreement ends the game,
the game is counted.

[jmeinh]

Now if you correctly claim a draw by threefold repetition and the arbiter is standing right next to you but rules against the claim, what happens is exactly this https://www.youtube.com/watch?v=3P0oKbJcQj0

Yes, in situations like this, a rule of this kind is difficult to apply.
And if it is true (which was claimed somewhere else, as far as I remember) that the superko rule in area rules is hardly ever followed in practice, I don't really see any need to continue thinking about an analogue of this chess rule in Go.

So much for the practice-oriented point of view.

From other points of view, for example that of making Go accessible to AI, the advantages of superko rules far outweigh the disadvantage of additional "mental" bookkeeping.

[RobertJasiek]

From other points of view, for example that of making Go accessible to AI, the advantages of superko rules far outweigh the disadvantage of additional "mental" bookkeeping.

Very good point. Human difficulties in every 5,000th game versus easy AI management on every move in every game.

[jann]

There is little difference for bots in treating repetition as draw or lose/forbidden. In any case Katago handles both rules without problem.

[John Fairbairn]

So there is no issue with passes not being allowed during the game.

They ARE allowed, and there ARE issues.

You need to start from the standpoint that Japanese rules are a pig's breakfast - and they know that. That's why they tried to revise them. As I indicated in another thread, one of the main issues, going back to the 1920s, was deciding how, unequivocally, a game ends. Various approaches with various terms for what is now the confirmation phase were debated. The debate was NOT about passes, but about the end of the game. Passes were just a tool. They were concerned with getting the confirmation phase right (and failed), not with defining passes.

The second point is that they tried to ape traditional practice. Passes allowed during a game were part of the tradition, so in that sense they were taken care of.

In another sense everything was also already taken care of by the introduction to the rules which says they have to be applied in a spirit of common sense. There was a case in an amateur Swiss event under Nihon Ki-in auspices in Japan where a 1-dan was drawn against a lowly kyu in Round 1. He took umbrage at this and so passed on his first move (and maybe others - can't remember) to turn the game into a handicap game. The kyu complained and was awarded the game. But not because of the pass (the dan player clearly thought he could play one, as you can see) - it was because he breached the spirit of the game and disrespected his opponent.

Now look at this local position from an almost completed game.

Capture.GIF (15.66 KiB) Viewed 22772 times

Black thinks the game is over and says Pass. Under your conception, White is not then allowed to play at once at A. That rule would not only be unthinkable to a pro, but it would go against one of the major tenets of Japanese go (and tradition) which they like to repeat endlessly as an advantage of their rules over Chinese rules. Namely, they believe that knowing whether a fill-in move is necessary is a mark of skill (reading ability - reading the moves, not reading the Laws of Go). They pour scorn on the Chinese rule that lets you fill in your own territory at no cost. This is a trivial example, but there was a complex one in the 22nd Honinbo (Sakata-Rin) involving a yose-ko, which was actually started after the game by the editor of Kido, who was a bit of a rules maven, and it made it to the newspapers. (That's almost as far as my memory goes - the days when I look these things up are long past, but as I recall it, no pass occurred during the game itself, but could have done if the players had spotted what the editor spotted.)

The above diagram example is, however, complicated by the fact that many White amateurs in that position would tell Black, after he passes, that he needs to make a move at A, and only if Black disagrees will White play at A. A variant procedure I have seen (often) is: "You need to make an extra move somewhere, Mr Black." Mr Black looks hard and can't see anything. So Mr White plays A. You can have a debate about what action is in the spirit of the game.

As I say, a pig's breakfast. My own view is that nearly all Japanese rules problems derive from amateurs trying to use professional rules. They are trying to insert a nail with a Swiss Army knife when all you need is a hammer - and common sense.

[CDavis7M]

Therefore, after n confirmation phases and resumptions, in resumed alternation eventually White must use plays to approach and remove the single black stones as follows...

That is a resolution. The Japanese Rules are so robust that they can ever handle this practically impossible situation with ease.

[jmeinh]

Well, let's change the thread title to "Really forbid a move after a pass during the game is BAD board game design"

[kvasir]

Now if you correctly claim a draw by threefold repetition and the arbiter is standing right next to you but rules against the claim, what happens is exactly this https://www.youtube.com/watch?v=3P0oKbJcQj0

Yes, in situations like this, a rule of this kind is difficult to apply.
And if it is true (which was claimed somewhere else, as far as I remember) that the superko rule in area rules is hardly ever followed in practice, I don't really see any need to continue thinking about an analogue of this chess rule in Go.

So much for the practice-oriented point of view.

From other points of view, for example that of making Go accessible to AI, the advantages of superko rules far outweigh the disadvantage of additional "mental" bookkeeping.

I won't dispute that superko can have some merits but it should not be an idle point that rules need to be something people are willing and able to use in practice.

Make the game more accessible to AI? Well obviously end the game after a fixed number of moves -- easy, simple, elegant and hardly requires more than two lines of code. Early stopping is used in practice.

[CDavis7M]

They ARE allowed, and there ARE issues.

Just to be clear, my point is that the Japanese Rules allow "alternating play" during the game and do not allow passing during the game. The Japanese Rules clearly allow (multiple) passes to resolve End-of-Game and Life & Death situations. I have no issue with a pass at this point. It's clearly in the Japanese Rules.

But I have seen no allowance in the Japanese Rules, no Japanese tradition, and no common sense reason why the rules must allow "pass in the middle of the game because I am so far ahead" as was suggested early on in this discussion. Presenting a practically impossible situations (as others have done) which can be handled by the End-of-Game and Life & Death rules after stopping the game is not a common sense reason to prove that passing must be allowed during the game.

Black thinks the game is over and says Pass. Under your conception, White is not then allowed to play at once at A. That rule would not only be unthinkable to a pro, but it would go against one of the major tenets of Japanese go (and tradition) which they like to repeat endlessly as an advantage of their rules over Chinese rules.

If Black thinks that he doesn't have any move left to play and white disagrees about the Life & Death status, then this is exactly the type of situation that is intended to be covered under Article 9-2 when confirming Life & Death. There is no "pass" needed. If the definition of Live Stone is not convincing, White can demand resumption and Black can pass.

Under your conception, White is not then allowed to play at once at A. That rule would not only be unthinkable to a pro, but it would go against one of the major tenets of Japanese go (and tradition) which they like to repeat endlessly as an advantage of their rules over Chinese rules. Namely, they believe that knowing whether a fill-in move is necessary is a mark of skill (reading ability - reading the moves, not reading the Laws of Go). They pour scorn on the Chinese rule that lets you fill in your own territory at no cost.

This is my understanding of the Japanese tradition as well. But for me, it supports my understanding that the most Japanese resolution for this gameplay situation is to not have the move played at all during the game. If black does not want to play at 'A' during the game, White can decide to proceed with filling point 'A' and its neighbor as dame, which will allow Black to recognize if a reinforcing play is needed. Everything works as it should.

By the way, 手入れ was the first term I looked at in the Nihon Kiin 小事典

[CDavis7M]

Well, let's change the thread title to "Really forbid a move after a pass during the game is BAD board game design"

That would be bad rules design. The Japanese Rules do not have this rule. The Japanese Rules allow alternating play. The Japanese Rules do not allow a play after a pass during normal game-play but do not explicitly forbid it. As with all game rules, anything not allowed by the rules is prohibited.

[CDavis7M]

From other points of view, for example that of making Go accessible to AI, the advantages of superko rules far outweigh the disadvantage of additional "mental" bookkeeping.

Everything is a trade off. Implementing super-ko in a computer program is good computer program design while implementing super-ko in a human-played board game is bad design. The improvement in a computer program outweighs the drawbacks. But just because a rule works better for computer analysis of game-play does not mean that the rule is a better rule for human game-play. In a tournament, these tradeoffs change and having a clear winner might be enough of an improvement to warrant the rule. But tournament rules should not be the same rules for normal play. Most humans would like to keep their phones in their pocket during normal play.

[Cassandra]

My own view is that nearly all Japanese rules problems derive from amateurs trying to use professional rules. They are trying to insert a nail with a Swiss Army knife when all you need is a hammer - and common sense.

Let me try to discuss in what sense J89 are "professional".

In my understanding, J89 were primarily designed to be applied on games between two (Japanese) professional Go players.

In my own words, the base lines for creating this revision of the J49 rules were:
(1) Comply with Japanese tradition.
(2) Find consistent solutions for all the earlier rules disputes in Japan, and prevent future cases.
(3) Find a compelling solution for use outside of Japan (by professionals and amateurs alike).

Being professionals in playing the game of Go, the authors succeeded with (1) and (2).

However, they failed with (3), lacking the last five percent of professional approach (/ attitude) to the creation of a set of rules.
They simply overlooked (being no amateurs) that things can happen in amateur games that professionals wouldn't even dream of. They also overlooked (being no Westerners) that Western attitude (being not aware of (1) and (2)) demands to take the legal text of the rules literally.

On the other hand, some responses in the Western world have been typically Western in that common sense has been completely left out (along with knowledge of probability theory).

But was it all a drama? Of course not.
After all, there are several approaches to creating rule sets based on Western thought patterns.

Are there any secret recipes as to how you can be sure to win under ruleset A against a player who has only known ruleset B until now? I don't think so. Therefore, the differences between the various rulesets can be only marginal, not affecting the character of the game.

--------------------------------------

In another thread I already mentioned that no AI on this planet would be able to solve Igo Hatsuyôron 120. Which is solely due to the fact that they were not designed for this purpose.
No AI would dream of this board position, just because it was not encountered while training the neural network.

But is this a drama? Of course not.
People still use the programmes for the purposes these have been developed for.

To overcome the above mentioned "weakness", lightvector used an approach to include positions of Igo Hatsuyôron 120 as material for the training of a specialised neural network.
This not only resulted in KataGo's ability to solve the problem, but also in an improvement on the (at that time) best known solution sequence.
Friday9i later continued the training of the specialised network, also with training material that emerged from our further analyses.


However, even with KataGo's specialised network, there was no guarantee that the correct continuation of a subvariation, which included several mistaken moves, would be found!
During the training, the neural network had simply learned to avoid these multiple-mistake-sequences, and therefore lacked sufficient experience with what might happen thereafter.

But was this a drama? Of course not.
We got new insights into the problem, which would have been unrevealed without.
Friday9i executed some kind of "enforced" training with these positions, resulting in KataGo's correct handling afterwards.


However, even with KataGo's specialised network, there was no guarantee that the correct continuation of a subvariation, which started with a "simple" (according to human understanding) valid change in the order of correct moves (but which added a lot of uncertainty to the position in AI understanding), would be found!
During the training, the neural network had simply learned to avoid that "too noisy" path, and therefore lacked sufficient experience with what might happen thereafter.

But was this a drama? Of course not.
We got new insights into the problem, which would have been unrevealed without.
Friday9i executed some kind of "enforced" training with this position, resulting in KataGo's correct handling afterwards.

--------------------------------------

There seems to be a general understanding that an AI only delivers "good" results for the application cases for which it has been trained. And that the quality of the results reflects the quality of the training material.

It would be advisable to apply this knowledge to the assessment of rulesets as well.

[CDavis7M]

$$c :b5: pass, :b7: pass
$$ ---------
$$ | . . . . . . . . . |
$$ | . . . . . . . . . |
$$ | . . 1 . 0 . 4 . . |
$$ | . . . . . . . . . |
$$ | . . 9 . , . 6 . . |
$$ | . . . . . . . . . |
$$ | . . 3 . 8 . 2 . . |
$$ | . . . . . . . . . |
$$ | . . . . . . . . . |
$$ ---------

Click Here To Show Diagram Code

[go]$$c :b5: pass, :b7: pass
$$ ---------
$$ | . . . . . . . . . |
$$ | . . . . . . . . . |
$$ | . . 1 . 0 . 4 . . |
$$ | . . . . . . . . . |
$$ | . . 9 . , . 6 . . |
$$ | . . . . . . . . . |
$$ | . . 3 . 8 . 2 . . |
$$ | . . . . . . . . . |
$$ | . . . . . . . . . |
$$ ---------[/go]

Here is an example of non-alternating play during the game. This is the situation that the Japanese Rules do not allow.

Or, can anyone try to find where the Japanese rules allow this type of non-alternating play during game-play, not during life-and-death confirmation or end-of-game dame filling & reinforcement. Or try to find some Japanese tradition of playing moves and then deciding to pass for a bit to allow non-alternating play, but then picking alternating play back up again.

[John Fairbairn]

The rules of go MUST be badly designed, since they give such a field day to rules mavens.

However, back to our muttons and the question of whether a single pass ends ordinary play (it doesn't). The Nihon Ki-in published a pamphlet in 2005 on go etiquette. It was in thee form of readers' questions and they gave the answers.

One question related to a game with clocks and sudden death. The reader had had the experience of playing a game in which he felt he was comfortably ahead, although he was very short of time and his opponent wasn't. The game came down to the last half-point and the reader said, "O-tsugi kudasai" (Please connect the ko) which is a common way to end the game in Japanese, and is equivalent to saying Pass.

But the opponent saw an opportunity to win the game on time and refused to connect but played somewhere else. The reader didn't say how his game finished, but he was clearly highly peeved.

The professional answering, Kudo Norio 9-dan, said this was an "egregious violation of good manners" (悪質なマナー違反) and the player should summon the referee, and if he, Kudo, were the referee he would declare the game an instant loss for the opponent. No mention of two passes, no mention of going on to a confirmation phase. Just "You're a lout, you're out."

This loutish behaviour is common in Europe, and I imagine everywhere else. Kudo suggested the way to avoid it, apart from calling the referee, was to ensure you left yourself about 5 minutes. The more obvious solution, to use byoyomi, has been slow in taking hold in Japan. He did NOT say, "You said Pass, so the game (ordinary play) was over."

Having met Kudo, and knowing he has refereed title games, I would trust him to know more about the rules than we do, even if he is "only a Japanese 9-dan."

[jmeinh]

Under your conception, White is not then allowed to play at once at A. That rule would not only be unthinkable to a pro, but it would go against one of the major tenets of Japanese go (and tradition) which they like to repeat endlessly as an advantage of their rules over Chinese rules. Namely, they believe that knowing whether a fill-in move is necessary is a mark of skill (reading ability - reading the moves, not reading the Laws of Go). They pour scorn on the Chinese rule that lets you fill in your own territory at no cost.

This is my understanding of the Japanese tradition as well. But for me, it supports my understanding that the most Japanese resolution for this gameplay situation is to not have the move played at all during the game. If black does not want to play at 'A' during the game, White can decide to proceed with filling point 'A' and its neighbor as dame, which will allow Black to recognize if a reinforcing play is needed. Everything works as it should.

Perhaps in this way Black can recognize that a reinforcing move is needed and be "forced" to do so, but at least this does not work as it should. Because part of the essence of Go as we know it is to have to make tactical decisions of this kind during the game - and not in a phase of collective analysis.
If the Japanese tradition really wanted it that way, that would be a good reason to break with that part of the tradition.