Life In 19x19

i can't seem to find the game in my game records, but i once had a position that was seki in double ko.
neither player could capture the other, unless they made a stupid mistake and filled.
the primary problem was, how are points counted for the group under aga rules?
my personal rule is: if a player surrounds an intersection, and can't be taken away by the opponent, then its a point.
but this rule breaks down with this position. either player can indefinitely capture and recapture.
its just that doing so doesn't improve or worsen the score.
hypothetically if the game isn't timed, it could go on forever.

phillip1882 wrote:i can't seem to find the game in my game records, but i once had a position that was seki in double ko.
neither player could capture the other, unless they made a stupid mistake and filled.

See this page on Sensei's Library. https://senseis.xmp.net/?DoubleKoSeki

the primary problem was, how are points counted for the group under aga rules?
my personal rule is: if a player surrounds an intersection, and can't be taken away by the opponent, then its a point.

That's my understanding of the AGA rules.

Edit: To be clear, if the players disagree about whether a point is surrounded by dead stones, they may resume play. But any superko ban remains in effect.

but this rule breaks down with this position. either player can indefinitely capture and recapture.

The AGA superko rule prevents indefinite repetition of double ko seki. See https://senseis.xmp.net/?SituationalSuperko

Edit: To be clear, the ko eyes in the double ko seki cannot be taken by taken away by the opponent by indefinite capture and recapture because the opponent is prevented from doing so by the superko rule at the end of the game. OC, the players may end the game without trying to play on forever.

Superko applies so endless repetition does not occur.

It is immaterial how things are counted - what matters is what scores.

Scoring applies at the game end after the final passes.

According to the rules, "Those empty points on the board which are entirely surrounded by live stones of a single color are considered the territory of the player of that color.". What the rules actually mean is: "Those empty points on the board which are entirely surrounded by stones of a single color are considered the territory of the player of that color." That is because of the rule "The game is over when the players agree on the status of all groups on the board, or, failing such agreement, if both players pass twice in succession. In this case any stones remaining on the board are deemed alive.". If the players disagree and after both players pass twice in succession (and White makes the last pass), any stones remaining on the board are deemed alive. An empty intersection (sorry, for the sake of confusion with the scoring unit, called "point" in these rules) adjacent to, what one might call, a ko stone is then entirely surrounded by stones of a single color and hence scored for their player.

The answers to this seem slightly misleading to me in that it implies AGA rules are important in the scheme of things. Are they? Who uses superko in real life?

I'm relying on memory, but I think the 2010 game with Gu Li was the only one to implement a superko ruling in the pro world and that was under Ing rules. Apart from that, have not all (and certainly most) triple kos been treated as void games - and that despite the fact that superko technically exists in Chinese rules?

If pros decide life is too short to worry about superkos, why should AGA rules try to buck the trend? I'm sure in practice US amateurs take the same pragmatic view as the pros anyway, which is what the poster seems to have done with his "personal" view. Long live persons!

Chinese Rules are a separate issue because not all of its authors might have been aware of the existence of superko but chosen a careless wording, which people aware of superko interpret as superko but the Chinese Rules authors have only meant to apply to sending-2-returning-1 while otherwise the referee ko rules take precedence. Last, I asked a Chinese professional and international referee who was not aware of the existence of the superko rules concept.

Apart from sending-2-returning-1, whose practical behaviour is agreed upon, the long cycle ko rules are rarely relevant at all: depending on the players, relevant shapes occur in 1:20,000 1:5,000 or for some rare players 1:1,000 games. Even among those shapes occurring at all, some have the same strategic behaviour. A genuine triple ko (composed of one, two or three shapes) may be the most frequent kind for which strategic behaviour differs.

Superko rules apply as such in New Zealand Rules (few players, few games), Simplified Ing Rules (in some of the European Go Congress main tournaments etc. but such has become infrequent after Ing sponsoring of the EGF has been abandoned), AGA Rules (in some AGA tournaments), French Rules (in some French tournaments) and BGA Rules (in some BGA tournaments), some other tournaments. This means that we have relatively few games for which there could be reports with different strategic behaviour of superko at all.

That there have been a few AGA tournament reports with superko not having been applied correctly (scandal attracting motivation of being reported) does not mean that otherwise in AGA or non-AGA tournaments superko would not have been applied correctly. Correct application does not attract reports easily. So witnesses are requires that then report and correctly so.

While I did witness a Ing 1991 ko dispute (and was readily called by the referee as an expert), I do not recall to have witnessed a game under superko with a position of strategically different superko behaviour. In my own (60,000 or 70,000?) games, I had two double ko sekis (but they alone would not create a different strategic behaviour) and one triple ko (double ko death and a basic teire ko elsewhere; interesting because I knew correct strategy of omitting reinforcement but could not play strategically correctly allowing my opponent many virtual ko threats due to sudden death thinking time, only a few remaining seconds and had to compromise in order to win the game) but in particular the latter was not played under superko.

Since Asian professionals in Asia play hardly any superko rule games, this is much more a matter of their practice than preference. I guess most Japanese pros would favour their applied rules tradition. Chinese professionals, as far as I have asked them, are roughly split 50:50. Korean professionals I expect in between. I forgot to ask the one Taiwanese professional I met.

phillip1882 wrote:hypothetically if the game isn't timed, it could go on forever.

In a general sense a single double ko seki is not a real perpetual cycle since - unless both players cooperate - it repeatedly triggers two successive passes (game stops).

John Fairbairn wrote:If pros decide life is too short to worry about superkos, why should AGA rules try to buck the trend? I'm sure in practice US amateurs take the same pragmatic view as the pros anyway, which is what the poster seems to have done with his "personal" view. Long live persons!

AFAICT, the AGA rules are derived from Ing's 1975 rules, which included a superko rule. Ing later changed his mind and came up with idea of fighting and disturbing kos, a double ko seki being an example of a disturbing ko.

I think the compatibility of widespread triple ko rules in Asia is no coincidence. The three large go playing countries share the history of the game, which includes knowledge of triple kos and void games. Changing well established things is hard, especially without reasons for such change. AGA had more freedom to invent/adopt new rules.

Thanks for the explanations about superko, Robert.

I don't know why the superko rule was first created. I guess that it is because void games, or ties, can't be dealt with within some kind of tournaments, like single elimination tournaments, for example.

But nowadays, superko is also necessary for artificial intelligence. A rule with endless repetition left to the referee to assess can't be translated algorithmically into a computer program.

RobertJasiek wrote:According to the rules, "Those empty points on the board which are entirely surrounded by live stones of a single color are considered the territory of the player of that color.". What the rules actually mean is: "Those empty points on the board which are entirely surrounded by stones of a single color are considered the territory of the player of that color." That is because of the rule "The game is over when the players agree on the status of all groups on the board, or, failing such agreement, if both players pass twice in succession. In this case any stones remaining on the board are deemed alive.".

That's right.
The french rule is interesting because if we leave aside the teaching parts, and read only the rules statements, the final score is defined without relying on such concepts as life or death.
The meaning of the AGA rule is thus the wording of the french rule.

What's interesting is that it allows complete novices, who discover the game by themselves, to try it and score their first game by themselves.
It is not possible for a novice player to tell who's the winner if the life and death status of each group must be assessed first. I have met two people already, fond of board games, who knew the game of go, but had never played it because when they tried, they failed to understand who was the winner.

Stating that any set of connected intersections that are surrounded by stones of the same color is a territory, without telling if these stones are dead or alive, also allows neural networks to score their self-learning games without any external intervention.

Pio2001 wrote:But nowadays, superko is also necessary for artificial intelligence. A rule with endless repetition left to the referee to assess can't be translated algorithmically into a computer program.

Triple ko draw is not much harder than triple ko forbidden. These programming difficulties are exaggerated, as shown by Katago. Even LZ only knows a rough approximation of some TT-like rules (it doesn't see enough history to understand all cycles and superkos).

The point is, in practice approximations are possible and sufficient. The superko rule itself is also only an imperfect approximation of the real game. See also the last Go AI World Championship where special button like rules were used, which many programs likely wasn't even aware of even though it could reverse some close games.

But nowadays, superko is also necessary for artificial intelligence.

No it isn't. You just make void game a valid result.

Triple ko draw is not much harder than triple ko forbidden. These programming difficulties are exaggerated, as shown by Katago.

Correct. I don't think anyone can reasonably argue even that superko is desirable, let alone necessary. If a bot is to be made capable of playing against humans (the vast majority) it has to be able to play the same rules as humans. In exactly the same way it has to be capable of accepting a jigo when the human sets the komi at 0.

It is insufficient for a void game to be a valid result when AI shall make decisions. Decisions must also be decidable. This is achieved by, e.g., specifying that the result "void" equals the result "score 0" for the sake of strategic decision-making.

Superko can be seen as desirable because it is the simplest rule text and smallest number of rule cases for preventing infinite game play.

Superko is not necessary for the sake of preventing infinite game play because alternative ko rulesets exist to achieve that. For each of them, the same applies. For this specific purpose, one can choose any ko ruleset achieving it but does not need to choose any particular. In particular, one does not need to choose a ko ruleset comprising a basic ko rule, a void result ko rule and a hypothetical ko rule, as is certain tradition.

The same rules as humans? 1) Humans have different rulesets. 2) Encoding some of them is more difficult than creating an AI go software for simple go rules:)

It is insufficient for a void game to be a valid result when AI shall make decisions.

Robert: in most cases of colloquial English, sollen != shall. Say something like "when AI is supposed to make decisions." (Or expected to, etc - anything but shall.)

And if that's not what you mean, I have absolutely no idea what you mean.

John Fairbairn wrote:

It is insufficient for a void game to be a valid result when AI shall make decisions.

Robert: in most cases of colloquial English, sollen != shall. Say something like "when AI is supposed to make decisions." (Or expected to, etc - anything but shall.)

And if that's not what you mean, I have absolutely no idea what you mean.

The indicative makes is probably good enough with when. Grade school grammar calls it present tense, but it is often used without regard to time. In this case we have a condition which could occur at any time.

When I was a kid, shall was on its way out in American English as indicating future tense. Now the pluperfect has virtually disappeared.

Shan't we remember Thatcher for her Queen's English ("We are the Prime Minister!") and me for non-colloquial speech?;) At (why, German, of course) school, we were taught shall = sollen | werden. For a few years after school, use of shall / shan't was much appreciated. Afterwards, it has become a sacrilege to dare using either. Only this Queen may use them;)

"is supposed to" is fine.