Good territory scoring rules for training computers?

[luigi]

Sounds like playing Go under true Japanese rules would make for a good Turing test.

[Bill Spight]

Back in the '90s, Lasker-Maas rules, Berlekamp's rules, and my rules, all of which have an encore (second phase, possibly optional) were devised to be played by humans. Back in the '70s I also wrote some rules to be used by humans. In the '60s Ikeda devised a number of territory rule sets with encores, also for human use.

Sure, but I wonder if any of these fit? Most of them change correct play (dame needs to be played, some even have further artifacts), and yours change the 1st phase (doesn't stop on two passes). Actually I'm not sure if it's even possible to met those conditions.

Not playing dame in the first phase under Japanese rules used to be the custom, and it was allowed by a lenient reading of the '89 rules, but after some end of game problems occurred, the custom has changed to filling the dame, I understand. Certainly, playing all the neutral points in the first phase is, except for rare anomalies that depend upon the strange seki rule, correct play. The only way that the J89 rules changed my game was to fill all the dame before passing, to avoid strange sekis. As far as I can tell, the changes in the rule sets I mentioned to correct play in the first phase depend upon the different scoring in the second phase, or in the changes in the ko rules. If you count points in seki, for instance, that can have a large difference in correct play in the first phase.

The third pass in my rules is a result of having passes lift ko or superko bans, which is to have kos resolved in the first phase. The Japanese '49 rules simply decreed that kos be resolved; the Japanese '89 rules let unresolved kos be played anew in hypothetical play, so that leaving them unresolved would be a bad idea, as a rule. I know of no rule set that reproduces the Japanese '49 rules with an encore, hypothetical or not. And I know of no rule set that reproduces the Japanese '89 rules with an actual encore.

[RobertJasiek]

Since Japanese-style rules with a single phase CAN easily be used for most human games

Sigh. Such a claim has no meaning. "Can be used" overlooks the possibility of pass-fights, which mean that a single phase does NOT reproduce human games - in almost all human games. If, however, you design rules with a single phase, with territory-like scoring and without pass-fights, they are not Japanese-style rules. Button rules are not single phase rules because there is the button. Territory scoring rules with a single phase and playout alternation are known to have frequent pass-fights. See Sensei's for examples. Modifying such rules to prohibit pass-fights creates highly complicated two-phase rules.

What is possible, see the Simplified Japanese Rules, is Japanese-style rules with a SINGLE PLAYOUT ALTERNATION (the second phase), and those are like almost all human games under Japanese rules and can be modified by adding exceptions for sekis etc. to be even closer. Computationally, playing well under such rules can be trained by AI programs, but we do not get 100% perfect play because AI can play suboptimally during playout; this is not a problem of the rules but a problem of better AI training.

"use two phases, and define the score as the territory score of the board position after the first two passes, with dead stones defined as strings with all stones on what is the opponent's pass-alive area after the second two passes".

Such would need to be worked out.

[Bill Spight]

Button rules are not single phase rules because there is the button.

That is a matter of definition. Button go is a kind of coupon or token go with only one or two tokens of a certain size. You do not have to define token go as having more than one phase.

Territory scoring rules with a single phase and playout alternation are known to have frequent pass-fights.

Robert has a broad definition of pass fight.

Modifying such rules to prohibit pass-fights creates highly complicated two-phase rules.

Original pass fights were of the following kind. Under AGA rules two consecutive passes end play, but a third pass is needed with territory counting if Black passes last, so that White makes the last pass. (The player who passes hands over a pass stone.) This produces area scoring with territory counting, because each player has the same number of stone on the board when the score is counted, so the difference in territory is the same as the difference in area. Some people thought that they could modify AGA rules to produce territory scoring by simply not requiring the third pass by White. This had the effect of players avoiding the second consecutive pass, if possible. They could do that by playing a sente move after an initial pass and then passing after the opponent's response. Then they would pass and force the opponent to make the last pass unless he could also interpose a sente play. Making the last pass at the cost of a pass stone is a disadvantage.

First, these rules are not simply territory rules, but naive territory rules. Second, the disadvantage of the last pass may be eliminated by making the last pass cost only ½ point instead of 1 pt. Then it does not matter who makes the last pass, and there is no reason for this kind of pass fight. That is a property of Double Button Go and one of Berlekamp's suggested rules. It is not especially complicated.

Edit: Not that having the last pass cost only ½ pt. makes AGA scoring territory scoring. It is still area scoring, but worth ½ pt. less for Black. But there is no pass fight.

Edit 2: That being the case, are these rules not naive territory rules, but not territory rules at all?

[moha]

Sure, but I wonder if any of these fit? Most of them change correct play (dame needs to be played, some even have further artifacts), and yours change the 1st phase (doesn't stop on two passes). Actually I'm not sure if it's even possible to met those conditions.

Not playing dame in the first phase under Japanese rules used to be the custom, and it was allowed by a lenient reading of the '89 rules, but after some end of game problems occurred, the custom has changed to filling the dame, I understand. Certainly, playing all the neutral points in the first phase is, except for rare anomalies that depend upon the strange seki rule, correct play. The only way that the J89 rules changed my game was to fill all the dame before passing, to avoid strange sekis. As far as I can tell, the changes in the rule sets I mentioned to correct play in the first phase depend upon the different scoring in the second phase, or in the changes in the ko rules. If you count points in seki, for instance, that can have a large difference in correct play in the first phase.

By not changing correct play I meant the players don't know in advance or prepare for that there will be a dispute or extra phase. They play by the assumption that the game can end the usual way with agreement. So the extra rules must be suitable to be applied unexpectedly, after a normal game stop, without this being any disadvantage for a player. Moves played assuming agreement must be ok even if dispute happens - hence dame fill is out.

Ikeda notes that between 1st and 2nd phase there is a point where playing worth nothing for either player, even letting the opponent move twice is no problem (that's why two passes happen). I think this needs to be taken further. The players consider all plays worthless on the above assumption of agreement. So when they realize there is no agreement, their opinion and strategic choice of moves may change.

I now think this is obviously possible, the rules only need three phases (with the 2nd still played in territory mode). The slight complication (in case of dispute) is a tiny price for allowing 99.9% of cases to go without ANY complication or (knowledge of) extra rules, nor any rules tampering with the first phase. And in the remaining cases the players can even be reminded of the procedure (by an online server for example - it's not too late to learn about the rules at this point! ).

There is also the minor case where the first pass happened before retaking in a ko, in lack of threat or dame (I guess this is one reason behind your stopping rule - I agree that passes lift bans, this is obvious), but I'm not sure if this needs special treatment (having the extra territory phase may already mitigate this to an extent). Even if necessary, such complicating seems better reserved to extra phases only, I think. The first phase is best untouched.

[Bill Spight]

@moha

I would like to reply without going comment by comment.

Traditionally, for centuries if not millennia, the game ended by agreement, not by passes, and dame were typically left unfilled. There were no written rules until relatively recently. Humans can handle that, with rare exceptions. But then you got the Segoe-Takahashi dispute, where the players did not reach agreement. But it was hardly because they were unfamiliar with 10,000 year kos. The root cause was the failure to reach agreement, the ko was just the pretext. (Nihon Kiin politics was also involved.)

How do you end the game without agreement? As rules became codified, the answer across the go world seems to be by passes. One of the first set of rules to be proposed, by Yasunaga Hajime, ended play by three consecutive passes. Most rules today end play by two consecutive passes. Ing rules require four passes, but few amateurs know why. It is because resumption of play in the encore after two passes starts with no ko bans. Similarly, the hypothetical encore in Japanese rules starts with no ko bans, and only hypothetical passes can lift ko bans. You and I may think that it is better just to have passes lift ko bans in the first phase, but we are in a small minority.

Another thing has happened with the adoption of written rules is that disputes are resolved by play, actual or hypothetical, and not by appeal to the opinion of a top play or players. And that means encores, actual or hypothetical. Unless some mistake has been made, actual encores occur below temperature 0 by territory scoring. The J89 rules avoid that, because hypothetical play addresses only questions of the life and death of stones. But the traditional ruling about Three Points Without Capturing makes perfect sense by play at temperature -1. If the Japanese rules allowed actual play at temperature -1, however, then a player might score a point by filling a false eye in a seki, thus violating the idea that there are no points in seki. Something had to give, and it was Three Points.

I get the impression that you would be happy to return to the days of the Japanese '49 rules, where games ended by agreement, dame were not played out, and disputes were rare. I probably would, too. (OTOH, ancient territory rules with a group tax are attractive, too. ) But those days are gone, and we seem to be stuck with encores, actual or hypothetical, which determine the final results.

[moha]

Thanks for nice summary again. For ending, I'm confident my tiny suggestion itself would not cause trouble. After first two passes, either agreement (99.9%), or play resumes first, without significant rules changes. There can be complications after this 2nd phase, but these are the same without it and can be dealt with as before. But there is no need to pollute the first phase, the main game itself, neither with dame fill, nor extra rules or special moves. (With territory scoring not even extra ko rules are necessary, beyond the simple ko rule.)

the traditional ruling about Three Points Without Capturing makes perfect sense by play at temperature -1. If the Japanese rules allowed actual play at temperature -1, however, then a player might score a point by filling a false eye in a seki, thus violating the idea that there are no points in seki. Something had to give, and it was Three Points.

I get the impression that you would be happy to return to the days of the Japanese '49 rules, where games ended by agreement, dame were not played out, and disputes were rare.

I like (two) passes, with no dame play. And I think the '89 rules are a significant improvement over '49. At least the logic is clear now: only completely clean captures are granted for free. OC, the exact procedure to determine what is a clean capture is up to debate, but this is a minor detail. (For human games it seems best to have two rules variants/options, one with hypothetical play after 2nd phase, the other with transition to area scored encores.) But if the first phase can - and it can! - be left untouched, no ruleset should go against this - and the 99.9%.

[RobertJasiek]

the '89 rules are a significant improvement over '49. At least the logic is clear now: only completely clean captures are granted for free. OC, the exact procedure to determine what is a clean capture is up to debate, but this is a minor detail.

The logic of the Japanese 1989 is clear now? The exact procedure is just a minor detail? Uhm. It is (almost) clear thanks to my 10 years of preliminary study of the Japanese 1989 Rules and 11 months of full time research resulting in the Japanese 2003 Rules and commentary on the Japanese 1989 Rules both explaining them clearly. However, the explanation is the opposite of a minor detail as it also contains a definition of strategy and a conceptual definition of life and death and paved the way to my definition of generalised ko.

http://home.snafu.de/jasiek/j2003.html
http://home.snafu.de/jasiek/j1989c.html
http://home.snafu.de/jasiek/ko.pdf