This is a ladder challenge. When scoring began, the dead stones were automatically recognised as dead by the system, rather than waiting for me to mark them as dead. There is nothing in the “Game information” popup that seems to indicate why the scoring is somewhat more automated than usual. I assume that’s because it’s a ladder challenge, but still I find it weird that sometimes groups are automatically recognised as dead, and sometimes they aren’t, and there is nothing in the “Game information” popup to tell me which one it’s going to be.
Anyway, this game was played with Japanese rules, and three neutral points were left unplayed at the end of the game: C9, E5, F4. I don’t like leaving dame unplayed, but this is a correspondence game, so playing out the dame would uselessly prolong the game by quite some time.
C9 is marked with a blue square. E5 and F4 are not marked at all. This marking was done automatically by the system immediately when scoring began, with no input from the players.
What does it mean? Is the blue square noting something special, like a distinction between dame in a seki and dame outside a seki? Does it affect scoring? Or is it just a bug in the display?
I think it’s probably true that they’re more or less the same. The blue is meant to mark dame I believe.
It might be some funny interaction with Katago estimating the status of points, that some are painted and marked as dame and others aren’t. Like maybe white plays some dame in most variations and black in others, but either could play the ones marked blue.
As in imagine Katago returned a number between 0 and 1 and if it’s 0 it’s black and if it’s 1 it’s white and if it’s 0.5 it could be either, and then some logic on top of that has to decide whether to paint a territory black or white using a flood fill or override it and call it all dame.
Anyway I do agree that it’d be better if one could consistently apply one convention or the other.
but either AI auto-scoring is not as smart as one might hope or it’s buggier than one might hope (or both)
e.g.
And on this point: (a bit off topic)
I thought there was some discussion about this maybe depending on players being on the game page at the time of the game ending - e.g. if both players are actually in the game page then autoscore is triggers but if not then it is not. it’s unusual I suppose for both players to be there in correspondence games but maybe not as unusual as you think if they both feel the game is about to finish and hang around to play the last move or two and pass.
That sounds like a very arbitrary way to decide whether to use the autoscore or not.
And it’s annoying that in some situations the system would decide to automatically switch to autoscore.
Is there a way to choose “Never autoscore, even if I’m on the gamepage”? I don’t see anything good about the autoscore, and I’d like to avoid it as much as possible.
Actually, what typically happens to me is that the opponent will hang around for the last ten moves… Then the game goes to scoring and the opponent leaves without accepting the score. Then I have to wait one or two or three days before the score is finally accepted.
I’m guessing this might be a consequence of the “Auto-advance to next game after making a move” option, which automatically whisks my opponents away from the scoring page without allowing them to accept the score. I hate this feature, and I hate that it’s activated by default, and this is just one more reason to hate it.
Blue square marks can also be used to mark neutral (non-territory) points within apparent territory, for scoring under Japanese rules.
Often players are too lazy to fill in all dame, which sometimes leaves teire points that should be filled later.
For example:
Since the dame at X was left unfilled, White was never forced to play a protecting move at the square. However, if both players have passed, then they can pretend that those two moves were played by putting a blue square at C2 to cause it not be counted for territory.
The blue squares can also be used to mark the eyes of a seki position, since those should not be counted as territory under Japanese rules.
I think autoscore is always supposed to activate, but maybe it only does if at least ONE of the scoring players is on the page, which causes a bug only in rengo.
In its defense, I’d like to say that with some limitations (borders are really closed and secure, and also no weird Japanese rules things like points in seki) it is basically 100% accurate. For my level (SDK) it’s way more reliable than my typical opponents, and I don’t think I’ve ever had a game that it couldn’t score properly.
I’m pretty sure that in most games that I’ve scored recently, I was on the page, and it wasn’t rengo, and autoscore didn’t activate. I marked the dead groups as dead myself.
Off-topic personal opinion about the autoscore
In my eyes, this statement is self-contradictory.
If you and your opponent agree that a group is alive, then it’s alive, and if you agree that it’s dead, then it’s dead. If you don’t agree, then play it out. The AI cannot be more reliable than your opponent at knowing what your opponent knows.
The AI might possibly know a way to kill one of your groups that your opponent think is alive, or the AI might know a way to make a group of your opponent live that your opponent think is dead, but that should be irrelevant to the game, because this game is between your opponent and you, not between you and an AI or between an AI and your opponent.
The autoscore is somewhat convenient because it saves you half a second of clicking on the dead groups manually, and when playing against a bot it’s very useful because it completely sidesteps the “accept scoring” phase which wouldn’t work well between a bot and a human, but if it disagrees with the players, then by definition, it’s wrong.
It won’t disagree with both of us, just my opponent. For example, there’s a seki, and they think they get to mark my stones dead. Or will just randomly click some stuff that doesn’t make sense. It’s only a few percent of players who act this way, but autoscore’s error rate has been basically zero for me.
In other words, if I ever disagree with my opponent, I’m always happy to just click autoscore on my side.
Well, it takes two players to play the game. Your opponent is a player too.
What about these situations:
You think your opponent’s group is dead. Your opponent says no, it isn’t.
What do you do then? Do you use the autoscore to force your opponent’s group to be dead?
Now imagine that it’s indeed possible to kill the opponent’s group, but the correct sequence to kill it is actually quite subtle and if you mess it up, they’ll live. You say your opponent’s group is dead, your opponent says no, it isn’t.
Do you use the autoscore to force your opponent’s group to be dead? If you do, I would argue that it’s cheating. You’re using the AI to magically-kill a group to avoid playing the killing sequence yourself. Notice that there is practically no difference between situation 1 and situation 2. If you and your opponent disagree on the status of a group, then obviously the status of the group is somewhat subtle. So, situation 1 was cheating too.
Now imagine that your opponent has a group that is dead; the sequence to kill it is relatively easy, you would definitely find it if you tried, but in the game you actually didn’t notice and so you thought it was alive. Now you arrive at scoring, and you click the autoscore. The autoscore magically marks the group dead. You notice that it’s dead (you hadn’t noticed before), you read the sequence out and you agree that yes, it’s dead.
Again, I would argue that this would be cheating. The autoscore helped you notice that your opponent’s group was killable, when you were ready to let it live.
So, yes, the autoscore’s “error rate”, from the point of view of a high-dan player, would be basically zero, but relying on this superstrong autoscore to score the game is cheating. You’re using an AI to finish the game in your place!
What about if the situations are reversed and you think your group is alive, opponent disagreed and you hit autoscore to make the determination and accept that outcome. Is that cheating? Surely you can’t “cheat” your way to a loss.
If your opponent thinks they can kill you then they should prove it. If they can’t and they use the autoscore then they’re cheating, using the AI to kill you.
If you click the autoscore to help your opponent kill your group, well, that’s just a weird way to end the game.
I mean, if you wanted the AI to finish the tsumego for you, you didn’t have to wait for the scoring phase. You could also agree with your opponent to stop the game at move 100 and let the AI evaluation decide who wins.
Sure, if you agree beforehand that you will accept the AI’s evaluation no matter whether it’s in your favour or not, then it’s not cheating. But it’s also not the game of go. It’s a go variant where you don’t have to finish the game yourself because the AI finishes it for you.
I think you’re exaggerating quite a bit here on how often the ai is likely being used to determine statuses of groups, versus marking what players likely already know. If you’re not implying this, then I’m probably not understanding what’s being said in the previous post regarding
^^ this comment in particular.
Probably in the high 90s of percents of games it’s fairly clear when a group has one eye, or no eyes or whichever and the autoscore is just quickly scoring the game the same way players would.
Some sekis are a bit trickier, and there was a slight issue with the autoscore marking the board in an unusual way when there’s aji but that is also only half the time it could be because it seemed to depend on who’s turn it was next. There’s a thread about autoscore and beginners for example I can look up, and possibly another thread about other behaviour.
Anyway it’s not to say the system shouldn’t be improved, more a question on what it should be improved to.
Just to go back to
that doesn’t really work in Japanese rules because it costs one player points to play it out which can change the result of the game.
Regarding
I read what @Feijoa was saying as “the autoscore marks the game the way I want to almost all of the time”, however their opponents can do something entirely random a small percentage of the time. For instance some players try to cheat in the scoring phase by unmarking dead groups or small patches of area and hope they get away with it. Or in other cases they don’t recognise a seki or not a seki. There’s still no “play it out” in Japanese rules, without a proper separate phase implemented for it.
With optimal play, as far as I understand this shouldn’t be the case, but it probably is the case with amateur games.
which is why area scoring is cooler
But still I feel that @ArsenLapin1 is right in that the person who believes they can save their stones has a right to prove themselves, and that it’s against the spirit of the game to hide behind “but the AI says your stones are dead”. Wanting to win on a technicality and losing the opportunity to learn why exactly some stones are dead or alive is also against the spirit of it, imho.
In some cases it might arguably be “rude”, by means of putting the pressure on the other player who shouldn’t have the burden of proof, akin to a case of speculative invasion. And I guess that’s quite subjective.
If instead it’s a competitive situation, such as a tournament, then it’s about who’s the best player. If defending costs you point, it means you lack some skill, and it arguably means your opponent had the skill to create a tricky situation where you would end up losing more points than you expected.
This is part of why I don’t personally enjoy pushing the competitive aspect too much, but if competition is the point of a specific match, there are no “low blows” other than the ones explicitly stated by the rules. Right?
If someone refuses to accept a group was dead with say for example a killable 5 or 6 point eyespace, you would lose a lot of points proving that it’s dead by removing it from the board.
You’d have to keep giving prisoners to shrink the 6 point eyespace down to a five point one, then to four, three two etc.
I don’t like using terms like straw man but literally this is something that’s just been invented to attack. I don’t think it was the case that anyone said this is what they do, nobody really disagrees that using the AI to gain information about the game wouldn’t be cheating.
It’s probably a very small number of cases and small number of players that probably do it, maybe even less than just outright cheating by changing the status of stones in the scoring phase.
So I’m not sure what or who it’s really arguing against.
I guess I’ll step back and list the salient points of the discussion as I see them.
ArsenLapin said his opinion is that the status of a group as dead or alive depends on whether the players agree on it, and that if the players don’t agree on it, they should play it out until they can.
Then he said something that I personally read as “if one player, hypothetically, refused to play it out based on the AI’s judgement, and wanted to force the AI’s judgement on the other player to their own advantage, that would be equivalent to cheating, specifically because they wouldn’t necessarily be able to play it out as well as the AI”.
At least in my interpretation, he never said “a lot of players use this to cheat, this is a problem”, and I would have no tools to agree or disagree with that.
So to me the discussion was more about the quasi-philosophical problem of “should players be forced to play it out if they disagree? Like, hypothetically?”
So I wasn’t arguing against, or “attacking”, anyone. What I was arguing against (and maybe not even arguing really, mostly just stating my opinion about what I believe to be a moderately subjective thing) was the notion that players shouldn’t be forced to play disagreements out.
It seemed to me that your point was essentially “if you play it out, the winner might (/would?) still have to lose points in Japanese rules, and there exists the AI scoring, so why not solve the problem by using the AI scoring”.
And my counterpoint to that was 1) if you’re good enough you shouldn’t lose points, and 2) using the AI feels against the spirit of the game. The second part is subjective, the first part is complicated but possibly arguable in an objective way.
So hopefully I’ve cleared myself, and I can go on to respond to the technical point:
Now, I’m not an expert or strong player, so I might be wrong, but I’ve read many times that area scoring is equivalent to territory scoring, and it affects play mostly by the fact that territory scoring punishes inefficient moves. One part of the explanation of this counterintuitive phenomenon, specifically applied to the kind of situation you’re describing, is this:
If the opponent has a dead group in your territory, in Japanese rules all of those stones are your prisoners. Every move they play to try to make it live will give you one point. Every time you respond with one move of yours, you lose one point if the stone you put down survives, and you lose the same one point if it instead gets captured. Your net loss is always zero, as long as you don’t play more moves than your opponent.
If you need to play more moves than they do in order to ensure their death, it means the group wasn’t really dead after all. If they pass, you pass. Do we agree it’s dead? No? Ok, keep playing. They play, you play. They pass, you pass. And so on.
I really can’t see a way where you can lose points as long as you don’t play inefficiently. At most, you can lose one point if it’s your turn to play after the first passing and disagreement, but that means it’s just what would have happened if the game had proceeded uninterrupted, so arguably that point was never really yours.
More in general, I believe that the only way to rigorously make sense of the usual scoring rules of Go, where you judge some stones dead etc., is that the score is supposed to be equivalent to the score that you would obtain by literally playing all legal moves that don’t require filling your own eyes.
If the two players kept playing as long as they can, making dumb invasions etc, in the end you would end up with a board completely filled with stones with the exception of two eyes for each group.
Each player’s score is the number of eyes plus the number of prisoners (plus komi). That score should be equivalent to the score usually calculated.
I don’t know about seki though, I don’t really have a good grasp on how different rulesets handle it. I guess with the above example winning a seki would be a matter of managing tempo, so it’s really difficult to predict who would win it, so the “approximate” scoring rulesets find some compromise.
I think these two paragraphs are at the core of the issue.
The way AGA/British/French rules do it is that passing costs one point if using territory scoring, and White must play the last move (so that both players have played the same number of moves). With these two rules, territory and area scoring are indeed equivalent, and passing is mostly equivalent to playing inside your own territory, or playing a doomed stone into your opponent’s territory.
However, Japanese rules don’t have this “passing costs one point” rule. So, if White has a group with 4 liberties inside Black’s territory, and Black wants to actually capture it, Black has to play 4 moves insider their own territory, which costs 4 points. If White also plays 4 moves that cost one point each during this time, then yes, the score doesn’t change. But what if White passes? After all, it’s Black who is arguing that the White stones are dead; as far as I know, no ruleset forces White to play “useless moves” while waiting for Black to capture the stones. In AGA/British/French rules using territory scoring, White can pass at the cost of one point per pass; but with Japanese rules, if White passes, Black will end up having lost 4 points.
On a small board, it might even be a problem that White doesn’t have any available useless moves; for instance, if White has several small groups with exactly two eyes per group, White can’t play in their own territory because that would fill their own eyes.
Yeah. The typical thing for me is that I arrive at a game to be scored with some strangely-marked spots, and I don’t know how they ended up like that. So I click autoscore, everything looks good, accept score, done. Or if it still looks incorrect after autoscore (this is rare if it happens at all), maybe I’ll learn how I was wrong.
OGS doesn’t support the scoring phase very well, and to me that’s not an interesting part of the game, so I’d rather just get the right score quickly and be done with it.
