Anti-escaping and Anti-stalling features

Sounds like a distinction without a difference. The board is scored incorrectly, and the reason, based on what the board shows, is what has been called the autoscore bug—a bug in the AI whether in autoscore mode or not.

If there’s a bug here, it’s that it highlights any territory at all. The board is not supposed to be getting scored since it didn’t go to scoring.

2 Likes

Nice players if they are only confused. This is so a complete failure of automation.

1 Like

If the situation were reversed and the winner had the unrecognized territory, the system would award the “server decision” to the wrong player due to this bug. Specifically, in the foregoing game, if white had had enough prisoners to win, he nevertheless would have lost, because the system does not recognize the bulk of his territory.

Unless I’m totally misunderstanding, there was no scoring done and no territory recognized, only a score estimate, which was done properly, leading to this result:

The confusing thing is that there are little black and white squares drawn on the board as if it was scored, but those have nothing at all to do with the result and are just a visual artifact. OGS should display the board in this case like it does for a resigned or timed-out game.

5 Likes

I had the same interpretation as Feijoa. It’s not a

rather the automation worked exactly as expected but it’s not reflected in the displayed board.

On the bug itself. Just to be clear, it’s that the dead stone detection element doesn’t run under some circumstances, not that the autoscore gets anything wrong exactly. I.e. the board is scored correctly for the stones that are on it but dead stones have not been removed.
And there is no need for the autoscore to even run in these server decision cases.

2 Likes

Seriously? The autoscore bug was first discovered in cases not involving dead stones. Then cases popped up involving dead stones. Some of these were due to score cheating, but most were not. The 2021 autoscore revision eliminated almost all of this problem. Then it began creeping back, largely involving unmarked dead stones (although IIRC, I have in my collection at least one recent case with no dead stones).

Implying that not marking dead stones and the adjacent territory is an acceptable scoring outcome is appalling. In the past, the scoring system had no problem marking dead stones as dead, except in comparatively rare cases where the AI failed to recognize large groups in complicated positions as dead. While it is true that the players can correct the board manually, many (especially beginners) do not know how or even that they are allowed to do so.

Having a system go from marking stray dead stones to not marking stray dead stones is hardly progress. It is a significant problem that should not be shrugged off, as it has been for years. If this systematic scoring failure is to be the status quo, then the compassionate action is for OGS to give newcomers a suitable pop-up warning about the frequent need to correct the poor automated scoring.

2 Likes

I think there’s a misunderstanding here.

  1. The result of the game (B+10.1) was determined by KataGo (called an AI) and is most likely good enough (the correct result is maybe B+10.5). It was not determined by the autoscore algorithm (not an AI). Those are two different systems.
  2. What the board shows is very misleading. The board shouldn’t show anything but the stones (or maybe KataGo’s territory estimation, just like when pressing Estimate Score after a player has resigned). What it shows is a wrongly scored game based on dead stones that were treated as being alive. It shows a result of B+82.5 (see the player cards), but that is not the official result.
8 Likes

This new rule got me quite upset. What happened to the 5 minute countdown? 30 seconds is way too short. I got disconnected today from a game with 30 minutes playtime each, we both had more than 25 minutes left on the clock and were at move 30 or so, so we were both commited to an hour of playing at least. I got a message on my OGS account from my opponent of my previous game, so I quickly checked the message while my opponent thinks about his next move, get back 35 seconds later and I was disconnected… with 25 minutes or so on the clock. I never left OGS. On another occasion I switched from playing on my phone to my laptop as i ran out of battery, also with plenty of time still on the clock, and yep i got disconnected. It is just silly to have this 30 seconds rule especially in longer live games that are agreed upon to last over an hour. The 5 minute countdown was so much better.

7 Likes

Yeah i agree, it doesnt feel very nice :<

Its not automatic rule, its more like a decicion by your opponent when you’ve been away from the game for over 30 seconds. But i know that doesnt make it feel any better ://

5 Likes

but

2 Likes

it is a strange feature…I read now that that button pops up only after a minute, i m not sure but even if it was a minute, the opponent should not have the power to win the game like that when he/she commited to a live game of an hour. I think it is wrong. I am all for catching people stalling the game though, but 30 seconds or 1 minute is just bizarre.

6 Likes

I recently played this game: mariayadri21 vs. PRHG
Why wasn’t I offered the option to claim victory by server decision?

I passed three times in a row, on each turn there was a 100% chance of victory, on each turn my expected score was greater than or equal to 52.9 points, and I began passing on turn 46 (9*9 / 2 = 40.5 turns)

1 Like

No idea.

But tbh I don’t think your opponent was stalling, it was just a clueless beginner trying a hopeless invasion probably unaware of the hopelessness.

1 Like

Actually, after the 2nd and 3rd passes, Black can live by playing H5 (and H4 or other moves after the 3rd pass). ^^

(even though it may not be enough to win the game, it’s an interesting tsumego :slight_smile: )

1 Like

I had window with game open. I created additional window near game window with profile of opponent. And I see that message in additional window. Will my opponent have victory button if I continue to have 2nd window open ? If no, then that message is misleading and should be fixed.

4 Likes

Do the server logs indicate that the ​ Accept predicted winner
button showed up in ​ deming cycle vs. pwsiegel ?


pwsiegel said it didn’t,
Use AI scoring to eliminate score cheating - #31 by pwsiegel
although what I remember of this thread makes me think was supposed to:

For White’s 3 passes between 142 and 149, and Black’s 2 turns between those 3 passes,
the AI analysis pane shows 100% for White and at least +10.5 for White.

1 Like

I am glad it took me several days to read this thread, as it allowed me time to mediate my positions and consider more positive aspects of the anti-stalling feature which I had not initially thought of in my reactionary first impression.

tldr, I am, contrary to my initial impressions, okay with the anti-stalling feature, some caveats notwithstanding, and am glad the anti-escaping feature was increased to 60 seconds, but wish it was reverted to the full 5 minutes.

Anti-Stalling Feature

If I understand the discussion in this thread correctly, attempting to use this feature before the game is effectively over and all dame and teire are filled, is an abuse of the feature and a reportable offense. If this is correct, I am very glad indeed that it is, and it is a huge part of why I am now okay with this feature.

There are many things which players ought not to do, but can, and many are far harder to catch than abusing this feature this way. Certain types of AI cheating, such as setting up a script browserside to notify you with no further information if your opponent’s last move lost more than 10 points AND a pseudorandom bit was equal to 1, would be, I imagine, an incredibly difficult, if not impossible, cheat to detect. How do we react to this possibility? Do we legalize it for everyone? Even make it easier? Do we stop playing online to avoid this? Do we require all OGS players to send a live video feed of themselves as they play? Do we require custom browser clients to be used? Or do we accept that while reasonable effort will be made to discourage such behavior and deal with it when detected, some, especially more sophisticated forms such as I described above, will fall through the cracks, and we will live with that? I think it’s the latter, and rightly so. Don’t condone it, don’t allow it, but make peace with the fact that it will happen anyway.

Abusing this feature is analogous, with one major difference: abusing this feature is incredibly easy to detect in comparison. So if we’re willing to put up with the reality of sophisticated cheating existing, I think we should likewise be willing to put up with the reality of abuse of this feature happening.

A major benefit I see in it, possibly alone outweighing the negatives, is the reduction in mod load whenever this feature works as intended. I can think of a game off the top of my head from a while back where this would have been useful, and changed a realtime mod request, into a non-urgent report which could be looked into at any time. Having thought about this for the past few days, my feeling is that, if as stated above it is reportable to use it before dame and teire are filled, it will cause significantly more good than harm.

Anti-Escaping Feature

I think we already had this with the 5 minute timer, and this just makes it more easily abused in exchange for dubious benefit.

Let’s say we’re playing 10m +3x30s byo-yomi. Let’s say you blitz so fast, that you play almost the entire game before reaching byo-yomi. Even still, you are commiting by playing this time control that your opponent may take it more seriously, use up their main time in, say, 30 moves, and then use most of the byo-yomi period on each turn for their remaining ~90 moves. That’s 10m + 45m spent by your opponent, and 10m spent by you, for a total commitment of at least 1h5m. And this is a relatively standard time control. In light of this, 5 minutes is a totally reasonable amount of time to give players to fix their Internet connection without needing to rush or panic (remember, their time is still ticking, which at this time control has a good chance, 100% if they’ve reached byo-yomi, to end the game before 5 minutes by timeout anyway, and if not, they’re penalized with the lost time and disruption).

But maybe you play longer time controls. Maybe you don’t want to rely on a disconnected opponent running out of time to end it before 5 minutes. Well, to get that, take a time control of 30s + 5x60s byo-yomi. Now if you blitz fast enough that your time used is less than what the opponent leaves unused at the end of each byo-yomi, you’re still committing to playing a game where the opponent, if taking it seriously, could spend 1m each move for ~120 moves, for a 2 hour game. Now waiting a paltry 5 minutes is even more reasonable, if that was possible.

At the sorts of time controls where a 5 minute disconnection timer is more relevant then running out of byo-yomi periods, 5 minutes is the least you can promise to a player suffering Internet trouble.

6 Likes

I just found out about the anti-stalling feature in this game elsantodel90 vs. jkim3902. I tend to always fill dame because I find it nicer and it does not take long (and tend to teach and recommend that to those I teach, although mainly focusing on over-the-board playing, where it is furthermore good idea to reduce counting mistakes). In this game in particular, there were only simple dame left and margin of victory is large so no cheating at all and everything is ok.

But, I will soon use OGS to run a beginner (at most 16kyu) tournament for our National Go Association (we hold one of those per year online, typically with games held over OGS). Beginners are OFTEN confused about when to end the game, they often pass with open borders and such. You can see all of the last two such tournaments’ games in Torneo Nakade. In particular, if you check this official tournament game, players ended quite prematurely (in chat, I advised them to avoid this for future games as if some complication happens we might need to fork the finished game to continue playing to resolve it, something which right now is quite easier to handle as a beginners’ tournament director when playing over-the-board instead of OGS). Luckily there were no disputes among players and it was not a close game, and the result happens to be ok, but the final score (margin of victory) was in a sense “extremely wrong” in that case.

So, I think that this new feature might unfortunately make my task as a tournament director for beginner players (a tiny bit) harder. Is there any way at all that I can disable it for a specific (can be unranked) game? [I assume no, would usually be a weird setting for normal OGS single games among strangers but some (tiny) fraction of OGS games are “tournament games with an actual tournament director in contact with players” where it would make sense to avoid any form of auto-score at all, to mimic over-the-board tournaments].

Also, can some admin officially confirm exactly WHEN is ending the game clicking the Anti-Stalling “computer autoscore” new feature unnacceptable and thus reportable as Samraku suggests on OGS in general? As it is not clear to me and there are many options.

Option 1: If there were any dame left (say, chinese rules 1 point), even those completely gote and not needing any kind of response, you can report. [e.g., with this option I would be able to report in my game above because I was still playing dame. I think most people would not consider this ground for report]
Option 2: If there were any “required teire move” left, so that blindly passing against dame would result in a loss of even a single point of score, even if it does not change the game result if the threat is carried out and executed, you can report [some people before mentioned this problem, and the line between Option 1 and Option 2 can get a bit blurry sometimes]
Option 3: If there were any “required teire move” left, so that blindly passing against dame would result in a change of the game result compared to autoscore, you can report [a lot of people before voiced concerns about this corner case as cheating, but I think the line between this and option 2 is extremely blurry]

Furthermore, even earlier than dames:

Option 4: If at move 181 or so, a player has amassed such a large advantage that passing three times in a row would still give certain victory (say, we are entering yose and one player killed and actually captured the opponent almost everywhere and has rock-solid groups), is passing three times and claiming victory while completely skipping yose (as the advantage is already huge) acceptable? Or should such a thing be reported by the opponent? (etiquette suggest the opponent should have resigned, but typically even if it is bad etiquette “you have the right to play until the proper end of the game, even in a hopelessly lost position”).

I feel that the spirit of the rule is in good faith, but even explaining when is it ok to do and when it is not ok, is already fairly tricky. I am also not sure if I should brief/warn the beginner tournament’s players about this new feature in advance, or not (As I might confuse them even more).

5 Likes

I would also like clear rules on when it’s reportable to repeatedly pass. I would suggest whenever there are any dame left, which also includes any time before dame-filling. One pass to see if your opponent passes back to go to scoring is okay, of course, but if they play, then if there are still dame left one would need to play in order to avoid being reported

4 Likes