Lets say I have 20 active games against equal ranked players.
10 of the games aren’t decided yet, for 5 I’m clearly winning and for 5 I’m losing (but haven’t resigned yet).
Now lets say I cannot play any moves for a week and I time out of all of those games.
If I come back:
At the moment: I will lose 1 “random” game. My rank is slightly lower then before. There is a 25% chance that the rank of an opponent I’d won against got the win and now has a win instead of a loss.
All games were decided by AI: I have 5 wins and 5 losses (my rank is the same), 5 of my opponents got their deserved win and 5 will have a loss for games.
The AI decides the games only as lose/annul: I have 15 annulled games and 5 losses.
All timeouts count as losses: I now have 20 losses.
Is my rank in case 3 now more accurate than in case 1 or 2?
Imagine we allow the AI to give the win to a person who timed out. This would potentially give people incentive to intentionally time out a game they’re ahead in. Because “winning a won game” can be a challenge too, and I don’t like this aspect of your proposal.
You assume that we can all detach all emotion from ranking, and merely rejoice in the fact that it’s accurate. I reserve the right to be sorely pissed if my opponent times out and gets a win if, even in the face of Leelas judgement in favor, I believe I could have won.
Leelas judgment does not take into account that players can excel in several aspects of the game - perhaps you are terrible at fuseki, but routinely gain 10+ points in the end game when playing against people of your own ranking who suck at end game but got fuseki down like a pro - these players will always be judged out by Leela who’s 98% certain she can pull a 2.5 point win.
Speaking from the perspective of someone who times out, sure, you’re right. Number 2 is the most accurate. However, we are not dealing with emotionless robots who value accuracy above all. I propose that a majority of people WILL consider it unfair punishment to lose a game by AI decision after their opponent timing out, and that in my opinion is orders of magnitude more important than whether or not objective accuracy is maximized.
Add to that the fact that we’re already concerning ourselves with whether or not people are gaming the serial timeout mechanics - we absolutely do not need to add a layer that under any circumstances might actively reward such behavior with a win.
Then I assume you agree that whether or not the mechanics FEEL fair is as important a factor as whether or not they are objectively as accurate as possible?
Do you agree that if either party feels unfairly treated, it should be the person who times out, since that actually is a well established losing condition, and they could have taken a series of steps to avoid timing out?
I guess the only question I have, and was confused about since my last comment was whether or not in the proposal ai’s are asked to judge every individual timeout, or only timeouts that are part of a serial timeout. [ I think my last post probably only matters if it’s checking single timeouts that aren’t part of a group of collective timeouts at once]
So say the new proposal takes effect in some shape or form.
Case 1: Imagine I’m playing 5 games, 4 are correspondence where I have a weeks time left and one is a ladder where I have only 3 days time. I forget to check over the weekend so I time out of only the ladder game. Does the ai decide this game? I’m definitely back in time to continue my other correspondence games and maybe I finish some before any other time outs can occur. I expect the usual being knocked out of the ladder etc happens regardless.
Case 2: same amount of games say. I forget to turn on vacation so I time out of all 5. Does the first game count as a loss as usual? Then since there could have been 3-4 days before the next 4 timeout at roughly the same time, does the ai decide all 4 of these games, or is the first one still counted as a loss because it’s the first of the batch of near simultaneous timeouts?
It seems case 1 is already covered in an okay manner at the moment, I get one loss I can rejoin the ladder at the bottom and then I go on playing as normal, and probably have the “recently timed out of a game” tag for a bit.
Just with case 2 I’m wondering what the timeframe for counting a bunch of games as one block of timeouts is, and which the ai should be asked to look at.
Like if you somehow managed to timeout of a live game and went away then timed out of a bunch of correspondence later on with no completed games in between, does the ai look at all of the correspondence ones or all of the correspondence -1 (minus the first one).
We are talking about "At the moment if a player times out in multiple correspondence games in a row only the first game is ranked and the others are silently annulled. "
Timeouts that are not serial are already losses, because timing out is a loss under the rules.
Serial timeouts get annulled due to concerned about effects on the rank pool if this were not the case.
This proposal is how to deal with the adverse effect on the opponents who otherwise would have had a win, and the adverse incentive this creates for people to time games out in batches.
I am against current series timeout rule, but I get over it.
Overtime, it does not matter.
Using AI is not fair at all, because it is too strong. Its 70% win rate is based on its own strength and is not a good indicator of me as a 3kyu to hold the lead to the end at all. Many times, I don’t even understand why I am at 70% win rate.
not trying to be rude but quite a few people, including myself, had already mentioned that the one who timed out should be allowed a draw at best, even if AI says he was winning.
we are not talking about whether AI is perfect, or does it accurately predict the game, or can you match AI’s move to capitalize on the win rate, etc.
In my humble opinion, the question is: do we have a better alternative than AI, that can solve the current problem. and the answer is no. ai is not perfect, but we don’t have anything better, yet.
any improvement, no matter how small, is better than no improvement. i mean, unless you think the current system is just fine.
It’s just a game. Correspondence games are not even serious one by design. Only live or in person games are truly fair in my opinion.
Again, even it is not fair, I actually agree timeout is a loss, it does not affect our rankings that much over time. And I also had timeout games where my opponents were winning. It all balances out in a long run.
I think useful debate on this topic seems to have run its course. I don’t see any reasonable additions or complaints after adding that it can’t decide in favor of the person timing out.
This seems like a fantastic idea and I hope it gets implemented soon!
I am aware. However, simply stating this as a fact is no way to have a proper discssion. Since flovo clarified that his proposal includes the possibility that AI declares the player who timed out the winner, I felt it was adequate to give a good argument against this.
I’ll just add this comment here to be aware (although lots of ogs mods/users are aware) that it might not be possible at present to use the ai to decide correspondence timeouts as part of serial timeouts if there’s handicap, since leela doesn’t seem to notice the reduce Komi and can output incorrect win percentages eg 80% win for white even though black is winning by 5.5. So white would be winning with normal Komi (6.5/7.5) but not with reduced 0.5 handicap Komi.
I know also that people have mentioned elsewhere that handicap games are in the minority on the server so it wouldn’t be a large issue anyway. It might just be worth noting for later in this topic as it is linked here by the github page
sorry to bring this back but is the issue resolved? a few of the OGS team members seem to like the idea and sounded like they would at least think about it.
On principle I don’t like using AI to decide game results, particularly if it means choosing to lose by timeout can actually make you win the game, effectively a superpower to force your opponent to resign. In general I get the feeling the OGS philosophy nowadays is too interventionist and trying to rewrite history of what actually happened naturally in the game. Go is a game where you put stones on a board under some pre-agreed rules and clock settings, have a battle of skill, and then you count the game if someone didn’t resign or run out of time before then. Admin intervention in that process should be limited. Losing on time sucks, I should know as I do it often: I lost a British championship game on time when I was ahead. But I blamed myself for playing too slowly, and the clock for not having an audible warning so lobbied to get that improved in future events, but the idea of appealing to an AI or human arbiter to award me the win because I was ahead on the board when I lost on time is so utterly alien and antithetical to fair play it didn’t even cross my mind.
The idea suggested in another thread that closing boundaries of territories is tedious so let’s not bother and allow marking of territory unclosed areas or ask an AI who won is utterly ridiculous. Or moderator annulling a game where someone complained after they missed an atari, with the rationale that 2 bad moves, a stupid sente and a dying in gote many moves prior were somehow “stalling” and against the rules and thus worthy of annulling the game (yes it was overturned, but that the idea of 2 silly moves in a game of several hundred could give an admin the right to change the game result shows they are too interventionist). Or moderators annulling the correct score of a game with unfinished boundaries that the players agreed upon due to player inexperience.
