It’s no secret that there are a number of people who are uncomfortable with being stuck at a level they use outside software to get ahead. I thought it would be worth it for a software engineer who works at OGS to develop software that is more powerful than what we have on the market. Its goal is to help claim managers get a decision faster and not wait a long time. If such a thing were possible here and on other sites, would you play Go?
We already have various tools of that nature, developed in OGS, which moderators use.
Slow decisions on this topic are because it’s a fundamentally difficult call to be certain, and there are very few moderators willing to suffer the abuse we receive when we make these decisions.
6 Likes
I understand, I think you should consider doing it once a month or two. Go over games, but all or those that are suspected of this will be deactivated.
“Suspected → annulled”
Thanks for the suggestion, I don’t think that would be popular.
4 Likes
We certainly get a lot of baseless accusations of cheating from sore losers haha
5 Likes
True, but I’m not sure the OP was meaning “accusations from reports/users” they were meaning “if the tools suspect AI use then annul the game”.
We have a different approach: if the tools suspect AI use we carefully scrutinize to remove any reasonable doubt before annulling the game and giving the person a final warning.
It’d be amusing to see the results of a poll asking “would you support annulling any games that tools suspect are AI influenced” 
Of course in practice, that poll would be useless because people would ask “how good are the tools and how do they work”, neither of which are questions that can be answered.
Except perhaps to say that the tools are not yet good enough that we trust them without scrutinizing.
4 Likes
I doubt this is the right approach to go scanning all the games. Surely too costly even if we got enough good and accepted tools.
Online go is developing its specific culture and a lot will come from education. For example I see many times streamers using score estimators (with the evaluation of what is owned by each player) during games, I mean even to check if their strategy is great when they are playing. For their pride they won’t use AI suggestions directly, too shameful but estimator looks to be less controversial. It’s maybe fun to watch but this is bad education on how one should act in a game. It’s no so easy to fight this trend in which players wants to use the new benefits directly when playing. Who leads? Is this alive? What’s the way to connect? I am searching something which was anyway not working? Something can give you answers, easy and quick to use. Mostly education will avoid people to use it.
5 Likes
Detecting AI use is difficult, there are always false positives and false negatives. If OGS were to develop a tool, I think that the following would be easier and more useful:
- Automatic detection of sandbaggers
- Automatic detection of users who register at a wrong rank (like novice players who register as “intermediate” or “advanced” players and annoy their opponent by playing nonsensical moves).
I’m setting up your resident. I think it’s really important to check this out if you had anyone who helps ogs develop want AI to suggest to you what to do to the player while you’re at it again in the future. If you had a load of requests then take this as an idea AI could tell you a player did something wrong suggest to you what to do from here I’m not talking about it
Very sorry but I have quite some difficulty to understand what you want to tell me. Are you using auto translation tools?
3 Likes
Yes. I just wanted to help them improve your decision-making ability regarding cheaters. AI will give you suggestions on what to do against a cheating player to get a decision within 10 minutes.
I would vote against automated cheat detection applied to all games.
First, it’s expensive. AI analysis uses CPU/GPU cycles and consumes electricity. Applying it to millions of games comes with a dollar cost. Implementation will also take up OGS developer time that could be better spent on other features.
Secondly, a small number of false positives will create a lot of bad feeling. And I don’t think it’s feasible for every single flagged game to be reviewed by a moderator.
Thirdly, and most importantly, there’s just not that much at stake! Most of the games here are basically social games. There’s no prize money and not that much prestige. We don’t need heavy anti-cheating systems and messaging that promotes a low-trust mindset.
If someone is cheating but they’re ranked at, say, 3 kyu (or really anything that’s not 9 dan) then they’re not very good at cheating! Or else they’re only using more subtle forms of assistance, or doing it rarely. In any case, the ranking system still works: another 3 kyu can play against them and have a 50% chance of winning. I think it’s pathetic on behalf of the cheater, and I’d prefer it if people don’t cheat, but I think the damage is minimal.
I’m reminded of xkcd: Self-Driving Issues
(Exceptions: if there are ever games on here with prize money; if there are complaints from lower ranked players being beaten up by suspected sandbaggers; if someone is putting out publicity of the form “my high OGS rank proves I am awesome and you should pay me lots of money for lessons” – I think our current setup can already handle these situations.)
5 Likes
They may cheat every other game, I’ve already seen that: strong reading skills in one game, followed by slow moves/bad shapes in the next game. That’s annoying for everyone: both opponents wanted to play against a 3k, but one of them plays against a 3d and the other one against a 8k.
1 Like
My performance was that erratic years before the invention of strong go-playing AI 
Joking aside, have we actually put a sample of such games through AI analysis or collected other evidence of cheating, or is that a subjective impression? My own go performance has been a little more stable since reaching dan level. But in badminton, where I’m playing a lower level, a few times I’ve come up against “unreasonably strong” players in C grade events, and then comfortably beaten the same players on a second meeting. Or beaten them ton the first occasion and been wiped out in the rematch. And I can’t blame that one on AI assistance. Human beings are just not that consistent in general. Could be either the opponent or myself having a good/bad day.
I’m not disputing the fact that some people cheat some of the time. Just wondering if we have any objective measures of how big a problem it is here.
2 Likes
Actually I only have one example of someone who probably cheated every other game. Knowing his real name, I had reasons to suspect he was cheating (won’t give details to respect privacy). I played against him online and got crushed, then looked at his game record. There was no need to run AI analysis, it was pretty clear his level was very inconsistent. Of course it doesn’t prove cheating, my suspicion was also based on other evidence.
2 Likes
You’d think that, but actually cheating and not doing enough to detect and counter it can put a lot of people off of playing online, or on the server in question.
I’ve heard a number of times, players lose interest in playing online leagues and tournaments precisely for that reason.
So if it becomes common knowledge that there’s a lot of cheaters on server X, people will just stop playing there.
If we weren’t consistent enough within some tolerance, then rating systems would have no purpose at all. However they are quite accurate at predicting who will win an even game. Of course there will be outliers, as there will in anything.
3 Likes
Uh doesnt OGS already do this (not as a cheating measure)
1 Like
I had 7 players who were suspected of cheating against me. One of them was using a robot from OGS. I beat him. He was in a tournament there and he was inviting.
It seems rather sad that with the advance of computer power, we end up discussing whether or not to use it to find out if somebody is being a tube. I think there are a better uses of AI to improve the customer experience here.
I agree with this one. What good does it do to punish cheaters as harshly and unforgiving as possible? For a moment, I will entertain the assumption that you know for sure that your opponent cheated and that a false positive is impossible. For that I ask you to exclude the occasional troll from your brain for a moment.
Then what? You may feel good for the moment that your opponent is banned and stigmatized by their games flagged as annulled because of cheating and play the next game. And the other human? Don’t they deserve some grace as well? When people cheat they often do it because they pressure themself into thinking they need to do this in order to keep their self-picture in place. The universe owes them a win, why then not use some assistance? Sure they will catch up in no time etc etc. And it is only downward from there.
Silly, even childish thoughts? Maybe. But they are present. And in this case no external punishment is enough, because it is an internal struggle and one part already hates oneself for cheating with a burning flame even if the other part is stronger for the moment.
I would feel very uneasy with making people so uneasy over Go. Therefore, I really like the compromise currently implemented. Give a harsh and strong warning if you suspect someone strongly and ask them to resign their cheated games. But if they follow through with it, let it be and accept them back into the community. Do not impose social stigma on someone who can change. Isn’t Go supposed to be a friendly game?
2 Likes