Moderation at OGS

As with any large open community, from time to time there will be users who's behaviour doesn't meet our vision for a friendly and respectful place to play Go.

"Moderation" is used to prevent this sort of thing from becoming unpleasant, and fix any issues that arise.

We have a range of moderation mechanisms in place:

• A small panel of "full" moderators.

"Full" means that these moderators have the "power" and authority to take most corrective actions available on OGS. They can issue warnings and suspensions, they can resolve appeals and they can annul (and un-annul) the result of games. They are responsible for handling the more "difficult" types of "Reports".

The main thing they can't do is change the result of a game (for example, no-one can change "who won").

The members of this panel are selected by the panel from time to time as needed - it's an "invitation only" panel. Addition of new "full" moderators is not done frequently.

• A larger panel of "Community Moderators" (CMs).

Community Moderators handle the more straightforward types of reports - reports that are about game-play, rather than the more "nasty" behaviour we sometimes regrettably see.

Community Moderators have the power to act on the reports that they handle. They can warn or suspend based on these reports. All Community Moderator action is by concensus - 3 CMs have to agree for the action to take place.

• A small "AI Detection" team.

The members of this team focus on the problem of "minimizing AI cheating" at OGS. Along with full moderators, they handle AI Use reports.

Reports and The Report System

"Reports" are the main way that you can alert us that moderation is needed.

You "file a report" by clicking on a player's name, then pressing the "Report" button.

It's important to tell us as much as you can about what you are reporting, and why you think it needs attention.

It is probably obvious to you, but the person handling the report sees hundreds of these - it may not be obvious to them and its time consuming to try to work it out!

Minimizing AI Cheating

OGS is committed to minimizing the occurrence of cheating by using AI assistance.

We have a dedicated team of people working on this challenge.

How do you detect AI cheating?

Regrettably, there are always people who can't resist the temptation of winning by using AI assistance. Preventing this can be likened to a "war": as detection gets better, cheaters work out how it is done and evade the new technique.

For this reason it should be obvious: we don't share how AI detection is done at OGS.

We can share that we have a range of techniques that we are continually improving. These involve tools that look at all the data we have available on this topic - all the data about suspected players' play, and all the data about what normal and cheated play looks like from our huge database of games, and elsewhere.

Every game played at OGS has at least a cursory pass of tool inspection for likelihood of cheating.

Human AI detectors proactively review this flow of information, and also automated tools look at it.

"You can't prove it was AI cheating"

It should also be obvious that we can never be 100% sure that AI assistance was used. Maybe "that person" played exactly the same as KataGo for the whole game just by luck, who can tell for sure :)

Because this is clearly true, and yet we have to decide ... we do decide. We have levels of indications about what AI looks like where we simply have to trust that this is "definitely AI". "Definitely" means "likely enough that we must act".

What do you do about AI cheating when you find it?

Warnings and Suspensions

First, we may issue a warning to the cheater.

A person who has played at OGS for some time without cheating will get a warning.

A newcomer who cheats straight away will probably not.

After a warning, there is suspension of the account.

Fixing the damage to ranks

When we find games that are "definitely" AI assisted, we take steps to remedy the damage done by this cheating.

• We annul games that are "definitely" AI assisted.
• We also annul enough games to correct any rating damage done by the cheater with their cheated rank.

It should be obvious that we will not answer the question "Why was this game chosen to be annuled?", other than "it fell in one of those categories".

If we were more specific, this would tell the cheaters how we are detecting, and help them to evade detection.

AI Cheat Detection FAQ

• Why was this game annulled?

"This game was annulled as part of AI cheating remediation"

That message is as much as we will ever say. You can read above the likely reasons why any game was chosen for annulment during AI cheating remediation.

We won't say more because that just helps the cheaters

• Which game of mine did you detect cheating in?

See above. For obvious reasons, we do not answer this question: answering that helps cheaters.

• Why was my AI Cheating report rejected, it's obviously AI cheating!

The most common answer to this question is "because you didn't explain why it is obvious to you".

We can't "convict" someone of AI cheating just because they beat you. The evidence of actual AI assistance needs to be clear.

We get hundreds of these reports, so we rely on the tools to tell us initially if they agree with you.

If you have spotted some AI-like moves, maybe the tools didn't spot it.

So tell us. If you describe in detail what the evidence is, then the AI detector who is processing the report will look at that.