I mean… technically. But we’re blurring the lines a bit 
5 Likes
I have looked at more AI reports than anyone else here: I can guarantee you that (or at least, in recent times).
And I’ve run appeals on those through Dan moderators.
They tell me “although the tools think this might be AI, humans do that actually” and “Yeah, this is almost certainly AI, humans don’t do that”.
This is in the context of other tools we have for detecting, which non-Dan mod like me can use: it’s the resort that we have to fall back to when the tools are not clear.
It works surprisingly well, still. I know this because of the kinds of admissions and appeals that we get after a decision.
If we couldn’t do this, we’d have to throw up our hands and say “we can’t differentiate AI use, so everyone have at it, we don’t care”.
We are not at that point yet, but we almost are.
This new development - making human-like AI cheating easier - may be the straw that breaks that camel’s back.
Who knows, maybe things will be better after that, when we stop caring about whether we’re defeating another person or not, and just worry about whether our play is good 
1 Like
Do you have a example of an AI move that a pro-level human wouldn’t make?
1 Like
That’s a question for Dan moderators like @mark5000 and @xhu98 who make these calls.
However, I think it’s not a good question to ask, and one they might (should?) be reluctant to answer, because any answer about “how we detect botting” can be used to hide botting.
(Although based on this discussion, maybe we should open the kimono, and forget about trying 
)
1 Like
We aren’t comparing AI with pros, we’re comparing AI with moves kyus or low dan amatuers makes.
2 Likes
I think it is never the question of “an AI move”, but of several characteristics of AI moves. Players who make the kind of moves that Ke Jie makes several times per game, so to say.
It’s like how I cannot recognise a text generated by GPT by looking at any of the individual words, but reading the whole paragraph makes it kind of obvious after you’ve read enough of those.
6 Likes
I don’t think this new feature requires changing OGS policy to investigate reports on botters who fail the botting Turing test.
If nobody notices that someone is botting, then I suppose they won’t get reported and they pass the botting Turing test.
And if the botter passes the botting Turing test and remains unsuspected of botting, but fails to do it consistently enough to mimic a human of the level that matches their OGS rating, I suppose they may still get reported for sandbagging and treated as such.
If the botter passes the botting turing test and also does it consistently with the same level of human-like settings, then perhaps it’s not worth it to spend much effort to detect them. Yes they technically violate OGS ToS, but other than that they do little harm. They won’t disrupt the rating system and nobody feels cheated.
6 Likes
Agreed - OGS policy has no reason to change and botting should still be prohibited.
We just need to accept that not all botting can be detected, which is nothing new, but also is not necessarily so worrying for casual games on a Go server, as the difficulty to identify a bot seems inversely proportional to the harm it causes for the opponent. A perfectly human bot that just plays slightly better would be impossible to investigate, but would also never be reported in the first place as the opponent would not notice anything different.
This is of course different in a competitive tournament setting, but nothing new here. You already couldn’t guarantee a no cheating environment for online tournaments, except through pretty invasive setups (cameras all around).
4 Likes
At least according to Lee Sedol back then 
(Discussing Game 1 of the jubango with Gu Li, but after it had concluded anyway)
I have a feeling the way pro players tend to discuss how AI changed things, some of that category of moves might probably be much more “normal looking”. That is moves an AI recommends that humans wouldn’t play probably out of a kind of stigma or learned judgement (the kick is bad, the early invasion is bad etc).
The Tesuji style moves people play if it works, the amazing moves people play if they find.
1 Like
I think the “Overton Window” for which moves are conceivable by humans has shifted significantly during the past 8 years.
Especially in the early game, many (most?) human players have incorporated moves in their normal play which would have been considered AI moves 8 years ago.
So nowadays AI-like moves during the opening shouldn’t raise much suspicion. I think only in the middle game you may still be able to point out inhuman moves (taking into account the level of the player), although even then I think you would need to spot multiple of such moves to build a case for botting based only on game content.
3 Likes
Still, I remember a lecture from InSeong a couple of years ago where he pointed out the potential effectiveness of exactly such 2nd line invasion moves. I admit that I had forgotten that lecture until now, but it’s conceivable that a studious 3k student of his incorporated such moves in their arsenal after watching that lecture.
2 Likes
Maybe I’m taking this out of context, but how is the cross on a dead stone a cheating tool? Are we talking about the scoring phase or maybe the scoring estimate tool? Sorry if this is a stupid question.
Correct, he’s referring to a different thing
1 Like
I used the form “look at the massive thread about xxxx … and here we are talking about yyyy” as a means of comparing the potential thread-size of (and implied complaining in) xxxx and yyyy, where xxxx is a trivial thing compared to yyyy.
2 Likes