I was happy to see that my earlier post on online cheating sparked a lively debate on L19. So far, several different measures for combatting online cheating have been proposed; personally I took a liking to the suggestion of proxy tournaments, described here in Marcel Grünauer’s post.
In unproctored tournament settings, it is always possible for a player to find a way to cheat:
When combatting cheating, it seems to me that the most important question is if the prevention mechanism is appropriate to the scenario at hand.
Besides the chosen cheating prevention mechanism, there should of course be a team of experienced referees who judge and analyse alleged cheating cases.
To my surprise, the general opinion on L19 seemed to be that it is not possible to catch cheaters from a small dataset, such as just one or two games. While smart cheaters are indeed hard to catch, in my experience a majority of cheating cases are fairly obvious. Let me introduce you to three anonymous examples.
In the first case, I was approached on ogs by a player with a Japanese username that translated to ‘god of go’. The player put on a surprisingly humble front, asking for a teaching game with a professional player. When I checked their profile, I noticed that their non-verbose username was actually one famously used by a Japanese top player on Asian servers.
I was getting suspicious, so I checked the player’s game list and noticed they are playing an opponent whom I knew to be a strong Japanese professional. I then checked that game and saw that the professional was losing badly; and the ‘god of go’ had even gone so far as to say ‘your play is bad, there is no way you can win’. I mentally gave them a 99.99% prior probability of cheating, on grounds of finding it unconceivable that a top-level human player would behave so badly.
Next I fired up my KataGo and checked the content of the game. On Lizzie, my current gui of choice, the ai’s #1 move candidate shows up on the board as a blue circle. Of the 50 moves our humble cheater played, 46 moves hit the blue circle, and two of the moves that didn’t were White’s first and second move.
Personally I think it would be more likely for me to win the jackpot in the Finnish national lottery than hit 46 out of 50 blues in a game; but I cannot deny the ‘possibility’ that there could be a stronger human than me who is more familiar with the ai than me and also has spare time to play around on ogs. Even then, my most generous estimate for the odds of this came to around 0.01% – and that’s before taking the prior 0.01% into account. Without thinking too much about it, I reported this ‘god of go’ to ogs admins.
The second case was even clearer, if possible. This time I was not approached by the god of go, but by Sai himself from Hikaru no Go (the exact username was not ‘Sai’, but directly referred to him). Sai, too, wanted a teaching game with me.
Sensing a pattern, I checked Sai’s list of games and saw they had a correspondence game going on with ‘god of go’. Analysing that game, I saw that Sai was actually winning even though the opponent was again obviously cheating – most of both players’ moves were in the blue. Let’s get this straight: here was a player who was beating an opponent probably three stones stronger than me, asking me for a teaching game.
This time, as an experiment, I first tried to get Sai to admit their cheating, but only got the all-too-common excuse along the lines of ‘I just studied with the ai really hard these past six months’. Finally, I just pressed the report button.
The first two cases were easy to figure out because of their context; but what about if we only have a game record or few with no context? In the third case, I’m looking at the first game record of the four uploaded in this post on L19 in a related thread.
All times are in Helsinki time (eet with summer time).
New pairing for the online league made every 1st and 3rd Saturday of the month.
Public lectures on Twitch every 2nd and 4th Saturday of the month at 1 pm.
Jeff and Mikko stream on Twitch on Fridays at 6 pm.