“Decision: case of using computer assistance in League A”

[RobertJasiek]

Good enough theory?

Perfectly not because every cheater is different.

[Bojanic]

Good enough theory?

Perfectly not because every cheater is different.

Yes of course, and I also think that there is more cheaters in PGETC and most of them will go unsanctioned.
But in case of those two games, it is good enough.

[Bill Spight]

As for the 98% matching evidence, you must understand that matching one of a bot's top three choices was chosen in order to generate impressive matching numbers, not through any theory of how a player might have cheated. (This motive may have been unconscious.) Also, restricting the possible matches to the fifty moves between moves 51 - 100 is also suspicious. In addition, it is confirmatory evidence instead of disconfirmatory evidence. IOW, it is not just unsound, it is crap.

So you think we should also include in analysis opening phase, which could easily show 100% similarities to Leela in many games?

I think that the analysis of game records should be done in terms of go, not in terms of matching any bot's choices. Why omit any play?

Consider Regan's approach. Chess engines are good enough that he is able to rate the difficulty of plays, giving them ELO ratings. That way he can eliminate plays that are too easy from consideration, not just consider a certain range of plays. Also, he can give an ELO rating for a player's play in a single game. We are far from being able to do that in go.

Now, playing well at chess will produce a lot of matches to the plays of top engines. But the point is not matching any engine's play, it is playing too well. Play a long enough game with no blunders or mistakes, and the odds are good that you are cheating. No human plays that well. But Regan requires extremely good odds for statistical evidence alone.

BTW, opening usually lasts up to move 30, not 50 as you claim.

You are confusing me with the people who came up with the method used.

And I am curious, what method would not show that Metta's two online games are very different to his two live games?

I await the evidence, in terms of go, not statistics. An analysis in terms of go might produce behavioral evidence, which is necessary, given the weakness of the statistical evidence.

[John Fairbairn]

tapir wrote:

I would much prefer unsound evidence (not too unsound, but the 98% in the case), occasional false positives and lenient punishment, that keeps the control systems simple and overall trust (we catch most cheaters + few innocents) higher.

maf wrote:

I understand your sentiment, but this will never be accepted. It is literally against a core value of Western society.

I think this method is in use in many places, even if a little disguised perhaps.

For example, a policeman told me that his almost 100% honest force had a brute in their ranks whom they found very useful. If they encountered a particularly obnoxious individual they couldn't handle by ordinary means, they would consign him to the hulk, who would beat the detritus out of him. This ensured few people then took the Michael out of the good police while limiting the bad stuff to a tiny handful of people whose only actual crime may, admittedly, have been that they were obnoxious.

This strategy is used even in trivial ways such as a baseball pitcher tossing at the head of a batter just to keep him "honest". Even parents sometimes use it with their kids: trampling over their human rights, terrifying them and scarring them for life by taking their smart phones away for a day just to make sure that the next time they need to ask for homework to be done, they just have mention the p word.

It sounds awful on paper, but in my experience it seems to work quite well. Maybe because, although it's open to abuse, it's a strategy used by basically good people who will draw the line at excess. So it might just work in go.

[theoldway]

And I am curious, what method would not show that Metta's two online games are very different to his two live games?

In addition to your question, I'm more interested in your conclusion. Are you suggesting that every player plays exactly the same way in every game?

This is far from reality. Every player can play in many ways and strength. We are not professional players, our game can vary depending on our opponent's strength, our opponent's mistake, our opponent's style, our mood, whether we slept well or not that day, basic time, byo yomi time, and so on.

I think a 4 dan can easily vary from 2 dan to 6 dan depending on hundreds of factors. 4 dan is simply his average strenght.

Not to mention the fact that online games are played around 1 every month, live games are usually played 2 or 3 every day. This aspect favors heavily strong players in live tournament and vanishes in online tournament.

[Bojanic]

I think a 4 dan can easily vary from 2 dan to 6 dan depending on hundreds of factors. 4 dan is simply his average strenght.

Actually I completely agree on you on this matter.
And we can see it Metta's WAGC games.

But that is why it is actually completely impossible to explain how could Metta played two entire games completely on level of Leela program. Only mistake - Leela's L&D mistake.
European pros in same event played several mistakes per game.

[tapir]

As for the 98% matching evidence, you must understand that matching one of a bot's top three choices was chosen in order to generate impressive matching numbers, not through any theory of how a player might have cheated. (This motive may have been unconscious.) And restricting the possible matches to the fifty moves between moves 51 - 100 is also suspicious. In addition, it is confirmatory evidence instead of disconfirmatory evidence. IOW, it is not just unsound, it is crap.

I fully understand how the 98% came about. The choice implies a theory of how the player cheated. (I.e. sometimes choose 2nd or 3rd move to make it not too obvious.) Yes, it is limited. Yes, it may be wrong. Still, afair no other game looked into in this thread came even remotely close in similarity.

What people don't seem to appreciate at all is that this isn't a scholarly discussion about the quality of the evidence, but a decision about how Go will be in the future. Making it almost impossible to catch a cheat (all the doubts piled up in this thread do exactly that) will only lead to more cheating and all pervading hypocrisy. I would like to see lenient (for the sake of the accused innocents and for human betterment), but swift sanctions.

I am also fairly sure that I won't ever be wrongly accused of cheating that way, this particular excuse seems something reserved for 4 dan upwards. I doubt anyone would give a 5 kyu the benefit of a doubt, even if he too would place 100% of his moves on the intersections of the board. Now please ponder what conclusions beginners will draw from all this.

[Tryss]

I am also fairly sure that I won't ever be wrongly accused of cheating that way, this particular excuse seems something reserved for 4 dan upwards. I doubt anyone would give a 5 kyu the benefit of a doubt, even if he too would place 100% of his moves on the intersections of the board. Now please ponder what conclusions beginners will draw from all this.

I'm currently 3k on KGS

I recently played a game where, according to LZ, I only made 3 mistakes, And only one after move 30 (and it was only a -1.5% winrate loss). Did I cheat?

[Bojanic]

Tryss,
If that is your level of game, why are you wasting your time on KGS?
You should play in some strong league, Honinbo or Judan would be OK.

[Javaness2]

Show us the game and your analysis, it can be an interesting example to study. What else is the sgf tag for here?

[theoldway]

But that is why it is actually completely impossible to explain how could Metta played two entire games completely on level of Leela program. Only mistake - Leela's L&D mistake.
European pros in same event played several mistakes per game.

Actually there are other PGETC players with several games almost completely Leela-like (even some famous and distinguished player). They are all cheaters? Or maybe in hundreds of PGETC games it is possible to observe these coincidences from time to time?

This is the main question we need to answer in the future.

[Tryss]

Tryss,
If that is your level of game, why are you wasting your time on KGS?
You should play in some strong league, Honinbo or Judan would be OK.

Where did I say I played like a pro player? I played like a 3k, my opponent just played badly in the oppening and blundered, and he never catch up with me, so from most of the match, my winrate was over 95% and didn't drop : I didn't make any mistake for LZ.

Now, what if he accused me of cheating?

[Gobang]

When I read this thread I see a lot of hand wringing, clutching at straws and missed points. People trying to glue together a toy that is broken.

As far as I am concerned this issue is not about a few games and one player who has been unfortunate enough to become the topic of this discussion. There are larger things at stake here, such as the future of online Go tournaments. We do not appear to be anywhere near a system that can reliably detect cheating in online games. While such a state of affairs exists, is it reasonable to have cash prizes and determine ranking based on online performance? Of course not.

[Javaness2]

Actually there are other PGETC players with several games almost completely Leela-like (even some famous and distinguished player). They are all cheaters? Or maybe in hundreds of PGETC games it is possible to observe these coincidences from time to time?

This is the main question we need to answer in the future.

Who are they, and which are the games you mention?

[theoldway]

Who are they, and which are the games you mention?

Are you seriously asking to expose more players, maybe false positive, to this crusade?

I envy some people certainty about cheaters and cheating, I'm still full of doubts.