Carlsen resigns on move 2

[JustinHorton]

For me it's maybe less about dishonesty and more about recklessness - people not knowing what a given tool is or what its limitations are, and yet piling in with premature conclusions anyway. Recklessness about their own degree of knowledge and for that matter about the reputation of other people.

[Geoff Chandler]

Hi All,

I too recall the Anand statement but searching for it could be a long job.
Vishy was playing in a charity event where his opponent admitted cheating
and this incident takes over any search regarding Anand and cheating.
Also it could be in print in a magazine, but it’s definitely out there.

I found a Sam Copeland reference to it (alas without a prime source) in an article regarding cheating in chess.

“Former World Champion Anand has said that simply being cued thrice in a game that a strong possibility exists would be enough to improve one's play by ~100 ELO points. “

https://www.chess.com/blog/SamCopeland/ ... t-at-chess

I’ve always taken this to mean if you are told (somehow) during a game to STOP! Something on! Then you dig in to find it.
When we solve puzzles in diagrams from actual games we know there is something happening
so we look and, hopefully, eventually solve it.
Sometimes, in the solution, we are told the player missed the combination which is very easily done if you
do not recognise a ‘hot’ position.

Being told there is something on means all you need do now(!) is find the combination you know
is there and dig it out as you would when solving a puzzle in a diagram.

[Paul Cooksey]

I decided to continue to posting. Justin is an Achilles heel for me. Because I dislike him so much I respond in ways I would not to other people, which I then regret. But since he has put me on his foes list, that might in itself solve my problem.

I saw some discussion of the concepts of evidence and proof and how to translate them from French on Twitter. Personally I think Lets Check is probably quite important in cheat detection, because it provides a different set of data to centipawn loss. I'd be surprised if not already in some of the models. But jumping to a conclusion from a small data set which I did not understand, would make me an idiot, whether it is lets check, centipawn loss or my own gut feeling.

I found googling "what percentage is beyond reasonable doubt" interesting. The standard of proof juries require if they don't like the look of you seems to be worryingly low.

I'm not sure what percentage sure I would want to statisticians to be before I accepted guilt in chess cheating. I am happy with less than 100%, even in the absence of other evidence. But probably do want significantly above 99%.

[Ian Thompson]

I'm not sure what percentage sure I would want to statisticians to be before I accepted guilt in chess cheating. I am happy with less than 100%, even in the absence of other evidence. But probably do want significantly above 99%.

In excess of 99.9%, I would hope. Here's a table of probabilities. I think Ken Regan has mentioned 5σ, which is 99.9999%.

[Roger de Coverly]

I. Personally I think Lets Check is probably quite important in cheat detection, because it provides a different set of data to centipawn loss.

There is a secret though. What Lets Check is measuring is whether there's a match to an engine, that's any engine. By general reasoning, the more engine analysis is loaded, the more likely it is that a human's move will match one of the many engine suggestions. Also if you use an old (not so good) engine, you may get a match that isn't there when a stronger engine is chosen. It was also discovered that one of the cheat hunters had uploaded analysis himself from an old engine in the attempt to discredit you know who.

The link is in one of today's posts.

[Paul Cooksey]

I'm not sure what percentage sure I would want to statisticians to be before I accepted guilt in chess cheating. I am happy with less than 100%, even in the absence of other evidence. But probably do want significantly above 99%.

In excess of 99.9%, I would hope. Here's a table of probabilities. I think Ken Regan has mentioned 5σ, which is 99.9999%.

I said percentage rather than sigma because I was trying to avoid jargon. But does make sense to consider this way.

Ken talked about sigma being 200 Elo points. But I wasn't sure if I could claim a linear, "performing 1000 Elo above expectation is 99.9999% proof of cheating". It seemed unlikely, but I haven't done the necessary reading yet.

Do you understand this Ian?

[Roger de Coverly]

But I wasn't sure if I could claim a linear, "performing 1000 Elo above expectation is 99.9999% proof of cheating".

Performing 1000 Elo above expectation also demonstrates a reasonably high probability that the expectation is wrong. Perhaps not 1000 points, but players in Congresses over the past year have regularly suffered against juniors rated 500 points lower.

It remains my view that cheat detection by "You are not strong enough to make those moves without assistance" is flawed logic if there is no obvious method of consultation being employed. If the supposed method of cheating involves an accomplice indicating important positions or suggesting moves, then find the accomplice!

[Paul Cooksey]

I understand the argument. But I was trying to make sure I was understanding the stats correctly, if only because the strongest engine is 3500, which makes a 5σ rule impossible for elite events not withstanding other objections.

[JustinHorton]

I found a Sam Copeland reference to it (alas without a prime source)

Alas indeed

“Former World Champion Anand has said that simply being cued thrice in a game that a strong possibility exists would be enough to improve one's play by ~100 ELO points.“

You can see the problem here, can't you? There's nothing to agree or disagree with here. You've just got a number with no explanation of how it's arrived at.

In some ways that's not my major gripe though: it's the idea that you could somehow be "cued thrice" unproblematically, like it was as easy as that. How's it going to be done? The suggestion in the blog piece itself is laughable.

[Matthew Turner]

Paul,
I would suspect that you have a very good understanding here (not the first time I have had this feeling ). I think it is important to recognize that when we are talking about 1 sigma relating to 200 Elo points in Regan's analysis, this is over a a particular number of moves, normally about 6 games.

[MartinCarpenter]

I understand the argument. But I was trying to make sure I was understanding the stats correctly, if only because the strongest engine is 3500, which makes a 5σ rule impossible for elite events not withstanding other objections.

You just need to stack up more games, maybe many more. Its a very simple methodology really.

There's a massive database of pre computer chess games/events which gives data regarding how often/to what extent people over perform in the short term absent any possible computer help.

He's then looking for anything that's a long way beyond anything seen in that database. At that point, given the obvious alternative option these days, you have a definite right to be very suspicious.

As various people way up thread noted/iirc quoted Carrunua - it sadly needs a lot of data, so will miss more subtle things.

[MartinCarpenter]

In some ways that's not my major gripe though: it's the idea that you could somehow be "cued thrice" unproblematically, like it was as easy as that. How's it going to be done? The suggestion in the blog piece itself is laughable.

It would be, in principled terms, utterly 'trivial' to do for online play. Your own room, your own computer etc etc. The safety measures in place are not - could not be - a particular challenge.

The more useful safety margin there is people not wanting to risk their hard earned FtF reputations/needing to back up their 'improved' play over the board. Anyone consistently playing a chunk stronger online would, I imagine, be looked at with great suspicion.
(Fairly or otherwise!).

It gets a lot more harder/strained if you want to do it in real life. Pretty trivial with an accomplice though.

[JustinHorton]

Pretty trivial with an accomplice though.

Why would you say so? The suggestion in the piece is a nonsense, so I'm sure it must be possible to do better, but I'm yet to come across a method that seems likely to succeed at high-level events for a good period of time.

[Roger de Coverly]

There's a massive database of pre computer chess games/events which gives data regarding how often/to what extent people over perform in the short term absent any possible computer help.

Comparisons to the pre-computer era should perhaps be treated cautiously. All players, or at least those who check books and engines before and after games have computer help these days. Some if it may rub off into move choices during games.

[Paul Cooksey]

There's a massive database of pre computer chess games/events which gives data regarding how often/to what extent people over perform in the short term absent any possible computer help.

Comparisons to the pre-computer era should perhaps be treated cautiously. All players, or at least those who check books and engines before and after games have computer help these days. Some if it may rub off into move choices during games.

I'm not sure I agree. Rowson eloquently explained why games by the computer generation might fail the eye test. But statistically, they are just data aren't they?

You do end up having to answer questions like was a 2700 Karpov in the 90s playing as well as a 2700 today, and probably giving answers that someone like me who is a Karpov fan dislikes. BUt seems like valid data to me still.