Cheating in chess

[JustinHorton]

I mean it's not even the report or anything, it's a letter, which contains a quote from the report, and that's it. It's not just that you don't know what they learned, you don't even know what they asked. You don't know whether they just accepted what they were told because it looked impressive, or whether they really interrogated the system and its operators. You don't know if they asked "if it's so infallible, how do you explain the cases of X, Y and Z?", let alone what answers they might have received. You don't know anything, except that some people said it's all hunky dory on a basis about which you know nothing.

Genuinely insulting trash.

[JustinHorton]

I also liked

This endorsement is not tied to the recently announced US Chess-Chess.com partnership

[Kevin Thurlow]

"This endorsement is not tied to the recently announced US Chess-Chess.com partnership"

Do they seriously expect anyone to believe that? It's not unknown for chess federations to favour individuals or groups, or even take large bribes. I note too, that any USCF member can appeal against a ban, so if you're not a USCF member, you are obviously guilty.

I'm not convinced that 1 in 1000 false positives is acceptable, even assuming that their claim is true.

[MartinCarpenter]

"This endorsement is not tied to the recently announced US Chess-Chess.com partnership"

Do they seriously expect anyone to believe that? It's not unknown for chess federations to favour individuals or groups, or even take large bribes. I note too, that any USCF member can appeal against a ban, so if you're not a USCF member, you are obviously guilty.

I'm not convinced that 1 in 1000 false positives is acceptable, even assuming that their claim is true.

That letter doesn't mention the false positive rate and its far from acceptable.

"The threshold ... to flag a player requires a level of certainty such that the chance of falsely accusing a player is less than one in one thousand."

That's the accuracy of the test. To get the false positive rate you need the actual rate of cheating in the overall population.

Assume 1% of people are cheating - then 1/10 accusations are going to be false. Doing that to people with real world consequences?!

I'd want ~1/1,000,000 accuracy off the statistics, especially given the ways they can be unreliable.

[John Hodgson]

One method of cheating that I have not seen discussed - though with the volume of posts I may have missed it - is cheating by moving the pieces on another board (at a longer time limit, of course).

This would be difficult to detect because the moves would still be human rather than computer-generated.

[Matthew Turner]

I think the letter is saying that out out of all the positive cases, in less than one case in a thousand is the player innocent. What does that mean? I tried running some numbers

Let’s say 20% of players cheat

So we have 8000 clean players and 2000 cheats

There is a very good detection system which picks up all of the cheats. Then to get to our 1 in a thousand false positive it would have to pick up 2 clean players. So that would mean that if you are a clean player there is a one in 4 thousand probability of a false accusation been made against you.
Given the size of these website that would be an awful lot of people, so I suspect they would be aiming for better than that and I note the letter says less than one in a thousand.

[Tim Spanton]

"This endorsement is not tied to the recently announced US Chess-Chess.com partnership"

Do they seriously expect anyone to believe that? It's not unknown for chess federations to favour individuals or groups, or even take large bribes. I note too, that any USCF member can appeal against a ban, so if you're not a USCF member, you are obviously guilty.

I'm not convinced that 1 in 1000 false positives is acceptable, even assuming that their claim is true.

I doubt if many court systems could run if 1 in 1000 false convictions meant they had to shut down

[Roger de Coverly]

I doubt if many court systems could run if 1 in 1000 false convictions meant they had to shut down

They would be shut down (hopefully) if they convicted 1 in 1000 without presenting any evidence.

The ECF grading list contains, or used to contain 15000 people. Assuming they all played clean, 1 in 1000 is 15 people wrongly accused.

[Tim Spanton]

I doubt if many court systems could run if 1 in 1000 false convictions meant they had to shut down

They would if they convicted 1 in 1000 without presenting any evidence.

But not all evidence is made public, eg psychiatric reports, social reports, cases of national security - just the main body of evidence. Similarly in a chess game the main body of evidence is the moves themselves.

[JustinHorton]

I doubt if many court systems could run if 1 in 1000 false convictions meant they had to shut down

On the other hand when you have appeals they're not usually heard by the same people who brought the case who in turn are not usually the same people who judged the case.

But not all evidence is made public, eg psychiatric reports, social reports, cases of national security

On the other hand in nearly all cases that evidence is in fact made available to the subject of those reports and the exceptions are both extremely exceptional and extremely controversial.

[Roger de Coverly]

Similarly in a chess game the main body of evidence is the moves themselves.

What do the moves show? The method used is the assertion that "a player of your rating isn't good enough to find those moves, therefore you must have had external assistance". That can even be true but also legitimate in the case of extended book knowledge. Note the dependence on rating as an input. The Elo method and similar alternatives uses the game result as its parameter and attempts to predict them. It's leap of faith to presume that rating directly correlates to move selection.

[Kevin Thurlow]

"What do the moves show? The method used is the assertion that "a player of your rating isn't good enough to find those moves, therefore you must have had external assistance". "

Yes, one of my old club colleagues, ended up with a grade sub 100 I think, due to ill-health and age, having peaked at 180 ish, but capable of playing much better on his day. Even towards the end of his life, there were flashes of brilliance...

[MartinCarpenter]

I think the letter is saying that out out of all the positive cases, in less than one case in a thousand is the player innocent.

It isn't absolutely clear, but I do think you have to stretch the language a bit to read it that way.

As I said too many posts back, something like a 1/1000 chance of flagging actually makes sense for the screening the blitz games on chess.com. Plenty of cheats and no 'real' consequences of falsely accusing anyone.

Even if they have done population estimate work, I wouldn't trust the 1/1000 figure at all. There'll surely be much more cheating in the population playing online blitz than playing serious long play games. With a consequently much higher false positive rate in the serious games.

There is a genuine debate to be had about what an acceptable rate of false positives actually is. I still can't see any real reason not to wait until the statistics are truly overwhelming.

[JustinHorton]

Besides, the claim is meaningless unless the means of arriving at it can be examined, this is just elementary.

[Stewart Reuben]

Justin Horton >That letter is pure garbage.<

Elegant put. But, were you referring to Alex McFarlane's letter.;The note from the USCF; or the letter from Roger Lancaster?

Is a puzzlement.