Analysis of Lichess' cheating detection with Machine Learning (ML) - a mis-use of ML & doesn't work

@odoaker2015 said in #69:
> An outsider does not possess and/or has access to any data that can establish who is or is not a false positive. That's one of the criticisms.

Your argument appears to be that there is no point to ever attempt to scientifically examine lichess’s mechanisms because not all data is public. I think thats a bad view, we should always try to better understand the efficiency of mechanisms, especially ones that are so consequential.

@dreadpresence said in #71:
> Your argument appears to be that there is no point to ever attempt to scientifically examine lichess’s mechanisms because not all data is public. I think thats a bad view, we should always try to better understand the efficiency of mechanisms, especially ones that are so consequential.

No, I didn't say that there is no point to ever attempt to scientifically examine lichess’s mechanisms. I am saying that no outsider can examine this because they lack the necessary data to do so. Lichess' actions affect other people who may have been wrongly flagged as cheaters, and whose appeal has no effect. Such a possible case is described in the video that I linked in post #33. I don't think that's how it should work.
The author of this thread has pointed this out several times.

@odoaker2015 said in #64:
> @IrwinCaladinResearch
> "A mechanism that accuses people of cheating cannot have 1-2% error rates."
>
> What error rate do you think it should be?

One thing I would like to know is whether IrwinCaladinResearch included on-line correspondence games.

The reason I ask is that I don't believe statistical based ML methods are capable of any degree of refinement with correspondence games because there is no way to factor in the time spent on determining moves. For example, a GM might spend only 5 seconds on their moves whereas a lowbie might spend 2 hours - and this is aside from the equalising nature equal access to non-engine resources. The other problem here is that it would be permittable for a player to make use of their own previously engine analysed games as well as Master games. If similar methods of cheat detection were used for on-line correspondence, they I believe you would get a heck of a lot of false-positives compared to say Blitz games.

The only time I have ever noticed players being banned was during a chessdotcom tournament of over 100+ players when I noticed that over 10% of the players eventually were banned. This compared to an extremely low % - something like 0.1% to 0.01% - for low level Blitz games. It was actually this observation that got me interested in how anti-cheat systems work and consequently why I stopped playing correspondence games.

Also note on chessdotcom over 25% of players play correspondence, not idea regards LiChess.

@AlexiHarvey

Cheating in correspondence chess games will not lead to being banned. So games like that don't matter.

@odoaker2015 said in #74:
> @AlexiHarvey
>
> Cheating in correspondence chess games will not lead to being banned. So games like that don't matter.

Uh yes it will?

I think so, but I don't know for sure. @Deadban

@odoaker2015 said in #76:
> I think so, but I don't know for sure. @Deadban

See lichess.org/page/fair-play.

@Deadban

So, if you cheat in a correspondence chess game it will lead to a permanent ban? And to a red marking?
Can you show one example of this?

@odoaker2015 said in #78:
> @Deadban
>
> So, if you cheat in a correspondence chess game it will lead to a permanent ban? And to a red marking?

Yes.

> Can you show one example of this?

No.

Let's ask a Lichess moderator! Does cheating in a correspondence chess game on Lichess result in the person being permanently banned and given a red mark? Will his opportunities to play be restricted? For example, that he is no longer allowed to play in rated tournaments? Or that he can only play against other cheats? It would be nice if a moderator could answer briefly. For clarification!