The independent home for discussions on the English Chess scene.
https://ecforum.org.uk/
I had no idea you were interested in important mattersGeoff Chandler wrote: ↑Sun Apr 18, 2021 11:18 amI'm actually doing my best, in my own sweet way, to say there are more important
matters
How can they possibly know (or even estimate) the rate of false positives?John Hodgson wrote: ↑Thu Apr 15, 2021 9:58 pmAccording to chess.com, "the rate of false positives detected by our algorithm is intentional."
Does anyone know what this rate is, and what we think this rate should be?
The fundamental difference though is that in order to assess the false positive rate of covid lateral flow testing you have options, such as testing with PCR and comparing the results.Roger de Coverly wrote: ↑Sun Apr 18, 2021 2:06 pmIt's a piece about covid testing, but the application of Bayes's Theorem seems also relevant to the question of how many false positives are being detected by the cheat finder methods.
As a summary, the smaller the proportion of people using engine assistance/ having covid, the greater the relative proportion of false accusations/ false diagnosis.
https://www.theguardian.com/world/2021/ ... robability
With a sufficiently large pool of trustworthy volunteers, they could attempt an experiment whereby they asked some volunteers to use engine assistance all the time some of them some of the time and others not at all. That would check the reliability or otherwise of their detection methods. In addition the part of their methods which uses the quality of play can be tested against pre engine historical games and contemporary correspondence games.Paolo Casaschi wrote: ↑Sun Apr 18, 2021 4:18 pm
There's no reliable way for chess.com or lichess to really know what the rate of false positive of their algorithms actually is.
Roger de Coverly wrote: ↑Sun Apr 18, 2021 5:08 pmWith a sufficiently large pool of trustworthy volunteers, they could attempt an experiment whereby they asked some volunteers to use engine assistance all the time some of them some of the time and others not at all. That would check the reliability or otherwise of their detection methods. In addition the part of their methods which uses the quality of play can be tested against pre engine historical games and contemporary correspondence games.Paolo Casaschi wrote: ↑Sun Apr 18, 2021 4:18 pm
There's no reliable way for chess.com or lichess to really know what the rate of false positive of their algorithms actually is.
All of Roger's suggestions could be done without involving chess.com or lichess. A sufficiently large group of volunteers could do it themselves. I'm surprised somebody hasn't done it already.
By "without involving chess.com or lichess ", I meant without telling them you were doing the analysis on their site and of their methods.JustinHorton wrote: ↑Sun Apr 18, 2021 7:04 pmI am missing something here - whose detection methods is this research using
But there is a credible explanation - he's deliberately losing.Matthew Turner wrote: ↑Sun Apr 18, 2021 8:37 amHuman observation of the games offers no credible explanation for the player's play.
That's not quite accurate. Our criteria is going to be who gets banned and who doesn't.JustinHorton wrote: ↑Sun Apr 18, 2021 7:26 pm... , our only criteria are going to be who gets caught and who doesn't ....
Yes, but for what purpose?Jonathan Bryant wrote: ↑Mon Apr 19, 2021 8:09 amBut there is a credible explanation - he's deliberately losing.Matthew Turner wrote: ↑Sun Apr 18, 2021 8:37 amHuman observation of the games offers no credible explanation for the player's play.