3 ms·

It sounds like cheating detection in chess is highly circumstantial, based on outlier data points. Given the massive volumes of chess data out there, I bet it

by awb 4y ago

It sounds like cheating detection in chess is highly circumstantial, based on outlier data points. Given the massive volumes of chess data out there, I bet it would be easy to make a bot that mirrors the win/loss and blunder/best move ratio of an average GM chess player’s trajectory. Instead of picking the best move all the time, maybe pick a random move out of the top 10 moves. And use a random number generator to determine if you make a blunder. And use an engine that’s appropriate to your current rating. It looks like Hans might have been picking the best moves from the best engine in those games where he got 100% accuracy, which is a dead giveaway. If you’re patient enough to sit through hours of slowly progressing your rating through a few wins, lots of draws a few losses, I bet you could do it undetected. And don’t give any interviews! Hans probably didn’t think about the trail of data he was leaving, or got greedy.

oldstrangers 4y ago
Except, chess.com has literal "strength profiles" of every top player. They have enough data to say, hey, this how strong you should be playing, anything outside that expected norm might be suspicious, especially over a long enough time period. It's a similar idea to the 'Biological Passport' seen in sports wherein the biological data of an athlete is stored and you can reference the data points against new test results to see any anomalies. Creating the "perfect cheating engine" would only be useful for unknown players, and in chess, it's literally impossible to be an unknown and play at a Super GM level (its happened, and of course they were immediately exposed for cheating). Now, if someone like Hikaru or Magnus decided to just automate themselves at their best in the form of a bot, yeah, it would be nearly impossible to detect them on data alone.
- awb 4y ago
  I was more thinking that cheaters would create a new account and play at a level that’s good but not uncommon like 1700 or 1800 and work their way up to GM status over a few years, careful to coincide their ascension with the historical pace, win rate, etc. of fellow GMs. That way there’s no anomalies in the data.
  - oldstrangers 4y ago
    Theoretically sure, but without a title they'd never really get to play in any online money tournaments, and to get a title they'd have to play over the board games where cheating would be a lot harder. So it would be a lot of work to be a very strong online only player that doesn't make any real money. As a proof of concept it would be interesting to create an AI that has an undetectable natural progression and human-like play style.

by awb 4y ago

oldstrangers 4y ago

Except, chess.com has literal "strength profiles" of every top player. They have enough data to say, hey, this how strong you should be playing, anything outside that expected norm might be suspicious, especially over a long enough time period. It's a similar idea to the 'Biological Passport' seen in sports wherein the biological data of an athlete is stored and you can reference the data points against new test results to see any anomalies. Creating the "perfect cheating engine" would only be useful for unknown players, and in chess, it's literally impossible to be an unknown and play at a Super GM level (its happened, and of course they were immediately exposed for cheating). Now, if someone like Hikaru or Magnus decided to just automate themselves at their best in the form of a bot, yeah, it would be nearly impossible to detect them on data alone.

awb 4y ago
I was more thinking that cheaters would create a new account and play at a level that’s good but not uncommon like 1700 or 1800 and work their way up to GM status over a few years, careful to coincide their ascension with the historical pace, win rate, etc. of fellow GMs. That way there’s no anomalies in the data.
- oldstrangers 4y ago
  Theoretically sure, but without a title they'd never really get to play in any online money tournaments, and to get a title they'd have to play over the board games where cheating would be a lot harder. So it would be a lot of work to be a very strong online only player that doesn't make any real money. As a proof of concept it would be interesting to create an AI that has an undetectable natural progression and human-like play style.