Saturday, 24 February 2018

It's not always how many moves

For a couple of different reasons, I've been looking over some the work I had previously done in the field of Anti-Cheating in Chess. While doing so, I came across a slightly different take on detecting engine use in online chess.
The simple approach is to compare the moves played with what a strong engine might play. However this generally only catches people who aren't smart enough to cover their tracks, although this is still quite a large number of guilty players. As discussed in this answer on Quora, it is often a shorter run of moves that is the giveaway, rather than the entire game (Note: This method was recognised as a possibility when I was a member of the FIDE Anti-Cheating Committee). Also mentioned in the answer are the conditions for turning on (or off) an engine during the game.
The other issue with move matching is that it returns differing results for different styles of games. The classic example of this were the respective performances of Mikhail Tal and Anatoly Karpov at the Montreal 1979 Super Tournament. In a retrospective examination of the games by Professor Kenneth Regen, Tal has the highest move match with modern chess engines, at a little under 70% (IIRC). Karpov had the lowest match of the players in the event, at least than 50%. Interestingly, they tied for first place with 12/18.
The explanation for this is due to the differing styles of players. Tal's games involved a lot of positions where the second or third best move was significantly worse than the best choice (due to the tactical nature of the positions), while Karpov's positions had a number of moves that were good, and it was a matter of his long term understanding of the position as to which one was chosen.

No comments: