Chess Player Strip Searched

[Roger de Coverly]

Not only is that the only fair way to do things you get a massive deterence effect too, and hopefully avoid the sorts of semi lynch mobs that we currently seem to be getting at times.

I don't see how it avoids a lynch mob. A game is played in which a established star is defeated quickly by a lower rated or lesser known player. All or nearly all of the winner's moves match an engine's first choice. It's completely legitimate that the reason for this is that most or all of the game was prepared earlier with engine assistance.

[Jonathan Bryant]

A game is played in which a established star is defeated quickly by a lower rated or lesser known player. All or nearly all of the winner's moves match an engine's first choice. It's completely legitimate that the reason for this is that most or all of the game was prepared earlier with engine assistance.

Well, yes. I'm sure we can all think of high profile accusations of cheating that were unwarrented.

I agree that false accusations of cheating are corrosive. Wide spread accusations/suspicion based on nothing would probably be more damaging than the cheating itself. All the more reason to have robust measures in place to catch the engine users, I'd say.

And as a group, as well as condemning cheating it wouldn't be at all out of place for chessers to respond "no, that's cobblers" whenever an somebody starts claiming "cheat" on the basis of a single game.

[PeterFarr]

Not only is that the only fair way to do things you get a massive deterence effect too, and hopefully avoid the sorts of semi lynch mobs that we currently seem to be getting at times.

I don't see how it avoids a lynch mob. A game is played in which a established star is defeated quickly by a lower rated or lesser known player. All or nearly all of the winner's moves match an engine's first choice. It's completely legitimate that the reason for this is that most or all of the game was prepared earlier with engine assistance.

As Jonathan said, you wouldn't base it on a single game, you analyse a series of games (including making the appropriate allowances for pre-game preparation).

Was it Freakonomics that had chapters on using statistical analysis to catch cheating sumo wrestlers and school teachers?

If you can use the method for that I don't see why you can't find a reliable way of doing it in che

Yes it was Freakonomics; worth noting also that the Sumo wrestling scandal was uncovered using only analysis of the scores and not the quality of play; with chess you can look for suspicious positive and negative results as a starter, and then analyse the underlying games.

[David Guthrie]

I'm also extermely sceptical that even long sequences of moves (or games) where a player's moves all match an engine's first choice moves proves very much.

However, maybe this doesn't matter too much if the detection methods are also looking for other less ambiguous signs of computer involvement (or non-involvement) i.e. checking games for moves no computer would make or checking games for moves that ONLY computers would make e.g.:

(1) Blunders or similar mistakes e.g. If the best move in a position is far superior to the alternatives (e.g. the first choice move has an evaluation of say +6, and the next 3 best moves are all -1), and the best move isn't played - its a pretty safe inference that a computer wasn't used for that move;

(2) Computer only moves. This category is bound to be more controversial but I think that in some special situations the computer's "fingerprints" are hard to mistake. e.g. Giving away material for no humanly discernable reason other than entering into a 8-piece tablebase position where mate in 49 is forced or similar. I'm sure that there must be others - the difficulty might be that you are likely to need very strong players involved to spot such signs as computers themselves might not so good at it. I don't know.

[PeterFarr]

Actually it might take a shorter time than one first thinks to determine computer-move matching.

Completely hypothecally, Let's say a very strong player has a 95% chance of matching the computer for any single move (o.k. an incredible simplification, but this is just an illustration).

After 40 moves, there is then only a 13% chance that the human will have followed the computer every time. After 3 x 40 move games the odds become 0.21%.

(n.b. my maths is very rusty, could be errors here).

Of course you can dispute the 95% starting assumption - and as David implies, the odds for any individual move would vary considerably; also one has to discount pre-game preparation somehow, but it still shouldn't take that long for a human to diverge from a computer.

It's maybe reminiscent of the old grains of rice on a chessboard 1 + 2 + 4 + 8.....64 story - it doesn't take that long to get to astronomic numbers.

[Roger de Coverly]

After 40 moves, there is then only a 13% chance that the human will have followed the computer every time. After 3 x 40 move games the odds become 0.21%..

The validity of multiplying probabilities in that manner is correct only if the moves are independent. In the context of a single game of chess, that's quite obviously false as moves form part of a plan, idea or technique. Also chess is a game of skill, so on paper at least, a competent player will attempt to select the best move, so no randomness enters into it.

There's a condition of prior knowledge, so if a player known to be a beginner makes a whole series of accurate moves, the inference that a computer engine was consulted during the game is more "likely" than if the same moves had been played by a GM. That's essentially the method the online chess sites try to use.

[Rob Thompson]

I think you need a more Bayesian approach - given that the human matches the computer with probability x, what is the probability that cheating is occurring. This is nowhere near as trivial as the case given above. Roger is also correct to point out that assumed independence is an issue, as is the apparent disregard for rote memorisation of theory (both opening and endgame).

[Jonathan Bryant]

... given that the human matches the computer with probability x ....

And presumably you'd have to vary x according to the human. Fairly obviously x for Mickey Adams would be higher than x for me. Roger's x being somewhere in between etc.

Endgame theory is an interesting one. Sometimes the 'correct' human move isn't the quickest win at all, but it conforms to an understandable plan. So you wouldn't expect a match in those cases.

On the other hand there are many many moves that are drawn according to tablebase but a person familiar with the theory would never play.

Match ups in these kind of cases might indicate a different kind of 'cheating signature' to engine use, I suppose.

[PeterFarr]

Yes of course my example is a gross simplification, I said that myself, and moves are not independent, but it doesn't invalidate the general point that it won't take very long for any unassisted human to diverge from the computer. That's all I'm saying.

Also the move independence point is only true over a certain (variable) number of moves - even for very strong players, it's only possible to fully plan a quite small number of moves ahead in most positions.

Finally, of course you don't disregard opening theory - I explicitly mentioned that in my earlier post this morning, but didn't bother repeating it. Ending theory ... mmm... maybe, but I'm not sure how often those rote situations really occur.

Yes, agree with Rob's better formulation of the question.

[Giulio Simeone]

I've already discussed this topic on other websites. If the moves of an human match the computer with a very high frequency, it's indeed likely that the human is a cheater. But that is of little help, for the simple reason that expert cheaters don't use the computer all the time, but only in 1,2,3 key moments each game. And you can't condemn a player just because he has made one or two good moves in a single game.

In the event of the Italian Championship Semifinals, my nation's Federation set up a sophisticated anti-cheating system, just based on computer matches, but that's simply ridicoulous. In order to really discover cheaters, the only thing you can do is to detect their cheating instruments. They have defeated terrorism in airports, they can also defeat cheating in chess. A bit of financial effort is needed, indeed.

[Paolo Casaschi]

In the event of the Italian Championship Semifinals, my nation's Federation set up a sophisticated anti-cheating system, just based on computer matches, but that's simply ridicoulous.

How is the FSI anti-cheating system working? What is actually tested?
Is there any description of the testing procedure from the FSI?

[MartinCarpenter]

Yes really subtle cheating is going to be (very) tricky and is a major long term problem. I think the game could survive that as a sport though. Athletics/cycling etc have more or less survived the endless drugs, although obviously damaged.

Total computer mirroring though is like letting people driving cars instead of running! Well worth stamping on from a very great height.

Detecting the potential physical instruments of cheating at events would be horribly hard to do. Well maybe OK for the very elite tournaments, but for a big open event?

[Giulio Simeone]

How is the FSI anti-cheating system working? What is actually tested?
Is there any description of the testing procedure from the FSI?

I also seeked for more detailed informations, but didn't find very much on the web. Probably the best person who can tell you is Ezio Montalbini, who ran the live broadcasting of the games and, probably, also the anti-cheating system. His e-mail address is [email protected].

[Matthew Turner]

Oops. Double post. Delete delete.

Was it Freakonomics that had chapters on using statistical analysis to catch cheating sumo wrestlers and school teachers?

If you can use the method for that I don't see why you can't find a reliable way of doing it in chess.

Freakonomics effectively showed that Sumo matches were 'fixed' because wrestlers who had already retained their place in the top division (almost always) lost their last bout against a wrestler who needed to win to avoid relegation. The Freakonomic method could in no way catch which particular wrestlers were 'cheating'. It is also the case that in a bout where one combatant is highly motivated and the other has very little to play for that past form might not be such a great indicator. As with most of Freakonomics it is a good story and quite interesting, but also a little superficial.

[PeterFarr]

Freakonomics effectively showed that Sumo matches were 'fixed' because wrestlers who had already retained their place in the top division (almost always) lost their last bout against a wrestler who needed to win to avoid relegation. The Freakonomic method could in no way catch which particular wrestlers were 'cheating'. It is also the case that in a bout where one combatant is highly motivated and the other has very little to play for that past form might not be such a great indicator. As with most of Freakonomics it is a good story and quite interesting, but also a little superficial.

Very true. I guess a chess analogy is that over a certain number of games you might "prove" (subject to the caveats people have expressed) that a player is a cheat, but you probably couldn't nail down cheating in any given game. You do need other evidence.

Picking up on something Jonathan Bryant said, it might be good to see a professional magician take a look at the Ivanov case; or perhaps ask a few magicians to try to construct their own cheating methods in tournament conditions.

As to whether cheats might only use computers 2-3 times a game, I suppose it's a question of what is enough cheating for a player to get whatever boost he is looking for - depending on the real strength and also the greed of the player. As the financial rewards in chess aren't that high, I tend to think you would typically need a fair bit of cheating over time for it to be worthwhile, which would increase the chance of eventually being caught out by statistics, but maybe that's too optimistic. The Don Azov Cup case in the Ken Regan article isn't much help in this regard - the cheating was far too obvious to be a fair test of the use of statistics; you couldn't expect a half-way competent cheat to be that thoughtless.