Chess Player Strip Searched

[Chris Rice]

Ken Regan interviewed by NPR's Scott Simon on cheating and the Katy Perry gambit.

SCOTT SIMON, HOST:

Ken Regan is a kind of chess detective. He's a computer scientist and an international chess master, who played with the likes of Bobby Fischer as a kid. Which gives him particular skills to help recognize cheating in chess, which, he says, is becoming more common. Ken Regan has created a new algorithm to help detect test cheating. He's profiled this month in "U.S. Chess" magazine and joins us now from Buffalo. Thanks for much for being with us.

KEN REGAN: Thank you.

SIMON: You know, I must say, when your work was first explained to me, there was a part of me that wondered if chess cheating was just a matter of somebody going, you know, hey, Katy Perry's over there, and upsetting the board.

REGAN: Yes, yes. Well, not quite. People don't distract and then take a rook off the board. That doesn't happen.

SIMON: So how does somebody cheat in chess?

REGAN: The most common way is having the game on your smart phone or handheld device and going into the bathroom surreptitiously to check it.

SIMON: So people are consulting their smart phones, because there are algorithms that will tell them what the propitious next move is?

REGAN: Yes. There are chess engines that are very strong - stronger than any human player, apparently even running on the reduced hardware of smart phone.

SIMON: I mean, shouldn't this be pretty easy to detect? I can't imagine that you need to be Sherlock Holmes to see if somebody is consulting their iPhone while they're playing a chess game?

REGAN: Well, that's been true. In some cases, people have had behavior that's suspected, and they've been followed and caught. But in the case in Dortmund, Germany, last August, it was very subtle. You had to notice that the person had his left hand in his pocket and was doing a little fingering with the fingers. If you're not watching closely, you might not see this.

SIMON: Is part of the rise in chess cheating due to the fact that - if I can put it this way - computers are becoming more accomplished at chess, too?

REGAN: Yes. Absolutely. I mean, you know, 17 years ago, Gary Kasparov fell to a supercomputer. But now the saying is that your iPad today is as good as a supercomputer was 20 years ago. So put to do the two together, and that kind of power, as well as improved chess algorithms, is at everyone's fingertips.

SIMON: Explain the kind of algorithm that you're talking about. Help guide us through it.

REGAN: Different chess positions have different character. Some have just a single move for you to stay in the game or a single move to keep your advantage. When that's the case, a strong chess player is highly likely to find such a move, as will the computer, so you'll get a match. So when there's a case where a person says, wow, this guy made a lot of moves that are just like the computer, I analyze the game and say, well, was it a forcing game? Were most of these moves ones that most people would find? And my model will spit out a number that says, yes, and, therefore, this is not a great deviation. In other cases, it'll say, whoa, wait a second. This is a much higher correspondence than these positions will would allow.

SIMON: Professor Regan, how do you analyze these games? How do you keep trap of track of them?

REGAN: Well, one of the great things about chess is the entire record of play is part of the public record. So I can feed the moves into a computer, and the computer will analyze them.

SIMON: Well, what are the odds of somebody being falsely accused?

REGAN: I deal with accusations, whispers, public statements, grouses that people make. And, usually, my model shows, no, this play really was within expectation. The other side is, yes, it's a great danger that the statistics might falsely accuse someone. As a failsafe, I have taken data - many millions of pages of data from the entire history of chess, including all the performances by Bobby Fischer and Gary Kasparov. So I have an idea of the distribution of what happens by nature.

SIMON: Ken Regan, a chest detective and computer science professor at the State University of New York at Buffalo. Thanks much for being with us.

REGAN: Thank you very much.

[Roger de Coverly]

The other side is, yes, it's a great danger that the statistics might falsely accuse someone.

The introduction to the paper on the FIDE site was somewhat lukewarm, suggesting that the proposals as written were unlikely to be implemented.

You might hope that it will not be possible to make credible accusations of cheating solely on the basis of move matching statistics. Over time, the matching could well increase thus invalidating historic benchmarks. I can see two drivers for this. On the human side, memory and pattern recognition can be important parts of chess success. So a player training with a computer is likely to have a higher matching rate than one who didn't. On the engine side, programmers are refining the way that engines select moves, in some cases to get a more human style with respect to long term considerations.

[Paul Dargan]

SIMON: Ken Regan, a chest detective and computer science professor at the State University of New York at Buffalo

New job title ... maybe it is something to do with the Katy Perry gambit after all?

Paul

[Paolo Casaschi]

REGAN: The most common way is having the game on your smart phone or handheld device and going into the bathroom surreptitiously to check it.

Good to hear this is considered the most common way of cheating; however the current proposal of statistical analysis seem to fail when the cheater only uses computer help for few moves during the game. As a result, it looks like the proposed paper seem to address only a very small portion of those most common cases, i.e. when a player goes to the bathroom for checking the smartphone suggestions AFTER ALMOST EVERY MOVE.
Do we really need super-computer analysis in order to raise suspicion about a game where a player goes to the bathroom AFTER ALMOST EVERY MOVE? Remember, by Regan's admission, if a player only uses computer's help for a few moves during a game then he would not be detected by the statistical analysis.

[Roger de Coverly]

As a result, it looks like the proposed paper seem to address only a very small portion of those most common cases, i.e. when a player goes to the bathroom for checking the smartphone suggestions AFTER ALMOST EVERY MOVE.

With the interview being aimed at the general public, I thought the point could have been made more strongly that consultation with external sources of information and advice is already forbidden in competition chess and always has been. Furthermore there has been a strong rule that phones should be switched off at the start of play and remain so for the entire duration of play. Or is that just a UK interpretation and US interpretations are more lax?

[MartinCarpenter]

No, you need some sort of stats to check cases where people have rigged up some much more sophisticated ('invisible') way of cheating and are playing computer suggestions most/every move (Like our favourite Bulgarian....).

People cheating by using mobiles on the toilet etc will be more common as its so much less effort to set up, but should mercifully be fairly easy to catch. Trivial for more serious events. Less overall impact too.

[PeterFarr]

Meanwhile, as everybody worries about electronic forms of cheating, the good old-fashioned way of obtaining GM titles by performing extraordinary feats in obscure provincial tournaments continues unabated. At least two out of the ten GM title applications up for consideration at the Tromso meeting are extremely interesting (or boring if you get fed up with the same names and methods recurring for years on end with nobody doing anything about it).

[Clive Blackburn]

REGAN: I deal with accusations, whispers, public statements, grouses that people make. And, usually, my model shows, no, this play really was within expectation. The other side is, yes, it's a great danger that the statistics might falsely accuse someone. As a failsafe, I have taken data - many millions of pages of data from the entire history of chess, including all the performances by Bobby Fischer and Gary Kasparov. So I have an idea of the distribution of what happens by nature.

How can it be a "failsafe" to use data from historic Grandmaster games?

REGAN seems to miss the fact that the top modern players train using computers, whereas Bobby Fischer never had that opportunity.

That must surely affect the distribution.

[MartinCarpenter]

That makes it rather useful still, as you absolutely know there's no computer cheating in any of that data So you can see what sorts of peak in performance/accuracy of move etc etc you get in a natural setting over an awful lot of games.

You have to know that to have a chance to find signals like computer stuff.

[Roger de Coverly]

So you can see what sorts of peak in performance/accuracy of move etc etc you get in a natural setting over an awful lot of games.

You can see what sort of variability you got before players had access to computers, but is that still relevant? In older games there's the possible effect of adjournments. Today's legitimate play isn't free of computer influence given their widespread use for training.

[MartinCarpenter]

For calibrating things to start of with yes, very much so Lots of factors which might affect things a bit but without some idea of the baseline data you'd just be guessing.

[KWRegan]

To answer Clive Blackburn and Roger de Coverley on this page (58): just like the sources I use (including ChessBase, TWIC, and OpeningMaster), the data set is heavily skewed toward recent years. The calibration in my 2011 papers used 2006--2009 as the master; now I am adding 2010--2013. I mentioned Fischer and Kasparov for popular recognition.

To answer Nick Faulks on page 57. the interview was in May 2013; I have voiced the same Anand / 150-Elo reference numerous times before and since. Regarding "isolated moves", literally the first thing I said to the group in Paris when I walked into the hotel was that my work would not catch someone cocking a yogurt spoon and hearing a spoon drop a minute later. I even dropped a spoon over my right shoulder for effect. Not to have heard things like these from me a few days later in Tallinn could only be a statistical deviation .

[JustinHorton]

Hi Ken. I confess I don't get the 150 figure. What's it based on? Is it come nominal situation where a player, in every game they play, secretly consults a computer once (but only once) in a game and selects the right moment to do so, such that a decisive move is available to them?

If so that seems to me to be a highly unlikely scenario for more than one reason (how would they know to pick the right moment? do such moments usually exist?) but as I don't know what it's based on, I'd be keen to clarify.

PS You won't remember me, but thirty years ago you gave a simul at Oxford University. I lost to you so quickly you invited me to have a second go (which I also lost). Had I consulted a computer at the right moment I would have known not to play ...b6 against a bishop on d3 and a queen on e2.

[KWRegan]

Hello, Justin---Oh for those days... The quote is something I recall being attributed to Anand in an article during the year after "Toiletgate", that is 2006--07. Unfortunately I did not save a copy of the page, and now I cannot find it via Google. It may have been on ChessBase.com. My role is only to concur that the estimate is reasonable.

Here's one way to view it: At the time of the interview my model had the equation (approximately) Elo = 3575 - 15,000*AE, where AE is the average error per move in units of "Pawns in Equal Positions" (hence also the reference to "3600" for perfect play in the article). You can call it 3575 - 150*ae if you prefer centipawn units. So saving 1 cp per move is worth 150 Elo. In a 40-move game, that means being able to avoid one 4/10-pawn error (large enough to show "purple" or now grey at ChessBomb.com, and about the difference between = vs. += or += vs. +/- in automated annotations) would confer 150 Elo benefit, all other things being equal. Finally, and importantly, we have to factor in the chance that you would actually make such an error---that's where dependence on your own skill comes into play. We can factor this dependence back out by supposing it's a move where you perceive two alternatives as "50-50", but the position is so complicated that you suspect at least 8/10 of a pawn value is at stake. This may seem a strange way to think about one isolated game, but the numbers work out that way based on thousands of games actually played.

[Chris Rice]

On the FIDE site they have published the FIDE Report that the Ethics Commission will be presenting to the General assembly in Tromso. Some great stuff here:
1) what looks like a withering attack on the work of the anti-cheating commission
2) Decision on cases referred to the Ethics Commssion includes the Kasparov/Leong agreement and the Agon agreement.


http://www.fide.com/images/stories/NEWS ... nex_31.pdf