Borislav Ivanov ain't playing so well this time around

[Simon T]

this document. (correction: the document is referring to the Zadar Open, not the Old Capital Open).

Thanks, that is a good resource (in contrast to those Lilov videos).

Statistical analyses are a sensible method to combat this problem. Much better than watching too many Bond movies and speculating about bionic contact lenses.

[User923005]

Statistical methods can show likely and unlikely.
It will be possible to convict someone who is innocent with such evidence.
Catching them with illegal electronic devices in use seems more reasonable to me.

[Simon T]

Catching a cheat in the act with physical proof is good but could be too difficult to catch enough of them. Small devices can be hidden and too intrusive a search would be needed to find them.

The advantage of statistical detection is that it's impossible to hide the evidence as the moves and their timings are necessarily made public. There's a chance of a miscarriage of justice but this can be reduced by adjustments to the methods.

[mwyoung]

Here's a rebuttal from Andrew Martin: http://www.youtube.com/watch?feature=pl ... zbPytRKY#!

Very tricky situation. I disagree with Martin's stance, but I understand his reluctance to damn Ivanov on the basis of Lilov's videos. The statistical evidence is pretty tough to refute, though. See also: Jens Kotainy, about whom a very lengthy (and very ironic) article in last month's "Schach" was written (with rebuttal by Jens, as well).

jb

I have to side with martin's video. I think he refuted all of Lilov's accusations. I don't know if Ivanov is cheating or not. But I can say Lilov's evidence is bunk. There is no contact lens technology on the market that can do what Lilov's claims. And Martin clearly shows that Lilov's claim that no human would make these moves evidence is also bunk. It is clear that Lilov has a biased point of you, and will make untrue claims as evidence to try and prove his point of you. There is no evidence, and until you do a detailed statistical analysis of many players you can not claim move matching with computers as evidence of cheating.

I have used Houdini to analyse many strong players games, and I have seen games were the human player had a very high correlation with Houdini. With some of these games I remember were played before computers were invented. . So much work would be needed to show this works as a valid method, and still it could not be used as hard evidence.

I would like to see Lilov or someone else make a video showing *HOW* Ivanov is cheating.

Please answer these question.

How is Ivanov transmitting moves?

How can Ivanov transmit moves so quickly in 10 sec or less the whole game, from Ivanov seeing the move played, then transmitting the move, then inputting the move into the computer programs, letting the computer calculate the move, reading the output from the chess engine, playing the move on board, then hitting his clock. In 10 seconds or less. That seems to me proof that he is not cheating. UNLESS you can demonstrate how this can be done, with no visible technology being seen. I am opened minded on this, but someone needs to show and demonstrate this in a video.

Were is the computer located that is running Houdini, a contact lens will not run a chess program.

[Jeremy Bernstein]

Here's a rebuttal from Andrew Martin: http://www.youtube.com/watch?feature=pl ... zbPytRKY#!

Very tricky situation. I disagree with Martin's stance, but I understand his reluctance to damn Ivanov on the basis of Lilov's videos. The statistical evidence is pretty tough to refute, though. See also: Jens Kotainy, about whom a very lengthy (and very ironic) article in last month's "Schach" was written (with rebuttal by Jens, as well).

jb

I have to side with martin's video. I think he refuted all of Lilov's accusations.

I think that's hyperbole, in that Lilov has made videos about numerous games and Martin chose a single one which isn't a great illustration of the kinds of "Houdini moves" that many of Ivanov's games have featured. My teacher, an IM, has only one question: how is he doing it? Examining the moves, he comes to exactly the same conclusion as FM Lilov (and this is a guy who, when analyzing my games (without a computer), very often picks the computer move as an alternative).

Valeri Lilov has made this personal, which I think is too bad, but I don't think that he's wrong, nor can Andrew Martin hand-wave the statistics away. Somewhere in the video he makes what I think is the essence of his argument: something like "if you can't prove it 100%, don't say it, because there's a chance you'll look stupid later". A defensible, logical stance, but one which hasn't served humanity well in the past.

But you're certainly entitled to your opinion.

Jeremy

[mwyoung]

Here's a rebuttal from Andrew Martin: http://www.youtube.com/watch?feature=pl ... zbPytRKY#!

Very tricky situation. I disagree with Martin's stance, but I understand his reluctance to damn Ivanov on the basis of Lilov's videos. The statistical evidence is pretty tough to refute, though. See also: Jens Kotainy, about whom a very lengthy (and very ironic) article in last month's "Schach" was written (with rebuttal by Jens, as well).

jb

I have to side with martin's video. I think he refuted all of Lilov's accusations.

I think that's hyperbole, in that Lilov has made videos about numerous games and Martin chose a single one which isn't a great illustration of the kinds of "Houdini moves" that many of Ivanov's games have featured. My teacher, an IM, has only one question: how is he doing it? Examining the moves, he comes to exactly the same conclusion as FM Lilov (and this is a guy who, when analyzing my games (without a computer), very often picks the computer move as an alternative).

Valeri Lilov has made this personal, which I think is too bad, but I don't think that he's wrong, nor can Andrew Martin hand-wave the statistics away. Somewhere in the video he makes what I think is the essence of his argument: something like "if you can't prove it 100%, don't say it, because there's a chance you'll look stupid later". A defensible, logical stance, but one which hasn't served humanity well in the past.

But you're certainly entitled to your opinion.

Jeremy

You can hand wave the statistics away, because no one has done any work to show that move matching is any kind of proof. That is just a fact.

Do other chess players also move match Houdini? The answer is no one knows.

If a player does move match Houidini does this prove a player is cheating. The answer is no one knows.

The guilt that your are putting on Ivanov is base 100% on speculation. That is the problem.

That is the point Martin is making, He may or may not be cheating.

But until you or someone else can show *How* Ivanov can cheat with no electronic device being visible, That works with Ivanov only looking at the chess board, and making moves like any other chess player. And can work very quickly and accurately in as little time as 10 seconds. All you will have is speculation and opinion, and much more is needed.

My speculation is the same as yours, something stinks. But I know there is also no proof of any kind.
And I know there is a chance my speculation could be wrong, because of what my speculation is based on.

If Ivanov is guilty, I would like to know how Ivanov is cheating.

[BB+]

What exactly is Regan's IPR (step 7 of Procedure)? He states an IPR of 3126-3258 for Ivanov (after ignoring the last 2 games perhaps), but in another paper [second link below, Section 5] they quote LK as saying R3 at depth 14 is maybe about 2750 (they use it at depth 13 in fact). So is this "IPR" just a Rybka/Houdini-matching indicator, rather than anything to do specifically with "rating"? It makes little sense for an engine rated 2750 to propose that someone played at the 3200 level from a data sample.

In the third link below, they say:

Code: Select all

A simple linear fit then yields the rule to produce the Elo rating for any (s, c),
which we call an “Intrinsic Performance Rating” (IPR) when the (s, c) are obtained by
analyzing the games of a particular event and player(s).
IPR = 3571 − 15413 · AEe . (6)
This expresses, incidentally, that at least from the vantage of RYBKA 3 run to reported
depth 13, perfect play has a rating under 3600. 

Which hardly leads me to think that they should call this "IPR", but rather something like "R3-correlation (weighted by eval loss) at depth 13" (linearly fit into a convenient number that "looks" like a rating, but is only loosely meaningful as such).

http://www.cse.buffalo.edu/~regan/chess ... report.pdf
http://www.cse.buffalo.edu/~regan/paper ... eHa11c.pdf
http://www.cse.buffalo.edu/~regan/papers/pdf/RMH11b.pdf

See also
http://www.cse.buffalo.edu/~regan/chess ... curial.txt

[is_revs]

Depending on the Position the ply level that Houdini can reach varies. It can reach above ply 20 on endgame positions or above ply 18 just after the opening. As the hardware supposed to be used by Ivanov
has not been discovered yet, we cannot say it would play at ply 14 and therefore at 2750 strenght.
Up to now for the software's IPR we can only guess that it could match the IPR shown in the tournament games.

What exactly is Regan's IPR (step 7 of Procedure)? He states an IPR of 3126-3258 for Ivanov (after ignoring the last 2 games perhaps), but in another paper [second link below, Section 5] they quote LK as saying R3 at depth 14 is maybe about 2750 (they use it at depth 13 in fact). So is this "IPR" just a Rybka/Houdini-matching indicator, rather than anything to do specifically with "rating"? It makes little sense for an engine rated 2750 to propose that someone played at the 3200 level from a data sample.

In the third link below, they say:

Code: Select all

A simple linear fit then yields the rule to produce the Elo rating for any (s, c),
which we call an “Intrinsic Performance Rating” (IPR) when the (s, c) are obtained by
analyzing the games of a particular event and player(s).
IPR = 3571 − 15413 · AEe . (6)
This expresses, incidentally, that at least from the vantage of RYBKA 3 run to reported
depth 13, perfect play has a rating under 3600. 

Which hardly leads me to think that they should call this "IPR", but rather something like "R3-correlation (weighted by eval loss) at depth 13" (linearly fit into a convenient number that "looks" like a rating, but is only loosely meaningful as such).

http://www.cse.buffalo.edu/~regan/chess ... report.pdf
http://www.cse.buffalo.edu/~regan/paper ... eHa11c.pdf
http://www.cse.buffalo.edu/~regan/papers/pdf/RMH11b.pdf

See also
http://www.cse.buffalo.edu/~regan/chess ... curial.txt

[pgn4web]

Somewhere in the video he makes what I think is the essence of his argument: something like "if you can't prove it 100%, don't say it, because there's a chance you'll look stupid later". A defensible, logical stance, but one which hasn't served humanity well in the past.

Yeah, setting this logical stance aside, for instance, eradicated the problem of witches: back in the middle ages they burned alive so many witches than none survived to our days. Well done.

[Jeremy Bernstein]

Somewhere in the video he makes what I think is the essence of his argument: something like "if you can't prove it 100%, don't say it, because there's a chance you'll look stupid later". A defensible, logical stance, but one which hasn't served humanity well in the past.

Yeah, setting this logical stance aside, for instance, eradicated the problem of witches: back in the middle ages they burned alive so many witches than none survived to our days. Well done.

That's a great point, especially if you equate banning suspected cheaters from tournament play with murder.

jb