Houdini move matching snares more chess cheats

[mwyoung]

Robert Houdart: "No human being can consistently play the #1 or #2 choice of Houdini. Detecting this pattern exposes the cheating, no further evidence is required."

I always suspected Paul Morphy of using a computer program to cheat, humans are not capable of playing the way computers play. When you see a game that move matches Houdini 100% no further evidence is required. Paul Morphy is guilty!


[Event "Paris"]
[Site "Paris"]
[Date "1858.??.??"]
[Round "?"]
[White "Paul Morphy"]
[Black "Duke Karl / Count Isouard"]
[Result "1-0"]
[ECO "C41"]
[Annotator "Houdini 3 Pro x64"]
[PlyCount "33"]
[EventDate "1858.??.??"]

1. e4 e5 2. Nf3 d6 3. d4 Bg4 4. dxe5 Bxf3 5. Qxf3 dxe5 6. Bc4 Nf6 7. Qb3 Qe7 {
1.43/18 last book move} 8. Nc3 {1.26/22} c6 {1.08/19} 9. Bg5 {0.99/19} b5 {2.
58/20} 10. Nxb5 {2.42/18} cxb5 {4.62/13} 11. Bxb5+ {4.46/14} Nbd7 {7.29/18} 12.
O-O-O {7.61/19} Rd8 {9.25/16} 13. Rxd7 {9.07/17} Rxd7 {11.25/18} 14. Rd1 {11.
43/20} Qe6 {15.62/9} 15. Bxd7+ {15.01/9} Nxd7 {#2/9} 16. Qb8+ {#1/1} Nxb8 {#1/
11} 17. Rd8# 1-0

[User923005]

Indeed.

Even with a test set like "Winning Chess Sacrifices" *most* strong engines will play identical moves.
Further, I would expect Super GM's to echo those very same moves almost all the time.

This sort of evidence is circumstantial to me.
Not necessarily to be ignored, and not to say it does not raise eyebrows. But it is not a proof of wrongdoing.

[is_revs]

Yeah Yeah, Morphy probably used a smartphone.

Some games of him which are considered GEMS of chess would certainly match

Remember Morphy was a GENIUS. Not a normal human chess player.

You have to get games from normal human players and find the 90%+ matching moves

Probably a human can have a high percentage matching in a single game.

But we are talking about a guy whom is doing it every single game. THAT is the diference

[Jeremy Bernstein]

You cannot use tactical games and forced sequences for your silly comparisons. You know this, but continue to insist.

[User923005]

I don't know if we know what is useful and what is not.
Empirical evidence is enough to raise suspicion. But the burden of proof is on the accusers and not the accused.
Houdini won't even match its own moves 90% of the time, so if a human matches the moves 90% of the time, that would really be shocking.
Even at that, I would consider such matching as circumstantial evidence.

[BB+]

1. e4 e5 2. Nf3 d6 3. d4 Bg4 4. dxe5 Bxf3 5. Qxf3 dxe5 6. Bc4 Nf6 7. Qb3 Qe7 {
1.43/18 last book move} 8. Nc3 {1.26/22} c6 {1.08/19} 9. Bg5 {0.99/19} b5 {2.
58/20} 10. Nxb5 {2.42/18} cxb5 {4.62/13} 11. Bxb5+ {4.46/14} Nbd7 {7.29/18} 12.
O-O-O {7.61/19} Rd8 {9.25/16} 13. Rxd7 {9.07/17} Rxd7 {11.25/18} 14. Rd1 {11.
43/20} Qe6 {15.62/9} 15. Bxd7+ {15.01/9} Nxd7 {#2/9} 16. Qb8+ {#1/1} Nxb8 {#1/
11} 17. Rd8# 1-0

As far as I can tell, by Regan's methodology, exactly 2 White moves (9 and 10) and one Black move (9) would be considered for comparison. From such a small sample pool, the z-statistic for Morphy (2 from 2, then to be adjusted for forcedness of Nxb5) does not seem to be much aberrant. For instance, in one of Ivanov's games (versus Kurajica), Regan states that Ivanov matched Houdini 3 at depth 17 in 16 from 16 relevant moves (any specific accounting of the weighting of said 16 moves is not given).

Regan's letter to the ACP has a discussion concerning how often one should expect such runs (phrased in terms of adjusted z-statistics), though he somewhat kookily (to me) wants to phrase the whole thing in terms of "Intrinsic Performance Rating" (which really means how often it matches a test engine), and furthermore introduces an exterior element into the analysis by assuming that Ivanov "should" play at the 2300 (or 2339) level [Regan himself notes that this could be considered prejudicial against Ivanov], and upon this additional assumption concludes the IPR of ~3000 is barely explicable (Regan's z-statistic for Ivanov in the MultiPV test were he rated 2700 would only be 3.21, compared to the "adjusted" 4.72 that he comes up with for a 2339 player). At least from the model Regan presents, I don't think Ivanov's move-matching to Houdini/Rybka at the Zadat Open is sufficiently extraordinary that it could not occur by chance from a strong human player -- but Regan goes further, and determines (from his model) that it is extremely unlikely that a presumably 2300 player would match Houdini/Rybka moves so often.

[User923005]

Nonsense.
What if he used Houdini to prepare some particular opening sequences before the match?
Given some particular opponents tendencies to play particular openings, such a thing seems downright likely to me.
Further, there is nothing wrong with doing that.

There is some sort of pretense here that all of the random variables have been isolated and that these events under study are completely independent variables.
Stuff and nonsense.
IMO-YMMV.

These studies, even when conducted correctly, do not show cause.
https://en.wikipedia.org/wiki/Correlati ... _causation

Now, the move matching stuff should trigger further investigation, if the correlation is unusual.

[BB+]

What if he used Houdini to prepare some particular opening sequences before the match?

Regan mentions this on page 5 of the Appendix of his ACP letter (side observations). In addition to his comment (which is rather oblique), it would also be strange (say), for moves 20-30 of the Kozul game to considered "preparation" (White's 9th move Qc1 is already a novelty says Regan), or similarly moves 15-25 with Kurajica (where Black's 10th move b6 is new according to Regan).

There is some sort of pretense here that all of the random variables have been isolated and that these events under study are completely independent variables.

Regan mentions this on page 7 of the Appendix of his ACP letter (third critique). He does not exactly solve the problem, however.

Now, the move matching stuff should trigger further investigation, if the correlation is unusual.

Regan says essentially the same in his conclusion.

[User923005]

In that case, our conclusions are the same.
I suppose that titles such as "snares chess cheats" are what torqued my bolts a few foot pounds over much.

[BB+]

In that case, our conclusions are the same.

Yes, I find Regan's writing to be quite couched, he does not exaggerate his methods, he discusses their limitations, etc. In fact, one of the stated "goals" he gave was: (c) To educate the playing public about the incidence of deviations that arise by chance, and their dependence on factors such as the forcing or non-forcing quality of their games. Well, good luck with that, at least for the general populace.

When I spoke with Jürgen Klüners (who is a Bundesliga arbiter) back in Jan 2012, the Natsidis incident was still recent enough, and he indicated that something like (a) [deterrence of prospective cheaters], particularly precluding "experimentation" in cheating (say in a last-round game, "just to see if/how it might work"), was a main goal. When I attended a Bundesliga match earlier this year, there was a ~300 Elo upset victory, and the loser (predictably) was saying that his opponent played a bunch of "computer moves" and such; his teammates seemed to calm him down before anything formal was done [and in fact his opponent simply made a speculative sac that really wasn't all that great, but maybe due to psychology, the defense was not the best, and 20-odd moves later, following mistakes by both sides, the opponent emerged with a won endgame].