Monday, January 2, 2012

RYBKA "CHEAT SCANDAL" ON CHESSBASE

At last ChessBase acknowledges there's been something "rotten" in Denmark about Rybka 4.1. (ChessBase and Convekta manufacture commercial versions of Rybka.)

A Dr. Soren Riis has begun outlining an article (paper?) in two parts. What he sets forth is less than exciting--even following code could be more interesting.

How he defends and/or proposes to defend Rybka's creator, IM Vasik Rajlich is enough to drive anyone to distraction or else it is a lesson on how to become an expert in ADD (attention deficit disorder). I "knew" in advance the line of defense would follow along the path of, "how much copying" is infringement? Basically, "what percentage of code is present in Rybka that was in Fruit?"

In a way that is an interesting question and does point to a problem with the investigating body's laws. Rajlich himself says that anything he copied was, in effect, fair use, and not that important to the case. Aha, any decent lawyer (is there such a thing?) would pounce on that remark. If it wasn't important, why was it included in the first place? Did he even state that in the coding? (I don't think he did.)

In an article in the most recent issue of Photoshop User, the subject of "fair use" comes up and the author rightly says that in many cases "fair use" is up to the courts to (actually) decide and that in a large number of cases we see "fair use" being claimed when in reality it isn't "fair" at all. On YouTube, for example, songs and videos are dotted everywhere with the "exclusion clause" of something like, "I am not the owner of this work but am putting it forth under article whatever in the copyright law as fair use and for educational purposes." The claim isn't vague, but it's intents to publish something created by someone else is waaaay over the top. All the "owner" of the copyright has to do is claim infringement and Google takes it down (you are aware Google owns YouTube aren't you?) Google is the pervasive company that was going to scan as many books and magazines and newspapers as it could, willy nilly, and store that data online for all to see and use regardless of who actually created the content. Publishers and authors, so far, have halted that.

The article on the ChessBase website itself needs drastic editing as many (75%?) statements are repeated again. The game positions are shown only once (thank God) but reading the article feels like deja vu all over! I estimated 75% because, unlike what Riis claims, unerring, undying accuracy to defend "copying" by "how much is enough," I am not in a court of law and not wasting my time counting words.

He (Riis) admits to the vitality and erudition of many (that number again) of the 32 programmers used on this "case" to determine how much Rajlich violated the ICGA's guidelines. Unless these guys have an axe to grind, they should have HUGE "weight" on this problem... that is, they recognized "naughty, naughty" WHEN THEY SAW IT (Riis tries to make light of this concept to some degree (vague again)).

As a former programmer myself (Fortran) I only have two questions:
1. Everyone codes differently no matter what this article says. We use different variable names, include comments (or not) and using loops, or whatfors, etc. may be similar but often there are various ways of "skinning the cat" so was Rajlich's stuff "straight copying or not?"
2. Even IF Rajlich did take common use code (say from a published search algorithm available for wide use) WHAT made his program superior to all the others in terms of rating strength? Was it a combination of using the BEST of OTHERS works or what? If the best was not plagiarism, that is, previously existing code (available to all), then there is no problem.

(Recall Kasparov's never ending complaint against Deep Blue's cheating. GM Joel Benjamin answers Kasparov in American Grandmaster. Kasparov was, to put it plainly, incredibly embarrassed when he called his 6th move of the last game a "fingerfehler." Does it get any weaker than this? Blaming his finger? No, blame the programming.)

But the fact that somewhere short of 32 "experts" deciding on this matter constitutes "proof" they come up with: "Rajlich, you've been a very bad boy," is most likely enough in an investigation of this type. Certainly the ATF had no damning, and absolute photo evidence of Mr. Anthrax when they said in a report that they believed Bruce Ivins was the one responsible for the anthrax deaths. Their report was very deep. One researcher admitted, as soon as she heard the case, the first person she thought of was Ivins. Only the "conspiracy theorists", who must be the same people in the questions of the Kennedy assassination, the no man on the moon blow up, etc. would be impossible to convince. (Although it's a shame that the "investigators of the Kennedy assassination" weren't available for the Warren investigation to call upon. It's tricky when the "government," in effect, investigates itself.)

The semantics of "how much" is supposed to allow avoidance of contempt. Who says so?

It would have been much easier for Rajlich, at the very beginning, to just include in the program, "I used Mr. X and Mr. Y's code for purposes of blah blah blah, but the implementation and organization is all mine and this program will kick anyone else's ass!" And until Houdini came along, it did. Now, can we ask at this point how "original" Houdini's code is?

bob@thinkerspressinc.com

1 comment:

  1. I'm new to this controversy (I learned of it this morning on Volokh), but I found Riis's defense utterly compelling. I also thought your post was extremely unfair to Rybka. Now that the entire four part defense has been published, have you changed your opinion?

    Taking Riis's piece at face value, I'd make four points:

    1. Long before their was any controversy, Rajlich very publicly acknowledged the enormous contributions of Fruit to his work. You appear to suggest otherwise in your last paragraph.

    2. Of the 34 members of the panel, only seven were actually involved in the deliberations, and most of them depended on the work of 1-3 members of the panel.

    3. The information provided by those 1-3 members contains gross inaccuracies. For example, page upon page of side by side comparisons between the Fruit code and code labeled "Rybka" were provided. These pages show a striking similarity between the two. Unfortunately, the code labeled "Rybka" is actually a reverse engineering of the Rybka source code which was deliberately designed to match the Fruit code as closely as possible.

    This was admitted by the committee member who created them: "The easiest way to show a layperson that the Fruit source matches the Rybka binary is to make our 'pseudo-Rybka source' match Fruit as closely as possible."

    Unfortunately, it is highly probable that at the time the committee made its decision, its members believed that the 'pseudo-Rybka source' was the real thing.

    4. There does not appear to be any conclusive evidence that even a single line of Fruit code was directly copied into Rybka. It is true that Riis starts out by arguing that even if some copying occured, it wouldn't be significant. Be he then goes on to thoroughly debunk the supposed evidence that copying occurred at all (and provides compelling evidence of differences that suggest it did not).

    ReplyDelete