Programmers Open Letter to ICGA on Rybka/Fruit

[thorstenczub]

but why is HW in the panel, is he a chess programmer ?

Not the Panel. He's on the Secretariat, which I guess is some sort of Council of Elders which arbitrates/manages any investigations: they don't perform any investigative function themselves, but (attempt to) gather any evidence, using the Panel as an investigative tool and present the collected findings to the IGCA board(?). I disagree with the choice of Harvey for such a task, given his rabid partisanism on these matters in the past, but that's apparently a done deal. Bob isn't exactly the least outspoken member of the CC community, either.

it remains the problem that harvey is in a situation where he is stretched between the interest of the company he works for (he is a member of the hiarcs team, a moderator in the hiarcs forum,
a sysop on chessbase playserver) and the decisions of the panel. its completely unimportant if he is in the panel or in the secretary. he is not objective. he is pro hiarcs. and that makes it impossible to do a job good, or the way one would expect somebody to do such a job.

he should resign this function.
his rigid method and how he judged about the "cloning" issue makes him a person not capable
for such a job.

you maybe forgot those emails ?

From:Harvey Williamson
To:mclane
Posted:Wed Mar 10, 2010 8:54 pm
Subject:Re: Your posts
mclane wrote:
that was a completely different time harvey.




yes you were not supporting clones then - how can you support engines with
unknown authors?

_________________
http://www.Hiarcs.com

From:mclane
To:Harvey Williamson
Posted:Wed Mar 10, 2010 8:58 pm
Subject:Re: Your posts
you now talk about rybka, aren't you ?

_________________
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....

From:Harvey Williamson
To:mclane
Posted:Wed Mar 10, 2010 9:06 pm
Subject:Re: Your posts
mclane wrote:
you now talk about rybka, aren't you ?


bye thorsten you are not welcome here for the moment

_________________
http://www.Hiarcs.com

Harvey is somebody not interested in truth.
he prefers to
HIDE the truth.
his membership in this panel shows that the panel is
not serious.
his interest is not the truth, it is to fit the interest of his
own, of chessbase or hiarcs company.

the emails he has send to me when he banned me showed the
signature of hiarcs company.
so he is not a private person separated from mark uniacke.
he works for the company. he bans people for hiarcs company.
he supported rajlich and
the chessbase business of selling rybka and banning all
people from playchess server who used "clone"-engines
that could hurt chessbase' business.

harvey williamson cannot be part of a panel of secretary that
SOLVES the rybka issue.

[Jeremy Bernstein]

it remains the problem that harvey is in a situation where he is stretched between the interest of the company he works for (he is a member of the hiarcs team, a moderator in the hiarcs forum,
a sysop on chessbase playserver) and the decisions of the panel. its completely unimportant if he is in the panel or in the secretary. he is not objective. he is pro hiarcs. and that makes it impossible to do a job good, or the way one would expect somebody to do such a job.

he should resign this function.
his rigid method and how he judged about the "cloning" issue makes him a person not capable
for such a job.

...

harvey williamson cannot be part of a panel of secretary that
SOLVES the rybka issue.

Well, think of it this way. Harvey represents the interests of commercial, closed-source computer chess. Bob represents the interests of academic, open-source computer chess. Mark Lefler is the author of NOW, founder of the chess programming wiki, not sure where he fits in, although it's worth noting that he was on Team Anand (which HIARCS supported). Not sure if that means that he works for HIARCS, or if he was doing other stuff.

I have accused Harvey of lots of things in the past, and we mostly don't see eye to eye, but I don't think he's anti-anything-specific except for software piracy, with a special place in the hell of his heart for chess engine clones. Of course he's pro-HIARCS and pro-commercial. But if you follow his emails and actions, you'll see that he's been consistent (also shrill, obtuse, blind, unjust, etc.) in his quixotic battle against what he perceives* to be the products of software piracy.

* Harvey doesn't have the technical chops, as far as I can tell, to actually make this determination himself, which is why his actions and words in the past have been so infuriating to those who actually go to the trouble to think about these issues, attempt to accumulate evidence, before starting name-calling, instituting bans, etc.

Every person with an engine that competes in tournaments stands to gain from Vas' alleged wrongdoing. I don't think there's a single person appropriate for the job that doesn't stand to gain somehow.

As long as Bob and Mark provide a foil to Harvey (and Harvey and Mark to Bob, and Bob and Harvey to Mark), he's probably no worse than any of the other choices. The biggest issue with him has been his ability to institute policy with no checks or oversight in the past, demonstrating his personal leaning toward dictatorial behavior. In this case, he has 2 equal partners who will (hopefully) ensure that he doesn't get out of hand.

Or would you prefer Conkie?

Jeremy

[BB+]

There are various names I might suggest if someone needs to step down, such as Gerd Isenberg (even though he signed the letter), or Remi Coulom (I can't imagine his being the Programmers Representative on the ICGA Board would be considered "conflicting", but I also can't imagine someone failing to bring it up ).

[BB+]

Indeed. My only reason for suggesting that an analysis of HIARCS or Shredder would be a good idea is to create some sort of "control" situation (I don't believe that HIARCS or Shredder are guilty of any code copying, from Fruit or anywhere else). Mostly so that "the public" could understand why this number and type of identical features is conspicuous, unusual and a reasonable basis for calling a piece of software a "clone" or "derivative".

I guess my opinion is that the "control" situation is already understood by most programmers.

Here is my "probabilisitic" analysis (YMMV): consider a post-Fruit 2.1 world in which ideas are free to take, but code is not. What the probability (or an upper bound for it) that two random programs would:
*) Have the same 20+ evaluation features: knight mobility, bishop mobility, trapped bishops, blocked bishops, halving opposite bishops in identical situations, rook mobility, rook open files, rook semi-open files, kings on/adjacent to such, 7th rank rook bonus under specific conditions, blocked rooks, queen mobility, 7th rank queen bonus (conditions as before), king safety by attacking squares next to king, doubled pawns, isolated pawns open file, isolated pawns closed file, backward pawns open file, backward pawns closed file, candidate pawns with 26-77-154-256 scaling, passed pawns with no self-blocker, no opp-blocker, free to advance (again same scaling) -- with only a few differences like nonlinear scaling in Rybka and also a side-to-move bonus there. [Furthermore, and the mobility was vanilla the evaluation was linear in it].

Certainly mobility was a big Fruit "idea", but I can't put the odds of this match at more than 1 in 10000.

*) Have the similar operations in root search (see Table 1), with the match of 7 common elements (some common to others, others not so much), in order.

Even with recognising that some of the Fruit parts are unnecessary and omitted in Rybka, I would put the odds at 1 in 100 or more.

*) Have the same relative PST scalings, for all 8 arrays of interest -- to be weighed against the slight differences in central pawns and the existence of zero-values in one program or the other (such as for pawns in the endgames). [Here I am accepting the Fruit "idea" of PST as legitimate to re-use, and looking only at specific numerical details].

Unless these arrays appeared elsewhere, I can't imagine this is more likely than 1 in 1000, likely more.

So I'm already at 1 in a billion, and I think that is even low by some factor. Adding in the other elements (like iterative deepening code) would increase this even more. As I say, this is seat-of-the-pants guesstimates, and YMMV (and I would interested if you came up with significantly different numbers). In a world w/o Fruit ideas, the numbers would be even larger.

How many post-Fruit engines have there been? Should we expect such a 1-in-a-billion (or more) match by now?

[BB+]

A philosophical statement from Aug 2004: http://www.stmintz.com/ccc/index.php?id=383344

Copying code, some might argue, is "smart" thing to do, because you benefit from the sweat of someone else, and still get credit. Society calls it plagiarism. However, to copy the bug, is "stupid".

[BB+]

Thanks for the data from Fruit history. I also got an email from Fabien about it. I will try to summarise briefly.

True, but ICGA aside, I'll point out that Rajlich didn't attempt to profit from Rybka until it had a significant Elo lead.

Fabien rejects the premise of this comparison, as he never prioritised the maximisation of Elo. For instance, after the WCCC in August, he was more involved with implementing things of "value to customers" (such as MultiPV, and robust UCI parsing to work with cantankerous GUIs). There was also work on a book generator and contempt for the WCCC itself. Some later work on hash-table management in analysis and cosmetic work on "good move" indicators was also done, but he's not sure that made the 2.2.x versions.

I guess one might say that, from Fabien's point of view (and this is my impression, not his words), that Rajlich took the Fruit 2.1 code and maximised Elo (some of this indeed via novel methods) and sold it, while he himself chose to work at least part of the time on other features -- so the argument that Rajlich had a significant Elo lead is a bit apples-to-oranges (or to pears, if you prefer).

[BB+]

[Fruit 2.1 and Rybka 1.0 Beta] Have the same 20+ evaluation features: knight mobility [...]

For comparison, early Rybkas used a decidedly different paradigm:

My new idea is this: Why not compute two different mobility scores for each side? It seems more flexible to compute a separate "offensive mobility" and "defensive mobility" for every piece. One could then use other components of the eval to decide upon the weightings for offensive vs. defensive mobility for each side.

I do something slightly similar. I score each piece according to how close it is to the enemy king, and to its own king. These scores are fed into a function which also takes into account the material on the board, holes in the pawn structure, and the difference in "attacking force". (For example if there are two pieces vs rook and pawn, the two pieces are a greater attacking force.)

[BB+]

In addition, we think the ICGA should in future insist that all authors of entries to ICGA events must submit to the ICGA the same executable(s), that is taking part in the ICGA event, where they can be stored for future analysis of potential derivative claims should they arise.

One problem with such storage is that systems do get hacked. Still, you'd then have to reverse-engineer the lot. [Ok, maybe just putting them all on a pirate site would be enough to black-eye the ICGA].

[Harvey Williamson]

his interest is not the truth, it is to fit the interest of his
own, of chessbase or hiarcs company.

the emails he has send to me when he banned me showed the
signature of hiarcs company.
so he is not a private person separated from mark uniacke.
he works for the company. he bans people for hiarcs company.
he supported rajlich and
the chessbase business of selling rybka and banning all
people from playchess server who used "clone"-engines
that could hurt chessbase' business.

harvey williamson cannot be part of a panel of secretary that
SOLVES the rybka issue.

One last time I am not an employee of Hiarcs or playchess. I have never been paid 1 penny by either.

[benstoker]

That's why there are THREE panelists. Did you know that committees in Congress are comprised of not just Republicans, but also Democrats? Isn't that amazing.

consider they take away the rybka championship titles.
and the 2nd programs get the title.

hiarcs will get a title. and harvey williamson is in the panel.
he is a member of the hiarcs team.

so this gives an interest conflict.

would the american system allow that the mother of a murderer is in the jury ?

Das stimmt. Aber, the "judicial" process is divided. There's a "trier of fact" and then the judge - or, here, a panel or body that will give an opinion.

Concern about the "trier of fact's" observation bias is important. Example: eyewitness to a car accident. The bias (race, status, etc.) of a witness can affect how that witness reports his/her observation of the accident. If there are no other witnesses to compare statements, then the bias of the sole observer may render that witness statement unreliable.

There is less concern over "opinion bias". Opinion is bias. It is the interpretation of the facts. See my remarks above on this. Experts will urge others to accept their openly biased opinions, and, in order to convince others, their argument must be good and address the criticism of other biased experts with contrary opinions.

But, back to the facts issue, and "observation bias". The facts involved here are not limited to a "sole observer". We don't have to rely on the statement of the only witness of the car accident, who happens to be a drunk, dope head, white supremacist, for example. We don't have to be concerned about a bad cop who plants a joint under a car seat to frame a black man -- that's a 'sole observer' problem. Wherever the observation of the facts is limited in time, space, and to a sole observer, you got problems.

Here, there is no sole observer. The facts are open for observation by everyone and their dogs. Exhibit A: Fruit source code. Exhibit B: Rybka object code. Exhibit C: Crafty source code, or other source code. All who care to do so can at any time and as many times as they wish observe the facts. It's really not any different than science. A scientist claims to have discovered cold fusion in a coke bottle. He publishes his "discovery" and describes his experiment. Science is objective. If there is any truth to what this scientist claims he observed, then his experiment must be repeatable by others observing the same facts. It doesn't even matter what the scientist's "observation bias" is; his factual claims are easily tested. [Exception: social "sciences"]

The excerpt below is from BB+'s report. Fadden produces a decompilation of Rybka object code. This is an observation of facts available to all. If Fadden is completely biased against Rybka, does it matter? It matters only if I know that this guy sees red when the wall is green. The point is that Fadden's factual assertions are not limited to a sole observer. I am free to observe the Rybka object code to confirm Fadden's observation.

Now, are you really worried that this panel/secretariat body will produce FALSE factual observations of the Rybka object code? That's not going to be the problem. All the evidence they gather will be observable by everyone! If they report "0x40702e: test %r15b,%r15b # r15 is "infinite" as Rybka disassembly, anyone can verify that.

The problem will be the interpretation and opinions about the facts. The problem will be settling on an opinion about the standards to apply to the facts. The definitions of "originality" and "clone"; how to distinguish permissible from impermissible source code copying/borrowing.

Code: Select all

The subsequent lines in Fadden's Rybka 1.0 Beta decompilation have:
// Rybka compares movetime with a double precision value: 0.0
if (movetime >= 0.0) {
time_limit_1 = 5 * movetime;
time_limit_2 = 1000 * movetime;
} else if (time > 0) {
time_max = time - 5000;
alloc = (time_max + inc * (movestogo - 1)) / movestogo;
if (alloc >= time_max) alloc = time_max;
time_limit_1 = alloc;
alloc = (time_max + inc * (movestogo - 1)) / 2;
if (alloc < time_limit_1) alloc = time_limit_1;
if (alloc > time_max) alloc = time_max;
time_limit_2 = alloc;
}