Stockfish: Our lawsuit against ChessBase

[Albert Silver]

No, neural nets are not just weights, there is not a "the neural net" or a "the NNUE" that you could take for granted. The designs, the tooling, the driving code and the precise set of numbers that made it to work are all creative work therefore subject to copyrights. And it is implementation specific and not a mathematical theorem or ground truth, there are endless instances of them offering specific characteristics and/or performance.

In this forum, there are people who would like to argue that search function is also "solved", there exists only one right way to do it and therefore it is free for all to copy paste.

Claiming NNUE nets are not copyrightable IMO is similar to claiming Blues music are not copyrightable, ymmv.

Thank you

You are welcome. With the lawsuit concluded, I would assume that you would now make your derivative work GPL compatible?

ChessBase took down your product, which I'd rather prefer they didn't.

The only part of the initial release that wasn't compatible, a misunderstanding on my part, 100% mea culpa, was embedding the NN weights into the binary. The weights were meant to be copyrighted and the search, available on Github, was there for the taking. Still is.

[noobpwnftw]

It is probably wise to pay more attention to what's inside the Makefile when you copy-paste. Now that it is also possible to consult the FOSS compliance officer, or even have it externally certified, 200% foolproof.

[MikeB]

No, neural nets are not just weights, there is not a "the neural net" or a "the NNUE" that you could take for granted. The designs, the tooling, the driving code and the precise set of numbers that made it to work are all creative work therefore subject to copyrights. And it is implementation specific and not a mathematical theorem or ground truth, there are endless instances of them offering specific characteristics and/or performance.

In this forum, there are people who would like to argue that search function is also "solved", there exists only one right way to do it and therefore it is free for all to copy paste.

Claiming NNUE nets are not copyrightable IMO is similar to claiming Blues music are not copyrightable, ymmv.

Thank you

You are welcome. With the lawsuit concluded, I would assume that you would now make your derivative work GPL compatible?

ChessBase took down your product, which I'd rather prefer they didn't.

The only part of the initial release that wasn't compatible, a misunderstanding on my part, 100% mea culpa, was embedding the NN weights into the binary. The weights were meant to be copyrighted and the search, available on Github, was there for the taking. Still is.

At the end of the day, whether legal or illegal, it did appear to be attempt by Chessbase to make money off of the backs who had volunteered their time to the Stockfish project. It left a sour taste in the mouth. I'm glad it was resolved and put to bed. I really don't blame you, I blame Chessbase. They should have known better. In hindsight, do you wish you had just released the net to the public and say, 'Check it out, it plays interesting chess"?

[Modern Times]

At the end of the day, whether legal or illegal, it did appear to be attempt by Chessbase to make money off of the backs who had volunteered their time to the Stockfish project. It left a sour taste in the mouth.

Well, the GPL does allow exactly that, for people or companies to commercialise and make money from open-source projects, as long as they comply with the terms of the GPL. So it shouldn't leave a sour taste, it is anticipated, expected and catered for.

[dkappe]

At the end of the day, whether legal or illegal, it did appear to be attempt by Chessbase to make money off of the backs who had volunteered their time to the Stockfish project. It left a sour taste in the mouth.

Well, the GPL does allow exactly that, for people or companies to commercialise and make money from open-source projects, as long as they comply with the terms of the GPL. So it shouldn't leave a sour taste, it is anticipated, expected and catered for.

Oof. The same misunderstanding and clarification of the GPL over and over. I know you are wasting your time.

[noobpwnftw]

What leaves a sour taste is that a bunch of hippies who asked for next to nothing are pioneering the computer chess ecosystem for more than half a decade, while the good ol commercial go to corporate, did nothing more than making random stunts to justify its obsolete business model.

[expositor]

This reminds me of that classic article What Colour are your bits? (https://ansuz.sooke.bc.ca/entry/23).

In general, one shouldn't put too much stock in a programmer's interpretation of the law (myself included!) because the world of human affairs is not particularly amenable to the ways that programmers like to reason about things.

That said, for US copyright law, people may be interested in reading about the Works subject to copyright law and Authorship, ownership, and work for hire sections of the Wiki article, the ruling on Feist Publications v Rural Telephone Service, and chapter 300 of the Compendium of U.S. Copyright Office Practices. I'm sure that most people already know all of this, but the links might be mildly convenient for anyone learning about the topic, like myself.

[syzygy]

As programming languages getting to higher levels, sometimes you can prompt in natural languages and it compiles into a binary, now the binary is not copyrightable because the involvement of human creativity is not sufficient? Or it is a mathematical function too much?

Reminds me of some OISC machines where there is this one instruction that does it all, so all programs unless handwritten in that situation is not copyrightable or what?

Pure object code will normally be essentially indistinguishable from its functionality, so in principle it should not be covered by copyright. However, courts will almost certainly consider that the copyright on the source code extends to the object code generated from it. (I thought the EU software copyright directive was explicit on this, but I seem to be wrong on that.)

If a program was handwritten directly in machine code and contains no expresson (such as copyright text or music), I would indeed argue that it is not copyrighted.

[syzygy]

The NN weights are not copyrighted. Just numbers not encoding any expression.

This seems questionable to me. Conventional object code also is just an ordered collection of numbers. So by this logic binaries of computer programs are not subject to copyright, and all of the GPL is just bollocks.

Copyright is only on expression, but as I also replied to noobpwnftw, I suspect that courts would let the copyright on the source code extend to the object code generated from it. I.e. if you copy object code generated from the source code, you infringe the copyright on the source code. Indeed this was probably the legislator's intention; both the EU and the US enacted provisions stating that computer programs are proctected as literary works. (The EU directive makes clear that the copyright does not cover the idea and principles underlying the program. It is still only about expression.)

A text stored in an electronic medium is just a collection of numbers. Or a JPEG encoded movie.

This counterargument goes absolutely nowhere (and I must have refuted it about 10 times by now). The numbers are not protected, but the expression they encode is. The expression encoded in electronic text or in a JPEG is not a mere idea or functionality.

Decompress and recompress a JPEG and you'll get a very different collection of numbers, but it is still the same copyright.

An NN is just a particular type of computer, and the collection of weights represents the program it executes.

The NN is just functionality, no expression. Therefore no copyright.

NNs can be used to store images (and then recall a particular one by showing it only a part of it). In this case it seems obvious to me that the NN weights would not have been able to shed off the copyright of the original image they can be asked to display. While all that was done is training the net on the images that it has learned to reproduce.

If you manage to use NNs as a functional alternative to JPEG, then sure, an NN encoding a specific image will be covered by the copyright on the image. If you manage to use an NN to encode the source code of a computer program, then same.

[syzygy]

Copyright is only on expression

Neural networks can be compared to chip designs:
https://en.wikipedia.org/wiki/Integrate ... protection

Because of the functional nature of the mask geometry, the designs cannot be effectively protected under copyright law (except perhaps as decorative art).

I already wrote above that a particular graphical representation of a neural network (with creative choice for colors and layout etc.) could be protected. But the set of weights does not contain such choices.

A mask geometry is clearly a computer program. The 6502 can now be emulated directly from a photo of the chip.

Perhaps legislators will come up with a special kind of intellectual property protection for NNs, but today this does not exist.