Stockfish: Our lawsuit against ChessBase

[syzygy]

This would appear to contradict the legal expert I linked ages ago (maybe not in this thread) who wrote that if someone commissions a work of art, then the copyright belongs to the person that commissioned it, not the creator of it - and this would very likely apply to someone generating art using a computer.

There are two questions:
- does a copyright exist?
- who owns it?

For a copyright to exist, there must be an expression of a human's free creativity. To be precise: the expression must be creative (it is not sufficient that the idea that is expressed is creative -- ideas cannot be copyrighted). The creativity could be creativity of the person doing the actual work, but it could also be the creativity of someone who gave instructons to that person on how to create the work.

Usually the human whose creativity is expressed will be the author and copyright holder, but copyrights can be transferred and national laws may determine e.g. that if the human is an employee and was carrying out his normal duties, then the employer is the author.

If the only expressed creativity came from an AI or a monkey, then there is no copyright to begin with.

A photographer using a camera to do the actual work will often have a copyright because he made some creative choices which to some extent controlled the expression shown in the photograph. If he tries to photograph the Mona Lisa as accurately as possible, he will not have a copyright because there is no expression of free creativity on the photographer's part (all choices he made are functional, namely to optimise the quality of the reproduction).

An average person who takes random pictures with his mobile phone might, strictly speaking, not have a copyright on his pictures because he is not using the camera as a tool to express his creativity but is just randomly toying around. (In practice he would probably be able to claim copyright in court.)

If an AI image generator gives the user sufficient control so that the user can use the image generator as a tool to express his own creativity, then the generated image would be copyrighted.

An AI image generator that takes a textual prompt and generates an image that expresses the idea encoded in the textual prompt would not produce a copyrighted image. Although the textual prompt itself might be copyrighted, the idea that it expresses is not. The textual prompt is copyrighted only if the user creatively expressed the idea in text. The image might creatively express the same idea, but that expression is the result of the AI's creativity, not the user's.

[Graham Banks]

This would appear to contradict the legal expert I linked ages ago (maybe not in this thread) who wrote that if someone commissions a work of art, then the copyright belongs to the person that commissioned it, not the creator of it - and this would very likely apply to someone generating art using a computer.

There are two questions:
- does a copyright exist?
- who owns it?

For a copyright to exist, there must be an expression of a human's free creativity. To be precise: the expression must be creative (it is not sufficient that the idea that is expressed is creative -- ideas cannot be copyrighted). The creativity could be creativity of the person doing the actual work, but it could also be the creativity of someone who gave instructons to that person on how to create the work.

Usually the human whose creativity is expressed will be the author and copyright holder, but copyrights can be transferred and national laws may determine e.g. that if the human is an employee and was carrying out his normal duties, then the employer is the author.

If the only expressed creativity came from an AI or a monkey, then there is no copyright to begin with.

A photographer using a camera to do the actual work will often have a copyright because he made some creative choices which to some extent controlled the expression shown in the photograph. If he tries to photograph the Mona Lisa as accurately as possible, he will not have a copyright because there is no expression of free creativity on the photographer's part (all choices he made are functional, namely to optimise the quality of the reproduction).

An average person who takes random pictures with his mobile phone might, strictly speaking, not have a copyright on his pictures because he is not using the camera as a tool to express his creativity but is just randomly toying around. (In practice he would probably be able to claim copyright in court.)

If an AI image generator gives the user sufficient control so that the user can use the image generator as a tool to express his own creativity, then the generated image would be copyrighted.

An AI image generator that takes a textual prompt and generates an image that expresses the idea encoded in the textual prompt would not produce a copyrighted image. Although the textual prompt itself might be copyrighted, the idea that it expresses is not. The textual prompt is copyrighted only if the user creatively expressed the idea in text. The image might creatively express the same idea, but that expression is the result of the AI's creativity, not the user's.

So is there an expression of human creativity in a neural net that is trained by the computer?
Just asking because I don't really understand the process required to come up with a neural net.

[Graham Banks]

This would appear to contradict the legal expert I linked ages ago (maybe not in this thread) who wrote that if someone commissions a work of art, then the copyright belongs to the person that commissioned it, not the creator of it - and this would very likely apply to someone generating art using a computer.

There are two questions:
- does a copyright exist?
- who owns it?

For a copyright to exist, there must be an expression of a human's free creativity. To be precise: the expression must be creative (it is not sufficient that the idea that is expressed is creative -- ideas cannot be copyrighted). The creativity could be creativity of the person doing the actual work, but it could also be the creativity of someone who gave instructons to that person on how to create the work.

Usually the human whose creativity is expressed will be the author and copyright holder, but copyrights can be transferred and national laws may determine e.g. that if the human is an employee and was carrying out his normal duties, then the employer is the author.

If the only expressed creativity came from an AI or a monkey, then there is no copyright to begin with.

A photographer using a camera to do the actual work will often have a copyright because he made some creative choices which to some extent controlled the expression shown in the photograph. If he tries to photograph the Mona Lisa as accurately as possible, he will not have a copyright because there is no expression of free creativity on the photographer's part (all choices he made are functional, namely to optimise the quality of the reproduction).

An average person who takes random pictures with his mobile phone might, strictly speaking, not have a copyright on his pictures because he is not using the camera as a tool to express his creativity but is just randomly toying around. (In practice he would probably be able to claim copyright in court.)

If an AI image generator gives the user sufficient control so that the user can use the image generator as a tool to express his own creativity, then the generated image would be copyrighted.

An AI image generator that takes a textual prompt and generates an image that expresses the idea encoded in the textual prompt would not produce a copyrighted image. Although the textual prompt itself might be copyrighted, the idea that it expresses is not. The textual prompt is copyrighted only if the user creatively expressed the idea in text. The image might creatively express the same idea, but that expression is the result of the AI's creativity, not the user's.

So is there an expression of human creativity in a neural net that is trained by the computer?
Just asking because I don't really understand the process required to come up with a neural net.

I'm guessing that the human creativity element comes in making the trainer and also if you're training the net purely on games played by your engine.

[syzygy]

So is there an expression of human creativity in a neural net that is trained by the computer?
Just asking because I don't really understand the process required to come up with a neural net.

A neural net is basically a mathematical function that, given an input, produces an output. The function is defined by a set of parameters called weights.

A neural net is typically trained by giving it pairs of (input, desired output). The training algorithm will gradually adjust the weights of the network such that the outputs it produces for the training inputs get closer to the desired outputs.

The hope is that the trained network will produce sensible outputs also on inputs on which it has not been trained. The idea is that the neural network has learned to recognise and evaluate certain patterns, and that this pattern recognition ability will also be valid for inputs it has never seen.

So with NNUE the inputs are chess positions and the outputs are evaluations. The network and the encoding of the input positions are designed in such a way that the neural network can be evaluated efficiently. Surprisingly (at least to me), this can be made to work very well.


The first problem with copyright on a neural net is that a neural net is a mathematical function, and mathematical functions are not things on which there can be copyright.
One could counterargue that a neural net is (also) a kind of computer program, and that computer programs are copyrightable. However, the copyright on a computer program protects the expression of the program (e.g. in source code), not the functionality. The weights of a neural network cannot be separated from the unprotectable function, as they are the function.

The second problem is that there will typically be no "free creativity" in training the net, at least not creativity that leaks through into the neural net. There might be cleverness involved in selecting the inputs and training parameters, but this cleverness is aimed at optimising the quality of the net, so it is functional.

To overcome the first problem, you could look at what the neural net produces instead of the neural net itself. If the neural network creates text, sound or images, then the output might be susceptible of copyright. But then you will probably run into the second problem, at least in so far as you are trying to argue that the creator of the neural network has a copyright on its output. (There are better chances for the person who created the particular input that led the neural network to create a particular output. This is what is discussed above.)


This is just my informed opinion. I cannot guarantee that courts will agree with me (and there will likely be some contradicting decisions until the dust settles on this topic).

[syzygy]

I'm guessing that the human creativity element comes in making the trainer and also if you're training the net purely on games played by your engine.

So my view is that NNUE nets (assuming you are mainly interested in those) are by definition free of copyright.

[noobpwnftw]

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.

[noobpwnftw]

Another interesting fact is that in some countries one cannot "give up" copyrights, therefore licenses like CC0 are created to grant as much freedom as it is legally possible. So in principle people would have to hire a lawyer and file a suit to argue and prove that they don't have copyrights over something for whatever reasons, lawyers do indeed know how to create work for themselves.

[Rebel]

Quality of the data set to create a NN matters a lot. And creating quality is a highly creative process and is still in childhood state. What does the learner like, what does the learner dislike, it's a learning process in itself. With so much invested energy I would hate to see if the nets I create (and I assume the same counts for others) do not fall under copyright.

[towforce]

...lawyers do indeed know how to create work for themselves.

This!

syzygy's line or argument is very good. I would, though, use the following line of argument to say that NNs should be copyrightable:

* NN is software

* software can be copyrighted

* it seems likely to me that software can STILL be copyrighted even if it was produced with the assistance of GitHub Co-pilot or Chat GPT. If in doubt, you could make tiny changes to the code (or NN weight set), and it would then have human input

* decisions about the architecture of the NN were probably made by a human

Another possible problem: if two different users both, independently, asked their AI system to "draw a picture of a cat sleeping by the fire", they might well both get the same image back. You would then have two people who both had a valid claim to the copyright over that image!

Off topic to this discussion: it is very clear to me that, in the modern world, both copyrights and patents should be shortened. As a quick example, you get cases where somebody has a patent over a life-saving drug or an invention that could lower carbon emissions, but the world cannot use them because the patent holder can overcharge for such a long period of time.

Off topic completely, but interesting to know: a recent study has shown that code written by Co-pilot is not as good as code written purely by humans. However, this might not always be the case in the future, and I haven't personally looked at the study's methodology. There do seem to be issues with code written by Chat GPT right now, but again I haven't personally tried generating code with it.

[smatovic]

Well, IMO something is missing, artists can not protect their work from being sucked into NN molochs, and people investing work/time/energy to create neural networks can not claim Copyright on them? Obviously both cases were not considered in the good old days...

--
Srdja