Ownership of AI-Generated Code Hotly Disputed

If an AI being inspired by GPL code that happened to be in its training data because someone other than the creator stuck it in a repo on GitHub is just fine, and if some code produced as a result is practically identical that is fine too, and the resulting code is not considered GPL any more, then the GPL and licences like it are worthless.

The only two designations that mean anything at that point are public domain and commercial. “free” as in “Free” rather than public domain means the same as public domain.

So there is no point releasing free (other than public domain) code, discouraging the act. If I want to control how my code is used at all protected commercial release becomes the only option.

This of course suits the commercial interest behind copilot just fine and dandy…

---

But, if commercial code ended up in the training set the same should apply because in terms of giving the right to use code licences like the GPL and commercial licences are no different: the licence gives the right to use the code. If passing it through an AI gives that right, bypassing the licence, for one case than it should for the other too. I wonder if MS would be happy for copilot's own code to be in the training set and for me to produce and sell something based on the output of an AI trained with their code?!

---

I think the AI systems like copilot should be considered the same as us wet-ware naturally formed intelligence systems in that respect: if it produces something based on code under a particular licence then that something should be subject to the terms of the licence. Ignorance of the licence is no excuse. If the AI can not be made aware of the correct licence and attribution for the code, so it can include that with its suggestions based on it, then that code should not be in the training set.

For decades MS complained about open source code potentially creating this very situation, just with only non-artificial intelligences in the mix, now they are hoping no one can call them on that because it is convenient for them to ignore the issue.

Here's my take from an audio DSP viewpoint.

A composer listens to 10,000 hours of music. One day she writes an original piece based on the annealed parameters (temporal, spectral, intensity, pitch, sequence...) of a million artists. However it sounds like another specific artist who sues.

It is not a cover, a remix, homage, or even forged in the genre... it's just accidentally a bit too like it. (compare: Banana Splits vs. Bob Marley - and - Huey Lewis vs. Ray Parker Jr. - which completely misses the real impact of "Pop Muzik" by M)

The question is, was she exposed to influencing materials incorporated without intent or was it plagiarism (intent to reproduce a derivative etc) ?

By contrast I can take a piece of music, break it down by analysis into melodies, chords, timings, and use FFT to extract the precise spectrums of instruments, feed those to a resynthesis engine that finds new synthesiser parameters to create an exact sound-alike and then deliberately recreate a piece "In the near style of artist X". With a little musical processing I can change the key, inversions, re-template the rhythm to a new swing... always pushing the derived piece into new territory until eventually it's barely recognisable.

Nonetheless, in the second case I have clearly intended to steal someone's idea and "make it my own" by automatable transformation.

In the former case it seems to be a "genuine labour". (whatever that means in 2023)

The genuine artist intends to make something through intellectual labour.

Maybe that's the real question. I mean, about "labour". If the cost of the labour tends to zero, does it really matter?

The easy thing to do is to compare outputs to inputs. This isn't technically hard (i.e. cosine similarity) but it is computationally difficult (i.e. cosine similarity of output to entire input). This would give us some weightings that show similarity. But this doesn't really tell us attribution, rather more a correlation. These are statistical models so there's reason to believe that this is okay.

Then there's inversion. This is processing data backwards through the model. This has different complexities compared to what type of model you're working with. GANs aren't great, diffusions are okay, normalizing flows are trivial (but good luck generating good images from NFs). If we can invert the model it is much easier to investigate and probe for contribution by looking at the distance of the latent generative variable to the location of the latent trained data. Basically you're looking at how different information is contributing to the overall output. This can also be done at every level in the network. Obviously this gets both technically challenging as well as computationally.

Another method would be using dataset reconstruction (this is outside my wheelhouse fwiw). This is where you try to recover the dataset from the final trained network. This too is complicated but there's plenty of papers showing progress in this space (lots of interest from privacy groups).

(TLDR-ish) There's other methods too. But basically what is being said is that there are ways to denote what and how much the training data contributed to the output of the model wherein we can then measure how similar the output is to the inputs (i.e. copying).

ZipCar, et. al.

> And your house?

AirBnB

> Let's not quibble about the particulars of me ever giving them back.

Symmetry.