I find this article pretty frustrating. I'm not going to do a point-by-point, because the response to every point is the same: Copilot is copying things. The mechanism by which it does so is...
I find this article pretty frustrating. I'm not going to do a point-by-point, because the response to every point is the same:
Copilot is copying things.
The mechanism by which it does so is completely irrelevant. If somebody copies my GPL code, I don't give a rat's ass how (and nor do the courts)—computer clipboard, file copy, handwritten notebook, or big pile of linear algebra, it doesn't matter at all, it is the fact of the copy that matters. And if they publish whatever project they copied my GPL code into with Copilot, it is my legal right to insist that they follow the terms of the GPL—i.e., they license their project under the same terms.
What happens if Copilot copies code under mutually incompatible licenses? I don't know the details of the licensing of their training data, but obviously, neither do they, so it's a safe bet there is a wide variety represented.
Much more broadly speaking, this debacle constitutes yet more proof (as if we needed more) that we don't understand how machine learning algorithms work well enough to actually use them (though of course the fact that they regurgitate large portions of their input data with minimal prompting is well-known and the Copilot developers should have thought of).
The point of the article, if I read it right, is that the copying that Copilot is doing neither is infringing, nor should be construed as infringing, on copyright, and therefore also on any...
The point of the article, if I read it right, is that the copying that Copilot is doing neither is infringing, nor should be construed as infringing, on copyright, and therefore also on any license supported by that right. So, no, it is not your right to insist that the output conforms to your license. (I'm not personally knowledgable about copyright, I'm just reiterating that point from the article.)
The article doesn't deny that Copilot (or machine learning algorithms in general) are copying things, so I find your saying that that a pretty weak rebuttal.
I have yet to read Reda’s article, and don’t think I’ll go through the pain of reading the actual paper, but am a lawyer and read Guadamuz’ article on the same topic. The crux, as you point it, is...
I have yet to read Reda’s article, and don’t think I’ll go through the pain of reading the actual paper, but am a lawyer and read Guadamuz’ article on the same topic.
The crux, as you point it, is indeed in whether if there was no license the use would constitute a copyright infringement or not. A license (by definition) is just specific rights granted to you by the licensor that go beyond what the law gives you.
(Then there is the question how much non-license, contractual obligations a FOSS license can carry, but that’s another topic and one that causes heated debates between US and EU lawyers.)
Julia is a former MEP and an advocate for digital rights issues, copyright among other things. She has been saying some quite contrary things to the prevailing opinion on Copilot and the copyright...
Julia is a former MEP and an advocate for digital rights issues, copyright among other things. She has been saying some quite contrary things to the prevailing opinion on Copilot and the copyright status of its outputs which I thought were interesting, so I was pleased to see she wrote an article expounding on that.
(Also, the previous post on Copilot was in ~comp but this is a bit broader scope so I thought ~tech would suit better.)
I find this article pretty frustrating. I'm not going to do a point-by-point, because the response to every point is the same:
Copilot is copying things.
The mechanism by which it does so is completely irrelevant. If somebody copies my GPL code, I don't give a rat's ass how (and nor do the courts)—computer clipboard, file copy, handwritten notebook, or big pile of linear algebra, it doesn't matter at all, it is the fact of the copy that matters. And if they publish whatever project they copied my GPL code into with Copilot, it is my legal right to insist that they follow the terms of the GPL—i.e., they license their project under the same terms.
What happens if Copilot copies code under mutually incompatible licenses? I don't know the details of the licensing of their training data, but obviously, neither do they, so it's a safe bet there is a wide variety represented.
Much more broadly speaking, this debacle constitutes yet more proof (as if we needed more) that we don't understand how machine learning algorithms work well enough to actually use them (though of course the fact that they regurgitate large portions of their input data with minimal prompting is well-known and the Copilot developers should have thought of).
The point of the article, if I read it right, is that the copying that Copilot is doing neither is infringing, nor should be construed as infringing, on copyright, and therefore also on any license supported by that right. So, no, it is not your right to insist that the output conforms to your license. (I'm not personally knowledgable about copyright, I'm just reiterating that point from the article.)
The article doesn't deny that Copilot (or machine learning algorithms in general) are copying things, so I find your saying that that a pretty weak rebuttal.
I have yet to read Reda’s article, and don’t think I’ll go through the pain of reading the actual paper, but am a lawyer and read Guadamuz’ article on the same topic.
The crux, as you point it, is indeed in whether if there was no license the use would constitute a copyright infringement or not. A license (by definition) is just specific rights granted to you by the licensor that go beyond what the law gives you.
(Then there is the question how much non-license, contractual obligations a FOSS license can carry, but that’s another topic and one that causes heated debates between US and EU lawyers.)
Julia is a former MEP and an advocate for digital rights issues, copyright among other things. She has been saying some quite contrary things to the prevailing opinion on Copilot and the copyright status of its outputs which I thought were interesting, so I was pleased to see she wrote an article expounding on that.
(Also, the previous post on Copilot was in ~comp but this is a bit broader scope so I thought ~tech would suit better.)
Another interesting read on this take, by Andres Guadamuz, an actual lawyer