GitHub Copilot and open source laundering
drewdevault.com
GitHub Copilot and open source laundering
1–10 of 29 posts
Re: GitHub Copilot and open source laundering
#2An interesting question would be over fair use. Generally, "transformative" works are covered under fair use. I can certainly see the argument that DL output is transformative in nature.
Not a lawyer take my comments with a grain of salt.
Re: GitHub Copilot and open source laundering
#3An interesting question would be over fair use. Generally, "transformative" works are covered under fair use. I can certainly see the argument that DL output is transformative in nature. Not a lawyer take my comments with a grain of salt.
Is plagiarism legal if I change the variable names?
Re: GitHub Copilot and open source laundering
#4An interesting question would be over fair use. Generally, "transformative" works are covered under fair use. I can certainly see the argument that DL output is transformative in nature. Not a lawyer take my comments with a grain of salt.
Copilot can certainly produce an output which appears "transformative". It can also produce verbatim copies of licensed works. The output exists along a spectrum from probably transformative to clearly non-transformative, and it offers no tools for the user to establish the provenance of its output. Fair use is evaluated on a case-by-case basis, and this would require that each program produced by Copilot is examined to determine the depth of the transformation.
It lives by this vagueness and is abused as a tool for laundering the source code, to wash off the license terms for free software under a suitably complex and inscrutable system that establishing this provenance is not realistic, even if the output would fall on the "non-transformative" end of the scale.
I can kind of understand the argument that GitHub itself is, perhaps, not liable for this (setting aside the fact that they use Copilot in their own works, and GitHub's own software would have to be examined under this lens), but it is certainly irresponsible of them and disrespectful of the community that it owes its success to, and it seems to me that its users would definitely be at a higher legal risk.
Re: GitHub Copilot and open source laundering
#5I find the proposed solutions in the article quite sensible and fair:
> Allow GitHub users and repositories to opt-out of being incorporated into the model. Better, allow them to opt-in. Do not tie this flag into unrelated projects like Software Heritage and the Internet Archive.
> Track the software licenses which are incorporated into the model and inform users of their obligations with respect to those licenses.
> Remove copyleft code from the model entirely, unless you want to make the model and its support code free software as well.
About the last point: Many of the larger corporations avoid copyleft code entirely or only use it in ways that avoids the viral nature of them. So I have to assume that Copilot would also be avoided for the same reasons by the same corporations until they can opt out.
Free software is already being exploited and violated in various ways by various actors and Copilot seems to be a tool that makes this even easier. I think it will be interesting to see where this issue goes.
Re: GitHub Copilot and open source laundering
#6I find the proposed solutions in the article quite sensible and fair: > Allow GitHub users and repositories to opt-out of being incorporated into the model. Better, allow them to opt-in. Do not tie this flag into unrelated projects like Software Heritage and the Internet Archive. > Track the software licenses which are incorporated into the model and inform users of their obligations with respect to those licenses. >…
I think opting out would be quite hard. Due to the nature of copyleft your code can end up in a lot of random projects (which are actually complying with the license) and those projects might not opt out.
Re: GitHub Copilot and open source laundering
#7An interesting question would be over fair use. Generally, "transformative" works are covered under fair use. I can certainly see the argument that DL output is transformative in nature. Not a lawyer take my comments with a grain of salt.
Many jurisdictions don't have a doctrine like fair use.
Re: GitHub Copilot and open source laundering
#8Just recently I was trying to implement a feature using an extremely complicated and obscure API (some Chrome internal C++ stuff) and the only place I found a working example was inside a GitHub project which was GPL.
Obviously if I took the code and modified it to suit my purposes I would've run afoul of the GPL, but if Copilot suggests the same snippet to me it'd be fine. So there's obviously an issue here.
Re: GitHub Copilot and open source laundering
#9Drew, that is a great post.
I have two comments and a question.
First, my comment is that GitHub claims the right to use any public code on the Internet, not just GitHub. This means that they claim the right to mine code on Sourcehut, or in my case, on my personal Gitea instance. I'm not sure your suggestion 1 covers that. I think it should.
Second, users could fork code from Sourcehut onto GitHub, and then even if GitHub only uses code on GitHub, they could then use that code. This is, of course, a consequence of using DVCS's and can't be worked around. Your first suggestion should also address this, in my opinion, by allowing original authors to opt-out.
Third, my question: you've surely seen the licenses I have been developing against Copilot ([1]) since you've emailed me about them. Do you think the way I've approached the anti-Copilot clause is still free and open? For reference, this is the clause in all of my licenses:
> In this license, the phrase “this software” includes the source code licensed under this license. It also includes the output of any execution of any algorithm that uses all or part of this software as all or part of its input, but only if the output itself is software.
(The rest of the license then says it applies to "this software.")
Re: GitHub Copilot and open source laundering
#10Just recently I was trying to implement a feature using an extremely complicated and obscure API (some Chrome internal C++ stuff) and the only place I found a working example was inside a GitHub project which was GPL. Obviously if I took the code and modified it to suit my purposes I would've run afoul of the GPL, but if Copilot suggests the same snippet to me it'd be fine. So there's obviously an issue here.
I think Copilot definitely is crossing some line that hasn't been defined yet, but your specific example has me thinking.
If I imagine some seasoned art expert standing next to me and giving me hints and ideas while I'm paining and he's using existing art as his examples, am I stealing the art? While Copilot recommends bits of code, it's not copying the entire project is it?