Re: Re: AI-generated code uncopyrightable in the US

Anna Vyalkova <[email protected]> writes:
 
>> Have there been any discussions elsewhere in the Gentoo project 
>> about this, given that copyright is the basis for various types 
>> of 
>> licensing (e.g. permissive and copyleft licenses)? Are there 
>> other 
>> jurisdictions which also regard AI-generated code as 
>> uncopyrightable? Is the use of AI-generated code simply so 
>> widespread at this point (e.g. as claimed by 
>> https://codeberg.org/small-hack/open-slopware) that trying to 
>> sort 
>> out the situation re. copyright and licensing is well beyond 
>> the 
>> capacity of both Gentoo in general and of package maintainers 
>> in 
>> particular?
>
> The basis for copyright law in most countries is the Berne 
> Convention.  
> There's an agency within the United Nations, the World 
> Intellectual 
> Property Organization, administrating international intellectual 
> property treaties. Here's what it says on generative AI:
> https://www.wipo.int/publications/en/details.jsp?id=4713
>
> I think it answers some of your questions.

Thanks for sharing that.

On page 11, it says:

> It is unclear whether new content generated by AI tools such as 
> text,
> images, or other creative works, can be protected by IP rights, 
> and if
> so, who owns those rights. Even if AI output is not IP 
> protected,
> there may be contractual provisions that govern its use. The 
> existence
> and ownership of IP rights in generative AI outputs is
> unclear. Businesses and organizations should seek contractual 
> clarity
> over ownership and consider using generative AI only in cases 
> where IP
> ownership in the outputs is not crucial for their business 
> model.

On the same page, under the heading of "Risks", it says:

> The IP laws of most countries were written without considering
> generative AI, leading to uncertainty over whether there can be 
> IP in
> AI outputs and who would own any such rights. This may not be an 
> issue
> for some IP rights, like trademarks, but there is widespread 
> concern
> for copyright.
> 
> Recent patent applications, naming an AI system, “DABUS”, as an
> inventor, have consistently been rejected in countries that have
> issued judgments because no human inventor has been 
> identified. It is
> not yet clear whether generative AI can make inventions without 
> human
> inventors or whether such inventions are patentable.
>
> The US Copyright Office has issued guidance on registering works
> containing material generated by AI, indicating that a creative
> contribution from a human is required. Decisions by the Office 
> suggest
> that a user’s text prompt alone may not establish copyright, as 
> the
> prompt merely “influences” the output. However, the Beijing 
> Internet
> Court has recently decided that a user owns the copyright in an
> AI-generated image because he adjusted the prompts and 
> parameters so
> that the image reflected his aesthetic choice and 
> judgment. These
> varying interpretations of copyright for AI-generated works 
> introduce
> legal uncertainty regarding global recognition of copyright in
> generative AI outputs.
>
> A few countries (such as India, Ireland, New Zealand, South 
> Africa,
> and the United Kingdom) provide copyright protection for
> “computer-generated works” without human authors. Ukraine has
> introduced rights for “non- original objects” generated by 
> computer
> programs.

Then the "Mitigations" section says:

> Review the terms and conditions of generative AI tools to 
> understand
> who owns the IP (if any) in outputs.
>
> Explore ways to enhance control or rights over outputs by
> incorporating IP elements like brand names and logos or 
> involving
> human creativity in modifying or creating new versions of the 
> outputs.
>
> Document the role of humans in the invention or creation
> process. Where possible, establish an agreement about who owns 
> the
> copyright in computer-generated works. The legal tests vary 
> between
> counties and may be difficult to apply, so an agreement improves
> certainty.
>
> When commissioning works, consider seeking a warranty that 
> generative
> AI has not been used. Consider using generative AI only where IP
> rights are not essential, such as for internal use, idea 
> generation,
> and for ephemeral uses such as (personal) social media post.

All of which says to me that distros and package maintainers are 
put in a difficult position, since a project's license depends on 
legal ability to assert copyright over the sources. Those who 
can't assert copyright can't offer the material under a particular 
license, which has implications for whether and how software can 
be distributed, and which on Gentoo obviously has implications for 
LICENSE in ebuilds.

It seems to me that the above suggests that a package maintainer 
needs to continually:

* Determine which jurisdictions are applicable to the sources 
  (possibly including consideration of the jurisdiction(s) of 
  specific contributors), and, if a jurisdiction is involved in 
  which gen AI is not copyrightable or potentially not 
  copyrightable:

* Review changes to the sources to check if anything is the output 
  of gen AI (which is not necessarily straightforward!);

* If gen AI output is involved, decide what to do about it. Remove 
  it to avoid potential copyright/licensing issues? Add a postinst 
  message to the effect of "Parts of this software might not be 
  copyrightable due to the use of gen AI, and thus the license 
  might not be applicable; use at your own risk"? Change LICENSE 
  to some previously-agreed-upon form, e.g. "MIT-AI"? Not do 
  anything in particular? Etc.

On the "Project:Licenses" and "License groups" pages on the wiki:

  https://wiki.gentoo.org/wiki/Project:Licenses

  https://wiki.gentoo.org/wiki/License_groups

we currently have:

> We are programmers, not lawyers. Our evaluation if a particular
> license is a free software license is only a guideline for 
> Gentoo
> developers and users. It is not a legal statement. There is also 
> no
> guarantee that a particular LICENSE variable in an ebuild 
> reflects
> reality. So don‘t rely on it, but check the license that is 
> included
> with the package itself.

Is that considered sufficient for package maintainers to not have 
to go through the sort of processes i mentioned above? And to 
point to if a user says "I've just found that half of the sources 
of [project] come from Claude, why wasn't I told?" 


Alexis.