[License-review] Adapting the license-review process to AI

[Stefano Maffulli]

On Tue, Sep 10, 2024 at 8:39 AM Roland Turner via License-review <
license-review at lists.opensource.org> wrote:

> There seems to me to be a serious batteries-not-included problem with this
> approach. I do get the need to evaluate terms more broadly than
> copyright-license terms[1] alone[2], but what you're describing appears to
> be abdicating an obvious role for OSI completely:
>
>    - Yes, we obviously can't become a certification body for every OS AI
>    system in the world. Scope, resourcing at every level, assumptions about
>    what F/OSS is about, etc., would all be invalidated by this approach.
>    - But limiting review and approval merely to component terms destroys
>    the meaning of OSI-approval for OSAID (because there would be no such
>    thing), in that it is by no means certain that the combination of a set of
>    compliant component terms will automatically yield a compliant whole[3].
>
>
I see the temptation to jump to this conclusion but I recommend everybody
to remember that the OSD looks easy to interpret today only because it's 26
years old. Also, when it came out there were hundreds of software packages
already using its principles.

For AI we don't have any of that, there isn't a body of widely-agreed-upon
"free as in freedom" work we can look at, there is no practice of openness,
therefore there are no examples we can use to generalize a 10-points
guideline to select AI systems to ship to freedom-deserving users (like
Debian did with its Free Software Guidelines).

The Open Source AI Definition should be seen as a set of widely agreed-upon
principles, a short version of the GNU Manifesto and the What is Free
Software page applied to AI, if you will. The companion document of the AI
Definition is the "Checklist" is still in its infancy, tied to a specific
kind of AI (GenAI) and to a specific view of the domain (the Linux
Foundation Model Openness Framework), still has a lot of room to evolve. I
expect the Checklist to grow as the collective knowledge of what makes an
Open Source AI grows. The conversation started here
https://discuss.opensource.org/t/concerns-and-feedback-on-anchoring-on-the-model-openness-framework/393/
but no conclusion yet.

Based on what I've seen so far in the wild, there is a simple shortcut to
identify an Open Source AI: Did the developers release all of their
training code and the code to create the dataset, with a detailed (peer
reviewed, maybe) research paper? Are the results of the training available
for anybody to use without limitations? If all answers are yes, I'd be
fairly certain that's an Open Source AI.

If any of these is no, you don't need lawyers: you need AI experts, not
generic ones but very specific experts of the AI being examined, to argue
if the technical details and code made available is sufficient to use,
study, share and modify. I think the practitioners will have to show the
path based on their practice.

> *Stefano*, is there a current documented position on the intended
> approach?
>
Not yet. If you think there needs to be one, I'd ask to contribute to the
discussion above on the forum.

Going back on topic, here is a summary of the answers to the original
questions:

- Do you believe that this group should be the one to review the new
documents, or should a separate group be created for it?

I didn't see strong arguments to create a separate group. People may
require to develop new competences though. I'd also add that we may want to
recruit people with skills this group may collectively lack at the moment.

- What new challenges do you expect to see in reviewing these licenses?

Consideration of rights other than copyright, patents, trademarks, moral
rights will likely be necessary (Josh). Because of this and because of how
fresh the field is, discussions are going to be harder (slower) to converge
as we work through multiple new ramifications for the first time. (Roland)

Some of the questions highlighted that there nomenclature is still evolving
and there are no clear standards when it comes to naming and versioning
conventions for these new terms. We don't even know what legal frameworks
apply to some of the components.

- Do you recommend any changes to the process in light of potential new
challenges?

Practical considerations like some kind of organized digital tracking of
verifications (Josh.) Also editorial changes to add references to OSAI and
the OSAID; and solicit additional information relevant for submissions
under the adopted OSAID specifically (Roland.)

Seems quite clear that the list of approved documents which are not
software licenses will need to be hosted separately on opensource.org, like
https://opensource.org/aiterms .

Any other answers to these questions?

thanks,
stef
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opensource.org/pipermail/license-review_lists.opensource.org/attachments/20240912/02d6673c/attachment.htm>