Re: <nettime> The Copy Far "AI" license

[rhatto via nettime-l <nettime-l@lists.nettime.org>]

On Tue, Sep 26, 2023 at 02:10:08PM +0300, ozgur k. wrote:
> On September 23, 2023 6:48:59 PM GMT+03:00, rhatto via nettime-l <nettime-l@lists.nettime.org> wrote:
> >Hello nettimers, this might be of interest:
> >
> >  Copy Far "AI" - A license close to copyleft, but far from the so-called
> >  "Artificial Intelligences": https://copyfarai.itcouldbewor.se
> >
> >ps: I'm not in the list (please Cc).
> 
> "[...]cannot be used for procedures known as "machine learning" and
> stylometric analysist[...]" https://copyfarai.itcouldbewor.se/ ;
> 
> for proprietary ones, it could be a nice strategy/statement, but how
> about floss and even copyleft floss ones? what if the model is/will be
> released under a copyleft license? who would benefit more from
> limiting the use for training copyleft floss uses as well? 

Supporting some sort of "FLOSS" "AIs" could be a use case.

But I wonder:

1. How this could be defined, and offer protections against scenarios
   where FLOSS "AIs" and data are part of a larger, proprietary system
   (similar to what happens with open source and "open core" nowadays),
   an hence defeating the license purpose.

2. How to keep the license text simple but supporting variant use cases
   like the one you suggest. One way is through forking, say maintaining
   something like a 'Copy Near "AI"' sister project.

3. Shall exactly FLOSS be the right scope regarding "AI" licensing?
   Seems like the important issue is not whether to support open/free
   versus proprietary, as these are categories mostly tied to
   _products_, whereas what could be encouraged are technologies
   related to fair _relationships_ (which are more like _processes_).

> i don't know if it is legally possible to enforce the output of
> copyleft ai tools to be copyleft as well. say, enforcing a program
> output from a copyleft ai system to be released as copyleft.

I don't know either. But the output of tools using share-alike data
as input would need to retain the same license.

And the tools themselves could be released under Affero GPLv3+, to
ensure that any changes should also be released (and not kept in
private).

> even if the copyleft method wouldn't be possible legally and
> technically for ai situations, isn't it a good.idea to support /
> encourage the development of floss models over proprietary ones
> through the statements as in this one?

That's a good point, and thanks for raising it.

When drafting Copy Far "AI", I decided to start focusing in the simpler
situation, so authors can state that using their work for "AIs" is
denied, at least until they can take a more nuanced position.

> --
> özgür k.
> gpg:A3E6 57AD E14D 1F66 A546 6101 BA42 0724 E750 C5AE

-- 
https://fluxo.info

Attachment: signature.asc
Description: PGP signature

-- 
# distributed via <nettime>: no commercial use without permission
# <nettime> is a moderated mailing list for net criticism,
# collaborative text filtering and cultural politics of the nets
# more info: https://www.nettime.org
# contact: nettime-l-owner@lists.nettime.org