FOSS APAS & FOSS-friendly ToS restrictions (was Re: Minutes of AI Assist Committee TUE 2022-06-22, 18:00-19:00 UTC)

Bradley M. Kuhn bkuhn at
Wed Jun 29 22:11:09 UTC 2022

2022-06-22 committee meeting minutes noted:
> > Some also noted we'd probably need to use terms of service to keep the thing
> > FOSS if we wanted to enforce a moral requirement (as opposed to, say, having
> > the copyleft license alone “take care of it”).  Using terms of service to
> > mandate software freedom has not been used in past, and may be another type
> > of activity that would dilute coalition between more and less strident
> > activists.

Just as a reminder: the Committee is debating lots of speculative ideas about
what could be, or what we might want to do as activists.  So, when you read
the minutes, keep in mind they aren't officially proposals — they're
discussion and ideas; we're publishing the minutes for transparency and to
spur discussion.

I say that because …

Joseph Turner wrote:
> IIUC, the terms of service would be used to mandate that this new FOSS
> Trainer be exclusively used on FOSS inputs to produce a FOSS Model. Is that
> correct?

… I don't know think anyone on the committee came to any conclusion on
specifics of how a ToS would be used here, just a general sense that ToS has
not been used before as a tool for software freedom, and the idea was floated
that perhaps such can more easily “be done” with ToS when copyleft
can't/shouldn't do those things.

If we open that door (which, as you can note in the minutes, some on the
committee immediately felt this runs immediately afoul of “fields of use” as
stated in the OSD/DSFG), probably a lot of options are present, but the
question would be whether they are ethical or not, and whether they violate
some other principle of software rights.

cf: the part in the minutes where we talked about whether “field of use” was
a fundamental principle or not, at least when the field of use is “writing
more proprietary software”.

To add something to that we didn't discuss in committee: one might argue that
copyleft itself is a “field of use” restriction since you cannot modify
copyleft software *into* proprietary software, so your “field of use” is, in
some sense, restricted.  I think it's historically and culturally significant
that DFSG/OSD were both written well after copyleft was established to be
clearly a reasonable way to promote and protect software freedom.  I posit
that had DFSG/OSD been written *before* copyleft was invented, *many* would
have argued that copyleft itself *is* a “field of use” restriction.  That's
what leads me to wonder whether or not “no field of use” restriction is more
“cargo cult” than it is a rigorously considered a priori principle of rights.
Perhaps there *is* some software where that really ought to prohibit a given
“field of use” … loosely defined.  For example, the GCC RTL Exception to GPL
exists as a tactical move, not because it's morally wrong for a compiler to
have a license that only allows it to compile code under a GPL-compatible
license.  What if GCC were to remove the RTL Exception entirely tomorrow
(which the GCC developers can do at any time for future code — after all, all
GPLv3 Additional Permissions can be removed by anyone to return to pure
GPLv3).  Would it be reasonable to say that GCC team had added a “field of
use” restriction to GCC?  If not, why not? 🤔

Anyway, that's just a thought experiment to explore on where the “field of
use principle” comes from and to think about whether it really ought to be an
absolute/a priori principle in the age of more advanced AI.

Regardless, as I wrote in
, I do believe that holding together the strained coalition around the
FSD/DFSG/OSD has value and we must take great care.  A ToS-based software
freedom protecting solution could easily backfire in this regard, just as
writing in other ethical requirements into (would-be) FOSS licenses does.

> *If* the primary cost to produce a FOSS Model is compute time and not
> developer time in creation and maintenance of the Trainer, then perhaps no
> TOS would be necessary to prevent the creation of a proprietary Model with
> the FOSS Trainer. IOW, if donations are given with the explicit goal of
> creating a FOSS Model (and Trainer), and we create a good FOSS Model, then
> perhaps the cost to create an alternative proprietary Model from the FOSS
> Trainer would be prohibitively high.

This logic doesn't seem to hold to me, because the cost is prohibitively high
for hobbyist and charities, but not for-profit companies and their trade
associations — all of whom are the key entities that either don't mind (or
actually *want*) more proprietary software in the world.  I think your
argument only works in a world where resources are distributed perfectly
equally to all … and that world has never existed in human history AFAIK.
Bradley M. Kuhn - he/him
Policy Fellow & Hacker-in-Residence at Software Freedom Conservancy
Become a Conservancy Sustainer today:

More information about the ai-assist mailing list