I believe this proposal to reject contributions generated by AI on a
line-by-line basis is counter-productive to the GNU and Free Software
community over the long term.
GNU Guix is an incredibly powerful primitive for developing free,
modifiable, and repeatable programs and operating systems.
The vampire capitalists who want to hoard the means of production
amongst themselves are not going to slow down developing proprietary
operating systems and computer services purely for profit, without
respect to user choice nor freedom to modify. They will try to
monopolize everything we do on our computers into their cloud.
However, this policy proposal will put great distance over time, in the
inevitable world of ubiquitous computing, between our resources in the
free software community, against corporate monopolies and overbearing
government control as proprietary software simply out-competes if we
eschew AI assistance. This proposal would constrain ourselves to using
the equivalent of hand tools, while technological progress continues at
an accelerating pace in the world of proprietary software with
terrawatts of energy behind it. It is a recipe, if carried out, to
consign free software to the rust bucket of computing history as it
becomes technologically obsolete over time if we do not make a wise
decision.
If we wish to have the GNU ethos alive and thrive over the coming
century, when our posterity will accept for granted the ubiquity of
artificial intelligence, including android robots in the home, in the
office, and in society, we need to embrace the technology, not reject it
outright. You or I might not have all the data or hardware to train an
android's AI model, but we nonetheless want the ability to control it's
source code, operating system, and kernel, or to pool our resources to
be able to fine tune it with the data that can be gathered and shared in
the community.
If we reject, out of the pride in our own craft, the contributions of
machines, we will certainly doom ourselves into the world of
obsolescence. The capitalist software pirates will succeed over time in
rewriting every last bit of GNU coreutils into non-GPL licensed
programs, porting the linux kernel to another language with another
license writ large similar to the recent Bun rewrite, and generally
outpacing the free software community who categorically reject AI-guided
software development.
I would hate to see the Guix project fade into obsolescence, if not
become forked into a lasting community divide, as we struggle to keep up
merging package updates, reviewing each others contributions, make large
and useful refactors, or trying to play turing test police against
otherwise valid and useful contributions made by machines. After all, AI
programs are still just computer programs, computer programs which we
all love to create, and we should not be prejudiced against them
bringing value to Free Software on principle. When we embrace AI, we
might find some pleasant surprises to be gained, such as proprietary
firmware which can be reverse-engineered, security vulnerabilities which
can be prevented, and a multitude of workflows which can be automated
and delegated to specialist programs to maintain a magnificent garden of
the Guix ecosystem.
For that reason, I do not think we should have rules to categorically
reject LLM-assisted contributions based on lines of code alone. By all
means, reject low quality AI slop and code of dubious copyright claims
which might create legal liabilities - but let contributions from other
computer programs be judged on the merits. Let us accelerate the
development and maintenance of GNU Guix and the free software ecosystem
using this fascinating new computing paradigm for good, rather than
greed, to help maintain free software as the gold standard of computing
in a competitive environment for a century to come, with Guix playing a
leading role in that future.
Sincerely,
Ryan Sundberg
