Appendix D.1 -- Independence Utopia
-----------------------------------

Collection of training data, and train AI models, is a very expensive process
especially for models comparable to the state-of-the-art ones.  This makes the
AI software freedom issue quite different from the free software issue at the
era of free software movement.

* At that time, individual developers can create free software on their own,
  such as Emacs. Then, they can define their own creation as "free software".

* At this time, individual developers and free software communities cannot
  train large AI models easily on their own.

It leads to a temporary paradox where people may want to define "free software
AI" while not being able to create one (that is useful enough instead of
educational enough), due to not being resourceful enough. The decision from the
proposal may lead to some public criticism due to this dilemma.


Appendix D.2 -- Potential Implications of This Proposal 
-------------------------------------------------------

Downside: Nearly no existing AI model that is useful enough can enter the main
section of Debian archive.  Namely, Debian will not be able to ship most AI
software without relying on downloading from the internet. That said,
downloading pre-trained AI models is a very common practice. Not shipping
pre-trained AI models in Debian archive will not be a problem for users.

Downside: Debian may receive public criticism for being too conservative and
not embracing the world trend. This is understandable when usefulness/fancyness
is ranked ahead of software freedom.

Upside: Within a short future, we do not need to face the technical problem of
handling 10+GB models in .deb packages. And we do not need to worry about
blowing up thousands of downstream Debian mirrors due to giant inflow of large
binary files.