Hi Hugo,
On Thu, 16 Jul 2026 at 12:06, Hugo Buddelmeijer via "Development of GNU Guix
and the GNU System distribution." <[email protected]> wrote:
> Second, I did (and still) do feel that it is possible to use tools like
> LLMs without harming conviviality; even enhancing it.
For what my opinion is worth, I don’t think LLMs might be a convivial
technology, never. Aside LLMs is about a monopolistic competitions,
from [1]:
Illich identified five main systemic threats to conviviality in
technologies:
(i) biological degradation, which refers to the negative
eco-systemic impacts of the development, production and use
of a technology;
(ii) radical monopoly, which refers to the over-dependence on, and
imposition of, the use of a technology, and to the way these
in turn over-constrain solutions and human thinking in a
problem domain;
(iii) polarization, which refers to the imbalances of power created
by the deployment and structure of a technology;
(iv) over-programming, which refers to excessive specialization,
constrained forms of learning and ensuing work alienation
induced by a technology, to the detriment of human autonomy
and human creativity;
(v) obsolescence, which refers to the excessive and engineered
rate of obsoleteness of a technology and of the associated
human knowledge for its development, use and maintenance.
And
(1) The ecological impact of LLMs is well-documented. Obviously most
of technologies have a biological impact, so the question is the
right balance of the grey: What bad do we accept for which good?
To me, the collateral damages from LLMs exceed the benefits.
(2) We already starts to view the radical monopoly of LLMs.
(3) Polarization about LLms seems more than clear.
(4) I start to receive emails about training from my University. For
example: Boost your grant application, or Explore genAI to boost
your creativity on future discoveries and innovations, etc.
(5) I don’t count the number of times: Try out the last LLM with more
parameters and specialized this or that and guess what, it’s just
so better!
Therefore, LLMs will never be “convivial” by Illich definition because
LLM ticks all the 5 threats boxes by its design.
It does not imply we cannot build some LLM variants trying to fit at
best some Common Good principles. However, as best as these LLM
variants would implement these Common Good principles, it would not lead
to a “convivial” technology by Illich definition.
Such Commond Good LLM could make easier some tasks, maybe it could
simplify some other tasks. However, it cannot be a “convivial”
technology by design, IMHO.
Cheers,
simon
1: Conviviality for Digital Degrowth
https://doi.org/10.4000/16954
