[OSGeo-Discuss] Request for a moratorium on AI/LLM use within OSGeo

[Jakob Miksch]

Hi,

My observation is that a huge portion of software developmemt is already created with GenAI. For some usecases like throwaway-scripts or rapid prototypes it is indeed useful and evidently works well.

Nobody knows how this will develop in future, but I guess these will be the two ends of the spectrum:  On the on hand software that can be understood and manually edited by humans **without GenAI**. On the other hand software mainly created by GenAI that is not supposed to be understood and manually edited by humans anymore.

As we all know, there are different interpretations of "Open Source". And vibecoded software can also be Open Source in a way that the sourcecode is open and has a free licence. But often the internals of these projects can hardly be understood by humans and require a LLM-subscription for contribution. For me this is similar to publishing a binary file under a free license.

My preference would be that OSGeo Open Source projects always puts the human in the center. This means humans should always be able to understand the code with their own brain. Also manually editing and running the software must be possible **without** usage of LLMs. This however does not necessarily forbid the responsible use of GenAI, which undoubtly can be helpful for many tasks as long it does not produce slop and can be understood by humans/maintainers with reasonable effort. 

Also I find human communication like issues, pull requests or docs must be written manually by humans. Responsible use of GenAI is fine for me in cases like translation, spell/grammar check as long the human always sits on the drivers seat.

Maybe there will evolve vibecoded "Open Source" projects, where mainly "AI"-agents collaborate with each other and only humans with a LLM-subscription can reasonably contribute. I dont think these kind of projects fit to OSGeo. 

Jakob

PS: I stumbled upon this song, which feels like the soundtrack of the current times in software development https://youtu.be/3TNpOD6bov8



Am 9. Oktober 2026 00:23:08 MESZ schrieb Javier Jimenez Shaw via Discuss <discuss at lists.osgeo.org>:
>Hi
>
>I am also worried with the impact of AI in our community and in the
>software development in general (and many other aspects as well). I find it
>very risky, and people going into it without any caution.
>Other technologies in the past were also dangerous at the beginning, and
>the society learned to deal with that risk. But the speed of introduction
>of those technologies was extremely slow compared with AI, letting us more
>time to learn those dangers and take the proper measures.
>To be clear: they are still risky. Not the same person uses a knife,
>dynamite or enriched uranium (just as an example). But still a knife, that
>we use daily, needs some training and safety measurements. We don't five a
>knife to a baby; we don't give drugs at the school door; but we are giving
>AI to anybody for anything. And extremely cheap.
>AI is fuelled by gazillions of dollars, as it never happened before in
>history. It is incredible. Just to introduce a very expensive technology
>(completely subsidized nowadays) that... do we really need it? Making us
>more productive is really a need? Is our live changing that much? No, we
>are all just more stressed, having to be more productive in a useless race
>against... what?
>
>Many people say "we check the code, we review the PRs". But don't be naïve:
>there is not way to seriously review all that code and read those horrible
>PRs.
>Others say "with LLMs I can finally contribute to that project in a
>programming language and/or topic I have no knowledge". Well, maybe you
>should not contribute there if you do not have the knowledge. Seriously.
>
>As a default position I also think we should be "better not AI". Are there
>good usages of LLMs? Yes. Are them the majority? In my opinion no. I prefer
>a moratorium until we have a better perspective. Until we digest the
>impacts of the current situation. And we do not overstress some projects
>with a flood of PRs and issues that are not helping.
>Recently I saw in a project an issue created by a new github user, and the
>day after the PR by other new github user. What are they doing? It was
>obviously coming from the same origin. The issue was itself telling the
>proposal. Why was not the same person doing the PR? Maybe because it was
>not a person?
>
>Why that hurry? Why do we need to be faster and faster?
>
>Javier.
>
>On Thu, 8 Oct 2026 at 20:05, Even Rouault via Discuss <
>discuss at lists.osgeo.org> wrote:
>
>> Hi,
>>
>> (sent both to discuss, and CC'ed to board mailing list for information.
>> Please keep the discussion only on -discuss)
>>
>> I don't think the current position of the foundation of not taking a firm
>> stance on the matter of AI/LLM, and let each group decide, can be held
>> anymore. It is becoming increasing clearer that penetration of AI within
>> our community is toxic, and require each group to go through the painful
>> process of defining its own position about the topic, which is very
>> complicated and drains a lot of energy among community members. Recently
>> TorchGeo (https://github.com/torchgeo/torchgeo/pull/4116) and QGIS (
>> https://github.com/qgis/QGIS-Enhancement-Proposals/pull/387) had to go
>> for a full ban of AI/LLM use, both for coding and communication, given the
>> difficulty to frame what exact use of that technology could be acceptable.
>> I've also noticed in the incubation committee that nearly all recent
>> submitted community projects have been largely vibe-coded, which seems
>> incompatible to the criteria we expect of community projects, but the
>> committee lacks clear guidance on how to proceed.
>>
>> Open source has always been about doing things differently from others,
>> and I believe that on AI, we must resist to the trend of embracing a
>> technology that has so many unknown consequences, and already very nocuous
>> ones. This falls under the precautionary principle.
>>
>> A moratorium on AI/LLM use in all OSGeo venues (projects, committees,
>> communication channels, conferences, etc.) seems to me to be to be the
>> prudent approach. At least as a by-default policy. Each group could
>> potentially reframe it, but by default, I believe groups should be
>> protected. In some time, when the implications of using those techs will be
>> better understood, and the landscape will have cleared up a bit, a
>> potential relaxation can be discussed.
>>
>> Even
>>
>> -- http://www.spatialys.com
>> My software is free, but my time generally not.
>> LLMs contribute to global warming and brain rot.
>> Let's guillotine them! "Ah ! ça ira, ça ira, ça ira !"
>> Initiator of https://fuckaifoundation.org
>>
>> _______________________________________________
>> Discuss mailing list
>> Discuss at lists.osgeo.org
>> https://lists.osgeo.org/mailman/listinfo/discuss
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/discuss/attachments/20261009/8e3986fb/attachment.htm>