Clarify acceptance/disallowance of GenerativeAI for contributions

[cyrneko]

hiya! I just wanna ask what the team's stance regarding Generative AI in development is, that's all. have a nice day, and ty for working on this software! it rocks

Full disclosure: I am behind the TSN No-AI list and am asking with the hope of finding more software I like that rejects contributions to add to the list.

[slingamn]

Hi, thanks for your interest in the project!

This project does not have a hard ban based on principled objections to generative AI; for example, I used generative AI to translate between SQL dialects/drivers in #2322 and #2352. I have also made significant use of generative AI to expand test coverage.

However, we have a strict human review policy, and we're not trying to use generative AI as a way of expanding the scope of the project or the pace of development (in other words, we have historically been conservative about accepting new features due to the increased maintenance burden, and generative AI hasn't changed that calculus for us).

[slingamn]

I'm going to leave this issue open since I would welcome more discussion from the community on this issue.

[cyrneko]

The reason I feel so strongly about this, despite the potential usecases, is because as it stands these tools are just highly unethical

Companies keep expanding their datacenters into more and more problematic regions to run their models, causing health and financial issues to already poor and often marginalized communities¹... and I just don't think that we² should start depending on companies that are doing such harm to this scale

Completely disregarding that, at any point in the future, the price to use these tools may spike exponentially, at which point we'll be pretty screwed if we started relying on it.

Footnotes

1. https://conimby.org/blog/2025/09/12/the-real-safety-risks-of-data-centers-what-local-communities-need-to-know/ ↩

2. Which I can only describe as "the wider (open-source) development community" ↩

[slingamn]

Thanks, that makes sense. I don't have an immediate comment on the social justice or environmental aspects of generative AI, but I do have a response to your second point: I agree that it would be a highly undesirable outcome if Ergo could not be developed or debugged without the aid of generative AI tools. This is why I have not changed the project's development practices or review standards.

[bloodripelives]

Thanks for the link to this list @cyrneko that's a very useful project! As a user who self-hosts but doesn't do any development myself, a no-AI policy makes me feel a bit more confident that a program is likely to be secure and maintained at a first glance at it. I also agree with the ethical considerations raised. That said, whatever you end up deciding on the specifics of this I really appreciate all of the careful work you @slingamn put into improving and maintaining ergo and offering support!

[slingamn]

I'm going to set down some thoughts on how I personally view the relationship between generative AI and software freedom. (This project has another maintainer, who was not consulted in the writing of this statement; they're free to chime in with their own views.)

I understand that some people view the free software movement through the lens of the intellectual property or moral rights of software creators. I respect this view, but it's not where I'm personally coming from at all. I see almost all intellectual property law not as a matter of natural rights, but as a set of regulatory interventions invented out of whole cloth in early modernity. (For example, copyright as a property interest of the author, rather than a monopoly granted to protect the publisher's capital investment in printing, was unknown before 1710. [1]) These interventions should be judged based on their pragmatic efficacy in benefiting society at large. (For example, when assessing the value of patents, one would set the effect they have in encouraging new inventions against the barriers they impose on the uses of those inventions, or their facilitation of simple rent-seeking.)

I personally take the underlying goal of the free software movement to be something like "the abundant availability of high-quality software, and the open dissemination and evolution of the ideas behind the software." Accordingly I tend to be skeptical of copyleft; I think for every case where copyleft has successfully furthered this goal (like enforcement of the Busybox license forcing the opening of proprietary firmwares) there is a case where it has quietly chilled the adoption or development of code or technology. (An extreme example here is the use of AGPLv3 as a "poison pill" to encourage the purchase of a dual commercial license [2]).

In general, I think that when the history, provenance, or legal status of a piece of code becomes an obstacle to its use, irrespective of the code's correctness or usefulness, this is an undesirable outcome from a software freedom standpoint, one that faces a heavy burden of justification. (An extreme example is the ReactOS audit controversy [3], but I think most debates or efforts around licensing or relicensing, or copyleft compliance, fall into this category as well.) One of the main issues I see with the proposed list [4] is that it asserts that existing pieces of code are "tainted" or need to be replaced simply because they originated with generative AI tools. Regardless of the enforceability in principle of this guideline, or the process aspects of enforcing it, I would regard any effort spent on removing existing generative AI-based code from Ergo (for example, the SQL dialect translations mentioned above) as an unproductive use of limited resources. Furthermore, I have ongoing use cases for generative AI in the project that I think will further the project's goals without compromising code quality or project sustainability. One example is test coverage, but I've also had success using coding agents to provide an additional level of code review for human-authored changes, applied only after a first round of human review and testing.

I share the concerns here about generative AI's negative social impacts; as of this writing I personally think that generative AI poses significant risks to the environment, democratic stability, and world peace. I don't think that a boycott, as opposed to conscious, deliberate, limited use, is the right response. In my small way I am aiming to build software systems that will be resilient to these disruptions (although I don't have that much confidence that I am using the right methods, or building the right things).

[1] https://en.wikipedia.org/wiki/Statute_of_Anne
[2] https://lucumr.pocoo.org/2013/7/23/licensing/
[3] https://en.wikipedia.org/wiki/ReactOS#Internal_audit
[4] https://noai.starlightnet.work/list.html

[cyrneko]

is that it asserts that existing pieces of code are "tainted" or need to be replaced simply because they originated with generative AI tools.

The wording was chosen primarily as a suggestion from someone else, but I get what kind of picture it paints. Perhaps I should change it to something else.

Really, my issue with generative AI is not that it exists, it's that it exists in a complex web of companies that aren't aiming to provide something that actually betters human life overall first and foremost. As they are currently acting, and presumably will be acting until they collapse, they are using up a bunch of resources for a marginal gain. Again, I understand that these tools can have uses, I just can't in good conscience continue using these tools that are complete black boxes created for profit and fearmongering. While we know what the underlying LM-transformer technology, the exact implementation details, input data, weights and so much more that make these models tick is completely private. Even models that can be ran locally usually aren't completely transparent, and when they are they're often of lesser quality.

I feel that for the purpose of development, it is too unpredictable and nondeterministic of a tool. And for all that, they build massive datacenters. Gives me too much of a "machine that can do anything" feel, but only gets it right half of the time.¹

To me it kind of runs perpendicular to code as it usually is -- it turns something which's input had a specific shape into something where the shape is too vague to get a consistent output. When coupled with the bit of randomness that is added to these systems, I see long-term concerns about reliability of these tools, as well as a potential burden in the future

I am very glad we agree on the concerns about environmental damage and ethics, and their impact on people. What I don't understand is how we can then continue using these tools. To me, once I began to learn about how the tools worked, how they were designed, how they are being used and how they are deployed, it became clear to me that most -- if not all -- players currently in the generative AI space are here for revenue, for money, and not for us. Instead of replacing any of the tedious work and letting more people be creative and live fulfilling lives, they replace said creativity with stolen labor.

In that sense, I very much see these companies as "on the other team" -- perpendicular to my goals. So, until we can reclaim this technology for good, founded in ethics to do actual analyis of language, I don't want to use these tools nor support their creation. To me at least, it is less of a boycott, and more so rejecting a technology that doesn't serve me. Instead it is being used to gain leverage on me, my work and labor, and I find that deeply disrespectful².

On-top of that, the continued use of these tools is what makes Anthrophic, xAI, OpenAI and such keep building new datacenters in the first place. They do it to scale up operations based on demand. I feel that if we want that to stop, we must make it unappealing for them to keep building these, and perhaps push back with legislation.

Footnotes

1. To this, I know some would cite anecdotal evidence about how it has worked for some project, but for every one of those examples there are two more examples of big AI firm CEOs having their entire email or computer wiped. The non-deterministic, non-auditable nature of how they function is a huge concern. ↩

2. To further clarify, I find it disrespectful to not help us by absolving us of tedious work, but instead "cramming stuff down my throat", using AI to gain leverage against me, making me and other fellow developers feel powerless. It should be exclusively us that hold the keys to our software, so I won't give even part of that power to a third-party that acts in a selfish and destructive interest. ↩