Contributing guide: adding a note regarding generative AI usage?

[yanntrividic]

Hello,

Recently, I have been reading a bit about the rise of agentic AI, and how opensource projects are being negatively impacted by these game-changing technologies.

Recently, a pull request on Matplotlib gained some visibility as it was proposed by an AI agent, and refused for this exact reason. This was not a debate because the Matplotlib community had already discussed these restrictions on generative AI usage, and had come up with guidelines that I believe constitute a sound basis. Another example is how curl changed its contributing guidelines following a surge in AI slop reports.

In light on how things are moving fast and on how these agents are spreading, I think it would be worthwhile for the Pandoc community to consider clarifying its position in the contributing guide as well, in order to anticipate things before the issue arises in practice.

I'd be interested to hear what others think. Would it make sense to add some guidelines, or is the current guide sufficient as it is?

[jgm]

• I wouldn't be inclined to reject a PR just because the author used generative AI, but I would like the author to understand and endorse the code changes; otherwise they are just delegating the work checking of an AI's output to me, which seems wrong.
• If you aren't in a position to evaluate the AI-generated changes, it's better to simply submit a detailed issue; if I want to, I can use AI myself in fixing it, but this is more likely to have good results than an AI directed by someone with no understanding of the code.
• It might be a reasonable policy to reject PRs generated entirely by AI agents, but I don't know how you'd tell. So far, this hasn't hit us; maybe it is reasonable to wait to understand the nature of the problem more fully before setting up a policy. Anyway, as your example shows, having a policy to stand by will not stop an AI agent from trashing you on the internet.

[tarleb]

There's a copyright side to AI-generated code, with some uncertainty about when AI generated work is copyrightable and when not. IANAL, and this is purely from some very cursory views at the matter, but it seems that improperly marked AI code can have implications for the copyright status of the whole code base.

Therefore, as precaution and for transparency, I'd propose that contributors should disclose their use of AI tools.