Add note on AI-assisted contributions in contributor code and PR template

[rhugonnet]

As discussed yesterday night, I'll come up with a first draft.

[dshean]

I found the recent GDAL mailing list discussion on this topic to be valuable. The resulting policy doc is here: https://gdal.org/en/stable/community/ai_tool_policy.html#ai-tool-policy

Personally, I think it is valuable to have a record of the AI models used for each contribution preserved in the commit history, and would not suggest requiring people to remove Claude from their PRs. If you are concerned about perceptions of developer contributions and academic credit, at this point, the community understands the role of AI code assistants, and the difference between human authors and AI authors.

For reference, Sliderule has major contributions from both Claude and Copilot, and Claude is now the #3 contributor to ASP.

[rhugonnet]

Indeed, the GDAL policy is quite nice, thanks a lot for sharing @dshean! 🙂
What do you think @belletva @adehecq, good for you? We could have a shorter version, and one that insists more on copyright/code conception as well?

Re contributor list: Interesting to get your perspective there. I did think about perception, as you mention, so it's good to hear examples with SlideRule/ASP.
But for me it's really about conceptual credit and use of the tools, in particular:

• We squash-merge into the main, so original commits are condensed into a single commit representing the PR, and it's the responsibility of the author/maintainers to review that PR code entirely so that it becomes their own, so the AI role should be minimal by then compared to an individual commit in the PR;
• Other completions (pre-commit) tools/linters also modify our code all the time in individual commits, but we don't list those;
• Many contributors use these tools in chat (not in agent), so they are not consistently acknowledged... Even though this will likely come eventually.

In short, we want this to be fully designed/checked by humans (and AI to be mostly a tool to do things faster, as for other tools), and so final authorship could reside solely with the human author? That was the reasoning behind this choice.
Not sure what other think on this; we mentioned this a bit in the dev meeting, but it was getting late. 😅