Re: Generative AI Policy (was Re: Experimental DSM-based terrain renderer with streamed orthophotos)

Thank you

Let's not become one of those projects/companies that had huge technical debt and unmaintainable dude due to AI crap slop!

--
Josh Davidson
________________________________
From: Stuart Buchanan <[email protected]>
Sent: Tuesday, June 2, 2026 8:23:00 AM
To: FlightGear developers discussions <[email protected]>
Subject: [Flightgear-devel] Generative AI Policy (was Re: Experimental DSM-based terrain renderer with streamed orthophotos)

Hi Florent,

On Tue, Jun 2, 2026 at 1:45 PM Florent Rougon via Flightgear-devel <[email protected]<mailto:[email protected]>> wrote:
Le 02/06/2026, Stuart Buchanan a écrit:

> I don't think we yet have a policy on using AI on the project, but my
> expectation is
> that you as the author need to stand behind any contribution.

We have this:

  https://docs.flightgear.org/contributors-guide/guidelines/generative-AI.html

We do, but we haven't formally adopted that text as a policy.  To save folks clicking on the link, here's what it says:

----

All contributions to FlightGear must be human authored. Tools such as Large Language Models, and more generally “AI”, may be used but contributors are fully accountable for their contributions and responsible for their quality. They must have reviewed line by line the output of any such tool and understand the changes they are proposing. So-called “AI slop” and automatically generated merge requests are unacceptable and will be rejected out of hand as they waste maintainer time.

----

I think this is a good time to discuss this policy and get consensus from folks.

The context here is that we should have a policy so that we are consistent and there are clear expectations for contributors.  We want to strike a balance between using AI to make volunteer time more effective, against the risk of AI slop, wasted time and creating an unmaintainable codebase that no-one understands.

We have a couple of examples which contrast approaches here:

Gerard's work is really impressive, and has been developed with incredible speed.  As he comments it has only had a limited human code review.   Should we expect that all code is reviewed by the contributor before being submitted as a merge request?  We currently have maintainers review all code going into the repo, which is valuable and regularly picks up bugs.

In contrast, I've been using AI as a tool to assist the development of the voxel-based clouds, and I'd estimate that it has made my time 4-10x more efficient in reviewing code, discussing approaches and fixing bugs.  That said, I've authored all the code and stand by it.  James H has still found plenty of bugs in it :).

Finally, we've already seen a bit of AI slop on gitlab which simply wastes volunteer time.

I appreciate that this is a part of a bigger conversation happening within software engineering as a discipline.  However, we operate within a slightly different environment here from a commercial software firm:  our focus is not on delivering value for shareholders, and we are constrained by volunteer time, with the caveat that we also want to encourage new contributions.

Best regards,

-Stuart

_______________________________________________
Flightgear-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/flightgear-devel