Re: [MAINTAINERS SUMMIT] Annotating patches containing AI-assisted code

["Paul E. McKenney" <paulmck@kernel.org>]

On Mon, Aug 11, 2025 at 05:02:19PM -0700, Luis Chamberlain wrote:
> On Mon, Aug 11, 2025 at 04:42:18PM -0700, Paul E. McKenney wrote:
> > On Mon, Aug 11, 2025 at 04:22:21PM -0700, Luis Chamberlain wrote:
> > > On Mon, Aug 11, 2025 at 03:51:48PM -0700, Paul E. McKenney wrote:
> > > > On Mon, Aug 11, 2025 at 03:11:47PM -0700, Luis Chamberlain wrote:
> > > > > c) If something like the Generated-by tag is used, and we trust it, then
> > > > >    if we do want to side against merging AI generated code, that's perhaps our
> > > > >    only chance at blocking that type of code. Its however not bullet proof.
> > > > 
> > > > Nothing is bullet proof.  ;-)
> > > 
> > > Agreed, and I think the legal concerns over AI code use are just as weak. I
> > > just don't see it holidng up long term.
> > 
> > That is quite possible.  But on what are you basing that legal opinion?
> 
> Its not a legal opinion. Its a personal opinion based on projections on
> growth, adoption, and personal risk analysis, and valuation for my own projects.
> At some point a project needs to take a positon on this, I had decide
> sooner for another project.

Your project, your opinion, so questions asked.  From me, anyway.

But...

> > > My expectations are that eventually foundation AI models will simply state they
> > > use permissively licensed code for training, and be done with these concerns.
> > > 
> > > Until then -- we just have wild speculations and I can't see any
> > > sensible case ending up in court wanting to avoid AI code in open source.
> > 
> > I don't know about open source, but they tell me that related cases are
> > already in court.  Yes, there was a recent decision that was favorable
> > to your position, which is great, but not necessarily either definitive
> > or final.
> 
> Indeed, its a risk assessment in the end.
> 
> Let us take an example. If one is using foundation models perhaps the
> ugliest position you could be in, is if you want to avoid GPL code on a
> non-GPL codebase. Since we don't have access to AI model training
> logistics, if we just work out of the code on Github the numbers I
> came up with was about 60% permissively licensed code, 25% GPL, 15%
> unclear. Give or take.
> 
> If you're using copyleft code though, well, the project is already open.
> So what's the risk assessment? Well who and why would they go after your
> project? My risk assessment for my projet is low, and due the high
> empirical value I already see in leveraging AI code, I think its worth to
> embrace.

Sadly, there is precedent for people going after copyleft projects.

> Eventually I predict foundation models will just take a position to
> annotate where what code they train their models on and I suspect that
> will be permissively licensed code. By the time this happens most of the
> code we know written by humans will have been replaced already.

Perhaps.  But on the other hand, there is a lot of code still in use
that was written by humans who have long since passed on.  So I am not
convinced that code replacement will happen all that quickly.

Or maybe you are saying that it will be a good long time before AI
projects implement the kind of traceability that we are discussing?

							Thanx, Paul