Some interesting points for sure! And a helpful post because I can respond to your points and illustrate better what I mean.
I don't think AI will be incorporated into OOTP (or many other games) as open-ended LLM projects. Rather, what I see developers doing is taking and compartmentalizing a proven AI capability, and then incorporating it as is, without any further development (until, perhaps, the next release).
So, take each of my bullet points, and imagine what AI will be capable of doing on Day X. Say, Feb. 21, 2025. Whatever that capability is — let's say at making logos — can be made part of the game, at which point the development of that AI will be frozen until it is updated. The AI will no longer "learn," but will simply be able to do what it could do when the game build was released. You tell it, "Make logos, but just make them like this or this or this (like you know how) and just make them sharp and cool."
Or take AI-generated crowd noise. AI can already simulate crowd noise, or a ball slapping a mitt. So you give the game little subroutines where it uses the knowledge it already has to make these noises, and it can only get as sophisticated as the programmer allows. In the game, it's not going to learn as it goes, but it can apply what it already knows. In that way, the developer maintains the same sort of control over the simulation that the developer has always had.
That's how I see all of these aspects of the game being impacted by AI — any game really — with the developers taking advantage of what amount to coding cheats that machines have learned how to do.
I completely agree that no one wants a rogue AI general manager who appears to be trying to make decisions like a human. The results would be too unpredictable, as you point out. Not sure when AI will be capable of that, but it's not part of my vision anyway, probably because I too am an old guy.
