

To me, what makes the difference is whether or not it’s trained on other people’s shit. The distinction between AI and an algorithm is pretty arbitrary, but I wouldn’t consider, for example, procedural generation via the wave function collapse algorithm to have the same moral implications as selling something using what most people would call AI-generated content.
I don’t have a problem with games creating their own models trained only on things they created. I believe charging money for anything using assets generated by a model trained on data they didn’t have the rights to should be illegal. If a model is trained on data that they do own the the rights to, but didn’t create, that’s a weird gray area where I think it shouldn’t be illegal to sell its results, but you should have to disclose that you used it.