I think the photographer example you put does touch upon an interesting point since there really were people who ridiculed photography as not art. And honestly the criteria I had said kinda would disqualify photography, which is unfair.

Would AI be able to create art if it really did understand how the pieces it’s putting together are part of what the user wants? I think it might be an useless question, because skeptics (like me) can keep shifting the goalposts of what understanding really means. So it’s unfalsifiable in a way. Some techbros claim AI can understand it because they are capable of minimizing a loss function. But I’m not satisfied by that because it amounts to making the claim that if a system performs a task well, the system has the property of having a cognitive understanding of the task. It’s a non sequitur, and I’ve seen AI enthusiasts make the same form of non sequitur a thousand times.

Maybe the conclusion we can draw from it is that trying to define what exactly is and isn’t art is hard, but clearly, the OP is not.