When an AI artist decides to publish the end product, they have made artistic determinations and filtering decisions along the way. They may have produced dozens or hundreds of images along the way to that final product. It was their artistic point of view which was making decisions about what tweaks to make and when they had actually reached the final product. It wasn’t a case of just "Hit a button and get the product’. They had to think of the concept and guide the process along to reach whatever their artistic vision was. They may not have been the ones painting the pixels, but they were making decisions about whether the pixel layout was artistic.
In some ways, it might be similar to how expert artists have apprentices produce their art. The expert may have never touched the actual final product. The expert comes up with the idea and guides the apprentices, but it’s the hands of the apprentices that produce the work, not the expert artist.
You did, and you reiterate it in the next sentence:
What I claimed was that if there’s something unique to creativity, then it is that it produces things that are not wholly determined by the input (nor random), but are choices giving rise to genuinely new information—something which AI is incapable of by its nature. There might not be any such thing, and again, you’re free to believe that, but contrary to your assertion, it’s not in conflict with the laws of physics that there is such a process.
This is saying ‘I never said that there is nothing unique about creativity, but there is nothing unique about creativity’, at least in the way I’ve used the terms, namely, that what is unique about it (if there is anything) is exactly that its determinations are not reducible to the input data (or randomness).
There was an early precedent in a graphic novel, Zarya of the Dawn (Bloomberg article), where the author wrote the words and a computer program generated the images. The Copyright Office ultimately granted copyright to the whole work, but not the individual images. Similarly, a collage is copyrightable even if the individual cutouts are all in the public domain.
For those curious on the state of U.S. law, the U.S. Copyright Office released a report on copyrightability of AI generated art last month. The Copyright Office is, of course, in the business of keeping up with current law.
It addresses a lot of remarks made here in Part II, pp. 18-27 (26-35 of the PDF). For example, sufficiently creative prompts may themselves be copyrightable, but that doesn’t imply the output is copyrightable. The Office compares a prompt to you telling a friend an idea for a work of art, and the friend then going and making it. Ideas are not copyrightable. There’s a very high threshold of creative expression that you have to contribute to the final product for you to receive all the rights to it, basically you have to retain total creative control of each aspect of the process itself, like an artist supervising craftsmen or a director supervising a film crew. The Office makes comparisons to Jackson Pollock, photography, &etc.
Of particular relevance to the OP, the Office described a feature in Midjourney that allows the author to iteratively move individual creative elements in the picture, such as a foreground object. It says these situations require a case-by-case determination. If the author exercises sufficient control over creative expression in the work, it becomes copyrightable.