@Manigarm This is an interesting point, and certainly correct.

It's also exactly how humans learn to become artists and writers - by studying, mimicking, and eventually adding to the existing body of work. We don't generally consider that theft, unless the copying is exact or deceptive.

Yet AI feels somehow different, much more like plagiarism. Perhaps it's that the ONLY input an ML system has is others' art, with no real-world human experience of its own to contribute.

@mattblaze
Honestly, a lot more artists do small scale sales in middle-class contexts than do sales to the rich, and in that context, their work is bought for a more utilitarian understanding of decorativeness+meaning, much closer to the illustration case. But what I mean around labor and value-accrual is on the other end. None of these systems work without the training data, and there derivative works that conversation compressed versions of that training data, and yet the value accrues entirely to an intermediary. Artists have the right to set their own rates for derivative works licensing. I'm not usually an IP maximalist, but there's a meaningful distinction between access to culture on an individual basis and the creation of a system intended to evolve to a point where the work of the people from which its creators are stealing is entirely replaced.

Something entirely based on theft from other people cannot be art. An original tuned prompt absolutely can have artistic merit, but the artistic merit and "artness" of the result rests almost entirely on a) the work of the ML engineering team, and b), much more heavily, on the source material. Now, there's of course a long tradition of artists working with general purpose software and having that software considered a tool or at most (e.g. with reactive projection mapping installations and touchdesigner or max) a medium, so we can ignore the first. However the last is and always will be an intrinsic component, massively more important than anything involved in prompt engineering. The prompt engineering has contributed almost none of the art and should receive almost none of the value.

It would be perfectly possible, if any of these companies cared, to create a fair licensing structure and modeling systems that could provide a proportional attribution distribution across the source material derived from each prompt, paid out in accordance with derivative work prices determined by the rights holders. Until they do this, it's nothing but theft.