Great article.
I have to admit that I'm more on the "style stealing" threat side of things, because we've never had a tool that could do it so proficiently. That said, I've always felt that Mohrbacher's application of AI is spot on and absolutely what I want from generative tools myself (I'm a professional composer and generative AI researcher in music). However, his perspective on what I'd call the "fair use" side of things is probably a bit skewed toward his own interests as an artist. When he says that "even if you could perfectly mimic styles, people will naturally end up mixing five different artists to create something new" I think that's true if you're speaking of artists, but not so true of industry. Industry will take whatever works for their end use, without regard for where it comes from. And if they can do it for free (or pennies on the dollar) they will. Also, if you look at the progress of this tech over just the last several months, it's not a question of "if" full style replication will be possible, it's only a question of "when" (arguably, it already is, depending on the style). The training procedures are, by definition, optimizing toward that goal—i.e., a style is a probability distribution from which samples (individual works) can be drawn. A better algorithm models that distribution more accurately and the samples it generates are closer approximations of the "cause" of the training data (i.e., the artist).
It's easy to take people who urge caution as being "alarmist", but I think that's a mischaracterization in a lot of cases (certainly in mine). I think we need to find a way of including the original content creators into the business models that these algorithms are creating, otherwise we'll have just another case of redistributing human effort and value up the same old pyramid...
Beyond that, I absolutely agree with Mohrbacher that artists creating original content (whether purely "analog" or through hybrid AI practices) will continue to create because that's what they do. And due to the data-driven nature of these algorithms, they will continue to define the future of what these generative models are capable of. But it's worth keeping in mind that, in doing so, they will be driving a machine that's creating huge amounts of real, concrete, monetary value. Leaving the artists who create the content that drives these technologies out of the monetary picture isn't a reasonable proposition, imho.
And no, I'm not talking about copyright. Copyright is, frankly, a joke and insufficient for the generative AI age.
