[kelseyfrog]

Just steel-manning the author for a bit: he attempts to position the value of a conceptual scissor between interpolative art and extrapolative art and that this is roughly commensurate with derivative and innovative(creative) art. Furthermore, most large models today are interpolative machines. Therefore most AI art is derivative(ie: not innovative/creative).

I have to question just what proportion of art today the author believes is derivative vs innovative, and if all derivative art were replaced by AI-generated art, would we really have lost anything? Additionally, in which ways, are human-created arts extrapolative? Surely it's not along all dimensions, so by which ones, to what degree? And finally, if extrapolative saliency is quantifiable, couldn't we just have models which extrapolate in the same way and produce innovative/creative works?

It's here that we run out of runway on the author's argument. It feels like an ontologically-based argument couched in behavioral terms. My opinion, and I've stood by this consistently, is that much of the discourse around AI critics(and AI deserves to be critiqued), is that ontological arguments are laundered through behaviorism either through lack of philosophical background, or unclarity of thought, and the result are muddy arguments, black and white thinking, and a host of other unhygienic epistomologies.