| > What are the limitations? Inability to procedurally generate highly stochastic forms or textures (for instance what a haboku painter, or any kindergartner, can get by splattering paint); reliance on a finite bank of source or "training" images; no way for the artist to turn an artistic vision into a meaningful input for an I/O program; no demonstrated ability or proposed algorithm for producing on its own (without human inputs) anything resembling any component of art aside from formulaic patterns. > AI us nothing but a human given no constraints. I think you have a key insight that productive constraints lie at the living core of the arts, and I agree that if some hypothetical entity that was a human with no constraints produced imaginative works, Kipling's devil[0] would finally get to make a point. I had a similar objection to the hype when AlphaGo beat Lee Sedol – Lee was playing under the constraints of human cognition, which the "AI" with its brute computing power did not even pretend to approximate, yet still overall held his own. That's where the beautiful or "living" quality of the game emerges. But no existing or proposed AI is anything like "a human without constraints." There are productive constraints, and their are counterproductive constraints. The 5-7-5 haiku form is a productive constraint; complete exclusion of the letter "e" from an novel[1] turned out to be an unproductive one. Having to include anything generated by Midjourney in an extremely unproductive constraint. [0] https://www.kiplingsociety.co.uk/poem/poems_conundrum.htm
[1] https://en.wikipedia.org/wiki/Gadsby_(novel) |