All of that is true, but as soon as you model subjectivity and embodiment with any kind of credibility it's game over.
The current generation of tools doesn't do that. It's essentially a very sophisticated parrot making speech-like noises it doesn't understand.
But a couple of generations from now, I think it's going to be much less straightforward.
Also, the thing about genius is that it's subjective. In the arts it's more or less synonymous with mastery of a medium with impactful novel insight.
You could argue that AI systems are well on their way to mastering visual media. It's not quite true, but that's because the people training these systems are not artists, and so far they're selecting work that looks a bit Social Media and Game-Ish rather than Art Museum and Contemporary Show-ish.
I don't see any reason in principle why that couldn't be fixed with better training.
So what about impactful novel insight? The point here is there's a kind of cultural and perceptual feedback process which selects certain works out of semi-random cultural noise. There are always a lot of artists making a lot of work, and most of it is not that interesting. Selecting interesting work doesn't require intent, it just needs a feedback loop.
So if you create a situation in which a community rates the art and selects certain works/algorithms/training sets over others, I suspect the impactful and novel insight will happen automatically.
Sentience or subjectivity are not required. In fact you'll get an automated version of what happens already, where different communities with different levels of education and sophistication select different kinds of work for their own reasons, and some are considered "works of genius" for reasons that may be as political and cultural as artistic.
The current generation of tools doesn't do that. It's essentially a very sophisticated parrot making speech-like noises it doesn't understand.
But a couple of generations from now, I think it's going to be much less straightforward.
Also, the thing about genius is that it's subjective. In the arts it's more or less synonymous with mastery of a medium with impactful novel insight.
You could argue that AI systems are well on their way to mastering visual media. It's not quite true, but that's because the people training these systems are not artists, and so far they're selecting work that looks a bit Social Media and Game-Ish rather than Art Museum and Contemporary Show-ish.
I don't see any reason in principle why that couldn't be fixed with better training.
So what about impactful novel insight? The point here is there's a kind of cultural and perceptual feedback process which selects certain works out of semi-random cultural noise. There are always a lot of artists making a lot of work, and most of it is not that interesting. Selecting interesting work doesn't require intent, it just needs a feedback loop.
So if you create a situation in which a community rates the art and selects certain works/algorithms/training sets over others, I suspect the impactful and novel insight will happen automatically.
Sentience or subjectivity are not required. In fact you'll get an automated version of what happens already, where different communities with different levels of education and sophistication select different kinds of work for their own reasons, and some are considered "works of genius" for reasons that may be as political and cultural as artistic.