Ask HN: Is AI not creative because its creators don't care about creativity?
5 comments
AI isn't smart because the people who make it aren't...
Nvm
Nvm
Can you be more specific? I don't see a tangible way to root this discussion except for speculation.
Crank up the temperature on the AI and you'll see plenty of creativity.
With creativity comes insanity, though. At least for your friendly neighborhood LLM.
With creativity comes insanity, though. At least for your friendly neighborhood LLM.
I think there are different aspects to what we call "creativity", and some are more easily mimicked in AI than others.
Of various creative arts, I am most familiar with music, so will use some musical examples, but I suspect this more or less applies to anything.
You could train an AI (or a person) on a bunch of Bach cantatas, and likely the trainee could start churning out similar music. More cantatas of a Bach style.
You could train an AI (or a person) on a bunch of 1980s pop music, and likely the trainee could start churning out similar music. More pop music of a 1980s style.
How likely is it, though, that you could train an AI (or a person) on music up through the 1970s, and (assume granting access to the requisite instruments) the trainee could develop the 1980s pop music sound?
Or training on all music up through the 1940s, how likely is it that the trainee could develop the Miles Davis / Gil Evans "cool jazz" subgenre?
Or for that matter, being trained on music up through circa 1900, how likely is it that the trainee could develop jazz as a whole? Or rock & roll?
I suppose in the "infinite monkeys" sort of scenario, maybe so, and, at that, AI may well have an advantage. If it just churned through all possible combinations of musical constructs, surely it would eventually develop jazz. But would this really constitute creativity? Or just randomness? Would the AI realize that it made something special in finally arriving at jazz? I don't think Miles just randomly recorded "Birth of the Cool".
Of course, it is the "more variations of the same" sort of creativity that most humans exhibit. It is rare that someone develops something significantly new, and even then, it is often multiple people playing off each other. I think one difference is, humans can realize there's something interesting down a particular path, and start trying to develop it on purpose. Can AI do that?
Perhaps what's missing is less creativity and moreso "taste"? But I think it all works together. Is creativity without taste just randomly trying things? But for an AI, you still need a human in the loop to say, "stop, that one there"? I'm not entirely convinced what AI is doing is truly "creative", but even if it mimics creativity well, it seems to me it is still lacking taste, and therein its attempt at creativity hits a wall.
I suppose it reasonable also to imagine that as soon as I hit the "add comment" button, a new AI model will come out that has mastered taste...
Of various creative arts, I am most familiar with music, so will use some musical examples, but I suspect this more or less applies to anything.
You could train an AI (or a person) on a bunch of Bach cantatas, and likely the trainee could start churning out similar music. More cantatas of a Bach style.
You could train an AI (or a person) on a bunch of 1980s pop music, and likely the trainee could start churning out similar music. More pop music of a 1980s style.
How likely is it, though, that you could train an AI (or a person) on music up through the 1970s, and (assume granting access to the requisite instruments) the trainee could develop the 1980s pop music sound?
Or training on all music up through the 1940s, how likely is it that the trainee could develop the Miles Davis / Gil Evans "cool jazz" subgenre?
Or for that matter, being trained on music up through circa 1900, how likely is it that the trainee could develop jazz as a whole? Or rock & roll?
I suppose in the "infinite monkeys" sort of scenario, maybe so, and, at that, AI may well have an advantage. If it just churned through all possible combinations of musical constructs, surely it would eventually develop jazz. But would this really constitute creativity? Or just randomness? Would the AI realize that it made something special in finally arriving at jazz? I don't think Miles just randomly recorded "Birth of the Cool".
Of course, it is the "more variations of the same" sort of creativity that most humans exhibit. It is rare that someone develops something significantly new, and even then, it is often multiple people playing off each other. I think one difference is, humans can realize there's something interesting down a particular path, and start trying to develop it on purpose. Can AI do that?
Perhaps what's missing is less creativity and moreso "taste"? But I think it all works together. Is creativity without taste just randomly trying things? But for an AI, you still need a human in the loop to say, "stop, that one there"? I'm not entirely convinced what AI is doing is truly "creative", but even if it mimics creativity well, it seems to me it is still lacking taste, and therein its attempt at creativity hits a wall.
I suppose it reasonable also to imagine that as soon as I hit the "add comment" button, a new AI model will come out that has mastered taste...
I think AI could learn creativity heuristics that can help it be more creative.
Maybe creative people could be paid to use an AI so that the AI could learn to be more creative from them.
Maybe creative people could be paid to use an AI so that the AI could learn to be more creative from them.
And many mathematically inclined AI researchers probably don’t care about — or even believe in — human creativity.
Maybe that’s why the AIs they build aren’t particularly creative?