briee
@briee

feel like i've seen AI discourse veer into dangerous waters lately. losing the threads that techies raised alarm bells about, like:

  • AI is trained on unethically acquired data
  • AI allows for laundering of bias and responsibility

and changing, among people with good intentions but little understanding, into

  • AI makes generative art, therefore generative art is bad and soulless
  • AI makes low-effort art, therefore work that appears low-effort is bad
  • AI violates copyright, therefore sampling and derivative works are bad

these aren't one-offs, but throughlines i've observed (admittedly anecdotal. i have no data). it's frustrating on many levels, but one that gets me: it accidentally buys into the AI hype cycle, giving AI credit for things that existed long beforehand

  • mozart rolled dice to splice together songs
  • consider the hatred for rothko and duchamp and the myth of the lazy modern artist
  • and i need not remind you of musical sampling, fanart culture, doujinshi culture

right now i'm using software that performs an evolutionary search to make a software synth's params loosely replicate a sample. before the AI bubble, this would be an experimental workflow, but now, it might set people off

i'm really hoping they're outliers


You must log in to comment.

in reply to @briee's post:

I see a lot of "it looks bad", which is also a really bad argument because, like, not wanting bad art to exist is kind of antithetical to wanting humans to make art. Humans make bad shit all the time for fun ON PURPOSE. It's a pretty inherently human thing to want to do. (Also really curious, what software are you talking about there at the end? I feel like I know what it is but I don't wanna nerd out about something 100% off topic by accident lmao)

Ok I just looked that up and it looks so fucking cool!!! I'm going to have to check it out sometime :D

I was thinking of some of the AI stuff they're adding to a lot of singing synths. About 6 years back to do a similar type of things with like figuring out how to generate nonexistent voice tones to reduce how much you'd have to record for a concatenative synth, made an entire singing synthesizer backend over the span of like 2 years that I was going to use to implement this, showed it to someone to try to get a job and they literally went "that's not that impressive". STILL VERY MAD ABOUT IT, but all my friends think it's pretty neat. Anyways there's a bunch of stuff like that out there now, at least from what I've seen, and it all looks so interesting I'd love to go back at it if I ever got the motivation lmao.

For my synth backend I was using WORLD, and I think the only thing I was doing different from a lot of other concatenative synths at that time that used WORLD was instead of rendering in chunks, which makes rendering times pretty reasonable, I was pre-analyzing all the audio files but ONLY saving their spectral envelope, loading those, splicing everything together all at once, generating the f0 and aperiodicity (I just filled that one with zeros bc it was faster than saving and loading it -- somehow still sounded fine) and rendering the entire thing all at once, so it took quite a bit longer but the vowel to vowel transitions sounded really good. you win some you lose some lmao.

Sorry for the long reply, this is unfortunately my special interest and if I don't talk about it enough I will explode