Comment by sharkjacobs
2 days ago
I'm not ideologically opposed to making music with AI, but the dream would be new songs which which showcase the new sounds and musical forms that AI enables, like Believe for autotune, or Rumble for electric guitar, or Autobahn for synths.
I want a friend to message me like "Hey, there's some interesting stuff happening in the AI music scene, check out these tracks".
But everything I've seen is pastiche, either novelty songs (hit song as different genre, or famous monologue from popular movie as pop song) or generic background music meant for algorithmic streaming playlists.
I'm a moderately frequent user of Suno and have encountered some unusual AI-generated effects that I've not come across before in more traditionally synthesized music.
One is when it attempts to generate vocals without a lyrics prompt. It's gibberish but just on the edge of comprehensibility. Sometimes it'll be entirely spoken word with no accompaniment. Very uncanny.
Another is transitioning between vocals and instrument in the same melody line. Like a humanesque voice holding a steady note at the end of a verse which seamlessly transitions into a saxophone sound and proceeds into a solo. Or vice versa, an instrumental morphing into a voice.
Finally is when the generation goes wrong and it starts spitting out absolute nonsensical sounds with no rhythm or melody, in a uniquely fragmented way I can't really describe. It feels like seeing the musical matrix, the inner thoughts of the AI.
Now I've written all that out and had a think about it, I'm tempted to sample these oddities and try to make something more structured out of them.
>Another is transitioning between vocals and instrument in the same melody line. Like a humanesque voice holding a steady note at the end of a verse which seamlessly transitions into a saxophone sound and proceeds into a solo. Or vice versa, an instrumental morphing into a voice.
I think there's something cool here, seamless morphing between sounds was one of the things they were trying (and failing, obviously) to do at IRCAM way back when. Finally we might be able to morph in something approaching perceptual space.
(and glitch is always interesting too, of course)
This sounds cool, but:
1. Pretty sure you can do all this already, it just takes some skill in certain programs and maybe recording special samples.
2. When the AI does it, it might sound weird and new, but I feel the lack of artistic vision makes these more artifacts than art. Even if it sounds interesting, there's no meaning to it.
Another is transitioning between vocals and instrument in the same melody line. Like a humanesque voice holding a steady note at the end of a verse which...
Cantabile: https://en.wikipedia.org/wiki/Cantabile
I'd be interested in hearing a few of those samples and what you made out of them!
Yes, it's really easy to argue against AI music and find that you're making exactly the same argument that was used against DJs and sampling in the 90s. "Real musicians" thought they were being ripped off by "non-musicians" who didn't know how to play an instrument, just a turntable or an AKAI. But it turns out that turntables and samplers are instruments if you allow people to get creative with them, and now we have entire genres of music that exist because sampling is legitimate (though the copyright wars did make it more expensive than the original guerrilla days). I mean, do we seriously think DJ Shadow isn't a musician?
You could regard AI as being literally just a very advanced form of sampling. I've seen and heard some very creative uses of AI tools, and it would be a terrible shame if that baby got thrown out with the bathwater.
Yeah, I want the Paul's Boutique of AI music, but I'm not interested in wading through the AI equivalent of a hundred thousand acoustic guitar covers of Wonderwall
This demonstrates where a lot of the mismatch in impressions of this tech arise. The thousandth amateur Wonderwall rendition is not at all interesting as a piece of recorded music, but for the performer (and those listening around them) it can be a fun and playful experience. The same could be said for AI generated music: it could be a fun and playful experience in the present moment, even if the resulting product is totally worthless to the market. This would still be a valuable thing for the human experience.
Arguably this is a return to a more traditional way of experiencing music from before the invention of recorded music. Before this, music was an entirely transient and often communal experience. Once the musician stops playing, the music is over. Songs from these times have largely unknown authors, and likely don't even have any single author or for that to even be a coherent concept. They were simply part of the shared culture that many had contributed to. Now music is owned by specific people and you can play back their performance as much as you like (for an increasingly insignificant price).
This tech may be a negative thing for the market of recorded music, but it needs to be argued that recorded music is the only authentic way to experience music, and that this is why that's how most people experience music currently, rather than that being an historical anomaly due to the technology available. Once you step away from treating music like it's only valid when it's a product for a market, the problems of AI music seem a lot less catastrophic.
We have had algorithmic composition for a long time. Here is a brief history of algorithmic composition from 1999 https://ccrma.stanford.edu/~blackrse/algorithm.html
What is meant by "AI Music" is not works by Iannis Xenakis or certain Autechre albums.
We should be defining all this better but we won't. It is also that there is no "AI music" equivalent of the amen break to invent new forms of art. The cultural structures and norms that made that possible no longer exist.
It really is the difference though between art and porn. A blurry distinction on paper but quite obvious in practice. Quite obvious in motivation.
Ever since the first electric guitarist who turned their amplifier input up way too high (*), artists have been ignoring the instructions that came with the technology and (mis)using it to make art, and they’ll do it with this one too.
(*) and before, but that’s an obvious example
> new sounds and musical forms
Has it done this? Or does it just make things that sound like what it's trained on?
That's why it's the dream.
I mean, even if it's just a pastiche machine, I do believe that people could use it to make new and interesting music, just like they did with sampling.
But yeah, music is so accessible and there is so much new music all the time that if all, or most, of what AI is being used for is to make even more of the same stuff we're already awash in then banning it is necessary curation.
What I'm hoping for is a (good) musician to take a small music model, train it on a selection of their influences and existing music and then add on a prompt tracklist. Something like a NN album that the fan can add their own prompts and get new music in the style the artist meant for that album.
I don't know if this really counts as art but there are existing AI music that are at least though provoking funny like https://kommandointernet.bandcamp.com/
It is German but there is really funny remixes combining Mallorca Party music (a German music genre of its own) with anti fascist themes. They manage to make fun of both without being IMHO totally alienating both subcultures, which could count as art. IMHO such things would not really be possible without AI. Would be really disappointed if this disappeared from BC.