Vinyl, then digital… now meet meta-DJing. OÍR by Moisés Horta Valenzuela imagines what would happen if one machine learning-powered DJ set could combine all those DJ sets streaming out of Berlin. And what happens? It slaps. It slaps so hard that someone asks for a track ID.
The receding tile walls. The DJ or DJs centered at the decks. Green and blue light. In a sense, HÖR, the Berlin streaming platform, was already algorithmic – taking over YouTube feeds and related social media posts as a parade of Berlin scenesters appeared almost as if they themselves were generated. (Disclosure: I played one in December 2019. Seems everyone does.)
That might explain why Moises’ work struck a nerve – unreal, trippy, yet also oozing just the vibe and groove you’d expect if you could somehow watch all HÖR videos at once. So this went viral – and I think his term “meta-DJing” is dead on:
Okay, let’s get this out of the way – no one should describe this as an “AI DJ.” There is no autonomous machine intelligence acting as a DJ. On the contrary, the mushy digital mash-up textures on offer here are unique, fresh, and distinctively sound like something that came from Moisés. Part analysis, part generative artwork, part creative remix, OÍR is simultaneously the intentional work of an artist and a machine reflection of a wide variety of streamed DJ sets.
Technically speaking, says Moisés, “the system is a compendium of OpenAI’s Jukebox, trained from scratch, StyleGAN2 for visuals.” “The mixing and DJ ‘transitions’ are done with a MIR [Music InformatioN Retrieval] ‘automatic mixing’ Python script,” he says.
But it’s worthwhile also understanding his artistic intention:
OÍR stems from my ongoing research on AI, sound synthesis, and electronic music.
Since starting my adventure into Deep Learning systems for music a couple of years ago, I’ve asked myself if Deep Learning (AI) is a tool or a medium?
Right now I’ve come to the conclusion that it can be both, and this is what
exactly I’m trying to explore with this project.
When we talk about Deep Learning as medium, there are three particular
processes engaged when working with generative systems: curation of the data, training and monitoring of the algorithm as it ‘learns,’ and generating new synthetic media. Rinse and repeat.
There are a couple of aspects that interest me from this process. Each time you train the AI algorithm, its weights and biases or what it has ‘learned’ change over time — depending on the data you are having it learn from. The algorithm generates patterns present in these vast amounts of images and music, as is the case of OÍR, and these can be changing as the ‘learning’ process continues.
So this quality of a constantly changing and morphing generative algorithm is exactly what I want to explore with OÍR, and what better way to do it than though electronic dance music and techno culture.
I chose a channel as the canvas for the first episode, or EPOCH, of OÍR with a selection from the archive from HÖR Berlin, because I feel this channel has done the amazing job of generating a collective culture, specifically within techno and electronic music. I wanted to explore which patterns are emerging from this culture – which patterns can be synthesized, both visual and sonic, from all these sets and different approximations of techno, over 1,400+ hours and counting.
My desire with this art project is not to automatize or replace DJ’s or
electronic musicians in any way, but rather have OÍR be a sort of ‘live generative archive’, as I did before with my album 𝕺𝖐𝖆𝖈𝖍𝖎𝖍𝖚𝖆𝖑𝖎 in relation to the Mexican 3ball electronic music genre, of certain cultural moments in electronic music which are increasingly existing on big tech platforms and the internet. By the way, OÍR means “to listen” in Spanish.
You could, and many developers do, automate DJing using algorithms and machine learning – track selection, mixing, and whatnot. But those are no fun. They remove context from what a DJ does, which is kind of the essence of what a DJ does – and why it’s been hard to be a DJ during a pandemic. And aside from that, they remove the thrill from the act of DJing. You could write an algorithm to play your Nintendo Switch for you, too, but … why?
What’s exciting about this is that it does something more in the tradition of DJing – it appropriates and abuses a technology intended for predictable playback, and uses it to transform culture. And even as machines choose what we see online, this really re-centers the human – it’s Moisés using the AI tools that conduct social media bending them to his own personal artistic creativity.
It really is DJing with the AI, in the way that DJing was already a meta-activity (playing records or digital music files).
I’m curious to see what happens next. Meanwhile, you can see more Hexorcismos from the CDM archives:
And more links: https://linktr.ee/hexorcismos
Plus also out this week is this terrific article:
Now, for now, please do stay alive, humans. Happy Tuesday.