Just a few years ago the financial gravity in the tech industry came from cryptocurrency. This was where a tiny group of VCs had chosen to put their money, which meant we all had to take notice whether or not we wanted to. Among my friends, there was a sense that this was the current Thing, but we talked about it with derision and humor. The underlying premise was interesting, but everything else seemed like a cash grab.
Partly because of this, I spent most of my non-work time cultivating other hobbies. I particularly took to music production, something I’d tried and failed to take on repeatedly in my 20s. I spent hours turning knobs and patching cables, mixing and EQing tracks, and building chord progressions that fit some mood I couldn’t articulate. I finished some songs that sounded bad; I polished them; I finished some songs I actually liked. I mixed an album. It was (and maybe this says something about the extent of accomplishment in my life) one of the most gratifying things I’ve ever done.
Then a little less than a year ago AI began to take off. Part of this was that same financial gravity being moved by those same VCs. AI had been around for a while after all. But what got people excited was the release of image generators.
Today the tech-focused conversations I have are imbued with a deep seriousness. Of particular interest is what tools like image generators mean for artists. Kyle Chayka recently wrote that “You could say that artists are losing their monopoly on being artists.” And then later, “anything generated by AI and instantly made public is by its nature utterly derivative.” These are effective distillations of two ends of the discourse. (I’m pretty sure he falls more toward the latter and that the first statement was not said with excitement.)
Personally I’ve found image generators to be extremely exciting. I agree that most of the work they create is worse than a Kinkade painting from Walmart, but I love their capacity to combine styles with extreme speed. A hypothetical artist maybe could have staged all the shots in the AI generated Jodorowski’s Tron, but it’s extremely unlikely it ever would have happened without Midjourney. Even more intriguing is Tim Molloy’s work in recreating Hasturon’s Dilemma, which robs no particular artist or style and is accompanied by a wikipedia-like, matter-of-fact description of a long-running sci fi soap opera that doesn’t exist. (Click in for the description.)
But when I’m confronted with AI audio generators, I get a little more queasy. There currently aren’t many of them, but the ones that exist allow people to plug in things like “deep house sunrise set” and get a 30 second clip of semi-confincing chillout music. Suddenly the words “utterly derivative” slot into place.
I recognize how hypocritical this is. If it’s visual art, something I can’t fathom doing, I’m overcome with a sense of possibility. But when confronted with an AI that takes all the work I’ve spent honing a craft and compressing it into matrix multiplication, I recoil.
So what’s fair? To answer that, I think I need to try to answer a different question. How is this (generative AI) different? What I mean by that is, how is this technology different from all the technologies that have come before? No one assumes photography isn’t art, but I can imagine a time (I’m not going to bother researching this) when painters looked at what photographers were doing and found it to be too quick and mechanical to be any form of creative expression. And what did drummers think of the 808 when it dominated music in the 80s? In its most basic form, the drum machine was a sad, repetitive imitation of a trained session musician. But programming those same consistent drums laid the ground work for a kind of pop music that wasn’t previously possible.
Then there’s sampling. Taking a previously recorded track (often without consent) and then looping it, has become one of the most common techniques in huge genres of electronic music. Hip-hop, house, and disco all did this to great effect. Some musicians at the time recoiled, but others (George Clinton, David Byrne) embraced it.
If what all these technologies did was simplify something that had once required skill, they also allowed artists to focus their energy elsewhere and create something entirely new. If you visit Tim Molloy’s instagram above, you’ll see he’s already a pretty skilled visual artist. What I find compelling about his work with AI is that he’s incorporated its quick, mechanical speed into a larger work of fiction. Instead of telling the generator to make “drawing of an alien, by Tim Molloy”, he’s gone for something more transformative.
So is AI different than the ability to capture a portrait perfectly without having to study human anatomy? Is it different than the ability record and replay music without having to learn and play it yourself? Hitting “play” on Spotify is hardly a creative act, but DJ’ing surely is.
I don’t really know if I’m ready to settle this question. It does feel like there’s something different about AI – especially when it blindly lets you pull from millions of other works without attribution. It feels like we’re about to see a wave of the most derivative “art” imaginable. Visit https://www.midjourney.com/showcase/recent/ and you’ll see hundreds of elf-like fantasy women, cute Pixar-ripoff creatures, and Painter-of-Light landscapes – all the products of a desire for instant gratification – one more work from an artist that never wanted it.
But I don’t think this sort of creation is zero sum. The existent of all this garbage makes compelling art all the easier to spot. And it’s very easy to say “picture of a fantasy landscape” and get exactly that. It’s much harder to ask for something specific; “4 trees with branches that spell out Nature” will never give me what I ask for.
So I guess I’m cautiously optimistic? Maybe I’ll incorporate AI into my music? Maybe there will always be an audience for songs that are more hand crafted? Hard to say. I’ll always have the process.
Just having visited the Rijksmuseum with it's unfortunately sold-out Vermeer exhibition, I had to settle for the permanent collection. It arcs through centuries of medieval religious art with really weird and janky perspective all the way to super high skillcap renaissance vanishing-point style stuff. You can feel hundreds of years of painters meticulously honing skill at perspective, light and shadow. Amazing.
Meanwhile, in the side of the museum that I couldn't get a ticket to, Vermeer just used a camera obscura to mechanically plot out perspective - bypassing all of that - investing his skill points differently. It must have been infuriating to all the others.
I wanna say, right now anyway, that I feel like AI is different from crypto at it’s base - mostly because “making things that evoke an aesthetic response” is different from “avoid what third parties want when making agreements”. My impulse is that one is generally good but could have bad externalities (people like to enjoy art and learn about trees, it’s a bummer if the art starves out better art or there are lies in the paragraph about trees) and one is generally bad but maybe occasionally good (in general I *should* recognize through taxation and regulation and common law that my agreements are part of a common social mesh, but it’s nice to evade social orders we need to duck under if we’re fleeing persecution). I feel like AI is just more like video streaming and HyperCard, another (possible) mind bicycle. This is super subjective and in no way career advice, but it just seems a little better in its promise, however and whatever it eventually delivers.