Play your part
Remixing my thinking on AI-generated art.
If you’re anything like me, your feeds have been flooded with a torrent of AI art. Photorealistic images of impossibly contradictory things like teapots on cats on dinosaurs; ethereal cyber-elves glowering at the Marvel Cinematic Universe; Burning Man vanlife by way of Dieter Rams.
The backlash has begun. Artists are filing class action lawsuits, claiming copyright violation. Provenance is complicated, and these tools use millions of original works and billions of humans’ uploads without credit. We have no idea what “fair use” looks like for AI.
Maybe the novelty has just worn off. Whatever the reason, the unrelenting hallucinations have become background radiation, and everyone’s switched their Lensa avatars back to actual photos, and is now busy sharing breathless advice on ChatGPT prompts instead. AI art is definitely here to stay, but its 15 minutes of populist fame may have peaked.
Except there are still a few people I follow who keep posting their art.
And it does seem like their art.
Every new instrument finds its virtuosos. The Roland 303, originally conceived as a bass accompaniment for musicians, launched techno and acid house (think Daft Punk’s Da Funk.) The TR-808 drum machine defined eighties music (think Marvin Gaye’s Sexual Healing, or Run-DMC’s Tricky, or New Kids On The Block’s Please Don’t Go Girl.)
The 303 had Josh Wink. The 808 had Afrika Bambaataa. And a software tool called AudioMulch had Girl Talk.
Girl Talk is Gregg Michael Gillis, a DJ who makes epic mashups built from hundreds of samples. He made money touring in legendary, energy-drink-fueled nights (and later at some pretty big festivals,) but released his bootleg mashups online for free because there was no way he could secure all of the licensing rights.
Gillis was a technologist, and his packed songs were catnip for the early days of online music sharing. There’s a realtime visualization of the 300+ samples from All Day. Over time, it became a badge of honour to be sampled by Girl Talk.
Samples have been a mainstay of music for decades.
Sometimes sampling is subtle, even unrecognizable, breathing new life into a few notes from a forgotten B-side the way Daft Punk or DJ Shadow do it. Sometimes sampling is blatant and derivative—think Pitbull using Toto’s Africa for the Aquaman theme song—because nostalgia sells.
Girl Talk is neither. It’s mashup. The samples aren’t hidden, they’re celebrated. You can hear all the parts—indeed, that’s one thing that makes them so fun. What’s new is the way they’re combined. The original artists would never have dreamt them up by themselves. The sum is greater than the parts.
That’s how I feel about some of the people posting art in my feed.
I know it’s not their work. And when they namedrop a particular artist in their prompt, the results are undoubtedly derivative. But sometimes the results are a brilliantly original mashup: Animals and musical instruments; video games and retail storefronts; cats and astronauts. These artists are coaxing, filtering, and curating, playing AI whack-a-mole, chasing an image they didn’t know existed but were certain they’d find. And it’s good: I’d buy prints of some of their work, or at the very least, make it my Lock Screen.
Maybe, I think, maybe they’ve just found their medium.
Maybe, I think, maybe they’re prompt virtuosos. Playing with ideas and contradictions and experimenting in ways that are novel and provocative and entertaining enough to their fellow humans for it to be a new art form.
Maybe, I think, maybe I’d watch a live performance.
Maybe, I think, maybe I’d go to a gallery. One where the artist spends six months crafting prompts around a theme—global warming, first love, Jodorowsky’s Tron—and then hosts a viewing for twenty nights, each with a new, random seed.
Maybe, just maybe, the original artists won’t mind being sampled. Maybe they’ll collaborate.
I’ve stopped being annoyed at these budding artists who push art at me. I’m happy they’re still in my feed. Now that the dilettantes have tired of the novelty of generative art that somehow all looks so bland and similar, maybe we’ll see the career of this new medium’s virtuosos.
Girl Talk didn’t have rights either, but somehow it all worked out.
Postscripts
I’m not saying anything about payment here. Artists aren’t being compensated for their work, and uploaders aren’t being credited for their training data, and that’s wrong.
It can be a prosthetic mind’s eye. I linked it above, but it’s worth pointing out that like Scott Kildall, I have Aphantasia. I literally cannot close my eyes and summon an image of my daughter’s face. Like Scott, text-to-image AI is a way to visualize externally what I cannot imagine myself.
Girl Talk’s latest album. Last year, a full decade after his last album Girl Talk released Full Court Press with Wiz Khalifa, Big K.R.I.T., and Smoke DZA. He’s no longer uploading can’t-license bootleg mashups; he’s streaming on Spotify.
Prompt, training data, and labels. There is a significant difference between including “Kurosawa” in the prompt, having Kurosawa’s art in the training set, and labelling something in the training set “Kurosawa.” Each has very different financial, ethical, inspirational, derivative, and artistic angles. We conflate them badly today.