Machine Learning
... last call at the No Future Future Club
For years until the trend changed, I walked by fake ‘Anti-Social Social Club’ hoodies in the Camden street market near my flat in London. Maybe it wasn’t the trend that made them disappear, but the plague, which made them true—we were in anti-social social clubs, grouping together six feet apart in public parks during two lockdowns. And so no one wanted one, anymore, the hoodies, because we wanted to pretend we were going back.
For months I have joked mordantly to friends that I want a ‘No Future Future Club’ hoodie. This is because my generation of scholars and would-be-layabout intellectuals and artists doesn’t feel like it has one. Even before we were all on strike on both American coasts and across Britain, the university was coming apart at the seams. Everyone had no job or three jobs, teaching a 5-5 adjunct load at last notice driving a 2007 Honda accord on shit interstates, or taking late subways, or topping it off with sex work, or applying for fellowships living in the spare room filled with their high school achievement awards in a once-comfortable suburb slowly growing dark around the edges. There were no jobs. At one point there were about five jobs, so we could lie to ourselves. But now there were no jobs. And we had to stop lying.
Scattered to our last postings, we tried to comfort each other on Twitter. Some people got out, into freelance or nonprofits, and made lives. Some didn’t, or couldn’t. I think about them, the suicides, a couple now, a trend, like that Ginsberg line that felt so hyperbolic it was almost parody: I saw the best minds of my generation/ destroyed by madness/ starving, hysterical. Except it didn’t feel hyperbolic now, it just felt true.
On nights when I can’t sleep and I’ve applied for all the things and I can’t see past the month, I play with what the internet mostly calls “AI art” on my phone. It’s not “artificial intelligence” in any general way, it’s actually machine learning, specifically a process called CLIP Diffusion that makes the images. It’s got a black-box kind of feel, a way like you’re touching the edges of concepts with your hands, but that’s probably because you can’t really picture a vector n-space, or the results of Gaussian noise unblurring, or the linear algebra you never learned because you fell in love with ink on vellum and the quiet reading rooms of manuscript libraries. That’s me-you, of course, maybe you-you stuck it out and became a topologist or logician, and has tenure somewhere and publications and a middle-class life and wonders about that novel at the same time me-you is up playing with machine learning.
I like using an app called Midjourney, which if you don’t know how to control it, tends toward Caspar David Friedrich run through a taffeta-cotton-candy factory. In the public channels, you can see the prompts that end up making second-rate videogame box fantasy art, and you feel a little bad even, for Caspar David Friedrich, and his young men standing on mountains waiting for trees to get hit by lightning. Is it strange that Schubert’s Winterreise is starting to resonate with me now? Maybe German Romanticism only makes sense when you’re desperate.
Anyway, Midjourney can be beautiful if you know how to whisper to it. “Prompting” is what people call it, entering words in the unnatural pidgin-English and technical terms of the software to bend it in your general direction. Because Midjourney style terms are often art historical, I am very good at prompting, better than even a lot of the tech people I know. It seems like a technical accomplishment; I tell people I know about machine learning. It’s the kind of thing that might just save me, a line on my CV that is the kind of knowledge that feels like it belongs in a future. But the truth is I can only get Midjourney to make me midnight Bauhaus coffeeshop dioramas, Tang dynasty robot porcelain, or a spaceship peeking through semi-abstracted Altdorfer trees because I’m a necromantic, somehow still historicist-materialist to the core. Machine learning algorithms for image diffusion are all trained on the past. They’re in the No Future Future Club too.
People talk about apps like Midjourney replacing artists, like they talk about GPT-4 replacing writers. Which is to say, I don’t see it happening. It’s not that the images it makes aren’t beautiful, or that they might not replace some stock illustration. It’s that they’re not intentional, there’s no there, there, unless the human doing the prompting thought about, inserted a conceptual frame to make it mean. Maybe I’m overly optimistic about what viewers ask of things. I don’t know. I know that I made the images for this Substack—its banners, and logos, on Midjourney myself as a kind of challenge. Then again, I’ve always made my own visual design, been particular about typography, and hand drawn sketches. I still draw, sometimes during lectures, sometimes alone.
But anyway, back to the No Future. It’s 3 AM and I’m lying in bed adjusting single prompt words and re-rolling randomizations. My friends are panicking on the internet. There are no jobs. We can’t see past the ellipsis of the dark. And I recite the dimensions of the image I want to myself in pixels, like one of those midnight monastic litanies, learning to talk to the machine.
