Listen to this story
In a tiny third-floor apartment in central London, the artist Anna Ridler was hunched over her desk. Next to the desk was a black PC tower whirring noisily and in front of her was a keyboard and a screen covered with a dense thicket of black-and-white code. As Ridler tapped in a few keystrokes, a window sprang open. In it were 16 images of tulips in jewel-like hues, arranged neatly in rows, like a page from a botanical textbook. Ridler inspected them skeptically. “Getting there,” she muttered.
It might not resemble the romantic image of an artist’s studio, but machine learning technology — and the computers that wield it — is becoming an important, even essential tool for many artists in the 21st century. When I met Ridler last month, she was at work revising a project for the forthcoming exhibition AI: More than Human at London’s Barbican Art Gallery. She had fed the machine with a series of digital photographs of tulips she took last year, then instructed an artificial intelligence (A.I.) algorithm to analyze and attempt to replicate them. Ridler’s aim was to make an infinite series of digital flowers — not copies, exactly, but automatically generated simulacra assembled from the fragments of the originals. Ridler wanted them to be plausible enough to fool the unwitting eye: “My version of a Dutch still life,” she explained, created for the age of computerized automation.
“But it’s art — it’s not meant to be real, is it?”
Except, technically, it was her computer that was doing the creating. Every few hours, the program would use a Generative Adversarial Network (GAN) to learn and refine what it was doing: teaching itself to draw, step by painstaking step. It was like watching a child learn the rudiments of art, but at terrifying speed.
Together, we watched as the hard drive whirred and spat out another batch of tulips, this one more eerily perfect than the last. The whole thing — the images, the technology — looked a little unreal.
Ridler smiled brightly. “But it’s art — it’s not meant to be real, is it?”
In recent months, the art world has been frantic with speculation — and more than a little consternation — about the possibilities and perils of A.I. In late October 2018, Christie’s New York became the first major auction house to sell a piece of A.I.-generated art. It apparently didn’t matter that the picture looked like a Rembrandt put through a washing machine — the work by the French collective Obvious sold for $432,500, 43 times its estimate.
In early March, Sotheby’s sold a diptych, by the German-based artist Mario Klingemann, entitled Memories of Passersby I: two ultra-high resolution video screens attached to a chestnut-wood box housing an “A.I. brain.” Having been trained on figurative portraits from the 17th century through the 19th century, the “brain” — an algorithm stored on a solid state drive — employs this data to paint a never-ending stream of portraits of its own.
Can computers really produce art? Isn’t culture one of the things that makes humans human? And what does all this do to our notions of what “culture” actually is?
Google has been pushing the boundaries of visual arts tech for years, starting in 2015 with Deep Dream, an A.I.-based image processing that gives existing pictures a surreal, hallucinogenic appearance, onwards. In late February, the British art dealer Aidan Meller commissioned the world’s first “humanoid A.I. robot artist” — an eerily lifelike automaton able to draw portraits using microchips embedded in its “eyes.” It’s scheduled to go on display in Oxford, England, in May.
It feels like a turning point not just for people in the art world but for anyone interested in what creativity will look like in the future. Can computers really produce art? Isn’t culture one of the things that makes humans, human? And what does all of this do to our notions of what “culture” actually is? Forget factory jobs lost to automation and cars that no longer need human drivers — eight decades after Walter Benjamin wondered whether art could survive the age of mechanical reproduction, are the robots coming for painters and sculptors, too?
Yes and no. No and yes. Ridler began exploring machine learning techniques as a way to extend concepts she had already taken on as a human artist: repetition and reproduction, the way images decay and become distorted, the role of fallible memory. She taught herself the general-purpose Python programming language and then created her first A.I.-based project, Fall of the House of Usher, in 2017. Using ink drawings she made in response to the 1928 silent horror film of the same name, she trained a neural net on her sketches so it could produce its own version of them. The result is a dreamlike replay of scenes from the movie, as if cloudy half-memories of it are dissolving before our eyes.
Another of Ridler’s work that year was called Drawing with Sound. While she sketched charcoal drawings, a computer trained on another neural net, and watching her via webcam, translated the marks into a throbbing chorus of human voices. It was somewhere between drawing, musical composition, and performance art — a powerful coupling of machine and human ingenuity.
As we talked, it became clear that Ridler regarded A.I. technology more as a technique or tool, rather than a replacement for anything she was doing. Yes, technically the machine was autonomous — but that’s far from genuine artistic independence. “Algorithms are hugely powerful these days, but they’re only as good as what they’re trained on,” Ridler said. “It’s the difference between the noun and the verb ‘draw’: computers can make drawings, but they can’t draw, you know? It’s enabling me to do things I wouldn’t normally be able to do, and I’m enabling it to do things it wouldn’t know how to do.”
Much like the land art created by the likes of Robert Smithson or Nancy Holt — in which artists carved striations in the earth or placed concrete structures in the desert before leaving wind and weather to do their work — Ridler saw her job as creating an interesting data set that the algorithm could then respond to. “You sculpt the land, set up the variables, then leave what you’ve created to the elements,” she said.
“What can I say? There are a ton of spreadsheets.”
As for the possibility that A.I. could put her out of a job, Ridler laughed. In her collaborations with A.I., she’s been doing most of the work, creative or otherwise. “The data set for Mosaic Virus was 10,000 photographs of individual tulips I bought from the market, each of which I had to take, process, and tag with metadata by hand,” she said. “It took months; I was dreaming tulips. And the code’s not that stable — if it goes wrong, you have to start the whole lot again.”
“What can I say? There are a ton of spreadsheets. My boyfriend thought going out with an artist would be way more exciting.”
Still, the possibilities of A.I.-generated art are dizzying, and changing all the time. In January, the Rhizomatiks research group in Japan unveiled a piece in which a dancer performed alongside an A.I.-generated doppelgänger that was projected on stage. The algorithm responded live to the human, creating a haunting duet.
Mario Klingemann, the most high-profile artist currently working with A.I., has pioneered what he calls “neurography,” a form of camera-less photography that links together trained algorithms to generate surreal results. His work that recently sold at Sotheby’s could hardly be described as radical, at least in terms of its output: It’s a series of figurative portraits, generated from a data set of portraits by old masters, but other works are more enterprising. Elsewhere, Klingemann has deployed imagery collected by electron microscopes and Instagram, and introduced deliberate glitches into his GANs to make them more unpredictable and spontaneous.
Klingemann, and others working with A.I., may seem more like curators than artists, selecting and assembling materials partly created by other forces. But the creativity of these projects is hard to deny.
Ben Vickers, curator and the chief technology officer at the forward-thinking Serpentine Galleries in London, says that the introduction of A.I. has created a fascinating moment for art — albeit one that’s difficult to read. On the one hand, argued Vickers, all this can feel shockingly new; on the other hand, we’ve been here before. Artists working with A.I., or earlier computer technologies, have often been treated with suspicion by mainstream museums and galleries, but so were conceptual artists of the 1960s and 1970s. Many of these free thinkers were ridiculed for turning away from “skill” and “craft” in favor of ideas. And they also challenged popular assumptions about what art should be, or how it could be made, in ways that many people found uncomfortable.
“There’s definitely a tension around where contemporary art currently stands,” Vickers said when we spoke. “But then maybe it’s part of a much wider moment: Technology has created this huge destabilization of what it means to be human.”
We may be nowhere near the kind of general A.I. that can create autonomous art — or autonomous anything else, for that matter — but the advances in neural networks over just the last six months are beginning to change our understanding of what art actually is.
“We are now in a situation where you can process every painting that Turner ever did, train the data set, and produce new Turners,” Vickers said. “You can automate art, at least in that narrow sense. But think of the 1960s, what Warhol was doing with pop art and mechanical reproduction — it’s not dissimilar.”
The Tokyo-based curator Maholo Uchida, who is cocurating the Barbican’s A.I. exhibition, agreed. “Think about Marcel Duchamp installing a toilet in a gallery in New York and calling it art,” she said. “It was so radical, and it happened in 1917! A.I. is still catching up with that.”
Where do we go from here? Uchida’s view is that A.I. techniques will eventually become just another technology for artists to draw on, though it is entirely possible that the creeping prevalence of A.I. in every area of life, from chatbots to smartphones, will encourage artists to go the other way and rediscover artisanal, handmade techniques. As we now live in an era of always-on digital streaming, our hunger for live experience — the one-off gig, the exclusive performance — is seemingly more powerful than ever before.
Still, A.I. might produce an entirely new species of art, as yet undreamed of. “Maybe art doesn’t need to just be for humans, to please our sense of what is beautiful,” Uchida said. “What would A.I. produce if it was making art for itself?”