Sebastian Smee
THE WASHINGTON POST – Can the art world live with Artifical Intelligence (AI)-generated art?
Relax, everyone. It already is. Artists have been doing amazing things with AI and its various predecessors for decades.
The work is only getting better, more interesting, more exciting.
Of course, it’s easy to see why people are freaking out. The worlds of AI and machine learning are changing things up with discombobulating speed. We suddenly have apps at our fingertips that can turn a simple verbal prompt into an image within seconds.
If you’re a graphic designer or illustrator working in certain commercial fields, it’s already clear that AI will be a major disruption.
Platforms like Midjourney and Stability Diffusion have built their businesses by scraping the internet for the data sets then used by their generators.
That material includes the work of artists and illustrators, almost none of whom have been asked for their consent, credited or compensated.
Defenders say the approach of AI companies falls under fair use because the results, like Picasso’s riffs on Manet or Delacroix, are transformative. But artists and illustrators feel violated and exploited. None of these apps, they point out, would work nearly as well without their skill and creativity, their life’s work.
These legal and ethical questions will take a while to sort out.
But in the meantime, if you’re concerned about the health of art as we know it, there’s little reason to think of AI as a threat.
Why? First, because the easier it is to get software to spew out digital imagery in response to a verbal prompt, the less interesting that imagery becomes.
The same thing happened with NFTs. Invented as a device to create artificial scarcity, they were so easy to make that they produced the opposite of scarcity: a deluge of supply and a subsequent loss of interest.
Second, because humans feel the pull of the physical. The more dominant the virtual becomes, the more we crave the physicality of art. That’s not just hopeful, old-style humanism.
It’s a clear phenomenon. Even as the digital possibilities get greater and more sophisticated, the art world has seen an undeniable rise in the popularity of physical materials – not only paint, but also ceramics, textiles and all kinds of sculpture, all of which are undergoing a noticeable revival.
I saw this with my own eyes at the Venice Biennale. I saw it at Art Basel Miami. I see it every week in museums and galleries.
Physical art pulses and glows before our screen-addled eyes with a kind of talismanic intensity.
So, if you’re an artist who makes sculpture, oil paintings, ceramics or textiles, if you’re into printmaking, watercolors or immersive, physical installations, you have nothing to fear.
Instead of thinking of AI-generated art as a doomsday development – a cluster-bomb thrown by Big Tech into the heart of the art world – you can think of it as something with its own fascinating history, intoxicating present and unknown future.
Something to be curious about.
Case in point: This winter, crowds have been congregating in front of an early masterpiece of AI-generated art called Unsupervised.
It’s showing, and has just been extended through April 15, on a big screen in the atrium of New York’s Museum of Modern Art. Most times of day, there are more people looking at it than there are in front of Van Gogh’s Starry Night.
The work is by Refik Anadol, an artist born in Turkyie and based in Los Angeles, whose work was used as backdrops at this year’s Grammy Awards. Anadol, 38, started making AI-generated art seven years ago during a residency at Google. (But in this field, as Anadol said, “seven years is like 70 years.”) He has a soft, round face, his default expression is a beaming smile, and he has an amazing ability to convert Cassandra-like prophecies and curly ethical questions into causes for quiet optimism.
Unsupervised, which is actually one of three works Anadol has on display at MoMA, uses machine learning to “interpret” the museum’s permanent collection or, as Anadol likes to see it, to “dream” about modern abstraction, about what might have been and what might be to come.
Just as AI-generating apps like Dall-E and Stable Diffusion “scrape” the internet for their source material, Anadol has fed 200 years’ worth of images from MoMA’s collection into an AI algorithm.
The result is a film in constant flux, with no beginning and no end. It shows one kind of abstract image morphing seamlessly into the next. Straight lines morph into sinuous curves before fading out and being replaced by complex matrices or fields of color. Dense, dark images, resembling liquid drops into a pool of mercury, shift into translucent, barely visible vertical curtains of orangy pink which, seconds later, have turned into a Jackson Pollock-like field of black scribble.
Constantly renewing itself, and constantly changing scale, the piece also responds to crowd movement, weather and other external stimuli, and it never repeats.
The only things detracting from the wondrous effect are the speed and extremity of the changes (which, like the Internet itself, can induce an overwhelming sense of arbitrariness) and the accompanying, New Age soundscape, which seems harmless enough until you plug into it and realise that it’s pure, manipulative kitsch.
According to MoMA curator Michelle Kuo, Anadol’s work “could not be further from some sort of input-output situation, like, ‘Show me a watch in the style of Van Gogh.'”
Indeed, Unsupervised makes those image-generating apps look like gimmicks. “Image prompting” with verbal prompts, he tells me, is “exciting but it’s not truly art making”.
Anadol began making “Unsupervised” by uploading lots of data. MoMA had made that possible in 2016 when it uploaded to the open-source program GitHub more than 140,000 records, representing all the works in the museum’s permanent collection and catalogued in its database. The records included such basic metadata as each work’s title, maker, medium, dimensions, date made and date acquired.
Anadol and his team in Los Angeles then trained the AI, using what he describes as a “deep level custom algorithm” that combines chance and control.
The first thing they did was remove the metadata categorizations. Although these were designed to be helpful to researchers, for Anadol’s purposes they were “a very human way of looking at things.” He wanted to know what would happen “if there were no categories, if everything became unified and could find a completely new form.”
Hence the work’s title. In AI, “unsupervised learning” identifies patterns without resort to labels and classifications.
The decision led to an artistic breakthrough. “When you don’t use labels,” Anadol said, “a painting in the collection can become a sculpture or even a video game.” (There are games in MoMA’s collection.) “It was a beautiful moment.” Scholars of modern art who see the work are fascinated by connections they wouldn’t otherwise have thought to make, he said.
“It’s creating a new intellectual discourse, unfolding new ways of seeing.”
Over many months, Anadol’s custom machine-learning model created an incredibly intricate map of MoMA’s collection.
This map, explains Kuo, “exists in exactly 1,024 dimensions. In between the clusters of information is a kind of empty or ‘latent’ space. ‘Dark matter,’ if you like. What Refik’s work actualises is flying through that dark galaxy of latent space and saying, ‘Nothing exists here, but what might exist here?’ That’s the dreaming aspect of what we’re seeing. You might think you’re seeing an artwork you know, but you’re not. You’re seeing what’s missing in the latent space.”
Human-machine loops are not new in art. Artists have always used technology to do things they could not do themselves or simply to see what would happen.
In the late 19th Century, John Singer Sargent needed a brush made from bristles, which he loaded with a particular quantity of viscous paint before dragging it across a primed canvas with different pressure and velocity to get results he wanted but could not entirely predict.
Decades later, Gerhard Richter, fascinated by the role chance plays in the way paint is applied to canvas, used a giant squeegee to drag huge dollops of coloured paint across his canvases.
A small quotient of unpredictability in the brushstrokes by Sargent turned into a large quotient of randomness in the work of Richter. Both artists were involved in a kind of human-machine feedback loop.
Anadol likens his AI algorithms to a “thinking brush”
The important thing, he said, is “designing the brush.” “Some people believe it’s a case of ‘Hey, here’s the data, here’s AI, voilà!'” he said. “But it’s actually more challenging when you start to have some control over the system instead of having something imposed on you. That’s where the true challenge of art creation comes in.”
Anadol is interested in the tension – or what he calls “this beautiful dance” – between chance and control. He works in a tradition that constitutes its own distinct narrative within the various histories of modernism told by MoMA.
On the control side, he traces that narrative back to the deeply deliberate application of pixelated paint by Georges Seurat. On the chance side, you can find antecedents to what he is doing in the work of Pollock, Marcel Duchamp, John Cage and Ellsworth Kelly.
As a student, Anadol immersed himself in the California Light and Space movement. He then studied systems and computer art. He enthusiastically acknowledges his debt to key figures such as Sol LeWitt (who famously said, “The idea is the machine that makes the work”) and such computer art pioneers as Vera Molnár, Peter Weibel, Casey Reas and Jeffrey Shaw.
Of course, Anadol is only one of thousands of artists today working with AI. Whether any of their creations transcend gee-whizzery and prove powerful enough to magnetize deeper meanings to it will become clearer with time. But like Christian Marclay’s celebrated The Clock (although not quite in the same league), Unsupervised is genuinely mesmerizing. It stimulates all kinds of musings on time and creativity and on the relationships between the general and the specific, the visible and the invisible.
Making the invisible visible, or at least using the visible to invoke the invisible, was the spiritual ambition of some of the earliest abstract artists, including Hilma af Klint and Wassily Kandinsky. Creating a language that might be universal was a utopian social ambition embedded in the work of their abstract contemporaries, Piet Mondrian and Kazimir Malevich.
Neither spiritual nor social utopianism fared well in the 20th Century. That’s why the history of modernism is often cast as the history of a failure. But these failures weren’t total. And developments in AI, if they don’t lead to disaster, may prompt us to rewrite those histories in a more positive light. Kuo told me that some of her curator colleagues watch Unsupervised and say, “‘Wow, it’s like it’s inventing abstraction before our very eyes!'”
While speaking about daydreaming, which he said is his favourite state, Anadol said, “It’s amazing when our minds transform something into something else.” That, at its simplest, is exactly what “Unsupervised” does, continuously. “It takes information and transforms it into new potentials,” he said.
The question now is, what are we going to do with so much potential?
