NEW YORK, NY.- Color spills into the lobby of the Museum of Modern Art, radiating from a 3D animation of sloshing, puffy liquid fizzing in ectoplasmic earth tones and rancid pastels. This is Refik Anadols Unsupervised, and the roiling shapes greeting visitors from a 24-foot-high wall of LEDs are the museums collection of art, design and photography analyzed, digitized, fed to an algorithm and returned as spectacular slurry.
Anadol, 37, is a digital artist known for using artificial intelligence to turn piles of data into spectral, abstract visuals, often at huge scale. To do this, he has partnered with the likes of Microsoft, MIT and the Jet Propulsion Laboratory.
In 2018, for example, with Googles help, Anadol projected mind-bending pictures based on the complete archive of the Los Angeles Philharmonic onto the exterior of Disney Hall. An upcoming show at Jeffrey Deitch in Los Angeles promises wavelike imagery stirred by observations of the Pacific Ocean.
This time, its MoMAs collection behind the globs and blobs. Informed by nearly every object in the museums collection, Unsupervised automates a modernist notion of progress: Input art history, output the next avant-garde. This context draws out some of the most interesting questions around big data, AI and creativity. How should we process the past? How can MoMA, the mausoleum of yesterdays radical art, stay relevant? Not least, the impulse to make computers do something artistic speaks to our anxiety around what makes us human: If software can create art, then whats left for us?
Its a fearsome question. Anadol doesnt answer it here. In his positivist vision, both progress and existential doubt have the breathless optimism of a TED Talk. Anadol describes his work as machine hallucinations and dreams, which is apt; the software interpolates the images its been given in a way that seems extra-real, vertiginous and wrong. In the argot of AI engineers, a program hallucinates when it generates falsehoods.
Unsupervised belongs to a burgeoning class of software known as generative adversarial networks, or GANs, where one part of the code tries to trick another part with imitations of whatever its been trained on, whether profile pictures or Seinfeld episodes. Here, software weaned on images and metadata names, dimensions, materials of MoMAs holdings generates what it determines are new works of art. These creations are tweaked using real-time data like local weather and the motion of visitors, and are displayed on the media wall in one of three morphing styles (officially, three distinct works): mottled blooms, skittering rays or molten Floam. A cosmic, swishing soundtrack issues from a pair of speakers.
In the past few years, as GAN-based programs like OpenAIs Dall-E (get it?) and Googles Deep Dream have become public, many users have tested the apps ability to reproduce or revise the greatest hits of art history, usually in the oversimplified way that immersive experiences riff on Vincent Van Gogh or Frida Kahlo.
Photographer Torbjørn Rødland, meanwhile, has used Dall-E to re-create masterpieces of photography and photo history from Henri Cartier-Bresson to William Eggleston that reside at MoMA, but with a more critical eye.
His experiments dont celebrate the technology but demonstrate its shortcomings and sullenness. Try asking a text-to-image AI for the decisive moment. Whatever the technology sutures together will include the unsatisfying shadow of Cartier-Bresson, the street photographer who coined the concept just as the contours of popular works appear when you type Picasso or Kusama, trail cam or fisheye skateboard video.
MoMA founding director Alfred Barr and first photo curator Beaumont Newhall started acquiring and exhibiting photography before the technical mediums status as fine art was settled. The museum is doing the same with digital media, even as similar skepticism toward machine-made imagery echoes around AI-generated work. (Anadol and MoMAs curators alluded to this reprise of history in a public conversation.)
In September, MoMA announced the sale of masterworks by Francis Bacon, Pierre Bonnard and others, under its care since 1990 (though not strictly in its collection), to endow its expansion into digital art. Theyve also dipped a toe into NFTs with alternate versions of Unsupervised.
Traces of the museums 82-year-old photo department are churning in Anadols data, including pictures by Rødland. Video games like Myst and The Sims, which Paola Antonelli, a senior curator, started accessioning in 2012, are represented there too (several are currently on view in a sharp collection show of interactive design), as are the succession of modernist paintings that used to hang where Unsupervised now mesmerizes patrons and their children.
Presumably, the project also incorporates provocative subjects sex, violence, racism the long and torturous 20th century as recorded by its artists. But Unsupervised is so abstract youd never know that Faith Ringgolds American People Series #20: Die, a bloody tableau of a 1960s uprising, may have provided shades of yellow, brown and red.
In fact, MoMA says the software is designed not to produce recognizable figures. (Other public-facing GAN products similarly avoid potentially sensitive material.) It could be anything. The works enveloping abstraction scuttles the possibility of conceptual dignity offered by an institution founded to embrace challenging new art.
Then, every few minutes, the big screen flickers, as if straining, and the soothing colors turn into an array of charts tracking things like energy difference and the current latent noise field. Under GAN video analysis there are thumbnails of the jade lumps, gold lozenges and coal-colored squares dreamed up by the software. Its like the way hacking is portrayed in a Hollywood film like Jurassic Park, with slick but simulated interfaces so as not to bore audiences with the reality of coding.
At MoMA, home to some of the smartest art in recent history, these theatrical peeks under the machines hood feel patronizing. Does it really need to flicker? Is it working that hard?
On the contrary, Unsupervised suggests that creating the next avant-garde should be easy, if you have enough computing power. Its as if AI can do our hallucinating and dreaming for us, minus the bad trips and nightmares. But the bad trips and nightmares are still there. Unsupervised is only a screen saver.
Refik Anadol: Unsupervised
Through March 5, Museum of Modern Art, 11 W. 53rd St., Manhattan; moma.org.
This article originally appeared in
The New York Times.