Which face is real?

Dirt is a daily email about entertainment.

Terry Nguyen, Dirt's senior staff writer, on Bennet Miller's AI-generated exhibition, Harry Potter by Balenciaga, and The Super Mario Bros. Movie.

In 2019, two academics from the University of Washington created a site to help visitors distinguish artificially-generated faces from human faces. WhichFaceIsReal.com is an image generator that presents you with two side-by-side photographs, and asks you to select the real person. The first time I “played” the generator, much to my distress, I mistook the synthetic face for a human one. But I began to pick up on visual clues after a few more tries. The key was to assess the whole image, rather than just the face. 

There are slight discrepancies within the AI-generated image that become more obvious the closer you look. The background, for example, is unusually distorted. Shadows or certain objects, like glasses or hair, have a distinctly liquified texture. I’m not as inclined as the creators of Which Face Is Real to trust that our visual processing systems will adapt and keep pace with software. Depending on who you ask, we seem to have already crossed over the digital Rubicon. Look no further than to the recent viral image of Pope Francis in a puffer. Computers are, to no surprise, only getting better at simulating reality. And if we are constantly skimming our sight—as most of us do online—rather than closely looking at an image, it’s easy to mistake the synthetic for the real. Yet, we retain hope that some base human instinct will prevail, that our eyes will alert us to the artifice within such images, provided we look closely and carefully. 

On Saturday, I went gallery-hopping on the Upper East Side. Galleries, unlike museums, don’t often display much contextual information about the artist or the work on their walls. The viewer typically has to seek out this extraneous information. I try not to read too much about a new show before seeing it as a general rule. I like to draw my own conclusions. It’s also fun to retain an element of surprise, to not let popular opinion (or a critic’s judgment) dampen your critical reception. 

My gallery partner and I had the suspicion that something was off with the prints at Bennett Miller’s show at Gagosian. (The show is at 976 Madison Ave until April 22.) But we couldn’t pinpoint exactly what. Once I scanned the gallery QR code for the press release, it confirmed our uncanny hunch: The “photographs” were artificially generated with DALL-E, a deep learning image generator by OpenAI. 

Miller, who directed Moneyball (2011) and Capote (2005), is a filmmaker who's investigating AI image generators for his next documentary project. From this five-year research period, Miller produced this series of aesthetically coherent, but otherwise equivocally empty images. The prints, as I’ll call them, impart an eeriness upon the gallery; the experience is not unlike entering a haunted house.

The images aren’t photographs because, to borrow Susan Sontag’s definition, there is no real experience or event being captured. There are no stakes attached to the subjects. Instead, they are amorphous vessels for a vibe, a fictive manipulation of feeling. A photo partly derives its emotion from the complicity of the witness. But what we are witnessing is a machine’s visual interpretation of a textual prompt. If the camera is “the ideal arm of consciousness in its acquisitive mood,” as Sontag has written, these image-generated prints are Miller’s banal attempt at replicating a historical consciousness. That alone isn’t interesting, nor is it “urgent” or “revolutionary,” despite the show summary claims. Miller offers no novel perspective or motive for his AI experimentations. 

The grainy, sepia tone dates the images to the early 20th century. There is a Depression-era or wartime feel to them, and many seem to be situated in the midst of disaster: a free-falling plane; a woman plummeting feet first to her death down a ravine; two figures embracing before a desolate landscape; a crouching, Grudge-like figure crawling off a cliff. The falling women recall Russell Corgi’s 1942 photograph “The Despondent Divorcee”: A woman, suspended mid-air in a chillingly balletic posture, seconds before her body hits the ground. Miller’s images are, seen without context, suicidal and frightening. Even so, my shock was tinged with a sense of disbelief: How can a photographer be so primed to shoot on the precipice of crisis? 

The first image I saw in the gallery was of a burning plane, moments from nose-diving into an empty field. It reminded me of Michael Ondaatje’s novel The English Patient, which is set during World War II. The shot was blurry, which led me to think that the photographer was in such a rush to capture the disaster that the focus was neglected. The second print was of a keeling human-like figure, whose right arm resembled a crow’s wing. My partner thought it was a felled scarecrow in a field; I thought it was a human in a Black Swan-like costume for a theater piece. It was too hazy to determine the unfolding situation. 

It didn’t take long, however, for us to sense the irreality behind the spectacle. Something clicked for us in the fourth print: a large-scale, sharply focused portrait of a Black woman (or was she a child?) with a disc-like tray floating in front of her chest. The object looked—and was positioned—like a light reflector, an accessory used by photographers to make sure the subject’s face is evenly lit. It looked a little too modern and out of place. The woman, too, looked unfinished; her gaze was sculptural and distinctly inhuman, devoid of any emotion. I remember being thrown off by the glassiness in her eyes. Later, when I became familiar with the Which Face Is Real generator, I recognized that this liquified immateriality was what cued me into her synthetic personhood. The format, though, was crucial in my seeing through the illusion. The exhibition environment encouraged close viewing.

Recently in The New Yorker, Kyle wrote about the “illusory realism” of AI-generated pop culture. People aren’t really alarmed by the experimentation in a video like “Harry Potter by Balenciaga.” They’re simply enjoying, or mindlessly consuming, the content without much deep thought. We’re only alarmed when we feel like we’ve been duped, as was the case with the Pope puffer. However, increased exposure to, and the proliferation of, AI-generated images will have an effect similar to photographic paresthesia. We may not be as shaken by their existence. In fact, we could even grow numb to them. Still, upon recognition, our skin will crawl. Our eyes will alert us to the off-ness of synthetic body parts, like warped fingers or glassy eyes. We’re going to have to learn to look closer. Seeing doesn’t necessarily mean believing.

PLAYBACK

Snippets of streaming news — and what we’re streaming.

• Variety releases its 2023 state of streaming report. 

• Alissa Wilkinson on Air, featuring Ben Affleck in his entrepreneur era. (Vox) 

• The Super Mario Bros. Movie apparently makes viewers want to play more Mario video games. But it doesn’t make a non-monetary case for a film franchise. (The Ringer)

MIXTAPE

Good links from the Dirtyverse.

• The weight-loss ad boom is just beginning. (Wall Street Journal)

• Claire Bishop on the proliferation of “research-based art,” a genre of art “characterized by a reliance on text and discourse to support an abundance of materials, distributed spatially.” It’s information overload in exhibition form. The audience likely doesn’t have time to sift or read through all the presented information. Bishop argues that, in the post-internet era, search—not research—becomes appropriated, like a readymade:

  • “Artists no longer undertake their own research but download, assemble, and recontextualize existing materials…”

• How the Michelin rating system affects a city’s dining scene. (Eater)

• The new Ralph Lauren store in Miami will accept crypto for purchases. The brand is partnering with Poolsuite to gift customers NFTs. (Vogue Business)

• I liked Dana Snitzsky’s short Chan is Missing review in End of the World.

  • “Identity is inflationary. The self is ever-growing new faces, new forms; but reaches nothing, like a vine growing into empty air.”

 🌱 JOIN THE DIRTYVERSE

• Join our Discord and talk Dirt-y with us. It’s free to join! Paid subscribers have access to all channels.

• Follow @dirtyverse on Twitter for the latest news and Spotify for monthly curated playlists.

• Shop for some in-demand Dirt merch.  🍄