Skip to main content
Tech + Life

It’s not Dali. It’s AI.

These shape-shifting portraits, created by machine learning, were among the first to sell at a traditional auction

By Tony Rehagen

The two strangers, one masculine, the other feminine, stare silently out of their frames from somewhere in the distant past. Their soft lines and classical, rounded facial features suggest that these are 17th- or 18th-Century oil paintings. But the portraits have no nameplates or signatures. Presumably, the identities behind these cherubic faces, and that of the painter, have been lost to time.

Then something changes. Almost imperceptibly, the bearded figure’s eye starts to droop and change hue. A third eye emerges from a shadow. The doughy face sags and begins to melt into a grotesquerie. Ears curl, lips twist, noses sink and reappear elsewhere in the amorphous sea of color. A new countenance emerges, briefly meets your eye, and grins. Then it, too, dissolves into an endless stream of strange new faces. These are not staid paintings, but framed 4K high-definition monitors. And this is not traditional and commonplace portraiture but rather a new, exciting — and possibly disturbing — glimpse into the future of art.

Memories of Passersby I is also not a standard video installation. The images are not pre-programmed, nor random; there is no internal database. Instead, the work originates with a computerized brain. Housed in a sleek retro-modern wooden cabinet wired to the two screens, a system of neural networks creates the flow of never-before-seen images in real time. Memories is a revolutionary work of artificial intelligence — every pixel created by an algorithm, based on the input of thousands of old paintings and a program that learns how to make them change.

Last year, the piece broke into the established art world when it became one of the first AI-generated works to sell at a major auction house — Sotheby’s, for £40,000, or about $50,400, to a private collector.

The emergence of a tireless, autonomous generator of original work raises provocative questions. Who is the creator, man or machine? Is technology merely an instrument of creation, or have we reached the point where the circuity is now capable of authentic creativity? For years and years, people have looked at the newest novel and abstract exhibitions and asked, “What is art?” Now they might fairly start to ask, “What is an artist?”


Mario Klingemann’s answer to that latter question is: I am an artist…I think. Whether people want to label him as “the creator” or “the architect” or “the coder” behind Memories, Klingemann, 50, came up with and executed the idea. “People say I’m an artist,” he says via Zoom from his apartment in Munich. “I claim to be one — though sometimes I feel like I have impostor syndrome.”

Klingemann’s reluctance to carry the title might be modesty, but it also might stem from his background, which is not that of a traditional artist. Though he’s the child of an engineer father and a painter mother, Klingemann hasn’t built his career as a straightforward marriage of technology and fine art. He always had big ideas and knew from an early age that he wanted to create, but he was never proficient with a pencil or brush, and he found attempts at photography unfulfilling. There was no art school in his future.

“Our physical world is limited. Now we have this second world, which is also kind of a dream world. It’s an almost endless space to explore.”

artist Mario Klingemann

He also loved computers and received his first, a Commodore 64, at age 13. He developed a knack for programming, but his creativity was hemmed in by the equipment: the binary, monochrome machines that were in his price range in the pre-internet 1980s. So computer science was out. Instead, Klingemann opted to go into advertising.

Though Klingemann worked as a copywriter, his time at the ad agency exposed him to the graphic interfaces of Macintosh computers and Photoshop software. He became a freelance graphic designer, making flyers and posters for techno music shows and eventually designing websites. At the same time, he started experimenting with Photoshop, created some of his own filters, and built his own programs to produce original graphics. It was Klingemann’s first “generative art” — created with autonomous non-human systems — though that term didn’t exist at the time.

“I didn’t dare to call myself an artist,” he says. “No one was doing this with computers. There were no role models. It was definitely what you’d call ‘generative art,’ but I didn’t know that.”

The technology was catching up with Klingemann’s imagination. He dove into Adobe Flash and started producing motion graphics and openers for TV soap operas. By the early 2000s, he was writing programs that produced moving imagery on a loop, using preloaded photographs from Flickr.

He kept coming back to the emerging power of algorithms. When properly written, they seemed to give art a life of its own. Sets of algorithms called neural networks, loosely modeled after the brain, can recognize patterns and, when fed the proper data, be trained to “think” for themselves and produce original imagery and sounds.

For Klingemann, Memories is a culmination of his use of machine learning to produce new images. He started with the data — thousands of portraits from the 17th through 19th centuries — which he uploaded into his AI model. Next, he programmed a Tinder-like application to teach the machine aesthetic preferences — based, in this instance, on the work of surrealist artists (hence the grotesques). The network processes this feedback on a loop, rapidly homing in on traits it deems pleasing and projecting those features as elements of the new faces on screen.

Just as human artists incorporate multiple influences into their work, so too does Klingemann’s creation. But the machine never sleeps, never stops creating, and never repeats itself. “Our physical world is limited,” says Klingemann. “Now we have this second world, which is also kind of a dream world. It’s an almost endless space to explore.”


But there are limits to AI — at least for now. Memories will never stop creating new portraits, but it will never produce a still life or a landscape. It cannot grow. The machine will never evolve beyond what Klingemann puts into it. Because of that, he and others believe, he is still the artist.

“There is a lot of effort involved in designing the neural networks and adjusting the large number of parameters to bootstrap the system,” says Ramon Lopez de Mantaras, research professor at the Artificial Intelligence Research Institute in Bellaterra, Spain. “[Klingemann] spent hundreds of hours setting up his systems and doing lots of trial-and-error adjustments. For him, the AI, in this case the neural networks, is a tool — a quite sophisticated tool, but a tool nonetheless.”

Ben Caras, a fine artist and professor of design and digital fabrication at Northeastern University, says Klingemann and other AI artists are just doing what people are doing across many other professions — using algorithms to perform their jobs in a different way. “We as humans use a collective consciousness to move through the world every day,” he says. “Is this art? Damn right, it’s art.” Caras says that what distinguishes any art, be it painting or sculpture or music, is the human response it evokes. And even though the machine in a work like Memories might be autonomously churning out the images that stimulate that human response, there’s always a human in control.

Humans have trained machines to essentially think for themselves, but only using the data and systems the creators have given them. Using cameras and recorders, some AI artists are already creating machines that can go beyond the preprogrammed data and continuously record new sights and sounds to expand the palettes with which they work. But thus far, science has been unable to endow the systems with perhaps the most important element of creative expression: Inspiration.

“That initial spark, that initiative, still comes from me,” says Klingemann. “The machine doesn’t have the urge to create.” His own drive, he says, comes from a place that a machine has no way of understanding. As long as a machine performs its function and the creator doesn’t pull the plug, it essentially lives forever. Humans don’t. “I want to leave something behind,” he says. “I want to put something out there that maybe stays longer than my body.”

Published on

Tony Rehagen is a writer based in St. Louis.

Images provided by Mario Klingemann

Tech + Life

A new film with a dead star? Thanks to CGI and deepfake technology, it’s possible

James Dean 2.0 may be coming to a theater near you

By Stav Dimitropoulos