clock menu more-arrow no yes

AI-produced artwork is moving from the fringes into the high art world

But is it replacing or just supplementing artists’ capacity for creative expression?

A portrait of a man that resembles an impressionistic painting but was created by artificial intelligence
This painting, “Portrait of Edmond de Belamy” from French collective Obvious, will be the first AI-generated art to be sold at prestigious auction house Christie’s.
Christie’s

In a month, prestigious auction house Christie’s will invite the upper crust of the art collector world to their Manhattan showroom to bid on a gold-framed impressionistic portrait of a European gentleman. It would be an ordinary affair in the world of high art, except for one detail: The portrait for sale was generated by AI, not a famous painter.

It will be the first AI-produced artwork to be sold at a major auction house — and is estimated to sell for $7,000 to $10,000.

AI-generated art has existed for years — remember those weird, trippy Google projects? — but in its nascent form has mostly been relegated to niche audiences and experiments. Now — with the Christie’s auction, an AI-specific show at the Grand Palais museum in Paris earlier this year and another at one of the largest contemporary commercial galleries in India where artwork was priced at over $30,000 — AI artwork is being appreciated and valued by the most elite art institutions.

So is it just novel? Or can AI replace human creativity?

“People expect the field of creativity as the last thing that will be mechanized, the thing that will be the most human,” Christie’s specialist Richard Lloyd told Recode. “They think, ‘Oh okay, I don’t mind if AI reads a CAT scan or answers my customer support questions,’ but when it actually starts writing songs or painting pictures — that’s in some ways its darkest or most radical intervention.”

The AI-generated portrait up for sale at Christie’s comes from Paris-based collective Obvious, made up of three 25-year-old best friends, Hugo Caselles-Dupré, Gauthier Vernier and Pierre Fautrel.

“For us, making portraits is the most striking way to convey your message of what you can do with an AI tool — and it’s not that complicated to explain,” said Caselles-Dupré, a PhD student in machine learning and former painter, in an interview with Recode.

To start, the group spent several months figuring out what general kind of image they wanted to create and the algorithm best suited to make it. They settled on making a series of portraits of a fictional aristocratic family using a generative adversarial network, or GAN. In a nod to the researcher who originally invented that algorithm, Ian Goodfellow, they named the series “La famille de Belamy” (“good fellow” is roughly translated to “bel ami” in French).

Then, the team fed the algorithm a database of 15,000 portraits from the 14th to 20th century. They used two tools to make the portrait, a generator and a discriminator. The former created an algorithmically generated image, and the latter distinguished if it was done by hand or by computation.

Once the discriminator tool could no longer distinguish if the portrait was man-made or not, the image was complete. The artists printed a single copy of the image on a canvas with a gold wooden frame, solidifying its physical existence as a piece of art.

“It’s an interesting piece, and it’s exciting there’s a larger embrace by a very influential art house to sell this work,” said Matthew Israel, an art historian and head curator of online art platform Artsy. Ultimately, though, Israel says that he views AI as another method — similar to the camera when it was first introduced — in an artist’s toolbox, rather than a technology that will replace artists altogether.

For others, the work represents more fundamental questions around creative agency.

“Who is responsible for this piece and what is that creative locus?” said Lloyd. “Is it Goodfellow, who wrote the algorithm? Is it the people who give the algorithm the initial dataset for it to learn from? Is it an amalgam of everything in that data that learns things?”

There’s only one signature on the bottom right corner of the painting: The mathematical equation for the algorithm itself.

This article originally appeared on Recode.net.