This morning, a portrait created by a Generative Adversarial Network (GAN) devised by the French art collective Obvious sold for $432,500 (including fees), vastly outpacing its pre-sale estimate of $7,000–10,000. After about seven minutes of bidding between three bidders on the phone, one in Christie’s New York saleroom, and one online, the work hammered down at around 11:20 am for more than 40 times its high estimate, with one of the phone bidders notching the winning bid.
The artwork, Portrait of Edmond Belamy (2018), is a deceptively generic, blurry, and seemingly unfinished portrait that seems to depict a portly man in a white shirt and black coat. It is one of a veritable family tree of portraits the trio of Hugo Caselles-Dupré, Pierre Fautrel and Gauthier Vernier developed by feeding a data set of 15,000 portrait paintings dating spanning the 14th century to the 20th to a GAN.
The artificial intelligence they developed begins by generating a portrait based on those in the data set, then tries to pick out the newly generated image out from amid a lineup of historical portraits. If the GAN can’t differentiate its own creation from the pre-existing portraits, the result is considered a success.
The work, which Christie’s marketed as the first portrait generated by an artificial intelligence to come to auction, was the final lot in this week’s Prints & Multiples auctions. At a panel on October 15th, Christie’s print specialists Richard Lloyd and Lindsay Griffith mused on the difficulty of categorizing such an artwork.
“Printmaking is ultimately a timeline of technological innovation,” Griffith said, before likening the advent of art created by artificial intelligence to the invention of photography. Whatever auction category they end up in, today’s sale suggests there is significant market interest in art generated by artificial intelligence.