14 “But Is It Art?”: GANs Enter the Art Market
Is the art market ready to embrace work made by artificial intelligence?
—Naomi Rea145
In October 2018, for the first time in its history, Christie’s in New York auctioned an artwork created by an AI—to be specific, a GAN. Portrait of Edmond de Belamy went on auction with an estimated price of $7,000 to $10,000. In the end, it sold to an anonymous phone bidder for an amazing $432,500.
The work is a portrait of an aristocratic-looking gentleman in a dark coat and white collar, but the features are mysteriously blurry, as if gazing out at the viewer from some other dimension, as shown in figure 14.1. The work is signed not by an artist but by the signature equation of the algorithm that spawned the painting.146
Portrait of Edmond de Belamy, 2018.
The creators of the work are a French collective called Obvious. To make it, they fed their GAN with fifteen thousand classical portraits from the WikiArt dataset, created between the fourteenth and twentieth centuries, then let the generator loose to create portraits that would fool the discriminator into thinking they were real art. They stopped the process at a point where the GAN produced a reasonably classical-looking portrait and in the end came up with eleven finished works. All are of the fictional Belamy family, Belamy being a play on bel ami, a loose translation of good fellow as a tribute to Ian Goodfellow, the inventor of GANs.
The event caused a media storm and a fair amount of controversy. A flurry of articles appeared declaring that AI had at last come into its own and been recognized as capable of producing true art—as if being auctioned by Christie’s validated it. Journalists proclaimed that this was the first piece of art “aesthetically and conceptually rich enough to hold the attention of the art world.”147 The argument is that unlike DeepDream, GANs can be used to produce complete new and dramatically different images.
The members of Obvious say they are conceptual artists whose goal is to democratize GANs and legitimize AI-produced art. They compare AI with the camera, which appeared to be a scientific instrument when it was first invented in the nineteenth century and only gradually revealed its artistic potential. They proclaim themselves pioneers of “GAN-ism” and give as their motto, “Creativity isn’t just for humans.”148
Richard Lloyd, head of prints and multiples at Christie’s, believes the auction will lay bare fundamental questions about art and creativity. Does art have to be made by a person to qualify as art? “Everybody has their own definition of a work of art,” he says. “If people find it emotionally charged and inspiring then it is.”149
Even before the auction, one art collector, Nicolas Laugero-Lasserre, had bought one of the eleven portraits directly from the collective. Le Comte de Belamy, which he bought for about $12,000 in February 2018, shows a bewigged and ruffed aristocratic gentleman, again mysteriously blurred.
But the event has been greeted with outrage in the AI world. Many people working in AI question why this particular work should have been chosen and accuse Obvious of being mere marketers, promoters of their work, not true artists.
Mario Klingemann, who has done pioneering work in using GANs to create art, says: “Pretty much everyone who is working seriously in this field is shaking their head in disbelief about the lack of judgment when it comes to featuring Obvious, and of course Christie’s decision to auction them out of all artists who work with neural networks.”150
Besides Klingemann, many other artists whom I feature in this book are producing more original and interesting work than Obvious’s Belamys. For a start, there is Ahmed Elgammal’s creative adversarial network (CAN), tailored to search for totally new styles, and Jake Elwes’s images, generated by the machine’s wanderings through latent space.
In addition, despite extravagant claims to the contrary, this is not the first time that AI art has been auctioned. There was an auction at the Gray Area Foundation for the Arts in February 2016, as mentioned earlier, and there have been several others, including a group exhibition entitled Gradient Descent at a gallery in New Delhi, which included contributions from several artists featured in this book.151 All these have helped establish AI art as a genre.
Some knotty questions arise as AI enters the fraught world of the art markets: Exactly who owns the artwork? Is it the artist, or is it the machine? Who owns the copyright?
The issue of copyright goes back to A. Michael Noll of Bell Labs, who, as mentioned previously, created one of the first works of computer art in the early 1960s.152 When he tried to copyright it, the Library of Congress refused, claiming that a work generated by a computer could not be considered art because computers were capable only of number crunching. Noll insisted that the computer’s output was created by a program written by a human being, and the Library of Congress eventually relented, making Noll’s work the first copyrighted piece of computer art.
Noll’s program was straightforward. He connected the dots produced by an equation called a Gaussian distribution from top to bottom in a random way. GANs, conversely, are much more complex. Most importantly, GANs assess their products, rejecting some and accepting others. Some people consider AI to be simply a tool, like a camera used by photographers who then alter their images using Adobe Photoshop. Jessica Fjeld, Assistant Director at the Cyberlaw Clinic at Harvard Law School, asserts that “humans are deeply involved with every aspect of the creation and training of today’s AI technologies, and this will continue to be true tomorrow and for the foreseeable future.”153 As I believe I’ve shown in this book, AI need not always be considered merely a tool for artists, musicians, and writers. I suspect that in the not unforeseeable future, machines will come to be considered artists, writers, and musicians in their own right.
Mozart’s father taught him music, but that does not make him the creator of his son’s music. Should the teachers who taught us take credit for our discoveries? As time passes, it will become more and more clear how this applies to computers. GANs are an excellent example of what may well come to pass.
Obvious were soon followed by another artist. On March 6, 2019, Sotheby’s in London auctioned an installation by Mario Klingemann. Applying his own alchemy to Pix2Pix, Klingemann’s Memories of Passersby I produces an ever-changing stream of mind-expanding male and female faces across two screens. Disappointingly his work went for only $51,000. Did the out-of-line payment for Portrait of Edmond de Belamy momentarily freeze the market or was Obvious’s good fortune simply a matter of novelty value? Nevertheless, the episode is indicative of the growing interest in AI art in the art world. (See my Guardian article: https://