[ Art and AI , episode 12]
The section on Art and AI is back, have you already read the other episodes ? We will talk about GAN !
Two years ago, on October 25, 2018, something incredible happened: London auction house Christie's managed to sell an image generated by a GAN for $ 432,500 . Why "incredible"? For two factors:
- because the GAN code used to create the painting was copied largely from an existing one
- because the sales value had been estimated between $ 7,000 and $ 10,000 , but media exposure based on some media falsehoods caused the final price to rise (many newspapers reported that the painting was "the first work of art created by an AI ", while, given the examples given in the previous paragraphs, we can say for sure that this is not true)
The origin of the painting
The painting in question is entitled Edmond de Belamy, part of a series of portraits entitled La Famille de Belamy. The print was created by a team called Obvious, a trio of 25-year-old French students whose goal is to “explain and democratize” AI through art. Over the past year, they have made a series of portraits depicting members of the fictional Belamy family, advertising their work through press releases created specifically to attract attention. The name “Belamy” is a tribute to Ian Goodfellow, inventor of GAN : it is a play on words, since “bel ami”, translated from French, means “good friend” (“good fellow”).
GAN experts say the code used to generate these prints is mostly the work of another artist and programmer : 19-year-old (at the time) Robbie Barrat, a recent graduate who shared his algorithms online via an open source license.
The members of Obvious do not deny that they have essentially borrowed Barrat's code, but they made it known some time after the presentation of the painting to the media attention.
Guilt and justifications
The fact greatly annoyed the GAN artistic community, which has been working for years in a collaborative and united way to redeem itself in front of the general public. Although seeing a portrait composed of an AI for sale at Christie's is a milestone that elevates the entire artistic community, it must not have been nice for artists to learn that whoever made it all possible, "stealing" the code to another artist, was someone from outside the community.
Robbie Barrat, the "robbed" artist, is one of the illuminating beacons in the GAN art world. He shares with others (on GitHub, an online community of programmers) the algorithms he uses to create his images, helping other artist-programmers put neural networks to work.
This is how Hugo Caselles-Dupré, of the Obvious team, managed to access Barrat's algorithms and use them to generate Belamy's portraits.
During an interview with "The Verge", Caselles-Dupré admitted that he borrowed parts of the code from Barrat , but made several changes:
Obvious's problems in changing the code have often been solved by Barrat himself: in the discussions on Github there are several conversations between Caselles-Dupré and Barrat in which the Frenchman asks the nineteen year old prodigy for help.
The reactions of the GAN artistic community
Tom White, a New Zealand artist, says the portrait is extremely similar to what Barrat's code, freely downloadable from GitHub, would produce without any modification.
The fact was expressed even Mario Klingemann (GAN Artist phenomenal, we had already spoken ), who thinks that 90% of "real work" on the offending part, has been done by Barrat. Art pioneer GAN also highlights the team's glaring technical shortcomings, which can be deduced from the low resolution and smearing of the picture:
Why did Christie's agree to sell this portrait?
Because Obvious has distorted reality, presenting the portrait to the world as if it were created by a sentient machine.
Their motto is "creativity is not just for man". A January press release sensationalistically said that "an artificial intelligence [had] succeeded in creating art." Shortly thereafter they sold their first portrait to a French collector and more and more articles began to circulate about the portrait entirely "created by artificial intelligence".
For GAN artists, these releases would be comparable to jokes because, according to them, they would give readers a false view on machine learning systems, showing them more autonomous than they really are.
Robbie Barrat claims he doesn't hold a grudge about the fact itself, but he fears that all this media attention on Obvious will overshadow other AI artists. According to Barrat, GAN networks are very far from the vision given by Obvious on their being autonomous , because it is always the man who encodes the networks, chooses the data with which to train them, trains it and chooses which are the best results.
Caselles-Dupré would never have expected such success for the portrait:
And instead, now, together with his team he has found himself in the History of Art.