The cost of Edmond de Belamy exceeded the initial estimate by 40 times.
The auction at Christie’s in New York lasted just under seven minutes, during which the buyer competed against potential customers from France (remotely, online), two representatives of businessmen (they were in touch by phone) and one person indoors.
When the hammer fell, the rates reached 350 thousand dollars. With all the fees, the cost was 432 thousand dollars. For comparison: the print of Andy Warhol, which hung opposite, was sold for 75 thousand dollars. The work of Roy Lichtenstein sold for 87 thousand dollars. Both prices include fees.
We would like to thank the AI community, especially those who pioneered this new technology, including Ian Goodfellow, the creator of the GAN algorithm, and the artist Robbie Barrat, who had a great influence on us.
We are grateful to Christie’s for opening this dialogue in the art community and are pleased to be part of this global conversation about the impact of this new technology on the creation of art.
The work, which was originally estimated at 7-10 thousand dollars, was the result of the collaboration of a French trio from a student of the machine learning department and two graduates of a business school. None of them had experience in art. The portrait did not use any color: it was the work of an algorithm that learned to imitate image sets covering the period from the 14th to the 20th century.
The auction was a test for Christie on the subject of market interest in art from artificial intelligence. But the most “strong” response after the auction announcement came from other artists who work with AI. Many of them said that the portrait was not original. They noted that the generative-adversary network – the technology used to create the portrait – has been used in art since 2015. For example, artists like Mario Klingemann , Anna Ridler and Robbie Barrath did this .
The algorithm consists of two parts. On the one hand, the generator, on the other hand, the discriminator. We fed the system a set of data from 15 thousand portraits, written from the 14th to the 20th century.
The generator creates a new image based on the set, then the discriminator tries to determine the difference between the image created by the person and the image created by the generator. The goal is to fool the discriminator so that he thinks that the new images are real portraits.
The participants of the trio admitted that they borrowed a part of the code from another person, who distributed it under an open license. It is unclear how much was borrowed, but experts say the amount was probably substantial. Also, ytgjyznyj, whether Robbie Barratt can claim ownership of the work, since his code was distributed under an open source license.