On October 25, 2018, Christie’s in New York sold “Portrait of Edmond de Belamy,” a blurry portrait of a fictional gentleman, for $432,500 including the buyer’s premium. The presale estimate had been $7,000 to $10,000, so the hammer result was roughly forty times the high estimate. It was the first artwork created with artificial intelligence to be offered by a major auction house.
The picture was produced by Obvious, a Paris collective of three friends - Hugo Caselles-Dupre, Pierre Fautrel, and Gauthier Vernier. As Christie’s own feature explained, they used a generative adversarial network, a pair of competing neural networks: “On one side is the Generator, on the other the Discriminator.” The generator invents new images while the discriminator tries to tell machine-made portraits from real ones, and the two improve each other until the discriminator can no longer distinguish them. The model was trained on a set of 15,000 portraits spanning the 14th to the 20th centuries, and the collective signed the canvas with part of the GAN loss function in place of a name.
The sale was not without controversy. Much of the open-source code behind the model had been written and published by the teenage AI artist Robbie Barrat, who received no formal credit, prompting a public debate about authorship and attribution in machine-generated art.
Why business readers should care: the Belamy sale was the moment the art market put a hard number on AI-made work, turning an academic technique into a headline-grabbing asset and forcing the questions - who is the author, and who gets paid - that still drive AI copyright fights today.