A machine-learning-generated printout has fetched $432,500 at auction house Christie’s in New York. That is more than forty times more than expected. A collective worked on it for months.
Although artificial intelligence can make art independently, the three men of the French collective Obvious chose a different approach. Using generative adversarial networks, they made a series of eleven portraits of a fictional family.
The portrait of ‘Edmond de Belamy’ brought in $432,500. It is signed with the mathematical formula describing the algorithm used in its generation. Machine learning student Hugo Caselles-Dupré explains to Motherboard that it took months to train the models, adjust parameters and select the correct input data.
The final result is the result of many iterations in which human taste still plays a major role. In generative adversarial networks, there are two competing models, one generating and one evaluating. The machine learning technique is more often used to generate realistic-looking human images, although Edmond de Belamy also somewhat resembles the result of a failed restoration of a fresco of Jesus, now more known as a meme, Motherboard notes.