The portrait in its gold-leaf frame reveals a corpulent gentlemen, maybe a Frenchman, and, judging by the dark frock coat and plain white collar he’s wearing, maybe a man of the church. There’s an unfinished quality to the work, in that the facial features are left somewhat blurred and there are blank areas on the canvas. A tag on the wall reveals that the model is someone named Edward Belamy, and the artist’s signature at the bottom right, rendered in cursive script, lets it be known that this work of art was created by Artificial Intelligence.

It went under the hammer at the Christie’s Prints & Multiples Sale that took place from the 23rd of October this year to the 25th. It sold for an extraordinary US$432 000, announcing the arrival of AI art. Artificial Intelligence is making itself felt in a number of industries as we near the end of 2018, most notably healthcare, automobiles, and entertainment, like the online slots Australia and the rest of the world now provide, and now the world of painting can be added to the list.

Obvious Art Uses a GAN to Create
The work is one of a group of portraits created by Obvious Art, a French cooperative enterprise consisting of Gauthier Vernier, Pierre Fautrel, and Hugo Caselles-Dupré. They all depict members of the fictional Belamy family, and are products of the collective’s exploration of the interface between art and AI. The method they use goes by the acronym GAN, or Generative Adversarial Network.

Caselles-Dupré revealed that GAN is made up of 2 parts, the Generator and the Discriminator. He stated that the collective fed the system with a data set of paintings created between the 14th and 20th centuries, and the Generator’s task was to create a new image based on these. The Discriminator then comes in to play, trying to spot the difference between an image created by humans and one that the Generator produced, and when it couldn’t, the new images were kept.

A Modern Version of 18th Century Portraits
One of the most interesting things about this portrait is that it is a definite departure from our idea of what 18th Century portraiture looks like. There is something very contemporary in the piece, with it even resembling some of Glenn Brown’s art history appropriations.

Caselles-Dupré explains that the distortions are an attribute of the model, since the Discriminator would be looking for features like faces and shoulders in the image, and for now was more easily tricked than the human eye by the absence of these.

A Difficult Genre for AI
Certainly this genre of art is a difficult one for Artificial Intelligence to tackle, since we are very attuned to the complexities and curves of a face in a way that machines just can’t be. This, however, was all part of the collective’s thinking.

Caselles-Dupré said that they had done some work with landscapes and nudes, and even tried feeding sets of famous painters’ works to the algorithm, but found that portraits worked best. The point they are trying to make with these creations is that algorithms are capable of copying creativity.