This Painting Was Made Completely By A Machine, But Will You Buy It For Rs 7.4 Lakh?

To look at that painting, it seems to the work of someone like Salvador Dali, or Frida Kahlo, with strong surrealist vibes.

Yet, it's actually a mix of various painting styles over the past 600 or so years. And not only is its painter not alive now, it never was to begin with.

Images courtesy: Obvious

The Portrait of Edmonde de Belamy isn't the representation of a visually impaired artist, or the fever dream of a tortured painter, it's actually the work of a neural network. The algorithm it runs on was designed by Paris-based art collective Obvious, which is is trying to fuse art and artificial intelligence.

This piece is just one of a series of 11 paintings of the fictitious Belamy family. And it's worth between $7,000 to $10,000 when it goes to auction next week.

Obvious designed an image-recognition neural network and fed it a dataset of 15,000 portraits made between the 14th and 20th centuries. They then had another part of the AI generate a new portrait, based on the styles of what it's learned.

The Belamy family tree

The system works using what's called a generative adversarial network (GAN) composed of two neural networks. The 'generator' attempts to create a portrait based on real ones it's seen, while the 'discriminator' compares it to its dataset to see if it can find similarities. If it does, the generator takes that knowledge and tries again, and so on and so forth. Eventually, you get to a point where the discriminator can't tell the difference between its counterpart's paintings, and that of humans'.

"The aim is to fool the discriminator into thinking that the new images are real-life portraits," Obvious co-founder Hugo Caselles-Dupré told auction house Christie's, where Edmond de Belamy will be sold. "Then we have a result."

The Compte (L) and Comptesse (R) de Belamy

The reason then that the painting looks so disjointed and odd is because it's trying to fool its own AI. "It is an attribute of the model that there is distortion," Caselles-Dupré said. "The discriminator is looking for the features of the image-a face, shoulders-and for now it is more easily fooled than a human eye."

Eventually though, who knows how AI may affect the world of art. Maybe It'll always be the AI's creator that gets the credit for a piece, or perhaps we'll get to a point where the algorithm itself will be considered the painter.

In either case, it's certainly going to very interesting when the Van Goghs and Picassos of our lifetime turn out to be machines.