This story is over 5 years old.


This Artist Is Teaching Neural Networks to Make Abstract Art

Gregory Chatonsky even let a program name the paintings.
Gregory Chatonsky, Purple Flowers, 2016. Images courtesy the artist

Can computers make art? Nearly a year after Google's DeepDream program flooded the internet with puppyslugs, it's still hard to tell. Artist Gregory Chatonsky, whose work deals with the technological catastrophes facing our culture, explores this question with his latest project, It's Not Really You. 

Chatonsky started with thousands of historical, contemporary, and abstract paintingsm which he fed to Facebook's Eyescream artificial neural network (ANN). It created a series of 13 works based on what it "learned" an abstract painting is supposed to look like. To add texture, Chatonsky fed those images to another ANN called "Neural Style," and fabricated the results with a 2.5D printer to give them the substance of actual paints. This represents taking the painting from concept to physical product. Perhaps the most fascinating part is the naming of each work. He showed the final products to an ANN called Neuraltalk 2, which gave each painting a name based on what it "saw," similar to the AI•Scry app. The results are fabulously abstract works that have mundane-yet-nonsensical titles like Colorful Salad, Train Cake, and Pizza Sitting on a Rock.


Gregory Chatonsky, Colorful Salad, 2016

Sure, this project doesn't answer the question of whether or not machines can make art, but it raises some good points. Chatonsky had no control of the paintings' content, but he did concoct the assembly line of programs and machines that fabricated them. In a gallery context, it might be impossible to tell whether these were created by a person or a computer. So, are the machines tools, or is the human pressing buttons actually just part of the machine?

Says Chatonsky, a former painter, "I was moved by the result. Some 'paintings' have an artistic quality. The machine tried to make something that looks like what we make, but it's not human. It's after humanity. We look at these paintings with the same uncanny feeling as a human-like robot."

When I pose this question to Chatonsky, he says I should be asking something else entirely. "With this project, I realized that the question is wrong. We still think that art is a creative act carried out by a sovereign person, the artist. But today, the work of a contemporary artist is linked to the network: are we creators or just followers of the zeitgeist by copying what we see on the internet, as an AI does? In the end, perhaps it is the contemporary artist who now works as an artificial intelligence. He accumulates data, he synthesizes this data to create something new that looks like contemporary art (art fair, galeries, museums). Different but not so much."


Check out Chatonsky's neural network-fabricated paintings, It's Not Really You, below.

Gregory Chatonsky, Cat Sitting on a Bench, 2016

Gregory Chatonsky, A Bird in the Water, 2016

Gregory Chatonsky, Train Cake, 2016

Gregory Chatonsky, Pizza Sitting on a Rock, 2016

So, do neural networks immitate art, or does art imitate neural networks? Let us know on Twitter at @CreatorsProject. See more of Gregory Chatonsky's work on his website.


Roam a 3D Mountain Range Made from Kim Kardashian's Face

Here Are Imaginary Fossils from a Post-Human Earth

We Got an Artificial Neural Network to Name Random Objects in the VICE Office