The neural networks on which the AI artists work are open-source software created by well-known scientists.
On a giant screen in the basement gallery of Nature Morte in south Delhi, a weird face is looking back at viewers, making uncanny expressions. The strange and morphing animation must worry artists because it’s not the work of a human being. Welcome to the world of art and artificial intelligence (AI).
Gradient Descent, an exhibition of works made by AI, challenges the basic concepts of creating art. At the centre of the debate—about art made by a human as opposed to that by AI—are issues like ‘labour’, ‘perception’ and ‘creativity’. The curators of the show, however, say these criteria also apply to art generated by AI. “AI is replacing physical and mechanical labour,” says Raghava KK, who has curated the exhibition with his brother Karthik Kalyanraman. “But what about creative and spiritual labour? I am not going to let AI replace the artist in me. It is only going to shape me as an artist,” he explains.
A new genre
AI practitioners say they want a serious conversation with the art world about introducing this new genre of art. At the Nature Morte gallery, the curators of Gradient Descent have chosen the works of seven artists from around the world. “It was very difficult to find AI artists,” says Kalyanraman. It took the brothers six months to find 90 artists practising AI art. They eventually identified 40 of them as fitting their definition of such an artist. “And only seven, we felt, truly pushed the boundaries,” adds Raghava.
79530 Self-Portraits, the work showing the weird face in a mammoth video installation, is by German AI artist Mario Klingemann. Considered one of the pioneers in AI art, Klingemann has been making art with computers for the past four years. “Before that, it was not technically possible,” says Klingemann, who has, in the past, exhibited his AI works at the Ars Electronica in Linz, Austria, which shows works and themes dealing with the future.
Klingemann’s work, which will travel to Ars Electronica later, is described by the artist as “a study of the human face”. In the world of AI art, that means creating three neural networks—computer systems designed by algorithms to recognise images—and training them on 6,000 portraits (of masters like Rembrandt). “I didn’t want to copy Rembrandt,” says Klingemann. So he trained one set of computer models made from the neural network to transform a sketch of the face and another set to change the sketch to a painting, both resulting in the creation of strange faces. “I see the face more like a canvas,” explains Klingemann.
Debate on creativity
For British AI artist Jake Elwes, it was about two different AI networks having a conversation. Closed Loop, the work of London-based Elwes, is an attempt to question creativity. “We believe only humans can create,” he says. In his work at Gradient Descent, Elwes has cut out humans. “We are just spectators,” he adds. For the work, the artist taught one computer network to write descriptions of images and another to generate images from words. The result is images on a screen and the computer’s response on a wall, with words like, ‘A gray bathroom floor’ and ‘Blue sky with no clouds’.
Anna Ridler, another British artist, brings her work The Fall of the House of Usher (an AI artwork based an Edgar Allen Poe’s short story of the same title about decay and collapse) to the exhibition. Ridler’s work mixes animation and drawings.
Harshit Agrawal, the only Indian AI artist who is part of the exhibition, trained a neural network to understand surgical procedures. His work, The Anatomy Lesson of Dr. Algorithm, is a series of images on the medium of archival pigment print in paper. “I fed 60,000 images of human dissection to the neural network,” says Agrawal.
The neural networks on which the AI artists work are open-source software created by well-known scientists. Some of the networks used are TensorFlow and PyTorch. “Everybody can use it,” says Klingemann. “AI is a very democratic way of creating artworks.” An artist needs powerful graphic cards, though, to negotiate the heavy files. AI practitioners say when they are creating art, the computer models are making constant decisions, just like an artist. Or are they? The debate is expected to continue. So are shows of art made by AI.
The exhibition Gradient Descent is on till September 15 at Nature Morte, New Delhi
Faizal Khan is a freelancer