An art exhibition at New Delhi's Nature Morte showcases art born as a result of the artists' encounters with machine learning. Gargi Gupta reports
Rare is the exhibition in an Indian gallery that can claim to be at the cutting edge of an art genre, to showcase a practice that is in its infancy even in the major art centres in the West. Gradient Descent at Nature Morte in New Delhi is one such exhibition, showcasing seven pioneering artists from around the world, who are using artificial intelligence (AI) as an integral part of their practice.
So how does it work?
As curators Karthik Kalyanaram and Raghava KK explain in their catalogue, "The kind of AI algorithm used by the artists is called a neural network: its design was meant to mimic the activity of biological neurons. To teach a neural network a visual language, the artist…only exposes it to a large number of examples (called training set) and allows it to draw its conclusions. After this 'training', the AI is ready to start creating the final work."
It sounds complicated but the results, as seen on the gallery walls, make for a compelling aesthetic experience, and show evidence of a painstaking process – two old-fashioned hallmarks for judging 'good' art.
Take Anna Ridler's The Fall of the House of Usher, based on sketches she drew by hand of stills from the first four minutes of a 1928 avant-garde film by James Sibley Watson that was based on Edgar Allan Poe's story of the same name. Ridler's 'training set' was the sketches, that the AI then uses to 'create' the entire film mimicking a style that was at once akin to her own, and yet dissimilar. In fact, so inspired was Ridler, a London-based artist who's had her works displayed at, among others, Tate Modern and V&A, by these machine-generated approximations/interpolations of her style, that she's come up with 12 illustrations in black ink wash done in the style – a case of man learning from machine!
Similarly, The Anatomy Lesson of Dr Algorithm, by Harshit Agrawal, the only Indian in the show (the name paralleling a famous painting by Rembrandt, The Anatomy Lesson of Dr Nicolaes Tulp), consists of 20 prints of what the AI algorithm imagines the inside of a body looks like. Here, Agarwal's 'training set' had been videos and images of human dissection. The result, enhanced by grainy pixelation of the final image, is disturbing and viscerally beautiful.
Mario Klingemann, perhaps the most eminent of the artists here and best known for his experiments with AI art, explores the fun, playful possibilities of the technology with his video, 79530 Self-Portraits. The work uses two 'neural networks' – one, images of Old Master portraits, and two, a webcam video of the artist's own face – which it tries to merge. The result? A grotesque, ever-changing, eerie series of images as the machine tries to transform the artist's face into the likeness of an Old Master portrait, only for the image to disintegrate and distort before it comes together again.
There's only so much, yet, that the machine can learn and do. But for how long?
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
 "artificial intelligence")
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}
)
){kind=small}
){kind=small}
)
)
)
)
)
){kind=small}
){kind=small}
){kind=small}
){kind=small}
){kind=small}