Mark II (Convolutional Neural Network)

Kinetic sculpture. Maple, Paper, Elastic Cord, Electronics. 2020 413 ppm
A photo of a room. On a wall next to a staircase hangs a sculpture made of wood, paper, and electronics. It looks like half diagram, half machine.
Photo: Brendan Getz

A physical instantiation of usually invisible algorithmic processes, the sculpture renders the learning process of a machine learning system visible, audible, and literally tangible to the viewer.

The Mark II is a functional convolutional artificial neural network — an artificially-intelligent algorithm that learns to distinguish letters. The machine is built in the shape its own diagram (as drawn by a researcher during its design). Mathematical precision in code contrasts with inevitable imperfections owed to the human body that manufactured the object by hand in natural materials.

The work challenges both abstract and biomorphic conceptions of artificial intelligence, offering instead a haptic encounter with contemporary AI technology. With its larger-than-human scale the sculpture radiates a totemic authority, alluding to the widespread glorification of artificial intelligence. The work enables the understanding of AI systems as human-made yet distinctly non-human, and as physically instantiated and materially grounded. It questions notions of “pure” intelligence and a scientific view from nowhere.

Credits

In collaboration with Prof. Yann LeCun, Dr. Jake Browning, Prof. Alfredo Canziani, and the ToftH program.

Code: https://github.com/philippschmitt/physical-neuralnet

Created during a residency in Yann LeCun's research group at NYU Center for Data Science awarded by the Berggruen Institute ToftH fellowship program.

Photo: Brendan Getz
Photo: Brendan Getz