On Training AI

This book explores how machines learn to see — and everything that resists such learning. Leaking Subjects and Bounding Boxes: On Training AI examines the processes through which artificial intelligence systems are trained to recognize, classify and organize the world, while simultaneously focusing on what escapes categorization: bodies, desires and fragments of human experience that refuse to fit neatly into systems.

The publication forms part of Elisa Giardina Papa’s long-term research into machine vision. In winter 2019, while working as a trainer for computer vision systems, the artist began collecting images that resisted organization — images that could not be clearly divided into signal and noise.

The book gathers visual fragments that challenge the very idea of strict separation. It is an invitation to think about “disorganized vision” — a way of seeing and existing that refuses the impulse to classify, normalize and create hierarchies. Here the blurry, excessive and unruly become acts of resistance against transparency and universality.