Leaking Subjects and Bounding Boxes: On Training AI

Leaking Subjects and Bounding Boxes addresses the ways in which machines are disciplined and trained to see. It is a visual account that documents the methods currently used to teach AI to capture, classify, and order the world but instead considers everything in our lives, embodiments, and desires that defies these normative modes of categorization. That is, it pauses to reflect upon that which continually flickers in and out of any possible taxonomization and exists only at the very edge of definition. This book is one outcome Elisa Giardina Papa’s ongoing visual and theoretical research into machinic vision. It presents images which the artist began collecting in the winter of 2019 while working as a human trainer for various AI vision systems. Among the thousands of training images she processed for her tasks as a “data-cleaner,” she collected those which seemed to resist AI’s orderly impulse. Segmenting, tracing, bounding-boxing, and labeling are key operations used to teach machines to separate data from data, signal from noise, and orderly things from disorderly ones. This collection of images is an invitation to reflect on what it means to practice a disorder of seeing and being which radically resists the normative impulse to divide and classify, to create hierarchies and produce difference. In other words, this book is an invitation to consider the ways in which all things that are opaque and promiscuous, heretical and unfaithful will perpetually upset orderly, idealized, transparent, and supposedly “universal” categories.