cover image Feeding the Machine: The Hidden Human Labor Powering A.I.

Feeding the Machine: The Hidden Human Labor Powering A.I.

James Muldoon, Mark Graham, and Callum Cant. Bloomsbury, $29.99 (368p) ISBN 978-1-63973-496-2

AI depends on the exploitation of artists, data annotators, and engineers, among others, according to this damning exposé. Graham (coauthor of The Digital Continent), an internet geography professor at the Oxford Internet Institute, teams up with Muldoon (Platform Socialism) and Cant (Riding for Deliveroo), both management professors at the University of Essex, to spotlight individuals whose labor powers AI. The authors describe the travails of a data annotator from rural Uganda who, for the equivalent of $1.16 per hour, works grueling shifts marking traffic lights, human faces, and other elements of interest in images that tech companies use to train software. Ethics have been sidelined in the AI gold rush, the authors contend, discussing a London machine learning engineer’s concern over the fact that her employer’s lack of guidance on handling sensitive topics when training the technology (e.g., “Should a particular event be described as a genocide?”) leaves often poorly paid data annotators to make morally freighted decisions. The grim real-life stories read like dystopian parables, such as the account of a European voice actor whose recordings were legally used without her consent to create an inexpensive synthetic clone whom she now competes with for business. Driven by striking reporting and finely observed profiles, this unsettles. Agent: Will Francis, Janklow & Nesbit Assoc. (Aug.)