Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence
Book Author: Kate Crawford
Kate Crawford dismantles the myth that artificial intelligence is “weightless code.” She traces every stage of an AI system’s life‑cycle—from the lithium blasted out of desert mines to the ghost‑workers labeling photos at a penny a click—and shows that what we casually call “the cloud” is really an empire of extraction. Chapter by chapter she maps the terrain:
Earth exposes how rare‑earth and battery metals power both data centers and electric cars, linking every trained model to literal holes in the ground and gigawatts of coal‑fired electricity macloo.com.
Labor follows the invisible armies on Amazon Mechanical Turk, Sama, and Appen who perform “Potemkin AI,” propping up chatbots and vision systems with underpaid human judgment macloo.com.
Data & Classification reveal how sprawling image corpora such as ImageNet encode world‑views—deciding whose faces, accents, and emotions are “ground truth,” and whose are noise macloo.com.
State connects military R&D, border surveillance, and predictive policing, illustrating how systems first built for drones and warzones drift into civic life macloo.com.
Space (Coda) even projects AI’s appetite beyond Earth, critiquing billionaire plans to mine asteroids once terrestrial resources run thin macloo.com.
Crawford’s credentials make her the ideal cartographer of this territory. As co‑founder of the AI Now Institute and a senior researcher at Microsoft, she toggles effortlessly between deep archival work, on‑the‑ground reportage, and insider knowledge of how models are really built Kate CrawfordMicrosoft. Her interdisciplinary lens—spanning political economy, environmental science, and critical race studies—adds rigor most technical authors lack.
Several cases linger: the Thacker Pass lithium mine that powers “green” mobility while poisoning Paiute‑Shoshone land; the South Carolina prison system deploying emotion‑recognition to flag “aggression” in phone calls; the Stanford “CheXNet” model that outperformed radiologists yet depended on unpaid interns to clean corrupted labels. Each vignette is meticulously sourced and serves Crawford’s larger thesis: AI is not synthetic intellect but an industrial process whose externalities have been outsourced to the world’s margins.
Why I Recommend This Book
Most AI commentary fixates on existential risk; Atlas of AI confronts the damage already here. For investors and builders who believe in “tech for good,” Crawford supplies the missing balance sheet: environmental debt, labor exploitation, and governance vacuums. Reading it forced me to scrutinize due‑diligence checklists—Are we pricing carbon intensity? Auditing dataset provenance? Valuing the well‑being of the annotation workforce? If you influence capital or code, this book is mandatory.
