Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence

Authors: Kate Crawford
Year: 2021
Source: Yale University Press

Contributions

Argued that AI is not immaterial or purely digital: it depends on rare earth mineral extraction, massive energy-consuming data centers, and underpaid human labor (crowdwork, content moderation)
Traced the material geography of AI from lithium mines in Nevada to GPU factories in Asia to data centers worldwide
Analyzed AI as infrastructure of power: who owns data, who benefits economically, and who bears the environmental and social costs
Proposed a political ecology approach to AI ethics that centers material and labor conditions rather than algorithmic fairness alone

Field research: site visits to lithium mines, Amazon fulfillment centers, and data centers
Historical analysis linking contemporary AI supply chains to longer histories of extraction and colonialism
Discourse analysis of AI company communications, patents, and policy documents
Synthesis of science and technology studies (STS) with political economy and critical geography

Showed that training a large AI model produces CO2 emissions comparable to the lifetime emissions of five cars
Documented that crowdworkers (e.g., Amazon Mechanical Turk) labeling training data often earn below minimum wage
Established that AI facial recognition systems are disproportionately deployed for surveillance in marginalized communities
Argued that the “intelligence” framing of AI conceals the enormous human labor and material resources that constitute it

Critical and activist framing may understate genuine benefits of AI applications in healthcare, science, and accessibility
Snapshot of 2021 landscape; the AI supply chain has continued to evolve rapidly since publication
Technical solutions to identified harms (renewable energy, fair pay) are underspecified
Primarily focused on US and Western supply chains; the global picture is more varied