The Foreign Service Journal, October 2021

78 OCTOBER 2021 | THE FOREIGN SERVICE JOURNAL analogy doesn’t survive the first chapter, Crawford nevertheless makes a convinc- ing case for understanding artificial intelligence as an “extractive industry” that depends on “exploiting energy and mineral resources from the planet and data at scale.” In this respect, Atlas of AI is well worth the read. Chapter by chapter, Crawford delivers a stunning critique of artificial intelligence institutions and practices, unveiling it as a “massive industrial formation” built on envi- ronmental degradation, unfair labor practices, intrusive data collection, unrestricted government surveillance activities and technology-sector profit margins. Beginning with a focus on the extrac- tion of rare earth minerals, water, coal and oil that fuel the technology sector’s energy intensive infrastructure, Craw- ford details the environmental costs of such practices. She then turns to the exploitation of the vast labor force that produces these elements, from miners to the shadow workforce of technology company employees. In her narrative, unrestricted physi- cal extraction and exploitation of the labor force are mirrored in unregulated data collection and manipulation. From images of people on the street, to personal photos posted online and the scraping of online conversations, Crawford tracks the unconstrained com- pilation of massive amounts of personal information and how this input is used to develop vast facial recognition data- bases and speech-based algorithms. That this personal information is collected and used without prior knowledge and consent is bad enough. Worse yet, as Crawford notes, this data is employed to build predictive and inter- pretive models of human behavior that are not based on systematic and respon- sible methods of data acquisition. The resultant classification systems reinforce harmful biases and stereotypes, and perpetuate patterns of political discrimi- nation and socioeconomic exclusion. These models of classification become, in Crawford’s view, part of a “far-ranging and centralizing normative logic that is used to determine how the world should be seen and evaluated.” Otherwise known as “affect detection,” this process draws on unfounded and often controversial notions about the way in which people express emotion to create a codified, machine learnable and trackable system of identification that strips individuals of their identity and culture. Beyond the risks inherent in the erasure of individuality and exploitation of personal information, this data col- lection process can become a dangerous tool when deployed at the intersection The so-called artificial intelligence revolutionhas significant consequences for national security, economic stability and social welfare.