AI and the Human Condition: Thinking Beyond Accuracy

“Accuracy” is often associated with the number of correct predictions generated by a machine-learning model and can be taken as a metric of performance of an AI system. Although its usefulness as such a metric is debatable, the notion of accuracy itself still organizes much of the thinking about AI. In an analysis of FORDISC, a database of skull measurements used to identify human remains, Iris Clever demonstrates how a focus on accuracy might struggle to account for the entwined relationship between humanity, science, and technology. She argues that because the ideology of nineteenth-century race sciences continues to configure the collection, organization, and reuse of data in FORDISC, improvements related to accuracy can be deeply misleading and further codify the global datafication of race in the sciences. How might a historical approach to “accuracy” enrich the ways that we understand one of the most fundamental concepts of statistical reckoning? This talk suggests how humanistic inquiry can help us think beyond accuracy. Iris Clever is a postdoctoral researcher at the Committee on Conceptual and Historical Studies of Science, University of Chicago. She has also been named an iSchool Research Fellow at the University of Illinois for 2024–26. Respondent: Cris Hughes, clinical associate professor of anthropology at the University of Illinois Urbana-Champaign