The Sociology of Technical Choices in Predictive AI

Abstract

Predictive AI models increasingly guide high-stakes institutional decisions across domains from criminal justice to education to finance. A rich body of interdisciplinary scholarship has emerged examining the technical choices made during the creation of these systems. This article synthesizes this emerging literature for a sociology audience, mapping key decision points in predictive AI development where diverse forms of sociological expertise can contribute meaningful insights. From how social problems are translated into prediction problems, to how models are developed and evaluated, to how their outputs are presented to decision-makers and subjects, we outline various ways sociologists across subfields and methodological specialities can engage with the technical aspects of predictive AI. We discuss how this engagement can strengthen theoretical frameworks, expose embedded policy choices, and bridge the gap between model development and use. By examining technical choices and design processes, this agenda can deepen understanding of the reciprocal relationship between AI and society while advancing broader sociological theory and research.