Introduction to thematic section on ‘social theory in an age of machine learning’

Abstract

The proliferation of machine learning (ML) systems, which are algorithmic assemblages designed to extract patterns from data and make predictions, is visibly transforming society and everyday life. OpenAI’s GPT-4 represents the latest development in this field, but even less remarkable ML systems have made significant inroads into important societal domains over the past decades. For instance, scholars have explored the deployment of ML systems in areas such as credit scoring (Kiviat 2019; Rona-Tas 2020), insurance (Cevolini and Esposito 2020), criminal justice (Brayne and Christin 2021), selfdriving cars (Bissell et al. 2020; Stilgoe 2018), social media (Fourcade and Johns 2020), warfare (Scharre 2018), and automated trading (Hansen 2020; Hansen and Borch 2021). A substantial and growing body of literature has examined the societal effects of these systems. Concerns have been raised about their potential biases (Zou and Schiebinger 2018), their contribution to racial and social inequalities (Benjamin 2019; Eubanks 2018; Noble 2018), and their transformative impact on subjectivity, everyday life, and labour markets (Shestakofsky 2017; Wajcman 2019). Scholars have also discussed the opacity of ML systems and the broader epistemological, ethical, and political implications they entail. These discussions have touched on established notions of accountability, expertise, liability, and more (Amoore 2020; Brighenti and Pavoni 2021; Burrell 2016; Coeckelbergh 2020; Collins 2018; Fazi 2020; Pasquale 2020; Svetlova 2021). Simultaneously, there is a growing recognition, partially fuelled by these studies, that the rise of ML may have profound implications for social theory. On one hand, ML’s use as a new methodological tool holds the promise of uncovering patterns in data that could prompt a reevaluation of established concepts used to describe the social world. While this promise may not yet be fully realized, some scholars are optimistic about ML’s potential to generate theories by extracting non-linear patterns in data (Edelmann et al. 2020; Evans and Aceves 2016). On the other hand, the functioning of ML systems necessitates a reconceptualization of human-centered social theory (Airoldi 2022; Borch 2023; Esposito 2017; Yolgörmez 2021). In certain domains, the actionable predictions of ML systems not only inform human decision-making but replace it entirely (Borch and Min 2023). This distinction sets them apart from previous algorithmic systems and raises questions about accountability, control, ethics, liability, and