Materializing futures: fairness and diversity in AI development as so...

An increasing number of cases of algorithmic bias has sparked a lively debate about normativity and values of algorithmic systems in the scientific community and the public (Benjamin, 2019; Crawford, 2021; Buolamwini & Gebru, 2018; Birhane 2021). This debate revolves around concepts such as transparency, accountability, and fairness. With our presentation we contribute to this ongoing discussion with a focus on the role of organizations in and as well as perspectives of public and private actors. Drawing on the analytical concept of "sociotechnical imaginaries'' developed by Sheila Jasanoff and Sang-Hyun Kim (2015), we show how organizations take part in the negotiation processes of the vision “of how life ought, or ought not, to be lived” (Jasanoff 2015: 4). Our presentation is based on two ethnographic in-depth studies in the context of AI development: one study was conducted in a private credit scoring company, where the ethnographer was involved in a project on fairness in credit scoring algorithms. The second study aimed at exploring implementation practices of a recommender system guided by the ideal of information diversity in a public service broadcasting. Here, the public actor anticipated diversity in recommendation systems and wanted to prevent be(coming) unconstitutional. We argue that these actors react to anticipated regulations, even if there are no manifest legal issues or if the regulation is not yet in place. By mobilizing popular discourses in expert communities and the general public the actors address an imagined future, in which these regulations might become problematic to their business model. By these anticipated regulations they try to establish their solutions as material discourses through which the sociotechnical imaginary of fair credit scoring or diverse media spaces is attained.