Data work as an organizing principle in developing AI

While data are often depicted as raw, neutral, and mere inputs to algorithms, we build on an emerging stream of research on data work viewing data as ambivalent, performative, and embedded entities, interwoven with organizing. We argue that in the process of developing AI, where epistemic uncertainty prevails as a key organizing challenge, data work serves as an organizing principle providing the logic through which behaviors are adopted, interpretations are made, and the collective efforts of domain experts and AI experts are coordinated. Prior research suggests that active involvement of both AI and domain experts is required for developing AI. Yet, domain experts and AI experts have distinct knowledge and understandings of domain specificities, meanings of data, and AI’s possibilities and limitations. Consequently, in AI initiatives, a key organizing challenge is epistemic uncertainty, i.e., ignorance of pertinent knowledge that is knowable in principle. We build a conceptual model deciphering three key mechanisms through which data work serves as an organizing principle supporting organizations to cope with epistemic uncertainty: cultivating knowledge interlace, triggering data-based effectuation, and facilitating multi-faceted delegations. These three mechanisms emerge when domain experts and AI experts work with and on data to define and shape trajectories of an AI initiative and make decisions about AI. This chapter contributes to the nascent body of research on data work by expounding the performative role of data as a relational entity, by providing a processual view on data’s interweaving with organizing, and by deciphering data work as a collectively accomplishment.