When Data Scientists Meet Domain Experts: Artificial Intelligence, Epistemic Conflict, and Knowledge Transformation

Predictive AI models developed by data scientists promise to transform organizational knowledge, yet in practice the predictions remain inscrutable to domain experts. Existing explanations situate the problems in the opacity of the AI tools, failed adoption, or jurisdictional conflict, but offer limited insight into how to make the knowledge accessible. Here we show how the problem emerges from the competing epistemic stances of data scientists and domain experts, which create conflict over how to produce and evaluate knowledge. We then ask how data scientists and domain experts address that conflict so that their work with AI satisfies each group’s standards for producing and evaluating knowledge. Drawing on an 18-month inductive study of three large Indian banks, we show that the data scientists produce new knowledge, but domain experts hold the power to decide whether it is deployed. Our findings distinguish between epistemic and jurisdictional conflict. We show that data scientists first try to address the epistemic conflict directly through assimilative practices that follow one epistemic stance or the other. When these practices fail to transform knowledge, data scientists learn to address the conflict indirectly through four bridging practices. These practices help the domain experts transform knowledge by embedding the data scientists’ predictions in the boundary objects and practices that the domain experts already use. Our findings reveal how power based on expertise rather than jurisdictional control can make possible the coproduction of knowledge.