Rethinking AI's Role in Science: Toward a Theory of Epistemic Co-Agency and Its Implications for Research Organizations

Aya Daruich, Mounya Chahboune, Lhoussein Baouzil

The question of how artificial intelligence is reshaping scientific knowledge production has moved, in a few short years, from theoretical speculation to practical urgency. Yet most organizational and management research still treats AI as a sophisticated tool — useful, even transformative, but ultimately passive in the epistemic sense. This paper argues that framing has become inadequate. Drawing on a theory-building conceptual approach, we develop the notion of AI as an epistemic co-agent: an active participant in generating, validating, and legitimating scientific knowledge alongside the human researcher. We organize this argument around a three-configuration typology — researcher as producer, collaborator, or supervisor — that evolves as AI autonomy rises. Five empirically testable propositions follow, and the paper ends by examining what this transition demands of institutions responsible for governing science.