Epistemic Algorithmic Drift: AI-Mediated Scholarly Infrastructures and the reproduction of Gendered and Intersectional Inequalities

Abstract

Artificial intelligence is reshaping scientific knowledge production in ways that risk reinforcing existing inequalities. This article introduces the concept of epistemic algorithmic drift to explain how AI-mediated scholarly infrastructures, trained on historically stratified data, generate cumulative and often imperceptible redistributions of epistemic visibility, amplifying gendered, linguistic, and geographic disparities. The contribution is theory-building: drawing on feminist epistemology, intersectionality theory, and critical algorithm studies, it conceptualizes drift as a self-reinforcing feedback loop between structured data and AI-mediated processes of discovery, evaluation, and recognition. In doing so, it distinguishes epistemic algorithmic drift from related notions such as algorithmic bias, concept drift, and path dependency. The article specifies empirically observable indicators, examines the process through an intersectionally informed perspective, and derives governance considerations grounded in its structural dynamics. It argues that addressing these inequalities requires longitudinal, intersectional, and infrastructural interventions rather than isolated technical fixes.