Beyond Bias: Studying ‘culture’ in LLMs and AI chatbots

This paper argues for a conceptual shift in the understanding of Large Language Models (LLMs) withinethnographic and organizational studies, proposing a framework that interprets biases in LLMs not asextrinsic flaws, but as intrinsic culture. Drawing from Bruno Latour's Actor-Network Theory (ANT), thisstudy conceptualizes LLMs as actants that participate dynamically within networks of human and nonhumanentities. By recognizing biases as reflective of the training data's cultural imprints, thisframework positions LLMs as embedded within the cultural, social, and ideological currents that shapeand are shaped by these technologies. The paper argues for a reconceptualization of AI's role in researchand practice, urging a methodological and ethical engagement that embraces the constitutive nature of‘AI culture’ in shaping organizational and societal dynamics.