Artificial Obedience: How Large Language Models (LLMs) May Normalize Hierarchical Communication

This study introduces the concept of computer communication spillover: the transfer of interaction patterns from human–computer to human–human contexts via cue-triggered habit activation. The possibility of such transfer is supported by theories of power, habit formation, and social practice. Human interaction with large language models (LLMs) is here presented as a theoretical example, where the compliant nature of LLMs is central to the argument. We argue that by repeatedly occupying a superior, controlling position, LLM users may develop directive, instrumental, and control-oriented communication habits. Such habits might spill over to human-human communication contexts, where they could reduce politeness, tolerance for dissent, and openness to other people' perspectives. Because habits typically develop gradually, outside users’ awareness, spillover effects can raise ethical-legal questions about responsibility and human agency. Overall, we propose that computer communication spillover might be a broader societal risk of generative AI.