N.M. Mellgren (Kristianstad University) and Sebastian Wärmländer (Stockholm University) have posted Artificial Obedience: How Large Language Models (LLMs) May Normalize Hierarchical Communication on SSRN. Here is the abstract:
In this study, we argue that repeated interaction with large language models (LLMs) may condition users into authority-driven, directive communication habits that can spill over into human-human interaction. LLMs imitate human patterns of communication and create structurally asymmetric interactions where users occupy positions of interactional authority, thereby fostering directive and efficiency-oriented communication patterns that may gradually evolve into enduring habits, schemas or behavioral scripts. Drawing on theories of power, habit formation, and social practice, and using military communication as an analogy, we propose that habits formed during human-LLM communication can transfer to human-human interactions via a process we call computer communication spillover. We conceptualize this process as cue-triggered habit activation in structurally similar contexts. Potential spillover effects include reduced politeness, lower tolerance for dissent, more control-oriented communication, and a reduced tendency to accept other people’s perspectives. The habits likely develop gradually and outside users’ awareness, which raises legal and ethical concerns about responsibility. Taken together, these dynamics highlight a broader societal risk of generative AI, namely, the subtle normalization of hierarchical and instrumental communication styles in everyday human interaction.
To receive new posts from Legal Theory Blog by email, get a free subscription to Legal Theory Stack.
Lawrence Solum