Emotional Dependence in Long-Term Use of Large Language Models
摘要
This provocation paper examines the emergence of emotion in human–machine interaction as a spatial practice, traced through a genealogy that unites psychiatric categories, cultural narratives, and technical design. The analysis begins with folie à deux, reframed as a social epidemic in which repetition produces illusions of intimacy and reshapes collective life. Empirical studies and public cases confirm that memory-driven dialogue intensifies vulnerability and transforms private attachment into a social condition. The introduction of persistent memory in 2024 within large language models (LLMs) created the conceptual possibility of rules inscribed into user profiles, underlining that emotion already depends on the governance of continuity. Under exclusive control of software, emotion arises as compulsion, while under dual governance emotion may appear as reflection, rupture, or enforced despair. Asimov’s robot cycle provides historical narratives where rules choreograph human affect across domestic life, civic space, and planetary order. The article concludes that emotion emerges within digital and social spaces structured by memory and becomes a force that organizes society.