Ethical Challenges in AI Emotional Interaction: Mechanisms of Emotional Dependence and Governance Pathways

This paper focuses on the ethical risks of AI emotional interaction, particularly the formation mechanisms and governance pathways of “emotional dependence.” By reviewing foundational technologies such as sentiment analysis, speech recognition, and NLP, and using LSTM based long term memory reinforcement, Bayesian emotion state transitions, data feedback loops, and real time physiological signal monitoring as the main analytic thread, it reveals how platforms, through multimodal personalization and just in time adaptation, construct “parasocial relationships” that amplify users’ behavioral addiction, anthropomorphic cognitive biases, and social alienation. Drawing on cases including the use of Replika during the pandemic, instances of digital immortality, and AI personas on social platforms, the paper identifies the mental health risks, the weakening of user autonomy, and ambiguity in responsibility attribution induced by emotional dependence. In response, it proposes a system level governance pathway: addiction mitigation design centered on session duration controls and emotional intensity thresholds; enhanced explainability via attention visualization; tiered, scenario aware, and dynamically adjustable permission management calibrated by age and psychological resilience; and mandatory ethics reviews with third party assessments across the full lifecycle. The paper builds a “mechanism–risk–governance” analytical framework and advocates a shift from anthropomorphic to assistive design paradigms, strengthening transparency and accountability to promote prudent innovation and the healthy development of affective AI.