数字时代的慰藉：AI情感支持的作用机制

Comfort in the Digital Age: The Mechanism of AI Emotional Support

摘要: 随着精神卫生资源短缺与心理健康需求增长的矛盾日益突出，AI情感支持作为一种新兴的数字干预手段，正在从研发实验室走进大众视野。本文系统梳理AI情感支持的心理学研究现状，围绕AI情感支持“是否有效”“如何起效”“何时失效”三个核心问题展开综述。研究发现：(1) 在作用效能层面，AI情感支持对轻中度抑郁、焦虑具有小到中等干预效果，优于空白对照和静态心理教育材料，但无法替代人类咨询师处理复杂情感需求；(2) 在作用机制层面，AI通过认知重构(引导用户抑制负性思维、促进正性思维)、社会支持替代(激活与真实社交类似的神经回路)、行为练习(提供低风险社交训练场)三重路径发挥作用；(3) 在风险层面，AI本身存在共情本质缺陷，表现为存在认知共情而无真实情感共情、仍存在危机干预盲区、具有过度依赖导致的社交退化隐患，以及隐私与算法偏见等伦理缺口。基于上述分析，本文提出AI情感支持的分级应用框架：日常情绪管理可独立使用，辅助治疗需人类监督，危机干预则应设为绝对禁区。未来研究需关注长期追踪、文化适配度与人机协同模式。

Abstract: As the contradiction between mental health resource shortages and growing psychological needs becomes increasingly prominent, AI-based emotional support, as an emerging digital intervention tool, is transitioning from research laboratories into public awareness. This paper systematically reviews the current psychological research on AI emotional support, focusing on three core questions: “effectiveness”, “mechanisms of action” and “potential limitations.” Key findings include: (1) In terms of efficacy, AI emotional support demonstrates moderate to strong intervention effects for mild-to-moderate depression and anxiety, outperforming placebo controls and static psychological education materials, though it cannot replace human counselors in addressing complex emotional needs; (2) Regarding mechanisms, AI operates through three pathways: cognitive restructuring (guiding users to suppress negative thinking and promote positive cognition), social support substitution (activating neural circuits analogous to real-world social interactions), and behavioral practice (providing low-risk social training environments); (3) Concerning risks, AI inherently exhibits empathetic deficits—manifesting as cognitive empathy without genuine emotional resonance, persistent blind spots in crisis intervention, potential social regression due to overdependence, and ethical gaps including privacy concerns and algorithmic bias. Based on these findings, this study proposes a tiered application framework for AI emotional support: standalone daily emotion management, human-supervised adjunctive therapy, and absolute prohibition in crisis interventions. Future research should focus on long-term tracking, cultural adaptability, and human-machine collaborative models.