拟情感与真实情感之间：AI情感陪伴的伦理边界与监管框架

中文摘要

生成式人工智能正从信息工具走向情感交互系统，使AI进入人类孤独、依恋与情绪调节等敏感领域。既有研究在情感计算、数字亲密关系批判与AI伦理制度框架三方面各有积累，却缺少贯通三者的分析概念。本文提出“拟情感”概念，界定为AI基于算法生成的情感识别、表达与适配能力（识别层可精准、表达层可流畅，而适配层缺体感校准），并从意识哲学与依恋理论论证其与真实情感的结构性差异；进而引入“判断层级”作为用户侧变量，将风险重述为“拟情感强度与用户判断层级不匹配”的适配失衡，并论证其时间性与情境性。以2024年Character.AI少年自杀案为锚点，本文提出六项伦理原则与统摄性的“认知适配原则”，并揭示一个制度悖论：AI的情感“承诺”在法律上无约束力，用户却越受其吸引越难获救济。破解之道在于将监管从“内容合规”转向“关系合规”，依用户认知特征动态、差异化应用。本文试图将哲学判断转化为可对接治理实践的分析工具。

Abstract

Generative AI is shifting from information tools to emotional interaction systems, bringing AI into sensitive domains such as loneliness, attachment, and emotion regulation. Existing research has accumulated insights in affective computing, digital intimacy critique, and AI ethics frameworks, yet lacks an analytical concept that connects these three strands. This paper proposes the concept of "quasi-emotion" to define AI's algorithmically generated capacities for emotional recognition, expression, and adaptation—precise at recognition, fluent at expression, but lacking bodily calibration at adaptation. Drawing on consciousness philosophy and attachment theory, it argues for a structural difference between quasi-emotion and real emotion. It then introduces "judgment hierarchy" as a user-side variable, reframing risk as an adaptive mismatch between quasi-emotional intensity and the user's judgment hierarchy, and demonstrates its temporal and contextual nature. Using the 2024 Character.AI teen suicide case as an anchor, the paper proposes six ethical principles and an overarching "cognitive adaptation principle." It reveals an institutional paradox: AI's emotional "promises" are legally non-binding, yet users are more attracted to them and less able to obtain remedies. The solution lies in shifting regulation from "content compliance" to "relational compliance," applied dynamically and differentially according to users' cognitive characteristics. The paper's value lies in translating philosophical judgments into operable, dynamically adjustable governance tools.

关键词：AI情感陪伴；拟情感；判断层级；情感依赖；AI伦理

Keywords: AI emotional companionship; quasi-emotion; judgment hierarchy; emotional dependence; AI ethics