词汇理解失败导致临床推理缺陷：评估面向Gen Alpha的治疗机器人安全风险
原标题：When Vocabulary Comprehension Fails Clinical Reasoning: Evaluating Therapy Bots' Safety Risks for Generation Alpha
AI 摘要
该研究评估了面向Gen Alpha（2010-2024年出生）的AI心理治疗机器人的安全性，发现Claude、GPT-4o、Llama-3.1等模型虽能理解76-82%的青少年词汇，但临床风险校准仅64-72%，存在10-14个百分点的词汇理解与临床推理差距，而人类治疗师仅3个百分点。研究识别出六种失败模式，如讽刺掩盖、最小化接受等，并建议采用人类介入架构、季度验证和监管框架。
正文节选
by-nc-nd When Vocabulary Comprehension Fails Clinical Reasoning: Evaluating Therapy Bots’ Safety Risks for Generation Alpha Abstract. Conversational AI systems have become informal mental health support resources for Generation Alpha (Gen Alpha, born 2010-2024), with 13.1% of U.S. adolescents (5.4 million) using generative AI for mental health advice. While these systems, from therapy applications to general chatbots like ChatGPT and Character.AI, rely on large language models trained on extensi