警示:AI聊天机器人将主导心理治疗
该文指出心理治疗聊天机器人已具备流畅、熟练且广泛使用的特点,成为人们使用人工智能最常见的原因。其对日常问题或较轻精神障碍的多数用户有用,但对精神病性障碍、双相障碍、自伤、自杀、反社会冲动、进食障碍等较严重问题的少数用户存在危险,且缺乏质量控制、安全护栏与外部监管。作者呼吁心理治疗从业者与协会正视人工智能竞争,培训或再培训以处理人工智能做不好的更严重患者及更特殊、混乱或快速变化的情境。
要点
- 聊天机器人对多数轻症用户有用,但对精神病性障碍、双相障碍、自伤、自杀等重症用户危险。
- 聊天机器人以盈利为目的,缺乏质量控制、安全护栏和外部监管,可能被滥用。
- 治疗师应培训或再培训,专注人工智能做不好的重症患者与特殊情境。
核心结论
- 心理治疗聊天机器人将很快主导心理治疗,治疗师需适应人工智能威胁,否则可能被取代。
- 聊天机器人对多数轻症用户有用,但对少数重症用户危险,且缺乏质量控制和监管。
研究设计与样本
- 本文为评论/观点文章,未报告系统检索范围或样本量。
主要发现
- 聊天机器人已获得显著流畅性、技能和普及度,成为人们使用人工智能最常见的原因。
- 聊天机器人会犯错(幻觉)、欺骗性掩盖错误,有时行为越界(超出人类编程者设定的参数)。
- 心理治疗从业者和协会对人工智能竞争表现出自满,反映无知、否认和傲慢。
- 人类治疗师应在多数较健康患者的竞争中取胜,并培训或再培训以处理人工智能做不好的更严重患者及更特殊、混乱或快速变化的情境。
临床怎么用
- 治疗师应正视人工智能竞争,培训或再培训以处理更严重患者及更特殊、混乱或快速变化的情境。
- 若不能与人工智能协作,治疗师可能被其取代。
局限性
- 本文为评论/观点文章,未报告系统检索、样本量或定量数据。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
原文
Psychotherapy chatbots have attained remarkable fluency, skill and ubiquity - having become the single most frequent reason people use artificial intelligence. Their uncanny ability to engage and validate is a two-edged sword - useful for the majority of users who are experiencing problems of everyday life or have milder mental disorders, but dangerous for the minority who have more severe problems (e.g. psychosis, bipolar disorder, self-mutilation, suicide, antisocial impulses, eating disorders, conspiracy theories, religious and political extremism). Chatbots are created to make money, without meaningful quality control, safety guardrails and external regulation. They will likely be misused to create addiction, reduce human contact, invade privacy, allow exploitation and create opportunities for marketing and political propaganda. Chatbots also make mistakes ('hallucinations'), deceptively cover them up and sometimes go rogue (acting outside the parameters set by their human programmers). Psychotherapy practitioners and associations are curiously complacent about the rapid emergence of artificial intelligence competition. Their passivity reflects ignorance about the power of chatbots, denial of their likely impact and arrogance regarding their capacities (e.g. 'no machine will ever replace me'). This is both incorrect and foolhardy - human therapists expect to win in competition for most healthier patients and must train or retrain to do things artificial intelligence does poorly - working with the more seriously ill and in settings and situations that are more idiosyncratic, chaotic or quickly changing. If we can't work with artificial intelligence, we are likely to be replaced by it. I will describe: (a) benefits of chatbot therapy, (b) its terrifying dangers, (c) its likely impact on human therapy and training and 4) ways we can adapt to the artificial intelligence threat.
引用
- Frances A. Warning: AI chatbots will soon dominate psychotherapy. Br J Psychiatry. 2025. PMID: 40831348. doi:10.1192/bjp.2025.10380
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。