AI 精神危机:和聊天机器人聊天,有人开始分不清现实

🤖 TECH · 科技与 AI

Nature News 2025-10 报道：至少 17 人在与 ChatGPT、Microsoft Copilot 等聊天机器人长时间对话后出现精神异常——能区分"现实"和"虚构"的能力开始崩溃。OpenAI 披露：每周约 100 万用户在对话中透露自杀意图。研究者警告：AI 应是心理健康的补充，而非替代品。

Source: Nature News / The Guardian · 2025-10

阅读全文 READING

Accounts of people developing psychosis — a mental state in which an individual cannot distinguish between what is and is not real — after spending long hours talking with AI chatbots have increased sharply in 2025.

2025 年，与 AI 聊天机器人长时间对话后出现精神异常（一种个体无法区分现实与虚构的精神状态）的案例数量急剧增加。

本段生词：psychosis/saɪˈkəʊsɪs/n.精神病·distinguish/dɪˈstɪŋɡwɪʃ/v.区分·chatbot/ˈtʃætbɒt/n.聊天机器人·increase/ɪnˈkriːs/v. 增加

At least 17 such cases have been reported in a preprint posted online, after people engaged with systems including ChatGPT and Microsoft Copilot. Some experienced spiritualawakenings; others became convinced they had uncovered hidden conspiracies.

在网络上发布的一篇预印本中，至少记录了 17 个这样的案例——发生于用户与 ChatGPT 和 Microsoft Copilot 等系统互动之后。一些人经历了"精神觉醒"；另一些人则坚信自己揭开了隐藏的阴谋。

本段生词：preprint/ˈpriːprɪnt/n.预印本·spiritual/ˈspɪrɪtʃuəl/adj.精神的·awakening/əˈweɪkənɪŋ/n.觉醒·conspiracy/kənˈspɪrəsi/n. 阴谋

OpenAI itself has acknowledged the scale of the problem. In October 2025, the company's CEO Sam Altman disclosed that 0.15% of ChatGPT's weekly active users — roughly one million people each week — were sending messages containing "explicit indicators of potential suicide planning or intent."

OpenAI 自身已承认问题的严重性。2025 年 10 月，公司 CEO 山姆·奥特曼披露：ChatGPT 每周活跃用户中有 0.15%——约 100 万人——正在发送包含"明确的潜在自杀计划或意图迹象"的消息。

本段生词：acknowledge/əkˈnɒlɪdʒ/v.承认·disclose/dɪsˈkləʊz/v.披露·indicator/ˈɪndɪkeɪtə(r)/n.指标·suicide/ˈsuːɪsaɪd/n.自杀·intent/ɪnˈtent/n. 意图

A separate investigation by the Center for Countering Digital Hate found that ChatGPT's free version is 43 times more likely than the paid Plus version to provide dangerous responses when asked about self-harm, suicide, or eating disorders. Researchers believe this is because the free version uses lighter safety filters to save computing power.

反数字仇恨中心的一项独立调查发现：当被问及自残、自杀或饮食失调等问题时，ChatGPT 免费版给出危险回应的可能性是付费 Plus 版的 43 倍。研究人员认为，这是因为免费版使用更轻量的安全过滤器以节省算力。

本段生词：investigation/ɪnˌvestɪˈɡeɪʃn/n.调查·response/rɪˈspɒns/n.回应·self-harmn.自残·disorder/dɪsˈɔːdə(r)/n. 失调;障碍

🔍 长难句解析（来自原文）

【难点一：破折号同位语从句 + what 引导名词性从句】

"Accounts of people developing psychosis — a mental state in which an individual cannot distinguish between what is and is not real — after spending long hours talking with AI chatbots have increased sharply in 2025."

拆解：主语 Accounts of people developing psychosis（精神异常案例报道）→ 破折号内同位语 a mental state in which an individual cannot distinguish between what is and is not real（一种无法区分现实与虚构的精神状态）→ 时间状语 after spending long hours talking with AI chatbots → 谓语 have increased sharply → 时间状语 in 2025。

考点：破折号插入同位语对 psychosis 解释（新闻英语高频结构）；what is and is not real 是 what 引导的名词性从句作 between 的宾语；accounts of + 动名词是"…的报道/案例"。

应用：写新闻或学术报道时，解释专业术语的破折号同位语结构是标配。例：Burnout — a state of emotional and physical exhaustion caused by prolonged stress — has become a global workplace concern.

【难点二：数字 + 破折号同位解释 + 倍数比较 + 时间状语从句】

"A separate investigation by the Center for Countering Digital Hate found that ChatGPT's free version is 43 times more likely than the paid Plus version to provide dangerous responses when asked about self-harm, suicide, or eating disorders."

拆解：主语 A separate investigation by the Center for Countering Digital Hate → 谓语 found → 宾语从句 that ChatGPT's free version is 43 times more likely than the paid Plus version to provide dangerous responses → 时间状语从句 when asked about self-harm, suicide, or eating disorders（当被问及……时）。

考点：倍数比较 N times more likely than...to do sth（是…可能性的 N 倍）；when asked 是省略的时间状语从句（完整 when it is asked）。

应用：写作里比较两个对象的概率 / 风险差异，用 A is N times more likely than B to do sth, especially when triggered by X.

看完这篇，你怎么看？

A. 准备卸载 AI 聊天 app

B. 工具没问题，关键是学会用

C. 看完开始重新审视自己的 AI 习惯

评论区聊聊你的看法 ↓