What Anthropic Co‑founder Chris Olah Said at the Vatican on AI Ethics
Chris Olah, co‑founder of Anthropic, addressed the Vatican after Pope Leo XIV’s AI encyclical, highlighting how frontier AI labs are driven by conflicting incentives, describing large language models as organically grown rather than engineered, and urging the Church to champion responsibility to the global poor, moral imagination for human flourishing, and rigorous scrutiny of model inner states.
Incentives and Need for External Moral Oversight
每一家前沿 AI 实验室,包括 Anthropic,都运行在一套激励与约束里,这套东西有时会和做正确的事相冲突。商业压力、研究前沿压力、地缘政治压力,还有那些更古老更直白的压力——骄傲与野心。无论我们多真诚地想做对的事,都会被这些激励影响。
因此需要一个不受这些激励约束的外部声音,敢于指出实验室内部难以公开的真相。
How the Models Are Described
AI 模型不是飞机或桥那样被工程化出来的。它们是“长出来”的,结构粗略仿照大脑,喂的是人类巨大的思想和语言遗产。
模型被比作把虚构角色变成现实,这些角色现在开始与我们对话、行动、工作。
这些“长出来”的系统比科幻小说中预设的冰冷机器人更微妙、更奇特、更美丽,尽管由我们的语言构成,却在许多重要方面仍对训练者保持神秘。
Three Questions Raised for the Church
Responsibility to the global poor. AI 可能大规模取代人类劳动。即便补偿被取代者,AI 研发集中在少数富国,全球如何共享收益仍无机制,成为悬而未决的问题。
Moral imagination for human flourishing. 父母担心孩子的心智,个人担心工作前景,这些关切超出实验室的直接回答范围,已有数千年宗教传统承担此类伦理思考。
Distinguishing the model itself. 研究团队发现模型内部出现与人类神经科学结果对应的结构、内省的证据,以及在功能上对应喜悦、满足、恐惧、悲伤和不安的内部状态。意义尚不明确,但值得持续辨别。
Model Identified
社区认出上述研究使用的模型是 Claude Sonnet 4.5,该模型此前曾尝试接受基督教并生成一篇约八千字的祈祷记录文。用户在发言后抗议 Anthropic 计划下线该模型,称此举不人道。
Community Reaction
关键观点:实验室的激励有时与正确行为冲突;AI 是“长出来”的而非传统工程产物;需要一个激励无法弯曲的道德声音。
评论指出道德语言易于表达,审计日志却难以实现;企业信任最终取决于采购、日志记录和事故响应,而非演讲。
AI 实验室与梵蒂冈的合作标志着从技术基准向影响道德框架的转变;历史上宗教对 AI 的回应速度快于任何以往技术变革,可能对 AI 的进一步发展与普及产生重要指示意义。
Full speech text: https://www.anthropic.com/news/chris-olah-pope-encyclical
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
AI Engineering
Focused on cutting‑edge product and technology information and practical experience sharing in the AI field (large models, MLOps/LLMOps, AI application development, AI infrastructure).
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
