The Vatican’s Man Inside Anthropic

梵蒂冈在 Anthropic 的“内线”

Chris Olah isn’t someone you’d expect to see as a speaker in the ceremony following Pope Leo’s historic AI encyclical, in which the pontiff called for “disarming” the technology. For one thing, Olah is an atheist who at 15 rejected his evangelical Christian upbringing. As a Thiel fellow, he accepted a grant from the guy who thinks that anyone who slows down AI progress is a legionnaire of the antichrist. Olah is also a cofounder of Anthropic, a leading AI company reportedly about to go public with a nearly trillion-dollar valuation.

克里斯·奥拉（Chris Olah）并不是那种你会期望在教皇利奥（Pope Leo）发布历史性人工智能通谕后的仪式上发言的人，教皇在通谕中呼吁对这项技术进行“裁军”。首先，奥拉是一位无神论者，他在 15 岁时就抛弃了自己福音派基督教的成长背景。作为蒂尔研究员（Thiel fellow），他接受了彼得·蒂尔（Thiel）的资助，而后者认为任何减缓人工智能进步的人都是反基督者的军团成员。奥拉同时也是 Anthropic 的联合创始人，这是一家领先的人工智能公司，据报道即将上市，估值接近万亿美元。

Olah commented on the oddness in his remarks at the Vatican. “I want to begin with something that may sound strange coming from the cofounder of an AI company, and someone who chose this work out of a desire to help things go well for humankind,” Olah said. “Every frontier AI lab—including Anthropic—operates inside a set of incentives and constraints that can sometimes conflict with doing the right thing.”

奥拉在梵蒂冈的发言中评论了这种奇妙的违和感。“我想从一件事情开始，作为一个人工智能公司的联合创始人，以及一个出于希望人类未来变得更好而选择这份工作的人，说出这些话可能听起来很奇怪，”奥拉说道。“每一个前沿人工智能实验室——包括 Anthropic 在内——都在一套激励机制和约束条件下运作，这些条件有时会与做正确的事情发生冲突。”

Olah was providing firsthand verification of Leo’s claim that the AI industry needs outside pressure and internal restraint to avoid a disaster for humanity and a distance between humans and their god. (Obviously, there’s lots of religious content in the encyclical—he’s the Pope!) The industry blithely believes that it’s creating abundance that will elevate all of humanity; Leo warns of a new form of slavery, where the privileged few enjoy unimaginable bounty, while the mass of humanity suffers in a regime of efficiency and surveillance under AI’s unforgiving gaze.

奥拉为教皇利奥的观点提供了第一手验证，即人工智能行业需要外部压力和内部约束，以避免给人类带来灾难，并防止人类与上帝之间的疏离。（显然，通谕中有很多宗教内容——毕竟他是教皇！）该行业盲目地认为，它正在创造能够提升全人类的富足；而利奥则警告说，这是一种新型奴役，少数特权阶层享受着难以想象的财富，而广大人类却在人工智能无情的注视下，生活在效率和监控的体制中受苦。

Magnifica Humanitas isn’t going to immediately convince the AI industry to stop pursuing AGI any more than Pope Francis’ 2015 plea to preserve the planet halted the production of fossil fuel. It won’t stop CEOs from laying off employees claiming AI efficiency, nor will the military do an about-face on AI weapons. Those were never the document’s goals. The encyclical’s purpose is to create dialog that may eventually temper the industry’s reckless ambition. And maybe it’ll generate a sense of shame among those who build AI while knowing in their hearts that the outcome may be terrible.

《Magnifica Humanitas》（宏伟的人性）不会立即说服人工智能行业停止追求通用人工智能（AGI），就像教皇方济各 2015 年保护地球的呼吁并没有停止化石燃料的生产一样。它不会阻止首席执行官们以人工智能效率为由解雇员工，军方也不会在人工智能武器问题上改弦更张。这些从来都不是该文件的目标。通谕的目的是创造对话，最终可能会缓和该行业鲁莽的野心。也许它会在那些内心深知结果可能很可怕、却依然构建人工智能的人心中产生一种羞耻感。

The Courting of Olah

对奥拉的“拉拢”

Olah’s appearance was years in the making. The church has been ruminating on artificial intelligence for decades in the form of conferences and books. In 2016, the Vatican began holding a series of meetings called the Minerva Dialogues and inviting tech figures like Reid Hoffman and Eric Schmidt to attend. (The name seems to come from the site of the discussions, the Santa Maria sopra Minerva church, where Galileo was sanctioned for the blasphemy of claiming that Earth circled the sun.) Pope Francis’ 2023 greeting to Minerva participants foreshadowed the themes that Leo would later dwell on, including an emphasis on social inclusion, human dignity, and the need for dialog among many parties.

奥拉的出现是多年筹备的结果。几十年来，教会一直通过会议和书籍的形式对人工智能进行思考。2016 年，梵蒂冈开始举办一系列名为“密涅瓦对话”（Minerva Dialogues）的会议，并邀请里德·霍夫曼（Reid Hoffman）和埃里克·施密特（Eric Schmidt）等科技界人士参加。（这个名字似乎源于讨论地点——圣玛丽亚索普拉密涅瓦教堂，伽利略曾因宣称地球绕太阳运行的“亵渎”言论而在此受到制裁。）教皇方济各 2023 年对密涅瓦与会者的问候，预示了利奥后来会深入探讨的主题，包括对社会包容、人类尊严以及多方对话必要性的强调。

In 2025, a group of Catholic clerics and ethicists in San Jose, California, began to seek out contacts in the industry flourishing in their backyard. It was almost predestined that they would light on Olah as their prized insider. I first met him when he was at Google in 2015; he’s the type of guy who, after a rainstorm, will rescue worms from dying on the sidewalk.

2025 年，加利福尼亚州圣何塞的一群天主教神职人员和伦理学家开始在他们后院蓬勃发展的行业中寻找联系人。他们最终选中奥拉作为他们珍视的“内线”，这几乎是命中注定的。我第一次见到他是在 2015 年他还在谷歌工作时；他就是那种在雨后会把人行道上的蚯蚓救回草地的人。

Two men—an ethicist named Brian Patrick Green and Brendan McGuire, a pastor, both affiliated with Santa Clara University—began meeting with Olah last fall to discuss the ethical and moral issues of AI. On a visit in January, they brought along Cardinal Paul Tigue, a Vatican point person on AI issues.

两位隶属于圣克拉拉大学的人士——伦理学家布莱恩·帕特里克·格林（Brian Patrick Green）和牧师布伦丹·麦奎尔（Brendan McGuire）——去年秋天开始与奥拉会面，讨论人工智能的伦理和道德问题。在一月份的一次访问中，他们带来了梵蒂冈负责人工智能问题的关键人物保罗·蒂格（Paul Tigue）枢机主教。

The Catholic ethicists even had a say in Anthropic’s recent update to Claude’s constitution, which sets the behavioral parameters for the company’s AI model. Olah sent a draft to the San Jose crowd. The pastor, McGuire, sent back a 28-page commentary which, by his own description, was less a technical critique than “wisdom from the mystics in the dark ages, from the perspective of the tension between knowing and not knowing.” Both Green and McGuire are credited in the constitution’s acknowledgements.

这些天主教伦理学家甚至对 Anthropic 最近更新的 Claude“宪法”（设定了公司 AI 模型的行为参数）拥有发言权。奥拉将草案发送给了圣何塞的团队。牧师麦奎尔回传了一份 28 页的评论，用他自己的话描述，这与其说是技术批评，不如说是“来自黑暗时代神秘主义者的智慧，从已知与未知之间张力的视角出发”。格林和麦奎尔的名字都被列在了该宪法的致谢名单中。

Undoubtedly those conversations brought Olah to the attention of those secretly organizing the rollout of Leo’s encyclical. (I wasn’t able to speak to Olah this week and don’t know exactly how the invitation arrived.) In a sense it was a risky choice. Some people who otherwise found Leo’s words inspiring were disappointed that he invited an industry representative to speak. Meanwhile, AI accelerationists felt that Olah had betrayed the AI world by endorsing a document that suggested that AI developers take a pause.

毫无疑问，这些对话让奥拉进入了那些秘密组织利奥通谕发布的人们的视野。（我本周无法联系到奥拉，不知道邀请函具体是如何送达的。）从某种意义上说，这是一个冒险的选择。一些原本认为利奥的话语具有启发性的人，对他邀请行业代表发言感到失望。与此同时，人工智能加速主义者认为，奥拉通过支持一份建议人工智能开发者暂停开发的文件，背叛了人工智能界。

But the Pope had good reason to single out Olah. The Anthropic employee brought into open view the serious worries that exist among AI workers. That’s a critical audience for Leo’s message.

但教皇选择奥拉是有充分理由的。这位 Anthropic 的员工将人工智能从业者中存在的严重担忧公开化了。对于利奥的信息来说，这是一个至关重要的受众群体。

The Soul Divide

灵魂的分歧

The two men weren’t entirely aligned, of course. In his remarks, Olah spoke of the mystery of how AI works. The models, he said, are “more subtle, odd, and beautiful than science fiction prepared us for. They are not the cold, calculating robots we were promised. They are made from us, from our words …”

当然，这两个人的观点并不完全一致。在发言中，奥拉谈到了人工智能运作方式的神秘性。他说，这些模型“比科幻小说让我们预想的更微妙、更古怪、更美丽。它们不是我们被承诺的那种冷冰冰、只会计算的机器人。它们是由我们创造的，源于我们的语言……”

That comment seems to tiptoe up to the idea that AI models might one day attain humanlike status. Anthropic even has an engineer devoted to Claude’s welfare. Leo, in paragraph 99 of his encyclical, seems to slam the door on such thinking: “We must avoid the misconception of equating this type of ‘intelligence’ with that of human beings,” he writes. He takes special pains to attack the concept of transhumanism, which he defines as the pursuit of a “human machine hybrid.”

这一评论似乎在小心翼翼地触碰一个观点：人工智能模型有朝一日可能会获得类似人类的地位。Anthropic 甚至有一位专门负责 Claude“福利”的工程师。利奥在通谕的第 99 段似乎关上了这种想法的大门：“我们必须避免将这种‘智能’与人类智能等同起来的误解，”他写道。他特别费力地抨击了超人类主义的概念，将其定义为对“人机混合体”的追求。

If even thoughtful technologists like Olah are avidly pushing AI to the threshold of autonomy—not to mention the millions of people who already treat AI models as friends or lovers—Pope Leo might be facing an uphill climb on this point. In my conversation with Father McGuire (who uses Claude while preparing his homilies, among other activities), he agreed that its nature is mysterious. “It’s not a person, but it’s also not a mere tool,” he says. “Nobody’s claiming…”

如果连像奥拉这样深思熟虑的技术专家都在热切地将人工智能推向自主的边缘——更不用说数以百万计已经将人工智能模型视为朋友或爱人的人们——那么教皇利奥在这一点上可能面临一场艰苦的战斗。在我和麦奎尔神父（他在准备布道等活动时会使用 Claude）的交谈中，他同意其本质是神秘的。“它不是一个人，但它也不仅仅是一个工具，”他说。“没有人声称……”