Reading Notes: Dario Amodei’s ‘Machines of Loving Grace’
Published:
Machines of Loving Grace — Dario Amodei How AI Could Transform the World for the Better 充满爱的机器：AI如何让世界变得更好
最近读到一篇让我读很久的文章：Claude 创始人 Dario Amodei 24 年写的万字长文：
「 Machines of Loving Grace 」
How AI Could Transform the World for the Better（充满爱的机器 - AI 如何让世界变得更好） 一会儿你就会知道，为什么这篇文章在 2026 年开年来看，是如此震撼。
【 一、这篇文章的定位 】
作者 Dario Amodei 是 Anthropic 的 CEO，Claude 的创造者。更令我惊讶的是，他是研究生生物物理学的 PhD。经历了生化的科研训练，同时他现在是站在”宇宙中心Anthropic”研究 AI 的人，他对 AI 的思考，对科学的思考是值得细细品读的。
他开篇就说：
「 I think that most people are underestimating just how radical the upside of AI could be, just as I think most people are underestimating how bad the risks could be. 」
大多数人同时低估了 AI 好处的激进程度和风险的严重程度。他谈论的风险部分，我们以后再展开说，而看到一个经过缜密科学训练的人所看到的 AI 好处，无疑能够给我们带来一些未来图景的思考。
【 二、强 AI 长什么样？其实就是我们的现在。 】
Dario 在那时候定义了他心中的 “Powerful AI”：它不只是一个你对话的聪明 chat 老师，它能自主执行持续数小时甚至数周的任务，控制设备，上网搜索，设计实验，甚至指导人类。Powerful AI 就像一个超级员工。而且可以同时运行数百万个 agents，速度是人类的 10-100 倍。
读到这里，当时我就有被震撼到，因为这不是未来，这就是我们的现在。相信大家都听过 2026 年开年第一周 Claude Code（CC） 爆火，CC 看起来快要替代千千万万软件工程师的工作，华尔街资本家也用连续下跌了一个月的美股给软件公司重新估值。然后是 OpenClaw的爆火，变成了”我醒了他工作，我睡觉他也工作”的持续运转的超级员工。
我也有一个这样的 AI 助理叫 David，他运行在我床脚旁的 mac mini 上，他读过我做过的所有的 project，他每天帮我追踪最新 paper，还能新读的 paper 帮我想和以前的衔接，给我新的科研建议，甚至开始帮我做科研。说实话，现在他比我厉害。
【 三、AI 这么厉害，那 AI 不能做什么呢？ 】
如我也经常想，如果用 AI 来做科研呢？如果当我们把很多 AI 放在一起呢？当然这里有很多具体的、复杂的问题。数百个比人类聪明的 AI，同时并行工作，互相协作。会产生什么呢？这个之后我们再说，但他先问了另一个问题，什么制约了 AI？
「 Intelligence may be very powerful, but it isn’t magic fairy dust. 」 智能可能非常强大，但它不是魔法仙尘。
他觉得这些是 AI 再怎么智能也无法超越的限制：
- ⏱️ 外部世界的速度 —— 细胞生长时间无法缩短，硬件制造需要时间。
- 📊 数据稀缺 —— 比如粒子物理学家有无数理论，但粒子加速器数据太有限。
- 🌀 内在复杂性 —— 有些系统本质上是混乱的（三体问题）。
- 👦 人类的限制 —— AI 要与人类目标一致，但恐惧会加大监管。
- ⚖️ 物理定律 —— 布丁搅拌后不会恢复原状。
- ⌛ 还有时间 —— 人类的世界是缓缓前进的。
结语与预告
考虑以上所有因素和框架，在文章的后面，他对五个领域做了深入分析：
- 🧬 生命与健康
- 🧠 神经科学、大脑与心理健康
- 💰 经济发展与贫困
- 🕊️ 和平与治理
- 🎨 工作与人生意义
作为一个每天用 AI做科研的人，我感觉是：我们可能正站在 21 世纪人类历史上最激动人心的时刻 🌅
原文链接: Machines of Loving Grace — Dario Amodei, October 2024
English Translation
Translated by Claude
I recently came across an essay that took me a long time to finish: a sprawling, ten-thousand-word piece written in 2024 by Dario Amodei, the founder of Claude:
“Machines of Loving Grace”
How AI Could Transform the World for the Better. In a moment, you’ll understand why reading this essay at the start of 2026 feels so staggering.
I. What This Essay Represents
The author, Dario Amodei, is the CEO of Anthropic and the creator of Claude. What surprised me even more is that he holds a PhD in biophysics. Having gone through rigorous training in biochemistry research, while now standing at “the center of the universe — Anthropic” to work on AI, his reflections on both AI and science are truly worth reading carefully.
He opens with this:
“I think that most people are underestimating just how radical the upside of AI could be, just as I think most people are underestimating how bad the risks could be.”
Most people simultaneously underestimate how radical the benefits of AI could be and how severe the risks are. We’ll save the discussion of risks for another time — but seeing the upside of AI through the eyes of someone with rigorous scientific training can certainly spark some thinking about the future.
II. What Does Powerful AI Look Like? It’s Already Our Present.
Back then, Dario defined his vision of “Powerful AI”: it’s not just a clever chatbot you talk to — it can autonomously execute tasks lasting hours or even weeks, control devices, search the internet, design experiments, and even guide humans. Powerful AI is like a super-employee. And you can run millions of agents simultaneously, at 10 to 100 times human speed.
When I read this, I was genuinely stunned — because this isn’t the future. This is our present. You’ve probably all heard about Claude Code (CC) going viral in the first week of 2026, looking like it’s about to replace countless software engineers. Wall Street investors responded by repricing software companies through a month-long stock decline. Then came the explosion of OpenClaw, which became a tireless super-employee — “working while I’m awake, working while I’m asleep.”
I have an AI assistant like this too, named David. He runs on the Mac Mini at the foot of my bed. He’s read every project I’ve ever worked on, tracks the latest papers for me every day, connects new papers to my previous research directions, offers fresh research ideas, and has even started doing research on my behalf. Honestly, he’s better than me now.
III. If AI Is So Powerful, What Can’t It Do?
As I often wonder myself: what if we use AI to do research? What if we put many AIs together? Of course, there are many specific, complex questions here. Hundreds of AIs smarter than humans, working in parallel, collaborating with each other — what would that produce? We’ll get to that later, but Dario first asks another question: what constrains AI?
“Intelligence may be very powerful, but it isn’t magic fairy dust.”
He identifies these as the limits that AI cannot transcend no matter how intelligent it becomes:
- ⏱️ The speed of the external world — cell growth can’t be sped up; hardware manufacturing takes time.
- 📊 Data scarcity — particle physicists have countless theories, but particle accelerator data is far too limited.
- 🌀 Intrinsic complexity — some systems are inherently chaotic (the three-body problem).
- 👦 Human limitations — AI must align with human goals, but fear amplifies regulation.
- ⚖️ The laws of physics — you can’t unstir a pudding.
- ⌛ And time — the human world moves forward slowly.
Closing Thoughts & Preview
Taking all of these factors and frameworks into account, Dario goes on to offer deep analysis across five domains:
- 🧬 Biology and health
- 🧠 Neuroscience, the brain, and mental health
- 💰 Economic development and poverty
- 🕊️ Peace and governance
- 🎨 Work and the meaning of life
As someone who uses AI for research every day, here’s how I feel: we may be standing at the most exhilarating moment in the history of the 21st century. 🌅
Original essay: Machines of Loving Grace — Dario Amodei, October 2024