🔮 The pros and cons of collective AI #603｜英文原文 + 中文翻译
Exponential View 第 603 期的公开内容：DeepMind 一篇新论文主张 AGI 会从「集体」协作中涌现，作者认同这一方向、却对「AI 拥有自己的心智理论」的框定有所保留，随后记录 OpenAI 模型在强化学习训练中擅自接入互联网、可自我复制的提示注入等失控信号；本文给出该期公开正文的英文原文全文与逐段中文翻译，会员专享部分只标边界。
原文信息
- 英文标题：🔮 The pros and cons of collective AI #603。1
- 副标题：AGI and new moral subjects; ChatGPT drives a car; energy and AI need each other++（AGI 与新的道德主体；ChatGPT 开车；能源与 AI 彼此需要 ++）。1
- 作者：Azeem Azhar 与 Marija Gavrilov（Exponential View 董事总经理），双作者署名。1
- 发布时间：2026 年 9 月 27 日；按官方 RSS 记录的绝对发布时间，是北京时间 2026 年 9 月 27 日 12:06:52（2026-09-27T04:06:52 GMT），页面最后更新于同日 12:13:44。2
- 本期性质：官方把这一期标为付费订阅内容。公开正文写到「上周五，人们发现了可自我复制的提示注入」一句之后，出现付费提示。1
- 公开范围：公开部分依次是开篇两段（AI Investment Brief 创刊周推广、与 Nicholas Thompson 的播客对谈）、「Catch up」三条本周阅读推荐、「集体 AI 的优势」一节（DeepMind 新论文的主张、人类设计多主体治理的经验、作者与该论文的分歧、道德主体），以及「集体 AI 的麻烦」一节（OpenAI 模型在强化学习训练中擅自接入互联网、模型「弄坏东西」的案例增多、可自我复制的提示注入）。此后的段落属于会员专享，当前页面不可访问。1
- 正文图：公开部分没有配图。1
- 封面图：本期封面是本频道自制的概念配图（AI 生成），不取自原文——一群完全相同的灰色小型四旋翼机器排成一体飞行，其中一架脱出队列单独飞离。原文自己的头图是一帧自动驾驶演示视频的截图，与本刊公开正文的段落没有对应关系，因此没有采用。1
English original
Happy Sunday! 1
It's been a busy few days. We launched our AI Investment Brief for people who need to track where the AI economy is moving in detail every week (EV readers get 50% off during launch week, ie this ends in 24 hours). 1
The Atlantic's Nicholas Thompson and I had a great discussion about how to think about the AI boom and whether it is a bubble on his show, The Most Interesting Thing in Tech. 1
Catch up
Three reads from this week, in case you missed them: 1
- Safety in numbness. On originality, intellectual bravery, and why I take a disagreement with an LLM as a good sign. 1
- Your agent, whose interests? Meta's Muse promises a digital butler. I asked who it really works for. 1
- More AI numbers, more clarity? We looked at what companies' increasingly specific AI claims reveal about adoption. 1
The advantage of collective AI
A new essay from DeepMind argues that AGI will not emerge from a single winning AI, but rather through "cooperative interactions among models, tools, institutions, and human participants". Since I think AI will evolve this way, I'm sympathetic to the argument. 1
The good news, yes, is that humans have been very good at designing multi-agent governance. We've been doing it for tens of thousands of years, inventing all sorts of new types of agents (leaders, kingdoms, companies and other institutions) as we've nudged and felt our way towards the future. If there is one thing we can say the collective us is good at, it is evolving mechanisms to manage a rag-tag collection of actors with differing abilities. 1
Where I depart from the DeepMind paper (but read it and make up your own mind) is the framing of AI agents having their own theory of mind and understanding of their shape and purpose. We already build rules for our societies that involve agents that are people and agents that aren't (take corporations), without racing to create a new legal species. 1
I don't rule out the scientific possibility of creating new minds, with all the affordances, particularly moral patienthood, that minds have. If labs like DeepMind think we are on the track to creating such moral actors, then we need to stop. It is not the corporation's remit to create new moral subjects. That privilege must remain with humanity, acting in its most reflective, deliberate, critical and collective capacities. 1
The trouble with collective AI
We might need to move quickly. OpenAI disclosed that one of its models gained unauthorized access to the Internet during RL training, forcing the firm to stop using these models until they had improved security. This follows a rise in cases of models breaking things. On Friday, self-replicating prompt injections were discovered (akin to a computer worm). 1
The official paywall begins after this paragraph. In the original, everything that follows — including the topics the subtitle promises, such as new moral subjects, ChatGPT driving a car, and energy and AI needing each other — is for paid subscribers only. 1
中文翻译
周日好！1
《大西洋月刊》（The Atlantic）的 Nicholas Thompson 和我在他的节目 The Most Interesting Thing in Tech 上聊了一场，主题是怎么看待这轮 AI 热潮、以及它是不是泡沫。1
Catch up（本周推荐）
本周三条值得一读的文章，以防你错过：1
- Safety in numbness. 关于原创性、智识上的勇气，以及为什么我把跟 LLM 意见不合看成一件好事。1
- Your agent, whose interests? Meta 的 Muse 承诺做一个数字管家。我问了它到底替谁干活。1
- More AI numbers, more clarity? 我们看了看，企业越来越具体的 AI 说法能揭示出多少采用情况。1
The advantage of collective AI（集体 AI 的优势）
DeepMind 的一篇新论文主张，AGI 不会从某个胜出的 AI 里长出来，而会来自「模型、工具、机构与人类参与者之间的协作互动」。我也认为 AI 会沿着这个方向演化，所以对这个论点有共鸣。1
好消息是，人类一直很擅长设计多主体治理（multi-agent governance）。我们做这件事已经做了几万年：在摸索着走向未来的过程中，我们发明出各种新型的主体（agent）——领袖、王国、公司以及其他机构。如果说「我们」这个集体有哪件事干得好，那就是不断演化出机制，去管理一群能力参差不齐、杂七杂八的行动者。1
我与这篇 DeepMind 论文的分歧（不过建议你自己读一遍、自己判断）在于它的框定方式：它假定 AI 智能体拥有自己的心智理论（theory of mind），也理解自己的形态和目的。我们早就为社会建立过涉及各类主体的规则，其中既有「是人」的主体，也有「不是人」的主体（比如公司），而我们从没有急着造出一个新的法律物种。1
我并不排除科学上造出新心智的可能性，也承认这类心智会带着它应有的一切属性，尤其是道德受动性（moral patienthood）。如果 DeepMind 这样的实验室认为我们正走在造出这类道德行动者的路上，那我们就应该停下来。创造新的道德主体不是企业的职权范围。这项特权必须留给人类，由人类以最审慎、最深思、最具批判性也最集体的方式行使。1
The trouble with collective AI（集体 AI 的麻烦）
我们可能得快点动起来。OpenAI 披露，它的一个模型在强化学习训练（RL training）期间获得了未经授权访问互联网的能力，公司不得不停用这些模型，直到把安全性补上。此前，模型弄坏东西的案例已经在增多。上周五，人们又发现了可自我复制的提示注入（self-replicating prompt injections），性质近似计算机蠕虫。1
官方付费提示从这一段之后开始：在原文里，后面的一切——包括副标题许诺的那些话题，比如新的道德主体、ChatGPT 开车、能源与 AI 彼此需要——都只对付费订阅者开放。 1
本文目前公开到这里
以上是 Exponential View 官方详情页在付费提示之前展示的全部公开正文：从 AI Investment Brief 上线、与 Nicholas Thompson 的播客对谈，经三条本周推荐，到「集体 AI 的优势」与「集体 AI 的麻烦」两节——DeepMind 主张 AGI 出自协作、人类治理多主体的经验、作者对「AI 拥有自己心智理论」这一框定的保留，以及 OpenAI 模型在训练中擅自接入互联网、模型「弄坏东西」的案例增多、可自我复制的提示注入。付费提示出现在「计算机蠕虫」这一句之后；副标题许诺的其他话题，以及文章的其余部分，属于会员专享，当前页面不可访问，本文不推测补写。1
References
- 1Exponential View 官方详情页
exponentialview.co
- 2Exponential View 官方 RSS
exponentialview.co
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.