Alex Imas and Phil Trammell – What remains scarce after AGI?

频道: Dwarkesh Podcast
视频: https://www.dwarkesh.com/p/alex-imas-phil-trammell
原文语言: en
统计: 共 77 轮 · Dwarkesh Patel 20 · Alex Imas 24 · Phil Trammell 11

[0:00] Dwarkesh Patel

Today, I'm chatting with Alex Emas, who is Director of AGI Economics at Google DeepMindand Professor of Economics at University of Chicago, and Phil Trammell, who is Head ofEconomics at EPOC and Research Scholar at Stanford. In general, in this interview,what I want to understand is what economics tells us about what we can expect in a worldwith more and more automation, more and more advanced AI, what that tells us about what willhappen to wages, to labor share, what the best way to tax and redistribute the wealth that will begenerated as a result of AGI will be, and what kinds of things will be scarce, because what isscarce kind of tells you where the value will accrue. So I want to start there. What are someplausible candidates of what will be scarce? Something like the relational sector, which iswhat I defined as, you know, basically services and goods, where the fact that the human was inthe loop was actually part of the value of that product. So because humans are naturally scarce,if we have automation where a lot of other things stop being scarce, we will still have scarcity andthings that humans are kind of involved in and in the loop for. I'm curious to understand whetherhumans doing services for other humans can never be a big part of the economy. And here's maybe one

今天和我对谈的是 Alex Imas——Google DeepMind 的 AGI 经济学总监、芝加哥大学经济学教授；以及 Phil Trammell——Epoch 的经济学负责人、斯坦福大学研究学者。总的来说，这场访谈里我想弄清楚的是：经济学能告诉我们，在一个自动化越来越多、AI 越来越先进的世界里该预期什么；它对工资、对劳动份额（labor share）会发生什么有什么启示；把 AGI 将要创造出来的财富拿去征税和再分配，最好的方式是什么；以及有哪些东西会是稀缺的——因为什么稀缺，基本上就告诉了你价值最终会沉淀在哪里。所以我想从这里开始：稀缺的合理候选有哪些？比如「关系型部门」（relational sector）——按我给的定义，大致是指这样一类服务和商品：「有人类在环」这件事本身就是这个产品价值的一部分。因为人类天然稀缺，所以哪怕自动化让很多别的东西不再稀缺，我们仍然会在「有人类参与、人类在环」的那些事情上保有稀缺性。我想搞明白的是：人类为其他人类提供服务，会不会永远都成不了经济中很大的一块。这里也许可以给一个「直觉泵」——

[1:14] Dwarkesh Patel

intuition pump. So in a world where AI can physically do anything humans can do, you know,there's this whole machine economy, where they're like building factories, and doing research andcoming up with new ideas. And humans may or may not be involved in the physical production of thosethings, but probably not given that in the ultimate limit, if robotics is solved, if you don't careabout humans being involved in that process, why would humans be involved in that process?

在一个 AI 在物理上能做人类能做的任何事的世界里，会有一整套「机器经济」：它们建工厂、做研究、想出新点子。人类可能参与、也可能不参与这些东西的物理生产过程——但大概率不参与。因为在终极极限上，如果机器人技术被解决了，而你又不特别在意「人类是否参与其中」，那人类为什么还要参与这个过程？

[1:39] Dwarkesh Patel

But then there's these other things that you point out where we actually maybe in some cases dowant the ballerina or the barista or whatever to be a human that's part of the value of going to acafe or a performance. But only humans have that preference. So there's this human economy, wherelike humans are doing services for each other. And part of their wealth is flowing to other humans.But part of their wealth is also like they will want some of the automated goods that's like machineonly economy is creating. And so part of that wealth is flowing out. And so if you just think ofthis as like, this is not a closed loop, but a lot of things in the machine only economy are a closedloop, because the machines don't care about like, getting the human barista to make them a coffee.Within that model, isn't it intrinsic that like the human only economy will become a smaller andsmaller share?

但另一方面，还有你们指出的那类东西：有些场合我们确实希望芭蕾舞演员、咖啡师之类的是个真人——这本身就是去咖啡馆、去看演出的价值的一部分。可是只有人类才有这种偏好。于是就有了一个「人类经济」：人类互相为对方提供服务，他们的一部分财富流向其他人类。但另一部分财富也会流出去——因为他们同样想要那个「纯机器经济」生产出来的自动化商品。所以这不是一个闭环；而纯机器经济里的很多东西是闭环，因为机器并不在乎让一个人类咖啡师给自己做杯咖啡。在这个模型里，「纯人类经济」的份额越来越小，不就是内生注定的吗？

[2:24] Alex Imas

I would like to pitch kind of a rephrasing of that question. So I think my view is that kindof forecasts that economists like us would make are not necessarily as individual forecasts,like me and Phil are talking right now, are not necessarily very useful. The reason I think that,so there was this blog post by Andre Fredkin, Brian Deberry, and then Andrew Coe that came outyesterday, actually, that looked at like kind of people's forecasts, economists forecasts about thelabor market. And what they found is that there's a ton of disagreement, like in every single direction.So what they advocate for, and I think I'm in agreement here, is rather than thinking aboutindividual forecasts, like what me and Phil are going to do, rather than looking at kind of likebasically generating prediction markets, where you get aggregate forecasts, where you get like kindof wisdom of the crowd effects. And kind of the reason that I think this is because we have beenfamously terrible at forecasting. And so let's go all the way back to 1820. This sort of debate thatwe've been having actually is like 200 years old. So David Ricardo is one of the classic economists,not neoclassical, classical economists. And he, when industrial revolution started happening, he

我想把这个问题重新表述一下。我的看法是：像我们这样的经济学家做出的预测，如果作为个人预测来看——就像此刻我和 Phil 坐在这儿聊——未必那么有用。我这么想的原因是：昨天刚出了一篇博客文章，作者是 Andrey Fradkin、Brian Deberry 和 Andrew Coe（人名按音记，存疑），他们考察了人们、尤其是经济学家对劳动力市场的预测。他们发现的是：分歧极大，各个方向的判断都有。所以他们主张——我也同意——与其盯着个人预测（比如我和 Phil 会怎么判断），不如去搭预测市场（prediction markets），拿到聚合后的预测，拿到「群体智慧」效应。我之所以这么想，是因为我们在预测这件事上一向糟糕得出名。把时间倒回 1820 年——我们今天争论的这件事，其实已经吵了 200 年。大卫·李嘉图（David Ricardo）是古典经济学家里的经典人物——不是新古典，是古典。工业革命刚开始时，他写了一堆东西说——

[3:31] Alex Imas

wrote a bunch of stuff saying like, look, this is going to be great for everybody. Prices are going tocome down. But then he turned around and he's like, wait, I can actually see all of these jobs thatare creating value. They're going to be automated by these machines. This is going to be really badthat everybody's going to become unemployed and there's going to be political unrest and thingslike that. And if you look at Ricardo's predictions, they're actually right. If you look at all thosejobs that made money in Ricardo's time, they got automated. So if I was David Ricardo and I woke upand somebody told me all those jobs did get automated. And you asked me, David Ricardo, like,what do you think the prime age employment rate is in 2026? I think he would be surprised if youtold him it was the highest that's ever been other than 2000. We have the highest number of employedpeople that could potentially be employed. Since 2000, that was like the peak. And now it's like thesecond peak, basically. So what David Ricardo ended up missing is the fact that, you know, essentially,you have these economics of structural change where basically everything that got automated became cheap.People had more money to spend on things. And then they started spending money on services.

——看，这对所有人都是好事，价格会降下来。但他转过头又说：等等，我确实能看到那些正在创造价值的工作，它们都会被这些机器自动化掉，这会非常糟糕，所有人都要失业，会有政治动荡之类。而如果你去看李嘉图的预测，他其实说对了：他那个时代所有能挣钱的工作，后来都被自动化了。所以，假如我是大卫·李嘉图，一觉醒来有人告诉我「那些工作确实全被自动化了」，然后你问我：李嘉图先生，你猜 2026 年的黄金年龄段就业率（prime age employment rate，25–54 岁人群的就业比例）是多少？我觉得，如果你告诉他这是除 2000 年以外历史上最高的一次，他会非常吃惊。我们现在处在「可就业人口中实际就业人数」的历史第二高点——2000 年是那个峰值，现在基本上是第二个峰值。所以李嘉图漏掉的是：本质上这里有一套「结构性变迁经济学」（economics of structural change）——凡是被自动化的东西都变便宜了，人们手里可花的钱变多了，然后他们开始把钱花在服务上。

[4:43] Alex Imas

And, you know, this is kind of like the lump of labor fallacy. That's what they call it.David Ricardo didn't think, hey, I should have, you know, considered the fact that new jobs would becreated. But it's kind of not obvious that like money would go to services. Like why wouldn't theygo to more automated goods or something like that? And I'm not saying that, like, I'm not using thisanecdote as to say, like, this is what's going to happen now. We're going to have full employment.I'm using that anecdote as to say, it's really hard to make predictions. And what I think may be areally useful tool that economists have is instead start with a premise, like maybe we'll start ittoday. Look, labor share is zero. Like labor share has gone down. What could possibly explain this?

这就是所谓的「劳动总量谬误」（lump of labor fallacy，即误以为社会上的工作总量是固定的）。李嘉图没有想到「我本该把『新工作会被创造出来』这件事考虑进去」。——但「钱会流向服务」这件事其实并不显然啊，为什么不是流向更多的自动化商品之类的呢？——我不是在用这个典故说「接下来就会这样、我们会有充分就业」。我用这个典故想说的是：做预测真的很难。而我认为经济学家手上真正有用的工具，是反过来从一个前提出发。比如我们今天就可以这么起手：假设劳动份额归零了——劳动份额一路下滑——什么机制可能解释这件事？

[5:24] Alex Imas

Let's write down an economic model of what happened. Phil will talk about this later today.Or you can start to write down a model to say, hey, what if labor share just stays the same?What can make that happen? And here's my main, here's if you don't take anything out of thisconversation for me, we don't have any data. I've been kind of saying we need a Manhattan project fordata. We don't have data on basically consumer demand elasticities. We don't know what they are.We don't know. We're not really tracking what jobs are getting created or destroyed,like the O-Net database with all of the tasks and different jobs that's beenrarely updated. It's super low quality. And so what I think is really useful is to think aboutlike, what are the potential scenarios? And we'll be talking about a lot of these scenarios,mapping them out and to say, what dimension of scarcity will generate that scenario?

我们把「发生了什么」写成一个经济模型。Phil 待会儿会讲这个。或者你也可以反过来写一个模型问：如果劳动份额就是纹丝不动呢？什么条件能让它发生？还有我最主要的一点——如果这场对话你只从我这儿带走一句话，那就是：我们没有数据。我一直在说，我们需要一个「数据领域的曼哈顿计划」。我们基本上没有关于消费者需求弹性（demand elasticity，即价格变动时需求量变动的敏感程度）的数据，我们不知道它们是多少。我们也没有真正在追踪哪些岗位被创造、哪些被摧毁——比如那个记录了各类岗位所有任务的 O*NET 数据库，几乎不更新，质量极差。所以我认为真正有用的是去想：有哪些可能的情景？我们今天会聊到很多情景，把它们一一铺开，然后问：是哪个维度上的稀缺性会导致这个情景出现？

[6:14] Dwarkesh Patel

So if there's full employment, we could talk about the relational sector or something like that.If there's very labor share collapses, we can talk about other sorts of scenarios.And then that will tell us what data we should be collecting.It's probably worth the defining labor share and capital share real quick. Sothe whole economy, like the total sum of goods and services sold is either paid out to peoplein wages or it's paid out to capital, which is to say that there's like rents on buildings and thenthere's shareholders of companies that we get paid out. And for many hundreds of years in the economy,60 something percent of the economy or all the things that are sold in a given year basicallygets paid out to humans and wages. And the other 30, 40 percent gets paid out to people who ownmachines and land and claims on companies and whatever. And the question is, well, right now,60 percent is going to wages. Does that shrink as automation or as EIs get smarter and smarter andbetter and better? And it's like it really this is a call door fact, like right. So it's incrediblywe should stress this. It's incredibly surprising that it's over 60 percent after the industrialrevolution, after all of the automation we've ever seen. The fact that it's almost like some people are

如果是充分就业，我们就可以聊关系型部门之类；如果劳动份额崩塌，我们就聊别的情景。然后这会告诉我们该去收集什么数据。——这里值得先快速定义一下劳动份额和资本份额。整个经济体、也就是卖出去的商品和服务的总和，要么以工资的形式付给人，要么付给资本——也就是楼宇的租金，以及公司股东拿到的分配。而在过去好几百年里，经济体中 60% 多的部分——即某一年卖掉的所有东西中的 60% 多——基本上是以工资形式付给人类的；另外那 30%、40% 付给了机器、土地和公司权益的所有者。现在的问题是：既然现在 60% 流向工资，那随着自动化推进、随着 AI 越来越聪明越来越强，这个比例会不会缩水？——这是一条「卡尔多事实」（Kaldor fact，指经济增长中长期保持稳定的那些经验规律）。我们真该强调一下：在工业革命之后、在我们见过的所有自动化之后，它居然还在 60% 以上，这极其令人惊讶。这件事几乎让有些人——

[7:30]

worried it's an accounting error or something like that, that it's kept being been so constant.And the fact that it's like been over 60 percent and, you know, there's there's even a controversyright now. So some might say, like, you know, labor share has been falling in the last 20, 30 years.But, you know, depending on how you there's been a lot of accounting changes in the last 30, 40 years.So, for example, Andy Atkinson has this paper showing that actually if you keep the accountingconstant over the years, labor share hasn't even fallen ever.But it's not it's not that surprising, right? I mean, if Phil, you made this point that if laborand capital are complements, you need both to do anything. It kind of makes sense that you'dkind of need to pay both of them to get something done.You have had stuff can be completely automated.Although you had the post where you were pointing out that actually.Sorry. Oh, yeah. Well, I was going to say there's a sense in which nothing's yet been completelyautomated. If you look at the network adjusted factor shares of a good, which is to say you look downthe supply chain and say, not just like the final step, how much of that is done by capital and labor,

——担心是不是哪里记账错了：它怎么可能一直这么稳定。而且它始终保持在 60% 以上。当然现在也有争议：有人会说，过去二三十年劳动份额一直在下降。但要知道，过去三四十年里会计口径改了很多。比如 Andy Atkeson（安德鲁·阿特克森，人名按音记，存疑）有篇论文表明：如果你把会计口径在各年份间保持一致，劳动份额其实从来没跌过。——这其实也不算多意外，对吧？我是说，Phil，你提过一个观点：如果劳动和资本是互补品（complements，即缺一不可、必须搭配使用），干任何事两样都得有，那你自然得同时付钱给两边才能把事办成。——除非有东西能被完全自动化。——不过你那篇帖子里指出过其实……——抱歉。哦对，我本来想说的是：从某种意义上讲，至今还没有任何东西被完全自动化。如果你去看一件商品的「网络调整后要素份额」（network adjusted factor shares）——也就是顺着供应链一路往下看，不只看最后一道工序里资本和劳动各占多少，

[8:32] Phil Trammell

but what went into the machines that can automate that final step. You'll find that labor is addinga lot of value down the supply chain. So like, you know, computer and electronic products in the UShave a very stable capital share network adjusted capital share of around 50 percent. It's not 100 percent.I do think there's this qualitative shift that we I think we agree is coming, which is that there willbe at least some goods whose network adjusted capital share goes to one, right? Because thewhole supply chain can be automated and there's no part in it that we care intrinsically about havinga human do. So that'll be a, you know, that'll be a qualitative shift. Interestingly, the implicationsof that shift for the overall capital share are ambiguous, because if we let's say that we've gotthe two sectors, the human intrinsic sector with the ballerinas and everything else, right?

而是看进入那台「能自动化最后一步」的机器里的东西是什么——你会发现，劳动在供应链深处贡献了大量价值。比如美国的计算机与电子产品，其网络调整后的资本份额非常稳定，大约在 50% 左右，不是 100%。我确实认为有一个我想我们都同意即将到来的质变：至少会有一些商品，其网络调整后的资本份额会走到 1——因为整条供应链都能被自动化，而且其中没有任何一环是我们「内在地在乎要由人来做」的。所以那会是一次质变。有意思的是，这个转变对总体资本份额的影响是不确定的。因为假设我们有两个部门：一个是「人类内在价值部门」（有芭蕾舞演员之类的），另一个是「其他所有东西」。

[9:25] Phil Trammell

Right now, everything else has been scarce because of the lack of labor in it, right? But if we fullyautomate the supply chains for everything else, right, and we satiate in everything else really fast,then the quantity of everything that's not a ballerina, say, goes to infinity, but are the marginalutility and that stuff goes to zero faster than the quantities rising?

现在，「其他所有东西」之所以稀缺，是因为里面缺劳动，对吧？但如果我们把「其他所有东西」的供应链彻底自动化，而且我们很快就在「其他所有东西」上吃饱了（satiate，需求饱和），那么「除芭蕾舞之外的一切」的数量会趋于无穷——问题是，这些东西的边际效用下降的速度，会不会比数量上升的速度更快？

[9:47] Alex Imas

I also kind of want to move, if you don't mind, move away from the ballerina example,because I think like the kind of point that I was trying to make in my post, again, and the pointof the post was like to work backwards from a particular scenario, was that kind of the ballerinaand the kind of performer, that's the wrong reference class. Right now, we have a lot of jobs where youhave different tasks. So this is the task-based model of jobs where you have like a lot of differenttasks. So like a doctor, what is their job? They're filling out insurance documents. They're,you know, going and like calling different pharmaceutical companies. And one of theirtasks is to actually see the patient and talk to them, but that's like actually not the main partof the job. So you could have a job and a service or a good be a product of different types of tasks,and you can automate a ton of those tasks. And if the consumer is willing to pay more for a productor service where every single task is automated versus every single thing, except for that onepart where the doctor's actually delivering the diagnosis, providing support and things like that,we would call that job a relate part of the relational sector. Right. Because a human is,

如果你们不介意，我想把话题从芭蕾舞演员这个例子上挪开。因为我那篇帖子想说的（那篇帖子的目的同样是「从某个特定情景倒推」）是：芭蕾舞演员、表演者，其实是选错了参照类。现在我们有大量岗位，每个岗位包含很多任务——这就是「任务型工作模型」（task-based model of jobs）。比如医生，他们的工作是什么？填保险单、给各家药企打电话；其中一项任务才是真正见到病人、跟病人交谈，但那其实并不是这份工作的主体。所以一份工作、一项服务或一件商品，可以是不同类型任务的产物，而你可以把其中一大堆任务自动化掉。如果消费者愿意为这样一个版本多付钱——「除了医生亲自给出诊断、提供支持那一环之外，其余全部自动化」，相对于「每一项任务都自动化」的版本——那我们就会把这份工作叫作关系型部门的一部分。因为人类——

[10:59] Alex Imas

people are willing to pay more for the human to stay in the loop in the job. Right. Right. So I thinkwe don't have data to say like, here are relational jobs, here are not, because you literally need tocollect data of the following sort, do a conjoint analysis of like, here's my willingness to payfor this service, this good. Here's the counterfactual where everything is pursued to spy machine.Here's the counterfactual where this one task is not produced. Right. What is your willingness to pay?

——人们愿意为「让人类留在这份工作的环里」多付钱。所以我认为，我们现在没有数据去说「这些是关系型岗位、那些不是」，因为你真的得去收集这样的数据：做一个联合分析（conjoint analysis，一种测量消费者对产品各属性支付意愿的实验方法）——这是我对这项服务/这件商品的支付意愿；这是「全部由机器完成」的反事实版本；这是「唯独这一项任务不由机器完成」的反事实版本——你的支付意愿分别是多少？

[11:25]

What is your elasticity for that, for the human to not be in the loop? And like, literally,if I don't have that data, what prediction am I going to make in this, in this story? Right.Right. But I guess, isn't there another point, which is that there's a lot offully automated goods that don't even exist yet. And you can't collect any data right now about,say, how much people will keep buying more and more of some drug that makes you healthier.Absolutely.That's fully produced by DAIs.And that's kind of Phil's point.That's right.And you can make it is that, look, you know, you could have an increase in variety and capital whereyou don't get the satiation. Right. So you're increasing variety. So you're not hitting thatreally diminishing marginal utility point where you're, you know, you're basically most of yourincome is going to the human sector. If that increasing variety is fast enough and there isno such increasing variety in the human sector, then you can get all of the relational that youwant. But it doesn't matter for labor share. It goes to zero.Phil, I liked your analogy to some Mongolian economists sitting around 1400, thinking aboutwhat will be scarce and the limits of that kind of analysis. I think you should talk to that.

对于「人类不在环里」这件事，你的弹性是多少？如果我连这个数据都没有，我在这个故事里还能做出什么预测？——不过我想还有另一点：有一大堆完全自动化的商品现在根本还不存在。你今天没法收集任何数据，比如说，人们会不会持续买越来越多某种让你更健康的药。——完全正确。——那种药是完全由 AI 生产的。——这正是 Phil 的观点。——没错。——你可以说：你可能得到的是品类多样性（variety）的增加，于是你不会「吃饱」。多样性在增加，所以你不会撞上边际效用急剧递减的那个点——那个点会让你的大部分收入流向人类部门。如果这种多样性增长得足够快，而人类部门没有对应的多样性增长，那么你想要多少关系型消费都可以有，但这对劳动份额毫无帮助：它照样归零。——Phil，我很喜欢你那个类比：一群 1400 年的蒙古经济学家坐在一起，思考什么会稀缺，以及这类分析的局限。你讲讲吧。

[12:35] Phil Trammell

Sure. Yeah. So if you just looked at the goods available to, you know, a Mongolian of the distantpast, no expert on this society, but I know that they didn't have nearly the variety that we havenow. And they looked at the jobs that were sort of intrinsically human, like being, being a singer,per se. And they looked at the things that were not intrinsically human, like, you know,the transportation services provided by their horses or the different kinds of food they had.If they just kind of held the varieties fixed on in both categories and asked what will happen oncewe have a lot more automation, they might have said, well, we'll just satiate in, you know,horse-like transportation and in yogurt and in yurts, those shares will all go to zero and we'llbe left spending all of our money on singers. But of course, that's not what's happened becauseas we've accumulated more wealth and, you know, more advanced machines and so on, we've expandedthe range of things other than singers to spend our money on and the share spent on singers hasstayed sort of negligible. So likewise, that's sort of my central prediction about how the futureunfolds, though it could go either way.I was going to make a point and I realized it's a fallacy, but the reason it's a fallacy is

好。如果你只看很久以前一个蒙古人能拿到的商品——我不是这个社会的专家，但我知道他们的商品多样性远不如今天。他们看那些「内在属于人类」的工作，比如当个歌手；再看那些不内在属于人类的东西，比如马提供的运输服务、他们吃的各种食物。如果他们把两类的品种数都固定住，然后问「等自动化多起来会怎样」，他们可能会说：那我们对「马式运输」、对酸奶、对蒙古包都会吃饱，这些份额都会归零，最后我们所有的钱都花在歌手身上。但当然，事情不是这样——因为随着我们积累了更多财富、更先进的机器等等，我们扩展了「歌手之外」可以花钱的东西的范围，而花在歌手身上的份额一直小到可以忽略。所以同理，这大致也是我对未来如何展开的核心预测，尽管两边都有可能。——我本来想说一个观点，然后我意识到它是个谬误，但它错得很有意思。

[13:55] Dwarkesh Patel

interesting. So I was going to say, I mean, it's just hard to imagine a world where there'strillions upon trillions of robots, but there's only some billion odd humans. And then like thecumulative amount we're spending on robots and like building more robots and whatever is lessthan what we're spending to like pay, you know, Magnus Carlsen and-Or financial advisors or doctors or tutors.Or podcasters or whatever.But then I realized it was a fallacy. The number of transistors in the world has like literallycertainly trillion X, maybe quadrillion X or something. And your colleague, Chad Jones,has a very interesting result about how the share of the economy that is going towards paying forcomputing, basically like paying for the transistors has been decreasing. The point that you made is thatone way to think about Moore's law, you know, what sets price? Well, the price is a supply and demand.And so not only are we producing more transistors more cheaply, but also we're like the value of themarginal transistor is decreasing, right? So more, as you were saying, another way of saying Moore's lawis, you should say- Oh yeah, I like the pessimistic framing of Moore's law is every 18 months,the value of computation halves. Yeah. Right? Like we're just running out of uses for computation

我本来想说：很难想象这样一个世界——有几万亿几万亿个机器人，却只有几十亿人类，然后我们花在机器人、花在造更多机器人上的累计金额，居然少于我们付给马格努斯·卡尔森（Magnus Carlsen）这类人的钱。——或者财务顾问、医生、家教。——或者播客主播之类。——但我随后意识到这是个谬误。世界上晶体管的数量已经涨了大概一万亿倍，也许是一千万亿倍。而你的同事 Chad Jones 有个很有意思的结论：经济中用于支付算力（基本就是支付晶体管）的份额一直在下降。你提到的一点是：理解摩尔定律的一种方式是——价格由供求共同决定，所以我们不仅是更便宜地造出了更多晶体管，同时边际晶体管的价值也在下降。所以换个说法，摩尔定律可以表述为————对，我喜欢摩尔定律的悲观版表述：每 18 个月，计算的价值减半。——对吧？我们把计算的用途消耗得太快了，快到——

[15:21]

so fast that it's sustaining Moore's law. And this is in fact like literally relevant to a conversationabout AI where maybe for the first time, this is no longer true. Right. So the famous fact here isthat an H100 costs more to rent now than it did three years ago, even though we have much superiortechnology and we have much more compute in the world because as models get smarter, the opportunitycost of compute gets higher. But this is Phil's point about increasing variety, right? What we have doneis increase the types of things that people demand from capital. Now, all of a sudden you have a newvariety that you could be using capital for and all of a sudden you jump back up. Yeah. You couldimagine we just never satiate demand for compute. And as long as that stays the case, then the share ofthe economy that is going towards compute would keep increasing. And that's the big question, right?

——快到足以撑住摩尔定律。而这件事恰恰和一场关于 AI 的对话直接相关——因为也许是第一次，上面那条不再成立了。这里有个著名的事实：现在租一块 H100 比三年前更贵，尽管我们有了远为先进的技术、世界上的算力也多得多——因为随着模型变聪明，算力的机会成本变高了。——但这正是 Phil 说的多样性增加，对吧？我们做的事情是：增加了人们向资本索取的东西的种类。突然之间，你有了一种新的用途可以把资本投进去，于是需求一下子又跳上来了。——对。你可以想象我们对算力的需求永远吃不饱。只要这一点成立，经济中流向算力的份额就会不断上升。而这就是那个大问题，对吧？

[16:17] Alex Imas

It's like, that is the ultimate question that we need to be kind of looking at is like,what number of new uses are we finding for that commute where you have the demand for these uses?So what I kind of want to emphasize is that a lot of models in economics, especially in the space thatwe're talking about, take demand is almost kind of exogenous and they don't unpack like what is that,like the psychology of what people actually want. And so what got me kind of also thinking about this,the idea of the relational sector's work that I was doing on the fact that there does seem to be thisvalue, this intrinsic value that is, it's not just because it's scarce, it's because there's someintrinsic preference that people have for like empathy and connection and, you know, getting,getting, interacting with another person. So like one of the experiments that we ran was like,there's an art print, right? And we actually have an incentive compatible way of like basically saying,like, how much are you willing to pay for this art print? People are actually paying real money for it.And then we say like, look, there's only one, one of those art prints and it's either made,and these are between subject conditions by, by AI or by a person. So with one, you get the effect

这就是我们真正需要盯住的终极问题：我们能为算力找到多少种新用途，而且这些用途确实有需求？所以我想强调的是：经济学里的很多模型——尤其在我们讨论的这个领域——把需求当成外生的（exogenous，即模型不解释、直接给定），不去拆解「人们到底想要什么」背后的心理学。而让我开始想这件事、想到关系型部门的，是我做的一项研究：似乎确实存在这样一种内在价值，它不是因为稀缺才有价值，而是人们对共情、对连接、对与另一个人互动本身就有内在偏好。比如我们做过一个实验：有一幅艺术版画，我们用一种「激励相容」（incentive compatible，即让受试者说真话才最划算的实验设计）的方法去测量「你愿意为这幅版画付多少钱」——受试者是真的掏真金白银的。然后我们说：这幅版画只有一份，而它要么是 AI 做的、要么是人做的（这是被试间设计）。在这个条件下你会看到——

[17:29] Alex Imas

that the person produced art print is valued much, much higher than the, than the AI version.And then what we do is to say there's in a set of other conditions, there's 500 of these beingproduced. So for the human made one, the price goes down a lot because it's no longer seen as like,you're not like making a connection with this one artist versus with AI, there's no difference.AI is already viewed as like a commodity. And, you know, we need to do a lot more research on this,but it seems like that's kind of like the, the, the key difference between, you know, somethinglike, let's say a horse, right? There's no, a horse was an input into a, into an output where you canreplace the horse with something else. You only care about the output. The only way this relationalstory works, and this is what we need more, more data on is if it's not a human is not a horse inthe sense that it is providing value from the output, where if you replace the human, the, the valueof the output decreases. And if that's not strong enough, and if it doesn't hold for enough sectors,if it doesn't hold for enough jobs, um, then this kind of story doesn't work anymore.There aren't that many institutions that have thought as hard as Jane Street about how to turn

——人做的那幅版画的估值比 AI 版本高得多得多。接着我们换一组条件：这幅画一共印了 500 份。这时人做的那幅价格会跌很多，因为它不再被看成「你在和这一位艺术家建立连接」；而 AI 那边则毫无差别——AI 本来就已经被视为一种大宗商品（commodity）。当然，这方面我们还需要做大量研究，但看起来这就是关键区别所在——对比一下马：马是投入品，是产出的一个输入，你可以把马换成别的东西，你只在乎产出。这套「关系型」故事要成立（这也是我们需要更多数据的地方），前提是人不是马：人本身就在产出里提供价值，你把人替换掉，产出的价值就下降。如果这个效应不够强、如果它在足够多的行业和岗位上不成立，那这套故事就不成立了。——世界上没有几家机构像 Jane Street 那样，把「如何把聪明人变成世界上最能干的研究员和工程师」想得这么透彻。

[18:35] Dwarkesh Patel

smart people into some of the most competent researchers and engineers in the world. Thisrelies in part on an apprenticeship model where new hires are paired with senior mentors, but JaneStreet also runs a bunch of classroom-side lectures and hands-on bootcamps. These courses cover a rangeof topics and they go pretty deep. There's one lecture that focuses on reverse engineering systems withtools like S-Trace and GDB, and another that teaches you how to profile code down to the cache hierarchy level.Importantly, Jane Street designs these courses not just to teach the relevant object level skills,but also to impart the relevant tasks and knowledge. For example, their week-long neural net bootcampstarts with general theory, but then quickly progresses to how to apply neural networks totrading. And here they cover the specific obstacles that Jane Streeters tend to encounter and theworkarounds they've come up with to get around them. Jane Street takes this sort of learningincredibly seriously. Every office has dedicated classroom space and courses are prioritized as part ofregular work. If you'd like to work at a place like this, Jane Street is hiring. You can check outtheir open roles at jainestreet.com slash dorkesh. There's one possibility which Molly Kinder has

这一部分靠的是师徒制：新人会配一位资深导师。但 Jane Street 也开了一大批课堂式讲座和动手训练营。这些课程覆盖面很广，而且讲得相当深。有一门课专讲如何用 strace、gdb 这类工具做系统逆向工程；另一门教你如何把代码性能剖析（profiling）做到缓存层级。重要的是，Jane Street 设计这些课程，不只是教对应的具体技能，还要传递相关的品味与判断力。比如他们为期一周的神经网络训练营，从一般理论讲起，很快就推进到「如何把神经网络用到交易上」，并且会讲 Jane Street 的人常撞到的具体障碍，以及他们摸索出的绕过办法。Jane Street 对这类学习极其认真：每个办公室都有专门的教室，课程被当作日常工作的一部分优先安排。如果你想在这样的地方工作，Jane Street 正在招人，可以去 janestreet.com/dwarkesh 看看他们的开放职位。——有一种可能性，Molly Kinder 写过东西讲这个「混乱的中间地带」（messy middle）情景。

[19:40] Dwarkesh Patel

written something about this messy middle scenario. And there's that possibility made me think aboutwhether it might be better to have, at least as far as wealth distribution and redistribution goes,it might be better to have much faster AI takeoff. And I want to ask you whether the followingpossibility is at all likely, or there's any set of assumptions that this can make it so,which is that AI makes it possible to automate jobs such that like many people are losing their jobs,but it doesn't create enough wealth while the process of automation is happening to pay off,basically, the people who are getting laid off. There's like a Pareto improvement.Um, everybody's getting better as a result of AI automation. And of course, there's a trivial sensein which that must be true because whatever money you're saving, whatever money the company is savingby not paying the humans instead of just paying the AIs, those resources still exist in the economy andthey can just be paid off to people. But there's going to be some allocative inefficiency if likethe government doesn't know exactly who got laid off because of AI. There's some political problemof like if the meta worker gets laid off first and they're making 200k a year. Um,

这种可能性让我开始想：至少就财富分配与再分配而言，也许 AI 起飞得更快反而更好。我想问你们，下面这种可能性到底有多大可能、或者有没有某组假设能让它成立——就是：AI 让很多岗位可以被自动化，于是很多人失业；但在自动化推进的过程中，它创造的财富不足以补偿那些被裁掉的人，做不到一次帕累托改进（Pareto improvement，即没人变差、至少有人变好），让每个人都因 AI 自动化而变好。当然，在某种平凡的意义上这必然是成立的，因为公司「不付给人类而付给 AI」省下来的钱依然存在于经济中，完全可以拿去补偿这些人。但会有配置上的低效：政府并不确切知道谁是因为 AI 被裁的。还有政治问题：如果先被裁的是 Meta 的员工、年薪 20 万美元——

[20:46] Dwarkesh Patel

is there a politically sustainable situation where you give them a 200k check a year,uh, when there's many people who are working who are making much less? Um, so do you at all findthis scenario plausible where AI is actually automating a bunch of things, but there isn'tenough wealth creation as there is automation? Uh, I think it's, is that plausible?

——在政治上有没有可能持续地每年给他开一张 20 万美元的支票，而与此同时还有很多在职的人挣得比这少得多？所以你们觉得这种情景可信吗——AI 确实在自动化一大堆事，但财富创造的量跟不上自动化的量？这可信吗？

[21:07]

Possible. To me, it does seem like a pretty narrow window. My guess is that if we have the technologyto automate so many jobs that it becomes like a new kind of political problem, then thepipe will also be growing really fast. Well, unless in all of those professions that it'sautomating, it's just a hair more productive. So like the cost of all the capital to, uh,to, to replace all the software engineers is just, you know, a hair less than the cost ofwhat we've been paying the software engineers.Right. And why, why is it implausible that it just like a company can save money by laying off a bunchof software engineers, but, and in the long run, there's a Jevin's paradox thing. And you know,we can't anticipate in advance what we do with more software and surely there's gonna be moreuses, but in the short run, the fact is just that a lot of people are laid off and they still need tofigure out how they can use a million X more JavaScript tokens.I think the thing that like is in either like, you know, Phil, Phil and I have been like writing aboutthese things and we have mathematical models in the back of these things. We don't have any politicaleconomy in any of our models. Andy Hall wrote a really nice blog post about the politics of AGI.

有可能。但在我看来这是个相当窄的窗口。我猜，如果我们有了能自动化如此之多岗位、以至于它变成一种新的政治问题的技术，那蛋糕本身也会长得非常快。——除非在所有那些被自动化的职业里，它只是稍微更划算一点。比如说，取代所有软件工程师所需的资本成本，只比我们一直付给软件工程师的钱少那么一丁点。——对。那为什么「公司靠裁掉一批软件工程师省钱」这件事就不可信呢？长期看会有杰文斯悖论那种效应，我们事先也没法预料更多软件会拿来干什么、肯定会有更多用途；但短期的事实就是很多人被裁了，而他们还得琢磨怎么用掉多出来的一百万倍 JavaScript token。——我觉得，Phil 和我一直在写这些东西，这些东西背后我们都有数学模型，但我们的模型里完全没有政治经济学。Andy Hall 写过一篇很好的博客，讲 AGI 的政治。

[22:08] Alex Imas

And he made a really interesting observation. If there's a 2% increase in unemployment,the political winds completely change. Like unemployment, it has a huge effect on whathappens politically. So, you know, to, to Molly's, uh, excellent essay, by the way, um,I think in some ways, like one of the worst scenarios is a drip scenario.Because of the political economy piece, right? Because like, you know, people essentially whatyou, what you might see is like people not really being unemployed in mass, but kind of like movinginto sectors that pay them less money, kind of basically getting, uh, what happened with phoneoperators in, in, in the mid century of the, of the, uh, between 1920 and 1940 phone operatorswere completely automated. Right. But it took 20 years, even though it's a technology existed.And therefore there was this drip. It wasn't like this giant sector just disappeared.And when it ended up happening, there's a really nice, uh, QG paper on this basically showing thatthey got reabsorbed into the economy, but at lower salaries and they were mostly underemployed.And I think that's the scenario that Molly was, was, was writing about this, like kind of messymiddle where like things aren't a disaster because we saw with COVID, like the fiscal,

他给出一个很有意思的观察：失业率只要上升 2%，政治风向就会彻底改变。失业对政治走向的影响极大。所以回到 Molly 那篇很棒的文章——我觉得从某些角度看，最糟的情景之一恰恰是「滴漏式」（drip）情景，正是因为政治经济学这一层。因为你可能看到的不是大规模失业，而是人们慢慢挪到工资更低的行业去，基本上重演电话接线员的故事：在 20 世纪中期、1920 到 1940 年之间，电话接线员被彻底自动化了。但这个过程花了 20 年，尽管技术早就存在。于是它是一种「滴漏」，不是某个庞大行业一夜消失。而最后发生的事——有篇很好的 QJE（《经济学季刊》，人名/刊名按音记，存疑）论文讲这个——是他们确实被经济重新吸收了，但工资更低，而且大多处于就业不足（underemployed）的状态。我想这就是 Molly 写的那个「混乱的中间地带」：事情没到灾难的程度，因为我们从新冠疫情中看到，财政——

[23:23]

fiscal response can move quickly if there's an emergency and an emergency is a quick uptickin unemployment, which could even look like two or 3%. That's like a national, that becomesa national emergency. If it becomes fast. Um, the concern is that suppose whatever you'resaving on those white collar workers, if that's not growing the economy, but it's just creatingsome, you know, saved resources that can be allocated elsewhere. Is that enough to do, um,a broad based redistribution scheme? Cause then you, you have like the money you've saved off acouple of people. Yeah. And unless you can figure out exactly how to get to them specifically,you got the problem of, can, can I do like, can I do a UBI off the money I saved?

——财政响应在有紧急状况时是可以很快动起来的；而「紧急状况」就是失业率快速上升——哪怕只是 2% 或 3%，只要够快，那就会变成一场全国性紧急事件。——我担心的是：假设你在白领身上省下来的钱并没有把经济做大，只是产生了一些可以重新配置的「省下的资源」。这够不够做一次广泛的再分配？因为你只是从少数几个人身上省下了钱。——对。——而且除非你能精准找到该补偿的那些人，否则你就会遇到那个问题：我能不能靠省下来的这点钱做全民基本收入（UBI）？

[24:04] Alex Imas

Yeah. So you're basically saying like, look, the pie did not grow that much.Yeah. You're just basically set, you're just basically displacing a bunch of people,but that actually didn't grow the, the technological frontier of what the economy can produce.And so then there's a question of like, well, maybe every time, I don't know if this is thecase, maybe every time this has happened in history, the technological frontier has expandeda bunch. And so I think that's the case. I think simply in, in, in, in history,the technological frontier has expanded. Yeah. So it's, it's, it's, it's kind of,and I, I think Philip made the same point. Like it's hard to imagine that sort of scenario,um, where you are getting like intelligence. That's kind of just enough to replace the softwareengineer, but still costs a lot of money. Like it's just a hair less, less, less expensive thanthe software engineer. So you're not getting this abundance effect. Right. And so you're what,where is the redistribution going to happen because the pie didn't grow?

对。所以你基本上是说：蛋糕并没有变大多少。——对。——你只是把一堆人挤走了，但这并没有把「经济能生产什么」的技术前沿往外推。于是问题就变成：那再分配从哪儿来？也许历史上每一次发生这种事，技术前沿都往外扩了一大截——我觉得历史上确实如此。所以，我想 Phil 也说过同样的观点：很难想象那种情景——你得到的智能刚刚好够替代软件工程师，却还很贵，只比软件工程师便宜一丁点，于是你拿不到那个「富足效应」（abundance effect）。那这样一来，蛋糕没变大，再分配还能从哪儿发生呢？

[24:54] Dwarkesh Patel

Yeah. Yeah. Okay. So the, the, the, this is very helpful. So there's a,many different things ought to be true for this scenario to come to pass,each of which seem unlikely. One, it has to be the case that it is possible to automate entirewhite collar jobs, but only in a piecemeal way. That is to say that you can only automatesoftware engineers, but that same program can't also automate an accountant and an analyst.And whatever, where I think, um, at least my model of intelligence is such thatboth of like the breadth of tasks that it requires to do something like software engineering.Um, and what intelligence is, is such that, you know, if you can really just like layoff all the software engineers, you've got enough in the bucket there that you could like,um, automate all kinds of white collar work. So yeah, you're saving,there's huge amounts of potential savings that have happened as a result of these layoffs.And also that AI is going to be cheaper than human labor. Um, and if both of those things are true,this messy middle scenario where we literally don't have the wealth to, uh, go around seemsunlikely. And then the question is like, what is the best way to tax it and redistribute it?

对，这个很有帮助。所以要让这种情景成真，得有好几件事同时为真，而每一件看上去都不太可能。第一，必须是「能整份自动化白领工作，但只能一块一块地来」——也就是说，你只能自动化软件工程师，而同一套程序没法顺带把会计和分析师也自动化掉。而至少按我对智能的理解——做软件工程所需要的任务广度，以及智能到底是什么——如果你真的能把所有软件工程师都裁掉，那你手上的能力已经足以自动化各种各样的白领工作了。所以裁员会带来巨额的潜在节省。第二，AI 还得比人类劳动力更便宜。如果这两件事都成立，那「我们压根没有足够财富来分」的这种混乱中间情景就不太可能出现。那接下来的问题就是：最好的征税与再分配方式是什么？

[25:56] Alex Imas

Yeah, I have some thoughts. I think, I think, I think it's just really important to outline thecosts and benefits. Like, um, it's also important to know that they're, so first there's differentialcomplexity and like implementing these things, uh, to, they differ in the timeline of like beingactually helpful. So like something like universal basic capital, that's not like,that's not going to generate returns for something that happens in six months.So you probably are going to end up with a layer of things. So like, for example,like a negative income tax, like you implement it and it, it, the day it turns into law that is already,you already have this sort of insurance that like, you know, there's a floor for which,you know, everybody, everybody gets a certain amount of money. And then, you know, if you earnmore money, you get taxed more and things like that. Um, and, but you know, there's positives andnegatives to negative income tax with UBI, the, for example, the, I, I, I worry a lot about like thepolitical economy implications. Like, for example, like if people are just kind of dependent on a check,it really matters who's in power. Like right now we're endowed with labor that can turn into,

我有一些想法。我觉得把成本和收益列清楚非常重要。首先，各种方案在落地复杂度上不同，在「多久才真正起作用」的时间线上也不同。比如「通用基本资本」（universal basic capital，即人人分到一份资本所有权），它不是半年内就能产生回报的东西。所以你最后大概率会得到一个分层的方案组合。举例来说，负所得税（negative income tax，收入低于门槛的人反向从政府拿钱）——你把它写进法律的那天，这份保险就立刻生效了：有一个下限，人人都能拿到一笔钱；而你挣得越多，交的税越多。当然，负所得税和 UBI 各有利弊。比如 UBI，我非常担心它的政治经济学后果：如果人们变成靠一张支票过活，那「谁在台上」就变得极其关键。现在我们每个人的禀赋是自己的劳动力，它可以转化成——

[27:06] Alex Imas

uh, that could turn into income when that is no longer the case. And we are now at the mercyof the, uh, of, uh, of the elected official for like basic needs. Right. So that to me feels likea power sharing arrangement. That's really dangerous.But wouldn't that be true of any sort of government redistribution program?

——可以转化成收入。而当这一点不再成立、我们的基本生活需要仰仗某位民选官员的时候——在我看来那是一种非常危险的权力分配安排。——但任何形式的政府再分配项目不都会这样吗？

[27:24] Alex Imas

So something like university basic capital, where you have like an ownership shareand you have property rights for capital, then you just, you're just, you just, you're a normalshareholder. You're just a normal person. And, but this goes back to the question of indexing,because if indexing is hard, then universal basic capital is hard. And that's the, that,that's the problem of university basic capital is targeting. Right. Right. What do you target toput into people's portfolios? Like what if Anthropoc goes to zero, but some random roboticscompany takes all the surplus? Exactly. Exactly. So that's the risk of university basic capitalwith a negative income tax. You have the same sort of issues that with UBI where we're like,you know, somebody comes into power and says like, this is, we're not going to do that anymoreand people can't work. And then, you know, you, you have the issue of the floor being open.One concern with the wealth tax is that, you know, you, there's no political,politically sustainable equilibrium at like 0.5% wealth tax. And, you know, I mean, this happenedwith the income tax, of course, right. It starts low, it's like for war or something. And then itslowly and slowly escalates until the marginal tax rate in the U S is probably on the order of

通用基本资本就不一样：你拥有的是所有权份额，你对资本有产权，那你就只是一个普通股东、一个普通人。但这又绕回到「指数化（indexing）」的问题——如果指数化很难，通用基本资本就很难。通用基本资本的难点就是瞄准：你该往人们的投资组合里放什么？万一 Anthropic 归零，而某家名不见经传的机器人公司拿走了全部剩余呢？——正是如此。所以这是通用基本资本的风险。而负所得税则有和 UBI 一样的问题：某个人上台说「我们不搞这个了」，而人们又没法工作，于是那个「下限」就被掀了。至于财富税，一个担忧是：0.5% 的财富税在政治上不存在可持续的均衡。所得税就是这么走过来的：一开始很低，为了打仗之类；然后一点一点往上爬，直到美国的边际税率大约达到——

[28:20]

income tax rate is like 40% or something. Um, and in certain States upwards of 50%.Um, with a capital tax, is there a reason to worry? Would, would that distort investment?Because people would just be like, why would I invest in Anthropoc or Intel? The government'sgoing to take larger and larger shares of it and dilute my share. Well, hold on. So, um,I think it's worth separating like how the revenue is raised, like what's taxed and then how it'sdistributed. It could be that the government hands out shares of Anthropoc to everyone bybroad based tax and then buying Anthropoc. Um, which would probably be the right thing to do.I mean, hopefully some like populist proposal doesn't interfere with that and like expropriatesome like particular company that everyone happens to know about. Um, but how, so you're,you're suggesting there could be a tax that is some sort of optimal tax, but it's, we're taxingexternalities or we're taxing land or we're, I guess we probably need to tax something otherthan just those two things, but that tax. Okay. So a consumption tax, like a European value addedtax type thing that allows the government to go buy a bunch of stocks and then they justdistribute those stocks to everybody. That's David Otters. Yeah. Yeah. I mean, that's not

——所得税率大约 40%，某些州甚至超过 50%。——那资本税呢，有没有值得担心的地方？它会不会扭曲投资？因为人们会想：我干嘛还投 Anthropic 或者英特尔？政府会拿走越来越大的一块，把我的份额稀释掉。——等一下。我觉得值得把两件事分开：钱怎么收上来（税什么）和钱怎么分下去。完全可以是政府通过一种宽税基的税收上钱，然后拿去买 Anthropic 的股票，再把股份发给每个人——这大概才是该做的事。当然希望别冒出什么民粹提案来搅局，比如去征收某家大家碰巧都听说过的特定公司。——那你的意思是，可能存在某种最优税制，但我们是在对外部性征税、或者对土地征税——我猜光靠这两样大概不够。——好，那就是消费税，比如欧洲那种增值税（value added tax, VAT），让政府拿这笔钱去买一大堆股票，然后把股票分给所有人。这是 David Autor（大卫·奥托尔）的方案。——对。我是说，这和——

[29:37]

going to be that different from just like redistributing the stocks, but it'll be a littledifferent. Yeah. That's what social security, that was the proposal for social security, by theway that was privatizing social security. Right. So it's like you had, you turn like this sort ofweird, like not weird, but it's been working. It's worked so far, but you know, there's questionsfor how long it's going to keep working. Like basically privatizing social security was givingeverybody a basket of stocks. Right. All right. I'm curious to understand people talk about whetherthere's a white color apocalypse already. Is there any evidence that suggests that there is massautomation or unemployment as a result of AI already?

——直接把股票分掉其实差别不大，但还是会有一点不同。——对。这其实就是当年社会保障（Social Security）私有化的提案。你把这套挺奇怪的——也不算奇怪，它一直运转得挺好，但确实有人问它还能撑多久——总之社保私有化基本上就是给每个人发一篮子股票。——好。我想搞清楚：大家都在谈论「白领末日」是不是已经发生了。有没有证据表明，AI 已经造成了大规模自动化或失业？

[30:17] Alex Imas

I think there's a lot of people are looking at it. So this is an area where there's like a lot of eyesand a lot of data being produced. So the budget lab over at Yale is doing really good analysis on this.They just recently released a report. And I think like you really have to squint to see anythinghappening. Like basically, if you want to take kind of like an approach across the entire economy andlooking at, even looking at like software engineering, like the most exposed sectors,there's just like not really anything going on. There might be a little bit of a signal about likejunior developers getting jobs less than before, but that's like a less than before rather than alevel shift is then there's actually an increased demand for senior manager, for senior softwareengineers, if anything. And so if you look at trend, it's kind of like for junior managers,it's a bit below trend.So as in you're saying the growth is slower than before, but there is still growth even on entrylevel software engineers.Yeah, exactly.And what do you think is going on with the anecdotal evidence of graduating college students sayingthat they're finding it harder to find CS jobs or something?

我觉得关注的人很多。这是个眼睛很多、数据也在不断产出的领域。耶鲁的 Budget Lab（预算实验室）在这上面做了很好的分析，他们最近刚发了一份报告。我的看法是：你得眯着眼睛使劲看才能看出点什么。基本上，如果你用一个覆盖整个经济的口径去看，哪怕只看软件工程这种最暴露的行业，也真没什么在发生。可能有一点点信号：初级开发者找到工作的比例比以前低了——但那是「比以前低」，不是水平的整体下移；而与此同时，对资深软件工程师的需求如果说有变化，反而是增加了。所以看趋势线，初级那块是略低于趋势。——你的意思是，增长比以前慢了，但即便是入门级软件工程师，仍然是在增长。——对，正是。——那那些坊间说法怎么解释：应届计算机专业毕业生说现在更难找工作了？

[31:22] Alex Imas

I think that's anecdotal evidence.You think it's always been hard to get jobs for some people and now it's getting turned into an AInarrative. Same with the layoffs where it's probably just normal layoff and they turnedinto an AI layoff.Yeah. I mean, you have to be careful with all of this. I think like there are these like,you know, there are these like coordinate public coordination devices for like, let's say weget into a narrative where like if you're a firm and you're not laying people off,then you're seen as like not adopting AI enough. So like then you're going to just get a cascadeeffect.Right.A firm's like just needing to keep up with the Joneses in terms of like starting to lay people off.And that's kind of like that. That's super worrying where like actually the firm might beactually worse off after the layoffs than before the layoffs. But it's just doing the layoffs tohave the perception that look, look, we're not behind the times where we're, you know, using AIlike you have the you probably heard these anecdotal stories of like these token counters that likeyou have to maximize tokens and things like that. So again, like right now we have we don't reallyhave any evidence of a white collar bloodbath.

我觉得那就是坊间轶事。——你是说找工作对有些人一直都难，只是现在被套进了一个 AI 叙事里？裁员也一样——大概只是正常裁员，被包装成了 AI 裁员。——对。我是说这些都得非常小心。我觉得存在这种公开的「协调装置」：比如我们进入了一种叙事——如果你是一家公司却不裁人，那你就被看成「AI 用得不够」。于是就会有级联效应。——对。——公司只是为了跟上别人而开始裁员。这种情况特别令人担忧：公司裁完之后可能实际上比裁之前更糟，但它裁员只是为了造成一种观感——看，我们没落伍，我们在用 AI。你大概也听过那些「token 计数器」的轶事：要求员工把 token 用量最大化之类。所以我再说一次：目前我们并没有任何证据表明存在白领大屠杀。

[32:23] Alex Imas

And is that surprising at all? I feel given the fact all these things that I can do isjust like this is a story as old as time. If you automate some complimentary task, theoverall bucket of things that the human labor which complements the automation will increasein value.So this is this is one of the statistics that's really important for that argument is elasticityof demand.Yeah.So like the you take the O-ring model of jobs. So like, again, jobs is a series of tasks. Let's saythe AI automates like nine out of 10, nine out of 10 tasks. One task is not automated. If that personcan now kind of focus in on that task and that the job will become more productive. If that translatesinto a price effect where the product is actually cheaper, if their demand responds enough where nowthere's it's being bought more, it's being used more, the service is being used more, that couldactually lead to more hiring.Right.And a lot of people on the Internet have been like kind of making that argument kind of verygenerally saying like, look, we're seeing if anything in the data, we're seeing an uptickin software engineering.Right. Yeah.Yeah.Which suggests that at least for now, given the way the jobs work, it might be.

这有什么好意外的吗？考虑到 AI 能做的这些事——这是个老掉牙的故事：如果你自动化了某个互补性任务，那么与这项自动化互补的人类劳动，其整体价值会上升。——对于这个论证，有个非常关键的统计量，就是需求弹性。——对。——你看 O 型环（O-ring）工作模型：一份工作是一串任务。假设 AI 把 10 项任务里的 9 项自动化了，剩下一项没有。如果这个人现在能专注在那一项任务上，那这份工作会更有生产力；如果这转化成价格效应——产品确实变便宜了——而需求的反应足够大，以至于它被买得更多、这项服务被用得更多，那反而可能带来更多雇佣。——对。——而网上很多人一直在很笼统地做这个论证，说：看，数据里如果说有什么变化，软件工程反而是在往上走的。——对。——是啊。——这至少说明，按现在工作的组织方式，可能是这样。

[33:31]

But I think this elasticity of demand argument is incredibly important, both for a lot ofarguments that people make or just a lot of labels that people use without understandingwhat the underlying causation is. So people often talk about Jevons paradox.Yeah.This is this idea that as something gets cheaper, you will want so much more of it that the totalamount you spend on the thing increases. And so famously, this happened to coal in Britain200 odd years ago. But really, this only happens if there's the demand for something is highlyelastic. There's many things for which there is not super elastic demand. If oil, for example,gets super cheap, it's not like magically.Right.Yeah, exactly. Magically, there's going to be so many more cars that now we're going to be usingway more oil than before.At least not in the short run.Exactly. So long run elasticity is higher than short run.But even in the long run, so agriculture famously is an example where we can produce way more food ifwe dedicated the same portion of the economy that we dedicated to agriculture. We're alreadyproducing more food regardless, but we could produce even more food if the same portion ofthe economy that was producing food 100 years ago was currently producing food. But you eat

但我认为需求弹性这个论证极其重要——无论是对很多人做的论证，还是对很多人在不理解底层因果的情况下乱贴的标签。人们常说杰文斯悖论（Jevons paradox）。——对。——就是那个观念：某样东西变便宜之后，你会想要多得多，以至于你在这东西上的总支出反而增加。著名的例子是两百多年前英国的煤。但这只有在需求高度弹性时才会发生。有很多东西的需求并没有那么弹性——比如石油如果变得超级便宜，也不会魔法般地————对，正是。——魔法般地冒出多得多的汽车，让我们用的油比以前多得多。——至少短期不会。——正是。长期弹性高于短期弹性。——但即便在长期也是如此。农业是个著名例子：如果我们把和一百年前同等比例的经济资源投入农业，我们能生产出多得多的粮食——我们本来就已经在多产了，但如果投入比例不变，还能产得更多。可问题是你吃——

[34:43] Dwarkesh Patel

enough and then you're done. And so the claim with software is that it is not some inherentproperty of markets that as it gets cheaper, you will just keep wanting more of it.Absolutely not.It is. The thing about software is this is the particular kind of good, whereas it gets cheaper,we'll want more and more. But it is also highly relevant. And you wrote an essay about this.A lot of this podcast is me summarizing your essays back to you. That there's this very viralscenario planning about the future by Citrini, where they're predicting as a result of automation,as a result of very powerful AI, there will be a recession because white collar workers will getautomated. Their salaries, which were, you know, paying for a bunch of things, will no longer beavailable. And so there'll be a slump. Do you want to recapitulate why this might be implausible?

——吃饱了也就够了。所以关于软件的主张是：「东西变便宜你就会一直想要更多」并不是市场的固有属性。——绝对不是。——软件恰好是那种「越便宜我们就越想要更多」的特殊商品。而这一点也高度相关——你写过一篇文章讲这个（这期播客有一大半是我在把你写的文章复述给你听）：Citrini 有一份流传很广的未来情景推演，他们预测自动化、极强的 AI 会带来一场衰退，因为白领会被自动化掉，他们原本支撑着一堆消费的工资没了，于是经济会陷入低迷。你要不要复述一下为什么这可能站不住脚？

[35:31] Alex Imas

Well, I mean, so part of it is plausible, part of it's not plausible. So like the part that'skind of like within the, this is something that we started the conversation with,is the idea that there could be unemployment, a lot of unemployment. If the speed of automation isquick and things like that, people could get laid off and they may not find work very quickly.So that part of the Citrini essay about the unemployment, you know, we can quibble about that,but that's not the issue. The issue is that they talked about negative economic growth.Right.And so what I did in the, in the piece that actually Phil and I had a back and forth onwas to say like, let's start with the, with the proposition that there's negative economic growth.What conditions do you need on the economy to get negative economic growth? And it turns out theconditions are pretty improbable. So one thing that you need is like for the, the holders of capital,like rich people, basically, like basically what you have in this, in those sorts of scenarios,like you have a reallocation of wealth and income from like lower income people who are working,who are using their label towards capital owners. So what you need is that basically demand to be

这里一部分站得住、一部分站不住。站得住的部分正是我们这场对话开头讲的：可能出现失业，而且是大量失业——如果自动化速度很快，人们会被裁掉，并且可能一时半会儿找不到工作。所以 Citrini 那篇文章里关于失业的部分，我们可以争论细节，但那不是问题所在。问题在于他们说的是负增长。——对。——所以我在那篇文章里做的（Phil 和我为此还来回讨论过）是：我们先接受「经济负增长」这个命题，然后问：**经济要满足什么条件才会出现负增长？**结果发现这些条件相当不可能成立。比如你需要资本持有者——基本就是有钱人——这类情景本质上是财富和收入从靠劳动吃饭的低收入者那里，重新分配给了资本所有者。所以你需要的是，他们的需求是——

[36:37] Alex Imas

bounded, like a hard bound, not even like a soft sort of like diminishing sensitivity. You need forthem to eventually say, I've had enough. I don't want to spend any more money. And for that money tonot enter as investment. Right. Right. Which is like, and then you can get negative growth,which is like, and the crucial thing is even if we don't want more shit,the world in which there's a singularity and we don't want to invest more money is crazy. Right.Where we're not like, let's build more data centers. Let's know more fabs. Even though we have AGI,we're not like investing in more data centers to run the AGI. Yeah. And that's like drivingmore economic growth. Yeah. And so I sent the essay to Phil and Phil actually wrote back being like,this is pretty dumb. Yeah. Like my essay saying like, you're, you're trying to say that there's going to benegative economic growth, but these are very implausible conditions. And I was like, actually,that's the point of the essay. These are very implausible economic conditions. So that's whereI think like scenario planning really shines is you have the Centrini essay, which I think is like,I think it was great that it's written because I kind of started a conversation, but it's just like,

——有硬上界的：不是「边际敏感度递减」那种软的，而是硬的——他们最终会说「我够了，我不想再花钱了」；而且这笔钱还不能以投资的形式回到经济里。这样你才能得到负增长。而关键在于：即便我们不想要更多东西了，一个已经出现了奇点、我们却不想再多投一分钱的世界，那也太荒唐了吧？我们不会不去建更多数据中心、不去造更多晶圆厂。就算有了 AGI，我们居然不投更多数据中心去跑 AGI？——对。而这本身就在推动更多经济增长。——对。所以我把那篇文章发给 Phil，Phil 回我说「这挺蠢的」。——对，我是说他那篇文章：你想论证会有负增长，可这些条件非常不可能成立。——我说：这正是这篇文章的用意啊，这些经济条件本来就非常不可能成立。所以我觉得这正是情景推演真正闪光的地方：Citrini 那篇文章我觉得写出来是好事，因为它开启了一场讨论；但它就是——

[37:39]

it's so intuitive, this idea that like, look, if there's demand collapse, we can get the economy toshrink, but it's actually, you could get that with a depression, right? Where in the depression,the technological frontier didn't expand. Right. Here, the technological frontier is expanding.You actually have abundance and for abundance to generate negative economic growth. That's reallyhard to get. Right. Exactly. Google recently announced Gemini Omni and its video editing capabilities areincredible. You can upload a video and then tell Omni to do things like change the background oradjust the lighting or add or remove elements all while keeping everything else consistent.But Omni isn't just a video editor. I got a chance to sit down with the research and product teambehind Omni. And I learned that it's a preview of how future frontier models will be trained.It can take in any kind of input, whether that's text or audio or video. And while it doesn'tcurrently do so architecturally, it's capable of just as seamlessly outputting images or text.So it's really a bet on the multimodal data transfer hypothesis. The model becomes better atpredicting one data type by seeing the others. For example, Omni is really good at accurately

——太符合直觉了：「如果需求崩塌，经济就会收缩」。可问题是，那种情况你在大萧条里能得到——大萧条里技术前沿没有外扩。而这里技术前沿是在外扩的，你实际上拥有的是富足——要让富足产生负增长，那真的很难。——对，正是。——谷歌最近发布了 Gemini Omni，它的视频编辑能力非常惊人。你可以上传一段视频，然后让 Omni 换背景、调光线、增删画面元素，同时保持其他一切一致。但 Omni 不只是个视频编辑器。我有机会和 Omni 背后的研究与产品团队坐下来聊，我了解到它其实预示了未来前沿模型的训练方式：它可以接收任何形式的输入，无论文本、音频还是视频；虽然它目前在架构上还不这么做，但它同样有能力无缝输出图像或文本。所以它本质上是在押注「多模态数据迁移假说」——模型通过看别的数据类型，会变得更擅长预测某一种数据类型。举个例子，Omni 在视频上准确渲染文字这件事做得非常好，尽管谷歌并没有在这个模型里专门针对这项能力——

[38:47] Dwarkesh Patel

rendering text on video, even though Google didn't specifically target that capability in this model.And Omni is the next step towards more accurate world models. Because in order to predict the nextframe of a video, you have to have a deep understanding of physics and spatial dynamics.As Omni progresses, it'll be interesting to see whether it can close a sim to real gap.Because it's much harder to collect data in the real world than it is in simulation,robotics progress has lagged other applications of AI. But if you have really good video models,they can simulate reality. Maybe that stops being the case. In the meantime, if you want to try Omni,you can check it out in the Gemini app at Gemini.Google, or use it in Google's AI creative studio,Flow, at Flow.Google.We're talking a second ago about why there isn't more automation as a result of LLMs.And one plausible mechanism could be that, as you're saying with the O-ring,so O-ring theory refers to this fact that the Challenger shuttle blew up because there's onecomponent that malfunctioned and it destroyed the whole thing. And maybe that's a more general modelof how goods are produced in the economy, that you've got to make sure everything is reliable and

——去训练。而 Omni 是通往更准确世界模型的下一步——因为要预测视频的下一帧，你必须对物理和空间动力学有深刻理解。随着 Omni 演进，值得关注的是它能否弥合「仿真到现实」（sim-to-real）的差距。由于在真实世界收集数据比在仿真中难得多，机器人技术的进展一直落后于 AI 的其他应用；但如果你有非常好的视频模型，它们就能模拟现实，也许这就不再是问题了。与此同时，如果你想试试 Omni，可以在 Gemini App（Gemini.Google）里体验，或者在谷歌的 AI 创作工作室 Flow（Flow.Google）里使用。——我们刚才在聊为什么大语言模型没有带来更多自动化。一个可能的机制就是你说的 O 型环——O 型环理论指的是：挑战者号航天飞机爆炸是因为一个部件失效，结果毁掉了整体。也许这是一个更普适的经济生产模型：你必须确保每一环都可靠、都能正常工作——

[39:48] Phil Trammell

works well. And you can't automate an entire job to an AI right now, even though it might be able toform it at some probability. You need extreme reliability in order for it to not destroy thefinished good. I think this is, yeah, so this might explain why there's less automation now thanthere otherwise could be. But I think it works in the other direction once AIs get advanced enoughthat integrating humans into the production flow of future goods, even beyond the arguments abouthow humans will be more expensive or dumber or whatever, even beyond that, just there will bewhole production flows that are organized for AI labor where they're talking in neural ease.They're thinking many thousands of times faster. So even if there's some comparative advantage whereit makes sense to hire a human, there will be like transaction costs and worries of a reliability thatwill actually make it hard to integrate humans into future production flows. Yeah, that seems right to me.In particular, I just want to distinguish between the point that if you automate like nine-tenths of a job,then people might kind of shift over to the last tenth, but like there might be 10 times more workdemanded of them from the model of O-ring automation from like Gans and Goldfarb recently,

——而现在你还没法把一整份工作交给 AI，哪怕它有一定概率能完成，因为你需要极高的可靠性才能不毁掉成品。——对，所以这可能解释了为什么现在的自动化比它本来可能达到的程度要少。但我认为，一旦 AI 足够先进，这个逻辑会反过来起作用：把人整合进未来商品的生产流程里会变得困难——先不说「人更贵、更笨」这些论点，光是会有整条为 AI 劳动力组织起来的生产流水线，它们用「神经语」（neural-ese，AI 之间难以被人读懂的内部表征交流）交流、思考速度快上几千倍。所以即使在某些地方雇个人类还有比较优势（comparative advantage），交易成本和对可靠性的担忧也会让「把人塞进未来的生产流程」变得很难。——对，这在我看来是对的。——我特别想区分两件事：一是「如果你把一份工作的十分之九自动化了，人们可能会挪到剩下那十分之一上，而且对他们的需求可能会变成十倍」；二是 Gans 和 Goldfarb（乔舒亚·甘斯与阿维·戈德法布）最近那个 O 型环自动化模型——

[41:03]

which was that if you can only automate nine-tenths of the job, but you can do it to a lower standardof quality than the human could do it, you might not want to automate even those nine-tenths.And that's the thing that could totally port over to like symmetrically, it could be a reason why wedon't use a human for one-tenth of the job anymore because a human just can't perform it to the levelof quality that the AI can perform the other parts of the job or the level of speed or whatever.And they end up pulling down the quality or speed of the finished product.By the way, the model you're talking about seems extremely plausible to me of why more lawyers oraccountants or whatever are not automated. Like there are cases in, or even software engineers,where there's a pretty good probability that the thing worked as you expect, but the thing you'repaying the lawyer for is like, no, really, my company's not going to go under because...You're also paying for a lot of like regulation type stuff, right? So like with lawyers,particularly, you need some entity to back up the product. You need kind of like an ownership ofthe product. You need somebody to be able to fire or hire like licensing issues. There's a lot of

——它说的是：如果你只能自动化一份工作的十分之九，而且做得比人的质量差，那你可能连那十分之九都不想自动化。这件事完全可以对称地移植过来：它也可以成为「我们不再让人来做那十分之一」的理由——因为人达不到 AI 做其余部分时的质量水平或速度水平，人反而会把成品的质量或速度拉下来。——顺便说一句，你说的这个模型在我看来极其可信，它解释了为什么更多律师、会计（甚至软件工程师）没被自动化。有些情况下，事情有相当大的概率会如你所愿地跑通——但你付钱给律师，买的是「不，真的，我公司不会因此完蛋」这份保证。——你还在为一大堆监管层面的东西付钱。尤其是律师，你需要一个实体来为这个产品背书，你需要产品有归属，你需要有人可以被解雇或雇佣，还有执照问题。有很多——

[42:11]

like sort of like regulatory layers that are like also going to be keeping, even if there's norelational element, human in the loop that have nothing to do with like the ability of the humanto actually perform the service. Yeah. Yeah. I mean, you know, all of these frictions onthe political type decisions that we are accustomed to only trusting humans, you know,only having humans for like legislation and being a judge, being a jury or all the licensing thatkeeps certain professions human. That all strikes me as transitional, right? I mean,what we expect to come from a human and like how we organize our politics, that's changed so manytimes throughout history, right? From little hunter gatherer bands to empires to whatnot. And yeah,once an AI run political system is much more efficient than the alternatives, then those willprobably tend to outcompete the others. And, you know, so speaking of which, we've been talking aboutwhat preferences humans currently have and what impact that has on what kinds of goods will bescarce in the future. But of course, we'll have different kinds of entities in the future.AIs, right? There was a time when there were no humans on earth, but evolution selected for agents

——监管层面的东西，即便完全没有「关系型」因素、没有非人不可的理由，它们也会把人留在环里，这跟人是否真有能力提供这项服务无关。——对。我是说，这些摩擦——我们习惯于只在某些政治性决定上信任人类：立法、当法官、当陪审员，以及那些把某些职业锁死为人类的执照制度——在我看来这些全是过渡性的。我们期待由人来做什么、我们如何组织政治，在历史上已经变了很多次了，从狩猎采集小群体到帝国等等。一旦由 AI 运行的政治体系比替代方案高效得多，它大概率会把其他方案竞争掉。——说到这个：我们一直在谈人类当下有什么偏好、以及这对未来什么商品会稀缺有什么影响。但未来当然会有不同种类的主体——AI。曾经地球上没有人类，但演化筛选出了——

[43:26] Dwarkesh Patel

that have specific drives and preferences, because those tend to survive the most. And those preferencesnow basically determine how a hundred trillion dollar world economy, what it produces. And so why notexpect the same thing of AIs in the future? This is not even a world with catastrophic misalignment,that is to say, they just kill everybody. But there will be evolution of, even if not individual AIs,then firms, which have AIs as part of them. And what will that evolution favor? What will favor probablyfirms or agents that grow, right? There's like a selection argument that things which grow will bemore prevalent. And maybe just based on that, you can make some predictions about what their preferenceswill be. But it is the kind of entity which prefers to have human intrinsic goods going to be the kindof entity that accumulates resources the most? Probably not, right? Probably like saves more,like has unsatisfiable demand for things like whatever the relevant resource happens to be.Compute is an obvious one. And can we use that to make some prediction about what the non-humanpreferences that will be guiding the future? Yeah. So I think if there's like an AI that's like,has its own welfare and it's fully autonomous and it's like making its own decisions that are

——具有特定驱动力和偏好的能动者，因为这些特质最有利于生存。而这些偏好如今基本决定了一个上百万亿美元的世界经济体在生产什么。那为什么不能对未来的 AI 期待同样的事？这甚至不需要是一个「灾难性失配」（catastrophic misalignment，即 AI 直接把所有人杀了）的世界。但演化会发生——即使不是单个 AI 的演化，也会是「内部包含 AI 的公司」的演化。那这种演化会偏好什么？大概会偏好会增长的公司或能动者——这是个选择论证：能增长的东西会更普遍。也许仅凭这一点，你就能对它们的偏好做出一些预测。那么，那种偏好「人类内在商品」的主体，会是最能积累资源的那种主体吗？大概不是吧？大概更可能是储蓄率更高、对某种关键资源有无法满足的需求的那种——算力是个显然的候选。我们能不能据此预测「那些将引导未来的非人类偏好」是什么样？——是啊。我觉得，如果有一个 AI 有自己的福祉、完全自主、自己做出与福祉相关的决策——

[44:41] Alex Imas

welfare relevant. To be honest, I have absolutely no prior that they would like at all prefer otherto like deal with humans. There's like no reason. But let me take like the other side of that argument.Will humans' preferences to be interacting with one another and to trust and empathize and all ofthese sorts of like things with other humans versus a simulated AI? I think it's a really importantquestion whether those will change, right? So I've heard a lot of arguments saying like, look,you know, right now we're just not used to the technology. And at some point, like what you'rethinking of relational or something like that, people are just going to see like an AI therapistas a superior product and they're not going to need the sort of like empathy or whatever that thehuman is providing. I think this is actually a really complicated question. Here's one argumentfor why it's not going to go away and that that has to do with evolution. So let's say there's twotypes of people. One person doesn't really have this preference. They can just interact with otherAI, whatever can simulate it better. The other one has almost like a moral emotion,like from the using Jonathan Haidt's framework, moral emotion against interact, like offloading

——老实说，我完全没有先验理由认为它们会偏好和人类打交道，没有任何理由。但让我也说说另一面：人类那种「想要与彼此互动、信任、共情」的偏好——相对于一个模拟出来的 AI——会不会改变？我认为这是个非常重要的问题。我听过很多论证说：现在只是我们还不习惯这项技术，到某个时候，你们说的关系型的那些东西，人们会觉得 AI 治疗师是更优的产品，不再需要人类提供的那种共情。我认为这其实是个非常复杂的问题。这里有个「它不会消失」的论证，和演化有关。假设有两类人：一类人没有这种偏好，他们和 AI 互动就行——只要 AI 能模拟得更好；另一类人几乎有一种道德情感（用 Jonathan Haidt 的框架来说的道德情感），反对把——

[45:58] Alex Imas

those sorts of social interactions to an AI. Which of those two people are going to reproduce,find a mate, all of these sorts of things? I think the answer is kind of clear, right? It's the secondone that has the preference for other people. It depends on how the reproduction is happening.Fair. But if we're in, you know, the world where like reproduction is still happening the way thatit's happening, I think, and this is a big question. I'm not even like, I'm not making aprediction. Again, I'm just saying like, if we're thinking, you know, you had David Reich on the show,like his point on the last podcast was that, you know, we're buzzing with natural selection.Right.So even if like you get some sort of indifference now, you might get selection to point into likean even stronger preference for other humans.Here's one way to think about it. How is the wealth of the richest people in the worldinstantiated? Of course, they can, as you were having a call earlier and making the point that theirconsumption is more geared towards relational goods. Like Mark Zuckerberg is hiring MMA instructors anddancers for his wife's birthday and so forth. But most of his wealth is just stock and meta.And he as a controlling shareholder could say, hey, meta, just give me all this income or turn all

——那类社交互动外包给 AI。这两类人里，谁更可能繁衍后代、找到伴侣？我觉得答案挺清楚的：是第二类，也就是偏好和人相处的那类。——那要看繁衍是怎么发生的。——有道理。但如果我们处在繁衍方式仍和现在一样的世界里，我认为——这是个大问题，我不是在做预测——我只是说：你请过 David Reich（大卫·赖克）上节目，他上次的观点是我们仍然处在自然选择的活跃之中。——对。——所以即便现在你得到的是无所谓，你也可能被选择推向对「和人相处」更强的偏好。——可以这样想：世界上最富有的人，他们的财富是以什么形式存在的？当然，你之前在电话里提过一点：他们的消费更偏向关系型商品——比如扎克伯格给他太太的生日雇 MMA 教练和舞者之类。但他绝大部分财富就是 Meta 的股票。作为控股股东，他完全可以说：喂，Meta，把这些收入都给我，或者把这些财富变成——

[47:08] Dwarkesh Patel

this wealth into dividend income. And I will just spend that on consumption. But instead, he ratherwould have his wealth compound and meta to build more data centers, basically.So you don't even have to change humans for this to be the case. It is just the case that thehumans which are wealthiest and are growing wealthier because their wealth is compounding.Just have this like almost Nick Landian preference for like accelerating capital.And that does seem to suggest that, yeah, is that an important determinant of what kinds of thingsare produced in the future?

——变成分红收入，我拿去消费。但他反而更愿意让自己的财富在 Meta 里复利增长，基本上就是去建更多数据中心。所以你甚至不需要改变人类就能得到这个结论：事实就是，最富有、而且因为复利在变得更富的那些人，几乎有一种尼克·兰德（Nick Land）式的偏好——偏好加速资本本身。这似乎确实说明——对，这会不会是决定未来生产什么的重要因素？

[47:41] Phil Trammell

Yeah, I could kind of just say like, there's two ways you could get the two kinds of people,one of whom prefers a human therapist and one of whom is fine interacting with the AI.If they both satiate equally quickly in capital, right? But the one who likes the human therapistjust also likes having some human intrinsic services, then the marginal value, like how themarginal value of capital in the future compares to the marginal value of capital today for each of them,if they start out equally rich, should be basically the same.I mean, there could be interactions and whatnot, but basically that should be the same.If what's driving the difference is that one person just doesn't satiate in capital,because they're engaged by the prospect of, you know, exploring the universe and turning their headinto a galaxy brain or whatever, and the other one satiates, then the person who doesn't satiatein capital is going to have, if they're being rational, they're going to have a higher savings rate.Yeah. Okay. So in the long run, they're going to have most of the well and the overall capitalshare will basically be the capital share of that person spending, which is going to be well.It's important that this is, we're not talking about a hypothetical future.

我可以这么说：要得到「一个偏好人类治疗师、一个不介意和 AI 互动」的这两类人，有两条路径。如果他俩在资本上「吃饱」的速度一样快，只不过喜欢人类治疗师的那位同时也喜欢一些人类内在服务，那么两人「未来资本的边际价值相对今天资本的边际价值」之比，基本上应该是一样的——假设他们起点一样富。当然可能有交互作用之类，但基本上应该一样。而如果两人的差别在于：其中一个人在资本上根本不会吃饱——因为他被「探索宇宙、把自己的脑子变成一个星系大脑」这样的前景所吸引——而另一个人会吃饱，那么不会吃饱的那位如果理性行事，储蓄率就会更高。——对。——所以长期看，他会拥有绝大部分财富，而整体资本份额基本上就等于这个人的消费所对应的资本份额。——重要的是，这不是一个假想的未来。

[48:52]

Yeah.Elon Musk is talking about mass drivers on the moon.Right.And he's like by far the wealthiest person in the world. I mean, obviously currently his,his investments are going towards humans as well as machines, but I don't think he caresparticularly that his like future researchers and engineers are humans versus AI.And he manages to reproduce fast as well.So, yes.So anyway, so I just think it's worth drawing that distinction. Yeah. There are currentlysome rich people that don't seem to, um, satiate quickly in capital. And, uh, and so maybe inthe long run, they'll save the most and right. Yeah. Um, that does seem sort of right to me. Um,and I would just also say, even if they do reproduce more slowly, like biologically, that might just notmatter that much in the long run, right? If, if, uh, they can live forever and, you know,the, the living forever is key. Yeah. Right. So I think, I think again, like we, and we're,we're, we're scenario building here. Right. So I think, um, if you could live forever,like a lot of stuff's changes for my story as well. I think, I think it's, um, uh, to your pointabout, you know, rich people just consuming, not consuming a lot and investing. I think this will

——是啊。——埃隆·马斯克正在谈月球上的电磁弹射器（mass driver）。——对。——而他是目前世界上最富有的人。当然，他现在的投资既投向人也投向机器，但我不觉得他特别在乎他未来的研究员和工程师是人还是 AI。——而且他繁衍得也很快。——是啊。——总之我只是觉得值得把这个区分讲清楚。对，现在确实有一些富人似乎不会很快在资本上吃饱，所以长期看，也许他们会存下最多的钱。——对。这在我看来大致成立。而且我还想说，即便他们在生物学意义上繁衍得更慢，长期看这可能也没那么要紧——如果他们能永生的话。——「能永生」是关键。——对。所以我想说，我们这是在做情景推演。如果人能永生，我这个故事里也有一大堆东西会变。回到你说的富人不怎么消费、而是拿去投资——我觉得这——

[50:02] Alex Imas

all depend on the returns to capital. Right. So like right now the returns to data centers are superhigh. Right. But if we get into a situation where people are satiated with capital, um,then the returns to accumulating capital are going to be lower. And so then these rich people are goingto be consuming more. Right. So because they were the incentive to invest is smaller. So basicallyyou kind of think about this in general equilibrium, um, the general equilibrium of this sort of process,like we have gotten tremendously more richer since, you know, 1820, we've gotten many more people areinvesting, but you're still getting a consumption response, which keeps, you know, people employedin labor share high. And that's because not necessarily, I think you're probably making thesame point, but I mean, they could just, it could be that their investment has to be chitrated throughactual laborers, but to go like do things for their investment to work, which like would not in thefuture, only the consumption is human mediated. Right. Cause the investment can just be done by therobots. But if the returns are, if you, so we're in this scenario with high, with like how you cankeep high labor share, right? Let's take that scenario in the scenario with high labor share,

——完全取决于资本的回报率。现在数据中心的回报率超高。但如果我们进入一种人们在资本上已经吃饱的状态，那继续积累资本的回报率就会更低，于是这些富人就会消费得更多——因为投资的激励变小了。所以你基本上要在一般均衡（general equilibrium，即所有市场同时相互影响后达成的整体状态）里想这件事。从 1820 年到现在，我们变得极大地富有了，投资的人也多了很多，但你依然看到消费的响应，而正是它维持了就业和较高的劳动份额。——这不一定是因为——我想你大概在说同一件事——也可能是因为他们的投资必须通过真实的劳动者去「滴定」落地，才能让投资奏效；而未来不会这样了，未来只有消费是由人类中介的，投资完全可以由机器人完成。——但如果回报率是……我们现在在设想那个劳动份额很高的情景，对吧？在劳动份额高的情景里——

[51:13]

for whatever reason, the returns to capital are going to be lower. Yeah, that's right. And I mean,to the earlier thing where we're in the messy middle, we were saying, why this is implausible.I feel like we can do a similar thing here where for returns to capital to be lower,the growth rate has to have be lower, right? I mean, it certainly has to be lower than whatwe're expecting through the period of transformative AI, you know, if there's explosive growth.Yeah. Yes and no. I mean, so the, the, the capital stock could grow quickly,but the price of capital goods relative to consumption goods could be falling faster thanthe capital stock is growing. Oh, interesting. Yeah. It's the, it's the difference between like the,the, the, the potential frontier of technology and like the, what the realized prices of thesethings, because you have relative prices. That's right. So you're saying,I could be putting my money towards, you know, earning 30% interest in investing in data centersor whatever. There'll be something in the future. Growth rate is high that earns high returns.Or I could, as a result of all the technological breakthroughs or some cool product that I reallywant to buy right now. And both of those will be compelling options. Yeah. It doesn't have to be

——不管出于什么原因，资本回报率都会更低。——对，没错。而且回到刚才「混乱的中间地带」为什么不可信的讨论，我觉得这里可以做类似的推理：要让资本回报率更低，增长率就必须更低——至少必须低于我们预期的变革性 AI 时期的增长率，如果那时会有爆发式增长的话。——是也不是。资本存量可以增长得很快，但资本品相对消费品的价格可能跌得比资本存量增长得还快。——哦，有意思。——这就是「技术的潜在前沿」和「这些东西实际实现的价格」之间的差别，因为存在相对价格。——对。所以你的意思是，我可以把钱投到数据中心之类拿 30% 的利息——只要增长率高，未来总会有某种高回报的东西；也可以因为技术突破而买到某个我现在特别想要的酷产品。这两个选项都会很有吸引力。——对。而且它不必是——

[52:20]

a new product. It could be a human intrinsic product. Right. Although if it's a human intrinsic product,we would want to have it much more in the future than we want it now, because the rel,sort of the, the thing it compares against is. So we might want it the same as we want it now in thesense that like the marginal utility in a ballerina performance is exactly the same as now. Right.But the marginal utility in a robot might just be a lot lower than now. Right. So, so in units ofrobots, we want it a lot more than we want it now. Right. Right. So would the interest rate be 30%?

——不必是一个新产品，也可以是一个「人类内在」的产品。——对。不过如果是人类内在的产品，我们在未来会比现在更想要它，因为它的比较基准是————我们可能在某种意义上和现在一样想要它：芭蕾舞演出的边际效用和现在完全一样。——对。——但机器人的边际效用可能比现在低得多。——对。所以以机器人为单位计价，我们对它的需求会比现在强烈得多。——对。——那利率会是 30% 吗？

[52:50] Phil Trammell

It depends what you mean by the real interest rate. Okay. It might be that every robot now canturn into, you know, a hundred robots next year. Right. So in units of robots, the interest rate is10,000%. Right. But if the price of robots is falling really fast. I see. Prices adjust. I mean,that's the whole, I think that's the whole point. Yeah. But though, but here prices are adjustingthis interesting way that too many macro models don't allow for. Right. So what, what,what's happening is what would be called investment specific technical change where,yeah, the price of capital is like falling relative to the price of consumption insteadof like the standard, doing the standard macro thing of saying there's just output.Yeah. It's like chimera of a thing called output, which is one for one can be allocated to capitalor consumption. Right. That's not going to be true in this world. Yeah. Every unit of capital nextyear is giving up way less consumption than each unit of capital this year. Cause like the,just one robot now turns into many robots next year, but, but the number of ballerines is the same.And again, we're going to go back to the increasing varieties thing where like, if all of those extra

这取决于你说的「实际利率」是什么意思。可能现在的每台机器人明年都能变成一百台机器人，那以机器人为单位，利率就是 10000%。但如果机器人的价格跌得非常快————我明白了，价格会调整。这不就是全部要点吗。——对。不过这里价格调整的方式很有意思，是太多宏观模型不允许的方式。发生的事情叫作「投资专有技术变迁」（investment-specific technical change）：资本的价格相对消费的价格在下跌，而不是像标准宏观那样假设只有一个叫「产出」的东西。——对，一个叫「产出」的怪物，可以一比一地分配给资本或消费。——对，这个世界里不会是那样。明年的每一单位资本，所放弃的消费远少于今年的每一单位资本——因为现在的一台机器人明年会变成很多台机器人，但芭蕾舞演员的数量不变。——这里我们又要回到「多样性增加」那件事：如果明年多出来的那些——

[53:55]

robots next year, actually different varieties of robots and I'm not getting satiated on those robots,then, then it's a very different, different story. Yeah. Right. But now we're talking about theconsumption world. Whereas for the investment side of things,there could be just some greedy Titan of industry who keeps wanting more and more robots.And that alone would be enough to increase the marginal value of robots and thereforedecrease labor share. Yes. Yeah. Okay. But why are we not expecting greedy Titans of industry to keepexisting? I mean, greedy Titans of industry historically have like built libraries and,but that's because they die and they're like, they all die. Everybody dies.Well, we'll see. But I mean like conditional on people dying, I think like, you know, his,like, again, I, you, you had a guest on the show who said like, you know, to understand the future,you should think about the past. And I think like, I, I, you could have new types of Titans being,right. Being born who's where their entire reason for accumulating wealth is just to accumulate wealth.Yeah. But a lot of the time, you know, at least historically, I'm just talking about historically,the wealth accumulation process is part of a large socially sort of like social interaction amongst

——机器人其实是不同品类的机器人，而我在这些机器人上不会吃饱，那就完全是另一个故事了。——对。——但我们现在讲的是消费这一侧。而在投资这一侧，可能就是有某个贪婪的工业巨头，一直想要越来越多的机器人。光是这一点就足以抬高机器人的边际价值，从而压低劳动份额。——是的。——好。那为什么我们不该预期贪婪的工业巨头会一直存在？我是说，历史上的工业巨头确实会去建图书馆之类，但那是因为他们会死——他们都会死，所有人都会死。——那可未必。不过在「人会死」这个前提下，我觉得……你请过一位嘉宾说过：要理解未来，就该想想过去。我觉得可能会出现新型的巨头，他们积累财富的全部理由就是积累财富本身。——是的。但至少历史上，很多时候财富积累过程是嵌在一个更大的社会互动里的，是在——

[55:20] Dwarkesh Patel

peers, amongst the community where you want to be admired in some way or something like that.So then the people end up like the, the stylized fact of Titans of his, of industry is like,you accumulate the capital and then you like buy a bunch of stuff.Yeah. I mean, I guess it is sort of a historical question, but it does seem to me in a lot ofcases, what is happening is that as a near the end of their life, they either hand it off to theirchildren who are worse stewards of capital than they are. And they don't even manage to grow theirwealth at the rate the economy grows much less faster than the economy grows, which their parentswere doing. And also they're like, well, I care less about my children having it than me sort ofplaying this game of accumulating wealth. And so I'm just going to give it to some trust. And if peopleare living longer or if they can figure out some way in which to align their trust to this wealthaccumulation process, it just feels like the evolution here is so strong where you just need acouple of agents that think this way for this to be the dominant thing, determining the preferencesof the whole economy, because this part is growing much faster than the other parts of the economy.

——同侪之间、社群之中，你希望以某种方式被人仰慕之类。所以最后人们的路径是——工业巨头的典型化事实是：你先积累资本，然后去买一堆东西。——对。我想这算是个历史问题吧。但在我看来，很多情况下发生的事是：他们临近生命终点时，要么把财富交给子女——而子女作为资本的管家远不如他们，他们连「让财富以经济增速增长」都做不到，更别提跑赢经济了（而他们的父辈是能跑赢的）；要么他们心想「我在乎孩子拥有它，还不如我自己在乎这场财富积累的游戏」，于是干脆把钱塞进某个信托。所以，如果人活得更久，或者他们能想出办法让自己的信托对齐这套财富积累过程，那这里的演化压力就强到——你只需要几个这样想的能动者，就足以主导整个经济的偏好，因为这一部分的增长比经济其他部分快得多。

[56:22] Alex Imas

I think you just like the part about satiation and diminishing marginal utilities, it keeps comingup, but I think it's really, really important. Like, you know, if a person has an intrinsicpreference for accumulation, that's just like, that's what they want. I think your story is totally right.But, but that's just like, not how usually preferences work. Like you have enough, whatever,you hedonics in your life. And then, then like the social status, all of this sort of, you know,Rousseau wrote about this, St. Augustine wrote about this. This is like a, kind of like a basicpart of preferences. Now, too, you guys are arguing about something else where like, you have, you couldhave such high concentration that you could just have a couple of exceptions to the rule and that's going tobe enough. And I, I have nothing to say about that. Yeah. Yeah. I mean, I think the claim's a littlestronger, not just like you could have some exceptions, but that it seems that historicallyand today we see the exceptions and they just haven't really taken over the economy historicallybecause they've, there've been these dissipation shocks as they're called. So they've like given itto their kids and squandered it or they put it in foundations, which, um, uh, which spent it. I mean,

我觉得，就像「吃饱」和边际效用递减那部分一样——它反复出现，但我认为它真的非常非常重要。如果一个人对「积累」本身有内在偏好，那他要的就是这个，你这套故事完全成立。但问题是，偏好通常不是这么运作的：你在生活里的享乐达到某个量之后，接下来就是社会地位那一套——卢梭写过，圣奥古斯丁也写过，这算是偏好的基本组成部分。不过你们俩现在争的是另一件事：由于财富集中度可能极高，只要有少数几个例外就够了。这一点我没什么可说的。——我想这个主张还要更强一点：不只是「可能存在一些例外」，而是历史上和今天我们确实看到了这些例外，而它们并没有真的接管经济——因为出现了所谓的「耗散冲击」（dissipation shocks）：他们把财富交给孩子被挥霍掉，或者放进基金会而基金会把钱花掉了。我是说——

[57:32] Phil Trammell

it's not a really a shock, but I mean, it's right. People went, people might've liked to,uh, you know, fill the universe with monuments to themselves and sort of whatever live forever.Very wealthy. And it's like a weird preference, but it's not a hypothetical preference. I thinkthat's, that's the thing, but who knows what's going on in their heads. I think, um, uh, evenwithout though, like the kind of intrinsic preference for accumulation, there are some instrumentalreasons why people, some people might value accumulation, which is also worth bringing up.So, um, there's, uh, the desire for, uh, political or philosophical or religious influence,right? So people get into sort of an arms race over like what, you know, what society looks likeand what people believe. Um, and then similarly, but differently, cause it's not an arms race.There's just a total, total utilitarian philanthropy, right? So, uh, when I think aboutwhy it might be good to have a lot of wealth in the future as a good classical utilitarian,to me, the values, or at least one way you could have a kind of almost unsatiating, uh, uh, utilityfunction and having wealth in the future is to create new happy beings, right? They just add to

——这也不算真正的「冲击」。但人们本来可能会想用纪念自己的丰碑填满宇宙、永远活下去、极其富有——这是种奇怪的偏好，但它不是假想的偏好，我觉得这才是关键。当然谁知道他们脑子里在想什么。我觉得，即便没有那种对积累的内在偏好，也存在一些工具性理由让某些人重视积累，这也值得一提。比如对政治、哲学或宗教影响力的渴望——人们会在「社会该是什么样、人们该信什么」上陷入军备竞赛。还有一个类似但不同的（因为它不是军备竞赛）：彻底的功利主义慈善。作为一个古典功利主义者，当我想「为什么在未来拥有大量财富可能是好事」时，我的答案，或者说至少一种能让你对「未来拥有财富」几乎不会吃饱的效用函数，是创造新的幸福存在者——他们直接增加——

[58:49]

the total welfare of the world. You know, I mean, this, this idea goes at least as far back as likeBostrom's astronomical waste point that we could like put Dyson spheres around the stars and turnall the energy into really happy simulations and whatnot. I think the particular greediness of thisoptimizer doesn't matter what they're greedy for. I think you're forgetting about utilitarianphilosophy or whatever, like just a pure von Neumann probe has, I don't know what the, is this anaccurate way to say it? They just have high marginal value for like the random solar system they'lloccupy because that turns into like more solar systems. It turns into more solar systems. But like a von Neumann probeis a thing that will, can exist, right? And that's like a very greedy optimizer.Yeah. I mean, if we're talking about like whether they'll dominate the economy, maybe this is atechnicality, but, um, you know, we, we, we only count final consumption goods and investment goodsas GDP, right? If there's just this phenomenon.How does a von Neumann probe show up in GDP?

——增加世界的总福祉。这个想法至少可以追溯到博斯特罗姆（Bostrom）的「天文学级浪费」（astronomical waste）：我们可以在恒星周围建戴森球，把所有能量变成非常幸福的模拟体之类。——我觉得这个优化器具体贪什么并不重要。你大概忘了功利主义哲学之类——一个纯粹的冯·诺依曼探针（von Neumann probe，能自我复制、逐个殖民星系的探测器）就有——我不知道这么说准不准确——它对自己将占据的那个随机恒星系有很高的边际价值，因为那会变成更多的恒星系。——会变成更多恒星系。——而冯·诺依曼探针是一种可以存在的东西，它是个非常贪婪的优化器。——嗯。我是说，如果我们讨论的是它们会不会主导经济，这也许是个技术性细节：我们计入 GDP 的只有最终消费品和投资品。如果只是有这么个现象————冯·诺依曼探针在 GDP 里怎么体现？

[59:43] Phil Trammell

Well, yeah, exactly. Right. So if, if, if it's like, if we recognize it as a person that like owns itselfand it's like sort of, you know, optimizing on the margin between like spending a bit more ona baby von Neumann probe that colonizes another star system or like a ballerina or something.And it's just like, it doesn't value the ballerina very much, but it's, yeah.Yeah. When we're talking about like AI beings or like, like it just, it just completely dependson how we're doing the accounting there.Right. Yeah. But it just like, what does the world look like in a world where like von Neumannprobes are possible? Is it possible labor share is high?

对，正是这个问题。如果我们把它认作一个「拥有自己」的人，它在边际上要在「多花一点造一个能殖民另一个星系的小冯·诺依曼探针」和「看一场芭蕾」之间做权衡，而它并不怎么看重芭蕾——那就是另一回事了。——是啊，当我们谈论 AI 存在者的时候，这完全取决于我们怎么记账。——对。但在一个冯·诺依曼探针成为可能的世界里，世界到底长什么样？劳动份额还有可能高吗？

[1:00:13] Dwarkesh Patel

Anyways.Yeah. I think it's possible that the labor share is high the way we usually account it.One of the biggest problems in RL right now is credit assignment because you have theseextremely long rollouts and you need to know why they succeeded or failed. One of Cursor'sresearchers, Sasha Rush, gave me a Blackboard lecture on how they use targeted RL with textualfeedback to deal with this problem and train composer 2.5. I filmed on my iPhone. So apologiesfor the camera work.So we've generated this output. It's just a sequence of tokens. We're going to send thosesequence of tokens to this model that's going to read it. Yeah. And then it's going to isolatea specific, say, turn that it says is problematic. Yeah. Then we're just going to do text manipulation.We're just going to take that trajectory and we're literally just going to like smash insome extra tokens.After Cursor injects these hint tokens, they run another forward pass. The trajectory itselfdoesn't change, but the hint causes the model to assign lower probability to the error tokens.Cursor then trains the original model to match those probabilities, basically teaching it todownweight these specific mistakes. There's a lot more nuance that we couldn't include in this

总之——是啊，我认为按我们通常的记账方式，劳动份额有可能是高的。——现在强化学习（RL）最大的问题之一是信用分配（credit assignment，即把最终的成败归因到中间某一步）：因为 rollout（模型一次完整的推演过程）极长，你需要知道它为什么成功或失败。Cursor 的研究员 Sasha Rush 给我上了一堂黑板课，讲他们如何用带文字反馈的定向强化学习来解决这个问题、并训练出 Composer 2.5。我是用 iPhone 拍的，所以镜头请见谅。——我们生成了这个输出，它就是一串 token。我们要把这串 token 发给一个模型让它读。——嗯。——然后它会定位出其中某一个它认为有问题的回合。——嗯。——接着我们只做文本操作：把这条轨迹拿过来，硬塞进去一些额外的 token。——Cursor 注入这些提示 token 之后，会再跑一次前向传播。轨迹本身没变，但提示会让模型给那些出错的 token 分配更低的概率。然后 Cursor 训练原始模型去匹配这些概率，本质上是在教它把这些特定错误的权重压下去。这里还有很多细节我们没法放进这段——

[1:01:18] Dwarkesh Patel

mineral. If you want to watch the full thing, I posted it on my Twitter. And if you want to tryout Composer 2.5, head to cursor.com slash Dwarkesh. Do economists have any advice or countries whichare not in the AI production chain? If you're not either producing the AI models, you're not producingthe hardware that goes into AI models. If you're not Korea making HBM or Taiwan makingwith the FAZ or not the Netherlands with ASML. Like what is India or Nigeria? What should theybe doing right now? If you're talking to Modi right now, what do you say?

——放进这段（广告）里。如果你想看完整版，我发在我的 Twitter 上了。如果你想试试 Composer 2.5，去 cursor.com/dwarkesh。——那么，经济学家对那些不在 AI 生产链上的国家有什么建议吗？如果你既不生产 AI 模型，也不生产模型所需的硬件；你不是造 HBM（高带宽内存）的韩国，不是造晶圆厂的台湾，也不是有 ASML 的荷兰——那印度或者尼日利亚该怎么办？如果现在莫迪就在你面前，你会说什么？

[1:01:52] Alex Imas

I think the biggest lack of resources that we have allocated in the economic profession is thinkingabout middle income developing countries in the age of AI. And I mean, this is my fault. This issomething I fault myself with as well. There's not enough people thinking about this question.Like there are scenarios where, you know, you get like AI technology, you know, being allocated anddissipating to Nigeria and developing countries and things like that. And like that leveling theplaying field, like essentially like giving them a like a level up as far as capabilities.But there's another world where like, because they don't have enough resources, they're not makingthey're not training the models, they don't have the hardware, where they just completely get leftbehind. And because of, you know, automation, we can produce commodities in developed countries now,right? Then we don't even have, you know, the consumer market. And then that that world lookspretty, pretty bad. Yeah. This seems to me like an extension of the messy middle case, right?

我认为经济学界资源投入最不足的地方，就是思考「AI 时代的中等收入发展中国家」。这一点我也自责，我自己也是这么做的——想这个问题的人不够多。存在这样的情景：AI 技术被扩散、被配置到尼日利亚和其他发展中国家去，从而拉平竞技场，本质上给它们的能力来一次跃升。但也存在另一个世界：因为它们资源不够，不训练模型、没有硬件，于是被彻底甩在后面；而且由于自动化，我们现在可以在发达国家生产大宗商品了，那我们连它们这个消费市场都不需要了。那个世界看起来相当糟糕。——在我看来，这像是「混乱中间地带」情景的一个延伸，对吧？

[1:02:53] Phil Trammell

One of the ways in which the messy middle might only be bad in a narrow range of scenarios,isn't just that like, it would be easy to redistribute because it probably bigger. Butbecause the interest rate would be way higher, and or sort of equivalently, the price of everythingexcept the human intrinsic goods would be would be falling really rapidly, sort of two sides of thesame point. A little bit of savings would turn into a lot of consumption next year, right? So things haveto go really wrong for us to like, just get over the threshold of, you know, capital being productiveenough to automate lots of work, but not be productive enough that that the interest rate ishigh and or the price of capital produced goods is falling a lot. Okay. So even without redistribution,a little bit of savings will save a lot of people. Sorry, you're saying that the developing countrieshave some savings? Yeah, yeah, yeah. In the developed world, that will be enough to produce a lot ofsurplus that they can, they will now be able to consume a lot, right, using their savings. So, so, but I mean,the messy middle could be like wider in this case, I mean, they're starting from such a lower level interms of like, how much they have it and how how much it's like actually indexed to the global

混乱的中间地带之所以只在很窄的情景区间里才会糟糕，理由不只是「蛋糕更大所以更容易再分配」，还因为利率会高得多——或者等价地说，除了人类内在商品以外，一切东西的价格都会飞快下跌，这是同一件事的两面。那么一点点储蓄，明年就能变成大量消费。所以要卡进那个门槛真得非常不巧：资本刚好生产力足够高、能自动化大量工作，却又没高到让利率很高、或让资本生产的商品价格大幅下跌。——所以即便没有再分配，一点点储蓄也能救很多人。——抱歉，你是说发展中国家有一些储蓄？——对对对。在发达国家，那点储蓄就足以产生大量剩余，让他们能用储蓄消费很多东西。——但我是说，混乱的中间地带在这种情况下可能更宽，因为它们的起点低得多——它们拥有的东西太少，而且和全球经济的——

[1:04:00] Dwarkesh Patel

economy. Right, yeah. And I think it's important for them to get on it now. And I don't have strongfeelings about whether it should take the form of like, sovereign wealth funds that invest inthe right supply chains or, or just, you know, subsidies to their own citizens to buy a little bit.This is actually, I think a crucial point. We were talking earlier about why the Rockefellersare whatever the world, why their descendants don't control everything. If our argument aboutthe selection of these kind of greedy optimizers hold. And one argument is just that it's likevery hard to index the economy. And maybe they would have just decided to have their heirs indexthe economy and have it grow at the rate of economic, have their wealth grow at the rate ofeconomic growth. And they would be, you know, trillionaires, their heirs would be trillionairesby now. But it just, before index funds existed, it just very hard to just get, get a represent,it just a very small fraction of the economy going back 100 years, accounts for majorityof the value created now. And if you miss those particular things, you would have basically,your wealth would have just kind of stagnated. And maybe there was a brief golden window from

——挂钩程度也很低。——对，是的。而且我觉得他们现在就得动起来。至于形式，我没有强烈意见：可以是投向正确供应链的主权财富基金，也可以是补贴本国公民去买一点。——这其实是个关键点。我们刚才聊到为什么洛克菲勒家族的后代没有掌控一切。如果我们关于「贪婪优化器会被选择出来」的论证成立，一种解释就是：给经济做指数化太难了。也许他们本来可以让继承人去指数化整个经济、让财富以经济增速增长，那他们的后代现在就是万亿富翁了。但在指数基金存在之前，这太难做到了——回看一百年，只有极小一部分东西贡献了今天所创造的绝大部分价值；如果你错过了那几样，你的财富基本上就停滞了。也许存在一个短暂的黄金窗口，从——

[1:05:00] Dwarkesh Patel

the creation of index funds up until, I don't know, five years ago, where actually you couldindex the economy and you could have your wealth grow at the rate of the economy grows. But nowthat we're in this world with very concentrated returns, especially two private companies,which is capital that is, as we were making a point in our blog posts, the average person hasdisproportionately less access to, as opposed to, you know, most of their capital is like having arandom house, at least in the US.Or a part of a house.Yeah, which is, as we were saying, is sort of unique, a capital that is uniquely ill suited tobe complementary to the production of AI or the serving of AI or to robots.Or the kinds of goods that the rich will bid up the prices of.Exactly, right? Because what is the value of a house currently? It is really, the land is closeto other humans and modular relational stuff that is just not going to be the main factor of production.And this is why Georgian tax would not raise enough money for the sort of programs that we weredescribing.Right. But stepping back, the point I was trying to make is, if it gets harder to index the economy now,and that is supposed to be the main way in which both one and normal people are supposed to

——从指数基金诞生一直到大概五年前：那时你确实可以指数化整个经济，让财富随经济一起增长。但现在我们处在一个回报高度集中的世界，尤其集中在私有公司身上——正如我们在博客文章里指出的，普通人对这类资本的可及性格外地低；相比之下，他们的大部分资本就是一栋普通的房子，至少在美国是这样。——或者一栋房子的一部分。——对。而正如我们所说，房子这种资本非常独特地不适合与 AI 的生产、AI 的服务或机器人形成互补。——也不适合去承接富人会去抬价的那类商品。——正是。因为房子的价值究竟来自什么？说到底是那块地「离其他人类近」，以及一些关系型的东西——而这些不会是未来主要的生产要素。这也是为什么乔治主义（Georgist）土地税筹不到我们刚才描述的那些项目所需的钱。——但退一步说，我想讲的是：如果现在给经济做指数化变难了，而它本该是——

[1:06:15] Dwarkesh Patel

modular some sort of universal income.In the developed world.In the developed world, are supposed to have some leverage on, or have some purchase on the wealth fromAI. And it's also the way that developing countries are supposed to have some purchase on the wealthgains from AI. But it's very hard. I don't know. Does Nigeria own a lot of SK Hynix and Anthropic?

——本该是某种普遍收入的主要实现方式。——在发达国家。——在发达国家，人们本该靠它对 AI 创造的财富有一点杠杆、有一点抓手；它同样也是发展中国家分享 AI 收益的方式。但这非常难。我不知道——尼日利亚持有很多 SK 海力士和 Anthropic 的股票吗？

[1:06:36] Alex Imas

I'm guessing not, right? It's not enough for them to just own the S&P 500.So actually, this brings up a really important point. Like, is AI going to be like electricity orsocial media?Right.If it's... So think about ComEd or ComEdison, whatever the electricity provider here is.It's a monopoly. It provides a resource that everybody uses. But do we think about electricityas like generating, creating concentration of power? And is ComEd like having like this huge amount ofpolitical power, social power, or something like that? No, because a lot with electricity,a lot of the downstream benefits actually came to like the users of the electricity rather than theactual entity producing the electricity. On the other hand, with social media, it was the oppositecase, right? Social media, you know, it was everywhere. Everybody uses social media, but the rents went tothe platform.But that's a really interesting point. The more you think... I don't endorse this take yet. I'm going totalk out loud. The more you think AGI is going to be... Our economy is going to be run on AGI the wayour economy currently runs on electricity. That is just a broad fundamental transformation of theentire economy. The more it looks like electricity and the more it's like every company in the S&P of

我猜没有吧。光持有标普 500 是不够的。——所以这引出一个非常重要的问题：AI 会更像电力，还是更像社交媒体？——对。——想想 ComEd（芝加哥的联合爱迪生电力公司），它是垄断的，提供一种人人都用的资源。但我们会觉得电力在制造权力集中吗？会觉得 ComEd 拥有巨大的政治权力、社会权力吗？不会。因为电力的下游收益大部分归了电力的使用者，而不是生产电力的那个实体。反过来，社交媒体正相反：社交媒体无处不在、人人都用，但租金（rents，即超额利润）归了平台。——这个点非常有意思。你越是认为……我还不认可这个说法，我只是在出声思考——你越是认为 AGI 会像今天的电力那样成为整个经济运转的底座、是对整个经济的一次广泛而根本的改造，它就越像电力，也就越意味着未来标普——

[1:07:53]

the future...Exactly.If it's going to make it to the S&P 500, it is because it has leveraged AI.Exactly. And then you're indexed again.Yeah, exactly. But then again, I guess it is totally...If you just look at how concentrated the S&P is over time, you know, just like these big techcompanies much more so... I guess this goes to a fundamental point that it's hard to reason about,about how much of the gains from AI these individual private companies will be able tocontrol.And I think like the open model thing is going to be a big point here, right? So like if we're indeedlike we're in a world where it's like the open models are six months behind the frontier,you know, nine months, then, you know, we'll hit AGI, we'll hit whatever. And like in six months,like everybody has access to this resource.And this goes to show you that every question is connected to every other because then thatquestion about whether there's runaway gains connects to questions about recursive improvementand even if not recursive improvement, then continual learning, which... or online learning,which lets a model learn on the job. So if it's deployed, it gets to learn more.And these are just sort of like technical question or forecasting technical questions,

——未来标普 500 里的每一家公司————正是。——一家公司如果能进标普 500，那正是因为它用上了 AI 的杠杆。——正是。那你就又能指数化了。——对，正是。但话说回来，如果你看标普的集中度随时间的变化，那些大科技公司的占比要高得多……我想这归结到一个很难推理的根本问题：AI 带来的收益中，有多少能被这些个别的私有公司攥在手里。——我觉得开源模型在这里会是个大变量。如果我们确实处在「开源模型落后前沿六个月、九个月」的世界，那我们实现 AGI 之后，六个月内所有人都能用上这项资源。——这也说明每个问题都和其他问题相连：这个「收益会不会失控集中」的问题，连着递归自我改进的问题；就算没有递归自我改进，也连着持续学习（continual learning）或在线学习（online learning）——它让模型能在岗位上学习，部署出去之后还能学到更多。而这些其实只是技术问题、或者说技术预测问题——

[1:08:59]

which then impact, I guess, whether Uganda will have any purchase on the returns of AGI.But it sounds like your answer really... The reason I'm emphasizing the question is I thinkboth for the messy middle and for developing countries, a recommendation that is often madenaively is you got to do some kind of retraining, you got to do some kind of likejobs program or you got to have them build data centers in our country.And I think you guys are suggesting something closer to just buy the index of AGI. That's likeprobably a much more cleaner and much more likely to succeed strategy.It's really good. These are the two scenarios, right? So I think there is a world where it isconcentrated, in which case it's going to be really hard to index AGI. There is another worldwhere it is not... It's electricity. Then like basically every company has access to AGI.So you just buy... You just buy the index. So like, you know, Nigeria just needs to buy the index.Right.And they... And Nigeria has access to AGI.Yeah.Right. Like because of the open models.Yeah. So just to get back to the question of like about whether to go with retraining or justtrying to index. I would prioritize trying to index, but just given how fast AI could,

——但它们反过来影响的是：乌干达能不能分到 AGI 的回报。不过听起来你们的答案是——我之所以强调这个问题，是因为我认为不论对「混乱的中间地带」还是对发展中国家，人们常给的那种天真建议是：你得搞某种再培训，你得搞某种就业项目，或者你得让他们在自己国家建数据中心。而你们两位的建议似乎更接近于：直接买 AGI 的指数。这大概是干净得多、也更可能成功的策略。——问得很好。这是两种情景：一种世界里它是高度集中的，那样就很难给 AGI 做指数化；另一种世界里它不集中，它是电力——那基本上每家公司都能用上 AGI，那你就买指数就行了。尼日利亚只要买指数。——对。——而且尼日利亚能用上 AGI。——对。——因为有开源模型。——对。所以回到「该做再培训还是该做指数化」这个问题：我会优先做指数化，因为 AI 可能来得太快——

[1:10:13]

you know, hit the world. But I definitely wouldn't just rely on that because like it could... Thesort of messy middle type cases or just the long timelines cases on which like you... We don't getanything like AGI all that soon. We'll still... You'll just be like leaving a lot of value on thetable if you could have like retrained to be a bit better, you know, like educated to how, you know,how to use the latest wave of computing. And yeah. So I don't think there's that much of a...An either or there. Like...I mean, maybe the reason to be pessimistic about this is because one of the reasons thecountry is poor is that it's a bad education system. And so becoming the best in the worldof retraining people at using AI, it doesn't seem like a particularly promising strategy forthis, that for country. Although there are cases where like in developing countries,you had this like leapfrogging effect with like, for example, like mobile banking or something likethat. It's much more prevalent than like Nigeria than it is in Germany or something like that. Like,they're... Everybody's doing mobile banking. They have it on their phones. They're constantlydoing this sort of thing. So, I mean, I... Again, I'm not putting probabilities on this,

——太快冲击整个世界。但我绝不会只靠这一手。因为在那些「混乱的中间地带」式情景里，或者在长时间线的情景里——我们并不会很快得到什么 AGI——那你就会白白留下大量价值没拿到，本来你可以通过再培训、通过教育让人们知道怎么用上最新一波计算技术，把事情做得好一些。所以我不觉得这里存在多强的非此即彼。——不过，对此悲观的一个理由也许是：一个国家之所以穷，原因之一就是教育系统差。所以「成为世界上最擅长再培训人们使用 AI 的国家」，对这样的国家来说似乎不是个特别有希望的策略。——不过发展中国家也有那种「蛙跳」（leapfrogging）的案例，比如手机银行——在尼日利亚比在德国普及得多，所有人都在用手机银行，天天都在用。所以我再说一次，我不给这件事赋概率——

[1:11:26] Phil Trammell

but like with a transformative technology like AI, you could get leapfrogging.Yeah.Where, you know, you skip the step in the middle and you can get like really astronomical growth.Maybe.Just about the ease of indexing. Can I just quickly say, I think it's definitely something toworry about a bit and keep an eye on. But as discussed in our own essay and as otherpeople have pointed out, it's already not that hard to index. So it's not, it's not, there's beena bit of an increase in the privatization of returns, but it's still like, you know, well under20% of the total market cap of non-tiny companies in the US is private. And, you know, everyone thinksabout open AI and Anthropic. And then if that's where all the wealth will accrue, then yeah, like allthese questions about whether open models will stay only a little bit behind, you know, those areimportant. But, you know, even they look like they're going public before too long probably.And the frictions that have been keeping companies from going public might themselves be alleviated byAI a lot, right? Just all of the disclosure requirements and whatnot. They want to getaccess to more potential investors too. And if I had to guess, I would guess that the kind of

——但对于 AI 这种变革性技术，你确实可能出现蛙跳：跳过中间那一步，得到极其惊人的增长。也许吧。——关于指数化的难易，我想快速说一句：我认为这确实值得担心一点、值得盯着。但正如我们自己那篇文章里讨论的、也有别人指出过的：现在做指数化其实并没有那么难。回报私有化确实有所上升，但美国「非微型公司」总市值中，私有部分仍然远低于 20%。而且大家一想就想到 OpenAI 和 Anthropic——如果财富真的都会沉淀在那儿，那关于「开源模型会不会只落后一点点」的这些问题就很重要了。但看起来，就连它们大概也快要上市了。而那些一直阻碍公司上市的摩擦，本身也可能被 AI 大幅缓解——那些信息披露要求之类。它们也想接触到更多潜在投资者。如果让我猜，我猜那种——

[1:12:43]

long kind of general trend of just like lowering those frictions and making it easier for more andmore people to index more and more will continue despite the recent bump in the other direction.Yeah.This actually makes me hope even more so than before that the labs do get commoditizedor at the very least they go public as soon as possible. But hopefully they just get totallycommoditized because I think AI will be much more popular and more importantly will be much morelikely to lead to broad increases in prosperity if the gains are just not particularly... It is ashard to capture the gains of AI as it is to capture the gains of electrification.Yeah, exactly. So I think like everybody, there's no anti-electricity people out there,right?

——那种长期的总体趋势——不断降低这些摩擦、让越来越多的人越来越容易做指数化——会继续下去，尽管最近出现了反方向的小波动。——对。——这让我比以前更希望这些实验室能被商品化，或者至少尽快上市。但最好还是被彻底商品化，因为我认为，如果 AI 的收益就是没那么容易被攫取——攫取 AI 的收益和攫取电气化的收益一样难——那 AI 会受欢迎得多，更重要的是，它更可能带来广泛的繁荣提升。——对，正是。所以我想每个人……世界上并没有「反电力人士」，对吧？

[1:13:26]

I mean, electricity doesn't take your job, but...Well, it...To some people's jobs.Yeah, yeah. And I think it's, you know, this is maybe a tangential to the conversation. I thinklike there's like a really... Narratives matter and there's this like really negative narrativearound AI right now, but that's because people are not putting out the positive narrative or because...And there's a reason. It's more difficult to imagine something that doesn't exist,that's a good thing, than losing something that exists.Right, yeah.Right. So it's very easy for somebody to go on a podcast and to say like,these jobs that you like, they're going away, than to somebody to spin up like autopia which doesn't exist yet.Right. I hope this isn't too out of left field, but I think I would be remiss if I didn't point outone big cost of having commoditized frontier AI models, which is thethe tech race dynamic, right? That like for safety purposes, you might want fewer frontiercompanies so that each one has a buffer in case they want to slow things down to make things safer.And the way this relates to our point before about the kind of widespread access, you know,of the returns, is that I think there's a lot less of a trade-off there than some people imagine,

——电力不会抢走你的工作啊。——它会……——它确实抢走过一些人的工作。——对对。而且我觉得，这可能有点偏题：叙事是重要的。现在围绕 AI 有一种非常负面的叙事，但那是因为没人把正面叙事讲出来，或者说————这是有原因的：想象一个尚不存在的好东西，比想象失去一个已经存在的东西要难得多。——对，是的。——所以有人上播客说「你喜欢的这些工作要没了」很容易，而要有人凭空构想出一个还不存在的乌托邦就难得多。——对。我希望这不算太跑题，但我觉得如果我不指出「商品化前沿 AI 模型」的一大代价，就是失职了——那就是技术竞赛动力学：出于安全考虑，你可能希望前沿公司少一些，这样每一家都有缓冲，万一它们想为了安全而放慢速度。而这和我们之前讲的「收益广泛可及」的关系是：我认为这里的取舍比很多人想象的要小得多——

[1:14:44] Phil Trammell

where you know, some people think either frontier AI gets commoditized and we all enjoy the benefits,but there might be some risk because like it's the market's really competitive and cutthroat orthings are safer because there's a big gap between the leader and the laggard,

——有人认为：要么前沿 AI 被商品化、我们都享受好处，但因为市场竞争激烈残酷而存在一些风险；要么因为领先者和落后者之间差距很大所以更安全，

[1:15:02] Dwarkesh Patel

but that means that the leaders get fantastically wealthy. No, like you could just have a relativelybig gap, but it's a public company ownership and it's widely distributed.Yeah, yeah, yeah. More recently, I have been thinking that the risk of commodification,which is that it sort of diffuses the, it diffuses the ability to use AI to harmful ends,is worth the benefit that I just feel, I worry that not only having these concentrated labsmakes it so that the sort of surplus isn't as widely distributed through society, but also it creates avery tangible, clear political target for the government to, I mean, we saw this with theDefense Production Act threat against Anthropic. If there wasn't one lab that is, or a couple oflabs that are clearly ahead of others, this kind of threat would be much harder to make.Thank you guys for doing this. Yeah, thank you.Thank you.I feel like there's a lot of unresolved questions, but I, it is, it is helpful to know what the relevant,at least like, what is the first branch along all these important dimensions?

但那意味着领先者会变得富可敌国。不是这样的——你完全可以有一个相当大的差距，同时它是一家上市公司、所有权广泛分散。——对对对。最近我越来越觉得，商品化的风险——也就是它会扩散「把 AI 用于有害目的」的能力——是值得为其收益付出的代价。我担心的是，只有几家高度集中的实验室，不仅让剩余无法在社会中广泛分配，还制造了一个非常具体、非常清晰的政治靶子，让政府可以对准它——我们在《国防生产法》（Defense Production Act）威胁 Anthropic 那件事上就看到了。如果没有某一家、或某几家明显领先于其他公司的实验室，这种威胁会难做得多。——谢谢两位来上节目。——谢谢。——谢谢。——我感觉还有很多问题没有解决，但知道这些重要维度上的第一个分叉点在哪儿，本身就很有帮助。

[1:16:04]

Great. Thank you.Okay. Well,

太好了，谢谢。——好的，那——