Transcript: "How will AI impact the economy?"

The text below is a transcript of the audio from Episode 52 of Onward, "How will AI impact the economy?"

___

Cardiff: Hello, and welcome to Onward the Fundrise Podcast, where you'll hear in-depth conversations about the big trends affecting the US and global economies. We are recording this on Thursday, December 18th, 2025, and before we start today's show, an evergreen reminder that this podcast is not investment advice and is intended for informational and entertainment purposes only

And now let's get on with the show. This is, the AI may take our jobs, but it'll never take our freedom to speculate about its effects on the economy. I'm Cardiff Garcia of the Economic Innovation Group, and I'm joined as always by co-host Ben Miller, CEO of Fundrise. Ben, this, this is the big one. This is the thing everybody's talking about right now.

This is the single biggest economic trend of the year. Are you ready for this?

Ben: Let's rumble.

Cardiff: It's also something that I should say that you've been thinking about for years. You know, technology, its effects on the economy, its effects on society, and also, of course, how it affects Fundrise's own approach to investing.

So I think it's worth pausing just to, just to discuss why you wanted to discuss this specific topic, uh, on this end of year episode.

Ben: Yes. So a lot of ink has been spilled on this question. And so what can we bring to the table that's new or or fresh? And I wanna argue that the question everybody's debating now, which is, is AI a bubble?

Cardiff: Is it, has it gone too far? Is it all gonna end, you

Ben: blow up? Is the price

Cardiff: everybody's gonna lose their money? Sure. That, that is the question that in fact everybody's talking about, will it crash the economy? If there is, if the bubble, if of all does exist, and then if it pops and you're saying, ask a different question.

Ben: Yeah, I'm saying that that's boring.

Cardiff: I'll be honest. I think it's pretty interesting because it's so, it's so full of tension, right? Like we're constantly on edge looking for the next indicator that says, yes, it's a bubble, or no, this is actually a thing. Um, but I,I suspect, I, I suspect, excuse me, that you bored of it just because there has been so much chatter about it.

Like this has been month after, month after month of people asking and trying to answer that very same question. Is that right?

Ben: Yeah. And, and I wanna argue, um, that it doesn't matter in the long run, like what matters. Like take the internet. The internet had a bubble in the early 90, like late nineties. Um, and it doesn't matter now, what mattered was the impact of the internet, like decades later, how it changed business and the economy and politics.

And so I feel like what I'd like to do now and generally is move past the question of is it a bubble? And say, forget. If it's a bubble, what happens after the a, a AI gets billed out. Because I think that there's some, um, basic things that are pretty obvious when you, once you play it out, and then some things we can debate.

And I, and I'm, I'm interested to see if we end up on different sides of some of these questions, but I thinkthat, like talking about if AI is a bubble, bubbles, like talking about the weather, like if you wait long enough, it'll change.

Cardiff: Sure.

Ben: It's a transient question. It's a question for investors and traders, but for the rest of the a hundred sub millions of people who will have AI change their lives, like, like how is it gonna change the, the economy?

I think it's gonna have very clear and I think pretty obvious effects on the economy. And then the other thing I can bring to the table is that because we've been building with ai, we've rolled it out across the company, we're an early adopter, we're investing in it. I think I have a better sense of what it's gonna do to the economy and to companies.

And I can talk about real specific examples of that.

Cardiff: Yeah, that'd be great. And, and also to discuss, you know, how, how it alters your own approach to managing Fundrise investments and also just managing Fundrise itself, the company, you know, your, your, your decisions, the team, your decisions as a CEO. I'm glad you brought up that analogy to the late nineties bubble.

The ideathat yes, it was an investment bubble. Yes. You know, it went too far and in the end we were still left with this extraordinary infrastructure, which set this. Stage for the technologies of the next couple of decades. And yet if you go back in time and you look at some of the companies that failed in the bubble, some of the technologies that they had put in place were actually kind of interesting and would end up playing out much later.

Like I, I think all the time about like cosmo.com, right? Delivering groceries to your, to your door and whatnot. Like that was a great idea. Way too far ahead of its time in a way. And the companies that did. Survive. The shakeout have gone on to do extraordinary things. Obviously Amazon, Microsoft, two of this big, you know, titans even now today.

Um, but the companies that failed are always on my mind because something like that could happen. And yet, regardless of who makes it and who gets shaken out, if in fact this is a bubble, if we're left with the technology afterwards, what are the effects gonna be? What'll be comparableto the IT build out of the late nineties?

Where did the analogies start and stop. But I, I love that we're talking about it in those terms because this is a different kind of situation. This is not a bubble driven by, I don't know, like mortgage debt or whatever. You know, this is, this is something where it's putting in place the infrastructure for what could be a transformative technology.

Maybe, maybe not. Obviously we'll find out, you know, if its effects are in fact as big as what happened in the aftermath of the nineties. I don't know. Um, but, but it's a fascinating way to think about this, you know?

Ben: I think that as a user or builder with ai, the idea that it's not gonna be enormously impactful is just off the table. For me. It's not even like one of the possible con concepts I consider like AI is definitely big. I think it's way bigger than the internet, not even close.

Cardiff: Really. Okay.

Ben: um, and so, so let, let, lemme see.

Let's try to walk through first thispremise. Let me just do the premise and then you and I can play out the debate. Okay. So, so moving past this question of whether or not it's a bubble, which is a question of saying whether or not it's a good investment and whether or not something's good. Investment is about like, uh, economics price.

So price question ultimately. Um, but if you move past that question and say, okay, what is going to happen over the next two to five years? Three to five years? So I think that if in almost all circumstances the US and the world's gonna build out trillions of dollars of AI data centers, and I'm gonna narrow it to the US 'cause it's just too hard for me to think about the world.

Just too complicated. So I just, so put some numbers on this and, and then we can talk about the, the cost and the consequences, but. By every, um, uh, analysis I've seen, and I've talked to people whobuild data centers, like I talked to some people building billions of dollars data centers. We'll build somewhere between eight to 10 gigawatts of data centers every year for the next five years.

We built five or six in 2025, and we'll build eight probably in 2026. And so you're talking about 40 to 50 gigawatts of AI data centers to be built by the end of this decade. And I think that like when people are debating if it's a bubble, they're asking if that's a good investment, but they are taking it for granted that it's gonna happen to some extent, you know, in the trillions.

And so, okay, so when people hear those numbers, I don't think they really know. What does that mean? What does 50 gigawatts of AI data centers mean? Like it doesn't mean

Cardiff: funny that you say it 'cause 'cause that's what I was gonna ask you. What does that mean? That means absolutely nothing to me. I don't mean that it's meaningless. What I mean is that I don't, I don't comprehend what's being said right now. I don't have the technical expertise.

Ben: So much inkspilled about it, but then you, but about whether or not that's a good idea rather than like, well, what does that mean if they do? Because I'm pretty sure they're going to, uh, and so, okay, so let's, let's talk about what that means. So there's a few different ways to look at that. So what does that mean in terms of total cost?

What does that mean in terms of, um, the, the economics needed to, to support that? And then what does it mean in terms of total output? So it's, so, so let's just say 50 gigawatts is about $2 trillion over, over, over five years. You know, we are spending up four or $500 billion a year. And so it seems very realistic.

Um, you need about $400 billion of economics to support. $2 trillion of CapEx spend of, of AI data center. So, so like, that's what you need. Maybe you, hopefully you get more, but like, for that, for that to, um, make in economic sense, you spend 2 trillion, the, the,the GPUs, the chips, they depreciate really quickly.

You know, some people say three years, some people say five years, but depreciate quickly. So you need like a lot of economic support. That, and then out of all of this, spend comes, tokens comes, you know, the ais that the out, the actual output that comes to your computer, that, that actually is used are these things called tokens.

And I feel like everybody knows what tokens are, but just in case, should we

Cardiff: No, it's don't, don't, don't assume that either. Yeah. Please, please tell. Tell

Ben: 'cause it's so funny 'cause it's like we're, everybody talked about the amount of money. But then they actually say like, okay, but what, what are they, what actually are they building? So if you think about it, like as a factory, they're building these giant AI factories.

Instead of producing cars or steel, they're producing tokens. And so a token is how AI thinks. Like I think people in our, and humans think in terms of language, right? When you, you think in maybe in terms of emotions, like, so the actual raw like building block ofai, that, that they think and produce is called a token.

And so, um, typically a token is about four letters. So like, um, the word cat is one token. The word running is like two tokens. New York City is three tokens, right? A picture of a light bulb is only one token, right? So like we can imagine, okay, since like these ais are these giant large language models.

Language is words, but they don't think in words. They think in these tokens. And, and tokens are how they, how they process things and how they then produce outputs for you. And so you're gonna, you're gonna, I'm gonna put some numbers on this. If you spend 50, uh, sorry. If you build 50 gigawatts of, of AI data centers and it costs you $2 trillion, that'll produce about a thousand trillion tokens per day.

Cardiff: Okay. What comes after trillions? Is it, is it Quadrillions? Quadrillions.

Ben: So, so, but I didn't, youknow, who the hell knows what

Cardiff: Yeah. Well, at that, at that point you might as well see. That's like saying, uh, bazillion, you know, gajillion or something like that. Yeah, I get it.

Ben: yeah, yeah. So, okay. So you're gonna spend all this money to get a thousand trillion tokens a day, you know, or, or about that. Could it? It's, there's some, a lot of assumptions inside of, you know, how efficient the models are, how efficient the chips are, and blah, blah, blah. So, so like, and that's about, um.

Four to 5 million tokens per adult person per day.

Cardiff: Yeah, it's a huge number. And by the way, to put this somewhat in economic terms for people who want a little bit of context there, you know, us GDP right now hovers around like $30 trillion. So if we're talking about a build out, um, that grows by hundreds of billions each year, you really are getting close to adding like whole percentage points of GDP growth of economic growth each year.

'cause of course, what matters from year to year is not whether you hit the number from the prior year,it's how much more or less you end up spending on investment. Right? And so if you incremental. Only keep growing that number so that it goes from whatever you just said. I'm, I'm sort of spit balling here, but 500 billion to 2 trillion or something like that.

You really are having a meaningful effect. You are meaningfully adding to GDP growth every year for the rest of the decade. So, you know, again, I don't want people to like, pinpoint us and say, oh, you said they were gonna spend exactly this amount. We don't know. This is speculative. But if we're talking about those magnitudes, we are talking about significant orders of magnitude.

We're not talking about, um, a kind of trivial addition to the economy. So if this continues and it, and it incrementally grows, it's a big increment. You know, it, it's a big deal.

Ben: And everybody's obsessed with the input, which is a huge amount of trillions of spend. And is it economic? And I'm just, I wanna talk about the output. So what do you get for spending $2 trillion? Like what's the output? And right now we, we can say we know it'stokens. Okay, so a hundred, a hundred, sorry, a thousand trillion tokens a day.

Or let's just say four to 5 million tokens per person. So now then the question is, okay, well you spend all this money to get this out, but what is 5 million tokens per day per person gets you? Right? Because that's like, that's the big question. You're not gonna spend trillions of dollars. F for something that's not valuable.

And so this is the thing I think people don't have any intuition around. They, most people don't even really know what a token is, let alone what does 5 trillion tokens get you? And that's where I think my experience building and using AI at Fundrise, 'cause we use, you know, we use millions and millions tokens a day.

And I can sort of say, okay, like if we're producing 5 million tokens per person per day, what, what would that mean to the economy? What would that mean to work? What would that mean to, to the types of, uh, of things AI could do? 'cause today, you know, there aren't 5 million tokens per person today.Um, and so that I think is the thing that note that I haven't really heard anybody talk about.

Cardiff: Yeah. The, the thing that comes to mind, uh, again, probably using a kind of economic approach to thinking through this, is you're talking about the output of all that spend. And I'm thinking also in terms of like, what will people pay for it? Do they want all that output, right? Like that has to be a part of the equation too.

When, when people talk about like bubbles and things like that, that's usually what they mean. But I think about it also in terms of diffusion, you know, whether or not there's, um, there's gonna be a lot more places that currently are not using it, start to use it. And the places that use it now, maybe on the periphery, you know, for marketing or to help with customer service or things like that, are they gonna start using it for more fundamental reasons?

And what I'll say there is that the surveys that have been taken show that there is some room to grow there. So I'll give you one simple example 'cause it ties directly into what you just said about the next few years. Uh, the Census Bureau conducts a survey, and in fact, likeour research director, uh, a guy named Nathan Golds, who's great, uh, where I work at, EIG, he used to work at Census and he played a big role in this specific survey, so he knows how to interpret it very well.

And as of right now, only about 17% of total US businesses say that they use AI for any purpose. So this could be either to make goods and services or on the periphery, um, which suggests to me a couple things. One is that use is still a little lower than people realize, but second, it means that there's a lot of space for expansion here.

You know, that if companies start finding better and better ways to use it. There's a lot of companies out there that can start using it. Okay. So again, this is somewhat speculative, but as you think that AI is gonna keep getting better and there's no reason to think it won't, that figure really could be a fast moving target in the future, in the next few years.

So that's, that's sort of what I would, what I would say in response to, to your comments about all this output that you're getting for it. Could people use it? I would say,yeah. A lot of people in a lot of businesses can't still find the room to use it if they, if the, if the use cases are there, you know?

Ben: Yeah, I think most people are skeptical because they haven't seen the use firsthand in any meaningful way. And so they're, they're sort of like, they're, they're wondering, well, is this bubble? 'cause I don't see, like, in my day-to-day life, a lot of changes from AI tokens. You know, I'm, I, maybe I use chat GT once in a while to ask a question or I see it in, in Google search, but I'm not seeing it in my day-today.

Life transforming how I do work. And so how do you justify the economics of this? And so I think that there's this gap, knowledge gap, right? The whole thing. What is it that the future is here is just unevenly distributed, right? Like there are, there are people flying in jets and at the same time there are people in agricultural like.

Back hinterlands of, of China, you know, using oxen to plow the road. So, so like,you know, some people will know about what's happening way before other people, and I think I'm just happened to be on, uh, closer to the edge, not the edge. 'cause I'm not a AI research lab, but I, I, I know what people at our company are doing with it and I can see exactly how it'll change people's lives.

Cardiff: I, I think it's accurate to characterize yourself in Fundrise as early movers. You might not be first movers, but you're early movers, especially in the space in which you operate. I think you, you've gotten to the point where you use AI inside of Fundrise, um, and you use it yourself for fundamental reasons, not just, you know, I think if you were to receive that survey Okay, on Fundrise's behalf, you would be one of the people saying, yes, we use it to produce goods and services.

Is that right? Am I, am

Ben: Yes. And I wanna, I wanna go through, because I wanna answer this question of, okay, if everybody had 5 million tokens a day, what would that mean? And I, and I think that's like, there's a few ways to do this sort of stylized, um, of how toanswer this question of what is the consequence of ai. And so one way to do it is this like, okay, so you spend this money, you get these tokens, what can the tokens do?

And I think that part of the reason why, um, there's this disconnect is the tokens are kinda a raw, raw input. So like, it's almost like a steel factory. Steel's not that useful, but like steel that's turned into girders for, for like, you know, building buildings or railroad ties or, or you know, cans for food, like the, the raw input is not as useful as the actual application.

And so there's this intermediate step that's happening right now, which is that you have to have built an application with AI that uses tokens as the input and the output actually some kind of useful work. And so I, we've done that. Uh, across the company. I wanna do some, some, share some examples and talk about how many tokens each one of these things actually uses to give you a sense of like, okay, what's the economic value of,you know, of 5 million tokens?

Because if economic value is very high, then you start thinking uhoh, like, maybe this is actually gonna be a huge impact and people are under appreciating it.

Cardiff: Yeah. And also it, it would then earn the right to, to have itself labeled general purpose technology. You know, the, the technology that doesn't just affect the immediate sector in which the technology was developed, but it essentially affects, you know, technologies inside of industries and sectors, all across the economy, changes everything.

Everybody has to rethink how they produce things. Um, and consumers end up also just altering their own habits, their consumption habits because of it. That's a general purpose technology. And if what you just said is what comes to pass, then yeah, that it will definitely qualify as that. So yeah, go go through one or two

Ben: Yeah. It's funny to me 'cause I, I, it's like people are debating and I'm like, no, it's, it's, it's already in the past. It's like old news. We, I know this is happening. It's de

Cardiff: this is a 2024

Ben: what the heck arewe talking about? Is it definitely a general, general purpose technology? It's definitely, it's so general purpose that I, and I think this is like a funny thing for most people, most normal people who are like not in the tech business.

This, they look at AI and they're both skeptical and actually afraid, and they think there's more fear of AI than any technology ever seen.

Cardiff: Yeah.

Ben: And I think they're right to be afraid. I mean, I, I think that it's gonna have a lot of negative and positive impacts. And I think people's instincts are not that far off when I've seen the impact, like in a, in the micro sense, like, so I'm gonna go through, lemme go through some, some examples and I'll come back to why this general purpose technology, uh, uh, theory that you were putting out there.

'cause I think it's, it's the most general purpose technology. Maybe electricity is more, it's hard. I'm gonna, I think so. But like what? Ai, AI is auto, it automates cognitive labor. Right. And the same way that we used, we automated factories, like thenumber of workers per factory has been falling for decades because you can automate, you know, making cars and making you know, shoes.

And that's what's gonna happen. And that's what is literally happening two day at, you know, at companies around white collar work or knowledge work. So here, let me give you, uh, a few examples. So, okay, so Fundrise, lemme do customer service first, but I, I could do customer service, product development, real estate analysis, accounting.

Uh, I could, I could do software development. I mean, there isn't a department that hasn't been impacted and it's really a question of the scale of impact. So, so let me go through some specific examples. So customer service. So Fundrise gets something like 6,000 customer service tickets a month, six or 7,000 a month.

Uh, half of those are handled by ai. And you know, the, the questions are like, anything from reset my password tohelp me understand this investment. So this is sort of broad set of, of kinds of questions people ask. So people are writing in emails or chat, and then like, depending on the question, it gets triaged to an ai from the ai.

I, I want to, I want to anthropomorphize the a the AI say, okay, it's like a person, so what does that person cost and how many people did it replace? Because it, it fundamentally replaced a lot of people. So, um, that it handles half our, our customer service and it costs 51,000. It costs us $51,000 a year.

That's how much we're spending for this AI customer service agent. It replaced six people and it, we, it uses about 2 million tokens a day.

Cardiff: Let me just pause on the, the replacement of people that you would need to do the same job that the AI is doing. The ai, AI costs $50,000, six peopledoing a job like that. Assume, and I'm just, this is a, just, just for a nice round number, a hundred grand per person, a salary and compensation, that kind of thing.

That's a 600,000, that's a 12 to one savings is essentially what just happened there, or one 12 savings I should say.

Ben: It's easy for people to imagine and they, they know that they're interacting with ai. I think more and more like I think, you know, you're gonna see Delta Airlines and, and lots of different customer service become ai and that what's interesting about it is that in some ways it's, it's, you know, not as good as a person in some way.

It's, it's far superior. So let's talk about how it's superior. 'cause I think there's dimensions where people aren't thinking about it and where I think it's becomes very powerful. So one is, it's 24 7, so you can get that customer service. All the time. You know, if you, if you have a flight, you miss your flight, it's midnight now.

You can get your flight changed, you can get the customer service all the time. There'sno wait time. So it's way faster. Um, in terms of how to affect the organization, I think this is again, something that we're, we'll repeat is it took away the easy work or if you wanted to say the clear work or the work that's like most routinized, the most like sort of like, uh, thing, repetitive work.

And, and it's an interesting, I think it, when you think about that, extrapolate that to the, to the whole country. On one hand that's great because what people, a team actually loves it. They got, they got rid of the work that was repetitive and boring and they kept the work. That's really interesting. And high value.

On the other hand, I think it, what's left is their harder work. And so it's like interesting, like it's nice during your day to have some easy work, some hard work, but if you use, if you're mostly stuck with the hard work, I think it's gonna have an interesting sort of like, like asurprising effect. So

Cardiff: certainly, it, it requires a different approach to the work itself. You have to sort of, um, you have to marshal your energy in a different way. If you have a day where half the day is sending kind of mindless emails with that kind of thing, and the other half is really hard thinking, you know, you could just about, you know, pull it off.

You know, if you, if you save your time, you know, if you manage your time the right way, if you have eight to 10 hours of really hard thinking work, like that's, that's a tough day. You know, you almost can't happen. Right? You need to space it out a little bit, um, because really hard work requires a different approach than the kind of routine stuff that you're describing

Ben: and one of the challenges that is that it's um, a lot of times really hard work. It's not like four hours is more valuable than two hours. Like it, it, a lot of times, this is something, I think there's something about the brain at fundraising, we're doing hard planning. We will do a session and be like, let's come back a week from now.

'cause literally a week from now we'll havethe answer and we won't even know how our brain came up with it. But it's some sort of like hidden cognitive analysis happening when I'm sleeping and, and when I'm in the shower. And so like, taking away a big chunk of the easy work isn't like all gravy for me.

So it's so, so this is a subtle thing and we can come back to some of these larger meta

Cardiff: is very common for creative work, especially work where you really need to figure out how to combine ideas and, and, and, you know, find new ways of doing things. You know, inventions, you think really hard about it. Then you step away, something clicks and you come back. It's all a little bit mysterious, but it's been shown again and again to really happen.

Um, and it's especially helpful for the hard problems to have that break in between. It's just a very different approach to work than what you're describing. Uh, routine work is like, you know?

Ben: Yeah. Yeah. And I, and I, I, I'm, I'm later. When we get to it, I think there's some interesting analogies that we've seen transformations like this from likeagricultural society to industrial society, and we're going through this transition and the work you did to being an agricultural society, you know, farming and the work you're doing, sitting in a desk is so different.

So we're in this, I think we're gonna go through a radical transformation, but let's come to that in the end. I just wanna give you more examples

Cardiff: I'd like to hear more examples, especially 'cause you just described customer service, but of course Fundrise is not primarily a customer service, uh, organization, even though of course it has a customer service component to it. Um, why don't you talk about the way it's used for the really nuts and bolts stuff like investment ideas, investment decisions, real estate analysis, stuff like that.

Ben: Okay. I wanted to do things that were more common across lots of organizations. Uh, let me come back to that one because I think that one is like, um, as a manager and a manager of an organization, one of the things I find is that, that there's two kinds of, um, um, management. There's figuring out, defining the work,and then there's doing the work once you figure out what to do.

And I, and I, obviously, I'm on the mostly figure out what to do job, um, but like, um. I think that the, doing the work is gonna get increasingly done by ai. Uh, and the figuring out the work is gonna be done by people and, and people who, who like uncertainty will do better in this sort of world. But anyways, so okay, let's, here's another one.

I just want to, I'm trying to give you a sense of what tokens are worth. Um, and so I skipped a piece, which is that the, the key to this analysis is that you need to turn raw tokens into an application and then when you have an application, both you can see the work it does and it's far more efficient.

Or even even tokens without an application almost can't get the job done. But once you root, you know, build the workflows and build the product. And so like another example is likecybersecurity. So we at Fundrise get billions of, uh, attacks all the time. There's sort of this 24 7 management of like access, all these things happening and sort of like, almost think of it as like watchdogs, like that sort of, and, and you have an IT team and the IT team does that kind of work.

They also like help, you know, help desk people are like, Hey, my computer's broken. I'm, I can't figure out how to, they're like, try turning it back on and off. You know, there, there's all these like work around it. And, um, and so we implemented it around. I mean, we had to build these workflows 'cause a lot of this stuff didn't totally exist, but we built these workflows around cybersecurity and can get kind of technical around it.

But it just, the point is that it, it does a lot of analysis looking at things that are happening, like attacks and like that and where people, like somebody turns on their computer, that computer is in Argentina, should it have been Argentina or not? Did the person go on vacation? Uh, that can be done by ai rather thansaying like, you know, some IT professionals, like trying to check to see that person put a notice in that they were gonna be on vacation and brought their computer with 'em or not.

Like, there's all these kind of things happening and so we use, that's something that's like highly valuable, saves hundreds of of hours a month and it's tokens a month.

Cardiff: That's a good example. It's definitely something that applies everywhere else too.

Ben: Another one is, I mean this is, everybody knows about this, but like we used to have three copywriters at the company and now we don't have any 'cause AI's so good at writing copy. Um, and so like, uh, I'll, I mean I can talk about product design, which again maybe is a little bit too specific to to, to software industry.

But when you write, when you create software product, like, you know, let's say, what should this website look like? Used to have designers and, and product managers, and now. A person can kind of like really figure out what it needs to look like. You know, they call it vibe codingand use that to give to the engineers.

So it's like a way to scope out what you're building. So what? And, and, and so what AI is really good at doing is doing things that are, that require, um, a lot of like, sort of greenfield new thinking fast. And so a lot of times with the software development, you're trying to figure out how to kind of attack a problem.

And so like I asked our software team and our CTO, my CTO is very earthy, like very grounded. And so he used to be total skeptic of ai. Absolutely. His skeptic of everything at first. 'cause that's like his nature, which makes him good. And um, you know, it's, I said if you two, two teams or a hundred software engineers, front end backend mobile engineers, how many one is ai, you know, codex or Cloud Code One doesn't.

What's the efficiency gain? And he's, it is probably 25 to 50%.Okay. So across these boards, right? We're getting the, the pattern is, and I and, and how many tokens do you need? Like I, um, I'll give you one other one. So I'm trying to give you this, this sense of what tokens are worth for work. 'cause tokens are cognitive work, right?

Is that, um, a software engineer built out like this sort of like, uh, uh, project for AI to do it has really good at like, kind of, again, routinized products, projects. So like, like a lot of times you have to upgrade your software. You go from like, you know, this is not exactly right, but like Microsoft Office 95 to Microsoft Host 2001.

There's like these changes to the front end and to the view migration. And so you can really, like, here's one that's really interesting, doge failed. Partly because it was trying to upgrade software at Social Security Administration and Medicare. That was written like with cobol, you can now do that migration withAI to Java, where before it would've been impo, it literally would've been impossible.

And now if you, if you build it out, it can literally do that migration from cobalt to, to Java. And you're, so you're talking about saving, I mean, you know, untold years of, of, of, of work. So, so anyways, okay. So, um, a typical project that might've taken our team eight hours, takes them maybe 30 minutes, and it costs 5 million tokens.

So an advanced software engineer is probably costing you hundreds of thousands dollars a day. A year. It saved them, you know, let's say six hours, maybe. It costs, and that was 5 million tokens and 'cause of token caching, blah, blah, blah. It costs $2 50 cents. So, so the, the, the, the, the bottom line is that, um, tokens in an application that's built to do a certain job, um,are highly valuable.

And so like 5 million to, like I would say, if you said you took every white collar worker in America, there's a hundred million workers, and you said, okay, each one of them got 5 million tokens. That's easily enough tokens to do half of their work. So, ha, I would, I mean like now of course you have a diffusion problem, adoption problem.

That's really the problem.

Cardiff: Or they could double how much work they do or how much output they themselves get out in the world. Right? Like it makes them more productive.

Ben: But the point is, I think that what I've been trying to do, and this is the, this is really the punchline, is how do you, how many tokens equal to a worker like an FTE, a full-time equivalent? Like how many tokens is, is, and I, and I'm, and I'm saying that I think about 2 million tokens is enough to do 50% of anybody, almost anybody'sjob.

Cardiff: So with enough tokens, basically you could replace a huge chunk of the white collar workforce. I think that's where you're going with this, right? Yeah.

Ben: No, Ken, I mean, I think that, like, it's not even a Ken. I think it's, it's going to

Cardiff: Or will or, yeah.

Ben: happen. It's only a question of diffusion. It's actually like, this is, I mean, we've, we've been doing that since we launched an adopted ai. We haven't hired a person since you're seeing that across the economy, no one's hiring anymore.

You don't need to hire, there hasn't, we haven't actually had as many people let go. It's just made the existing team so much more effective. And so I think that we're like, uh, we are the future of what companies will like, we just need a lot less people to do a lot more work.

Cardiff: Yeah, there's an interesting next order effect there

Ben: Mm-hmm.

Cardiff: which is, let's say for a minute that the economy starts operating the way you've just described it, which is not hiring as many people as you would've hired in the absence of this technology,still getting as much or more than ever done with the same number of people, right?

Companies become more efficient, more innovative, more productive, you know? Take that to the next level. What that also means is that the people who continue working for Fundrise. Make more money, right? You're making more product. You're selling more product, like you're more efficient. So with that, more money, with that added money you're making, uh, you'll then look for things to spend it on, right?

Then you'll be able to afford things that in the past you may not have been able to afford, and you'll spend it on parts of the economy that are not being affected as much by this in terms of the number of people that are employed, right? And so this is just like a basic sectoral shift, maybe accelerated, maybe rapid, but the nature of the economy.

Could change because of it, but it doesn't mean that like everybody's condemned to unemployment forever and ever. It's gonna have all kinds of interesting effects. It mightchange the value of different skill sets. I often joke about this with our chief economist, Adam Ek, you know, saying that like people will look for other things to spend their money on.

A lot of things that people like to spend on involve, like interpersonal interaction, a lot of like, you know, live performances, things like that. And he makes the point that we came up with automated pianos, you know, a hundred and something years ago, but people still go to concerts where they see human beings play pianos.

Human piano players are employed in bars and hotels and things like that, and we like seeing other people perform human level things. It just might be a little bit more of a performance based thing. And if you think about the economy now, anyways, it already has a heavy component of performance attached to it more than in the past.

You know, I'm not just influencers and things like that. I mean, everything. Look at us, look at you. And I like, we're on video now, right? Like we're, we're in a sense on a podcast, like in a sense, like we're, we're doing something that is being performed even though we could technically type in, you know, into a document somethingpretty quickly and send it to Google, LM or whatever, and it would spit out some version of this.

But we're doing it. And I think people appreciate being able to hear you discuss, like, you know, your version of like, what a what? Where the economy's headed, where Fundrise is headed, that kind of thing, rather than having it filtered through the voice of a computer or whatever. You know what I mean? So like I, I just, I think about all these different things and I think we always have to remember to go to that next level effect and not just stop at not as many people are gonna have jobs here or in other parts of the white collar labor force.

Ben: Yeah. Yeah. You're getting to like the, okay, let's talk about what it means for the economy, but I'm trying to get us over the hump that people haven't experienced yet, or, or don't know about, which is that AI is going to do white collar work.

Cardiff: It's coming.

Ben: It's, it's a fact. It's not a, a hypothesis. It's already happening.

If you, you can adopt it. You can, if you spend, if you have smartpeople who can spend all the time building out the, the, the, the routines with ai, it will take on work and it's gonna do meaningful part of people's work. Tens of percent, tens of percent. So that's tens of millions of people of work. And you, you can look at this sort of like a bunch of different ways, which I have.

So you get to pass this question of, is it gonna affect the society? And you start saying, okay, what is the effect gonna be? And it's definitely a cog, it's definitely a general purpose technology. And let's talk about general purpose technology and how, how is it different than past general purpose technologies?

Because I think it's more general because it automates cognitive labor and other technologies in the past really didn't have. All of the boxes checked. So the, the, the things that, the things that make it so wild and I I, is that itessentially, it's like capital that behaves like labor. You have a huge CapEx investment, it's highly flexible and then it scales to at, at nearly, you know, to nearly infinity.

Like it's, it's, um, it's another way to look at, it's very fungible. So if you have tokens, right? Tokens be used in different industries. So you can be in healthcare or you could use an industry, you could use a token in finance you used at different companies, be used at different jobs, like, you know, a lawyer or an administrator.

We use the different tasks. A writer or coding, we've talked about it. You can use 24 7 and also it can be used in any language to any nation. So it's super generalized. Like a, a, a person. You may be super competent, but you can't change industries, you can't change jobs, you can't change. It's hard to change companies.

There's only so many tasks you're able to do. It's very, very, very general. And I think that's something that makes it different than other technologies in thepast.

Cardiff: The possibility that it will be doing cognitive tasks that in the past possibly were not automatable, you know, and by. Way, not just routine, uh, cognitive tasks as, as you were describing earlier, if it gets into the non-routine realm, you know, if it gets beyond just, I don't know, fixing contracts or giving you kind of bog standard generic pros, um, or doing, you know, rope customer service type stuff.

If it gets into the more kind of creative space, you know, the, the space that involves connecting ideas together, um, making decisions really well informed, smart decisions, then we're really talking about something that can do tasks that in the past just we would've, would've thought were kind of, uh, you know, kind of impossible to, to take over,

Ben: Yeah, but I think the magic is to think of it as a copilot or as a pair. An AI plus a really comant person is worth 10 people. Not to think of it aspurely automated. Like, it, it, it, you know, Cardiff, you know, if you were a maniac and you wanted to produce like 10 podcasts a week, you might be able to do that with ai.

It would've been impossible before.

Cardiff: Wouldn't unleash that on the world for anything you.

Ben: But, so that's, but then in software, you know, you can, you can write more software, you can do more customer service. You can, you can like, uh, there's all this analysis that's possible. Um, and so it's, I think that's the, the way to think about it is to think of it is how many, how many tokens does it take to, to, to automate 10% of your job, 20% of your job, 50% of your job, you know, and then you can get to like 90% like we did with customer service.

But I don't, I think that you have these, you know, this, this diminishing rapidly diminishing returns and you don't need to get to a hundred percent for it to be extremely effective. And so just to use 50% as I think is a fairly good number, a hundred million people, 50% automated, that's 50 million AIworkers or an impact of, of 50 million new in the United States alone.

Um, so I think that, so I think what I wanna move to this next thing is, so if you say, okay, I think that's true, I think that AI is gonna be, uh, very, very much a general purpose technology. What's the economic impact? Because I think that like, for whatever reason, it's so obvious and yet most people are focused on the bubble question or their job.

Is it going to get rid of their job? But not what does it mean for interest rates? What's it mean for growth? What's it mean for politics? I mean, there's some really, really obvious things that I feel like are just not talked about much.

Cardiff: Okay. Do you wanna, do you wanna shift into that, uh, into that phase of the chat?

Ben: Do you think I convinced you that it's like, uh, you're not, maybe not fully, but more than before? Maybe you're already there.

Cardiff: let, let, let me, let me put it this way. Like with a task like this, the question is, will this technology transform the economy not just from where it is now, but at a faster pace than priortechnologies had transformed it? I think there's a very, very strong possibility that it will, right. I just, I can't, this is one of those things where like, I, I have the luxury, as I've told you this before, of just being uncertain, of just waiting to see what happens.

But I, I think it would be so silly to dismiss the possibility that just two or three years after we learned that we had the ability to essentially talk to our freaking computers and it would answer in a way that was totally unrecognizable, uh, that this won't have some pretty impressive, uh, effects across the economy.

I think there's a really good chance of that. So, uh, lemme put it this way. I'm very impressed. I will say, with all the different ways that. That you and Fundrise use the technology. How quickly will other companies follow in your footsteps, right? Um, will they learn about all these different ways of using this technology?

Will they also make similar decisions of, Hey, we're producing more than ever because we can use this with the same number of people. So this is great from our standpoint, we don'tneed to increase, you know, the, the total number of people that we employ to keep, you know, to keep doing really well. Um, that I find very convincing.

I find that. Terrific, frankly. And I, and I think the more of this kind of experimentation, the closer we'll get to learning exactly how transformative it will be. But I, I think it would be crazy to dismiss that possibility outright. And a lot of people still do, a lot of people essentially say that this is all just an LLM situation and that Well, yeah, but they'll always just take the aggregate of the knowledge that's already out there.

Um, I disagree with that. I think that there, there is the prospect for this technology to be used in some new, interesting, um, and, and quite profound ways, and I'm very excited about it. By the way, I have maybe less fear than you do that this will suddenly lead to a big, sharp reduction in the number of people employed.

Right. A call across the

Ben: Well, let's,

Cardiff: I'm a little more optimistic there, but let's, let's go

Ben: I'm notsaying that I have a different, I have a slightly different take on that, but I, for the people who are skeptical. You haven't seen it firsthand. It kind of reminds me of COVID until, you know, somebody who died. It not, it's not real to you. I mean, it's, it's like that's how a lot of people are. And then there's people who are early adopters who, who don't, who can get there before it's like, so obvious that it's too late.

So you wanna be, you wanna be an early adopter of things that are really significant. You don't, you don't wanna be late to the party. So, uh, okay. The impact on the, on the economy. So as you said, there are, there are still some really big open questions. I absolutely agree. I think the biggest one is the adoption curve.

How long does it take to change people? People are the, are the bottleneck, uh, people, organizations, governments, all these things are slow to adopt. There's an open question of, of how much it AI progresses. It's still progressing, it's getting better. Andum, and there's a little bit of a question of like, how, how do they build 50 50, you know, megawatts, sorry, 50 gigawatts, you know, 30 gigawatts or, or, or way too much.

'cause they could do too much as, as, I'm not saying that's not possible, but okay, as you move on, they're gonna, this all happens to some extent. So I think before we talk about the impact, I think we gotta agree on the baseline because I feel like people forget what the status quo is and they, and they get lost in the sort of theory of it.

And I, I'm gonna argue that what's, what AI is going to do is in part already what's happening. And I know you already know this, but let me just put some of the stats on, on here. 'cause I think people. Uh, who aren't economic nerds like you and me may not track the labor share of GDP and things like that.

Cardiff: Hey, this is all important. This all matters. Yeah. Let's, let's, uh, let's give that to the

Ben: Okay? So if you, uh, want to go check this, you can go look at the, uh, uh, uh, federalReserve. Uh, St. Louis has this thing called Fred, and you can go look at Google labor share of, of GDP or labor share of income, and so labor versus capital as a share of the total amount of income or, or, or, or GDP of the country.

And it was pretty much at around 62% from 1950 to year 2000. It's, it was pretty stable. As labor's share of the GDP until about 2000? You could, you could, you can sort of say, well, maybe it was early nineties, but it really looks like technology has changed. I think it's technology. We don't know exactly what the other, other reasons might be, but, but, uh, labor's share of income has gone from about 62, 60 3% to about 58%.

So labor has a smaller share of income and it's been trending down for the last 25 years. Right.

Cardiff: that the leading sort of contenders for what could be behind it. Our technology is one for sure. Automation is one.Um, the, uh, the rise of globalization, possibly the China shock, things of that nature may have played a role. Simple methodological, you know, issues with it. The rise of income inequality, okay.

Which appears to have actually either, either plateaued or possibly even reversed a little bit over the course of the last five or 10 years, a conversation for another time. But that could be, that could have played part of the role, um, earlier, but of course that gets back into what caused that. So, yeah, I, I think technology and, and, and trade are the two big, um, are the two big contenders for what could be explaining that.

Ben: Okay. Another, uh, uh, stat that you can also go look at is, is corporate profits as a percentage share of GDP. So how much of GDP is captured by corporate profits? So it was about 5% of GDP from 1940 to year 2000 and corporates corporate profits is a share of GDP has been rising since 2000and has more than doubled in the last 25 years.

So went from 5% to about 11, 11.5% this year. And so, um, I think that is clearly about technology. You know, it takes less workers and less and, and there's, you know, things are capital light, more efficient, you know, more automations, lots of things driving that. But having corporate profits be a greater share of GDP is a really interesting and I think very core correlated phenomenon to their labor share of GDPI.

Cardiff: sort of, uh, the inverse, inverse, you know, concept. Right? Not perfectly, but, but something like that. Yeah.

Ben: Yeah. But it's, it's, okay. So, um, one other, just see when I talk about the, those, um, those trends, I think you think of those trends as linear, but actually if you could look at those, both, both of those data sets, you see that it stair stepped. So you see, you see thatthere is, um, usually every recession, every downturn year 2000, 2008, 2020, there's a, there's a shift and there's a shift to a lower percentage of, of, um, labor as a share of GDP or higher corporate profits.

So it's stair steps around, even stair steps around the early nineties. So there's like these, like, um, what used to happen before 2000 is there'd be a stair step down and then it would recover. And so there'd be, you know, labor share of GDP would decline. 'cause obviously people are getting laid off in a recession and they would recover back up.

Or corporate profits as a share of GDP would decline during a recession, then come back or, or actually improves as a share and come back up. But what's happened since 2000 is that after a recession, it, it plateaued a new low in 2000 2008. 2000 and 2020. And so that, what I, what I think ishappening is that there's like, um, during these downturns companies, organizations are forced to change and they adopt things like telehealth.

We saw during COVID, they adopt these things all at once during a recession where the technology's been there, but they really didn't have a necessity to adopt these efficiency driving technologies.

Cardiff: Yeah. I should note that a similar kind of trend is seen in manufacturing employment where there's a kind of shallow recovery when the economy's doing well, then there's like a massive steep collapse during a recession and then again a shallow recovery that never quite gets back to the prior peak. And the thinking there is exactly what you just said, which is that very often a downturn will force companies to do a rethink or a lot of companies that maybe wanted to put in place automation technologies but didn't wanna like piss off their workforces.

Right. Then have what you might refer to as an excuse to doexactly. That, that, hey, we're in a recession. We need to let some people go. Everybody else is doing it. So there's the cover of like the social proof of, of letting people go. And so they make the changes then that perhaps they had wanted to make earlier.

'cause frankly, it's not, it's not easy for companies, you know, or for, I should say, individual managers, I guess, to let, to let people go. And it can be a pr problem for the companies. Like it's very often they wanna make these changes, but the triggering effect ends up being a recession. I would add one more point because I think it, it is, it is intertwined with every, everything you just described, which is that ever since the years two thousands, the recoveries from the downturns have been quite unimpressive from a labor market standpoint, with the exception of the most recent one after COVID.

So after the 2000 bubble popped, the recovery was famously known as the jobless recovery. And that only lasted until we got to the financial crisis, which was a massive catastrophic. You know, problem for the labor market. And the labor market didn't really get tightagain until roughly 20 18, 20 19, and then bam ran into COVID.

So you're talking about multiple decades where the labor market was not tight, it was just a really bad negotiating position for workers all throughout that period. That's a very long time for that to be the case. Um, and I, I think that played a role in all of the trends you're describing. Certainly the decline in labor share, um, you know, the gain in the corporate share and so forth.

Ben: Right. And so, because it, it's, it's, there's these episodes episodic or these stair steps and we, and the way you described it, right, where people, where companies and organizations make these changes 'cause they have to, or the permission to, that's why I think it's mostly technology. It's not all technology, but it's mostly technology in the broadest sense of the term.

Technology gives me adoptions of changing in process and things like that. So, and so you already have this baseline trend of labor, share of income,declining corporate share of profits, corporate profits share of, uh, of GDP increasing. And I think that trend not only continues because of all, all sorts of technology, not just ai.

I think that AI. Well, virtually guarantees it continues and then I think it accelerates that trend. And so I wanted to put some, this is a classic, you know, us doing scenario planning, but just take that trend and extrapolate it out and then take that trend and just magnify it. And I think that it's clear to me that it's, that trend is like not about to reverse, like what we're talking about automating white collar workers.

The way that we automated blue collar workers is like a, a, uh, magnification of the trend. So, so let's just do the trend continues versus trend accelerates, right? So we're at 58%, or let's do corporate profits. 'cause corporate profits means big stock market, right?Stock market. So, okay, so corporate profits are 11.5%.

It basically doubled. In about 25 years. So you do that again, corporate profits are 15, 16% by 2035. So you have this sort of, you know, increase of another few percent. And same thing with labor share profits. So you have this sort of decline and then you say, okay, what if that actually is even more, and corporate share, corporate profits go to 17, 18%.

Labor share of income goes to 50 52%. And so you can, it's some, I think that like, unless AI is somehow some kind of weird mirage, which I don't think it is, 'cause I've lived it, like I think we, I think the economic impact of AI is some kind of band within that trend. And I don't, and anyway, you, and, and this is why I think when you look at the impact of ai, when you go down the list, it's, it'sonly a question of magnitude.

It's not a question of direction.

Cardiff: I would also say it, it is also a question of how policy responds. Yeah, that's a big deal. And I'm talking primarily about how the Fed responds to the trends that emerge. And, you know, depending on the size of the shock, which in, in the scenario you just described is a massive, a massive, uh, new trend in the economy, I think it would accurately be characterized as a shock.

Do fiscal policy makers also respond? Keeping in mind that all of the added production that we would end up getting from ai, like could potentially be very good. Again, depending on the policy response for the fiscal coffers, it could be very good for, you know, um, it could be very good for the, the sort of.

Maneuverability that the government has for how to respond. You know, like all of these things could be actually quite healthy, you know, if, if the economy becomes way more efficient because of ai, and the Fed doesn't rush to likego back to a super tight labor market because of the growth of other sectors, whatever, what you're talking about is a huge disinflationary trend, right?

Which gives policymakers a lot of room to maneuver at that point. And so it also, by the way, it'll just naturally lead to, again, in the absence of something else happening, like lower interest rates, because that's what's happens in disinflation, which, you know, again, more maneuverability for policymakers because this is a favorable technological shock.

And I'm not talking about workers right now. I'm, I'm talking primarily about added productivity growth. How it affects workers, again, does actually depend a lot on how the economy adjusts and how policy responds. You see what I mean?

Ben: Yeah, for sure. I feel like, again, everybody's obsessed with this question of how it affects jobs, especially their job, and they're missing this big picture. So let's put, let's put this question of how it affects workers aside for a minute and come back to that last, because I think that's, again, that's the most contingent.

But some of these things I think are much less contingent, much more likely. So I think that the, you said it, but technology and in particular, AI is highly disinflationary and I, I believe that people are sleeping on this 'cause everybody's obsessed with inflation looking backwards, looking at COVID, looking at all the money printed.

We're about to put trillions of dollars into an extremely deflationary technology and I think it's going to lower CPI, especially CPI run services. So wage growth and, and, and, um, and the general things that have been driving, uh, inflation, I think are gonna get muted. And that's, I think gonna create a lot of it said room for policy.

So I think that's, I think that's, there's no question about that. And it's just a question of magnitude. And the other ones just go down the list. Uh, GDP growth clearly very positive for, for growth. It's just aquestion of magnitude. Um, corporate profits, stock market, it's very good for that. It's just a question of magnitude before I get to unemployment.

Let's talk about, um, treasury rates and fiscal debts. I think that, um, I think it lowers the 10 year treasury 'cause lower inflation, lower inflation means lower treasury and I think it drives more taxes back to the, uh, government 'cause it's tons of growth. And so it's good for deficits. So I think it's across everything but labor.

It's very, very positive.

Cardiff: Yeah, by the way, there, there was, there is another sense in which that analogy to the nineties, uh, can be applied here, which was the nineties. Were also a time of, of rapid productivity growth. And so let's say we get back to that place and there was a moment where Alan Greenspan, you know, uh, essentially was being pressured to raise interest rates at one point, but he could see the productivity enhancements that were being made and he kind of said, Nope,we can leave interest rates where they are.

I don't have. To raise rates here, he could keep rates low and see that essentially we would not get a big boost to inflation because productivity growth was in fact as impressive as he had anticipated it to be. And if something like that happens again, this could affect how the Fed responds. And I would say then it would be fine for workers effectively if policy responds the right way.

Okay. That you're right. That is contingent. But it's worth noting that the nineties were a time of tremendously low unemployment, even though all these productivity, um, enhancements were being made. And even though that was a time of astonishing innovation and implementation of things like that could happen again, you know, and budget surpluses.

So it's all connected.

Ben: Yeah. And the, and the, and the response to the technology potential in the nineties was a huge investment. Everybody saw, saw it, got excited and invested in it. And, and that investment ended up paying off. So, and that's what's happening again. And just afew other sort of second order consequences that drive interest rates down even more is all that excess profits, all that excess, uh, um, um, returns ends up as capital abundance and capital Abundance means lower rates.

Cardiff: Famously back into the, the sort of chase for yield, that kind of thing,

Ben: Yeah. More money means less, less return is the supply demand. And so, so I think it's, there's a very, you know, for us to, who we also do real estate, I think there's a very positive story for real estate. 'cause rates are gonna come down as a result of ai, uh, growth will be higher. And I think that the last question is, what does it mean for labor?

What does it mean for policy? And I wanted to do a, um, look at a single industry and talk about how it could, that industry could end up succeeding with good policy or failing. Because I think it's interesting 'cause I think a lot of people's initial response to this is alittle bit of fear and dislike.

'cause they don't, they don't like the idea of having AI take a lot of their job. That's how they, I think this emotional response when what I think is actually there's, for most, most people, it's very positive for them. But let's do, let me just do how it goes bad as a policy and how it goes well. And I think it's easiest to see that in the healthcare industry.

So lemme just do healthcare and then I wanna hear your thoughts. So healthcare, there are 20 million people work in healthcare. It's cost $4.5 trillion a year. And if you go look at the work, there's huge part of healthcare is administrative costs. Everybody knows that like documentation and the billing and the claims and the nonsense is a big problem in healthcare.

Of the 20 million workers, something like five or 6 million of them are, are in the administrative part of it. And only 7 million are doctors and nurses and tech tech, you know, the, thetechs. And so there are millions of healthcare workers that do administrative work and people find healthcare inflation infuriating and they don't like the healthcare process.

And so you could see a world. I mean absolutely like clear world where AI starts to automate a lot of the billing and the doctors taking notes and the claims, and all of a sudden doctors have more time spent with patients and the cost of healthcare comes down and okay, there are people in the healthcare industry who end up being displaced.

You know, maybe I could easily imagine 10%. That's a couple million people get displaced. But for the rest of society, having the cost of healthcare and the practice of healthcare get way improved, 'cause of AI is a massive benefit.

Cardiff: I have two responses on healthcare 'cause I think a lot of what you said is dead on, but healthcare is a tricky one, right? So first it is absolutely thecase that the healthcare sector could use a ton of efficiency enhancements. So there's just absolutely no question about it. They also have a problem with like occupational licensing issues where nurses aren't allowed to do things that they could do because doctors want that turf for themselves.

Stuff like that. Efficiency enhancements are definitely a thing that could happen there, but the government's very intertwined with the healthcare sector. There are unbelievably arcane rules that still govern what can be done there. So even if the technology to help healthcare is invented, exists already, et cetera.

What you need to sort, there is like the human barriers, the human made barriers, you know, the, the, the laws, the rules, the regulations and so forth. And the healthcare sector is a tough one to get that right. Even though I think the technology still will break through in at least some places. Right? Second, healthcare is also kind of tough because it is subject to what economists refer to as the bamal effect, which is that a lot of healthcare workers, including doctors are bydefinition supposed to be low productivity in the sense that a lot of us like high touch doctors.

We like being able to have the doctor take their time with us, you know, give us really good service, personalized service. You can't always accelerate that, which means that if it remains a lower productivity sector than other sectors, then wages are gonna go higher because you have to incentivize people to go work in the sector to keep up with the wages in the, in the higher productivity sectors, the non-healthcare sectors.

So you're gonna have to. Pay higher wages and higher wages means higher costs for consumers, right? Higher nurses, higher wages for nurses and doctors means higher costs for consumers. So healthcare is tough for those two big reasons, but I also still think you're right. I still think that the introduction of miraculous new technologies, if we get them, can still make a big difference, even in spite of those barriers.

I just think those barriers are big ones,

Ben: Yeah. Yeah. But I, I feel like when I talk about AI causing, uh, uh,job displacement, all sorts of problems, and, and people respond negatively, but then I say healthcare, they're like, oh, that'd be great. I'd love to see the cost of healthcare from down. I mean, I think you could probably save half a trillion and trillion dollars a year.

10, 20% is totally realistic with ai. So much of the, of the work and the information and the, and the processing is very, very conducive to AI gets very routine repetitive. Um, and the economics for AI are high. And so you can sort of, when you take it outta the theory and make it practical and people say, yeah, hell's yeah, I want, I want healthcare to come down, you know, 'cause of AI and Okay, displaces some people, but clearly that's, it's a net benefit for society.

So now all of a sudden, now what? Now people get why, oh, this is really good. And then as you said, second they get why the policy gets in the way.And you could have kind of the worst of all worlds, which is AI plus inflation. And that this is my, this is my like, um, you know, scenario planning where you have like continue the trend.

It gets better, it gets worse. And the worse is that the costs fall because AI makes things more efficient. But that doesn't mean that the falling costs will get passed through to the consumer. It doesn't mean prices fall because of, you know, the, because of the policy response. And the policy response is literally like people not liking ai.

And, and healthcare industry's easier for me to imagine healthcare industry saying, oh, we can't have AI 'cause we gotta protect the patients and only doctors can do this work. And they clog up the change to protect their job. And then you end up with AI like, um, saving a lot of, a lot of like, money for the people, the, the corporations who own, you know, the hospitals or the insurance companies, but doesn't actually drive lower costs to theconsumer.

So it's like kind of a worst of all worlds because of bad policy. It's kind of like the, um, what's happening with the dock workers, you know, or the, or the, where you're not allowed to have driverless trucks or driverless cars 'cause it's gonna displace jobs. Like it's preventing the technological change is sort of like, you get technological change anyways, but it ends up being really inefficient and it's like kinda the worst of all worlds.

Cardiff: by the way, those are like, you know, driverless vehicles. Like that's, that's machine learning technology, right? Like that is, so we're, we're talking about the, basically the same thing here. And that's a great point though, because we do have to get to the idea that there's gonna be some societal and perhaps policy pushback on these changes and that may have already started and that that really could slow the diffusion of these technologies.

And there's something in particular that I'm worried about, which is that a lot of the jobs that at least are gonna be perceived to be at risk and might end up being at risk are the jobs that are held now by. You know, very well educated,influential types people that went to all the fancy schools, and they're friends with the politicians and they're friends with the corporate CEOs and whatnot, and they're the ones who might end up driving the pushback.

And these are people that have a lot of power, you know, societally, like they, they actually might succeed in elevating new barriers to the adoption of these technologies. So that is another, that is another potential outcome here, which is the technology evolves and it, and it gets better and better, but the actual ability of the economy of society to diffuse the technology to start using it, um.

Might be slowed by these kind of artificial barriers that have not yet been erected, but might if the politics goes far enough. And you're already seeing it, by the way, you're already seeing it. I think it was Bernie Sanders that just said he wanted to put some kind of a moratorium on the build out of data centers.

I can't remember the exact details of the idea. Um, but like the, it's happening, right? It's starting and it might, it might getplaces. And that that will also be something that stops these technologies from, from, you know,

Ben: Not stop. This is, this is, this is the part that it's, no, it's in, on, uh, the impact ends up being highly uneven. And so it ends up certain sectors or people are protected and certain sectors are not. And I, and that's like, I think that's very similar to the kinds of problems we have today. And what's caused is a lot of, uh, populous rage and, and distrust of experts.

And so I'm, I'm worried that, um, not that AI is not real and not gonna have massive gains. I think those two things are, are near certainty. I'm worried that the policy, you know, the, the political, the cultural response to AI is gonna be so negative that we get kind of more of thesame of what's been happening with how, let's talk about three scenarios.

Win-win, win, lose, lose, lose. Right. So the win-lose we're talking about is like what happened to blue collar workers and you know, we just, we let the ai uh, automated factories, devastated towns, people in the blue cities didn't care, you know, like oi, opioid addiction, all these things happen because an uneven effect and 'cause of politics and policy response was bad.

And, and said, I think, I think another one that is, is not as clear, but similar to social media, had a very uneven effect and, um, clearly like social media as a technological breakthrough. But it, I think our policy around it has been captive of the, um, big tech and we've had trouble really regulating it.

And I think we, and as a result, I think people, a lot of people arevery angry about it. And they, and it has affected people's view of big tech generally. And so those are to me, kind of examples of win-lose, win-wins, as you said of the internet. You know, the initial internet in 1990s. Quite cloud technology things where it's just mostly win-win.

I.

Cardiff: Low unemployment, by the way, is part, is a crucial part of this, and generally widespread prosperity, which I'm glad you brought up that example of what happened to a lot of blue collar towns in the aftermath of the internet bubble and also in the aftermath of like the trade shock with China. You know, the, the thinking there and the, the best research sort of shows that overall these trends were good for the US economy writ large, unfortunately.

The burden of adjustment fell disproportionately on a lot of specific towns, which did not recover the way that economists had sort of expected them to. Like those local labor markets didn't adapt. You had a ton of places that simply never got better, never got better. A lot of, a lot of, uh,workers essentially were, you know, out of luck for a very long time unless they were willing to move.

But then you're wrenching people away from their homes and they did not have too much like political and societal and economic power to respond. And the response in that case would not have been, I think, to say no more technology. No more trade with the rest of the world. The correct response is how do we help out these places that have lost out on this trend?

How do we spread the pro the prosperity? And we didn't. And by the way, there are also other historical examples of big productivity shocks, positive productivity shocks where the effects were deeply uneven. Right? Go back to the 19th century. This happened a lot actually. It took a very long time after the onset of the industrial revolution itself, before workers themselves were no longer, um, you know, falling behind the owners of capital.

So like that's definitely something to pay attention to. I think it's something that you're, uh, that you're pointing to in your scenario where, you know,the, the sort of corporate share goes way up. Um, or, you know, or the labor share goes down. That is the thing to worry about, that there's a lot of people that are gonna miss out on this.

A lot of people who are gonna really suffer if you let it continue for too long, which is why the policy response matters so much.

Ben: Fundamental technology that is ai is it, it turns capital into labor. So shifts power and economics from labor to capital, which is literally a token or a data center. And, uh, and I don't think we have any idea what the policy response ought to be. And instead of spending all our time talking about whether or not, you know, it's a bubble or whether or not, uh, it's gonna happen, I think we need to move beyond that question and move to like, okay, it's gonna happen, or at least we better be prepared for it to

Cardiff: At least be prepared.

Ben: At least be prepared. I'm, I feel like that's like, uh. You know, it's, I think that's like, I mean, you know, whatever, our politicians are in their eighties, so they're gonna be slow to slow tosee the future, but, um,

Cardiff: won't be there forever,

Ben: yeah, but so what, so 'cause it's like, the parts, I think the problem with your, your, your point about this, um, no unemployment is that with massive wealth inequality with a strong stock market and, and you know, people losing a job that they really love to go work at Walmart.

You know, these are just, these are not good social solutions, even if they're good economic solutions in theory. And I think our country doesn't have any ideas of what to do about this problem. And it's gonna happen. I think it's gonna, the automation of manufacturing, everybody talks about it. You and I have just, you just mentioned it, you know, we went from like.

I think something like 16 million manufactured workers, manufacturing workers to 12 million manufacturing workers over like 12, 20 years or something. Like it was a decline. I think there's like 4 million less manufacturer, uh, workers in the manufacturing industry over that long period from the China ShockAutomation.

And I'm talking about 'cause of ai, like tens of millions of workers in a much shorter period. So this is like, this is what we need to figure out. And I'm just, I just wish everybody would stop talking about how much money they're gonna make or lose in the stock market on this. And actually talking about like, the effect on hundreds of millions of American workers and globally.

Globally. I mean, so that's, that's the, that's what I wanted to, to make this episode about. Hopefully it's

Cardiff: Do you have, uh, this is, this is maybe the for, for a future episode. Do you have a preferred suite of policies that you would wanna push for, uh, to respond to a scenario like that? If it, if it turns out that way?

Ben: I think I have a spectrum of possible positives. I think one is that AI U, sorry, the US is way ahead of the world on ai and I think I can imagine a world where the US is producing more AI workersthan anybody and we're exporting them. And so like France and Germany and India and Brazil actually using American AI workers speaking their language, like I think AI worker export of American AI workers is like a really interesting new phenomenon.

And I also wanna point out the fears that China produces hundreds of millions of AI workers and, and everybody uses China's AI workers. There's

Cardiff: to be clear for the audience. When you say AI workers, you mean literally non-human workers? You're not talking about workers who, you're not talking about human workers who work in the AI space. You're talking about workers who now exist as ais. Right. These are not, these are not human beings you're talking about.

Ben: right. I mean, I, and I think in practice it'll look a lot more like big tech people using Uber or using Netflix. You know, they use Netflix. They don't use Chinese, Netflix, and, but it's just, it's in, so it's an application that produces a job like anAI lawyer. And, and so I think America could end up being a huge exporter of, of ai, and that could be really positive for America.

Uh, that's a positive. Another positive is that the number of workers per year in America is declining. More people are retiring more. Baby boomers are retiring every year. Uh, I think approximately 5 million baby boomer baby boomers retire a year, and we only add two to 3 million workers, you know, labor, you know, young people entering the workforce every year.

So there's a, there's actually a labor shortage in the, at the present, we're losing a couple million, one to 2 million, um, workers a year just on a pure like, flow of,

Cardiff: We're also cracking down on immigration, which does not help with

Ben: right? So we actually, maybe we need more workers, maybe AI workers actually like, is not gonna take away jobs. Actually gonna, like gonna add, I think as you said,like people are afraid of it.

I'm seeing at the company, I do think that like, um, if you ask the machine, which is what we call Chad, GBT, and Claude, if you ask the machine what's gonna happen, uh, they, it says, um, that actually doesn't cause a lot of unemployment. Unemployment goes up as a result of, you know, this phenomenon. But actually it suppresses new hiring.

So instead of, instead of hiring 20 million people over the next decade, you maybe only hire like 1 million. And so it's much more about suppressing new hiring than it is about, um, unemployment and layoffs, which is what we're already seeing in the economy. It, I don't think is totally related, but I think that what we're seeing in the economy today is what we're gonna see as the new normal for the next like five, 10 years.

Cardiff: Yeah, I mean, I, I would. Definitely caution against assuming that what's happening in the economy right now is necessarily down to ai. I think, uh, I think a lot of that just has to do withbasic economic cycle swings, right? Like we've seen the unemployment rate kind of tick up over the course of the last year.

You know, we've also just introduced a lot of shocks to the economy this year, the crackdown on immigration, the trade wars. Um, these things are all intertwined and a little bit tough to, a little bit tough to disentangle. Um, we're, we're firmly in the realm of like sort of speculating into the near, into the near future, near to medium term future, rather than what's

Ben: for sure. But if, if AI causes us to, um, not hire as much, then actually I think the biggest impact will be on young people. The biggest impact of AI is that if you're me. I'm gonna see lots of gains. But if you're, you know, my son and you're graduating from college, you're gonna struggle 'cause you're just not, companies aren't hiring as much 'cause AI does so much of the work.

And so the easiest people to retrain are young people. And so I think that there's like a, an opportunity and say, okay, well what, what would, if I'm, if I'm talking toSam Altman from OpenAI or, or the DIO from philanthropic, I'm like, you guys should start like AI schools and people should go to AI school and, 'cause most companies are bad at adoption.

They don't know how to do it. And they would hire somebody who would help them adopt ai. And I think, I think I could see millions of people going to AI school and millions of those people getting hired and having a huge positive effect for the AI companies, for the society, for culture, for, for companies.

And um, and I think that that's like the kind of way we turn this from like. A, a, a place of fear to a place of, of, uh, positivity

Cardiff: Opportunity. Yeah. The, the, the point you're making there is that the tasks that matter to the economy may change, which is a very different story from mass Disp employment, right? Mass, you know, all suddenly tons of people can get jobs, that kind of thing. If people can see it coming than they can, you know, adapt forwhatever, new economic sectors might become more important and emerge out of this.

Um, and. In terms of how to actually use the new technologies that are coming, start training in advance for those as well. How quickly an adjustment like that can happen is tough. If people have been, you know, training to do something all their lives. If, if suddenly 70% of lawyers are unemployable, that's tough.

If you're in your third year of law school and you have a ton of student debt and you were hoping, uh, that like a big corporate law job would sort of help you pay it down and so forth, you're gonna be in a tough place, right? But over time, that kind of adjustment. Can't happen where then you realize, okay, well, like going to law school might be a bad idea.

Or law schools themselves end up teaching you something that's actually valuable. Um, or, you know, the, the amount of debt you're willing to take for it. Uh, changes. Like people, people will get new information from which they can make new decisions about what they do. We have to be careful not to assume that everything stays sort of fixed and it crashes up against this new reality.

Youknow, adaptability is a big part of the way out of this mess. If, if it, if it does, uh,

Ben: Right. That's, but that's my positive take is that, um, unlike, you know, somebody who's, you know, in their forties and they didn't, they were coal mining their whole life. It's really hard to retrain that person. A, a white collar worker in their twenties is probably the most adaptive employee in the, in the world.

And so I, I think that who gets impacted the most and who's the most willing to learn new skills, change industries, change jobs, drive the future. I think that could come together in a, in a surprising win-win.

Cardiff: What about forties era podcasters? Forties age podcasters? Their

Ben: it. Forget it.

Cardiff: Their toast? I'll wait for the universal basic income and that'll be it.

Ben: Well, iff, this was fabulous. I don't know if you, if you have anything to add to this,

Cardiff: No, I, I would say this, that, you know, the, the, we've covered quite a spectrum here and what I like aboutit is that if you position people who work in AI at one extreme end of the spectrum, I think a lot of 'em are even more extreme than you are. They, they're people who are like, this is gonna lead to absolute economic catastrophe for the vast majority of workers who are gonna be replaced.

And we need to put in place some kind of a, like a universal basic income to set them up. And then you take a lot of economists on the other end where they very often kind of use models that assume the technology's fixed, and they're trying to go from there because they don't want to enter into the, um, they don't want to enter into the realm of like.

Trying to speculate on like how the technology will, will change over time. Um, and so they maybe are under appreciating the dynamism in the sector, how it can change, how the technologies can get better. We've kind of covered a lot of what's in between those two places. Right. And maybe I'm a little bit closer to the, let's wait and see.

It may not be that bad and you're closer to the, this is gonna be enormous, right? But we've,I think we've, we've covered the full suite of the spectrum and I would just encourage everybody to just read as much as they can about this. Like, listen to this episode, but go seek out sources that you disagree with on this.

You know, like I, I'm certainly, uh, compelled by a lot of what you said, Ben, to like look for the more, I guess, optimistic scenarios from the standpoint of the technology, you know, but people who are really like heady about it and think it's gonna be enormous, should also read some of the pessimistic takes that are out there and really try to inform themselves deeply.

'cause one way or the other, I think we will still be talking about this a lot. Over the next few years. I don't think that it's just like gonna go away no matter what happens. There's no scenario where this doesn't become important. Either it becomes the thing that you just said or it becomes a disappointment and we have other issues to deal with.

Right. So like in keep yourself informed on this so that at least you know what's going on. It is, it is. I think the most important thing to keep in, to keep track of over the course of the next couple ofyears in the US economy. Doesn't mean it's the only thing, but I would consider it to be the topic of the moment, maybe of the whole decade is what we'll see.

Ben: Yeah, sorry, you, you triggered me on one thing, which is that if it's successful, it's deflationary and if AI collapses, it's deflationary

Cardiff: that's possibly right.

Ben: because if

Cardiff: Not necessarily.

Ben: if the AI bubble collapses, the only thing driving growth in America today.

Cardiff: Yeah. It's disinflationary though for number one, assuming no policy response that drives inflation back up, which I think, you know, we could expect to happen, but it's also disinflationary for two different reasons and not all reasons for disinflation are good, right? There's the no demand in the economy.

Everybody's out of a job, nobody has money to spend, and therefore there's disinflation and then there's the, oh my God. Companies are getting so much better at creating an abundance of things that we can all enjoy, that it's disinflationary. I think clearly there's one world where that disinflation is better,but I do wanna be clear, like it doesn't mean we'll get disinflation.

The policy response drives so much of this,

Ben: Yeah, that's, this is a, just to roll it up here or sum it up since we're gone. Longer than normal. Yeah. The two extremes of it's doomsday or it's like, uh, it's a normal thing like a cloud and it's just gonna diffuse, is I think it's not serving us. I think it's somewhere in the middle. And, and trying to embrace it as an individual, being an an adopter of it, using it, you can become an expert compared to everyone else pretty fast.

And as a society, I think is, if we adopt it, it's, it's the future and it gives us so much opportunity for abundance. And I feel like generally people are scared of it or negative. It's not politically popular. And, and, and this sort of like constructive future is, is just not how people are talking about it or enough people are talking about it.

Cardiff: Well, for sure. This is, uh, this is atopic where people have very strong points of view, so if you disagree with anything said here, definitely yell at Ben about it.

Ben: Yes, please do. I'm gonna disagree with you. So, Ari Cardiff,

Cardiff: All right. Thanks Ben. This was, this was, this was fun. You have been listening to Onward, the Fundrise podcast, featuring Ben Miller, CEO of Fundrise. I'm Cardiff Garcia of the Economic Innovation Group and host of the New Bazaar podcast. We invite you again to please send your comments and questions to onward@fundrise.com.

And if you like what you heard, rate and review us on Apple Podcasts and be sure to follow us. Wherever you listen to podcasts. Finally, for more information on Fundrise sponsored investment products, including relevant legal disclaimers, please check out our show notes. Thanks so much for listening, and we'll see you next episode.