This message comes from Insparity, providing HR services and technology from payroll, benefits, and HR compliance to talent development. Learn more at insparity.com/hrmatters.
This is Planet Money from NPR.
Hey everyone, for the past couple weeks, the Planet Money team has been on the road on book tour. Tonight, live from New York, the Planet Money book launch.
I'm in San Francisco. What's going on?
So excited to be in Boston.
Hello everyone. Thank you.
We say hello Seattle. Is that a thing we get to do on book tour?
Hello.
We have been crisscrossing America doing live shows to help promote the new Planet Money book. We've been in New York, DC, Boston, San Francisco, Seattle, Los Angeles. And this is a rare thing that we don't get to do very often. We don't get to meet and talk with so many of you. In DC, we played this game with the audience to demonstrate a concept known as the wisdom of the crowds. This is the idea that if you ask a large group of people to guess the price of something like a piece of art, the average will actually be pretty close to correct.
You all guessed that this piece of art is worth Oh, come on, guys. $693 trillion. I don't think that's right. It wasn't a perfect study. In New York, our hosts Amanda Ranchic and Mary Childs answered audience questions about how we choose what stories to report or not report.
It was weird to look at the list of episodes that I thought I was going to make and one of them was just cocaine.
We also have a running thread where we just yell drugs at each other cuz we keep wanting to make a how does a drug become legal episode.
Yes. But the lawyers have said no. And at every stop, we have gotten the true delight of meeting so many of our listeners turned readers, some of whom it seems like maybe should be writing their own economics books.
My name is Ryan and I'm really glad you were talking about cargo shipping because I'd like to hear your thoughts on the Jones Act, which was briefly and for the long for those who aren't like maritime shipping nerds, I'll just contextualize briefly. The Jones Act is a subsection.
Yeah, actually, please do do our job for us. I love this.
Now, we do have a few stops left on this book tour, but obviously we can't make it to every city in America. Can't actually make it to most of them really. And so, we thought, not everyone can come to the book tour. Let's bring the book tour to everyone.
Hello and welcome to Planet Money.
Hello and welcome to Planet Money.
Hello and welcome to Planet Money.
Hello and welcome to Planet Money.
Oh, wow. That's good.
Hello and welcome to Planet Money. I'm Erica Baris. In every city on the tour, we've been doing these interviews with special guests. They've been Planet Money regulars like economists Raj Shetty and Emily Auster and some people in the news like the co-founder of Anthropic, Jack Clark.
By April 2027, AI systems should be able to do tasks that might take a person 150 hours. No, that what is that? That's almost a month's worth of work, which it requires strange things to happen in the economy.
Yeah, but like what task is it doing that's going like what task are you thinking of?
What's the last bit of work that took you a month?
Writing these questions for you took more time than I wanted.
And also people like economist Daryl Farweather who's trying to help regular people understand the massively complex housing market.
So I think a good metaphor for the housing market is musical chairs. It's like the people who already own houses are sitting in their chairs. And whenever there's a new spring housing market, some people get up and they move and they select a new chair. And if you're not adding more chairs in, you can't get first-time home buyers into the market so easily. Today on the show, two stops from our live book tour, San Francisco and Seattle, where we'll talk about AI anxiety and how you, yes you, could be the key to fixing the housing crisis.
This message comes from Insparity, providing HR services and technology from payroll, benefits, and HR compliance to talent development. Learn more at insparity.com/hrmatters.
This message comes from Instacart. Everyone prefers things a certain way, like groceries. If you want groceries just how you like them, you got to try Instacart. They have a new preference picker that lets you pick how ripe or unripe you want your bananas. Shoppers can see your preferences upfront, helping guide their choices. Because when it comes to groceries, the details matter. Instacart, get groceries just how you like.
Every episode of NPR's It's Been a Minute podcast starts with a question about how culture shapes our lives. Are we spending too much on other people's weddings? Is social media bad for your mental health? We're here for your right to be curious, one big question at a time. Follow It's Been a Minute wherever you get your podcasts.
So, first stop, San Francisco, where we got to talk with the co-founder of Anthropic, Jack Clark. As you know, Anthropic and their AI models, Claude and now Claude Mythos, are raising all sorts of thorny questions about what AI can do and what it can't. And we were, of course, going to ask Jack about all of that. But we wanted to start with some writing. He published online several years before starting Anthropic, a series of short stories that give some perspective into how he thinks about the future. Here he is with Planet Money host Kenny Malone.
Is it fair to call these science fiction?
It is fair to call them science fiction.
I'd like to give people a little taste of what's in here. So, the one I want you to read that feels very Planet Money, it imagines in the year 2025, we'll forgive you cuz you're it was a little you're a little wrong on that one, but in the year 2025, you imagine a factory in China. It is overseen by a robot factory line and AI is overseeing the robots. And then AI is also trying to imagine new types of products. And then if you just want to read when new candidate products are found, it runs through a series of tuned expert systems. And if it gets a high enough score, simulates the product in a high-fidelity simulation. If it passes these tests, then a candidate product is produced, airlifted by drone to a nearby human focus group, and if it satisfies their criteria, is sold on an eBay-like auction site frequented by the factory's thousands of distributors. A bidding process takes place, and in a few days or weeks, data comes back about how the product succeeds in the market. If it does better than the existing product, more parts of the factory are dedicated to creating new products in this style, and the improvisation line begins exploring the latent space of the new product.
Okay, so we imagine a world where AI is able to both control the robots that make the product, innovate and design the products, ship them off to human focus groups, and then sell them to humans. So here is my question for you. Where are we the humans in this story? Are we off in a field dancing, eating dates in the sun? And maybe more importantly, how do we have money to buy the products that we didn't help make? Sort of a serious question.
That's a good it's a good question. So, I think I think if AI goes as far as people think, you actually need to reconceptualize how how capitalism in the largest possible sense works. I think if you end up in a world where you have a closed loop production system with just machine to machine to machine to machine and then people buy stuff, people need money.
That is the thing. Yep. So you need to tax the robots and AI companies significantly and you need to somehow find a way to reallocate money from this machine economy to the human economy.
Okay. Now, one of the things that occurred to me in reading this book is Can people get this by the way? Is this like a thing?
You can get it for free online at jackclark.net. But I haven't figured out how to sell the book yet. Each time I try and make an actual book to sell, I get sidetracked. And the last time was when I was starting Anthropic and then I got busy.
Surely someone will publish your book, Jack. I'm
It's on my list. It plagues me. I have a Google Doc that I stare at on my free moments.
Having gone through a series of reporting on on books and how they get made, I suspect you'd be an attractive candidate to a publisher. One of the things that did occur to me in reading your book is that science fiction writers by definition are imagining a future. They're predicting a future. But you may be genuinely the only one that I can think of who is simultaneously shaping the future. And we are all living in the world that your technology is creating. And so I do genuinely wonder if that changes how I should read this book. Like, should I think of your writing as a blueprint, as instructions, as a warning, as advice on how to live in Jack Clark's future?
Yes. There's there's a there's a band Jawbreaker, and they have a lyric which says, 'My fiction beats the hell out of my truth.' And so, the newsletter is mostly me writing about AI research papers, but the fiction stories are where I'm basically trying to grapple with what's happening and my moral and ethical responsibility in it. So you can read them and I view it as like messages in a bottle I'm trying to throw out of this semi-frightening AI lab which I'm I'm a principal character in.
Are you and so you are warning us and you're warning us. You're not warning us. I guess that's why you said yes.
It's very it's exciting. I think it's very exciting.
That's that's about the response I expected to be honest.
Eerie applause. Um, but it's change, right? And with that comes some amount of trepidation, right?
Yeah. Well, yes. For me, a lot of trepidation personally. That may just be me. I don't know. We talked earlier today. This was a very funny interaction. So, Jack, I met Jack briefly early today, and this is the greatest reason to run out of room. He's like, I must run and make some predictions. And I was like, what is happening? And you paused to briefly explain that you had made a series of predictions essentially that would come true in 2026 of September. and they had they had done very well and you had to go make another set. So what were your predictions?
Okay, there's two.
We see any of these on Polymarket though we know where they came from.
Well, one I think again if you do the time thing it doesn't seem wild that by April 2027 AI systems should be able to do tasks that might take a person 150 hours. No, that what is that? That's almost a month's worth of work which it requires strange things to happen in the economy.
Yeah, but like what task is it doing that's going like what task are you thinking of?
What's the last bit of work that took you a month?
Writing these questions for you took more time than I wanted.
Yeah. Yeah. So tasks like this.
Yeah. Um no, I mean tasks that might take a person a really really long time might be designing a complicated piece of circuitry, doing a very very involved research project involving reading and cross referencing many things, building a piece of software. The kinds of tasks which we currently pay extremely talented people large amounts of money to do are exactly the type of tasks which AI systems are kind of creeping into. So that has some implications.
The other prediction we made was you know we've talked about how the majority of the code at Anthropic is written by AI systems like Claude. Well all of us in the room thought by April maybe 100% of the code is which led to us say what are we doing? And I think that we invented a guild system where we would sit around analyzing and critiquing the code that Claude writes and verifying that it's correct. And I can't work out if this is like a really fun country club style job or if it paints some different picture.
I don't know, Jack. I find it all I waffle between horrified and optimistic. And I find I'm I'm just Is that Is that kind of where everyone else is?
I I find myself staring into the face of change that we haven't experienced in perhaps a century. And I I don't know how we'll respond. Um a 100 years ago, CS Lewis wrote a book called The Screwtape Letters, which is about a demon in hell whose job is writing letters to their boss, a senior demon, a manager demon, about their attempts to corrupt a person on Earth. And they have a letter where they say, 'Terra, wonderful news. My human has just got an invention called the telephone.' And the telephone means that whenever my person is alone and would otherwise introspect on themselves and be drawn closer to God, they will instead pick up the phone and call their friends, which brings them closer to hell. Now, what what's what does that get right and wrong? What it gets right is that things like phones actually massively changed how people relate to one another. Um, but I don't think that we're all like doomed to hell because we have phones.
Well, I realize now I picked the wrong example, but but it gets at the shape of this, right? Like, we are going to experience immense change and we're looking at it and we're trying to think through what it means for how we relate to us as people, and it will change how we relate to one another.
Yeah. Uh, I want to shift gears from your writing for a second because, uh, you made a lot of news lately. Yeah. I mean, you know, exciting. I mean, you've made a lot of news lately, but most recently, there's a new AI model called Mythos. Please correct me if I get any of this wrong. Uh, so Mythos, and that seems to be really good at hacking if someone wanted to use it for that. And so, you have not made this model public yet. You've given it to something like 40 or 50 companies to, I guess, gird themselves from the tool that you made.
Okay. So that's companies like Apple and Google and Amazon. And as far as we could tell, the only financial institution that I've seen reported is JP Morgan maybe has it to test. But uh I believe Treasury Secretary Scott Bessent has called you asking for this. So question one is how have these phone calls been? Like
I I have a high tolerance for weird phone calls at this point in time. Um, I think the nature of the conversations we're having is, uh, we have been saying for years that AI systems would get better and they would keep getting better until they got better than people at most tasks. And it started in coding last year.
Don't be surprised when they get better at other tasks as well. And um, that doesn't go down super well on the call. So then you frame it as well, what do we do about this? And I think the challenge is
This isn't a special model, right? This is representative.
This isn't your hacking model, is it?
This isn't our It's our regular Claude.
It's really good at hacking.
Claude turns out to be good at hacking now as well.
That's cool. It looks cool.
I I know. I know. It's Look, we contain multitudes. We can be optimistic and terrified at the same time.
