Transcription
There are a bunch of jobs, millions of jobs that are going to disappear.
How soon? Already? It's already happening.
This is Eric, Stanford economist who saw AI coming before almost anyone. He spent 30 years measuring what technology does to jobs. And he says we've just turned the corner. But what happens next depends on what we do right now.
I think it's going to be even bigger than most people realize. The industrial revolution allowed machines to augment muscle power. Now we're doing the same thing for our brains, our minds.
If intelligence is automated, what is left for humans to make money with?
The next decade, if we play our cards right, will be the best decade in human history by far. Or this could be like one of the worst 10 years ever.
What can someone like me do? I I think what you really need to do is >> you just had this lab paper called canaries in the coal mine that shows that AI has already wiped out 16% of entry-level jobs but only for people under 25. Can we talk about that?
Sure. It's not just under people under 25 is also specifically in the most exposed occupations. You can rank all the occupations in the economy by whether AI can affect them. So we did that and we looked at the most exposed occupations and that's the number you just quoted about 16% less employment for young people up to age 25. But it's also worth noting that in the other end of the spectrum the least exposed occupations like home health aids there's actually growing employment.
Also for older workers growing employment and perhaps most interestingly for people using AI to augment what they're doing versus automate what they're doing. we could kind of look at the kinds of prompts they were using that also those people also did significantly better. So I don't want to sugarcoat it. The core folks who are using AI to automate their jobs in places like coding and call centers that are highly exposed there was doubledigit declines in employment and since we published that paper we've continued to track it and the effects just getting bigger every month.
What are these most exposed fields? So um coding is obviously dead center call centers parts of sales marketing it's actually we find that the most useful way to do it is look at tasks as opposed to entire occupations so every job is a bundle of tasks like um you know Jeff Hinton the famous uh uh deep learning researcher talked about radiologists being replaced but you know radiologists actually do 26 distinct tasks we've recorded one of them is reading medical images that one's getting you know done by machines but they also sometimes conduct physical exams they review lab data they coordinate care with other physicians those are not nearly as affected by LLMs if you look at all the occupations in the economy there's not a single one where uh LLM just run the table and can do everything in each case there's parts of the job that LLM can help with writing memos you know doing emails looking at labs um there's others where LM can't help you know they don't lift a box or drive a car at least not Yeah,
for those uh most endangered fields, how much is AI doing in terms of tasks? Is it like close to 80% or
Well, it you know, so every that's the other thing within tasks is varying quite a bit as well. So in coding, it's it's happened so fast with agents and like I teach my course at Stanford even last year the students all did projects and they presented at the end of the class typically like PowerPoint presentations. this year, every single student, every single project they have to have running code because whether or not they were a coder before or not, everybody's a coder now. Everybody's a coder now. That's a good message for your for your listeners. If you use tools like Replet or cursor or cloud code, you can just have an idea, you describe it and cloud code or or replet will help create it. So, they all presented actually just Friday we had our final presentations. 20 teams presented it. So, that's something where it's doing a lot. uh call centers. I did a paper on call centers and when I wrote that a couple of years ago uh with Lindseay Raymond and Danielle Lee um we found that the LMS were mainly helping the human agents answer questions and the human always did the actual discussion with the person calling in. Now we're working with the same company and a big percentage of the questions are being directly answered by the agent by the AI agent I should say.
So they're employing less people. not clear actually that's another really interesting thing is that um you know it sort of seems intuitive that when AI can do a task you need fewer people but that's actually not always true in some cases like with farmers and other uh categories you do see falling employment and I mentioned with with the coders we fall falling employment but in other cases when a person becomes more productive and AI does parts of their job >> that actually leads companies to hire more of them and if I can get a little bit wonky I'm going to explain a little economics here.
Please. Yeah.
So, so the way I think of it is is through the lens of what we call demand curves, which is a downward sloping curve. So, if you compare price on the vertical axis and quantity on the horizontal axis, then um lower prices lead to more quantity. We all kind of have intuition that if you cut the price, more people can buy.
But the steepness varies a lot and it matters a lot. If it's very very steep then a lower price leads to only a small increase in quantity. So you end up earning less money.
But sometimes demand curves are very flat. Economists call that an elastic demand curve. And then a small decrease in price leads to a big increase in quantity. Like when uh jet engines made air travel cheaper, it didn't mean that we spend less on air travel. You and I and lots of other people fly a lot more than people did 50 years ago because flying is just so much cheaper than it used to be and end up spending more than you did before.
Roughly half the economy is in categories where you have falling spending as the price goes down. But the more interesting part is the half of the economy where lower prices lead to more spending. And that's a really important message I think is that as AI makes things more efficient, it's definitely destroying jobs and eliminating income in some places, but it's also creating opportunities and lots of other ones. That creation part is where I'm focusing my energy. That's what my course at Stanford is about. I have a master class that teaches people how to lean in to that creation part of the economy. I mentioned some of the changes in employment, a little bit in productivity, but it's really not a dramatic change yet. um we're watching it carefully to see whether or not um it will start taking off more. I there's a real contrast. We also created something called the AI index, the Stanford AI index, which tracks some of the raw capabilities like you know all these benchmark tests like how well can it do on a math test or uh read a document and and on those it's doing really well. So the raw capabilities are skyrocketing, but the economic impact is pretty muted right now. That gap between the capabilities and what's actually happening is a big opportunity. I think over the next few years, businesses are going to kind of close that. That's why I teach the master class. I also have a startup called Work Helix where we're very focused on teaching companies how to use those amazing capabilities to boost productivity, profits, sales. It's not happening as much yet as it should be, but over the next few years, I think we'll see a lot more.
So when when you say it is not happening, does that mean that companies use it in a way that creates, I don't know, AI slop or things they can't use? Why is it even happening?
So So part of it is they're Yeah. They're creating AI slop or they're using it in things that aren't that important. I was at one company, they did a a big hackathon where everybody was making stuff and they were so excited. You know, the winning one was this person who used uh LLMs to make lunch menus. And I was like, "Oh, that's kind of fun, but is that really like the core value of your company to have better lunch menus?" So, they need to connect it to real business problems. And it takes a while to figure out what those opportunities are and then execute well. Now, to be fair, this happens every time there's a powerful new technology. Like I studied in my PhD work, I studied how electricity rolled out 100 years ago in American factories. Believe it or not, it took about 30 years between when they first introduced electric motors in American factories and when you saw, you know, significant productivity gains. People like Paul David looked at the production records now 30 years.
That's insane.
That's insane. But that's true. That's what the data show.
How long will it take us with AI?
It's going to be a lot faster. U but it's not going to be overnight. I was just visiting Deep Mind uh about 10 days ago and they were telling me Yeah. in London and they were telling me um how oh my god with the next you know 18 months 24 months we're going to have all these capabilities and I believe them I don't know I mean you know they're the experts on the capabilities but I say it's going to take a lot longer for that to translate into business value because you need to change your business processes you need to reskill your workforce sometimes you need to invent new uh products and services takes a while I'm sure it's not going to be 30 years like it was with electricity or you know 50 years with the steam mentioned I mean some of these early technologies took a long time. This time I think it's going to be you more like 3 to 5 years. I think we're already actually seeing some inklings of it turning up. I actually made a bet with uh one of my economist friends Bob Gordon but he's kind of an AI skeptic and he said look you know AI is overblown and I said I'm on the other side of that. I think AI is anything believe it or not I think it's underhyped. I think it's going to be even bigger than most people realize. So, we made a friendly wager that by the end of the 2020s, by the year 2030, we actually made this bet the beginning of the 20 uh 20s, productivity is going to be significantly higher than what the Bureau of Labor Statistics is predicting. So, I think the official government statistics are way lowballing what's going to happen. And and that's going to be great news. If we can get this higher productivity, it's going to help with the budget deficit. It's going to be help with poverty. It's going to help us with health care. Uh we're going to have a lot more uh wealth than we would otherwise have. Um, I'm already a little bit ahead in that bet. Um, and I think that the best is going to happen in the next three or four years.
Eric just made a bet that the real payoff from AI shows up in the next 2 3 years and it goes to whoever actually puts it to work. And that's actually a perfect place to pause for a second. Lately, I care less about which model is best and more about turning AI into a system that actually helps me run the business. So, I've been testing Denspark with exactly that in mind. Every model is already inside it. GPT, Claude, Gemini, plus the image and video ones like Nana, Banana, VO, and Cling. So, I'm not picking favorites or paying for five different subscriptions. Jensen Spark crossed 250 million in annual recurring revenue in 12 months. And honestly, the models are just the starting point. The real value is what you build on top of them. Take a simple example. My team lives in spreadsheets. So, I opened AI Sheets and typed, "Rank my last 30 Silicon Valley Girl episodes by views, tagged each one by topic, and tell me which three topics outperform." A few seconds later, I had every episode ranked, sorted by topic, and three clear winners to make more of. That's exactly what Eric's betting on. Put AI to work on real decisions and act on them. That's where it turned into a system. I run that breakdown every week. So instead of retyping the prompt, I saved it as a skill. Now it's a reusable tool my whole team runs with one click and nobody ever rebuilt it from scratch. It feels less like an AI tool and more like an operating system for your business. And there is also design. I took our current futurep proof newsletter logo and turned it into a banner in a few clicks. None of this replaces anyone on my team, but what it does, it takes the repetitive work off our plate so we get our hours back for the parts that matter. I'm still finding new workflows to turn into skills. So, if you try just one thing, make it this. Pick one task you repeat every week and turn it into a skill. And if you're new, you can try Pro Tier deep research and AI web app building through the get started bonus. The link is in the description. And you have this uh report with ADP that private employees added 122,000 new jobs in May. What kind of jobs are they? Are they connected with AI? For people who are watching who are like, okay, I'm very technical. What is my next step? Do you is there any data that shows that uh you need to become a generalist or an entrepreneur within your workspace? Cuz we're talking about this, but is there something that's proving that?
You know, generalist and specialist is is one lens. I actually have a different way of thinking about it. So when I look at it, I think almost every project can be divided into three parts. There's defining the question, there's executing it once you've got it defined and then there's evaluating it. Did it really give you what you wanted? How do you need to change things? And through most of history, you know, people did all three parts. There wasn't anyone else, right? Um but now AI agents are getting really good at that middle one, executing once you've got it defined. So I hope all of your listeners are are you know playing around with cloud code or these other tools and they'll see that once you ask the right question these tools will execute and generate software. So in the near future and today it's already happening for folks at at work helix and a lot of our clients most people their job will be managing agents not just one agent but like a whole fleet of agents each person will be kind of like the CEO of a bunch of agents and their job is going to be at the first and third parts that is asking the right questions and evaluating which is a lot of what a CEO does right >> and if you can think about okay what's the right question like the FTEES what are the problems that really need to be solved that adds a lot of value and then once you can scope it out now the agent does it but let's be realistic these agents sometimes they hallucinate they mess up or what often happens is you know you think you asked the right question then the agent does and you look back say oh I guess you did what I literally asked but that's not really what I meant and then you iterate and you go back and you change the question so that's the evaluation part um that's the future of work I think is figuring out how to ask questions and evaluate and have the agents do a lot of the execution And I think it can be learned. I think it can be taught. I think it's a skill that more and more people are going to have to have.
How do you learn that to start deploying agents for your work?
That's a great way. So, everybody should start deploying if they haven't already. Um, but you know, when I teach at at Stanford, you know, we do a lot of things by the Socratic method. You know, my students don't love it when I cold call on them, but I ask them to, you know, think on their feet and define the problem. uh they do homework and they they have to like figure out um how to scope something. So it's not just okay let me write the problem for you and you just carry out the steps kind of like a cookbook. That's the old way of learning. The new way of learning is you give them a much more unstructured set of issues and they figure out okay what's the core question here and you like anything you practice you get better at it and you get to get to be pretty good at it.
The art of understanding the problem and understanding which answer is correct.
That's exactly it. Yeah,
and it takes a special mix of skills, you know. So, I think if you only have technical skills, you're going to miss on understanding the problem. If you only have, you know, uh people skills or, you know, uh domain knowledge, you may not understand where the technology can help. But if you combine the two, that's where you really add the most value.
Yeah. So, basically becoming a generalist, right? Because you also have >> becoming a generalist, >> you also have to have this academic knowledge because otherwise, how how do you know that this is a correct or an incorrect answer?
I think so. Yeah. I mean, you know, some people, you know, they have this idea there's rigor on one end, you know, theory on one end and there's relevance or practicality on the other end. That's not the way I think about it. I think of these two as being very synergistic and if you can combine rigor with relevance, that's the motto of MIT where I I used to work, men's at manos in Latin, mind in hand, um that's where you get the biggest value by combining those two things together.
A lot of people have a dream of going to Stanford, but maybe they're I don't know 12 years old. I had this dream when I was a kid. Do you think it's it will still be a valid dream in 10 years from what you're seeing?
I hope so. I have a job there.
From what you're saying, how relevant is education?
It's changing. Honestly, it's changing quite a bit. And I think, you know, the kinds of courses where they're just kind of teaching a cookbook. This is how you invert a matrix. This is the step-by-step process for, you know, for doing whatever. I think those are going to disappear. They'll become less valuable because, uh, AI tools will do them. I'm going to name some jobs. Well, good paying jobs. Would you tell people to spend years becoming them or no? Junior software engineer that pays 95k a year.
No. Unfortunately, that's one that's very much in the bullseye of being replaced because especially because you said the junior part. We see that in the data they're disappearing. Um, if you had said senior, I would be much more positive.
Do where are they going when they're disappearing? What happens to them?
You mean the jobs or the people?
People,
they need to find something else to do. So, one of the things they do is they learn to do more of the senior stuff. So, I was working with Infosys, one of the big companies,
and they said they're actually hiring as many junior people as before, but instead of having them do the sort of routine work that the LLMs and the agents can do, they're actually having them spend a lot more time >> training and learning the big picture project management stuff. They used to kind of learn that by osmosis, just by hanging around and and hoping that it would rub off on them. now they're explicitly teaching them sometimes using AI as a as a tool. So um you know it's it's a different mindset. Most companies to be frank are not that forward-looking and I think they're going to be hurt because you know they had like this pyramid most companies have this pyramid like you know a law firm software engineering we got a bunch of junior people and then some of them work their way up and become middle management and senior. Now if you get rid of the base of the pyramid, it becomes like a a diamond.
Then um where are those junior those middle managers going to come from and where the senior people going to come from? And too many companies are being shortsighted about that. I think Infosys is doing it right and saying, you know, we're still going to hire those because we need the the people with more taste and experience.
And that's uh the question that a lot of um people are having these days. How do I become senior if there is no position where I can be a junior for a few months at least?
I'll tell you something. It's it's a societal problem is a bit of a prisoners dilemma I think um or a coordination problem economists call it because for every company individually maybe it's you know it's privately okay to just like save the cost not hire the junior people but as a society you need to have those people have jobs and learn the skills. So we need to, you know, I'm glad emphasis is doing it on their own, but we also need to come up with some societal solutions. You know, for me, I think part of that is is public investment in education and training.
Mid-level marketing manager 115. Okay.
Sorry, that's another one that I'm I'm not really seeing a lot of value. We see a lot of LMS being able to do that. Now, to be fair, in each of these jobs, there's bits and pieces of them that are more immune. you know, some of the project management, the taste part, but the core part of the job is kind of in the bullseye.
Okay. Parallegal.
Oh my gosh, it's even worse.
What a list.
Look, I don't want to sugar coat it. My job's not here to like paint a happy story. There are a bunch of jobs, millions of jobs that are going to disappear.
How soon?
Already? It's already happening in our in our canaries data. Look, that that's again, that's only half the story. The bigger story is all the new jobs being created. Technology has always been destroying jobs, it's always been creating jobs. And you know, while we have this job destruction on one side, we're having new creation. And no society has ever succeeded by trying to hang on to the old jobs, you know, the coal miners or whatever that sometimes uh get talked about or or or these jobs you just mentioned. Um every society has succeeded by leaning in to dynamism, to re-education, to training, and to embracing um that kind of flexibility. Um there's a real instinct among politicians, among union leaders, among uh workers sometimes to try to just like, oh, you know, just just freeze the old way of doing things. That hasn't worked for a country. It wouldn't work for a company. It doesn't work as an individual.
Mhm. It's just the speed at which it's happening these days is much much faster.
Totally fair. And we don't not have in place the resources and the investment to help with the transition.
Yeah. We're still figuring out. No, I mean, look, >> we've seen this movie before, unfortunately, with globalization and free trade. And I have to confess as an economist, I'm one of the people who said, "Hey, free trade is great. It's going to make the pie bigger." Yes, there'll be some disruption. There'll be winners and losers, but with a bigger pie, we can make basically everyone better off. Well, we did the first part. We did the free trade, but we didn't do the second part where we helped out the people who were hurt. And now there's this huge backlash like a title wave of anti-globalization, anti-free trade. Tariffs are like the highest they've been in most of a century. It's from economist perspective. It's a catastrophe. But in a way we brought it upon ourselves by not being careful enough to point out you need to compensate and and retrain people. If you just unleash all this uh disruption without a plan for managing the transition, you're going to get a backlash. And what's happening with AI, I think, is 10 times bigger. Um, and, uh, we're going to, we're already seeing a backlash. Um, I urge my friends in the tech industry, political leaders to work on, you know, smoothing that transition. You can't ignore it.
Totally. Let's, uh, wrap up with a radiologist that gets 350K.
Radiologist. Okay, this is a good one. I love radiologist because this is such an iconic story. You know, Jeff Hinton back in like 2017, he looked at what deep learning could do, read medical images, and he famously said, he's he's one of the smartest guys. I want to give him credit, but he got this one really wrong. He famously said, you know, we should stop hiring radiologists. It's over for them. AI can do that. However, we now have more radiologists than ever. There's almost a shortage of radiologists. They're being well paid. Why is that? Well, it's a couple of things. First off, reading medical images is only part of a radiologist's job. They have these 25 other tasks that they do. And so when you make one part more efficient, it actually increases the demand for the other parts. And the related part of it is that the elasticity of demand for medical images is very high. What that means is that as you make it more efficient, you actually have more demand. Yeah. Like if I have a little bit of a sore shoulder and it cost me $2,000 like an MRI, I'm like, "Ah, now I'm gonna do it." Cost $200. Yeah. I'll go I'll go have it checked out. And so what we've seen is that making things more efficient led to more demand. And um there's a lot of people who could use more medical care. So I think that's one of the areas where in general we're going to have growth is in medical care. AI is going to make it more efficient, but that doesn't necessarily mean we'll spend less, we'll spend more. And you know, I actually think that's good news because it means more people are going to be helped and we're going to have, you know, maybe twice as much spending, but four times as much cures, four times as much uh benefits.
So, it's a good job.
It's I think it's it's it's been a good job. Yeah. And it probably will be for a while.
Okay. This is the part that actually worries me a little. Everything Eric just walked through, which jobs shrink, which ones grow, and the whole economy is shifting under our feet is a lot to sit with. And the thing people always ask me after a conversation like this, okay, but what do I actually do? Where do I even start? That's literally what my newsletter is for. Every week, I take what I learn from podcasts like this one, from my own experiments with different agents and models and turn it into the real moves. What to learn, what to build, how to end up on the right side of the shift. My newsletter is called Futureproof. It's free and it's very, very practical. The link is in the description. Subscribe and start deploying AI in your life.
You know, ironically, I think a lot of the liberal arts have become more valuable. Philosophy, like even art appreciation, you know, in a future world where we have abundance, and I don't know for sure we're going to get there, but if we do, then, you know, learning how to appreciate art and music. You said you were a singer earlier. Yeah.
Are you going to sing for us a little bit?
Maybe. um you know that actually is is a great thing for universities to do and so it it's kind you know it's a little um contrarian view but I think one of the things that universities should think about doing is going back to the way they were like a few hundred years ago you know a lot of universities really started off as being you know liberal arts philosophy religion art uh music and um uh history that stuff I think is going to always be important
that makes total sense that's developing taste basically developing taste. Exactly. And and you know, I I mostly took like nerdy uh math courses, but I'm so glad I took some music appreciation courses and I honestly like can hear music differently. Like you you literally hear things that you wouldn't otherwise hear before you took the course.
And you can taste things if you go to wine tasting here in Napa like you know, you know, you can like learn to like recognize new kinds of tastes. You can see things in in art that you didn't see before.
It's like opening up your eyes.
Okay. Now let's talk about um this AI revolution as an economist right you've studied all the previous AI re uh all the previous revolutions and we've had the recent one semi-reent industrial revolution didn't happen as fast apart from
you're taking the long view I like how you called the industrial revolution a recent one yeah
well it's one of the >> in the greater scheme of >> in the greater scheme and in terms of impact so I think the one that we can talk about when we try to compare to AI is industrial revolution
a greater comparison Yeah,
but that happened much slower apart from speed. What else is different this time?
Well, the main thing you So Andy McAfee and I wrote this book called the second machine age, which everyone should go out and buy and read. Um, the second machine age explains all this. And the basic idea is that the fir the industrial revolution was this first amazing transition in our world. Up until then, most people their living standards just barely moved. Their parents, grandparents, great-grandparents, they all lived close to poverty. That was just life. And you know, the average family didn't change. With the industrial revolution, we started seeing economic growth skyrocket. Well, so right now we're like 30 to 50 times richer than our ancestors a couple hundred years ago. And the reason for that is the industrial revolution allowed machines to augment muscle power. So instead of, you know, humans or or cows, you know, providing muscle power, you had steam engines. And it just unleashed this an amazing explosion of productivity growth. couple percent per year, which may not sound like much, but when you compound it, it's like I said, 30 to 50 times richer. That was a real, it was kind of like a singularity, the first singularity, where we transitioned from like stagnant growth to much faster growth. The current era is what we call the second machine age, because now we're doing the same thing for our brains, our minds. We're augmenting them. And in my view, that's going to be at least as big. It's going to be bigger. It's going to be faster. is going to affect a much bigger share of the economy. Most workers in the United States and other advanced economies are doing cognitive work like you know most of what your job is is not like lifting boxes it's you know communicating ideas mine too and uh even you know even people who like they're doing a lot of physical work they're also usually doing a lot of cognitive work as well
so AI is going to be even bigger than the industrial revolution it's clearly happening a lot faster and uh that's the good news the bad news is like we were talking before we're not really prepared for the size of this this tidal wave of change.
So people make money these days because they have this scarce resource resource which is intelligence. If intelligence is automated, what is left for humans to make money with?
Oh my god, that's the trillion dollar question. And I don't think there's a clean answer, but you're totally right. Like people like me, I kind of prize intelligence because, you know, it's helped me make a lot of money and it's kind of where I get my my status from. But AI is going to um you know, have intelligence on demand. So one thing that's going to be more valuable is initiative or agency. Uh what my closing class my students will remember me saying I think whenever they hear the words AI they should think of >> amplifying intention not artificial intelligence because what it does is it takes your agency your intention and it amplifies it. If you don't have any it doesn't do much for you but if you've got a plan this can totally amplify it. So the people in the future are the ones with a lot of high agency. The second thing I think that will be increasingly important is human connection. You know, um when AI was able to defeat humans at chess, that was not the end of chess playing for humans. People today play chess more than they did before. My son, Xander, he likes to play chess. And I asked him, do you play against machines or humans? He, well, humans, of course. It's no fun to play against machines. And you know there's this Knicks basketball game last night that millions of people watched. I don't think it would have been nearly as fun if it was a bunch of machines playing each other. So in the future we will value things that are certified human that are you know authentic that real people are creating. I think that's another big area. Uh a third area that at least for a little window will be valuable is is just like physical work. I mean AI is getting very good at cognitive work and if you are a plumber, a carpenter, if you have you know uh particular skills, uh that's something that turns out is harder for machines to do. That said, I think the windows closing on that one. Um and then the fourth category I would say is all the things I haven't thought of. Um every time in history that we have tried to think of what the future holds, we've always way underestimated. If you and I were having this conversation 200 years ago, we'd be like, "Well, all the farmers, you know, they're going to disappear. 90% of people are farmers." I'm pretty sure we wouldn't have thought of, you know, podcaster or, you know, all the other jobs that exist today. And there will be new ones that are invented and created. And and and it's not necessarily my job to invent those. You know whose job it is? It's your viewers. It's it's entrepreneurs. And here in Silicon Valley, people are constantly trying out new ideas. A lot of them are really dumb, honestly. And some of the really dumb ideas turn out to be brilliant later who you've turned out that oh my god, you know, space data centers. Well, maybe that can work. I I don't know. Um, and so we have an ecosystem here. I had a had uh brunch with a VC this morning and she was telling me that, you know, all of her payoff is just from like five or 10% of her investments or less. And the other ones, you know, they don't pan out. And thank God we've got an ecosystem where people like her are willing to take those gamles and the entrepreneurs willing to take those gamles and they try out things and America is leading the world in this kind of innovation of inventing new things and I'm looking forward to seeing what they invent next.
Yeah. Well, we're always good with coming up with new things, new bottlenecks and things to solve. That's the definition of of a human.
Yeah, that's that's our superpower. You know, I had I know you had Reed Hoffman on this before and he told me something really valuable. you asked this question about what will what will humans do and he said a human superpower is improv improvisation and um you know you you define the problem really well and the machine can do it but if there's something unexpected that comes up you know then the human figures out how to do it actually if you have time he had told me this funny little little example that really crystallized it for me he said imagine that you have like an ordinary person from my class had to play chess against the world's best chess computer and the game was in 30 days and you know whoever wins you know great that the loser dies. Um he said that he wasn't sure but he thought there'd be a decent chance that the human would win. Not because the human could play chess better but let's face it if that human was life or death >> they would probably figure out some way to short circuit. You know maybe there'd be a virus in there. Maybe there'd be a a lightning bolt that day. you know, something water would spill in the wrong way and they would just they'd figure something out and they would they would find a way to win and that's what humans are good at doing.
How do you see um resource distribution when it's not companies hiring humans? What is it?
I'm super worried about this. You heard me earlier say that I'm optimistic about growth and I think we're going to have higher productivity growth, a lot more wealth creation. I'm concerned that that's going to be very concentrated, more concentrated than it is right now. It's not inevitability. We have choices going forward. Um, and one of the things I want people to think about is what kind of values we have and what kind of future we want to create. I would like to see a world where we not only have prosperity but shared prosperity. But one scenario that worries me is AI will automate a lot of work, a lot of jobs and people will be entrepreneurial. But if it becomes too focused in just a few companies or one big government-owned entity then all the wealth and power gets concentrated >> and we need to you know plan for a future where lots of people can participate and where everybody has a stake in the society. I don't think either of those paths is inevitable. Um but I do worry that we are right now on a bit of a path towards that growing concentration of economic wealth and therefore political power and we need to we need to be mindful of that.
What can someone like me do or someone who doesn't have a podcast like how can they make sure they participate um by buy stocks?
Well literally one of the reasons I created companies no I think well stocks is a bit but I think what you really need to do is create the value and that's why I created the master class. That's what I teach in my my Stanford class is how can you use AI to create new goods and services. Not to be a rule follower who, you know, just does stuff because you're going to be replaced by a machine if you do that, but how can you be one of those people who ask the right questions? How can you use AI to create new products and services? And in a world where there's more entrepreneurship and value creation, then I think we continue to have widely dispersed economic power. Um but if everybody's just following instructions, then we're going to have that concentration of wealth. So that's the number one thing. Another thing, look, I think we do have to look at, you know, different kinds of redistribution. It's not my first choice, but we need to have it as as a as a backup plan that if we have a lot of concentration of wealth, then we need to have things like universal basic income and, you know, progressive income taxes, wealth taxes. is I know a lot of my Silicon Valley friends are going to yell at me for that. But I think that you you don't want to have all the wealth and power too concentrated. It's not in anyone's interest including the billionaires. People will come after them with pitchforks and and so we want to have a world where everyone can participate and in the end people create more value. You know, I've visited some of these um you know, developing countries or parts of Latin America where wealthy people live in gated communities with these walls and they have like machine guns and you know, they have their own schools, their own doctors and private police forces.
No, it's not fun for anybody. I had a a friend, she uh she lived in Brazil and she said she and all of her rich friends were in prison. I said, "What do you mean you're not in prison?" She said, "No, we're a prison of our own creation. I sit I sit behind these walls and when I go out I have guards on either side of me because it's just like the society is not safe for me. Totally. And you know I don't think anybody wants to live in a world like that.
Yeah. It's just interesting when we talk about this problem it feels like it's up to those large corporations, governments and on the individual level yes you could become an entrepreneur but it's not like everyone is entrepreneurial.
Let me push back on that a little bit. Honestly I think a lot more people could be entrepreneurial than they are right now. You know, a few hundred years ago, most people were kind of farmer entrepreneurs. They ran their own thing and then we created these societies with big corporations where people became kind of like cogs and, you know, create a lot of wealth. But I think we make potentially and I'm not for sure, but I think we could try to go back to a world where a lot of us are initiative, our agency became more important. And you know, I really think using these tools like we show in the master class is is exactly what you want to do is is figure out how to do it. I think almost everybody has some area where they see problems that other people don't see, where they understand some needs and opportunities. And you know, you can just take a Saturday afternoon and just brainstorm with a sheet of paper or with one of the LLMs helping you all the types of things you might be able to create and and try some of them out. And it the neat thing is that it's so low cost to give it a try. If it doesn't work, then you try something else. And for most of it, it's kind of fun. Honestly, I think it's more fun creating new things than it is just following instructions. So I I would encourage probably every one of your listeners to at least give it a try.
Yeah, that makes total sense. That's what the the purpose of this channel is honestly to inspire people to >> you. Yeah, you are doing it and and we need more people like you. We need more people listening to to this show to to give it a shot and and have it work and it it'll not only be good for them, it'll be good for all the people. If you want more conversations like this with the people who can see where the economy is going before the rest of us and what they'd actually do about it, subscribe to Silicon Valley Girl for more. What about the whole concept? Because I studied economics and you know we all studied market economies. Do you think we're going to switch to this new AI economy where money loses value? When you think about this like in 10 years, what do you think it's going to be?
It could be it could be different. you know the we need an economist who can think through what the economics of the future is. I'm trying to help play that role. You know, Adam Smith helped define the market economy and John Maynard Kanes helped update it in the early 20th century. I think for the 21st century, we're going to need some new economic rules to understand it. Um AI AI agents, we're going to have billions or trillions of them. We're going to have a lot of routine work done automatically. um the kinds of things that worked in the old mo old old market economy won't necessarily work going forward. I mean one way I think about it as I learned in my my PhD program is you can think of a market as a big information processor. It takes all this information about prices and quantities and aggregates it and allocates resources. You can also think of a of an organization like a big company as an information processor. Both of them are information processors based on 20th
century technology. Now we're going to have a millionfold more powerful information processing AI. It would be a miracle if those two institutions just stayed the way they are. I'm pretty sure they're going to change. Exactly how, I'm not sure.
Um, you asked about money, particularly. I, I think it's very likely that we will have a world where our basic needs, you know, the base of Maslow's hierarchy, um, will be taken care of, and we'll be able to just like you gave me some water here for free. You didn't charge me for it. Thank you. Um, you know, it'll be like that for most goods and services. It'll be just like, why would you charge something for something that can just be made by robots for free?
Now, there will still be things that are scarce. One obvious thing is status because it's kind of zero-sum. It's like a hierarchy. Um, you know, or there'll be a few physical things like, you know, I want to go to the far side of Pluto or something, you know, that would still be expensive. Um, but a lot of basic needs will be taken care of, and then we'll have to figure out, you know, what the economy is, what our new status hierarchies are. Some people will, you know, get status from being great entrepreneurs. Uh, some will be from getting lots of citations in academic literature. Some will be great snowboarders or video gamers or, you know, movie stars. There'll be lots of different ways you can get status and, and I think for better or worse, we humans are kind of wired for that. And the real job of the future economy is to steer all that status competition into something productive. You know, be like Einstein or be like, uh, Pasteur and cure some diseases rather than zero-sum status that doesn't really help anybody.
Totally. Do you think GDP is going to explode in five years?
Depends how you measure it. So, traditional GDP is getting to be a worse and worse measure of what's really happening. I do think welfare and productivity is going to explode, and probably conventional GDP will capture a big part of it. But the thing is that, um, GDP measures all the things that are bought and sold in the economy. So when something has zero price, with few exceptions, it has zero weight in GDP.
And think of all the free goods we have like Wikipedia, YouTube, you know, most users of ChatGPT are free. Um, that doesn't show up in GDP.
By the chosen valuations of those, those companies? Well, a little bit. That's not really GDP either. Yeah, but, but yeah. So, so they think somewhere, it's not like, no. Um, it, it goes to well-being, but it doesn't necessarily show up in, in any measure of GDP.
A little bit of it is in electricity. Um.
I wonder if those, set aside the ones that we can do those separately, but let's just look at like a Wikipedia, something that's totally free. Like that doesn't show up in any stock value. But the average person, we've measured this, values Wikipedia, you know, way more than Encyclopedia Britannica. They value it at like $10 a month. If I have to go back and check the numbers. So there's, you know, billions of dollars being created. And there's lots of other free things like that. And some of it does show up in advertising, stock, elsewhere, but, um, I've studied this and, and, and, and most of it is just invisible in GDP. So we need a new measure. Happiness, like Nordic countries where they have free education, free healthcare, they measure happiness.
It would measure, so some of it shows up in happiness, and that really is the ultimate measure. And so there's one measure there, these happiness measures where they ask people on a scale of one to ten, how happy you are. And yeah, you know, my country, Denmark, usually does pretty well. Um, so that's partly, but, you know, that's pretty coarse, like one to ten, like are you a 6.2 or 6.3? I mean, it's kind of, um.
So we've developed a new measure. We call it GDPB. The B stands for benefits. And what we do is for every good, we ask, you know, even if you're getting it for free, how much would I have to pay you to stop using it? If I paid you $50, would you stop using, uh, Wikipedia for the next month? Some people say yes, some people say no. What if I paid you $2? Um, how about ChatGPT? How about Google search? How about email? And so we've done this for 600 goods and services, and we now have kind of a ranking of how much consumer surplus, how much value people are getting from all these goods. And it's staggering. There's trillions of dollars from free goods that are not otherwise being measured in our economy. I think for the 21st century, we need to lean more on tools like GDPB and be able to understand where the real value is. So, we're in the process of rolling this out in such a way that we, we'll still have traditional GDP, which is what the, where you spend the money, but increasingly we want to start paying attention to GDPB, which is where you're getting the value. And those are two different things. There may be things you spend zero on and you get a lot of value. There may be things you spend a lot of money on and you're not getting a lot of value. They're two different things.
Roughly how much value is created by Wikipedia versus ChatGPT versus, you know, bacon and eggs, doing all those things? Like, like, just like for LLMs, we did this and we just published this. So for LLMs, like chatbots, the amount of value just in the past nine months has gone up by like 70%. And that's partly because people value each LLM more than they did, uh, nine months ago. It's also partly because more and more people are using it, and so we're just getting these, are creating a huge increase in, in, uh, welfare in the economy.
Is it $125 a month? I think the number that people.
It varies. So here's the thing is like, different people have different values. So our approach allows it to be heterogeneous. So there's some people who value $125 a month or even $1,000 a month. There's other people who value it at $10 or zero. So you get a whole demand curve of them, and the total area under that is the value created. Um, you know, a few people who value it a lot add, add some of it, and then a lot of people who value it a little bit add some, and you get the total value is the, is the, is the sum of all those.
What's the number for you? How much would you pay to never, to not touch AI this month?
Oh my god, it almost, I mean, for me, it's, it's tens of thousands. You know, somebody because I, I, it's my life. Like it's, I use it every day. I use it every night until too late at night. You know, I'm working with Claude co-work and and testing out different research ideas.
Uh, I use it for fun when I plan things. Anytime I land in a new city, I have it give me advice on which restaurants to go to. Uh, it's just so integrated into my life. It would be like, uh, tearing off my left arm.
Is there a use case that can be very inspiring for people who haven't tried using AI deeply enough, if they only use it like search?
Here's a kind of meta way of doing it. Sit down with it and ask it how I can use it in my life.
But if they haven't used it enough, I don't think there's like enough.
No, no, no, no. You have, you have, you have the conversation. So what you do is, is you ask ChatGPT or Claude, say, "Hey, tell me how you can be useful to me and ask me questions." You can literally say, "Keep asking me questions. Interview me." And it'll say, "Okay, you know, what's your job? You know, do you have kids? You know, whatever. Um, what are some of the problems you worried about last week?" And it'll have a conversation with you. And then I've done this, by the way. It'll come up with like 10 recommended things that you can be using it for.
What would you never delegate to AI?
What would I never delegate to AI? Um.
There's nothing like that anymore.
I know. I, something pops in my head, I say, no, I could sit doing that. You know, delegate entirely. You know, there are some really like life or death decisions. I, I use it before I go to the doctor and it gives me some, um, thing questions to ask, but at the end of the day, I still want to have a real human make the call and, you know, they're just not good enough. They, they have the issues. Um, and, um, I think it's, it's usually a partnership. Like, so I almost never 100% delegate something to AI. It, for me, it's always co-working and collaboration where I'll interact with the AI and it will give me some ideas and then I'll overrule some and I'll agree with some, and it's kind of a, a partner.
How much more productive have you become in the past few years?
I think I've become a lot more productive. I'm not sure it would show up in official GDP statistics, but I feel like the.
The amount of papers maybe you can, can you attract that?
A little bit. I think it's also like the quality. I'm working on some more interesting problems that I probably wouldn't have. Um, yeah, and my citations have gone up, but that's just because I think I just say the word AI and people, you know, cite me, you know, from my daily work, I feel like I'm being much more productive. I, I, like I'll give you a little more concrete example. You know, as a professor, like a, a pretty common routine is I'll meet with grad students. We'll talk about a research project. I'll say, "Hey, why don't you do this? You know, look at this data and see what the answer is." And then they come back, we meet like once a week, and they show me what they found and like, "Oh, that's interesting. Well, this part doesn't make sense. Why don't you go back and double check that or let's explore this?" And we kind of had this weekly cycle, and we, you know, we move forward and after like, you know, 10 weeks or 20 weeks, you know, we, we figure out what the answer is or we think we do, and we write a paper. Now, that cycle is like almost instant. I will sit with Claude co-work and I'll ask him and say, well, what do the data show? And, you know, five or 10 minutes later, it'll pull up the data and I'll say, "Oh, wait, that doesn't seem right. You know, you should double check this part." And then we'll go back and, and I have a similar kind of conversation. In some ways, it's worse than grad students, but no offense to my wonderful grad students. In some ways, it's better. Like, it's much faster, and it, it sometimes can track down different kinds of data. You have to know about its strengths and weaknesses, but that cycle time is just so much faster.
Yeah, it's fascinating with the speed, but also something that I'm noticing myself. Yes, I'm becoming more productive. Yes, the speed is faster. But there hasn't been this change that's, I don't know, almost dramatic. For like, I was just talking about this with my peers, like, for example, like COVID happened, right? That dramatically changed our lives. With AI, we're talking about this dramatic change. For some people, yes, it is because they've been laid off. But we will never know if that's AI or not because a lot of companies just use AI as a word. But we haven't cured cancer yet. No, uh, self-driving is, yes, it's cool, but it's in San Francisco, and it's still like, it's rolling out, but it's, it's regulation. When do you think we're going to see something that's going to be mind-blowing for all of us and we're going to say, "Oh, wow. This, this is where I see the impact."
I think over the next three to five years, people are going to see more and more mind-blowing things. There's little ones already happening. There are some, um, breakthroughs in medicine, and there are some, you know, you mentioned like cars, and companies are beginning to use it. I, I agree 100% though that it hasn't nearly had the economic impact or the impact on work, um, that you might expect given the magnitude of the technology. And that's back to that J-curve idea. It just, everything takes longer than the technologists think, but it is happening. It is coming, and, um, by 2030, I don't think there'll be any question that this is transformative of the economy, um, but these things happen step by step. So we're somewhere.
Do you think we're down here in the J-curve?
I think we're turning the corner. You know, that's why we created the takeoff tracker. If you go to the AI Economic Indicators at Stanford, you know, we have these, these metrics, and every month we're updating them. And there's a few of them like, we have these different categories: no evidence, mild evidence, strong evidence. And, you know, there's one or two that show strong evidence, there's, uh, three or four that show mild evidence, and all the rest show no evidence yet. But I, I'm pretty confident, well, we'll see is every month we're going to sort of be moving more and more into the mild or the strong evidence category, and then it might start happening really suddenly. You know, there's this thing that, that we say in The Second Machine Age, my book, is the thing about exponentials is that things happen slowly and then suddenly. And we're just entering the suddenly part. We aren't in the suddenly part yet, but we're getting there.
Wow. Okay. This, this makes me very excited. A little bit scared, cuz we never.
That's the right thing. I'm excited and scared too. No, look, if you're not both excited and scared, you're missing at least half the story.
Yeah. Okay. My last question. If my daughter, she's four, five years old, just turned five, asks me tomorrow.
Yeah.
What is my life going to look like in 30 years?
Yeah.
What would you.
Oh, nobody knows. 30 years. No, I think it's going to be hard enough either even five or 10 years. Look, I think the next decade, if we play our cards right, will be the best decade in human history by far. There'll be more wealth creation than ever before. We're going to have noticeable improvements in longevity. I mentioned I was over at, you know, Google, and I was talking to, I'm sorry, at DeepMind. I was talking to Demis Hassabis. He thinks that, uh, they'll start curing a majority of diseases within 10 years. I hope it's right. That sounds ambitious, but your daughter will see that. So that's the good news.
I also think there's a future that could, this could be like one of the worst 10 years ever. I have to be honest that, like, there's the potential for catastrophic risk. You know, viruses being created in the lab and released. Uh, AIs, uh, taking over social media and manipulating people, for mass centralization of power. We already see AI-powered drones like hunting down people. These are like so tragic. The dystopians, you know, these drones like chasing a soldier. It's like, oh my god. And, you know, doesn't matter which side of the war I'm on, I kind of sympathize with the human being chased by the drone. So all those things are also possible. Um, the thing I would say is that we have a tremendous amount of agency. And so we should think less about what will happen to us and what AI will do, and more about what we want to use AI for. AI is a tool, and a message I keep hammering over and over is that when tools become more powerful, that means by definition we have more agency. We have more power to change the world. So we need to really think, you know, be philosophers and think about our values. What kind of world do we want to shape? And don't take it for granted that AI is just going to steer us one way or the other. We still have the agency right now, and we should be steering that technology towards one of those more beneficial futures and being damn careful to avoid those catastrophic futures. They, I totally think they're possible.
The doomers are not wrong that there's a real risk there. Um, they are wrong if they think those are inevitable. Um, because we'll have choices. And so I've been working with, you know, the labs and with politicians to do what we can to share shape us towards that future of shared prosperity.
Fingers crossed we're going to land on the positive scenario. I keep telling my daughters that they won't have as many problems as I have. I mean, like they will have different ones, but the ones that I'm having, they're probably not going to have them.
That's probably true.
Thank you so much, Eric. This was so insightful.
Oh my god, it was such fun talking to you. Thank you for having me on.
Amazing.