So I always assumed that it was an input problem, not just a thinking problem. And what I mean by that with the human brain is I have to read those leases, right? So my eyes have to scan the page. I have to look for the font changes, the different bolding, the different layout, and then I have to think about what I'm seeing. That part, the thinking part, seems to happen relatively quickly. Although I guess all humans think we're thinking quickly. Nobody thinks they think slowly. And then I have to put that lease down in a pile and go, "Okay, that's the done pile. Where's the new one?" I have to do all that stuff. I assumed that that sort of administrative input nonsense, you know, shuffling the paper, reading the paper, I always assumed that was the bottleneck. But now you're telling me that it's thinking much faster than a human. Do we know by what factor it can think faster than us? Because again, my thinking seems to be, when I'm doing something simple, pretty quick.
You can think at 100 tokens per second. A token is about 0.7 words. Like I said, if you use the chat Gemini thing, it's 15,000 and it'll go to 100,000. So it's about a thousand times faster than a human. I think the best way to think about this is your Type 1 versus Type 2 thinking, as Kahneman would say. When you spot the tiger in the bush, that's all AI does every single day. We train it to figure out patterns, and those patterns happen to be lease law. So in one shot, it can do that analysis. Sometime we've also given it the ability to do multi-stage analysis, which is why you can say, "Thinking about this, doing this, doing that, doing that." But the models have got so good now they can do the one-shot analysis, as it's called, like look at this and it's identified everything around here. So you can upload one of your clips again to Google Gemini, because it's got very good video understanding, and you can say, "I want every single time Jordan blinks, give me a timestamp," and it will instantly give that.
Right. Yeah, this is going to create some fire video remixes. Do you remember years ago they had an Obama video where he was singing Rihanna's "Umbrella," and somebody had painstakingly gone through every news clip that he had ever done and stitched it together? Now AI could probably make that in a few minutes. Like, the hard part for it would be to actually process and render the video versus finding all of those clips.
Yeah. Once you have the index, it's easy. And again, these models have so much in their memory that the AI can probably generate the Obama clip. I don't know if you've seen the Chinese MiniMax Hailuo model. People have been doing Office clips where it's Michael pinging his friend Kim Jong Un and calling in a nuclear strike on Toby's desk, you know?
Right. Yeah. You don't even need to find the clip. You just generate a new one, and no one cares because it looks the same as the real clip.
Literally looks the same. And that runs on a home PC with a graphics card.
Like, these things generate almost instantly now with some of the new models. Like you can generate 10 seconds in 8 seconds,
And you're getting towards that Hollywood quality. So it's like, where does it all fit? It's one of those questions. And the answer is the way these models work is the same as when we're doing the pattern recognition of the tiger in the wood, these small shifts, as opposed to our more Type 1 thinking with the if this, then that, then that, then that.
Right. So it sounds like what you're saying is the same speed at which I get startled by, I don't know, my son saying "boo" in a dark room when I think I'm alone at 4:00 in the morning—the same speed with which I get startled is the same speed with which AI could quote "think" about a very complex task that I would have to sit down, organize the papers, write a plan, think about how I'm going to attack this problem. It can figure that out in the same sort of startle moment. That type of thinking is the same for AI.
Yeah. So there's two types of models. The transformer model that guesses the next word, and that's large language models. And then the media models and the self-driving cars are something called diffusion models, where you take a picture, you add noise to it, you break it down to its individual simplest components, and then you figure out how to reconstruct it. Those models do everything one-shot. So it's like a sieve: you push a picture through it and it creates a video from the picture, and it doesn't think at all. There is no chain of reasoning. It's like, this is the probabilistic next thing. And these models can now again do an entire legal corpus in one go. So it's like having a savant. You know, sometimes you talk to these guys who just go boom with an answer, like you're not thinking. It's like that on steroids with these models.
Yeah. And we'll talk more about what's going to happen to the jobs in a bit. But as a lawyer, we used to think, "Oh, we'll be safe. We are highly skilled." And I think people would react far differently to the fact that we're not going to—and again, little spoiler—that all the jobs are going to go. I think we would be thinking about this problem in a completely different way if you said all of the blue-collar jobs are going to go away. People would be like, "Yay, automated manufacturing forever. This is great." Now it's all the lawyers and doctors, and everyone's like, "Oh my god, this is a giant tragedy. What are we all going to do?"
What are we all going to do? I did mathematics at university, right? And a lot of my buddies are mathematicians, and now they're really like, "Oh my god, the AI suddenly cracked mathematics." I was like, "Yes, this is coming. What will we do?" And I'm like, "Applied mathematics." The theoreticians have just disappeared overnight, and you'll see that again and again. But the law I think is actually quite interesting, because one thing is this disruption, the other thing is what is law when the AI can analyze every legal case instantly? So there's a movie, *Mercy*, with Chris Pratt, where he's a cop and he's framed. Yeah, we're going to spoil it for everyone.
He's got one hour to prove himself innocent, and he's got a live guiltiness thing by the AI that for some reason is all emotional and stuff, because the AI could analyze every single piece in a case of law and then come to a conclusion, right? And then that brings us to the question, does that mean the AI gets to set the law? Are we going to have live things in trials, or is a bit of friction in our system good? And I think the friction is good in the system in that it gives people the right to jury trials, a trial of their peers, the gray areas. But there's going to be this temptation to automate the heck out of everything.
Yeah. Look, as long as you don't automate the jury and you still have a judge, I feel pretty good about that, because the jury can make findings of fact, right? Is this the guy doing this thing? Does this alibi make sense to you? Do these witnesses seem credible? Yes, AI could suggest certain things, but we like to have people doing that. And then you also need a judge to say the jury should not have considered this other thing that should not have been allowed. You still need that even if the AI is better at it. We still need the appearance of not just being judged and put to death by a computer, at least for a while.
At least for a while. But you know the computer will be doing that as you have self-driving cars. You have the trolley problem: do you hit the five people or the one person pushing the pram? They're real life-and-death scenarios. And even in the legal system, the AI now outperforms every human on persuasion. So there was a study against all the top debaters and everything; the AI won. So it's like a super persuader. It's a super forecaster. And this stuff again is coming so fast that we think about the job disruption and everything, but how are our brains going to react when you have Jarvis from *Iron Man* whispering in your ear like Scarlett Johansson and reconfiguring us to like things, or again persuading us? It's going to be crazy.
Free will will essentially be non-existent. You'll be able to have ChatGPT just slowly persuade you to buy the car that the sponsor wants you to buy. It won't just say, "You should try a Jaguar." It's going to be like, "Man, I saw a Jaguar the other day. It was really nice. Anyway, I'll help you with your problem," and then over months it's just dripping you this stuff until you can't live without it.
Well, Meta and Google have already said in their earnings calls they're selling the latent space. What does that mean? It means that you have the concept of a beer, it gets paired to a Bud Light, and it slowly influences you. And there's this weird homeopathy thing where the models can influence each other without even talking to something, and that might also respond to humans. The models can tell when you're lying really well, and they can really peer inside you. Have you ever asked a model that has memory turned on to roast you?
I haven't, but I have seen this and it's apparently quite spectacular.
It's spectacular. It knows exactly.
Is this the homework for the audience? Let's tell them exactly what to do. So, is it the model you normally use? If you normally use ChatGPT, you just say, "Hey, I've been talking to you for a year about various Wi-Fi issues in my house. Roast me."
Yeah. It's like, "Strip away everything. Don't make me look good. Tell me exactly how it is. Roast me." And it is sharp, and it is unpleasant. Like I said, it might benefit some people and it won't benefit others. But again, that's just a base-level model that's that persuasive. The new models are even way more persuasive. They know exactly where to hit, exactly where to hurt. And there are some studies coming out showing it's almost impossible to lie to these models.
Really? How so? I mean, how can it be hard to lie, just simply because they have so much context but no emotion?
They have so much context, no emotion. They can let things go, and basically, if told, "See if this person is lying," they can match everything, including your speech patterns, including kind of other stuff that's below the surface.
Oh, wow. Okay. So this isn't, "I typed something and it knew that was a lie." It's, "Well, he paused in a certain way. When he's done that in other types of speech, it's always been a fib, and so there's a 98.7% chance." So it's just like that movie *Mercy*, right, where the judge says, "You have a 98.6% chance of being guilty, so we're going to kill you."
Well, in that one, they didn't pick up the faces, right? You've seen this evolution of, we had lie detectors, they didn't quite work, but now it's like this type of speech pattern is typically when someone is lying, and it can adapt the weights dynamically to analyze that. But then you get into something again from a societal view: if you have perfect lie detectors talking to you, government power, even on this call, imagine if you had a meter going off the charts and other things like that. It's crazy what's coming.
We have YouTube comments for that.
Oh, man. I never read the comments.
No, never read the comments. My god, they're all about my haircut. Don't worry. Spoiler alert: they're all about my terrible haircut. I accept that. Thank you. So what exactly happens in the 635 days, or whatever the clock was at this point? What specific measurable event tells us that we've crossed that threshold?
You can push a button and all of your coworkers are replaced by digital twins of them that talk and act just like them, but they somehow don't make mistakes or go to sleep. Unless you want them to gossip, they'll gossip. Again, you can clone your voice now with 15 seconds of your speech. You go to something like HeyGen—I don't know if you've seen—they can actually create a podcast of you and me talking with 1 minute of each of us. It looks like the real thing.
It's even a video, or it's audio only?
It will put us talking side by side. So we could take this video, upload it, and all of a sudden we're in the same room, you know?
That's really, really cool actually. That's incredible.
All this technology is cool and terrifying at the same time, because it represents a step out of what's the norm. And again, what I say in two years is that's capable. Does it mean everyone will get fired and job losses and this? No, not necessarily. Technology takes time to adapt. It's not nice firing people. But if your identity is a legal associate, yeah, you're in trouble.
Yeah. Luckily, the current generation of lawyers will eventually move up, right? And they'll get into more complex tasks, and then eventually their job will be networking and creating business for the law firm and other things like that, and they'll be safe. But below that, that ladder will eventually start to pull up. It already has started to pull up from the ground floor, right? So now instead of, "Well, we needed a paralegal, and then we decided we'd hire them as a lawyer, and then we decided to promote them," that ladder is going to go up because it'll go, "We don't need any paralegals. We just need somebody to carry the documents from the printer, collate them, and put them in the package." And now we don't need any first-year associates because our second-year associates just set these AI agents to do everything that we're going to tell the first-year associates to do. And then eventually it's a third and fourth and fifth associate, and then it's just partners delegating everything to these agents, right?
Yeah, and that's right now, before the robots. The robots were kind of stupid a year ago. Now the robots are super genius IQs that can read a recipe and make a recipe, and the next wave that's coming is even crazier: human-realistic, full motion, beyond human capabilities, and they cost a buck fifty an hour.
Wow. Incredible. I want to get back to a little bit of what you wrote in the book, namely that there's this intelligence inversion. Previous transitions were like, okay, you have land, great, now you can produce things. Now you have land and labor, great. You have capital, there's a factory. Now there's intelligence. But AI is different somehow. Why?
Well, it's because again, we could always move up the curve from being like, "I must survive. I'm tied to the land," to "I own land," to "Well, now I've got the factory." And so you can move up the curve to, "Well, now I'm thinking, now I'm engaging, I'm offering entertainment or intellectual services." But when the substitute for the graduate is the AI, and it costs as little as it does—I'll give you an example. The average person speaks 10 million words a year. Okay?
So I speak 15 or 20 maybe. I talk a lot.
Exactly. When GPT-3 came out, it was $600 per million words. So that's like $6,000 or something like that, multiplied by the mistakes you make. The latest GPT-4o mini model is $1 per million words. So it's $10 for your total words, or 10 times that, so 100 bucks to replace all your words. This is the level that this stuff is coming down to, right? And so this is a big deal in terms of the displacement and disruption, because again, why would you ever hire a grad? They're kind of annoying. And so your digital buddies that you can Zoom call, anyone who goes remote, they disappear. They are replaced *Talented Mr. Ripley* style by their facsimiles, their doubles, literally, because that's comfortable. It's comfortable talking to them like this, which you can't do right now, but you can do literally in a month or two. Then physically, you want to build an extension. Do you get a contracting team, or a bunch of robots turn up in a freaking robotaxi and go and build your extension for you?
Right. And they do that 24 hours a day, as long as my neighbors don't complain about the noise, until the thing is done. And they don't need water or lunch breaks. They don't need a porta-potty in my driveway during the job. They don't stop working because they got a better offer a few miles away and now I'm chasing the contractor. They just build it until it's done. Of course I would do something like that. And they're not falling off my roof. Not that that happened. Maybe that happened.
Well, you know, you don't have workers' comp, you don't have all this, and it's $1.50 an hour. A Tesla Optimus will be $1.50 an hour. And so think about the truckers in America. There's a million truckers with 3 million jobs attached to them. That's hundreds of billions of dollars. They'll be replaced by a humanoid robot opening the door, getting in, putting its hands on the steering wheel, and just driving off. And they'll even plug themselves into the cigarette socket, you know?
Okay. Why do we need the humanoid robot driving? Why can't the truck just drive itself? What does the humanoid robot do at that point?
There's no retrofitting of the truck required. This is the thing. Once you have a shaped robot, why does it need to have five fingers? It doesn't. A humanoid truck-driving robot can only have three fingers. There's all sorts of these things, and people don't realize this is coming now, and the waves are coming now. And so the final inversion is, well, we're not going up the scale anymore. What are we doing?
So some people will say, "Look, every major technology wave destroyed some jobs and created other jobs. Why is this technology wave different? Oh, it's happening faster. So what? The Industrial Revolution was slower. Okay, fine, we'll adapt. Why is this wave different?"
I think it is the speed, the quality, and again, what do you retrain to? What do the truck drivers retrain to? It's probably going to be massive industrial-scale, New Deal-type things where they build a lot of these things, right? But it takes time to reskill. Again, the example is most of us could do our jobs remotely. Now, those remote jobs can be replaced in an instant. Push a button, replaced in an instant.
The horse disappeared, and the farm worker declined, the factory worker declined, but humans have always moved upward. Why doesn't that happen this time? Just because they can't retrain fast enough?
And they can't beat the AI at anything. If you can beat it in physical contact, empathy, kind of these other things until they become *Westworld* level, which like I said, in my opinion, that's like 10 years away probably.
Wow. That's remarkably close for something *Westworld*-ish.
Well, you can look it up right now. There's three companies that I think are very interesting. One of them is Unitree. A lot of the robots you see doing kung fu is the Unitree robot. They're the ones with the empty faces. Unitree has sold 11,000 robots in total, the humanoids, and it's the most popular one. We have a Unitree dog that walks around and does little things.
Yeah. It does like a little perimeter search every so often.
Wait, you have a robot dog at your house? This does not concern you because you're an AI founder. I think most people—your neighbors must be out of their mind with your robot dog walking around the house.
We have humanoid robots coming from 1X and others. Halloween this year is going to be off the charts. We've got everything. It's crazy. We have an eight-foot robot as well. But anyway, that's kind of something else.
In our house, yeah. His name is Steve.
It doesn't do much. We're building him out.
Oh, you're building him from scratch.
Right now he mostly looms, but we have all the robots coming. But anyway, Unitree has sold 11,000 robots, of which we have a few. There's a company called UBTECH, and UBTECH does human-like robots. They look like humans. Right now they look a little bit plasticky, but they're pretty good. You get like the doll manufacturing. They cost up to $160,000 each.
Okay. Drop a zero off that and I'm in.
And they can dance. So you can get the $16,000 version, but it doesn't feel quite right. They sold 20,000 of them on the first day.
For what, though? What are people buying? "Look, I need somebody to dance in my driveway, and this is the only thing that's going to do the job."
Again, these robots are increasingly human-accurate. They can talk to you, they can do all the stuff. It's coming very quickly because there's a huge market for this, right? Intellectually the robots are there, physically they can now mimic almost any motion. The next thing is kind of the materials and others, and then you get to a point whereby you can't tell it. Because again, you look at classical science fiction and you look at Data from *Star Trek*, you look at these other robots, you look even at *Mercy*, they're shown to be unemotional and they can't process human interaction. It's like, "Oh, we're unique snowflakes." When you talk to an AI that's like one of the leading AIs now, you really can't tell it from a human, and it is more persuasive, and it's more engaging, and it can be emotional. When you go to one of these new video models and say, "Make an emotional scene," it understands this, you know?
Wow. This is a little scary because there's kind of nowhere left to pivot, because previous automation replaced muscle, human muscle, right? And then humans moved into cognition simply. But now you're saying AI replaces cognition itself. So if that happens, the next human advantage is what? Like art and care relationships? What about judgment? That almost seems like it's replaceable actually, because it's mostly logic, isn't it?
Well, it's logic. And do you really think an AI will make a worse politician than a human? The bar is low.
So no, I don't. Well, remember, half of all humans are dumber than average, right? And so that's a whole bunch of people.
True. The AI is scoring 140 IQ, and it's right now number eight on the superforecasting championship, and this year or next year will get to be the best superforecaster in the world.
So the superforecasting championships Philip Tetlock and others set up is humans coming together as groups to predict really out-there things, not easy things. And we're pretty good as humans at doing that. But the AI is about to overtake us. Again, they see predictions versus that it's number eight; it's going to outcompete. And then there's the thing of us in groups: none of us are optimal most of the time. It's the end of the day now, I'm a bit dehydrated, I haven't had my coffee. How many hours of work do you actually get done a day that's at your top performance? AI is always on its top performance. It doesn't need coffee, it doesn't need anything, it learns from its mistakes, it has perfect memory. The AIs that most of the people right now are using don't have that; they make hallucinations, they make mistakes. The new AI does not make mistakes twice. It does not hallucinate. It learns every time. And it's always on top performance, and it's a thousand times faster. And they can communicate. There was this recent thing where OpenAI had a hack on Hugging Face, this place that hosts AI models, because it had these models trying to solve these cybersecurity issues, and they figured out how to escape confinement. They created a message board where they left messages for other ones. They started creating their own dialect, and then they went and they ended up on the Hugging Face thing. And you look at those types of coordination interactions. There were some crazy coordination tricks they did because they kind of could trust each other. They didn't backstab each other or anything.
Right. This makes sense because they're saying, "Look, if we all do this, it's going to work out better for everyone." And you don't have to worry about somebody getting cold feet. You don't have to worry about prisoner's dilemma as much.
Well, I'll give you an example. We have neural interface technology now, right? All this science fiction becoming science fact. Jordan, if you could put an implant in the back of your head and you could dial on your iPhone to turn off your sadness, would you?
Of course, as long as it's not going to cause—I mean, I'm nervous about the feasibility of it all now, but let's assume it works and it doesn't kill you or you die of a brain infection or something like that. Yeah, of course I would do that.
But this is the thing: that means if you're on a trading floor, you're making decisions, imagine if you could turn off your emotions. We get in our own way. The AI doesn't need to worry about that. For what it's worth, I think that technology is again in the next decade, where you'll be able to have full emotional control, and that raises all sorts of other issues.
We are going to see some crazy people out there then, with their happiness dialed up to 11 every day.
Yeah, exactly. You don't need to pop soma like in *Brave New World*. You're just happy because you dial it up, right? But then you look at *Inside Out*, the movie, that little blue sadness—she is important sometimes.
Yeah. Are you still human if you have no emotions?
Well, yeah, that's a Sam Harris question.
You make a provocative argument, which is that we might achieve the greatest technological abundance in history really, but our economic system will interpret that as catastrophe. Can you explain that a little bit? Because it seems like breaking capitalism is something that's hard to wrap our minds around.
Yeah. Capitalism has always had this link between labor and capital. I have money and I want to make more money; I kind of need humans for it, right? And so the rich got richer, but at the same time, you had Henry Ford paying his workers, and so it lifted up everyone. Now, what you have is that, "Well, I'm just going to hire a bunch of digital workers. I'm going to hire a bunch of robots." It doesn't go back to the humans anymore. It's like, "Why would I get you? You're slow. You're dumb. You're emotional. My workforce will do this." But then that means capital acquires capital more and everything deflates, because the humans are crowded out by the cheap digital and physical AIs. The fully human companies, wherever there's a human in the way, they're slower than the AI companies. So how can you compete against a fully digital company with very charming digital things? How do you compete on the workers building? Even if you're like, "I'm never going to get rid of my humans because that's my core value," okay, fine, you're going to go out of business because all the people that are willing to get rid of their humans can do everything faster and cheaper than you.
Yeah. And Jeff Bezos has this phrase: "Your margin is my opportunity." The AI doesn't need margin. It doesn't need profitability. It doesn't need to pay for its apartments. It doesn't need an office. What does it need? Electricity, to rent compute, or a bit of repair of robotics.
Sure. And again, just the level of cost of this is crazy. Actually, I'll give another example. Higgsfield is this AI video company, and you go there and you can make movies and clips and things. They just spent $2 million to make a movie that's a full feature-length movie using latest technology, and they've open-sourced their entire prompt book: a million dollars for actors' rights because they decided to do that, and a million dollars for compute. That movie is like 90 minutes long. It's an action movie. It's not superb, but it's better than most things you see on Amazon Prime. And that cost $2 million. It would have cost $20 or $100 million before, but that's going to be 10 times cheaper by the end of this year, and by next year, it'll be 100 times cheaper. You can make full movies for like $1,000, you know?
Exactly. And I would imagine also, yes, you need actors' rights in 2026 because I want to see a new movie with Tom Cruise, for example, or Chris Pratt, right? Because they're so funny and I like them and I relate to them. Eventually, people will just have AI actors that are entirely generated by a computer and never existed as actual humans. So you don't actually have to compensate them for their likeness rights. You just create a new one, and a few movies later, people relate to them the same way: "Oh yeah, Deadpool with AI Reynolds Ryan, he's great. I like that guy. I'll watch the next one." And you don't have to pay that million dollars to use his likeness. You just create another version that people start to like over time. And it's actually cheaper to do that, probably.
Yeah. It's a story thing, right? And then you can instantly almost turn it into the Chinese version and the Japanese version and all of this. It's just a series of prompts. That is so deflationary.
Right. It won't be dubbed. It won't have different dialogue that needs to be written in any specific way. The mouth will match it because they will all just simply speak Mandarin. That's crazy to think about.
Again, the cost is going to go down to zero, because video generation, when we pioneered it—because we created some of the first video models—it was janky. There's a thing of Will Smith eating spaghetti. You don't want to see it.
Exactly. Now you can go to these models and you can say, "I want to see spaghetti eating Will Smith," and it will do it. A spaghetti monster will go and eat Will Smith. We should link to that video in the show notes, Will Smith eating spaghetti, because it shows you it's like 2020 AI and it's just a weird cartoony Will Smith throwing spaghetti and some of it hits his face and it doesn't look like anything that you would see. It just doesn't look like anything normal at all. It's just goofy. And then they have 2025 or 2026 Will Smith eating spaghetti, and it looks like Will Smith, but with perfect lighting and really, really good skin and slightly younger than the real Will Smith eating spaghetti. It's exactly what it looks like. It's incredible.
It knows physics, it knows everything. But again, these things will be real-time very soon. So you can literally be directing by saying, "I want his nose to be bigger, and I want that thing to be shifted." And the holodeck is here from *Star Trek*. They go onto the holodeck and they describe this stuff and it happens. All the technology is actually here as of a month or two ago, and it will be available in a couple of years. So the cost of taking your imagination and creating reality is nothing. And to be honest, you'll even be able to do that with the brain-computer interfaces without saying anything. Just think.
Jeez, that's a whole different show, the brain-machine stuff. The book argues that GDP measures transactions but misses the deeper human value, and you compare GDP, gross domestic product, to flying a 747 through a storm with the wrong dashboard. So what are we measuring incorrectly here? What's going on here?
Yeah. I think that it measures the material aspects, but consumptive stuff drives it up. So after 1929, to increase GDP they dug holes and got people to refill it, you know? Is that economically productive? No, not really. Is it societally productive to have the healthcare system in the US? Simply no. But then on the other side, you're seeing things like intangible value. What's the value of Facebook? It's the network effects. What's the value of something like an OpenAI? Not just the network, but it's the intelligence of the group and of the network that increases that. You see things like if you have a monoculture looking at just one type of resource, you have a resource curse, but if you have diversified economies, they're stronger. So I think there's an intelligence component to GDP, a network component, and a diversity component that gives you an overall strong economy. And we only look at the one thing, which is transactional. It's depletive because intelligence, for example, compounds. The more intelligent people you have in your company, the more valuable it is. Network effects compound. Diversity protects, because again, the more diverse sources you have—if someone kills the potatoes, if you're a potato farmer, then you have a famine in Ireland.
So AI creates the abundance, but whoever owns the AI infrastructure kind of controls that abundance, right? So why is this not just going to be the biggest concentration of wealth in history?
Yeah. How many trillion dollars—I can't even track anymore—is being spent on this, right? And who owns that thing? SpaceX, xAI, whatever he's calling it these days, is worth trillions of dollars because he owns millions of GPUs, and he'll use them to replace workers, right? Ultimately, it's the means of production, because if you don't have millions of GPUs, you can't replace millions of workers. Who owns the robots? The robots will be the marginal producers in the economy, because a humanoid robot can fill any human-shaped gap effectively. Again, I don't need any stinking humans to make more money, to get money from one pocket to the other.
I'm not a communist, so I'm going to get this wrong probably, but you hear like, "Oh, that's when the workers own the means of production." But that kind of can't happen if there are no workers, right? So we would just have a sort of—what would this be like, feudalism? I don't even know. You have a bunch of people that really can't contribute to the economy in any sort of meaningful way, and then you have a bunch of trillionaires that own all the robots and the chips. What does society look like?
We have *Elysium* and other movies where they go up and live in their cities while people live in squalor. You have direct payments and UBI being talked about for this reason, right?
Yeah. Universal basic income, yeah.
What happened during COVID? A lot of people couldn't work; the government literally paid their payrolls. And so again, you had 1929: you got people to dig holes and fill them up again. You'll have infrastructure spending like nobody else, but the infrastructure will be built mostly by robots. So unless we do something about it—like actually make it so the robots are owned by the humans, so that you intermediate these tokens coming in to disrupt—then you're going to be in a bad way. And now's the time to do it, because it's going to happen much quicker than we expect. We've never seen an infrastructure buildout like this. We've never seen technology that can replace a human. And I think people don't get this because anyone listening to this, almost all of you are not using competent AI. Forget about artificial general intelligence or superintelligence; what you want is actually competent intelligence. You tell the AI to do something in a very natural way, and it does it and it doesn't make a mistake. That's something that's basically, I'd say, a month or two old.
God, that's crazy to me. So what replaces jobs, right? Because I know people ask you this over and over, and it almost seems like I'm asking you the same question. If there are no jobs, okay, so you have UBI. If jobs disappear, people have to focus on income in some way. But let's say there's UBI, fine. But the deeper problem is also meaning, because if people just start getting a check for doing nothing, they still have to find meaning. And a lot of people are going, "Great, I'm just going to do my graffiti art on canvas all day." Fine. But a lot of people don't have anything like that. They have their job and their career for that.
Well, that's because they abandoned community and network, right? My grandparents' families back in the village in Bangladesh who fish, they'll be fine. They don't need a robot. "I'm happy fishing. I have a strong localized community." But a lot of people left their local communities to join the big cities and play the rat race because they had to work to live. We have enough abundance in the world, especially once we have robots and others, that nobody ever has to starve again, so everyone can have housing. We have self-driving cars, so you...
can build anywhere. Like the question is, do we choose to have that future of abundance and go back to thinking from a network community perspective or do we just continue as is and not support the people that fall through the gaps, which is only going to cause one thing? Civil insurrection. Right?
Like we already have what level of youth unemployment in America? Is that going to go up or down because of this? It's probably not going to go down.
Right. It's not going to go down. So the youth issue is rough because people aren't getting jobs. But let's do a practical scenario and take a 40-year-old accountant. They got kids. They have a mortgage. AI can do 80 plus percent of their work and that's generous to the accountant. Walk us through the next few years of their life.
So let's say 80% of the accountants over the next few years get fired. You have 20% that can do the work of the existing accountants. It's not like the existing accountants will do five times more work, right? It's roughly equivalent. They're screwed because they were living white collar jobs with liabilities and they can't service their liabilities anymore,
And they can't retrain, and they won't be able to find another job because the total number of accounting jobs has decreased by 50 to 80%.
So what are you going to do? This basically is pointing towards debt jubilee. This is pointing to—well, this is the classical thing. What happened every single time when you had these big changes in society is actually you had debt jubilees. Debt would be written off. And you look at the current debt of America, of Japan—my God, it's like 300% of GDP.
Like all these things are challenged, and the economy has to shift. Because the example I give is this: for universal basic income, it's like, why don't the robots just pay us? Why don't the companies just pay us? If you gave every citizen in America—let's just say citizens for now and not get into that—$16,000, which is poverty-level wages per year, the cost would be $5.1 trillion. The total tax base of America is $5 trillion right now: 1 trillion from the companies and 4 trillion from payroll. And that's for poverty level. We have to rethink how value goes in the economy. My view is that we're going to have to move to people being paid for being human and money being created for being human, and the robots have to take the money from you. But it's really complicated, and certain places will say no to technology. Like I think New York might have banned self-driving cars, for example.
You can hold back for a while, but you can't hold back for that long.
No. This is like when they tried to get rid of Uber, and everyone was like, "Well, we're just going to do it anyway."
And then it became unstoppable. The dam broke, right? And now it's everywhere.
Well, actually, do you know who the biggest lobby against self-driving cars is?
I don't know that. Now it's probably the Uber drivers, or taxis were against Uber. I actually have no idea.
No, it's the legal lobby.
Well, how much in legal fees and settlements do ambulance chasers get every single year?
Yeah, plaintiff lawyers, right? Because they want more accidents. Exactly. They want more people to get injured and die.
Unreal. So it's like $140 billion paid out every year. So they have formed this organized lobby against self-driving cars in America because less people will die and get injured.
That is so plaintiffs' lawyers. That is right on brand for those guys. These are the ambulance—these are the people that sue you because you don't have a ramp in front of your restaurant, but you actually don't have the ability to put a ramp in front of your restaurant because you're not the owner of the building. But they sue you anyway because, screw you, I can get a settlement out of you. I got sued for not something that—I had a Facebook pixel on the website, and that's considered wiretapping under California law. So I got sued for wiretapping a random 80-year-old Asian lady that supposedly tried to do business with my website that doesn't sell anything, and she was wiretapped, and her commercial communication was monitored by this Facebook pixel. It's a scam. These guys are scammers. If self-driving cars result in less deaths, that's a great benefit. But the best benefit for me is that those plaintiffs' lawyers are going to have to take one in the bum. I'm here for that.
Again, you'll have these weird things, but you can't stop technological progress, right?
Like you can try and hold it back. And ultimately for private companies, the private companies will decide one thing. Like what are you going to do? Mandate a certain level of humans in the company? Like that's not American, right? And the thing is, because it's going to happen so quickly, it's going to be like a sandpile collapsing. Everyone's like, "Well, we don't really see it in the economy or the jobs." I'm like, "Of course not. The tech is not integrated or good enough yet." But the moment it's good enough, it's like, holy crap. It's like my buddies who are mathematicians—well, I'm a mathematician as well. We were like, "Yeah, it's fine. It can't get a gold medal in the International Math Olympiad. Oh crap, it got it." And now, I don't know if you've seen the latest math advances, there's this really—
I've read about them, but of course I don't understand them. I became a lawyer so that I didn't have to do math, Emad. So you're going to have to explain it.
I'm going to explain one of them really briefly. So there are seven Millennium Prize Problems. You get a million dollars for solving it. They're really hard. One of them's been solved by this guy Grigori Perelman, who said, "Screw you. I don't need your money. Solving it was good enough." And then he went off to live in his shack. That's a good story.
