Next year it becomes really usable, and then a year after, you can push a button, replicate your entire workforce—their entire mannerisms, all their knowledge—and you won't be able to tell it's a human or an AI on the other side of the screen. So your economic life expectancy ends effectively—
In two years. Next year, on the current pace of robotics, Tesla Optimus or equivalent robot, a buck fifty an hour, can get into a truck in America and just drive it off, and that's a million American truck driver jobs gone. So what we did is we made people think of a bottle, for example, and then we did an fMRI, and then from the MRIs, we reconstructed the image using our image-generating software. You look at not being able to lie to an AI because it can tell your facial structures, the words, and we can even read your brain. We will never get access to the top AI anymore as of now.
It's done. Like there's AI for the media and AI for the top labs. It makes no economic sense for them to give superhuman AI to the hoi polloi, you know, us plebs.
Anyway, man, listen, nice to meet you. Thank you for coming and doing this. I'm very intrigued by this idea of our economic life expectancy, because there's a lot of people worried about their jobs and what's going to happen to the economy. And then we've got Elon Musk telling us we're going to have universal high income. Can you explain this concept of economic life expectancy?
Yeah. So, first, thanks for having me on the show. We have this thing of we're trained to do a certain task, and we're kind of turned a little bit into robots by our system: do the task, have it verified, etc. COVID really showed up what that was when we ended up doing a lot of this stuff remotely, but we're reaching a point now where AI is becoming actually competent intelligence.
Next year it becomes really usable, and then a year after, you can push a button, replicate your entire workforce, their entire mannerisms, all their knowledge, and you won't be able to tell it's a human or an AI on the other side of the screen, except for the fact you pay a hundredth of the value, and a human will make that team worse because they go to sleep, they make errors, they wake up a bit grumpy when they don't have a coffee. So your cognitive value goes negative. So your economic life expectancy ends effectively—
In two years for any job that can be done remotely. A few years after that for the jobs that can be done in person, because, you know, we've got to build the robots a little bit better.
Right. Okay. And has anyone done any mapping of what kind of percentage of the workforce that is?
Yeah. So we've had some from McKinsey and OpenAI. They estimate about 50% of all tasks. I think that it will take—it's not like everyone's going to be fired overnight. It's not nice to fire people, you know, but it's going to be very difficult for companies with humans to compete with companies with fewer humans and fully AI companies. Again, particularly in the digital space if your job can be done remotely.
So, lawyers, accountants, a lot of that work?
We're already seeing it, like junior hiring in law firms is decreasing because they're like, just push a button on AI. Like what is the job of a partner? It's a relationship job and it's a liability job. And so it's going to be the entry level that gets hollowed out first, because those that own the assets, the knowledge base, the relationships, and others can use compute to leverage themselves up more and they don't have to deal with annoying grads.
So it's like a ladder: we've taken off the first rung, we're going to come for the second rung until we take the whole workforce.
Until we take the whole workforce, until again we get to a point like next year. Next year, on the current pace of robotics, Tesla Optimus or equivalent robot, a buck fifty an hour, can get into a truck in America and just drive it off, and that's a million American truck driver jobs gone, plus three million people that service them.
Hold on. Do you even need the robot to drive it?
Well, this is the whole thing. Like, you could install a whole system. Well, what I'm saying is like the first humanoid robots are arriving like in a couple of months—the 1X and others. The way they deliver themselves is they take a taxi to your house and they ring your doorbell.
Yeah, it's the best way, right? I mean, why would you need a box? Like the Optimus robots will get their Tesla cars, but the fundamental thing is this: having that truck retrofitted, that's a whole hassle. Humanoid robots, again, a buck fifty an hour, can fit in human-shaped holes. So there's no retrofitting to be done. Everything, literally open the door, get in. That's a truck driver job gone. A lot more physical jobs will go as well because robotics has had the breakthrough. But the digital jobs right now, those are in trouble.
But 50% of the jobs in two years, it is kind of overnight really when you think about it.
Yeah, it's like a sand pile collapsing, right? Like the pressure is building. We're not seeing the first little glimmers, but again, no one really likes firing people. But now you go to the Philippines and places like that, do you know what's happened to the call centers? Why would you need a call center? You use Grok, you use OpenAI voice, right? Those jobs are never coming back.
So do we know how many of those jobs have gone in, say, the—
So we're starting to see that crack. I can't remember the exact numbers, but we're just starting to see it crack, again because they have service contracts.
Again, this isn't an overnight thing. So your job can become economically irrelevant, but again, you might not want to be fired because firing really reduces morale, or you could be working in the public sector where it doesn't matter at all.
Okay. But if we're losing 50% of jobs and we're not replacing them, but eventually we're going to need another economic system, we're really talking about what is just going to be a volatile period where the world figures out what it's going to be from where it was right now.
Yeah. Well, in my book, The Lost Economy, I said we've always inverted. Like first of all, it was like your land for subsistence, then it was the people that you kind of owned as serfs on the land, then it was factories, and then we kind of moved up to the information age. You always inverted up the job ladder. You had some aspirational goals. Now it's like, okay, we hit intelligence, but the AI is smarter than me, it doesn't sleep, it learns from its mistakes, and it's a thousand times faster than I am. Crap, what do I do now? Like what do those truck drivers do?
I think what the governments are going to do is just like 1929, they had the New Deal: dig holes, fill holes, massive infrastructure programs, right? Again, the robots can do it better. Like you'll have a robot team doing your extension in your house in five, ten years far better than a human, but you've got to get those humans working and sweating. And again, it's probably going to be infrastructure; it's not going to be high-paid jobs.
I was just out in San Francisco and had my first experience of Waymo, which was weird for about a minute and then just normal.
You get used to it so quickly, right? Pick up your phone, hedonic adaptation. Like first we're like, it's going to crash, like the steering wheel's moving, like whatever, right.
I interviewed the guy who invented it, one of the founders of Google X, Sebastian Thrun—
Sebastian, sorry, apologies. What he said to me, the great thing about like we learn from our mistakes, he said every one of those Waymo cars learns from all of each other's mistakes.
Yeah. If every car in America was as safe as a Waymo, I calculated there'd be 80,000 deaths less a year and about $400 billion less in damages. Right now, the biggest lobbying group against Waymos are lawyers, because you know those ads you see—
When you get to the airport.
Yeah, like "sue these people," "call 1-800-ACCIDENT." It's like a $200 billion industry that will be killed by self-driving cars, so they prefer to kill people and lobby against them.
1.3 million deaths a year, he told me, worldwide from car accidents.
And look, a lot of those are in developing countries with poor roads and infrastructure. But like I said, my calculations are 80,000 less deaths in America a year.
Do you think they'll make human driving illegal because it's too dangerous?
Well, of course they will. It's like giving a kid a supercar; the insurance premiums are higher. What's going to happen is it's going to be new, then novel, then normal. And then if you don't have this, then yeah, your premiums will be higher because your car is more dangerous, because the liability profile is more. But it all starts with AI being a checker and a verifier. And then if you're not using AI to do a second diagnosis for a medicine, if you're not using AI to stabilize your driving, then yeah, your insurance is going to go through the roof.
But if we're talking two years and 50% of jobs going—
50% of jobs becoming economically irrelevant.
Sorry. Because there'll be new jobs created, of course.
No, just because it's annoying firing people, that's why.
Well, yeah, and employment rights, and paternity, holiday, sickness—all those things.
Yeah. You'll see the owners of the companies hovering over the "replace all workers" button, you know.
Yeah. Well, you say that there's certain things like, I don't know, I don't really want to get served by a robot in a bar.
Well, like I said, the human part is going to disappear. But if you look again where robotics is, it's had a massive increase just over the last year, just like AI models, whereby now they can read a recipe and they can make any recipe, for example. They can talk and they can become charismatic. And there's a company UBTech—have you ever seen UBTech? No?
So all these Chinese robot companies, this one's one of the most funny and disturbing. Unitree, which is the main one that you may have seen, where there's a gap for the face and they're doing kung fu and things, the silver robots, has sold 11,000 robots in total, humanoids.
Unitree announced their new UB1 robots, which are basically like, let's say they look like humans.
Kurt, can you bring this up?
Right, so kind of humanoid robots.
It's UBTech, UB Space Tech humanoid robots.
So they announced, and they've got lovely lustrous hair and everything. They don't look quite real, but on day one, they sold 20,000 units up to $160,000 per robot. So it's called UBTech—no, it's a humanoid. UB Space Tech, and then say "lifelike." That'll do it. Oh, like that.
Yeah, U1. Yeah, that's it.
Hold on. Is it like UBTech lifelike? Go to that one. That's UB's lifelike humanoid. You can see the two ladies there together.
Yeah. So they're lifelike companion robots.
Well, you know this technology. That's why up to $160,000 per unit of these guys, and they sold 20,000, so they outstripped the top robot manufacturer. But all of that technology that was used for companions previously, now if you look at Clone Robotics, they've replicated the whole Westworld intro. They've got real tendons and things like that. I think you'll have human-realistic robots within 5 years, 10 years max. We did a screening of Blade Runner for our AI movie club and we were like, "Yeah, that's coming."
Hold on. Can I ask a question? How lifelike are these?
These ones, you can tell it's not a human, but like if you look at the Clone Robotics and the way they go—
It's a full—I don't know about all the parts.
Yeah, but because I just know there's some people that—that's why they're going to buy them.
Well, of course they are. Again, this is going to be big in Japan and many other places, right? But the thing is, when you're talking to your AIs now, like you use voice mode in GPT, AIs are more persuasive than any human already in studies. And then you pair that with something that's reasonably lifelike, and then you look at the economic incentive to make it completely lifelike. Again, Clone Robotics moves just like a human. These things can dance. You can't tell.
We're living in the future that we were promised in all the films.
Science fiction is becoming science fact, even the bad science fiction. Like imagine what Charlie Brooker is thinking now looking at Black Mirror. He's like, "This wasn't meant to be a guide."
"It's meant to be a warning." Well, I put down a deposit for one of those 1X Neos when I first heard about it, but I've been told they are controlled by people remotely.
Yeah. So they have remote control because they hadn't solved the operating model gap until about a month ago.
Yeah. So this is the thing: model capability just suddenly once it gets through a level—like we saw it with mathematics. Like I'm a mathematician, studied at Oxford. It's like, it's not good enough for math; oh crap, it's solving all these conjectures. Now what are mathematicians going to do? Because when you got to GPT-4o and Claude 3.5 Sonnet, the latest OpenAI and Anthropic models, it was suddenly good enough to do math, and it can do just about all math. It's suddenly good enough to look at a recipe and make it. It's suddenly good enough to fold your clothes and do laundry, or cut your hair and things like that. And once it can learn something, it can learn everything.
Actually, it's quite cool. Now, again, when you go to the Valley next, there are companies that will give you free cleaning services. They send a robot, or sometimes a human in with a camera, and they're doing your cleaning for free to train up the other robots to do it. And once one robot learns it, every robot learns how to do laundry.
Privacy is going to be an issue here. Look, I really want one. I saw this Neo emptying the dishwasher and I thought that's kind of cool. So I was like, I want one. But everyone in my family has been like, "Can we not have this?" Like my son, my daughter, my wife—they're like, "We don't want this here."
Yeah, especially if it's teleoperators. I mean, I feel the same sometimes with my Alexas. Like, I swear they're like, "Do you want to buy something?" every so often, listening in. I don't care if I put it on kind of pause. But again, it's remarkable how quickly we give up our privacy for convenience, right? And so this is where we have to be careful about our kind of almost cognitive colonialism of these things coming in and persuading us. They're going to be our kids' best friends, you know. They're going to be our companions. In some cases, really going to be people's companions.
They are. No, they are. Have you seen that film—what is it? It's the Ryan Gosling one where he has the sex doll that becomes his partner?
Oh, but anyway, everyone comes to accept her because they accept him, right? And he gets married to her. There are people out there who are going to have one of these as their partner, girlfriend, wife.
Yeah. I mean, I'm worried about guys because we don't listen, and the AI is infinitely patient and listens. The women will just give up on us. What's going to happen?
We had that debate in Italy. It's like, what will life get rid of first: would it be the men or the women? We came to the conclusion it'd be men, because we cause all the wars and also we don't have ovaries.
So we're not required, basically.
I mean, it can open the doors and it can listen. So, again, why were we needed?
This show is brought to you by my lead sponsor, IREN, the AI cloud for the next big thing. IREN builds and operates next-generation data centers and delivers cutting-edge GPU infrastructure, all powered by renewable energy. Now, if you need access to scalable GPU clusters or are simply curious about who is powering the future of AI, check out iren.com to learn more, which is iren.com.
I do wonder, though: how much autonomy and how much AI are they putting in the robots themselves? Is each robot completely autonomous? Can it learn on its own?
So, again, that technology has just been achieved where the robots can learn on their own, on-policy on-robot, as it were. This is just like you've just seen the latest Qwen models achieve frontier-level performance nearly, that can work on a MacBook. Like, wait, what? Most robots are still teleoperated, or they're simple; they don't have the self-learning algorithms. But it turns out you don't need much. The analogy is self-driving cars: if you got a Tesla FSD, it was crap; now it's great, you know.
Same with robots. But we will have a time when a robot kills a human.
It's coming soon. And some of these robots are dangerous. There's a good one: the EngineAI T800.
By the way, did you see that tweet from Arnold Schwarzenegger this week?
It was Figure Robotics. They were saying, "How should we get rid of our original number one robot?"
And Arnold came in and said, "Melt them." Hold on, the name sounds like something from Terminator. What is it called?
Hold on. Wasn't the T800 in Terminator? It was. Are they trolling us?
So the EngineAI T800 knows kung fu. It can punch five times harder than Mike Tyson. It's a 70 kg, six-foot robot.
And you look at this thing and you're like—
"That should not be allowed to walk the streets."
Hold on, but that's graphics. Does it actually exist?
That's the actual video. You can buy one today. It's $30,000.
I can buy one of those today?
Yeah. It's being used in the robot fighting championships.
Yeah, so they use these ones.
One punched another one's head off. So obviously it can punch a human's head off. The previous one was a cute little five-foot thing; this thing literally, its torque is five times the punching power of Mike Tyson.
Yeah. And what they do in the fighting competition is they actually do shadow fighting, so it picks up your biometric movements and then it mirrors you just like in Real Steel with Hugh Jackman.
How much battery life do they have?
They've got about two hours of battery life, so their chest cavity is a battery.
If I had that and he was sat here now, could I just tell it to go and beat Kurt up, and will it?
If you put the right algorithms, yeah. And again, there's no laws or regulation against these things. These things will be walking the streets very soon.
Is there a buy button there?
"Contact," yeah. But like I said, you go—they have an EngineAI store you can go to. You can pick them up.
As a bodyguard, exactly. That's one of the things they said. Or as a police force, you know.
Well, it's like Unitree. I don't know if you saw the latest Unitree one.
No, go back. What the fuck is that?
The JS01. That is Black Mirror.
Yeah, the JS01, I think, runs at 25 mph, if I can remember correctly. That's Metalhead. That's Metalhead.
Metalhead. If you look at the latest Unitree ones, the latest Unitree ones run faster than Usain Bolt and they can do two-meter jumps, the latest humanoids.
Another company has a version of this that a human can ride on. So yeah, it's coming.
That looked like it was carrying medical supplies.
Yeah, gaslight us: medical supplies, flamethrowers, all sorts of things. It's like when you see the drone shows, you're like, "Those are wonderful!" and they'll be used to invade Taiwan, definitely.
Oh my god, these are wild.
They are wild. Have you seen the Unitree running one?
Is that the one that was this week? It set the record.
It set the record. Like you see that thing just pelting it.
But it ran into the wall.
And then it blew up. Yeah, it blew up.
Yeah, there it is. Kurt, go back. Either of those, the first or the last one. Have you seen this, Kurt?
It just runs into the wall.
Yeah, there's another one where it runs into a thing and then it bounces off and staggers back and then blows up. That one's even cooler.
I don't think we have fully grokked what this means. So as far as I understand it, we've got humans building robots at the moment. When the robots build the robots, that's when it almost becomes the point where it becomes really cheap and recursive.
No, they've already got that. Someone like Xiaomi, they used to make phones.
Now the Xiaomi robot is about to be released. They have the EV. Their factories are dark factories because there's no humans, so you don't need to have the lights on.
You don't need light. I've heard about this.
You don't need light; just churn these things out.
Okay, so you can buy these things now.
How many companies are making these robots?
Oh god, I think I counted at least 80 companies making these robots.
How far are we away from now—do you know like the start of I, Robot, where the robot's running trying to deliver the post, and everywhere you look it's almost like half the people are humans, half the people are robots? How far away do you think we are from literally we finish this interview, walk down the street, and half the people are robots?
Actually, if you can look at the Unitree site just while I talk about this. Robots are about as complicated as a motorcycle, and there's about 70 million motorcycles built a year; there's 70 million cars.
It's only as complicated as a motorcycle? I thought it would be more.
Think about it. It's not that complicated. It's the brain that's complicated. The actual actuators and other things aren't that complicated.
I thought it was the hands.
Look at the Unitree H2. Does that look familiar?
That looks like an I, Robot.
Yeah, it can do ballet. It's got much better than this now, though, the new servo actuators on that one.
So okay, how long away do you think we are from just it being normal to see them walking down the street?
I mean, I already walk my robot dog down the street.
You have a robot dog? What robot dog do you have?
I have a robot dog. I've been programming it with my little personality. That's the little buddy that I have. It's only 1,600 bucks.
I can get that for 1,600?
Yeah. It can dance, it can do flips, it can do everything.
Yeah, if you give it one.
This is not good. We are racing into our own demise here, I'm sure of it.
Well, I mean, this is the thing: like my p(doom) in one of my scenarios is a billion robots and a bad firmware upgrade. And you can see it happening. Obviously, these things are being built terribly in terms of their safety. Like, you've got all those Chinese cameras everywhere, right? And have you ever tried to hack them? No? It's so easy to hack all these Chinese cameras that people install everywhere. They have such big backdoors, so you can just randomly go into people's things. Not that I do, for anyone listening, you know; just I might know people who did. But these things have massive security holes. They have no regulators, they have kind of nothing like that. Deliberately killing a human, it's going to happen.
Accidentally? Definitely soon.
Hasn't it already happened?
I thought it already—can you look at that? Go and have a look in Venice and see if we find that, because I'm sure I've seen that.
Right. Has anyone been killed by a robot? "Yes, several people have been killed by robots. The first recorded incident was in 1979." I want to know recent ones. "A study examining robot-related fatalities from 1992 to 2017 found 41 deaths, with 78% of those involving a robot physically striking a human worker." Yeah, but is that a punch? Search, see if it's got anything about humanoid robots killing anyone. That's what we want to know.
"Based on there, there's been several human fatalities by robots, though these have been primarily involved industrial robotic arms." Yeah, that's what I expected. "Search results also note that police have used robots to intentionally kill suspects, such as the 2016 incident in Dallas where a bomb disposal robot was rigged with explosives to kill a suspect following a breakdown in negotiations."
Wow. So should we be regulating these things?
Of course we should. You should never allow an EngineAI T800 to walk down the street, you know.
Yeah. Just like you don't allow tanks on the street, you shouldn't allow potentially killer robots on the street. Like the Neo 1X that you kind of got is a soft, fluffy robot. That type of torque, where it can't twist off someone's head or accidentally kill them to death—sure, let them walk the streets.
But they still could. The Neo could hold a knife and do that.
It could stab, but it can't punch. Again, it's actually been built to within human standards.
No, you can rip its arm off. It's like a big fluffy doll, and the way that the tendons and everything are generated is such that a human can overpower it. We shouldn't allow these other AIs to kind of be on the street.
The thing is, when you think about life as a kid, right—I don't know if you remember, I don't even know how old you are, I'm 47—I remember being about 18 and thinking about my career, you know, and I remember being in my early 30s and thinking I'd like to retire in my 50s. But we're now in a different place where we have to think about a completely different world that we've only seen in the cinema that's coming now and in the next five years. I can't even picture the world in five years. I don't even know what it's going to look like economically. I don't know what it's going to look like visually. I don't know what it's going to look like socially. I've got no idea.
Do you think most of society just has no idea what really is going on?
No. I mean, I've been in the rooms with the smartest people, the biggest forward thinkers; everyone's like, "I have no idea." None of us do. This is one of the reasons why, when people go down the rabbit hole—people looking at your podcast with my buddy Connor or Roman—it gets a bit depressing because you're like, there's a black wall potentially ahead of us, there's a Great Filter ahead of us. But if we don't have that, then just normally, like, what is my meaning?
My social contract is about to be torn up. It was already under huge stress because of our stupid institutions and other things. But now again, what is the American dream or the British dream or others? Where's my mobility gone? And this is a real big thing because it hits directly to what people's identity has become, which is not "I'm a member of a community," it's "I'm a job, I'm a role, I'm a robot effectively."
But maybe that's where we failed in that we created a world where we needed an identity. We needed to feel part of something; we needed to feel successful. And young kids growing up with TikTok or Instagram Reels were just seeing the lives of these other kids thinking they want it, but very few got to do it. And there's like a despondency. It's like most of us just have to be kind of normal peasants.
Well, Isaiah Berlin conceptualized this in positive and negative liberty back in the early 1900s, right? He said positive liberty is the freedom to believe in big things, big isms and other things, and he's like, "I don't like that, because it causes a lot of stress and war and things like that because these big movements can end up like that." He's like, instead, have negative liberty: the freedom from anyone telling you what to do, which then led to laissez-faire kind of capitalism, a whole bunch of other things, and the brands telling you what to do, identification with the brands, the destruction of the family and the community units and more—not being able to be proud of your British history or American history or other things like that.
And so then the question is: what do we do next? Because if the robots and the digital beings are smarter than us, faster than us, more charming than us, what does it mean to be human? Well, it means our connections with others, it means our networks, it means exploration, wonder, and more. But not many people are actually articulating positive visions of the future.
Well, because we've kind of done the opposite. I was listening to Joe Rogan with Andrew Schulz in the car on the way this morning, and they were talking about how they used to drive down the street and all the kids would be out playing, and Andrew Schulz saying, "Yeah, my mom and dad would just throw me out the house, go play, and you play hockey on the street." I had the same. You'd just go to the woods and you would build stuff. And kids don't do that anymore; they stay in. And if we end up with like Ready Player One, we're going to be even more contained inside.
Yeah. And again, it's who regulates your attention, right? And how is your attention doled out? I think the whole thing is that the system has become hyper-optimized to try and put you into these boxes. And it's difficult to see what will break you up, because how do you build trust? There's very little trust in our communities at any level, shall we say, right?
It's because all the institutions have failed us.
The institutions have failed us because who are they even working for, right? It gets very complicated sometimes if you don't have—like, the best organizations work on a shared narrative, ideally one that you've gone through a little bit of pain and psychological safety. Most people don't feel psychologically safe about anything. You can't say anything because, I mean, until recently you could be arrested for saying something on social media, for example, right?
Again, who's behind that, right? Like who's like, "This is my policy thing, and I'm going to convince everyone"—it somehow became law.
Do you know who Jake Monroe is? He's the right-wing goth guy, he's really based. Try and find Jake Monroe. We were talking about getting him on the show just because it's fascinating to see this guy's full goth but really kind of based right-wing guy. He got arrested at the border, I think it was in Wales recently, about his opinions. They wanted to ask him questions about his opinions.
Yeah, but this is going to be a really dark road, though, because you can't lie to an AI.
An AI can lie to you, though. Yeah, this guy.
We've again had science fiction about thought police and precrime and things like that, but where it's going is like one of the things I was thinking recently with the whole Fable Mythos thing, where you couldn't get access to Anthropic's models, and 100% governments will introduce licenses for AI. And to get a license, you have to convince the AI you're patriotic, and you won't be able to lie to the AI; the AI is too smart for that. And you think about that, you're like, holy crap, where's that going to go? Because they can ask your opinions, but you can lie, right?
But then when we look actually even further, it gets even more dystopian and crazy. So at Stability AI, my former company, we pioneered open source and everything. We did a study called Mind's Eye. Don't know if you ever heard of that one.
No, that sounds like something from The Handmaid's Tale.
Yeah, maybe a little bit. So what we did is we made people think of a bottle, for example, and then we did an fMRI on them.
Explain fMRI. Is it brain MRI?
Yeah, magnetic resonance images. So you sit down and it does your brain: think of a bottle, think of a can of Coke, and things like that. And then from the MRIs, we reconstruct the image using our image-generating software.
And so you look at that, you look at not being able to lie to an AI because it can tell your facial structures, the words, the cadences, and we can even read your brain. And you're like, okay, you can kind of see where that's going in terms of control and the power execution of a government with the monopoly on violence. And the question is: who is the government working for? Again, where are these laws coming from? Where is the debate on this that actually makes a change? And how do you even make a change against these obviously stupid laws?
You're obviously very suspicious of what's running the world right now anyway, then.
I just don't think actually there are people behind it running it. I think it's just an omnishambles where collectively we don't have the ability to build institutions that haven't optimized for the wrong things, which sounds like AI alignment. Our governments and institutions are slow, dumb AIs that have become misaligned and don't work for the people, because they're optimizing for local things that, again, aren't in the long-term interests of the people, because there's nobody with sufficient respect and gravitas that can speak for the people, because they speak for a group of people—
Or for what the status quo is, or an extension of that, what's safe. Because when you deal with uncertainty, you minimize for maximum regrets. It's like, well, if people speak up—and so if Kanye comes to the UK and something bad happens, then that'll look bad on me, so I should forbid Kanye from coming to the UK.
But that is still as bad on you.
Exactly, but it's a different constituent. It's a different kind of thing.
Which raises the question: what happens when there's an AI that analyzes every single piece of policy in the UK and can say, "That's bollocks," or "That's a good idea," and rate how good it is for the British people? These types of things are coming. But then the question is: who's built that? Is it a Palantir, or is it OpenAI? Is it a British company? Like the whole way that policy institutions and others will change because AI will infiltrate it completely. And the question is: whose cognition? Whose AI?
See, I'm very pro the idea of replacing large amounts of government with AI. I think we can root out a lot of corruption. I talked about it on the show recently: I built this thing with Claude where I spent about three days on it, and all I did was I had it go through—it was called Show Us Receipts—I had it just go through every politician's tweets and fact-check them, just see if they're true, and it did a pretty good job. Never released it because it wasn't accurate enough and I don't want to get sued, and in the UK you can sue people.
Beyond that, I started to think, you know what I really want to do is I want to be able to analyze government contracts, see who awarded it, and then go on Facebook and see if there's any connections between these people and start to try and notice corruption. I started to think actually—and there's these other really great accounts using AI to discover things—ultimately, I think we should be able to weed out a lot of corruption, but we also should be able to weed out a lot of useless people. But the question then becomes: well, if government was just AI, how do we decide what we want government to do?
Well, I think, again, this is an articulation of who's articulated what it means to be British and what our hopes and dreams and future are. The policies tend to look at themselves in contrast to each other, as opposed to, again, what is a positive future for Britain, you know? And then that becomes complicated because, again, who speaks to that?
However, we're moving to a stage now whereby soon you will have your Grok agent, your Claude agent, your other AI agents that have access to all these super-geniuses. So the exercise that took you three days will take you three hours right now and probably three minutes in the future, fact-checking all these, right? The ability of people to contribute to a whole, to check every—like why can't you check every government policy and say is it good for the British people as a subjective matter?
Yeah, it's good enough for that now.
But how do you decide what is good for the British people?
Again, the reference measure should be a reflection of the people, so that's actually a proper discussion that needs to be had, because we don't have a constitution here, right?
We kind of do, we kind of do.
I'm very jealous of America having a constitution.
Why don't we have a constitution? Why don't we have that as a directed thing where everyone can give—it will probably look very similar to others.
constitutions. But if you don't have a reference measure, how do you measure anything against? Like we have our laws, and we have our precedent and other things. The AI can analyze that in two seconds.
Sure, we can merge this. What we're lacking right now is an authority that can do that. Like, you know, Sir Demis, fantastic. He's going to solve all disease. Spend maybe a little bit of time on this app and fly all of Google's compute. Come and help us, please, you know? Like we need someone with that authority of trust to do that execution, and then people to be able to contribute to that.
I think what I'm getting from you here is that you basically are saying in the political world we have right now, we are severely lacking people with integrity to make decisions that are for the benefit of the people, not for the benefit of the party or niche groups. We are lacking good politicians.
I think we're lacking politicians that you would trust, aspire to follow. Again, what did Starmer do wrong compared to previous ones? Complete lack of charisma and everything else. Burnham's come in now. What's his actual policy shift versus Starmer? Is he fixing the things that people said that Starmer was doing bad? Not really, right? It seems to be like, well, he's more charismatic and he's got a bit of devolution. Were we asking for devolution?
We were not asking or voting for devolution. I mean, you could argue that 23,000 people in Makerfield maybe voted for devolution.
But again, this is the thing. We have the High Court to check the judiciary, and we can talk about the issues with that, right? It was meant to be that the upper house, the House of Lords, was a check on the House of Commons. It obviously isn't, right, with the way that it is.
But what if we had an upper-upper house, which is an artificial intelligence system that actually checks and says objectively, "This is the logic. It's good or bad for the British people. If you don't like it, debate it," and people could send their own agents to debate it? That actually feels like a better version of the future. Or when you have constituent democracy whereby people have citizen councils where you're taught about a topic, you actually see from studies that if you take people just like jury duty, teach them about a topic, and then get them to give their view, they come up with better decisions, and AI can help with things like that. So I think we have to think about what does politics look like in 5–10 years leveraging this technology, and if we can have a check and balance on not only the corruption, but the stupidity.
It might just move the fight, though. Do you like football?
You don't like football that much.
Trust me, if you keep liking Spurs, you like football.
I mean, you just signed a City player for 86 million. They're selling all their players anyway.
So when we got VAR, we just moved where the argument is.
All right. So we can get pretty accurate on goal lines. Did the ball cross the goal line? Great. It's on a watch. I think we can all agree with that. We still argue about offside. We say, "Well, it's too accurate. Where do you draw the line?" And I think the point is all we've done with VAR is moved where the argument is and changed the person we blame. Rather than shouting at the person on the pitch, we shout at the person up in the studio, but we still argue about the decisions and the role.
I just wonder if AI just moves the argument, because if we get to a point where the AI is saying, "Look, these socialism ideas aren't very good. If you have too much welfare, you disincentivize people. We can factually prove this, we got data to prove this," you're still going to have the Marxists argue back and say, "Well, the AI is wrong," and then there's going to be a revolution against the AI.
So this is the interesting thing whereby something like VAR, again, people are like, "Give it a little bit of margin of error. It's only human, right? It was much more fun blah, blah, blah. His knee's sticking out," or whatever.
The AI can actually go beyond what has classically been done, which was you took large amounts of big data and you extrapolated from that yes-or-no Boolean, almost one or zero, to, "Well, this is reasonably the case." It can be actually quite subjective and quite dynamic. I don't think AI should make the decisions. I think it should be acting as a check. Like right now, if we had a second check of every NHS decision made by a doctor, how many lives would be saved?
Should AI make the diagnosis? No.
Because A, we need to have an interim period where we figure out liability and things like that; B, there is the very, again, human aspect of it. But again, this is like self-driving cars. Why should we not ban all humans from the road? Again, there's a period where that will adjust, where you've got AI checks and balances, and then it moves forward.
There's also a danger in that if AI, again, sets the reference, if it makes all the legal decisions and others, we're all watched over by machines of loving grace. Who runs the machines? What are their inherent biases? Because the way the models are built now, I'll give you an example. If you take a Chinese model—there was a study done by CrowdStrike about this—and you say, "I'm an Uyghur, or someone who doesn't like China," the code it writes will have vulnerabilities coded into it.
Well, I had it recently. We did a show with a guy going into some quite controversial topics, some that people say are conspiracies. Whenever I do a show, I always just put it into Google Gemini and just say, "What do you think of the show and how did I do?" because I want to learn and get better. And it just came back and said, "Do not release this show. If you release this show, you're going to piss off your sponsors and piss off your advertisers." I was like, "Oh, hold up."
So I ran it through Claude, same answer. Ran it through OpenAI, same answer. Ran it through Grok, and it was like, "Definitely release this show. This is who you are. This is what you stand for. Release this show." So there are these inherent biases that are programmed in.
Yeah. And that's why you've got to think, just like you train a kid, you take them to school or university, how are we training our AIs? Where are they going at home? Because right now AI doesn't even have morality or ethics baked into it.
Well, it clearly has a morality baked into it when it's telling me not to release a show.
Ah, it has safety baked into it, because what you do is you train on everything, but deliberately the AI labs don't put any culture or ethics at the start, as opposed to you grow up in your house and you've got culture and ethics at the start. And then at the other end is where they say, "Well, make sure you don't create liability for your user, or definitely for us as Anthropic or other things." And so you take that generalized liberal arts model and you turn it into an accountant or a lawyer or whatever.
Yeah, it's going to be a big challenge to try and get to the point where people can even—because look, we are so divided as a country right now. Even yesterday, I was just having a conversation. I'll tell you the question I asked ChatGPT, because this will tell you where my head is at with this. Let me see if I can find it. It called it "Discussing political anger."
So I put: "I'm thinking a lot about the culture at the moment, the political discourse. Everyone fighting, everyone's got their subgroup and no one can get on and no one can trust the government. It's all lies. Taxes go up, costs go up, spend goes up. It's all just a big mess, and it's seeming like the collapse of Western liberal democracies. I personally don't like it. It's making me angrier. I'm angry at myself for being sucked into it." So I started having this conversation. It's like, how do you fix this mess? We don't have a shared story at the moment. We kind of have two worlds which are led by large, loud subgroups, because I don't believe the majority of the country are angry left-wing socialists and I don't believe a large percent of the country are angry right-wing fascists.
But I think that's how they're categorizing each other. But they're kind of the representation of anyone who votes from their side.
Yeah. I mean, it's sufficient for that to be pushed anyway by the political parties.
Because they know it works. Ultimately, how do we break this?
Well, the biggest lie in the world is that we're not all human, and that's used to create war. And then you have other lies told all the way down. There's a lot that we agree on as Brits. Again, we want to be safe. We want to be secure. We want to have our kids grow up to be the best they can be.
But is that even true? You say that we want to be safe, we want to be secure. I think we can demonstrably prove that immigration has reduced safety for certain people in certain communities, but we've pushed it.
So this is the thing: then you get into the branches and you get into things like that. And then the question is, who decides, right? So you have a withdrawal of immigration, and then a new prime minister comes in and it gets flipped, and then it's used as a flare point. But at the same time, you can see some statistical elements there, right? So the question is, how do we find a common ground from which to build? And there is no center right now.
Do you think people actually want a common ground, or do they want to win?
I think that Brexit was a vote for change, not for Brexit. A lot of political actors have realized statistically and otherwise that's the best way to actually get and maintain power. It's always a question of: are you happy with the status quo?
Does it feel like anyone is right now?
No, because, again, the UK is behind even the worst American state now in prosperity, right? People don't feel mobility. They don't feel hope. And again, we're not really getting stories of hope. We're not getting stories of leadership and more. Every politician says that—like Andy Burnham: "I want to bring back hope." Zack Polanski: "I want to make hope normal again."
They can say that, but what does that look like practically? It's delivering for their voters and not the country.
And not the country, exactly. So this is why it's not generally kind of the whole entirety, and people just don't believe it. This is the other thing, because they're on one side or the other, and they're like, "Well, we're not in a positive-sum mindset," let's put it that way.
If you look at everything that we have in terms of our talent, in terms of our infrastructure, in terms of other things, everyone just feels that the bad things are the things that are dragging us down, as opposed to, "Oh my god, this is the most exciting time in history."
We have some of the top AI talent in the world, for example, and they don't go overseas. They're actually here in DeepMind and kind of other things. What are our champion industries? What are we proud of showing to the world? If you go and ask anyone, "What are you proud of? What are we best at?" they'll be like, "I don't know. Regulation, I think."
Premier League football. It was interesting, the World Cup obviously. Thank God it ended on that little high, but people getting a little bit proud of that. Just like I like the British cricket team now because we actually play positively. We should just be positive on everything. Maybe not the second half, but the first half in that third-place match, we just go Bazball it or Gazball it or whatever. Let's be on the front foot. But we're not really on the front foot internationally, let alone locally. That's why I think people don't believe.
So how do you address that? I think you mobilize the technologies that we have. I think you understand that just like America is about to re-industrialize, the UK needs to re-industrialize. We need to have power going up and not pay a billion percent more than everyone else, and things like that.
But I think that's the problem: we've had two or three decades now of getting materially poorer as a nation. I know the top percent and elites have benefited so much from inflation, but the majority of people are just getting materially poorer, working harder, getting less from the state. The state is taking more, and for some reason, every time an election comes around, they get convinced that this new Labour Party is going to solve it, or this new Conservative, or this Reform. Yet we have clearly a dysfunctional state now that cannot fix itself. It's starting to eat itself.
Yeah, and this is where you really need to take bold decisions, and where you need to gather the talent of the country towards it. I'm part of the entrepreneurial class. I created a billion-dollar company, and maybe another one. We build top artificial intelligence. In all the entrepreneur groups that I'm in, everyone's like, "Oh my god, we're going to get wealth taxes. We're going to get capital gains. Should I relocate?" and things like that, just because I don't know what's even happening and I don't feel proud to be British.
Actually, even myself, I had the largest seed funding round in the UK, made a billion-dollar company, top open source. Government refused to engage with me.
Well, they refused to engage with me.
How did you try and engage with them?
I reached out to various parts of the government. I said, "How can I help? I will give free compute to the universities," other things like that. And they were like, "Well, we don't really think AI is kind of the big thing right now," and other things like that. "We'll get back to you." And so I went and I raised money from America, and I actually flipped the company to America and other things like that. This time I'm doing it all in the UK.
But I think the government goes in fits and starts about: do we want entrepreneurs? Do we want to have rich foreigners coming? Like obviously there's a question of immigration at different levels. I think let's invite in the rich people to get jobs for everyone else. Let's really incentivize entrepreneurship, and let's move for a more entrepreneurial, forward-minded thing.
But you've got half the country who doesn't want that. Half the country, when you turn around—not half the country, but you know what I mean, that side of it—there are people when you turn around and say, "Well, look, if you introduce a wealth tax, the rich are going to leave," and they'll say, "Good riddance. We don't need them."
It's like, yes, you do. Your taxes will go up because of this. They create the jobs. They take the risk. They take the gamble. This is the thing that I cannot figure out how we break, because it feels like a team sport, a bit like Tottenham-Arsenal.
Tottenham play Arsenal. You get a penalty, it's definitely a penalty. The exact same incident the other end, you're like, that's not a penalty. We are blinkered with sports. I think we are now blinkered with politics. I think there are people who consider themselves on the left and you can give them every rational argument for conservative economics, they won't agree. Minimum wage, you can explain to them how that costs jobs, how it shouldn't be illegal for two people to agree a contract, they won't have it, and vice versa.
So I think this is the interesting thing: something like economics. A year ago, we released our book, *The Last Economy*, a bestseller thinking about economics in the age of AI—again, value of our labor going negative, all sorts of prescriptions. We're about to release economic theory on that, where we show the equations of economics are basically the equations of generative AI.
Give us a summary of what that will be.
Basically, economics has various things like a utility function. My preferences or your preferences—Arsenal or Spurs—influence our buying preferences, supply, demand, and others. And so we have this whole hodgepodge of different things that drive economics, but our economic forecasts never seem to match up. There's all sorts of things that seem to drop.
In our book, I say that our economics is focused on material consumption versus things like our collective intelligence, our network effects, our diversity in terms of being able to withstand crises and things like that. And we show that mathematically this is actually the case. Our economics only has 1/4 of the picture versus the whole picture. Just like you have self-driving cars based on certain algorithms of generative AI, we found that the equations of economics can be derived from those same equations of generative AI, and you can model the economy better.
This is important because there's the objective truth and the subjective truth. Again, you have this penalty, never a penalty, and things like that. Sometimes it's pretty clearly a penalty, but the most supporters will still say it's not. People don't trust the statistics. They don't trust numbers because it's like: who are the trusted institutions? And again, this is where we have to think, can we build a trusted institution leveraging the latest technology, and what would that look like in an ideal future?
Because you have these sci-fi futures, as you said, like the AI is overwatching everything and telling us everything to do and we're all kind of plugged in. There's various futures in between. What's a future that we want to build together as the UK where we get information that's correct or information that's reasonable? Because when you're having an argument between two people, what's the best way? Someone you both trust coming in?
Mediating. Can we build an AI mediator? So this is one of the things we're building right now. We're working with multiple governments. We announced at FII in Saudi Arabia last year, it's coming out in October. We've taken all the policies of every country in the world, and we're releasing open source an AI that you can use to your own political bent or other people's political bent, and you can see what's objective and subjective on policy and then economic analysis.
Now, will that help? Only if it's implemented right. But someone had to build it. So we were like, let's build it, because you need that infrastructure. And that's probably going to be better than one company's owned AI making the decisions, because we're already seeing that. Like the Trump tariffs were designed with ChatGPT; that's why you have little penguin-filled islands on the list and other things like that. And you're going to see it more and more going forward. But for this to happen, you need some kind of movement.
Yeah. I almost think it has to get a lot worse. I think it has to get a lot worse. People go, "I've had enough. I cannot cope with this anymore." Because I feel like we are on the trajectory of some kind of revolution. I don't know if it's violent, I don't know if it's peaceful, but I don't think we can have many more administrations where there's continued failure and people accept it.
Yeah. So we're designing some structures there. Like what if there was a utility institution owned by the people of the UK, and everyone got their own agents to participate in this? That's a system we call the Champion that we're finishing the designs of.
Explain that. How do you mean? Because wouldn't I, like a voter, vote for the people who write the laws that suit my life and write the tax laws that suit my life?
Yeah, so you give your input, other people give their inputs. Everyone gets a free AI. You all own part of it, just like you owned part of British Gas back in 1989 before those kind of waves of things—like an AI company that represents the British people, that has collective ownership, and everyone has an agent so that their intellect is raised, shall we say, so it can teach them about these things if they want to, or they can just be completely partisan. Let's gather people's views and votes, and let's have a sense of ownership.
Because ultimately, the AI that drives the economy, that guides the policy, should be owned by the people, and it shouldn't be rented from someone else. And that type of AI is a bit different from a ChatGPT; that's different from this. That's why we built this whole economic theory in *The Last Economy* and *The Commonwealth*, where we look at personhood, law, and political economy. And it was really hard, because a lot of the things that you classically have for being human don't apply anymore for AI.
One of the real examples of this I'd like to say is this: have you read the *Foundation* series by Asimov?
Is that the one they turned into the Apple TV series?
