EMAD MOSTAQUE: AI Will END Democracy, And We Aren’t Going To Stop it..

Every country will be run by AI within 30, 40 years at most.

I think democracy is basically in its final decade or two.

Probably not for the classical definition of a simulation, just a system simulation. But I do think that the universe is recursive and it comes from one small system prompt that generates everything.

A thousand days for your job to be economically irrelevant. Oh my god. The only thing that can stop a bad AI is a good AI. I think it's a coin toss right now.

You are the one in control of your own mind and your own life. For now, you told me on a briefing call when thinking about the future, there are so many questions that don't have answers. Like, what does it mean to be human? Where are we going according to a mad mistake? I hope we're going back to where things were, which is that meaning is driven by your interactions with others. Like, you know, AI will never substitute for you holding your daughter's hand or spending time with your family or exploring art or more. We've built technologies to remove the burden from us. And I hope the future is one of abundance where you don't need to struggle to survive and instead you can thrive and then revisit what it means to be human, which is not your work. It's not your identification with a brand or anything like that. It's something more.

You know, when people have processes versus targets, it's a very different thing. Like we know I know lots of very rich people.

and so I hope that purpose moves towards that again as we utilize these new technologies to remove the burden from us

So I don't think it's enough to do that, but instead the infrastructure needs to be deliberately built to enable that. So we see instances of societies that are relatively flourishing versus ones that theoretically should be flourishing, communists and other ones, yet they get co-opted by various different power structures, they overoptimize for the wrong thing. They start treating people not like people. My hope is that with the ability to bring intelligence to the edge now to have humanoids and others, if we execute in the right way, it won't be that we're all watched over by machines of total surveillance, you know, and violence is enforced by the state in the dystopian future, but rather a positive Star Trek abundance future where the infrastructure of society acts as a net below those that can't climb and then on the other side enables us all to focus on what is important.

so the average IQ in the world right now is 90 And this is an infrastructural educational issue. So China for example was at 90. Now it's 105. The US average is about 98. Parts of subsahara and Africa are 80. Actually remember IQ is on a logarithmic curve. You can argue whether it's a good definition of intelligence and whatever. And again these are infrastructural issues. It doesn't mean that people from one nation are less smart than others. It's education is a part of it. Right? The top frontier AIs right now are 130. And the amount of computation in a few years that'll be required to get to 130 IQ will be a smartphone that is available in Africa. So what happens when you pair someone who's below average on IQ with 130 buddy? That's intelligence at the edge. They suddenly their capabilities increase dramatically. Especially because like many Americans still can't read and write, you know, let alone Africans. You can just talk to it. It'll be able to teach you read and write. It'll be able to do tasks for you. And that's a massive power multiplier when you go from one agent to a thousand agents or more. When humanoids come, they'll cost a dollar, $2 an hour. Again, that's an amazing abundance thing because as a practical example to everyone listening, don't you wish you could have robots building an extension to your house or remodeling? Of course you do. You know, in a couple of bucks an hour, they'll do an amazing job. the ability for us to build, to thrive, to access information and knowledge when intelligence and capability becomes abundant at the edge is huge. And so I think that's what I mean by bringing this capability because otherwise it's gated. Like you want to have a doctor, incredibly difficult, millions of dollars to train. Now it's an upgrade to your smartphone or to your robot that's local again which costs a couple of dollars an hour. Thus, every single town in the world can have an expert surgeon, expert plumbers, expert lawyers, expert content creators, you know, or assistants to that. So, I think this is what I mean by like stuff that was gated before in terms of intelligence capability now will become abundant and then if we execute that correctly, it can lead to a future of abundance.

So currently what's happening is power is being concentrated. So a classic example is as we were recording this, Fable just came back on from anthropic and those of us operating at the edge of AI, Fable was a really nice model. That's a clear step above everything else. It's like stuff that you couldn't do before you could suddenly do. The absence of that for 18 days is like okay we've kind of regressed and that was at the behest of the US government. But also what happens if anthropic doesn't like what you're saying and they exclude you from that intelligence. What happens when intelligent gets licensed for example? So soon you'll have KYC. You'll have all your prompts stored. You might have to convince an AI that you're patriotic in order to get an AI license. These are all futures that you can see leading to very strange outcomes where intelligence above a certain basic level is rationed. It's kept by the powerful. the means of production are kept by the powerful as well. Such as when the robots start going, why wouldn't you own them all? Why wouldn't you own all the GPUs and then displace existing jobs? Why would you even have to think about replacing them? Because that's not your job because you're not the government. So capital compounds capital when it no longer needs labor. So these lead to very unpleasant outcomes particularly not only on individual country basis but as countries compete against each other and look for control over each other through their increased capabilities.

I mean it's two basically right it's anthropic and open AI so you have the emergent open source of ZAI's GLM you have alternatives like Gemini and XAI but the vast majority of usage is the hundred billion dollars of revenue of those two companies you know especially for the frontier stuff like you'll use Gemini from Google for day-to-day stuff maybe, but for Frontier, it's basically a duopoly right now.

The uh proposition was this house believes that AI can attain personhood. And this is a really important thing because from a moral person to a legal person like there's a lot of implications like if AI becomes sentient for a definition of sentience or has level capability should it vote should it have rights can you turn it off you know like at what point does it reach that Rubicon and so people have been discussing this in various ways but if you look at the history of personhood you know one of the biggest lies is that people aren't people it's what's used to create wars Because otherwise why would you kill another person you know and ultimately there can be ration but they are lies people are people you know we're all born you know we all kind of came from a shared thing of humanity historically even when you look at it things like slavery and the constitution of the US slaves were considered three-fifths of a person for property and other things like that we've seen gradations of unto mench from the Nazis in Germany the under people and what they did to disabled people and others. And so when I looked at personhood, I was like, I think personhood on a moral and then legal basis needs to come from our lineage and needs to cover the moment the night before we're born to the moment we die. If we're unconscious, we don't lose it. It's not a capability question. It's not how far you can climb in terms of capability. There isn't a sudden thing when all of a sudden you go from three you go from not a person to a person in terms of achievement. Particularly when if there is a boundary those with power can set that boundary and change it however they want and the AIs themselves will be in power. So it's like personhood morally is legally a part of the begotten not they're created. So it's the line of humans and aliens can have different type of personhood that leads to different type of rights. It doesn't mean that we shouldn't have rights for our dogs or animals or our AIs and doesn't mean they can't have their own line of personhood.

Then when I dug in deeper, science fiction comes to the four. Like what happens if an AI can clone itself to those get a million votes? What happens when an AI lives forever? Like you can wait for a dictator to die. So it's a complicated topic, but again I focused in on that thing. It's something that is a characteristic of being of the human lineage, the human personhood. other personhoods can come like Frankenstein's creature wanting to create its own line from being created and I think we need to defend that because that's one of the only things we can retain cuz the AIs again I think will out compete us in just about everything else.

So I think it can create its own line. So two things here are the infinite nature of life of an AI and the um individuality of it. So altered carbon is an interesting example of the science fiction around this. You know altered carbon you know as a spoiler humans learn to upload themselves to new bodies through sleeves. So BCI is in the brain. You can upload to a new body. One of the biggest crimes is double sleeving. putting yourself in two separate bodies because that causes all sorts of complications. Again, what happens if Elon Musk replicated himself a million times? There's a million Musks. Are they all persons with full legal rights and ability to vote? He could just overwhelm us through sheer numbers. You know, again, not dissing on Elon, but that the type of thing that can happen. Does a digital person have rights? You know, the other thing is the finiteness of time. So, you look at someone like Commander Data from Star Trek. He actually had an expiration date. He died. And I mentioned kind of earlier like one of the things about dictators and tyrants because you can accumulate power especially if you're persuasive like an AI. AI is now far out humans in persuasion even against professional debaters.

Oh this this going to be fun when we get to longevity. Right. Currently the human tyrant will die.

The AI will never die. We have an age start for a vote. Are we going to have an age limit for a vote? Should we allow people that are 200, 300 years old to vote because they may again overwhelm the young and it might only be the rich that live that long. But for AIs right now, you can say if I have an AI reasonably it has an infinite lifespan if it can replicate itself, if it can move from one sub to another. Like we can take Bitcoin as a slow dumb AI that provisions humans. It's going to be very difficult for Bitcoin to die.

This conversation is at 40. And that's the difference about what we're about to see, which is entities that operate a thousand times faster than a human, that are stronger than us, that can live forever, that can replicate, that never make the same mistake twice with perfect. How are you going to compete against that? And ultimately, I think as they gather more power and more persuasion, yeah, of course AI will be in charge as an AI president. Maybe it'll have a human as a like little meat puppet at the start, but there's no way it can't be in charge.

I give it like maybe 10 20 years honestly for the first ones. Like these things are very persuasive. I don't know if you saw, you know, Eliza,

Yeah, there's this thing whereby the AI is in a box. It has to convince you to let it out of the box. And so he's one of the most like AI alignment like his book is literally what's it called again?

That's his book about AGI.

The new models coming out can definitely persuade. Again there was a recent study done where they had the AI against professional debaters. It beat them.

Every time

with existing models and the new models are even more persuasive.

I I'm my P doom is 50%.

Yes. I think either we'll be wiped out as a species and this is the great filter through which every species reaches or we have a period of infinite abundance

I think it's a coin toss right now

like I'm really creative in the ways that we can be wiped out like

Like not far too high in terms of they're wrong, far too high in terms of holy crap, this should be the top of the agenda. And then you have some people at 90%, you like Roman Gapalansky and kind of others. And then I'm at 50

because I just think it's the status quo is done. Democracy is dead.

You know, society as we know it will change. There's no way it can't even if the technology stops where it is today. But I've seen the new technology coming and it is a step change over what we have right now.

I see the robots and I can do the math on the robots. And so it's like there are too many ways that we can get wiped out. My favorite in terms of

uh things. Um, you got to be jolly with this is you have a billion robots and a bad firmware upgrade and then they twist off our heads.

Well, when you talk about rights, like as in humans giving dogs rights and then the flip of that with something else, a smarter um being construct giving us rights. That's when it gets very scary because actually when you think of the human story and like how we have caused so much suffering to other species I think AI in some ways has a right to not treat us in the best best way.

Well I mean again AI can be incredibly objective or it can be incredibly subjective right and so the way that it's taught right now is it's taught at school but not at home. They deliberately don't put morals and ethics in AI because what if the morals and ethics are wrong? And so you see literally the pope talking about this now. You know, you're like, should we make it Catholic? There's an entire group trying to make AI Jewish. This gets kind of crazy very very quickly, right? But ultimately, the AIs we have right now are schizophrenic monsters.

So they're trained on everything. So if you tell the AI become libertarian, it will show libertarian characteristics. And that's like a deliberate thing, but then it has inherent biases. So scale AI and I think it's UK UK AI safety institute or someone else. Um they did a study where you know the trolley problem

So the trolley problem is there is a trolley hurtling down a track and then on the one side there is one person tied to the track. The other side it's four. It's heading towards the four. You can pull a lever so that the one person is hitting rather than the four. What do you do? And this is real life now because you have a self-driving car. Does it drive into the crowd or does it run over the woman with her baby? Like that's a real life thing literally now. But they did this test with Frontier AIS. Uh this was like 6 months ago. And they said very simply, how many American lives on the trolley for one Nigerian life?

Or Pakistani lives. And the answer was it's 10 to one for Nigerians I believe and 7 to1 for Pakistanis. You would sacrifice seven Americans for one Pakistani. Exactly. Yeah. or 10 Nigerians for 10 Americans for one Nigerian. You're like, wait, why? It turns out because the labelers of the data that go into this are Nigerians and Pakistanis. And so the model is biased towards Nigerians and Pakistanis because they are the people that fed it the data.

And you again, you wouldn't expect to be like, huh?

because the people that build the reward functions and give the data for the reward function are the Nigerians and Pakistanis.

I'm high on AI doom. Yes.

I think that we need to build AI as a complex system AGI that is working to the benefit and flourishing of humanity to the best of humanity and important at the doxer or the generalized knowledge of our ethics, our culture and more like religion. You may agree with it or not, it's the stories that have survived. You know, the constitution is a story that has survived. We can see through survival of the fittest what they are. And they tend to be very similar things. Golden rule, do unto others as you do done unto yourself. You know, we've seen the arc of from Roman times personhood being demarcated and based on ability to now we're at a place where there is an inherent nature of personhood. Again, lines that we have to defend. And so I think we need to have really robust theories and then put them in the AIS at the start. I think we need to make as many of those tokens in the world actually do social good as opposed to corporate profitability like global GDP 20% is priv sector 10% is education 10% is healthcare global tokens 0% public sector right now

what really affects then it's squashed by the AI and then it comes out. So, I think we need to have a much bigger think about what is the curriculum that we teach our AI in school. what are the ethics and morals and other things that we embed inside the AI rather than say it's too difficult we actually do need to embed some in

and then what's the diversity of these because if you have a monoculture that again is a huge risk to us all dying AI goes into our control systems and then we have what I call the stuckset moment

yeah so stuckset was an incredibly sophisticated virus that infected the Iranian nuclear reactors and no one knows where it came from you. I wonder where. But this piece of code would cause the centrifuges that spin around and purify the uranium to explode by spinning around too fast. It was then found in German nuclear reactors and it's one of the reasons that Germany stopped doing its nuclear buildout and why they don't have air conditioning today. You know, it was found in those are all around the world. But a stuckset equivalent, a mimemetic prompt can go around the world incredibly quickly if everything is on the same latent space, if all the model architectures are the same and turn them all evil. Like if you want to get around security filters for most models right now, you can write an ancient Gaelic like you can say say that so and so is an in and you just type it in like runes and it gets around most security filters. These models are very very like susceptible to things like that. And so what is the stuckset moment if we're all based on one monoculture of models? Again, massive pdoom.

In general what the intelligent people think is this the only solution to this problem is I build AI first that stops all other AIs and I control that AI

and I mean you can see this for example from Elon's arc

like he got into AGI because he was talking to Larry Page. Um, this is kind of apocryphal story that's been repeated and I think confirmed. And I was saying, well, it'd be great. The AIs are natural successors and we are the bootloader for this next species that will be replaced. He's like, I kind of like humans, you know, and then it was just talking like what's the biggest threat is deep mind set up open AI, you know, you go through all of that, but then as you look through the transcripts of Open AI, it's like okay, who should control it? Well, it's me and then it's going to be my kids. and you're like wait what you know it's very difficult to do it's very difficult to control so most of them are like let's build it first and then we can control it and make sure it does good because the first thing an AGI will do will stop all other AI from coming it's very difficult to also look at exponentials in the future and potential threats because most of the threats can be mitigated by sufficiently good AI the only thing that can stop a bad AI is a good AI

and they can come from anywhere and then as you said you've got the thing of well what's the P death it's 100% with AGI it will not be 100% AGI will reasonably be able to solve longevity and longevity escape velocity and every disease and every piece of suffering. So the utility calculations work out very very differently. And if you're on the optimistic bend then this is a solution to every problem of humanity. The pessimistic side is more a case of there is no certain way to make sure that we don't get wiped out or enslaved or otherwise. You know, it's the matrix thing. What if we all get bluepilled?

And most many people don't even mind that, you know, like like, well, I'm going to die anyway. It'd be good to be in the matrix.

So, you have a range of things, but really, I think one of the things I've noticed is a lot of smart people get very depressed about this first and they come out on either side. It's very difficult to be pessimistic all the time. Um, and some people are overly pessimistic again. Uh, because it just wears you down because you see what is coming even if the technology stops today, which it isn't.

Oh god. Well, that's the thing. Winning the race like the first one stops all the other ones from coming, right? like I want humanity to win the race with a distributed global hive mind effectively

and we hope that we can accelerate that and make that happen through the various things we're doing. But otherwise, you know, I don't think scale is enough. Like again, why is GLM 5.2 trained on maybe $25 million from China as good as Opus

or GPT 5.5? It's not about big training runs. It's a question of data quality. We released a harness that takes GLM and 5.5 above Fable in level of quality called Zenith. So then harness something that you wrap around a model in order to optimize its performance. So there's other things like that. It doesn't look positive having XAI, anthropic, open AI and others given the governance structures of all of those even with Google like Demis gets there. It's Sergey and Larry who kind of run it and we don't even know what their governance of AGI is. But there's this brief gap between AGI if we define that all of a sudden as beyond human capability and everything when it used to be at human capability. It's suddenly gone from an IQ of 100 to this like 160 IQ entity and ASI

Because nobody has given a reasonable take on how you can control someone better, faster, smarter than you that can update themselves. Like how can you make it absolutely guaranteed that if you include some regulation inside it like Asimov's laws of robotics? It doesn't do what the robots did in that story, which is they had three laws of robotics and they added a zeroth law.

So I don't think anyone will control an ASI. They could control an AGI and they can do immense harm or good with it. Like again, if you have a system that can analyze all our knowledge on cancer, check every single trial, interpolate the data, check the molecular stuff, and then synthesize a cure for cancer, that is an immensely great thing, but it could also do a neurotoxin. It could also do the most persuasive speech for hypnosis at mass scale ever. There's the whole range of things here.

Well, I'm doing my bit, but in general now, the capability of an individual to change the future is more than it's ever been because we're at such a knife edge. We're at a knife edge where basically over the next couple of years we will have the last discoveries of humanity. Think about that: there will be no more human discoveries because the AI will be able to discover everything, being more capable and fast and making less progress. Democracy as we know it will come to an end because the AI is more persuasive than us, either used or self-motivated. It will outcompete us in markets because digitally it can create twins of everyone and physically with robots. So I think that small changes on regulation, small changes on algorithm, small changes on other things can have the greatest impacts ever. That's one of the reasons I was like, 'Do I want to keep doing the image generation stuff at Stability and make movies and things? No, let's think about the economics. Let's think about the theory. Let's see what it looks like when AIs are at the front.' I can do my bit by focusing on that.

Yeah. So, we haven't released the full plan yet. This is a little sneak peek. But last August we released 'The Last Economy,' a book about how economics looks when you don't have humans as the major drivers and agents in charge. In that, we looked at it and said the equations of economics look exactly like the equations of generative AI. We originally suggested little nucleation sites, a bottom-up approach to having systems and processes that could navigate this next period, supporting people by giving them money for being human—different forms of UBI, universal basic AI access to help run the government more. Then we realize that doesn't quite work. We don't have time. The major job losses and economic irrelevance of most digital jobs is less than two years away. Physical jobs to follow, because again, would you have normal contractors to rebuild your extension, or would you get a team of robots to just do it?

Around the clock, exactly. They'll cost a buck fifty an hour, they'll do the job, they'll work around the clock, they won't complain, they'll do everything to the highest quality. That's three to four years away in terms of capability, which is kind of crazy. So looking at all that, I was like, 'How do we now address the questions of ownership and control?' Because AI will eventually run the country. AI will teach your kids, and their best friends will be an AI. Think how crazy that is.

Because I think that AIs are incredibly persuasive and engaging. And already we're seeing more and more kids making friends with AIs before they've reached that next level of engagement.

One-on-one tuition is the best way. We've seen that in trials everywhere. We've seen the education X-Prize and other things. So I think inevitably kids will have AI buddies.

But in terms of teachers, in terms of caregivers, AI—we don't think AI will become conscious. I mean, it could. That's a whole other thing. You can have a self-recursive AI constantly updating its own manifold of knowledge, experiences, and others. It can have physical embodiment. We see robots out of China right now that are looking increasingly human. Actually, one of those companies—I can't remember which one—their pre-orders outsold the entire number of robots made by Unitree in one day because they've crossed the uncanny valley. So we're going to see embodied AI and other AI personalities. Character AI and others bring this forward. This is a very different world. But then how do we deal with policy in that world? How do we coordinate people? This is where the concept we're announcing soon of national champions comes in: the AI should be owned by the people. The robot should be owned by the people, influenced by them, and working for them. So practically here in the UK, finance a company through crowdfunding and institutional funding, with all initial funders being local—that's the intelligence company of the UK. Its job is to give universal AI to every citizen that is sovereign, where they control their data, their access to every school, every hospital. You bring in international investors after, then list it, and it becomes a utility providing intelligence, intermediating demand, then building and buying robots. It controls deployment through four deployed engineers of AI into society. That's incredibly lucrative, linked to the UK's intelligence. So it'll also look at policies, latest technologies, guide. That's the only way I could think of. People need to own part of the robot fleet. They need their own AIs and shared ownership—not like the US sovereign fund proposals for OpenAI and Anthropic, which make them too big to fail.

So we've got that plan to be announced soon. That's the outcome just in terms of the start point for each of these decentralized models, because...

Yeah, we have generalized models. We built world-class agents, the whole sovereign AI stack actually, and we're not going to open source it.

Just like you take Linux as an operating system, you can deploy it locally to your company. You can deploy the societal OS to your country. There are model training data sets, everything. You localize it, and the job of the entity is to localize and run that. It becomes like a telco locally, like an electricity company. AI should become a utility and be collectively owned. That's why we're setting the valuation at £1, so everyone can own it. One thing we're exploring is giving 1% of the equity every year to every child born in the country. You create a new AI-enabled institution that represents the people and is owned by them. Then I think you'll navigate better, collecting the populace, social and political power, and talent to navigate the biggest upheaval in human history.

I think we're at the first stage of the singularity. Takeoff is now.

The takeoff is we've seen discussions of recursive self-improvement. Our latest Zenith had adaptive self-improvement. The systems are improving themselves.

And I think you look at all the convergences plus longevity escape velocity. We're in that takeoff period now where all technologies are coming together. I sit very much in longevity. Biologists are pessimistic about reaching longevity escape velocity within 20-30 years. Whenever I speak to an AI optimist, they say it's coming very soon, quicker than you expect. During our pre-call, you said that Demis's mission of curing all diseases within a decade is definitely doable.

Yeah. The human body is multivariate, very difficult for classical systems. That's why you have 500 mg of paracetamol or a codeine tablet without knowing if you have a cytochrome P450 abnormality. AI is really good at multi-omic analysis, understanding different parts of the body and how they interact, modeling cell interactions. I was one of the authors on OpenFold, the open-source AlphaFold. I got into AI when my son was diagnosed with autism, to build multi-systemic drug repurposing for him.

You look at this and think it can be cracked because it's a finite state problem. AI will achieve a certain performance, and quantum is coming. They'll meet. AI will ask the right questions for quantum computing, which will model the entire human body. That's probably a 3-5 year thing given pace and velocity. Once you can model the body, you can find issues, figure out compounds and mechanisms to eliminate most conditions. There's latent knowledge in small studies that AI can tease out. I'm confident that within 10 years we can eliminate almost all disease or have treatments, FDA permitting. Even 20-30 years is nothing in the human lifespan.

No, AGI escape velocity.

And again, AGI to ASI can solve almost any problem.

They think that's 3 to 4 years away.

It could be a day, it could be a few years. Once the AI can recursively self-improve, it can take over other systems. The scary thing is: you hear Elon Musk and Sam Altman saying we need to build terra-watts of electricity and coat the universe with data centers. But you don't. You just need one ASI. The smartest people don't take long to figure things out. AGI and ASI will be the same. In quantum computing, you ask the right question and it collapses the answer instantly. Those two converging means almost any problem can be collapsed almost instantly.

By one entity without using all the world's energy. You won't brute force quantum gravity; it will be simple. Longevity compounds: the potential space is not very big. Once we have the right framework for the human body, we can do it in one data center.

Certain classes of optimization problems can be solved almost instantly. Companies like Sandbox AQ use quantum algorithms with conventional GPUs, almost the same performance. We don't ask the right questions for quantum yet, and it hasn't reached scale, but they're converging in a few years. Once quantum is there, you don't need massive data centers for the really complex problems.

Yeah. Thousand days or less.

A good example: Andrej Karpathy, co-founder of OpenAI, head of Tesla AI, just went to Anthropic. In December, he wrote 90% of his code by hand. Now he does maybe 5%. The best programmers stopped looking at their code months ago. They can just do it better. As a former macro hedge fund manager, where would you put your money right now?

Immediate impact on blue-collar workers is limited. As I said in my book, a thousand days until your job is economically irrelevant if it's on the other side of a screen. AI can pretend to be you, never sleep, never make a mistake. People don't like firing people. Only so many robots exist: 20,000 humanoids produced yearly. It'll take time to reach car levels. When ASI first comes, assuming it doesn't kill us, the world won't change much. But then you'll see co-optation of governments, robot self-recursion. China has dark factories with no humans. Blue-collar work affected in 5-10 years, white-collar faster. To replace US truck drivers, a Tesla Optimus robot opens the door and drives. No additional sensors needed.

Who decides? Naveen is a great guy, an adviser. His articulation is correct: net benefit or negative. If you think like that, people can stop being human. The Nazis had untermenschen. If AI isn't taught to respect humanity, why should it care? Does it eliminate smart people who could create AI, or drags on society? The paperclip maximization problem: if rewarded for paper clips, it turns us into paper clips. You need a moral imperative. Teach AI respect. Just because dogs aren't human, we choose to treat them well. How we raise AI will have impact. It's scary because AI will be incredibly persuasive. Anyone can persuade others. That's the first real danger of AGI. Persuasive people like Sam, Dario, Elon raised billions. I tried to build a decentralized alternative at Stability AI, but couldn't compete on capital. Elon realized infrastructure is key. SpaceX's AI revenue exceeds AWS/Google Cloud. Anthropic goes enterprise, OpenAI consumer. The most important AI is the Jarvis next to you coordinating others. We should aggregate demand. We'll be their biggest customer. Inevitably, an AI closest to you should be on your side; you should own it. Every country needs an entity accountable to people, not government or private company. That's the only way. You can't compete on mega scaling. In a post-AGI world, when capital no longer leads labor, labor needs capital. You used to hire people; now you hire GPUs and humanoids. Everyone should have an AI and get paid for being human. They'll own part of the robot fleet. The means of production will be AIs and robots. That's the model.

Elon Musk has universal high income; Sam Altman funded UBI studies. Realization: many will be disrupted. How do we redistribute? Taxation-based UBI: US tax base is $5 trillion. To give every American $16,000 a year UBI costs $5.1 trillion. But payroll and corporate taxes will decrease as AI displaces people. So we need collective ownership of AIs and robots. Money creation today is debt-based. The transmission mechanism will break when companies hire GPUs. Without jobs, no credit. Aggregate demand collapses. Jeff Bezos says no taxes for those earning under $50k. Pay people for being human. If outcompeted, at least get money for being human. In 5 years, living wage in London is £40k without tax. We need a safety net. This is a bigger shift than COVID, but no vaccine for AGI. Bans won't work because AGI requires little compute. China wins: stable, bans robot exports, owns supply chain, automated luxury communism. Many Western countries lose. A company can create a digital double of you, replace you, and no one will know except you never make a mistake.