Nobody Is On Your Side in the Age of AI - Emad Mostaque | PostAGI Episode 003

Thank you for having me.

Yeah, I mean it's a nice big topic, right? People's definitions of AGI have been shifting dramatically. You know, like when I started in AI years and years ago, it was like human capability, right? And human capability and the Turing test that we haven't heard of for a long time isn't that amazing, but it is very general. So, you know, just like people's IQ is on this plus distribution, right? Half of all people approximately are under 100 IQ. Now, we have these 120 IQ AIs that can do just about anything. We're like, well, AGI isn't here yet. Like what is AGI? So we take the definition of an AGI now apparently as being better than any given human at anything. What happens at that point? Well, at that point all knowledge work, you are already the dumbest person in the room. You'll definitely be the dumbest person on the team in the room then because the ability of an AI to one-shot almost any piece of knowledge work will be almost instant. And we're anchored in that we don't really understand what that is until we use AIs that are a thousand words a second or 10,000 words a second. Like the example I give is Tallus, T-A-L-L-U-S. They have this thing Chat Jimmy and it's a Llama 8B model, but it goes at 16,000 tokens a second. So you type in anything and it's instant. It's 20 nanoseconds of response. It's not like Claude that you wait for. And they're etching DeepSeek, GLM level models onto the silicon. That's how they can be so fast. Then you realize it's going to be smarter than me and it's going to be one microsecond of time and that's before we even get to quantum computing. Like there isn't a quantum computing problem that takes days or hours. It's not like it crunches through. So in that world, in that economy, any knowledge-based work is basically you are a negative because it can trade, it can build websites, it can do other things better than you can. And we're already seeing that today. A year ago, it was very difficult to make a good website as a practical example. Now if you go and you say I want a website for Shri Ramik, make it go hard, you will get a website that goes hard. You know, it'll look pretty good and again that process takes like tens of thousands of dollars down to one prompt and almost certainly it'll be one-shot just like again we've experienced so many of these AI conversations going from long long long down to one-shot. So the value and demand for negative value human labor cognitively is like zero. There isn't a market for that. Then beyond that you have robots. Now yesterday you had a company Genesis AI coming out using a company called Wuji Hands and it's like playing the piano and it's picking up stuff randomly with just 200,000 hours of training. Robotics is basically solved now for manual dexterity. And if you look at the cost of a robot it's about $150 an hour all in including depreciation. Like you think about the Unitree robots when they first came out to now they're doing like kung fu and all sorts of things and now they got these clever little hands. That's like...

Exactly. Like it's the same hardware. It's just the brain got smarter. But the brain that's driving these things, someone just figured out a way to train them to play tennis and they can already like get to human level on tennis just by watching tennis being played. Like they are general few-shot learners now and so physical activity as well. They can do just about anything because what these models have is economies of scope. You train the models, you build a robot once and it's not like it is a plumbing robot. The plumbing robot can also do open heart surgery. Can also fly a fighter jet. Like the actual amount of compute I think a year or two ago to actually beat a human with a fighter jet is a Raspberry Pi level. There's a company that did that, you know, and so we're not so special anymore. And again, this economic future is one whereby in the private sector, we will almost certainly be out-competed almost everywhere except for the places that really require humans. And even then, there is this question like we have this AI movie night in a few weeks in London. We're showing Blade Runner, right? I can tell you the companies from China and Japan are getting really good with human level skin and eyes and movement and things like that. I think we're less than 10 years away from I can't tell it's a robot on the street.

Synthetics. Exactly. I mean, all the science fiction is becoming science fact at the same time, right? And so even things like pop stars and celebrities and things like that, like who's a human and who's a robot and who's a fully digital one, you know, like you have all these vloggers already and things like that. It's going to get very strange. That's the only thing I can kind of say. But again if it's something intellectual or physical I don't think you compete with an AI.

So I mean again we've kind of seen when people get out-competed many times in history, right? And you always pivot somewhere and we pivoted up the tree and have these inversions. Now it's where do you pivot to from intelligence? It's nowhere really. Instead it's like the flip side is this is a fantastic potential thing whereby why should I work if the robots work for me? You know, what if they're communally owned? What if I get money because I am human versus from needing to work? But then you still need to have meaning. You still need to have other things. Like in ancient Greece, you got paid obols to be a citizen and participate democratically. You know, you don't get paid to participate democratically right now. You don't get paid to improve your community. This is a question of the social contract because what you're really going towards is that the idiot tax for everything has decreased. The idiot tax is this thing that Elon Musk came up with like you know he wanted like I want to build rockets to get to Mars and then he tried to buy a rocket. It was really hard. And so he's like, I'm going to build my rocket. And they looked at all the parts and there was this idiot tax on every single part. Like I want to have this specific cog that'll be $20,000. Like wait, what? It's the price that you pay. Every part of the supply chain. You will be able to build the extension in your house with a team of robots for basically the cost of materials in a couple of years. And your wife will be much happier. You know, like you'll be able to build anything or make anything for almost nothing but the cost of materials, which is just fantastic. And that leads us to this Star Trek future cuz why shouldn't it be abundant, right? But that then does call into question our current conceptualizations of capitalism and capital and other things whereby how do we get to communal ownership? How do we get people paid for being human? You know, how do we go and protect the people at the bottom that are going to be displaced like this? Like 1 million truck drivers in America, right? Let's say that the average all-in cost is $100,000 per truck driver. That's a hundred billion dollars a year, right? They will be replaced by a Tesla Optimus robot opening the door to the truck and getting into the truck. And there's no extra like things needed. That's hundred billion worth of jobs and probably $400 billion worth of like diners and other things just displaced all at once. And they're not going to become vibe coders, you know? It's just not going to happen. So, I think that we probably have more and more public sector. We need to think about new ways of distributing money, capital, ownership of this infrastructure and think about what the positive visions of the future are like because again you shouldn't have to work I think to live but if you want to thrive then you do need to put some effort in and the society should reward thriving that's positive for society but again that's a very different social contract.

Yeah, like our current level of living is way beyond 50 years or 100 years ago, right? So the floor has been rising like thankfully there are less children deaths, there's less hunger, we still haven't eliminated it though but we should really be thinking how can we leverage this technology to eliminate all those things and make sure everyone can have a good life. You know again there is this surplus that's coming whereby so many things that were difficult to coordinate, there wasn't knowledge of medicine at the edge, right? Like you should have more, do you need doctors, nurses or do you have like local health practitioners that have AR goggles and can help solve most these things? Do you make it so that your entire biometrics are portable, self-sovereign, and you can plug in and analyze anywhere, right? Like these will save lives. A very practical example, for example, we should have an app that goes on every single medical, hospital, and doctor computer and does a second diagnosis for every single thing. And a lot of people's lives will be saved. People's outcomes will be better. There's no downward costs. So we need to think I think in this way. I said the public sector can expand but the government can shrink if we protocolize this which I think is a really interesting thing because again if you think about the slowest dumbest AI is their government, right? Like there are some crazy things like the Los Angeles to San Francisco railway I think they spent like hundred billion on it or $30 billion and they haven't laid a single track. I noted last year that the total amount they've spent on like 1 kilometer of track was more than had been spent on all the LLM training in the world. And it gives you an idea here in the UK we were building a high-speed rail track and £150 million was put aside to do soundproofing because of six bats, you know, and you look at things like that you're like wait what? But then there's homeless people and this and that. So at the same time I think that it is public sector but it's more like public good. We don't think about the government going and controlling everything or 1929 Great Depression style jobs programs and things. We can be more original about that. But again, this tipping point last August when my book came out, I said about a thousand days cuz it had been about a thousand days since ChatGPT and I could see the directions. We're about a year or two before your information knowledge job becomes economically irrelevant. Doesn't mean that you're going to be fired cuz it's not nice to fire people. But you can definitely be replaced by an AI avatar, you know, and the physical jobs are just a few years from that. In fact, I think you can replace a truck driver in America with a $150 an hour AI next year.

Yeah I think that one of the key things here is these are really difficult questions and we've got no consensus around many things, right? They said flourishing like obviously we should be looking at flourishing. What does that mean? And who actually is trying to measure that and who's actually trying to achieve that? No real organizations are aiming for that. Like you have private companies with their fiduciaries, right? You have governments looking to get re-elected and not move the needle unless it's something stressful. So you don't really have entities that are there to look after the big picture and the little picture. Like I think again this is where protocol design can be very interesting cuz you can optimize especially with intelligence that but we're at the very start of actually having actually intelligent DAOs and things like that that can constitutionally check outcomes and I think that's an area that could be great because you can ask an AI does this bill contribute to flourishing. I think probably the biggest thing full stop that will do this is you know something we've been working on which is an open-source AI system we call it SAGE, the Sovereign AI Governance Engine for policy makers globally because all policy will be assisted by AI and then to be honest probably made by AI and it should be done in an open transparent way because already things like the Trump tariffs were clearly done by ChatGPT that's why they were levied on islands with just penguins on, you know there are lots of tells there and if you can make it so that you can have trusted AI advisor to formulate policy, help with decision-making and things, you're more likely to come to the outside because you can embed in that. For example, in America, I want the policy advisor of Seattle, you know, Washington to optimize for the wealth and health and liberty of the citizens of Washington state. You know, life, liberty, and the pursuit of happiness. And it can actually judge whether a policy does that or not because they're capable of looking at all the policies federally and locally and analyzing those and that creates an interesting Schelling point. If you don't have that, then you know you're really trying to go bottom up with a lot of these things. But this coming wave is going to impact billions of well, it's going to impact everyone, right? Like maybe like my old village in Bangladesh, it's not going to affect that much. Who cares about robots? You know, they're self-sustaining. It's all fine. But certainly here in the west where we have these knowledge-based jobs and other things like it's on the one hand really great to be a graduate today because if you're in deep with these technologies you'll get paid hundreds of thousands of dollars right but for the average graduate it's terrible now cuz who's hiring anymore? It's about to stop and we can see that like freezing up. So there needs to be a policy response that isn't ban AI or isn't like I believe in New York Mandani said no self-driving cars that's so self-defeating right like come on just like they banned Ubers for many years like but you've got to have some sort of coordination system and that's why I think you need to have policy advice from a protocol effectively.

Yeah, I think it's very difficult to stop the ocean, right? Like ultimately the wave is coming as Mustafa Suleyman said and they stopped Uber for a long time. They're going to stop self-driving, but ultimately people want to have these technologies cuz they make life better and easier. When you introduce legislation to slow this stuff down locally, like the dock workers of America last year, there is no automation allowed really in ports because of what they introduced. That's hard. But what that really is is it's a tax on citizens. It's a tax on your pocket. And so it's just a different type of tax. Like it doesn't sound like it, but again, to get from A to B, you have the option of a self-driving or you have the option of paying a taxi. Well, if you've got no other option then you have to pay the taxi, which is paying the tolls, it's paying the taxes. Ultimately, that leads to stagnation in an economy in certain areas. But it's inherently quite local when it comes to information knowledge work, which is the interesting thing. Like again, you can ban robots from building houses. That just means that other people would be able to build houses better than your state, right? You have to pay the construction workers and they're protected and they can charge the excess tax because they're bonded and they're part of the glorious association of construction workers or whatever they're called, right? Whereas things like digitally, it's a very different story. You know, how can you compete with your company on the global stage if a company in India is doing it far better and you can't tell it's from India, you know, so we're going to lose to those using AI better and AI-first companies that don't have legacy issues in the private sector. But again, these legislations, these tariffs are kind of a pseudo public sector anyway. It's just the way the capital flows is just a bit different. I think again it's sensible to slow down some of these big things because again how are you going to do it but then you're losing competitiveness as a country like on an industrial basis. China's ramping up the electricity and they've got fully dark factories and soon they'll be autonomous as a country like we all know about the demographic issues of China with the one child policy, robots will fill in that gap. I wouldn't be surprised if in 5, 10 years China just stops exporting robots and says the rest of the world you just deal with it, you know we're good. We got all the energy we need kind of, we got our factories but you'll definitely lose competitiveness.

Well exactly but even then I think this whole carte blanche thing is also wrong because intelligence is about compression. You know the smartest people you know do they take ages of time to figure out stuff or they really quick and disturbing?

They're disturbingly quick right? Like you're like how did you figure that out? How did you do that? Like what's going on?