The Truth About AI They Don't Want You To Know - Sebastian Mallaby

0:00This episode is brought to you by Indeed. Stop waiting around for the perfect candidate. Instead, use Indeed Sponsored Jobs to find the right people with the right skills fast. It's a simple way to make sure your listing is the first candidate C. According to Indeed data, Sponsored Jobs have four times more applicants than non-sponsored jobs. So go build your dream team today with Indeed. Get a$75 Sponsored Job Credit at Indeed.com slash podcast. Terms and conditions apply. Trading at Schwab is now powered by Ameritrade, giving you even more specialized support than ever before. Like access to the trade desk, our team of passionate traders ready to tackle anything from the most complex trading questions to a simple strategy gut check.

0:42Need assistance? No problem. Get 24-7 professional answers and live help and access support by phone, email and in-platform chat. That's how Schwab is here for you, to help you trade brilliantly. Learn more at schwab.com slash trading. Morning decisions. How about a creamy mocha frappuccino drink? Or sweet vanilla? Smooth caramel, maybe. Or white chocolate mocha. Whichever you choose, delicious coffee awaits. Find Starbucks frappuccino drinks wherever you buy your groceries. Jeffrey Hinton, I've had the opportunity to interview him. He's, of course, known as the godfather of AI. He fears for human survival.

1:20How seriously should the rest of us take the argument about the existential risk? So I go up to Toronto one day to see Geoffrey Hinton, and we discuss it. I say, why are you so worried? Because why would the machines want to attack us? And he goes, okay. Imagine you've got a really intelligent AI, and you're worried that some enemy of yours is going to hack it with a cyber attack. Now you have to protect your AI. Humans are not smart enough to do anti-cyber attack in real time. So you have to empower your own AI, that if it sees an attack coming, it counterattacks, defends itself, it survives.

1:54So now you've given your AI a survival instinct and it's smarter than you. Are you sure you're feeling so comfortable now, Sebastian? No, like, I'm not so comfortable. I don't wake up every morning worrying about existential risk, but I do think intellectually, analytically, it's impossible to say this isn't the risk. I think we should take it seriously. Welcome to Open Book. I am your host, Anthony Scaramucci. Joining us today is bestselling author, Sebastian Malaby. He's written an incredible book. and the title of the book is The Infinity Machine. Demis Hassabis, Deep Mind and the Quest for Super Intelligence and Sebastian is the Paul A.

2:31Volcker Senior Fellow for International Economics at the Council of Foreign Relations. My law school classmate, Mike Froman, is running the Council of Foreign Relations. Now, small world. That's right. Let's go. First of all, it's great to have you back. Thank you. We've done this before. You've written about a lot of fun things for me. I loved your book about the Federal Reserve and Greenspan and the whole thing, and the hedge fund book. I learned more about the hedge fund industry from you than being in the hedge fund industry for 30 years. So I give you a lot of credit for that as well. But this book is about the future.

3:09I felt like I was reading a book from 2032. You're explaining to me or explaining to the reader what's going to happen. but what drew you to Demis Hassabis if I'm pronouncing his name right and I apologize if I hear you he's the co-founder and CEO of Google DeepMind but this young man has an incredible story I mean I say young man because he starts out as a 13 year old chess prodigy I know he's 50 now but but this guy's a true genius Sebastian and so tell me why you've focused on him and of course you weave in this very broad narrative about what's going on in the world of artificial intelligence.

3:47Well, one of the incredible things about Demis Asabis is that he had this vision to build a really powerful artificial intelligence when he was still in his teens, and it was the mid-1990s. So get that. I mean, AI couldn't even recognize a picture of a cat until 2012. So this guy is like 15, 17 years early. And so through his life and his thinking, and the incredible evolution of his ideas. One can tell the story of him making a modern AI through his story. And so I always love to do both the figure and the landscape behind the figure. You need a character. You also need a big meaty theme. And Demis provided that.

4:31Are you impressed with the progress of AI? I feel like when I'm using AI and I'm sort of addicted to it now, I use ChatGBT, Gemini. I use Claude. I feel like I've picked up three or four partners who are 250 IQ partners that can more or less help me on just about anything. Are you impressed with how far they've gotten so far? Or what's your thought on where we are in terms of the progress of this incredible software? I mean, I think the progress has been amazing. um if you think back to chat gpt in november of 2022 um you had a system that hallucinated all the time you know you would ask it you know tell me about my wife and it would award her a gratuitous phd which she doesn't have just to be you know it's hallucinating and then about six months later the hallucination was massively reduced and then these systems got very very long memory so you in a whole Tolstoy novel into the context window and query the system about that.

5:37Then they could do video and they could do pictures. Then they could do reasoning, mathematics, logic, science, all that stuff. Next thing is they're becoming agentic so they can actually take care of actions for you. Then the new thing over the horizon is going to be world models that teach these systems about how the physical world looks. And that's going to be fantastic for robotics. So, you know, yeah, I think the progress is super fast. And it's going exponential, right? I mean, there's going to be super intelligence here before we know it. And we'll talk a little bit about my fear of that and perhaps maybe yours.

6:14But let's go to this fascinating character, Mr. Hassabis. Is that how you pronounce his name? It's actually Hassabis, but you're not fair enough. Hassabis. Okay, so he's a chess prodigy. He's a game designer. He's a neuroscientist. He's an entrepreneur. Um, you spent 30 hours interviewing him. What surprised you the most about him once you got past the public persona? Yeah, well, it's completely true that I spent this huge amount of time. I stopped counting at 30 hours. Um, uh, really I've written about lots of fascinating, interesting, smart people, but he is on a different level. you know we would meet in this london pub near his home and there was a kind of secret staircase at the back we would go up and it would be this dusty room we sit there for two hours and he would just riff on intelligence and life neuroscience and computer science which movies he loved why he you know why he thinks what he thinks and how he thinks it's just fascinating because it really goes all the way from physics to football.

7:21And so, I think if you ask the question, what was surprising? One of the things was that I thought going in, the project here would be partly to describe what it feels like for a scientist to have his hands on the 21st century equivalent of the nuclear material. This guy's inventing this artificial intelligence, which could be super dangerous. Does he sleep? I mean, is he worried? What's that like? And I was a bit nervous about posing that question directly to somebody. Hey, what's it like to feel that you might be kidding us all? But he brings it up himself. He's obsessed with the parallels between nuclear power and AI power.

8:05And that was a big surprise to me. You know, you say, you know, because I've obviously in preparation for this, I finished the book last night, by the way. I powered through it because I didn't want to show up this morning without reading the whole thing. And then I did look at some of the other things that you've done. You say it's not a Silicon Valley story. What do you mean by that, Sebastian? Well, you know, surprisingly, this is a person who grew up in North London and built his company in London, recruited and retained scientists in London. And so it isn't a Silicon Valley company. He did raise capital by coming out to the Valley and getting the first check out of Peter Thiel, then raised money from Elon Musk, then raised a ton of money by selling his company to Google.

8:52So it's not like totally disconnected from Silicon Valley, but I think there is a different mentality of doing deep science as opposed to always hustling for the next product. And this is a very long horizon project, right? He starts the company in 2010. That's a fully five years before Sam Altman had the idea for OpenAI. And, you know, he's working on even inventing what a machine intelligence could be like. he studied neuroscience and got a phd in neuroscience because he wanted to understand human intelligence as the existence proof that intelligence could even exist right um and so he was really beginning at the very beginning of this story and and that's not something silicon valley does so much but the silicon valley is more about a where's something where we can build a product and ship it in a couple of years and and this was very different well i mean listen i mean And, you know, he, I mean, you do a wonderful job of explaining him and he's a non-Silicon Valley guy, right?

9:53He sold his business, if I remember correctly, less for less money and less control than he was actually offered. It's like, that's almost unheard of in tech. And so what does that tell you about how he thinks and his philosophy? Well, you know, he told me from the start that he was not motivated by money. and of course i was always a bit skeptical initially but then you hear these stories like first of all when he was motivated by money sebastian by the way you didn't know i mean let's go back to him i'm sorry somebody tells me they're not motivated by money i don't believe them either i think we could do a whole different podcast about you you know anthony you're you're fascinating and no no i just i just when someone says they're not motivated by money my first reaction is okay that's a bunch of i'll use a joe biden term as a bunch of malarkey but I believe him and I believe you.

10:44So listen, tell me how we... So the first example... How did you get over the skepticism, I guess, is what I'm trying to get at, you know? Sure. I mean, the first example was, you know, when he was 18, he had already coded up a video game which sold more than 5 million copies. And so his boss at the video game studio was desperate to keep him because he was this genius that could like, you know, just generate tons of revenue. And so he said, listen, you don't need to go to college. forget that here's a check if you stay with me cash this check and it was basically more than one million dollars in today's money for a 17 or 18 year old who came from a modest family no money in the background right and yeah in stanford people drop out all the time and you know they would have said sure you know i'm off to y combinator and you know the hell with college no no damien sasabi says i prefer knowledge to money and he goes to college and he never caches that check for more than a million bucks.

11:40Next thing happens, as you said, Anthony, you know, he was offered by Mark Zuckerberg of Facebook more money to sell the company to Facebook. He turned it down because he thought Zuck was a bit flaky. He wasn't for real when it came to understanding why AI mattered. Whereas he thought Larry Page was a kind of more cerebral computer scientist, could have been a professor in a different life. So he sold to Google instead. it. And then you fast forward and at the end of my research with him, I sat there in that pub and I said, okay, you say you don't care about money, but let's just check this. What kind of car does your family drive?

12:16He goes, I think it's 11 years old. I can't remember what it's called. Okay. Where do you live? Have you moved houses since you sold the company to Google? No, I've been in the same house 12 years. All right. Second home, forget it. Yuck. You must be joking. Well, what do you spend your money on? He says, well, I do have one extravagance um i did buy a couple of season tickets to my favorite soccer club football club liverpool and and i said okay how much that cost i think it was about three thousand pounds right so five thousand dollars i mean that's the full extent of his extravagance as far as i can tell so to me it's impressive because it speaks to he he's almost like an ai monk sebastian you know he He had this religious calling.

13:02When I was finishing your book, I'm like, okay, this is a very unique person. And maybe because he was raised in North London, Sebastian, culturally, he's different from these guys that are Americans that end up in Silicon Valley. But I want to go to DeepMind for a second. And I would like you to talk about it for people that don't follow this closely. What exactly is DeepMind? What are they trying to achieve? So yeah, DeepMind is the company that Dem is set up in London in 2010. And the mission was not modest. It was solve intelligence to solve everything else. So the idea was, we're going to create these crazy intelligent AI systems.

13:46And then we're going to figure out the deep mysteries of physics, biology, and material sciences. And that will allow us to unlock tons of fantastic new medicines. It will allow us to do much better material science. we can reduce climate change because we can get a bad resource of energy any major bottleneck in human flourishing you can think of that's what we're going to solve so this was the vision um you know it's easy to have a big vision but then you have to deliver and i think what's super impressive about him is that he combined this very very pushy long-range goal with like a practical day-to-day, okay, how do I build the next step in the ladder?

14:26And the key insight he had was there was two kinds of artificial intelligence. One is called deep learning, where you basically suck up tons of data and learn from the data. The other is called reinforcement learning, where you do experiments, you trial and error, and you get feedback. Did my action work? Did it not produce a good result? And he combined the two, and that was the magic that set him on the road to be the first to build AI. So Jeffrey Hinton, I've had the opportunity to interview him. He's, of course, known as the godfather of AI. He fears for human survival, right? You know this, right?

15:04And you know that apocalyptic stance. How seriously should the rest of us take the argument about the existential risk? Well, for the first couple of years of my research, I was wandering around with this happy idea that, sure, the machines will be smarter than us, but they won't have a motive, a reason to damage us, to harm us. Because we are evolved over centuries to survive. The survival instinct is like the strongest thing inside us. Machines don't have that. So why would they attack us? You know, whatever. So I go up to Toronto one day to see Geoffrey Hinton. He's a professor at the university there, and I sit in his kitchen for two hours, and we discuss it.

15:46I say, why are you so worried? Because why would the machines want to attack us? And he goes, okay, imagine you've got a really intelligent AI, and you're worried that some enemy of yours is going to hack it with a cyber attack. Now you have to protect your AI, but humans are not smart enough to do anti-cyber attack in real time. So you have to empower your own AI, that if it sees an attack coming. It counterattacks, defends itself. It survives. So now you've given your AI a survival instinct and it's smarter than you. Are you sure you're feeling so comfortable now, Sebastian? No, I'm not so comfortable.

16:27So I, look, I don't wake up every morning worrying about existential risk because I'm not wired that way. But I do think intellectually, analytically, it's impossible to say this isn't the risk. I think we should take it seriously. Okay. I'm going to say it back to you because I'm going to say it to you in a way that I interpreted it from your book and I want to get your reaction. It came across to me that this man understands how the brain works and he understands that it's an electrical circuit and a result of which he can create an external electrical circuit that will work faster and more intelligently than the brain.

17:05And so therefore it is an additional layer of evolvement. So said differently, these centuries or millennials that you're talking about of survival, he's putting it into the new brain, which is this electronic circuit. And this brain is on a voltage. It's an electronic circuit. And so a result of which it will be exponentially more powerful than anything that we've ever seen. And so when I closed this book last night when I closed it, I said, I don't think the people, Sebastian, I don't think the people around the world understand it, which is why I'd like to, you know, and I told Caddy, I bought a hundred of these books.

17:47I've been handing it out to people, but I want people to understand what's going on. So you have a few minutes here. I want you to give me the message. Okay. Let's say you're talking to my 89 year old mother, or you're talking to somebody that's not college educated about this what is it that they need to know look i said i think you said it beautifully just then um if you can understand the brain is an electrical circuit you can build this external version and it's going to be super powerful so i think what we need to know is that it's coming um i think it's helpful to understand what it is and how it was built and so in some sense reading my book as a form of therapy.

18:28As we get our mind around the future, we need to understand who is this person who created it? Why did he do it? And maybe the fact that, you know, I think he is a sincerely good person who wants to develop the upside and do it for science. And he's already invented a system that unravels all of the shapes and proteins in nature, which is going to accelerate new medicines and stuff. There is a positive side to all this, so we shouldn't freak out too much. But we should be realistic that lots of analytical tasks that you and I can do in front of a computer, those are going to be done better, faster, cheaper by machines.

19:04So I think we need to double down on the human, the human side of us, which is good at human-to-human interaction, in some sense this conversation is an example of this. And just remember, for example, that machines have been better than humans at chess since 1997. And yet more people play chess and more humans watch humans playing chess than ever before in history. So I think we're going to find out new ways of flourishing. But it will mean, you know, doubling down on what you love, figuring out the people you love, that human to human side, and then we'll be okay. Okay. So I want to go back now, okay?

19:46More money than gosh. Take you back probably 15 years now, right? I want you to think about the research that you did for that book. And I want you to now overlay this book because I think there's a relationship between capital and transformative technology. And I want you to relate those two books together. And I guess the core question is Is AI different from what came before in terms of the allocation of capital? Well, strictly on the capital market side, I would say the big thing about AI is it's so capital intensive. You need so much money. And so if you look at what OpenAI is doing, and I pick them because, you know, they are not owned by a big company like Google, so they have to go to the capital markets and raise money.

20:32And they're doing that through private transactions. but the size of these transactions is the biggest in the history of capitalism. It used to be that the biggest ever fundraiser in an IPO with Saudi Aramco, they raised less than$30 billion. Now these guys have raised$41 billion last year. They had a headline deal just now,$120 billion. I think there's a lot of smoke and mirrors in that, but still, it's way bigger than any private fundraising as well. So I think just the scale of the capital is off the charts. is going to change the shape of global interest rates. Interest rates will go up for everybody because all of the capital is being sucked up by these guys.

21:10The cost of power, electricity goes up for everybody because these guys are hogging so much electricity. So that's a sort of strictly capital markets point. I would say, though, that hedge fund people are brilliantly exploiting the rules in the financial market. And they're finding kinks and opportunities and sort of inefficiencies. and they are trading those out. And there's a lot of intellectual creativity that goes into that. And when I wrote my book, More Money Than God, you know, it was kind of like a, not a whodunit, but a howdunit. How did these people do it? How did they think up the ideas that generated profits in the stock market?

21:48Demis Sabis with AI is not playing within a set of rules. He is writing a new set of rules. Entirely new way we're going to bring up our kids who, you know, maybe don't learn to write because this machine can write for them. A new way of us working because some of our jobs will be taken. A new way of how we think of ourselves as being human because now we have this rival form of cognition in a machine. So I think there's a difference between playing in a rule set and then inventing a different set of rules. Okay, so I wasn't frightened by this stuff. I have to confess to you that using it, I'm like, okay, this is awesome.

22:26I read your book and I'm a tiny bit frightened now. You know, I'm not apocalyptic about it, but I want you to explain to our viewers and listeners what AGI is. What is artificial general intelligence? And then who gets there? Who gets there first? And does it matter, Sebastian, who gets there first? Well, I think first of all, AGI, you can, you know, people have different definitions. It's very powerful intelligence. intelligence maybe we could define this where most tasks that humans do on a computer can be done better faster cheaper by an AI and we're not quite there yet but we already have systems you can talk to them you can ask them questions they are artificial they are very general in what they know and they're very intelligent so we're getting there and I think in you know two three years it will no longer be a debate.

23:21We will have AGI. Now, does it matter who gets there first? Sure. I mean, there's always a technological aspect to geopolitical and military rivalry. If you imagine the US military without any semiconductors, it wouldn't be much. And so sure, it does matter who gets there, us or China. On the other hand, we have to remember, it's not just creating the code of the AI that matters. It's like putting it into specific applications. So it's going to be powerful for a business only once they figure out how to integrate it into their workflow. Powerful for the military only once they figure out how they're going to use it.

24:02And that's just beginning. And I think that might take a little bit longer than the two or three years I just mentioned for AGI. You know, I have to ask this geopolitical question because it's interacting. It's interacting in the Iranian war. Reports are the Americans are very happy with the way they were able to deploy AI and they are also sneering at the way the Chinese were able to deploy AI in the war. They sort of feel like the Chinese are more behind us now than they originally thought. And so I would like to get your reaction to that sentence. Is that a true sentence or is that war propaganda?

24:45and then secondarily, where does Europe fit into the whole equation? Yeah. So the first part, what do you think? So I think the American debate should not underestimate the power of the rivals. And I just went to China for eight days because my book launched first in China. They're super fast, everything, and they got the book out really fast. So I was there for eight days and I was introduced to all of these people who were like top AI researchers in Beijing or like the chief executive of some tech company and they're implementing AI. These guys are very good, very energetic. And in fact, when you go to a Chinese tech company these days, it kind of feels like you're visiting a Silicon Valley company.

25:24I mean, down to the fact that they have an HR officer who's telling you about the gender balance and the top staff. And they're kind of like planting trees because they're doing good stuff for the environment. I mean, really, it's pretty indistinguishable from the US. And the technology, I was taken in a completely autonomous car, which drove beautifully, which is made by Huawei. They showed me the phones they have. They don't just fold once like this, like a Google Pixel phone. They fold over three times. And there are all these new software tricks they have. You look at their cloud services.

26:02Actually, I visited NVIDIA in Washington, D.C. just yesterday. and I said, you know, my impression was Huawei was super head-on-cloud in terms of delivering useful technology to like, you know, fixing trains, doing better mining technology, doing logistics better. You know, does anyone in the U.S. do it quite that well? And they're like, no, Huawei is sui generis. Huawei is a combination of the consumer products of Apple, the automobile stuff of Waymo, and then, you know, kind of Azure cloud all wrapped into one. So, you know, we should not underestimate these people. And then in Europe, yeah, you know, Europe's always been laughed at as the laggard in tech.

26:44But guess what? The lithography machines that etch the circuits onto silicon chips are made by a Dutch company with German subsidiaries. It's a European outfit. And they have a pretty much monopoly on the cutting edge chips, on how you build them. So I think the U.S. needs to get real about the fact that, sure, it is the leading country in the world. It still has the best tech, but not by much. And don't be arrogant about it. You've changed my view, okay? I have now a different view of the future. And also, I actually have a different view of my kids' future. So before I tell you mine, I'd like to hear yours, okay?

27:27Has it changed your view, this work that you've done three, three years of deep research inside this story? And has it changed your view of the future and the future of your children? Yeah. I mean, I think it just makes me understand that it's super uncertain how things develop. You know, I gave you the story earlier that, you know, people are going to adapt. We're human beings and we are good at finding projects to keep ourselves busy and kind of deriving purpose from these projects. So at some sort of emotional, existential level, I think we will find things to do. It'll be okay. But there are going to be massive frictions along the way.

28:06And as we know from trade disruptions, you can have a trade shock and a whole bunch of people lose their jobs. And those people, maybe the next generation figures it out and they're okay, but you get people laid off and it's really tough for them and they don't find a new job. So I'm worried. A lot of transition. And, you know, one of the best lines I heard is you may not lose your job immediately to AI, but you're going to lose, you're going to lose it to the person that knows how to use AI. So, right. All right. So you've done this before with me. So I've come up with the five words that I'd like you to comment on.