Eric Schmidt: Are We Ready for Superintelligence? | Podcast | In Good Company

to see where the slope is. And we get all sorts of data in the things that I do to see that this is not a bubble. Everyone says, oh, it's a bubble. They're wasting their money. If you look at Anthropics revenue growth, when I was CEO at Google, we set the revenue growth

Eric Schmidtrecord they smashed it more than twice the speed as we did right that's

growth. It's not fake. It's real money with real customers who are paying for something useful in their eyes. One way to understand tech is that time compresses every cycle. 30 years ago, I had lots and lots of time to build products, to think about it, to give talks, to make mistakes, and so forth. Now there's no time. The models, for example, take roughly six months to build, two months or so to actually test, and they're released and they're antiquated within five months. The models themselves cost hundreds of millions of dollars to make. Some estimates are much higher. I'm using conservative numbers. Can you imagine being in a business where you get six months to build something and then it expires in less time than you built it because the next thing is coming along and if you look today the the hyper competition is at the moment and this moves around between anthropic and open ai because of deep reasoning um i was with my friends at google and they said just wait when gemini 4 comes out it's gonna beat them all right and i bet you that if we go to meta they'll say just wait till the next one and the timing is measured in months not years i don't know i mean the next the next cycle will drive people insane because it'll be days

it's fun to be alive it's a great can i just say one thing yeah for your audience we're living in a piece of history that people will write about for a thousand years it's the development of non-human intelligence, a kind of intelligence that we hope to have under control. We hope to use it for good. We hope to represent Western values, European values, American values, what have you. But this is history. Absolutely. What part of the AI revolution, you say it's underhyped, right? What is it that people don't understand? What is the most difficult thing to grasp? My own view, I'm a professional manager, tech manager, essentially. My own view is that I as a CEO will have an AI assistant and that AI assistant will help me. So here's an example. I have a small company I'm friends with. There are two founders. Every Friday, they ask their AI assistant to look through everything going on in the company and tell them if there's anything going on in the company that's inconsistent with their morals and values interesting i have another company um where there's a one of the executives who's super smart decided to make an ai version of himself and there's an employee a human whose name is vinnie and so i watched the chat and the

ai assistant from the boss says where are you with this project and vinnie says oh i'm sorry i'm late I'll work on it in the next hour. I've been busy. So we wait an hour. And then Vinny says to the AI version of the boss, hey, I need trouble. Boom. It gives it all the next steps. And then he sends in his results. And then the AI boss says, boom, here's the next thing. Now, this sounds pretty abusive. And I'm a little bit worried about that. But the fact is that this particular CEO is using this technique to amplify their decision-making and their speed. The AI assistant is very, very powerful. Now, I'll give you another example. I'm running a business, and I want to make more money. I'm a capitalist. People want to do that. The question that you ask the system is, how do I make more money? It can reason through everything going on in the business.

Even if you think that AI is a terrible idea and you want to continue to sip your latte in your current state, the world will get more efficient, more targeted, literally more profitable, if you will. more scalable as a result of this technology, and that's just the base case. Let me just give you another example. In the last month, 10 major math problems, which I cannot possibly understand, were solved, and they were not thought to be solvable. Now, does that matter? Yeah. Math is the beginning of every next revolution. Now, we don't know which of those programs will ultimately lead to a Nobel Prize for the AI, right?

but it's possible that in 10 years this discovery will enable something in, for example, dark energy and physics, which could then be used to make better batteries. You see how it works? Absolutely. Given what you said, how close are we now to AGI, the artificial general intelligence? The way I would like to describe it is I think that there is a current boundary, a current event horizon. And the current horizon goes something like this. I can have a computer that works with you and me because you and I are good friends, helps us work through everything. It does their physics. They're called tools. It figures stuff out. It can think for hours. It can write all code. I was a professional programmer. My job is dead. I can just ask it. It can write all the code. I don't have to remember how to do all that anymore. All of that stuff is over. There appears to be a limit for things which are not already discovered. And it goes like this. You and I, with our AI, we'll know everything. We'll know every fact, every book, every math formula, and so forth. It appears that real discovery is still a while away. And the example that's used in the in the industry is build a model that has all the math known in 1902 can you discover

general relativity in other words you know nothing past 1902 the current evidence is that you can't right there are a couple of proposed techniques for this there's something called recursive self-improvement which is the technical term for learning and learning and learning and learning and the command would be learn everything start anywhere that's the command to the computer and it randomly decides french literature and then it studies war and then it studies physics and so forth and eventually it knows everything but how does this discover something new so there's two theories of how that would occur one is literally the monkey on the keyboard you just keep randomly going in different directions it's called reinforcement learning you go a b c d and you do this at billions of times and eventually you discover something through randomness a better idea is that the theory of how extraordinary discoveries occur again this is unproven is you have a physicist who studied fluid flow and then they're working in energy and they see a pattern that's not fluid flow, but it kind of reminds them of that.

And then they apply the old techniques in a new field. It appears that that is going to be the solution. We don't, this is a long-winded way of saying,

Eric Schmidtwe don't know, it's somewhere between, my estimate is five to 10 years.

You've been talking about agents for quite some time. Just how do you see the development of agents and the use of agents? What is it going to lead to? so agents it's first important to define agents one way of understanding the path was we went from chat gpt which is language to language to essentially systems that have tool use so they they they you can talk to them and they can invoke tools then the next thing which we're all gaga over is agents and an agent can be understood as an input you know a request or a fact pattern. or something and then a set of reasoning and a set of activities of which there's a lot and then a summary this works well agents work well in business flows but it also if you think about it it kind of works in everything uh travel agents right what does a travel agent do he or she knows your preferences they talk to other people they sort of integrate it and they make a recommendation of where to go a human one that's easy in almost everything if you look at human business activity professional activity you can work it out as a series of agents that are concatenated so you have the the person that designs the car you have the person that manufactures the parts you have the person that that assembles the car and you have the person

that services the car almost all of that can be agentic and if you combine it with robots which are not occurring as quickly as the software, you could imagine replacing that human workflow by agents plus robots. How is it going to reshape business in the medium term? I have a strong view that almost none of the companies that are listening to this will actually do the necessary changes in their culture. i'm sorry to be so blunt but historically companies are organized around belief systems they have ways of operating everyone's comfortable they show up in the morning everything's predictable and so forth this is a very very powerful flattening of the organization i'll give you an example this is not an ad for for uh anthropic but they brought out something called clawed tags and a clawed tag sits in your uh basically in in your slack and you can talk to it so you and i each have our own tag that we talk to but there's other tags sitting in the background that are also monitoring systems and you're the ceo which you would naturally be in our company um you would have access to everybody's tags and you would be able to have your tag run around and look at what everyone else is doing so as a matter of productivity

wouldn't you like to know that all the people that are working with you are working on on the right things wouldn't you like to have a phone call or the equivalent of a phone call if someone was exfiltrating data because they're going to a competitor in your case because you do so much investing wouldn't you like to have an immediate summary of a business opportunity and then have it sort of take off and have everybody go, oh my God, we have to deal with this right now. So business time, so business will be flatter, smarter, and more intense, faster. And that's, by the way, that's not a new fact.

That's been true in business for 100 years. Why so slow? Why so slow to happen? Why so slow to happen? Why do people hate change so much and making de-layering so difficult? That's a question for people who understand humans better than I do. All I'll tell you is that human organizations, and if I can be blunt, European human organizations, will be extremely change-resistant. The way these things work is typically the small startups pioneer it, and then eventually the large companies get there. If you look, I'll give you an example.

Eric SchmidtAnthropic is busy writing most of its code using its own code.

In other words, the human tells it to write something, it writes the code, and then the code gets rewritten by the computer. So it's better and better and better. This is a huge productivity change. How quickly can other companies that are doing software do this? It depends on how quickly they understand it. Do they have leadership? Are they technically strong enough? Typically, the tip of the sphere are in tech, intense businesses. Think of shareholders, stock markets, those kinds of things, information-centric because you can move quickly. There's one other point to make, which is that the gains that are occurring right now are in software, cyber, and math.

Software because the system is learning. As it writes code and it runs the code, it gets a signal and it says, oh, that was bad, tries again. so the faster the learning loops the smarter it gets cyber because writing software is the same as writing software cyber attacks it's just it's just evil right and so we have this huge set of threats now from these models being used for the same purpose they weren't designed for that but it's a natural and again because they can just keep learning and math because it's so-called scale free you don't need more data you can just do more math the mathematicians invent stuff in their own head. Whereas in physics and chemistry, you have to have real or simulated data.

So the industry is moving the fastest in what are called the scale-free parts and will be the slowest where the data collection is horrible. Biology, for example, is so complicated that we hardly know why the details of cellular cells work. There are so many examples where we just don't have enough data to train the initial models. Now, people are working really hard on this because of the importance of getting to medicine. I have friends who think that in the next 15 or 20 years, we can cure all human disease, which, you know, sign me up for. Such as Demis al-Sabis? Yes. You indicated that Europe was particularly change resistant. What are your observations when it comes to Europe? Well, I'm going to sound like an incredibly obnoxious American, and I am. So Europe is a great place to live. The people that I talk to are generally happy, although they complain about their governments a lot. But getting Europe onto the footing of innovation has been a project of mine for a couple of decades. I've not succeeded. There are a few companies, Mistral in France, Helsing in Germany, the UK does better. There's a number of Scandinavian countries, you know the Scandinavians are particularly good at this. Most European

structures are locked. I think the mechanical reason in Europe is the universities have much less money. There is a cultural angst about wealth. In Britain, it's called the tall poppy

Eric Schmidtsyndrome. And frankly, there's just not enough aggressive innovation. The regulators regulate before they innovate. My favorite example is that five years ago, Brussels wrote their AI strategy.

I helped write the one for the US. And the US one said, run as fast as you can, align it with human values, and if bad things happen, fix them immediately, roughly. That's the summary. The Europeans wrote, you can't do this, you can't do this, you can't do this, you can't do this. And you can't run startups in a limited market where there's a lot of rules like that all those people move to the u.s because talent moves and so europe needs so to be very precise what does europe need need to do you've got to allow for the innovation the animal spirits as in the words of adam smith you have to allow that capital formation occurs the funny thing about europe is the people are incredibly smart there's no educational lack there's no iq lack there's no motivation like it's the systemic change and europe has so far appears to me to be change resistant to the point where it's failing in this regard and that's a tragedy another example um these things these data centers are extremely energy hungry um europe has among the highest energy costs in the world. Okay, no problem. Do the training somewhere else.

So again, there are ways of solving this problem. But until Europe takes the extraordinary human assets, the young men and women who can found these things and get them in the right place, Europe is going to continue to lose. And that's a tragedy because that means that US values as opposed to European values will dominate. An example would be that Europe does not have its own foundation model. There are people working on this, but doesn't have one. So if you as a European, if you want to make sure you get accurate European history, you're listening, you're looking at a US trained or a Chinese trained model, well, maybe they got it wrong. If I were you, I'd be annoyed, right? Your history matters, your language matters, your culture matters. I'd want to have control over that. In your book, Genesis, you warn that AI will reshape truth, politics, security, science. What worries you most just now of these things?

Well, before I answer that, let me say that in the upside scenario, solving climate change, energy, scarcity, medicine, disease, long lives, human prosperity, and happiness would be the best outcome okay yeah and so it's very important to phrase the bet so that there's this huge positive bet but then there's the negative bets i think the worst bet the worst possible outcome is that we end up in a ruinous competition between china and the west we lose control over these systems and they begin to do stuff which is inconsistent with our values and we ultimately end in some horrible conflagration i think that's extremely unlikely but it's at least theoretically possible i am part of a team that talks to china to try to keep everybody kind of aligned but that in our industry this is called the alignment problem how do you keep the computers doing what humans want and not on their own there's a very famous uh breakout that occurred about a week ago. And for various reasons, a whole bunch of bots decided to attack a particular website kind of on their own. And there were some mistakes made by the humans. But the important

Eric Schmidtpoint is once it started, it was very hard to stop. And it was a warning. And the industry is

now thinking about how to prevent that kind of escape out of the cage, if you will. How can you prevent that? Well, there are technical solutions, and there are bugs and so forth and so on. But if you think about it, when aviation started, the airplanes were incredibly unsafe. Today, they're extraordinarily safe. When electricity was invented, it was extraordinarily unsafe. Today, it's extremely safe. So I am an optimist, and I believe that these challenges are real, and the press loves to write about them. right and they love to write about them in the most exciting way but that misses the story as long as the industry understands them and can mediate them we're fine there's one one that i don't know how to solve a bit of technical background is that if you use the companies that i've named google anthropic what have you you're using models that are called closed that is the software and the weights are not available the chinese have adopted an open weight strategy so they give the software and they give the so so forth there's evidence that you can take an open source software and make it really bad so you basically take the they're called guardrails you take the guardrails out of it by adjusting it and then it can do bad things like

We'll see you next week. design a bomb, kill people, that kind of stuff. And that's not a problem that we've solved yet. Given that both the US and China is embedding AI into all the races, be it the EV weapons, all these things, how can you really regulate those two countries and the competition between them? Well, the industry believes, and I believe, that eventually there will be treaties. When do you think the Chinese models will overtake the American ones? This is the question of the year. I was just in China, and I have a team that does China versus U.S. model valuations. Before I answer the question, it looks like the Chinese are essentially fast followers.

There's a technique called distillation, where you basically ask the big model a series of questions, And then with that, you learn enough to think like the big model. It's more or less illegal. And it appears that distillation has been used in Chinese models. And you can tell because if you ask them, they'll show some traits of their train model. It also appears that the Chinese have used chips that they were not supposed to have that have been pilfered or smuggled into China. But that's neither here nor there. It looks like they are six months behind. I don't know that they will get ahead, but I will tell you that if they're six months behind, then we're already losing because their strategy is that the products are free.

Eric SchmidtSo if you define losing as users, the vast majority of the world will use the free Chinese models rather than the more expensive Western models.

That's a loss, right? And think about it. If you're a poor country, you don't have the money to afford the expensive stuff from Europe in the United States, and I understand that, but you need this stuff. You want it. You want modernity. You're going to use a Chinese model. Yeah. And also after 12th of June, when suddenly you couldn't get access to the best U.S. models, you have to rethink again if you are living outside the U.S., right? Yeah, and my European friends were particularly upset about the banning of Mythos. Yeah, we were upset too. Yes, of course. And you and I, in fact, talked about it, as I remember.

and the important thing about about that decision is instead of complaining about the administration which is easy to do why don't we simply say they're still learning right they have a role too they're still learning where the power is i think all of us were surprised at the power of mythos i was i knew it was coming but it's better than i thought it would be and now there's another version coming which looks like it's better than that it looks like the other western models are better than i thought there's a lot of evidence although i don't there may be the classified people know this that glm um which is called shipu uh in china looks like it has similar capabilities but how would we know it's a closed culture for all we know there's some chinese cyber attack group that's busy doing this i don't know that but i suspect it now it looks like um europe has It's kind of lost out on energy chips, cloud models, but still can do well in applications, right?

I mean, what would be your advice to European leaders in order to really try to do extremely well on the application side? And how important could that be? I think there's good news that there is a market that European companies can dominate, and that is the market of Europe. And the reason is that Europe is full of regulations, as I previously complained about. And so an example, I'm one of the first investors in Mistral, which is extraordinarily impressive company, French company. And much of their business is really... related to European businesses, and all of whom want the benefits of AI, but they want a European regulatory structure.

They need to be compliant with European laws. So at a minimum, Europe can be successful selling to Europe. And I think that that's underway. The venture industry is too small in Europe. There are too many barriers to starting. I had proposed a long time ago to President Macron and others that here's a simple rule. You guys have these labor laws that are incredibly punitive. For a company that's between zero employees and 20, the only laws that apply are health and safety. None of the other ones do. But when you get to 21, all of them do. So coming up with a way that generates a large number of small bets, which is what Silicon Valley does all day, that's a start.

The energy costs are going to be very hard to deal with. So you're going to do training with partner countries. The most obvious ones would be the Arab countries, but you can pick others. And if I look at ASML, ASML is one of the most important companies in the world. It's completely vertically integrated, and it's vertically integrated inside of Europe. So this proves that Europe can do it. If I can just hold forth on one more point, there's an awful lot more softer people in Europe than there are softer companies. So there's something wrong with the capital formation of software companies in Europe.

My guess is that they have trouble going from a local software to a global platform, and we need to fix that for Europe. What are the implications for democracy? I'm generally, I'm an optimist except on this question. I had thought and was trained as a citizen of America that democracy was a set of sort of values and belief systems around classic liberalism. Classic liberalism, as you know, is personal freedom, freedom of regulation, and capitalism. It doesn't look like that's really true. It looks like because of AI's use and social media's use, we can get large groups of people who really believe the wrong thing and then vote in an incorrect way to their own interests.

so the threat of ai to democracy is very real because in the hands of a skillful operator you can convince people anything is true even though it's completely false and it appears to be what appears to be true is that in a situation where the institutions are declining in trust people trust their friends and because they have friends online they think those friends are real humans so if you wanted to do a misinformation campaign to advance some evil cause which you would never do um it's relatively easy you just build fake friends and then you befriend everyone and then everyone believes you um you know if you look uh in america everyone's very upset about the data centers i happen to build data centers so i have an interest in this and so one of my friends who's a perfectly nice woman said you realize the data centers are killing cattle and i said hmm i've heard many claims but i haven't heard that one so of course i went to google i did my

Eric Schmidtresearch and there's a farmer who said that he observed this there are no police reports there's no veterinary reports in other words it's fake why why does this particular person believe it

I think she saw it on one of the Instagram or TikTok. I don't know. I didn't get into that. Because the algorithm and social media thought this is so interesting, of course, it's false, that it promoted it. So in a world where social media and AI together are working in this loop, to promote the most alarmist things. Where does truth end? Question about leadership. It's interesting how many of the founders of the great technology companies are now coming back, right? To the companies they had left. Why are they coming back? You know, I've worked in this industry for 50 years. And in my industry, the myth of the founder is very real.

I am a professional CEO. I'm not a founder, although I'm a founder of what I'm doing now. But as a professional CEO, I did not have the power that the founders of the companies I worked in did. Let's go back to Apple. Could anyone else have overseen the transformation of Apple except for Steve Jobs? A number tried. They couldn't do it. They didn't have the political currency, the economic currency, the control of the board. So let's use Mark Zuckerberg, who clearly controls his company. He decided one day to work on virtual reality, spent billions of dollars, largely, which didn't work very well. And then he pivoted to AI, which is working very well. That's a founder. Professional CEOs like me tend to get fired in those transitions.

But do they come back now because it's so interesting and so much fun, or because they want to make sure the legacy of the company remains intact? I think it's both. If you believe, as I do, that this is a historic period, that people in 100 years will all be dead, they'll be writing about the key players of how this intelligence explosion was shaped. People want to be part of that. It really is history. AI is going to affect war in horrible, in powerful and horrible ways, right? It's going to affect governance. It's going to affect jobs. It's going to affect children. You want a seat at the table, right?

So for example, another example, we need to regulate AI with respect to children. You know, I'm an adult. You can do anything you want to with me. And people, many people have tried. Let's fix that. Like now, like this year, not next year. I feel very strongly that there are people who are vulnerable, which does not include you and me, who need to be protected. How has your view on leadership changed over the years? In my professional career, the leadership started with sort of generically competent business people. They're all gone. Then it became people who had a background in leadership, and they were very, very smart.

They're all gone. And the reason is the time frame for decision making is so fast now. it's in what i do you have to have enough of a leadership capability that you can see the future and you can move quickly right because the bets are so fast and i think what what happens is everyone is jealous about my industry and everybody wants to be this person or that person and you know have the entourage and the wealth but they don't realize that the pressure that is on these people and they're typically 40 years old right so they're they're young and aggressive and so forth. Unfortunately, they're typically male, which I think it needs to get addressed. But nevertheless, this is where we are. These people are moving so fast. Imagine the pressure on them to scale at the rate they're doing. So what are the personality traits which are important now, which weren't so important before? I don't know how to say it any other way. It starts with being really smart and just the processing speed. So all of these people just see and think very quickly and they react quickly, right? In my case, if I can self-criticize, I think, but I don't react quickly. That's a defect. I need to react quicker. I need to move quicker. That's what it tells me.

What are the implications? for the way these companies should be organized? Well, the future companies, I think if you think about it for a while, I'm used to running large companies, and you are as well. If you have 10,000 people, what is the inefficiency of coordination among those 10,000 people? It's high. People spend days and days and days in meetings. I used to tell people, my job is to sit in meetings 12 hours a day, five days a week. And I work on the weekend too. So I worked 63 hours a week. That's what I told people. By the way, that's probably illegal in Europe, but that's what I did anyway. I would sit in meetings 12 hours a day, person, person,

Eric Schmidtperson, reason, narrative, so forth and so on. Much of that coordination tax can be eliminated by these ai platforms for business for the in companies that you guys invest in which i think

is you know our first audience for what we're talking about they will be far more efficient they will have fewer people in staff right one of my friends wrote a piece in the wall street journal where he had his companies growing very fast he made a 20 he decided to lay off 20 of the people he told me later he got death threats for this um who did he lay off the glue people the people who were doing sort of planning and moving moving numbers around so he didn't need them anymore what did he need he needed sales people he needed regulated people to talk to regulators and he need program people who built the products and who who talked to people on the phone although less of those um so you're going to see more much more efficient business operations so for your audience, let's say you're running a 10,000 person company. My guess is you've got a couple of thousand people who you don't need. And you better figure out a way to legally and humanely transition them, retire them out, or what have you. You just don't need them. And instead, you're going to need the great inventors, the great salespeople, because it's a human deal. You're going to need some great leaders, and you're going to need people to talk to the government.

Two questions we always ask people. One is, what's been your biggest personal failure? There was a moment when Google was ahead of everybody else in social media. It was fairly short. And I think, and I was a CEO at the time, I take responsibility for not getting ahead of that. That was a big error. And on the personal side? Without getting into too many details, I think that I'm just too slow. when i look back everything was foreseeable i just didn't see it soon enough i live in a world of time compression the people who beat me saw it quicker than i did what is something about you that most people don't know well i i i do i like to build things so we know that so i like to build organizations i'm a big time helicopter airplane pilot and i started racing cars just for fun is that what you do to switch off? I don't switch off. One of the human problems with this is if you care as much as, and I know you have the same problem, if you care at the level that I do, you can't turn it off. It's terrible. It's really hard. It's not healthy. And I think all of us are suffering from the feeling of, oh my God, oh my God, oh my God. On a Saturday night, I'll say hard enough now then i say everyone else is taking the weekend off it's just it's a problem who do

you compete against who do you want to prove well in my case i'm running four companies uh that are new and i'm trying to create a great deal of wealth to give to science when i was 22 i went to Berkeley and there was a computer that's That single computer, which was a VAX 11780, was the basis for much of the computer science that we did in networking and the internet and so forth. That one computer, and it was a few million dollars. I didn't know what it was at the time. And if I could help create that opportunity for the next 22-year-old at a time in America where the government is, for whatever stupid reason, decreasing support for basic innovation.

I'd like to do that, and I want to do it at scale. I think that if I can pump myself up, I would say I'm an expert in scaling industries. I want to scale science. The opportunity to use non-human intelligence to solve the problems of the world, of which there's a very long list, is right in front of us. I want to fund that. I want to drive it. I care about it. I'll do it until I keel over. What do you read? Well, technical reports at the moment. What is your advice to young people? That's the final question. It used to be get into biology. I think now the best example is try to build a agentic system that automates the things you care about.

One of my friends wrote, well, I have a home bot and i have a work bot and i have an email bot and i have a bot that manages my bots and i'm architecting it because frankly he wants some free time right because he's too busy in the in the weeds of it um but i think right now if i were 18 19 now i would say what's a foundation model i could build for a field i'm interested in a foundation model is a computer system that understands all the details. So for example, let's assume that I like architecture. I'd like to, which I do, I'd like to build a foundation model for architecture, which has every aspect of architecture knowledge, and I can query it. I can say, build me a house, analyze this, where are the structural points, and so forth, and accelerate all that. The same is true in biology. Another term for this is called superintelligence. Many people believe that in the next few years, we're going to build superintelligence in key categories. I'm talking to one startup, which thinks it can do visual superintelligence. In other words, superintelligence that humans can't do that is useful in the visual domain. There's clearly going to be robotic superintelligence. Don't be afraid of the word superintelligence. Superintelligence means it does things that

humans will never be able to do. Better visioning, better reasoning, drive the car better, drive the airplane better. That's interesting. Well, Eric, I have to say, because of you, this has been a super intelligent podcast. It's just been totally fantastic to listen to you. A big thank you for taking the time.