Eric Schmidt: The Implications of Superintelligence

0:00My own view is that the AI boom is underhyped because people do not appreciate, although it seems crazy. these tools should be powerful enough to materially change the essentially productivity curve of humans at a scale that we've never seen before, more than, for example, clean water and electricity. So the arrival of a new intelligence is a very, very big deal for humans. That's Eric Schmidt, best known as the CEO who helped transform Google from a promising startup into one of the best businesses in human history. But Eric's story doesn't end with search. In recent years, he's become one of the most important voices in the global conversation about artificial intelligence, helping leaders make sense of what's coming and why it matters.

0:51This is Mike Maples Jr. of Floodgate, and it's go time with Eric Schmidt.

0:58This is Mike Maples Jr., and welcome to the Pattern Breakers podcast, where we explore why some founders radically change the future and how they stand apart. Together, we'll learn about the counterintuitive mindsets and actions behind their remarkable success. Brace yourself for a world where chaos is welcome, naysayers are often a positive signal, and movements galvanize misfits who transform the impossible to the inevitable.

1:35Eric Schmidt doesn't just help build the future, he interrogates it. In The Age of AI and Genesis, co-authored with Henry Kissinger, he goes beyond examining what AI can do. And he asks, what happens to human judgment when machines produce knowledge we can't explain? What becomes of dignity, diplomacy, and democracy when intelligence is decoupled from morality? In our episode, Eric explores how AI is transforming human agency, global stability, and the very nature of innovation. From personal superintelligence to geopolitical risk, this conversation dives deep into the moral, philosophical, and strategic stakes of AI's rise, and what it means for the future of humanity.

2:23Let's catch up with them.

2:32Eric Schmidt, welcome to the podcast. I'm so happy to be with you, Mike. I'm excited that you would take the time to do this because I've been spending a lot of time on AI, like a lot of folks. And you've written a couple of pretty interesting books on it that I've devoured, both The Age of AI with Henry Kissinger and Dan Huttenlacher, and then Genesis with Henry Kissinger and Craig Mundy. So I think it'd be good to just talk about the circumstances of each book, what caused you to partner with these people, and what you were trying to convey with each book. You know, it's interesting that for some interesting reason in my life, I guess, I ended up being best friends with Dr.

3:13Kissinger over 15 years. And when he first heard Demis Hassabis speak about AI, a light bulb went off in his brain because he had worked on Kant when he was an undergraduate at Harvard before any of us were alive. And he saw the parallels of the impact of this technology on humanity. Combining my strengths, Henry's strengths, and then Dan and Craig made perfect sense because we wanted strong technical people as well as policy people, and plus, obviously, the greatest diplomat alive at the time. That's what drove those. And what's interesting is that Dr. Kissinger Henry spent his time really accepting that AI was going to happen and what would happen to society.

3:59And our industry, you're very much a part of this, often invent things without understanding the secondary impacts on society. And it's generally agreed that social media has had deleterious effects that were not foreseen when social media took off 15 years ago, and that we collectively did not see them coming. My idea with these books was to say, let's see what's coming and try to get ready for it. And I'm more convinced than ever that at the rate of innovation of artificial intelligence, society will change quite rapidly, well ahead of where the governments, the laws, and the sort of moral principles really are.

4:41I've felt this a lot, that one of the positives about tinkering is that the tinkerers tend to get to decide what's going to happen in the future, right? Like the aeronautics experts didn't get to decide what planes were going to be and when they were going to happen. A couple of guys in Kitty Hawk did, a couple of bicycle mechanics tinkering with this stuff. But one of the challenges of that is that not that many people I know in the technology industry are very philosophically grounded. You know, not many people in the tech industry studied Kant or Sartre or Aristotle or Thomas Aquinas or some of these people throughout history who've really had some powerfully profound ideas about what the meaning of life is and what a good life is and what ethics are and those types of things.

5:29And so I'm curious if there are philosophers that have resonated with you that would be valuable for people in the technology industry writ large to understand, right, as part of their ethical grounding as they think about building the future? Well, when we worked on the books, we spent quite a bit of time looking at the theories of Kant, which have to do with human identity. Who am I? What is my identity? How do I interact with the world? And also a philosopher named Spengler, who essentially tried to develop a mathematical order of the world. And he was one of these early people who was a mathematician who basically said, there is clearly a mathematical order for the world around us and human behavior.

6:12And by the way, he failed. And he was ultimately, because it was at the time of anti-religious stuff, he was actually essentially jailed over his beliefs that they were secular as opposed to godlike powers. And he ultimately renounced his mathematical ideas and proclaimed his faith in God for whatever reason. So there's no obvious philosophical thought that we can apply to what we're talking about today. I personally like deism, which is a religion from a few hundred years ago, which is essentially self-reliance. And one of the things that happens with the online world is we're ultimately getting people who are traveling in their own ships in the ocean.

6:57And we're not operating like flotillas. We're actually operating as our own ships. Occasionally, we meet up with other ships and so forth. And so you end up with a situation where the hunger for connection and immediacy is so strong, you end up being lonelier than before, which is a very strange conundrum. the key message for your audience is that this stuff is going to happen really fast now how fast in the next year we're going to have essentially AI mathematicians so you can have a million mathematicians all they do is need electricity AI programmers all they need is electricity they don't need food, they don't need pizza all the usual things that programmers eat and the combination of math and programming is kind of the basis of everything digital.

7:46And so in the next years after that, you're going to see these enormous changes in physics and chemistry and material science leading to incredible discoveries and also acceleration of war, much more time-sensitive decisions, cyber biohacks, all those kinds of things. We document them, as you know, in our books, but that's coming. It's not like it's maybe coming. I'm telling you it's coming and it's coming in the next 12 months. This has lots of implications. So for example, what happens to the distribution of wealth? What happens to jobs? What happens to people's identity? And to me, the most interesting question is what does it mean to be human in an age where these things are superpowers?

8:34So I'll give you an example. In five years, probably, we're going to have the equivalent of your phone have 90 % of the world's best artist, designer, mathematician, physicist, chemist, so forth, in your pocket, serving you. Now, what are you going to do with that? Well, you could ask it to invent some new math or some new game. You could ask it to invent an attack on your enemy. or you could imagine that it would do comedy at a scale that's never been seen before in order to harass your opponents. Because there's so many people in the world and each person would have access to this, you don't have the restraints of industrial morals, institution, and power.

9:20You have it's a free-for-all. And in a free-for-all, behavior is determined not by law but by culture. So we don't have a way of discussing how that free-for-all will work. We can't even agree on basic rights like privacy. And the epoch that this innovation is arriving into is a new one, because it's an epoch where humans and a non-human intelligence coexist. That coexistence will become as important or more important than China versus the US, because it touches everything. I guess I've kind of been around the computer industry for a long time, right? My dad was even in it, you know, since the 60s.

10:02And I remember when the personal computer came out, it wasn't so much a substitute for mainframes, which is where the power was. It was that you could apply computation to things you would have never dreamed to apply it to. And then the internet comes out and you do the same with communications bandwidth and networks. I've been doing a lot of things with the AI where if you have abundant cognition, you can just run analyses that you would never dream of asking a person to run. It's almost like you have a trillion people in the world willing to do it rather than a few billion people in the world.

10:36And let me just interrupt that your father was a good friend of mine and he was an incredibly great human being. Just so we get the facts straight. Yeah, on this we agree. Many things we agree. This is one of those things we agree on. Yeah. You know, I'm very interested in the automation boundary between human and robotics in manufacturing. It's a question I don't and I'm not an expert in it. So you sit down and each of these systems now has this thing called deep research. And, you know, you ask it a relatively straightforward question and then you watch it and it's spending, I don't know, 10 minutes.

11:13Do you have any idea how much computation 10 minutes of a supercomputer is? It's insane, right? It's available to me. It produces this result. I then send it around to the company saying, you know, is there anything new here? And yeah, there were some new insights and so forth. These are professionals. So it gives you a sense of the power and unification of knowledge if you ask the right question. The key here is we collectively want humanity to innovate very quickly to solve the hard problems in humanity, which I will define as education, health care, war, climate change, things like that. And the same tools can also be used to destroy them.

11:53So an example would be, you could ask it, and I don't recommend this, what would be the most effective claims to deny climate change? It would produce them. Show me a way to attack France and disable their nuclear facilities. And it would say, well, I don't really speak nuclear. So then you'd say, well, tell me how to break into their servers. And then you'd lead it through it. Another example is in biology. We've known for a long time how to build biological pathogens, which are bad. That information is generally available, but the only people who can understand it are PhDs in biology. Thank God there are very few evil ones.

12:34There are very few Osama bin Laden type people. Thank God. But this tool would allow a less than brilliant evil person to follow a recipe. So what happens is another query would be, show me how to do the following evil thing. And it would give you the recipe. And it would say, well, you have to buy this and you have to buy that. And then you have to mix it this and so forth. It's like a kitchen. And we just published an emerging biothreat analysis in Washington a week ago from the government, from the Congress, which goes into these threats in some detail. I'm not suggesting the world is over.

13:10What I'm suggesting is it's a new power center where this new power needs to be used correctly and the bad parts of it need to be policed in some appropriate way. And that's a really good use of all of our time. Yeah, and it's interesting to me because part of what I think you're highlighting is AI allows very high agency people to basically force multiply their capabilities. You can have high agency good guys and high agency bad guys. And as you were telling that example, one of the things that I internalized is that part of what the good guys are going to have to do is imagine what a high agency bad person would be and then what would be signals that they're engaging in those activities.

13:56If you need to buy certain things to attack a nuclear reactor, then you'd want to use the AI to help the good guys find the people who are trying to get those things, signals that somebody's engaging in those activities before they can pull them off. No, I agree with that. In my military work, I learned a term called offense and defense dominance. So offensively dominant means that the offense tends to win and the defense is not strong enough. Defense dominant means that the offense can attack, but the defense is strong enough that it can repel and get stronger because of the attack. So it turns out that there are some problems that are offense dominant, which favor the offense.

14:42The most obvious one is biology. You can do a lot of damage in biology before the antidote arrives. The question you have in that scenario is, how do you control that? But it gets worse. And I'll give you the example that I don't think anyone has a solution for. So you're the good guy and I'm the bad guy. and we are in a race for super intelligence, which is intelligence greater than that of any human. It's more than Einstein. And you're ahead of me. And for purposes of argument, I'm six months behind. Now, six months is not very much, except that this is a network effect business. And network effect businesses, which I think most people don't understand, have the property that the network gets stronger and you get more and more market share.

15:31And ultimately, they tend to end up at about a 90 % outcome. And the reason is that you just get smarter faster than anyone else. You get more information, more resources, and so forth. Most of the tech businesses operate in network effect positions. They ultimately end up with 90 % market share. You know this from Microsoft. It's true in many, many fields. I know that you're ahead of me. And I also know that when you invent super intelligence, your ability to innovate ahead of me will be infinite. it. What are my choices? So I'm the bad guy. The first thing I'm going to try to do is steal your knowledge.

16:08Second thing I'm going to try to do is infiltrate with spies your team. The third thing I'm going to try to do is to change your algorithm. It's called adversarial attacks, where I attack your model. Let's say all of those fail. What's my next choice? Bomb your data center. That escalation path is not unlike what we faced in the early 1950s with nuclear power. But in nuclear, you were counting weapons, whereas in AI, there's no weapons to count. What you're doing is you're understanding innovation. The good news is this is still a theoretical scenario, but I think it's true that in the next five to 10 years, we'll have the ability to build systems that are more super intelligent than humans, they are probably destabilizing to the nation state.

17:01If somebody is going to say, this is not okay, I'm going to bomb your data center. And I'm obviously not advocating that. Now, why bomb the data centers? They're really big. Furthermore, you know exactly where they are. Furthermore, all you have to do is bomb their transmission lines and you create havoc. And no one is busy burying their data centers and their transmissions lines against military attack from an opponent. I'm not advocating in this any way, but I'm using this as an example of the kind of getting ahead thinking. I want to get ahead of this problem. I want a doctrine ahead of the first attack.

17:39And, you know, there's historical precedence for that, right? The internet itself was in many ways a defensively dominant technology architecture, right? It was designed to be anti-fragile at its design center. Probably we need to be thinking that way about AI more. In fact, I and Dan Hendricks and Alex Wang wrote a paper on superintelligence where we actually talked about this anti-fragile question. We call them wicked hard problems. You can imagine situations, for example, where the chips all report what they're doing and where they are. so you know what's happening. The U.S. government has been putting export controls on chips to China.

18:19That's helpful, right? They most recently this week banned the H20 chip, which led to NVIDIA taking a$5 billion write-off. That's helpful from a national security perspective to slow this process down. I think, you know, going back to your original questions which have to do with innovation. Why is this happening? It's happening because too many people have too much money and it's too sexy not to work on. A normal person would say, Mike and Eric, my God, why are you doing this? Why don't you do a pause? And the answer is there's too many people, too many players, too many nations to do a pause.

18:59There's too much at stake for every player. So what we have to do is we have to do a better job of solving this, right? By coming up with stable solutions. And, you know, the nuclear umbrella is not perfect, but the reason we're alive today is because people had good judgment around limitations, notification, weapons inspections, and things like that to keep the pressure level low. Dr. Kissinger was extremely worried about the following scenario that what would happen with China is that you'd have a conflagration and then you'd a small one and then you have another one. He would always explain it as it was like World War One.

19:41When the Sarajevo maneuver occurred in World War One, no one understood that within a few months they would have a world war that was at the time extraordinarily horrific for human life. We're playing with fire in the sense that it's not like the Chinese were happy that we were 80 % of the GDP of the world 50 years ago. They want their share too. That competition is natural in human behavior. The competition is fine. The violence and the destruction is not. You see this with Russia and Ukraine. Yeah. And also, I think what you're kind of alluding to is part of why we didn't have nuclear war was we had diplomats and people who had theories about, They applied game theory logic to avoiding nuclear war, and they understood the philosophical underpinnings of certain moves you could make that would increase the odds of war happening and certain moves that you could make that would decrease the odds of war happening.

20:39And that's an important thing to model out, right? It's important to have an intentional strategy related to that rather than just see what happens. Yeah. And an example, and we talked about this in the first book with Kissinger, we spent a lot of time talking about the origin of nuclear containment and why this was relevant to AI. And the key point in the book is that the notion of mutually assured destruction has not actually been tested. Thank God. Thank God it's not been tested. And we don't want to run that test. So there may be things we encounter with AI where we don't actually want to run the test to see if the doctrine works.

21:20We just want to believe that it works. And that ultimately, our ability to manage this transition will, that we collectively in the world come to some agreement on where the limits are. The most obvious limit, by the way, with China is a no surprise rule. My own view is that we can find that path in AI to watch for really bad things. And a good example is that we have a list of things in AI that we're worried about. One is exfiltration of the model, where the model decides to make copies of itself on its own. In other words, reproduction. Another one is what is called continuous recursive learning on its own, where we don't know what it's learning.

22:00Another one is if you have agents and the agents decide to talk not in English, but in a language they invent that we don't understand. Our view collectively is that those are points to say, stop and try to do this in such a way that we understand what's going on. So one of the great dangers here is that you get some system that's operating on its own that does something which, in its opinion, is reasonable, which ultimately creates some huge conflict. So that's called agency. It's really important that we, and understandability, it's really important that we have human agency and control, human alignment, if you will, as well as interpretability and understanding exactly what it's doing.

22:41We don't fully understand what these models are doing, but we're getting a better understanding of it. I should say, by the way, if you look at DeepSeek R1, OpenAI 03, Gemini has a similar one they've just brought out. If you look at the planning stuff that they're doing, like solving complicated problems, it will blow you away. Its ability to solve problems is well beyond my ability at my best, let alone now. In my specific area, that's an example of the arrival of an intelligence better than I would ever be. So I'm curious, and this may take us a little bit back into the past, but Google has obviously been huge in the development of AI.

23:23Was there a time that that occurred to you for the first time? And what was that? When was that moment where you're like, okay, this is going to be core to everything? Well, it's interesting that Larry Page was a student in AI. And he always wanted to do AI things. And we created a lab, Larry and Sergey created a lab called Google X. And they hired a couple of people who are very well known now, including Andrew Ng out of Stanford. And they did something where they took all of the YouTube videos and they tried to classify them. And in that classification, they discovered the concept of a cat, not a cat, but the concept of a cat, just from reading the videos without any text training and so forth.

24:07And that was the first result that I'm aware of, of the use of AI at scale and so forth. And this was at the time of this thing called ImageNet, which was a competition to get better. This is 2011, 2012. We then went into a process where we thought, this is interesting. let's see if we can apply this to our ad system. So they created ad systems, again, very tactical, that increased revenue. We then came across a set of companies, which included one of the core teams that invented deep learning. This is Jeffrey Hinton. We bought them. And we also bought Deep Mind. These were all at the time, relatively expensive without a certain notion of revenue.

24:51But the idea from the management was that this technology would eventually strengthen search and ads. With DeepMind, what happened was it was essentially a gaming company at the time working on new algorithms and reinforcement learning. And they had this thing called they wanted to win the Go game. Go is a game that's played largely in Asia, very important, very hard to win. Computers couldn't handle it computationally. and so i they announced they were going to fight or have a match if you will with a korean and i i went i had no idea if these guys could do it it was interesting was and because of who i was i got to sit on the korean side and the computer side so the koreans are going in with a translator we're going to beat that computer no one can beat us you know we've done this for years we're super proud and the and the computer people were like no problem we've got this one way to understand the the algorithm did is it was entirely optimized to increase the probability of winning every move.

25:53It would make decisions that made no sense, but there were small increases in probability from 50%. You start off at 50%, right? In a two-person game. And I'm looking at the probability inching up. And on the other side, I go visit the Koreans and they're all scheming. What a stupid move. I've never seen such an idiot. You know, we've got it hands down. And as you know, the best Korean in the world lost. A year later, we did the same thing in China. Same thing, same human arrogance. And this was the best player in the entire world, even better than the Koreans. And the Chinese were so upset that in the middle of the match, which was broadcast to a few hundred million people, they shut the video off.

26:38That was, for me, the moment. A year later, Google produced a paper called the Transformer Architecture, Attention is All You Need, which was an algorithm that allowed for scaling. Now, why did all this occur? Well, we were talking earlier about how does innovation occur? It's a little bit of randomness. This team wanted this, this team one of that. And it wasn't at all clear that the transformer paper, which was just published, would become the seminal paper to create the entire industry. GPT-1, the T is transformers, was a small system. And then two was slightly better. And then in OpenAI's case, my friends were there and they told me, they turned it on and they said, oh my God, this thing can write.

27:25right it was a real discovery and sam altman today says that it was a real invention of humanity that you could use these algorithms that had been pioneered over a decade to produce this kind of analysis and i think that set off the current boom we had no idea during those years of what was going to happen but it was super interesting and we had a sense that At a minimum, it could help search in ads. Yeah, and I'm curious, Eric, because I've seen this time and again that it's like breakthroughs. You can't have a recipe for them because they're undiscovered. And so there's this paradox where most of the breakthroughs are discovered because somebody's pursuing something they're obsessively interested in for its own sake.