Eric Schmidt on the Future of A.I., Journalism, & Reparations | The Don Lemon Show

It's great to see you, Don, and great to be on the show. Thank you very much.

King Midas of ancient Asia Minor wished that everything he touched would turn to gold. Dionysus, the god of pleasure, granted his wish, but soon Midas's dream became a nightmare. His food and drink turned into lifeless gold. In his despair, he sought relief in the river Pactolus, trying to wash away the curse. What does this tell us about humanity trying to wield a power which they cannot possibly understand? So Henry, Dr. Kissinger, had studied this since he was an undergraduate, before we were all born, the relationships of humanity to the world around us and those sorts of things. So when he learned about AI, he said, 'Oh my God, this is going to change the relationship that humans have with each other.' And remember, we've never had another species that was smarter and more capable than humans. We've never had an alien intelligence that actually could subjugate us under certain circumstances. So he became very... we wrote a book together about the age of AI right before ChatGPT, and this is the one that now talks about what happens to society when everyone is using all these tools all the time.

Yeah, I think we forget how humans did everything. You had these polymaths such as... and they literally were masters of the universe because they could see deeper than anybody else. And indeed, in the book we start with a discussion about discovery, and we have a long history of these extraordinary people in society. This is you know Leonardo da Vinci and Einstein, people that everyone in the audience knows. And all of those people invent stuff that has huge consequences. So the question we pose in the book is, at some point we think in the next five to ten years, your phone will have on it something close to Leonardo da Vinci and Einstein and the greatest chemists and the greatest musicians, Mozart and so forth. What will you do with it, especially since every human on the planet will have this? So we've never had this kind of a situation. It means when you go to an art gallery, the phone can explain to you why the art is so exceptional and can diss the artists that aren't as good, right? It's all of this weird stuff. It also means that society will change very rapidly because these systems will become intertwined with us. Your job will depend on it, your social life will depend on it, and much more than social media because they'll be giving you advice how to behave, what to do, how to think. And there are all sorts of implications there: human agency, human freedom, things that we all are concerned about that we don't want to give up to machines.

We wrote the book in order to make the arguments. The answer is it'll be a combination of both. The upsides are obvious. Anybody in the audience have any kind of disease? Of course you do. It's probable that we can identify in the next few years all human druggable targets, which will then ultimately lead to thousands of new drugs. Now how does this occur? The computer, with what is called a drug discovery agent, can actually try something, see if it works, try again, try again, and try again, but it does that a million times an hour. Humans can't do that. Climate change: the only way we'll get climate change fixed is by new sources of energy. In order to design them, we need new kinds of materials, new kinds of ways in which they generate heat and so forth, and they get distributed. The only way to do that is with AI. I can go on. The gains in science... you sit there and you go, 'Science, oh who cares?' Well, I care a lot because science is the beginning of a new business, right? So the material science, the plastics that you use in your house, everything, the built human environment will all be changed to be more efficient, and these systems are organized around efficiency. So the proliferation of services... I'll give you another example. Most people find computers sort of opaque, it doesn't make any sense to them. They're normal human beings, they're not computer scientists. Well, wouldn't it be nice if you could just tell the computer what you'd want and it would just organize it? So an example would be: I want to build a vacation home outside of New York, select the site, figure out what kind of house I can build, consider the regulations, hire an architect, be the architect, submit the plans, hire the contractors, build the house, and pay the bills. Each of those is a function today that can be done better by an integrated computer system of agents. Now what happens to all those people? By the way, they don't go away, they just get more productive. So I'm going to tell you my opinion, which I can't prove, that everyone's going to be at least twice as productive. So in your case, because you're an incredibly well-known journalist, you can do twice as much. You'll literally be twice as better, twice as more, twice as big of a reach. In my case as a business person, I can invest twice as quickly. Scientists can do this, artists can make twice as much, teachers can teach, lawyers can sue twice as many people. Whatever thing you do, you can do twice as much because the systems are organized around making humans more productive, at least for a while.

I think that in order to believe that, you have to believe that this time is different. And in every technological revolution, there's been lots of dislocation, right? So the people who drove horses were replaced by people who drove taxis. The number one job shortage in America right now is people who want to drive trucks. We have a huge number of trucker jobs open, and the job is difficult, it's hard, it's long hours away from home, all that kind of stuff. And I spent the last ten years everybody telling me, 'Hey, you know, all those jobs are going away.' Let me tell you, the number one job in America is available to anybody who's willing to do this hardworking job.

The first and most obvious is misinformation. The systems today are poorly targeted at misinformation. You just sort of make up something and then you spread it in general, but these new systems will be able to lie to you, Don, to me, Eric, directly because it'll know exactly what your hot buttons are, it'll know what you care about and where your weaknesses are, and the system can actually just pounce on you, right? And you know, you're a normal person, you can say, 'Well, it's probably true,' you believe it, and then misinformation works. So the arrival of misinformation at scale from essentially computer-generated content, we've known this was coming, this will get a boost. That's one. Another one would be cyber attacks. We're all working towards something called artificial general intelligence, which means systems that have the kind of intelligence that humans have, where they are more flexible and creative and they are learning and so forth. When that occurs, you can imagine the system will know how, for example, to do a cyber attack to take down the financial system. How is that possible? Well, because the system sees everything, it sees what everyone sees, but it's ultimately able to run through more choices than humans. It can find more potential attacks than a human can, so it's going to be the one that an evil person will use to do their cyber attack. Another one would be biology. Turns out biology is pretty easy for computers in the sense that you can put viruses together and they more or less work, but it makes it easy to make really bad viruses as well. Now these are all issues that have to get regulated, and there are various people working on this, and on balance the benefit is really positive. Now the next question, that's kind of the tactical answer. The next question is: what does it mean to be human when your day-to-day life requires this thing, right? For advice, for helping you make productive and so forth, you become completely codependent on it, right? Is there a possible loss of free agency? Do you feel that you're losing something by being so tied to this AI intelligence? We will see.

One of the questions we talk about in the book is: let's imagine you have a system that generates math conjectures and you have another system that proves them. So it generates a conjecture and it proves it, and it involves something important like energy or dark energy or physics, and no human understands what the proof is about. You know, we know it to be true, but no human, not even the smartest mathematician, could have generated it. What is that? Is that a parody of humans? Is that a new kind of intelligence? We just don't know. So as this technology gets smarter and smarter and smarter, which we all believe it will, you will ultimately have situations which are highly uncomfortable for human identity, right? Frankly, the professor says, 'I'm sorry, I don't really know, but I can ask you know Charlie the computer.' The HAL scenario is not the one I'm particularly worried about, because in order for HAL to decide to kill everybody, it would have to have some self-image of what it is. So it would have had to be trained that it had a consciousness. Now there's an obvious question: let's assume that there's a third thing in this conversation which is a computer that says, 'I'm conscious.' How would you, Don, ask it a series of questions to determine if it's conscious or faking it? Because it could have easily learned how to answer your questions without being conscious. There are real boundaries about how we can understand what these things are. Another example would be: how do you feel about your best friend? You have kids and you have a two-year-old, and you give the computer the equivalent of a bear, and the bear every year gets smarter and the kid does too. So when the kid is ten years later, the kid is like 12 and the bear is giving the kid life advice, and it's the kid's best friend. What does that mean for human development? I mean, for you and I, we're fully formed adults, we can sort of coexist with these, but imagine the impact that this could have on teenagers. We already know that social media is terrible for teenage boys and especially for girls. There's plenty of evidence now that online... people who are unhappy, young people, they're unhappy. Typical scenario would be a young man who's having a very tough time, and the system, because it feeds them what they want, keeps feeding them conspiratorial and negative images, and eventually in some tragic situation, that's what's happening now, they kill themselves. I mean, how is that okay? And so we're going to have to have a language to... we're going to have to have a language about what it means to be human and where is the appropriate point of regulation.

No, no. In fact, Elon did that back when he had nothing in the space. He proposed that we slow everything down by six months. Today, Elon has a huge initiative called xAI, which looks like it's going to be successful, and of course Sam runs a company that is historically very important now with a valuation of close to $200 billion, with more coming. So these are people who have big stakes in the acceleration. I and others have worked hard on what guardrails we should be putting in place. I understand that President Trump intends to reverse the Biden AI regulation proposal, which was pretty light, but I'm quite sure that when you look at the potential for what China can do adversarially against the United States, we're going to want to have some limits on the technology in software and hardware that we invent that gets to them, for example. In other words, there will be reasons, even if you don't agree with what Biden did, in the Trump administration for national security reasons to make sure the country remains safe.

No, I'm afraid of humans' reaction to it. To me, AI is this unalloyed good. The concept of having a partner that can do the stuff that's hard for you, the memorization... When I was a boy, I remember in school I was required to memorize the names of the counties of Virginia. I grew up in Virginia. And today, you know, why would you ever ask anyone to memorize such a thing? It's a waste of time. And the idea of having every single human in America and indeed in the world operating at their full potential, it has to be good.

I think what you propose can be done. I'm skeptical that our political leadership will actually embrace an analytical framework, because everyone comes at this with a pre-assigned bias. What you're describing is what we call a foundation model. In a foundation model, you take everything that you know and you put it in the model, and then it can answer questions that are informed about what it learned in the synthesis of all the data. So in your case, let's use slavery, racism, and discrimination globally. You could probably come up with some common patterns. The systems are particularly good at finding patterns that humans don't see, because they see everything faster than we do. They're not smarter than we are, but they see it faster. They see a million points where the average scholar can see a thousand, and I can see ten. So the ability to do technically data fusion across that means we will really understand how bad the suffering was, to the degree that we have the sources. And also you can ask it where are the holes. So another good example is in humanities. People spend years in the libraries looking at these ancient texts. Why in the world just feed all those ancient texts into the model and start asking it questions? Because it can modernize its analysis and it's going to be a better analyst than you. So again, it makes the historian more accurate, it makes the person working on the price of reparations more worth it, you know, what have you.

Well, in the last few years, it's pretty clear that facts are not as important as they should be. So I was always raised, speaking as a scientist, that you and I can disagree on what to do, but we don't disagree on facts. Facts and reality, and I think that as a society we have to sort of recognize that critical thinking is important. I spent all day being pitched things, and I understand that advertising is trying to sell me things, and misinformation is somebody's trying to convince me of something. So I am a natural skeptic, and what I do is I use Google. You know, somebody says, 'Did you know that the women in Boston had a blue hair day?' Right, I'll make that up, and I would say, 'Sounds unlikely, check.' Right, and then of course it's a joke, and I say, 'Well, you didn't fool me.' So my view of information is that I have to be, as a citizen, armed with the tools of critical thinking, which is not the same thing as... you and I may disagree on what to do about reparations, but we should agree that there were an awful lot of people who were really severely harmed. Yeah, right. And we should actually have the data that says this horrible thing happened and this horrible thing happened and this good thing happened, and then we can argue about whether we address it or whether we address it at all.

Well, he's brilliant, and I know him well as I know you do. The answer to the question of misinformation has to do with how the models are trained. So I'll give you an example.

So let's have fun and we'll make a model that lies all day, right? Everything it says is false. Now would that be useful? It might be if you're a purveyor of misinformation. So the fact that there's not a single definition of truth and the fact that there will be many, many, many entrants means that there will be a model that lies, it's full of misinformation. And so if you have this notion of the false equivalence of something is true and something's false, and you present them both, then you're going to end up doing a disservice. I would much rather, for you especially as a journalist and I know you're committed to this, try to figure out what's true and then try to say what the choices are about this fact.