Sean Carroll: The Nature of the Universe, Life, and Intelligence | Lex Fridman Podcast #26

The following is a conversation with Sean Carroll. He's a theoretical physicist at Caltech, specializing in quantum mechanics, gravity, and cosmology. He's the author of several popular books, one on the Arrow of Time called From Eternity to Here, one on the Higgs Boson called Particle at the End of the Universe, and one on science and philosophy called the Big Picture on the origins of Life, meaning, and the Universe itself. He has an upcoming book on quantum mechanics that you can pre order now called Something Deeply Hidden. He writes one of my favorite blogs on his website, preposterousuniverse.com I recommend clicking on the Greatest Hits link that lists accessible interesting posts on the Arrow of Time, Dark Matter, Dark energy, the Big Bang, general relativity, string theory, quantum mechanics, and the Big Meta questions about the philosophy of science, God, ethics, politics, academia, and much, much more. Finally, and perhaps most famously, he's the host of a podcast called Mindscape that you should subscribe to and support on Patreon. Along with the Joe Rogan Experience, Sam Harris's Making Sense, and Dan Carlin's Hardcore History, Sean's Mindscape podcast is one of my favorite ways to learn new ideas or explore different perspectives and ideas that I thought I understood. It was truly an honor to meet and and spend a couple hours with Sean. It's a bit heartbreaking to say that for the first time ever, the audio recorder for this podcast died in the middle of our conversation. There are technical reasons for this having to do with Phantom power that I now understand and will avoid. It took me one hour to notice and fix the problem.

So much like the universe is 68%.

Dark energy, roughly the same amount from this conversation was lost except in the memories of the two people involved and and in my notes. I'm sure we'll talk again and continue this conversation on this podcast or on Sean's. And of course I look forward to it. This is the Artificial Intelligence podcast. If you enjoy it, subscribe on YouTube, iTunes, support it on Patreon, or simply connect with me on Twitter lexfriedman. And now, here's my conversation with Sean.

Carol, what do you think is more interesting and impactful? Understanding how the universe works at a fundamental level, or understanding how the human mind works?

Of course this is a crazy, meaningless, unanswerable question in some sense because they're both very interesting and there's no absolute scale of interestingness that we can rate them on. There's the glib answers that says the human brain is part of the universe and therefore understanding the universe is more fundamental than Understanding the human brain.

But do you really believe that once we understand the fundamental way the universe works at the particle level, the forces, we would be able to understand how the mind works?

No, certainly not. We cannot understand how ice cream works just from understanding how particles work. Right. So I'm a big believer in emergence. I'm a big believer that there are different ways of talking about the world beyond just the most fundamental microscopic one. When we talk about tables and chairs and planets and people, we're not talking the language of particle physics and cosmology. So. But understanding the universe, you didn't say just at the most fundamental level. Right. So understanding the universe at all levels is part of that. I do think, you know, to be a little bit more fair to the question, there probably are general principles of complexity, biology, information processing, memory, knowledge, creativity, that go beyond just the human brain. Right. And maybe one could count understanding those as part of understanding the universe. The human brain, as far as we know, is the most complex thing in the universe. So there's. It's certainly absurd to think that by understanding the fundamental laws of particle physics, you get any direct insight on how the brain works.

But then there's this step from the fundamentals of particle physics to information processing, which a lot of physicists and philosophers maybe a little bit carelessly take when they talk about artificial intelligence. Do you think of the universe as a kind of a computational device?

No. To be like the honest answer, there is no. There's a sense in which the universe processes information clearly. There's a sense in which the universe is like a computer. Clearly. But in some sense, I think I tried to say this once on my blog and no one agreed with me. But the universe is more like a computation than a computer, because the universe happens once. A computer is a general purpose machine that you can ask it different questions, even a pocket calculator, and it's set up to answer certain kinds of questions. The universe isn't that. So information processing happens in the universe, but it's not what the universe is. And I know your MIT colleague Seth Lloyd feels very differently about this. Right.

Well, you're thinking of the universe as a closed system.

I am.

So. So what makes a computer more like a. Like a PC, like a computing machine, is that there's a human that everyone comes up to it and moves the mouse around. So input gives it input gives it input. And you're. And that's why you're saying it's just a computation, a deterministic thing that's just unrolling but the immense complexity of it is nevertheless like processing. There's a state, and then it changes with good rules. And there's a sense for a lot of people that if the brain operates, the human brain operates within that world, then it's simply just a small subset of that. And so there's no reason we can't build arbitrarily great intelligences.

Yeah.

Do you think of intelligence in this way?

Intelligence is tricky. I don't have a definition of it offhand. So I remember this panel discussion that I saw on YouTube. I wasn't there, but Seth Lloyd was on the panel, and so was Martin Rees, the famous astrophysicist. And Seth gave his shtick for why the universe is a computer. And he explained this, and Martin Reese said, so what is not a computer? Seth is like, oh, that's a good question. I'm not sure. Because if you have a sufficiently broad definition of what a computer is, then everything is right. And the similarly, or the analogy gains force when it excludes some things. Is the moon going around the Earth performing a computation? I can come up with definitions in which the answer is yes, but it's not a very useful computation. I think that it's absolutely helpful to think about the universe in certain situations, certain contexts, as an information processing device. I'm even guilty of writing a paper called Quantum Circuit cosmology, where we modeled the whole universe as a quantum circuit.

As a circuit.

As a circuit, yeah.

And with qubits, kind of thing.

With qubits, basically, Right? Yeah. So qubits becoming more and more entangled. So do we want to digress a little bit? Because it's kind of fun. So here's a mystery about the universe that is so deep and profound that nobody talks about it. Space expands, right? And we talk about, in a certain region of space, a certain number of degrees of freedom, a certain number of ways that the quantum fields and the particles in that region can arrange themselves that that number of degrees of freedom in a region of space is arguably finite. We actually don't know how many there are, but there's a very good argument that says it's a finite number. So as the universe expands and space gets bigger, are there more degrees of freedom? If it's an infinite number, it doesn't really matter. Infinity times two is still infinity. But if it's a finite number, then there's more space, so there's more degrees of freedom. So where did they come from? That would mean the universe is not a closed system. There's more degrees of freedom popping into existence. So what we suggested was that there are more degrees of freedom, and it's not that they're not there to start, but they're not entangled to start. So the universe that you and I know of, the universe, the three dimensions around us that we see, we said those are the entangled degrees of freedom making up space time. As the universe expands, There are a whole bunch of qubits in their zero state that become entangled with the rest of space time through the action of these quantum circuits.

So what does it mean that there is now more degrees of freedom as they become more entangled? Yeah, so as the universe expands.

That's right. So there's more and more degrees of freedom that are entangled that are playing the role of part of the entangled spacetime structure. So the basic, the underlying philosophy is that space time itself arises from the entanglement of some fundamental quantum degrees of freedom.

Wow. Okay, so at which point is most of the entanglement happening? Are we talking about close to the Big bang? Are we talking about throughout the time of the.

Yeah. So the idea is that at the Big Bang, Almost all the degrees of freedom that the universe could have were there, but they were unentangled with anything else. And that's a reflection of the fact that the Big bang had a low entropy. It was a very simple, very small place. And as space expands, More and more degrees of freedom Become entangled with the rest of the world.

Well, I have to ask John Carroll, what do you think of the thought experiment from Nick Bostrom that we're living in a simulation? So I think. Let me contextualize that a little bit more. I think people don't actually take this thought experiment. I think it's quite interesting. It's not very useful, but it's quite interesting from the perspective of AI. A lot of the learning that can be done Usually happens in simulation from artificial examples. And so it's a constructive question to ask how difficult is our real world to simulate? Right. Which is kind of a dual part of. If we're living in a simulation and somebody built that simulation, if you were to try to do it yourself, how hard would it be?

So obviously we could be living in a simulation. If you just want the physical possibility, then I completely agree that it's physically possible. I don't think that we actually are. So take this one piece of data into consideration. You know, we. We live in a big universe. Okay. There's 2 trillion galaxies in our observable universe with 200 billion stars in each galaxy, etc. It would seem to be a waste of resources to have a universe that big going on just to do a simulation. So in other words, I want to be a good Bayesian. I want to ask, under this hypothesis, what do I expect to see? So the first thing I would say is I wouldn't expect to see a universe that was that big. Okay? The second thing is I wouldn't expect the resolution of the universe to be as good as it is. So it's always possible that if our superhuman simulators only have finite resources, that they don't render the entire universe right. That the part that is out there, the two trillion galaxies, is. Isn't actually being simulated fully. Okay, but then the obvious extrapolation of that is that only I am being simulated fully. Like the rest of you are just part non player characters. Right. I'm the only thing that is real. The rest of you are just chatbots beyond this wall. I see the wall, but there is literally nothing on the other side of the wall that is sort of the Bayesian prediction. That's what it would be like to do an efficient simulation of me. So none of that seems quite realistic. I hear the argument that it's just possible and easy to simulate lots of things. I don't see any evidence from what we know about our universe that we look like a simulated universe. Now maybe you can say, well, we don't know what it would look like, but that's just abandoning your Bayesian responsibilities. Your job is to say, under this theory, here's what you would expect to see.

Yeah. So certainly if you think about simulation as a thing that's like a video game where only a small subset is being rendered, but say the entire, all the laws of physics, the entire closed system of the quote, unquote universe, it had a creator.

Yeah. It's always possible.

Right. So that's not useful to think about when you're thinking about physics. The way Nick Bostrom phrases it, if it's possible to simulate the universe, eventually we'll do it. Right? You can use that, by the way.

For a lot of things.

Well, yeah, but I guess the question is how hard is it to create a universe?

I wrote a little blog post about this and maybe I'm missing something, but there's an argument that says not only that it might be possible to simulate a universe, but probably if you imagine that you actually attribute consciousness and agency to the little things that we're simulating, to our little artificial beings, there's probably a Lot more of them than there are ordinary organic beings in the universe or there will be in the future, right? So there's an argument that not only is being a simulation possible, it's probable because in the space of all living consciousnesses, most of them are being simulated, right? Most of them are not at the top level. I think that argument must be wrong because it follows from that argument that if we're simulated, but we can also simulate other things. Well, but if we can simulate other things, they can simulate other things, right? If we give them enough power and resolution and ultimately we'll reach a bottom because the laws of physics in our universe have a bottom made of atoms and so forth. So there will be the cheapest possible simulations. And if you believe the original argument, you should conclude that we should be in the cheapest possible simulation because that's where most people are. But we don't look like that. It doesn't look at all like we're at the edge of resolution. You know, that we're 16 bit, you know, things, but it seems much easier to make much lower level things than we are. So. And also I question the whole approach to the anthropic principle that says we are typical observers in the universe. I think that that's not actually. I think that there's a lot of selection that we can do that we're typical within things we already know, but not typical within all the universe.

So do you think there is intelligent life? However you would like to define intelligent life out there in the universe, My.

Guess is that there is not intelligent life in the observable universe other than.

Us.

Simply on the basis of the fact that the likely number of other intelligent species in the observable universe, there's two likely numbers, zero or billions. And if there had been billions, we would have noticed already for there to be literally like a small number, like, you know, Star Trek, there's, you know, a dozen intelligent civilizations in our galaxy, but not a billion. That's weird that. That's sort of bizarre to me. It's easy for me to imagine that there are zero others because there's just a big bottleneck to making multicellular life or techn. Life or whatever. It's very hard for me to imagine that there's a whole bunch out there that have somehow remained hidden from us.

The question I'd like to ask is what would intelligent life look like? What I mean by that question and where it's going is what if intelligent life is just fundamental in some very big ways different than the One that is on Earth, that there's all kinds of intelligent life that operates at different scales of both size and temporal.

Right. That's a great possibility because I think we should be humble about what intelligence is, what life is. We don't even agree on what life is, much less what intelligent life is. Right. So that, that's an argument for humility, saying there could be intelligent life of a very different character. Right. Like you could imagine that dolphins are intelligent but never invent space travel because they live in the ocean and they don't have thumbs. Right? So they never invent technology, they never meant smelting. Maybe the universe is full of intelligent species that just don't make technology.

Right.

That's compatible with the data. I think, and I think maybe, maybe what you're pointing at is even more out there versions of intelligence. You know, intelligence in intermolecular clouds or on the surface of a neutron star, or in between the galaxies in giant things where the equivalent of a heartbeat is 100 million years. On the one hand, yes, we should be very open minded about those things. On the other hand, all of us share the same laws of physics. There might be something about the laws of physics, even though we don't currently know exactly what that thing would be, that makes meters and years the right length and time scales for intelligent life. Maybe not. But we're made of atoms. Atoms have a certain size, we orbit stars, stars have a certain lifetime. It's not impossible to me that there's a sweet spot for intelligent life that we find ourselves in. So I'm open minded either way. I'm open minded either being humble and there's all sorts of different kinds of life, or no, there's a reason we just don't know it yet. Why life like ours is the kind of life that's out there.

Yeah, I'm of two minds too, but I often wonder if our brains is just designed to quite obviously to operate and see the world in these timescales and we're almost blind and the tools we've created for detecting things are blind to the kind of observation needed to see intelligent life at other scales.

Well, I'm totally open to that. So here's another argument I would make. We have looked for intelligent life, but we've looked for it in the dumbest way we can. Right, by turning radio telescopes to the sky. And why in the world would a super advanced civilization randomly beam out radio signals wastefully in all directions into the universe? That just doesn't make any sense. Especially because in Order to think that you would actually contact another civilization, you would have to do it forever. You have to keep doing it for millions of years. That sounds like a waste of resources. If you thought that there were other solar systems with planets around them where maybe intelligent life didn't yet exist, but might someday, you wouldn't try to talk to it with radio waves. You would send a spacecraft out there and you would park it around there and it would be like, from our point of view, be like 2001 where there's a monolith.

Monolith.

So there could be an artifact. In fact, the other way works also, right? There could be artifacts in our solar system that have been put there by other technologically advanced civilizations and that's how we will eventually contact them. We just haven't explored the solar system well enough yet to find them. The reason why we don't think about that is because we're young and impatient. Right. Like it would take more than my lifetime to actually send something to another star system and wait for it and then come back. But if we start thinking on hundreds of thousands of years or million year time scales, that's of kind, clearly the right thing to do.

Are you excited by the thing that Elon Musk is doing with SpaceX in general space, but the idea of space exploration, even though you're or your species is young and impatient?

Yeah. No. I do think that space travel is crucially important long term, even to other star systems. And I think that many people overestimate the difficulty because they say, look, if you travel 1% the speed of light to another star system, we'll be dead before we get there. Right. And I think that it's much easier. And therefore when they write their science fiction stories, they imagine we go faster than the speed of light because otherwise they're too impatient.

Right.

We're not going to go faster than the speed of light, but we could easily imagine that the human lifespan gets extended to thousands of years. And once you do that, then the stars are much closer, effectively. Right. What's a hundred year trip? Right. So I think that that's going to be the future, the far future, not, not my lifetime once again, but baby steps.

And unless your lifetime gets extended, well.

It'S in a race against time. Right. A friend of mine who actually thinks about these things said, you know, you and I are going to die, but I don't know about our grandchildren. That's, that's, I don't know. Predicting the future is hard, but that's at least a Plausible scenario. And so, yeah, no, I think that as we discussed earlier, there are threats to the earth, known and unknown. Right. Having spread humanity and biology elsewhere is a really important long term goal.

What kind of questions can science not currently answer, but might soon? When you think about the problems and the mysteries before us that may be within reach of science, I think an.

Obvious one is the origin of life. We don't know how that happened. There's a difficulty in knowing how it happened historically actually, you know, literally on Earth. But starting life from non life is something I kind of think we're close to.

Right.

We're really, really think. So, like how, how difficult is it to start life?

Well, I, I've talked to people, including on the podcast, about this. You know, life requires three things, life as we know it. So there's a difference with life, which who knows what it is, and life as we know it, which we can talk about with some intelligence. So life as we know it requires compartmentalization. You need like a little membrane around your cell metabolism. You take in food and eat it and let that make you do things. And then replication. Okay, so you need to have some information about who you are that you pass down to future generations in the lab. Compartmentalization seems pretty easy. Not hard to make lipid bilayers that come into little cellular walls pretty easily. Metabolism and replication are hard, but replication we're close to. People have made RNA like molecules in the lab that I think the state of the art is they're not able to make one molecule that reproduces itself, but they're able to make two molecules that reproduce each other. So that's okay, that's pretty close. Metabolism is harder, believe it or not, even though it's sort of the most obvious thing, but you want some sort of controlled metabolism. And the actual cellular machinery in our bodies is quite complicated. It's hard to see it just popping into existence all by itself. It probably took a while, but we're making progress. In fact, I don't think we're spending nearly enough money on it. If I were the nsf, I would flood this area with money because it would change our view of the world. If we could actually make life in the lab and understand how it was made originally here on Earth.

And I'm sure it would have some ripple effects that help cure diseases and so on. I mean, just. That's right.

So synthetic biology is a wonderful big frontier where we're making cells right now. The best way to do that is to borrow heavily from existing biology. Right Craig Venter, several years ago, created an artificial cell, but all he did was. Not all he did. It was a tremendous accomplishment, but all he did was take out the DNA from a cell and put in entirely new DNA and let it boot up and go.

What about the leap to creating intelligent life on Earth, however, again, we define intelligence, of course, but let's just even say Homo sapiens, the, The. The modern, the modern intelligence in our human brain. Do you have a sense of what's involved in that leap and how big of a leap that is?

So AI would count in this? Or you really want life? You want, really an organism in some sense?

AI would count, I think. Okay, yeah, of course. Of course.

AI would count, but, well, let's say artificial consciousness, right. So I do not think we are on the threshold of creating artificial consciousness. I think it's possible. I'm not, again, very educated about how close we are, but my impression is not that we're really close because we understand how little we understand of consciousness and what it is. So if we don't have any idea what it is, it's hard to imagine we're on the threshold of making it ourselves. But it's doable, it's possible. I don't see any obstacles in principle. So, yeah, I would hold out some.

Interest in that happening eventually, I think, in general consciousness. I think we'll be just surprised how easy consciousness is once we create intelligence.

Yeah.

I think consciousness is a thing that. That's just something we all fake.

Well, good. No, actually, I like this idea that, in fact, consciousness is way less mysterious than we think because we're all, at every time, at every moment, less conscious than we think we are. Right. We can fool things. And I think that plus the idea that you not only have artificial intelligent systems, but you put them in a body.

Right.

Give them a robot body. That will help the faking a lot.

Yeah, I think. I think creating consciousness in artificial consciousness is as simple as asking a Roomba to say, I'm conscious and refusing to be talked out of it.

Could be. It could be.

And I mean, I'm almost being silly, but that's what we do.

Yeah.

That's what we do with each other. This is the kind of. That consciousness is also a social construct, and a lot of our ideas of intelligence is a social construct. And so reaching that bar involves something that's beyond. That's not necessarily. Doesn't necessarily involve the fundamental understanding of how you go from electrons to neurons to cognition.

No, actually, I think that is an extremely good Point. And in fact, what it suggests is, yeah, you referred to Kate Darling, who I had on the podcast, and who does these experiments with very simple robots. But they look like animals and they can look like they're experiencing pain. And we human beings react very negatively to these little robots looking like they're experiencing pain. And what you want to say is, yeah, but they're just robots. It's not really painful.

Right.

It's just some electrons going around. But then you realize you and I are just electrons going around and that's what pain is also. And so what I would have an easy time imagining is that there is a spectrum between these simple little robots that Kate works with and a human being where there are things that sort of, by some strict definition, Turing test level thing, are not conscious, but nevertheless walk and talk like their conscious. And it could be that the future is. I mean, Siri is close. Right. And so it might be the future has a lot more agents like that. And in fact, rather than someday going, aha, we have consciousness, we'll just creep up on it with more and more accurate reflections of what we expect.

And in the future, maybe the present, for example, we haven't met before, and you're basically assuming that I'm human.

I give it a high probability at.

This time because the. Yeah, but in the future there might be question marks around that, right?

Yeah, no, absolutely. Certainly. Videos are almost to the point where you shouldn't trust them already. Photos you can't trust.

Right.

Videos is easier to trust, but we're getting worse. We're getting better at faking them, right? Yeah. So physical embodied people. What's so hard about faking that? This is very depressing, this conversation we're having right now.

To me, it's exciting.

You're doing it, so it's exciting to you, but it's a sobering thought. We're very bad, right? At imagining what the next 50 years are going to be like when we're in the middle of a phase transition as we are right now.

Yeah. And in general, I'm not blind to all the threats. I am excited by the power of technology to solve, to protect us against the threats as they evolve. I'm not as much as Steven Pinker optimistic about the world, but in everything I've seen, all of the brilliant people in the world that I've met are good people. So the army of the good, in terms of the development of technology is large.

Okay. You're way more optimistic than I am. I think that goodness and badness are Equally distributed among intelligent and unintelligent people. I don't see much of a correlation there.

Interesting. Neither of us have proof.

Yeah, exactly. Again, the opinions are free, right?

Nor definitions of good and evil. We come without definitions or without data. Opinions. So what kind of questions can science not currently answer, may never be able to answer, in your view?

Well, the obvious one is, what is good and bad? What is right and wrong? I think that there are questions that science tells us what happens, what the world is and what it does. It doesn't say what the world should do or what we should do, because we're part of the world, but we are part of the world, and we have the ability to feel like something's right, something's wrong. And to make a very long story very short, I think that the idea of moral philosophy is systematizing our intuitions of what is right and what is wrong. And science might be able to predict ahead of time what we will do, but it won't ever be able to judge whether we should have done it or not.

So, you know, you're kind of unique in terms of scientists. It doesn't have to do with podcasts, but even just reaching out, I think you referred to as sort of doing interdisciplinary science. So you reach out and talk to people that are outside of your discipline, which I always hope that's what science was for. In fact, I was a little disillusioned when I realized that academia is very siloed.

Yeah.

And so the question is, how, at your own level, how do you prepare for these conversations? How do you think about these conversations? How do you open your mind enough to have these conversations? And it may be a little bit broader. How can you advise other scientists to have these kinds of conversations? Not at the podcast. The fact that you're doing a podcast is awesome. Other people get to hear them, but it's also good to have it without mics in general.

That's a good question, but a tough one to answer. I think about, you know, a guy I know is a personal trainer, and he was asked on a podcast, how do we, you know, psych ourselves up to do a workout? How do we make that discipline to go and work out? And he's like, why are you asking me? Like, I can't stop working out? Like, I don't need to psych myself up, so. And likewise, you know, he asked me, like, how do you get to have interdisciplinary conversations and all sorts of different things with all sorts of different people? I'm like, that's what makes Me go right. I couldn't stop doing that. I did that long before any of them were recorded. In fact, a lot of the motivation for starting recording it was making sure I would read all these books that I had purchased, all these books I wanted to read. Not enough time to read them. And now if I have the motivation, because I'm going to interview Pat Churchland, I'm going to finally read her book. Her book, you know, and it's absolutely true that academia is extraordinarily siloed. Right. We don't talk to people. We rarely do. And in fact, when we do, it's punished. You know, like the people who do it successfully generally first became very successful within their little silo discipline, and only then did they start expanding out. If you're a young person, you know, I have graduate students. I try to be very, very candid with them about this, that most graduate students do not become faculty members. It's a tough road. And so live the life you want to live, but do it with your eyes open about what it does to your job chances. And the more broad you are and the less time you spend hyper specializing in your field, the lower your job chances are. That's just an academic reality. It's terrible. I don't like it, but it's a reality. And for some people, that's fine. There's plenty of people who are wonderful scientists who have zero interest in branching out and talking to anyone outside their field. But it is disillusioning to me. Some of the romantic notion I had of the intellectual academic life is belied by the reality of it. The idea that we should reach out beyond our discipline and that is a positive good, is just so rare in universities that it may as well not exist at all.

But that said, even though you're saying you're doing it like the personal trainer because you just can't help it, you're also an inspiration to others. Like I could speak for myself, you know, I also have a career I'm thinking about. Right? Yeah. And without your podcast, I may have not have been doing this at all. Right. So it makes me realize that these kinds of conversations is kind of what science is about in many ways. The reason we write papers, this exchange of ideas, is it's much harder to do interdisciplinary papers, I would say.

Yeah, right.

And conversations are easier. So conversations is a beginning. And in the field of AI that it's obvious that we should think outside of pure computer vision competitions and a particular data sets. We should think about the broader impact of how this can be, reaching out to physics, to psychology, to neuroscience, and having these conversations. So you're an inspiration.

Well, thank you very much.

Never sweet, but never know how the world changes. I mean, the, the fact that this stuff is out there and I've a huge number of people come up to me, grad students, really loving the podcast, inspired by it. And they will probably have that there'll be ripple effects when they become faculty and so on. So we can end on a balance between pessimism and optimism. And Sean, thank you so much for talking today. It was awesome.

No, Lex, thank you very much for this conversation. It was great.