Sean Carroll's Mindscape: Science, Society, Philosophy, Culture, Arts, and Ideas - 75 | Max Tegmark on Reality, Simulation, and the Multiverse

Episode Date: December 2, 2019

We've talked a lot recently about the Many Worlds of quantum mechanics. That's one kind of multiverse that physicists often contemplate. There is also the cosmological multiverse, which we talked about with Brian Greene. Today's guest, Max Tegmark, has thought a great deal about both of those ideas, as well as a more ambitious and speculative one: the Mathematical Multiverse, in which we imagine that every mathematical structure is real, and the universe we perceive is just one such mathematical structure. And there's yet another possibility, that what we experience as "reality" is just a simulation inside computers operated by some advanced civilization. Max has thought about all of these possibilities at a deep level, as his research has ranged from physical cosmology to foundations of quantum mechanics and now to applied artificial intelligence. Strap in and be ready for a wild ride. Max Tegmark received his Ph.D. in physics from the University of California, Berkeley. He is currently professor of physics at the Massachusetts Institute of Technology. He has played an important role analyzing data from large-scale structure and the cosmic microwave background. He is the author of Our Mathematical Universe and Life 2.0: Being Human in the Age of Artificial Intelligence. He is a co-founder of the Foundational Questions Institute and the Future of Life Institute. Web site Google Scholar page Wikipedia Amazon author page Talk on AI Twitter

Your social media feed delivers plenty of advice. But it doesn't know you. It doesn't ask questions. It doesn't give physical exams or order tests. Doctors do. At the American Medical Association, we believe the best care starts with a real conversation with someone who understands the science and your unique health.

So stay curious. Ask questions. But when it's time to make decisions, make them with a doctor. Learn more at AMA Health vs.hype.org. That's AMAHealthVShype.org. Indeed, sponsor jobs gets you quality candidates when you need them most. Spend less time searching and more time actually interviewing candidates who check all your boxes. Less stress, less time, more results.

When you need the right person to cut through the chaos, this is a job for Indeed sponsored jobs. And listeners of this show will get a $75 sponsor job credit to help get your job the premium status it deserves at Indeed.com slash podcast. Terms and conditions apply. Need to hire? This is a job for indeed sponsored jobs. Hello, everyone, and welcome to the Mindscape podcast. I'm your host, Sean Carroll. And today is one of those episodes that brings you a guest who people have been hoping and expecting to get on the podcast for a long time.

That would be Max Tagmark, cosmologist and physicist, professor at MIT. I've known Max for a long time. In addition to being a physicist, he's also one of the co-founders of the Foundational Questions Institute, which we talked a little bit about with its other co-founder, Anthony Aguirre, a few podcasts back, and also a co-founder of the Future of Life Institute, as well as a well-known author and public speaker. Max sort of made his bones, physics-wise, by studying the large-scale structure of the universe and the cosmic microwave background. He's worked with some of the largest and most important experimental projects, as an analyst, as a theorist, taking the data

from these projects and asking what we can learn about the universe. But his interests are very broad. He's worked in the foundations of quantum mechanics, and most recently, he's been interested in artificial intelligence. His most recent book is called Life 3.0, being human in the age of artificial intelligence, and we will talk a little bit about his most recent work in creating artificial intelligence that acts like a physicist. But his previous book was called Our Mathematical Universe, and that put forward a theory of a very, very big multiverse, bigger multiverse than most cosmologists ever think about, one in which all mathematical structures are somehow real. So that'll be the focus of most of today's conversation, working up to from our universe to the

multiverse, why you would think that, what it all means, the big cosmic questions that we'd like to dig our teeth into here at the Mindscape podcast. Wellness looks different at every stage. The right support makes all the difference. Shake up your routine with vital proteins collagen peptides. With 20 grams of collagen sourced from grass-fed, pasture-raised bovine, it helps support Healthy hair, skin, nails, bones, and joints. Made with no artificial sweeteners, it's a clean way to fuel your body.

So your wellness stays effortless wherever the day takes you. Vital Proteins. Stay vital. Visit VitalProtene's.com to get started. These statements have not been evaluated by the Food and Drug Administration. These products are not intended to diagnose, treat, cure, or prevent any disease. When people turn to telehealth or weight loss, they're looking for real support. That's why more people are choosing orderly meds.com. Orderly meds connects you with real doctors and access to proven GLP1 medications like

semaglutide and terseptatide. No guessing, just a more supportive experience, and all ship directly to your door in discrete packaging. Do your research, ask questions, then visit orderlymeds.com slash podcast for an exclusive offer. That's orderlymeds.com slash podcast. Individual results may vary on medical advice, eligibility required, C-Sight for details. Remember, for those of you who don't know, there is a web page, preposterousuniverse.com slash podcast. And there you can get show notes for all the episodes, including complete transcripts. You can also click on a link to go to Patreon and pledge money when every podcast comes out. The people who pledge money on Patreon get episodes without any ads in them, and once a month they get to ask questions that are answered in an Ask Me Anything

episodes. So it's a worthwhile way to spend your $1 per week on the Mindscape podcast. And whether or not your Patreon supporter, I'm always very grateful for all the nice comments, all the reviews, all the ways in which people have told me that they really enjoy the Mindscape. podcast. So let's go. Next tech, Mark, welcome to the Mindscape podcast. Thank you. It's a pleasure to be here. Now, of the many things you're known for, the multiverse is certainly one of them. I just wrote a book on quantum mechanics in many worlds. So one of the very first questions

I get, I'm going to give to you. What is the multiverse? What is the relationship to many worlds? What are all these words really mean? So before we start talking about other universes, let's be clear on what we mean by our universe. We don't mean all of space. When we, in astrophysics say our universe, we actually just mean the spherical part of space from which light has had time to reach us so far during a 13.8 billion years since our big bang. And it's huge. It has 10 to the power, 78 particles in it and huge numbers of galaxies, but certainly not infinite, right? And if you call that our universe or observable universe and space goes on for beyond that which is what's the simplest theory of what traded

our big bang predicts inflation theory then there are other universes just like it i call that level one so but other universe in that sense it means more of the same but further away out of contact exactly there just could be other people living on some galaxy a hundred billion light years away when they look around they don't see us because light from us hasn't reached them and someone might say, oh, that theory you have, that Earth exists with Sean Carroll is just philosophical. Unprovable, yeah. But it's all part of the same space.

And then, if you take seriously this inflation theory, the simplest versions of them actually predict that space isn't just big and much bigger than we can see, but actually infinite. And moreover, that the distribution of stuff in space started out kind of randomly from place to place. which means that no matter how unlikely it is that the particles would have started out in such a way that Sean Carroll and Max Tegmark would come along 13.8 billion years later and have this conversation, the probability isn't zero, right, since it happened. And that means if you roll the dice again an infinite number of times, it will happen again somewhere else. And you can even crudely estimate how far you have to go until you get to the nearest cost. of this universe. It's about a Google Plex meters where Google Plex is one with a Google zeros and a Google is one with 100 zeros after it. Sounds pretty crazy, but actually

much less crazy than what you conclude if you take this inflation theory seriously, because it actually says that there are parts of this space, most likely, which aren't just really far away so we can't get there. Where the laws of physics you would learn in school, would be the same, but what would you learn in history class would be different, you know, because basically the particles started out differently, so things happen differently. But where you might actually learn different things in physics class, too. I call that the level two multiverse, and the way you might get convinced to take that seriously is if you both trust people like Alan, Alan Goose and Andre Linday who say that

this inflation thingy happen, and you also take seriously the idea that the laws of physics might have more than one solution for what a uniform space can look like. We know that's the case for water. It can be solid ice or liquid water or steam. Spring theorists say the space we live in might be like that too, except there might be more than three ways, maybe more than a Google ways it could be. But this is a pretty general phenomena in physics.

If you have some complicated equations, they can have multiple solutions. And inflation is such a violent process that it doesn't just create a lot of space, but it really sort of jostles things around so violently that it's quite possible that it created not just lots of one solution for how space can be, but huge amounts of many different solutions. And we might be tricked into thinking that the solution that we live in is the only kind because it's all we ever see.

But that would be just as dumb a conclusion as if some fish in the Caribbean concludes that all water is in liquid form. because it's all it's seen. If it swam up to Greenland, they would see icebergs and realize that, hey, there's more. So that's the level two multiverse. It's more diverse. Actually, let me dwell on something about the level one multiverse,

because I think it does get glossed over a little bit. Level one multiverse, let's just imagine space is infinite, right? The universe is open. And this is something that cosmologists have contemplated ever since we've been doing the expanding universe, right? And just from that one idea, it more or less follows that there's an infinite copy of people like you. Almost, I would say. You could still have an idea which Newton and Einstein originally flirted with.

Maybe space went on forever, like even Euclid thought 2,000 years ago. But there was only stuff in a small region of it. And then the rest was just this infinite void where nothing was happening. But inflation torpedoes that idea because it's... It says that actually all of the space is created with stuff in it. And it's also not just the same thing tiled over and over again, but it starts out kind of randomly in such a way that all initial conditions get realized somewhere. So I agree with you.

If you put together these two ideas, infinite space, and that you do all the possible initial conditions for them, then you're really forced into this conclusion that everything gets realized, which is quite shocking when you first contemplated. Because long before we, if you travel into space, we get to that other conversation between Sean Carroll and Max Tegmark or Googleplex meters away. But you get to other ones where you're interviewing Max Schmegmark

and Max Pegmark about his golf playing and all sorts of other variations of our lives. And some people really don't like that. And this is a conversation that could have been had in the 1920s, right? this is not like modern super duper crazy cosmology. Like the idea of an open infinite universe could have been taken more seriously. Yeah. And it was pretty violently attacked even before that. You know, anyone who listens to this, if you go to Rome, go to Campo dei Fiore and there's

a statue there of Giordana Bruno, who was burnt at the stake right there, the year 1600, for talking about an infinite universe with other solar systems and so on. For doing a bunch of things including that. There's historians who want to nitpick about exactly the reasons why his tongue was cut out and he was burned at the stake. But that was definitely one of them. Well, I would argue that there are no scientific claims which should get burnt at the stake. Wellness looks different at every stage. The right support makes all the difference.

Shake up your routine with vital proteins collagen peptides. With 20 grams of collagen sourced from grass-fed, pasteur-raised bovine, it helps support healthy hair-skin nails, and joints. Made with no artificial sweeteners, it's a clean way to fuel your body. So your wellness stays effortless wherever the day takes you. Vital Proteins. Stay vital. Visit VitalProtene's.com to get started. These statements have not been evaluated by the Food and Drug Administration. These products are not intended to diagnose, treat, cure, or prevent any disease. When people turn to telehealth for weight loss, they're looking for real support. That's why more people are choosing orderly meds.com. Orderly meds connects you with real doctors and

access to proven GLP1 medications like semaglutide and terseptitide. No guessing, just a more supportive experience, and all shift directly to your door in discrete packaging. Do your research. Ask questions. Then visit orderly meds.com slash podcast for an exclusive offer. That's orderly meds.com slash podcast. Individual results may vary now medical advice, eligibility required seaside for details. Okay, but good. So level one already has some weird philosophical implications in this many copies of me. Then level two is that there's also many different versions of the local laws of physics far away. And just to add to that, I think it's important to remember that the things that we were taught to be laws of physics as fundamental, the more we study them, we realize that a lot of those are actually not fundamental, but just more, that information is more like just part of your address in the sense. Like if you have a T-shirt one day, which has the theory of everything on it, which hopefully you Sean will discover one day, right?

Would you expect to have a big eight on there? And if I ask you, what's the eighth for? You're like, well, eight, that's the number of planets in the solar system. That will be pretty weird, right? Because you'll be like, Max, there are other solar systems out there too, with two planets, five planets. So the eighth isn't telling us anything fundamental about space. It's telling us something about our address in space. We don't think that the average temperature of the air on Earth is a fundamental quantity of nature.

We don't. Environmental. We don't. But we did think. for a long time that the number 1836 was kind of fundamental. That's how many times heavier the proton is than the electron, right? Fundamental. And then a lot of our stream theory friends tell us actually to have like a Google different

solutions. And in one of them, it's 1836. You can probably come up with another one where that number is 2019. So maybe that's also just part of our address. In fact, there are so far 32 pure numbers that we've found, dimensionalist numbers with decimals that seem to describe everything we, in theory, can try to predict about our world. Maybe they're all just part of our address also.

Yep. All right, good. We need a level three multiverse then, clearly. Yeah, so this one comes not from studying the very big, what's out there in our physical space, but from studying very small things, elementary particles they were all made out of. where in 1950s, this Princeton grad student, Hugh Everett, came up with this incredibly radical suggestion. Earlier, people had already realized that electrons, for example, and other subatomic particles seem to be schizophrenic

in the way, in the sense that it can be in many places at once. And people had really bent over backwards to try to explain that away. For example, the Danish physicist Niels Bohr said basically, there are certain, questions you're not allowed to ask about nature. As a Swedish person, I am honor-bound to always make fun of the Danes. And I think, I think Hamlet was on to something when he said there's something rotten in the state of Denmark. But Everett said, no, no, that's BS. What's actually going on is sure these particles can be in several places once, but that means that since we are made of those particles, we can also be in several places at once. And then instead of just freaking out and having a

stiff drink, he continued thinking through what it would feel like if it really were that way. And he realized that if I make a measurement of a particle that's in two places at once, and in advance I've decided that if it's here, I'm going to go for a drink, and if it's in the other place, then I'm going to watch Netflix instead. What that's going to feel like is there are going to be two different versions of max now, and one has a drink, one watches Netflix, and they're going to be. unaware of each other. And it's going to feel to them that there was just something random that happened. They discovered the particle was this way or that way, and then life went on. And we actually

have known for many years that seem to be kind of random in quantum physics. And he said, that is the explanation. Everett said, randomness is just the way it feels subjectively whenever you get cloned, so to speak. And people still argue fiercely about this, as you know, over a hundred years after quantum mechanics. But I think it's important to realize that this weirdness is so fundamental. It has nothing to necessarily even with quantum mechanics. Because if we just go across the river here to the MGH hospital and they have a cloning device purely classically, where they just sedate you and then they make an actual copy.

They put all these atoms together in exactly the same pattern. And then you're told in advance that the next day, One of you is going to wake up in room one and the other cop is going to wake up in room two. The next morning, both of the Shans wake up and they go out of the hospital room and look at the number on the door, right? What do you predict that you're going to experience? Do you predict that you're going to experience a one or a two? What would you say? I would say it's 50-50.

I have an argument for this. It's going to feel random to you, though, right? Yes, exactly. And there's been nothing you can do to predict your subjective future experience. Although the doctors would claim there was nothing random that actually happened. All they did was cloned you. I think personally that everyone was on to something here. It's quite likely that yet again,

our reality is really bigger than we thought, and that there is nothing truly random actually happening in nature and the things that seem random come from us having been cloned in this sense. Ironically, even if this probably sounds even weirder than the level one and level two, when you look at the kind of diversity you get in this, which I call the level three multiverse, it's actually smaller or no bigger than the other ones. It's just to get all the same parallel world sort of all over again in a different way.

You get much faster in some sense. I mean, it's happening all the time in this room, right? The universe is just branching off the pieces. Yeah. And in some way you're also more connected to some of those parallel branches because if you are upset that you got a parking ticket, you might realize that there's another you that actually was identical to you. It was up until maybe 10 minutes ago and quite recently only branched out. You can take more solace maybe.

Does that make you feel better? Really? I think it takes the pressure off to get things right all the time. It doesn't make me feel any different, for better or for worse. But, you know. It does. I reflected a lot on this before my first son was born because I was hoping things were going to go well. And then I started thinking, I have to align my emotions with my scientific views. Because what do I mean when I say, I'm hoping?

Yeah. If there are parallel universes where things are going to go great and things where they're going to end in tragedy, right? So then maybe I thought, maybe what I'm really hoping is I'm going to be in the parallel universe where things go well with this birth. But then I realized, no, that's stupid because it's going to be in all of them. Yeah, exactly. There'll be a version of you and all of them. Maybe I mean that I hope the probability of, I hope that in most universes it's going to go well. But that's also BS because if you're really smart, that's just the number you could calculate

from quantum mechanics, what fraction of the world goes well. So how can I hope for something, for some fraction being large? Just like saying, I hope that pie is bigger than three or something. Either it is or it isn't. But this is just a free will question. These would all be questions in a deterministic Newtonian universe, right, where you just don't know what the future holds. Is there still the possibility of hope?

So I think in the end I ended up kind of where you are. It feels pretty cool to be in this universe. And sometimes it's useful to have the heuristics like hope, even if you know you're kind of on thin ice when you say things like that. Hope comes from imperfect information. Yeah. Okay, so that's level three. That's the Everettian multiple worlds of quantum mechanics.

and that's separate from the level two cosmological multiverse where things are different. And the typical speculative theoretical physicist stops at three, but you will not be stopped. You're going to go on. You're going to say that there's a level four. That's right. So, of course, most people didn't even go to level one. Geronabu got burned and not just on the job market, literally. that's become more normalized since now that we can see so far with our telescopes and since inflation

has caught on even level two has gotten a bit more normalized moved from the ridiculed to taken seriously albeit albeit controversial even level three it's interesting to see how it's gone from being first utterly ignored to being violently attacked being actually regularly voted as one of the top interpretations if you talk to people who study quantum foundations. So it's sort of respectably controversial in the sense that people

argue about it at conferences rather than just in bars. The level four on the other hand... Sorry, before we get... Now you've raised the whole other issue that I've got to get right. So I personally would put a greater probability on level three really existing than on level two. I think that there's better reason to believe in the Everettian multiverse than there is in the cosmological multiverse. And maybe very roughly 95% for Everett and 50% for the cosmological multiverse.

Do you have different credences yourself? At the risk of being boring, I think I would agree with you on this because there are fewer assumptions behind the Avarian one. Really, all you need is to assume minimalism. It's just the Schrodinger equation that have there in my office. with no little footnote with ifs and buts and this is not valid when someone observes which we're not going to define for you. It's just the equation, boom. Whereas to get the level two multiverse, you really need to assume both inflation and some fundamental theory that has multiple solutions like string or loop quantum gravity.

And neither of those two are things that we should be anywhere near, you know, 90% sure of at this point. Okay, good. Just checking. All right. So we can get on a little board. Yeah, so level four, this is an extremely unpopular kind of multivorice, which... This is your chance to convince me. There's very few people who take it seriously.

You have 100,000 listeners ready to be convinced. Yeah, so the idea here is that... So the way I got into this was, first of all, by just being fascinated by the effectiveness of mathematics in describing physical world so well. and many people in the physics community, especially theorists like you, Sean, will say, our universe is really accurately described by mathematics,

approximated at least, or maybe it's even exactly described. But then you have to ask, what is the difference really between being perfectly described by math and just being math? After all, mathematicians, they also have as language of mathematics

that they use to describe mathematical objects like the cube, like the sphere, like some colabi-yao-manifold, whatnot. And what modern mathematicians have

increasingly done is try to strip out everything but just this brute formalism where they just describe different kinds of mathematical objects. You can make a type of mathematical family tree, which I had fun doing in my book, all these different mathematical structures here.

are the integers, here are the real numbers, the complex numbers, here is quaternions, here is this three-plus one-dimensional pseudo-rimanian manifold. You can go on and put in things with all sorts of weird names, but they're all fundamentally the same kind of thing. Things that can be described purely abstractly with mathematical symbols and no human sort of baggage. And if you look at math that way, at math space in that way, there's just very disturbing question, which is, why would, in Stephen Hawking's words, Stephen asked, why would God breathe fire into one of these equations and make a universe for them to describe, and not others? Or, like, what breaks the symmetry? Why is this particular one that's supposed to describe our world

different from all the others? So the level four multiverse comes from this very, very, radical conjecture that actually there is none of that metaphysics. There is no special fire breathing needed or whatever. All of these mathematical objects exist in exactly the same sort of way. And we just happen to be living in one of them. One of them that's complex enough that parts of it can feel subjectively aware, what they're going on, what's going on there and sort of studying physics. Or to just break it down to plain English. What it's really saying is that we live, live in a mathematical structure, a mathematical object, which is something which has only mathematical properties. And here, I mean, you have to stop and say, well, that sounds completely nutty, right?

What do I mean? Do you see the moose here? We have a moose on the table. Yeah, I was given a moose doll. It's a gift today. Not a true moose. What do I mean this has only mathematical properties? Look at it's made of cloth that has colorful garb. It's squibish. It's squibed. It's A little bit blue and moose-like. Those properties, I just describe, don't sound very mathematical at all, right? But when we look at it, Sean, through our physics eyes, we see that it's actually a blob of quirks and electrons. And what properties does an electron actually have?

It has the property minus one, one-half, one, and so on. And we physicists have made up these nerdy names for these properties. properties like electric charge, spin, lepton number. But it's just to be humans who invented that language, calling them that. They are really just numbers. And you know as well as I do that the only difference between an electron and a top quark is what numbers its properties are. We have not discovered any other properties that they actually have.

So that's the stuff in space. All the different particles in the standard model, you've written so much my stuff about in your books are all described by just by sets of numbers what about the space that they're in like what properties the space have i think i actually have your old nerdy non-popular great my unpopular book yes space has for example the property three that's a number and we have a nerdy name for that too we call it the dimensionality of space it's the maximum number of fingers i can put in space that are all perpendicular to each other but the name dimensionality is just the human language thing. The property is three. We also discovered that it has some other properties, like curvature and topology that Einstein was interested in.

But those are all mathematical properties too. And as far as we know today in physics, we have never discovered any properties of either space or the stuff in space yet that are actually non-mathematical. And then it starts to feel a little bit less. insane that maybe we are living in a mathematical object. It's not so different from if you were a character living in a video game and you started to analyze how your world worked. You would secretly be discovering just the mathematical workings of the code, right?

Yeah, that's an intelligently designed environment, but yes. It is. Yep. I mean, I guess we can talk about probability that we are living in a simulation also later. But if that is in fact the case that we're living in a mathematical object, then What about the other mathematical objects that happen to also be well-defined? Why should we get on our high horses and say they can't exist as much as ours can? Maybe they're mostly still born because you're not complicated enough or whatnot. But you could say the same thing even about those other less controversial kinds of universe. Anyway, so that's the level four multiverse in a nutshell.

It's simply the math space, space of all mathematical. objects that exist. And if this is true, the first prediction it makes is that all the properties of our universe, even the ones we haven't yet been able to understand to do with intelligence and consciousness, for example, should ultimately also succumb to mathematical description. But just to be clear, maybe it sneaked under the radar there, we can imagine the space of all mathematical structures, but you're making the extra statement that they're all equally real and we just live in one of them. Is that fair?

That's right. That's the level four multi- That's right. I really want to strip out as much human baggage as possible from this and saying that something is real or not. What does it even mean?

I think it means something. Well, it means something if you have this sort of very pagan kind of idea that this moose is real because it's made of stuff that's real, right? But then we come and look as

physicists again and just made of all these stuff is just these elementary particles and what are their properties they're all mathematical so you can what is there here really beyond all these mathematical properties i've yet to be convinced i have a lot of usually when people start trying to persuade me of things beyond that it starts to feel very much like some sort of new agey thing that they just want there to be something fluffy and fuzzy and non-mathematical because you don't think mathematics is cool enough or they say things like,

oh, I can't believe it's just mathematics. I object to that derogatory use of the word just. Sure. I will definitely go along with that, but I do want to distinguish, I think maybe I should face it as a question, can we sensibly distinguish between the idea that our universe is just a mathematical structure

and that all mathematical structures are equally real, no matter what you might mean by the word real, but at least there's no difference in the level of reality between ours and others. Otherwise, it's not really a multiverse. You're right. There are two separate questions there. First of all, does our universe have some non-mathematical properties or not? We don't know yet. If you had gone back 400 years and had Galileo on your podcast, then it would have seemed like, duh, of course it has non-mathematical properties. Because the only stuff that we could even predict at all with physics was motion. Galea could throw a hazelnut and a grape and he could tell you that they're going to move in

this parabola shape, Y-C-squared. But he couldn't tell you why the grape was green and the hazelnut was brown because that seemed non-mathematical. He couldn't predict why the grape was soft and the hazelnut was hard because that seemed non-mathematical. It was to do with the mysterious nature of stuff. But then we got Maxwell's equation. But then we got Maxwell's equation. which helped us understand light and colors. And even that part of reality succumbed to the mathematical description. And then we got quantum mechanics, which explains now why the grape is soft and the hazelnut is hard.

And we've gone from having a situation where the whole world seemed just mostly magical and the only thing we could predict about it was motion mathematically to a situation where we can actually apply math to almost everything except maybe intelligent. and consciousness. They're the final holdouts. Well, I did have a podcast interview with Philip Goff, who is a pan-psychist, and the name of his book was literally Galileo's error, because he says that Galileo made an error in making things too mathematical. Now, we'll plug your book, the mathematical universe, which says the opposite. So people can buy both books and decide which one is more persuasive. I changed the title in the last second from the mathematical universe

to our mathematical universe, because I thought saying that would be very arrogant. You say, the solar system. It sort of implies that we're the only one, right? That's true. Whereas if you think seriously that there could be other universes, this is ours. Free is great, but only if it's useful. Free credit scores from some apps can differ by as much as 100 points from your actual FICO score that 90% of top lenders use when you apply for a credit card, personal loan, car loan, or mortgage. That can meet a higher interest rate, a bigger monthly payment, or worse. Denied. My FICO gives you your actual FICO score, the score lenders use, straight from the company

that created it. For the moments that matter, get the score that matters, your FICO score. Visit myfico.com and get started for free today. Wellness looks different at every stage. The right support makes all the difference. Shake up your routine with vital proteins collagen peptides. With 20 grams of collagen sourced from grass-fed, pasture-raised bovine, it helps support healthy hair, skin, nails, bones, and joints. Made with no artificial sweeteners, it's a clean way to fuel your body. So your wellness stays effortless wherever the day takes you. Vital proteins. Stay vital. Visit VitalProtene's.com to get started. These statements have not been evaluated by the Food and Drug Administration.

These products are not intended to diagnose, treat, cure, or prevent any disease. Okay, but wait, but we got distracted because you were making sales pitch for something I already agree with, namely that our universe is mathematical. But what about the others? Hmm, yes. The reality level of the other. How do you make this jump from saying there is, so there's one mathematical structure which happens to be existing also physically?

That's this one. how do you then get to the much stronger conclusion that all the other mathematical ones also have a physical existence? So the first part, I call it the mathematical universe hypothesis, which is that our external reality out there, which I believe exists and continue to exist even if we all died, that it is purely mathematical. And I call out the external reality hypothesis that exists. and then I have an argument for why it should be mathematical. But then there's a separate logical step, which you raise here. How do you get from that to the idea that actually all the other ones also exist? I probably shouldn't bore you with getting into that too much in this podcast.

But what I argue in the book is you basically just have to add a whole, you can take some mathematical structure and just study it very carefully. What are all the properties that I had? right? And you can say this here is this thing. It's going to be self-aware. It has information processing like this. And if you're really smart, you can figure out how AI works and consciousness or whatever. You could figure out even how the parts, some cells where parts of this will subjectively experience themselves in there.

And you might find one that looks just like ours. You might find some other ones. And then you come to this question. Now, if you're going to claim that that this one, here that corresponds to our universe exists physically also. And some other one doesn't. What do you actually mean by that? You're positing that there is some additional property that this has, which has nothing to do

with the math, which is whether it exists physically or not somehow. But this is nowhere to be seen in the math. It's a little bit like if you try to convince somebody that they have a soul and they ask you, okay, does my soul weigh anything? No. Can I measure its existence with these particle detectors? And you say no. And then they ask you, well, does the soul actually push my particles around and make me act differently somehow? And you're like, no. The soul has no effect at all on your particles, but it still exists. Then wouldn't you say that you're making yourself a little bit vulnerable to Occam's razor? They could just say, I'm going to take everything you say, Sean, and just shave off the

soul part because it doesn't have any additional explanatory value. I feel that, so just as someone might say that the soul there is like a little fairy dust that's been sprinkled on, but it doesn't really add anything explanatory. You could say the same thing about this idea that there's this extra magical property that a mathematical structure has called physical existence, but it doesn't have any effect on the mathematical relations that hold within it at all. and I can ask you, well, does that say anything about whether these two lions that are parallel are going to cross or not? And you say, no.

Does it have any effect on whether this self-aware object here is going to experience X or Y? No. So it has no effect on any other relations in there. That to me also is just as vulnerable to Occam's razor now, because it sounds like just philosophical BS really, that has no consequences on any of the mathematical relations. And I would just assume my guess is that it is just philosophical hogwash, and we should just accept the fact that regardless of whether something is a computer simulation or exist in some other way, you know, all that matters about it is,

is how the, what the relations are within it. That's what determines how it feels like to live in it. So there you have it basically, a mathematical universe. And if you, this leads very naturally into this question about whether we're living in a computer simulation also. Because if you and I were arguing about whether we're in a simulation right now, right? And I'm like, no, there's no way because this moose here, it's real. Feels real, right. It's physically real.

And if we were in a computer simulation, the moose would not feel real. How would you respond to my argument? Yeah, that's a dopey argument, right. I'm claiming that my argument is basically the same argument, that the way I experience the Moose, Max, the computer game character here, has only to do with what the mathematical properties are of this computer simulation.

It has nothing to do with whether it's being simulated on a Mac or a Windows machine or a Unix machine. It's just internal mathematical properties of this simulation that matter. I would say it's the same with the mathematical structure. All that matters are the relations in it, not whether some philosopher says that it's sort of made of real stuff or not. So I certainly, forget whether I buy it or not, that's irrelevant, but I understand the argument that it is absolutely a simpler idea

to imagine that all mathematical structures exist than to imagine that, we can conceive of all mathematical structures and we are one of them. There's a lot of information that goes into pinpointing which one we are, right? Which you're happy to get rid of. But then I think, and at the risk of getting a little bit technical here, it would seem that to make sense of this idea, you need some kind of way of comparing the different kinds of universes, the different kinds of mathematical structures.

There are so many that look like this, so many that look like that. And I would also think that once you had that way of comparing, there's far more universes that kind of look like ours and then explode into chaos a second later, right, than to keep up this very, very specific obeisance to the laws of physics. So doesn't your theory make a prediction that we're all going to cease to exist within a nanosecond? As it both my brain would. That's a very good question. I think the answer is no for reasons I'll explain in a second. a sec but first of all I'm totally open to the possibility that this idea is wrong and I'm very interested in testing it in various ways I think to me this idea is more of an inspiration for

a research program to try to find more mathematical regularities because we're never going to find any if we don't look for them right if we say consciousness is off limits because it's not mathematical we're not going to discover any equations that might actually help explain it, say. In terms of your question about whether the most generic thing we should expect is just random weird stuff taking place, I have a suspicion that just like in cosmology, when we study infinite spaces and so on, we stress out a lot about the so-called measure problem to which things exist. are more often or to a greater extent than others, that there is something very similar to that unsolved problem here. And I think it has going to have a lot to do with complexity. In some sense,

maybe very simple mathematical structures that are simple to describe, get more weight. If you were just randomly bored and simulated different universes on your future supercomputer, the simpler ones would get a lot more attention than the more complicated ones because they could be evaluated much faster. And I sometimes wonder this is a very quarter-baked idea if the reason why our universe seems so remarkably simple when you look at fundamental laws might have to do with that. That simplicity is very much favored. And if that's true, then we shouldn't just expect the past has been simple, but we should also

expected the future, which should be kind of simple, because it's much more complicated to write down to define a simulation algorithm or a mathematical structure, where suddenly a pink giraffe appears randomly in my office, right, than to just specify the standard model that we have today, where that's probably not going to happen. I'm worried that you're, I mean, sure, as far as that goes, yes, if there were an additional principle that would somehow favor simplicity, you might be able to account for our universe. It's simple enough. It's a little complicated, but simple enough you might hope for that. But I do worry that that's another metaphysical principle that you were just trying to avoid mere moments ago. You've gone away from the pure, pristine beauty of

everything mathematical exists. That's all there is to it. So first of all, I try to be very humble. So I'm not saying I believe that any other ways of the universe exists. My job as a scientist isn't to believe stuff. I'm happy to make bets, though. One has credences. One's a good basis. Yeah. And second, whether you feel that something is frivolous and full of extra assumptions

or complicated or not, depends a lot on what you mean by simple. Many people criticize any kind of parallel universe as being wasteful because they waste a lot of atoms. Like, I don't buy this idea that space is. is infinite or bigger than we can see because, oh, my gosh, there's all these extra atoms. A different perspective is that simplicity is you should be measured by how simple the mathematical equations are. It turns out it's much easier to write down some physics equations actually give us this really

big space than to write inflation that makes a really small space, which is exactly as big as we can see and then magically stops. And it's the same thing we've seen in quantum mechanics where it turns out to be much easier to have the shreddinger equation that just has all these parallel worlds than to add in a bunch of extra stuff that's supposed to get rid of all the parallel worlds. And I feel in the same sense that the level four multiverse is actually simpler. We were joking by the t-shirt earlier when you write down the theory of everything, right? So if you did that 2,000 years ago, oh boy, would you have to have a small font? You have to say, well, you know, we live in a solar system

as a blah blah blah and then this is what the garlic looks like this is what a watermelon and this copper and periodic table well if you just fast forward a little bit they could wouldn't have to specify the whole periodic table they could just write down this equation for QCD the quantum chromodynamics and three numbers and from them you can compute in principle the whole periodic table and books full of numbers about how atoms shine light and different colors and stuff so more simplicity now and I think

by going to the level 2 multiverse you don't even have to write down numbers like that proton electron mass ratio of 1836 because that doesn't have to go on the t-shirt that now just got it's your address you find out

if you want to know where you're living you go measure that and now you know what part of the level 2 multiverse you're in the level 4 multiverse economize even more because without it, you have to put on your t-shirt, at least the equations for string theory or loop quantum gravity or whatever the correct quantum gravity theory is. You have to put that.

In the level four multiverse, you can just wear a black t-shirt from nothing on it because all of the mathematical structures exist, which you can try to catalog with a future computer program, exists, and that's kind of all there's to it. I mean, I don't want to dwell on this too much, but it sounds a lot to me like Borges. library, right? Like Borges said, what if there's this library that literally contains books

with every possible sequence of symbols. And somewhere in that library are all the works of Shakespeare and the Bible, and for that matter, every book ever written in the future, right? The problem is finding them. You know, the problem is that the card catalog is the same size at the library. Most of those books are utter nonsense. Yeah, and there's another problem with it. There's really no structure in it. There's no sense in which simple books are more important than complicated books. So the average generic book is just pure garbage. What's so different about mathematical structures is it's actually very hard to write down a system of axioms and theorem-proving rules that are even consistent at all. So the great mathematician David Hilbert once said that

mathematical existence is simply freedom from contradiction. And it turns out to be just really hard to do that. So there's, you know, some people like to say that you know something exists, if you kick it and it kicks back. Mathematical structures are a lot like that. If you, if you, the Plato, for example, was really interested in regular 3D shapes. And he discovered that there are five of them. They call it the platonic solids,

the cube, mycosahedron, et cetera, right? And no matter how hard he tried, he couldn't just make up a sixth one. There just isn't one. He could make up names for them. But couldn't make up more. In that sense, mathematics kind of kicks back. They discovered the dodecahedron made of 12 pentagons

and thought that was so cool that they didn't tell anyone. They thought it had magic powers, and you see it sometimes in art museums in the corner of some paintings. Discovering that shape was a little bit like when we discovered the planet Neptune. It felt like it was out there all along, and now, ah, we found it. So the math space is much more interesting than Borges' library in the sense that there's a lot of stuff in it. But it's absolutely not the case of just everything exists. It's not like some people sometimes criticize social sciences for being very loose and loosey-goosey about things.

If you just take some random mathematical structure and just add one more axiom to it often like pops the balloon and the whole thing. just collapses to nothing. This June, the world comes to Los Angeles. Kick off FIFA World Cup 2026 at the FIFA Fan Festival at the iconic Los Angeles Memorial Coliseum. Watch matches live on Giant Scream. Feel every goal with thousands of fans. And celebrate with music, culture, and flavors from around the world. Join us June 11th through 14th opening weekend as the tournament kicks off in Los Angeles.

Tickets are just $10 and kids under 12 were free. Get yours now at Los Angeles FWC. Your first sip should do more than simply start the day. Elevate your scoop with vital proteins collagen peptides advanced. Featuring collagen plus hyluronic acid and vitamin C. It helps support healthy hair, skin, nails, bones, and joints. Adding extra wellness support to your daily routine. So your wellness stays effortless wherever the day takes you.

Vital Proteins. Stay vital. Visit VitalProtines.com to get started. These statements have not been evaluated by the Food and Drug Administration. these products are not intended to diagnose, treat, cure, or prevent any disease. So, would it be, you're leaning on simplicity quite a bit. You've also alluded to the idea that everything that we do see in our physical universe is just running on some computer somewhere in super advanced civilization. Is that simpler or more complicated? I would actually bet against the fact that we're living in a simulation, not because I think it's impossible, but because I think it's less likely.

So Nick Bostrom, or Bostrom, as we call him in Swedish, popularized this so-called simulation argument. What he basically said was, we're building ever better computers when they will be able to have much more simulated minds than real minds, so therefore we should expect ourselves to be a simulated mine. I claim that that argument is actually flawed in the last step. And you can see it sort of start falling apart already by, just taking it one step further.

Suppose you agree with it. So we're assimilated. There is some sort of basement universe with actual physical stuff, but we're not in there. We're in one of their computer simulations. Nothing prevents us from making the same argument again. In our simulated world, we're going to have more simulated computers, and they're going to do doubly simulated minds, and they'll outnumber us, so therefore we're doubly simulated. And now we can do it again.

We're triply simulated, and now you're simulated a trillion times. are you getting a sinking feeling at this point? Well, I mean, go ahead. Finish the argument. Is it just that we should be infinitely simulated? Is that the argument? That's what the simulation argument logically implies. They were basically infinitely simulated,

which already feels a little bit fishy. I had the flip side of that argument, which is that according to that logic, the large majority of beings should be in the lowest resolution simulations. right? Because they're just the easiest to make and they would win. But empirically, we don't see, I mean, it's easy for us to make simulations. And so we could easily have lower resolution than what we see around us. Therefore, there's some logical flaw. I like that too. There was even an interesting paper a while back where they had tried to look for evidence that were simulated by finding weird noise and things. They didn't find any, by the way, for readers out there. Yeah. I think where the argument goes wrong is very interesting. What it's saying is you should have to be.

this basement reality, it has some laws of physics in it, where there's real stuff. And in that basement reality, if you can there make the argument that most minds in there are going to be simulated, sure, now you aren't onto something. Maybe, but we don't know they were in the basement reality, right? If we conclude in the end they were simulated, then we have immediately violated that assumption. And we have no idea in that case what basement reality were actually simulated in. The thing that would matter, the only thing that could make this argument work is if you knew for sure that you were in the basement reality so you can make some statements about what's actually probable. Yeah. But just to be very clear, you're totally on board with the idea that maybe not everyone listening is on board.

I'm on your side here. Life in a simulation would be just as real as life in the basement reality. Right. I mean, you could simulate things. There's no substrate dependence. The people in the simulation could, in principle, have all the hopes and dreams and consciousness that you and I have. I want to be humble here. I mean, there are many people who think that there is more, there's something mysterious about intelligence and think that it can only exist in biological organisms such as us that exists sort of in the basement reality.

But I am not very fond of that viewpoint. I call it carbon chauvinism. the idea that you can only be smart if you're made of carbon atoms. I think it's in fact exactly the opposite of carbon chauvinism is giving us the whole AI revolution. The opposite of carbon chauvinism is the idea that intelligence and consciousness are all about information processing. And it just doesn't matter whether the information is processed by carbon atoms in neurons and brains or by silicon atoms in today's technology or some other kind of particles. and tomorrow's technology. If you accept that, that it's all about information processing,

then of course, an entity in a really good simulation with the right kind of information processing is going to feel the same way they would feel if that information processing happened on some other substrate. And just like they will have no idea of knowing whether they're simulated on a Mac or a Windows machine or a Linux machine, they will have no idea of way to tell whether they're actually simulated at all or living in an actual physical universe that has those laws.

And so would you say, so not to put words in your mouth, tell me if this is accurate, it's possible, it's conceivable that we are living in somebody else's simulation. So among people who buy that, there's their response that, but I don't care, and there's a response that, but I think it's unlikely for the following law. reason. It sounds like you're saying you think it's unlikely, but you would be interested in caring if you thought that it was likely. Oh, totally. I agree with all you said. I think we might

be. I have not seen a compelling argument for why we are, and I think there's a little bit of experimental evidence against, since it seems like the quality of this one is a bit higher than it would need to be. Although sometimes when I pay attention to politics, I take that back to the quality. Maybe isn't that high after all. But that aside, if you still have a lingering doubt that you are simulated, I mean, the advice is pretty clear. Just live a really interesting life. So those running the simulation don't get bored and shut you down.

How do we know what's interesting to them? This is the question. Well, so let's flip it around. But you seem to be doing it right since you haven't been turned off yet. As far as I know. I mean, the thing about being in a simulation is they can always hit a save point and shut down the simulation and start you up. years later, who knows how many times that's happened in our lifetime, right?

Indeed. The clock in our universe need not be the clock in their universe. This is where it gets scary. But let's turn it around. If the simulation theory is true, then there are super beings in the basement reality who have built a computer in simulating us. And then the intelligent design that you despise so much is true. That's right, exactly.

But then what about the other way around? What about us building simulations with AIs in them? I had Melanie Mitchell in the podcast recently, a bit of a skeptic. about AI's ability to reach human levels of intelligence, at least near term, not in principle, but how close we are. What is your feeling to how close we are? Because this is kind of what you're doing these days, right? You're doing more AI than cosmology. Yeah, my research group here at MIT, we've been doing AI research for the last bunch of years. And so if you look at minority papers, they're published in the AI journals and the AI conferences. And by the way, is that more

a response to the state of cosmology or the state of AI or the state of max? Oh, that's a good question. I think it's a little bit of both, the first two, actually. Ever since I was a teenager, been just really fascinated by the biggest questions. The bigger, the better. And I remember thinking back then that the biggest mysteries of all were our universe out there and our universe in there in your head, right? So I spent the first 25 years of my career around the outer universe, super exciting. And in recent years, I've just felt ever more intrigued by the,

the second part. And I realize that not only that this pace is picking up so much that it's really fun now, but also that we physicists actually have a lot to contribute to that field in ways we can come back to. But to get me back on track, you asked a question before you asked my personal. Sorry, what is your statement? What is your feeling about the state of the progress? What's more interesting than what I think is what AI researchers have said, across the world in recent surveys, which is that most of them guess AI will succeed within

a matter of decades to get AGI, which is an acronym for artificial general intelligence, which just means AI that can do all intellectual tasks, at least as well as us. Superintelligence is something way beyond that in turn. Are we going to, is that actually going to happen? There's, of course, a lot of people who dismiss it and say, forget about it. it's not going to happen ever because it's impossible, because intelligence is something mysterious somehow and all those people who say it's about information or working up the wrong tree. Maybe they're right. I would guess that they're wrong.

And just as in any field, you have a genuine scientific controversy about how fast or slow things are going to be. But I think it would be a mistake to dismiss it being possible, that it's possible. Sure, we have many embarrassing examples of scientists being over-optimistic of how quickly things are going to happen. Remember when you and I were both starting grad school, Fusion was 40 years away. Oh, yeah. It's still 40 years away. But on the other hand, we thought we'd never know the density of the universe. Yeah, and now we know.

Hard to predict the future. 1.6, 10.7 square times 10 times 10 times 10 times 10,000, 26 kilograms square meter. and we've gone for about the age of our universe from arguing whether it's 10 or 20 billion years to arguing about whether it's 13.7 or 13.8. So that's progress. And so it's very dangerous to say

something is impossible when you're on a planet with a lot of smart, motivated people. For example, Ernest Rutherford, one of your physics heroes, grandfather of nuclear physics, famously said that this idea of getting nuclear energy, moonshine. Do you know how long it took

from when he said that in a lecture in London until Leo Zelard invented the neutron chain reaction. Couldn't have been that long, a decade? Less than 24 hours. Oh, okay. So we have to be, the bottom line I think is we just have to have an open mind. It could happen in decades, could be a bit sooner, could be much, take much longer.

But I think if it's a few decades away, that already makes it arguably even more urgent than thinking about climate change in your retirement because this is going to be a much bigger deal, obviously, if it happens. If some human you don't like very much, think about your least favorite politician or unscrupulous business tycoon has this and can use it to take control over Earth and impose what they think is cool on you, you know, it might not be so stoked about that, right? Alternatively. So, sorry, just to, I don't want that to go by too quickly for the people out there.

I mean, this is the worry of artificial general intelligence that it will give either the AI itself or the people controlling that AI some version of superpowers that will be hard to control for the rest of us and therefore we should worry a lot about AI gone wrong. That's right. And I think the reason I decided to start my AI book, like 3.0, with a little fiction story,

was exactly because of this. Hollywood gets worried always about the wrong things. It's always robots because they have to have something they can film because it's a visual medium. Intelligence itself can confer great power. Why is it that humans have more power on this planet than tigers? Is it because we have sharper claws or bigger biceps? No, it's because we're smarter, right? And that means, of course, that if someone can amplify their intelligence with vastly stronger AI,

it can confer enormous power on them. They can pretty quickly outsmart others on the stock market and out patent them and out argue them on your blog and whatever and have a lot of influence. And is that good or bad? Well, that entirely depends on how it's controlled and with what goals. I think like any technology, of course, AI will be a double-edged sword.

People sometimes ask me if I'm against AI or for it. I usually counter by asking them if they're, about fire, like whether they against it or for it. Like, duh. It's dumb question. The interesting question is, what can you do to get the upside of fire, but not the downside? Turns out there's a lot you can do. That's why we have fire extinguishers and fire escapes and we teach our kids to not, and blah, blah, blah, blah, you know. That to me is the most interesting question, important question of our time, I think. How are we going to harness this technology to make sure it gets used to cure cancer and lift people out of poverty and help life flourish like never before,

hopefully not just during the next election cycle, but for billions of years and maybe not just on this planet, but throughout much of our universe. It could be awesome. On the other hand, you don't have to think too long to envision scenarios where it's not so great. For starters, if you just envision your least favorite politician using it to impose their will on everybody. you don't even have to get the things

where the AI itself sort of breaks out and takes over to a lot of people you probably wouldn't want to control you So I had the following worry which I haven't really put in front of AI experts such as yourself yet If it's going to be AI at all

It is going to need the ability to learn and therefore to change itself So isn't there an intrinsic issue with the idea of controlling it Like you can tell it not to do something but if it decides to change its mind, what are we going to do about it? This is a very active research in AI today,

how to actually take deep learning systems, which have learned stuff and where you don't really understand in detail how they work. And seeing, these are some way nonetheless, by having better architectures, you actually prove that they're always going to do certain things. It breaks into three different nerd problems,

which are all unsolved, but which we all also face. as parents, I would say. We want the AI first to understand our goals and then adopt our goals and then retain them. You know, our little babies first are too dumb to understand most of our goals. And then they're teenagers and don't want to adopt our goals. But fortunately, our kids spend several years of this in-between period where they're smart enough to understand our goals and hopefully malleable enough that we have a little chance at persuading them.

They're weak enough that we can control them. Yeah. Yeah. And that's what education, child, that's what parenting is, is, is smart enough. fundamentally about. But even if we succeed in getting AI to understand our goals and adopt them,

how can we make sure if they keep getting smarter that they retain them? Like Philip and Alexander, you can see them on those pictures there when they were still small enough that they thought Lego was the coolest thing. That was a big priority

for them. Now all those Legos are gathering dust in the basement. because they're 18 and 20. If we create powerful AI that has this goal to really take care and protect humanity, we don't want it to become as bored with that as they got with Legos, right? So you would also like to have some sort of recursive algorithm where you're guaranteed that there are certain goals that's always going to retain. And the interesting news here is that there's been a lot of technical progress, actually.

we put a big effort in five years ago with this Future Life Institute to bring together AI researchers and others at conferences to talk about these technical questions and with the goal of mainstreaming that research or more about what happened and it's been very exciting to see actually how much is taking place.

Stuart Russell, for example, famous AI professor who just has a new book out called Human Compatible where he actually doesn't just once again lay out these challenges but actually proposes what he considers a cool solution And the good news is we still have time. Maybe we have a few decades to get this right. My message is on one hand, you don't freak out and panic.

But on the other hand, work hard on these technical questions now so we can get the solutions when we need them. Because it might take decades, right? Rather than start working on them the night before someone's switching. which is on an AGI. Well, okay, we're running low on time, but I do want to give, this leads very well into you giving a little sales pitch for the work that you're doing now that I heard you talk about at the most recent foundational questions institute, where you're training AIs to be theoretical physicists, right, to find the laws of nature by looking at big data sets, not just find patterns, but really find quantitative laws. Tell us what that project is and how it's going. So we call it my grad students and I the intelligible intelligence project. So as you know, most of the progress in AI these days has moved away from this old paradigm

where a human programs in the intelligence to a new paradigm where the computer just learns it itself, usually better at the cost of you having no clue on the darn thing works. He just knows better than you. And that's fine if the stakes are very, small, but if this is the thing flying your airplane or your Boeing 737 max or controlling the nuclear arsenal or something, you'd like to have a higher bar on really trusting the thing than just some dude telling you that I trained this on 15 terabytes of data and I worked every time. And I'm quite convinced, actually, we can make a lot of progress. I think there's a

common misconception, almost like people almost worship the mysteriousness of neural nets and say part of the magic of them that makes them so powerful is they're inscrutable. I think that's not true.

I think the power of them doesn't come from their inscrutability, but from their differentiability in nerd speak, but in plain English from the fact that they're basically a black box with a lot of knobs that you can continually turn, tune until they work really, really great.

So they can learn for themselves from data. And what I would really like to see is where we combine the best aspect of this new paradigm of machine learning with the old school thing with little programs that are human understandable and simple. We humans actually know it's possible because we do it in our brains, right? if someone teaches you to play baseball or whatever you know you get pretty good at it by training your neural network but then if someone asks you how does a ball move exactly you can explain in english you can say it moves in this shape called a parabola here's the equation for it yeah and if we can do it and if you

don't think that intelligence is something mysterious limited the bioorganisms there's no reason we couldn't also develop AI tools that can do the same you can first use this power of deep learning to figure out how to do stuff in an obscure, unintelligible way, and then rather than to stop at that point, do some additional things, try to simplify down to something that can be explained and understood. As you mentioned, we've had some progress doing this with simple physics problems, and if we could make more progress, I think there's a real hope here that we can get systems that we trust better,

trust for the simple reason that we actually can understand how they're working. The real threat from AI systems today is not that they're going to turn evil like in some lousy Hollywood movie, but that we they just turn very competent and

accomplished things we didn't want them to accomplish because you hadn't fully understood them. Boeing put this very simple system into their 737 max things and they hadn't understood how it worked, it didn't work so well for them. Knight Capital had this trading algorithm that they hadn't fully understood and it lost them $10 million a minute

kept going for 44 minutes until someone were like, oh, shut this tapping off. So we're putting AI in charge of ever more decisions and infrastructure that just affects people's lives. And I think we should and can raise the bar and making sure that these are things we can actually understand. And since we're running at a time, can I say something a little bit optimistic. I'd like to end always on an optimistic note. The risks get so much airtime because of Hollywood movies and so on.

And yes, there are very serious risks. I mean, you founded something called the Future of Life Institute. Which sounds a little bit hopeful. I don't know. But of course there. It begs the question of whether or not life is something that we might not have a future for. Well, of course, if we create something more intelligent than us, you don't have to think very long about it until you realize that, yeah, someone we don't like could use it to do also the bad things.

Of course, things could go to hell in the handbasket. But that's just the nature of any powerful technology, right? The interesting question for me is not the quibble about whether you should worry or not. The interesting question is, what can we do right now to maximize the chances that this goes well? First of all, what we have to do with some nerdy work that I could discuss. for example, try to build systems that we can trust because we can actually understand them better. Second, we have to really have a conversation about what kind of future we actually want.

We're here in my office at MIT, right? At this very table, I often have students coming in for career advice. I always start by asking them, where do you want to be in the future? And if she looks at me and all she can say is, oh, maybe I'll get more. ordered. Maybe I'll get cancer. Terrible strategy for career planning. I want her to come in, their eyes shining and saying, Max, this is where I want to be. And then we can talk about strategies for actually getting there and all the pitfalls that she should avoid, right? And that's what we should do as a species also. But are we? No, we're doing exactly the

opposite. We're doing precisely this sort of silly thing that's, fortunately, I hope, she won't do. We go to the movies and watch sci-fi about the future and it's almost always dystopian, right? Which makes us paralyzed with fear basically and we really need to have a serious conversation about what kind of future, high-tech future are we truly excited about? We also want to not get murdered along the way. Of course. But it has to start with a shared positive vision. It's so cool that you get people in China and in Russia and in Kansas and in L.A. all on board about it saying this is a future. for humanity that we all agree we're all going to be much better off we would like this to happen

that fosters collaboration and that's the time to start making lists of the things that can go wrong that you have to then make sure you avoid if you can't start by even giving the positive vision they're not going to want if you're going to feel it's a waste of time talking about those things right in 1945 people in europe you know where i was born looked at each other and said oops we screwed up again, you know. And something really remarkable happened. Some people decided to see if they could come up with a really positive vision for a change, for how they were going to use this ever-improving tech for not just making ever more powerful

weapons, but maybe create a society that was awesome, where everybody could have free health care, free college education, free pensions, and a society where everybody was better off. And I'm pretty sure back in those days, they got a lot of. of, they got pooed a lot and go smoke some weed, hug a tree, you know, it's never going to work. But the positive vision was compelling enough and realistic enough that they got enough support that it started to happen in countries like Sweden, England, Germany, and soon

and by now it's happened in all rich industrial countries except the one they were in. And that's the power of a positive vision, right? And I would like to see a new vision. which is much more ambitious. It's powered by AI for how we can have a truly awesome future for everybody on the planet, where we amplify our own intelligence with artificial intelligence to solve all the problems that we're stumped on now. And we think hard about also how we're going to not just make the tech, but how we're going to make sure that it's deployed to help everybody,

rather than just to help me, Max, who owns all the tech, to take over and screw all the rest of you, if we can have such a shared vision, then I think it's really much, much more likely that we're going to get it. All right, I love it. I love the optimistic way to end. I do want to foresee the future

where you and your students build little AI agents which get conscious enough that they can deny that we exist. I refuse to believe they live in a simulation. Would you invite them on your podcast? I would. I would love to have that. All right. Max Take, Mark.

Thanks so much for being on the podcast. Thank you.