Donald Hoffman: Reality is an Illusion - How Evolution Hid the Truth | Lex Fridman Podcast #293

Donald Hoffman 0:00 Whatever reality is, it's not what you see. W- what you see is, is just an adaptive fiction.

Lex Fridman 0:12 The following is a conversation with Donald Hoffman, Professor of Cognitive Sciences at UC Irvine, focusing his research on evolutionary psychology, visual perception, and consciousness. He's the author of over 120 scientific papers on these topics, and his most recent book titled The Case Against Reality: Why Evolution Hid the Truth from Our Eyes. I think some of the most interesting ideas in this world, like those of Donald Hoffman's, attempt to shake the foundation of our understanding of reality, and thus, they take a long time to internalize deeply, so proceed with caution. Questioning the fabric of reality can lead you to either madness or the truth. And the funny thing is, you won't know which is which. This is the Lex Fridman Podcast. To support it, please check out our sponsors in the description, and now, dear friends, here's Donald Hoffman.

Lex Fridman 1:12 In your book, The Case Against Reality: Why Evolution Hid the Truth from Our Eyes, you make the bold claim that the world we see with our eyes is not real. It's not even an abstraction of objective reality, it is completely detached from, uh, objective reality. Can you explain this idea?

Donald Hoffman 1:30 Right, so this is a theorem from evolution by natural selection. So, the technical question that I and my team asked was, "What is the probability that natural selection would shape sensory systems to see true properties of objective reality?" And to our surprise, we found that the answer is precisely zero, except for one, one kind of structure that we can go into if you want to. But for, for any generic structure that you might think the world might have, a total order, a topology, metric, the probability is precisely zero that natural selection would shape any sensory system of any organism to see any aspect of objective reality. So, in that sense, uh, what we're seeing is what we need to see to stay alive long enough to reproduce. So, in other words, we're seeing what we need to guide adaptive behavior, full stop.

Lex Fridman 2:23 So, the evolutionary process, the process that took us from the origin of life on Earth-

Donald Hoffman 2:30 (clears throat)

Lex Fridman 2:30 ... to the humans that we are today, that process does not maximize for truth, it maximizes for fitness, as you say, "Fitness beats truth." And fitness does not have to be connected to truth, is the claim. And that's where you have a- an approach towards zero of probability that we have evolved, human cognition, human consciousness, whatever it is, the magic that makes our mind work, evolved not for its ability to see the truth of reality, but its ability to survive in the environment.

Donald Hoffman 3:09 That's exactly right. So, most of us intuitively think that surely the way that evolution will make our senses more fit is to make them tell us more truths, or at least the truths we need to know-

Lex Fridman 3:22 Yeah.

Donald Hoffman 3:22 ... about objective reality, the truths we need in our niche. That's the standard view, and it was the view I took. I mean, that's, that's sort of what we're taught or just even assume. It's just sort of like the intelligent assumption that we would all make. But we don't have to just wave our hands. Uh, evolution by natural selection is a mathematically precise theory. Uh, John Maynard Smith, uh, in the '70s, uh, created evolutionary game theory, and we have evolutionary graph theory and even genetic algorithms that we can use to study this. And so we don't have to wave our hands, it's- it's a matter of theorem and proof and/or simulation before you get to theorems and proofs. And, uh, a couple graduate students of mine, Justin Maurach and Brian Marion, um, did some wonderful simulations that tipped me off that there was something going on here, and then I went to a mathematician, Chetan Prakash and Maneesh Singh and, uh, some other friends of mine, uh, Chris Fields and... But Chetan was the real mathematician in- behind all this and he's proved several theorems that, uh, uniformly indicate that, um, with one exception, which has to do with probability measures, um, there's no, uh, the probability is zero. Uh, the- the reason there's an exception for probability measures, so-called sigma algebras or, or, um, s- s- sigmatitive classes, is that for any scientific theory, uh, there is the assumption th- that needs to be made that the whatever structure the s- whatever probabilistic structure the world may have is not unrelated to the probabilistic structure n- of our perceptions. If they were completely unrelated, then no science would be possible. So in, so this is technically the- the map from reality to our senses has to be a so-called measurable map, has to preserve sigma algebras. But that means it could be infinite to one, and it could collapse all sorts of, of event information. But other than that there's, there's no requirement in standard evolutionary theory for, uh, fitness payoff functions, for example, to preserve any specific structures of objective reality. So, you can ask the technical question, this is one of the avenues we took. Um, if you look at all the fitness payoffs from, um, whatever world structure you might want to i- imagine, so a world with say a- a total order on it. So it's got end states and they're totally ordered. And then you can have, uh, a set of maps from that world into a set of payoffs, say from zero to 1,000 or whatever you want your payoffs to be, and you can just literally count all the payoff functions, and just do the combinatorics and count them. Then you can ask a precise question, how many of those payoff functions preserve the total order? If that's what you're lo- ... Or how many preserve the topology?... and you just count them and divide. So, so the number that are homomorphisms versus the total number, and then take the limit as the number of states in the world and the number of payoff values goes very large. And when you do that, you get zero every time.

Lex Fridman 6:21 O- okay. You've (laughs) ... There's a million things to ask here. But first of all, just in case, uh, people are not familiar with your work, let's sort of linger on the big bold statement here-

Donald Hoffman 6:35 Sure. Mm-hmm.

Lex Fridman 6:36 ... which is, the thing we see with our eyes is not some kind of limited window into reality, it is completely detached from reality, likely completely detached from reality. You're saying 100% likely. (sighs) Okay. So none of this is real in the way we think is real, in the way we have this intuition there's, um, like this table is some kind of abstraction, but underneath it all, there's atoms, and there's an entire century of physics that describes the functioning of those atoms and the quarks that make them up. There's, uh, m- many Nobel Prizes-

Donald Hoffman 7:17 (laughs)

Lex Fridman 7:17 ... about particles and fields and all that kinda stuff that, uh, slowly builds up to something that's perceivable to us, both with our eyes, with our different senses as this table. Then there's also ideas of chemistry that over layers of abstraction, from DNA to embryos, the cells that make the human body.

Donald Hoffman 7:42 Right.

Lex Fridman 7:43 So all of that is not real.

Donald Hoffman 7:46 It's a real experience, and it's a real adaptive set of perceptions. So it's an adaptive set of perceptions, full stop. The- we want to think that the perceptions-

Lex Fridman 7:57 So the perceptions are real.

Donald Hoffman 7:58 So, so their perceptions are real as perceptions, right? They, they are... We, we are having our perceptions, but we've assumed that there's, uh, a, a pretty tight relationship between our perceptions and reality. If I look up and see the moon, then there is something that, uh, exists in space and time that, uh, matches, um, what I perceive. And all I'm saying is that if you take evolution by natural selection seriously, then that is precluded, that our perceptions are there, they're there to guide adaptive behavior, full stop. They're not there to show you the truth. In fact, the way I think about it is they're there to hide the truth because the truth is too complicated. It's just like if you're trying to, you know, use your laptop to write an email, right? What you're doing is toggling voltages in the computer, but good luck trying to do it that way. That's... We... The reason why we have a user interface is because we don't want to know that "truth." The diodes and resistors and all that, that terrible hardware. If you had to know all that truth, it would... you know, your friends wouldn't hear from you . So you... so what evolution gave us was perceptions that guide adaptive behavior, and part of that process, it turns out, means hiding the truth and giving you, um, uh, eye candy.

Lex Fridman 9:16 So what's the difference between hiding the truth and forming abstractions, uh, layers upon layers of abstractions over these, over low level voltages and transistors and, uh, chips and, uh, programming languages from Assembly to Python that then leads you to be able to have an interface like Chrome where you open up another set of JavaScript and HTML, uh, programming languages that lead you to have a graphical user interface on which you can then send your friends an email? Is that completely detached from the zeros and ones that are firing away inside the computer?

Donald Hoffman 10:01 It's not, though, of course, when I talk about the user interface on your desktop, um, there's this whole sophisticated backstory to it, right? That, that the hardware and the software that's allowing that to happen. Evolution doesn't tell us the backstory, right? So the theory of evolution is not going to be adequate to tell you what is that backstory. It's gonna say that whatever reality is, and that's the interesting thing, it says whatever reality is, you don't see it. You see a user interface, but it doesn't tell you what that user interface is, how it's built, right? Now, we can, we can try to look at certain aspects of the interface, but already we're gonna look at that and go real... okay, before I would look at neurons and I was assuming that I was seeing something that was, uh, at least partially true, and now I'm realizing it, it could be like looking at the pixels on my desktop, uh, or icons on my desktop, and good luck, you know, going from that to the data structures and then the voltages and the... I mean, good luck. It, it... there's just n- no way. So what's interesting about this is that our scientific theories are precise enough and rigorous enough to tell us certain limits, but... and even limits of the theories themselves, but they're not going to tell us what the next move is. And that's where scientific creativity comes in. So the, the stuff that I'm saying here, for example, um, is not alien to physicists. The physicists are saying precisely the same thing that space time is doomed. We've assumed that space time is fundamental. We've assumed that for, for several centuries, and it's been very useful. So all the things that you are mentioning, the particles and all the work that's been done, that's all been done in space time. But now physicists are saying space time is doomed. There's no such thing as space time fundamentally in the laws of physics, and that comes actually out of gravity together with quantum field theory. It just comes right out of it. It's, it's, it's a theorem of, of, of those two theories put together, but it doesn't tell you what's behind it. So the physicists are... know that their, their best theories, Einstein's gravity and quantum field theory put together, is...... entail that spacetime cannot be fundamental and therefore, particles in spacetime cannot be fundamental. They're just irreducible representations of the symmetries of spacetime. That's what they are. So we have, so spacetime, so we put the two together. We put together what the physicists are discovering, and we can talk about how they do that, and then we, uh, the new discoveries from evolution by natural selection. Both of these discoveries are really in the last 20 years, and what both are saying

Donald Hoffman 12:40 is, um, spacetime has had a good ride.

Lex Fridman 12:43 (chuckles)

Donald Hoffman 12:43 It's been very useful. Reductionism has been useful, but it's over and it's time for us to go beyond.

Lex Fridman 12:48 When you say spacetime is doomed, is it the space? Is the, is the, is it the time? Is it the very hard-coded specification of four dimensions? Um, or are, are you specifically referring to the kind of, um, perceptual domain that humans operate in, which is spacetime? You think like there's a 3D, um, like our world is three-dimensional and time progresses forward, therefore three dimensions plus one, 4D. What, uh, what, what exactly do you mean by spacetime? What, when, (laughs) what, what do you mean by spacetime is doomed?

Donald Hoffman 13:24 Great. Great. So this is, by the way, not my quote, this is from, for example, Nima Arkani-Hamed at the Institute for Advanced Study at Princeton, Ed Witten also there, David Gross, Nobel Prize winner. So this is not just something that cognitive scientists, this is what the physicists are saying.

Lex Fridman 13:40 Yeah. The physicists, they're spacetime, uh, skeptics.

Donald Hoffman 13:45 Well, yeah, they're saying-

Lex Fridman 13:46 (laughs)

Donald Hoffman 13:46 ... that, and I can say exactly why they think it's doomed, but what they're saying is that, you know, 'cause your question was, what, what aspect of spacetime, what are we talking about here? It's both space and time, their union into spacetime, a- as in Einstein's theory, that's doomed.

Lex Fridman 14:00 Mm-hmm.

Donald Hoffman 14:01 And they're, they're basically saying that, uh, even quantum theory, uh, this is with Nima Arkani-Hamed especially. So Hilbert spaces will not be fundamental either. So that, that where the notion of Hilbert space, which is really critical to quantum field theory, quantum information theory, uh, that's not going to figure in the fundamental new laws of physics. So what they're looking for is some new mathematical structures beyond spacetime, beyond, you know, Einstein's four-dimensional spacetime or super symmetric version, you know, geometric algebra signature, two comma four kind of. Uh, there are different ways that you can represent it, but they're finding new structures. And, and, by the way, they're succeeding now. They're finding, they found something called the amplituhedron, this is Nima and his colleagues, the, the cosmological polytope. These are... So the- there are these like polytopes, these polyhedra in, in multi-dimensions, generalizations of simplices that are coding for, for example, the scattering amplitudes of, of processes in the Large Hadron Collider and other, other colliders. So they're finding that if they let go of spacetime completely, they're finding new ways of computing these scattering amplitudes that turn literally billions of terms into one term. When you do it in space and time, because it's the wrong framework, it's- it's- it's just a user interface from, that's now from the evolutionary point of view, it's just user interface, it's not a deep insight into the nature of reality. So it's missing deep symmetry is something called a dual conformal symmetry, which turns out to be true of the scattering data, but you can't see it in spacetime, and is making the compli- the computations way too complicated because you're trying to compute all the loops and Feynman diagrams and all the Feynman integrals. So see, the Feynman approach to the scattering amplitudes is trying to enforce two critical properties of spacetime, locality and unitarity. And so by, when you enforce those, you get all these loops and multiple, you know, different levels of loops. And for each of those, you have to add new terms to your computation. But when you do it outside of spacetime, you don't have the notion of unitarity, you don't have the notion of locality, you have something deeper and it's capturing some symmetries that are actually true of the data. And, but then when you look at the geometry of the facets of these polytopes, then certain of them will code for unitarity and, uh, locality. So it actually comes out of the structure of these deep polytopes. So what we're finding is there's this whole new world now beyond spacetime that is making explicit symmetries that are true of the data that cannot be seen in spacetime and that is turning the computations from billions of terms to one or two or a handful of terms. So we're getting insights into symmetries and we're... And all of a sudden the math is becoming simple because we're not doing something silly. We're not adding up all these loops in spacetime. We're doing something far deeper. But they don't know what this world is about. Also, you know, they're in an interesting position where we know that spacetime is doomed. And I, I, I should probably tell you why it's doomed, uh, what they're saying about why it's doomed, but, but they need a flashlight to look beyond spacetime. What is, what flashlight are we gonna use to look into the dark beyond spacetime? Because Einstein's theory and quantum theory can't tell us what's beyond them. All they can do is tell us that when you put us together, spacetime is doomed at 10 to the minus 33 centimeters, 10 to the minus 43 seconds, beyond that, spacetime doesn't even make sense. It just has no operational definition. So, but it doesn't tell you what's beyond. And so they're, they're just looking for deep structures, like guessing. It's really fun. So these really brilliant guys, generic, we have brilliant men and women who are doing this work-

Lex Fridman 17:48 Mm-hmm.

Donald Hoffman 17:48 ... uh, physicists, are making guesses about these structures, informed guesses, because they're trying to ask, "Well, okay, what deeper structure could give us the stuff that we're seeing in spacetime, but without certain commitments that we have to make in spacetime, like locality So they make these brilliant guesses. And of course, most of the time you're gonna be wrong. But once you get one or two, that start to pay off. And then you get some lucky breaks. So they got a lucky break back in 1986. Um, a couple of mathematicians named Park and Taylor.... took the scattering amplitude for two gluons coming in at high energy and four gluons going out at low energy. So, that kind of scattering thing. So, so like, apparently for people w- who are into this, that's sort of something that happens so often you need to be able to find it and get rid of those, 'cause you already know about that and you need to... So, you needed to compute them. It was billions of terms. And they couldn't do it, even for the supercomputers, couldn't do that for the many billions or millions of times per second they needed to do it. So, they, they begged, you know, the experimentals begged the theorist, "Please, can you... I mean, you gotta..." And so Park and Taylor took the billions of terms, hundreds of pages, and miracli- miraculously turned it into nine, and then a little bit later, they guessed one term expression that turned out to be equivalent. So, billions of terms reduced to one term, the so-called famous Park-Taylor formula, 1986. And that was like, okay, where did that come from? What... This is a pointer into a deep realm beyond space and time, but, but no one... I mean, what can you do with it? And they thought maybe it was a one-off, but then other formulas started coming up and then eventually, Nima Arkani-Hamed and his team found this thing called the amplituhedron which really sort of captures the whole, a, a big part of the whole ball of wax. Um, I- I'm sure they would say, "No, there's plenty more to do." So, so I won't say they did it all, by any means. They're looking at the cosmological polytope as well. So, what's remarkable to me is that two pillars of modern science, quantum field theory with gravity on the one hand, and evolution by natural selection on the other, just in the last 20 years have very clearly said spacetime has had a good run, reductionism has been a fantastic methodology, so we had a great ontology of spacetime, a great methodology of reductionism. Now it's time for a new trick.

Lex Fridman 20:09 (laughs)

Donald Hoffman 20:10 But now you need to go deeper. And, and show... But by the way, this isn't, doesn't mean we throw away everything we've done. N- not by a long shot. Every new idea that we come up with beyond spacetime must project precisely into spacetime and it better give us back everything that we know and love in spacetime or generalizations. Or it's not gonna be taken seriously, and it shouldn't be. So, so we have a strong constraint on whatever we're going to do beyond spacetime. It needs to project into spacetime. And whatever this deeper theory is, it may not itself have evolution by natural selection. This may not be part of this deeper realm. But when we take the, whatever that thing is beyond spacetime and project it into spacetime, it has to look like evolution by natural selection, or it's wrong. So, so that's, so that's a strong constraint on, on this work.

Lex Fridman 20:57 So, even the evolution by natural selection and, um, quantum field theory, uh, could be interfaces into something that n- that doesn't look anything like... Like you mentioned evol- I mean, it's interesting to think that evolution might be a very crappy interface into something much deeper.

Donald Hoffman 21:18 That's right. They're both telling us that the framework that you've had can only go so far. And it has to stop. And there's something beyond. And that framework, th- the very framework that is, is spacetime itself. Now, of course, evolution by natural selection is not telling us, uh, about like Einstein's relativistic spacetime. So, that was another question you asked a little bit earlier. It's telling us more about our perceptual space and time, which, um, we have used as the basis for creating first a Newtonian space versus time as a mathemat- mathematical extension of our perceptions. And then Einstein then took that and extended it even further. So, the relationship between what evolution is telling us and what the physicists are telling us is that in some sense, the Newton and Einstein spacetime are formulated as sort of rigorous extensions of our perceptual space, um, making it mathematically rigorous and f- and laying out the symmetries that, that, that they find there. So, that's sort of the relationship between them. So, it's the perceptual spacetime that evolution is telling us is just a, a user interface, effectively, and then the physicists are finding that even the mathematical extension of that into the Einsteinian formulation has to be as well, um, not the final story. There's something deeper.

Lex Fridman 22:38 So, let me ask you about reductionism and interfaces. As we march forward from Newtonian physics, uh, to quantum mechanics, these are all, in your view, interfaces. Um, are we getting closer to objective reality? How do we know... If these interfaces in the process of science, the reason we like those interfaces is because they're predictive of some aspects, strongly predictive about some aspects of our reality. Is that completely deviating from our understanding of that reality? Or is it helping us get closer and closer and closer?

Donald Hoffman 23:22 Well, of course, one critical constraint on all of our theories is that they are empirically tested and pass the experiments that we have for them. So, so no one's arguing against experiments being important and wanting to test all of our, our current theories and, uh, any new theories on that. So, that's, that's, that's all there. But we have good reason to believe that science will never get a theory of everything. In a sense-

Lex Fridman 23:52 Everything, everything.

Donald Hoffman 23:53 Everything, everything. Right. A final theory of everything, right. I think that my, my own take is, for what it's worth, is that Godel's incompleteness theorem sort of points us in that direction. That e- even with mathematics, uh, any finite axiomatization that's sophisticated enough to be able to do arithmetic, it's easy to show that there'll be, um, statements that are true that can't be proven, can't be deduced from within that framework. And if you add the new statements to your axioms, then there'll be always new statements that are true but can't be proven with the new axiom system. And s-... the best scientific theories, um, in physics for example, and also now evolution, are mathematical. So our theories are gonna be- a- a- they're gonna have their own assumptions, and, um, they'll be mathematically precise, and there'll be theories perhaps of everything except those assumptions 'cause the assumptions are, we say, "Please grant me these assumptions. If you grant me these assumptions, then I can explain this other stuff." But, so you have the assumptions that, um, are like miracles as far as the theory is concerned. They're not explained. They're the starting points for explanation. And then you have the mathematical structure of the theory itself which will have the girdle limits. And so, my- my take is that, um, reality, whatever it is, is always going to transcend any conceptual theory that we can come up with.

Lex Fridman 25:22 There's- there's always gonna be mystery at the edges. Uh, (laughs) -

Donald Hoffman 25:27 Right.

Lex Fridman 25:27 ... uh, contradictions and all that kind of stuff. Okay. (sighs)

Donald Hoffman 25:31 And truths.

Lex Fridman 25:32 So there's a- this idea that was brought up in the- in the financial space of, uh, settlement of transactions. It's o- often talked about in cryptocurrency especially. So you could do, you know, money, cash, is not connected to anything. Uh, it used to be connected to gold, to physical reality, but then you can use money to exchange, uh, to exchange value, trans- transact. Uh, so when- when it was in the gold standard, eh, the money would represent some stable, uh, component of reality. Isn't it more effective to avoid things like hyperinflation if we generalize that idea? Isn't it better to connect your, uh, whatever we humans are doing in the social interaction space with each other, isn't it better, uh, from an evolutionary perspective to connect it to some degree to reality so that the- the- the transactions are settled with something that's universal as opposed to us constantly operating in something that's a complete illusion? Isn't it easy to hyperinflate that? (laughs) Like, where n- where you really-

Donald Hoffman 26:43 (laughs)

Lex Fridman 26:43 ... deviate very, very far away from, um, from the underlying reality? Or do you not, never get in trouble for this? Can you just completely drift far, far away from the underlying reality and never get in trouble?

Donald Hoffman 27:01 That's a great question. S- y- on the financial side, there's two levels at least th- that we could take your question. One- one is strictly like evolutionary psychology of financial systems. Um, and that's- that's pretty interesting. Um, and there, the decentralized idea, the De- DeFi kind of idea-

Lex Fridman 27:17 Mm-hmm.

Donald Hoffman 27:17 ... in- in cryptocurrencies may make good sense from just an evolutionary psychology point of view. Having, you know, human nature being what it is, putting a lot of faith in a few central controllers, um, depends a lot on the veracity of those c- and trustworthiness of those few central controllers. And we have ample evidence time and again that, um, that's often betrayed. So it- it makes good evolutionary sense, I would say, to have a decentralized... I mean, democracy is a step in that direction, right? We're- we don't- we don't have a monarch now telling us what to do. We decentralize things, right? Because if the monarch, if you have Marcus Aurelius as your emperor, you're great. If you have Nero, it's not so great. And so we don't want that. So democracy is a step in that direction. But- but I think the DeFi thing is- is e- an even bigger step and is- is going to even make the democratization even- even greater. So- so that's one level of it.

Lex Fridman 28:14 Also, the fact that power corrupts and absolute power corrupts absolutely is also an evolu- uh, a consequence of, uh, evolution. (laughs)

Donald Hoffman 28:24 Right.

Lex Fridman 28:24 That's also a feature, I think, right?

Donald Hoffman 28:27 Right.

Lex Fridman 28:27 You can argue from the long span of living organisms, it's nice for power to corrupt, for you to, eh, so, uh, madmen and women throughout history might be useful to teach us a lesson (laughs) about ourselves.

Donald Hoffman 28:43 We can learn from our negative example, right?

Lex Fridman 28:44 Exactly. (laughs)

Donald Hoffman 28:45 Right, right. (laughs) Right. So power does corrupt, and I think the, you can think about that again from an evolutionary point of view. But I think that your question was a little deeper w- when that was, is, does the evolutionary interface idea sort of unhinge science from- from some kind of important test for the theories, right?

Lex Fridman 29:08 Yeah.

Donald Hoffman 29:08 We don't want... It- it, uh, doesn't mean that anything goes i- in scientific theory, but there's no, if there, if we don't see the truth, is there no way to tether our theories and- and test them? And- and I think ... there, there's no problem there. We- we can only test things i- in terms of what we can measure with our senses in space and time. So w- we're going to have to continue to do experiments and, but we'll go to re-, we're gonna understand a little bit differently what those experiments are. W- w- we had thought that when we see a pointer on a, uh, some machine in a, you know, experiment, that the machine exists, the pointer exists, and the values exist even when no one is looking at them, and that they're an objective truth. And- and, uh, our best theories are telling us, no. The pointers, pointers are just pointers, and that's what you have to rely on for making your judgments. Um, but, um, the, even the pointers themselves are not the objective reality. So- and- and I think Godel is telling us that not that, um, anything goes, but, you know, as you develop new axiom systems, you will find out what goes within that axiom system and what- what testable predictions you can make. So I don't think we're- we're untethered. We- we continue to do experiments. What I think w- we won't have that we want is a conceptual understanding t-... that gives us a theory of everything that's final and complete. I, I, I think that this is, to put it another way, this is job security for scientists.

Lex Fridman 30:43 (laughs)

Donald Hoffman 30:44 Our, our job will never be done. This is job security for neuroscience because before we thought that when we looked in the brain we saw neurons and neural networks and, and, uh, you know, action potentials and, and synapses and so forth and that's, that was it. That, that was the reality. Now we have to reverse engineer that. We have to say what is beyond space-time? What is going on? What is a dynamical system beyond space-time that when we project it into Einstein space-time gives us things that look like neurons and neural networks and synapses? That's ... So we have to reverse engineer it, so there's gonna be lots more work for neuroscience. It's gonna be far more complicated and, and difficult and challenging. But, but that's wonderful. That's what we need to do. We thought neurons exist when they aren't perceived and, and they don't. In the same way that if I show you ... When I say they don't exist, I, I should be very, very concrete. If I draw on a piece of paper a little sketch, um, of, of something that is called the Necker cube, it's just a little line drawing of a cube. Right? So on a flat piece of paper. If I execute it well and I show it to you, you'll see a 3D cube and you'll see it flipped. Sometimes you'll see one face in front, sometimes you'll see the other face in front. But if I ask you, you know, "Which face is in front when you don't look?" You know, the answer is, "Well, neither face is in front 'cause there's no cube."

Lex Fridman 32:01 (laughs)

Donald Hoffman 32:01 It's, it's a flat piece of paper.

Lex Fridman 32:02 Yeah.

Donald Hoffman 32:03 So when you look at the piece of paper, you perceptually create the cube and when you look at it then, you fix one face to be in front and one face to be out. So that's what I mean when I say it doesn't exist. Space-time itself is like the cube. It's a data structure that your sensory systems construct, whatever your sensory systems mean now, because we now have to even not... even take that for granted.

Lex Fridman 32:27 Mm-hmm.

Donald Hoffman 32:27 But there are perceptions that you construct on the fly and, uh, they're data structures in a computer science sense, and you garbage collect them when you don't need them so you create them and garbage collect them.

Lex Fridman 32:37 But is it possible that it's mapped well in some concrete predictable way to objective reality? The sheet of paper, this two-dimensional space or we can talk about space-time maps in some way that we, uh, maybe don't yet understand but will one day understand wh- what that mapping is, but it maps reliably. It is tethered in that way.

Donald Hoffman 33:03 Well, yes, and, and so the new theories that the physicists are finding beyond space-time have that kind of tethering. So they're, they show precisely how you start with an amplituhedron and how you project this high dimensional structure into the four dimensions of space-time. So there's a precise procedure that, that relates the two and they're doing the same thing with the cosmological polytopes. So, so there are the... they're the ones that are making the most, uh, you know, concrete and, and fun advances going beyond space-time and they're, they're, they're tethering it, right? Th- they say this is precisely the mathematical projection from this deeper structure into space-time. One, one thing I'll say about f- as a non-physicist what I find interesting is that they're finding just geometry but there's no notion of dynamics. Right now, they're just finding these static geometric structures which is impressive. I'm ... So I'm not putting them down. This is... What they're doing is unbelievably complicated and, and brilliant and, uh, um, adventurous all the... and it's, it's all those things.

Lex Fridman 34:08 And beautiful-

Donald Hoffman 34:09 And, and beautiful, yeah, right.

Lex Fridman 34:09 ... uh, uh, from a human aesthetic perspective because geometry's beautiful.

Donald Hoffman 34:12 It's, it's... Absolutely. And it's fi- they're finding symmetries that are true of the data that can't be seen in space-time. But I'm looking for a theory beyond space-time that's a dynamical theory. I would love to find, and we can talk about that at some point, a, a theory of consciousness in which the dynamics of consciousness itself will give rise to the geometry that the physicists are finding beyond space-time. If we can do that, then we'd have a completely different way of looking at how consciousness is related to what we call the brain or the physical world more generally, right? Right now, all of my brilliant colleagues, well, 99% of them are trying to... they're, they're assuming space-time is fundamental, they're assuming that particles are fundamental, quarks, gluons, leptons and so forth. Elements, atoms and so forth are fundamental and that therefore neurons and brains are part of objective reality and that somehow when you get matter this complicated enough, it will somehow generate conscious experiences by its functional properties or if you're a panpsychist, um, maybe y- you... in addition to the physical properties of particles you add, uh, you know, a consciousness, um, p- property as well and then you have... you c- combine these physical and conscious properties to get more complicated ones but they're all doing it within space-time.

Lex Fridman 35:34 Mm-hmm.

Donald Hoffman 35:36 All of the work that's being done on consciousness and its relationship to the brain has all assumed something that our fit... our best theories are telling us is doomed, space-time.

Lex Fridman 35:46 Why does that particular assumption bother you the most? So you bring up space-time. Uh, uh, I mean, that's just one useful interface we've used for a long time. Uh, surely there's other interfaces. Is space-time just the... one of the big ones that you... to build up people's intuition about the fact that they just do assume a lot of things strongly? Or, or it i- i- is it in fact a fundamental flaw in the way we see the world?

Donald Hoffman 36:17 Well, everything else that we think we know are things in space-time.

Lex Fridman 36:22 (laughs) Sure.

Donald Hoffman 36:24 And so if, if you... when you say space-time is doomed, this is a shot to the heart-

Lex Fridman 36:29 Yeah.

Donald Hoffman 36:30 ... of the whole framework, the c- whole conceptual framework that we've had in science.... not, not to the scientific method, but to the, the fundamental ontology and also the fundamental methodology, the ontology of space-time and its contents, and the methodology of reductionism, which is that as we go to smaller scales in space-time, we will find more and more fundamental laws. And that's been very useful for, for space and time for centuries, reductionism for centuries. But, but now we realize that, um, that's

Donald Hoffman 37:04 over. Reductionism is in fact dead, as is space-time.

Lex Fridman 37:08 What exactly is reductionism? What is the process of reductionism that is different than, uh, some of the physicists that you mentioned there trying to think, trying to let go of the assumption of space-time, looking beyond? Isn't that still trying to come up with a simple model that explains this whole thing? Isn't it still reducing?

Donald Hoffman 37:29 It's a wonderful question because it really helps to clarify two different notions, which is scientific explanation on the one hand and a particular kind of scientific explanation on the other, which is the reductionist. So the reductionist explanation is saying, "I will start with, with things that are smaller in space-time and, and therefore more fundamental, and, where the laws are more fundamental."

Lex Fridman 37:50 Mm-hmm.

Donald Hoffman 37:51 So we go, uh, to just smaller and smaller scales. Whereas w- in science more generally, we just say, like when Einstein did the special theory of relativity, he's saying, "Let me have a couple postulates. I will assume that the speed of light is universal for, for all, um, observers, uh, in uniform motion, um, and that the laws of physics, so for, for, for, you know, uniform motion are, are those." That's not a reductionist. That, those are saying, "Grant me these assumptions. I can build this entire concept of space-time out of it." It's not a reductionist thing. You're not going to smaller and smaller scales of space. You're, you're coming up with these deep, deep principles. Same thing with his theory of gravity, right? It's, it's the falling elevator ide- idea, right? So this is not a reductionist kind of thing. It's, it's, it's, it's, it's something different. So s-

Lex Fridman 38:40 So simplification is a bigger thing than just reductionism?

Donald Hoffman 38:44 That, yeah, r- reductionism has been a particularly useful kind of scientific explanation, for example, in thermodynamics, right, where the notion that we have of heat, some ma- macroscopic thing like temperature and heat. It turns out that, you know, Boltzmann and others discovered, well, hey, you know, if we go to smaller and smaller scales, we find these things called molecules or atoms. And if we think of them as bouncing around and having some kind of, um, energy, then, um, what we call heat is, is a, is really can be reduced to, to that. And, and so that's a particularly useful kind of e- um, reduction, is a useful kind of scientific explanation that works within a range of scales within space-time. But we know now precisely where that has to stop, at 10 to the minus 33 centimeters and 10 to the minus 43 seconds. And I would be impressed if it was 10 to the minus 33 trillion centimeters. I'm not terribly impressed at 10 to the minus 33 centimeters.

Lex Fridman 39:41 (laughs) I don't even know how to comprehend either of those numbers, frankly. Uh, do... Just a small aside-

Donald Hoffman 39:49 (laughs)

Lex Fridman 39:49 ... 'cause I am a computer science person, I also find cellular automata beautiful.

Donald Hoffman 39:53 Yes.

Lex Fridman 39:54 And, uh, so you have somebody like, uh, Stephen Wolfram, who recently has been very excitedly exploring, um, a proposal for a data structure that could be, uh, um, the numbers that would make you a little bit happier in terms of scale, 'cause they're very, very, very, very tiny. Um, so do you like this space of exploration, of really thinking, letting go of space-time, letting go of everything, and trying to think what kind of data structures could be underneath this whole mess?

Donald Hoffman 40:23 That's right. So if they're thinking about these as outside of space-time, then that's th- that's what we have to do. That's what our best theories are telling us. You now have to think outside of space-time. Now, of course, I should back up and say we know that Einstein surpassed Newton, right? But that doesn't mean that there's not good work to do on Newton. There's all sorts of Newtonian physics that takes us to the moon and so forth, and there's lots of good problems that we want to d- solve with Newtonian physics. The same thing will be true of space-time. We'll, we'll still... It's not like we're gonna stop using space-time. We'll continue to do all sorts of good work there. But for, for those scientists who are really looking to go deeper, to actually find the next... You know, just like what Einstein did to Newton, what, what are we gonna do to Einstein? How do we get beyond Einstein and quantum theory to something deeper? Then we have to actually let go. And, and if we're gonna do, like, this a- a- automata kind of approach, it's, it's critical that it's not automata in space-time. It's automata prior to space-time from which we're going to show how space-time emerges. If you're doing automata within space-time, well, that might be a fun model, but it's not the radical new step that we need.

Lex Fridman 41:33 Yeah, so the space-time emerges from that whatever system. And y- uh, like you're saying, it's, it's a dynamical system. Do we even have an understanding of what dynamical means when we go beyond the, the... When, when you start to think about dynamics, that could mean a lot of things. Even causality could mean a lot of things if, if we, if we realize that everything's an interface. (laughs) Uh, like what...

Donald Hoffman 41:59 Right.

Lex Fridman 41:59 How much do we really know is an interesting question, 'cause you, you brought up neurons. I gotta ask you an another, yet another tangent. There's a paper, I remember a while ago, looking at... called, uh, Could a Neuroscientist Understand a Microprocessor? And I just enjoyed that thought experiment-

Donald Hoffman 42:14 Mm-hmm.

Lex Fridman 42:14 ... that they provided, which is they basically... It's a couple of, uh, neuroscientists, Eric Jonas and Konrad Kording, uh, who use the tools of neuroscience to analyze a microprocessor, so a computer, computer chip.

Donald Hoffman 42:30 Yeah, if we lesion it here, what happens, and so forth. And if you go and-

Lex Fridman 42:33 Yeah.

Donald Hoffman 42:33 ... lesion in, uh, you know, computer, it's very, very clear that lesion experiments on computers are not gonna give you a lot of insight into-

Lex Fridman 42:39 Yeah.

Donald Hoffman 42:39 ... how it works.

Lex Fridman 42:40 And also the measurement devices and the kind of, sort of... Just using the basic approaches of neuroscience, collecting the data, uh, trying to intuit about the underlying function of it. And that helps you understand that...... our scientific exploration of concepts, um, depending on the field, uh, are, are, maybe in the very, very early stages. I wouldn't say in, leads us astray. Uh, perhaps it does sometimes. But it's not a, uh, it's not anywhere close to some fundamental mechanism that actually makes a thing work. I don't know if you can sort of comment on that-

Donald Hoffman 43:18 Sure.

Lex Fridman 43:18 ... in terms of using neuroscience to understand the human mind and neurons. Are we really far away, potentially, from, uh, understanding in the way we understand the transistors enough to be able to build a computer? So one, one, uh, one thing about understanding is you can understand for fun. The other one is to, to understand so you could build things.

Donald Hoffman 43:45 Right.

Lex Fridman 43:45 And, and that's when you really have to understand, (laughs) uh-

Donald Hoffman 43:49 Exactly. In fact, what got me into the field that I'm at MIT was, um, work by David Marr on this very topic. So David Marr was a professor at MIT, but he'd done his PhD in neuroscience, studying just the architectures of the brain. But he realized that hi- his, his work, it was on the cerebellum. Um, he, he realized that his work, as, as, as rigorous as it was, left him unsatisfied because he didn't know what the cerebellum was for.

Lex Fridman 44:18 (laughs) Yeah.

Donald Hoffman 44:19 And, and, and why it had that architecture. And so, he, he went to MIT, and he was in the AI lab there. And, and, uh, he, he said he had this three level approach that really grabbed my attention. So I, when I was an undergrad at UCLA, I read one of his papers in a, in a class and said, "Who is this guy?" Because he said, "You have to have a computational theory." What is being computed and why?

Lex Fridman 44:39 Mm-hmm.

Donald Hoffman 44:40 An algorithm, how is it being computed? What, what are the preci- preci- algorithms? And then the hardware, how does it get instantiated in the hardware? And so, to really do neuroscience, he argued, we needed to have, um, understanding at all those levels. And I thought that- That really got me. I had loved the neuroscience, but I realized this guy was saying, "If you can't build it, you don't understand it effectively." And so that's why I went to MIT. And I, I had the pl- pleasure of working with David until he, he died as, just a year and a half later. So there was- there's been that idea that, you know, with neuroscience, we, we have to have, in some sense, a top-down model of what, what, what's being computed and why, that we would then go after... And same thing with the, you know, trying to reverse engineer, uh, you know, a computing system like your laptop. We, we really wouldn't, we really need to understand what the user interface is about and why we have, what are keys on the, the keyboard for and so forth. You need to know why to really understand all the circuitry and what it's, what it's for. Now, we don't... Evolution by natural selection does not tell us the deeper question th- that we're asking. The, the answer to the deeper question, which is why. What, what if, why... What is, what's this deeper reality and what's it up to and why? It, all it tells us is that whatever reality is, it's not what you see. What you see is, is just an adaptive fiction.

Lex Fridman 46:12 So just to linger on this fascinating bold question that shakes you out of your dream state. Does this fiction still help you in building intuitions as literary fiction does about reality? The reason we read literary fiction, uh, is it helps us build intuitions and an understanding in indirect ways, sneak up to the difficult questions of human nature. Great fiction. Same with this observed reality. Um, does this interface that we get, this fictional interface help us build intuition about deeper truths of how this whole mess works?

Donald Hoffman 46:55 Well, I think that each theory that we propose will give its own answer to that question, right? So when the physicists are proposing, um, these structures like the amplituhedron and cosmological polytope, associahedron, and so forth beyond space time, we can then ask your question for those specific structures and say, "How much information, for example, does evolution by natural selection and the, the kinds of sensory, uh, systems that we have right now give us about this deeper reality and, and, and why did we evolve this way?" We can try to answer that qu- that question from within the deep. So there's not gonna be a general answer. I think we're gonna... What we'll have to do is pause at these new deeper theories and then try to answer your question within the framework of those deeper theories knowing full well that there'll be an even deeper theory.

Lex Fridman 47:47 So is it, is this paralyzing though? 'Cause, uh, how do we know we're not completely adrift, uh, out to sea, lost forever from... So, like, that our theories are completely lost. So if, if, if it's all, um... If we can never truly deeply introspect to the bottom, if it's always just turtles on top of turtles infinitely, um, isn't that paralyzing for a scientific mind?

Donald Hoffman 48:18 Well, it's interesting that you say introspect to the bottom-

Lex Fridman 48:22 (laughs)

Donald Hoffman 48:23 ... (laughs) because there, there is that one, there is one... I mean, again, this is in the same spirit of what I said before, which is, it depends on what answer you give to what's beyond space time what answer we would give to your question, right?

Lex Fridman 48:35 Mm-hmm.

Donald Hoffman 48:35 So but one answer that, um, is interesting to explore is something that spiritual traditions have said for thousands of years, but haven't said precisely, so we can't take it seriously in science until it's made precise, but we might be able to make it precise. An- and that is that, um, they've, they've also said something like, um, space and time aren't fundamental. They're Maya. They're, they're illusion. And, but, but that, um...... if you look inside, if you introspect, uh, and let go of all of your particular perceptions. Uh, you will come to something that's beyond con- conceptual thought, and that is a co- they claim, uh, being in contact with the deep ground of being that, that transcends any particular conceptual understanding. If that is correct, and I'm not saying it's correct, uh, but, and I'm not saying it's not correct. I'm just saying if that's correct, then it would be the case that as scientists, because w- we also are in touch with this ground of being, we would then not be able to conceptually understand ourselves all the way. But we could know ourselves just by being ourselves. And so we would d- there would be a sense in which there is a fundamental grounding to the whole enterprise, because we're not separate from the enterprise. This is the opposite of third, the, the impersonal third-person science. This, this would make science go-

Lex Fridman 49:58 Deeply personal. (laughs)

Donald Hoffman 49:59 ... personal all the way down. And, and, but, but nevertheless scientific because the scientific method would still be what we would use all the way down for the conceptual understanding.

Lex Fridman 50:10 Unfortunately, you still don't know if you went all the way down. It's possible that this kind of whatever consciousness is, and we'll talk about it, is getting, um, (laughs) the cliché statement of be yourself, uh, is, is, it, it is somehow digging at a deeper truth of reality. But you still don't know when you get to the bottom. You know, a lot of people they'll take psychedelic drugs, and they'll say, "Well, that takes my mind to certain places where it feels like that is revealing some deeper truth of reality." But you still, uh, it could be interfaces on top of interfaces. That's, that's, um, in your view of this, you really don't know. I mean, it's Gadou's Incompleteness is that you really don't know.

Donald Hoffman 50:55 My own view on it, for what it's worth, but for, 'cause I don't know the right answer. But my own view on it right now is that it, um, it's never-ending. I think that there will never, that this is great, as I said before, great, um, job security for science. And that we, i- if this is true and if, if consciousness is somehow important or fundamental in the universe, this may be an important fundamental fact about consciousness itself that, that it's a never-ending exploration that's going on in some sense.

Lex Fridman 51:27 Well, well, that's interesting. Let me push back on the job security.

Donald Hoffman 51:31 Okay.

Lex Fridman 51:34 So maybe as we understand this kind of idea deeper and deeper, we understand that the pursuit is not a fruitful one. That maybe we need to, maybe that's why we don't see aliens everywhere is you get smarter and smarter and smarter, you realize that like exploration is, uh, there's other fun ways to spend your time than exploring. You could be, um, you could be sort of living maximally in some way that's not exploration. Um, you know, I could... there's all kinds of video games you can construct and put yourself inside of them that don't involve you going outside of the game world. That's, um, you know, feeling, for, for my human perspective, what seems to be fun is challenging yourself and overcoming those challenges. So you can constantly artificially generate challenges for yourself like Sisyphus and his boulder-

Donald Hoffman 52:27 (laughs) Right.

Lex Fridman 52:28 ... just, and, and that's it. So the scientific method that's always reaching out to the stars, that's always trying to figure out the puzzle upon a puzzle, the, the tur- we're always trying to get to the bottom turtle. Um, maybe if we can build more and more the intuition that that's a infinite pursuit, we get, um, we agree to start deviating from that pursuit and start enjoying the, the here and now versus the-

Donald Hoffman 52:54 Hmm.

Lex Fridman 52:54 ... looking out into the unknown always. Maybe that's a looking out into the unknown is a, um, early, uh, activity for a species (laughs) -

Donald Hoffman 53:06 (laughs)

Lex Fridman 53:06 ... that's evolved. I'm just sort of saying, uh, pushing back 'cause you probably got a lot of scientists e- excited in terms of job security. I could, I could envision where it's not job security, where scientists become more and more useless. Uh, maybe they're like the holders of the ancient wisdom, uh, that's, that allows us to study our own history, but n- not much more than that. Just to- (laughs)

Donald Hoffman 53:33 Well, well, it's, that's-

Lex Fridman 53:33 ... fun pushback.

Donald Hoffman 53:34 That's good pushback. I, I'll, I'll put one in there for the scientists again.

Lex Fridman 53:38 Yes. (laughs)

Donald Hoffman 53:39 Uh, but, but, but sure. But then I'll take the other s- side too. So when, uh, Faraday did all of his experiments, right? With magnets and electricity and so forth, it came with all this wonderful empirical data. And James Clerk Maxwell looked at it and wrote down a few equations, which we can now write down in a single equation, the Maxwell equation if we use geometric algebra, just one equation. That opened up unbelievable technologies where, you know, people are Zooming and talking to each other around the world. Um, the whole electronics industry. There was something that transformed our lives in a very positive way with the theories beyond space-time. Here's one potential. Right now, most of the galaxies that we see, um, we can see them, but we know that we could never get to them no matter how fast we traveled. They're going away from us, uh, uh, the speed of light or beyond. So we can't, we can't ever get to them. So there's all this beautiful real estate that's just smiling and waving at us and we can never get to it.

Lex Fridman 54:44 (laughs) Yeah.

Donald Hoffman 54:44 But that's if we go through space-time. But if we recognize that space-time is just a data structure, it's not fundamental. We're not little things inside space-time. Space-time is a little data structure in our perceptions.

Lex Fridman 54:59 Mm-hmm.

Donald Hoffman 55:00 It's just the other way around. Once we understand that and, and we get equations for the stuff that's beyond space-time-... maybe we won't have to go through space-time, maybe we can go around it. Maybe I can go to Proxima Centauri and not go through space, I can just go right there directly. It's a data structure, we can start to play with it. So, so I think that my... for what it's worth, my take would be that, that the endless sequence of theories that we could contemplate building will lead to an endless sequence of new remarkable insights into the potentialities, the possibilities-

Lex Fridman 55:41 Mm-hmm.

Donald Hoffman 55:41 ... that would, that would, that would seem miraculous to us, and that we will be motivated to continue the exploration partly, um, just for the technological innovations that, that come out. But you're... the other thing that you mentioned though, what about just being? What if we, what if we decide t- instead of all this doing and exploring, what about being? My guess is that the best scientists will do both, and that the act of being will be a place where they get many of their ideas, and that they then pull into the conceptual realm. And I think many of the best scientists, you th- like Einstein comes to mind, right? Where these guys say, "Look, I didn't come up with these ideas by a conceptual analysis. I was thinking in vague images and I was... it was just something non-conceptual, and then it took me a long, long time to pull it out into concepts, and then longer to put it into math. But the real insights didn't come from just slavishly wri- you know, playing with equations. They came from a, a deeper place." And so the- there, there may be this going back and forth between the complete non-conceptual where there's e- essentially no end to the wisdom, and then conceptual systems where there's the girdle limits, um, that we have to that. And that may be if, if consciousness is, is important and fundamental, that may be what consciousness, at least part of what consciousness is about, is this discovering itself, discovering its possibilities so to speak. We can talk about what that might mean. Um, by going from the non-conceptual to the conceptual and back, back and forth.

Lex Fridman 57:23 To get better and better and better at being.

Donald Hoffman 57:26 Right.

Lex Fridman 57:27 Let me ask you, uh, ju- just to linger on the evolutionary...

Lex Fridman 57:30 c- because you mentioned evolutionary game theory, and that's really where you... the perspective from which you come, uh, uh, to form the case against reality.

Donald Hoffman 57:40 Right.

Lex Fridman 57:40 Uh, at which point in our evolutionary history did we start to deviate the most from reality? Is it, uh, is it way before life even originated on Earth? Is it, um, in the early development from bacteria and so on? Or is it when some inklings of what we think of as intelligence or maybe even, uh, complex consciousness, uh, started to emerge? So where did this deviation... Um, just like with the interfaces on, in a computer, you know, you start with transistors and then you have, uh, assembly, and then you have, uh, C, C++, then you have Python, then you have GUIs, all that k-... you have layers upon layers. When did we start to deviate?

Donald Hoffman 58:33 Well, David Marr, again my advisor at MIT, in his book Vision, suggested that the more primitive sensory systems were less realistic, less veridical. But that by the time you got to something as complicated as the humans, we were actually tr- estimating the true shapes and distances to objects and so forth. So, so his point of view, and I think it was probably... a- is not an un- uncommon view among my colleagues, that, that, yeah, the sensory systems of lower creatures may just not be complicated enough to give them much, much truth. Um, but as you get t- you know, to 86 billion neurons, you can now compute the truth, or at least the parts of the truth that we need. When w- when I look at evolutionary game theory, um, one of my graduate students, Justin Mauch, did some simulations using genetic algorithms. So there he was just exploring... Um, we started off with random organisms, random sen-

Lex Fridman 59:33 Wow.

Donald Hoffman 59:33 ... sensory genetics and random actions, and first generation was unbelievably... there were... it was a foraging situation, they were foraging for resources, and most of them s- set, you know, stayed in one place, didn't do anything important, and, and... but we could then just look at how the genes evolved, and, and what we found was... what, what, what, what he found was that, uh, basically you never even saw the, the, the truth organisms even come on the stage. They, they, they, they... if they came, th- they were gone in one generation. They just, they just weren't... so they, they, they came and gone, th- they came and went, uh, even just in one generation. They, they just are not good enough. The ones that were just tracking... their senses just were tracking the fitness payoffs, were, were far more, um, fit than, than, um, the truth seekers. So from a... so an answer at, at one level... now I'm gonna give an answer at a deeper level, but just with evolutionary game theory, because my attitude as a scientist is, um, I don't believe any of our theories. I take them very, very seriously, I study them, I look at their implications, but none of them are the gospel. They're just the latest ideas that we have. And, you know... so the reason I study evolutionary game theory is because that's the best tool we have right now in this area. There's, there's, there is n- nothing else that competes. And so as a scientist it's my responsibility to take the best tools and see what they mean. And the same thing the physicists are doing. They're, they're taking the best tools and looking at what, what they entail. But I don't... I, I think that science now has enough experience to realize that we should not...... believe our theories, i- in the sense that we've now arrived. I- in 1890, it was a lot of physicists thought we'd arrived. They were discouraging, um, bright young students from going into physics because it was all done. And that's precisely the wrong attitude.

Lex Fridman 1:01:31 Yeah.

Donald Hoffman 1:01:31 Forever.

Lex Fridman 1:01:32 (laughs)

Donald Hoffman 1:01:32 It's the wrong attitude forever. It, we, the attitude we should have is, I, a- a century from now they'll be looking at us and laughing at what we didn't know, and we just have to assume that that's going to be the case. A c- just, just know that everything that we think is so brilliant right now, our final theory of everything, a century from now they'll look at us like we look at the physicists of 1890 and go, "How could they have been so dumb?"

Lex Fridman 1:01:55 Yeah.

Donald Hoffman 1:01:56 So, so I don't wanna make that mistake. So, so I'm not doctrinaire about any of our current scientific theories. I am doctrinaire about this. We should use the best tools we have right now.

Lex Fridman 1:02:08 Mm-hmm.

Donald Hoffman 1:02:09 That's what we've got.

Lex Fridman 1:02:09 And with, with humility. Well, l- so let me ask you about game theory. There's, um... I, I love game theory and, uh, evolution game theory, um, but I'm always suspicious of it, um, like economics. Um, when you construct models, it's too easy to construct things, uh, that oversimplify just, um, because we, our human brains enjoy the over- the simplification of constructing a few variables that somehow represent organisms or represent people and running a simulation that then allows you to build up intuition, and then it feels really good because you get, can get some really deep and surprising intuitions. But how do you know, uh, your models aren't... The assumptions underlying your models aren't some fundamentally flawed, and because of that, your conclusions are fundamentally flawed? So, I guess-

Donald Hoffman 1:03:04 Yes.

Lex Fridman 1:03:04 ... my question is, what are the limits in your use of game theory, evolutionary game theory, your experience with it, what are the limits of game theory?

Donald Hoffman 1:03:12 So I've gotten some pushback from professional colleagues and friends who have tried to rerun simulations and tried to... I mean, the, the idea that we don't see the truth is not comfortable, and so many of my colleagues are very interested in trying to show that we're wrong.

Lex Fridman 1:03:26 Mm-hmm.

Donald Hoffman 1:03:26 And so the idea would be to say that somehow we did something, as you're suggesting, maybe something special that wasn't completely general. Um, we've got some little special part of the whole search space in evolutionary game theory in which this happens to be true, but more generally organisms would evolve to see the truth. So the, the best pushback we've gotten is from a team at Yale, and, uh, they suggested that, um, if you use thousands of payoff functions... So we, in, uh, our simulations we just use a couple-

Lex Fridman 1:03:56 Mm-hmm.

Donald Hoffman 1:03:57 ... one or two, 'cause it was through our first simulations, right? So that would be a limit. We had one or two payoff functions. We showed th- the result in those, at least for the genetic algorithms. And they said if you have 20,000 of them, then we can find these conditions in which, um, truth see- seeing organisms would be the ones that, that evolved and, and survived. And so we looked at, at their simulations and, and it, it certainly is the case that you can find special cases in which truth can evolve. So when I say its probability is zero, it doesn't mean it can't happen.

Lex Fridman 1:04:30 Mm-hmm.

Donald Hoffman 1:04:30 It, it can happen. In fact, it could happen infinitely often. It's just probability is zero. So if probability zero things can happen infinitely often.

Lex Fridman 1:04:37 Wha- when you say probability zero, you mean probability close to zero?

Donald Hoffman 1:04:42 To be very, very precise, so for example, if I have a unit square, uh, on the plane, um, and I use a measure in which the, um, well, on... a probability measure in which the area of a region is this probability.

Lex Fridman 1:04:56 Mm-hmm.

Donald Hoffman 1:04:58 Then if I draw a curve in that unit square, it has me- measured precisely zero. L- precisely, not approximately, precisely zero. And yet it has infinitely many points. So there is an object that, for that probability measure, has probability of zero, and yet there is infinitely, infinitely many points on it. So that's what I, what I mean by, when I say that the things that are probability zero can happen infinitely often in principle.

Lex Fridman 1:05:21 Yeah, but infinity as far as I... and I look outside often. I walk around, and I look at people. I have never seen infinity in real life.

Donald Hoffman 1:05:32 That's an interesting i- issue.

Lex Fridman 1:05:35 Because I've been looking. I've been looking. I just, I don't notice it, infinitely small or the infinitely big. And so the tools of mathematics, you could sort of apply the same kind of criticism that it is a very convenient interface into our reality.

Donald Hoffman 1:05:49 Well, that's a big debate in mathematics, the intuitionists versus the ones who take, for example, the real numbers, uh, as, as, as, as real, and that, that's a fun dis- discussion. Nicholas Giesswein has... A physicist had really interesting work recently on how if you go with intuitionist mathematics, uh, y- you could effectively quantize Newton, and you find that, uh, Newtonian theory and, and quantum theory aren't that different once you go with a, uh-

Lex Fridman 1:06:16 (laughs) That's funny.

Donald Hoffman 1:06:16 It's, it's really quite interesting.

Lex Fridman 1:06:18 Yeah.

Donald Hoffman 1:06:18 So, so the, the issue you raise is a very, very deep one and one that I think we should take quite seriously, which is, you know, how should we think about the reality of, you know, contras hierarchy, aleph one, aleph two, and all these, all these different, um, infinities versus, um, just, um, a more algorithmic approach, right? So where it's everything is computable in some sense and everything is finite, as big as you want but, but nevertheless finite, um...

Lex Fridman 1:06:50 So yeah. They bo- ultimately boils down to whether the world is discrete or continuous in some general sense. And again, we can't really know, but there's just, um, a mind-breaking thought, uh, just common sense reasoning that something can happen and is yet probability of it happening is 0%.... that doesn't, that doesn't, uh, compute for common sense computer.

Donald Hoffman 1:07:18 Right. Um, this is where you have to be a, be a, a, a sharp mathematician to really ... And I'm not.

Lex Fridman 1:07:23 Sharp is one word.

Donald Hoffman 1:07:24 (laughs)

Lex Fridman 1:07:25 What I'm saying is, commonsense computers, I, I mean that, uh, in, in a very kind of ... in a positive sense, uh, because we've been, been talking about perception systems and interfaces. If we were ... if we are to reason about the world, we have to use the best interfaces we got. And I'm not exactly sure, uh, that game theory is the best interface we've got for this.

Donald Hoffman 1:07:52 Oh. Right.

Lex Fridman 1:07:53 And, and, uh, applications of mathematics, uh, tricks and tools of mathematics to game theory is the best we got when we are thinking about the nature of reality and fitness functions and evolution, period.

Donald Hoffman 1:08:07 Right. Well, that's a, a f- fair rejoinder, and I think that, um, that, that was the tool that we used. And if s- someone says, "Here's a better mathematical tool, and here's why." This is, this mathematical tool better captures the essence of Darwin's idea. John Maynard Smith didn't quite get it with evolutionary game theory. There's this better ... this thing. Now, there are tools like evolutionary graph theory, which generalize evolutionary game theory, and then there's quantum game theory. So, so you can y- you can use, uh, quantum tools, like entanglement, for example, as, as a resource in games, that, that change the very nature of, of the solutions, of the, the optimal solutions of, of the game theoretic.

Lex Fridman 1:08:52 Well, the, the, the work from Yale is really interesting. It's a really interesting challenge of that kind of ... of these ideas where, okay, if you have a very large number of fitness functions, or let's say you have a nearly infinite number of fitness functions, or a growing number of fitness functions, what, what kind of interesting things, uh, start emerging, uh, emerging if, if you are to be an organism? If to be an organism that adapts means having to deal with an ensemble of fitness functions? (laughs)

Donald Hoffman 1:09:23 R- right. And so we've actually redone some of our own work based on theirs, and this is the, the back and forth that we expect-

Lex Fridman 1:09:31 Right.

Donald Hoffman 1:09:31 ... uh, in science, right? And what we found was that they ... in their simulations, they were assuming that you couldn't carve the world up into objects, that, uh ... And so we said, "Well, let's relax that assumption." Allow organisms to create data structures that we might call objects, and an object would be you take ... you, you would do hierarchical clustering-

Lex Fridman 1:09:51 Mm-hmm.

Donald Hoffman 1:09:51 ... of your fitness payoff functions, the ones that have similar shapes. If, if you have 20,000 of them, maybe these 50 are all very, very similar. So I can take all the perception action, fitness stuff and make that into a data structure, and we'll call that a unit or an object. And a- as soon as we did that, then all of their results went away. It turned out they were the special case, and that the, the organisms that were allowed to only see ... that, that, that were shaped to see only fitness payoffs were the ones that were ... So, so the idea is that objects then ... What are objects from an evolutionary point of view? This bottle, um, we, we thought that when I saw a bottle, uh, it was because I was seeing a true object that existed whether or not it was perceived. Evolutionary theories suggest a different interpretation. I'm seeing a data structure that is encoding a convenient way of looking at various fitness payoffs. I can use this for drinking.

Lex Fridman 1:10:47 Mm-hmm.

Donald Hoffman 1:10:48 I could use it as a weapon, but not a very good one. I could beat someone over the head with it.

Lex Fridman 1:10:52 Yeah.

Donald Hoffman 1:10:52 Um, if my goal is mating, this is pointless.

Lex Fridman 1:10:56 Mm-hmm.

Donald Hoffman 1:10:56 So I'm seeing for ... What I'm coding here is all sorts of actions and the payoffs that I could get. When I pick up an apple, now I'm getting a different set of actions and pay- and, and payoffs for ... When I pick up a rock, I'm getting ... So for every object, what I'm getting is a different set of payoff functions, um, and, and act- rec- with various actions. And so once you a- allow that, then what you find is once, once again that, uh, truth goes extinct, and the organisms that just get an interface are the ones that, that win.

Lex Fridman 1:11:29 But the question, uh ... I'm just sneaking up on ... This is fascinating. W- from where do fitness functions originate? What gives birth to the fitness functions? So if there's a giant black box of ... that just keeps giving you fitness functions, what are we trying to optimize? You said that water, um, has different, um, uses than an apple. So there's these objects. What are we trying to optimize, and w- why is not reality a really good generator of fitness functions?

Donald Hoffman 1:12:05 So each theory makes its own assumptions and says, "Grant me this, and I'll explain that."

Lex Fridman 1:12:09 Mm-hmm.

Donald Hoffman 1:12:09 So evolutionary game theory says, "Grant me fitness payoffs." Right? And, "Grant me strategies with payoffs, and I can write down the matrix for this strategy and rec- with that strategy, these are the payoffs that come up." If you grant me that, then I can start to, to explain a lot of things. Now you can ask for deeper questions, like, okay, s- "How does physics evolve biology, and where do these fitness payoffs-

Lex Fridman 1:12:33 Mm-hmm.

Donald Hoffman 1:12:34 ... come from?" Right? Now, that would be a ... That, that's a, a completely different enterprise, and, of course, evolutionary game theory then would be not the, the right tool for that. It would have to be a deeper tool that shows where evolutionary game theory comes from. My own take is that there's gonna be a problem in doing that, because space-time isn't fundamental. It's just a user interface. And that the distinction that we make between living and non-living is not a fundamental distinction. It's an artifact of the limits of our interface.... right? So this is a new wrinkle, and this is an important wrinkle. Wha- it's so nice to take space and time as fundamental because if something looks like it's inanimate, it's, it's inanimate, and we can just say it's, it's not living. Now, it's much more complicated. Certain things are obviously living. I'm talking with you. I, uh, uh, there's, I'm obviously interacting with some- something that's alive and, and conscious.

Lex Fridman 1:13:38 I think we've let go of the word "obviously" in this conversation.

Donald Hoffman 1:13:42 (laughs)

Lex Fridman 1:13:42 I think nothing is obvious.

Donald Hoffman 1:13:43 Nothing's obvious, that's right. But when we get down to, you know, to, like, you know, an ant, it's obviously living, but, well, I'll say it, it appears to be living.

Lex Fridman 1:13:52 Yeah.

Donald Hoffman 1:13:52 But when we get down to a virus, now people wonder. And when we get down to, you know, protons, people say it's not living. And, and my attitude is, look, I have a user interface. The interface is there to hide certain aspects of reality and others to, uh, to ex- well, to, i- i- it's an uneven representation, put it that way. Certain things just get completely hidden. Dark matter and dark energy are most of the energy and matter that's out there. Our interface just plain flat-out hides them.

Lex Fridman 1:14:22 Mm-hmm.

Donald Hoffman 1:14:23 The only way we get some hint is because gravitational things are going wrong within our... So, so most things are outside of our, of our interface. The distinction between living and non-living is not fundamental. It's an artifact of our interface. So if... So this is the... If we really, really want to understand where evolution comes from, to answer the question, the deep question you asked, I think the, the right way we're gonna have to do that is to come up with a deeper theory than spacetime in which there may not be the notion of time and show that whatever this dynamics of that deeper theory is... And by the way, I'll talk about how you could have dynamics without time. But the, the dynamics of this deeper theory, when we project it into, uh, is in certain ways, then we do get spacetime, and we get what appears to be evolution by natural selection. So I would love to see evolution by natural selection, nature red in tooth and claw, people fighting and animals fighting for resources and the whole bit come out of a deeper theory in which m- perhaps it's all cooperation. There's no re- no limited resources and so forth. But as a result of projection, you get space and time. And as a result of projection, you get nature red in tooth and claw, right?

Lex Fridman 1:15:35 Right.

Donald Hoffman 1:15:35 The appearance of it. But it's all an artifact of the interface.

Lex Fridman 1:15:39 I like this idea that, uh, the line between living and non-living is very important 'cause that's, that's a thing that would emerge before you have evolution, the idea of death. Um, so that seems to be an important component of natural selection if... and if that emerged because that's also, um, you know, asking the question, I guess, that I asked, where do fitness come, uh, fitness functions come from? That's like asking the, the old meaning-of-life question, right? Is that what's the why? Why, why? (laughs) And one of the big underlying whys, okay, you can start with evolution on Earth, but without living, without life and death, without the line between the living and the dead, you don't have evolution. So what if underneath it, there's no such thing as the living and the dead? (laughs) There's no...

Donald Hoffman 1:16:36 (coughs)

Lex Fridman 1:16:36 Uh, like this concept of an organism period. There's a living organism that's defined by a volume in spacetime that somehow interacts, that, uh, over time maintains its integrity somehow, uh, has some kind of history. It has a wall of some kind. The outside world, the environment, and then inside, there's an organism.

Donald Hoffman 1:17:00 Mm-hmm.

Lex Fridman 1:17:00 So you're defining an organism, and also you're defining that organism by the fact that it can move, and it can be... come alive, which you kind of think of as moving, combined with the fact that it's keeping itself separate from the environment. So you can point out that thing is living, and then it can also die. Uh, that seems to be a all very powerful components of spacetime that enable you to have something like natural selection and, and evolution. Um...

Donald Hoffman 1:17:31 Well, and there's a lot of interesting work, some of it, um, by collaborators of Karl Friston and, and others where they-

Lex Fridman 1:17:36 Yes.

Donald Hoffman 1:17:37 ... they have, um, uh, like Bayes net kind of stuff that they build on and the notion of a Markov blanket. So you have some states within this network that are inside the blanket. Then you have the blanket and then the states outside the blanket. And the states inside this Markov blanket are conditionally independent of the states outside the blank- blanket conditioned on the blanket. And what they're, what they're looking at is that the dynamics inside of the states inside the Markov blanket seem to be trying to estimate properties of the outside and, and, and react to them in a way. So it seems like you're doing probabilistic inferences in ways that might be able to, to, to keep you alive. So there's interesting work going on in, in that direction. But, but, but what I'm saying is, is something slightly different, and that is, uh, like, when I, when I look at you, all I see is skin, hair, and eyes, right? That's all I see.

Lex Fridman 1:18:27 Mm-hmm.

Donald Hoffman 1:18:27 But I know that there's a deeper reality. I believe that there's a much deeper reality. There's the whole world of your experiences, your thoughts, your hopes, your dreams. I- in some sense, the face that I see is, is just a symbol that I create, right? And as soon as I look away, the, I, I delete that symbol, but I don't delete you. I don't delete the conscious experience, the, the whole world of your... So I, I'm, I'm only deleting an interface symbol, but that interface symbol is a portal, so to speak. Uh, not a perfect portal, but, uh, a genuine portal into your beliefs, into your conscious experiences, into... That's why we can have a conversation. We, we genuinely... What your consciousness is genuinely affecting mine, and mine is genuinely affecting yours through these icons which, which, uh-... I create on the fly. I, I mean, I create your face when I look, I delete it. I don't create you, your consciousness. That's there all the time. Um, but, but I do... So now when, when I look at a cat, I'm creating something that I still call living and I s- I still think is conscious. When I look at an ant, I create something that I still would call living, but maybe not conscious. When I, when I look at something I call a virus, now I'm not even sure I would call it living. And when I look at a proton, I would say I'm, I don't even think... It, it's n- it's not alive at all. It could be that I'm nevertheless interacting with something that's just as conscious as you. I'm not saying the proton is conscious. The face that I'm creating when I look at you, that face is not conscious. That face is a data structure m- in, in me. That face is n- it's, it's an experience, it's not an experiencer. Sim- similarly, a proton is, is, is something that I create, you know, when I look or do a collision in the Large Hadron Collider or something like that. But what is behind the entity in space time? So the... I've got this space time interface and I just got this entity that I call a proton. What is the reality behind it? Well, the physicists are finding these big, big structures. Amplituhedron, the associahedron, cos- what's behind those? Could be consciousness, y- y- what I'm playing with. In which case, when I'm interacting with a proton, I could be interacting with consciousness. Again, to be very, very clear, 'cause it's easy to, uh, misunder... I'm not saying a proton is conscious. Just like I'm not saying your face is conscious. Your face is a symbol I create and then delete as I look and... So your face is not conscious, but I know that that face in my interface, the, the Lex Fridman face that I create is an interface symbol that's a genuine portal into your consciousness. The portal is, is less clear for a cat, even less clea- clear for an ant. And by the time we get down to a proton, the portal is not clear at all. But that doesn't mean I'm not interacting with consciousness, it just means my interface gave up and there's some, some deeper reality that we have to go after. So, so that... So your question really forces out a, a big part of this whole approach that I'm talking about.

Lex Fridman 1:21:29 So it's this portal to conscious- I wonder why (laughs) you can't... Your portal is not as good to a cat, to the cat's consciousness than it is to human?

Donald Hoffman 1:21:40 (coughs)

Lex Fridman 1:21:40 Does it have to be... Have to do with the fact that you're human and just similar organisms, organisms of similar cal- complexity are able to create portals better to each other? Or is it just, as you get more and more complex you get better and better portals?

Donald Hoffman 1:21:57 Well, let me answer one a- one aspect-

Lex Fridman 1:21:59 Yes.

Donald Hoffman 1:21:59 ... of it that, that I'm more confident about, then I'll speculate on that. Why is, why is it that the portal is so bad with protons? Well... And, and, and, and elementary particles more generally? So quarks, leptons and gluons and so forth.

Lex Fridman 1:22:11 (clears throat)

Donald Hoffman 1:22:12 Well, the reason for that is because those are just symmetries of space time. More technically, they're irreducible representations of the Poincaré group of space time. So they're just literally representations of the data structure of space time that we're using. So that's why they're not very much insightful. They're, they're just almost entirely tied to the data structure itself. There's, there's not much... The- they're telling you only something about the data structure, not behind the data structure. It's only when we get to higher levels that we're starting to, in some sense, build portals to what's behind space time.

Lex Fridman 1:22:52 Sure. Yeah. So there's more and more, um, complexity built on top of the interface of space time with a cat.

Donald Hoffman 1:23:02 So you can actually build a portal, right? (laughs)

Lex Fridman 1:23:03 Yeah. (laughs) Yeah. Uh, d- yeah. R- Right. (sighs) Ah, yeah. I... This interface of face and hair and so on, skin. There's some syncing going on between humans, though where we sync... Like, you're l- you're getting a pretty good representation of the ideas in my head and starting to get a foggy view of, uh, my memories in my head. Even though, you know, this is the first time we're talking, you start to project your own memories. You start to solve, like, a giant hierarchy of puzzles about a human 'cause we're all... Uh, there's a lot of similarities. A lot of it rhymes. So you start to make a lot of inferences and you build up this model of a person. You have a pretty sophisticated model of what's going on, uh, underneath. Again, I just, um... I wonder if it's possible to construct these models about each other and nevertheless be very distant from an underlying reality. The syncing-

Donald Hoffman 1:24:08 Yeah. There's a lot of work on this. So there's some interesting work called signaling games where they, they look at how people can coordinate and, uh, come to communicate. There's some interesting work that was done by some colleagues and friends of mine, um, Louis Nahreins, uh, Natalia Komarova and Kimberley Jamieson, where they, um, were looking at e- evolving color words. So you have a, a circle of colors, you know, this is the color circle. And they wanted to see if they could get people to cooperate in how they carve the color circle up into t- units of words. And so they had a game, theoretic kind of thing, that they'd had people do. And what they found was that when they included... So y- most people are trichromats, you know, three kinds of cone photo receptors. But there are some, lot of men, 7% of men have, are dichromats, they might be missing the, the red cone photo receptor. They found that the dichromats had an outsized influence on the final ways that the whole space of colors was carved up and, and labels attached.

Lex Fridman 1:25:14 Mm-hmm.

Donald Hoffman 1:25:15 You needed to be able to include the dichromats in the conversation, and so they had a bigger influence on how you made the boundaries of the language. And I thought that was a really interesting kind of insight that, that there's going to be, again, a game, perhaps a game where evolutionary or genetic-... algorithm kind of thing that goes on in terms of learning to communicate in ways that, that are useful. And, uh, so, yeah, you can use game theory to, to actually explore that, or signaling games. Um, there's a lot of brilliant work on that. I'm not doing it, but there's work out there.

Lex Fridman 1:25:47 So if it's okay, let us tackle once more, and perhaps several more times after,

Lex Fridman 1:25:53 the big topic of consciousness.

Donald Hoffman 1:25:55 Okay.

Lex Fridman 1:25:57 This, this very beautiful, powerful things that perhaps is the thing that makes us human. What is it? What's the role of consciousness in, um, let's say even just the thing we've been talking about, which is the formation of this interface? Um, any kind of ways you want to kind of start-

Donald Hoffman 1:26:15 Sure.

Lex Fridman 1:26:16 ... uh, trocking, talking about it?

Donald Hoffman 1:26:18 Well, let me s- say first what most of my colleagues say. 99% are, again, assuming that space-time is fundamental, particles in space-time, matter is fundamental. And most are reductionist. And so the standard approach to consciousness is to figure out what complicated systems of matter with the right functional properties could possibly lead to the emergence of consciousness. That's the general idea, right? So maybe you have to have neurons. Maybe only if you have neurons, but that might not be enough. They have to, certain kinds of complexity in their, their organization and their dynamics, certain kind of network abilities, for example. So there's, there are