Joscha Bach: Artificial Consciousness and the Nature of Reality | Lex Fridman Podcast #101

Lex Fridman 0:00 The following is a conversation with Joscha Bach, VP of research at the AI Foundation, with a history of research positions at MIT and Harvard. Joscha is one of the most unique and brilliant people in the artificial intelligence community, exploring the workings of the human mind, intelligence, consciousness, life on Earth, and the possibly simulated fabric of our universe. I can see myself talking to Joscha many times in the future. Quick summary of the ads. Two sponsors, ExpressVPN and Cash App. Please consider supporting the podcast by signing up at expressvpn.com/lexpod and downloading Cash App and using code LEXPODCAST. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube, review it with five stars on Apple Podcasts, support it on Patreon, or simply connect with me on Twitter @LexFridman. Since this comes up more often than I ever would have imagined, I challenge you to try to figure out how to spell my last name without using the letter E. An- and it'll probably be the correct way. (laughs) As usual, I'll do a few minutes of ads now and never any ads in the middle that can break the flow of the conversation. This show is sponsored by ExpressVPN. Get it at expressvpn.com/lexpod to support this podcast and to get an extra three months free on a one-year package. I've been using ExpressVPN for many years. I love it. I think ExpressVPN is the best VPN out there. They told me to say it, but I think it actually happens to be true. It doesn't log your data, it's crazy fast, and it's easy to use. Literally, just one big power on button. Again, for obvious reasons, it's really important that they don't log your data. It works on Linux and everywhere else too. Shout out to my favorite flavor of Linux, Ubuntu Mate 2004. Once again, get it at expressvpn.com/lexpod to support this podcast and to get a extra three months free on a one-year package. This show is presented by Cash App, the number one finance app in the App Store. When you get it, use code LEXPODCAST. Cash App lets you send money to friends, buy Bitcoin, and invest in the stock market with as little as $1. Since Cash App does fractional share trading, let me mention that the order execution algorithm that works behind the scenes to create the abstraction of the fractional orders is an algorithmic marvel. So big props to the Cash App engineers for taking a step up to the next layer of abstraction over the stock market, making trading more accessible for new investors and diversification much easier. So again, if you get Cash App from the App Store or Google Play and use the code LEXPODCAST, you get $10 and Cash App will also donate $10 to FIRST, an organization that is helping advance robotics and STEM education for young people around the world. And now here's my conversation with Joscha Bach.

Lex Fridman 3:14 As you've said, you grew up in a forest in East Germany, just as we were talking about off mic, to parents who are artists. And now I think, at least to me, you've become one of the most unique thinkers in the AI world. So can we try to reverse engineer your mind a little bit? What were the key philosopher's, scientist ideas, maybe even movies or just realizations that had a impact on you when you were growing up that kind of led to the trajectory or were the key sort of crossroads in the trajectory of your intellectual development?

Joscha Bach 3:49 My father, uh, came from a long tradition of architects, a distant, uh, branch of the Bach family, and, um, so basically he was technically a nerd, and nerds need to interface in society with nonstandard ways. Sometimes I define a nerd as somebody who thinks that the purpose of communication is to submit your ideas to peer review, and, um, normal people understand that the primary purpose of communication is to negotiate alignment, and these purposes tend to conflict, which means that nerds have to learn how to interact with society at large.

Lex Fridman 4:27 Who is the reviewer in, uh, the nerd's view of communication?

Joscha Bach 4:32 Everybody who you consider to be a peer. So whatever hapless individual is to, uh, around, well, you would try to make him or her the gift of information. (laughs)

Lex Fridman 4:43 (laughs) Okay. So you're... Now, by the way, my research w- uh, malinformed me. So you're w- architect or artist? Or do you see those two as the same?

Joscha Bach 4:52 Um, uh, so he, he did study architecture, but, um, basically my grandfather made the wrong decision. He married an aristocrat and, uh, was drawn into, into the war, and, uh, he came back after 15 years. So basically, my father, uh, was not, um, parented by a nerd but, but by somebody who tried to tell him what to do and expected him to do what he was told.

Lex Fridman 5:16 Yeah.

Joscha Bach 5:17 And he was unable to. He's unable to, uh, do things if he's not intrinsically motivated. So in some sense, my grandmother broke her son, and her son responded by, um, when he became an architect, to become an artist. So he built Hundertwasser architecture. He built houses without right angles. He built lots of things that didn't work in the more brutalist traditions of Eastern Germany. And so he, uh, bought an old water mill, moved out to the countryside, and did only what he wanted to do, which was art. Eastern Germany was perfect for Boheme because, um, you had complete material safety. Food was heavily subsidized. Healthcare was free. You didn't have to worry about rent or pensions or anything. So-

Lex Fridman 5:56 So it's the socialized communist side of Germany.

Joscha Bach 5:59 Yes. And the other thing is, uh, it was almost impossible not to be in political disagreement with your government, which is very productive for artists. So everything that you do is intrinsically meaningful because it will always touch on the deeper currents of society, of culture and be in conflict with it and tension with it, and you will always have to define yourself with respect to this.

Lex Fridman 6:18 So what impact did your father, this-... outside the bo- uh, outside of the box thinker against the government, against the world artists have (overlapping 00:00:09) an impact on your life.

Joscha Bach 6:26 He was actually not a thinker. He was somebody who only got self-aware to the degree that he needed to make himself functional.

Lex Fridman 6:33 I see.

Joscha Bach 6:33 So in some sense, he's- he was also in- uh, late 1960s and, um, he was in some sense a hippie, so he became a one-person cult. He lived out there in his kingdom. He built big sculpture gardens and, um, in- started many avenues of art and so on, and convinced a, um, woman to live with him. She was also an architect, and she adored him and decided to share her life with him. And I basically grew up in a big cave full of books. I'm almost feral. And I was bored out there. It was very, very beautiful, very quiet, and quite lonely. So I started to read. And by the time I came to school, I've read everything until fourth grade and then some. And there was not a real way for me to relate to the outside world. And I couldn't quite put my finger on why, and today I know it was because I was a nerd obviously, and I was the only nerd around so there w- was- were no other kids like me. And there was nobody interested in physics or computing or mathematics and so on. And, um, this village school that I went to, it was basically a nice school, um, kids were nice to me, I was not beaten up, but I also didn't make many friends or build deep relationships. That only happened in- starting from ninth grade when I went to- to a school for mathematics and physics.

Lex Fridman 7:49 Do you remember any key books from this moment that like-

Joscha Bach 7:51 Yes, yes, I basically read everything. So-

Lex Fridman 7:53 (laughs)

Joscha Bach 7:53 ... uh, I went to the library and I worked my way through the, uh, children's and young adult sections, and then I read a lot of, uh, science fiction, for instance, Stanislaw Lem, um, basically the great author of Cybernetics, has influenced me. Back then, I didn't see him as a big influence because everything that he wrote seemed to be so natural to me. It's only later that I contrasted it with what other people wrote. Uh, another thing that was very influential on me were the classical philosophers and, um, also the of romanticism, so German poetry and art, um, Drost-Hülshoff and Heine and, um, up to Hesse and so on. And-

Lex Fridman 8:30 Hesse, I love Hesse. So at which point do the classical philosophers end? At this point, we're in the 21st century, what's- what's the latest classical philosopher? Does this stretch through even as far as Nietzsche, or is this- are we talking about Plato and Aristotle and-

Joscha Bach 8:46 I think that Nietzsche is the classical equivalent of a shit poster.

Lex Fridman 8:50 (laughs)

Joscha Bach 8:51 So, uh, he's very smart and easy to read-

Lex Fridman 8:54 So he's a- yeah, classical troll. Okay.

Joscha Bach 8:54 ... and, yeah, but he's not so much trolling others, he's trolling himself because he was at odds with the world. Largely his romantic relationships didn't work out, he got angry, and he basically became an nihilist. And, um-

Lex Fridman 9:06 Isn't that a- isn't that a beautiful way to be as an intellectual, is to constantly be trolling yourself, to be in that conflict, in- in that... no? (laughs) In that- in that tension?

Joscha Bach 9:15 No, I think it's a lack of self-awareness. At some point, you have to understand the comedy of your own situation. If you take yourself seriously, uh, and- and you are not functional, it ends in tragedy as it did for Nietzsche.

Lex Fridman 9:26 I think- you think he took himself too seriously in that- in that tension.

Joscha Bach 9:30 Yeah.

Lex Fridman 9:30 So to be able to laugh at yourself-

Joscha Bach 9:30 And as you find the same thing in- in Hesse and so on, this Steppenwolf syndrome is classic adolescence where you basically feel misunderstood by the world and you don't understand that all the misunderstandings are the result of your own lack of self-awareness because you think that you are a prototypical human and the others around you should behave the same way as you expect them based on your innate instincts and it doesn't work out, and you become a transcendentalist to, uh, deal with that. And, uh, so it's very, very understandable and have great sympathies for this, to the degree that I can have sympathy for my own intellectual history. But, uh, you have to grow out of it. (laughs)

Lex Fridman 10:04 So as an int- (laughs) So as an intellectual, a life well lived, a journey well traveled is one where you don't take yourself seriously from your perspective.

Joscha Bach 10:11 No, I think that you are neither serious or not serious, uh, y- yourself because you need to become unimportant as a subject. That is, if you are a philosopher, belief is not a verb.

Lex Fridman 10:24 Yeah.

Joscha Bach 10:24 Uh, you don't do this for the audience and you don't do it for yourself. You have to submit to the things that are possibly true and you have to follow wherever your inquiry leads, but it's not about you. It has nothing to do with you.

Lex Fridman 10:36 So do you think, uh, then

Lex Fridman 10:39 people like Ayn Rand believed sort of an idea of there's objective truth? So do you- what's your sense in the philosophical... w- if you remove yourself as objective from the picture, you think it's possible to actually discover ideas that are true or are we just in a mesh of relative concepts that are neither true nor false, it's just a giant mess?

Joscha Bach 10:58 You cannot define, uh, objective truth without understanding the nature of truth in the first place. So what does the brain mean by saying that discover something is truth? So for instance, uh, a model can be predictive or not predictive.

Lex Fridman 11:10 Right.

Joscha Bach 11:11 Um, then there can be a sense in which a mathematical statement can be true because it's defined as true under certain conditions. So it's basically a particular state that, uh, a variable can have an assembled game. And then, uh, you can have a correspondence between systems and talk about truths, which is again, a type of model correspondence. And there also seems to be a particular kind of ground truths. So for instance, we are confronted with the enormity of something existing at all, right? That's-

Lex Fridman 11:39 Right.

Joscha Bach 11:40 ... stunning when you realize something exists rather than nothing. And this seems to be true, right? There is some abs- absolute truth in- in the fact that something seems to be happening.

Lex Fridman 11:50 Yeah, that sh- that- that to me is a showstopper. I can just think about that idea and be amazed by that idea for the rest of my life and not go any farther 'cause I don't even know the answer to that. Why does anything exist at all?

Joscha Bach 12:01 Well, the easiest answer is existence is the default, right? So this is the lowest number of bits that you would need to encode this.

Lex Fridman 12:07 Whose answer... who provides that?

Joscha Bach 12:08 The simplest answer-

Lex Fridman 12:08 The simplest answer.

Joscha Bach 12:09 ... to this is that existence is the default. So-

Lex Fridman 12:11 What about non-existence? I mean, that seems... (laughs)

Joscha Bach 12:14 Non-existence might not be a meaningful notion in this sense. So in some sense if everything that can exist exists, for something to exist it probably needs to be implementable. The e- and only thing that can be implemented is finite automata, so maybe the whole of existence is a superposition of all finite automata and we are in some region of the fractal that has the properties that it can contain us.

Lex Fridman 12:34 Uh, what does it mean to be a superposition of- of finite auto- so I under- uh...... super position of all pos- like, all possible rules.

Joscha Bach 12:43 Imagine that every automaton is basically an operator-

Lex Fridman 12:45 Yeah.

Joscha Bach 12:46 ... that acts on some substrate and as a result you get emergent patterns.

Lex Fridman 12:50 What's a substrate?

Joscha Bach 12:52 Um-

Lex Fridman 12:52 What's a substrate?

Joscha Bach 12:52 ... I have no idea to know, so it's basic-

Lex Fridman 12:54 What's... Some substrate.

Joscha Bach 12:55 It's some s- something that can, uh, store information.

Lex Fridman 12:58 Something that can store information, there's an automaton op- operator.

Joscha Bach 13:00 Something that can hold state.

Lex Fridman 13:01 Still, doesn't make sense to me the why that exists at all. I could just sit there with a, with a beer or a, or a vodka and just enjoy the fact, pondering the why.

Joscha Bach 13:11 May not have a why. The- this might be the- the wrong direction to ask-

Lex Fridman 13:15 It is.

Joscha Bach 13:15 ... into this. So there could be no relation in- in the why direction without asking for a purpose or for a cause. It doesn't mean that everything has to have a purpose or a cause, right?

Lex Fridman 13:26 So we mentioned some philosophers in that early... just taking a brief step back into- into that.

Joscha Bach 13:31 Okay. So we asked ourselves when did classical philosophy end? I think for Germany it largely ended with the first revolution. That's basically when we c-

Lex Fridman 13:39 Which one is that?

Joscha Bach 13:40 Uh, this was, um, when we ended the monarchy and, uh, started a democracy. And at this point, we basically came up with a new form of government that didn't have a good sense of the- this new organism that society wanted to be and in a way it decapitated the universities. So the universities went on through modernism like a headless chicken.

Lex Fridman 13:59 (laughs)

Joscha Bach 14:00 At the same time democracy failed in Germany and we got fascism as a result.

Lex Fridman 14:04 Yeah.

Joscha Bach 14:04 And it burnt down things in a similar way as Stalinism burnt down intellectual traditions in Russia. And, uh, Germany, both Germanies have not recovered from this. Eastern Germany had this vulgar dialectic materialism and Western Germany didn't get much more edgy than Habermas. So in- in some sense, both countries lost their intellectual traditions, and killing off and driving out the Jews didn't help.

Lex Fridman 14:26 (laughs) Yeah, so that was the end, that was the end of really rigorous... what- what you would say is classical, classical philosophy.

Joscha Bach 14:34 There's also this thing that, um, in some sense the low-hanging fruits in philosophy were mostly wrapped. And the last big things that we discovered was the constructivist turn in mathematics, so to understand that the parts of mathematics that work are computation.

Lex Fridman 14:52 Mm-hmm.

Joscha Bach 14:52 It was a very significant discovery in the first half of the 20th century and it hasn't f- fully permeated philosophy and, um, even physics yet. Physicists checked out the code libraries from mathematics before constructivism became universal.

Lex Fridman 15:07 What's constructivism? What, uh, are you referring to Gödel's Incompleteness Theorem?

Joscha Bach 15:11 Yes.

Lex Fridman 15:11 That kind of... those kinds of ideas.

Joscha Bach 15:11 So basically, uh, Gödel himself I think didn't get it yet. Hilbert, uh, could get it.

Lex Fridman 15:16 (laughs)

Joscha Bach 15:16 Hilbert saw that, for instance, a ????? set theoretic experiments in mathematics led into contradictions and he noticed that, uh, with the current semantics, we cannot build a computer in mathematics that runs mathematics without crashing.

Lex Fridman 15:29 Right.

Joscha Bach 15:30 And, uh, Gödel prove... could prove this. And so what Gödel could show is using classical mathematical semantics, you run into contradictions and because Gödel strongly believed in these semantics and... more than in what he could observe and so on, he was shocked. It basically, uh, shook his world to the core because in some sense he felt that the world has to be implemented in classical mathematics.

Lex Fridman 15:50 Yes.

Joscha Bach 15:51 And, um, for Turing it wasn't quite so bad. I think that Turing could see that, uh, the solution is to understand that quasi ma- mathematics was computation all along which means you, uh... for instance, pi in classical mathematics is a value.

Lex Fridman 16:05 Yes.

Joscha Bach 16:05 It's also a function, but it's the same thing and-

Lex Fridman 16:08 Yes.

Joscha Bach 16:08 ... in computation, a function is only a value when you can compute it. And if you cannot compute the last digit of pi, you only have a function. You can plug this function into your local sun, let it run until the sun burns out, this is it, this is the last digit of pi you will know. But it also means there can be no process in the physical universe or on any physically realized computer that depends on having known the last digit of pi.

Lex Fridman 16:31 Yes.

Joscha Bach 16:31 Which means there are parts of physics that are defined in such a way that cannot strictly be true because assuming that this could be true leads into contradictions.

Lex Fridman 16:39 So I think, uh, putting computation at the center of the- the worldview is actually the right way to think about it.

Joscha Bach 16:45 Yes. And Wittgenstein could see it and Wittgenstein basically preempted the logiciest program of AI that Minsky started later, 30 years later. Um, Turing was actually a pupil of Wittgenstein and-

Lex Fridman 16:57 Really?

Joscha Bach 16:57 Yeah.

Lex Fridman 16:57 So I didn't know there was any connection between Turing and Wittgenstein.

Joscha Bach 16:59 Yeah, yeah. And Wittgenstein even canceled some classes when Turing was not present because he thought it was not worth spending the time on-

Lex Fridman 17:05 Oh, interesting.

Joscha Bach 17:07 ... with the others. (laughs)

Lex Fridman 17:07 Interesting.

Joscha Bach 17:07 And if you read the Tractatus, it's a very beautiful book, like basically one thought on 75 pages. It's very non-typical for philosophy because it doesn't have, um, arguments in it and it doesn't have references in it. It's just one thought that he's not intending to convince anybody. He just says it's mostly for people that had the same insight as me. Just spell it out. And this insight is there is a way in which mathematics and philosophy ought to meet. Mathematics tries to understand the domain of all languages by starting with those that are so formalizable that you can prove all the properties of the statements that you make. But the price that you pay is that your language is very, very simple so it's very hard to say something meaningful in mathematics.

Lex Fridman 17:48 Yes.

Joscha Bach 17:49 And it looks complicated to people but it's far less complicated than what our brain is casually doing all the time-

Lex Fridman 17:54 That's right.

Joscha Bach 17:54 ... when it makes sense of reality.

Lex Fridman 17:56 That's right.

Joscha Bach 17:56 And, uh, philosophy is coming from the top so it's mostly starting from natural languages, with vaguely defined concepts and the hope is that mathematics and philosophy can meet at some point. And Wittgenstein was trying to make them meet and he already understood that for instance you could express everything with the NAND calculus, that you could reduce the entire logic to NAND gates as we do on our modern computers. So in some sense he already understood Turing universality before Turing spelled it out. I think he... when he wrote the Tractatus, he didn't understand yet that the idea was so important and significant and I suspect then when Turing wrote it out nobody cared that much. Ur- Turing was not that famous, uh, when he lived. It was mostly his work in decrypting, uh, the German codes that made him famous and or gave him some notoriety but this saint status that he has to computer science right now and AI is something that I think he acquired later.

Lex Fridman 18:47 That's kind of interesting. Do you think of computation and computer science, and you kind of represent that to me as maybe that's the modern day... you in a sense are the new philosopher by sort of the computer scientist-... who dares to ask the bigger questions that philosophy originally started is the new philos- is the new philosopher.

Joscha Bach 19:07 Certainly not me, I think. I'm mostly still this child that g- grows up in a very beautiful valley and looks at the world from the outside and tries to understand what's going on. And my teachers tell me things, and they largely don't make sense.

Lex Fridman 19:19 Right.

Joscha Bach 19:19 So, I have to make my own models. So, I have to f- discover the foundations of what the others are saying. I have to try to fix them, have to be charitable. I try to understand what they must have, uh, thought originally or what their teachers or their teachers' teachers must have thought until everything got lost in translation, and how to make sense of the reality that we are in. And whenever I have an original idea, I'm usually late to the party by, say, 400 years. And the only thing that's good is that the parties get smaller and smaller the older I get and the more I explore. (laughs)

Lex Fridman 19:47 (laughs) The par- the parties get smaller?

Joscha Bach 19:49 And more exclusive. (laughs)

Lex Fridman 19:50 And more exclusive.

Joscha Bach 19:51 (laughs)

Lex Fridman 19:51 So, uh, I- it seems like one of the key qualities of your upbringing was that you were not tethered, whether it's because your parents or in general, maybe your, something within your, within your mind, some genetic m- material, that you were not tethered to the ideas of the general populous.

Joscha Bach 20:09 Mm-hmm.

Lex Fridman 20:09 Which is actually a unique property. We're kind of, uh, thr- you know, the education system and whatever from... Not education system, just existing in this world forces certain sets of ideas onto you. Can you, uh, disentangle that? Why were you, why are you not so tethered? Even in your work today, you seem to not care about perhaps, uh, a best paper in Europe's, right? Being tethered to particular, uh, things that current, today, in this year, people seem to value as a thing you put on your CV and resume. You're a little bit more outside of that world, outside of the world of ideas that people are especially focused in, the benchmarks of today, the things. What's, can you disentangle that? 'Cause I think that's inspiring, and if there were more people like that, we might be able to solve some of the bigger problems that sort of AI dreams to solve.

Joscha Bach 21:06 Um, there's a big danger in this because in a way, you are expected to marry into an intellectual tradition and within this tradition, into a particular school. If everybody comes up with their own paradigms, the whole thing is not communicative as an enterprise.

Lex Fridman 21:20 Right.

Joscha Bach 21:20 Right? So, in some sense, you need a healthy balance. You need paradigmatic thinkers and you need people that work within given paradigms. Basically, scientists today define themselves largely by methods, and it's almost a disease that we think as a scientist that somebody who was, uh, con- convinced by their guidance counselor that they should, uh, join a particular discipline, and then they find a good mentor-

Lex Fridman 21:41 Yes.

Joscha Bach 21:41 ... to learn the right methods, and then they're lucky enough, uh, and privileged enough to join the right team, and then they will, their name will show up on influential papers. But we also see that there are diminishing returns with this approach. And when our field, computer science, uh, and AI started, uh, most of the people that joined this field had interesting opinions. And today's thinkers in AI either don't have interesting opinions at all or these opinions are inconsequential for what they're actually doing, because what they're doing is they apply the state-of-the-art methods with a small epsilon. And, uh-

Lex Fridman 22:15 I see.

Joscha Bach 22:15 This is, um, often a good idea if, if you f- think that this is the best way to make progress. And for me, it's, uh, first of all, very boring. If somebody else can do it, why should I do it? Right?

Lex Fridman 22:26 Yes.

Joscha Bach 22:26 So, uh, if the, if the current methods of machine learning lead to s- uh, strong AI, why should I be doing it, right? I will just wait until they're done and wait until they do this on the beach or read interesting books or write some and have fun. But, uh, if you don't think that we are currently doing the right thing-

Lex Fridman 22:43 Mm-hmm.

Joscha Bach 22:43 ... if we are missing, um, some perspectives, then it's required to think outside of the box. It's also required to understand the boxes. But it's, it's necessary to understand what worked and what didn't work and for what reasons. So, you have to be willing to ask new questions and design new methods whenever you want to answer them, and you have to be willing to dismiss the existing methods if you think that they're not going to yield the right answers. It's very bad career advice to do that.

Lex Fridman 23:14 So, maybe to briefly stay for one more time in the, the early days, when would you say for you was the dream, before we dive into the discussions, uh, that we just almost started, uh, when was the dream to understand or maybe to create human level intelligence born for you?

Joscha Bach 23:35 I think that you, uh, can see AI largely today as, um, advanced, um, information processing. If you would change the acronym of AI into that, most people in the field would be happy.

Lex Fridman 23:48 Yeah.

Joscha Bach 23:48 It would not change anything what they're doing. We're automating statistics and, uh, many of the statistical models are more advanced than what statisticians had in the past, and it's pretty good work. It's very productive. And the, the other aspect of AI is, is philosophical project, and this philosophical project is very risky and, uh, very few people work on it, and it's not clear if it succeeds.

Lex Fridman 24:10 So, first of all, let's, this is, you, you, you keep throwing, uh, sort of a lot of really interesting ideas and I have to pick which ones we go with.

Joscha Bach 24:17 (laughs)

Lex Fridman 24:17 But, uh, sort of, uh, first of all, you use the term information processing, just information processing, as if it's, um, it's the mere, it's the muck of existence, as if it's the epitome of, of ex- like, the, that, that the entirety of the universe might be information processing-

Joscha Bach 24:37 Yes.

Lex Fridman 24:37 ... that consciousness and intelligence might be an information problem. So, that, maybe you can comment on if that's, if the advanced information processing is, uh, is a limiting kind of, uh, realm of ideas. And then the other one is, what do you mean by the philosophical project?

Joscha Bach 24:52 So, um, I suspect that general intelligence is the...... result of trying to solve general problems. So, uh, intelligence, I think, is the ability to model. It's not necessarily goal-directed rationality or something. Many intelligent people are bad at this. But, uh, it's the ability to be presented with a number of patterns and see a structure in those patterns and be able to predict the next set of patterns, right? To, to make sense of things. And, uh, some problems are very general. Usually intelligence serves control, so you make these models for a particular purpose of interacting as an agent with the world and getting certain results. But it's, um, the intelligence itself is, in a sense, instrumental to something. But by itself, it's just the ability to make models. And some of the problems are so general that the system that makes them needs to understand what itself is and how it relates to the environment. So as a child, for instance, you notice you do certain things despite y- you perceiving yourself as wanting different things. So you become aware of your own psychology. You become aware of the fact that you have complex structure in yourself and you need to, uh, model yourself to reverse engineer yourself to be able to predict how you will react to certain situations and how you deal with yourself in relationship to your environment. And this process, if this project, if you reverse engineer yourself and your relationship to reality and the nature of the universe that can contain you, if you go all the way, this is basically the project of AI. Or you could say the project of AI is a very important component in it. The true Turing test, in a way, is you ask a system what is intelligence. If that s- system is able to explain what it is, uh, how it works, then you wou- should, uh, assign it the property of being intelligent in this general sense. So the test that Turing was administering, in a way, I don't think that, uh, he couldn't see it, but he didn't express it yet in the original 1950 paper, is that, uh, he was trying to find out a- whether he was generally intelligent. Because in order to-

Lex Fridman 26:49 Yeah. (laughs)

Joscha Bach 26:49 ... take this test, the wrap is, of course, you need to be able to understand what that system is saying. And we don't yet know if we can build an AI. We don't yet know if we are generally intelligent. Basically, you win the Turing test by building an AI. (laughs)

Lex Fridman 27:01 Yes. So it, so in a sense, hidden within the Turing test is a kind of recursive test.

Joscha Bach 27:07 Yes, it's a test on us.

Lex Fridman 27:09 Yeah.

Joscha Bach 27:09 The Turing test is basically a test of the conjecture whether people are intelligent enough to understand themselves.

Lex Fridman 27:16 Okay. But you also mentioned a little bit of a self-awareness, so... And in the project of AI, do you think this kind of emergent self-awareness is one of the fundamental aspects of intelligence? So as opposed to goal-oriented, as you said, kind of puzzle-solving, is coming to grips with the idea that you're an agent in the world, and like how you've-

Joscha Bach 27:39 I find that many highly intelligent people are not very self-aware. Right? So, uh, self-awareness and intelligence are not the same thing. And you can also be self-aware if you have good priors especially-

Lex Fridman 27:50 Yeah.

Joscha Bach 27:50 ... without being especially intelligent. So you don't need to be very good at solving puzzles if the system that you are already implements the solution.

Lex Fridman 27:59 But I, I do find intelligence... So can you kind of... You mentioned children, right?

Joscha Bach 28:03 Mm-hmm.

Lex Fridman 28:03 Uh, i- is that the fundamental project of a- AI, is to create the learning system that's able to exist in the world? So you kinda drew a difference between self-awareness and, uh, intelligence.

Joscha Bach 28:18 Mm-hmm.

Lex Fridman 28:18 And yet you said that the self-awareness seems to be important for children.

Joscha Bach 28:23 Uh, so I call this ability to make sense of the world and your own place in it, so to under- able, make you able to understand what you are doing in this world sentience. And I would distinguish sentience from intelligence-

Lex Fridman 28:35 Okay.

Joscha Bach 28:35 ... because sentience is the, uh, possessing certain classes of models. And intelligence is the way to get to these models if you don't already have them.

Lex Fridman 28:44 I see. So (sighs) can you, can you maybe pause a bit and try to, uh, answer the question that we just said we may not be able to answer? And it might be a recursive meta question of what is intelligence?

Joscha Bach 29:00 Uh, I think that intelligence is the ability to make models.

Lex Fridman 29:04 So models. It's, I think it's useful as examples, very popular now, neural networks form representations of a large-scale dataset. They, they, they form models of those datasets.

Joscha Bach 29:20 Mm-hmm.

Lex Fridman 29:20 When you say models and look at today's neural networks, what are the difference of how you're thinking about what is intelligent in saying that intelligence is the process of making models?

Joscha Bach 29:31 There are two aspects to, to this question. One is the representation. Is the representation adequate for the domain that we want to represent? And the other one is, is the type of the model that you arrive at adequate? So basically-

Lex Fridman 29:45 Mm-hmm.

Joscha Bach 29:45 ... are you modeling the correct domain? And I think in both of these cases, modern AI is lacking still. And I think that I'm not saying anything new here. I'm not criticizing the field. Most of the people that design our paradigms are aware of that. And so one aspect that we are missing is unified learning. When we learn, we, uh, at some point discover that everything that we sense is part of the same object, which means we learn it all into one model. And we call this model the universe. So experience of the world that we are embedded on is not a secret direct via to physical reality. Physical reality is a weird quantum graph that we can never experience or get access to. But it has these properties that it can create certain patterns that our systemic interface to the world. And we make sense of these patterns, and the relationship between the patterns that we discover is what we call the physical universe. So at some point in our development as a, a nervous system, we discover that everything that we relate to in, uh, in the world can be mapped to a region in the same three-dimensional space, by and large. We now know in physics that this is not quite true. The world is not actually three-dimensional, but the world that we are entangled with, at the level which we are entangled with, is largely a flat three-dimensional space.... and so this is the model that our brain is intuitively making. And this is, I think, what, uh, gave rise to this intuition of res extensa, of this material world, this material domain. It's one of the mental domains, but it's just a class of all models that relate to this environment, this three-dimensional physics engine in which we are embedded.

Lex Fridman 31:19 Physics engine in which we are embedded, I love that.

Joscha Bach 31:22 Right?

Lex Fridman 31:22 Ju- ju- ju- just slowly pause. So, uh, th- uh, the, the, the quantum graph, I think you called, which is the real world which we can never get access to, there's a bunch of questions I wanna sort of disentangle them, but maybe, um, uh, one, one useful one, one of your re- recent talks I looked at, can you just describe the basics? Can you talk about

Lex Fridman 31:45 what is dualism? What is idealism? What is materialism? What is functionalism? And what connects with you most in terms of... because you just mentioned there's a reality we don't have access to.

Joscha Bach 31:54 Mm-hmm.

Lex Fridman 31:55 Okay. What does that even mean? (laughs) And why don't we get access to it? Aren't we part of that reality? Why don't we, why can't we access it?

Joscha Bach 32:03 So, the particular trajectory that mostly exists in the West is the result of our indoctrination by a cult for 2,000 years.

Lex Fridman 32:11 A cult? Which one?

Joscha Bach 32:12 Yes.

Lex Fridman 32:12 Oh, 2,000 years.

Joscha Bach 32:12 The Ca- the Catholic cult mostly.

Lex Fridman 32:13 Yes.

Joscha Bach 32:14 And for better or worse, right? It has created, uh, or defined many of the modes of interaction that we have that hav- has created the society, but it has also, in some sense, um, scarred our rationality. And the intuition that exists, if you would translate, uh, the mythology of the Catholic Church into the modern world is that the world in which you and me interact is something like a multiplayer role-playing adventure.

Lex Fridman 32:41 Yes.

Joscha Bach 32:41 And the money and the objects that we have in this world, this is all not real. Or as, uh, Eastern philosophers would say, it's Maya. It's just, um, stuff that is, appears to be meaningful, and, uh, this embedding in this meaning, and if you believe in it, is Samsara. This-

Lex Fridman 32:56 Yeah.

Joscha Bach 32:57 It's th- uh, basically the identification with the needs of the mundane, secular everyday existence. And, uh, the, uh, Catholics also introduced the notion of higher meaning, the sacred. And this existed before, but eventually the natural shape of God is the platonic form of the civilization that you are part of. It's basically the superorganism that is formed by the individuals as an intentional agent. And, uh, basically the Catholics used a relatively crude mythology to implement software on the minds of people and get the software synchronized to make them work in lockstep-

Lex Fridman 33:30 The software synchronized.

Joscha Bach 33:30 ... to basically get, uh, to get this God, uh, online and-

Lex Fridman 33:34 Yeah.

Joscha Bach 33:34 ... to make it efficient and effective. And, uh, I think G- uh, God technically is just a self that spans multiple brains as opposed to your and my self, which mostly exists just on one brain, right?

Lex Fridman 33:45 Yeah.

Joscha Bach 33:45 And, uh, so in some sense you can construct a self functionally as a function that is implemented by brains that exists across brains. And this is a god with a small G.

Lex Fridman 33:55 That's one of the, if you l- uh, Yuval Harari kind of talking about, th- this is one of the nice features of our brains, it seems to, that we can, uh, all download the same piece of software, like God in this case, and kind of share it ???

Joscha Bach 34:08 Yeah, so mathem- you give everybody a spec and the mathematical constraints, uh, that, uh, are, uh, intrinsic to information processing make sure that given the same spec, you come up with a compatible structure.

Lex Fridman 34:20 Okay. So that's, there's this space of ideas that we all share, and we think that's kind of the mind, but that's separate from... Uh, th- the idea is, fr- from, from Christianity, for, from religion, is that there's a separate thing between the mind and-

Joscha Bach 34:35 There is a real world.

Lex Fridman 34:36 ... there's a real world.

Joscha Bach 34:36 And this real world is, uh, the world in which God exists. God is the coder of the multiplayer adventure, so to speak, and-

Lex Fridman 34:43 Yeah.

Joscha Bach 34:43 ... uh, we are all players in this game. And, um-

Lex Fridman 34:46 And that's dualism. You, you would say-

Joscha Bach 34:48 Yes. But the, the dualist aspect-

Lex Fridman 34:50 ... so dualism is like... yeah.

Joscha Bach 34:50 ... is because the mental realm is, uh, exists in a different implementation than the physical realm. And the mental realm is real. And, uh, a lot of people have this intuition that there is this real room in which you and me talk and speak right now.

Lex Fridman 35:03 Yeah.

Joscha Bach 35:03 Then comes a layer of physics and abstract rules and so on. And then comes another real room where our souls are, and our, uh, true form is in the thing that gives us phenomenal experience. And this is of course a very confused notion that you would get. And it's basically, it's the result of connecting materialism and, uh, idealism in, in the wrong way.

Lex Fridman 35:25 So, okay. I apologize, but I think it's really helpful if we just try to define, uh, try to define terms. Like, what is dualism? What is idealism? What is materialism? For people who don't know.

Joscha Bach 35:35 So the idea of dualism in our, uh, cultural tradition is that there are two substances, a mental substance and a physical substance, and they interact by different rules. And the physical world is basically causally closed and, uh, is built on a low level causal structure. So there's basically a bottom level that is causally closed that's entirely mechanical.

Lex Fridman 35:56 Mm-hmm.

Joscha Bach 35:56 And mechanical in the widest sense. So it's computational. There's basically a physical world in which information flows around, and physics describes the laws of how information flows around in this world.

Lex Fridman 36:06 Would you, would you compare it to like a computer where you have hardware and software?

Joscha Bach 36:10 A computer is a generalization of information flowing around. Basically, th- what Turing discovered that there is uni- universal principle, you can define this universal machine, um, that is able to perform all the computations. So all these machines have the same power. This, this means that you can always define a translation between them as long as they have unlimited memory, um, to be able to perform each other's computations.

Lex Fridman 36:34 So would you then say that materialism is this whole world is just the hardware and idealism is this whole world is just the software? Or how-

Joscha Bach 36:42 Uh, not quite.

Lex Fridman 36:42 We think.

Joscha Bach 36:42 I think that most idealists don't have a notion of software yet, because software also comes down to information processing, right? So, uh, what you notice is the only thing that is real to you and me is this experiential world in which things matter, in which things have taste, in which things have color, phenomenal content and so on. And you realize that-

Lex Fridman 37:00 Oh, there you are bringing up consciousness. Okay. Very good.

Joscha Bach 37:02 And this is distinct from the physical world-

Lex Fridman 37:04 Mm-hmm.

Joscha Bach 37:04 ... in which, uh, things have values, uh, in, only in an abstract sense. And you only look at cold...... patterns moving around, so how does anything feel like something? And this connection between the two things is very puzzling to a lot of people, and of course too many philosophers. So idealism starts out with the notion that mind is primary, materialism, things that matter is primary.

Lex Fridman 37:26 Yes.

Joscha Bach 37:26 And so, uh, for the idealist, uh, the material patterns that we see, uh, play- playing out are part of the dream that the mind is dreaming. And, uh, we exist in a mind on a higher plane of existence, if you want. And for, uh, the materialist, there is only the- this material thing, and that generates some models and, um, we are the result of these models. And in some sense, I don't think that we should understand, if you understand it properly, materialism and, uh, idealism is a dichotomy, but there's two different aspects of the same thing. So the weird thing is we don't exist in the physical world. We do exist inside of a story that the brain tells itself.

Lex Fridman 38:06 O- okay. Uh, let me, uh, let my- my information processing, uh, take th- (laughs) ta- take that in. We don't exist in the physical world, we exist in the narrative...

Joscha Bach 38:18 Basically, a brain cannot feel anything. A neuron cannot feel anything. They're physical things. Physical systems are unable to experience anything. But it would be very useful for the brain or for the organism to know what it would be like to be a person and to feel something.

Lex Fridman 38:31 Yeah.

Joscha Bach 38:31 So the brain creates a simulacrum of such a person that it uses to model the interactions of the persons, the best model of what that brain, this organism, thinks it is in relationship to its environment. So it creates that model. It's a story, a multimedia novel that the brain is continuously writing and updating.

Lex Fridman 38:48 But you also kind of said that... you said that we kind of exist in that, uh, in that story.

Joscha Bach 38:53 In that story, yes.

Lex Fridman 38:53 In that story.

Joscha Bach 38:55 Yeah, ?

Lex Fridman 38:55 What is real in any of this? So like, there's a... again, these terms are... you- you kind of said there's a quantum graph. I mean, what is, what is this whole thing running on then? Is- is the story, uh... and is it completely fundamentally impossible to get access to it? Because isn't the story supposed to... isn't the brain in a- in- in something, in, um, existing in some kind of context?

Joscha Bach 39:24 So what we can identify as computer scientists, we can engineer systems and test our theories this way that, uh, might have the necessary and sufficient properties to produce the phenomena that we are observing, which is there is a self in a virtual world-

Lex Fridman 39:40 Yeah.

Joscha Bach 39:40 ... that is generated in somebody's neocortex who, uh... that is contained in the skull of this primate here. And when I point at this, this indexicality is of course wrong, but I do create something that is likely to, uh, give rise to patterns on your retina that allow you to interpret what I'm saying, right?

Lex Fridman 39:57 Mm-hmm.

Joscha Bach 39:58 But I- uh, we both know that the world that you and me are seeing is not the real physical world. What we are seeing is a virtual reality generated in your brain to explain the patterns on your retina.

Lex Fridman 40:08 How close is it to the real world? That's kind of the- the question. Is it, um... when you have, when you have like people like Donald Hoffman, let's say that like that you're really far away. The- the thing we're seeing, you and I now, that interface we have is very far away from anything. Like, we don't even have anything close like to the sense of what the real world is or is it a very surfaced piece of, uh, architecture?

Joscha Bach 40:32 Imagine you look at the Mandelbrot fractal, right? This famous thing that Bernhard Mandelbrot discovered.

Lex Fridman 40:37 Yes, Mandelbrot, yeah.

Joscha Bach 40:38 If you- uh, you see an overall shape in there, right? But you know the- if you truly understand it, you know it's two lines of code. It's, uh, basically an- a series that is being, uh, tested for complex numbers in- in the, uh, complex number plane for every point. And for those where the series is diverging, uh, you, uh, paint this black.

Lex Fridman 40:59 Yes.

Joscha Bach 40:59 And where it's converging, you don't. And, uh, for- you get the intermediate colors, um, by, um, uh, checking how far it diverges.

Lex Fridman 41:09 Yes.

Joscha Bach 41:09 Right? This is- gives you this shape of this fractal. But imagine you live inside of this fractal and you don't have access to where you are in the fractal or, uh, you have not discovered the generator function even, right? So what you see is all of- all I can see right now is a spiral and this spiral moves a little bit to the right. Is this an accurate model of reality? Yes, it is, right? It is an adeq- adequate description. You know that there is actually no spiral in the Mandelbrot fractal, it only appears to- like this to an observer that is interpreting things as a two dimensional space and then define certain irregularities in there at a certain scale that it currently observes. Because if you zoom in, the spiral might disappear and turn out to be something different at a different resolution, right?

Lex Fridman 41:48 Yes.

Joscha Bach 41:49 So at this level, you have the spiral, and then you discover the spiral moves to the right, at some point it disappears. So you have a singularity. At this point, your model is no longer valid. You cannot predict what happens beyond the singularity. But you can observe again, and you will see it hit another spiral. And at this point, it disappeared. So may- we now have a second order law. And if you make 30 layers of these laws, then you have a description of the world that is similar to the one that we come up with when we describe the reality around us. It's reasonably predictive, it does not cut to the core of it-

Lex Fridman 42:18 So you-

Joscha Bach 42:18 ... does not explain how it's being generated-

Lex Fridman 42:20 Yes.

Joscha Bach 42:20 ... how it actually works.

Lex Fridman 42:21 The-

Joscha Bach 42:21 But it's relatively good to explain the universe that we are entangled with.

Lex Fridman 42:25 But- but you don't think the tools of computer science or the tools of physics could get- could step outside, see the whole drawing, and get at the basic mechanism of how the pattern, the spirals are generated?

Joscha Bach 42:36 Um, imagine you would find yourself embedded into a Mandelbrot fractal and you try to figure out what works and you- you have somehow a Turing machine with enough memory to think. And as a result, you've, uh, come to this idea, it must be some kind of automaton. And maybe you just enumerate all the possible automata until you get to the one that produces your reality.

Lex Fridman 42:56 Yeah.

Joscha Bach 42:56 So you can identify necessary and sufficient condition. For instance, we discover that mathematics itself is the domain of all languages.

Lex Fridman 43:04 Yeah.

Joscha Bach 43:04 And then we see that most of the domains of mathematics that we have discovered are, in some sense, describing the same fractals. This is what category theory is obsessed about, that you can map these different domains to each other. So there are not that many fractals. And some of these have interesting structure and symmetry breaks, and, uh, so you can discover what region of this global fractal you might be embedded in from first principles.

Lex Fridman 43:28 Yes.

Joscha Bach 43:28 But the only way you can get there is from first principles. So basically, your understanding of the universe has to start with automata and then number theory and then spaces and so on.

Lex Fridman 43:37 Yeah, I think, uh, like Stephen Wolfram still dreams that he's a- that he'll be able to arrive at the fundamental rules of the cellular automata or the generalization of which is behind our universe.

Joscha Bach 43:48 Yeah.

Lex Fridman 43:48 It's...You've said, on this topic, you said in a recent conversation that, quote, "Some people think that a simulation can't be conscious and only a physical system can, but they got it completely backward. A physical system cannot be conscious. Only a simulation can be conscious."

Joscha Bach 44:06 Yeah.

Lex Fridman 44:06 "Consciousness is a simulated property of the simulated self." Just- just like you said, the mind is kind of the- the- we'll call it story, narrative. There's a simulation in our- so our mind is essentially a simulation?

Joscha Bach 44:19 Uh, usually, um, I try to use the terminology so that the mind is basically the principles that produce the simulation. It's the software that is implemented by your brain.

Lex Fridman 44:29 Yeah.

Joscha Bach 44:30 And the mind is creating both the universe that we are in and the self, the i- idea of a person that is on the other side of attention and is embedded in this world.

Lex Fridman 44:40 Why is that important, that idea of a self? Uh, why is that an important feature in the simulation?

Joscha Bach 44:46 Uh, it's basically a result of, um, the purpose that the mind has. It's a tool for modeling, right? We are not actually monkeys. We are side effects of the regulation needs of monkeys. And (laughs) what the monkey has to regulate is, uh, the relationship of an organism to an outside world that is, uh, in large part also consisting of other organisms. And as a result, it basically has regulation targets that it, uh, tries to get to. These regulation targets start with priors. They are basically like unconventional reflexes that we are more or less born with, and then we can reverse engineer them to make them more consistent, and then we get more detailed models about how the world works and how to interact with it. And s- uh, so these priors that you commit to are largely target values that our needs would approach, set points. And this deviation to the set point creates some urge, some tension, and we find ourselves living inside of feedback loops, right? The consciousness emerges over dimensions of disagreements with the universe, things where you care, uh, things are not the way they should be, that you need to regulate. And so in some sense, the sense itself is the result of all the identifications that you're having. An identification is a regulation target that you're committing to, it's a dimension that you care about, that you think is important, and this is also what locks you in. If you g- uh, let go of these commitments, of these identifications, you get free. There's nothing that you have to do anymore. And if you let go of all of them, you're completely free and you can enter nirvana because you are done.

Lex Fridman 46:14 (laughs) And actually, this is a good time to pause and say, uh, thank you to, um, sort of a friend of mine, Gustaf Soderström, who introduced me to your work. I wanted to give him a shout-out. Uh, uh, he's a brilliant guy, and I think the AI community is actually quite amazing, and Gustaf is a good representative of that. Y- y- you are as well, so I'm- I'm glad... First of all, I'm glad the internet exists and YouTube exists where I can, uh, watch your talks and then get to your b- book and study your writing and- and think about... Uh, you know, that's- that's amazing. Okay, but, uh, the- you've kind of described me sort of this, uh, emergent phenomena of consciousness from the simulation,

Lex Fridman 46:51 so what about the hard problem of consciousness? The- the- can you just linger on it? Like, w- why does it still feel... Like, I- I understand you're kinda... the self is an important part of the simulation, but why does the simulation feel like something?

Joscha Bach 47:11 So if you look at a book by, say, George R. R. Martin, where the characters have plausible psychology-

Lex Fridman 47:16 Yeah.

Joscha Bach 47:16 ... and they stand on a hill, uh, because they want to conquer the city below the hill, and they're ????? it, and they look at the color of the sky, and they are apprehensive and feel, uh, empowered and all these things, why do they have these emotions? It's because it's written into the story, right? It's written into the story because there's an adequate model of the person that predicts what they're going to do next. And the same thing is hap- uh, uh, too for us. So it's basically a story that our brain is writing. It's not written in words, it's written in, uh, perceptual content, basically multimedia content, and it's a model of what the person would feel if it existed, so it's a virtual person, and you and me happen to be this virtual person. So if this virtual person gets access to the language center and talks about the sky being blue, and this is us.

Lex Fridman 48:02 But hold on a second, do I exist in your simulation? Like-

Joscha Bach 48:06 You d- you do exist in an, uh, almost similar way as me, so there are internal states that I, uh, that are less accessible, uh, for me, um, th- that you have and so on, and your- my model might not be completely adequate. There are also things that I might perceive about you that you don't perceive, but in some sense, uh, both you and me are some puppets, two puppets that enact this, uh, play in my mind, and I identify with one of them because I can control one of the puppet directly, and with the other one, I can create, uh, things in between. So for instance, we can go on an interaction that even leads to a coupling, to a feedback loop, so we can think things together in a certain way or feel things together, but, uh, this coupling is itself not a physical phenomenon, it's entirely a software phenomenon. It's the result of two different implementations interacting with each other.

Lex Fridman 48:55 So, that's interesting. So, are- are you suggesting... I- I- d- like, the way you think about it, is the entirety of existence a simulation and we're kind of... each mind is a little sub-simulation that... Like, why don't you... why doesn't your mind have access to my mind's full state? Like-

Joscha Bach 49:19 For the same reason that my mind doesn't have access to its own full state.

Lex Fridman 49:23 So what, uh, I mean-

Joscha Bach 49:26 There is no trick involved. So basically when I s- know something about myself, it's because I made a model.

Lex Fridman 49:32 Yes, but there's-

Joscha Bach 49:32 So one part of your brain is tasked with modeling what other parts of your brain are doing.

Lex Fridman 49:36 Yes, but there seems to be an incredible consistency about this world in the physical sense.

Joscha Bach 49:42 Mm-hmm.

Lex Fridman 49:42 That there's repeatable experiments and so on.

Joscha Bach 49:44 Yeah.

Lex Fridman 49:44 How does that fit into our silly Descendants of Apes simulation of the world? So why is it so repea- why is everything so repeatable? And, uh, not everything, there- there's a lot of fundamental physics experiments that are repeatable-... for a long time, all over the place, and so on. The laws of physics-

Joscha Bach 50:03 Yeah.

Lex Fridman 50:04 ... how, how does that fit in, do you think?

Joscha Bach 50:05 It seems that the parts of the world that are not deterministic, uh, are not long lived. So if you build a system, any kind of, uh, automaton... So if you build simulations of something, uh, you'll notice that, uh, the phenomena that endure are those that give rise to stable dynamics. So basically, if you see anything that is complex in the world, it's the result of, usually of some control, of some feedback that keeps it stable around certain attractors. And the things that are not stable, that, uh, don't give rise to certain harmonic patterns and so on, they tend to get weeded out over time. So, uh, if we are in a region of the universe that sustains complexity, which is required to implement minds like ours, um, this is going to be a region of the universe that is very tightly controlled and controllable. So it's going to have lots of interesting symmetries and also symmetry breaks that allow to... The creation of structure.

Lex Fridman 51:00 But they exist where? So th- th- there's such an interesting idea that our mind is a simulation that's constructing the narrative. My, my question is l-

Lex Fridman 51:09 just to try to understand how that fits with this, with the entirety of the universe. You're saying that there's a, a region of this universe that allows enough complexity to create creatures like us, but what's the connection between the, the brain, the mind, and the broader universe? Which comes first? Which is more fundamental? Is the, is the mind the starting point and the universe is emergent? Is the universe the starting point and the minds are emergent?

Joscha Bach 51:37 Um, I think quite clearly the latter. It's at least a much easier explanation-

Lex Fridman 51:41 Yeah.

Joscha Bach 51:41 ... because it allows us to make causal models. And I don't see any way to construct an inverse causality. I mean-

Lex Fridman 51:47 So like, what happens when you die to your mind simulation?

Joscha Bach 51:51 Um, my implementation ceases. So basically, the thing that implements myself will no longer be present.

Lex Fridman 51:57 Got it.

Joscha Bach 51:57 Which means if I am not implemented on the minds of other people, the thing that I identify with is... The, the weird thing is I don't actually have an identity beyond the identity that I construct. If I was the Dalai Lama, uh, he identifies as a form of government. So basically, the Dalai Lama gets reborn, not because he's confused, but, uh, because he is, uh, not identifying as a human being. He runs on a human being. He's basically a governmental software-

Lex Fridman 52:25 Right.

Joscha Bach 52:25 ... that is instantiated in every new generation anew. So his advisors will pick someone who does this in the next generation. So if you identify with this, you are no longer human and you don't die in the sense that what dies is only the body of the human that you run on.

Lex Fridman 52:40 Yeah.

Joscha Bach 52:40 Uh, he... To kill the Dalai Lama, you would have to kill his tradition. And, uh, if we look at ourselves, we realize that we are... To a small part like this, most of us. So for instance, if you have children, you realize something lives on-

Lex Fridman 52:52 Yeah.

Joscha Bach 52:52 ... in them or if you spark an idea in the world, something lives on, or if you identify with a society around you. Because you are in part that, you're not just this human being.

Lex Fridman 53:01 Yeah, so in a, in a sense, you are kind of like a Dalai Lama, in the sense that, uh, you, Joshua Bach, is just a collection of ideas. So like, you have this operating system on which a bunch of ideas live and interact, and then once you die they kind of... Part... Some of them, uh, jump off the ship.

Joscha Bach 53:19 You put it, uh, put it the other way. Identity is a software state, it's a construction. It's not physically real. You, you know, identity is not a physical concept. It's basically a representation of different objects on the same world line.

Lex Fridman 53:32 But identity li- lives and dies. Are you attached... This is, it's, uh... What's the fundamental thing? Is it the ideas that come together to form an identity or is each individual identity actually a fundamental thing?

Joscha Bach 53:46 It's a representation that you can get agency over if you care. So basically, you can choose what you identify with if you want to.

Lex Fridman 53:53 No, but it just seems, um, if, if the mind is not r- real, if it's not-

Joscha Bach 53:59 Mm-hmm.

Lex Fridman 53:59 ... that the, the birth and death is not, um, a crucial part of it.

Joscha Bach 54:04 Mm-hmm.

Lex Fridman 54:05 Is... Well, may- maybe I'm s- uh, silly. I'm, maybe I'm s... Attached to this whole biological organism, but it seems that the physical... Being a physical object in this world is, is, um, an important aspect of birth and death. Like, it feels like it has to be physical to die. It feels like simulations don't have to die.

Joscha Bach 54:30 The physics that we experience is not the real physics. There is, for instance-

Lex Fridman 54:33 Right.

Joscha Bach 54:33 ... no color and sound in the real world. Color and sound are types of representations that you get if you want to, uh, model reality with oscillators, right? So colors and sound, in some sense, have octaves-

Lex Fridman 54:44 Yes.

Joscha Bach 54:44 ... and it's because they are represented probably with oscillators, right? So that's why colors form a circle of hues. And, uh, colors have harmonics, sounds have harmonics as a result of synchronizing oscillators in, in the brain, right? So the world that we subjectively interact with is f- fundamentally the result of the representation mechanisms in our brain. They are mathematically, to some degree, universal. There are certain regularities that you can discover in the patterns and not others-

Lex Fridman 55:10 Yes.

Joscha Bach 55:10 ... but the patterns that we get, this is not the real world. The world that we interact with is al- always made of too many parts to count, right? So the, uh, when you look at this table and so on, it's consisting of so many moli- uh, molecules and atoms that you cannot count them. So you only look at the aggregate dynamics, at limit dynamics. If you had almost infinitely many patterns, uh, particles, what would be the dynamics of the table? And this is roughly what you get. So geometry that we are interacting with is the result of discovering those operators that work in the limit that you get by building an infinite series that converges. For those parts where it converges, it's geometry, for those parts where it doesn't converge, it's chaos.

Lex Fridman 55:49 Right, and then so all of that is filtered through k- the con... Sort of the consciousness that's emergent in our narrative. So the, the consciousness gives it color, gives it feeling, gives it flavor.

Joscha Bach 56:00 So, uh, I think the...... feeling, um, flavor and so on is given by the relationship that a feature has to all the other features. It's basically a giant relational graph that is our subjective universe. The color is given by those aspects of the representation, uh, or the- this experiential color where you care about, where you have identifications, where something means something, where you are the inside of a feedback loop and the dimensions of- of caring are basically dimensions of this motivational system that we emerge

Joscha Bach 56:29 over.

Lex Fridman 56:29 The- the meaning of the relations-

Joscha Bach 56:31 Mm-hmm.

Lex Fridman 56:32 ... the graph. Can you elaborate on that a little bit? Like, where does the... Maybe we can even step back and ask the question of what is consciousness to be sort of more systematic. Like, what- what- what do you- how do you think about consciousness? What is consciousness?

Joscha Bach 56:47 I think that consciousness is largely a model of the contents of your attention. It's a mechanism that has evolved for a certain type of learning. At the moment, our machine learning systems largely work by, uh, building chains of weighted sums of real numbers with some, um, non-linearity. And, uh, you will learn by, um, piping an error signal through these different chained layers and adjusting the weights in these weighted sums. And you can approximate mo- most polynums with this, uh, if you have enough training data. But the price is you need to change a lot of these weights. Basically, the error, uh, is piped backwards into the system until it c- accumulates at certain junctures in the network and, uh, everything else evens out statistically. And only at these junctures, this is where you had the actual error in the network and you make the change there. This is a very slow process and our brains don't have enough time for that because we don't get old enough to play Go the way that our machines learn to play Go. So instead, what we do is an attention-based learning. We pinpoint the probable region in the network where we, uh, can make an improvement and then, uh, we store the- this binding state together with the expected outcome in a protocol. And this ability to make indexed memories for the purpose of learning to revisit these, uh, commitments later, this requires a memory of the contents of our attention. Another aspect is when I construct my reality, I make mistakes. So I see things that turn out to be reflections or shadows and so on, which means I have to be able to point out which features of my perception gave rise to the, uh, present construction of reality. So the system needs to pay attention to the, uh, features that are currently in its focus, and it also needs to pay attention to whether it pays attention itself, in part because the attentional system gets trained with the same mechanism, so it's reflexive, but also in part because your attention lapses if you don't pay attention to the attention itself.

Lex Fridman 58:43 (laughs)

Joscha Bach 58:44 Right? So is the thing that I'm currently seeing just a- a dream that my brain has spun off into some kind of daydream or am I still paying attention to my percepts? So you have to periodically go back and see whether you're still paying attention, and if you have this loop and you make it tight enough between the system becoming aware of the contents of its attention and the fact that it's paying attention itself and makes attention the object of its attention, I think this is the loop over which we wake up.

Lex Fridman 59:09 So there's this (laughs) ... so there's this attentional mechanism that's somehow self-referential that's fundamental to what consciousness is.

Joscha Bach 59:17 Mm-hmm.

Lex Fridman 59:17 So just, uh, ask you a question, I don't know how much you're familiar with the recent breakthroughs in natural language processing. They use attentional mechanisms, they use something called transformers to, uh, learn patterns in sentences by allowing the network to focus its attention to particular parts of a sentence at each individual... So, like, parameterize and make it learnable the dynamics of a sentence by having, like, a little window into the- into the sentence. Wh- do you think that's, like, a little step towards... that eventually would- will take us to the attentional mechanisms from which consciousness could emerge?

Joscha Bach 1:00:00 Not quite. I think it models only one aspect of attention. In the early days of automated, uh, language translation, there was a- an example that I found particularly funny where somebody tried to translate a text from English into German, and it was, "A bat broke the window." And, uh, the, uh, translation in German was, "Eine Fledermaus zerbrach das Fenster mit einem Baseballschläger." So to- to translate back into English, "A bat," uh, uh, this flying mammal-

Lex Fridman 1:00:29 Yeah.

Joscha Bach 1:00:29 ... "broke the window with a baseball bat."

Lex Fridman 1:00:31 Yes.

Joscha Bach 1:00:32 And, uh, it seemed to be the most similar to this program because it somehow maximized, uh, the possibility of translating the concept "bat" into German in the same sentence. And this is some- a mistake that the transformer model is not doing because it's tracking identity. And the attentional mechanism in the transformer model is basically putting its finger on individual concepts and make sure that these concepts, um, pop up later in the text-

Lex Fridman 1:00:57 Yeah.

Joscha Bach 1:00:57 ... and, uh, tracks basically the individuals through the text. And it's why, uh, the system can learn things that other systems couldn't before it, which, uh, makes it, for instance, possible to write a text where it talks about the scientist and the scientist has a name and has a pronoun and, uh, it gets a consistent story about that thing. What it does not do, it doesn't fully integrate this, so this meaning falls apart at some point, it loses track of this context. It does not yet understand that everything that it says has to refer to the same universe, and this is where this thing falls apart. But the, uh, attention in a transformer model does not go beyond tracking identity and tracking identity is an important part of attention, but it's a different, very specific attentional mechanism, um, and it's not the one that gives rise to the type of consciousness that we have.

Lex Fridman 1:01:42 Okay. Just to linger on what- what do you mean by "identity" in the context of language?

Joscha Bach 1:01:47 So, uh, when you talk about language, we have different words that can refer to the same concept.

Lex Fridman 1:01:52 Got it.

Joscha Bach 1:01:53 And in the sense that-

Lex Fridman 1:01:53 So space of concepts. So, uh-

Joscha Bach 1:01:55 Yes. And, uh, it can also be, uh, in a nominal sense or, uh, in an lexical sense that you say, uh, "This word does not only refer to this class of objects, but it refers to a definite object, to some kind of agent that weaves their way through- through the story and is only referred by different ways in the language."... so the language is basically a projection from, uh, a conceptual representation from a scene that is evolving into a discrete string of symbols. And, uh, what the transformer is able to do, it learns, um, aspects of this projection mechanism that other models couldn't learn.

Lex Fridman 1:02:32 So, have you ever seen an artificial intelligence or any kind of construction idea that allows for unlike neural networks or perhaps within neural networks that's able to form something where the space of concepts continues to be integrated? So what you're describing, building a knowl- knowledge base, building this consistent larger and larger sets of ideas that would then allow for deeper understanding.

Joscha Bach 1:02:59 Uh, Wittgenstein thought that we can build everything from language, from basically a logical grammatical construct. And I think to some degree this was also what Minsky believed. So that's why he focused so much on common sense reasoning and so on. And, um, a project that was inspired by him was Cyc.

Lex Fridman 1:03:17 Cyc.

Joscha Bach 1:03:17 Um, that was basically-

Lex Fridman 1:03:19 That's still going on.

Joscha Bach 1:03:19 Yes. Of course, uh, ideas don't die, (laughs) only people die. And-

Lex Fridman 1:03:25 That, that's true. But, uh-

Joscha Bach 1:03:27 And also is a productive project, it's just probably not one that is going to, uh, converge to general intelligence. The thing that Wittgenstein couldn't solve, and he looked at this in his, uh, book at the end of his life, Philosophical Investigations, was, uh, the notion of images. So images play an important role in Tractatus. The Tractatus is an attempt to basically turn philosophy into logical probing language, to design a logical language in which you can do actual philosophy that's rich enough for doing this. And the difficulty was to deal with perceptual content. And eventually I think he decided that he was not able to solve it. And I think this preempted the failure of the logicians program in AI. And the solution as we see it today is we need more general function approximation. There are functions, geometric functions, that we learn to approximate that cannot be efficiently expressed and computed in a grammatical language. We can, of course, build automata that go via number theory and so on and, uh, to learn the new algebra and then compute an approximation of this geometry. But, uh, to create language in geometry is not an efficient way to think about it.

Lex Fridman 1:04:32 So functional, uh, well, you kind of just said that neural networks are sort of... the approach that neural networks takes is actually more general than the, than what can be expressed through language.

Joscha Bach 1:04:45 Yes. So what can be e- efficiently expressed through language at the data rates at which we process grammatical language.

Lex Fridman 1:04:52 Okay. So you don't think... so you don't think language is... so you disagree with Wittgenstein that language is not fundamental too?

Joscha Bach 1:04:59 I agree with Whit- Wittgenstein. I just, uh, agree with the late Wittgenstein.

Lex Fridman 1:05:03 (laughs)

Joscha Bach 1:05:03 And, uh, I also agree with the beauty of the early Wittgenstein. I think that the, the Tractatus itself is probably the most beautiful philosophical text that was written in the 20th century.

Lex Fridman 1:05:14 But, uh, but language is not fundamental to cognition and intelligence and consciousness?

Joscha Bach 1:05:18 So, uh, I think that language is a particular way, or the natural language that we're using is a particular level of abstraction that we use to communicate with each other. But, uh, the languages in which we, uh, express geometry are not grammatical languages in the same sense, so they work slightly different, they're more general expressions of functions. And I think the general nature of a model is you have a bunch of parameters, um, these are- have a range because, uh, the variances of the world, and you have relationships between them which are constraints, which say if certain parameters have these values, then other parameters have to have the following values. And, um, this is a very early insight in computer science, and I, I think the- some of the earliest formulations is the Boltzmann machine. And the problem with the Boltzmann machine is that, uh, while it has a measure of whether it's good, it's basically the energy on the system, the amount of tension that you have left in the constraints, where the constraints don't quite match, um, it's very difficult to, despite having this global measure, to train it. Because if you- as, as soon as you add more than trivially few elements, uh, parameters into the system, it's very difficult to get it settled in the right architecture. And so we... uh, the solution that Hinton and Zanichovsky found was, um, to use a restricted Boltzmann machine which uses the hidden links, the internal links in, in the Boltzmann machine, and only has based the input and output layer. But this limits the expressivity of the Boltzmann machine, so now he builds a network of small, of these primitive Boltzmann machines. And in some sense you can see a almost continuous development from this to the deep learning models that we're using today, even though we don't use Boltzmann machines at, at this point. But the idea of the Boltzmann machine is you take this model, you clamp some of the values to perception, and this forces the entire machine to go into a state that is compatible with the states that you currently perceive, and this state is your model of the world. Right? So I, I think it's a very general way of thinking about models. But we have to, uh, use a different approach to make it work. And this is we have to find different networks that train the Boltzmann machine. So the mechanism that trains the Boltzmann machine and the mechanism that makes the Boltzmann machine settle into its state are distinct from the constrained architecture of the Boltzmann machine itself.

Lex Fridman 1:07:33 The, the kind of mechanism that we want to develop, you're saying?

Joscha Bach 1:07:36 Yes. So this, the direction in which I think our research is going to go is going to... for instance, what you notice in perception is our perceptual models of the world are not probabilistic, but possibilistic, which means-

Lex Fridman 1:07:49 What's that mean?

Joscha Bach 1:07:49 ... you should be able to perceive things that are improbable but possible. Right?

Lex Fridman 1:07:54 Interesting.

Joscha Bach 1:07:55 A perceptual state is valid not if it's probable, but if it's possible, if it's co- coherent.

Lex Fridman 1:08:00 Yeah.

Joscha Bach 1:08:00 So if you see a tiger coming after you, you should be able to see this even if it's unlikely.

Lex Fridman 1:08:05 Mm-hmm.

Joscha Bach 1:08:05 And, uh, the probability is necessary for convergence of the model. So given the state of possibilities that is very, very large and a set of perceptual features, how should you change the state of- states of the model to get it to converge with your perception?

Lex Fridman 1:08:21 Yeah.But the space of, uh, the space of ideas that are coherent with the context that you're sensing is perhaps not as large. I mean, that- that's perhaps pretty small.

Joscha Bach 1:08:35 The degree of coherence that you, uh, need to achieve depends, of course, how deep your models go. There is, uh, for instance, politics is very simple when you know very little about game theory and human nature.

Lex Fridman 1:08:45 Yes.

Joscha Bach 1:08:45 So, the younger you are, the more obvious it is how politics should work, right?

Lex Fridman 1:08:49 Yes.

Joscha Bach 1:08:49 And, uh, because you get a coherent aesthetics from relatively few inputs. And, uh, the more layers you model, the more layers you model reality, the harder it- it gets to satisfy all the constraints.

Lex Fridman 1:09:01 So,

Lex Fridman 1:09:02 you know, the current neural networks are a fundamentally supervised learning system with a feedforward neural network, use back-propagation to learn.

Joscha Bach 1:09:10 Mm-hmm.

Lex Fridman 1:09:10 What's your intuition about what kind of mechanisms might we move towards to improve the learning procedure?

Joscha Bach 1:09:18 Uh, I think one big aspect is going to be meta-learning, and architecture search starts in this direction. In some sense, the first wave of AI, classical AI worked by identifying a problem and a possible solution and implementing the solution, right? Program that plays chess. And, um, right now we are in the second wave of AI. So instead of writing the algorithm that, uh, implements the solution, we write an algorithm that automatically searches for an algorithm that implements the solution. So the learning system, in some sense, is an algorithm that itself discovers the algorithm that solves the problem, like Go. Go is too hard to implement it by, the solution by hand, but we can implement an algorithm that finds the solution.

Lex Fridman 1:09:56 Yeah, so that, yeah-

Joscha Bach 1:09:57 So now let's move to the third stage, right? The third stage would be meta-learning, find an algorithm that discovers a learning algorithm for the given domain. Our brain is probably not a learning system, but a meta-learning system. This is one way of looking at what we are doing. There is another way if you look at the way our brain is, for instance, implemented. There is no central control that tells all the neurons how to fire up.

Lex Fridman 1:10:19 Yes.

Joscha Bach 1:10:19 Instead, every neuron is an individual reinforcement learning agent. Every neuron is a single-celled organism that is quite complicated and in some sense quite motivated to get fed.

Lex Fridman 1:10:29 Yeah.

Joscha Bach 1:10:29 And it gets fed if it fires on average at the right time.

Lex Fridman 1:10:32 Yes.

Joscha Bach 1:10:32 And the, um, the right time depends on the context that the, uh, this neuron exists in, which is the electrical and chemical environment that it has. So it basically has to, uh, learn a function over its environment that tells us when to fire to get fed. Or if you see it as a reinforcement learning agent, every neuron is, in some sense, making a hypothesis when it sends a signal and tries to pipe a signal through the universe and tries to get positive feedback for it. And the entire thing is set up in such a way that it's robustly self-organizing into a brain.

Lex Fridman 1:11:04 Mm-hmm.

Joscha Bach 1:11:05 Which means you start out with different neuron types, um, that have different priors in, uh, on which hypothesis to test on how to get its reward. And you put them into different concentrations in a certain spatial alignment, and, uh, then you entrain it in a particular order. And as a result, you get a well-organized brain.

Lex Fridman 1:11:22 Yeah, so, okay, so the brain is a meta-learning system, uh, with a bunch of, uh, with reinforcement learning agents. And what... I think you said, but, uh, just to clarify, where do the lo- there's no centralized, uh, government that tells you, "Here's a loss function, here's a loss function, here's a loss function." Like what, uh, who is, uh, who says what's the objective of what's being optimized?

Joscha Bach 1:11:49 There are also governments which impose loss functions on different parts of the brain. So we have differential attention. Some areas in your brain get specially rewarded when you look at faces.

Lex Fridman 1:11:58 Yeah.

Joscha Bach 1:11:58 If you don't have that, you will get prosopagnosia, which basically me- the inability to tell people apart by their faces.

Lex Fridman 1:12:04 Yeah.

Joscha Bach 1:12:04 So, uh-

Lex Fridman 1:12:06 And the reason that happens is because it was, uh, had an evolutionary advantage. So like evolution comes into play here about...

Joscha Bach 1:12:11 Yeah, but it's basically an extraordinary attention that we have for faces. I don't think that people with prosopagnosia have a, per se, a defective brain. The, uh, the brain just has an average attention for faces. So people with prosopagnosia don't look at faces more than they look at cups. So the level at which they resolve the geometry of faces is not higher than the one that, than for cups. And people that don't have prosopagnosia look, uh, obsessively at faces, right? For you and me, it's impossible to move through a crowd without scanning the faces. And as a result, we make insanely detailed models of faces that allow us to discern mental states of people.

Lex Fridman 1:12:45 So obviously, we don't know 99% of the details of this meta-learning system that's our mind. Okay. H- but still we took a leap from something much dumber-

Joscha Bach 1:12:57 Mm-hmm.

Lex Fridman 1:12:57 ... to that from, uh, through the evolutionary process. Can you, first of all, maybe say h- how hard do you, how big of a leap is that from our brain, from our ape ancestors, to multi-cell organisms? And, um, is there something we can think about, about th- as we start to think about how to engineer intelligence, is there something we can learn from evolution?

Joscha Bach 1:13:25 In some sense, uh, life exists because of the market opportunity of controlled chemical reactions. We compete with dumb chemical reactions, and we win in some areas against this dumb combustion because we can harness those entropy gradients where you need to add a little bit of energy in a specific way to harvest more energy.

Lex Fridman 1:13:43 So we out-competed combustion? (laughs)

Joscha Bach 1:13:45 Yes, in many regions we do. And we try very hard because, uh, when we are in direct competition, we lose, right?

Lex Fridman 1:13:51 Yeah.

Joscha Bach 1:13:51 So, uh, because, uh, the combustion is going to close the entropy gradients much faster than we can run.

Lex Fridman 1:13:56 Yes, you got it. That's abso- that's, that's quite a ????? notion, yeah.

Joscha Bach 1:14:00 So, uh, yeah, so basically we do this because every cell has a Turing machine built into it. It's like literally read/write head on a tape. (laughs)

Lex Fridman 1:14:07 Yes, yeah. (laughs)

Joscha Bach 1:14:08 And, uh, so everything that's more complicated than a molecule that just, uh, is, um, a vortex around attractors, uh, th- that needs a Turing machine in it for its regulation. And, uh, then you bind cells together and you get next level organization, an organism where the cells together implement some kind of software.... and, uh, for me, a very interesting discovery in the last year was the word spirit because I realized that what spirit actually means, it's an operating system for an autonomous robot. And when the word was invented, people needed this word, but they didn't have, uh, robots that they built themselves yet. The only autonomous robots that were known were people, animals, plants, ecosystems, cities, and so on, and they all had spirits. And it makes sense to say that the plant has an operating system, right? If you pinch the plant in one area, then it's going to have repercussions throughout the plant. Everything in the plant is in some sense connected into some global aesthetics, like in other organisms. An organism is not a collection of cells, it's a function that, uh, tells cells how to behave. And this function is not implemented as some kind of supernatural thing, like some morphogenetic field, it is an emergent result of the interactions of the, each cell with each other cell, right?

Lex Fridman 1:15:21 Oh, got it. So, you're, what you're saying is the organism is a function that tells what's, what, what, uh-

Joscha Bach 1:15:29 Tells the cells what to do.

Lex Fridman 1:15:30 ... tells what to do, and the function ar- is emer- emerges from the interaction of the cells.

Joscha Bach 1:15:37 Yes. So it's basically a description of what the plant is doing in terms of microstates.

Lex Fridman 1:15:43 Yeah.

Joscha Bach 1:15:44 And, uh, the microstates, the physical implementation, uh, are too many of them to describe them, so the, uh, software that we use to describe what the plant is doing, this spirit of the plant is the software, the operating system of the plant, right? Uh, this is a way in which we, the observers, make sense of the plant.

Lex Fridman 1:16:01 Yes. Uh, can you, can-

Joscha Bach 1:16:01 And the same is true for people. So people have spirits, which is their operating system in a way, right? And there is aspects of that operating system that relate to how your body functions, and others how you socially interact, how you interact with yourself and so on. And, uh, we make models of that spirit, and we think it's a loaded term because it's from a pre-scientific age, but we, it took the scientific age a long time to rediscover a term that is pretty much the same thing. And I suspect that the differences that we still see between the old word and the new word are translation errors that-

Lex Fridman 1:16:33 (laughs)

Joscha Bach 1:16:33 ... traveled over the centuries.

Lex Fridman 1:16:35 Uh, can you, can you actually linger on that? Like, w- why do you say that spirit... Just to clarify because I'm a little bit confused. So, th- the, the, the word spirit is a powerful thing, but why did you say in the last year or so that you discovered this? Uh, uh, do you mean the same old traditional idea of a spirit? Or do you mean it's-

Joscha Bach 1:16:51 I tried to find out what people mean by spirit. When people say spirituality in the US, it usually is the, refers to the phantom limb that they develop in the absence of culture. And a culture is, in some sense, you could say the spirit of a society that is long game. This thing that, uh, is, becomes self-aware at a level b- above the individuals where you say, "If you don't do the following things, then the grand, grand, grand grandchildren of our children will now have nothing to eat."

Lex Fridman 1:17:18 Yes.

Joscha Bach 1:17:19 So, uh, if you take this long scope where you try to maximize the lengths of the game that you are playing as a species to realize that you're part of a larger thing that you cannot fully control, you probably just need to submit to the, uh, ecosphere instead of trying to completely control it, right? There needs to be a- a certain level at which we can exist as a species if we want to endure. And our culture is not sustaining this anymore. We basically made this bet with the Industrial Revolution that we can control everything, and the modernist societies with basically unfettered growth led to a situation, uh, in which we depend on the ability to control the entire planet. And, uh, since we are not able to do that, as it seems, um, this culture will die.

Lex Fridman 1:18:02 (laughs)

Joscha Bach 1:18:02 And we, we realize that it doesn't have a future, right? We called our children Generation Z. (laughs) It's not very optimistic thing to do.

Lex Fridman 1:18:09 (laughs)

Lex Fridman 1:18:10 Yeah, uh, so y- you can have this kind of intuition that our civilization, you said culture, but you really mean the s- the spirit of the civilization, the, uh, the entirety of the civilization may not exist for long.

Joscha Bach 1:18:26 Yeah. So-

Lex Fridman 1:18:27 What's your... Can you, can you untangle that? What's your intuition behind