#75 – Marcus Hutter: Universal Artificial Intelligence, AIXI, and AGI

Marcus Hutter is a senior research scientist at DeepMind and professor at Australian National University. Throughout his career of research, including with Jürgen Schmidhuber and Shane Legg, he has proposed a lot of interesting ideas in and around the field of artificial general intelligence, including the development of the AIXI model which is a mathematical approach to AGI that incorporates ideas of Kolmogorov complexity, Solomonoff induction, and reinforcement learning. EPISODE LINKS: Hutter Prize: http://prize.hutter1.net Marcus web: http://www.hutter1.net Books mentioned: – Universal AI: https://amzn.to/2waIAuw – AI: A Modern Approach: https://amzn.to/3camxnY – Reinforcement Learning: https://amzn.to/2PoANj9 – Theory of Knowledge: https://amzn.to/3a6Vp7x This conversation is part of the Artificial Intelligence podcast. If you would like to get more information about this podcast go to https://lexfridman.com/ai or connect with @lexfridman on Twitter https://twitter.com/lexfridman, LinkedIn https://www.linkedin.com/in/lexfridman/, Facebook https://www.facebook.com/lexfridman, Medium https://medium.com/@lexfridman, or YouTube https://www.youtube.com/lexfridman where you can watch the video versions of these conversations. If you enjoy the podcast, please rate it 5 stars on Apple Podcasts https://podcasts.apple.com/us/podcast/artificial-intelligence/id1434243584, follow on Spotify https://open.spotify.com/show/2MAi0BvDc6GTFvKFPXnkCL, or support it on Patreon https://www.patreon.com/lexfridman. This episode is presented by Cash App. Download it (App Store https://apps.apple.com/us/app/cash-app/id711923939, Google Play https://play.google.com/store/apps/details?id=com.squareup.cash&hl=en_US), use code “LexPodcast”. Here’s the outline of the episode. On some podcast players you should be able to click the timestamp to jump to that time. OUTLINE: 00:00 – Introduction 03:32 – Universe as a computer 05:48 – Occam’s razor 09:26 – Solomonoff induction 15:05 – Kolmogorov complexity 20:06 – Cellular automata 26:03 – What is intelligence? 35:26 – AIXI – Universal Artificial Intelligence 1:05:24 – Where do rewards come from? 1:12:14 – Reward function for human existence 1:13:32 – Bounded rationality 1:16:07 – Approximation in AIXI 1:18:01 – Godel machines 1:21:51 – Consciousness 1:27:15 – AGI community 1:32:36 – Book recommendations 1:36:07 – Two moments to relive (past and future)

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Lex Fridman?

  The following is a conversation with Marcus Hutter, senior research scientist at Google DeepMind. Throughout his career of research, including with Jürgen Schmitt-Huber and Shane Legg, he has proposed a lot of interesting ideas in and around the field of artificial general intelligence, including the development of AIXI, spelled A-I-X-I, model, which is a mathematical approach to AGI that incorporates ideas of Kolmogorov complexity, Solomonoff induction, and reinforcement learning. In 2006, Marcus launched the 50,000 euro Hutter Prize for Lossless Compression of Human Knowledge.

• 0:41Lex Fridman?

  The idea behind this prize is that the ability to compress well is closely related to intelligence. This, to me, is a profound idea. Specifically, if you can compress the first 100 megabytes or 1 gigabyte of Wikipedia better than your predecessors, your compressor likely has to also be smarter. The intention of this prize is to encourage the development of intelligent compressors as a path to AGI. In conjunction with his podcast release just a few days ago, Marcus announced a 10x increase in several aspects of this prize, including the money, to 500,000 euros. The better your compressor works relative to the previous winners, the higher fraction

• 1:26Lex Fridman?

  of that prize money is awarded to you. You can learn more about it if you Google simply "Hutter Prize." I'm a big fan of benchmarks for developing AI systems, and the Hutter Prize may indeed be one that will spark some good ideas for approaches that will make progress on the path of developing AGI systems. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube, give it five stars on Apple Podcasts, support it on Patreon, or simply connect with me on Twitter at @LexFriedman, spelled F-R-I-D-M-A-N.

• 2:02Lex Fridman?

  As usual, I'll do one or two minutes of ads now and never any ads in the middle that can break the flow of the conversation. I hope that works for you and doesn't hurt the listening experience. This show is presented by Cash App, the number one finance app in the App Store. When you get it, use code LEX PODCAST. Cash App lets you send money to friends, buy Bitcoin, and invest in the stock market with as little as $1. Brokerage services are provided by Cash App Investing, a subsidiary of Square and member SIPC. Since Cash App allows you to send and receive money digitally, peer to peer, and security in all digital transactions is very important, let me mention the PCI

• 2:44Lex Fridman?

  Data Security Standard that Cash App is compliant with. I'm a big fan of standards for safety and security. PCI DSS is a good example of that, where a bunch of competitors got together and agreed that there needs to be a global standard around the security of transactions. Now, we just need to do the same for autonomous vehicles and AI systems in general. So again, if you get Cash App from the App Store or Google Play and use the code LEX PODCAST, you'll get $10 and Cash App will also donate $10 to FIRST, one of my favorite organizations that is helping to advance robotics and STEM education for young people around the world. And now, here's my conversation

• 3:29Lex Fridman?

  with Marcus Hutter.

• 3:50Lex Fridman?

  Do you think of the universe as a computer or maybe an information processing system? Let's go with a big question first.

• 3:56Speaker 01

  Okay, I have a big question first. I think it's a very interesting hypothesis or idea. And I have a background in physics. So I know a little bit about physical theories, the standard model of particle physics and general relativity theory. And they are amazing and describe virtually everything in the universe. And they're all, in a sense, computable theories. I mean, they're very hard to compute. And, you know, it's very elegant, simple theories which describe virtually everything in the universe. So there's a strong indication that somehow the universe is computable. But it's a plausible hypothesis.

• 4:35Lex Fridman?

  So what, what do you think, just like you said, general relativity, quantum field theory? Why do you think that the laws of physics are so nice and beautiful and simple and compressible? Do you think our universe was designed, is naturally this way? Are we just focusing on the parts that are especially compressible? Our human minds just enjoy something about that simplicity? And in fact, there's other things that are not so compressible.

• 5:04Speaker 01

  No, I strongly believe and I'm pretty convinced that the universe is inherently beautiful, elegant and simple and described by these equations. And we're not just picking that. I mean, if there were some phenomena which cannot be neatly described, scientists would try that, right? And, you know, there's biology which is more messy, but we understand that it's an emergent phenomena. And, you know, it's complex systems, but they still follow the same rules, right? Of quantum electrodynamics, all of chemistry follows that and we know that. I mean, we cannot compute everything because we have limited computational resources. No, I think it's not a bias of the humans, but it's objectively simple. I mean, of course, you never know, you know, maybe there's some corners very far out in the universe

• 5:45Speaker 01

  or super, super tiny below the nucleus of, of, of atoms or, well, parallel universes where, which are not nice and simple, but there's no evidence for that. And we should apply Occam's razor and, you know, choose the simplest tree consistent with it. But although it's a little bit self-referential.

• 6:05Lex Fridman?

  So maybe a quick pause. What is Occam's razor?

• 6:08Speaker 01

  comes razor says that you should not multiply entities beyond necessity, which sort of if you translate it to proper English means uh and and you know in the scientific context means that if you have two theories or hypotheses or models which equally well describe the phenomenon you have study or the data, you should choose the more simple one.

• 6:31Lex Fridman?

  So that's just a principle or sort of that's not like a provable law perhaps perhaps we'll we'll we'll we'll kind of discuss it and think about it, but what's the intuition of why the simpler answer is the one that is likelier to be more correct descriptor of whatever we're talking about?

• 6:52Speaker 01

  I believe that Occam's razor is probably the most important principle in science. I mean, of course, we lead logical deduction, we do experimental design, yeah, but um science is about finding uh understanding the world, finding models of the world, and we can come up with crazy complex models which, you know, explain everything but predict nothing, but um the simple model seem to have predictive power and it's a valid question why? Yeah, and um there are two answers to that. You can just accept it that is the principle of science and we use this principle and it seems to be successful. We don't know why, but it just happens to be. Or you can try,

• 7:34Speaker 01

  you know, find another principle which explains Occam's razor. And if we start with the assumption that the world is governed by simple rules, then um there's a bias towards simplicity and applying Occam's razor um is the mechanism to finding these rules and actually in a more quantitative sense, and we come back to that later in terms of solemn deduction, you can rigorously prove that you if you assume that the world is simple, then Occam's razor is the best you can do in a certain sense.

• 8:06Lex Fridman?

  So I apologize for the romanticized question, but uh why do you think outside of its effectiveness? Why do we do you think we find simplicity so appealing as human beings? Why does it just? Why does E equals mc squared seem so?

• 8:23Lex Fridman?

  Beautiful to us, humans.

• 8:25Speaker 01

  I guess mostly, in general, many things can be explained by an evolutionary argument. And you know, there's some artifacts in humans, which you know are just artifacts and not evolutionary necessary. But with this beauty and simplicity, it's, I believe, at least the core is about, like science, finding regularities in the world, understanding the world, which is necessary for survival, right? You know, if I look at a bush, right, and I just see noise, and there is a tiger, right, and eats me, then I'm dead. But if I try to find a pattern, and we know that humans are prone to

• 9:08Speaker 01

  find more patterns in data than they are, you know, like the Mars face and all these things. But this bias towards finding patterns, even if they are none, but, I mean, it's best, of course, if they are, yeah, helps us for survival.

• 9:24Lex Fridman?

  Yeah, that's fascinating. I haven't thought really about the, I thought I just loved science, but they're indeed from, in terms of just for survival purposes, there is an evolutionary argument for why, why we find the work of Einstein so beautiful. Maybe a quick, small tangent, could you describe what Solomon of induction is?

• 9:48Speaker 01

  Yeah, so that's a theory which I claim, and Resolomonov sort of claimed, you know, a long time ago, that this solves the big philosophical problem of induction. And I believe the claim is essentially true. And what it does is the following. So, okay, for the picky listener, induction can be interpreted narrowly and wildly. Narrow means inferring models from data.

• 10:18Speaker 01

  And widely means also then using these models for doing predictions. So prediction is also part of of the induction. So I'm a little sloppy sort of with the terminology. And maybe that comes from Ray Solomon of, you know, being sloppy. Maybe I shouldn't say that.

• 10:33Speaker 01

  He can't complain anymore. So let me explain a little bit this theory.

• 10:37Lex Fridman?

  Yeah.

• 10:38Speaker 01

  In simple terms, so assume we have a data sequence. Make it very simple, the simplest one, say, one, one, one, one, one. And you see if one hundred ones. What do you think comes next? The natural answer, I must speed up a little bit. The natural answer is, of course, you know, one. Okay. And the question is, why? Okay. Well, we see a pattern there. Yeah, okay, there's a one and we repeat it. And why should it suddenly after hundred ones be different? So what we're looking for is simple explanations or models for the data we have. And now the question is, a model has to be presented in a certain language. In which language do we use? In science, we want formal languages and we can use mathematics or we can use programs on a computer.

• 11:21Speaker 01

  So abstractly on a Turing machine, for instance, or it can be a general purpose computer. So. And there, of course, lots of models of you can say, maybe it's a hundred ones and then a hundred zeros and a hundred ones. That's a model, right? But they are simpler models. There's a model, print one loop. It also explains the data. And if you push it to the extreme, you are looking for the shortest program, which, if you run this program, reproduces the data you have. It will not stop, it will continue naturally. And this you take for your prediction. And on the sequence of ones, it very plausible, right? That print one loop is the shortest program. We can give some more complex examples like one, two, three, four, five. What comes next? The short program is again, you know, counter.

• 12:06Speaker 01

  And so that is roughly speaking how Solomon's induction works. The extra twist is that it can also deal with noisy data. So if you have, for instance, a coin flip, say a biased coin, which comes up had with 60% probability.

• 12:22Speaker 01

  Then it will predict. It will learn and figure this out. And after a while it predict, oh, the next coin flip will be had with probability 60%. So it's the stochastic version of that.

• 12:32Lex Fridman?

  But the goal is, the dream is always the search for the short program.

• 12:36Speaker 01

  Yes, yeah. Well, in Solomon of Induction, precisely what you do is, so you combine, so looking for the shortest program is like applying Opax Razer, like looking for the simplest theory. There's also Epicurus' principle, which says, if you have multiple hypotheses which equally well describe your data, don't discard any of them, keep all of them around, you never know. And you can put that together and say, okay, I have a bias towards simplicity, but it don't rule out the larger models. And technically, what we do is, we weigh the shorter models higher and the longer models lower. And you use a Bayesian techniques, you have a prior, and which is precisely two to the minus the complexity of the program.

• 13:20Speaker 01

  And you weigh all this hypothesis and takes this mixture, and then you get also this stochasticity in.

• 13:25Lex Fridman?

  Yeah, like many of your ideas, that's just a beautiful idea of weighing based on the simplicity of the program. I love that. That seems to me maybe a very human-centric concept, seems to be a very appealing way of discovering good programs in this world. You've used the term compression quite a bit. I think it's a beautiful idea. Sort of, we just talked about simplicity and maybe science or just all of our intellectual pursuits is basically the attempt to compress the complexity all around us into something simple. So what does this word mean to you, compression?

• 14:07Speaker 01

  I essentially have already explained it. So it, compression means for me, finding short programs for the data or the phenomenon at hand. You could interpret it more widely as, you know, finding simple theories, which can be mathematical theories, or maybe even informal, like, you know, just in words. Compression means finding short descriptions, explanations, programs for the data.

• 14:32Lex Fridman?

  Do you see science as a kind of? Our human attempt at compression. So we're speaking more generally because when you say programs, you're kind of zooming in on a particular sort of almost like a computer science, artificial intelligence focus. But do you see all of human endeavor as a kind of compression?

• 14:52Speaker 01

  Well, at least all of science I see as an endeavor of compression, not all of humanity, maybe. And well, there are also some other aspects of science like experimental design. Right? I mean, we we create experiments specifically to get extra knowledge. And this is that isn't part of the decision making process. But once we have the data to understand the data is essentially compression. So I don't see any difference between compression,

• 15:20Speaker 01

  understanding and prediction.

• 15:23Lex Fridman?

  So we're jumping around topics a little bit, but returning back to simplicity, a fascinating concept of Kolmogorov complexity. So in your sense, do most objects in the in our mathematical universe have high Kolmogorov complexity? And maybe what is, first of all, what is Kolmogorov complexity?

• 15:43Speaker 01

  Okay, Kolmogorov complexity is a notion of simplicity or complexity. And it takes the compression view to the extreme. So I explained before that if you have some data sequence, just think about a file on a computer and best sort of, you know, just a string of bits. And if you and we have data compresses like we compress big files into zip files with certain compressors. And you can also produce self extracting AKFs. That means there's an executable. If you run it, it reproduces the original file without needing an extra decompressor. It's just a decompressor plus the AKF together in one. And now there are better and worse compressors. And you can ask, what is the ultimate compressor?

• 16:28Speaker 01

  So what is the shortest possible self extracting AKF you could produce for certain data set? Yeah, which reproduces the data set. And the length of this is called the Kolmogorov complexity. And arguably, that is the information content in the data set. I mean, if the data set is very redundant or very boring, you can compress it very well. So the information content. Should be low and you know it is low according to this definition.

• 16:54Lex Fridman?

  So it's the length of the shortest program that summarizes the data?

• 16:58Speaker 01

  Yes, yeah.

• 16:59Lex Fridman?

  And what's your sense of our sort of universe? When we think about the different the different objects in our universe that we try concepts or whatever at every level, do they have higher or low Kolmogorov complexity? So what's the hope? Do we have a lot of hope in be able to summarize much of our world?

• 17:22Speaker 01

  Ah, that's a tricky and difficult question. So as I said before, I believe that the whole universe, based on the evidence we have, is very simple. So it has a very short description.

• 17:36Lex Fridman?

  Sorry, to linger on that, the whole universe. What does that mean? Do you mean at the very basic fundamental level in order to create the universe?

• 17:46Speaker 01

  Yes, yeah. So you need a very short program and you run it.

• 17:50Lex Fridman?

  To get the thing going.

• 17:51Speaker 01

  To get the thing going and then it will reproduce our universe. Um, there's a problem with noise. Um, we can come back to that later, possibly.

• 17:59Lex Fridman?

  Is noise a problem or is it a bug or a feature?

• 18:03Speaker 01

  I would say it makes our life as a scientist really, really much harder. I mean, think about without noise, we wouldn't need all of the statistics.

• 18:13Lex Fridman?

  But then maybe we wouldn't feel like there's a free will. Maybe we need that for the

• 18:18Speaker 01

  This is an illusion that noise can give you free will.

• 18:22Lex Fridman?

  That's at least in that way, it's a feature.

• 18:24Speaker 01

  But also, if you don't have noise, you have chaotic phenomena which are effectively like noise. So we can't, you know, get away with statistics even then. I mean, think about rolling a dice and you know, forget about quantum mechanics and you know exactly how you you throw it, but I mean, it's still so hard to compute the trajectory that effectively it is best to model it, you know, as you know, coming out with a number with probability one over six. But from this set of philosophical Kolmogorov complexity perspective, if we didn't have noise, then arguably you could describe the whole universe as Um, well, as a standard model plus generativity. I mean, we don't have a theory of everything yet,

• 19:07Speaker 01

  but sort of assuming we are close to it or have it, yeah. Plus the initial conditions, which may hopefully be simple, and then you just run it and then you would reproduce the universe. Um, but that's brought by noise or by chaotic um systems or by initial conditions, which, you know, may be complex. So now if we don't take the whole universe, but just um a subset, you know, just take planet Earth. Planet Earth cannot be compressed, you know, into a couple of equations. Um, this is a hugely complex system.

• 19:36Lex Fridman?

  So interesting. So when you look at the window, like the whole thing might be simple, but when you just take a small window, then uh.

• 19:43Speaker 01

  It may become complex and that may be counterintuitive, but um there's a very nice analogy. Um, the, the book, uh, the library of all books. So imagine you have a normal library with interesting books and you go there, great, lots of information and huge, quite complex, yeah. So now I create a library which contains all possible books, say of five hundred pages. So the first book just has A, A, A, A, A over all the pages. The next book, A, A, A, and ends with B and so on. I create this library of all books. I can write a super short program which creates this library. So this library which has all books has zero information content. And you take a subset of this library and suddenly you have a lot of information in there.

• 20:22Lex Fridman?

  So that's fascinating. Uh, I think one of the most beautiful object, mathematical objects that at least today seems to be understudied or undertalked about is cellular automata. Uh, what lessons do you draw from sort of the game of life or cellular automata, where you start with the simple rules, just like you're describing with the universe, and somehow complexity emerges? Do you feel like you have an intuitive grasp on the behavior, the fascinating behavior of such systems where some, like you said, some chaotic behavior could happen, some complexity could emerge, some, it could die out in some very rigid structures? Do you have a sense about uh cellular automata that somehow transfers maybe

• 21:05Lex Fridman?

  to the bigger questions of our universe?

• 21:08Speaker 01

  Yeah, the cellular automata and especially the converse game of life. Um. It's really great because this rule is so simple. You can explain it to every child and even by hand you can simulate a little bit. And you see this beautiful patterns emerge. And people have proven, you know, that is even Turing complete. You cannot just use a computer to simulate game of life, but you can also use game of life to simulate any computer. Um. That is truly amazing. And it's it's the prime example probably to demonstrate that very simple rules can lead to very rich phenomena. And people, you know, sometimes, you know, how can, how is chemistry and biology so rich? I mean, this can't be based on simple rules, yeah. But no, we know quantum electrodynamics describes all of chemistry.

• 21:53Speaker 01

  And um, and we come later back to that. I claim intelligence can be explained or described in one single equation, this very rich phenomenon. Um. You asked also about whether, you know, I understand this phenomenon and it's probably not. Um. And there's this saying, you never understand really things, you just get used to them. And

• 22:18Speaker 01

  I think I was pretty used used to sell a lot of tomatoes. So you believe that you understand now why this phenomenon happens, but I give you a different example. I didn't play too much with this Converse game of life, but um, a little bit more um, with fractals and with the Mandelbrot set and you know, these beautiful, you know, patterns. Just just look Mandelbrot set. Um. And well, when the computers were really slow and I just had a black and white monitor and I programmed my own programs on a, in a assembler too.

• 22:46Lex Fridman?

  A assembler, wow. Wow, you're legit.

• 22:51Speaker 01

  To get these practice on the screen and it was mesmerized and much later. So I returned to this, you know, every couple of years. And then I tried to understand what is going on and you can understand a little bit. So I tried to derive the locations, you know, they are this um, um, um, circles and the apple shape. And then you have um, smaller um, Mandelbrot sets recursively in this set. And there's a way to mathematically by solving high order polynomials to figure out where these centers are and what size they are approximately. And by sort of mathematically approaching this problem, you slowly get a feeling of

• 23:32Speaker 01

  um, why things are like they are and that sort of isn't, you know.

• 23:39Speaker 01

  First step to understanding why this rich phenomenon.

• 23:42Lex Fridman?

  Do you think it's possible, what's your intuition? Do you think it's possible to reverse engineer and find the short program that generated the, these fractals sort of by looking at the fractals?

• 23:53Speaker 01

  Well, in principle, yes. Yeah. So, I mean, in principle, what you can do is you take, you know, any data set, you know, you take these fractals or you take whatever, your data set, whatever you have. Say a picture of Converse Game of Life. And you run through all programs. You take a program of size one, two, three, four, and all these programs, run them all in parallel in so-called dovetailing fashion. Give them computational resources, first one fifty percent, second one half resources and so on, and let them run. Wait until they halt, give an output, compare it to your data. And if some of these programs produce the correct data, then you stop and then you have already some program. It may be a long program because it's faster. And then you continue and you get shorter and shorter programs until you eventually

• 24:38Speaker 01

  find the shortest program. The interesting thing, you can never know whether it's the shortest program because there could be an even shorter program, which is just even slower. And you just have to wait, yeah? But asymptotically, and actually after finite time, you have the shortest program. So this is a theoretical but completely impractical way of finding the underlying structure in every data set. And that is a Solomonoff induction, Dars and Kolmogorov complexity. In practice, of course, we have to approach the problem more intelligently. And then if you take resource limitations into account, there's, for instance, the field of pseudo-random numbers, yeah?

• 25:21Speaker 01

  And these are random numbers, so these are deterministic sequences. But no algorithm, which is fast, fast means runs in polynomial time, can detect that it's actually deterministic. So we can produce interesting, I mean, random numbers maybe not that interesting, but just an example, we can produce complex-looking data. And we can then prove that no fast algorithm can detect the underlying pattern.

• 25:49Lex Fridman?

  Which is unfortunately, that's a big challenge for our search for simple programs in the space of artificial intelligence, perhaps.

• 25:59Speaker 01

  Yes, it definitely is one of the vision intelligence and it's quite surprising that it's, I can't say easy, I mean physicists work really hard to find these theories, but apparently it was possible for human minds to find these simple rules in the universe. It could have been different, right?

• 26:16Lex Fridman?

  It could have been different. It's awe inspiring.

• 26:22Lex Fridman?

  So let me ask another absurdly big question. What is intelligence? In your view?

• 26:30Speaker 01

  So I have, of course, a definition.

• 26:34Lex Fridman?

  I wasn't sure what you're gonna say, because you could have just as easy said, "I have no clue."

• 26:38Speaker 01

  Which many people would say, but I'm not modest in this question.

• 26:44Speaker 01

  So the informal version, which I worked out together with Shane Leck, who co-founded DeepMind, is that intelligence measures an agent's ability to perform well in a wide range of environments. So that doesn't sound very impressive, but these words have been very carefully chosen and there is a mathematical theory behind that and we come back to that later. And if you look at this definition by itself, it seems like, yeah, okay, but it seems a lot of things are missing. But if you think it through, then you realize that most, and I claim all of the other traits,

• 27:28Speaker 01

  at least of rational intelligence, which we usually associate with intelligence, are emergent phenomena from this definition, like creativity, memorization, planning, knowledge. You all need that in order to perform well in a wide range of environments. So you don't have to explicitly mention that in a definition.

• 27:47Lex Fridman?

  Interesting. So yeah, so the consciousness, abstract reasoning, all these kinds of things are just emerging phenomena that help you in towards, can you say the definition again? So multiple environments, did you mention the word goals?

• 28:03Speaker 01

  No, but we have an alternative definition. Instead of performing well, you can just replace it by goal. So intelligence measures an agent's ability to achieve goals in a wide range of environments. That's more or less.

• 28:14Lex Fridman?

  But interesting because in there, there's an injection of the word goals. So we want to specify, there should be a goal.

• 28:20Speaker 01

  Yeah, but perform well is sort of, what does it mean? It's the same problem.

• 28:24Lex Fridman?

  Yeah, there's a little bit of gray area, but it's much closer to something that could be formalized. Are, in your view, are humans, where do humans fit into that definition? Are they general intelligence systems that are able to perform in, like, how good are they at fulfilling that definition, at performing well in multiple environments?

• 28:48Speaker 01

  Yeah, that's a big question. I mean, the humans are performing best among all.

• 28:54Lex Fridman?

  Species on earth.

• 28:55Speaker 01

  Species we know, we know of, yeah.

• 28:58Lex Fridman?

  Depends. You could say that trees and plants are doing a better job. They'll probably outlast us. So.

• 29:04Speaker 01

  Yeah, but they are in a much more narrow environment, right? I mean, you just, you know, have a little bit of air pollution and these trees die and we can adapt, right? We build houses, we build filters, we, we, we do geoengineering.

• 29:17Lex Fridman?

  So the multiple environment part.

• 29:18Speaker 01

  Yeah, that is very important. Yes. So that distinguished narrow intelligence from wide intelligence, also in the AI research.

• 29:26Lex Fridman?

  So let me ask the Alan Turing question. Can machines think? Can machines be intelligent? So in your view, I have to kind of ask, the answer is probably yes, but I want to kind of hear your thoughts on it. Can machines be made to fulfill this definition of intelligence, to achieve intelligence?

• 29:48Speaker 01

  Well, we are sort of getting there and, you know, on a small scale, we are already there. The wide range of environments is missing. But we have self-driving cars, we have programs that play Go and chess, we have speech recognition. So it's pretty amazing, but you can, you know, these are narrow environments.

• 30:07Speaker 01

  But if you look at AlphaZero, that was also developed by DeepMind. I mean, got famous with AlphaGo and then came AlphaZero a year later. That was truly amazing. So reinforcement learning algorithm, which is able just by self-play to play chess and then also Go. And I mean, yes, they're both games, but they're quite different games. And you know, there's, you didn't, don't feed them the rules of the game. And the most remarkable thing, which is still a mystery to me, that usually for any decent chess programmer, I don't know much about Go, you need opening books and end game tables and so on to, and nothing in there, nothing was put in there.

• 30:47Lex Fridman?

  Especially with AlphaZero, the self-play mechanism, starting from scratch, being able to learn actually new strategies is a.

• 30:56Speaker 01

  Yeah, it rediscovered, you know, all these famous openings within four hours by itself. What I was really happy about, I'm a terrible chess player, but I like Queen Gambi. And AlphaZero figured out that this is the best opening.

• 31:12Lex Fridman?

  Finally.

• 31:14Lex Fridman?

  Somebody proved you correct.

• 31:17Speaker 01

  So yes, to answer your question, yes, I believe that general intelligence is possible. And it also, I mean, it depends how you define it. Do you say AGI, with general intelligence, artificial general intelligence, only refers to if you achieve human level or a sub-human level, but quite broad? Is it also general intelligence? So we have to distinguish or it's only superhuman intelligence, general artificial intelligence?

• 31:42Lex Fridman?

  Is there a test in your mind, like the Turing test for natural language or some other test that would impress the heck out of you? That would kind of cross the line of your sense of intelligence within the framework that you said?

• 31:57Speaker 01

  Well, the Turing test, well, has been criticized a lot, but I think it's not as bad as some people think. And some people think it's too strong. So it tests not just for a system to be intelligent, but it also has to fake human.

• 32:12Lex Fridman?

  Deception.

• 32:13Speaker 01

  Deception, right? Which is, you know, much harder. And on the other hand, they say it's too weak, yeah, because it just maybe fakes, you know, emotions or intelligent behavior. It's not real. But I don't think that's the problem or big problem. So if, if you would pass the Turing test. Um. So a conversation over terminal with a bot for an hour um or maybe a day or so and you can fool a human into, you know, not knowing whether this is a human or not, so that's the Turing test, yeah. I would be truly impressed. And we have this annual competition, the Lepner Prize. And I mean, it started with Eliza, that was the first conversational program.

• 32:55Speaker 01

  And um, what is it called, the Japanese Mitsuku or so, that's the winner of the last, you know, couple of years.

• 33:01Lex Fridman?

  Yeah.

• 33:02Speaker 01

  And well,

• 33:02Lex Fridman?

  Quite impressive.

• 33:03Speaker 01

  Yeah, it's quite impressive. And then Google has developed Mina, right? Just, just recently, that's an open domain conversational bot, uh, just a couple of weeks ago, I think.

• 33:15Lex Fridman?

  Yeah, I kind of like the metric that uh, sort of the Alexa Prize has proposed. I mean, maybe it's obvious to you, it wasn't to me, of setting sort of a length of a conversation, like you, you want the bot to be sufficiently interesting that you'd want to keep talking to it for like twenty minutes. And that's, that's a surprisingly uh, effective in aggregate metric. Because you really, like, nobody has the patience to be able to talk to a bot that's not interesting and intelligent and witty and is able to go on to different tangents, jump domains, be able to, you know, say something interesting to maintain your attention.

• 33:54Speaker 01

  And maybe many humans who've also failed this test.

• 33:56Lex Fridman?

  That's the, that's unfortunately, we set, just like with autonomous vehicles, with chatbots, we also set a bar that's way too hard, hard to reach.

• 34:05Speaker 01

  I said, you know, the Turing test is not as bad as some people believe. But what is um, really not useful about the Turing test, it gives us no guidance um, how to develop these systems in the first place. Of course, you know, we can develop them by trial and error and, you know, do whatever and then run the test and see whether it works or not. But um, a mathematical definition of intelligence gives us um, you know, an objective which we can then analyze by, you know, theoretical tools or computational um, and, you know, maybe even prove how close we are. And we will come back to that later, um, with the ICSI model. So um, I mentioned the compression, right? So in, in natural language processing,

• 34:51Speaker 01

  They achieved amazing results and one way to test is of course, you know, take the system, you train it and then you, you know, see how well it performs on the task. But a lot of performance measurement is done by so-called perplexity, which is essentially the same as complexity or compression length. So the NLP community develops new systems and then they measure the compression length and then they have ranking and leaks because there's a strong correlation between compressing well and then the system's performing well at the task at hand. It's not perfect, but it's good enough for them as as an intermediate aim.

• 35:30Lex Fridman?

  So you mean a measure, so this is kind of almost returning to the common goal of complexity. So you're saying good compression usually means good intelligence?

• 35:42Speaker 01

  Yes.

• 35:43Lex Fridman?

  So you mentioned you're one of the, one of the only people who dared boldly to try to formalize the idea of artificial general intelligence, to have a mathematical framework for intelligence, just like as we mentioned, termed IAXI. So let me ask the basic question, what is IAXI?

• 36:10Speaker 01

  Okay, so let me first say what it stands for because

• 36:15Lex Fridman?

  What it stands for, actually, that's probably the more basic question.

• 36:17Speaker 01

  Yeah. The first question is usually how, how it's pronounced, but finally I put it on the website how it's pronounced, so you figured it out.

• 36:25Lex Fridman?

  Yeah.

• 36:26Speaker 01

  The name comes from AI, artificial intelligence, and the Xi is the Greek letter Xi, which are used for Solomonoff's distribution for quite stupid reasons, which I'm not willing to repeat here in front of camera.

• 36:44Speaker 01

  So it just happened to be more or less arbitrary choice of xi. But it also has nice other interpretations. So there are actions and perceptions in this model, right? An agent has actions and perceptions and over time. So this is A index i, X index i. So this is action at time i and then followed by perception at time i.

• 37:07Lex Fridman?

  We'll go with that. I'll edit out the first part.

• 37:09Speaker 01

  I'm just kidding. I have some more interpretations. So at some point, maybe five years ago or 10 years ago, I discovered in

• 37:19Speaker 01

  in Barcelona, it was on a big church. There was in, you know, in stone engraved some text. And the word ixi appeared there a couple of times.

• 37:30Speaker 01

  I was very surprised and and and happy about that. And I looked it up. So it is Catalan language and it means with some interpretation of that's it. That's the right thing to do. Yeah. Heurica.

• 37:42Lex Fridman?

  Oh, so it's almost like destined somehow came

• 37:48Lex Fridman?

  came to you in a dream.

• 37:49Speaker 01

  So And similar, there's a Chinese word, ixi, also written like ixi. If you transcribe that to Pinyin. And the final one is that is AI crossed with induction because that is and it's going more to the content now. So good old fashioned AI is more about, you know, planning and known deterministic world. And induction is more about often your I R D data and inferring models. And essentially what this AIXI model does is combining these two.

• 38:14Lex Fridman?

  And I actually also recently, I think, heard that in Japanese AI means love. So so if you can combine Xi somehow with that, I think we can.

• 38:26Lex Fridman?

  There might be some interesting ideas there. So ixi, let's then take the next step. Can you maybe talk at the big level of what is this mathematical framework?

• 38:36Speaker 01

  Yeah, so it consists essentially of two parts. One is the learning and induction and prediction part. And the other one is the planning part. So let's come first to the learning induction prediction part, which essentially I explained already before. So what we need for any agent to act well is that it can somehow predict what happens. I mean, if you have no idea what your actions do, how can you decide which actions are good or not? So you need to have some model of what your actions affect. So what you do is you have some experience. You build models like scientists, you know, of your experience. Then you hope these models are roughly correct and then you use these models for prediction.

• 39:21Lex Fridman?

  And a model is, sorry to interrupt, and a model is based on the, your perception of the world, how your actions will affect that world.

• 39:27Speaker 01

  That's not.

• 39:29Lex Fridman?

  So what, how do you think about?

• 39:30Speaker 01

  That's not the important part, but it is technically important, but at this stage, we can just think about predicting, say, uh, stock market data, weather data or IQ sequences, one, two, three, four, five, what comes next? Yeah. So, um, of course our actions affect what we're doing, but I come back to that in a second.

• 39:47Lex Fridman?

  So, so, and I'll keep just interrupting. So, uh, just to draw a line between prediction and planning. Or what do you mean by prediction in this, in this way? It's trying to predict the, the environment without your long-term action in that environment? What is prediction?

• 40:06Speaker 01

  Okay. If you want to put the actions in now, okay, then let's put in now. Yeah. Um, so. We don't have to put them now.

• 40:13Lex Fridman?

  Yeah, yeah. Scratch it, scratch it. Dumb question. Okay.

• 40:15Speaker 01

  So the simplest form of prediction is that you just have data which you passively observe.

• 40:21Lex Fridman?

  Yes.

• 40:22Speaker 01

  And you want to predict what happens without, you know, interfering. As I said, weather forecasting, stock market, uh, IQ sequences or, uh, just anything. Okay? And Solomons' theory of induction based on compression. So you look for the shortest program which describes your data sequence. And then you take this program, run it, which reproduces your data sequence by definition. And then you let it continue running and then it will produce some predictions. And you can rigorously prove that for any prediction task, this is essentially the best possible predictor. Of course, if there's a prediction task, um, or a task which is unpredictable, like, you know, your fair coin flips.

• 41:04Speaker 01

  Yeah, I cannot predict the next fair coin flip. What Solomons does is says, okay, next head is probably 50%. It's the best you can do. So if something is unpredictable, Solomons will also not magically predict it. But if there is some pattern and predictability, then Solomons induction will figure that out eventually. And not just eventually, but rather quickly and you can have proof convergence rates.

• 41:28Speaker 01

  Whatever your data is. So that is pure magic in a sense. What's the catch? Well, the catch is that it's not computable and we come back to that later. You cannot just implement it, you know, even with Google resources here and run it and, you know, predict the stock market and become rich. I mean. Ray Solomonoff already tried it at the time.

• 41:45Lex Fridman?

  But so the the basic task is you're in the environment and you're interacting with the environment to try to learn a model of that environment. And the model is in the space of all these programs and your goal is to get a bunch of programs that are simple.

• 41:58Speaker 01

  And so let's let's go to the actions now. But actually, good that you asked. Usually I skip this part, although that is also a minor contribution which I did. So the action part, but I usually sort of just jump to the decision part. So let me explain to the action part now. Thanks for asking.

• 42:11Lex Fridman?

  Yes.

• 42:12Speaker 01

  So you have to modify it a little bit by now not just predicting a sequence which just comes to you. But you have an observation, then you act somehow. And then you want to predict the next observation based on the past observation and your action. Then you take the next action. You don't care about predicting it because you're doing it. And then you get the next observation and you want, well, before you get it, you want to predict it again based on your past action and observation sequence. It just condition extra on your actions. There's an interesting alternative that you also try to predict your own actions.

• 42:53Speaker 01

  If you want.

• 42:53Lex Fridman?

  In the past or the future?

• 42:55Speaker 01

  Your future actions.

• 42:57Lex Fridman?

  That's interesting.

• 42:59Lex Fridman?

  Wait, let me wrap. I think my brain is broke.

• 43:03Speaker 01

  We should maybe discuss that later after I've explained the ICSI model. That's an interesting variation.

• 43:07Lex Fridman?

  But that is a really interesting variation. And a quick comment, I don't know if you want to insert that in here. But you're looking at the, in terms of observations, you're looking at the entire, the big history, the long history of the observations.

• 43:20Speaker 01

  Exactly. That's very important, the whole history from birth sort of of the agent. And we can come back to that also why this is important here. Often, you know, in RL, you have MDPs, Markov decision processes, which are much more limiting. Okay, so now we can predict. Conditioned on actions, so even if influenced environment. But prediction is not all we want to do, right? We also want to act really in the world. And the question is how to choose the actions. And we don't want to greedily choose the actions. You know, just, you know, what is best in the next time step. And we first, I should say, you know, what is, you know, how do we measure performance? So we measure performance by giving the agent reward. That's the so-called reinforcement learning framework.

• 44:03Speaker 01

  So every time step, you can give it a positive reward or negative reward. Or maybe no reward, it could be very scarce, right? Like if you play chess, just at the end of the game, you give plus one for winning or minus one for losing. So in the AICE framework, that's completely sufficient. So occasionally you give a reward signal and you ask the agent to maximize reward, but not greedily, sort of, you know, the next one, next one, because that's very bad in the long run if you're greedy. So, but over the lifetime of the agent. So let's assume the agent lives for M time steps. Let's say dies in sort of hundred years, sharp. That's just, you know, the simple model to explain. So it looks at the future reward sum and ask what is my action sequence? Or actually more precisely my policy, which leads in expectation,

• 44:47Speaker 01

  because I don't know the world, to the maximum reward sum. Let me give you an analogy. In chess, for instance, we know how to play optimally in theory. It's just a minimax strategy. I play the move which seems best to me under the assumption that the opponent plays the move which is best for him. So best, so worst for me. Under the assumption that he, I play again the best move. And then you have this expectimax tree to the end of the game. And then you back propagate and then you get the best possible move. So that is the optimal strategy which for Neumann already figured out a long time ago

• 45:23Speaker 01

  for playing adversarial games. Luckily, or maybe unluckily for the theory, it becomes harder. The world is not always adversarial. So it can be, if there are humans even cooperative here, or nature is usually, I mean, the dead nature is stochastic, you know, things just happen randomly. Or I don't care about you. So what you have to take into account is the noise, yeah? And not necessarily adversality. So you replace the minimum on the opponent's side by an expectation, which is general enough to include also adversarial cases. So now instead of a minimax strategy, you have an expectimax strategy. So far so good, so that is well known, it's called sequential decision theory.

• 46:05Speaker 01

  But the question is, on which probability distribution do you base that? If I have the true probability distribution, like say I play begemen, right? There's dice and there's certain randomness involved. Yeah, I can calculate probabilities, I feed it in the expectimax or the sequential decision tree, come up with the optimal decision if I have enough compute. But in the, for the real world, we don't know that, you know? What is the probability the driver in front of me breaks? I don't know, yeah, so depends on all kinds of things and especially new situations, I don't know. So this is this unknown thing about prediction and there's where Solomonoff comes in. So what you do is in sequential decision tree, you just replace the true distribution, which we don't know, by this universal distribution.

• 46:50Speaker 01

  I didn't explicitly talk about it, but this is used for universal prediction and plug it into the sequential decision mechanism. And then you get the best of both worlds. You have a long-term planning agent, but it doesn't need to know anything about the world because the Solomonoff induction part learns.

• 47:09Lex Fridman?

  Can you explicitly try to describe the universal distribution and how Solomonoff induction plays a role here? I'm trying to understand.

• 47:18Speaker 01

  So what it does it? So in the simplest case, I said, take the shortest program describing your data, run it, have a prediction which would be deterministic.

• 47:26Lex Fridman?

  Yes.

• 47:27Speaker 01

  Okay. But you should not just take the shortest program, but also consider the longer ones, but give it lower a priori probability. So in the Bayesian framework, you say a priori, any distribution,

• 47:43Speaker 01

  which is a model or a stochastic program, has a certain a priori probability, which is two to the minus, and y two to the minus length, you know, I could explain, length of this program. So longer programs are punished, yeah, a priori. And then you multiply it with the so-called likelihood function, yeah? Which is, as the name suggests, is how likely is this model given the data at hand? So if you have a very wrong model, it's very unlikely that this model is true. And so it is very small number. So even if the model is simple, it gets penalized by that. And what you do is then you take just the sum, or this is the average over it. And this gives you a probability distribution. So called universal distribution or Solomonoff distribution.

• 48:28Lex Fridman?

  So it's weighed by the simplicity of the program and the likelihood?

• 48:31Speaker 01

  Yes.

• 48:32Lex Fridman?

  It's kind of a nice idea.

• 48:34Speaker 01

  Yeah.

• 48:35Lex Fridman?

  So, okay, uh, and then, uh, you said there's you're planning N or M or forgot the letter steps into the future. So how difficult is that problem? What's involved there?

• 48:46Speaker 01

  Okay.

• 48:47Lex Fridman?

  It's a basic optimization problem. What are we talking about?

• 48:49Speaker 01

  Yeah, so you have a planning problem up to horizon M and that's exponential time in in the horizon M, which is, I mean, it's computable, but intractable. I mean, even for chess, it's already intractable to do that exactly. And, you know, for.

• 49:02Lex Fridman?

  But it could be also discounted kind of framework where.

• 49:06Speaker 01

  Yeah. So, so having a hard horizon, you know, at hundred years is just for simplicity of discussing the model and also sometimes the master's simple.

• 49:16Speaker 01

  But there are lots of variations. It's actually a quite interesting parameter. It's it's there's nothing really problematic about it, but it's very interesting. So for instance, you think, no, let's let's let's let the parameter M tend to infinity, right? You want an agent which lives forever, right? If you do it narrowly, you have two problems. First, the mathematics breaks down because you have an infinite reward sum, which may give infinity and getting reward zero point one every time step is infinity and giving reward one every time step is infinity. So equally good. Not really what we want. Other problem is that if you have an infinite life, you can be lazy for as long as you want for 10 years.

• 49:55Lex Fridman?

  Yeah.

• 49:56Speaker 01

  And then catch up with the same expected reward. And, you know, think about yourself or, you know, or maybe, you know, some friends or so, if they knew they lived forever, you know, why work hard now? You know, just enjoy your life, you know, and then catch up later. So that's another problem with the infinite horizon. And you mentioned, yes, we can go to discounting. But then the standard discounting is so called geometric discounting. So a dollar today is about worth as much as, you know, one dollar and five cents tomorrow. So if you do this so called geometric discounting, you have introduced an effective horizon. So the agent is now motivated to look ahead a certain amount of time effectively. It's like a moving horizon.

• 50:38Speaker 01

  And for any fixed effective horizon, there is a problem to solve which requires larger horizon. So if I look ahead, you know, five time steps, I'm a terrible chess player, right? I need to look ahead longer. If I play Go, I probably have to look ahead even longer. So for every problem, um, for every horizon, there is a problem which this horizon cannot solve.

• 51:00Lex Fridman?

  Yes.

• 51:01Speaker 01

  But I introduced the so-called near harmonic horizon, which goes down with one over T rather than exponentially T, which produces an agent which effectively looks into the future proportional to its age. So if it's five years old, it plans for five years. If it's hundred years old, it then plans for hundred years.

• 51:17Lex Fridman?

  Interesting.

• 51:17Speaker 01

  And it's a little bit similar to humans too, right? My children don't plan ahead very long, but then we get adult, we play ahead more longer. Maybe when we get old, very old, I mean, we know that we don't live forever, you know, maybe then our horizon shrinks again.

• 51:31Lex Fridman?

  So that, oh, that, so that's really interesting. So adjusting, adjusting the horizon, what, is there some mathematical benefit of that? Of, or is it just a nice, uh, I mean intuitively, empirically, it would probably be a good idea to sort of push the horizon back to, extend the horizon as you experience more of the world. But is there some mathematical conclusions here that are beneficial?

• 51:54Speaker 01

  With so long model of predictions, the prediction part, we have extremely strong finite time, but finite data results. So you have so and so much data, then you lose so and so much. So the theory is really great. With the ICSI model, with the planning part, many results are only asymptotic, um, which, well, this is.

• 52:14Lex Fridman?

  What does asymptotic mean?

• 52:15Speaker 01

  Asymptotic means you can prove, for instance, that in the long run, if the agent, you know, acts long enough, then, you know, it performs optimal or some nice thing happens. So, but you don't know how fast it converges, yeah? So it may converge fast, but we're just not able to prove it because it's a difficult problem. Or maybe there's a bug, um, in the, in the, in the model so that it's really dead slow. Yeah. So, so that is what asymptotic means, sort of eventually, but we don't know how fast. And if I give the agent a fixed horizon M, yeah, then. I cannot prove asymptotic results, right? So I mean, sort of if it dies in a hundred years, then in a hundred years it's over. I cannot say eventually. So this is the advantage of the discounting, that I can prove asymptotic results.

• 53:00Lex Fridman?

  So just to clarify, so, so I, okay, I made, I've built up a model. Well, now in the moment, I've have this way of looking several steps ahead. How do I pick what action I will take?

• 53:16Speaker 01

  It's like with the playing chess, right? You do this minimax. In this case here, do you expect the max based on the Solomon of distribution? Um, you propagate back. Um, and then while an action falls out, the action which maximizes the future expected reward under Solomon of distribution. And then you just take this action.

• 53:35Lex Fridman?

  And then repeat.

• 53:37Speaker 01

  And then you get a new observation and you feed it in this action observation. Then you repeat.

• 53:40Lex Fridman?

  And the reward, so on.

• 53:42Speaker 01

  Yeah, so you're a road to, yeah.

• 53:44Lex Fridman?

  And then maybe you can even predict your own action. I love that idea. But, okay, this big framework. What is it, uh, is, is, I mean,

• 53:54Lex Fridman?

  uh, it's kind of a beautiful mathematical framework to think about artificial general intelligence. What can you, what does it help you intuit about, um, how to build such systems? Or maybe from another perspective, what, how, what does it help us to, in understanding AGI?

• 54:14Speaker 01

  So when I started in the field, I was always interested in two things. One was, you know, AGI. Um, the name didn't exist then, which was called general AI or, um, strong AI. Um, and the physics theory of everything. So I switched back and forth between computer science and physics quite often.

• 54:32Lex Fridman?

  You said the theory of everything.

• 54:33Speaker 01

  Theory of everything, yeah, this is like.

• 54:34Lex Fridman?

  It was basically the two biggest problems before all, all of humanity.

• 54:40Speaker 01

  Yeah, I can explain if you wanted some later time, you know, why I'm interested in these two questions.

• 54:47Lex Fridman?

  Can I ask you, and a small tangent? If, if, uh, if one to be, it was one to be solved, which one would you? If one, if you were, if an apple fell in your head and there was a brilliant insight and you could arrive at the solution to one, would it be AGI or the theory of everything?

• 55:06Speaker 01

  Uh, definitely AGI because once the AGI problem is solved, they can ask the AGI to solve the other problem for me.

• 55:13Speaker 01

  Yeah, brilliantly put. Okay, so, uh, so as you were saying about it. Okay, so, and the reason why I didn't settle, I mean, this thought about, you know, once you have solved AGI, it solves all kinds of other, not just the theory of every problem, but all kinds of use, more useful problems to humanity is very appealing to many people. And, you know, I had this thought also, but um, I was quite disappointed with the state of the art of the field of AI. There was some theory, you know, about logical reasoning, but I was never convinced that this will fly. And then there was this more heuristic approaches with neural networks, and I didn't like these heuristics. So, and also I didn't have any

• 55:56Speaker 01

  good idea myself. So that's the reason why I toggled back and forth quite some while and even worked for four and a half years in a company developing software, something completely unrelated. But then I had this idea about the ICSI model. And um, so what it gives you, it gives you a gold standard. So I have proven that this is the most intelligent agents which anybody could build, built in quotation mark, because it's just mathematical and you need infinite compute. Yeah, but this is the limit. And this is completely specified. It's not just a framework. And, you know, every year, tens of frameworks are developed,

• 56:38Speaker 01

  which are just skeletons, and then pieces are missing. And usually these missing pieces, you know, turn out to be really, really difficult. And so this is completely and uniquely defined. And we can analyze that mathematically. And um, we have also developed some approximations. I can talk about that a little bit later. That would be sort of the top down approach, like say for Neumann's minimax theory, that's the theoretical optimal play of games. And now we need to approximate it, put heuristics in, prune the tree, blah, blah, blah, and so on. So we can do that also with the ICSI model, but for general AI. It can also inspire those, and most of, most researchers go bottom up, right? They have their systems that try

• 57:19Speaker 01

  to make it more general, more intelligent. It can inspire in which direction to go.

• 57:25Lex Fridman?

  What do you mean by that?

• 57:26Speaker 01

  So if you have some choice to make, right? So how should I evaluate my system if I can't do cross-validation? How should I do my learning if my standard regularization doesn't work well? Yeah, so the answer is always this, we have a system which does everything that's ICSI. It's just, you know, completely in the ivory tower, completely useless from a practical point of view. But you can look at it and see, ah, yeah, maybe, you know, I can take some aspects and, you know, instead of Kolmogorov complexity, they just take some compressors which has been developed so far. And for the planning, well, we have UCT, which has also, you know, been used in Go. And at least it's inspired me a lot to have this formal definition.

• 58:11Speaker 01

  And if you look at other fields, you know, like I always come back to physics because I have a physics background. Think about the phenomenon of energy, that was a long time a mysterious concept. And at some point it was completely formalized. And that really helped a lot. And you can point out a lot of these things which were first mysterious and vague, and then they have been rigorously formalized. Speed and acceleration has been confused, right? Until it was formally defined, yeah, there was a time like this. And people, you know, often, you know, who don't have any background, you know, still confuse it.

• 58:45Speaker 01

  So, and this ICSI model or the intelligence definitions, which is sort of the dual to it, we come back to that later, formalizes the notion of intelligence uniquely and rigorously.

• 58:56Lex Fridman?

  So in a sense, it serves as kind of the light at the end of the tunnel.

• 59:00Speaker 01

  Yes, yeah.

• 59:01Lex Fridman?

  So, I mean, there's a million questions I could ask her. So maybe the kind of, okay, let's feel around in the dark a little bit. So there's been here a deep mind, but in general, been a lot of breakthrough ideas, just like we've been saying around reinforcement learning. So how do you see the progress in reinforcement learning is different? Like which subset of ICSI does it occupy? The current, like you said, the maybe the Markov assumption is made quite often in reinforcement learning. The, there's, there's other assumptions made in order to make the system work. What do you see as the difference, connection between reinforcement learning and ICSI?

• 59:44Speaker 01

  Yeah, so the major difference is that essentially all other approaches, they make stronger assumptions. So in reinforcement learning, the Markov assumption is that the next state or next observation only depends on the previous observation and not the whole history, which makes, of course, the mathematics much easier rather than dealing with histories. Of course, they profit from it also because then you have algorithms that run on current computers and do something practically useful. But for general AI, all the assumptions which are made by other approaches, we know already now they are limiting. So for instance, usually you need an ergodicity assumption in the MTP

• 1:00:27Speaker 01

  frameworks in order to learn. Ergodicity essentially means that you can recover from your mistakes and that they are not traps in the environment. And if you make this assumption, then essentially you can, you know, go back to a previous state, go there a couple of times and then learn what statistics and what this state is like and then in the long run perform well in this state, yeah. But there are no fundamental problems. But in real life, we know, you know, there can be one single action, you know, one second of being inattentive while driving a car fast, you know, can ruin the rest of my life. I can become quadriplegic or whatever. So there's no recovery anymore. So the real world is not ergodic, I always say, you know, there are traps and there

• 1:01:11Speaker 01

  are situations where you are not recover from. And very little theory has been developed for this case.

• 1:01:19Lex Fridman?

  What about, what do you see in the context of IAXI as the role of exploration? Sort of, you mentioned, you know, in the in the real world, we get into trouble when we make the wrong decisions and really pay for it. But exploration is seems to be fundamentally important for learning about this world, for gaining new knowledge. So is exploration baked in? Another way to ask it, what are the parameters of this of IAXI that can be controlled?

• 1:01:53Speaker 01

  Yeah, I say the good thing is that there are no parameters to control. Some other people try knobs to to control and you can do that. I mean, you can modify axes or that you have some knobs to play with if you want to. But the exploration is directly baked in. And that comes from the Bayesian learning and the long term planning. So these together already imply exploration. You can nicely and explicitly prove that for simple problems like so called banded problems. Where you say, to give a real world example, say you have two medical treatments, A and B.

• 1:02:37Speaker 01

  You don't know the effectiveness. You try A a little bit, B a little bit, but you don't want to harm too many patients. So you have to sort of trade off exploring, yeah. And at some point you want to explore and you can do the mathematics and figure out the optimal strategy.

• 1:02:55Speaker 01

  It's called Bayesian agents, they're also non Bayesian agents. But it shows that this Bayesian framework by taking a prior over possible worlds. Doing the Bayesian mixture, then the base optimal decision with long term planning that is important. Automatically implies exploration also to the proper extent, not too much exploration and not too little. In this very simple settings. In the AICSI model, I was also able to prove that that is a self optimizing theorem or asymptotic optimality theorem. Although they're only asymptotic, not finite time bounds.

• 1:03:27Lex Fridman?

  So it seems like the long term planning is a really important, but the long term part of the planning is really important.

• 1:03:32Speaker 01

  Yes.

• 1:03:33Lex Fridman?

  And also, I mean, maybe a quick tangent. How important do you think is removing the Markov assumption and looking at the full history? Sort of intuitively, of course, it's important, but is it like fundamentally transformative to the entirety of the problem? What's your sense of it like? Cause we all, we make that assumption quite often. It's just throwing away the past.

• 1:03:57Speaker 01

  No, I think it's absolutely crucial. The question is whether there's a way to deal with it in a more heuristic and.

• 1:04:07Speaker 01

  Still sufficiently well way. So um I have to come up with an example and fly. But you know, you have some, you know, key event in your life, you know, a long time ago, you know, in some city or something, you realize, you know, that's a really dangerous street or whatever, right? Yeah. And you want to remember that forever, right? Um, in case you come back there.

• 1:04:27Lex Fridman?

  Kind of a selective kind of memory. So you remember that all the important events in the past, but somehow selecting the importance is.

• 1:04:34Speaker 01

  That's very hard. Yeah. And I'm not concerned about, you know, just storing the whole history. Just you can calculate, um, you know, human life says thirty or a hundred years doesn't matter, right? Um, uh, how much uh data comes in through the vision system and the auditory system? You compress it a little bit, um, in this case, lossily and store it. Um, we are soon in the means of just storing it, yeah. But um, you still need to the selection um for the planning part and the compression for the understanding part. The raw storage, I'm really not concerned about. Um, and I think we should just store, if we develop an agent, um, preferably just just store all the interaction history.

• 1:05:16Speaker 01

  And then um, you build, of course, models on top of it and you compress it. Um, and you are selective, but occasionally you go back to the old data and reanalyze it um based on your new experience you have. You know, sometimes you you're in school, you learn all these things you think is totally useless and you know, much later you realize, oh, they were not, you know, so useless as you thought.

• 1:05:39Lex Fridman?

  I'm looking at you linear algebra. Right. So maybe let me ask about objective functions because that rewards. It seems to be an important part. The rewards are kind of given to the system. Uh, for a lot of people, the um, the specification of the objective function i- is a key part of intelligence. The the agent itself figuring out what is important. What what do you think about that? Is it possible within IAXI framework to yourself discover the reward based on which you should operate?

• 1:06:22Speaker 01

  Okay, that will be a long answer. Um, so, um, and that is a very interesting question and I, uh, I'm asked a lot about this question. Where do the rewards come from? And, uh, that depends, yeah? So, um, and then, you know, I give you now a couple of answers. So if we want to build agents, now let's start simple. So let's assume we want to build an agent based on the ICSI, um, model which performs a particular task. Let's start with something super simple like, I mean, super simple like playing chess, yeah, or Go or something, yeah? Then you just, you know, the reward is, you know, winning the game is plus one, losing the game is minus one. Done, uh, you apply this agent, if you have enough compute, you let it self-play.

• 1:07:07Speaker 01

  Um, and it will learn the rules of the game, will play perfect chess after some while, problem solved, okay? So if you have more complicated problems, um, then, um, you may believe that you have the right reward, but it's not. So a nice cute example is elevator control, that is also in Rich Sutton's book, which is a great book, by the way. Um, so you control the elevator and you think, well, maybe the reward should be coupled to how long people wait in front of the elevator. You know, long wait is bad. You program it and you do it. And what happens is the elevator eagerly picks up all the people but never drops them off.

• 1:07:48Speaker 01

  So then you realize, oh, maybe the time in the elevator also counts, so you minimize the sum, yeah? Yeah. And the elevator does that, but never picks up the people in the 10th floor, in the top floor, because in expectation it's not worth it, just let them stay. Yeah.

• 1:08:03Speaker 01

  So, um, so even in apparently simple problems, you can make mistakes, yeah? And that's, um, um, what in, in more serious contexts, say, AGI safety researchers consider. So now let's go back to, um, general agents. So assume we want to build an agent which is generally useful to humans, yes? We have a household robot, yeah? Um, and it should do all kinds of tasks. So in this case. The human should give the reward on the fly. I mean, maybe it's pre-trained in the factory and that there's some sort of internal reward for, you know, the battery level or whatever, yeah, but um so it, you know, it does the dishes badly, you know, you punish the robot, it does it good, you reward the robot and then to train it to a new task, you know, like a child, right?

• 1:08:46Speaker 01

  So um you need the human in the loop if you want a system which is useful to the human. And as long as this agent stays sub-human level, um that should work reasonably well, apart from, you know, these examples. I mean, it becomes critical if they become, you know, on a human level. It's like with children, small children, you have reasonably well under control, they become older, um the reward technique doesn't work so well anymore. Um so then finally, um so this would be agents which are just, you could say slaves to the humans, yeah? So if you are more ambitious and just say, we want to build a new species of intelligent beings, we put them on a new planet and we want them to develop this planet or whatever.

• 1:09:30Speaker 01

  So we don't give them any reward. So what could we do? And you could try to, you know, come up with some reward functions like, you know, it should maintain itself, the robot, it should um maybe multiply, build more robots, right? Um and um you know, maybe well, all kinds of things which you find useful, but that's pretty hard, right? You know, what what does self-maintenance mean? You know, what does it mean to build a copy? Should it be exact copy, an approximate copy? And so that's really hard. But um Laurent also, also at DeepMind, uh developed a beautiful model. So he just took the AICs model and coupled the rewards to information gain.

• 1:10:12Speaker 01

  So he said the reward is proportional to how much the agent had learned about the world. And you can rigorously formally uniquely define that in terms of other versions, okay? So if you put that in, you get a completely autonomous agent. And actually, interestingly, for this agent, we can prove much stronger result than for the general agent, which is also nice. And if you let this agent loose, it will be, in a sense, the optimal scientist. It is absolutely curious to learn as much as possible about the world. And of course, it will also have a lot of instrumental goals, right? In order to learn, it needs to at least survive, right? That that agent is not good for anything. So it needs to have self-preservation. And if it builds small helpers acquiring more information, it will do that.

• 1:10:56Speaker 01

  Yeah, if exploration, space exploration or whatever is necessary, right? To gathering information and develop it. So it has a lot of instrumental goals following on this information gain. And this agent is completely autonomous of us. No rewards necessary anymore.

• 1:11:13Lex Fridman?

  Yeah, of course, it could find a way to game the concept of information and get stuck in that library that you mentioned beforehand with a with a very large number of books.

• 1:11:26Speaker 01

  The first agent had this problem. It would get stuck in front of an old TV screen, which has just had wide noise.

• 1:11:32Lex Fridman?

  Yeah, wide noise, yeah.

• 1:11:34Speaker 01

  But the second version can deal with at least stochasticity. Well.

• 1:11:39Lex Fridman?

  Yeah. What about curiosity, this kind of word? Curiosity, creativity? Is that kind of the reward function being of getting new information? Is that similar to idea of kind of injecting exploration for its own sake inside the reward function? Do you find this at all appealing, interesting?

• 1:12:02Speaker 01

  I think that's a nice definition. Curiosity is reward. Sorry, curiosity is exploration for its own sake. Yeah, I would accept that. But most curiosity, well, in humans and especially in children, yeah, is not just for its own sake, but for actually learning about the environment and for behaving better. So I would, I think most curiosity is tied in the end towards performing better.

• 1:12:32Lex Fridman?

  Well, OK, so if intelligent systems need to have this reward function, let me, you're an intelligent system.

• 1:12:41Lex Fridman?

  Currently passing the Turing test quite effectively. What's the reward function of our human intelligence existence? What's the reward function that Marcus Hutter is operating under?

• 1:12:55Speaker 01

  Okay, to the first question, the biological reward function is to survive and to spread. And very few humans are able to overcome this biological reward function. But we live in a very nice world where we have lots of spare time and can still survive and spread. So we can develop arbitrary other interests, which is quite interesting.

• 1:13:20Lex Fridman?

  On top of that.

• 1:13:21Speaker 01

  On top of that, yeah. But the survival and spreading is, I would say, the goal or the reward function of humans, the core one.

• 1:13:32Lex Fridman?

  I like how you avoided answering the second question, which a good intelligence system would.

• 1:13:37Speaker 01

  So my.

• 1:13:38Lex Fridman?

  Your own meaning of life and the reward function.

• 1:13:41Speaker 01

  My own meaning of life and reward function is to find an AGI to build it.

• 1:13:48Lex Fridman?

  Beautifully put. Okay. Let's dissect ICSI even further. So one of the assumptions is infinity keeps creeping up everywhere.

• 1:14:00Lex Fridman?

  Which, what are your thoughts on bounded rationality? And the nature of our existence and intelligent systems is that we're operating always under constraints, under limited time, limited resources. How does that, how do you think about that within the ICSI framework? Within trying to create an AGI system that operates under these constraints?

• 1:14:24Speaker 01

  Yeah, that is one of the criticisms about ICSI, that it ignores computation completely. And some people believe that intelligence is inherently tied to what's bounded resources.

• 1:14:37Lex Fridman?

  What do you think on this one point? Do you think it's, do you think the bounded resources are fundamental to intelligence?

• 1:14:45Speaker 01

  I would say that an intelligent notion which ignores computational limits is extremely useful. A good intelligent notion which includes these resources would be even more useful, but we don't have that yet. And so look at other fields outside of computer science. Computational aspects never play a fundamental role. You develop biological models for cells, something in physics, these theories, I mean, become more and more crazy and harder and harder to compute. Well, in the end, of course, we need to do something with this model, but this more nuisance than a feature. And I'm sometimes wondering if artificial intelligence would not sit in a computer science department, but in a philosophy

• 1:15:30Speaker 01

  department, then this computational focus would be probably significantly less. I mean, think about the induction problem is more in the philosophy department. There's a version of paper cares about, you know, how long it takes to compute the answer. That is completely secondary. Of course, once we have figured out the first problem, so intelligence without computational resources, then the next and very good question is, could we improve it by including computational resources, but nobody was able to do that so far in an even halfway satisfactory manner?

• 1:16:06Lex Fridman?

  I like that, that in the long run, the right department to belong to is philosophy. That's, it's, it's actually quite a deep idea of, or even to, at least to think about big picture philosophical questions, big picture questions, even in the computer science department. But you've mentioned approximation, sort of, there's a lot of infinity, a lot of huge resources needed. Are there approximations to ICSI that, within the ICSI framework, that are useful?

• 1:16:37Speaker 01

  Yeah, we have developed a couple of approximations. Um, and um, what we do there is that the Solomon induction part, which was, you know, find the shortest program describing your data, which just replaces by standard data compressors, right? And the better compressors get, you know, the better this part will become. Um, we focused on a particular compressor called Context Reweighting, which is uh pretty amazing, not so well known. And has beautiful theoretical properties, also works reasonably well in practice. So we use that for the approximation of the induction and the learning and the prediction part.

• 1:17:15Speaker 01

  And um for the planning part, um, we essentially just took the ideas from a computer girl from two thousand and six. It was Java, Sepespari, also now DeepMind, uh, who developed the so-called um UCT algorithm, upper confidence bound for trees algorithm on top of the Monte Carlo tree search. So we approximate this planning part um by sampling. And um it's successful on some small toy problems. Um, we don't want to lose the generality, right? And that's sort of the handicap, right? If you want to be general, you have to give up something.

• 1:17:56Speaker 01

  So, but this single agent was able to play, you know, small games like Kuhn poker and tic-tac-toe and um and and even Pac-Man. Um, um, in the same architecture, no change. The agent doesn't know the rules of the game, really nothing, all by self or by player with these environments.

• 1:18:17Lex Fridman?

  So, um, Jürgen Schmidhuber proposed something called Ghetto Machines, which is a self-improving program that rewrites its own code.

• 1:18:27Lex Fridman?

  What, sort of mathematically or philosophically, what's the relationship in your eyes, if you're familiar with it, between IAXI and the Ghetto Machines?

• 1:18:35Speaker 01

  Yeah, familiar with it, he developed it while I was in his lab. Yeah, so the Gödel machine, to explain it briefly, you give it a task. It could be a simple task as you know, finding prime factors in numbers, right? You can formally write it down. There's a very slow algorithm to do that. Just try all the factors, yeah? Or play chess, right? Optimally, you write the algorithm to minimize to the end of the game. So you write down what the Gödel machine should do. Then it will take part of its resources to run this program and other part of the sources to improve this program. And when it finds an improved version, which provably computes the same answer. So that's the key part, yeah?

• 1:19:19Speaker 01

  It needs to prove by itself that this change of program still satisfies the original specification. And if it does so, then it replaces the original program by the improved program. And by definition, it does the same job, but just faster. Okay? And then, you know, it proves over it and over it. And it's developing a way that all parts of this Gödel machine can self-improve, but it stays provably consistent with the original specification. So from this perspective, it has nothing to do with ICSI. But if you would now put ICSI as the starting axioms in, it would run ICSI, but you know, that takes forever.

• 1:20:02Speaker 01

  But then if it finds a provable speedup of ICSI, it would replace it by this and this and this. And maybe eventually it comes up with a model which is still the ICSI model. It cannot be, I mean, just for the knowledgeable reader, ICSI is incomputable. And there can prove that therefore there cannot be a computable, exact algorithm computer. There needs to be some approximations. And this is not dealt with the Gödel machine. So you have to do something about it. But there's the ICSI model, which is finitely computable, which we could put in.

• 1:20:34Lex Fridman?

  Which part of ICSI is non-computable?

• 1:20:36Speaker 01

  The Solomon of induction part.

• 1:20:38Lex Fridman?

  The induction, okay, so.

• 1:20:39Speaker 01

  But there is ways of getting computable approximations of the ICSI model.

• 1:20:46Speaker 01

  So then it's at least computable. It is still way beyond any resources anybody will ever have. But then the Gödel machine could sort of improve it further and further in an exact way.

• 1:20:55Lex Fridman?

  So is it theoretically possible that the Gödel machine process could improve? Isn't...

• 1:21:04Lex Fridman?

  Isn't AIC already optimal?

• 1:21:09Speaker 01

  It is optimal in terms of the reward collected over its interaction cycles, but it takes infinite time to produce one action. And the world, you know, continues whether you want it or not, yeah? So the model is assuming you had an oracle, which, you know, solved this problem. And then in the next hundred milliseconds or the reaction time you need gives the answer, then AIC is optimal.

• 1:21:35Speaker 01

  It's optimal in sense of data, also from learning efficiency and data efficiency, but not in terms of computation time.

• 1:21:43Lex Fridman?

  And then the GPT machine in theory, but probably not provably could make it go faster.

• 1:21:48Speaker 01

  Yes.

• 1:21:49Lex Fridman?

  Okay. Interesting. Those two components are super interesting. The sort of the the perfect intelligence combined with self-improvement. Sort of provable self-improvement in the sense you're always like you're always getting the correct answer and you're improving.

• 1:22:07Speaker 01

  Yep.

• 1:22:07Lex Fridman?

  Beautiful ideas. Okay. So you've also mentioned that different kinds of things in in the chase of solving this reward sort of optimizing for the goal. Interesting human things could emerge. So is there a place for consciousness within AIC? Where where does maybe you can comment? Because I suppose we humans are just another instantiation of AIC agents and we seem to have consciousness.

• 1:22:38Speaker 01

  You say humans are an instantiation of an AIC agent?

• 1:22:40Lex Fridman?

  Yes.

• 1:22:41Speaker 01

  Oh, that would be amazing. But I think that's not true even for the smartest and most rational humans. I think maybe we are very crude approximations.

• 1:22:50Lex Fridman?

  Interesting. I mean, I tend to believe, again, I'm Russian. So I tend to believe our flaws are part of the optimal. So the we tend to laugh off and criticize our flaws. And I tend to think that that's actually close to optimal behavior.

• 1:23:08Speaker 01

  But some flaws, if you think more carefully about it, are actually not flaws. Yeah, but I think there are still enough flaws.

• 1:23:16Lex Fridman?

  I don't know. It's unclear. As a student of history, I think all the suffering that we've endured as a civilization, it's possible that that's the optimal amount of suffering we need to endure to minimize long-term suffering.

• 1:23:32Speaker 01

  That's your Russian background, I think.

• 1:23:34Lex Fridman?

  That's the Russian. Whether humans are or not instantiations of an ICSI agent, do you think there's a consciousness of something that could emerge in a computational formal framework like ICSI?

• 1:23:47Speaker 01

  Let me also ask you a question. Do you think I'm conscious?

• 1:23:54Lex Fridman?

  That's a good question. You're

• 1:23:58Lex Fridman?

  that that tie is confusing me, but I think so.

• 1:24:01Speaker 01

  You think that makes me unconscious because it strangles me?

• 1:24:04Lex Fridman?

  If if an agent were to solve the imitation game posed by Turing, I think that would be dressed similarly to you. That because there's a there's a kind of flamboyant, interesting, complex behavior pattern that sells that you're human and you're conscious. But why do you ask?

• 1:24:23Speaker 01

  Was it a yes or was it a no?

• 1:24:25Lex Fridman?

  Yes, I think you're I think you're conscious, yes.

• 1:24:29Speaker 01

  Yeah. So and you explain sort of somehow why, but you infer that from my behavior, right? You can never be sure about that. And I think the same thing will happen with any intelligent agent we develop, if it behaves in a way sufficiently close to humans, or maybe even not humans. I mean, you know, maybe a dog is also sometimes a little bit self-conscious, right? So so if it behaves in a way where we attribute typically consciousness, we would attribute consciousness to these intelligent systems and, you know, ICSI probably in particular, that of course doesn't answer the question whether it's really conscious. And that's the, you know, the big hard problem of consciousness. You know, maybe I'm a zombie. I mean, not the

• 1:25:13Speaker 01

  movie zombie, but the philosophical zombie.

• 1:25:16Lex Fridman?

  Is to you the display of consciousness close enough to consciousness from a perspective of AGI that the distinction of the hard problem of consciousness is not an interesting one?

• 1:25:28Speaker 01

  I think we don't have to worry about the consciousness problem, especially the hard problem for developing AGI. I think, you know, we progress. At some point, we have, you know, solved all the technical problems and this system will behave intelligent and then super intelligent. And this consciousness will emerge. I mean, definitely it will display behavior which we will interpret as conscious. And then it's a philosophical question. Did this consciousness really emerge? Or is it a zombie which just, you know, fakes everything? We still don't have to figure that out, although it may be interesting. At least from a philosophical point of view, it's very interesting, but it may also be sort of practically interesting. You know, there's some people saying, you know, if it's just faking consciousness and feelings, you know,

• 1:26:13Speaker 01

  then we don't need to be concerned about, you know, rights. But if it's real conscious and has feelings, then we need to be concerned. Yeah.

• 1:26:23Lex Fridman?

  I can't wait till the day where AI systems exhibit consciousness because it'll truly be some of the hardest ethical questions of what we do with that.

• 1:26:33Speaker 01

  It is rather easy to build systems which people ascribe consciousness. And I give you an analogy. I mean, remember, maybe it was before you were born, the Tamagotchi?

• 1:26:44Lex Fridman?

  Yeah, before you were born. How dare you, sir?

• 1:26:48Speaker 01

  Why? That's the, that you're young, right?

• 1:26:50Lex Fridman?

  Yes, that's good. Thank you, thank you very much. But I was also in the Soviet Union. We didn't have, uh, we didn't have any of those fun things.

• 1:26:58Speaker 01

  But you have heard about this Tamagotchi, which was, you know, really, really primitive. Actually, for the time it was, and you know, you could raise, you know, this, and, and, and, and kids got so attached to it and, you know, didn't want to let it die. And would have probably, if we would have asked, you know, the children, you know, do you think this Tamagotchi is conscious?

• 1:27:17Lex Fridman?

  They would have said yes.

• 1:27:18Speaker 01

  Yes, I would, yes.

• 1:27:19Lex Fridman?

  I think that's kind of a beautiful thing, actually. Because that consciousness, ascribing consciousness seems to create a deeper connection.

• 1:27:27Speaker 01

  Yep.

• 1:27:28Lex Fridman?

  Which is a powerful thing. But we'll have to be careful on the ethics side of that. Well, let me ask about the AGI community broadly. You kind of represent some of the most serious work on AGI, at least earlier. DeepMind represents a serious work on AGI these days. But why, in your sense, is the AGI community so small or has been so small until maybe DeepMind came along? Like, why aren't more people seriously working on human-level and superhuman-level intelligence from a formal perspective?

• 1:28:05Speaker 01

  Okay, from a formal perspective, that's sort of, you know, an extra point. So I think there are a couple of reasons. I mean, AI came in waves, right? You know, AI winters and AI summers, and then there were big promises which were not fulfilled. And people got disappointed. And that narrow AI solving particular problems, which seemed to require intelligence, was always, to some extent, successful. And there were improvements, small steps. And if you build something which is, you know, useful for society or industrial useful, then there's a lot of funding. So I guess it was in parts the money,

• 1:28:47Speaker 01

  which drives people to develop specific systems solving specific tasks. But you would think that, you know, at least in university, you should be able to do ivory tower research. And that was probably better a long time ago. But even nowadays, there's quite some pressure of doing applied research or translational research. And, you know, it's harder to get grants as a theorist. So that also drives people away. It's maybe also harder attacking the general intelligence problem. So I think enough people, I mean, maybe a small number, were still interested in formalizing intelligence and thinking of general intelligence.

• 1:29:30Speaker 01

  But, you know, not much came up, right? Or not much great stuff came up.

• 1:29:37Lex Fridman?

  So what do you think? We talked about the formal, big light at the end of the tunnel. But from the engineering perspective, what do you think it takes to build an AGI system? Is that in? I don't know if that's a stupid question or a distinct question from everything we've been talking about IAXE, but what do you see as the steps that are necessary to take to start to try to build something?

• 1:30:00Speaker 01

  So you want a blueprint now and then you go off and do it?

• 1:30:03Lex Fridman?

  That's the whole point of this conversation, trying to squeeze that in there. Now, is there, I mean, what's your intuition? Is it, is it in the robotics space or something that has a body and tries to explore the world? Is in the reinforcement learning space? Like the efforts with AlphaZero and AlphaStar that are kind of exploring how you can solve it through in the simulation in the gaming world? Is there stuff in sort of the, all the transformer work in natural language processing? Sort of maybe attacking the open domain dialogue? Like what, where do you see a promising pathways?

• 1:30:39Speaker 01

  Let me pick the embodiment, maybe. So embodiment is important, yes and no. I don't believe that we need a physical robot walking or rolling around, interacting with the real world in order to achieve AGI. And I think it's more of a distraction probably than helpful. It's sort of confusing the body with the mind. For industrial applications or near-term applications, of course, we need robotters for all kinds of things, yeah? But for solving the big problem, at least at this stage,

• 1:31:23Speaker 01

  I think it's not necessary. But the answer is also yes, that I think the most promising approach is that you have an agent. And you know, that can be a virtual agent, you know, in a computer, interacting with an environment, possibly in a 3D simulated environment like in many computer games. And you train and learn the agent. Even if you don't intend to later put it sort of, you know, this algorithm in a, in a robot brain and leave it forever in the virtual reality, getting experience in a

• 1:31:59Speaker 01

  Although it's just simulated 3D world, is possibly, and I say possibly, important to understand things on a similar level as humans do, especially if the agent or primarily if the agent wants, needs to interact with the humans, right? And if you talk about objects on top of each other in space and flying and cars and so on, and the agent has no experience with even virtual 3D worlds, it's probably hard to grasp. But if you develop an abstract agent, say we take the mathematical path and we just want to build an agent which can prove theorems and becomes a better and better mathematician, then this agent needs to be able to reason in very abstract spaces

• 1:32:43Speaker 01

  and then maybe sort of putting it into 3D environment, simulated or is even harmful. It should sort of, you put it in, I don't know, an environment which it creates itself or so.

• 1:32:54Lex Fridman?

  It seems like you have an interesting, rich, complex trajectory through life in terms of your journey of ideas. So it's interesting to ask what books, technical fiction, philosophical, books, ideas, people had a transformative effect. Books are most interesting because maybe people could also read those books and see if they could be inspired as well.

• 1:33:17Speaker 01

  Yeah, luckily I asked books and not singular book. It's very hard and I tried to pin down one book.

• 1:33:25Lex Fridman?

  Yeah.

• 1:33:25Speaker 01

  And I can't do that at the end. So the most, the books which were most transformative for me or which I can most highly recommend to people interested in AI.

• 1:33:39Lex Fridman?

  Both perhaps.

• 1:33:40Speaker 01

  Yeah, both, yeah, yeah. I would always start with Russell and Norbert, Artificial Intelligence, A Modern Approach. That's the AI Bible. It's an amazing book. It's very broad. It covers, you know, all approaches to AI. And even if you focus on one approach, I think that is the minimum you should know about the other approaches out there. So that should be your first book.

• 1:34:03Lex Fridman?

  Fourth edition should be coming out soon.

• 1:34:05Speaker 01

  Oh, okay, interesting.

• 1:34:07Lex Fridman?

  There's a deep learning chapter now, so there must be, written by Ian Goodfellow, okay.

• 1:34:13Speaker 01

  And then the next book I would recommend, the reinforcement learning book by Satya and Barto. That's a beautiful book.

• 1:34:23Speaker 01

  If there's any problem with the book, it makes RL. Feel and look much easier than it actually is. It's very gentle book. It's very nice to read, the exercises to do. You can very quickly, you know, get some RL systems to run. You know, very toy problems, but um it's a lot of fun. And you in very, in a couple of days, you feel, you know, you know what RL is about. But it's much harder than the book. Yeah. Come on now, it's an awesome book. Yeah, that is, yeah. And um maybe, I mean, there's so many books out there. If you like the information theoretic approach, then there's Komo-Golf Complexity uh by Leon Vitani. But uh probably, you know, some, some short article is enough.

• 1:35:08Speaker 01

  Uh you don't need to read the whole book, but it's, it's a great book. And if we have to mention one all time favorite book, so different flavor. That's a book which is used in the International Backlerat for high school students in several countries. That's from Nicholas Alchen, Theory of Knowledge, second edition or first, not the third, please. The third one, they put, they took out all the fun. Okay. So uh this asked all the interesting, or to me, interesting philosophical questions about how we acquire knowledge from all perspectives, you know, from math, from art, from physics. Um and ask, how can we know um anything?

• 1:35:53Speaker 01

  And the book is called Theory of Knowledge. From which, it's almost like a philosophical exploration of uh how we get knowledge from anything. Yes, yeah, I mean, can religion tell us, you know, about something about the world? Can science tell us something about the world? Can mathematics, or is it just playing with symbols? Um and, and you know, and it's open ended questions. And I mean, it's for high school students, so they have then resources from Hitchhiker's Guide to the Galaxy and from Star Wars and the Chicken Across the Road, yeah. And it's, it's, it's fun to read and but it's also quite deep.

• 1:36:25Lex Fridman?

  If you could live one day of your life over again, because it made you truly happy, or maybe like we said with the books, it was truly transformative. What day, what moment would you choose? Does something pop into your mind?

• 1:36:39Speaker 01

  Does it need to be a day in the past? Or can it be a day in the future?

• 1:36:43Lex Fridman?

  Well, spacetime is an emergent phenomenon, so it's all the same anyway.

• 1:36:47Speaker 01

  Okay. Okay, from the past.

• 1:36:51Lex Fridman?

  You're really good at saying from the future. I love it.

• 1:36:54Speaker 01

  No, I will tell you from the future.

• 1:36:55Lex Fridman?

  Okay.

• 1:36:56Speaker 01

  So from the past, I would say when I discovered my axi model. I mean, it was not in one day, but it was one moment where I realized come a go of complexity and didn't even know that it existed, but I rediscovered sort of this compression idea myself. But immediately I knew I can't be the first one, but I had this idea. And then I knew about sequential decision tree and I knew if I put it together, this is the right thing. And yeah, still, when I think back about this moment, I'm I'm super excited about it.

• 1:37:29Lex Fridman?

  Was there, was there any more details and context that moment? Did an apple fall on your head? Where, so like if you look at Ian Goodfellow talking about gans, there was beer involved. Is there, is there some more context of what sparked your thought? Or was it just?

• 1:37:48Speaker 01

  No, it was much more mundane. So I worked in this company. So in this sense, the four and a half years was not completely wasted. So, and I worked on an image interpolation problem. And I developed a quite neat new interpolation techniques and they got patented. And then, you know, which happens quite often. I got sort of overboard and thought about, you know, yeah, that's pretty good, but it's not the best. So what is the best possible way of doing interpolation? And then I thought, yeah, you, you, you want the simplest picture, which is if you coarse grain it, recovers your original picture. And then I, you know, thought about the simplicity concept more in quantitative terms and, you know, then everything developed.

• 1:38:32Lex Fridman?

  And somehow the full beautiful mix of also being a physicist and thinking about the big picture of it then led you to probably

• 1:38:40Speaker 01

  Yeah, yeah. So as a physicist, I was probably trained not to always think in computational terms, you know, just ignore that and think about the fundamental properties which you want to have.

• 1:38:51Lex Fridman?

  So what about if you could really live one day in the future? What would that be?

• 1:38:57Speaker 01

  When I solve the AGI problem.

• 1:39:00Speaker 01

  In practice, in practice. So in theory, I have solved it with the ICSI model, but in practice.

• 1:39:06Speaker 01

  And then I ask the first question.

• 1:39:08Lex Fridman?

  What would be the first question?

• 1:39:10Speaker 01

  What's the meaning of life?

• 1:39:13Lex Fridman?

  I don't think there's a better way to end it. Thank you so much for talking today. It's a huge honor to finally meet you.

• 1:39:18Speaker 01

  Yeah, thank you too. It was a pleasure of mine, so too.

• 1:39:22Lex Fridman?

  Thanks for listening to this conversation with Marcus Hutter. And thank you to our presenting sponsor, Cash App. Download it, use code LEXPODCAST, you'll get $10 and $10 will go to FIRST, an organization that inspires and educates young minds to become science and technology innovators of tomorrow. If you enjoy this podcast, subscribe on YouTube, give it five stars on Apple Podcasts, support on Patreon, or simply connect with me on Twitter at Lex Fridman. And now let me leave you with some words of wisdom from Albert Einstein. The measure of intelligence is the ability to change. Thank you for listening and hope to see you next time.