François Chollet: Measures of Intelligence | Lex Fridman Podcast #120

Lex Fridman 0:00 The following is a conversation with Francois Chollet, his second time on the podcast. He's both a world-class engineer and a philosopher in the realm of deep learning and artificial intelligence. This time, we talk a lot about his paper titled On the Measure of Intelligence that discusses how we might define and measure general intelligence in our computing machinery. Quick summary of the sponsors: Babbel, MasterClass, and Cash App. Click the sponsor links in the description to get a discount and to support this podcast. As a side note, let me say that the serious, rigorous, scientific study of artificial general intelligence is a rare thing. The mainstream machine learning community works on very narrow AI with very narrow benchmarks. This is very good for incremental and sometimes big incremental progress. On the other hand, the outside-the-mainstream, renegade, you could say, AGI community works on approaches that verge on the philosophical and even the literary without big public benchmarks. Walking the line between the two worlds is a rare breed, but it doesn't have to be. I ran the AGI series at MIT as an attempt to inspire more people to walk this line. DeepMind and OpenAI for a time, and still on occasion, walk this line. Francois Chollet does as well. I hope to also. It's a beautiful dream to work towards and to make real one day. If you enjoy this thing, subscribe on YouTube, review it with five stars on Apple Podcasts, follow on Spotify, support on Patreon, or connect with me on Twitter @LexFridman. As usual, I'll do a few minutes of ads now and no ads in the middle. I try to make these interesting, but I give you timestamps so you can skip. But still, please do check out the sponsors by clicking the links in the description. It's the best way to support this podcast. This show is sponsored by Babbel, an app and website that gets you speaking in a new language within weeks. Go to babbel.com and use code LEX to get three months free. They offer 14 languages, including Spanish, French, Italian, German, and yes, Russian. Daily lessons are 10 to 15 minutes, super easy, effective, designed by over 100 language experts. Let me read a few lines from the Russian poem, (Russian) , by Alexander Blok, that you'll start to understand if you sign up to Babbel. Ночь. Улица. Фонарь. Аптека. Бессмысленный и тусклый свет. Живи еще хоть четверть века. Все будет так. Исхода нет. Now, I say that you'll start to understand this poem because Russian starts with a language and ends with vodka. Now, the latter part is definitely not endorsed or provided by Babbel and will probably lose me the sponsorship, although it hasn't yet. But once you graduate with Babbel, you can enroll in my advanced course of late-night Russian conversation over vodka. No app for that yet. So get started by visiting babbel.com and use code LEX to get three months free. This show is also sponsored by MasterClass. Sign up at masterclass.com/lex to get a discount and to support this podcast. When I first heard about MasterClass, I thought it was too good to be true. I still think it's too good to be true. For $180 a year, you get an all-access pass to watch courses from, to list some of my favorites, Chris Hadfield on space exploration; hope to have him on this podcast one day. Neil deGrasse Tyson on scientific thinking and communication. Neil too. Will Wright, creator of SimCity and Sims on game design. Carlos Santana on guitar. Garry Kasparov on chess. Daniel Negreanu on poker and many more. Chris Hadfield explaining how rockets work and the experience of being launched to space alone is worth the money. By the way, you can watch it on basically any device. Once again, sign up at masterclass.com/lex to get a discount and to support this podcast. This show, finally, is presented by Cash App, the number one finance app in the App Store. When you get it, use code LEXPODCAST. Cash App lets you send money to friends, buy Bitcoin, and invest in the stock market with as little as $1. Since Cash App allows you to send and receive money digitally, let me mention a surprising fact related to physical money. Of all the currency in the world, roughly 8% of it is actually physical money. The other 92% of the money only exists digitally, and that's only going to increase. So again, if you get Cash App from the App Store or Google Play and use code LEXPODCAST, you get 10 bucks. And Cash App will also donate $10 to FIRST, an organization that is helping to advance robotics and STEM education for young people around the world. And now, here's my conversation with Francois Chollet.

Lex Fridman 5:05 What philosophers, thinkers, or ideas had a big impact on you growing up? And today?

François Chollet 5:10 So one author that had a big impact, uh, on me when I, I read, uh, his books as a teenager was Jean Piaget, who is a Swiss psychologist, is considered, uh, to be the father of developmental psychology. And he has a large body of work about, um, basically how intelligence, uh, develops, uh, in children. And so, uh, it's very old work, like most of it is from the 1930s, 1940s. Uh, so it's not quite up-to-date. Uh, it's actually superseded by many, uh, newer developments in developmental psychology. But to me, it was, it was very, uh, very interesting, very striking, and actually shaped the early ways in which, uh, I started to think about the mind and the development of intelligence as a teenager.

Lex Fridman 5:56 His actual ideas or the way he thought about it or just the fact that you could think about the developing mind at all?

François Chollet 6:01 I guess both. Jean Piaget is, uh, the author that's really introduced me to the notion that intelligence and the mind is something that you construct throughout, throughout your life and that you, uh, that children, uh, uh, construct it in stages. And I thought that was a very interesting idea which is, you know, of course, very relevant, uh, to AI, to building artificial minds....

François Chollet 6:23 another book that I, I read around the same time that had a big impact on me, uh, and, and there was actually a, a little bit of overlap with Jean Piaget as well, and I read it around the same time, uh, is, uh, Jeff Hawkins' On Intelligence-

Lex Fridman 6:37 Hmm.

François Chollet 6:38 ... which is a classic. And he, he has this vision of the mind as a multi-scale hierarchy of temporal prediction modules, and these ideas really resonated with me, like the, the notion of, uh, a, a modular hierarchy, um, of, you know, potentially, um, of compression functions or prediction functions. I thought it was really, really interesting, uh, and it sh- shaped, uh, uh, the way I started thinking about how to build minds.

Lex Fridman 7:09 The hi- the hierarchical nature, the which aspect? Also, he's a neuroscientist, so he was thinking-

François Chollet 7:15 Yes.

Lex Fridman 7:16 ... of actual... He was basically talking about how our mind works.

François Chollet 7:20 Yeah, the notion that cognition is prediction was an idea that was kinda new to me at the time, and that I- that I really loved at the time. And, yeah, and the notion that, uh, th- that there are multiple scales of processing, uh, in the brain.

Lex Fridman 7:35 The hierarchy?

François Chollet 7:36 Yes. Uh-

Lex Fridman 7:36 This was before-

François Chollet 7:37 ... hierar-

Lex Fridman 7:37 ... deep learning?

François Chollet 7:38 These ideas of hierarchies in AI have been around for a long time. Uh, even before On Intelligence, I mean, they've, they've been around since the 1980s. Um, and yeah, that was before deep learning, but of course, I, I think these ideas really found, uh, uh, their practical implementation in deep learning.

Lex Fridman 7:58 What about the memory side of things? I think he was talking about knowledge representation. Do you think about memory a lot? One way you could think of neural networks as a, is a kind of memory. You're memorizing things, but it doesn't seem to be the kind of memory that's in our brains, or it doesn't have the same rich complexity and long-term nature that's in our brains.

François Chollet 8:20 Yes. The brain is more of a sparse access memory, so that you can actually retrieve, um, very precisely, like bits of your experience.

Lex Fridman 8:30 The retrieval aspect, you can like introspect, you can ask yourself questions, I guess.

François Chollet 8:36 Yes. You can program your own memory, and language is actually, uh, the tool you use to do that. I think language is a kind of a operating system for the mind, and you use language, uh... Well, one of the uses of language is as a query that you run over your own memory. You use words as keys to retrieve specific experiences or specific concepts, specific thoughts. Like language is a way you store thoughts, not just in writing in the, in the physical world, but also in your own mind, and it's also how you retrieve them. Like imagine if you didn't have language, then you would have to... You would not have really have, um, a self internally triggered, uh, way of retrieving past thoughts. You would have to rely on external experiences. For instance, you, you see a specific sight, you smell a specific smell, and that brings up memories, but you would not really have a, a, a way to deliberately, deliberately access these memories with our language.

Lex Fridman 9:32 Well, the interesting thing you mentioned is you can also program th- the memory. You can change it probably with language.

François Chollet 9:40 Yeah, using language. Yes.

Lex Fridman 9:41 Well, let me ask you a Chomsky question, which is like... First of all, do you think language is like fundamental? Like, uh, there's turtles. What's at the bottom of the turtles? They don't go... It can't be turtles all the way down. Is language at the bottom of cognition of everything? Is like language the fundamental aspect of like what it means to be a thinking thing?

François Chollet 10:10 No, I don't think so. I think language is-

Lex Fridman 10:12 You disagree with Noam Chomsky?

François Chollet 10:14 Yes.

Lex Fridman 10:15 (laughs)

François Chollet 10:15 I think language is a layer on top of cognition. So-

Lex Fridman 10:18 Okay.

François Chollet 10:18 ... it i- it is fundamental to cognition in the sense that, to, to use a computing metaphor, I see language as the operating system, uh, of the brain, of the-

Lex Fridman 10:28 Gotcha.

François Chollet 10:28 ... human mind.

Lex Fridman 10:29 Yeah.

François Chollet 10:29 And the operating system, you know, is a layer on top of the computer. The computer exists before the operating system, but the operating system is how you make it truly useful.

Lex Fridman 10:39 And the operating system is most likely Windows, not, not Linux, 'cause it's, um... Language is messy.

François Chollet 10:46 Yeah, it's messy, and it's, uh, it's, um, pretty difficult to, uh, uh, uh, inspect it, introspect it.

Lex Fridman 10:53 How do you think about language? Like, uh, we use actually sort of human interpretable language, but is there something like, uh, deeper that's closer to like, like logical type of statements? Um, like... Yeah, what is the nature o- of language do you think? Like is there something deeper than like the syntactic rules we construct? Is there something that doesn't require utterances or, uh,-

François Chollet 11:24 So a-

Lex Fridman 11:24 ... writing or so on?

François Chollet 11:25 ... are you asking about the possibility that th- that there could exist, uh, languages for thinking that are not made of words?

Lex Fridman 11:32 Yeah. Yeah, I thought so.

François Chollet 11:34 Yeah, I think so. I think... So, uh, the mind is layers, right? And language is almost like the, the outermost, the uppermost layer. Um, but before we think in words, I think we think in, in terms of, uh, uh, motion and space, and we think in terms of, uh, physical actions. And I think, uh, uh, babies in particular probably express these thoughts in terms of, um, the actions, uh, that they've seen of the... or that they can perform, and in terms of the- in, in terms of motions of objects in their environment before they start thinking in terms of words.

Lex Fridman 12:10 It's amazing to think about that as the building blocks of language. So like, the kind of actions and ways that babies see the world as like more fundamental than the beautiful Shakespearean language you construct on top of it.

François Chollet 12:27 Mm-hmm.

Lex Fridman 12:28 ... and we, we probably don't have any idea what that looks like, right? Like, what... 'cause it's important for then trying to cons- engineer it into AI systems.

François Chollet 12:38 I think visual analogies and motion is a fundamental building block of the mind, and you s- you actually see it reflected in language. Like, language is full of, uh,

François Chollet 12:50 special metaphors. And when you think about things, uh, I consider myself very much as a, as a visual thinker. You, you often, uh, express these thoughts, um, by using things like, uh, visualizing concepts, um, in, uh, in 2D space, or like, you solve problems by imag- ima- imagining yourself, uh, navigating, uh, a, a, a concept space. I, I don't know if, if you have this sort of experience.

Lex Fridman 13:18 You said visualizing concept space, so like... So I certainly think about... I certainly meth- I certainly visualize mathematical concepts, but you mean like co- in concept space? Visually, you're embedding ideas (laughs) into som- into three-dimensional space you didn't explore with your mind, essentially?

François Chollet 13:38 Usually more like 2D, but yeah.

Lex Fridman 13:40 2D?

François Chollet 13:40 Yeah.

Lex Fridman 13:42 (laughs) You're a flatlander. You're, um... Okay. No. I d- I, I... I do not. I always have to, uh... Before I jump from concept to concept, I have to put it back down (laughs) on pa- uh, it has to be on paper. I can only travel on, uh, 2D paper, not inside my mind.

François Chollet 14:03 Mm-hmm.

Lex Fridman 14:03 You're able to move inside your mind.

François Chollet 14:05 But even if you're writing, like, a paper for instance, don't you have, like, a, a spatial representation of your paper? Like, you, you visualize where ideas lie topologically in relationship to other ideas, kind of like a subway map of the ideas in your paper?

Lex Fridman 14:22 Yeah. That's true. I mean, there, there is, um... In papers, I don't know about you, but, uh, there feels like there's a destination. Um, there's a, there's a key idea that you want to arrive at, and a lot of it is in, in the fog, and you're trying to kind of... It's almost like, um, um... What's that called when, um, you do a path planning search from both directions, from the start and from the end? (laughs) And then you find, you do like shortest path but like, uh, you know, in game-playing you do this with like A star from both sides.

François Chollet 15:00 Mm-hmm.

Lex Fridman 15:01 And that's-

François Chollet 15:01 And you see where, where they join.

Lex Fridman 15:03 Yeah. So you kind of do, uh, at least for me, I think, like, first of all just exploring from the start, from like, uh, first principles. What do I know? Uh, what can I start proving from that, right? And then from the destination, if, uh, you start backtracking, like if, if I want to show some kind of sets of ideas, what would it take to show them? And you kind of backtrack. But like, yeah, I, I don't think I'm doing all that in my mind though. Like, I'm putting it down on paper.

François Chollet 15:33 Do you use mind maps to organize your ideas?

Lex Fridman 15:35 No. No.

François Chollet 15:35 Yeah. I like mind maps. I, I'm that kind of person.

Lex Fridman 15:37 Okay. So let's get into this 'cause it's... I've, I've been so jealous of people, I haven't really tried it. I've been jealous of people that seem to, like... They get like this fire of passion in their eyes, 'cause everything starts making sense. It's like, uh, Tom Cruise in the movie when he's like moving stuff around. Uh, some of the most brilliant people I know use mind maps. I haven't tried really. Can you explain what the hell a mind map is?

François Chollet 16:01 I guess a mind map is a way to make kind of like the mess inside your mind to just put it on paper, so that you gain more control over it. It's a way to organize things on paper and, uh, as, as, as kind of like a consequence for organizing things on paper, you start being more organized inside, inside your own mind.

Lex Fridman 16:20 So what, what does that look like? You put... Like, do you have an example? Like, what, what do, what do you... What's the first thing you write on paper? What's the second thing you write?

François Chollet 16:29 I mean, typically, uh, you, you draw a mind map to organize the way you think about a topic. So you would start by writing down, like, the, the key concept about that topic. Like, you would write intelligence or something, and then you would start adding, uh, associative connections. Like, what do you think about when you think about intelligence? What do you think are the key elements of intelligence? So maybe you would have language, for instance, and you'd have motion. And so you would start drawing nodes with these things. And then you would see, what do you think about when you think about motion? And so on. And, and you would go like that, like a tree.

Lex Fridman 17:00 It's, it's, uh, is it a tree or... A tr- a tree mostly, or is it a graph too? Like, a tree. I guess it's a tree.

François Chollet 17:05 Oh, it's, it's more of a graph than a tree. And, um, and it's not limited to just, you know, uh, uh, writing down words. You can also, uh, uh, draw things. And it's not, it's not supposed to be purely hierarchical, right? Like, you can, um... The point is that you can start... Once, once you start writing it down, you can start reorganizing it so that it makes more sense, so that it's, uh, connected in a more effective way.

Lex Fridman 17:30 See, but I- I'm so OCD that... You just mentioned intelligence, and then language and motion. I would start becoming paranoid that the categorization is imperfect. Like, that I, I would become paralyzed with the mind map that like, this may not be... So like, the... Even though you're just doing associative kind of connections, there's an implied hierarchy that's emerging, and I would start becoming paranoid that it's not the proper hierarchy. So you're not just... One way to see mind maps is you're putting thoughts on paper. It's like a stream of consciousness. But then, you can also start getting paranoid, well, if this is the right hierarchy.

François Chollet 18:15 Sure.

Lex Fridman 18:15 Like-

François Chollet 18:15 Which... But it's a mind map. It's your mind map. You're free to draw anything you want. You're free to draw any connection you want, and you can just make a different mind map if, if you think the central node is not the right node.

Lex Fridman 18:26 Yeah.

François Chollet 18:26 So-

Lex Fridman 18:26 I suppose there's a fear of being w- wrong.

François Chollet 18:29 If, if you want to-

Lex Fridman 18:31 Or so.

François Chollet 18:31 If you want to organize your ideas by writing down what you think, which I think is, is, is very effective, like...... h- how do you know what you think about something if you, if you don't write it down? Right? Uh, if you do that, the thing is that it imposes, uh, much more, uh, syntactic structure over your ideas-

Lex Fridman 18:49 Yeah.

François Chollet 18:50 ... which is not required with a mind map. So, a mind map is kind of like a lower level, more freehand way of orga- organizing your thoughts and once you've drawn it, then you can start, uh, uh, actually voicing your thoughts in terms of, you know, paragraphs and sentences.

Lex Fridman 19:05 It's a two-dimensional aspect of layout too, right?

François Chollet 19:08 Yeah.

Lex Fridman 19:09 Uh, and, uh, it's, it's a kind of flower, I guess? You start ... there's usually, you want to start with a central concept?

François Chollet 19:15 Yes.

Lex Fridman 19:16 And then you move out?

François Chollet 19:16 Yeah. Typically, it's, it ends up more like a subway map, so it ends up more like a graph, a topological graph.

Lex Fridman 19:22 Without a root node?

François Chollet 19:23 There are ... So, like in a subway map, there are some nodes that are more connected than others, and there are some nodes-

Lex Fridman 19:28 All right.

François Chollet 19:28 ... that are more important than others. All right. So, there are destinations, but it's, it's not gonna be purely like a tree, for instance.

Lex Fridman 19:36 Yeah. It's fascinating to think that ... if there is something to that about our, uh, about the way our mind thinks. By the way, I just kind of remembered obvious thing that I have probably thousands of documents in Google Doc at this point that are bullet point lists, uh, which is you can probably map a mind map to a bullet point list. It's the same. It's, uh ... well, no, it's not. It's a tree. It's a tree, yeah.

François Chollet 20:06 Mm-hmm.

Lex Fridman 20:06 So, I, I create trees, but also they don't have the visual element.

François Chollet 20:10 Mm-hmm. Mm-hmm.

Lex Fridman 20:10 Like, um, I guess I'm comfortable with the structure. It feels like-

François Chollet 20:14 Well-

Lex Fridman 20:14 ... it, it ... the narrowness, the constraints feel more comforting.

François Chollet 20:18 If you have thousands of documents with your own thoughts in Google Docs, why, why don't you write, uh, some kind of search engine? Like, maybe a, a mind map, um, a piece of software, mind mapping software-

Lex Fridman 20:31 A generative-

François Chollet 20:31 ... where you write down a concept and then it gives you sentences or paragraphs from your thousand Google Docs document that match this concept.

Lex Fridman 20:41 The, the problem is it's so deeply ... unlike mind maps, uh, it's so deeply rooted in natural language. So, it's not, um, it's not semantically searchable, I would say, 'cause the categories are very ... you kind of mentioned intelligence, language, and motion. They're very strong semantic. Like, it feels like the mind map forces you to be semantically clear and specific. The bullet points list I have are, are, are sparse, disparate thoughts that, uh, poetically represent a category like motion, as opposed to saying motion.

François Chollet 21:24 Mm-hmm.

Lex Fridman 21:25 (laughs) So, unfortunately, it's ... that's the same problem with the internet. That's why the idea of semantic web is difficult to get. It's, um ... most language on the internet is a, is a giant mess of natural language that's hard to interpret, which ... so do you think, uh, do you think there's something to mind maps as, um ... you actually originally brought it up as when we were talking about kind of cognition and language. Do you think there's something to mind maps about how our brain actually deals, like, think, reasons about things?

François Chollet 22:01 It's possible. I think it's reasonable to assume that there is some level of topological processing in the brain, that the brain is, um, very associative in nature. And I also believe that, uh, a topological space, uh, is a better medium, um, to encode thoughts than a, uh, a geometric space. Than a ... so I think-

Lex Fridman 22:28 What's the difference between a topological and a geometric space?

François Chollet 22:31 Well, um, if you're talking about topologies, uh, then points, uh, are either connected or not. So, a topology is more like a subway map and, uh, geometry is when you're interested, uh, in the distance between things. And in subway maps, you don't really have the concept of distance. You only have the concept of whether there is a train going from station A to station, uh, B. Um, and, uh, what we do in deep learning is it's, we're, we're actually dealing with, uh, geometric spaces. We're dealing with concept vectors, word vectors, um, that have a, a distance between them-

Lex Fridman 23:03 The, the distance is a product of thing.

François Chollet 23:03 ... which is expressed in terms of dot product. Um, we are not s- ... we are not really building topological models usually.

Lex Fridman 23:10 I think you're absolutely right. Like, distance is a, uh, a fundamental importance in deep learning. I mean, it's, it's the continuous aspect of it to me. Yeah.

François Chollet 23:19 Yes. Because everything is a vector, and everything has to be a vector, because everything has to be differentiable.

Lex Fridman 23:24 Yeah.

François Chollet 23:24 If your space is discrete, it's no longer differentiable. You cannot do deep learning in it anymore. Well, you could, but you could only do it by embedding it in a, a bigger continuous space. So, if you do, uh, uh, topology in the cont- in the context of deep learning, you have to do it by embedding your topology in a geometry. Right?

Lex Fridman 23:42 Yeah. Well, let me, uh, let me zoom out for a second. Uh, let's get into your paper on The Measure of Intelligence that, uh, did you put out in 2019?

François Chollet 23:52 Yes.

Lex Fridman 23:53 Okay. Yeah.

François Chollet 23:54 November.

Lex Fridman 23:55 November. Yeah. Remember 2019? That was a, it was a different time.

François Chollet 24:01 Yeah.

Lex Fridman 24:02 (laughs)

François Chollet 24:02 I remember. I still remember.

Lex Fridman 24:05 (laughs) It feels like a different, uh, different, different world.

François Chollet 24:09 You could travel. You could, you know, actually go outside and see friends.

Lex Fridman 24:15 Yeah. Let me ask the most absurd question. I think, uh, there's some non-zero probability that there'll be a textbook one day, like 200 years from now, on artificial intelligence or it will be called, like, just Intelligence 'cause humans will already be gone, and it'll be your picture with a quote. This is ... you know, one of the early, uh, biological systems would consider the nature of intelligence, and there'll be, like, a definition of how they thought about intelligence, which is one of the things you do in your paper on Measure of Intelligence is to ask, like-... well, what is intelligence and, and, uh, how to test for intelligence and so on. So, is there a spiffy quote about what is intelligence? What is the definition of intelligence, a- according to François Chollet?

François Chollet 25:06 Yeah. So, d- do you think-

Lex Fridman 25:08 (laughs)

François Chollet 25:09 ... the, the, the super intelligent AIs of the future will want to remember us the way we remember humans from the past and do you think they would be, you know, they won't be ashamed of having a biological origin?

Lex Fridman 25:22 Uh, no. I, I think it will be a niche topic, it won't be that interesting, but it'll be, it'll be like the people that study in certain contexts, like historical civilization that no longer exists, the Aztecs and so on, that- that's how it'll be seen. And it'll be a study in the c- also the context on social media, there'll be hashtags about the atrocity committed to human beings, um, when, when the, when the robots finally got rid of them. Like, it was a mistake. It'll be seen as a, as a giant mistake, but ultimately, eh, in the name of progress and it created a better world because humans were, uh, over-consuming the resources and ult- they were not very rational and were destructive in the end, in terms of productivity and, uh, putting more love in the world. And so within that context, there'll be a chapter about these biological systems.

François Chollet 26:17 You seem to have a very detailed vision of that future. You should write, uh, a sci-fi novel about it.

Lex Fridman 26:22 I s- I'm, I'm working, I'm, uh, I'm working on a, on a sci-fi novel currently, yes. (laughs)

François Chollet 26:27 Yeah, so-

Lex Fridman 26:28 Self-published, yeah.

François Chollet 26:29 The definition of intelligence. So, intelligence is the efficiency with which, uh, you acquire new skills at tasks that you did not previously, uh, know about, that you did not prepare for. All right? So, it is not... Intelligence is not skill itself. It's not what you know, it's not what you can do. It's how well and how efficiently you can learn new things.

Lex Fridman 26:54 New things?

François Chollet 26:55 Yes.

Lex Fridman 26:56 I- the idea of newness there seems to be fundamentally important.

François Chollet 27:01 Yes. So, you would see intelligence, uh, on display, for instance, uh, whenever you see, uh, a human being or, you know, an AI creature adapt to a new environment that it di- it has not seen before, that its creators did not anticipate. Uh, when you see adaptation, when you see improvisation, when you see generalization, that's intelligence. Uh, in reverse, if you have a system that when you put it in a slightly new environment, it cannot adapt, it cannot improvise, it cannot deviate from what's it's, it's hardcoded to do or, um, what, uh, what it has, uh, been trained to do, um, that is a system that is not intelligent. There's actually a quote from, uh, uh, Einstein that captures, uh, uh, this idea, which is, "The measure of intelligence is the ability to change." I, I like that quote. I think it, uh, captures at least part of this idea.

Lex Fridman 27:55 You know, there might be something interesting about the difference between your definition and Einstein's. I mean, he's just being Einstein (laughs) and clever. But acquisition of, um, new ability to deal with new things versus ability to just change, what's the difference between those two things? So, just change in itself, do you think there's something to that? Just being able to change?

François Chollet 28:26 Yes, being able to adapt. So not, not change, but certainly, uh, uh, change this direction, being able to adapt yourself to your environment.

Lex Fridman 28:37 Whatever the environment is.

François Chollet 28:38 That's, that's a big part of intelligence, yes. And intelligence is more precisely, you know, how efficiently you're able to adapt, how efficiently you're able to basically master your environment. How efficiently, uh, you can acquire new skills. And I think there's a, there's a big distinction to be drawn between, uh, intelligence, which is a process, and the output of that process, which is skill. Um, so for instance, if you have, uh, a very smart human programmer, uh, that considers the game of chess and that writes down, uh, a static program that can play chess, then the intelligence is the process of developing that program. But the program itself is just encoding, um, the output artifact of that process.

Lex Fridman 29:28 Right.

François Chollet 29:28 The program itself is not intelligent, and the way you tell it's not intelligent is that if you put it in a different context, you ask it to play Go or something, it's not gonna be able to perform well without human involvement, because the source of intelligence, the entity that is capable of that process, is the human programmer.

Lex Fridman 29:44 Is the programmer.

François Chollet 29:44 So, we should, uh, um, be able to tell the difference between the process and its output. We should not confuse, uh, the output and the process. It's the same as, you know, do not confuse, uh, a road-building company and one specific road, because one specific road takes you from point A to point B, but a road-building company can take you from... Can make a path from anywhere to anywhere else.

Lex Fridman 30:09 Yeah, that's beautifully put, but it's also, to play devil's advocate a, uh, a little bit, you know, um, it's possible that there's something more fundamental than us humans. So, you kind of said the programmer creates, uh, the difference between the, the acquirer of the skill and the skill itself. There could be something, like you could argue the universe is more intelligent. Like, the, the deep... The base intelligence of, um, that we should be trying to measure is something that created humans. We should be measuring God or (laughs) the s- the source of the universe, as opposed to... Like, there's, there could be a deeper intelligence.

François Chollet 30:55 Sure, I mean, you-

Lex Fridman 30:55 There's always deeper intelligence, I guess.

François Chollet 30:57 You, you can argue that, but that does not take anything away from the fact that humans are intelligence.

Lex Fridman 31:01 Right.

François Chollet 31:01 And you can tell that because they are capable of adaptation and, and generality.

Lex Fridman 31:07 Got it.

François Chollet 31:07 Um, and you see that, uh, particularly in the fact that, uh-Uh, humans are capable of handling, uh, uh, situations and tasks that are quite different from anything that any of our evolutionary ancestors has ever encountered. So, we are capable of generalizing very much out of distribution if you consider all- all evolutionary history as being in- in a way of training data. Like-

Lex Fridman 31:33 Of course, evolutionary biologists would argue that we're not going too far out of the distribution. We're like mapping the skills we've learned previously, desperately trying to like jam them into like these new situations.

François Chollet 31:47 I mean, there's definitely a little bit- a little bit of that, but it's pretty clear to me that we're able to, uh... You know, most of the things we do, uh, any given day in our modern civilization are things that are very, very different from what, you know, our ancestors a million years ago would have been doing in- in- in a given day. And the e- our environment is very different. So, I agree that, um, everything we do, we do it with cognitive building blocks that we acquired, uh, over the course of evolution, right? And that anchors, um, our cognition to a certain context, which is the human condition very much. Uh, but still our mind is capable of a- a pretty remarkable degree of generality, far beyond anything we can, uh, create in artificial systems today. Like the- the- the degree in which the mind can generalize from its evolutionary history, uh, can generalize a way from its evolutionary history is much greater than the degree to which a deep learning system today can generalize away from its training data.

Lex Fridman 32:51 And like the key point you're making which I think is quite beautiful is like, we shouldn't measure, if we're talking about measurement, we shouldn't measure the skill, we should measure like the creation of the new skill, the ability to create that new skill.

François Chollet 33:06 Yes.

Lex Fridman 33:06 But there, it's tempting, like... (sighs) It's weird because the skill is a little bit of a small window into the- into the system. So, whenever you have a lot of skills, I mean, it's tempting to measure the skills.

François Chollet 33:21 Yes. I mean, the skill is the, uh, only thing you can objectively, uh, measure. But yeah. So, the- the thing to keep in mind is that when you see skill, uh, in a human, um, it gives you a strong signal that that human is intelligent because you know they weren't born with that skill typically. Like you see a very s- you see a very strong chess player, maybe you're- you're a very strong chess player yourself. And-

Lex Fridman 33:47 I- I think you're- you're- you're just saying that 'cause I'm Russian and now- now you're- you're prejudiced, you assume all Russians are good at chess.

François Chollet 33:55 Yeah, I'm- I'm biased, exactly.

Lex Fridman 33:56 I'm biased. Yeah. Well, you- (laughs)

François Chollet 33:57 That's right. Cultural bias. Um, so if you see a very strong chess player, you know they weren't born, uh, knowing how to play chess, so they had to acquire that skill with their limited resources, with their limited lifetime. And, you know, they did that because they are generally intelligent, and so they- they may as well have acquired any other skill, you know they have this potential. And on the other hand, if you see a- a computer playing chess, you cannot make these same assumptions, because you cannot, you know, just assume the computer is generally intelligent. The computer may be born, uh, knowing how to play chess in the sense that it may have been programmed by a human that, uh, has understood chess for the computer and- and it has just encoded, um, the output of that understanding in a static program, and that program, uh, is not intelligent.

Lex Fridman 34:49 So, let- let's zoom out just for a second and say like, what is the goal of the On the Measure of Intelligence paper? Like, what do you hope to achieve with it?

François Chollet 34:59 So, the goal of the paper is to clear up some long-standing misunderstandings about the way we've been, uh, conceptualizing intelligence in the AI community, and, uh, in the way we've been evaluating progress in AI. Um, there's been a lot of progress recently in machine learning and people are- are, you know, extrapolating from that progress that we are about to solve general intelligence. And if you want to be able to evaluate these statements, you need to precisely define what you're talking about when you're talking about general intelligence, and you need, um, a formal way, a reliable way to measure how much, uh, intelligence, how much general intelligence a system possesses. And i- ideally, this measure of intelligence should be, uh, actionable, so it should not just describe, uh, what intelligence is. It shou- should not just be a binary indicator that tells you the system is intelligent, uh, or- or- or it isn't. Um, it should be actionable. It should have, uh, explanatory power, right? So you could use it as a feedback signal. It would show you, uh, the way towards building more intelligent systems.

Lex Fridman 36:13 So, at the first level, you draw a distinction between two divergent views of intelligence, of, um, as we just talked about, intelligence is a collection of task sp- task-specific skills, and a general learning ability. So, what's the difference between kind of this memorization of skills and a general learning ability? We talked about it a little bit, but can you try to linger on this topic for a bit?

François Chollet 36:43 Yeah. So, the first part of the paper, uh, is, uh, an assessment of the different ways, uh, we've been thinking about intelligence, and the different ways we've been, uh, evaluating progress in AI, and, uh, the history of cognitive sciences has been shaped by two views of the human mind. And one view is the evolutionary psychology view in which the mind is, uh, a collection of fairly static, uh, special-purpose ad hoc mechanisms that have been hardcoded by evolution, uh, over our- our history as a species for a very long time. And, um...Early, uh, uh, AI researchers, people like, uh, Marvin Minsky for instance, they clearly subscribed to this view, and they saw, they saw the mind as a kind of, you know, collection of static programs, uh, uh, similar to the programs they would, they would run on, like, mainframe computers. And in fact, they, I think they very much understood the mind, uh, through the metaphor of the mainframe computer because it was the tool they, they were working with, right? And so you had these static programs, this collection of, uh, different static programs operating over a database-like memory, and in this picture learning was not very important. Uh, learning was considered to be just memorization, and in fact, uh, uh, learning is basically not featured in AI textbooks until, uh, uh, the 1980s, uh, with the rise of machine learning.

Lex Fridman 38:16 It's kind of fun to think about that learning was the outcast, like the- the weird people working on learning. Like the mainstream AI world was, um... I mean, I don't know what the best term is, but it's non-learning. It- it was seen as like reasoning-

François Chollet 38:36 Yes.

Lex Fridman 38:36 ... would not be learning-based.

François Chollet 38:38 Yes. It was seen, it was considered the mind was a collection of programs that were, uh, uh, primarily logical in nature, and that all you needed to do to create a mind was to write down these programs and they would operate over knowledge which will be stored in some kind of database, and as long as your database would encompass, you know, everything about the world and your logical rules were, uh, uh, comprehensive, then you would have a mind. So the other view of the mind is the brain are, uh, as, uh, sort of blank slates, right? This is a very old idea. You find it in, uh, uh, John Locke's writings. This is the- the tabula rasa. Um, and this is this idea that the mind is some kind of like information sponge, um, that starts, uh, uh, empty, that starts blank, and that absorbs, uh, uh, knowledge and skills from experience, right? So it's, uh, it's a sponge that reflects the complexity of the world, the complexity of, uh, of your life experience essentially, that everything you know and everything you can do is a reflection of something you found in the outside world essentially. So this is an idea that's very old, uh, that was not very popular for instance in the- in the 1970s, but that had gained a lot of vitality recently with the rise of, uh, connectionism, in particular deep learning. And so today deep learning is the dominant paradigm in AI, and, uh, I feel like lots of, uh, AI researchers are conceptualizing the mind, uh, via a deep learning metaphor. Like they see, uh, the mind as a kind of a- a randomly initialized neural network that starts blank when you're born and then that, uh, gets trained via, uh, exposure to training data, that acquires knowledge and skills via exposure to training data.

Lex Fridman 40:30 By the way, uh, it's a small tangent. I feel like people who are thinking about intelligence are not conceptualizing it that way. I actually haven't met too many people who believe that a neural network will be able to reason, who- who seriously think that rigorously, because I think it's an actually interesting worldview, and- and we'll talk about it more, but it's been impressive what the c- uh, what neural networks have been able to accomplish, and it's... I, to me, I don't know, you might disagree, but it's an open question whether like, like scaling size eventually might lead to incredible results to us mere humans will appear as if it's general.

François Chollet 41:17 I mean, if you- if you ask people who are seriously thinking about intelligence, they will definitely not say that all you need to do is- is, like the mind is just a neural network. Uh, however, it's actually a view that's- that's very popular I think in the deep learning community that many people are kind of, uh, uh, conceptually, you know, intellectually lazy about it.

Lex Fridman 41:37 Right. It's a-

François Chollet 41:37 Uh...

Lex Fridman 41:38 But, uh, what... I guess what I'm saying exactly right is, uh, I- I'm y- I haven't met many people and I think it would be interesting, uh, to meet a person who is not intellectually lazy about this particular topic and still believes that neural networks will go all the way. I think Yann LeCun-

François Chollet 41:55 I believe-

Lex Fridman 41:55 ... is probably closest to that with self-supervised-

François Chollet 41:57 There are definitely people who argue that, uh, current deep learning techniques are already the way to, uh, general artificial intelligence and that all you need to do is to scale it up to all the available training data. And that's... If you look at the- the waves that OpenAI's GPT-3 model has made, you see echoes of this idea.

Lex Fridman 42:22 So on that topic,

Lex Fridman 42:24 GPT-3, similar to GPT-2 actually, have captivated some part of the imagination of the public. There's just a bunch of hype of different kind that's... I would say it's emergent, it's not artificially manufactured, it's just like people just get excited for some strange reason. In- in the case of GPT-3 which is funny that there's I believe a couple of months delay from release to hype. Maybe I'm not, um, uh, historically correct on that, but it feels like there was a little bit of a- a lack of hype and then there's a phase shift into- into hype. But nevertheless, there's a bunch of cool applications that seem to captivate the imagination of the public about what this language model that's trained in unsupervised way without any fine-tuning is able to achieve. So what do you make of that? What are your thoughts about GPT-3?

François Chollet 43:23 Yeah. So I think what's interesting about GPT-3...... is the idea that it may be able to learn new tasks in, um, after just being shown a few examples. So, I think if it's actually capable of doing that, that's novel and that's very interesting and that's something we should investigate. Uh, that said, I must say I'm not entirely convinced that we have shown it's, uh, it's capable of doing that. Uh, it's very likely, given the amount of data that the model is trained on, that what it's actually doing is pattern matching, uh, a new task you give it with a task that it's been, uh, exposed to in its training data. It's just recognizing the task instead of just developing a model of the task, right? Um-

Lex Fridman 44:05 But there's, um... Sorry to interrupt. There, there's parallels to what you said before, which is it's possible to see GPT-3 as, like, the prompts it's given as a kind of SQL query into this thing that it's learned, similar to what you said before, which is language is used to query the memory.

François Chollet 44:23 Yes.

Lex Fridman 44:23 So, is it possible that a neural network is a giant memorization thing, but then it'll... if it gets sufficiently giant, it'll memorize sufficiently large amounts of things in the world where it becomes... where intelligence becomes a querying machine?

François Chollet 44:39 Mm-hmm. I think it's possible that, uh, a significant chunk of intelligence is this giant associative memory. Uh, I definitely don't believe that intelligence is just a giant associative memory, but it may well be, uh, a big component.

Lex Fridman 44:57 So, do you think GPT-3, 4, 5, GPT-10 will eventually... Like, what do you think... where's the ceiling? Do you think it'll be able to reason? Um... No, that's a bad question. Uh, (laughs) like, what is the ceiling is the better question. W- w- what-

François Chollet 45:17 How well is it gonna scale? How good is GPT-N going to be?

Lex Fridman 45:21 Yeah.

François Chollet 45:22 So, I believe GPT-N is gonna-

Lex Fridman 45:25 GPT-N. (laughs)

François Chollet 45:26 ... is gonna improve on the strength, uh, of GPT-2 and 3, which is it, it will be able to generate, you know, uh, ever-more plausible text, uh, in context-

Lex Fridman 45:37 Just monotonically increasing-

François Chollet 45:38 ... given a prompt.

Lex Fridman 45:39 ... performance. (laughs)

François Chollet 45:40 Um, yes. If you train, if you train a bigger model on more data, then, uh, uh, your text will be, uh, increasingly more, uh, context aware and increasingly more, uh, plausible in the same way that GPT-3 is, is much better at generating plausible texts compared to GPT-2. Um, but that said, I don't think just scaling up, uh, the model to more transformer layers and more training data is gonna address the flaws of GPT-3, which is that it can generate plausible texts, but that text is not constrained by anything else other than plausibility. So, in particular, it's not constrained by, uh, factualness, uh, or even consistency, which is why it's very easy to get GPT-3 to, to generate statements that are, uh, uh, factually untrue, uh, or to generate statements that are even self-contradictory, right? Uh, because it's, uh, it's, it's only goal is plausibility, and it has no other constraints. It's not constrained to be self-consistent, for instance, right? And so, for this reason, one thing that I thought was very interesting with GPT-3 is that you can predetermine the answer it will give you by asking the question in a specific way, because it's very responsive to the way you ask the question since it has-

Lex Fridman 46:58 Mm-hmm.

François Chollet 46:58 ... no understanding of the content of the question.

Lex Fridman 47:01 Right. The-

François Chollet 47:03 And if you, if you ask the same question in two different ways that are basically adversarially, uh, engineered to produce a certain answer, you will get, uh, uh, two different answers, two contradictory answers.

Lex Fridman 47:15 It's very susceptible to adversarial attacks, essentially.

François Chollet 47:18 Potentially, yes. So, in, in general, the problem with these models, these generative models, is that they, they're very good at generating plausible text, but that's just, that's just not enough, right? Um, uh, you need, uh... W- I think one, one, one avenue that would be very, uh, interesting to make progress is to make it possible, uh, to write programs over the latent space that these models operate on, so that you would rely on these, uh, uh, uh, self-supervised models to generate a sort of, like, pool, uh, of knowledge and concepts and common sense. And then you would be able to write explicit, uh, uh, reasoning programs over it. Uh, because the, the current problem with GPT-3 is that you... it's, it's... it can be quite difficult to get it to do what you want to do. Uh, if you want to, uh, turn GPT-3 into products, you need to put constraints on it, uh, you need to, um, force it to obey certain rules. So, you need a way to program it explicitly.

Lex Fridman 48:22 Yeah. So if you look at its ability to do program synthesis, it generates, like you said, something that's plausible.

François Chollet 48:29 Yeah. So if you, if you try to make it generate programs, it will perform well, uh, for any program that, that it has seen in its, in its training data. But because, uh, program space is not interpretive, right, um, uh, it's not gonna be able to generalize to problems it hasn't seen before.

Lex Fridman 48:48 Now, that's currently... Do you think sort of an absurd but I think useful, um, I guess, intuition builder is, uh, you know, the, the GPT-3 has 175 billion parameters. The human brain has a hundred... has about a thousand times that or, or more in terms of number of synapses. Do you think, um... obviously, very different kinds of things, but there is some degree of, uh, similarity. Do you think... what, what do you think GPT will look like when it has 100 trillion parameters?You- you think our conversation might be-

François Chollet 49:36 So-

Lex Fridman 49:36 ... in nature, different? Like, the... 'cause you've criticized-

François Chollet 49:40 In nature-

Lex Fridman 49:40 ... GPT-3 very effectively now. Do you think...

François Chollet 49:45 No, I don't think so. So, the, the... t- to begin with, the bottleneck with scaling up GPT-3 and GPT models, uh, uh, generative pre-trained transformer models, is not gonna be the size of the model or how long it- it takes to train it. The bottleneck is gonna be the training data because OpenAI's already training GPT-3 on a crawl of basically the entire web. Right? And that's a lot of data. So, you could imagine training on more data than that. Like, Google could train on more data than that.

Lex Fridman 50:14 Mm-hmm.

François Chollet 50:14 But it would still be only incrementally more data. And I- I don't r- recall exactly how much more data GPT-3 was trained on compared to GPT-2, but it's probably at least like 100 or maybe even 1,000 X. I don't have the exact number. Uh, you're not gonna be able to train a model on 100 more data than what- what you're alread- what- what you're already doing.

Lex Fridman 50:34 So, that's- that's brilliant. So, it's not... You know, it's easier to think of compute as a bottleneck and then arguing that we can remove that bottleneck, but...

François Chollet 50:41 We can remove the compute bottleneck. I don't think it's a big problem. If you look at the- at the pace at which we've, uh, uh, uh, improved the efficiency, uh, of deep learning models, uh, uh, in the past, uh, few years, uh, I'm not worried about, uh, training time bottlenecks or model size bottlenecks. Uh, the- the bottleneck in the case of this, uh, uh, generative transformer models, is absolutely the training data.

Lex Fridman 51:05 What about the quality of the data? So...

François Chollet 51:07 So, yeah. So, the quality of the data is, uh, is an interesting point. The thing is, if you're gonna want to use these models, uh, in real products, um, then you- you want to feed them, uh, data that's as high quality, as factual, I would say as unbiased as possible, but, you know, there's- there's not really such a thing as, uh, unbiased data in the first place. But you probably don't want to- to train it, uh, uh, on Reddit, for instance. It sounds li- it sounds like a bad plan. So, from my personal experience working with, uh, uh, uh, large-scale deep learning models, so at some point, I was working on a- a- a model at Google that's trained, uh, on, uh, uh, like 350 million, uh, labeled images, uh, its image classification model. That's a lot of images. That's like, probably most publicly available images on the web at the time. And, uh, it was a very, uh, noisy data set because the labels were not originally, uh, annotated by hand, by humans. They were, uh, automatically derived from like tags, and social media, uh, or just keywords, uh, in- in the same page as the image response and so on. So, it was very noisy. And it turned out that you could, uh, easily get a better model, uh, not just by training... Like, if you train on more, uh, of the noisy data, you get an incrementally better model, but you- you- you- you very quickly hit diminishing returns. On the other hand, if you train on smaller data set with higher quality annotations, quality that are, uh, uh, annotations that are actually, uh, made by humans, you get a better model. And it also takes, you know, less time to train it.

Lex Fridman 52:49 Uh, yeah, that's fascinating. It's the self-supervised learning, if there's a way to get better doing the automated labeling.

François Chollet 52:58 Yeah. So, uh, you can enrich or refine your labels, uh, in an automated way. That's correct.

Lex Fridman 53:07 Do you have a hope for, um... I don't know if you're familiar with the idea of a semantic web? Is, so semantic web, just for people who are not familiar, and... is, uh, is the idea of being able to convert the internet (laughs) or- or be able to attach like semantic meaning to the words on the internet, the s- the sentences, the paragraphs, to be able to contr- convert information on the internet or some fraction of the internet into something that's interpretable by machines. That was kind of a dream for, um, I think, the- the semantic web papers in the '90s.

François Chollet 53:46 Mm-hmm.

Lex Fridman 53:48 It's kind of the dream that, you know, the internet is full of rich, exciting information. Even in just looking at Wikipedia, we should be able to use that as data for machines.

François Chollet 53:57 Yeah.

Lex Fridman 53:57 And so far-

François Chollet 53:58 And that information is not, is not really in a format that's available to machines. So, no, I don't think the semantic web will ever work simply because it would be a lot of work, right, to make, uh... to provide that information in structured form. And there is not really any incentive for anyone to provide that work. Uh, so I think the- the way forward to make, um, the knowledge on the web available to machines is actually, uh, uh, something closer to unsupervised deep learning.

Lex Fridman 54:29 Yeah.

François Chollet 54:29 So, GPT-3 is actually a bigger step in the direction of making the knowledge of the web available to machines than the semantic web was.

Lex Fridman 54:36 Yeah. Perhaps. In- in a human-centric sense, it- it feels like GPT-3 hasn't learned anything that could be used to reason. But, uh, that might be just the early days.

François Chollet 54:53 Yeah. I- I think that's correct. I think the forms of reasoning that you- that you see it perform are basically just reproducing patterns that it has seen in string data. So, of course, if you train on, uh, the entire web, then you can produce an illusion of reasoning in many different situations, but it will break down if it's presented with- with a novel, uh, situation.

Lex Fridman 55:15 That's the open question between the illusion of reasoning and actual reasoning, yeah.

François Chollet 55:18 Yes. The power to adapt to something that is genuinely new, because the thing is, even... imagine you had, uh, um... you could train on every bit of data ever generated, uh, in the history of humanity.Uh, it remains that, that model would be capable of, of, uh, uh, anticipating, uh, many different possible situations. But it remains that the future is gonna be something different. Like, um, for instance, if you train a GPT-3 model, uh, on, on data from the year 2002 for instance, and then use it today, it's gonna be missing many things. It's gonna be missing many common sense facts about the world. It's even gonna be missing vocabulary and so on.

Lex Fridman 56:05 Yeah, it's interesting that, uh, GPT-3 even doesn't have, I think any information about the coronavirus. Uh- (laughs)

François Chollet 56:13 Yes. Which is why, you know, uh, uh, a system that's, uh, you, you tell that a system is intelligent when it's capable to adapt. So intelligence is gonna require a small amount of continuous learning, but it's also gonna require some amount of improvisation. Like it's not enough to assume that what you're gonna be asked to do is something that you've seen before or something that is a simple interpolation of things you've seen before.

Lex Fridman 56:42 Yeah.

François Chollet 56:43 In fact, that model breaks down for, uh, even, even very, tasks that look relatively simple from a distance, like L5 self-driving for instance. Google had a, had a paper, uh, uh, a couple years back showing that something like 30 million, uh, different road situations were actually completely insufficient to train a driving model. It wasn't even L2, right? And that's a lot of data. That's a lot more data than the, the 20 or 30 hours of driving that a human needs, uh, to learn to drive given the, the knowledge they've already accumulated.

Lex Fridman 57:22 Well, let me ask you on that topic, uh, Elon Musk, Tesla Autopilot. It's one of the only companies I believe is really pushing for a learning-based approach. Are you, you're skeptical that that kind of network can achieve level four?

François Chollet 57:39 L4, uh, is probably achievable. L5 probably not.

Lex Fridman 57:44 What's the distinction there? Is L5 is completely, you can just fall asleep?

François Chollet 57:49 Yeah, L5 is basically human level.

Lex Fridman 57:51 Well, with driving we have to be careful saying human level 'cause like that's the most-

François Chollet 57:55 Yeah, there are all kinds of drivers. (laughs)

Lex Fridman 57:56 (laughs) Yeah, that's the clearest example of like, you know, cars will most likely be much safer than humans in situ- in many situations where humans fail. It's the vice versa question.

François Chollet 58:09 So I, I, I'll tell you, you know, the thing is the, the amounts of training data you would need to anticipate for pretty much every possible situation you, you run into in the real world, uh, is such that it's not entirely unrealistic to think that at some point in the future we'll develop a system that's trained on enough data, especially, uh, uh, provided that we can, uh, simulate a lot of that data. We don't necessarily need actual, uh, actual cars on the road, uh, for everything. Uh, but it's a massive effort, and it turns out you can create a system that's much more adaptative, uh, that can generalize much better if you just add, um, uh, explicit models of the surroundings of the car, uh, and if you use deep learning for what it's good at, which is to provide perceptive information. So, in general, deep learning is, is, uh, a way to encode perception and a way to encode intuition. But it's not a good medium for, uh, any sort of, uh, explicit reasoning, and, uh, in AI systems today, uh, strong generalization tends to come from, um, explicit models, tend to come from abstractions in the human mind that are encoded in program form, uh, by human engineer, right?

Lex Fridman 59:29 Yeah.

François Chollet 59:29 These are the abstractions that can actually generalize, not the sort of, uh, weak abstraction that is learned by a neural network.

Lex Fridman 59:34 Yeah, and the question is how much, uh, how much reasoning, how much strong abstractions are required to solve particular tasks like driving? That's, that's the question. Or human life, existence, how much, how much strong obs- uh, abstractions does existence require? But more specifically on hu- on driving, that's, that seems to be, that seems to be a coupled question about intelligence is like, uh, how much intelligence, like how do you build an intelligent system and, uh, the coupled problem, how hard is this problem? How much intelligence does this problem actually require? So we're, um, we get to cheat, right, 'cause we get to look at the problem. Like it's not like you get to close your eyes and completely new to driving. We get to do what we do as human beings which is, uh, for the majority of our life before we ever learn quote unquote "to drive" we get to watch other cars and other people drive. We get to be in cars, we get to watch, we get to g- again, see movies about cars. We get to, you know, we get to observe all that stuff, and that's similar to what neural networks are doing is getting a lot of data and the, the, the question is, yeah, how much is, uh, how many leaps of reasoning genius is required (laughs) to be-

François Chollet 1:00:58 Well-

Lex Fridman 1:00:58 ... able to actually effectively drive.

François Chollet 1:00:59 I think this example of driving, I mean, sure, um, uh, you've seen a lot of cars, uh, in your life before you learned to drive. But let's say you've learned to drive in Silicon Valley and now, uh, you rent a car in Tokyo. Well, now everyone is driving on the other side of the road and the signs are different and the roads are more narrow and so on. So it's a very, very different environment and, uh, a smart human, even an average human should be able to just zero shot it, to just be operational-

Lex Fridman 1:01:32 Zero shot.

François Chollet 1:01:32 ... in this, in this very different environment-

Lex Fridman 1:01:33 Yeah.

François Chollet 1:01:34 ... right away despite having had no contact with the novel complexity that is contained in this environment.... right? And that is novel complexity. It's not just, uh, um, interpolation, uh, over the situations that you've encountered previously, like learning to drive in the US, right?

Lex Fridman 1:01:55 I, I would say, the reason I ask is, one of the most interesting tests of intelligence we have today, actively, which is driving, in terms of having an impact on the world. Like, when do you think we'll pass that test of intelligence?

François Chollet 1:02:09 So, I, I don't think driving is that much of a test of intelligence because again, there is no task for which skill at that task demonstrates intelligence, unless, uh, it's a kind of meta-task that involves acquiring, uh, new skills. So, I don't think... I think you can actually solve driving without having, uh, uh, any- any real amount of intelligence. For instance, if you really did have infinite training data, um, you could just literally train an end-to-end deep learning model at just driving, provided infinite training data. The only problem, uh, with the whole idea is, um, collecting datasets that's sufficiently comprehensive that covers the very long tail of possible situations you might encounter. And it's really just a scale problem. So, I think the- there's nothing fundamentally wrong, uh, uh, with this plan, with this idea. It's just that, um, it strikes me as a fairly inefficient thing to do, because you run into this, uh, this, uh, uh, scaling issue with diminishing returns. Whereas if instead you took a more, uh, manual engineering approach where you, uh, uh, use deep learning modules, uh, in combination, uh, with, um, engineering and explicit model of the surrounding of the cars and you, and you bridge the two in a clever way, your model will actually start generalizing much earlier and more effectively than the end-to-end deep learning model. So, why would you not go with the more manual engineering oriented approach? Like, even if- if you created, uh, that system, either the end-to-end deep learning model system that's around infinite data or, uh, the slightly, uh, uh, uh, more human system, I- I don't think achieving L5 would demonstrate, uh, a general intelligence or intelligence of any generality at all. Again, the only possible test, uh, of generality in AI would be a test that looks at skill acquisition over unknown tasks.

Lex Fridman 1:04:14 But then-

François Chollet 1:04:14 For instance, you could take your L5-

Lex Fridman 1:04:16 Mm-hmm.

François Chollet 1:04:16 ... uh, driver and ask it to- to learn to- to pilot a- a- a commercial airplane, for instance. And then you would look at how much human involvement is required and how much-

Lex Fridman 1:04:26 Right.

François Chollet 1:04:26 ... training data is required, uh, for the system to learn to pilot an airplane. And, uh, that- that gives you a measure of how intelligent that system really is.

Lex Fridman 1:04:35 Yeah. Well, I mean, that- that's a big leap. I get you. But, uh, I'm more interested as a problem, I would see... To me, driving is a black box that can generate novel situations at some rate, like what people call edge cases. Like, so it does have newness that keeps being... Like, we're confronted, let's say, once a month.

François Chollet 1:04:59 It is a very long tail. Yes.

Lex Fridman 1:05:00 It's a long tail.

François Chollet 1:05:01 That doesn't mean you cannot solve it, uh, just by- by training a statistical model on lots of data.

Lex Fridman 1:05:08 Huge amount of data.

François Chollet 1:05:09 It's- it's really a matter of scale.

Lex Fridman 1:05:11 But I guess what I'm saying is, if you have a vehicle that achieves level five, it is going to be able to deal with new situations, or, I mean, the data is so large that the rate of new situations is very low.

François Chollet 1:05:32 Yes.

Lex Fridman 1:05:33 That's not intelligence. So, uh, if we go back to your kind of definition of intelligence, it's the efficiency.

François Chollet 1:05:39 With which you can adapt to new situations, to truly new situations, not situations you've seen before, right?

Lex Fridman 1:05:45 Right.

François Chollet 1:05:45 Not situations that could be anticipated by your creators, by the creators of the system, but truly new situations. The efficiency with which you acquire new skills. If you require... If in o- in order to pick up a new skill, you require, um, very extensive training datasets of most possible situations that can- that can occur in the practice of that skill, then the system is not intelligent. It is mostly just a, um, a lookup table.

Lex Fridman 1:06:15 Yeah. Well-

François Chollet 1:06:16 And likewise, if, uh, in order to acquire a skill, you need a human engineer to write down, uh, a bunch of rules that cover most or every possible situation, likewise, the system is- is not intelligent. The system is merely the output artifact, uh, of, uh, a process that- that happens, that happens in the minds, uh, of- of the engineers that are creating it, right? It is encoding, uh, an abstraction that's produced by the human mind, and intelligence with that would actually be, uh, the process of producing- of autonomously producing this obstruction.

Lex Fridman 1:06:56 Yeah.

François Chollet 1:06:57 Not... Like, if you take an obstruction and you encode it on a piece of paper or in a computer program, the obstruction itself is not intelligent. What's intelligent is the- the agent that's capable of producing these obstructions, right?

Lex Fridman 1:07:12 Yeah. It feels like there's a little bit of a gray area. Like, because you're basically saying that deep learning forms obstructions too, but those obstructions do not seem to be effective for generalizing far outside of the things that it's already seen, but generalize a little bit.

François Chollet 1:07:31 Yeah, absolutely. No, deep learning does generalize a little bit. Like, generalization is not- is not binary, it's more, like, a spectrum.

Lex Fridman 1:07:38 Yeah. And there's a certain point, it's a gray area, but there's a certain point where there's an impressive degree of generalization that happens-... no, like, the- I guess exactly what you were saying is, uh, intelligence is, um, how efficiently you're able to generalize far outside of the distribution of things you've seen already.

François Chollet 1:08:03 Yes.

Lex Fridman 1:08:03 So it's both, like the-

François Chollet 1:08:05 So it's-

Lex Fridman 1:08:05 ... the distance of how far you can... like, how new, how radically new something is and how efficiently-

François Chollet 1:08:11 Yes, absolutely.

Lex Fridman 1:08:11 ... you're able to deal with that.

François Chollet 1:08:12 So, you- you- you can think of, uh, intelligence as a measure of an information conversion ratio. Like, imagine a- a space of possible situations, and, um, you've covered some of them. Uh, so you have some amount of information, uh, about your space of possible situations that's provided by the situations you already know, and that's, on the other hand, also provided by, um, the prior knowledge that the system brings to the table, the prior knowledge that's embedded in the system. So, the system starts with some information, right? About the problem, about the task. And it's about going from that information to a program, what we would call a- a skill program, a behavioral program, that can cover a large area of possible situation space. Um, and essentially the ratio between that area and the amount of information you start with, uh, is intelligence. So a- a very smart agent, uh, can make efficient uses of very little, uh, information about a new problem and very little prior knowledge as well, to cover a very large area of- of potential situations in that problem without, uh, uh, knowing what these future new situations are- are going to be.

Lex Fridman 1:09:31 So, one of the other big things you talk about in- in the paper, we've talked about it a little bit already but let's talk about it some more, is, uh, the actual tests of intelligence. So, if we look at, like, human and machine intelligence, do you think tests of intelligence should be different for humans and machines, or how we think about testing of intelligence? Are these fundamentally the same kind of, uh, intelligences that we're after and therefore the tests should be similar?

François Chollet 1:10:03 So if your goal is to create, uh, uh, AIs that are, uh, more human-like, then it will be super valuable, obviously, to have a test, uh, that's- that's universal, that applies to both, uh, AIs, uh, and humans so that you can- you could establish a comparison, uh, between the two that you could tell exactly how, uh, intelligent, in terms of human intelligence, uh, a given system is. So that said, the constraints, uh, that apply to artificial intelligence and to human intelligence are very different, and your test should, uh, uh, account, uh, for this difference, um, because if yo- if you get artificial systems, it's always possible, uh, for an experimenter to buy, uh, arbitrary levels of skill at arbitrary tasks, either by, um, injecting hardcoded prior knowledge, uh, into the system via rules, uh, uh, and so on, uh, that come from the human mind, from the minds of the programmers, and also, um, buying, uh, higher levels of skill just by training on more data.

Lex Fridman 1:11:15 Mm-hmm.

François Chollet 1:11:15 Uh, for instance, you could generate an infinity of different Go games and you could train a- a- a Go-playing system, uh, that way, but you could not directly compare it to human Go-playing skills because a human that plays Go had to develop that skill in a very constrained environment. They had a limited amount of time, they had a limited amount of energy, uh, and of course, uh, uh, this started from a different set of priors. This started from, uh, um, you know, innate, uh, uh, human priors. Um, so I think if you want to compare the intelligence of two systems, like the intelligence of an AI and the intelligence, uh, of a human, you have to, um, control for priors. You have to, uh, start from the- the- the same set of knowledge priors about the task and you have to control for- for experience, uh, that is to say for training data.

Lex Fridman 1:12:11 So prior- what's priors?

François Chollet 1:12:15 So prior is whatever information, uh, you have about a given task before you start learning about this task.

Lex Fridman 1:12:23 And how's that different from experience?

François Chollet 1:12:25 Well, experience is acquired, right? So for instance, if you're- if you're trying to play Go, uh, your experience with Go is all the Go games you've played or- or- or you've seen or you've simulated in your mind, let's say. And, uh, your priors are things like, well, Go- Go is a game on a- on a 2D grid, uh, and we have lots of hardcoded, uh, priors about, uh- uh- uh, the organization of 2D space.

Lex Fridman 1:12:53 And r- so rules of how the- the dynamics of this, the physics of this game in this 2D space?

François Chollet 1:13:00 Yes.

Lex Fridman 1:13:00 And the idea that you have ... that what winning is. (laughs)

François Chollet 1:13:04 Yes, exactly.

Lex Fridman 1:13:05 So like-

François Chollet 1:13:05 And all ot- other board games can also share some similarities to Go, and if you've played these board games then, uh, with respect to the game of Go, that would be part of your priors about the game.

Lex Fridman 1:13:16 Well, it's interesting to think about the game of Go as how many priors are actually brought to the table. When you look at, um, self-play reinforcement learning based mechanisms that do learning, it seems like the number of priors is pretty low.

François Chollet 1:13:31 Yes.

Lex Fridman 1:13:31 But you're saying you should be ex- you should be-

François Chollet 1:13:32 There is a 2D special prior though, in the covenant.

Lex Fridman 1:13:35 Right. But you should be clear at making those priors explicit.

François Chollet 1:13:40 Yes. Uh, so in particular, I think if your- if your goal is to measure a human-like form of intelligence-... then you should clearly establish that you want, uh, the AI you're testing to start from, uh, the same set of priors that humans start with.

Lex Fridman 1:13:57 Right. So,

Lex Fridman 1:13:59 I mean, uh, t- to me personally but I think to a lot of people, the human side of things is very interesting. So, testing intelligence for humans, what, um, what do you think is a good test of human intelligence?

François Chollet 1:14:14 Well, that's the question that psychometrics is- is interested in. And-

Lex Fridman 1:14:19 What's-

François Chollet 1:14:19 ... there's an entire subfield of psychology, uh, that deals with this question.

Lex Fridman 1:14:23 So, what's psychometrics?

François Chollet 1:14:25 So, psychometrics is the subfield of psychology that- that tries to, uh, measure or quantify aspects of the human mind. So, in particular, cognitive abilities, intelligence, and personality traits as well.

Lex Fridman 1:14:39 So, uh, like, what are... it might be a weird question, but what are, like, the first principles of the- of psychometrics that is- operates on, you know... (laughs) what- what- what are the priors it brings to the table?

François Chollet 1:14:55 So, uh, it's a field with a- with a fairly long history. Um, it's- s- so, you know, psychology sometimes gets a- a bad reputation for not having very reproducible, uh, results and so on. Uh, psychometrics is actually some fairly solidly reproducible results. So, the ideal goals, uh, of the field is, you know, a test should be- be reliable, which is a- a- a notion tied to reproducibility. It should be valid, uh, meaning that it should actually measure what you says- what you say it measures. Um, so for instance, if you're- if you're saying that you're measuring intelligence, then your test results should be correlated with things that y- you expect to be correlated with intelligence, like success in school or success in the workplace and so on. Should be standardized, meaning that, uh, you can administer your test to many different people in the same conditions. Uh, and it should be free from bias, meaning that, for instance, uh, if your- if- if your test involves, uh, the English language, then you have to be aware that, uh, this creates a bias, uh, against people who have English as their second language or people who can't speak English at all. Um, so of course, these, uh, these principles for creating psychometric tests are, um, very much 90-year-old. I don't think every psychometric test is, uh- is really either reliable, uh, um, valid, or are free from bias. But at least through- the field is aware, uh, of these weaknesses and is- is trying to address them.

Lex Fridman 1:16:27 So, it's kind of interesting. Um, ultimately, you're only able to measure, like you said previously, the skill, but you're trying to f- do a bunch of measures of different skills that correlate-

François Chollet 1:16:38 Yes.

Lex Fridman 1:16:38 ... as you mentioned strongly with some general concept of cognitive ability.

François Chollet 1:16:43 Yes. Yes.

Lex Fridman 1:16:44 So, what's the g-factor?

François Chollet 1:16:46 So, right. There are many different kinds of- of tests is- tests of, uh, intelligence and, uh, each of them is interested in- in, uh, uh, different aspects of intelligence, you know. Some of them will deal with language, some of them will deal with, uh, uh, uh, spatial vision, maybe mental rotations, numbers and so on. When you run these very different tests at scale, what you start seeing is that there are clusters of correlations, uh, among test results. So, for instance, if you look at, uh, uh, homework at school, um, you will see that people who do well at math are also likely, statistically, to do well in physics.

Lex Fridman 1:17:25 Mm-hmm.

François Chollet 1:17:25 And what's more, uh, they are- they are also people who do well at math and physics are also statistically likely to do well in things that sound completely unrelated, like writing an- an English essay, for instance.

Lex Fridman 1:17:38 Mm-hmm.

François Chollet 1:17:38 And so when you see clusters of correlations, uh, uh, in s- in statistical terms, you would explain them with a latent variable, and the latent variable that would, for instance, explain, uh, the relationship between being good at math and being good at physics would be, uh, cognitive ability, right?

Lex Fridman 1:17:57 Mm-hmm.

François Chollet 1:17:57 And the g-factor is the- the latent variable that explains, uh, the fact that every test of intelligence that you can come up with results on that- on- on- on this test end up being correlated. So, there is some, uh, single, uh, uh, unique variable, uh, that- that explains this correlation, so that's the g-factor. So, it's a statistical construct. It's not really something you can directly measure, for instance, in a- in a person. Um-

Lex Fridman 1:18:25 But it's there.

François Chollet 1:18:26 But it's there. It's there, it's there at scale, and that's also one thing I want to, uh, mention about psychometrics. Like, you know, when- when you talk about measuring intelligence in- in humans, for instance, some people get a little bit worried. They will say, "No, that sounds dangerous. Maybe that sounds potentially discriminatory," and so on. And they're not wrong. And the thing is... so personally, I'm not interested in psychometrics, uh, as a way to characterize one, uh, individual person. Like if, uh- if I get your psychometric personality assessment or your IQ, I don't think that actually tells me much, uh, about you as a person. I think psychometrics is most useful, uh, as a statistical tool, so it's most useful at scale. Uh, it's most useful when you start getting test results for a large number of people and you start, uh, cross-correlating these test results because that gives you information about the structure, uh, of the human mind, in particular about the structure of human cognitive abilities. So, at scale, psychometrics paints a certain picture of the human mind and that's interesting, and that's what's driven to AI, the structure of human cognitive abilities.

Lex Fridman 1:19:41 Yeah. It g- gives you an insight into... I mean, to me, I remember when I learned about g-factor, it seemed, um...... it, it seemed like it would be impossible for i- even- i- it to be real, even as a statistical variable. Like, it felt, uh, kind of like astrology. Like, it's like wishful thinking among psychologists. But, uh, the more I learned, uh, I realized that there is some... I mean, I'm not sure what to make about human beings, the fact that the g factor is a thing, that there's a commonality across all of the human species-

François Chollet 1:20:13 Yes.

Lex Fridman 1:20:13 ... that there does seem to be a, a strong correlation between cognitive abilities.

François Chollet 1:20:17 Yeah.

Lex Fridman 1:20:17 That's kind of fascinating, actually.

François Chollet 1:20:19 Yeah. So, human cognitive abilities have, uh, a structure. Like the, the most mainstream theory of the structure of cognitive abilities is called, uh, uh, CHC Theory. It's, uh, Cattell, Horn, Carroll is the name of, uh, the three psychologists who contributed key pieces of it. And it describes, uh, cognitive abilities as a hierarchy with three levels. And at the top, you have the g factor. Then you have broad cognitive abilities, uh, for instance, fluid intelligence, right, um, that, that encompass, um, uh, a broad set of possible, uh, kinds of tasks that are all k- uh, uh, related. And then you have, uh, narrow cognitive abilities at the last level, which is, uh, closer to task-specific skill. And, uh, th- there are actually different theories, uh, of the structure of cognitive abilities, they just emerge from different statistical analysis of, uh, IQ test results. Uh, but they all describe, uh, a hierarchy with, uh, a kind of g factor, uh, at the top. And you're right that the g factor is ... it's not quite real in the sense that it's not something you can observe and measure like your height, for instance. But it's real in the sense that you, you see it in, in a statistical analysis of the data, right? One thing I want to mention is that the fact that there is a g factor does not really mean that human intelligence is, uh, general in a strong sense. It does not mean human intelligence can, can be applied to any problem at all and that someone w- who has a high IQ is gonna be able to solve any problem at all. That's not quite what it means. I think, um, one, uh, one, uh, popular analogy to understand it is the sports, uh, analogy. Uh, if you consider the concept of, uh, physical fitness, it's a concept that's very similar to intelligence because it's a useful concept, it's, uh, something you can intuitively understand. Some people are, are fit, uh, maybe like you. Some people are not as fit, maybe like me. Um-

Lex Fridman 1:22:20 But none of us can fly. (laughs)

François Chollet 1:22:23 Absolutely. So-

Lex Fridman 1:22:24 It's, it's constrained to a specific set of skills.

François Chollet 1:22:24 Even, even if you're very fit, that doesn't mean you can do, uh, uh, anything at all in any environment. You, you obviously cannot fly, you cannot, uh, survive at the bottom of, of the ocean and so on. And i- if you were a scientist and you want- you wanted to precisely define and measure physical fitness in humans, then you would come up with a, a battery, uh, of tests. Uh, like you would, you know, have running 100 meter, uh, playing soccer, playing table tennis, swimming, and so on. And, uh, if you run these tests over many different people, you would start seeing correlations in test results. For instance, people who are good at soccer are also good at, at sprinting, right? And, uh, you would explain these correlations with, uh, physical abilities that are, uh, strictly analogous to cognitive abilities, right? And then you would start also observing, uh, correlations between, uh, biological, uh, uh, characteristics, like maybe lung volume is correlated with being a, a, a fast runner, for instance, uh, in the same way that there, there are neurophysical, uh, uh, correlates, uh, of cognitive abilities, right? And at the top of the, uh, hierarchy of physical abilities that you would be able to observe, you would have a, a g factor, a physical g factor, which would map to physical fitness, right? And, uh, as you just said, that doesn't mean that, uh, people with a, with high physical fitness can fly. It doesn't mean, uh, human morphology and human physiology is universal. It's actually super specialized. We can only do the things, um, that we were, uh, evolved to do. Right? Like, we, we are not appropriate to, to, to... You, you could not exist on, on Venus or Mars or in the void of space or the bottom of the ocean. So, that said, one thing that's really striking and remarkable, um, is that, uh, uh, our morphology, uh, generalizes far beyond the environments that we evolved for. Like, in a way, you could say we evolved to run a- after prey in the savanna, right?

Lex Fridman 1:24:32 Mm-hmm.

François Chollet 1:24:32 That's very much where our, uh, human morphology comes from. And that said, we, we can, we can do a lot of things that are, that are completely unrelated to that. We can climb mountains, we can, we can swim across lakes. Uh, we can play, uh, table tennis. I mean, table tennis is very different from what we were evolved to do, right? Uh, so our morphology, our bodies, our sensorimotor affordances are of a degree of generality that is absolutely remarkable, right? And I think cognition is very similar to that. Our cognitive abilities have a degree of generality that goes far beyond what the mind was initially supposed to do, which is why we can, you know, play music and write novels and, and, and go to Mars and do all, all kinds of crazy things. Uh, but it's not universal in the same way that human morphology and our body, uh, is not appropriate for actually most of the universe by volume. In the same way you could say that the human mind is not really appropriate for most of problem space, potential problem space, uh, by volume. So, we have very strong, uh, cognitive biases actually that mean that there are certain types of problems that we handle very well and certain, certain types of problem that we are, uh, completely inadapted for. So, that, that's really how I would interpret, uh, the g factor. It's not a sign of strong generality. Uh, um, it's, it's really just a broader, the broadest cognitive ability. Uh, but our abilities, whether we are talking about sensorimotor abilities, uh, or cognitive abilities, they, they still, they remain very specialized in the human condition, right?

Lex Fridman 1:26:12 ... within the constraints of the human cognition, they're general. (laughs)

François Chollet 1:26:18 Yes, absolutely.

Lex Fridman 1:26:19 So, but the constraints, as you're saying, are very limited-

François Chollet 1:26:21 What, what's, I think what's, yeah-

Lex Fridman 1:26:23 ... very limiting.

François Chollet 1:26:24 So, we, we evolved our cognition and our body, evolved in, in very specific environments. Because our environment was so viable, fast changing, and so unpredictable, part of the constraints that, that drove our evolution is generality itself. So, we were in a way evolved to, to be able to improvise in all kinds of, of physical or cognitive environments, right?

Lex Fridman 1:26:46 Yeah.

François Chollet 1:26:46 Um, and for this reason, it turns out that, uh, the, the minds and bodies that we ended up with, uh, can be applied to much, much broader scope than what they were evolved for, right? And that's truly remarkable. And that goes, that's a degree of generalization that is far beyond anything you can see in artificial systems today. Right? Um, that said, it, it does not mean that, that, uh, uh, human intelligence is anywhere universal.

Lex Fridman 1:27:16 It's, yeah, it's not general.

Lex Fridman 1:27:18 You know, it's a kind of exciting topic for people, even out, you know, outside of artificial intelligence is IQ tests. They're, I think it's Mensa, whatever, there's different degrees of difficulty for questions. We talked about this offline a little bit too, about sort of difficult questions. You know, what makes a question on an IQ test, um, more difficult or less difficult, do you think?

François Chollet 1:27:43 So, the, the thing to keep in mind is that there's no such thing as a question that's intrinsically difficult. It has to be difficult with respects to the things you already know and the things you can already do, right? So, in, uh, in terms of, uh, an IQ test question, typically we'd have, uh, it would be structured, for instance, as a set of demonstration input and output pairs, right? And then you would be given, uh, a test input, a prompt, and you w- you, you would need to recognize or produce the corresponding output. And in that narrow context, you could say a difficult question, uh, is a question where, um, the input prompt is very surprising and unexpected given the, the training examples.

Lex Fridman 1:28:36 Just even the nature of the patterns that you're observing in the input prompt.

François Chollet 1:28:40 Yes. For instance, let's say you have a, a, a rotation problem. You must rotate a shape by 90 degrees. If I give you two examples and then I give you one, one, uh, prompt, which is actually one of the two training examples, then there is zero generalization difficulty for the task. It's actually a trivial task. You, you just recognize that it's one, one of the training examples and you produce the same answer. Now, if it's, uh, if it's a more complex shape, there is, you know, a little bit more generalization, but it remains that you are still doing the same thing at test time as you were, uh, being demonstrated at, at training time. A difficult task is a task that will require some amount of, uh, uh, test time adaptation, some amount of, uh, improvisation. Right? So, uh, consider, I don't know, uh, you're, you're teaching a class on like quantum physics or something. Um, if, uh, if you wanted to kinda test the understanding that students have of the material, you would come up with, uh, an exam, uh, that's very different from anything they've seen, like on the internet when they were cramming. On the other hand, if you wanted to make it easy, you would just give them something that's, uh, very similar to the, the mock exams that, that they've, that they've taken. Something that's just a simple interpolation of questions that they've al- they've already seen. And so, that would be an easy exam. It's very similar to what you've been trained on. And a difficult exam is one