Joe Rogan Experience #1188 - Lex Fridman

1. 0:00 – 4:02

   Lex’s origin story: understanding the mind by trying to build it

   1. LFLex Fridman0:00

      (laughs)

   2. JRJoe Rogan0:01

      (laughs) Four, three, two, one. Hello, Lex.

   3. LFLex Fridman0:09

      Hey, Joe.

   4. JRJoe Rogan0:09

      We're here, man. What's going on?

   5. LFLex Fridman0:11

      We're here. Mecca.

   6. JRJoe Rogan0:11

      Thanks for doing this. You brought notes. You're seriously prepared.

   7. LFLex Fridman0:15

      When you're jumping out of a plane, it's best to bring a parachute. This is my parachute.

   8. JRJoe Rogan0:20

      I, I understand. Yeah. Um, how long have you been working in artificial intelligence?

   9. LFLex Fridman0:26

      My whole life, I think.

   10. JRJoe Rogan0:27

       Really?

   11. LFLex Fridman0:27

       So I've, uh, when I was a kid, wanted to become a psychiatrist. I wanted to understand the human mind. I think the human mind is the most beautiful mystery that our entire civilization has taken on exploring through science. I think, you look up at the stars and you look at the universe out there, you had Neil deGrasse Tyson here, it's an amazing, beautiful scientific journey that we're taking on in exploring the stars, but the mind, to me, is a bigger mystery and more fascinating. And it's been the thing I've been fascinated by from the very beginning of my life, and just I think all of human civilization has been wondering, you know, what is in this- inside this thing, the hundred trillion connections that are just firing all the time, somehow making the magic happen to where you and I can look at each other, make words, all the fear, love, life, death that happens is all because of this thing in here. And understanding why is fascinating. And what I early on understood is that one of the best ways, for me at least, to understand the human mind is to try to build it, and that's what artificial intelligence-

   12. JRJoe Rogan1:49

       Ah.

   13. LFLex Fridman1:49

       ... is, you know, i- it's, it's not enough to s- from a psychology perspective to study, from a psychiatry perspective to i- investigate from the outside. The best way to understand is to do.

   14. JRJoe Rogan2:05

       So, you mean almost like reverse engineering a brain.

   15. LFLex Fridman2:09

       There's some stuff, exactly, reverse engineering the brain, there's some stuff that you can't understand until you try to do it. You can hypothesize your... I mean, we're both martial artists from various, uh, directions, you can hypothesize about what is the best martial art, but until you get in the ring, like what the UFC did, and test ideas is when you first realize that the touch of death that I've seen some YouTube videos on, that you perhaps cannot kill a person with a single touch, or your mind, or telepathy, that there are certain things that work, wrestling works, punching works. Okay, can we make it better? Can we create something like a touch of death? Can we figure out how to turn the hips, how to deliver a punch in the way that does do a significant amount of damage? And then you've, at that moment, when you start to try to do it, and you face some of the people that are trying to do the same thing, that's the scientific process. And you try, you actually begin to understand what is intelligence, and you begin to also understand how little we understand. It's like, uh, Richard Feynman, who I'm dressed after today-

   16. JRJoe Rogan3:29

       Are you? (laughs)

   17. LFLex Fridman3:31

       (laughs) He's a physicist. I'm not sure if you're familiar with him.

   18. JRJoe Rogan3:32

       Sure, yeah.

   19. LFLex Fridman3:34

       Yeah. Yeah, he always used to wear this exact thing, so I, I feel, I feel pretty badass wearing it. Uh-

   20. JRJoe Rogan3:39

       "If you think you know astrophysics, you don't know astrophysics."

   21. LFLex Fridman3:42

       That's right.

   22. JRJoe Rogan3:43

       Well, he said it about quantum physics, right?

   23. LFLex Fridman3:44

       Quantum physics.

   24. JRJoe Rogan3:45

       Yeah.

   25. LFLex Fridman3:45

       That's right. That's right. Uh, so he was a, a quantum physicist. And he kind of, uh, I remember hearing him talk about s- that... understanding our, the nature of the universe, uh, of reality could be like an onion. We don't know.

   26. JRJoe Rogan4:01

       Mm.

2. 4:02 – 7:30

   AlphaGo, creativity, and what it means for AI progress

   1. LFLex Fridman4:02

      But it could be like an onion to where you think you know, you're studying a layer of an onion, and then you peel it away and there's more, and you keep doing it and there's an infinite number of layers. With intelligence, there's the same kind of component to where we think we know, we got it, we figured it out, we figured out how to beat the human world champion in chess. We solved intelligence. And then we try the next thing. Wait a minute, Go is really difficult to solve as a game. And then you say, "Okay, it's, uh, I've" I came up when the game of Go was impossible for artificial intelligence systems to beat, and have now recently have been beaten. And-

   2. JRJoe Rogan4:37

      Within the last, like, five years, right?

   3. LFLex Fridman4:38

      ... in the next, the last five years. There's a lot of technical fascinating things of why that victory is interesting and important for artificial intelligence.

   4. JRJoe Rogan4:46

      It requires creativity, correct?

   5. LFLex Fridman4:48

      It r- it does not.

   6. JRJoe Rogan4:50

      No?

   7. LFLex Fridman4:50

      It a- it just exhibits creativity.

   8. JRJoe Rogan4:52

      Oh.

   9. LFLex Fridman4:52

      So, uh, the technical aspects of why AlphaGo from Google DeepMind that, that was the v- the designers and the builders of the system that was the victor, they did a few very interesting technical things where essentially you develop a neural network, this is this type of artificial intelligence system, that looks at a board of Go, has a lot of elements on it, has black and white pieces, and is able to tell you how good is this situation and how can I make it better? And that idea... So chess players can do this. I'm not actually that familiar with the game of Go so I can't speak t- I'm Russian, so chess to us is romanticized. It's a beautiful game. Uh, I think that there... you look at a board and all your previous experiences, all the things you've developed over tens of years of practice and thinking, you get this instinct of what is the right path to, to follow and that's exactly what the neural network is doing. And some of the in- uh, some of the paths it has come up with are-... surprising to other world champions. So in that sense it says, "Whoa, this thing's exhibiting creativity," because it's coming up with solutions that are something that's outside the box, thinking from the perspective of the human.

   10. JRJoe Rogan6:16

       When... Why, why do you differentiate between requires creativity and exhibits creativity?

   11. LFLex Fridman6:23

       I think, one, because we don't really understand what creativity is.

   12. JRJoe Rogan6:27

       Mm.

   13. LFLex Fridman6:27

       So, it, it's almost... It, it's on the level of concepts such as consciousness. For example, the question which there's a lot of thinking about whether creating something intelligent requires consciousness, requires for us to be actual living beings aware of our own existence. In the same way, does doing something like building an autonomous vehicle, that's the area where I work in, does that require creativity? Does that even require something like consciousness and self-awareness? I mean, I'm sure in LA, there's some degree of creativity required to navigate traffic. And in that sense, you, you start to think, are there solutions that are outside of the box an AI system needs to create? It, it's... Once you start to build it, you realize that to us humans, certain things appear creative, certain things don't, certain things we take for granted, certain things we find beautiful, and certain things we're like, "Yeah, yeah, that's, that's boring."

3. 7:30 – 13:04

   Creativity, the muse, and AI as ‘forging the gods’

   1. JRJoe Rogan7:30

      Well, there's creativity in different levels, right? There's creativity like to write The Stand, the Stephen King novel.

   2. LFLex Fridman7:35

      Yeah, that's great.

   3. JRJoe Rogan7:36

      That requires creativity. There's something about his... He's creating these stories. He's giving voices to these characters. He's developing these scenarios and these dramatic sequences in the book that's gonna get you really sucked in. That's, that's almost undeniable creativity, right?

   4. LFLex Fridman7:55

      Is it? So...

   5. JRJoe Rogan7:56

      Is it? Yeah.

   6. LFLex Fridman7:57

      Yeah. It's... He's imagining a world.

   7. JRJoe Rogan7:58

      Mm-hmm.

   8. LFLex Fridman7:59

      What is it, always set in, uh, New Hampshire or Massachusetts or s-

   9. JRJoe Rogan8:02

      Well, I would say Maine.

   10. LFLex Fridman8:03

       Maine. Yeah, that's right.

   11. JRJoe Rogan8:04

       Yeah.

   12. LFLex Fridman8:04

       So he's imagining a world, and imagining the emotion of different levels surrounding that world. Yeah, that's, that's creative. Although a few- there's a few really good books, including his own, that talks about writing. Uh-

   13. JRJoe Rogan8:17

       Yeah, he's got a great book on writing.

   14. LFLex Fridman8:18

       And an au-

   15. JRJoe Rogan8:19

       Au- It's actually called On Writing, Stephen King-

   16. LFLex Fridman8:20

       On writing.

   17. JRJoe Rogan8:20

       ... On Writing. Yeah.

   18. LFLex Fridman8:21

       If there's anyone who can write a book on writing, it should be Stephen King.

   19. JRJoe Rogan8:24

       Mm-hmm.

   20. LFLex Fridman8:25

       Uh, s- I think Steven Pressfield, I hope I'm not saying that wrong-

   21. JRJoe Rogan8:28

       Mm-hmm. The War of Art.

   22. LFLex Fridman8:28

       The War of Art.

   23. JRJoe Rogan8:29

       Yeah.

   24. LFLex Fridman8:29

       Beautiful book. And I would say if, from my recollection, they don't necessarily talk about creativity very much, that it's really hard work of putting in the hours of every day of just grinding it out.

   25. JRJoe Rogan8:42

       Well, Pressfield talks about the muse. Pressfield th- speaks of it almost in like a strange mystical, mystical sort of connection to the unknown.

   26. LFLex Fridman8:53

       Right.

   27. JRJoe Rogan8:54

       'Cause he, he, uh, almost... I'm not even exactly sure if he believes in the muse, but he... I think, if I could put words in his mouth... I have met him. He's a great guy. He was on the podcast once. I think the way he treats it is that if you decide the muse is real and you show up every day and you write as if the muse is real, you get the benefits of the muse being real.

   28. LFLex Fridman9:19

       That's right.

   29. JRJoe Rogan9:20

       Whether or not there's actually a muse that's giving you these wonderful ideas...

   30. LFLex Fridman9:24

       And what is the muse? So I think of artificial intelligence the same way. There's a quote by, uh, Pamela, uh, McCorduck from 1979 book that I really like. Uh, when... She talks about the history of artificial intelligence, "AI began with an ancient wish to forge the gods." And to me, gods, broadly speaking, or religions, represents... It's kind of like the muse. It represents the limits of possibility, the limits of our imagination. So it's this thing that we don't quite understand, that is the muse, that is God. This... Uh, us, us chimps are very narrow in our ability to perceive and understand the world, and there's clearly a much bigger, beautiful, mysterious world out there, and God or the muse represents that world. And for many people, I think throughout history, and especially in the, in the past sort of 100 years, artificial intelligence has become to represent that a little bit, to the thing which we don't understand and we crave, we're both terrified and we crave in creating this thing that is greater, that is able to understand the world better than us. In that, in that sense, artificial intelligence is the desire to create the muse, this other, this imaginary thing. And I think the... one of the beautiful things, if you talk about for everybody from Elon Musk to Sam Harris to all the people thinking about this, is that there's a mix of fear of that, of, of that unknown, of creating that unknown, and an excitement for it, because there's something in human nature that desires creating that. Because like I said, creating is how you understand.
4. 13:04 – 26:31

   Sci‑fi realism, ‘cut-the-shit’ moments, and why movies get AI wrong

   1. JRJoe Rogan13:04

      Did you see Alien: Covenant?

   2. LFLex Fridman13:07

      Is that a sci-fi movie?

   3. JRJoe Rogan13:08

      Yeah.

   4. LFLex Fridman13:09

      (laughs) No.

   5. JRJoe Rogan13:10

      Have you ever seen any of the Alien films?

   6. LFLex Fridman13:12

      I, uh, so I grew up in the Soviet Union where the, um (laughs) we didn't watch too many movies, so I need to catch up, I-

   7. JRJoe Rogan13:21

      Well, you should catch up on that one in particular because a lot of it has to do with artificial intelligence. There's actually a- a battle between, spoiler alert, two different but identical, um, artificially intelligent synthetic beings-

   8. LFLex Fridman13:37

      Mm-hmm.

   9. JRJoe Rogan13:37

      ... that are there to aid the people on the ship. One of them is very creative and one of them is not. And the one that is not has to save them from the one that is. Spoiler alert. Uh, I don't wanna tell you who wins.

   10. LFLex Fridman13:52

       Right.

   11. JRJoe Rogan13:52

       But there's- there's a really fascinating, um, scene at the very beginning of the movie where the creator of this artificially intelligent being is discussing its existence with the being itself and the being is trying to figure out who made him.

   12. LFLex Fridman14:10

       Mm-hmm.

   13. JRJoe Rogan14:10

       And- and it's this really fascinating moment and this being winds up being a- a bit of a problem because it possesses creativity and it- it has the ability to think for itself and may, uh, they've- they found it to be a problem-

   14. LFLex Fridman14:29

       Mm-hmm.

   15. JRJoe Rogan14:29

       ... so they made a different version of it which was not able to create and the one that was not able to create was much more of- more of a servant and so there's this battle between these two. I think you would find it quite fascinating. It's a really good movie.

   16. LFLex Fridman14:43

       Yeah, the same kinda theme carries through, uh, Ex Machina and-

   17. JRJoe Rogan14:47

       Mm-hmm, yeah.

   18. LFLex Fridman14:48

       ... and 2001: Space Odyssey, sort of-

   19. JRJoe Rogan14:50

       You've seen Ex Machina?

   20. LFLex Fridman14:51

       Yeah, I s- I've seen it. So because of your, I've listened to your podcast and because, because of it I've watched it a second time. Because the first time I watched it, a Neil deGrasse Tyson moment where I, it was, y- you said there's cut be- cut the-

   21. JRJoe Rogan15:05

       Cut the shit?

   22. LFLex Fridman15:05

       ... cut the shit moments.

   23. JRJoe Rogan15:06

       Yes (laughs) .

   24. LFLex Fridman15:08

       For me, for me, the, it, the, the movie opening is, everyth- everything about it was, uh, I- I was rolling my eyes the first time.

   25. JRJoe Rogan15:15

       Why were you rolling your eyes? What was the cut the shit moment?

   26. LFLex Fridman15:20

       So, uh, that's a general bad tendency that I'd like to talk about amongst people who are scientists that are actually trying to do stuff, uh, they're trying to build the- the thing. Uh, it's- it's very tempting to roll your eyes and tune out-

   27. JRJoe Rogan15:34

       Mm-hmm.

   28. LFLex Fridman15:34

       ... in a lot of aspects of artificial intelligence discussion and so on. For me, there's real reasons to roll your eyes and there's just... Well, let me, uh, let me just describe it. So the- this person in Ex Machina, no spoiler alerts, uh, is in the middle, what, like a Jurassic Park-type situation where he's like in the middle of a land that he owns?

   29. JRJoe Rogan15:56

       Yeah, we don't really know where it is-

   30. LFLex Fridman15:58

       Right.
5. 26:31 – 29:26

   Internet discourse vs real conversation, and the split camps on AGI

   1. JRJoe Rogan26:31

      I think very complex discussions should be had with people in person. That's what I think.

   2. LFLex Fridman26:36

      In person, yeah.

   3. JRJoe Rogan26:36

      And I think that when you allow comments, just random anonymous comments to enter into your consciousness, like you're, you're taking risks. And you may, um, you may run into a bunch of really brilliant ideas that are, um, you know, coming from people that are considerate, that have thought these things through, or you might just run into a river of assholes.

   4. LFLex Fridman27:00

      (laughs)

   5. JRJoe Rogan27:01

      And it's entirely possible. And I ... I, I peeked into my comments today on Twitter, and I was like, "What in the fuck?"

   6. LFLex Fridman27:07

      Yeah.

   7. JRJoe Rogan27:07

      I started reading like a couple of them, some just morons, and I'm like, all right, about some ... Shit, I didn't even know what the fuck they were talking about. But, but that's the risk you take when you dive in. You're gonna get people that are disproportionately upset. You're gonna get people that are disproportionate, you know, delusional or what- whatever it is in regards to your position on something.

   8. LFLex Fridman27:29

      Right.

   9. JRJoe Rogan27:29

      Or whether or not they even understand your position.

   10. LFLex Fridman27:31

       Right.

   11. JRJoe Rogan27:31

       They'll argue something that's an incorrect interpretation of your position.

   12. LFLex Fridman27:36

       Yeah. And you've actually, uh, from what I've heard, you've actually been to this podcast and so on, really good at being open-minded. And that's something I try to preach as well. S- so in AI discussions, when you're talking about AGI and talking about ... So there's a difference between narrow AI and general artificial intelligence. Narrow AI is the kind of things that are, uh, the kind of tools that are being applied now and being quite effective, and then there's general AI which is a broad categorization of concepts that are human-level or superhuman-level intelligence. And there ... When you talk about AGI, artificial general intelligence, there seems to be two camps of people: ones who are really working deep in it, like that's the camp I kind of sit in, and a lot of those folks tend to roll their eyes and just not engage into any discussions of the future. Their idea is saying, "It's really hard to do what we're doing, and it's just really hard to see how this becomes intelligent." And then there's another group of people who say, "Yeah, but be- you're being very shortsighted that you may not be able to do s- much now, but the exponential, the hard take-off, uh, overnight, it can become super intelligent and then it'll be too late to think about."

   13. JRJoe Rogan28:56

       Mm-hmm.

   14. LFLex Fridman28:56

       Now the problem with those two camps, as with any camps, Democrat or Republican, any camps, is they don't seem ... They be- seem to be talking past each other, as opposed to both have really interesting ideas. If, if you go back to the analogy of, uh, touch of death, uh, of this, this idea of MMA, right? So I'm the ... In this analogy, I'm gonna put myself in the UFC for a second, uh, in this analogy, I'm, you know, like ranked in the top 20,

6. 29:26 – 45:46

   Martial arts as a model for truth-testing (Aikido, Wing Chun, and humility)

   1. LFLex Fridman29:26

      I'm working really hard and my dream is to become a world champion, I'm training three times a day, I'm really working, I'm an engineer, I'm trying to build my skills up. And then there is other folks that come along like Steven Seagal and so on that kind of talk about other kinds of martial arts, other ideas of how you can do certain things. And I think, I think Steven Seagal and ... could ... is al- mi- might be onto something. I think we really need to be open-minded like, uh, An- Anderson Silva, I think, uh, talks to Steven Seagal, or somebody talks to Steven Seagal, right?

   2. JRJoe Rogan29:58

      The ... Well, Anderson Silva thinks Steven Seagal is ... I'm gonna, I'm gonna put this in a respectful way. He ... Anderson Silva has a wonderful sense of humor.

   3. LFLex Fridman30:12

      Mm-hmm.

   4. JRJoe Rogan30:13

      And Anderson Silva is very playful, and he thought it would be hilarious if-... if people believed that he was learning all of his martial arts from Steven Seagal.

   5. LFLex Fridman30:23

      From Steven Seagal, got it.

   6. JRJoe Rogan30:24

      He also loves Steven Seagal movies, legitimately, so treated him with a great deal of respect.

   7. LFLex Fridman30:30

      Right.

   8. JRJoe Rogan30:30

      He also recognizes that Steven Seagal actually is a master of Aikido. He really does understand Aikido and was one of the very first Westerners that was teaching in Japan, speaks fluent Japanese, sp- was teaching at a dojo in Japan, and is, you know, a, a legitimate master of Aikido.

   9. LFLex Fridman30:52

      Right.

   10. JRJoe Rogan30:53

       The problem with Aikido is it's, it's one of those martial arts that has merit i- in a, in a vacuum. Like, if you, if you're in a world where there's no p- NCAA wrestlers or no judo players or no Brazilian jiu-jitsu black belts or no, um, Muay Thai kickboxers-

   11. LFLex Fridman31:15

       Mm-hmm.

   12. JRJoe Rogan31:15

       ... there might be something to that Aikido stuff. But in the world where all those other martial arts exist, and we've examined all the intricacies of hand-to-hand combat, it falls horribly short.

   13. LFLex Fridman31:28

       Well, see, this is the point I'm trying to make. You just said that, "We've investigated, uh, all the intricacies."

   14. JRJoe Rogan31:35

       Yeah.

   15. LFLex Fridman31:35

       You said, "All the intricacies of hand-to-hand combat."

   16. JRJoe Rogan31:38

       Mm-hmm.

   17. LFLex Fridman31:38

       I mean, you're just speaking, but you wanna open your mind to the possibility that Aikido has, uh-

   18. JRJoe Rogan31:45

       Some techniques that are effective.

   19. LFLex Fridman31:46

       ... some techniques that are effective.

   20. JRJoe Rogan31:47

       Yeah, when I say all, that's, you're, you're correct. That's not a, uh, correct way of describing it.

   21. LFLex Fridman31:52

       Right.

   22. JRJoe Rogan31:53

       'Cause there's always new moves that are being ... Like, for instance, um, in this, uh, recent fight between Anthony Pettis and Tony Ferguson, Tony Ferguson actually used Wing Chun in a fight.

   23. LFLex Fridman32:05

       Mm-hmm.

   24. JRJoe Rogan32:05

       He, he trapped one of Anthony Pettis' hands and hit him with an elbow.

   25. LFLex Fridman32:10

       Right.

   26. JRJoe Rogan32:11

       He, uh, basically used a technique that you would use on a Wing Chun dummy-

   27. LFLex Fridman32:15

       Right.

   28. JRJoe Rogan32:15

       ... and he did it in an actual-

   29. LFLex Fridman32:16

       In an actual fight.

   30. JRJoe Rogan32:17

       ... world-class mixed martial arts fight. And I remember watching it, "Wow," going, "This crazy motherfucker actually pulled that off."
7. 45:46 – 47:53

   Near-term AI risks: bias, fairness, and why training data matters

   1. JRJoe Rogan45:46

      Let's go over those.

   2. LFLex Fridman45:47

      Yep. Fairness. So the w- the more and more we put decisions about our lives into the hands of artificial intelligence systems, whether you get a loan or, uh, an autonomous vehicle context, or in terms of, uh, re- recommending jobs for you on LinkedIn or all these kinds of things, the idea of fairness becomes a bias in, in these machine learning systems, becomes a really big threat. Because the way current neuro, uh, the way current artificial intelligence systems function is they train on data. So there's no way to, for them to somehow gain a greater intelligence than our, than the data we provide them with. So we provide them with actual data and so they carry over, if we're not careful, the biases in that data, the, the discrimination that's inherent in our current society as, as represented by the data. So they'll, they'll just carry that forward.

   3. JRJoe Rogan46:48

      How so?

   4. LFLex Fridman46:50

      Uh, so there's people working on this, uh, more so to s- uh, to show really the negative impacts, uh, in terms of getting a loan or whether to say whether this particular human being should be convicted or not of a crime. Or there's, there's ideas there that can carry... You know, in our criminal system, there's discrimination and if you use data from that criminal system to then assist deciders, judges, juries, lawyers in making this incriminating, in, in making a decision of what kind of penalty a person gets, they're gonna carry that forward.

   5. JRJoe Rogan47:29

      So you mean like racial, economic biases?

   6. LFLex Fridman47:31

      Racial, economic, yeah.

   7. JRJoe Rogan47:33

      Um, geographical?

   8. LFLex Fridman47:35

      And that's a... Sort of I don't study that e- exact problem, but it's, it's you're aware of it because of the tools we're using. It only... So the two ways... So I'd like to talk about neural networks with-

   9. JRJoe Rogan47:51

      Okay.

   10. LFLex Fridman47:51

       ... with Joe. (laughs)

   11. JRJoe Rogan47:52

       Sure. Let's do it.

8. 47:53 – 51:24

   How neural networks learn: datasets vs simulation (and why reality is hard)

   1. LFLex Fridman47:53

      Okay. So the current approaches are there's been a lot of, uh, demonstrated improvements, exciting new improvements in our advancements of our artificial intelligence and those are, for the most part, have to do with neural networks, something that's been around since the 1940s, has gone through two AI winters where everyone was super hyped and then super bummed and super hyped again and bummed again and now we're in this other hype cycle. And what neural networks are is these collections of interconnected simple compute units, they're all similar. It's kind of, like, it's inspired by our own brain. We have a bunch of little neurons interconnected and the idea is y- these interconnections are really dominant and random, but if you feed it with some data-... they'll learn to connect just like they do in our brain, in a way that interprets that data. They form representations of that data and can make decisions. But there's only two ways to train those neural networks that we have now. One is we have to provide a large dataset. If you want that neural network to tell the difference between a cat and a dog, you have to give it 10,000 images of a cat and 10,000 images of a dog. You need to give it those images. And who tells you what a picture of a cat and a dog is? It's humans, so it has to be annotated. So as teachers of these artificial intelligence systems, we have to collect this data, we have to invest significant amount of effort and co- annotate that data, and then we teach neural networks, uh, to make that prediction. The, what's not obvious there is how poor of a method there is to achieve any kind of greater degree of intelligence. You're just not able to get very far besides very specific narrow tasks of cat versus dog or, uh, should I give this person a loan or not, these kind of simple, simple tasks. I would argue autonomous vehicles are actually beyond the scope of that kind of approach. And then the other realm of where neural networks can be trained is if you can simulate that world. So if the world is simple enough or it's conducive to be formalized sufficiently to where you can simulate it, so a game of chess is just, it's- it's- there's rules. Game of Go, there's rules, so you can simulate it. The- the big exciting thing about Google DeepMind is that they were able to beat the world champion by doing something called competitive self-play, uh, which is to have two systems play against each other. They don't need the human. They play against each other. But that only works, and that's a beautiful idea and super powerful and really interesting and surprising, but that only works on things like games and simulation. So now if I wanted to, uh... Sorry to be going to analogies of like UFC for example. (laughs) I- if I wanted to train a system to become the world champion, uh, beat, uh, what's his name, Nurmagomedov, right? I could play the UFC game. I- I could create sys- that. I could create two neural networks that play, use competitive self-play to play in that virtual world and they could become state of the art, the best fighter ever in that game. But transferring that to the physical world, we don't know how to

9. 51:24 – 57:18

   Boston Dynamics fear vs reality: control algorithms, not ‘learning’ (yet)

   1. LFLex Fridman51:24

      do that. We don't know how to teach systems to do stuff in the real world. So some of the stuff that freaks you out often is Boston Dynamics robots.

   2. JRJoe Rogan51:32

      Ugh.

   3. LFLex Fridman51:33

      Yeah. Those, that- (laughs)

   4. JRJoe Rogan51:34

      Every day I go to the Instagram page and I just go, "What the fuck are you guys doing?"

   5. LFLex Fridman51:38

      So, uh-

   6. JRJoe Rogan51:39

      Engineering our demise.

   7. LFLex Fridman51:41

      (laughs) Mark Rober, uh, CEO is, uh, spoke at the class, uh, I taught. He is a, he calls himself a bad boy of robotics so he- he's having a little fun with it.

   8. JRJoe Rogan51:51

      He should definitely stop doing that.

   9. LFLex Fridman51:52

      (laughs)

   10. JRJoe Rogan51:52

       Don't call yourself a bad boy of anything.

   11. LFLex Fridman51:55

       That's true.

   12. JRJoe Rogan51:55

       How old is he?

   13. LFLex Fridman51:56

       (laughs)

   14. GUGuest51:57

       (laughs)

   15. LFLex Fridman51:58

       He's, he... (laughs) Okay, he's one of the greatest roboticists of our generation.

   16. JRJoe Rogan52:01

       That's great.

   17. LFLex Fridman52:01

       So-

   18. JRJoe Rogan52:02

       That's wonderful. However-

   19. LFLex Fridman52:03

       You just can't call yourself a-

   20. JRJoe Rogan52:04

       Don't call yourself a bad boy, bro.

   21. LFLex Fridman52:06

       Okay, so you-

   22. JRJoe Rogan52:09

       (laughs)

   23. GUGuest52:09

       (laughs)

   24. LFLex Fridman52:09

       (laughs) So you're not the bad boy of MMA?

   25. JRJoe Rogan52:11

       Definitely not.

   26. GUGuest52:11

       (laughs)

   27. LFLex Fridman52:12

       Okay. All right.

   28. JRJoe Rogan52:14

       (laughs)

   29. LFLex Fridman52:14

       (laughs) Um, hm. But-

   30. JRJoe Rogan52:16

       I'm not even the bad man.
10. 57:18 – 1:06:42

    Autonomous driving’s ‘onion layers’: from DARPA to LA traffic, LiDAR vs cameras

    1. LFLex Fridman57:18

       Uh, so first of all, that, I would argue, is we're quite far away from still, but that's within 10, 20 years.

    2. JRJoe Rogan57:25

       Well, what, how much can it do now?

    3. LFLex Fridman57:27

       It can, uh, stay inside the lane on the highway-

    4. JRJoe Rogan57:30

       Mm-hmm.

    5. LFLex Fridman57:30

       ... or on different roads, and it can change lanes. And what's being pushed now is they're trying to be able to enter and exit a highway.

    6. JRJoe Rogan57:38

       Hm.

    7. LFLex Fridman57:38

       So it's ba- some basic highway driving. It doesn't stop at traffic lights. It doesn't stop at stop signs. And it doesn't interact with the complex, irrational human beings, pedestrians-

    8. JRJoe Rogan57:52

       Mm-hmm.

    9. LFLex Fridman57:52

       ... cyclists, cars. That's, this is the-

    10. JRJoe Rogan57:55

        So-

    11. LFLex Fridman57:55

        ... onion I talked about, is we first, in 2005, the DARPA Grand Challenge, DARPA organized, uh, this challenge in the desert. It says, "Let's go across the desert. Let's see if we can build an autonomous vehicle that goes across the desert." 2004, they did the first one, and everybody failed. We're talking about some of the smartest people in the world really tried, uh, and, and failed. And so they did again in 2005. There's a few... Stanford won. There's a really badass guy from CMU, Red, that I think he's, like, a Marine. He led the team there. And they succeeded. The four teams finished. Stanford won. That was in the desert. And there was this feeling that we solved autonomous driving, but that's that onion, because you then, "Okay, what's the next step? We got a car that travels across a desert autonomously. What's the next?" So then in 2007, they did the Urban Grand Challe- the Urban Challenge, DARPA Urban Challenge, where you drove around the city a little bit. And again, super hard problem, people took it on. Uh, CMU won that one, uh, the Stanford s- second I believe. And, and then, then there was definitely a feeling like, "Yeah, we, now that we had a car drive around the city, it's, it's definitely solved." The problem is those cars were traveling super slow, first of all, and second of all, there's no pedestrians. There, there's no, it wasn't a real city. It was an art- artificial. It's just basically having to stop at different signs. Again, one other layer of the onion. And then you say, "Okay, when we actually have to put this car in a city like LA, how are we gonna make this work?" Because if there's no cars in the street and no pedestrians in the street, driving around is still hard, but doable and I think solvable in the next five years. When you put pedestrians, everybody jaywalks. (laughs)

    12. JRJoe Rogan59:47

        Mm-hmm.

    13. LFLex Fridman59:48

        If, if you put human beings into, into this interaction, it becomes much, much harder. Now, it's not impossible, and I think it's very doable, and with completely new interesting ideas, including revolutionizing infrastructure and rethinking of transportation in general, it's possible to do in the next five, 10 years, maybe 20. But it's not easy, like everybody says. And-

    14. JRJoe Rogan1:00:12

        But does anybody say it's easy?

    15. LFLex Fridman1:00:14

        Yeah. This, uh, s- the, this, there's a lot of hype between autonomous, uh, behind autonomous vehicles. Elon Musk himself and other people have promised autonomous vehicles. Th- that timeline has already passed. There's been going on, "In 2018, we'll have autonomous vehicles." Now, Ford, GM-

    16. JRJoe Rogan1:00:32

        Well, they're, they're semi-autonomous now, right? So-

    17. LFLex Fridman1:00:35

        Yeah.

    18. JRJoe Rogan1:00:35

        ... they, I know they do, they can brake for pedestrians. Like, if they see pedestrians, they're supposed to brake for them and avoid them. Right?

    19. LFLex Fridman1:00:42

        Th- that's part of the... Technically, no.

    20. JRJoe Rogan1:00:45

        Wasn't that an issue with an Uber car that hit a pedestrian that was o- operating autonomously?

    21. LFLex Fridman1:00:49

        That's right.

    22. JRJoe Rogan1:00:49

        Someone, a homeless person stepped out off of a median right into traffic and it, it nailed it, and then they found out it didn't have-... just one of the settings wasn't in place.

    23. LFLex Fridman1:00:59

        That's right. But that was an autonomous vehicle being tested in Arizona.

    24. JRJoe Rogan1:01:02

        Mm-hmm.

    25. LFLex Fridman1:01:03

        And, uh, unfortunately there was a fatality. A person, a person died.

    26. JRJoe Rogan1:01:06

        Yeah.

    27. LFLex Fridman1:01:06

        A pedestrian was killed. So what happened there, that's the, that's the thing I'm saying is really hard. That's full autonomy. That's technically when the car, you can remove the steering wheel and the car just drives itself and take care of-

    28. JRJoe Rogan1:01:20

        Right.

    29. LFLex Fridman1:01:20

        ... everything. Everything I've seen, everything we're studying, so we're studying drivers in Tesla vehicles, we're building our own vehicles. It seems that it'll be a long way off before we can solve the fully autonomous pr- driving problem.

    30. JRJoe Rogan1:01:34

        Because of pedestrians?
11. 1:06:42 – 1:51:23

    Predicting AI’s future: why experts are often wrong (and why smartphones matter)

    1. LFLex Fridman1:06:42

       And the thing... There's, there's a few examples that I brought along just 'cause I enjoy these predictions. So the, the, the, the... of how bad we are at predicting stuff. From the very engineers, the very guys and, and gals-... like me sitting before you, made some of the worst predictions in history, in, in terms of both pessimistic and optimistic. The Wright Brothers, one of the Wright Brothers, before they flew in 1903, predicted two years before that it'll be 50 years. "I confess that in 1901..." That's one of the brothers talking. "I said to my brother Orville, 'That man would not fly for 50 years.' Two years later, we ourselves were making flights. This demonstration of my inability as a prophet gave me such shock that I have ever since distrusted myself and have refrained from all prediction." That's one of the Wright Brothers, one of the people working on it. Two years-

    2. JRJoe Rogan1:07:41

       So that's a pessimistic estimation versus an optimistic exa- explanation.

    3. LFLex Fridman1:07:46

       Yep.

    4. JRJoe Rogan1:07:46

       Or a predication.

    5. LFLex Fridman1:07:47

       Exactly. And, and the same with, uh, Albert Einstein, for me, made these kind of pessimistic observations. Uh, for me, three years before the, the, the first critical chain reaction as part of the... He led the-

    6. JRJoe Rogan1:08:00

       Mm-hmm.

    7. LFLex Fridman1:08:00

       ... the nuclear de- development of the bomb. He, he said that it would... He has 90% confidence that it's impossible three years before. Okay, so that's on the pessimistic side. On the optimistic side, the history of AI is laden with, with optimistic predictions. Uh, the... In, uh, 1965, one of the seminal people in AI, Herbert Simon, said, "Machines will be capable within 20 years of doing any work a man can do." He also said, "Within 10 years, the digital computer will be the world's chess champion." That's in '58. And we didn't do that until '90-something, '98, so 40 years ago.

    8. JRJoe Rogan1:08:34

       Yeah, but that's one person, right?

    9. LFLex Fridman1:08:36

       Right.

    10. JRJoe Rogan1:08:36

        I mean, that's a guy taking a, a stab in the dark based on what data? I mean, what's he, what's he basing this off of?

    11. LFLex Fridman1:08:42

        Our imagination.

    12. JRJoe Rogan1:08:43

        Right, but you have more data points now though, don't you think?

    13. LFLex Fridman1:08:45

        No.

    14. JRJoe Rogan1:08:45

        In terms of... No?

    15. LFLex Fridman1:08:46

        Not about the future. That's the thing.

    16. JRJoe Rogan1:08:48

        Hmm, not about the future but about what's possible right now.

    17. LFLex Fridman1:08:53

        Right. And if you look at... The past is a really bad predictor of the future.

    18. JRJoe Rogan1:08:58

        Mm-hmm.

    19. LFLex Fridman1:08:58

        If you look at the past, what we've done, the i- immense advancement of technology has giving us, in many ways, optimism about what's possible.

    20. JRJoe Rogan1:09:08

        Right.

    21. LFLex Fridman1:09:09

        But exactly what is possible, we're not good at. So, I am much more confident that the world will look very fascinatingly different in the future. Whether AI will be part of that world is unclear. It could be we will all live in a virtual reality world. For... Or, for example, one of the things I'm really think about is, to me, a s- a s- a really dumb AI on one billion smartphones is potentially more impactful than a super intelligent AI on one smartphone.

    22. JRJoe Rogan1:09:45

        Hmm.

    23. LFLex Fridman1:09:46

        The, the, the, the fact that everybody now has smartphones, this kind of access to information, the way we communicate, the, the globalization of everything, the potential impact there of just even subtle improvements in AI could be in, in... could completely change the fabric of our society in a way where these discussions about an ex machina type lady walking around would be silly. Because we'll all be either living on Mars or living in virtual reality, or, you know, there's, there's so many exciting possibilities.

    24. JRJoe Rogan1:10:19

        Right.

    25. LFLex Fridman1:10:19

        And what I believe in is we have to think about them, we have to talk about them. Technology's always the source of danger, of risk. The... All of the biggest things that threatened our civilization at the small and large scale, all are connected to misuse of technology we develop. And at the same time, it's that very technology that will empower us and save us. So there's, uh, Max Tegmark, brilliant guy, Life 3.0. I recommend people read his book on the artificial general intelligence. He talks about the race, that there- there's a, there's a race that can't be stopped. One is the development of technology and the other is the development of our wisdom of how to stop or how to control the technology, and then this, this kind of race. And our wisdom is now... is always, like, one step behind. And then that's why we need to invest in it and keep, sort of, keep always thinking about new ideas. So right now, we're talking about AI. We don't know what it's going to look like in five years. We have to keep thinking about, we have to, uh, through simulation, explore different ideas, through conferences, have debates, come up with different, uh, approaches, uh, of how to solve particular problems, like I said, with bias or how to solve deepfakes where you fake... You can make Donald Trump or president, former President Obama say anything, or you can have Facebook advertisement har- hyper-targeted, uh, advertisements. How we can deal with those situations and constantly have this race of wisdom versus, uh, the development of technology, but not to sit and think, "Well, look at the, you know, look at the development of technology. Imagine what it could do in 50 years and we're all screwed." Because that's important to sort of be nervous about it in that way, but it's not conducive to what do we do about it. And the people that know what to do about it are the people trying to build this technology, building this future one step at a time.

    26. JRJoe Rogan1:12:33

        What do you mean by know what to do about it? 'Cause like l- let's, let's put it in terms of, uh, Elon Musk.

    27. LFLex Fridman1:12:38

        Right.

    28. JRJoe Rogan1:12:39

        Like Elon Musk is terrified of, of artificial intelligence 'cause he thinks by the time it becomes sentient, it'll be too late. It'll be smarter than us and we'll, we'll have essentially created our successors.

    29. LFLex Fridman1:12:50

        Yes. And let me quote Joe Rogan and say, "That's just one guy."

    30. JRJoe Rogan1:12:55

        Yeah. Well, Sam Harris thinks the same thing.

Episode duration: 2:55:44