WEBVTT 00:00.000 --> 00:05.060 As part of MIT course 6S 099 Artificial General Intelligence, I've gotten the chance to sit 00:05.060 --> 00:06.740 down with Max Tagmark. 00:06.740 --> 00:13.780 He is a professor here at MIT, he's a physicist, spent a large part of his career studying the 00:13.780 --> 00:20.660 mysteries of our cosmological universe, but he's also studied and delved into the beneficial 00:20.660 --> 00:25.860 possibilities and the existential risks of artificial intelligence. 00:25.860 --> 00:32.220 Amongst many other things, he's the cofounder of the Future of Life Institute, author of 00:32.220 --> 00:35.140 two books, both of which I highly recommend. 00:35.140 --> 00:40.220 First, our mathematical universe, second is Life 3.0. 00:40.220 --> 00:45.060 He's truly an out of the box thinker and a fun personality, so I really enjoy talking 00:45.060 --> 00:46.060 to him. 00:46.060 --> 00:49.500 If you'd like to see more of these videos in the future, please subscribe and also click 00:49.500 --> 00:52.980 the little bell icon to make sure you don't miss any videos. 00:52.980 --> 01:00.260 Also, Twitter, LinkedIn, AGI.MIT.IDU, if you want to watch other lectures or conversations 01:00.260 --> 01:01.260 like this one. 01:01.260 --> 01:07.980 Better yet, go read Max's book, Life 3.0, chapter 7 on goals is my favorite. 01:07.980 --> 01:12.300 It's really where philosophy and engineering come together and it opens with a quote by 01:12.300 --> 01:18.460 Dostoevsky, the mystery of human existence lies not in just staying alive, but in finding 01:18.460 --> 01:20.300 something to live for. 01:20.300 --> 01:27.100 Lastly, I believe that every failure rewards us with an opportunity to learn, in that sense 01:27.100 --> 01:33.060 I've been very fortunate to fail in so many new and exciting ways and this conversation 01:33.060 --> 01:34.060 was no different. 01:34.060 --> 01:41.260 I've learned about something called Radio Frequency Interference, RFI, look it up. 01:41.260 --> 01:45.500 Apparently music and conversations from local radio stations can bleed into the audio that 01:45.500 --> 01:49.380 you're recording in such a way that almost completely ruins that audio. 01:49.380 --> 01:52.460 It's an exceptionally difficult sound source to remove. 01:52.460 --> 01:59.620 So, I've gotten the opportunity to learn how to avoid RFI in the future during recording 01:59.620 --> 02:00.620 sessions. 02:00.620 --> 02:06.260 I've also gotten the opportunity to learn how to use Adobe Audition and iZotope RX6 02:06.260 --> 02:11.740 to do some audio repair. 02:11.740 --> 02:14.940 Of course, this is an exceptionally difficult noise to remove. 02:14.940 --> 02:20.380 I am an engineer, I'm not an audio engineer, neither is anybody else in our group, but 02:20.380 --> 02:21.780 we did our best. 02:21.780 --> 02:26.780 Nevertheless, I thank you for your patience and I hope you're still able to enjoy this 02:26.780 --> 02:27.780 conversation. 02:27.780 --> 02:31.460 Do you think there's intelligent life out there in the universe? 02:31.460 --> 02:33.420 Let's open up with an easy question. 02:33.420 --> 02:36.260 I have a minority view here actually. 02:36.260 --> 02:41.180 When I give public lectures, I often ask for show of hands who thinks there's intelligent 02:41.180 --> 02:47.060 life out there somewhere else and almost everyone puts their hands up and when I ask why, they'll 02:47.060 --> 02:52.060 be like, oh, there's so many galaxies out there, there's got to be. 02:52.060 --> 02:54.660 But I'm a number nerd, right? 02:54.660 --> 02:59.180 So when you look more carefully at it, it's not so clear at all. 02:59.180 --> 03:03.140 When we talk about our universe, first of all, we don't mean all of space. 03:03.140 --> 03:05.900 We actually mean, I don't know, you can throw me the universe if you want, it's behind you 03:05.900 --> 03:06.900 there. 03:06.900 --> 03:14.540 We simply mean the spherical region of space from which light has had time to reach us 03:14.540 --> 03:19.460 so far during the 13.8 billion years since our big bang. 03:19.460 --> 03:23.020 There's more space here, but this is what we call a universe because that's all we have 03:23.020 --> 03:24.140 access to. 03:24.140 --> 03:31.220 So is there intelligent life here that's gotten to the point of building telescopes and computers? 03:31.220 --> 03:39.500 My guess is no, actually, the probability of it happening on any given planet is some 03:39.500 --> 03:42.860 number we don't know what it is. 03:42.860 --> 03:49.340 And what we do know is that the number can't be super high because there's over a billion 03:49.340 --> 03:54.780 Earth like planets in the Milky Way galaxy alone, many of which are billions of years 03:54.780 --> 04:01.740 older than Earth, and aside from some UFO believers, you know, there isn't much evidence 04:01.740 --> 04:05.740 that any super advanced civilization has come here at all. 04:05.740 --> 04:08.700 And so that's the famous Fermi paradox, right? 04:08.700 --> 04:13.620 And then if you work the numbers, what you find is that if you have no clue what the 04:13.620 --> 04:18.500 probability is of getting life on a given planet, so it could be 10 to the minus 10, 04:18.500 --> 04:23.620 10 to the minus 20, or 10 to the minus two, or any power of 10 is sort of equally likely 04:23.620 --> 04:27.700 if you want to be really open minded, that translates into it being equally likely that 04:27.700 --> 04:34.700 our nearest neighbor is 10 to the 16 meters away, 10 to the 17 meters away, 10 to the 04:34.700 --> 04:35.700 18. 04:35.700 --> 04:42.860 Now, by the time you get much less than 10 to the 16 already, we pretty much know there 04:42.860 --> 04:46.220 is nothing else that's close. 04:46.220 --> 04:49.740 And when you get because it would have discovered us, they, yeah, they would have discovered 04:49.740 --> 04:53.540 us longer or if they're really close, we would have probably noted some engineering projects 04:53.540 --> 04:54.540 that they're doing. 04:54.540 --> 05:00.140 And if it's beyond 10 to the 26 meters, that's already outside of here. 05:00.140 --> 05:06.340 So my guess is actually that there are, we are the only life in here that's gotten the 05:06.340 --> 05:14.020 point of building advanced tech, which I think is very, puts a lot of responsibility on our 05:14.020 --> 05:18.140 shoulders, not screw up, you know, I think people who take for granted that it's okay 05:18.140 --> 05:23.300 for us to screw up, have an accidental nuclear war or go extinct somehow because there's 05:23.300 --> 05:27.460 a sort of Star Trek like situation out there where some other life forms are going to come 05:27.460 --> 05:30.380 and bail us out and it doesn't matter so much. 05:30.380 --> 05:33.380 I think they're leveling us into a false sense of security. 05:33.380 --> 05:37.540 I think it's much more prudent to say, let's be really grateful for this amazing opportunity 05:37.540 --> 05:44.180 we've had and make the best of it just in case it is down to us. 05:44.180 --> 05:50.220 So from a physics perspective, do you think intelligent life, so it's unique from a sort 05:50.220 --> 05:55.860 of statistical view of the size of the universe, but from the basic matter of the universe, 05:55.860 --> 06:00.100 how difficult is it for intelligent life to come about with the kind of advanced tech 06:00.100 --> 06:06.300 building life is implied in your statement that it's really difficult to create something 06:06.300 --> 06:07.620 like a human species? 06:07.620 --> 06:14.740 Well, I think what we know is that going from no life to having life that can do our level 06:14.740 --> 06:21.140 of tech, there's some sort of to going beyond that than actually settling our whole universe 06:21.140 --> 06:22.300 with life. 06:22.300 --> 06:30.700 There's some road major roadblock there, which is some great filter as I just sometimes called 06:30.700 --> 06:37.180 which, which tough to get through, it's either that that roadblock is either behind us or 06:37.180 --> 06:38.620 in front of us. 06:38.620 --> 06:40.980 I'm hoping very much that it's behind us. 06:40.980 --> 06:46.900 I'm super excited every time we get a new report from NASA saying they failed to find 06:46.900 --> 06:53.260 any life on Mars, because that suggests that the hard part, maybe it was getting the first 06:53.260 --> 06:59.540 ribosome or some some very low level kind of stepping stone. 06:59.540 --> 07:03.620 So they were home free because if that's true, then the future is really only limited by 07:03.620 --> 07:04.620 our own imagination. 07:04.620 --> 07:11.460 It would be much suckier if it turns out that this level of life is kind of a diamond dozen, 07:11.460 --> 07:12.780 but maybe there's some other problem. 07:12.780 --> 07:17.220 Like as soon as a civilization gets advanced technology within 100 years, they get into 07:17.220 --> 07:21.740 some stupid fight with themselves and poof, you know, that would be a bummer. 07:21.740 --> 07:22.740 Yeah. 07:22.740 --> 07:28.980 So you've explored the mysteries of the universe, the cosmological universe, the one that's 07:28.980 --> 07:36.340 between us today, I think you've also begun to explore the other universe, which is sort 07:36.340 --> 07:42.860 of the mystery, the mysterious universe of the mind of intelligence, of intelligent life. 07:42.860 --> 07:48.260 So is there a common thread between your interests or the way you think about space and intelligence? 07:48.260 --> 07:49.260 Oh, yeah. 07:49.260 --> 07:57.700 When I was a teenager, I was already very fascinated by the biggest questions and I felt that the 07:57.700 --> 08:03.660 two biggest mysteries of all in science were our universe out there and our universe in 08:03.660 --> 08:04.660 here. 08:04.660 --> 08:05.660 Yeah. 08:05.660 --> 08:11.260 So it's quite natural after having spent a quarter of a century on my career thinking 08:11.260 --> 08:12.260 a lot about this one. 08:12.260 --> 08:15.980 And now I'm indulging in the luxury of doing research on this one. 08:15.980 --> 08:17.660 It's just so cool. 08:17.660 --> 08:25.260 I feel the time is ripe now for you transparently deepening our understanding of this. 08:25.260 --> 08:26.420 Just start exploring this one. 08:26.420 --> 08:32.500 Yeah, because I think a lot of people view intelligence as something mysterious that 08:32.500 --> 08:38.340 can only exist in biological organisms like us and therefore dismiss all talk about artificial 08:38.340 --> 08:41.260 general intelligence is science fiction. 08:41.260 --> 08:47.260 But from my perspective as a physicist, I am a blob of quirks and electrons moving around 08:47.260 --> 08:50.180 in a certain pattern and processing information in certain ways. 08:50.180 --> 08:53.580 And this is also a blob of quirks and electrons. 08:53.580 --> 08:57.860 I'm not smarter than the water bottle because I'm made of different kind of quirks. 08:57.860 --> 09:02.220 I'm made of up quirks and down quirks exact same kind as this. 09:02.220 --> 09:07.020 It's a there's no secret sauce, I think in me, it's it's all about the pattern of the 09:07.020 --> 09:08.820 information processing. 09:08.820 --> 09:16.020 And this means that there's no law of physics saying that we can't create technology, which 09:16.020 --> 09:21.740 can help us by being incredibly intelligent and help us crack mysteries that we couldn't. 09:21.740 --> 09:25.580 In other words, I think we've really only seen the tip of the intelligence iceberg so 09:25.580 --> 09:26.580 far. 09:26.580 --> 09:27.580 Yeah. 09:27.580 --> 09:34.380 So the perceptronium, yeah, so you coined this amazing term, it's a hypothetical state 09:34.380 --> 09:39.420 of matter, sort of thinking from a physics perspective, what is the kind of matter that 09:39.420 --> 09:44.500 can help as you're saying, subjective experience emerge, consciousness emerge. 09:44.500 --> 09:50.140 So how do you think about consciousness from this physics perspective? 09:50.140 --> 09:51.980 Very good question. 09:51.980 --> 10:03.060 So, again, I think many people have underestimated our ability to make progress on this by convincing 10:03.060 --> 10:08.500 themselves it's hopeless because somehow we're missing some ingredient that we need. 10:08.500 --> 10:13.020 There's some new consciousness particle or whatever. 10:13.020 --> 10:19.660 I happen to think that we're not missing anything and that it's not the interesting thing about 10:19.660 --> 10:25.900 consciousness that gives us this amazing subjective experience of colors and sounds and emotions 10:25.900 --> 10:32.300 and so on is rather something at the higher level about the patterns of information processing. 10:32.300 --> 10:38.300 And that's why I like to think about this idea of perceptronium. 10:38.300 --> 10:44.220 What does it mean for an arbitrary physical system to be conscious in terms of what its 10:44.220 --> 10:47.100 particles are doing or its information is doing? 10:47.100 --> 10:52.300 I hate carbon chauvinism, this attitude, you have to be made of carbon atoms to be smart 10:52.300 --> 10:53.300 or conscious. 10:53.300 --> 10:58.180 So something about the information processing that this kind of matter performs. 10:58.180 --> 11:02.700 Yeah, and you can see I have my favorite equations here describing various fundamental 11:02.700 --> 11:04.660 aspects of the world. 11:04.660 --> 11:09.620 I think one day, maybe someone who's watching this will come up with the equations that 11:09.620 --> 11:12.140 information processing has to satisfy to be conscious. 11:12.140 --> 11:19.580 And I'm quite convinced there is big discovery to be made there because let's face it, we 11:19.580 --> 11:25.900 know that some information processing is conscious because we are conscious. 11:25.900 --> 11:28.980 But we also know that a lot of information processing is not conscious. 11:28.980 --> 11:32.980 Most of the information processing happening in your brain right now is not conscious. 11:32.980 --> 11:38.380 There are like 10 megabytes per second coming in even just through your visual system. 11:38.380 --> 11:42.940 You're not conscious about your heartbeat regulation or most things. 11:42.940 --> 11:47.300 Even if I just ask you to read what it says here, you look at it and then, oh, now you 11:47.300 --> 11:48.300 know what it said. 11:48.300 --> 11:51.820 But you're not aware of how the computation actually happened. 11:51.820 --> 11:57.020 Your consciousness is like the CEO that got an email at the end with the final answer. 11:57.020 --> 12:01.140 So what is it that makes a difference? 12:01.140 --> 12:06.620 I think that's both a great science mystery, we're actually studying it a little bit in 12:06.620 --> 12:12.260 my lab here at MIT, but I also think it's a really urgent question to answer. 12:12.260 --> 12:16.460 For starters, I mean, if you're an emergency room doctor and you have an unresponsive patient 12:16.460 --> 12:24.180 coming in, wouldn't it be great if in addition to having a CT scanner, you had a conscious 12:24.180 --> 12:30.780 scanner that could figure out whether this person is actually having locked in syndrome 12:30.780 --> 12:33.580 or is actually comatose. 12:33.580 --> 12:40.740 And in the future, imagine if we build robots or the machine that we can have really good 12:40.740 --> 12:45.100 conversations with, I think it's very likely to happen, right? 12:45.100 --> 12:50.020 Wouldn't you want to know if your home helper robot is actually experiencing anything or 12:50.020 --> 12:52.980 just like a zombie? 12:52.980 --> 12:53.980 Would you prefer it? 12:53.980 --> 12:54.980 What would you prefer? 12:54.980 --> 12:57.820 Would you prefer that it's actually unconscious so that you don't have to feel guilty about 12:57.820 --> 12:59.980 switching it off or giving boring chores? 12:59.980 --> 13:02.380 What would you prefer? 13:02.380 --> 13:09.780 Well, certainly we would prefer, I would prefer the appearance of consciousness, but the question 13:09.780 --> 13:15.300 is whether the appearance of consciousness is different than consciousness itself. 13:15.300 --> 13:21.420 And sort of ask that as a question, do you think we need to understand what consciousness 13:21.420 --> 13:28.420 is, solve the hard problem of consciousness in order to build something like an AGI system? 13:28.420 --> 13:29.420 No. 13:29.420 --> 13:31.140 I don't think that. 13:31.140 --> 13:36.220 I think we will probably be able to build things even if we don't answer that question. 13:36.220 --> 13:41.100 But if we want to make sure that what happens is a good thing, we better solve it first. 13:41.100 --> 13:47.220 So it's a wonderful controversy you're raising there, where you have basically three points 13:47.220 --> 13:50.220 of view about the hard problem. 13:50.220 --> 13:55.060 There are two different points of view that both conclude that the hard problem of consciousness 13:55.060 --> 13:56.060 is BS. 13:56.060 --> 14:01.100 On one hand, you have some people like Daniel Dennett who say that consciousness is just 14:01.100 --> 14:05.140 BS because consciousness is the same thing as intelligence. 14:05.140 --> 14:06.580 There's no difference. 14:06.580 --> 14:13.620 So anything which acts conscious is conscious, just like we are. 14:13.620 --> 14:18.820 And then there are also a lot of people, including many top AI researchers I know, who say, oh, 14:18.820 --> 14:22.820 consciousness is just bullshit because of course machines can never be conscious. 14:22.820 --> 14:28.020 They're always going to skiddy zombies, never have to feel guilty about how you treat them. 14:28.020 --> 14:35.380 And then there's a third group of people, including Giulio Tononi, for example, and another, and 14:35.380 --> 14:40.020 Gustav Koch and a number of others, I would put myself on this middle camp who say that 14:40.020 --> 14:44.260 actually some information processing is conscious and some is not. 14:44.260 --> 14:49.380 So let's find the equation which can be used to determine which it is. 14:49.380 --> 14:53.980 And I think we've just been a little bit lazy kind of running away from this problem for 14:53.980 --> 14:55.100 a long time. 14:55.100 --> 15:01.940 It's been almost taboo to even mention the C word in a lot of circles because, but we 15:01.940 --> 15:03.700 should stop making excuses. 15:03.700 --> 15:10.940 This is a science question and there are ways we can even test any theory that makes predictions 15:10.940 --> 15:12.140 for this. 15:12.140 --> 15:16.060 And coming back to this helper robot, I mean, so you said you would want your helper robot 15:16.060 --> 15:21.340 to certainly act conscious and treat you, like have conversations with you and stuff. 15:21.340 --> 15:24.860 But wouldn't you, would you feel a little bit creeped out if you realized that it was 15:24.860 --> 15:31.700 just a glossed up tape recorder, you know, that was just zombie and it's a faking emotion? 15:31.700 --> 15:37.220 Would you prefer that it actually had an experience or would you prefer that it's actually not 15:37.220 --> 15:42.300 experiencing anything so you feel, you don't have to feel guilty about what you do to it? 15:42.300 --> 15:46.580 It's such a difficult question because, you know, it's like when you're in a relationship 15:46.580 --> 15:49.860 and you say, well, I love you and the other person said I love you back. 15:49.860 --> 15:53.860 It's like asking, well, do they really love you back or are they just saying they love 15:53.860 --> 15:54.860 you back? 15:54.860 --> 15:59.620 Don't you really want them to actually love you? 15:59.620 --> 16:08.100 It's hard to, it's hard to really know the difference between everything seeming like 16:08.100 --> 16:14.820 there's consciousness present, there's intelligence present, there's affection, passion, love, 16:14.820 --> 16:16.180 and it actually being there. 16:16.180 --> 16:17.180 I'm not sure. 16:17.180 --> 16:18.180 Do you have... 16:18.180 --> 16:19.180 Can I ask you a question about this? 16:19.180 --> 16:20.180 Yes. 16:20.180 --> 16:21.180 To make it a bit more pointed. 16:21.180 --> 16:23.140 So Mass General Hospital is right across the river, right? 16:23.140 --> 16:29.180 Suppose you're going in for a medical procedure and they're like, you know, for anesthesia 16:29.180 --> 16:32.180 what we're going to do is we're going to give you muscle relaxance so you won't be able 16:32.180 --> 16:36.140 to move and you're going to feel excruciating pain during the whole surgery but you won't 16:36.140 --> 16:37.660 be able to do anything about it. 16:37.660 --> 16:42.020 But then we're going to give you this drug that erases your memory of it. 16:42.020 --> 16:45.420 Would you be cool about that? 16:45.420 --> 16:51.100 What's the difference that you're conscious about it or not if there's no behavioral change, 16:51.100 --> 16:52.100 right? 16:52.100 --> 16:53.100 Right. 16:53.100 --> 16:55.220 And that's a really clear way to put it. 16:55.220 --> 17:01.100 Yeah, it feels like in that sense, experiencing it is a valuable quality. 17:01.100 --> 17:09.140 So actually being able to have subjective experiences, at least in that case, is valuable. 17:09.140 --> 17:14.060 And I think we humans have a little bit of a bad track record also of making these self 17:14.060 --> 17:17.940 serving arguments that other entities aren't conscious. 17:17.940 --> 17:20.700 You know, people often say, oh, these animals can't feel pain. 17:20.700 --> 17:21.700 Right. 17:21.700 --> 17:25.580 It's okay to boil lobsters because we asked them if it hurt and they didn't say anything. 17:25.580 --> 17:29.180 And now there was just a paper out saying lobsters did do feel pain when you boil them 17:29.180 --> 17:31.180 and they're banning it in Switzerland. 17:31.180 --> 17:36.300 And we did this with slaves too often and said, oh, they don't mind. 17:36.300 --> 17:41.180 They don't maybe aren't conscious or women don't have souls or whatever. 17:41.180 --> 17:46.540 So I'm a little bit nervous when I hear people just take as an axiom that machines can't 17:46.540 --> 17:48.900 have experience ever. 17:48.900 --> 17:52.500 I think this is just a really fascinating science question is what it is. 17:52.500 --> 17:57.420 Let's research it and try to figure out what it is that makes the difference between unconscious 17:57.420 --> 18:01.220 intelligent behavior and conscious intelligent behavior. 18:01.220 --> 18:07.140 So in terms of, so if you think of a Boston Dynamics human or robot being sort of with 18:07.140 --> 18:13.420 a broom being pushed around, it starts pushing on a consciousness question. 18:13.420 --> 18:20.060 So let me ask, do you think an AGI system, like a few neuroscientists believe needs to 18:20.060 --> 18:25.860 have a physical embodiment, needs to have a body or something like a body? 18:25.860 --> 18:28.340 No, I don't think so. 18:28.340 --> 18:30.620 You mean to have a conscious experience? 18:30.620 --> 18:33.140 To have consciousness. 18:33.140 --> 18:37.860 I do think it helps a lot to have a physical embodiment to learn the kind of things about 18:37.860 --> 18:42.820 the world that are important to us humans for sure. 18:42.820 --> 18:47.460 But I don't think the physical embodiment is necessary after you've learned it. 18:47.460 --> 18:48.860 Just have the experience. 18:48.860 --> 18:51.500 Think about it when you're dreaming, right? 18:51.500 --> 18:55.500 Your eyes are closed, you're not getting any sensory input, you're not behaving or moving 18:55.500 --> 18:59.780 in any way, but there's still an experience there, right? 18:59.780 --> 19:03.220 And so clearly the experience that you have when you see something cool in your dreams 19:03.220 --> 19:08.660 isn't coming from your eyes, it's just the information processing itself in your brain, 19:08.660 --> 19:11.100 which is that experience, right? 19:11.100 --> 19:16.660 But if I put it another way, I'll say because it comes from neuroscience is the reason you 19:16.660 --> 19:24.620 want to have a body and a physical, something like a physical system is because you want 19:24.620 --> 19:27.100 to be able to preserve something. 19:27.100 --> 19:35.740 In order to have a self, you could argue, you'd need to have some kind of embodiment 19:35.740 --> 19:38.180 of self to want to preserve. 19:38.180 --> 19:45.940 Well, now we're getting a little bit anthropomorphic, anthropomorphizing things, maybe talking about 19:45.940 --> 19:47.820 self preservation instincts. 19:47.820 --> 19:50.700 We are evolved organisms, right? 19:50.700 --> 19:57.020 So Darwinian evolution endowed us and other evolved organisms with self preservation instinct 19:57.020 --> 20:03.100 because those that didn't have those self preservation genes got cleaned out of the gene pool. 20:03.100 --> 20:09.180 But if you build an artificial general intelligence, the mind space that you can design is much, 20:09.180 --> 20:14.500 much larger than just a specific subset of minds that can evolve that have. 20:14.500 --> 20:19.260 So an AGI mind doesn't necessarily have to have any self preservation instinct. 20:19.260 --> 20:24.100 It also doesn't necessarily have to be so individualistic as us. 20:24.100 --> 20:28.140 Like imagine if you could just, first of all, we're also very afraid of death, you know, 20:28.140 --> 20:32.180 as opposed to you could back yourself up every five minutes and then your airplane is about 20:32.180 --> 20:33.180 to crash. 20:33.180 --> 20:37.340 You're like, shucks, I'm just, I'm going to lose the last five minutes of experiences 20:37.340 --> 20:41.580 since my last cloud backup, dang, you know, it's not as big a deal. 20:41.580 --> 20:47.380 Or if we could just copy experiences between our minds easily, like which we could easily 20:47.380 --> 20:50.620 do if we were silicon based, right? 20:50.620 --> 20:55.860 Then maybe we would feel a little bit more like a hive mind, actually, that maybe it's 20:55.860 --> 21:01.220 the, so, so there's, so I don't think we should take for granted at all that AGI will have 21:01.220 --> 21:06.820 to have any of those sort of competitive as alpha male instincts. 21:06.820 --> 21:07.820 Right. 21:07.820 --> 21:12.820 On the other hand, you know, this is really interesting because I think some people go 21:12.820 --> 21:17.900 too far and say, of course, we don't have to have any concerns either that advanced 21:17.900 --> 21:22.700 AI will have those instincts because we can build anything we want. 21:22.700 --> 21:27.420 That there's, there's a very nice set of arguments going back to Steve Omohandro and 21:27.420 --> 21:32.900 Nick Bostrom and others just pointing out that when we build machines, we normally build 21:32.900 --> 21:37.700 them with some kind of goal, you know, win this chess game, drive this car safely or 21:37.700 --> 21:38.700 whatever. 21:38.700 --> 21:42.540 And as soon as you put in a goal into machine, especially if it's kind of open ended goal 21:42.540 --> 21:48.460 and the machine is very intelligent, it'll break that down into a bunch of sub goals. 21:48.460 --> 21:53.500 And one of those goals will almost always be self preservation because if it breaks 21:53.500 --> 21:56.140 or dies in the process, it's not going to accomplish the goal, right? 21:56.140 --> 21:59.540 Like, suppose you just build a little, you have a little robot and you tell it to go 21:59.540 --> 22:05.460 down the store market here and, and get you some food, make you cook your Italian dinner, 22:05.460 --> 22:09.540 you know, and then someone mugs it and tries to break it on the way. 22:09.540 --> 22:15.380 That robot has an incentive to not get destroyed and defend itself for a runaway because otherwise 22:15.380 --> 22:17.780 it's going to fail and cooking your dinner. 22:17.780 --> 22:22.940 It's not afraid of death, but it really wants to complete the dinner cooking goal. 22:22.940 --> 22:24.780 So it will have a self preservation instinct. 22:24.780 --> 22:26.820 It will continue being a functional agent. 22:26.820 --> 22:27.820 Yeah. 22:27.820 --> 22:35.860 And, and, and similarly, if you give any kind of more ambitious goal to an AGI, it's very 22:35.860 --> 22:39.940 likely they want to acquire more resources so it can do that better. 22:39.940 --> 22:44.500 And it's exactly from those sort of sub goals that we might not have intended that some 22:44.500 --> 22:50.740 of the concerns about AGI safety come, you give it some goal that seems completely harmless. 22:50.740 --> 22:55.540 And then before you realize it, it's also trying to do these other things which you 22:55.540 --> 22:59.220 didn't want it to do and it's maybe smarter than us. 22:59.220 --> 23:08.220 So, so, and let me pause just because I am in a very kind of human centric way, see fear 23:08.220 --> 23:11.900 of death as a valuable motivator. 23:11.900 --> 23:17.220 So you don't think you think that's an artifact of evolution. 23:17.220 --> 23:21.980 So that's the kind of mind space evolution created that we're sort of almost obsessed 23:21.980 --> 23:22.980 about self preservation. 23:22.980 --> 23:23.980 Yeah. 23:23.980 --> 23:29.500 Some kind of genetic well, you don't think that's necessary to be afraid of death. 23:29.500 --> 23:34.980 So not just a kind of sub goal of self preservation just so you can keep doing the thing, but 23:34.980 --> 23:42.980 more fundamentally sort of have the finite thing like this ends for you at some point. 23:42.980 --> 23:43.980 Interesting. 23:43.980 --> 23:47.500 Do I think it's necessary for what precisely? 23:47.500 --> 23:51.020 For intelligence, but also for consciousness. 23:51.020 --> 23:58.220 So for those for both, do you think really like a finite death and the fear of it is 23:58.220 --> 24:01.020 important? 24:01.020 --> 24:06.980 So before I can answer, before we can agree on whether it's necessary for intelligence 24:06.980 --> 24:10.660 or for consciousness, we should be clear on how we define those two words because a lot 24:10.660 --> 24:13.340 are really smart people define them in very different ways. 24:13.340 --> 24:18.500 I was in this on this panel with AI experts and they couldn't, they couldn't agree on 24:18.500 --> 24:20.180 how to define intelligence even. 24:20.180 --> 24:24.860 So I define intelligence simply as the ability to accomplish complex goals. 24:24.860 --> 24:30.740 I like your broad definition because again, I don't want to be a carbon chauvinist. 24:30.740 --> 24:36.580 And in that case, no, certainly it doesn't require fear of death. 24:36.580 --> 24:40.100 I would say AlphaGo AlphaZero is quite intelligent. 24:40.100 --> 24:44.260 I don't think AlphaZero has any fear of being turned off because it doesn't understand the 24:44.260 --> 24:52.180 concept of even and similarly consciousness, I mean, you can certainly imagine a very simple 24:52.180 --> 24:57.660 kind of experience if certain plants have any kind of experience, I don't think they're 24:57.660 --> 25:00.940 very afraid of dying or there's nothing they can do about it anyway much. 25:00.940 --> 25:08.420 So there wasn't that much value and but more seriously, I think if you ask not just about 25:08.420 --> 25:15.460 being conscious, but maybe having what you would, we might call an exciting life for 25:15.460 --> 25:23.300 you for your passion and really appreciate the things, maybe there, somehow, maybe there 25:23.300 --> 25:29.180 perhaps it does help having a backdrop that, hey, it's finite, you know, let's make the 25:29.180 --> 25:31.380 most of this, let's live to the fullest. 25:31.380 --> 25:36.220 So if you knew you were going to just live forever, do you think you would change your 25:36.220 --> 25:40.500 career? Yeah, I mean, in some perspective, it would 25:40.500 --> 25:44.020 be an incredibly boring life living forever. 25:44.020 --> 25:49.740 So in the sort of loose, subjective terms that you said of something exciting and something 25:49.740 --> 25:55.180 in this that other humans would understand, I think, is yeah, it seems that the finiteness 25:55.180 --> 25:56.660 of it is important. 25:56.660 --> 26:02.420 Well, the good news I have for you then is based on what we understand about cosmology, 26:02.420 --> 26:10.460 things in our universe is probably finite, although big crunch or big or big, what's 26:10.460 --> 26:11.460 the extent of the infinite? 26:11.460 --> 26:16.820 Yeah, we could have a big chill or a big crunch or a big rip or death, the big snap or death 26:16.820 --> 26:17.820 bubbles. 26:17.820 --> 26:20.140 All of them are more than a billion years away. 26:20.140 --> 26:29.500 So we should we certainly have vastly more time than our ancestors thought, but still 26:29.500 --> 26:35.580 pretty hard to squeeze in an infinite number of compute cycles, even though there are some 26:35.580 --> 26:37.820 loopholes that just might be possible. 26:37.820 --> 26:44.620 But I think, you know, some people like to say that you should live as if you're about 26:44.620 --> 26:48.100 to you're going to die in five years or so, and that's sort of optimal. 26:48.100 --> 26:54.740 Maybe it's a good as some we should build our civilization as if it's all finite to 26:54.740 --> 26:55.740 be on the safe side. 26:55.740 --> 27:02.020 Right, exactly. So you mentioned in defining intelligence as the ability to solve complex 27:02.020 --> 27:03.020 goals. 27:03.020 --> 27:04.940 So where would you draw a line? 27:04.940 --> 27:10.940 How would you try to define human level intelligence and super human level intelligence? 27:10.940 --> 27:13.380 Where is consciousness part of that definition? 27:13.380 --> 27:16.860 No, consciousness does not come into this definition. 27:16.860 --> 27:21.580 So so I think of intelligence as it's a spectrum, but there are very many different kinds of 27:21.580 --> 27:22.580 goals you can have. 27:22.580 --> 27:27.140 You have a goal to be a good chess player, a good goal player, a good car driver, a good 27:27.140 --> 27:31.260 investor, good poet, etc. 27:31.260 --> 27:35.740 So intelligence that bind by its very nature, isn't something you can measure, but it's 27:35.740 --> 27:39.900 one number, some overall goodness, no, no, there are some people who are more better 27:39.900 --> 27:42.540 at this, some people are better at that. 27:42.540 --> 27:48.380 Right now we have machines that are much better than us at some very narrow tasks like multiplying 27:48.380 --> 27:57.620 large numbers fast, memorizing large databases, playing chess, playing go, soon driving cars. 27:57.620 --> 28:03.340 But there's still no machine that can match a human child in general intelligence. 28:03.340 --> 28:08.420 But artificial general intelligence, AGI, the name of your course, of course, that 28:08.420 --> 28:16.460 is by its very definition, the quest to build a machine that can do everything as well as 28:16.460 --> 28:17.460 we can. 28:17.460 --> 28:24.060 Up to the old Holy Grail of AI from back to its inception in the 60s. 28:24.060 --> 28:27.500 If that ever happens, of course, I think it's going to be the biggest transition in the 28:27.500 --> 28:33.860 history of life on Earth, but it doesn't necessarily have to wait the big impact until machines 28:33.860 --> 28:35.780 are better than us at knitting. 28:35.780 --> 28:41.940 The really big change doesn't come exactly at the moment they're better than us at everything. 28:41.940 --> 28:45.820 The really big change comes, first, their big change is when they start becoming better 28:45.820 --> 28:51.140 at us at doing most of the jobs that we do, because that takes away much of the demand 28:51.140 --> 28:53.380 for human labor. 28:53.380 --> 29:01.300 And then the really warping change comes when they become better than us at AI research. 29:01.300 --> 29:07.900 Because right now, the time scale of AI research is limited by the human research and development 29:07.900 --> 29:14.100 cycle of years, typically, along the take from one release of some software or iPhone 29:14.100 --> 29:16.300 or whatever to the next. 29:16.300 --> 29:25.820 But once Google can replace 40,000 engineers by 40,000 equivalent pieces of software or 29:25.820 --> 29:29.660 whatever, then there's no reason that has to be years. 29:29.660 --> 29:32.020 It can be, in principle, much faster. 29:32.020 --> 29:38.900 And the time scale of future progress in AI and all of science and technology will be 29:38.900 --> 29:40.980 driven by machines, not humans. 29:40.980 --> 29:49.660 So it's this simple point, which gives right this incredibly fun controversy about whether 29:49.660 --> 29:54.540 there can be intelligence explosion, so called singularity, as Werner Winge called it. 29:54.540 --> 30:00.060 The idea, as articulated by I.J. Good, is obviously way back fifties, but you can see 30:00.060 --> 30:07.220 Alan Turing and others thought about it even earlier. 30:07.220 --> 30:12.980 You asked me what exactly what I define human level intelligence. 30:12.980 --> 30:18.540 So the glib answer is just to say something which is better than us at all cognitive tasks 30:18.540 --> 30:21.980 or better than any human at all cognitive tasks. 30:21.980 --> 30:25.900 But the really interesting bar, I think, goes a little bit lower than that, actually. 30:25.900 --> 30:33.260 It's when they're better than us at AI programming and general learning so that they can, if 30:33.260 --> 30:37.340 they want to, get better than us at anything by just starting out. 30:37.340 --> 30:43.100 So there better is a key word and better is towards this kind of spectrum of the complexity 30:43.100 --> 30:45.740 of goals it's able to accomplish. 30:45.740 --> 30:53.060 So another way to, and that's certainly a very clear definition of human love. 30:53.060 --> 30:56.300 So there's, it's almost like a sea that's rising, you can do more and more and more 30:56.300 --> 30:57.300 things. 30:57.300 --> 30:59.900 It's actually a graphic that you show, it's really nice way to put it. 30:59.900 --> 31:04.340 So there's some peaks and there's an ocean level elevating and you solve more and more 31:04.340 --> 31:05.340 problems. 31:05.340 --> 31:09.220 But, you know, just kind of to take a pause and we took a bunch of questions and a lot 31:09.220 --> 31:14.380 of social networks and a bunch of people asked a sort of a slightly different direction 31:14.380 --> 31:22.260 on creativity and on things that perhaps aren't a peak. 31:22.260 --> 31:28.620 It's, you know, human beings are flawed and perhaps better means having being having contradiction 31:28.620 --> 31:30.260 being flawed in some way. 31:30.260 --> 31:34.980 So let me sort of, yeah, start and start easy, first of all. 31:34.980 --> 31:36.620 So you have a lot of cool equations. 31:36.620 --> 31:39.660 Let me ask, what's your favorite equation, first of all? 31:39.660 --> 31:43.580 I know they're all like your children, but which one is that? 31:43.580 --> 31:49.060 This is the Shreddinger equation, it's the master key of quantum mechanics of the micro 31:49.060 --> 31:50.060 world. 31:50.060 --> 31:55.340 So this equation can take everything to do with atoms and all the fuels and all the 31:55.340 --> 32:04.020 way up to… Yeah, so, okay, so quantum mechanics is certainly a beautiful mysterious formulation 32:04.020 --> 32:05.020 of our world. 32:05.020 --> 32:10.740 So I'd like to sort of ask you, just as an example, it perhaps doesn't have the same 32:10.740 --> 32:17.420 beauty as physics does, but in mathematics abstract, the Andrew Wiles who proved the 32:17.420 --> 32:19.460 Fermat's last theory. 32:19.460 --> 32:24.180 So he just saw this recently and it kind of caught my eye a little bit. 32:24.180 --> 32:27.980 This is 358 years after it was conjectured. 32:27.980 --> 32:32.940 So this very simple formulation, everybody tried to prove it, everybody failed. 32:32.940 --> 32:38.820 And so here's this guy comes along and eventually proves it and then fails to prove it and then 32:38.820 --> 32:41.340 proves it again in 94. 32:41.340 --> 32:45.940 And he said like the moment when everything connected into place, in an interview he said 32:45.940 --> 32:47.980 it was so indescribably beautiful. 32:47.980 --> 32:53.580 That moment when you finally realize the connecting piece of two conjectures, he said it was so 32:53.580 --> 32:56.940 indescribably beautiful, it was so simple and so elegant. 32:56.940 --> 33:01.540 I couldn't understand how I'd missed it and I just stared at it in disbelief for 20 33:01.540 --> 33:02.540 minutes. 33:02.540 --> 33:08.100 Then during the day I walked around the department and I keep coming back to my desk looking 33:08.100 --> 33:09.820 to see if it was still there. 33:09.820 --> 33:10.820 It was still there. 33:10.820 --> 33:11.820 I couldn't contain myself. 33:11.820 --> 33:12.820 I was so excited. 33:12.820 --> 33:16.180 It was the most important moment of my working life. 33:16.180 --> 33:18.940 Nothing I ever do again will mean as much. 33:18.940 --> 33:24.860 So that particular moment and it kind of made me think of what would it take? 33:24.860 --> 33:28.380 And I think we have all been there at small levels. 33:28.380 --> 33:34.820 Maybe let me ask, have you had a moment like that in your life where you just had an idea 33:34.820 --> 33:40.060 it's like, wow, yes. 33:40.060 --> 33:44.700 I wouldn't mention myself in the same breath as Andrew Wiles, but I certainly had a number 33:44.700 --> 33:54.820 of aha moments when I realized something very cool about physics just completely made 33:54.820 --> 33:55.820 my head explode. 33:55.820 --> 33:59.580 In fact, some of my favorite discoveries I made later, I later realized that they had 33:59.580 --> 34:03.340 been discovered earlier by someone who's sometimes got quite famous for it. 34:03.340 --> 34:07.460 So there's too late for me to even publish it, but that doesn't diminish in any way. 34:07.460 --> 34:12.340 The emotional experience you have when you realize it like, wow. 34:12.340 --> 34:17.460 So what would it take in that moment, that wow, that was yours in that moment? 34:17.460 --> 34:23.420 So what do you think it takes for an intelligent system, an AGI system, an AI system to have 34:23.420 --> 34:24.980 a moment like that? 34:24.980 --> 34:29.420 It's a tricky question because there are actually two parts to it, right? 34:29.420 --> 34:37.260 One of them is, can it accomplish that proof, can it prove that you can never write A to 34:37.260 --> 34:46.420 the N plus B to the N equals 3 to the N for all integers, etc., etc., when N is bigger 34:46.420 --> 34:49.420 than 2. 34:49.420 --> 34:51.580 That's simply the question about intelligence. 34:51.580 --> 34:54.420 Can you build machines that are that intelligent? 34:54.420 --> 34:59.860 And I think by the time we get a machine that can independently come up with that level 34:59.860 --> 35:03.460 of proofs, probably quite close to AGI. 35:03.460 --> 35:07.860 But the second question is a question about consciousness. 35:07.860 --> 35:13.060 When will we, how likely is it that such a machine would actually have any experience 35:13.060 --> 35:16.500 at all as opposed to just being like a zombie? 35:16.500 --> 35:22.940 And would we expect it to have some sort of emotional response to this or anything at 35:22.940 --> 35:31.140 all akin to human emotion where when it accomplishes its machine goal, it views it as something 35:31.140 --> 35:39.260 very positive and sublime and deeply meaningful. 35:39.260 --> 35:45.260 I would certainly hope that if in the future we do create machines that are our peers or 35:45.260 --> 35:53.700 even our descendants, I would certainly hope that they do have this sort of sublime appreciation 35:53.700 --> 36:06.020 of life in a way, my absolutely worst nightmare would be that at some point in the future, 36:06.020 --> 36:10.620 the distant future, maybe our cosmos is teeming with all this post biological life, doing 36:10.620 --> 36:13.180 all the seemingly cool stuff. 36:13.180 --> 36:20.660 And maybe the last humans by the time our species eventually fizzles out will be like, 36:20.660 --> 36:26.140 well, that's okay, because we're so proud of our descendants here and look, my worst 36:26.140 --> 36:30.580 nightmare is that we haven't solved the consciousness problem. 36:30.580 --> 36:34.100 And we haven't realized that these are all the zombies, they're not aware of anything 36:34.100 --> 36:37.900 anymore than a tape recorder, as in any kind of experience. 36:37.900 --> 36:41.660 So the whole thing has just become a play for empty benches. 36:41.660 --> 36:44.700 That would be like the ultimate zombie apocalypse to me. 36:44.700 --> 36:52.900 So I would much rather, in that case, that we have these beings which can really appreciate 36:52.900 --> 36:57.060 how amazing it is. 36:57.060 --> 37:02.260 And in that picture, what would be the role of creativity, what a few people ask about 37:02.260 --> 37:03.260 creativity? 37:03.260 --> 37:04.260 Yeah. 37:04.260 --> 37:08.700 And do you think, when you think about intelligence, I mean, certainly the story you told at the 37:08.700 --> 37:14.100 beginning of your book involved, you know, creating movies and so on, sort of making 37:14.100 --> 37:18.580 money, you know, you can make a lot of money in our modern world with music and movies. 37:18.580 --> 37:23.100 So if you are an intelligent system, you may want to get good at that. 37:23.100 --> 37:26.300 But that's not necessarily what I mean by creativity. 37:26.300 --> 37:32.620 Is it important on that complex goals where the sea is rising for there to be something 37:32.620 --> 37:39.940 creative, or am I being very human centric and thinking creativity somehow special relative 37:39.940 --> 37:41.940 to intelligence? 37:41.940 --> 37:50.940 My hunch is that we should think of creativity simply as an aspect of intelligence. 37:50.940 --> 37:57.820 And we have to be very careful with human vanity. 37:57.820 --> 38:01.540 We have this tendency to very often want to say, as soon as machines can do something, 38:01.540 --> 38:05.980 we try to diminish it and say, oh, but that's not like real intelligence, you know, is 38:05.980 --> 38:12.620 it not creative or this or that, the other thing, if we ask ourselves to write down a 38:12.620 --> 38:18.500 definition of what we actually mean by being creative, what we mean by Andrew Wiles, what 38:18.500 --> 38:23.660 he did there, for example, don't we often mean that someone takes a very unexpected 38:23.660 --> 38:26.060 leap? 38:26.060 --> 38:33.740 It's not like taking 573 and multiplying by 224 by just a step of straightforward cookbook 38:33.740 --> 38:36.500 like rules, right? 38:36.500 --> 38:40.660 You can maybe make a connection between two things that people have never thought was 38:40.660 --> 38:41.660 connected. 38:41.660 --> 38:42.660 It's very surprising. 38:42.660 --> 38:44.300 Something like that. 38:44.300 --> 38:50.660 I think this is an aspect of intelligence, and this is actually one of the most important 38:50.660 --> 38:53.260 aspects of it. 38:53.260 --> 38:57.940 Maybe the reason we humans tend to be better at it than traditional computers is because 38:57.940 --> 39:02.020 it's something that comes more naturally if you're a neural network than if you're a 39:02.020 --> 39:05.820 traditional logic gates based computer machine. 39:05.820 --> 39:11.900 We physically have all these connections, and if you activate here, activate here, activate 39:11.900 --> 39:20.980 here, it ping, you know, my hunch is that if we ever build a machine where you could 39:20.980 --> 39:31.020 just give it the task, hey, hey, you say, hey, you know, I just realized I want to travel 39:31.020 --> 39:32.380 around the world instead this month. 39:32.380 --> 39:34.700 Can you teach my AGI course for me? 39:34.700 --> 39:36.100 And it's like, okay, I'll do it. 39:36.100 --> 39:39.860 And it does everything that you would have done and it improvises and stuff. 39:39.860 --> 39:42.860 That would in my mind involve a lot of creativity. 39:42.860 --> 39:45.660 Yeah, so it's actually a beautiful way to put it. 39:45.660 --> 39:54.540 I think we do try to grasp at the definition of intelligence as everything we don't understand 39:54.540 --> 39:57.580 how to build. 39:57.580 --> 40:02.180 So we as humans try to find things that we have and machines don't have, and maybe creativity 40:02.180 --> 40:05.940 is just one of the things, one of the words we used to describe that. 40:05.940 --> 40:06.940 That's a really interesting way to put it. 40:06.940 --> 40:09.820 I don't think we need to be that defensive. 40:09.820 --> 40:14.700 I don't think anything good comes out of saying, we're somehow special, you know, it's 40:14.700 --> 40:27.540 very wise, there are many examples in history of where trying to pretend they were somehow 40:27.540 --> 40:36.220 superior to all other intelligent beings has led to pretty bad results, right? 40:36.220 --> 40:39.700 Nazi Germany, they said that they were somehow superior to other people. 40:39.700 --> 40:44.580 Today, we still do a lot of cruelty to animals by saying they were so superior somehow on 40:44.580 --> 40:50.500 the other, they can't feel pain, slavery was justified by the same kind of really weak 40:50.500 --> 40:52.420 arguments. 40:52.420 --> 40:58.700 And I don't think if we actually go ahead and build artificial general intelligence, 40:58.700 --> 41:01.100 it can do things better than us. 41:01.100 --> 41:08.980 I don't think we should try to found our self worth on some sort of bogus claims of superiority 41:08.980 --> 41:11.940 in terms of our intelligence. 41:11.940 --> 41:21.780 I think we should instead find our calling and the meaning of life from the experiences 41:21.780 --> 41:22.780 that we have. 41:22.780 --> 41:23.780 Right. 41:23.780 --> 41:30.260 You know, I can have very meaningful experiences even if there are other people who are smarter 41:30.260 --> 41:35.860 than me, you know, when I go to faculty meeting here and I was talking about something and 41:35.860 --> 41:39.420 then I certainly realized, oh, he has an old prize, he has an old prize, he has an old 41:39.420 --> 41:40.420 prize. 41:40.420 --> 41:41.420 Yeah. 41:41.420 --> 41:47.660 You know, it doesn't make me enjoy life any less or enjoy talking to those people less. 41:47.660 --> 41:49.780 Of course not. 41:49.780 --> 41:57.420 And contrary to that, I feel very honored and privileged to get to interact with other 41:57.420 --> 42:00.820 very intelligent beings that are better than me and a lot of stuff. 42:00.820 --> 42:05.420 So I don't think there's any reason why we can't have the same approach with intelligent 42:05.420 --> 42:06.420 machines. 42:06.420 --> 42:08.900 That's a really interesting, so people don't often think about that. 42:08.900 --> 42:14.380 They think about if there's machines that are more intelligent, you naturally think 42:14.380 --> 42:19.100 that that's not going to be a beneficial type of intelligence. 42:19.100 --> 42:24.060 You don't realize it could be, you know, like peers with no ball prizes that would be just 42:24.060 --> 42:25.060 fun to talk with. 42:25.060 --> 42:30.580 And they might be clever about certain topics and you can have fun having a few drinks with 42:30.580 --> 42:31.580 them. 42:31.580 --> 42:38.620 Well, also, you know, another example we can all relate to why it doesn't have to be a 42:38.620 --> 42:42.580 terrible thing to be impressed, the presence of people who are even smarter than us all 42:42.580 --> 42:47.980 around is when you and I were both two years old, I mean, our parents were much more intelligent 42:47.980 --> 42:48.980 than us. 42:48.980 --> 42:49.980 Right. 42:49.980 --> 42:50.980 Worked out okay. 42:50.980 --> 42:54.140 Because their goals were aligned with our goals. 42:54.140 --> 43:01.380 And that I think is really the number one key issue we have to solve if we value align 43:01.380 --> 43:07.380 the value alignment problem exactly because people who see too many Hollywood movies with 43:07.380 --> 43:12.260 lousy science fiction plot lines, they worry about the wrong thing, right? 43:12.260 --> 43:16.500 They worry about some machine suddenly turning evil. 43:16.500 --> 43:21.500 It's not malice that we should that is the concern. 43:21.500 --> 43:23.000 It's competence. 43:23.000 --> 43:29.580 By definition, intelligence makes you makes you very competent if you have a more intelligent 43:29.580 --> 43:35.300 goal playing machine computer playing as a less intelligent one and when we define intelligence 43:35.300 --> 43:37.740 as the ability to accomplish go winning, right? 43:37.740 --> 43:40.780 It's going to be the more intelligent one that wins. 43:40.780 --> 43:47.860 And if you have a human and then you have an AGI that's more intelligent in all ways 43:47.860 --> 43:50.500 and they have different goals, guess who's going to get their way, right? 43:50.500 --> 43:58.060 So I was just reading about this particular rhinoceros species that was driven extinct 43:58.060 --> 43:59.060 just a few years ago. 43:59.060 --> 44:05.740 Alan Bummer is looking at this cute picture of a mommy rhinoceros with its child, you 44:05.740 --> 44:09.140 know, and why did we humans drive it to extinction? 44:09.140 --> 44:12.860 It wasn't because we were evil rhino haters as a whole. 44:12.860 --> 44:16.380 It was just because we our goals weren't aligned with those of the rhinoceros and it didn't 44:16.380 --> 44:19.660 work out so well for the rhinoceros because we were more intelligent, right? 44:19.660 --> 44:27.220 So I think it's just so important that if we ever do build AGI before we unleash anything, 44:27.220 --> 44:37.380 we have to make sure that it learns to understand our goals, that it adopts our goals and retains 44:37.380 --> 44:38.380 those goals. 44:38.380 --> 44:45.740 So the cool interesting problem there is being able, us as human beings, trying to formulate 44:45.740 --> 44:47.240 our values. 44:47.240 --> 44:52.540 So you know, you could think of the United States Constitution as a way that people sat 44:52.540 --> 44:59.780 down at the time a bunch of white men, which is a good example, I should say. 44:59.780 --> 45:03.460 They formulated the goals for this country and a lot of people agree that those goals 45:03.460 --> 45:05.540 actually held up pretty well. 45:05.540 --> 45:09.600 It's an interesting formulation of values and failed miserably in other ways. 45:09.600 --> 45:15.500 So for the value alignment problem and the solution to it, we have to be able to put 45:15.500 --> 45:23.420 on paper or in a program, human values, how difficult do you think that is? 45:23.420 --> 45:24.420 Very. 45:24.420 --> 45:25.980 But it's so important. 45:25.980 --> 45:30.340 We really have to give it our best and it's difficult for two separate reasons. 45:30.340 --> 45:37.660 There's the technical value alignment problem of figuring out just how to make machines 45:37.660 --> 45:40.660 understand our goals, adopt them and retain them. 45:40.660 --> 45:46.140 And then there's the separate part of it, the philosophical part, whose values anyway. 45:46.140 --> 45:51.700 And since we, it's not like we have any great consensus on this planet on values, what mechanism 45:51.700 --> 45:56.780 should we create then to aggregate and decide, okay, what's a good compromise? 45:56.780 --> 46:01.260 That second discussion can't just be left the tech nerds like myself, right? 46:01.260 --> 46:02.260 That's right. 46:02.260 --> 46:06.820 And if we refuse to talk about it and then AGI gets built, who's going to be actually 46:06.820 --> 46:10.660 making the decision about whose values, it's going to be a bunch of dudes in some tech 46:10.660 --> 46:12.380 company, right? 46:12.380 --> 46:18.420 And are they necessarily so representative of all of humankind that we want to just 46:18.420 --> 46:19.580 endorse it to them? 46:19.580 --> 46:25.220 Are they even uniquely qualified to speak to future human happiness just because they're 46:25.220 --> 46:26.460 good at programming AI? 46:26.460 --> 46:30.380 I'd much rather have this be a really inclusive conversation. 46:30.380 --> 46:32.700 But do you think it's possible? 46:32.700 --> 46:38.820 You create a beautiful vision that includes sort of the diversity, cultural diversity 46:38.820 --> 46:43.900 and various perspectives on discussing rights, freedoms, human dignity. 46:43.900 --> 46:46.620 But how hard is it to come to that consensus? 46:46.620 --> 46:52.140 Do you think it's certainly a really important thing that we should all try to do, but do 46:52.140 --> 46:54.460 you think it's feasible? 46:54.460 --> 47:01.660 I think there's no better way to guarantee failure than to refuse to talk about it or 47:01.660 --> 47:02.980 refuse to try. 47:02.980 --> 47:08.060 And I also think it's a really bad strategy to say, okay, let's first have a discussion 47:08.060 --> 47:09.060 for a long time. 47:09.060 --> 47:13.540 And then once we reach complete consensus, then we'll try to load it into some machine. 47:13.540 --> 47:16.980 No, we shouldn't let perfect be the enemy of good. 47:16.980 --> 47:22.140 Instead, we should start with the kindergarten ethics that pretty much everybody agrees on 47:22.140 --> 47:24.580 and put that into our machines now. 47:24.580 --> 47:26.100 We're not doing that even. 47:26.100 --> 47:32.980 Look at anyone who builds a passenger aircraft wants it to never under any circumstances 47:32.980 --> 47:35.900 fly into a building or mountain, right? 47:35.900 --> 47:38.860 Yet the September 11 hijackers were able to do that. 47:38.860 --> 47:44.220 And even more embarrassingly, Andreas Lubitz, this depressed German wings pilot, when he 47:44.220 --> 47:50.220 flew his passenger jet into the Alps, killing over 100 people, he just told the autopilot 47:50.220 --> 47:51.220 to do it. 47:51.220 --> 47:55.140 He told the freaking computer to change the altitude to 100 meters. 47:55.140 --> 48:01.820 And even though it had the GPS maps, everything, the computer was like, okay, no, so we should 48:01.820 --> 48:07.300 take those very basic values, though, where the problem is not that we don't agree. 48:07.300 --> 48:12.460 The problem is just we've been too lazy to try to put it into our machines and make sure 48:12.460 --> 48:17.460 that from now on, airplanes will just, which all have computers in them, but we'll just 48:17.460 --> 48:19.820 never just refuse to do something like that. 48:19.820 --> 48:25.580 We go into safe mode, maybe lock the cockpit door, go to the nearest airport, and there's 48:25.580 --> 48:31.340 so much other technology in our world as well now where it's really coming quite timely 48:31.340 --> 48:34.300 to put in some sort of very basic values like this. 48:34.300 --> 48:41.460 Even in cars, we've had enough vehicle terrorism attacks by now where people have driven trucks 48:41.460 --> 48:47.300 and vans into pedestrians that it's not at all a crazy idea to just have that hardwired 48:47.300 --> 48:51.420 into the car, because yeah, there are a lot of, there's always going to be people who 48:51.420 --> 48:55.620 for some reason want to harm others, but most of those people don't have the technical 48:55.620 --> 48:58.620 expertise to figure out how to work around something like that. 48:58.620 --> 49:01.780 So if the car just won't do it, it helps. 49:01.780 --> 49:02.940 So let's start there. 49:02.940 --> 49:05.020 So there's a lot of, that's a great point. 49:05.020 --> 49:06.900 So not chasing perfect. 49:06.900 --> 49:10.780 There's a lot of things that most of the world agrees on. 49:10.780 --> 49:11.940 Yeah, let's start there. 49:11.940 --> 49:12.940 Let's start there. 49:12.940 --> 49:18.140 And then once we start there, we'll also get into the habit of having these kind of conversations 49:18.140 --> 49:21.940 about, okay, what else should we put in here and have these discussions? 49:21.940 --> 49:24.100 This should be a gradual process then. 49:24.100 --> 49:25.100 Great. 49:25.100 --> 49:31.380 So, but that also means describing these things and describing it to a machine. 49:31.380 --> 49:35.620 So one thing, we had a few conversations with Steven Wolfram. 49:35.620 --> 49:37.140 I'm not sure if you're familiar with Steven Wolfram. 49:37.140 --> 49:38.500 Oh yeah, I know him quite well. 49:38.500 --> 49:43.380 So he has, you know, he works with a bunch of things, but you know, cellular automata, 49:43.380 --> 49:47.660 these simple computable things, these computation systems. 49:47.660 --> 49:52.380 And he kind of mentioned that, you know, we probably have already within these systems 49:52.380 --> 49:59.580 already something that's AGI, meaning like we just don't know it because we can't talk 49:59.580 --> 50:00.580 to it. 50:00.580 --> 50:06.380 So if you give me this chance to try it, to try to at least form a question out of this, 50:06.380 --> 50:12.780 because I think it's an interesting idea to think that we can have intelligent systems, 50:12.780 --> 50:17.260 but we don't know how to describe something to them and they can't communicate with us. 50:17.260 --> 50:21.220 I know you're doing a little bit of work in explainable AI, trying to get AI to explain 50:21.220 --> 50:22.220 itself. 50:22.220 --> 50:28.340 So what are your thoughts of natural language processing or some kind of other communication? 50:28.340 --> 50:30.220 How does the AI explain something to us? 50:30.220 --> 50:33.740 How do we explain something to it, to machines? 50:33.740 --> 50:35.420 Or you think of it differently? 50:35.420 --> 50:40.100 So there are two separate parts to your question there. 50:40.100 --> 50:43.900 One of them has to do with communication, which is super interesting and I'll get to 50:43.900 --> 50:44.900 that in a sec. 50:44.900 --> 50:50.100 The other is whether we already have AGI, we just haven't noticed it. 50:50.100 --> 50:54.340 There, I beg to differ. 50:54.340 --> 50:58.420 And don't think there's anything in any cellular automaton or anything or the internet itself 50:58.420 --> 51:05.400 or whatever that has artificial general intelligence in that it didn't really do exactly everything 51:05.400 --> 51:06.980 we humans can do better. 51:06.980 --> 51:14.100 I think the day that happens, when that happens, we will very soon notice and we'll probably 51:14.100 --> 51:17.980 notice even before because in a very, very big way. 51:17.980 --> 51:18.980 For the second part though. 51:18.980 --> 51:20.700 Can I just, sorry. 51:20.700 --> 51:30.260 Because you have this beautiful way to formulate in consciousness as information processing 51:30.260 --> 51:33.740 and you can think of intelligence and information processing and you can think of the entire 51:33.740 --> 51:34.740 universe. 51:34.740 --> 51:40.220 These particles and these systems roaming around that have this information processing 51:40.220 --> 51:47.500 power, you don't think there is something with the power to process information in the 51:47.500 --> 51:55.460 way that we human beings do that's out there that needs to be sort of connected to. 51:55.460 --> 51:59.980 It seems a little bit philosophical perhaps, but there's something compelling to the idea 51:59.980 --> 52:06.100 that the power is already there, the focus should be more on being able to communicate 52:06.100 --> 52:07.100 with it. 52:07.100 --> 52:15.340 Well, I agree that in a certain sense, the hardware processing power is already out there 52:15.340 --> 52:21.180 because our universe itself can think of it as being a computer already. 52:21.180 --> 52:25.540 It's constantly computing what water waves, how it devolved the water waves and the river 52:25.540 --> 52:29.860 Charles and how to move the air molecules around that Seth Lloyd has pointed out. 52:29.860 --> 52:33.940 My colleague here that you can even in a very rigorous way think of our entire universe 52:33.940 --> 52:35.660 is just being a quantum computer. 52:35.660 --> 52:40.900 It's pretty clear that our universe supports this amazing processing power because you 52:40.900 --> 52:46.580 can even within this physics computer that we live in, we can even build actual laptops 52:46.580 --> 52:47.580 and stuff. 52:47.580 --> 52:49.140 So clearly the power is there. 52:49.140 --> 52:53.420 It's just that most of the compute power that nature has, it's in my opinion kind of wasting 52:53.420 --> 52:57.140 on boring stuff like simulating yet another ocean wave somewhere where no one is even 52:57.140 --> 52:58.140 looking. 52:58.140 --> 53:03.820 So in a sense, what life does, what we are doing when we build computers is we're rechanneling 53:03.820 --> 53:09.380 all this compute that nature is doing anyway into doing things that are more interesting 53:09.380 --> 53:14.220 than just yet another ocean wave and do something cool here. 53:14.220 --> 53:21.100 So the raw hardware power is there for sure, and even just computing what's going to happen 53:21.100 --> 53:25.540 for the next five seconds in this water ball, you know, it takes a ridiculous amount of 53:25.540 --> 53:28.060 compute if you do it on a human computer. 53:28.060 --> 53:30.040 This water ball just did it. 53:30.040 --> 53:36.020 But that does not mean that this water ball has AGI and this because AGI means it should 53:36.020 --> 53:40.300 also be able to like I've written my book done this interview. 53:40.300 --> 53:42.100 And I don't think it's just communication problems. 53:42.100 --> 53:47.020 I don't think it can do it. 53:47.020 --> 53:51.780 So Buddhists say when they watch the water and that there is some beauty, that there's 53:51.780 --> 53:55.380 some depth and beauty in nature that they can communicate with. 53:55.380 --> 54:01.180 Communication is also very important because I mean, look, part of my job is being a teacher 54:01.180 --> 54:09.940 and I know some very intelligent professors even who just have a better hard time communicating. 54:09.940 --> 54:14.620 They come up with all these brilliant ideas, but to communicate with somebody else, you 54:14.620 --> 54:17.140 have to also be able to simulate their own mind. 54:17.140 --> 54:18.140 Yes. 54:18.140 --> 54:22.020 And build well enough and understand that model of their mind that you can say things 54:22.020 --> 54:24.500 that they will understand. 54:24.500 --> 54:26.700 And that's quite difficult. 54:26.700 --> 54:31.620 And that's why today it's so frustrating if you have a computer that makes some cancer 54:31.620 --> 54:36.260 diagnosis and you ask it, well, why are you saying I should have a surgery? 54:36.260 --> 54:43.620 And if you don't want to reply, I was trained on five terabytes of data and this is my diagnosis 54:43.620 --> 54:49.220 boop, boop, beep, beep, doesn't really instill a lot of confidence, right? 54:49.220 --> 54:54.420 So I think we have a lot of work to do on communication there. 54:54.420 --> 54:59.380 So what kind of, I think you're doing a little bit of work in explainable AI. 54:59.380 --> 55:01.340 What do you think are the most promising avenues? 55:01.340 --> 55:07.100 Is it mostly about sort of the Alexa problem of natural language processing of being able 55:07.100 --> 55:13.220 to actually use human interpretable methods of communication? 55:13.220 --> 55:17.500 So being able to talk to a system and talk back to you, or is there some more fundamental 55:17.500 --> 55:18.500 problems to be solved? 55:18.500 --> 55:21.180 I think it's all of the above. 55:21.180 --> 55:27.180 The natural language processing is obviously important, but there are also more nerdy fundamental 55:27.180 --> 55:28.180 problems. 55:28.180 --> 55:39.180 Like if you take, you play chess, Russian, I have to, when did you learn Russian? 55:39.180 --> 55:45.700 I speak Russian very poorly, but I bought a book, teach yourself Russian, I read a lot, 55:45.700 --> 55:47.700 but it was very difficult. 55:47.700 --> 55:48.700 Wow. 55:48.700 --> 55:49.700 That's why I speak so poorly. 55:49.700 --> 55:51.700 How many languages do you know? 55:51.700 --> 55:52.700 Wow. 55:52.700 --> 55:53.700 That's really impressive. 55:53.700 --> 55:54.700 I don't know. 55:54.700 --> 55:58.740 My wife has some calculations, but my point was, if you played chess, have you looked 55:58.740 --> 56:00.260 at the AlphaZero games? 56:00.260 --> 56:01.260 Yeah. 56:01.260 --> 56:02.260 Oh, the actual games now. 56:02.260 --> 56:03.260 Check it out. 56:03.260 --> 56:09.900 Some of them are just mind blowing, really beautiful. 56:09.900 --> 56:12.460 If you ask, how did it do that? 56:12.460 --> 56:14.500 You got that. 56:14.500 --> 56:20.540 Talk to Demis Osabis, others from DeepMind, all they'll ultimately be able to give you 56:20.540 --> 56:26.940 is big tables of numbers, matrices that define the neural network, and you can stare at these 56:26.940 --> 56:32.980 tables numbers till your face turned blue, and you're not going to understand much about 56:32.980 --> 56:35.860 why it made that move. 56:35.860 --> 56:40.540 Even if you have a natural language processing that can tell you in human language about, 56:40.540 --> 56:44.180 oh, five, seven, point two, eight, still not going to really help. 56:44.180 --> 56:50.660 I think there's a whole spectrum of fun challenges there involved in taking computation that 56:50.660 --> 56:59.940 does intelligent things and transforming it into something equally good, equally intelligent, 56:59.940 --> 57:02.060 but that's more understandable. 57:02.060 --> 57:08.180 I think that's really valuable because I think as we put machines in charge of ever more 57:08.180 --> 57:13.540 infrastructure in our world, the power grid, the trading on the stock market, weapon systems, 57:13.540 --> 57:19.620 and so on, it's absolutely crucial that we can trust these AIs that do all we want and 57:19.620 --> 57:25.860 trust really comes from understanding in a very fundamental way. 57:25.860 --> 57:29.940 That's why I'm working on this, because I think the more if we're going to have some 57:29.940 --> 57:34.700 hope of ensuring that machines have adopted our goals and that they're going to retain 57:34.700 --> 57:41.260 them, that kind of trust, I think, needs to be based on things you can actually understand, 57:41.260 --> 57:47.140 preferably even improve theorems on, even with a self driving car, right? 57:47.140 --> 57:51.020 If someone just tells you it's been trained on tons of data and never crashed, it's less 57:51.020 --> 57:54.460 reassuring than if someone actually has a proof. 57:54.460 --> 57:58.820 Maybe it's a computer verified proof, but still it says that under no circumstances 57:58.820 --> 58:02.420 is this car just going to swerve into oncoming traffic. 58:02.420 --> 58:09.460 And that kind of information helps build trust and helps build the alignment of goals, at 58:09.460 --> 58:12.300 least awareness that your goals, your values are aligned. 58:12.300 --> 58:17.620 And I think even in the very short term, if you look at how today, this absolutely pathetic 58:17.620 --> 58:25.980 state of cybersecurity that we have, where is it, 3 billion Yahoo accounts are packed 58:25.980 --> 58:34.300 and almost every American's credit card and so on, you know, why is this happening? 58:34.300 --> 58:39.940 It's ultimately happening because we have software that nobody fully understood how 58:39.940 --> 58:41.460 it worked. 58:41.460 --> 58:45.100 That's why the bugs hadn't been found, right? 58:45.100 --> 58:50.340 And I think AI can be used very effectively for offense for hacking, but it can also be 58:50.340 --> 59:00.580 used for defense, hopefully, automating verifiability and creating systems that are built in different 59:00.580 --> 59:03.140 ways so you can actually prove things about them. 59:03.140 --> 59:05.460 And it's important. 59:05.460 --> 59:09.740 So speaking of software that nobody understands how it works, of course, a bunch of people 59:09.740 --> 59:14.820 ask about your paper about your thoughts of why does deep and cheap learning work so well? 59:14.820 --> 59:19.280 That's the paper, but what are your thoughts on deep learning, these kind of simplified 59:19.280 --> 59:26.620 models of our own brains that have been able to do some successful perception work, pattern 59:26.620 --> 59:30.940 recognition work, and now with AlphaZero and so on, do some clever things? 59:30.940 --> 59:35.740 What are your thoughts about the promise limitations of this piece? 59:35.740 --> 59:37.140 Great. 59:37.140 --> 59:44.300 I think there are a number of very important insights, very important lessons we can always 59:44.300 --> 59:47.340 draw from these kind of successes. 59:47.340 --> 59:50.460 One of them is when you look at the human brain, you see it's very complicated, a tenth 59:50.460 --> 59:54.140 of 11 neurons, and there are all these different kinds of neurons, and yada yada, and there's 59:54.140 --> 59:57.980 been this long debate about whether the fact that we have dozens of different kinds is 59:57.980 --> 1:00:01.580 actually necessary for intelligence. 1:00:01.580 --> 1:00:06.500 We can now, I think, quite convincingly answer that question of no, it's enough to have just 1:00:06.500 --> 1:00:07.500 one kind. 1:00:07.500 --> 1:00:11.780 If you look under the hood of AlphaZero, there's only one kind of neuron, and it's ridiculously 1:00:11.780 --> 1:00:15.060 simple, a simple mathematical thing. 1:00:15.060 --> 1:00:21.380 So it's just like in physics, if you have a gas with waves in it, it's not the detailed 1:00:21.380 --> 1:00:24.380 nature of the molecules that matter. 1:00:24.380 --> 1:00:27.060 It's the collective behavior, somehow. 1:00:27.060 --> 1:00:33.060 Similarly, it's this higher level structure of the network that matters, not that you 1:00:33.060 --> 1:00:34.060 have 20 kinds of neurons. 1:00:34.060 --> 1:00:41.740 I think our brain is such a complicated mess because it wasn't evolved just to be intelligent, 1:00:41.740 --> 1:00:51.740 it was evolved to also be self assembling, and self repairing, and evolutionarily attainable. 1:00:51.740 --> 1:00:53.660 And patches and so on. 1:00:53.660 --> 1:00:58.700 So I think it's pretty, my hunch is that we're going to understand how to build AGI before 1:00:58.700 --> 1:01:01.060 we fully understand how our brains work. 1:01:01.060 --> 1:01:06.260 Just like we understood how to build flying machines long before we were able to build 1:01:06.260 --> 1:01:07.260 a mechanical bird. 1:01:07.260 --> 1:01:08.260 Yeah, that's right. 1:01:08.260 --> 1:01:15.300 You've given the example of mechanical birds and airplanes, and airplanes do a pretty good 1:01:15.300 --> 1:01:18.620 job of flying without really mimicking bird flight. 1:01:18.620 --> 1:01:23.180 And even now, after 100 years later, did you see the TED talk with this German group of 1:01:23.180 --> 1:01:24.180 mechanical birds? 1:01:24.180 --> 1:01:25.180 I did not. 1:01:25.180 --> 1:01:26.180 I've heard you mention it. 1:01:26.180 --> 1:01:27.180 Check it out. 1:01:27.180 --> 1:01:28.180 It's amazing. 1:01:28.180 --> 1:01:30.180 But even after that, we still don't fly in mechanical birds because it turned out the 1:01:30.180 --> 1:01:34.580 way we came up with simpler, and it's better for our purposes, and I think it might be the 1:01:34.580 --> 1:01:35.580 same there. 1:01:35.580 --> 1:01:38.140 So that's one lesson. 1:01:38.140 --> 1:01:42.020 Another lesson is one of what our paper was about. 1:01:42.020 --> 1:01:47.420 Well, first, as a physicist thought, it was fascinating how there's a very close mathematical 1:01:47.420 --> 1:01:50.900 relationship, actually, between our artificial neural networks. 1:01:50.900 --> 1:01:56.580 And a lot of things that we've studied for in physics go by nerdy names like the renormalization 1:01:56.580 --> 1:02:01.100 group equation and Hamiltonians and yada, yada, yada. 1:02:01.100 --> 1:02:11.380 And when you look a little more closely at this, you have, at first, I was like, well, 1:02:11.380 --> 1:02:18.700 there's something crazy here that doesn't make sense because we know that if you even 1:02:18.700 --> 1:02:23.380 want to build a super simple neural network to tell apart cat pictures and dog pictures, 1:02:23.380 --> 1:02:27.260 right, that you can do that very, very well now. 1:02:27.260 --> 1:02:31.540 But if you think about it a little bit, you convince yourself it must be impossible because 1:02:31.540 --> 1:02:36.420 if I have one megapixel, even if each pixel is just black or white, there's two to the 1:02:36.420 --> 1:02:40.900 power of one million possible images, which is way more than there are atoms in our universe. 1:02:40.900 --> 1:02:47.220 So in order to, and then for each one of those, I have to assign a number, which is the probability 1:02:47.220 --> 1:02:49.100 that it's a dog. 1:02:49.100 --> 1:02:55.900 So an arbitrary function of images is a list of more numbers than there are atoms in our 1:02:55.900 --> 1:02:56.900 universe. 1:02:56.900 --> 1:03:02.220 So clearly, I can't store that under the hood of my, my GPU or my, my computer yet somehow 1:03:02.220 --> 1:03:03.220 works. 1:03:03.220 --> 1:03:04.220 So what does that mean? 1:03:04.220 --> 1:03:12.940 Well, it means that out of all of the problems that you could try to solve with a neural network, 1:03:12.940 --> 1:03:17.940 almost all of them are impossible to solve with a reasonably sized one. 1:03:17.940 --> 1:03:24.820 But then what we showed in our paper was, was that the, the fraction, the kind of problems, 1:03:24.820 --> 1:03:29.740 the fraction of all the problems that you could possibly pose that the, that we actually 1:03:29.740 --> 1:03:34.980 care about given the laws of physics is also an infinitesimally tiny little part. 1:03:34.980 --> 1:03:37.180 And amazingly, they're basically the same part. 1:03:37.180 --> 1:03:38.180 Yeah. 1:03:38.180 --> 1:03:41.180 It's almost like our world was created for, I mean, they kind of come together. 1:03:41.180 --> 1:03:42.180 Yeah. 1:03:42.180 --> 1:03:44.940 You, but you could say maybe where the world created the world that the world was created 1:03:44.940 --> 1:03:50.300 for us, but I have a more modest interpretation, which is that instead evolution endowments 1:03:50.300 --> 1:03:54.700 with neural networks, precisely for that reason, because this particular architecture has 1:03:54.700 --> 1:04:02.380 opposed to the one in your laptop is very, very well adapted to solving the kind of problems 1:04:02.380 --> 1:04:05.540 that nature kept presenting our ancestors with, right? 1:04:05.540 --> 1:04:09.380 So it makes sense that why do we have a brain in the first place? 1:04:09.380 --> 1:04:12.940 It's to be able to make predictions about the future and so on. 1:04:12.940 --> 1:04:17.580 So if we had a sucky system, which could never solve it, it wouldn't have a lot. 1:04:17.580 --> 1:04:23.420 So, but it's, so this is, this is a, I think a very beautiful fact. 1:04:23.420 --> 1:04:24.420 Yeah. 1:04:24.420 --> 1:04:28.780 And you also realize that there's, there, that we, there've been, it's been earlier 1:04:28.780 --> 1:04:34.140 work on, on why deeper networks are good, but we were able to show an additional cool 1:04:34.140 --> 1:04:40.260 fact there, which is that even incredibly simple problems, like suppose I give you a 1:04:40.260 --> 1:04:45.020 thousand numbers and ask you to multiply them together and you can write a few lines of 1:04:45.020 --> 1:04:46.820 code, boom, done, trivial. 1:04:46.820 --> 1:04:52.580 If you just try to do that with a neural network that has only one single hidden layer in it, 1:04:52.580 --> 1:04:59.940 you can do it, but you're going to need two to the power of thousand neurons to multiply 1:04:59.940 --> 1:05:03.260 a thousand numbers, which is again, more neurons than their atoms in our universe. 1:05:03.260 --> 1:05:05.740 So that's fascinating. 1:05:05.740 --> 1:05:11.580 But if you allow, if you allow yourself, make it a deep network of many layers, you only 1:05:11.580 --> 1:05:15.340 need four thousand neurons, it's perfectly feasible. 1:05:15.340 --> 1:05:17.500 So that's really interesting. 1:05:17.500 --> 1:05:18.500 Yeah. 1:05:18.500 --> 1:05:19.500 Yeah. 1:05:19.500 --> 1:05:24.460 So architecture type, I mean, you mentioned Schrodinger's equation and what are your thoughts 1:05:24.460 --> 1:05:32.860 about quantum computing and the role of this kind of computational unit in creating an 1:05:32.860 --> 1:05:34.900 intelligent system? 1:05:34.900 --> 1:05:41.100 In some Hollywood movies that I don't mention my name because I don't want to spoil them. 1:05:41.100 --> 1:05:46.820 The way they get AGI is building a quantum computer because the word quantum sounds 1:05:46.820 --> 1:05:47.820 cool and so on. 1:05:47.820 --> 1:05:48.820 That's right. 1:05:48.820 --> 1:05:54.940 But first of all, I think we don't need quantum computers to build AGI. 1:05:54.940 --> 1:06:01.740 I suspect your brain is not quantum computer in any found sense. 1:06:01.740 --> 1:06:03.460 So you don't even wrote a paper about that. 1:06:03.460 --> 1:06:09.060 Many years ago, I calculated the so called decoherence time that how long it takes until 1:06:09.060 --> 1:06:16.900 the quantum computerness of what your neurons are doing gets erased by just random noise 1:06:16.900 --> 1:06:21.420 from the environment and it's about 10 to the minus 21 seconds. 1:06:21.420 --> 1:06:27.420 So as cool as it would be to have a quantum computer in my head, I don't think that fast. 1:06:27.420 --> 1:06:35.820 On the other hand, there are very cool things you could do with quantum computers or I think 1:06:35.820 --> 1:06:40.780 we'll be able to do soon when we get bigger ones that might actually help machine learning 1:06:40.780 --> 1:06:43.180 do even better than the brain. 1:06:43.180 --> 1:06:58.620 So for example, one, this is just a moonshot, but hey, learning is very much same thing 1:06:58.620 --> 1:07:00.860 as search. 1:07:00.860 --> 1:07:05.460 If you're trying to train a neural network to get really learned to do something really 1:07:05.460 --> 1:07:10.820 well, you have some loss function, you have a bunch of knobs you can turn represented 1:07:10.820 --> 1:07:14.420 by a bunch of numbers and you're trying to tweak them so that it becomes as good as possible 1:07:14.420 --> 1:07:15.420 at this thing. 1:07:15.420 --> 1:07:22.580 So if you think of a landscape with some valley, where each dimension of the landscape corresponds 1:07:22.580 --> 1:07:25.780 to some number you can change, you're trying to find the minimum. 1:07:25.780 --> 1:07:29.980 And it's well known that if you have a very high dimensional landscape, complicated things, 1:07:29.980 --> 1:07:34.140 it's super hard to find the minimum. 1:07:34.140 --> 1:07:37.500 Quantum mechanics is amazingly good at this. 1:07:37.500 --> 1:07:42.980 If I want to know what's the lowest energy state this water can possibly have incredibly 1:07:42.980 --> 1:07:47.860 hard to compute, but nature will happily figure this out for you if you just cool it down, 1:07:47.860 --> 1:07:50.860 make it very, very cold. 1:07:50.860 --> 1:07:55.260 If you put a ball somewhere, it'll roll down to its minimum and this happens metaphorically 1:07:55.260 --> 1:07:57.620 at the energy landscape too. 1:07:57.620 --> 1:08:02.940 And quantum mechanics even uses some clever tricks which today's machine learning systems 1:08:02.940 --> 1:08:03.940 don't. 1:08:03.940 --> 1:08:07.940 If you're trying to find the minimum and you get stuck in the little local minimum here 1:08:07.940 --> 1:08:14.180 in quantum mechanics, you can actually tunnel through the barrier and get unstuck again. 1:08:14.180 --> 1:08:15.420 And that's really interesting. 1:08:15.420 --> 1:08:16.420 Yeah. 1:08:16.420 --> 1:08:22.940 So maybe for example, we'll one day use quantum computers that help train neural networks 1:08:22.940 --> 1:08:23.940 better. 1:08:23.940 --> 1:08:24.940 That's really interesting. 1:08:24.940 --> 1:08:25.940 Okay. 1:08:25.940 --> 1:08:32.020 So as a component of kind of the learning process, for example, let me ask sort of wrapping 1:08:32.020 --> 1:08:34.060 up here a little bit. 1:08:34.060 --> 1:08:40.540 Let me return to the questions of our human nature and love, as I mentioned. 1:08:40.540 --> 1:08:48.020 So do you think you mentioned sort of a helper robot that you could think of also personal 1:08:48.020 --> 1:08:49.020 robots. 1:08:49.020 --> 1:08:55.300 Do you think the way we human beings fall in love and get connected to each other is 1:08:55.300 --> 1:09:00.420 possible to achieve in an AI system and human level AI intelligence system. 1:09:00.420 --> 1:09:06.100 Do you think we would ever see that kind of connection or, you know, in all this discussion 1:09:06.100 --> 1:09:11.460 about solving complex goals, as this kind of human social connection, do you think that's 1:09:11.460 --> 1:09:16.460 one of the goals on the peaks and valleys that were the raising sea levels that we'd 1:09:16.460 --> 1:09:17.460 be able to achieve? 1:09:17.460 --> 1:09:22.180 Or do you think that's something that's ultimately, or at least in the short term, relative to 1:09:22.180 --> 1:09:23.620 the other goals is not achievable? 1:09:23.620 --> 1:09:25.220 I think it's all possible. 1:09:25.220 --> 1:09:31.780 And I mean, in recent, there's a very wide range of guesses, as you know, among AI researchers 1:09:31.780 --> 1:09:35.300 when we're going to get AGI. 1:09:35.300 --> 1:09:39.620 Some people, you know, like our friend Rodney Brooks said, it's going to be hundreds of 1:09:39.620 --> 1:09:41.140 years at least. 1:09:41.140 --> 1:09:44.780 And then there are many others that think it's going to happen relatively much sooner. 1:09:44.780 --> 1:09:52.140 Recent polls, maybe half or so, AI researchers think we're going to get AGI within decades. 1:09:52.140 --> 1:09:56.260 So if that happens, of course, then I think these things are all possible. 1:09:56.260 --> 1:10:01.860 But in terms of whether it will happen, I think we shouldn't spend so much time asking, 1:10:01.860 --> 1:10:04.260 what do we think will happen in the future? 1:10:04.260 --> 1:10:08.980 As if we are just some sort of pathetic, passive bystanders, you know, waiting for the future 1:10:08.980 --> 1:10:12.740 to happen to us, hey, we're the ones creating this future, right? 1:10:12.740 --> 1:10:18.340 So we should be proactive about it and ask ourselves what sort of future we would like 1:10:18.340 --> 1:10:19.340 to have happen. 1:10:19.340 --> 1:10:20.340 That's right. 1:10:20.340 --> 1:10:21.340 Trying to make it like that. 1:10:21.340 --> 1:10:25.660 Well, what I prefer is some sort of incredibly boring zombie like future where there's all 1:10:25.660 --> 1:10:30.220 these mechanical things happening and there's no passion, no emotion, no experience, maybe 1:10:30.220 --> 1:10:31.220 even. 1:10:31.220 --> 1:10:35.740 No, I would, of course, much rather prefer it if all the things that we find that we 1:10:35.740 --> 1:10:44.180 value the most about humanity are a subjective experience, passion, inspiration, love, you 1:10:44.180 --> 1:10:50.780 know, if we can create a future where those things do exist. 1:10:50.780 --> 1:10:56.500 You know, I think ultimately it's not our universe giving meaning to us, it's us giving 1:10:56.500 --> 1:10:58.500 meaning to our universe. 1:10:58.500 --> 1:11:03.620 And if we build more advanced intelligence, let's make sure we build it in such a way 1:11:03.620 --> 1:11:09.100 that meaning is part of it. 1:11:09.100 --> 1:11:13.900 A lot of people that seriously study this problem and think of it from different angles have 1:11:13.900 --> 1:11:20.140 trouble in the majority of cases, if they think through that happen, are the ones that 1:11:20.140 --> 1:11:22.620 are not beneficial to humanity. 1:11:22.620 --> 1:11:27.260 And so, yeah, so what are your thoughts? 1:11:27.260 --> 1:11:33.820 What should people, you know, I really don't like people to be terrified, what's the way 1:11:33.820 --> 1:11:38.660 for people to think about it in a way that, in a way we can solve it and we can make it 1:11:38.660 --> 1:11:39.660 better. 1:11:39.660 --> 1:11:40.660 Yeah. 1:11:40.660 --> 1:11:44.780 No, I don't think panicking is going to help in any way, it's not going to increase chances 1:11:44.780 --> 1:11:46.060 of things going well either. 1:11:46.060 --> 1:11:49.340 Even if you are in a situation where there is a real threat, does it help if everybody 1:11:49.340 --> 1:11:50.620 just freaks out? 1:11:50.620 --> 1:11:51.620 Right. 1:11:51.620 --> 1:11:53.620 No, of course not. 1:11:53.620 --> 1:11:59.740 I think, yeah, there are, of course, ways in which things can go horribly wrong. 1:11:59.740 --> 1:12:04.460 First of all, it's important when we think about this thing, this, about the problems 1:12:04.460 --> 1:12:08.780 and risks, to also remember how huge the upsides can be if we get it right. 1:12:08.780 --> 1:12:13.420 Everything we love about society and civilization is a product of intelligence. 1:12:13.420 --> 1:12:17.980 So if we can amplify our intelligence with machine intelligence and not anymore lose 1:12:17.980 --> 1:12:23.380 our loved ones, what we're told is an uncurable disease and things like this, of course, we 1:12:23.380 --> 1:12:24.940 should aspire to that. 1:12:24.940 --> 1:12:28.700 So that can be a motivator, I think, reminding yourselves that the reason we try to solve 1:12:28.700 --> 1:12:34.140 problems is not just because we're trying to avoid gloom, but because we're trying to 1:12:34.140 --> 1:12:35.900 do something great. 1:12:35.900 --> 1:12:43.340 But then in terms of the risks, I think the really important question is to ask, what 1:12:43.340 --> 1:12:47.740 can we do today that will actually help make the outcome good, right? 1:12:47.740 --> 1:12:52.700 And dismissing the risk is not one of them, you know, I find it quite funny often when 1:12:52.700 --> 1:13:01.540 I'm in discussion panels about these things, how the people who work for companies will 1:13:01.540 --> 1:13:05.100 always be like, oh, nothing to worry about, nothing to worry about, nothing to worry about. 1:13:05.100 --> 1:13:09.980 And it's always, it's only academics sometimes express concerns. 1:13:09.980 --> 1:13:10.980 That's not surprising at all. 1:13:10.980 --> 1:13:17.500 If you think about it, often Sinclair quipped, right, that it's hard to make a man believe 1:13:17.500 --> 1:13:20.620 in something when his income depends on not believing in it. 1:13:20.620 --> 1:13:25.580 And frankly, we know a lot of these people in companies that they're just as concerned 1:13:25.580 --> 1:13:26.580 as anyone else. 1:13:26.580 --> 1:13:30.300 But if you're the CEO of a company, that's not something you want to go on record saying 1:13:30.300 --> 1:13:34.980 when you have silly journalists who are going to put a picture of a Terminator robot when 1:13:34.980 --> 1:13:35.980 they quote you. 1:13:35.980 --> 1:13:39.380 So, so the issues are real. 1:13:39.380 --> 1:13:45.660 And the way I think about what the issue is, is basically, you know, the real choice we 1:13:45.660 --> 1:13:51.980 have is, first of all, are we going to dismiss this, the risks and say, well, you know, let's 1:13:51.980 --> 1:13:57.140 just go ahead and build machines that can do everything we can do better and cheaper, 1:13:57.140 --> 1:14:00.940 you know, let's just make ourselves obsolete as fast as possible or what could possibly 1:14:00.940 --> 1:14:01.940 go wrong. 1:14:01.940 --> 1:14:02.940 Right. 1:14:02.940 --> 1:14:03.940 That's one attitude. 1:14:03.940 --> 1:14:09.380 The opposite attitude that I think is to say, it's incredible potential, you know, let's 1:14:09.380 --> 1:14:14.900 think about what kind of future we're really, really excited about. 1:14:14.900 --> 1:14:18.700 What are the shared goals that we can really aspire towards? 1:14:18.700 --> 1:14:22.100 And then let's think really hard about how we can actually get there. 1:14:22.100 --> 1:14:23.100 So start with it. 1:14:23.100 --> 1:14:24.460 Don't start thinking about the risks. 1:14:24.460 --> 1:14:26.940 Start thinking about the goals. 1:14:26.940 --> 1:14:30.540 And then when you do that, then you can think about the obstacles you want to avoid, right? 1:14:30.540 --> 1:14:34.420 I often get students coming in right here into my office for career advice. 1:14:34.420 --> 1:14:38.060 Always ask them this very question, where do you want to be in the future? 1:14:38.060 --> 1:14:42.580 If all she can say is, oh, maybe I'll have cancer, maybe I'll run over by a truck. 1:14:42.580 --> 1:14:44.420 Focus on the obstacles instead of the goal. 1:14:44.420 --> 1:14:49.340 She's just going to end up a hypochondriac paranoid, whereas if she comes in and fire 1:14:49.340 --> 1:14:54.060 in her eyes and is like, I want to be there, and then we can talk about the obstacles and 1:14:54.060 --> 1:14:56.100 see how we can circumvent them. 1:14:56.100 --> 1:14:59.100 That's I think a much, much healthier attitude. 1:14:59.100 --> 1:15:01.540 And that's really what we're in. 1:15:01.540 --> 1:15:09.420 And I feel it's very challenging to come up with a vision for the future, which we're 1:15:09.420 --> 1:15:10.660 unequivocally excited about. 1:15:10.660 --> 1:15:14.300 I'm not just talking now in the vague terms like, yeah, let's cure cancer. 1:15:14.300 --> 1:15:18.500 I'm talking about what kind of society do we want to create? 1:15:18.500 --> 1:15:25.380 What do we want it to mean to be human in the age of AI, in the age of AGI? 1:15:25.380 --> 1:15:31.460 So if we can have this conversation, broad, inclusive conversation, and gradually start 1:15:31.460 --> 1:15:38.100 converging towards some future with some direction at least that we want to steer towards, right? 1:15:38.100 --> 1:15:42.340 Then we'll be much more motivated to constructively take on the obstacles. 1:15:42.340 --> 1:15:54.260 And I think if I wrap this up in a more succinct way, I think we can all agree already now that 1:15:54.260 --> 1:16:05.540 we should aspire to build AGI that doesn't overpower us, but that empowers us. 1:16:05.540 --> 1:16:10.820 And think of the many various ways that can do that, whether that's from my side of the 1:16:10.820 --> 1:16:12.860 world of autonomous vehicles. 1:16:12.860 --> 1:16:17.020 I'm personally actually from the camp that believes this human level intelligence is 1:16:17.020 --> 1:16:22.780 required to achieve something like vehicles that would actually be something we would 1:16:22.780 --> 1:16:25.380 enjoy using and being part of. 1:16:25.380 --> 1:16:26.380 So that's the one example. 1:16:26.380 --> 1:16:31.140 And certainly there's a lot of other types of robots and medicine and so on. 1:16:31.140 --> 1:16:35.300 So focusing on those and then coming up with the obstacles, coming up with the ways that 1:16:35.300 --> 1:16:38.420 that can go wrong and solving those one at a time. 1:16:38.420 --> 1:16:42.980 And just because you can build an autonomous vehicle, even if you could build one that 1:16:42.980 --> 1:16:47.500 would drive this final AGI, maybe there are some things in life that we would actually 1:16:47.500 --> 1:16:48.500 want to do ourselves. 1:16:48.500 --> 1:16:49.500 That's right. 1:16:49.500 --> 1:16:50.500 Right? 1:16:50.500 --> 1:16:54.660 Like, for example, if you think of our society as a whole, there are some things that we 1:16:54.660 --> 1:16:57.540 find very meaningful to do. 1:16:57.540 --> 1:17:02.100 And that doesn't mean we have to stop doing them just because machines can do them better. 1:17:02.100 --> 1:17:06.660 I'm not going to stop playing tennis just the day someone builds a tennis robot and 1:17:06.660 --> 1:17:07.660 beat me. 1:17:07.660 --> 1:17:09.900 People are still playing chess and even go. 1:17:09.900 --> 1:17:10.900 Yeah. 1:17:10.900 --> 1:17:19.100 And in this very near term, even some people are advocating basic income, replace jobs. 1:17:19.100 --> 1:17:22.780 But if the government is going to be willing to just hand out cash to people for doing 1:17:22.780 --> 1:17:27.660 nothing, then one should also seriously consider whether the government should also just hire 1:17:27.660 --> 1:17:33.380 a lot more teachers and nurses and the kind of jobs which people often find great fulfillment 1:17:33.380 --> 1:17:34.380 in doing, right? 1:17:34.380 --> 1:17:38.900 We get very tired of hearing politicians saying, oh, we can't afford hiring more teachers, 1:17:38.900 --> 1:17:41.700 but we're going to maybe have basic income. 1:17:41.700 --> 1:17:46.340 If we can have more serious research and thought into what gives meaning to our lives, the 1:17:46.340 --> 1:17:50.700 jobs give so much more than income, right? 1:17:50.700 --> 1:18:00.020 And then think about, in the future, what are the roles that we want to have people 1:18:00.020 --> 1:18:03.180 continually feeling empowered by machines? 1:18:03.180 --> 1:18:08.900 And I think sort of, I come from the Russia, from the Soviet Union, and I think for a lot 1:18:08.900 --> 1:18:14.100 of people in the 20th century, going to the moon, going to space was an inspiring thing. 1:18:14.100 --> 1:18:21.300 I feel like the universe of the mind, so AI, understanding, creating intelligence is that 1:18:21.300 --> 1:18:23.380 for the 21st century. 1:18:23.380 --> 1:18:26.740 So it's really surprising, and I've heard you mention this, it's really surprising to 1:18:26.740 --> 1:18:31.940 me both on the research funding side that it's not funded as greatly as it could be. 1:18:31.940 --> 1:18:36.500 But most importantly, on the politician side, that it's not part of the public discourse 1:18:36.500 --> 1:18:44.300 except in killer bots, terminator kind of view, that people are not yet, I think, perhaps 1:18:44.300 --> 1:18:48.260 excited by the possible positive future that we can build together. 1:18:48.260 --> 1:18:54.660 So we should be, because politicians usually just focus on the next election cycle, right? 1:18:54.660 --> 1:18:59.340 The single most important thing I feel we humans have learned in the entire history of science 1:18:59.340 --> 1:19:07.460 is they were the masters of underestimation, underestimated the size of our cosmos, again 1:19:07.460 --> 1:19:11.380 and again, realizing that everything we thought existed was just a small part of something 1:19:11.380 --> 1:19:12.380 grander, right? 1:19:12.380 --> 1:19:18.580 Planet, solar system, the galaxy, clusters of galaxies, the universe. 1:19:18.580 --> 1:19:25.700 And we now know that we have the future has just so much more potential than our ancestors 1:19:25.700 --> 1:19:27.820 could ever have dreamt of. 1:19:27.820 --> 1:19:39.820 This cosmos, imagine if all of Earth was completely devoid of life except for Cambridge, Massachusetts. 1:19:39.820 --> 1:19:44.220 Wouldn't it be kind of lame if all we ever aspired to was to stay in Cambridge, Massachusetts 1:19:44.220 --> 1:19:49.660 forever and then go extinct in one week, even though Earth was going to continue on for 1:19:49.660 --> 1:19:50.660 longer? 1:19:50.660 --> 1:19:57.300 That sort of attitude I think we have now on the cosmic scale, we can, life can flourish 1:19:57.300 --> 1:20:00.820 on Earth, not for four years, but for billions of years. 1:20:00.820 --> 1:20:06.340 I can even tell you about how to move it out of harm's way when the sun gets too hot. 1:20:06.340 --> 1:20:11.900 And then we have so much more resources out here, which today, maybe there are a lot of 1:20:11.900 --> 1:20:19.380 other planets with bacteria or cow like life on them, but most of this, all this opportunity 1:20:19.380 --> 1:20:25.380 seems as far as we can tell to be largely dead, like the Sahara Desert, and yet we have the 1:20:25.380 --> 1:20:30.380 opportunity to help life flourish around this for billions of years. 1:20:30.380 --> 1:20:37.420 So like, let's quit squabbling about whether some little border should be drawn one mile 1:20:37.420 --> 1:20:43.380 to the left or right and look up into the skies and realize, hey, we can do such incredible 1:20:43.380 --> 1:20:44.380 things. 1:20:44.380 --> 1:20:45.380 Yeah. 1:20:45.380 --> 1:20:49.980 And that's I think why it's really exciting that you and others are connected with some 1:20:49.980 --> 1:20:54.740 of the work Elon Musk is doing because he's literally going out into that space, really 1:20:54.740 --> 1:20:56.260 exploring our universe. 1:20:56.260 --> 1:20:57.260 And it's wonderful. 1:20:57.260 --> 1:21:02.340 That is exactly why Elon Musk is so misunderstood, right? 1:21:02.340 --> 1:21:05.300 Misconstrued with some kind of pessimistic doomsayer. 1:21:05.300 --> 1:21:10.860 The reason he cares so much about AI safety is because he more than almost anyone else 1:21:10.860 --> 1:21:13.340 appreciates these amazing opportunities. 1:21:13.340 --> 1:21:16.340 It will squander if we wipe out here on Earth. 1:21:16.340 --> 1:21:22.740 We're not just going to wipe out the next generation, but all generations and this incredible 1:21:22.740 --> 1:21:25.580 opportunity that's out there and that would be really be a waste. 1:21:25.580 --> 1:21:32.740 And AI, for people who think that there would be better to do without technology, let me 1:21:32.740 --> 1:21:37.740 just mention that if we don't improve our technology, the question isn't whether humanity 1:21:37.740 --> 1:21:38.740 is going to go extinct. 1:21:38.740 --> 1:21:43.620 The question is just whether we're going to get taken out by the next big asteroid or 1:21:43.620 --> 1:21:49.540 the next super volcano or something else dumb that we could easily prevent with more tech, 1:21:49.540 --> 1:21:50.540 right? 1:21:50.540 --> 1:21:56.220 If we want life to flourish throughout the cosmos, AI is the key to it. 1:21:56.220 --> 1:22:04.780 As I mentioned in a lot of detail in my book, even many of the most inspired sci fi writers 1:22:04.780 --> 1:22:11.580 I feel have totally underestimated the opportunities for space travel, especially to other galaxies, 1:22:11.580 --> 1:22:17.100 because they weren't thinking about the possibility of AGI, which just makes it so much easier. 1:22:17.100 --> 1:22:18.100 Right. 1:22:18.100 --> 1:22:25.900 Yeah, so that goes to a view of AGI that enables our progress, that enables a better life. 1:22:25.900 --> 1:22:30.060 So that's a beautiful way to put it and something to strive for. 1:22:30.060 --> 1:22:31.580 So Max, thank you so much. 1:22:31.580 --> 1:22:32.580 Thank you for your time today. 1:22:32.580 --> 1:22:33.580 It's been awesome. 1:22:33.580 --> 1:22:34.580 Thank you so much. 1:22:34.580 --> 1:22:35.580 Thanks. 1:22:35.580 --> 1:22:36.580 Merci beaucoup. 1:22:36.580 --> 1:22:49.100 Thank you so much for your time today and thank you so much for your time and for your 1:22:49.100 --> 1:22:50.100 time. 1:22:50.100 --> 1:22:51.100 Thank you. 1:22:51.100 --> 1:22:52.100 Thank you. 1:22:52.100 --> 1:22:53.100 Bye. 1:22:53.100 --> 1:22:54.100 Bye. 1:22:54.100 --> 1:22:55.100 Bye. 1:22:55.100 --> 1:22:56.100 Bye. 1:22:56.100 --> 1:22:57.100 Bye. 1:22:57.100 --> 1:22:58.100 Bye. 1:22:58.100 --> 1:22:59.100 Bye. 1:22:59.100 --> 1:23:00.100 Bye.