WEBVTT - The Story: Denial Isn’t the Answer w/ Geoffrey Hinton 0:00:00.320 --> 0:00:03.000 Thanks for tune into Tech Stuff. If you don't recognize 0:00:03.000 --> 0:00:06.120 my voice, my name is Osvalshin and I'm here because 0:00:06.160 --> 0:00:09.240 the inimitable Jonathan Strickland has passed the baton to Cara 0:00:09.360 --> 0:00:12.680 Price and myself to host Tech Stuff. The show will 0:00:12.720 --> 0:00:15.480 remain your home for all things tech, and all the 0:00:15.480 --> 0:00:17.960 old episodes will remain available in this feed. 0:00:18.440 --> 0:00:19.200 Thanks for listening. 0:00:20.360 --> 0:00:23.120 I was in a hotel in California and I saw 0:00:23.160 --> 0:00:26.120 that the phone lit up, and I thought, who's calling 0:00:26.120 --> 0:00:27.840 me at one o'clock in the morning, And then this 0:00:27.920 --> 0:00:30.520 Swedish voice came on. Then they said I won the 0:00:30.560 --> 0:00:33.840 Nobel Prize in Physics, and I thought, this is very odd. 0:00:33.880 --> 0:00:36.519 I don't do physics, so that's when I thought it 0:00:36.600 --> 0:00:37.280 might be a prank. 0:00:38.360 --> 0:00:42.280 Jeffrey Hinton won the twenty twenty four Nobel Prize in Physics, 0:00:42.880 --> 0:00:46.519 an honor held by Albert Einstein and Marie Curie. A 0:00:46.560 --> 0:00:50.000 certain j Robert Oppenheimer were shortlisted but never won. 0:00:50.400 --> 0:00:52.519 My big hope was to win the Nobel Prize in 0:00:52.520 --> 0:00:55.080 Physiology or Medicine for figuring out how the brain worked. 0:00:55.960 --> 0:00:58.120 And what I didn't realize is you could fail to 0:00:58.120 --> 0:00:59.640 figure out how the brain worked and still get a 0:00:59.680 --> 0:01:00.640 Noble Prize anyway. 0:01:02.560 --> 0:01:06.000 Welcome to tech stuff. This is the story with our guests, 0:01:06.000 --> 0:01:09.880 the Nobel Laureate Jeffrey Hinton. Each week on Wednesdays, we 0:01:10.000 --> 0:01:12.400 bring you an in depth interview with someone who's at 0:01:12.400 --> 0:01:15.800 the forefront of technology or who can unlock a world 0:01:15.840 --> 0:01:19.520 where tech is at its most fascinating. His recent Nobel 0:01:19.560 --> 0:01:23.959 Prize win was for quote foundational discoveries and inventions that 0:01:24.120 --> 0:01:29.200 enable machine learning with artificial neural networks. Now, artificial neural 0:01:29.240 --> 0:01:33.319 networks are learning models inspired by the network of neurons 0:01:33.360 --> 0:01:36.679 present in the human brain, and Hinton's desire to figure 0:01:36.720 --> 0:01:39.759 out the brain was a key inspiration for his pioneering 0:01:39.800 --> 0:01:44.399 work on AI. I was particularly fascinated by Hinton because 0:01:44.480 --> 0:01:47.960 his work went completely against the mainstream of computer science 0:01:48.040 --> 0:01:51.880 for decades, and yet he stuck to his guns. It's 0:01:51.920 --> 0:01:55.800 an incredible story of dedication in the face of personal loss. 0:01:56.720 --> 0:02:01.680 Also fascinating is Hinton's relationship to his own creation. Here's 0:02:01.720 --> 0:02:04.120 what he said at the Nobel Prize banquet. 0:02:04.800 --> 0:02:08.399 There is also a longer term existential threat that will 0:02:08.400 --> 0:02:12.000 arise when we create digital beings that are more intelligent 0:02:12.080 --> 0:02:16.040 than ourselves. We have no idea whether we can stay 0:02:16.040 --> 0:02:20.120 in control. But we now have evidence that if they 0:02:20.160 --> 0:02:24.280 are created by companies motivated by short term profits, our 0:02:24.400 --> 0:02:28.799 safety will not be the top priority. We urgently need 0:02:28.880 --> 0:02:32.239 research on how to prevent these new beings from wanting 0:02:32.320 --> 0:02:37.240 to take control. They are no longer science fiction, thank you. 0:02:40.200 --> 0:02:42.560 So when I got the opportunity to sit down with 0:02:42.639 --> 0:02:46.200 Jeffrey Hinton, I wanted to know how he went from 0:02:46.240 --> 0:02:49.680 someone who wanted to understand the relationship between the mind 0:02:49.760 --> 0:02:53.240 and the brain to someone who paved the path for 0:02:53.320 --> 0:02:56.519 AI as we know it. And he obliged me by 0:02:56.560 --> 0:03:01.240 telling me about his trajectory from student to researcher, to professor, 0:03:01.520 --> 0:03:05.280 to Google employee to finally AI safety advocate. 0:03:06.280 --> 0:03:09.239 Am I riding thinking with all due respect to Steve 0:03:09.360 --> 0:03:11.520 Jobs and Bill Gates and marks up a bug that 0:03:12.160 --> 0:03:15.320 you were, in a sense the original college dropout, not. 0:03:15.360 --> 0:03:17.280 In the sense they dropped out, Because what I did 0:03:17.400 --> 0:03:19.600 was I I went to Cambridge and after a month 0:03:19.639 --> 0:03:21.800 I dropped out, but then I went back the next year. 0:03:22.160 --> 0:03:25.480 And then while I was doing my PhD, I dropped out, 0:03:26.680 --> 0:03:30.480 but then I finished it. So I'm not like them. 0:03:30.760 --> 0:03:33.679 I went back. I'm a failed to drop out. 0:03:34.120 --> 0:03:37.400 Failed to drop out, that's good. What were the reasons for? 0:03:38.640 --> 0:03:42.800 Was it uncertainty or ambivalence or curiosity or so. 0:03:42.880 --> 0:03:45.800 When I first went to Cambridge, it was the first 0:03:45.840 --> 0:03:49.320 time I'd ever lived away from home, and I'd always 0:03:49.560 --> 0:03:51.480 my image of myself was that it was one of 0:03:51.480 --> 0:03:54.200 the clever ones. And when I went to Cambridge, I 0:03:54.240 --> 0:03:57.840 wasn't one of the clever ones. Everybody was clever, and 0:03:57.920 --> 0:04:00.560 I found it very stressful. I worked very, very hard, 0:04:00.720 --> 0:04:03.440 so I was working like twelve hours a day doing science, 0:04:04.280 --> 0:04:07.240 and the combination of working very hard to keep up 0:04:07.800 --> 0:04:10.760 and living away from home for the first time was 0:04:10.920 --> 0:04:11.800 just too much for me. 0:04:12.520 --> 0:04:14.400 So there was a fantastic profile of you in the 0:04:14.440 --> 0:04:18.200 New Yorker which said, quote, Hinton tried different feels but 0:04:18.320 --> 0:04:20.560 was dismayed to find that he was never the brightest 0:04:20.560 --> 0:04:22.960 student in any given class, which made me smile. But 0:04:23.040 --> 0:04:26.719 I guess dismay and stress are quite a close cousins, 0:04:27.040 --> 0:04:30.560 And I suppose the stakes for you were high given 0:04:30.600 --> 0:04:32.760 the family you came from. Can you speak a little 0:04:32.760 --> 0:04:33.159 bit about that. 0:04:33.839 --> 0:04:36.080 Yes, I had a lot of pressure from my father 0:04:36.160 --> 0:04:40.479 to be an academic success, and my mother sort of 0:04:41.040 --> 0:04:44.919 went along with it. So from an early age I 0:04:45.000 --> 0:04:48.000 realized that's what I had to achieve, and that's a 0:04:48.000 --> 0:04:48.600 lot of pressure. 0:04:49.040 --> 0:04:51.760 How did they exert that pressure? How are you aware 0:04:51.760 --> 0:04:52.040 of it? 0:04:53.360 --> 0:04:57.080 My father was a slightly strange character. He grew up 0:04:57.120 --> 0:05:02.040 in Mexico during all the revolutions without a mother, somewhat odd. 0:05:04.279 --> 0:05:06.480 Every morning when I went to school, not every morning, 0:05:06.480 --> 0:05:09.320 but quite often, as I left, you would say, get 0:05:09.320 --> 0:05:11.840 in their pitching. If you work very hard, when you're 0:05:11.839 --> 0:05:13.839 twice as old as me, you might be half as good. 0:05:14.800 --> 0:05:17.000 Well that's sort of pressure. 0:05:18.279 --> 0:05:20.120 Did you find that motivating? 0:05:21.200 --> 0:05:24.560 I found it irritating, but I think it probably was motivating. 0:05:25.800 --> 0:05:27.360 He very inconsiderately. 0:05:27.400 --> 0:05:31.479 He died while I was writing my thesis, and he 0:05:31.600 --> 0:05:33.839 never saw me being a success. 0:05:34.600 --> 0:05:37.320 You were at Cambridge, you left briefly and you came back. 0:05:38.400 --> 0:05:42.039 I think you settled on experimental psychology as your degree. 0:05:42.040 --> 0:05:44.440 In the end, I was doing natural sciences. 0:05:45.120 --> 0:05:50.279 I started off doing physics chemist ring crystalline state because 0:05:50.279 --> 0:05:53.159 of the success in decoding the structure of DNA. Crystalline 0:05:53.160 --> 0:05:56.640 state was a big thing, and I left after a month. 0:05:56.880 --> 0:06:00.480 Then I reapplied to do architecture. I've always liked oarchitecture, 0:06:01.400 --> 0:06:03.480 and after a day I decided I was never going 0:06:03.520 --> 0:06:06.479 to be any good architecture because I wasn't I wasn't 0:06:06.560 --> 0:06:09.840 artistic enough. I loved the engineering, but the artistic bit 0:06:09.839 --> 0:06:13.000 I couldn't do very well. So I switched back to science. 0:06:13.440 --> 0:06:16.920 But then I did physics, chemistry and physiology, and I 0:06:16.960 --> 0:06:20.200 really liked physiology. I'd never been allowed to do biology 0:06:20.240 --> 0:06:20.760 at school. 0:06:21.760 --> 0:06:22.960 My father wouldn't allow it. 0:06:23.480 --> 0:06:26.919 Why not, Ah well, he said, if you do biology, 0:06:26.920 --> 0:06:30.359 they'll teach you genetics. And he was a Stalinist and 0:06:30.360 --> 0:06:34.600 he didn't believe in genetics. Now, he was also a 0:06:34.640 --> 0:06:39.159 fellow of the British Royal Society in Biology who didn't 0:06:39.160 --> 0:06:41.000 believe in genetics. 0:06:41.520 --> 0:06:46.680 Gosh, complicated man, Yes, but I mean he really you 0:06:46.720 --> 0:06:50.240 won't exaggerated him. He said to you, you can't study biology. 0:06:50.560 --> 0:06:50.800 Now. 0:06:50.920 --> 0:06:53.760 He had other reasons which weren't so bad, which is, 0:06:54.120 --> 0:06:57.719 you can always pick up biology when you're older. What 0:06:57.839 --> 0:06:59.800 you can't pick up when you're older is math. And 0:07:00.760 --> 0:07:02.880 I think that's probably true, and so that was a 0:07:02.960 --> 0:07:03.839 more valid reason. 0:07:04.360 --> 0:07:07.479 Yeah, yeah, So what did you end up graduating from 0:07:07.480 --> 0:07:10.040 Cambridge with a degree in psychology? 0:07:10.320 --> 0:07:14.880 So I did physics, chemistry and physiology for a year, 0:07:15.840 --> 0:07:17.440 and I did very well in physics. I got a 0:07:17.480 --> 0:07:20.800 first in physics. That's obviously a good predictor of a 0:07:20.840 --> 0:07:21.720 Nobel Price. 0:07:22.400 --> 0:07:23.880 Said, your tutors would have been there. 0:07:24.960 --> 0:07:29.760 Then I dropped it all and did philosophy, philosophy, philosophy. 0:07:29.800 --> 0:07:33.640 I did a year of philosophy and I developed strong antibodies, 0:07:34.640 --> 0:07:37.600 and then I switched to psychology, and so my final 0:07:37.640 --> 0:07:38.720 degree was in psychology. 0:07:39.160 --> 0:07:42.520 MHM. And was there a single question or set of 0:07:42.640 --> 0:07:44.960 questions that you were in search of. 0:07:45.400 --> 0:07:49.040 Yes, I wanted to know how the how the brain worked, 0:07:49.040 --> 0:07:51.280 and how the mind worked, and what the relationship was. 0:07:52.640 --> 0:07:55.000 I decided fairly early on that you're never going to 0:07:55.040 --> 0:07:56.920 understand the mind unless you understood the brain. 0:07:58.280 --> 0:07:59.800 Was that a popular view at the time. 0:08:00.200 --> 0:08:01.000 No, not really. 0:08:01.320 --> 0:08:03.920 There was this kind of functionalist view that basically the 0:08:04.000 --> 0:08:07.280 view that came from computer software, which is that the 0:08:07.360 --> 0:08:10.760 software is totally different from the hardware, and what the 0:08:10.800 --> 0:08:13.440 mind is all about is software, the heuristics you use 0:08:13.480 --> 0:08:15.600 and the way you represent things. The hardware has got 0:08:15.640 --> 0:08:19.200 nothing to do with it. That's a completely crazy view, 0:08:19.200 --> 0:08:22.040 but it seemed very plausible at the time, and so 0:08:22.120 --> 0:08:25.400 the computers we designed so that we could program them, 0:08:25.720 --> 0:08:29.440 had programs being a completely separate domain from hardware. 0:08:30.360 --> 0:08:32.080 But that's not how the real brain is. 0:08:33.440 --> 0:08:37.719 Was as funny this constant dance between us expecting our 0:08:37.720 --> 0:08:40.560 computers to be like us, and then expecting us to 0:08:40.640 --> 0:08:44.320 be like our computers, right as a kind of continual 0:08:44.400 --> 0:08:45.800 dance between those two things. 0:08:46.320 --> 0:08:49.360 Yes, well, you always try and understand things in terms 0:08:49.400 --> 0:08:54.280 of the latest technology. So when telephones were new, the 0:08:54.320 --> 0:08:56.559 brain was clearly a very large telephone switchboard. 0:08:58.800 --> 0:09:01.160 But is it different this time in our AI has 0:09:01.800 --> 0:09:04.720 taken off and become ubiquitous. Do you think this is 0:09:04.840 --> 0:09:07.199 indeed more than the telephone? 0:09:07.480 --> 0:09:11.040 Yes, I think these artimicial neural networks for training are 0:09:11.320 --> 0:09:14.280 in many respects quite like real neural networks. Obviously, the 0:09:14.320 --> 0:09:17.559 neurons are much simpler. There's all sorts of properties that 0:09:17.600 --> 0:09:21.200 are different in the brain, but basically they're working in 0:09:21.240 --> 0:09:24.240 the same way. They learn things by changing connection strengths 0:09:24.240 --> 0:09:26.520 between neurons, just like the brain does. 0:09:27.400 --> 0:09:29.559 And when does that question for you of wanting to 0:09:29.640 --> 0:09:32.840 understand how the brain works really really begin? 0:09:33.080 --> 0:09:36.280 When I was at high school, so even before I 0:09:36.280 --> 0:09:39.199 went to Cambridge, I had a very bright friend who 0:09:39.240 --> 0:09:42.120 was always smarter than me called Inmund Harvey, who came 0:09:42.160 --> 0:09:46.400 into school one day and said, maybe memories in the 0:09:46.400 --> 0:09:48.760 brain are spread out over the whole brain. They're not 0:09:48.840 --> 0:09:52.960 localized like a hologram, because holograms when you so he 0:09:53.040 --> 0:09:55.080 was trying to understand memories in the brain in terms 0:09:55.120 --> 0:09:57.520 of this new technology of holograms. 0:09:57.280 --> 0:09:58.679 And you were stimulated by this idea. 0:09:58.720 --> 0:09:59.599 I was very stimulated by that. 0:09:59.679 --> 0:10:03.440 It's very interesting idea, and ever since then I've thought 0:10:03.480 --> 0:10:05.760 a lot about how memories are represented in the brain. 0:10:06.520 --> 0:10:08.800 And then that also led into well how does the 0:10:08.840 --> 0:10:09.720 brain learn stuff? 0:10:11.559 --> 0:10:14.439 Coming up how Jeffrey Hinton ended up at Google. 0:10:14.920 --> 0:10:23.319 Stay with us, so bear with me, but I need 0:10:23.360 --> 0:10:27.800 to try and summarize what the Boltziman machine is because 0:10:27.840 --> 0:10:31.040 the Nobel Prize Committee credited the Boltzman machine with your win. 0:10:32.000 --> 0:10:35.600 According to their press release, the Boltzman machine can quote 0:10:36.000 --> 0:10:39.840 learn to recognize characteristic elements in a given type of data. 0:10:40.640 --> 0:10:43.400 The machine is trained by feeding it examples that are 0:10:43.520 --> 0:10:47.280 very likely to arise. The Boltzmer machine can be used 0:10:47.280 --> 0:10:50.960 to classify images or create new examples of the type 0:10:51.000 --> 0:10:54.199 of pattern on which it was trained. Hinton has built 0:10:54.240 --> 0:10:58.400 upon this work helping initiate the current explosive development of 0:10:58.440 --> 0:11:02.840 machine learning. Now, in fact, the timeline right, you graduated 0:11:02.840 --> 0:11:05.920 from Cambridge, you worked on your PhD, and in the 0:11:05.960 --> 0:11:08.880 eighties you wound up at Connigie Mellon and that's where 0:11:08.880 --> 0:11:11.160 you work on the bolt of machine really took off. 0:11:12.080 --> 0:11:14.480 Uh yes, just before I went to Carnegie Mellon. 0:11:14.679 --> 0:11:16.760 Now am I writing thinking? At the time, there was 0:11:16.800 --> 0:11:20.840 a kind of debate that you were on the unpopular 0:11:20.960 --> 0:11:23.280 side of about artificial intelligence. 0:11:23.640 --> 0:11:25.839 Okay, so there's two things to say about that. From 0:11:25.920 --> 0:11:29.839 the earliest days of AI in the nineteen fifties, there 0:11:29.840 --> 0:11:32.640 were these two approaches, two kind of paradigms. So how 0:11:32.679 --> 0:11:36.160 you build an intelligence system. One was inspired by the brain, 0:11:36.200 --> 0:11:40.000 that was neural networks. Then the other paradigm was no, no, no, no. 0:11:40.840 --> 0:11:44.319 Logic is a paradigm for intelligence. Intelligence is all about reasoning. 0:11:45.080 --> 0:11:47.439 Learning comes later once we figured out how reasoning works, 0:11:47.760 --> 0:11:50.640 and reasoning depends on having the right representations for things. 0:11:51.160 --> 0:11:54.520 So we have to figure out what kind of logically 0:11:54.600 --> 0:11:58.040 unambiguous language the brain is using so that it can 0:11:58.240 --> 0:12:01.720 use rules to do reasoning. It's a completely different approach 0:12:02.080 --> 0:12:05.200 and it's very unbiological because reasoning something comes very late, 0:12:06.360 --> 0:12:09.920 and actually from most of the second half of the 0:12:09.960 --> 0:12:13.600 last century, neural networks weren't seen as AI. AI was 0:12:13.640 --> 0:12:16.240 believing you have symbolic rules in your head and then 0:12:16.320 --> 0:12:20.240 manipulated using rules, using to screet symbolic rules. 0:12:20.080 --> 0:12:21.720 And what was what were neural? They were seen as 0:12:21.720 --> 0:12:23.160 statistics or physically. 0:12:23.200 --> 0:12:26.040 They were seen as this crazy idea that you could 0:12:26.040 --> 0:12:28.480 take this random neural network and just learn everything, which 0:12:28.520 --> 0:12:29.440 was obviously hopeless. 0:12:29.440 --> 0:12:30.680 And what what gave you this conviction? 0:12:32.040 --> 0:12:34.840 Well, the brain has to learn somehow, and of course 0:12:34.840 --> 0:12:38.200 there's a lot of innate structure whir i'd in, which 0:12:38.200 --> 0:12:40.160 explains why a goat can sort of drop out of 0:12:40.160 --> 0:12:44.440 the womb and five minutes later it's walking. But so 0:12:44.480 --> 0:12:47.360 stuff like learning to read, that's all learned. That's not innate, 0:12:48.280 --> 0:12:49.760 and we have to figure out how that works, and 0:12:49.800 --> 0:12:51.080 it's not symbolic rules. 0:12:51.480 --> 0:12:55.560 This conviction that you have now, did you always have? 0:12:56.720 --> 0:12:57.040 Yes? 0:13:00.000 --> 0:13:02.360 Sorry to say it seemed to me it was obviously right. 0:13:03.120 --> 0:13:04.760 I think part of that is I was sent to 0:13:04.760 --> 0:13:08.000 a private school when I was seven from an atheist family, 0:13:09.240 --> 0:13:13.720 and they started telling me about religion and all this rubbish, 0:13:14.400 --> 0:13:18.760 all the stuff that was obviously nonsense, and I had 0:13:18.760 --> 0:13:20.600 the experience of being the only kid at school who 0:13:20.600 --> 0:13:23.040 thought this was all nonsense and turning out to be right. 0:13:24.280 --> 0:13:25.880 And that's very good training for a scientist. 0:13:26.679 --> 0:13:29.000 I brought it up earlier, but there was a wonderful 0:13:29.000 --> 0:13:31.880 profile of you in the New Yorker titled why the 0:13:31.920 --> 0:13:35.760 Godfather of Ai fears what He's built, And there was 0:13:35.760 --> 0:13:39.439 a quote that I found quite stunning where you said 0:13:39.880 --> 0:13:42.280 I was dead in the water at forty six. 0:13:43.320 --> 0:13:50.160 Yes, my wife and I adopted to children, to babies, 0:13:52.440 --> 0:13:58.120 and then my wife got ovarying cancer. But she also, 0:13:59.040 --> 0:14:01.080 even though she was a science she started believing in 0:14:01.120 --> 0:14:05.040 homeopathy and she tried treating her very in cancer with homeopathy, 0:14:05.720 --> 0:14:08.400 and so she died and. 0:14:09.760 --> 0:14:10.520 I was left with. 0:14:10.520 --> 0:14:12.480 Two young children, one of three and one of five, 0:14:15.320 --> 0:14:22.600 who were both very upset, as was I, and I 0:14:22.760 --> 0:14:25.840 began to appreciate what life is like for female academics, 0:14:27.240 --> 0:14:28.240 which is impossible. 0:14:28.840 --> 0:14:29.840 You can't. 0:14:30.400 --> 0:14:35.320 Looking after small children makes it very very hard to 0:14:35.400 --> 0:14:39.040 spend long periods of time thinking about your current idea. 0:14:40.520 --> 0:14:41.680 It's just very difficult. 0:14:42.600 --> 0:14:46.320 So you were bereaved, you were looking after two small children, 0:14:46.440 --> 0:14:49.320 and then yes, that moment in twenty twelve when you 0:14:49.360 --> 0:14:55.040 publish a paper called ImageNet classification with deep convolutional neural networks, 0:14:55.120 --> 0:14:58.680 which to the layman doesn't sound like something that would 0:14:58.760 --> 0:15:02.520 change the world and how we live, but it did well. 0:15:02.560 --> 0:15:06.400 It changed the views of people in computer vision and 0:15:06.440 --> 0:15:08.760 the views of people in other areas of computer science. 0:15:09.440 --> 0:15:13.800 It basically showed neural networks actually do work. Now, people 0:15:13.840 --> 0:15:15.920 have showed that before, but it hadn't convinced people the 0:15:16.000 --> 0:15:16.440 same way. 0:15:16.760 --> 0:15:19.680 See, you published that paper in twenty twelve with Iliot 0:15:19.760 --> 0:15:21.920 sutch Keva, who went on to be a very important 0:15:21.960 --> 0:15:23.840 figure at open AI and I want to talk more 0:15:23.840 --> 0:15:27.080 about him. But within a few days of the publication 0:15:27.120 --> 0:15:30.600 of that paper, you had an offer of millions and 0:15:30.600 --> 0:15:33.720 millions of dollars to move to China. 0:15:33.960 --> 0:15:37.000 Ah. Yeah, either to move to China or to let 0:15:37.000 --> 0:15:38.720 them invest in our group. 0:15:39.280 --> 0:15:41.200 I think it was a bit longer than a few days, 0:15:41.200 --> 0:15:42.840 but it was. It was that fall for sure. 0:15:43.080 --> 0:15:45.280 Did you kind of know, Okay, we're going to publish 0:15:45.520 --> 0:15:49.680 and this is going to change everything or are you 0:15:49.720 --> 0:15:51.040 surprised by this response. 0:15:51.520 --> 0:15:53.400 We thought it would have a big effect. We didn't 0:15:53.400 --> 0:15:54.760 realize quite how big. 0:15:55.600 --> 0:15:59.120 And when you got the call from Baidu, Chinese tech company, 0:16:00.040 --> 0:16:02.080 What did you think, I mean, it must have been tempting. 0:16:02.760 --> 0:16:07.800 H yes, I think they said they could fund our group, 0:16:07.920 --> 0:16:10.000 or they could we could go and work for them, 0:16:10.600 --> 0:16:14.200 or various possible alternatives. And in the end I just 0:16:14.240 --> 0:16:16.160 asked them, well, how much money are we talking about? 0:16:17.000 --> 0:16:20.680 Are you talking about like ten million dollars? And they said, yes. 0:16:21.120 --> 0:16:22.160 You localed yourself. 0:16:23.440 --> 0:16:24.600 I didn't think it at the time. 0:16:25.720 --> 0:16:27.240 I thought of the biggest number I could think of, 0:16:27.280 --> 0:16:32.240 it was five million dollars and doubled it. So once 0:16:32.240 --> 0:16:34.520 they said that, I realized we had no idea how 0:16:34.600 --> 0:16:37.760 much we were worth. And so at that point we 0:16:37.840 --> 0:16:39.320 decided to set up an auction. 0:16:39.720 --> 0:16:41.720 What does it mean to set up an auction for 0:16:41.800 --> 0:16:43.200 an academic paper? 0:16:43.480 --> 0:16:47.560 The three of us founded a company. The company that 0:16:47.640 --> 0:16:50.680 belonged to the three of us owned these six pandent applications, 0:16:51.800 --> 0:16:54.360 so then we had something to sell, but mainly we 0:16:54.360 --> 0:16:57.560 were selling ourselves. But I insisted that I still be 0:16:57.600 --> 0:17:00.520 an academic so I could continue to advise my current students. 0:17:01.400 --> 0:17:04.120 That was a big problem because Google had never done 0:17:04.160 --> 0:17:06.400 that before. They were one of the companies. 0:17:06.440 --> 0:17:08.760 The bid for us, so by do Google. 0:17:08.840 --> 0:17:12.560 Microsoft and deep Mind, which wasn't owned by Google then. 0:17:13.000 --> 0:17:16.359 And not only do you invent this well together with 0:17:16.400 --> 0:17:19.359 your colleagues, this advancement of neural nets, but you also 0:17:19.400 --> 0:17:24.320 invented your own process to sell this company essentially, right. 0:17:24.720 --> 0:17:30.119 Yeah, we decided we just have an auction by Gmail, 0:17:31.000 --> 0:17:34.480 and you'd send me Gmail with your bid, and the 0:17:35.080 --> 0:17:37.440 time of the bid would be the timestamp on the Gmail. 0:17:38.560 --> 0:17:40.200 I would then immediately send the bid to all the 0:17:40.240 --> 0:17:44.400 other bidders, and you had to raise by a million dollars, 0:17:44.720 --> 0:17:47.480 and if there was no bid within an hour of 0:17:47.520 --> 0:17:48.960 the last bid, that was the end of the auction. 0:17:50.320 --> 0:17:52.320 I was amazed to see that the bids would come 0:17:52.320 --> 0:17:55.560 in fifty nine minutes after the previous bid. Would be 0:17:55.560 --> 0:17:58.440 sitting there think okay, it's over, and then fifty nine 0:17:58.440 --> 0:17:59.720 minutes later a bid would come in. 0:18:00.880 --> 0:18:02.200 Only everybody trusted Gmail. 0:18:02.880 --> 0:18:05.800 I had worked at Google over the summer, and that 0:18:05.920 --> 0:18:08.199 meant two things. One, I did trust that they wouldn't 0:18:08.200 --> 0:18:11.320 read my Gmail. I knew they were very serious about that. 0:18:12.280 --> 0:18:15.960 And also I really wanted Google to win the competition 0:18:16.000 --> 0:18:18.840 because I really like Jeff Dean, who ran the brain 0:18:18.880 --> 0:18:22.640 team at Google. So I wanted Google to win, and 0:18:23.359 --> 0:18:26.240 in the end we slightly fudged it. So after deep 0:18:26.280 --> 0:18:29.119 minded Microsoft had dropped out, it was between Google and 0:18:29.119 --> 0:18:32.800 by Do, and we were scared by You would win, 0:18:33.520 --> 0:18:35.159 and I didn't want to go to China. 0:18:36.960 --> 0:18:39.240 I couldn't travel at that point, well, so I felt 0:18:39.240 --> 0:18:39.960 I wouldn't. 0:18:39.680 --> 0:18:43.040 Really understand what was going on in a Chinese company. 0:18:43.160 --> 0:18:46.919 So it got to forty four million, and we just 0:18:47.000 --> 0:18:50.120 said we've got an offer we can't refuse, and it's 0:18:50.160 --> 0:18:52.520 the end of the auction. And the offer we couldn't 0:18:52.560 --> 0:18:54.320 refuse was to work with Jeff Deane. 0:18:54.960 --> 0:18:58.240 So you got what you wanted in the sense of. 0:18:57.920 --> 0:19:00.200 We've got more money than we could imagine, and we 0:19:00.280 --> 0:19:03.040 got to work at the company that I wanted most 0:19:03.080 --> 0:19:03.600 to work. 0:19:03.440 --> 0:19:10.119 With when we come back. Nobel Laureate Jeffrey Hinton on 0:19:10.200 --> 0:19:27.520 why he advocates for AI safety. In twenty eighteen, you 0:19:27.640 --> 0:19:29.639 receive the Touring Award, which is kind of like the 0:19:29.680 --> 0:19:33.320 Nobel Prize for Computer Science. But also in twenty eighteen, 0:19:33.800 --> 0:19:36.399 you were widowed for a second time. You lost your 0:19:36.440 --> 0:19:40.000 wife Rosalind in nineteen ninety four, and then twenty four 0:19:40.040 --> 0:19:44.200 years later your wife Jackie passed away also from cancer. 0:19:45.119 --> 0:19:52.560 Yes, that was that was difficult. I didn't have My 0:19:52.640 --> 0:19:54.840 children were much older then, so I didn't have the 0:19:54.840 --> 0:19:57.520 problem of having to cope with young children at the 0:19:57.520 --> 0:20:01.720 same time as everything else. And Google was very understanding. 0:20:02.560 --> 0:20:04.160 Part of my deal had been that I would spend 0:20:04.200 --> 0:20:09.720 three months a year in Silicon Valley, and they let 0:20:09.760 --> 0:20:11.000 me out of that and said I could spend my 0:20:11.040 --> 0:20:14.800 whole time in Toronto, and they helped me set up 0:20:14.800 --> 0:20:19.000 a small lab doing basic research in Toronto, so that 0:20:19.119 --> 0:20:23.359 was much less stressful, so I could be with my wife. 0:20:23.640 --> 0:20:26.479 She had cancer as well, she got pancreatic cancer. 0:20:29.359 --> 0:20:32.520 One of the most striking things, again in the New 0:20:32.560 --> 0:20:38.760 Yorker piece, was the way you talked about observing the 0:20:38.800 --> 0:20:45.760 way in which Rosalind and Jackie approached their cancer as 0:20:45.800 --> 0:20:50.760 a mental model for how to think about the implications 0:20:50.760 --> 0:20:52.040 of artificial intelligence. 0:20:52.680 --> 0:20:55.359 Yeah, that's a rather sort of dark scenario. 0:20:57.440 --> 0:20:59.840 So occasionally when I think, well, lis stuff probably will 0:20:59.840 --> 0:21:05.679 take over, which I sometimes think, then there's the issue 0:21:05.720 --> 0:21:10.800 of should you go into denial and say no, no, no, no, 0:21:10.840 --> 0:21:15.080 this can't possibly happen, which is what my first wife did. Actually, 0:21:15.119 --> 0:21:17.040 she wasn't my first wife I was married a long 0:21:17.040 --> 0:21:25.680 time before that, just briefly, and so Ross went into 0:21:25.800 --> 0:21:31.040 denial and Jackie was very realistic about it. And maybe 0:21:31.080 --> 0:21:34.440 we should be very realistic about the possibility machines will 0:21:34.480 --> 0:21:37.040 take over. We should do our best to make sure 0:21:37.080 --> 0:21:40.680 that doesn't happen, but we should also think about whether 0:21:40.720 --> 0:21:43.440 there's ways of making that if that does happen, whether 0:21:43.480 --> 0:21:46.280 there's ways of making it not so bad, whether people 0:21:46.280 --> 0:21:48.639 could still be around even if the machines were in control, 0:21:48.680 --> 0:21:49.240 for example. 0:21:49.640 --> 0:21:51.760 Yeah, I mean there's something. I mean, you use the 0:21:51.800 --> 0:21:56.800 word dark, but you've seen your life's work in many 0:21:56.800 --> 0:22:02.080 ways kind of fruition, haunted by these thoughts of how 0:22:02.119 --> 0:22:06.680 to deal with something as awful as the terminal council diagnosis. 0:22:07.200 --> 0:22:09.919 Yeah, you have to be careful what you wish for. 0:22:11.440 --> 0:22:17.040 Yeah, I mean my mentioned Oppenheimer at the beginning in 0:22:17.160 --> 0:22:20.040 terms of a Nobel Well, he wasn't almost a physics gloriate. 0:22:20.080 --> 0:22:20.800 He wasn't in the end. 0:22:22.000 --> 0:22:24.800 Yes, it's rather absurd, isn't it That I've got a 0:22:24.880 --> 0:22:26.800