1 00:00:00,120 --> 00:00:02,440 Speaker 1: Hey, it's us here. I'm traveling this week and next 2 00:00:02,440 --> 00:00:04,680 Speaker 1: week is the fourth of July, so we won't be 3 00:00:04,680 --> 00:00:07,960 Speaker 1: publishing a Week in Tech episode for the next two Fridays, 4 00:00:08,119 --> 00:00:10,680 Speaker 1: but you'll hear from all of our favorite tech contributors 5 00:00:10,760 --> 00:00:14,040 Speaker 1: once again starting July tenth. In the meantime, I wanted 6 00:00:14,040 --> 00:00:17,080 Speaker 1: to share an episode that demystified the man behind one 7 00:00:17,120 --> 00:00:20,480 Speaker 1: of the most influential AI labs in the world, Google 8 00:00:20,520 --> 00:00:21,040 Speaker 1: Deep Mind. 9 00:00:21,480 --> 00:00:47,000 Speaker 2: Hope you enjoy. 10 00:00:38,159 --> 00:00:40,920 Speaker 1: Welcome to Tech Stuff. I'm os Voloscan, and today we 11 00:00:41,000 --> 00:00:44,280 Speaker 1: get the opportunity to go behind the curtain at Google's 12 00:00:44,360 --> 00:00:47,959 Speaker 1: Deep Mind. For almost three years, in the upstairs room 13 00:00:48,000 --> 00:00:51,880 Speaker 1: of a pub in North London, journalist Sebastian Malabi met 14 00:00:51,880 --> 00:00:55,959 Speaker 1: regularly with the company's CEO and co founder, Demis Hasabis. 15 00:00:56,680 --> 00:01:02,480 Speaker 1: They spoke about artificial intelligence, philosophy, neuroscience, motivation and consequence, 16 00:01:03,320 --> 00:01:06,559 Speaker 1: all against the backdrop of an increasingly intense three way 17 00:01:06,680 --> 00:01:10,400 Speaker 1: race between Open AI Andthropic and Google to win the 18 00:01:10,480 --> 00:01:14,520 Speaker 1: race towards Agi so Ba see. Congratulations on your new book, 19 00:01:14,840 --> 00:01:17,240 Speaker 1: The Infinity Machine and welcome to tex Stuff. 20 00:01:17,440 --> 00:01:18,600 Speaker 3: Thank you ask great to be here. 21 00:01:19,120 --> 00:01:21,240 Speaker 1: You begin the book with a quote from one of 22 00:01:21,280 --> 00:01:24,399 Speaker 1: the scientists who work on the Manhattan Project who said, 23 00:01:24,880 --> 00:01:27,520 Speaker 1: what we are creating now is a monster whose influence 24 00:01:27,600 --> 00:01:30,280 Speaker 1: is going to change history. Yet it would be impossible 25 00:01:30,480 --> 00:01:33,080 Speaker 1: not to see it through. The energy source which is 26 00:01:33,120 --> 00:01:36,959 Speaker 1: now being made available will make scientists the most hated 27 00:01:37,319 --> 00:01:39,959 Speaker 1: and the most wanted citizens of any country. 28 00:01:40,400 --> 00:01:43,919 Speaker 3: We're reliving that now with AI. Agree. I mean, I 29 00:01:43,959 --> 00:01:47,880 Speaker 3: began this project wanting to capture the tingling sensation of 30 00:01:48,000 --> 00:01:51,600 Speaker 3: human beings like demisis ABIs creating the new version of 31 00:01:51,880 --> 00:01:56,320 Speaker 3: atomic weapons. Right, This incredibly powerful AI technology that has 32 00:01:56,440 --> 00:02:00,480 Speaker 3: enormous upsides could also be very, very dangerous. The prize 33 00:02:00,560 --> 00:02:02,040 Speaker 3: was I didn't have to bring it up to them. 34 00:02:02,080 --> 00:02:03,480 Speaker 3: They brought it up to me. I mean, it's so 35 00:02:03,560 --> 00:02:05,880 Speaker 3: much on their minds. And so that's why I put 36 00:02:05,920 --> 00:02:08,679 Speaker 3: this quotation about the Manhattan Project at the start of 37 00:02:08,720 --> 00:02:11,400 Speaker 3: the book, because it's kind of the It sums up 38 00:02:11,800 --> 00:02:14,600 Speaker 3: one of the main threads, which is this, you know, 39 00:02:14,800 --> 00:02:20,400 Speaker 3: scientists can't resist inventing something which is exciting technically, and 40 00:02:20,440 --> 00:02:22,520 Speaker 3: then they're going to be the most hated and most 41 00:02:22,520 --> 00:02:24,760 Speaker 3: wanted people in the country. 42 00:02:25,160 --> 00:02:29,240 Speaker 1: You described the question of motivation when it comes to 43 00:02:29,280 --> 00:02:33,320 Speaker 1: Demis hanging in the air like the mushroom cloud over 44 00:02:33,360 --> 00:02:39,040 Speaker 1: Los Alamos, very arresting visual image. Why that image and 45 00:02:39,440 --> 00:02:40,840 Speaker 1: what did you want to stand in the end about 46 00:02:40,880 --> 00:02:41,440 Speaker 1: his motivation? 47 00:02:42,000 --> 00:02:43,760 Speaker 3: I think when you look at that picture of the 48 00:02:43,880 --> 00:02:48,760 Speaker 3: mushroom cloud over Los Alamos, you're kind of thinking both wow, 49 00:02:49,200 --> 00:02:52,840 Speaker 3: but also why why did human beings do this? You 50 00:02:52,880 --> 00:02:55,960 Speaker 3: know it's so destructive? Why did you do it? And 51 00:02:56,120 --> 00:02:59,240 Speaker 3: I guess you know. Part of the thread in my 52 00:02:59,280 --> 00:03:03,040 Speaker 3: book is he had a series of ideas about how 53 00:03:03,080 --> 00:03:06,520 Speaker 3: he could build AIE and make it safe for humanity 54 00:03:06,560 --> 00:03:10,200 Speaker 3: and beneficial for humanity, and one by one these ideas 55 00:03:10,280 --> 00:03:13,560 Speaker 3: become unraveled as they collide with reality. So the story 56 00:03:13,600 --> 00:03:16,799 Speaker 3: of demisisabass in some ways, you know a story. I 57 00:03:16,840 --> 00:03:19,000 Speaker 3: think there's two categories of screw up in the world. Right. 58 00:03:19,919 --> 00:03:24,440 Speaker 3: Sometimes you get something where basically idiots are in charge, 59 00:03:24,800 --> 00:03:27,040 Speaker 3: they don't understand what they're doing, and they make a 60 00:03:27,080 --> 00:03:30,920 Speaker 3: humongous mistake Iran War for example. Right Then you have 61 00:03:30,960 --> 00:03:34,720 Speaker 3: another much more interesting category of screw up, and that 62 00:03:34,880 --> 00:03:38,480 Speaker 3: is where intelligent people know from the beginning exactly what 63 00:03:38,520 --> 00:03:40,680 Speaker 3: they're doing. They can see the risks, they think they 64 00:03:40,680 --> 00:03:43,560 Speaker 3: can manage them, but then forces which are larger than them. 65 00:03:44,120 --> 00:03:46,560 Speaker 3: In this case with AI, it's a race dynamic between 66 00:03:46,600 --> 00:03:50,080 Speaker 3: multiple labs and multiple countries take over and they can 67 00:03:50,120 --> 00:03:55,280 Speaker 3: no longer control the technology that they've invented. And I 68 00:03:55,320 --> 00:03:58,320 Speaker 3: think those episodes where you couldn't just switch out the 69 00:03:58,320 --> 00:04:02,920 Speaker 3: individuals and have a better act come where the individual 70 00:04:03,000 --> 00:04:07,080 Speaker 3: is good, sincerely, good, intelligent, thoughtful, has foresight, and yet 71 00:04:07,120 --> 00:04:08,840 Speaker 3: you still end up in a bad place. That's what's 72 00:04:08,880 --> 00:04:12,720 Speaker 3: really fascinating. When Demis sold his company deep Mind to 73 00:04:12,840 --> 00:04:16,160 Speaker 3: Google in twenty fourteen, there was a condition which was 74 00:04:16,320 --> 00:04:18,880 Speaker 3: this will never be used for weapons. Well, you know 75 00:04:19,000 --> 00:04:22,440 Speaker 3: now it's twenty twenty six it is being used for weapons. 76 00:04:22,560 --> 00:04:25,600 Speaker 3: And you know, so in time after time he tried 77 00:04:25,640 --> 00:04:29,080 Speaker 3: to draw lines in the sand, and they've all been erased. 78 00:04:29,240 --> 00:04:32,680 Speaker 1: It's harder and harder to say what these AI companies 79 00:04:32,760 --> 00:04:36,479 Speaker 1: are today. I mean, for a moment last year, OpenAI 80 00:04:36,680 --> 00:04:39,880 Speaker 1: was the most popular social video app company in the world, 81 00:04:39,920 --> 00:04:42,560 Speaker 1: and now it doesn't do Sora anymore. But like, what 82 00:04:42,720 --> 00:04:44,880 Speaker 1: is your working definition of what deep Mind is. 83 00:04:45,440 --> 00:04:48,280 Speaker 3: I mean, it's a laboratory for the invention of machine intelligence, 84 00:04:48,440 --> 00:04:52,080 Speaker 3: and machine intelligence is a very compacious thing. You know, 85 00:04:52,120 --> 00:04:57,240 Speaker 3: it goes from text to video to images to a 86 00:04:57,320 --> 00:05:01,479 Speaker 3: system like Alpha fold, which divined all the shapes of 87 00:05:01,520 --> 00:05:04,960 Speaker 3: proteins in nature, and one demis the Nobel Price. So 88 00:05:05,080 --> 00:05:08,440 Speaker 3: it's a huge field, and indeed the creation of it 89 00:05:08,480 --> 00:05:11,760 Speaker 3: is a huge thing because you bring in experts in neuroscience, 90 00:05:11,880 --> 00:05:15,920 Speaker 3: experts in chemistry, experts in physics, experts, computer science, ethical 91 00:05:16,240 --> 00:05:19,520 Speaker 3: experts who can philosophize about the personality that an AI 92 00:05:19,560 --> 00:05:22,920 Speaker 3: system should have. I mean, it's a very multidisciplinary thing, 93 00:05:22,960 --> 00:05:24,839 Speaker 3: which is part of what makes the story fascinating. 94 00:05:24,960 --> 00:05:26,400 Speaker 1: When did you first meet Demis Now? 95 00:05:26,480 --> 00:05:30,040 Speaker 3: I first met him when I'd being interested in, you know, 96 00:05:30,080 --> 00:05:32,800 Speaker 3: technology generally. My last book was about Silicon Valley and 97 00:05:32,880 --> 00:05:36,839 Speaker 3: venture capital, and in the process of writing that book, 98 00:05:37,360 --> 00:05:40,040 Speaker 3: I would go to tech conferences in Europe and there 99 00:05:40,080 --> 00:05:43,240 Speaker 3: would be this sort of diminutive figure with a big, 100 00:05:43,279 --> 00:05:46,200 Speaker 3: big smile and sort of a kind of boyish charm, 101 00:05:46,360 --> 00:05:49,600 Speaker 3: really unassuming kind of guy with you know, a sort 102 00:05:49,640 --> 00:05:53,200 Speaker 3: of round neck sweater and his hair falling forward in 103 00:05:53,240 --> 00:05:56,240 Speaker 3: a fringe, and he would get up on the stage 104 00:05:56,560 --> 00:05:59,000 Speaker 3: with a big grin and sort of just almost as 105 00:05:59,040 --> 00:06:01,360 Speaker 3: if he was talking about how he was about to 106 00:06:01,600 --> 00:06:03,920 Speaker 3: wash the dishes after lunch. You know, he would in 107 00:06:03,960 --> 00:06:07,400 Speaker 3: a very plain spoken way, talk about well, when I 108 00:06:07,440 --> 00:06:09,360 Speaker 3: was a child, I had two ambitions. One was to 109 00:06:09,400 --> 00:06:11,799 Speaker 3: understand all of science and the other was to understand 110 00:06:11,800 --> 00:06:15,440 Speaker 3: all of philosophy. So I resolved this dilemma by deciding 111 00:06:15,440 --> 00:06:17,559 Speaker 3: to build AI, which would help me to understand both. 112 00:06:17,960 --> 00:06:22,320 Speaker 3: And so he had this mind blowing mental reach combined 113 00:06:22,360 --> 00:06:26,880 Speaker 3: with this totally approachable nextdoor friend kind of attitude. 114 00:06:27,080 --> 00:06:29,720 Speaker 1: And when did you have the idea to pitch him 115 00:06:29,760 --> 00:06:32,159 Speaker 1: on being a biographical subject for you? 116 00:06:32,440 --> 00:06:34,520 Speaker 3: So after finishing my last book, The Power Law about 117 00:06:34,560 --> 00:06:36,520 Speaker 3: Venture Capital, I was thinking, in you know, this is 118 00:06:36,520 --> 00:06:39,599 Speaker 3: now mid twenty twenty two, what would be a good 119 00:06:39,680 --> 00:06:42,679 Speaker 3: next subject. And because I had met Demis several times 120 00:06:42,680 --> 00:06:45,159 Speaker 3: and I kind of followed what Deep Mind was up to. 121 00:06:45,240 --> 00:06:48,479 Speaker 3: I knew about the protein folding system, I knew about 122 00:06:48,560 --> 00:06:51,599 Speaker 3: Alpha Go, the go playing system before that, and so forth, 123 00:06:52,240 --> 00:06:54,719 Speaker 3: and I had a sense that it would probably go 124 00:06:54,800 --> 00:06:57,760 Speaker 3: from the fringe to the mainstream at some point in 125 00:06:57,760 --> 00:06:59,719 Speaker 3: the next year or so. And then it took me 126 00:06:59,760 --> 00:07:03,320 Speaker 3: if few months after that conversation inside my head to 127 00:07:03,560 --> 00:07:05,640 Speaker 3: get my act together and listen to every podcast that 128 00:07:05,680 --> 00:07:07,920 Speaker 3: Demis had ever done, read all his lectures, really think 129 00:07:07,960 --> 00:07:10,680 Speaker 3: my way into his brain, and then go and see 130 00:07:10,720 --> 00:07:13,400 Speaker 3: him to pitch him on giving me a ton of 131 00:07:13,440 --> 00:07:15,440 Speaker 3: time because I need a lot of time with people 132 00:07:15,720 --> 00:07:17,920 Speaker 3: if I'm going to write a book with him. And 133 00:07:19,280 --> 00:07:20,560 Speaker 3: I said, look, you know, I want to write this 134 00:07:20,560 --> 00:07:23,120 Speaker 3: book about you. And it seems to me Demists that 135 00:07:23,120 --> 00:07:26,520 Speaker 3: you may not want a book about you, but you've 136 00:07:26,560 --> 00:07:29,080 Speaker 3: said repeatedly in all of your lectures that AI will 137 00:07:29,120 --> 00:07:32,920 Speaker 3: be the most important invention in all of human history. Demists. 138 00:07:33,160 --> 00:07:36,880 Speaker 3: So that means if you're the creator of this AI, 139 00:07:37,320 --> 00:07:40,080 Speaker 3: you must be one of the most important people in 140 00:07:40,160 --> 00:07:42,440 Speaker 3: human history. And if that's the case, you don't have 141 00:07:42,440 --> 00:07:45,600 Speaker 3: a choice. Somebody's going to write a book, right And furthermore, 142 00:07:45,600 --> 00:07:48,640 Speaker 3: you should welcome this because if you're going to invent 143 00:07:48,680 --> 00:07:52,920 Speaker 3: a technology that is going to disrupt people's lives so thoroughly, 144 00:07:53,040 --> 00:07:55,680 Speaker 3: you know your job will be different, how you raise 145 00:07:55,720 --> 00:07:58,000 Speaker 3: your children will be different, how you think yourself as 146 00:07:58,040 --> 00:08:00,840 Speaker 3: a human will be different because you now have this 147 00:08:00,920 --> 00:08:04,720 Speaker 3: different source of intelligence competing with you. You can't disrupt 148 00:08:04,720 --> 00:08:07,480 Speaker 3: people from head to toe and then not explain them 149 00:08:07,680 --> 00:08:10,680 Speaker 3: why you did it. You need to explain your motives, right, 150 00:08:11,360 --> 00:08:14,840 Speaker 3: and that's that's the project I'm proposing to you. And 151 00:08:15,000 --> 00:08:17,120 Speaker 3: he thought about it, and he seemed well disposed to this, 152 00:08:17,440 --> 00:08:21,280 Speaker 3: and then one week later Chatchipt came out. 153 00:08:21,440 --> 00:08:22,360 Speaker 1: Oh my goodness, and. 154 00:08:22,360 --> 00:08:24,720 Speaker 3: My expectation of the technology going from the fringe to 155 00:08:24,760 --> 00:08:27,000 Speaker 3: the mainstream happened a whole lot quicker than I expected. 156 00:08:27,920 --> 00:08:30,520 Speaker 1: And I mean, obviously a lot of the book of 157 00:08:30,800 --> 00:08:33,559 Speaker 1: your interest is in the technology, how it might change 158 00:08:33,559 --> 00:08:37,640 Speaker 1: the world, the kind of financing and deal making shenanigans 159 00:08:37,640 --> 00:08:40,480 Speaker 1: that made deep mind in many ways what it is today. 160 00:08:40,520 --> 00:08:45,400 Speaker 1: But what other personal side of an extraordinary biographical portrait 161 00:08:45,559 --> 00:08:51,280 Speaker 1: to very specific parents, a prodigious talent for chess, which 162 00:08:51,280 --> 00:08:52,960 Speaker 1: he then gave up because he thought it wasn't in 163 00:08:52,960 --> 00:08:55,439 Speaker 1: some sense consequential enough. I mean, did you know when 164 00:08:55,480 --> 00:08:57,319 Speaker 1: you got it, you knew that AI was the next thing, 165 00:08:57,400 --> 00:08:59,400 Speaker 1: and you know there he was. But did you know 166 00:08:59,400 --> 00:09:02,559 Speaker 1: what an extra personal story he had before you really 167 00:09:02,600 --> 00:09:03,400 Speaker 1: got into it with him. 168 00:09:03,520 --> 00:09:05,880 Speaker 3: No? I didn't, And in fact, I remember very clearly 169 00:09:07,000 --> 00:09:11,080 Speaker 3: too early experiences in the first discussions. You know, one 170 00:09:11,280 --> 00:09:14,400 Speaker 3: was I was going to have this dinner I told 171 00:09:14,400 --> 00:09:17,000 Speaker 3: you about and he told me to read a book 172 00:09:17,040 --> 00:09:19,800 Speaker 3: before I came to the dinner and the book was 173 00:09:19,920 --> 00:09:22,320 Speaker 3: Enders a game. Now, this is a science fiction story 174 00:09:22,960 --> 00:09:27,160 Speaker 3: about a sort of diminutive boy genius hero who has 175 00:09:27,200 --> 00:09:31,720 Speaker 3: to save planet from invading space aliens. And he's at 176 00:09:31,760 --> 00:09:33,840 Speaker 3: the end of the book he saves all of humanity 177 00:09:33,840 --> 00:09:36,800 Speaker 3: from space aliens. And Demis said to me, well, I 178 00:09:36,800 --> 00:09:38,880 Speaker 3: wanted you to read this book because I really identify 179 00:09:39,000 --> 00:09:42,680 Speaker 3: with that character Ender. And I'm thinking, wait, so you're 180 00:09:42,720 --> 00:09:44,920 Speaker 3: telling me you're the savior of humanity. I mean, even 181 00:09:44,960 --> 00:09:47,600 Speaker 3: if you think that Demis, maybe you shouldn't be announcing 182 00:09:47,600 --> 00:09:49,040 Speaker 3: it to the person who's about to write a book 183 00:09:49,040 --> 00:09:51,400 Speaker 3: about you. I mean, surely that's too messianic, to over 184 00:09:51,400 --> 00:09:54,640 Speaker 3: the top, too ridiculous. But he's right out there with it. 185 00:09:54,679 --> 00:09:56,400 Speaker 3: You mean, that is how he thinks, and he's not 186 00:09:56,440 --> 00:09:59,640 Speaker 3: ashamed to tell you. And so that was pretty extraordinary. 187 00:09:59,679 --> 00:10:02,320 Speaker 3: And then and the second thing was I went to 188 00:10:02,320 --> 00:10:05,319 Speaker 3: see Shane leg his Scientific co founder, and he told 189 00:10:05,360 --> 00:10:07,360 Speaker 3: me the story about how. I said, you know, what 190 00:10:07,400 --> 00:10:09,880 Speaker 3: was it light to work with Demis? And he said, well, 191 00:10:10,120 --> 00:10:13,439 Speaker 3: you know, Demis has crazy determination. I said, well, what 192 00:10:13,520 --> 00:10:15,960 Speaker 3: do you mean He said, well, you know, one day, 193 00:10:17,120 --> 00:10:19,439 Speaker 3: according to Demis, his dad said to him during the 194 00:10:19,559 --> 00:10:22,559 Speaker 3: chess period of his life. Listen, you're going to go 195 00:10:22,600 --> 00:10:24,880 Speaker 3: play chess today. You know you just have to try 196 00:10:24,920 --> 00:10:27,400 Speaker 3: your best. Now, when I say that to my son, 197 00:10:27,480 --> 00:10:28,920 Speaker 3: I mean, you know, it's fine to lose so long 198 00:10:28,920 --> 00:10:31,200 Speaker 3: as you try your best. The way that Demis apparently 199 00:10:31,240 --> 00:10:34,560 Speaker 3: interpreted it, according to Shane, was you have to try 200 00:10:34,600 --> 00:10:37,599 Speaker 3: your absolute, absolute, absolute best. And it's like running a 201 00:10:37,679 --> 00:10:39,920 Speaker 3: race and at the end of the marathon you fall 202 00:10:39,960 --> 00:10:43,120 Speaker 3: over the tape and you're on the ground and you 203 00:10:43,120 --> 00:10:45,440 Speaker 3: have to be taken to hospital because you're almost dead. 204 00:10:45,640 --> 00:10:48,160 Speaker 3: And if you haven't been taken to hospital, it means 205 00:10:48,200 --> 00:10:51,439 Speaker 3: you didn't try hard enough. That is what try your 206 00:10:51,440 --> 00:10:55,920 Speaker 3: best meant to Demis aged about ten or twelve, And 207 00:10:56,000 --> 00:10:58,240 Speaker 3: I went to see Demis the next time and I 208 00:10:58,320 --> 00:11:01,120 Speaker 3: replayed this back to him and said, is that really true? 209 00:11:01,160 --> 00:11:03,199 Speaker 3: Is that how you interpreted? He said, oh, yeah, absolutely. 210 00:11:03,720 --> 00:11:05,960 Speaker 3: You know you have to give it every single drop 211 00:11:06,280 --> 00:11:06,880 Speaker 3: all the time. 212 00:11:07,200 --> 00:11:10,520 Speaker 1: And there's an amazing moment where Demis describes talking about 213 00:11:10,640 --> 00:11:15,560 Speaker 1: hearing nature or science screaming at him and him struggling 214 00:11:15,640 --> 00:11:16,800 Speaker 1: to hear and to understand. 215 00:11:17,200 --> 00:11:20,600 Speaker 3: Yeah, that was the most extreme expression of his desire 216 00:11:20,679 --> 00:11:23,720 Speaker 3: to invent AI. So one day I was with him 217 00:11:23,720 --> 00:11:26,720 Speaker 3: in Hampstead Heath, which is a park in North London. 218 00:11:26,840 --> 00:11:28,520 Speaker 3: Not in the pub, Not in the pub. This time 219 00:11:29,200 --> 00:11:30,560 Speaker 3: it was a nice day, so we went for this 220 00:11:30,640 --> 00:11:33,280 Speaker 3: cafe instead, and you know, there he was. There was 221 00:11:33,360 --> 00:11:35,400 Speaker 3: kind of a classic English scene. There was somebody in 222 00:11:35,400 --> 00:11:37,600 Speaker 3: front of me who was on his cell phone. You're 223 00:11:37,600 --> 00:11:39,679 Speaker 3: doing some sort of sales job, and two women behind 224 00:11:39,720 --> 00:11:43,200 Speaker 3: me talking about their friend who had a medical incident 225 00:11:43,240 --> 00:11:44,840 Speaker 3: and had to get to a hospital. So all these 226 00:11:45,080 --> 00:11:47,760 Speaker 3: quididian noises in the background, and there is demis a 227 00:11:47,800 --> 00:11:50,000 Speaker 3: sabists looking at me, talking about the creation of this 228 00:11:50,080 --> 00:11:52,920 Speaker 3: godlike machine and saying that when he's up at two 229 00:11:52,960 --> 00:11:55,880 Speaker 3: in the morning at his desk at home thinking about this, 230 00:11:56,280 --> 00:12:00,280 Speaker 3: he can sort of feel reality summoning him, streaming at him, 231 00:12:00,400 --> 00:12:03,439 Speaker 3: understand me, understand me, And you know he would then 232 00:12:03,480 --> 00:12:07,280 Speaker 3: slam the table and say, look, Sebastian, this table it's 233 00:12:07,320 --> 00:12:10,200 Speaker 3: made of atoms, buzzing around with electrons. Why should it 234 00:12:10,200 --> 00:12:12,600 Speaker 3: be solid? Why should that laptop you've got there, why 235 00:12:12,640 --> 00:12:15,040 Speaker 3: should it you know, pieces of sand and metal? How 236 00:12:15,040 --> 00:12:17,439 Speaker 3: could that turn into something which can think? I mean, 237 00:12:17,480 --> 00:12:20,160 Speaker 3: what's going on? Here. There must be some intelligent force 238 00:12:20,280 --> 00:12:23,440 Speaker 3: designing all these things. And so he kind of basically 239 00:12:23,480 --> 00:12:27,560 Speaker 3: told me that inventing AI and understanding the universe is 240 00:12:27,600 --> 00:12:29,920 Speaker 3: like getting closer to what he thinks of as God. 241 00:12:30,480 --> 00:12:31,560 Speaker 1: So he's a religious man. 242 00:12:31,679 --> 00:12:33,640 Speaker 3: I don't know if he would agree with religious because 243 00:12:33,640 --> 00:12:37,160 Speaker 3: he doesn't go to organized religious services, but he's spiritual. 244 00:12:37,240 --> 00:12:38,640 Speaker 3: I would say, interesting. 245 00:12:39,440 --> 00:12:41,439 Speaker 1: And I mean that scene you described could be a 246 00:12:41,440 --> 00:12:45,360 Speaker 1: scene from Oppenheimer, right, I mean, it's so cinematic. Did 247 00:12:45,400 --> 00:12:47,400 Speaker 1: you ever think is he doing this for me? Or 248 00:12:47,480 --> 00:12:51,319 Speaker 1: is he crazy? Or is it just absolutely captivating? The 249 00:12:51,400 --> 00:12:53,640 Speaker 1: energy in the sense of purpose that he brings to this. 250 00:12:53,840 --> 00:12:57,280 Speaker 3: He just exudes both energy and intelligence but also storytelling 251 00:12:57,360 --> 00:13:00,080 Speaker 3: natural talent. It's just amazing. I mean, you know, one 252 00:13:00,160 --> 00:13:03,080 Speaker 3: time I asked him about his first office in London 253 00:13:03,120 --> 00:13:05,959 Speaker 3: and Russell Square, which is a sort of storied square, 254 00:13:06,000 --> 00:13:08,400 Speaker 3: you know, near the British Museum and so forth. And 255 00:13:08,559 --> 00:13:10,679 Speaker 3: you know, normally, as a writer, you are somebody to 256 00:13:10,720 --> 00:13:15,480 Speaker 3: recapture the emotion of opening their first office fifteen years ago. 257 00:13:16,120 --> 00:13:17,959 Speaker 3: It's fifteen years ago. They're going to say, oh, yeah, 258 00:13:17,960 --> 00:13:19,600 Speaker 3: it was cool. You know, That's all you'll get out 259 00:13:19,640 --> 00:13:22,920 Speaker 3: of them. But demis just flows with stories. So he said, well, 260 00:13:22,960 --> 00:13:25,120 Speaker 3: you know, I was in the attic that's where the 261 00:13:25,120 --> 00:13:26,640 Speaker 3: office was, and of course you had to come down 262 00:13:26,679 --> 00:13:28,440 Speaker 3: the stairs. They were all Rickety's. So I came down 263 00:13:28,600 --> 00:13:30,679 Speaker 3: ding ding ding ding ding, bang bang bang, And then 264 00:13:30,720 --> 00:13:32,200 Speaker 3: I come out on the square and there's these beautiful 265 00:13:32,200 --> 00:13:34,040 Speaker 3: trees in front of me. Blond to the right. If 266 00:13:34,080 --> 00:13:36,040 Speaker 3: you just go three doors down, Sebastian, that's where you 267 00:13:36,080 --> 00:13:41,080 Speaker 3: see the London Mathematical Society where Turing invented the origins 268 00:13:41,120 --> 00:13:44,040 Speaker 3: of computer science, which we are now completing. And then 269 00:13:44,320 --> 00:13:47,160 Speaker 3: if you go beyond that to the level crossing black white, 270 00:13:47,200 --> 00:13:50,360 Speaker 3: black white crossing the street, the pedestrian crossing, that is 271 00:13:50,440 --> 00:13:54,800 Speaker 3: where the Hungarian nuclear scientist Zilad had the idea for 272 00:13:54,840 --> 00:13:57,560 Speaker 3: a nuclear chain reaction back in the nineteen thirties, which 273 00:13:57,640 --> 00:14:00,320 Speaker 3: led to the atom bomb. And of course we are 274 00:14:00,360 --> 00:14:02,640 Speaker 3: now creating the equivalent of the atom bomb with Ai. 275 00:14:03,120 --> 00:14:06,960 Speaker 3: What a subject, Yeah, I mean, he is such a storyteller. 276 00:14:06,960 --> 00:14:08,480 Speaker 1: And I heard that he has a sense of humor 277 00:14:08,559 --> 00:14:10,280 Speaker 1: or perhaps a sense of humour about himself in some 278 00:14:10,320 --> 00:14:12,760 Speaker 1: ways as well. Didn't he say when he lost the 279 00:14:13,600 --> 00:14:16,480 Speaker 1: table football competition? The office that his soul was on fire. 280 00:14:16,720 --> 00:14:19,080 Speaker 3: He did say that, yes, you know, one can mock 281 00:14:19,160 --> 00:14:23,000 Speaker 3: him for being too competitive and taking trivial things like 282 00:14:23,040 --> 00:14:25,280 Speaker 3: table football seriously, but he actually really feels it. 283 00:14:25,520 --> 00:14:28,760 Speaker 1: We talked about the sort of mushroom cloud of motivation. 284 00:14:29,440 --> 00:14:30,840 Speaker 1: One of the things that doesn't seem to be so 285 00:14:31,040 --> 00:14:34,040 Speaker 1: motivating to him is money. I mean, there's a story 286 00:14:34,080 --> 00:14:37,400 Speaker 1: about the offer as an eighteen year old of a 287 00:14:37,400 --> 00:14:41,880 Speaker 1: half million pounds to join game development studio right, which 288 00:14:41,880 --> 00:14:44,400 Speaker 1: he turned out right, despite coming from me. I know 289 00:14:44,440 --> 00:14:47,080 Speaker 1: his mother had gone through homelessness her in her youth. 290 00:14:47,120 --> 00:14:49,520 Speaker 1: I mean, was that a hard decision for him? Why 291 00:14:49,520 --> 00:14:50,120 Speaker 1: did he make it? 292 00:14:50,360 --> 00:14:52,720 Speaker 3: He said it was completely easy in today's money. It 293 00:14:52,840 --> 00:14:55,040 Speaker 3: was well over a million dollars that he was being offered. 294 00:14:55,440 --> 00:14:57,680 Speaker 3: He was eighteen. As you say, his parents were not rich. 295 00:14:58,200 --> 00:15:01,360 Speaker 3: I mean, you know, any self respecting sort of Stanford 296 00:15:01,440 --> 00:15:04,440 Speaker 3: character would at this point of fall, you know, taking 297 00:15:04,440 --> 00:15:07,280 Speaker 3: the money dropped out of Stanford and you know, written 298 00:15:07,320 --> 00:15:10,240 Speaker 3: off into the sunset with a loot. No, Demis is different. 299 00:15:10,280 --> 00:15:13,600 Speaker 3: Demis wanted to understand science. That was his primary motivation. 300 00:15:13,640 --> 00:15:15,400 Speaker 3: That's what he's up when he's up at two o'clock 301 00:15:15,440 --> 00:15:17,840 Speaker 3: in the morning. He's thinking, how do I understand nature? 302 00:15:18,240 --> 00:15:20,760 Speaker 3: And so he turned down the cash to go and 303 00:15:20,800 --> 00:15:22,240 Speaker 3: study computer science instead. 304 00:15:22,480 --> 00:15:26,360 Speaker 1: Fast forward a few years and he meets Peter Teel, 305 00:15:27,480 --> 00:15:31,480 Speaker 1: who gives him a A plus for science fiction and 306 00:15:31,520 --> 00:15:34,680 Speaker 1: an F for business model, but nonetheless a size skiins 307 00:15:34,680 --> 00:15:35,080 Speaker 1: some money. 308 00:15:36,320 --> 00:15:38,560 Speaker 3: Well, actually there's a fraudulent slip there. You said an 309 00:15:38,560 --> 00:15:40,520 Speaker 3: A plus for science fiction. I think you met an 310 00:15:40,520 --> 00:15:41,840 Speaker 3: A plus for science. 311 00:15:42,160 --> 00:15:45,160 Speaker 1: Science fiction, but science fiction. 312 00:15:45,360 --> 00:15:47,720 Speaker 3: Maybe that would have been better, because in fact Demis 313 00:15:47,800 --> 00:15:50,960 Speaker 3: was spinning this vision and this is twenty ten, right. 314 00:15:51,000 --> 00:15:54,080 Speaker 3: He was saying, I'm going to invent very powerful AI. 315 00:15:54,480 --> 00:15:58,280 Speaker 3: This is at a time when AI literally couldn't recognize 316 00:15:58,280 --> 00:16:01,960 Speaker 3: the photograph of a cat. Nothing was working, and you 317 00:16:01,960 --> 00:16:03,440 Speaker 3: have this character coming and say, oh, I'm going to 318 00:16:03,440 --> 00:16:06,680 Speaker 3: create artificial general intelligence. It was nuts. So it kind 319 00:16:06,720 --> 00:16:07,640 Speaker 3: of was science fiction. 320 00:16:08,320 --> 00:16:11,720 Speaker 1: What was his entree to the world of technology investors? 321 00:16:11,720 --> 00:16:13,920 Speaker 1: And when did deep Mind actually start as a company. 322 00:16:14,040 --> 00:16:16,400 Speaker 3: Deep Mind started in twenty ten, having raised the money 323 00:16:16,400 --> 00:16:20,000 Speaker 3: from Peter Teel. The entree is very interesting because in 324 00:16:20,000 --> 00:16:23,000 Speaker 3: fact what happened was, you know, Demis had done a 325 00:16:23,040 --> 00:16:26,080 Speaker 3: small games company before, and he made some money. Wasn't 326 00:16:26,080 --> 00:16:29,440 Speaker 3: a terrific success, but nonetheless it wasn't total failure. And 327 00:16:29,480 --> 00:16:31,400 Speaker 3: he went back to the same investors. They all said, 328 00:16:31,400 --> 00:16:34,240 Speaker 3: you must be joking. There's no product if you're doing 329 00:16:34,280 --> 00:16:36,680 Speaker 3: AI and not putting money into that. So then he 330 00:16:36,720 --> 00:16:39,800 Speaker 3: had to think again, and his entree into the world 331 00:16:39,800 --> 00:16:43,400 Speaker 3: of Peter Teel and what's called the Singularity summits, where 332 00:16:43,440 --> 00:16:46,960 Speaker 3: all these very early believers in AI would gather, people 333 00:16:47,000 --> 00:16:50,320 Speaker 3: like Ray Kurtzweil, and they would dream about a future 334 00:16:50,640 --> 00:16:54,120 Speaker 3: of an AI that totally did not exist. And when 335 00:16:54,160 --> 00:16:56,120 Speaker 3: they got up on the stage, actually they did often 336 00:16:56,320 --> 00:17:01,400 Speaker 3: draw more on science fiction novels than on science when 337 00:17:01,440 --> 00:17:04,159 Speaker 3: they were kind of imagining a future with AI. And 338 00:17:04,320 --> 00:17:08,480 Speaker 3: so in this strange cauldron of mythology and reality with 339 00:17:08,600 --> 00:17:12,760 Speaker 3: all kinds of weirdos trotting about, demis a Sabots, who 340 00:17:12,760 --> 00:17:15,000 Speaker 3: by this point has a computer science degree and a 341 00:17:15,000 --> 00:17:18,280 Speaker 3: PhD in neuroscience as a proper scientist, shows up and 342 00:17:18,320 --> 00:17:20,679 Speaker 3: he's asked by a journalist what do you think of 343 00:17:21,040 --> 00:17:24,720 Speaker 3: the Singularity conference? Are you a Singularitarian? And he says 344 00:17:25,400 --> 00:17:28,040 Speaker 3: it's a bit Californian for me, and he's, oh, you 345 00:17:28,040 --> 00:17:30,720 Speaker 3: could sort of feel the kind of anxiety of being 346 00:17:30,760 --> 00:17:32,960 Speaker 3: seen in this crowd. But that's why you had to 347 00:17:33,000 --> 00:17:35,360 Speaker 3: go to meet Peter tele And then when he met 348 00:17:35,400 --> 00:17:37,879 Speaker 3: Peter Til, he had this clever trick. Peter Teel is 349 00:17:37,880 --> 00:17:40,600 Speaker 3: a chess player. Demis is a chess player. So rather 350 00:17:40,640 --> 00:17:43,840 Speaker 3: than pitch Peter tele On some idea about a company, 351 00:17:43,840 --> 00:17:45,640 Speaker 3: and he said, well, I think the interesting thing about 352 00:17:45,760 --> 00:17:50,080 Speaker 3: chess is that the knight and the Bishop are supremely 353 00:17:50,080 --> 00:17:52,719 Speaker 3: well balanced, and it's in that tension between those two 354 00:17:52,760 --> 00:17:55,160 Speaker 3: pieces that much of the joy of the game resides. 355 00:17:55,720 --> 00:17:58,080 Speaker 3: So p t is like, WHOA, that's a conversation I 356 00:17:58,119 --> 00:18:00,760 Speaker 3: want to pursue. And so that got him and got 357 00:18:00,760 --> 00:18:03,640 Speaker 3: Demssus Habits an invitation to Peter Teel's house the next day, 358 00:18:04,200 --> 00:18:06,080 Speaker 3: and then that's when he pitched him on Deep Mind 359 00:18:06,119 --> 00:18:08,600 Speaker 3: and got the money he needed to start the company. 360 00:18:08,920 --> 00:18:14,840 Speaker 1: And then Fast forty twenty thirteen and which the excerpt 361 00:18:14,840 --> 00:18:16,600 Speaker 1: in the Wall Street Journal of your book tells the 362 00:18:16,640 --> 00:18:20,399 Speaker 1: story of a birthday party for Elon Musk, replete with 363 00:18:20,440 --> 00:18:24,040 Speaker 1: all kinds of costumes and strange things and fake battlements. 364 00:18:24,080 --> 00:18:28,920 Speaker 1: But this is, perhaps apart from the founding, the most 365 00:18:28,960 --> 00:18:31,879 Speaker 1: crucial moment in Deep Minds Genesis as a company. 366 00:18:31,960 --> 00:18:34,640 Speaker 3: Right, yeah, that's right. So you know, by this point, 367 00:18:35,240 --> 00:18:39,440 Speaker 3: Demis had raised three rounds of venture capital, including from 368 00:18:39,440 --> 00:18:42,000 Speaker 3: Elon Musk, and you know, there are various people could 369 00:18:42,040 --> 00:18:44,480 Speaker 3: come in, but it was a total pain in the neck. 370 00:18:44,520 --> 00:18:47,879 Speaker 3: He hated it. You know, he would sometimes have this expression, 371 00:18:48,119 --> 00:18:50,320 Speaker 3: I don't want this part of my brain to expand. 372 00:18:50,840 --> 00:18:53,840 Speaker 3: He wanted to be doing science, and so what he 373 00:18:53,920 --> 00:18:57,439 Speaker 3: wanted was to be liberated from this hamster wheel of fundraising. 374 00:18:58,280 --> 00:19:01,399 Speaker 3: And along at this party, this birthday party that Enol 375 00:19:01,480 --> 00:19:05,960 Speaker 3: Musk had, along comes Larry Page from Google who's also there, 376 00:19:06,680 --> 00:19:08,840 Speaker 3: and says, let's go for a walk, and they walk 377 00:19:08,880 --> 00:19:12,679 Speaker 3: around the castle grounds, and in this bizarre setting, Larry 378 00:19:12,680 --> 00:19:15,200 Speaker 3: Page says to him, well, you know, you could spend 379 00:19:15,200 --> 00:19:18,480 Speaker 3: your career building another company like Google. That's fine, but 380 00:19:18,560 --> 00:19:21,480 Speaker 3: if you really want to do science, just join Google 381 00:19:21,760 --> 00:19:25,479 Speaker 3: and we'll give you the resources, use our platform, and 382 00:19:25,520 --> 00:19:28,080 Speaker 3: you'll be able to do what you really love. And 383 00:19:28,359 --> 00:19:31,880 Speaker 3: Demis not only agreed with that pitch in the sense 384 00:19:31,920 --> 00:19:34,399 Speaker 3: that yes, he preferred to do science than to be 385 00:19:34,440 --> 00:19:37,840 Speaker 3: a billionaire, but he felt that Larry Page himself would 386 00:19:37,840 --> 00:19:41,560 Speaker 3: have accepted that pitch, that Larry Page cared about science. 387 00:19:41,560 --> 00:19:43,840 Speaker 3: He could have been a standard professor of computer science. 388 00:19:44,160 --> 00:19:47,840 Speaker 3: So Demis really identified with Larry Page, and that was 389 00:19:47,840 --> 00:19:48,840 Speaker 3: why he sold to Google. 390 00:19:48,920 --> 00:19:52,399 Speaker 1: And Paige had his eye on Demis or this was impulsive. 391 00:19:52,680 --> 00:19:55,479 Speaker 1: Had he planned out his chess game for this party 392 00:19:55,600 --> 00:19:57,040 Speaker 1: like Demis had three years before. 393 00:19:57,080 --> 00:19:59,320 Speaker 3: He had totally planned the chess game. He'd been thinking 394 00:19:59,359 --> 00:20:03,720 Speaker 3: for a while about buying up nascent AI companies, and 395 00:20:03,760 --> 00:20:07,359 Speaker 3: he'd bought the boutique founded by the Toronto professor Jeffrey 396 00:20:07,400 --> 00:20:11,360 Speaker 3: Hinton together with Iliasaskeva and one other person. Uh, and 397 00:20:11,680 --> 00:20:14,200 Speaker 3: so he was in a buying mode. 398 00:20:14,880 --> 00:20:16,719 Speaker 1: It came into that was twenty twelve, right the image 399 00:20:16,720 --> 00:20:17,880 Speaker 1: net team exactly. 400 00:20:17,880 --> 00:20:19,960 Speaker 3: He bought the image net team, and then the next 401 00:20:20,000 --> 00:20:23,800 Speaker 3: obvious person to buy was Demis and DeepMind because they 402 00:20:23,840 --> 00:20:27,520 Speaker 3: had a different approach to AI. It wasn't just deep learning, 403 00:20:28,080 --> 00:20:31,240 Speaker 3: which is the image net secret source, which is kind 404 00:20:31,240 --> 00:20:34,720 Speaker 3: of packet pattern recognition learning from data. It was also 405 00:20:34,760 --> 00:20:38,800 Speaker 3: what's called reinforcement learning, which is learning through trial and 406 00:20:38,920 --> 00:20:41,359 Speaker 3: error in a simulation. So you have a game like 407 00:20:41,400 --> 00:20:45,200 Speaker 3: the Atari games or go later on and you try 408 00:20:45,240 --> 00:20:47,520 Speaker 3: lots of different the computer trysts and moves ces whies 409 00:20:47,560 --> 00:20:50,200 Speaker 3: run works and then learns through trial and error, and 410 00:20:50,240 --> 00:20:53,159 Speaker 3: in some ways. Another strand in my book is the 411 00:20:53,560 --> 00:20:58,240 Speaker 3: interplay between deep learning on the one hand and reinforcement 412 00:20:58,320 --> 00:21:01,800 Speaker 3: learning on the other hand. And these two fields of 413 00:21:01,920 --> 00:21:05,640 Speaker 3: artificial intelligence, you know, have their different moments in the sun. 414 00:21:05,720 --> 00:21:07,359 Speaker 3: As the story progresses. 415 00:21:07,680 --> 00:21:09,720 Speaker 1: Hinton talked, He came on tech Stuff and talked about 416 00:21:09,760 --> 00:21:12,719 Speaker 1: how he ran an auction to sell image neet with Google, 417 00:21:12,800 --> 00:21:15,359 Speaker 1: Microsoft and by Doo. But in the end, all he 418 00:21:15,400 --> 00:21:19,320 Speaker 1: really wanted was to go to Google for for demis. 419 00:21:19,880 --> 00:21:23,240 Speaker 1: He was being courted as well by others, including a 420 00:21:23,280 --> 00:21:26,480 Speaker 1: dinner at Mark Zuckerberg's house in the I guess weeks 421 00:21:26,520 --> 00:21:29,320 Speaker 1: or months after this first meeting with Larry Page at 422 00:21:29,400 --> 00:21:35,000 Speaker 1: Elon Maas's birthday party, and he submitted Mark Zuckerberg to 423 00:21:35,040 --> 00:21:36,760 Speaker 1: a test at this dinner. 424 00:21:36,600 --> 00:21:41,200 Speaker 3: Right, Yeah, that's right. So the test was a bit subtle. Predictably, 425 00:21:41,280 --> 00:21:44,200 Speaker 3: they sit down to dinner and Mark Zuckerberg, who's longing 426 00:21:44,240 --> 00:21:47,280 Speaker 3: to buy deep Mind to get one over Google. 427 00:21:48,040 --> 00:21:50,240 Speaker 1: And this was not recently, this was ten years ago. 428 00:21:50,280 --> 00:21:55,000 Speaker 3: It was twenty thirteen. So Mark Zuckerberg says, well, I 429 00:21:55,000 --> 00:21:59,119 Speaker 3: think AI is the most important technology in human history. 430 00:21:59,160 --> 00:22:01,560 Speaker 3: It's extraordinary, and you know, I really hope you agree 431 00:22:01,600 --> 00:22:06,120 Speaker 3: to join me at Facebook because you know, we could 432 00:22:06,200 --> 00:22:08,480 Speaker 3: just do great things together. Blah blah blah blah. And 433 00:22:08,520 --> 00:22:10,800 Speaker 3: then you know, the conversation moves on, time goes by, 434 00:22:10,840 --> 00:22:14,200 Speaker 3: and then Demis slyly says, you know, three D printing 435 00:22:14,240 --> 00:22:17,760 Speaker 3: is extraordinary, and Zuckerberg goes, yeah, I agree, you know, 436 00:22:17,920 --> 00:22:20,200 Speaker 3: incredible that that's just going to unlock so many things. 437 00:22:20,200 --> 00:22:23,200 Speaker 3: And then a bit later, Demis says, you know, artificial reality, 438 00:22:23,840 --> 00:22:28,040 Speaker 3: that really is going to be transformative, and Zuckerberg's like, yeah, 439 00:22:28,040 --> 00:22:30,760 Speaker 3: it's transformative. It's so exciting. I'm so excited by that. 440 00:22:31,320 --> 00:22:33,600 Speaker 3: And then Demis's mind is whearing. Is said, Okay, he's 441 00:22:33,600 --> 00:22:36,960 Speaker 3: a bullshit artist. He does not believe that AI is 442 00:22:37,000 --> 00:22:38,959 Speaker 3: the most important thing ever, he does not get it. 443 00:22:39,040 --> 00:22:41,400 Speaker 3: I'm selling to Google, Forget forget. 444 00:22:41,040 --> 00:22:43,040 Speaker 1: Facebook, even though the more money on the table. 445 00:22:43,760 --> 00:22:45,840 Speaker 3: Yeah, that's right. In fact, Facebook was offering to make 446 00:22:45,840 --> 00:22:48,320 Speaker 3: Demis a lot richer, but he was consistent throughout his career. 447 00:22:48,400 --> 00:22:50,879 Speaker 3: Demis in turning down the money not to go to 448 00:22:51,040 --> 00:22:54,200 Speaker 3: Cambridge University, turning down the money to sell to Facebook. 449 00:22:54,400 --> 00:22:56,200 Speaker 3: It's not about the money for him. It's really about 450 00:22:56,200 --> 00:22:57,000 Speaker 3: the science. 451 00:22:56,640 --> 00:22:59,040 Speaker 1: And you mentioned him using his scientific method to see 452 00:22:59,040 --> 00:23:02,679 Speaker 1: two three years into the future. Instead, Facebook went with 453 00:23:02,960 --> 00:23:05,920 Speaker 1: Yan Lacun and gave him plenty of resources, and I 454 00:23:05,960 --> 00:23:08,280 Speaker 1: think he was trying to poach some of Dems's employees. 455 00:23:08,840 --> 00:23:10,840 Speaker 1: Demist told them that Google deal was going to happen, 456 00:23:10,960 --> 00:23:13,200 Speaker 1: to sit and therefore to sit tight, and they did. 457 00:23:13,240 --> 00:23:16,520 Speaker 1: But you know, fast forward to twenty twenty six and 458 00:23:16,880 --> 00:23:21,080 Speaker 1: Yanklan has sentually been dumped from from Meta and Demis 459 00:23:21,160 --> 00:23:24,000 Speaker 1: is where does he sit in Google? Is he is he? 460 00:23:24,119 --> 00:23:26,840 Speaker 1: Is he the successor to Sundar? Is he the you know, 461 00:23:26,880 --> 00:23:29,240 Speaker 1: the the ego and the ID? I mean, what is 462 00:23:29,240 --> 00:23:30,480 Speaker 1: his roll in Google today? 463 00:23:31,119 --> 00:23:33,640 Speaker 3: Well, what his rold is today is to be the 464 00:23:33,760 --> 00:23:36,439 Speaker 3: chief executive of Google Deep Mind, which is the AI 465 00:23:36,520 --> 00:23:39,159 Speaker 3: engine which is basically powering all the new products in Google. 466 00:23:39,720 --> 00:23:42,600 Speaker 3: So he's super important. Sundai is the chief executive of 467 00:23:42,600 --> 00:23:46,280 Speaker 3: Albret and Google, and I would argue that the relationship 468 00:23:46,320 --> 00:23:49,320 Speaker 3: between Sundar and Demis is the most important relationship in 469 00:23:49,400 --> 00:23:53,960 Speaker 3: business anywhere at the moment. Because Sundar has Demiss back, 470 00:23:54,720 --> 00:23:57,399 Speaker 3: Sundar gives him the resources, Sundar takes care of the 471 00:23:57,480 --> 00:24:00,439 Speaker 3: kind of all that kind of corporate leaderships staff that 472 00:24:00,440 --> 00:24:02,480 Speaker 3: Demis is good at, but it's really not what he 473 00:24:02,480 --> 00:24:05,919 Speaker 3: wants to do full time. And that gives Demis the 474 00:24:05,920 --> 00:24:09,800 Speaker 3: oxygen to pursue AI to the fullest of his abilities, 475 00:24:09,880 --> 00:24:13,399 Speaker 3: which are considerable you know, in the future, if Sundar 476 00:24:13,480 --> 00:24:15,359 Speaker 3: were to go, I don't think that's happening anytime soon, 477 00:24:15,440 --> 00:24:17,840 Speaker 3: by the way, but I think if he were to go, 478 00:24:17,920 --> 00:24:20,040 Speaker 3: you know, Demis would obviously be talked about as a candidate. 479 00:24:20,080 --> 00:24:23,000 Speaker 3: And it's a really interesting question because he is on 480 00:24:23,040 --> 00:24:27,320 Speaker 3: the one hand, somebody who is a leader, has vision, 481 00:24:27,560 --> 00:24:31,560 Speaker 3: can motivate people, would have the credibility to lead Google 482 00:24:32,240 --> 00:24:34,000 Speaker 3: as an AI company. I mean, how often do you 483 00:24:34,000 --> 00:24:36,800 Speaker 3: get somebody who's the CEO and also has a Nobel price. 484 00:24:36,840 --> 00:24:39,440 Speaker 3: That would be quite something. But at the same time, 485 00:24:39,440 --> 00:24:41,040 Speaker 3: Demis has a side to him that wants to be 486 00:24:41,119 --> 00:24:43,959 Speaker 3: a pure scientist that talks to me about you know, 487 00:24:44,200 --> 00:24:46,359 Speaker 3: there's too much noise in Silicon Valley. I want to 488 00:24:46,359 --> 00:24:49,440 Speaker 3: go and think I want to have a research professorship 489 00:24:49,480 --> 00:24:53,359 Speaker 3: at Princeton. That's where Oppenheimer went after the Manhattan Project. 490 00:24:53,560 --> 00:24:56,399 Speaker 3: That's where Einstein went, That's where I should be. You know, 491 00:24:56,440 --> 00:24:59,040 Speaker 3: he has that kind of you know, retreat to the 492 00:24:59,080 --> 00:25:03,560 Speaker 3: idyll of abstract contemplation side to him. And he's so 493 00:25:03,640 --> 00:25:06,040 Speaker 3: good at both of these things. It's what makes him exceptional. 494 00:25:06,040 --> 00:25:09,639 Speaker 3: I mean, if you mentioned Jan Lukun no, very good scientist, 495 00:25:09,760 --> 00:25:13,040 Speaker 3: but clearly not a great operator inside business. You know, 496 00:25:13,040 --> 00:25:15,920 Speaker 3: when we could talk about Simultman, a great business operator 497 00:25:16,080 --> 00:25:18,199 Speaker 3: but not a scientist, dropped out of Stanford, doesn't have 498 00:25:18,200 --> 00:25:21,720 Speaker 3: a degree. You know, it's very rare to find both 499 00:25:21,880 --> 00:25:22,800 Speaker 3: in the same person. 500 00:25:31,400 --> 00:25:35,680 Speaker 1: After the break is Demis an evil genius stay with us. 501 00:25:50,960 --> 00:25:55,520 Speaker 1: You mentioned earlier in the conversation this kind of journey 502 00:25:56,359 --> 00:26:01,280 Speaker 1: Demis had been on where one sort of safety mechanism 503 00:26:01,320 --> 00:26:05,760 Speaker 1: after another that he believed in fell away and thus 504 00:26:05,840 --> 00:26:11,480 Speaker 1: these kind of metaphors about the atomic bomb. But ironically, 505 00:26:12,280 --> 00:26:16,320 Speaker 1: in some sense, the kind of safety to the wayside 506 00:26:17,119 --> 00:26:21,399 Speaker 1: race that we're in today with Ai was kicked off 507 00:26:21,960 --> 00:26:24,639 Speaker 1: by Demis's desire for a safety board. 508 00:26:25,560 --> 00:26:28,479 Speaker 3: Yes, that's a good irony, you're right. So what happened 509 00:26:28,600 --> 00:26:31,119 Speaker 3: was that, you know, Demis sold the company to Google 510 00:26:31,359 --> 00:26:34,320 Speaker 3: in twenty fourteen, and one of the conditions was there 511 00:26:34,320 --> 00:26:37,360 Speaker 3: had to be a safety oversight board whereby Google would 512 00:26:37,359 --> 00:26:39,879 Speaker 3: allow Deep Mind to sort of appoint some you know, 513 00:26:39,960 --> 00:26:45,280 Speaker 3: important philosophers or other people of independent stature to make 514 00:26:45,320 --> 00:26:49,880 Speaker 3: a final decision on when AI would be deployed into 515 00:26:49,880 --> 00:26:52,159 Speaker 3: the world. And the idea was this is AI is 516 00:26:52,160 --> 00:26:54,199 Speaker 3: too big just to let the corporate board of Google 517 00:26:54,600 --> 00:26:56,040 Speaker 3: do whatever it wants with it. You know, there has 518 00:26:56,040 --> 00:26:59,320 Speaker 3: to be a check. So the first of these safety 519 00:26:59,359 --> 00:27:03,359 Speaker 3: meetings arranged and Demis had the idea, we'll invite Elon 520 00:27:03,480 --> 00:27:07,439 Speaker 3: Musk to chair it, and he invited Rieed Hoffmann and 521 00:27:07,520 --> 00:27:11,360 Speaker 3: various other people and they all met at SpaceX and 522 00:27:11,640 --> 00:27:15,520 Speaker 3: basically what happens. Elon Musk sat there listening, absorbed all 523 00:27:15,840 --> 00:27:18,840 Speaker 3: the presentations from deep Mind about their plans to build AI, 524 00:27:19,640 --> 00:27:23,720 Speaker 3: and a few months later he announces open Ai, which 525 00:27:23,760 --> 00:27:26,720 Speaker 3: is going to be the rival company. And so all 526 00:27:26,720 --> 00:27:29,679 Speaker 3: of a sudden, there's Singleton vision, the idea that you know, 527 00:27:29,760 --> 00:27:33,480 Speaker 3: only one AI lab would shepherd AI into the world 528 00:27:33,480 --> 00:27:37,879 Speaker 3: on behalf of all humanity. That just is by the wayside, 529 00:27:37,920 --> 00:27:40,960 Speaker 3: And you've now got two competing labs, and the race 530 00:27:41,040 --> 00:27:42,440 Speaker 3: dynamic begins to set in. 531 00:27:42,960 --> 00:27:45,240 Speaker 1: How did Demis feel about what Elon did? 532 00:27:45,520 --> 00:27:49,560 Speaker 3: Betrayed? Elon had sat there listening to all his plans, 533 00:27:50,640 --> 00:27:53,080 Speaker 3: and he'd been invited to chair that meeting in good 534 00:27:53,080 --> 00:27:56,320 Speaker 3: faith to ensure safety for the world, which, of course 535 00:27:56,400 --> 00:27:58,480 Speaker 3: is what at the time Elon was a big duma 536 00:27:59,160 --> 00:28:02,840 Speaker 3: and was constantly talking about AI safety and existential risk, 537 00:28:03,240 --> 00:28:07,320 Speaker 3: and so the idea that rather than uniting with Deep 538 00:28:07,359 --> 00:28:10,199 Speaker 3: Mind and Google in a single effort to make the 539 00:28:10,280 --> 00:28:14,080 Speaker 3: technology safe, Elon Musk preferred to go off and start 540 00:28:14,119 --> 00:28:17,159 Speaker 3: a rival in open AI to Demis. This was a 541 00:28:17,160 --> 00:28:20,880 Speaker 3: total betrayal. Of course, Elon thought of this as Demis 542 00:28:20,920 --> 00:28:26,200 Speaker 3: is dangerous, he's an evil genius, and therefore I need 543 00:28:26,240 --> 00:28:28,680 Speaker 3: to be the one because you know, all of these 544 00:28:28,720 --> 00:28:32,040 Speaker 3: actors they basically say, I know that I'm a good person. Yeah, 545 00:28:32,520 --> 00:28:34,800 Speaker 3: if I'm the leader of the AI race, I will 546 00:28:34,800 --> 00:28:37,440 Speaker 3: make it safe because I'm good. But those other guys 547 00:28:37,440 --> 00:28:41,320 Speaker 3: over there, you can't trust those guys because you know whatever. Now, 548 00:28:41,400 --> 00:28:44,760 Speaker 3: if you quizzed Elon Musk about why did he say 549 00:28:44,760 --> 00:28:47,800 Speaker 3: that Demis was an Elon was it was an evil genius? 550 00:28:47,880 --> 00:28:53,040 Speaker 1: Your term for a Freudian slip, Elon evil genius? 551 00:28:53,560 --> 00:28:56,160 Speaker 3: Why was demisis and evil genius? Well, the only good reason, 552 00:28:56,240 --> 00:28:59,400 Speaker 3: or not a good reason, but a reason was apparently Demis, 553 00:28:59,440 --> 00:29:01,960 Speaker 3: in his gay design days, had worked on a game 554 00:29:02,080 --> 00:29:06,040 Speaker 3: called Evil Genius, which is a pretty thin basis on 555 00:29:06,080 --> 00:29:08,400 Speaker 3: which to call him an evil genius, but whatever I mean, 556 00:29:09,480 --> 00:29:10,000 Speaker 3: they all. 557 00:29:09,920 --> 00:29:16,360 Speaker 1: Had association Sean Elbows. So then that this is twenty 558 00:29:16,960 --> 00:29:19,480 Speaker 1: this is meeting is in twenty sixteen, twenty fifteen, twenty fifteen, 559 00:29:19,520 --> 00:29:21,680 Speaker 1: and when is the Alpha Go moment? 560 00:29:21,880 --> 00:29:25,200 Speaker 3: Twenty sixteen? Okay, So coming out of that moment when 561 00:29:25,960 --> 00:29:31,520 Speaker 3: Elon Musk decides to set up open Ai, Demis decides, well, 562 00:29:31,600 --> 00:29:34,360 Speaker 3: I'm just going to accelerate as fast as possible. And 563 00:29:34,560 --> 00:29:37,560 Speaker 3: the first thing he manages to score is this victory 564 00:29:37,640 --> 00:29:42,120 Speaker 3: over the Korean Go champion, Lisa Dol And it's a 565 00:29:42,200 --> 00:29:45,880 Speaker 3: huge exhibition match in South Korea with all the media 566 00:29:46,040 --> 00:29:48,680 Speaker 3: in attendance, and it's kind of an it's not quite 567 00:29:48,720 --> 00:29:52,640 Speaker 3: chatchy pt but it's it's a moment when Ai had 568 00:29:52,680 --> 00:29:56,680 Speaker 3: what one might call the Kaspar of Deep Blue moment 569 00:29:57,240 --> 00:30:00,560 Speaker 3: in nineteen ninety seven, first time the human champion get defeated, 570 00:30:01,520 --> 00:30:04,800 Speaker 3: and then twenty sixteen, so that nineteen years later, the 571 00:30:04,800 --> 00:30:05,800 Speaker 3: same thing happens with Go. 572 00:30:06,240 --> 00:30:08,920 Speaker 1: And two hundred million people tune in and the defeated 573 00:30:09,000 --> 00:30:13,680 Speaker 1: Korean player apologizes to humanity. It's a huge moment, but 574 00:30:13,720 --> 00:30:16,240 Speaker 1: it's nothing like the chat Chipet moment six years later. 575 00:30:16,720 --> 00:30:21,720 Speaker 3: Yeah, because go people watched. Whereas chat ChiPT you used it. 576 00:30:21,720 --> 00:30:23,280 Speaker 3: It was personal, it was visceral. 577 00:30:23,880 --> 00:30:26,480 Speaker 1: And within a week of you pitching demish on the book, 578 00:30:26,720 --> 00:30:27,760 Speaker 1: Chatchipet came out. 579 00:30:27,880 --> 00:30:29,680 Speaker 3: That's right. And I went to see him right after that, 580 00:30:30,280 --> 00:30:33,920 Speaker 3: and he said, you know, this is war. Those guys 581 00:30:33,920 --> 00:30:37,800 Speaker 3: have parked their tanks in our front yard actually said 582 00:30:38,440 --> 00:30:42,000 Speaker 3: on our lawn but translating for American ordience in our 583 00:30:42,000 --> 00:30:46,840 Speaker 3: front yard. And so you could see that competitive glint 584 00:30:46,840 --> 00:30:48,960 Speaker 3: in his eye, and you knew he was going to 585 00:30:48,960 --> 00:30:49,760 Speaker 3: try and fight back. 586 00:30:50,160 --> 00:30:52,680 Speaker 1: Was he self aware about the risk of using that 587 00:30:52,800 --> 00:30:58,120 Speaker 1: language even for himself given all these Manhattan Project analogies. 588 00:30:57,880 --> 00:31:02,240 Speaker 3: You know, he's a person with many different dimensions, and 589 00:31:02,400 --> 00:31:05,400 Speaker 3: he's both capable of worrying about safety and also using 590 00:31:05,480 --> 00:31:09,640 Speaker 3: military metaphors to express this determination to crush the opposition. 591 00:31:10,240 --> 00:31:11,600 Speaker 3: And I think actually it's going to be a business 592 00:31:11,600 --> 00:31:16,320 Speaker 3: school case study of how DeepMind made the comeback because 593 00:31:16,640 --> 00:31:20,560 Speaker 3: they merged deep Mind the London Lab with Google Brain, 594 00:31:20,760 --> 00:31:25,080 Speaker 3: the Mountain View, Google AI lab. Normally, mergers are super difficult, 595 00:31:25,080 --> 00:31:28,080 Speaker 3: they don't work. And here was a merger you had 596 00:31:28,120 --> 00:31:30,800 Speaker 3: to do in the middle of an AI race which 597 00:31:30,800 --> 00:31:33,680 Speaker 3: had been kicked off by chatchapt You had eight time 598 00:31:33,800 --> 00:31:36,960 Speaker 3: zones between California and London. You had a record of 599 00:31:37,040 --> 00:31:40,400 Speaker 3: bitter rivalry between the AI scientists from Google and the 600 00:31:40,400 --> 00:31:43,000 Speaker 3: ones from deep Mind. And yet they pulled it off. 601 00:31:43,040 --> 00:31:45,640 Speaker 3: They did the merger, they blended the cultures, and within 602 00:31:45,680 --> 00:31:47,520 Speaker 3: two and a half years they had a model that 603 00:31:47,720 --> 00:31:49,520 Speaker 3: was outclassing open AI models. 604 00:31:49,840 --> 00:31:54,280 Speaker 1: See that's his extraordinary I remember when the chetchipt moment happened, 605 00:31:54,480 --> 00:31:58,520 Speaker 1: and I would say up until twenty twenty beginning of 606 00:31:58,560 --> 00:32:00,800 Speaker 1: twenty twenty five, people were saying Google is down and 607 00:32:00,840 --> 00:32:04,520 Speaker 1: out Google might be over. I mean, you knew because 608 00:32:04,520 --> 00:32:06,600 Speaker 1: you were reporting along the way that probably wasn't true. 609 00:32:06,640 --> 00:32:09,160 Speaker 1: But what the what the indications that you saw that 610 00:32:09,280 --> 00:32:11,760 Speaker 1: the rest of the world didn't that convinced you along 611 00:32:11,800 --> 00:32:14,920 Speaker 1: the way that that Demis and deep Mind might might 612 00:32:14,960 --> 00:32:17,440 Speaker 1: be roaring back into do you put them in the 613 00:32:17,440 --> 00:32:17,880 Speaker 1: first place? 614 00:32:17,960 --> 00:32:20,440 Speaker 3: Now, I think it's sort of a pretty close race 615 00:32:20,560 --> 00:32:24,800 Speaker 3: between the Gemini model from Demis and then Claude is 616 00:32:24,840 --> 00:32:27,360 Speaker 3: doing really well, at the moment, the anthropic model. People 617 00:32:27,400 --> 00:32:30,000 Speaker 3: love it for coding and so forth, So you know, 618 00:32:30,680 --> 00:32:32,680 Speaker 3: I'm not sure that it's I think the race is 619 00:32:32,680 --> 00:32:35,840 Speaker 3: still ongoing. What I would say, though, is that you know, 620 00:32:35,880 --> 00:32:38,600 Speaker 3: I'm on record as having written in The New York Times. 621 00:32:38,320 --> 00:32:39,880 Speaker 1: That are around of money. 622 00:32:39,920 --> 00:32:42,360 Speaker 3: Right, probably run ound of money. I mean, they may 623 00:32:42,400 --> 00:32:45,080 Speaker 3: put it out, but basically, in fact, since I wrote 624 00:32:45,080 --> 00:32:49,160 Speaker 3: that piece, they do seem to have focused their business 625 00:32:49,200 --> 00:32:52,160 Speaker 3: quite a bit by giving up on Soora for example, 626 00:32:52,240 --> 00:32:55,280 Speaker 3: Sura was a classic money losing idea. You know, it 627 00:32:55,320 --> 00:32:58,720 Speaker 3: costs enormous amounts to generate video, but people don't pay 628 00:32:58,760 --> 00:33:01,840 Speaker 3: you to generate, so quite rightly, they can do it. 629 00:33:02,440 --> 00:33:05,440 Speaker 3: So maybe they can cut costs enough to survive. But 630 00:33:06,920 --> 00:33:10,040 Speaker 3: they have huge cash need and they do not have 631 00:33:10,080 --> 00:33:12,280 Speaker 3: Google's deep pockets behind them, unlike Demos. 632 00:33:12,440 --> 00:33:14,760 Speaker 1: So Demis is kind of winning. But he said to you, 633 00:33:14,840 --> 00:33:17,800 Speaker 1: it doesn't necessarily feel like that, right, he said, this 634 00:33:17,880 --> 00:33:20,640 Speaker 1: is a paradoxical moment. It should feel amazing, but it 635 00:33:20,680 --> 00:33:22,520 Speaker 1: doesn't feel how I thought it would feel. 636 00:33:23,000 --> 00:33:26,120 Speaker 3: Yeah, because early on he had this rather naive idea 637 00:33:26,120 --> 00:33:28,760 Speaker 3: that there would be one lab building AI and so 638 00:33:28,800 --> 00:33:30,960 Speaker 3: you could take your time about releasing the models, and 639 00:33:31,480 --> 00:33:33,280 Speaker 3: you know, if you were worried about safety, you could 640 00:33:33,320 --> 00:33:35,800 Speaker 3: just take another six months to test them. And now 641 00:33:35,840 --> 00:33:38,480 Speaker 3: you have this race, and you know, the Chinese have 642 00:33:38,560 --> 00:33:40,680 Speaker 3: plenty of models, and the other thing, it's not just 643 00:33:40,720 --> 00:33:43,240 Speaker 3: a race, it's actually also the open source nature of 644 00:33:43,280 --> 00:33:46,480 Speaker 3: these models, where they're being released out into the wild, 645 00:33:47,320 --> 00:33:51,480 Speaker 3: and some weird group can just download the model, have 646 00:33:51,560 --> 00:33:54,120 Speaker 3: it on their own computer, and then you can't pull 647 00:33:54,160 --> 00:33:56,640 Speaker 3: it back anymore. And so there was a big cyber 648 00:33:56,680 --> 00:34:00,760 Speaker 3: attack in Mexico recently where all of the actual records 649 00:34:00,760 --> 00:34:04,680 Speaker 3: were stolen, and Anthropic realized that its clawed model was 650 00:34:04,720 --> 00:34:08,399 Speaker 3: being used. But because that model is proprietary, they could 651 00:34:08,400 --> 00:34:11,800 Speaker 3: immediately shut off access and stop the attack. You couldn't 652 00:34:11,840 --> 00:34:14,359 Speaker 3: do that with an open weight, open source model. And 653 00:34:14,440 --> 00:34:17,560 Speaker 3: yet we have open weight, you know, that's being put 654 00:34:17,600 --> 00:34:20,279 Speaker 3: out there, both by Meta and by the Chinese and 655 00:34:20,320 --> 00:34:23,560 Speaker 3: by Mestride in France. A lot of open source models 656 00:34:23,600 --> 00:34:27,360 Speaker 3: are out there, and so in many ways the way 657 00:34:27,440 --> 00:34:32,080 Speaker 3: AI is being deployed is frightening. The obvious safety measures 658 00:34:32,080 --> 00:34:35,000 Speaker 3: one might take are not happening. In addition to banning 659 00:34:35,040 --> 00:34:38,319 Speaker 3: open source, I think there should be much more powerful 660 00:34:38,880 --> 00:34:42,640 Speaker 3: sort of government oversight, so that, just like with a pharmaceutical, 661 00:34:42,680 --> 00:34:45,200 Speaker 3: before you release it to be used in people, has 662 00:34:45,239 --> 00:34:47,680 Speaker 3: to go through clinical trials. So too, I think there 663 00:34:47,719 --> 00:34:49,239 Speaker 3: should be a sort of equivalent of the Food and 664 00:34:49,280 --> 00:34:53,960 Speaker 3: Drug Administration, an AI agency that can actually veto the 665 00:34:54,000 --> 00:34:57,160 Speaker 3: release of really powerful models. And we don't have that, 666 00:34:57,400 --> 00:34:59,680 Speaker 3: and we should have that, and we should be negotiating 667 00:34:59,719 --> 00:35:01,600 Speaker 3: with Chin about doing it in both places at once, 668 00:35:02,080 --> 00:35:04,160 Speaker 3: because this is a global race and both sides have 669 00:35:04,200 --> 00:35:07,200 Speaker 3: to slow down. I was in China recently for eight 670 00:35:07,280 --> 00:35:11,960 Speaker 3: days because they always published books faster, so I was 671 00:35:12,320 --> 00:35:17,239 Speaker 3: meeting AI leaders, both from industry and from academia, and 672 00:35:17,280 --> 00:35:19,440 Speaker 3: I was surprised by how much they do talk about safety. 673 00:35:20,160 --> 00:35:22,200 Speaker 3: So I think there is a discussion to be had 674 00:35:22,200 --> 00:35:26,960 Speaker 3: with the Chinese about safety, but the US administration of 675 00:35:27,000 --> 00:35:28,360 Speaker 3: this moment doesn't want to do that. 676 00:35:28,800 --> 00:35:33,200 Speaker 1: I mean, coming back to the Manhattan Project again, Demis 677 00:35:33,239 --> 00:35:35,600 Speaker 1: has said I think that he thinks this may end 678 00:35:35,640 --> 00:35:37,920 Speaker 1: in a bunker And what does he mean by that? 679 00:35:37,960 --> 00:35:44,480 Speaker 1: And has he primed himself psychologically for an ai Hiroshima 680 00:35:44,520 --> 00:35:46,960 Speaker 1: that he may feel in some sense responsible for. 681 00:35:47,600 --> 00:35:50,600 Speaker 3: Yeah, I mean, when I was doing the research interfering, 682 00:35:50,600 --> 00:35:53,080 Speaker 3: not just Demis, but all the scientists that he works with, 683 00:35:53,120 --> 00:35:55,319 Speaker 3: you know, one hundred or something of them. In Deep Mind. 684 00:35:55,760 --> 00:35:58,239 Speaker 3: I would hear these references to the bunker come up, 685 00:35:58,280 --> 00:36:02,120 Speaker 3: and I assumed it wasn't literally, you know, a real thing. 686 00:36:02,160 --> 00:36:04,799 Speaker 3: That Demis wanted to disappear into a bunker at the 687 00:36:04,800 --> 00:36:07,760 Speaker 3: moment when he thought the AI models were coming dangerously powerful. 688 00:36:08,400 --> 00:36:10,680 Speaker 3: And I would have these dinners every six months with 689 00:36:11,440 --> 00:36:14,239 Speaker 3: a friend who had been at Deep Mind but had left, 690 00:36:15,160 --> 00:36:18,799 Speaker 3: and I tested this on him one evening and I said, yeah, 691 00:36:18,800 --> 00:36:21,480 Speaker 3: surely this is just a metaphor bunker. He can't be serious. 692 00:36:22,160 --> 00:36:25,239 Speaker 3: And this guy said, well, actually, you know, I had 693 00:36:25,239 --> 00:36:28,719 Speaker 3: my bag packed. It was serious. There was actually this 694 00:36:28,840 --> 00:36:33,359 Speaker 3: vision that AI would become so powerful that bad guys 695 00:36:33,400 --> 00:36:35,319 Speaker 3: would try and get it off you. So you had 696 00:36:35,320 --> 00:36:37,480 Speaker 3: to hide in some place a bit like Los Animals 697 00:36:38,440 --> 00:36:42,360 Speaker 3: and develop in seat of isolation and secret, and also 698 00:36:42,360 --> 00:36:45,400 Speaker 3: be isolated because you needed maximum focus on the science 699 00:36:45,440 --> 00:36:47,600 Speaker 3: to get it right when you were at this moment 700 00:36:47,640 --> 00:36:50,320 Speaker 3: of maximum danger because the model was suddenly very powerful. 701 00:36:51,320 --> 00:36:53,840 Speaker 3: And that was his vision. Now I think today he 702 00:36:53,880 --> 00:36:57,560 Speaker 3: doesn't believe that anymore, because we're so far from a 703 00:36:57,600 --> 00:37:02,600 Speaker 3: single lab, you know, midwifing AI. So I think now 704 00:37:02,680 --> 00:37:06,360 Speaker 3: he's more inclined to speak of some version of the 705 00:37:06,400 --> 00:37:09,680 Speaker 3: Center for European Nuclear Research SERAN, which is a sort 706 00:37:09,680 --> 00:37:14,960 Speaker 3: of technical agency that oversees nuclear power on a multinational basis. 707 00:37:15,040 --> 00:37:17,920 Speaker 3: I think he would like some sort of global body 708 00:37:18,840 --> 00:37:21,960 Speaker 3: to impose rules on what kind of AI should be 709 00:37:22,040 --> 00:37:25,560 Speaker 3: let out into the wild. But you know, at the 710 00:37:25,560 --> 00:37:28,280 Speaker 3: same time, he knows that politically that's not on the cards, 711 00:37:28,840 --> 00:37:31,920 Speaker 3: and he has a sense of timing about when you 712 00:37:31,920 --> 00:37:36,759 Speaker 3: should raise these issues, and so you know, whereas Dariama 713 00:37:36,840 --> 00:37:40,560 Speaker 3: Day took on the Pentagon by trying to assert safety 714 00:37:40,560 --> 00:37:43,719 Speaker 3: principles and then just got rolled, I think their miss 715 00:37:43,760 --> 00:37:46,359 Speaker 3: when he does that, is going to feel that he's 716 00:37:46,400 --> 00:37:49,040 Speaker 3: got the door is half open, and he can give 717 00:37:49,040 --> 00:37:51,200 Speaker 3: it a push and we'll see. You know, of course, 718 00:37:51,200 --> 00:37:54,799 Speaker 3: sometimes people keep their capital drive for so long that 719 00:37:54,840 --> 00:37:57,240 Speaker 3: they never use it. But we'll see if the moment 720 00:37:57,239 --> 00:37:59,719 Speaker 3: comes when he does use it, it'll be very interesting. Bess. 721 00:37:59,920 --> 00:38:01,920 Speaker 1: Just close. There was a great review of your book 722 00:38:02,239 --> 00:38:06,719 Speaker 1: in the Financial Times which ends with this, Whether and 723 00:38:06,840 --> 00:38:11,680 Speaker 1: how Demis ever achieves AGI will form the defining chapters 724 00:38:12,040 --> 00:38:16,080 Speaker 1: of his extraordinary and unfinished biography. What did you think 725 00:38:16,080 --> 00:38:18,279 Speaker 1: about that? And what is the next chapter for him? 726 00:38:18,280 --> 00:38:19,560 Speaker 1: And will you write another follow up book? 727 00:38:19,560 --> 00:38:20,279 Speaker 2: Do you think? You know? 728 00:38:20,320 --> 00:38:22,319 Speaker 3: I tend not to write follow ups about the same thing, 729 00:38:22,400 --> 00:38:25,120 Speaker 3: the same person. I prefer to plower of fresh ground. 730 00:38:25,800 --> 00:38:28,440 Speaker 3: But look, I mean, you know, Demis is turming fifty 731 00:38:28,520 --> 00:38:31,719 Speaker 3: this year. He's got a lot of runway. I'm sure 732 00:38:31,760 --> 00:38:34,840 Speaker 3: he'll do more incredible things in the future. So probably 733 00:38:34,840 --> 00:38:38,279 Speaker 3: I am offering an interim report. But the advantage you know, 734 00:38:38,320 --> 00:38:40,360 Speaker 3: if you wait, I did this before with Alan Greenspann. 735 00:38:40,400 --> 00:38:43,680 Speaker 3: I wrote the definitive biography after he retired, and by 736 00:38:43,680 --> 00:38:46,520 Speaker 3: that time, you know, people are interested, but less so 737 00:38:46,640 --> 00:38:49,200 Speaker 3: than when he's stood in the seat. I think capturing 738 00:38:49,320 --> 00:38:52,719 Speaker 3: a portrait of you know, the most interesting figure in 739 00:38:52,800 --> 00:38:55,400 Speaker 3: artificial intelligence in real time while he's still in the 740 00:38:55,400 --> 00:38:57,799 Speaker 3: seat and he's still doing it is sometimes the fun 741 00:38:57,840 --> 00:39:00,360 Speaker 3: of it, right, I mean, who wants to for the 742 00:39:00,400 --> 00:39:03,719 Speaker 3: definitive biography in twenty years time? But well, next for him, 743 00:39:03,760 --> 00:39:05,680 Speaker 3: you know, I think he's going to carry on running 744 00:39:05,840 --> 00:39:09,080 Speaker 3: Google Deep Mind. There's going to be more agentic models 745 00:39:09,080 --> 00:39:11,800 Speaker 3: coming out this year. There will be you know, world 746 00:39:11,880 --> 00:39:16,120 Speaker 3: models and more robotics coming, there will probably be much 747 00:39:16,160 --> 00:39:19,239 Speaker 3: more AI for science, both in terms of drug discovery 748 00:39:19,280 --> 00:39:22,640 Speaker 3: and in terms of you know, material sciences, chemistry and 749 00:39:22,640 --> 00:39:26,200 Speaker 3: so forth. So I think, you know, one day, I remember, 750 00:39:26,480 --> 00:39:29,959 Speaker 3: towards the end of my time interviewing him, he showed 751 00:39:30,040 --> 00:39:31,920 Speaker 3: up at the pub and he had a backpack and 752 00:39:31,960 --> 00:39:34,960 Speaker 3: he'd pished something out of it, and he got this 753 00:39:35,000 --> 00:39:37,279 Speaker 3: little box out and he said, I got to show 754 00:39:37,320 --> 00:39:39,960 Speaker 3: you this, and he opened the box up and inside 755 00:39:40,080 --> 00:39:44,600 Speaker 3: was the Nobel Prize medal, And either at that meeting 756 00:39:44,800 --> 00:39:47,359 Speaker 3: or another one, he said to me, I wonder if 757 00:39:47,400 --> 00:39:49,640 Speaker 3: I can get another one. He's not over yet. 758 00:39:50,560 --> 00:39:51,560 Speaker 1: It's a fasting amount of me. 759 00:39:51,640 --> 00:40:09,280 Speaker 3: Thank you. It's been great fun to talk for tech Stuff. 760 00:40:09,320 --> 00:40:10,200 Speaker 2: I'm oz Voloscian. 761 00:40:10,640 --> 00:40:13,560 Speaker 1: This episode was produced by Eliza Dennis and Melissa Slaughter. 762 00:40:14,200 --> 00:40:17,080 Speaker 1: It was executive produced by me Julia Nutter and Kate 763 00:40:17,080 --> 00:40:21,520 Speaker 1: Osborne for Kaleidoscope and Katrina Norvell for iHeart Podcasts. The 764 00:40:21,600 --> 00:40:24,719 Speaker 1: engineer is Paul Bowman and Jack Insley makes this episode. 765 00:40:25,160 --> 00:40:26,759 Speaker 2: Kyle Murdoch wrote our theme song. 766 00:40:27,440 --> 00:40:29,839 Speaker 1: Please rate, review and reach out to us at tech 767 00:40:29,880 --> 00:40:30,919 Speaker 1: Stuff podcast at. 768 00:40:30,840 --> 00:40:32,080 Speaker 2: Gmail dot com. 769 00:40:32,239 --> 00:40:33,840 Speaker 1: We also love to hear what you think our panels 770 00:40:33,880 --> 00:40:34,799 Speaker 1: should cover next time