1 00:00:02,000 --> 00:00:07,000 2 00:00:02,654 --> 00:00:04,656 [grand orchestral fanfare playing] 3 00:00:08,000 --> 00:00:13,000 4 00:00:27,636 --> 00:00:29,681 ♪ ♪ 5 00:00:41,389 --> 00:00:43,217 [birds chirping and calling] 6 00:00:48,657 --> 00:00:50,833 -[burbling] -[trilling] 7 00:00:55,055 --> 00:00:57,057 [VCR clicking and whirring] 8 00:01:00,582 --> 00:01:02,018 [projector clicking] 9 00:01:04,847 --> 00:01:09,286 Now, the present-day computers are complete morons, 10 00:01:09,330 --> 00:01:12,115 but this will not be true in another generation. 11 00:01:12,159 --> 00:01:14,857 They will start to think, and eventually they will 12 00:01:14,900 --> 00:01:17,599 completely outthink their makers. 13 00:01:17,642 --> 00:01:20,254 Is this depressing? I don't see why it should be. 14 00:01:20,297 --> 00:01:24,823 I suspect that organic or biological evolution 15 00:01:24,867 --> 00:01:26,303 has about come to its end. 16 00:01:27,739 --> 00:01:29,524 [Daniel chuckles] 17 00:01:29,567 --> 00:01:30,655 CAROLINE: You don't want me to be the narrator. 18 00:01:30,699 --> 00:01:32,353 I don't have a good voice. 19 00:01:32,396 --> 00:01:34,006 DANIEL: You have a great voice. Just do it. 20 00:01:34,050 --> 00:01:36,183 It's just a-- we're just trying it out. 21 00:01:36,226 --> 00:01:37,967 CAROLINE: Do I get paid for this shit? 22 00:01:38,010 --> 00:01:40,230 DANIEL [laughing]: Well, we'll see how well you do. 23 00:01:40,274 --> 00:01:41,579 [Caroline laughing] 24 00:01:41,623 --> 00:01:43,451 -♪ ♪ -[dog barks] 25 00:01:46,454 --> 00:01:47,890 CAROLINE: I need to call my parents. 26 00:01:47,933 --> 00:01:49,065 [engine starts] 27 00:01:49,109 --> 00:01:51,023 Sometimes we rush into things 28 00:01:51,067 --> 00:01:52,851 without thinking them through. 29 00:01:52,895 --> 00:01:54,723 DANIEL: Yeah, can you sh... you want to shoot? That's... 30 00:01:54,766 --> 00:01:57,204 CAROLINE: Like when Daniel and Caroline got married 31 00:01:57,247 --> 00:02:00,163 151 days after they met. 32 00:02:00,207 --> 00:02:01,817 Let's rewind. 33 00:02:03,079 --> 00:02:04,776 [applause] 34 00:02:04,820 --> 00:02:07,170 Have you been thinking about buying a home computer? 35 00:02:07,214 --> 00:02:09,303 CAROLINE: In 1993, when Daniel was born, 36 00:02:09,346 --> 00:02:12,132 his parents didn't even have a computer in the house. 37 00:02:12,175 --> 00:02:13,785 But he remembers when they finally got one. 38 00:02:13,829 --> 00:02:15,700 What a computer is to me is 39 00:02:15,744 --> 00:02:18,355 the equivalent of a bicycle for our minds. 40 00:02:18,399 --> 00:02:20,227 Hello. 41 00:02:20,270 --> 00:02:22,185 CAROLINE: His computer helped unleash his creativity. 42 00:02:22,229 --> 00:02:23,926 DANIEL: Rolling. Go. 43 00:02:23,969 --> 00:02:26,581 CAROLINE: He used his dad's iMac to edit videos. 44 00:02:26,624 --> 00:02:28,322 Yeah! 45 00:02:28,365 --> 00:02:29,932 CAROLINE: And even make little animations. 46 00:02:29,975 --> 00:02:31,934 [tires squealing, crash] 47 00:02:31,977 --> 00:02:35,067 In 1998, when Caroline was a little girl... 48 00:02:35,111 --> 00:02:36,721 Hi, Mommy. 49 00:02:36,765 --> 00:02:38,158 ...only nerds knew what the Internet was. 50 00:02:38,201 --> 00:02:39,811 And what about this Internet thing? 51 00:02:39,855 --> 00:02:41,335 -Do you, do you know anything about that? -Sure. 52 00:02:41,378 --> 00:02:42,771 -[audience laughter] -CAROLINE: But soon, 53 00:02:42,814 --> 00:02:44,903 -everyone was on the... -...Internets. 54 00:02:44,947 --> 00:02:46,992 ...and computers were beating humans at chess. 55 00:02:47,036 --> 00:02:49,995 ANNOUNCER: Deep Blue-- Kasparov has resigned! 56 00:02:50,039 --> 00:02:52,607 CAROLINE: And by the time Caroline went to college, 57 00:02:52,650 --> 00:02:55,087 computers were in everyone's pocket. 58 00:02:55,131 --> 00:02:58,395 Now Daniel could make movies with his phone. 59 00:02:58,439 --> 00:03:01,485 He grew up to be an artist and a filmmaker... 60 00:03:01,529 --> 00:03:03,922 -Good night. -...and so did Caroline. 61 00:03:03,966 --> 00:03:05,707 Cut. Let's go into a close-up. That was perfect. 62 00:03:05,750 --> 00:03:07,274 By then, computers were connecting people 63 00:03:07,317 --> 00:03:08,971 and changing the world in ways 64 00:03:09,014 --> 00:03:10,668 that we never could have imagined. 65 00:03:10,712 --> 00:03:12,322 All the money raised by the Ice Bucket Challenge... 66 00:03:12,366 --> 00:03:14,194 CAROLINE: Some of it was good. 67 00:03:14,237 --> 00:03:15,499 [gasping] 68 00:03:15,543 --> 00:03:17,109 Some not so good. 69 00:03:17,153 --> 00:03:19,286 Anxiety, self-harm, suicide... 70 00:03:19,329 --> 00:03:21,201 But it's clear now that we didn't do enough 71 00:03:21,244 --> 00:03:24,552 to prevent these tools from being used for harm as well. 72 00:03:24,595 --> 00:03:26,380 CAROLINE: And as the world got more and more focused 73 00:03:26,423 --> 00:03:28,033 -on their computers, -[phone dings] 74 00:03:28,077 --> 00:03:30,210 Daniel focused on his artwork. 75 00:03:32,037 --> 00:03:35,606 Wherever he went, he always had a sketchbook, 76 00:03:35,650 --> 00:03:38,392 including the night he met Caroline. 77 00:03:38,435 --> 00:03:41,656 He drew a terrible picture of her, and 20 minutes later, 78 00:03:41,699 --> 00:03:44,136 he boldly pronounced they were going to get married, 79 00:03:44,180 --> 00:03:47,575 which, as you know, they did, 151 days later. 80 00:03:47,618 --> 00:03:49,925 WOMAN: ...now pronounce you husband and wife. 81 00:03:49,968 --> 00:03:51,666 CAROLINE: They moved into a cute house... 82 00:03:51,709 --> 00:03:52,841 -DANIEL: Moose, hey! Come here! -...and got a dog 83 00:03:52,884 --> 00:03:54,669 as dumb as Daniel. 84 00:03:54,712 --> 00:03:57,280 Meanwhile, computers were writing entire essays 85 00:03:57,324 --> 00:03:59,326 based on simple questions like, 86 00:03:59,369 --> 00:04:03,155 "How hard would it be to build a shack in my backyard?" 87 00:04:03,199 --> 00:04:07,072 And so Daniel built a shack in his backyard. 88 00:04:07,116 --> 00:04:09,945 But just as he sat down to start working on his next movie, 89 00:04:09,988 --> 00:04:13,644 he learned that computers could now write screenplays. 90 00:04:14,863 --> 00:04:16,560 I mean, they were bad, 91 00:04:16,604 --> 00:04:18,954 but they were getting better really fast. 92 00:04:18,997 --> 00:04:21,652 They could create new images and videos from scratch, 93 00:04:21,696 --> 00:04:24,786 and some of them could even pass the bar exam. 94 00:04:24,829 --> 00:04:28,833 Not just pass the bar but be in the top ten percentile. 95 00:04:28,877 --> 00:04:30,531 CAROLINE: It was confusing. 96 00:04:30,574 --> 00:04:32,141 [sighs] 97 00:04:32,184 --> 00:04:35,057 Daniel just wanted a bicycle for his mind, 98 00:04:35,100 --> 00:04:38,626 but computers had become a self-driving rocket ship. 99 00:04:38,669 --> 00:04:40,236 [whooshing] 100 00:04:40,280 --> 00:04:41,977 ["We R in Control" by Neil Young playing] 101 00:04:42,020 --> 00:04:44,762 Pioneers in the field of artificial intelligence 102 00:04:44,806 --> 00:04:46,503 are pleading with Congress 103 00:04:46,547 --> 00:04:48,505 to pass safety rules before it's too late. 104 00:04:48,549 --> 00:04:50,551 CAROLINE: Now it felt like the whole world 105 00:04:50,594 --> 00:04:52,988 was rushing into something without thinking it through, 106 00:04:53,031 --> 00:04:56,339 and everyone had an opinion. 107 00:04:56,383 --> 00:05:00,125 Was it gonna destroy the world or save humanity? 108 00:05:00,169 --> 00:05:02,389 There was too much information, 109 00:05:02,432 --> 00:05:04,652 which made him anxious, which made Caroline anxious... 110 00:05:04,695 --> 00:05:06,001 [exasperated scream] 111 00:05:06,044 --> 00:05:07,785 I don't have kids yet, but I worry 112 00:05:07,829 --> 00:05:10,179 what the world would look like if I decide to. [chuckles] 113 00:05:10,222 --> 00:05:12,355 CAROLINE: He was starting to spin out. 114 00:05:12,399 --> 00:05:14,052 [indistinct chatter] 115 00:05:14,096 --> 00:05:16,925 It was becoming a mountain of anxiety. 116 00:05:19,144 --> 00:05:21,364 -MAN: It's horrible. -It's freakin' Siri. 117 00:05:22,670 --> 00:05:25,586 CAROLINE: And so, Daniel decided to go out 118 00:05:25,629 --> 00:05:27,979 and find someone who could explain this to him, 119 00:05:28,023 --> 00:05:30,155 so he could stop thinking about it 120 00:05:30,199 --> 00:05:33,463 and he and Caroline could get on with their lives. 121 00:05:33,507 --> 00:05:36,814 ♪ We are in control, we are in control ♪ 122 00:05:36,858 --> 00:05:38,729 ♪ We are in control ♪ 123 00:05:38,773 --> 00:05:40,514 ♪ Chemical computer thinking battery ♪ 124 00:05:40,557 --> 00:05:42,559 ♪ We are in control, we are... ♪ 125 00:05:42,603 --> 00:05:45,693 This endeavor would turn out to be hopelessly naive 126 00:05:45,736 --> 00:05:49,044 and kick off the most confusing year of his life. 127 00:05:49,087 --> 00:05:50,698 ♪ We are in control... ♪ 128 00:05:50,741 --> 00:05:52,787 But as we know, we sometimes rush into things 129 00:05:52,830 --> 00:05:54,615 without thinking them through. 130 00:05:54,658 --> 00:05:56,356 ♪ Chemical computer thinking battery... ♪ 131 00:05:56,399 --> 00:05:57,922 Oh, my gosh. What is happening? 132 00:05:57,966 --> 00:05:59,924 CAROLINE: He had questions. 133 00:05:59,968 --> 00:06:01,578 Questions only the smartest nerds could answer. 134 00:06:01,622 --> 00:06:04,146 Submitting to the interrogator. 135 00:06:04,189 --> 00:06:06,017 CAROLINE: Questions like: 136 00:06:06,061 --> 00:06:08,585 Was this the apocalypse or did he actually have reason 137 00:06:08,629 --> 00:06:11,109 -to be optimistic? -Yes. 138 00:06:11,153 --> 00:06:13,764 -♪ Chemical computer thinking battery. ♪ -[song ends] 139 00:06:13,808 --> 00:06:16,158 CAROLINE: Uh, is that good? 140 00:06:16,201 --> 00:06:17,594 DANIEL: Yeah. 141 00:06:17,638 --> 00:06:20,162 So, to begin, 142 00:06:20,205 --> 00:06:22,120 what is artificial intelligence? 143 00:06:22,164 --> 00:06:24,688 I know that must be annoying for you, that-that question, 144 00:06:24,732 --> 00:06:26,473 but I do think it's important. 145 00:06:26,516 --> 00:06:29,780 So... AI... 146 00:06:29,824 --> 00:06:31,826 [stammers] Um... 147 00:06:31,869 --> 00:06:33,654 -Yeah... -Uh, hmm. 148 00:06:33,697 --> 00:06:34,742 [laughs]: That's a good question. 149 00:06:34,785 --> 00:06:36,178 -Yeah, um... -Um... 150 00:06:36,221 --> 00:06:37,658 DANIEL: What is AI? 151 00:06:37,701 --> 00:06:39,094 [laughs] 152 00:06:39,137 --> 00:06:40,748 I love that that's the first question, 153 00:06:40,791 --> 00:06:43,141 'cause there is not a clear and consistent answer. 154 00:06:43,185 --> 00:06:45,274 Artificial intelligence is 155 00:06:45,317 --> 00:06:50,714 a kind of intentionally and maybe uselessly broad term. 156 00:06:50,758 --> 00:06:53,325 DAVID EVAN HARRIS: It's a machine 157 00:06:53,369 --> 00:06:55,545 doing things that we previously only thought 158 00:06:55,589 --> 00:06:57,242 that people could do: 159 00:06:57,286 --> 00:07:01,116 making recommendations, decisions and predictions. 160 00:07:01,159 --> 00:07:06,774 AI is the, uh, application of computer science 161 00:07:06,817 --> 00:07:09,124 to solving cognitive problems. 162 00:07:09,167 --> 00:07:13,302 Okay, so when I picture AI, 163 00:07:13,345 --> 00:07:16,740 it's sort of like this magical computer box... 164 00:07:16,784 --> 00:07:18,786 ♪ ♪ 165 00:07:20,527 --> 00:07:24,095 ...just, like, floating in, like, inert space. 166 00:07:24,139 --> 00:07:26,446 And no matter how many times people try 167 00:07:26,489 --> 00:07:29,710 and explain this to me, I just don't get how... 168 00:07:29,753 --> 00:07:32,495 how it's understanding all of these things 169 00:07:32,539 --> 00:07:35,672 and how it's feeling like intelligence. 170 00:07:35,716 --> 00:07:37,152 [beeping, electronic chattering] 171 00:07:37,195 --> 00:07:39,241 And that's kind of nerve-racking. 172 00:07:41,243 --> 00:07:45,552 AZA RASKIN: What is this new generation of AI? 173 00:07:47,205 --> 00:07:50,078 This AI that is different than every other generation? 174 00:07:50,121 --> 00:07:52,733 Like, no one ever talked about, like, 175 00:07:52,776 --> 00:07:56,911 Siri taking over the world or causing catastrophes. 176 00:07:56,954 --> 00:07:58,478 -WOMAN: Hi, Siri? -MAN: ...want to. 177 00:07:58,521 --> 00:08:00,523 WOMAN: Hello? Siri? 178 00:08:00,567 --> 00:08:02,569 Hello? Hey, Siri! 179 00:08:02,612 --> 00:08:05,746 Or the voice in Google Maps, which mispronounces road names, 180 00:08:05,789 --> 00:08:07,487 like, breaking society. 181 00:08:07,530 --> 00:08:10,620 MAN: Google Maps says this is a road. 182 00:08:10,664 --> 00:08:13,057 [laughing]: But I think I'm in a river. 183 00:08:13,101 --> 00:08:14,624 This is definitely a river, innit? 184 00:08:14,668 --> 00:08:17,192 Something changed with ChatGPT coming out. 185 00:08:18,193 --> 00:08:19,934 People understood-- no, no, no-- 186 00:08:19,977 --> 00:08:21,457 this technology's insanely valuable, 187 00:08:21,501 --> 00:08:23,633 it's insanely powerful and also insanely scary. 188 00:08:23,677 --> 00:08:25,548 Okay, listen to this. Very creepy. 189 00:08:25,592 --> 00:08:28,595 A new artificial intelligence tool is going viral 190 00:08:28,638 --> 00:08:32,294 for cranking out entire essays in a matter of seconds. 191 00:08:35,950 --> 00:08:41,172 AI dwarfs the power of all other technologies combined. 192 00:08:41,216 --> 00:08:43,697 DANIEL: "AI dwarfs the power 193 00:08:43,740 --> 00:08:46,047 of all other technologies combined." 194 00:08:46,090 --> 00:08:47,222 Yeah. 195 00:08:47,265 --> 00:08:48,702 Do you think that's true? 196 00:08:48,745 --> 00:08:50,747 Yes. 197 00:08:50,791 --> 00:08:53,184 Tell me about-- How? How? 198 00:08:53,228 --> 00:08:56,710 I think, to paint that picture, it's really important 199 00:08:56,753 --> 00:09:00,627 to understand what today's state-of-the-art systems 200 00:09:00,670 --> 00:09:02,150 look like and how they're built. 201 00:09:02,193 --> 00:09:05,762 [laughs] This is quite a... quite a setup. 202 00:09:05,806 --> 00:09:07,242 So, one thing that 203 00:09:07,285 --> 00:09:09,679 not a lot of people realize is that 204 00:09:09,723 --> 00:09:13,683 systems like ChatGPT aren't programmed by any human. 205 00:09:13,727 --> 00:09:15,206 What do you mean? 206 00:09:15,250 --> 00:09:16,991 Instead, it's something like th-they're grown. 207 00:09:17,034 --> 00:09:19,167 We kind of give them raw resources, like, 208 00:09:19,210 --> 00:09:20,995 "Here's a lot of computational resources. 209 00:09:21,038 --> 00:09:22,170 Here's a lot of data." 210 00:09:22,213 --> 00:09:23,867 Under the hood, it's math, 211 00:09:23,911 --> 00:09:27,305 and the math is actually surprisingly straightforward. 212 00:09:27,349 --> 00:09:31,092 DANIEL: So ChatGPT is a kind of AI but it's not all of AI? 213 00:09:31,135 --> 00:09:32,659 Totally. ChatGPT is just the beginning, 214 00:09:32,702 --> 00:09:34,095 but it's a good place to start. 215 00:09:34,138 --> 00:09:37,794 But I still don't know what AI is. 216 00:09:37,838 --> 00:09:40,362 To understand AI, it begins with understanding 217 00:09:40,405 --> 00:09:43,147 that intelligence is about recognizing patterns. 218 00:09:43,191 --> 00:09:44,845 -Patterns. -Patterns. -Patterns. 219 00:09:44,888 --> 00:09:48,544 COTRA: It is shown trillions of words of text 220 00:09:48,588 --> 00:09:51,286 across millions of documents in the Internet. 221 00:09:51,329 --> 00:09:52,766 RANDIMA FERNANDO: It started with text. 222 00:09:52,809 --> 00:09:55,899 And what they did was they took textbooks, 223 00:09:55,943 --> 00:09:59,511 and they took poems and essays and instruction manuals. 224 00:09:59,555 --> 00:10:00,861 DAVID EVAN HARRIS: They can do things like 225 00:10:00,904 --> 00:10:03,124 digest the entire Internet, 226 00:10:03,167 --> 00:10:06,736 every single word that's ever been written by a person. 227 00:10:06,780 --> 00:10:09,957 Reddit threads and social media and all of Wikipedia. 228 00:10:10,000 --> 00:10:13,613 More data than anybody could ever read in several lifetimes. 229 00:10:13,656 --> 00:10:15,484 And they gave this system one job. 230 00:10:15,527 --> 00:10:18,487 Figure out the patterns and structure of that information 231 00:10:18,530 --> 00:10:20,750 and use that to make predictions about 232 00:10:20,794 --> 00:10:23,231 what word should come next in a sentence. 233 00:10:23,274 --> 00:10:24,972 When you say "patterns in a sentence," 234 00:10:25,015 --> 00:10:26,756 what are you talking about? 235 00:10:26,800 --> 00:10:29,367 So, it's everything from, like, the really simple things, 236 00:10:29,411 --> 00:10:31,979 like most sentences end with a period, 237 00:10:32,022 --> 00:10:34,895 all the way up to the more conceptual things, like: 238 00:10:34,938 --> 00:10:36,766 What is a sonnet? 239 00:10:36,810 --> 00:10:38,333 It's a type of poem, and it has some particular structure. 240 00:10:39,421 --> 00:10:41,249 RASKIN: So, it then looks at 241 00:10:41,292 --> 00:10:43,599 all of that data, all of that text... 242 00:10:43,643 --> 00:10:46,036 COTRA: And over trillions and trillions of tries, 243 00:10:46,080 --> 00:10:48,778 each time it gets something right or wrong, 244 00:10:48,822 --> 00:10:51,259 it's given a little bit of positive reinforcement 245 00:10:51,302 --> 00:10:53,391 when it guesses the next word correctly, 246 00:10:53,435 --> 00:10:55,698 and it's given a little bit of negative reinforcement 247 00:10:55,742 --> 00:10:59,354 when it guesses the next word incorrectly. 248 00:10:59,397 --> 00:11:01,573 And at the end of it, you have a system that 249 00:11:01,617 --> 00:11:05,055 speaks really good English as a side effect 250 00:11:05,099 --> 00:11:06,796 of being really, really good at predicting 251 00:11:06,840 --> 00:11:09,059 the word that comes next in a piece of text. 252 00:11:09,103 --> 00:11:12,019 AUTOMATED VOICE [in French]: Actually, my French isn't too bad either. 253 00:11:12,062 --> 00:11:14,021 [in English]: It uses all of those patterns it has learned 254 00:11:14,064 --> 00:11:15,675 to be able to make a prediction about 255 00:11:15,718 --> 00:11:17,241 what the answer should be, 256 00:11:17,285 --> 00:11:18,808 then it gives you that as the output. 257 00:11:18,852 --> 00:11:21,028 AUTOMATED VOICE: It's a little oversimplified, 258 00:11:21,071 --> 00:11:23,073 but I think people will get it. 259 00:11:23,117 --> 00:11:24,945 So, so that's all it does? 260 00:11:24,988 --> 00:11:26,816 Yeah. It doesn't seem like it would be that complicated, 261 00:11:26,860 --> 00:11:28,035 but actually you have to know 262 00:11:28,078 --> 00:11:29,645 a huge amount of things 263 00:11:29,689 --> 00:11:31,386 in order to actually succeed at that. 264 00:11:32,517 --> 00:11:34,258 RASKIN: If you say to ChatGPT, 265 00:11:34,302 --> 00:11:36,696 "Write me a Shakespearean sonnet about my dog," 266 00:11:36,739 --> 00:11:38,785 it has to know what dogs are. 267 00:11:38,828 --> 00:11:40,438 -It has to know what you love about your dog.-[barking] 268 00:11:40,482 --> 00:11:43,006 It has to know who Shakespeare is, 269 00:11:43,050 --> 00:11:45,574 that sonnets rhyme, that they have a structure, 270 00:11:45,617 --> 00:11:47,750 that words have sounds that can rhyme. 271 00:11:47,794 --> 00:11:49,273 It takes a lot. 272 00:11:49,317 --> 00:11:50,884 [voices overlapping in various languages] 273 00:11:50,927 --> 00:11:52,929 Holy shit, you can talk to your computer now. 274 00:11:52,973 --> 00:11:56,063 That was just not true three years ago. 275 00:11:56,106 --> 00:11:57,629 RASKIN: Yes, and this is the really important part. 276 00:11:57,673 --> 00:12:00,589 The same process that lets AI 277 00:12:00,632 --> 00:12:03,157 uncover and manipulate the patterns of text 278 00:12:03,200 --> 00:12:05,768 is the same process that lets it uncover 279 00:12:05,812 --> 00:12:10,512 the patterns of the entire universe and everything in it. 280 00:12:10,555 --> 00:12:12,862 FERNANDO: There are patterns and images and sound 281 00:12:12,906 --> 00:12:15,952 in computer code and DNA and music and physics 282 00:12:15,996 --> 00:12:18,476 and fashion and building design 283 00:12:18,520 --> 00:12:21,566 [distorted]: and in human voices and human faces, 284 00:12:21,610 --> 00:12:23,917 [normally]: really, truly everywhere. 285 00:12:23,960 --> 00:12:26,006 -Everywhere. -Everywhere. -Everywhere. -Everywhere. 286 00:12:26,049 --> 00:12:27,747 If you have learned those patterns, 287 00:12:27,790 --> 00:12:29,661 you can generate new kinds of songs. 288 00:12:29,705 --> 00:12:31,141 You can generate new videos. 289 00:12:31,185 --> 00:12:32,621 And that's why, if you give it 290 00:12:32,664 --> 00:12:34,710 a three-second recording of your grandmother, 291 00:12:34,754 --> 00:12:36,538 it can speak back in her voice. 292 00:12:36,581 --> 00:12:37,931 Oh, my God. 293 00:12:37,974 --> 00:12:40,585 [repeating]: Oh, my God. Oh, my God. 294 00:12:40,629 --> 00:12:42,849 What will they think of next? 295 00:12:42,892 --> 00:12:44,024 It's moving very, very quickly. 296 00:12:44,067 --> 00:12:45,808 An American AI start-up 297 00:12:45,852 --> 00:12:47,592 has released its latest model. 298 00:12:47,636 --> 00:12:50,770 That company is Anthropic, and it has just unveiled 299 00:12:50,813 --> 00:12:53,773 the latest versions of its AI assistant Claude. 300 00:12:53,816 --> 00:12:57,167 GLENN BECK: So, the xAI team was there to unveil Grok 4. 301 00:12:57,211 --> 00:13:00,127 MAN: Google released one just last week. Gemini is... 302 00:13:00,170 --> 00:13:02,999 We've gone from GPT-2 just a couple years ago, 303 00:13:03,043 --> 00:13:05,523 which could barely write a coherent paragraph, 304 00:13:05,567 --> 00:13:07,961 to GPT-4, which can pass the bar exam. 305 00:13:08,004 --> 00:13:10,354 And all they had to do to get there 306 00:13:10,398 --> 00:13:12,966 was essentially add more data and more compute. 307 00:13:13,009 --> 00:13:16,273 -These people who are building this... -COTRA: Yeah. 308 00:13:16,317 --> 00:13:18,014 ...they're just throwing more... 309 00:13:18,058 --> 00:13:21,801 More physical computers, more of the same kinds of data. 310 00:13:21,844 --> 00:13:24,542 Because the more computing power you add, 311 00:13:24,586 --> 00:13:28,024 the more complex intellectual tasks they can do. 312 00:13:28,068 --> 00:13:30,157 So, the more weather data you give it, 313 00:13:30,200 --> 00:13:31,898 the better it can make predictions about 314 00:13:31,941 --> 00:13:34,204 where a hurricane might go. 315 00:13:34,248 --> 00:13:36,337 RASKIN: And the more patterns of tumors and bones 316 00:13:36,380 --> 00:13:38,861 and tissues an AI has seen, then the better able it is 317 00:13:38,905 --> 00:13:41,951 to detect a tumor in a new CT scan. 318 00:13:41,995 --> 00:13:43,648 Better even than a human doctor. 319 00:13:43,692 --> 00:13:46,956 AI that's already being deployed for the military 320 00:13:47,000 --> 00:13:49,132 can already use satellite imagery, 321 00:13:49,176 --> 00:13:51,787 troop movements, communications to determine, 322 00:13:51,831 --> 00:13:53,354 -sometimes days in advance, -[civil defense siren blaring] 323 00:13:53,397 --> 00:13:55,008 where an attack is going to happen, 324 00:13:55,051 --> 00:13:57,401 like where an enemy is going to strike. 325 00:13:57,445 --> 00:13:59,447 TRISTAN HARRIS: This whole space is moving so fast 326 00:13:59,490 --> 00:14:01,666 that any example you put in this movie 327 00:14:01,710 --> 00:14:04,582 will feel absolutely clumsy by the time it comes out. 328 00:14:04,626 --> 00:14:06,671 [beeping, clicking] 329 00:14:09,326 --> 00:14:11,198 -[applause] -[upbeat music playing] 330 00:14:11,241 --> 00:14:14,114 RASKIN: These models are being released 331 00:14:14,157 --> 00:14:17,073 before anyone knows what they're even capable of. 332 00:14:17,117 --> 00:14:22,513 GPT-3.5 was released and out to 100 million people plus 333 00:14:22,557 --> 00:14:25,952 before some researchers discovered that it could do 334 00:14:25,995 --> 00:14:28,737 research-grade chemistry better than models 335 00:14:28,780 --> 00:14:32,784 that were trained specifically to do research-grade chemistry. 336 00:14:32,828 --> 00:14:35,439 Something is happening in there that the people 337 00:14:35,483 --> 00:14:38,268 who are building them don't fully understand. 338 00:14:38,312 --> 00:14:40,967 Basically, it just analyzes the data by itself, 339 00:14:41,010 --> 00:14:45,058 and as it does that, it just teaches itself 340 00:14:45,101 --> 00:14:47,974 various things that we often didn't intend. 341 00:14:48,017 --> 00:14:49,932 So, for instance, it reads a lot online, 342 00:14:49,976 --> 00:14:52,543 and then at some point, it just learns how to do arithmetic. 343 00:14:52,587 --> 00:14:54,458 KIDS: One, two. 344 00:14:54,502 --> 00:14:56,112 HENDRYCKS: And then at some point, it starts to learn 345 00:14:56,156 --> 00:14:58,549 how to answer advanced physics questions. 346 00:14:58,593 --> 00:15:00,769 We didn't program that in it whatsoever. 347 00:15:00,812 --> 00:15:02,423 It just learned by itself. 348 00:15:05,382 --> 00:15:07,254 TRISTAN HARRIS: An AI is like a digital brain. 349 00:15:07,297 --> 00:15:08,690 But just like a human brain, 350 00:15:08,733 --> 00:15:10,518 if you did a brain scan on a human brain, 351 00:15:10,561 --> 00:15:13,086 would you know everything that person was capable of? 352 00:15:13,129 --> 00:15:15,131 You can't know that just from the brain scan. 353 00:15:15,175 --> 00:15:17,917 It's just, like, a bunch of numbers and, like, 354 00:15:17,960 --> 00:15:20,180 the multiplications that are happening that-that, like, 355 00:15:20,223 --> 00:15:22,225 the best machine learning researcher in the world 356 00:15:22,269 --> 00:15:24,314 could look at and, like, have no idea what was happening. 357 00:15:30,973 --> 00:15:32,714 DANIEL: That chair right there. 358 00:15:32,757 --> 00:15:35,847 -Is that okay for you? -Yes. 359 00:15:35,891 --> 00:15:37,675 DANIEL: So, that's kind of mind-boggling. 360 00:15:37,719 --> 00:15:39,068 Okay? Like, it's taking over the world, 361 00:15:39,112 --> 00:15:41,070 and we don't even know how it works. 362 00:15:41,114 --> 00:15:43,116 -Is that right? -Mm. 363 00:15:43,159 --> 00:15:46,946 We do understand a number of important things, 364 00:15:46,989 --> 00:15:50,471 but we don't have a very good grasp on 365 00:15:50,514 --> 00:15:53,213 why they provide specific answers to questions. 366 00:15:53,256 --> 00:15:58,609 It is a problem because we are on a path t-to build machines, 367 00:15:58,653 --> 00:16:03,092 based on these principles, that could be smarter than us 368 00:16:03,136 --> 00:16:06,139 and thus potentially have a lot of power. 369 00:16:12,580 --> 00:16:15,278 One of the most cited computer scientists in history. 370 00:16:15,322 --> 00:16:16,932 SUTSKEVER: I actually find it a little difficult 371 00:16:16,976 --> 00:16:18,499 to talk about my own role. 372 00:16:18,542 --> 00:16:21,067 Really much prefer when other people do it. 373 00:16:21,110 --> 00:16:23,895 Ilya joining was the... was-was the linchpin for 374 00:16:23,939 --> 00:16:25,506 OpenAI being ultimately successful. 375 00:16:25,549 --> 00:16:27,203 I think it's just going to be 376 00:16:27,247 --> 00:16:30,598 some kind of a vast, dramatic and unimaginable impact. 377 00:16:30,641 --> 00:16:32,861 I don't know if you've spent any time on YouTube, 378 00:16:32,904 --> 00:16:34,950 but you can kind of feel the speed already, right? 379 00:16:34,994 --> 00:16:36,256 You know what I mean? 380 00:16:36,299 --> 00:16:38,345 SHANE LEGG: This is just the warmup. 381 00:16:38,388 --> 00:16:40,434 The really powerful systems are still coming, 382 00:16:40,477 --> 00:16:42,262 and they're gonna be coming quite soon. 383 00:16:47,963 --> 00:16:51,575 AGI stands for "artificial general intelligence." 384 00:16:51,619 --> 00:16:53,795 Uh, systems that are basically... 385 00:16:58,191 --> 00:17:00,236 And this is, like, you know, seems to be, like, 386 00:17:00,280 --> 00:17:03,022 the holy grail of AI? 387 00:17:03,065 --> 00:17:05,154 When you can simulate a human mind 388 00:17:05,198 --> 00:17:07,852 that is doing human cognition and can do reasoning, 389 00:17:07,896 --> 00:17:10,942 that is a new sort of tier of AI 390 00:17:10,986 --> 00:17:14,250 that we have to distinguish from previous AI. 391 00:17:14,294 --> 00:17:16,600 When that happens, by the way, that's when 392 00:17:16,644 --> 00:17:21,040 you would hire one of those AGIs instead of a person. 393 00:17:21,083 --> 00:17:24,478 Most jobs in our economy it can do. 394 00:17:24,521 --> 00:17:25,914 It can work 24 hours a day, 395 00:17:25,957 --> 00:17:28,177 never gets tired, never gets bored. 396 00:17:28,221 --> 00:17:30,136 They don't need to sleep. They don't need breaks. 397 00:17:30,179 --> 00:17:31,876 They're, like, not gonna join a union. 398 00:17:31,920 --> 00:17:33,878 Won't complain, won't whistleblow. 399 00:17:33,922 --> 00:17:35,793 More than 100 times cheaper 400 00:17:35,837 --> 00:17:38,231 than humans working at m-minimum wage. 401 00:17:38,274 --> 00:17:39,493 Not only will they be doing everything, 402 00:17:39,536 --> 00:17:41,016 but they'll be doing it faster. 403 00:17:41,060 --> 00:17:43,062 TRISTAN HARRIS: The same intelligence 404 00:17:43,105 --> 00:17:45,368 that powers that can also look at the patterns and movements 405 00:17:45,412 --> 00:17:47,849 and articulating muscles and, you know, robotics. 406 00:17:47,892 --> 00:17:50,069 And so it's not just gonna automate desk jobs. 407 00:17:50,112 --> 00:17:51,722 That's just the beginning. 408 00:17:51,766 --> 00:17:55,552 It will automate all physical labor. 409 00:17:55,596 --> 00:17:56,988 LEIKE: There's no way 410 00:17:57,032 --> 00:17:59,165 humans are gonna compete with them. 411 00:17:59,208 --> 00:18:01,210 ♪ ♪ 412 00:18:05,606 --> 00:18:09,262 It is hard to conceptualize the impact of AGI. 413 00:18:09,305 --> 00:18:11,046 [electronic chatter] 414 00:18:11,090 --> 00:18:12,830 But I think it's going to be something very big 415 00:18:12,874 --> 00:18:15,181 and drastic and radical. 416 00:18:17,226 --> 00:18:19,010 DANIEL: You think this is one of the most 417 00:18:19,054 --> 00:18:20,882 consequential moments in human history? 418 00:18:20,925 --> 00:18:23,319 Yeah. Yeah, that's-- I mean, what else would be? 419 00:18:23,363 --> 00:18:25,452 I mean, like, there's the Industrial Revolution. 420 00:18:25,495 --> 00:18:27,193 MALO BOURGON: You know, it'll make the Industrial Revolution 421 00:18:27,236 --> 00:18:29,847 look like small beans. 422 00:18:29,891 --> 00:18:32,111 TRISTAN HARRIS: AGI is an inflection point 423 00:18:32,154 --> 00:18:34,025 because it means you can accelerate 424 00:18:34,069 --> 00:18:37,986 all other intellectual fields all at the same time. 425 00:18:38,029 --> 00:18:39,814 Like, if you make an advance in rocketry, 426 00:18:39,857 --> 00:18:42,033 that doesn't advance biology and medicine. 427 00:18:43,905 --> 00:18:45,515 If you make an advance in medicine, 428 00:18:45,559 --> 00:18:47,126 that doesn't advance rocketry. 429 00:18:48,866 --> 00:18:51,304 But if you make an advance in artificial intelligence, 430 00:18:51,347 --> 00:18:54,133 that advances all scientific and technological fields 431 00:18:54,176 --> 00:18:55,612 all at the same time. 432 00:18:55,656 --> 00:18:57,092 That's why, for a long time, 433 00:18:57,136 --> 00:18:59,181 Google DeepMind's mission statement was... 434 00:18:59,225 --> 00:19:00,965 -Step one, solve intelligence. -LEX FRIDMAN: Yeah. 435 00:19:01,009 --> 00:19:02,750 -Step two, use it to solve everything else. -Yes. 436 00:19:02,793 --> 00:19:05,144 That's why AI dwarfs the power 437 00:19:05,187 --> 00:19:07,494 of all other technologies combined. 438 00:19:07,537 --> 00:19:09,060 It will transform everything. 439 00:19:09,104 --> 00:19:10,845 So, uh, it'll be at least as big as 440 00:19:10,888 --> 00:19:13,761 the Industrial Revolution, possibly, you know, bigger, 441 00:19:13,804 --> 00:19:16,285 more like the advent of electricity or even fire. 442 00:19:16,329 --> 00:19:18,635 VIDEO NARRATOR: The caveman literally held aloft 443 00:19:18,679 --> 00:19:21,508 the torch of civilization. 444 00:19:23,727 --> 00:19:25,425 KOKOTAJLO: It is generally thought that 445 00:19:25,468 --> 00:19:28,167 around the time of AGI, we'll have AIs that can 446 00:19:28,210 --> 00:19:31,518 do all or most of the AI research process 447 00:19:31,561 --> 00:19:34,434 and, of course, can do it faster and cheaper. 448 00:19:34,477 --> 00:19:35,913 [beeping] 449 00:19:35,957 --> 00:19:37,306 LEIKE: It can copy itself. 450 00:19:37,350 --> 00:19:39,352 A thousand times, a million times, 451 00:19:39,395 --> 00:19:41,049 and, like, now you have a million copies 452 00:19:41,092 --> 00:19:43,573 all working in parallel. 453 00:19:43,617 --> 00:19:46,576 RASKIN: When it learns how to make its code faster, 454 00:19:46,620 --> 00:19:48,665 make its code more efficient, 455 00:19:48,709 --> 00:19:51,059 obviously that becomes, like, a-a runaway loop. 456 00:19:51,102 --> 00:19:53,192 AGI isn't, like, the end. 457 00:19:53,235 --> 00:19:54,932 It's just the beginning. 458 00:19:54,976 --> 00:19:57,979 It's the beginning of an incredibly rapid explosion 459 00:19:58,022 --> 00:20:00,024 of scientific progress, 460 00:20:00,068 --> 00:20:02,201 and in particular, scientific progress in AI. 461 00:20:02,244 --> 00:20:04,377 And when they're smarter than us, too, 462 00:20:04,420 --> 00:20:06,248 and substantially faster than us, 463 00:20:06,292 --> 00:20:08,685 and they're getting faster each year, exponentially, 464 00:20:08,729 --> 00:20:12,123 those are the ones that can potentially become superhuman, 465 00:20:12,167 --> 00:20:14,343 uh, possibly this decade. 466 00:20:14,387 --> 00:20:15,866 Sorry, did you say 467 00:20:15,910 --> 00:20:18,478 "become superhuman, maybe in this decade"? 468 00:20:18,521 --> 00:20:22,003 Yeah. I mean, I think, uh, a lot of people who are 469 00:20:22,046 --> 00:20:24,484 actually building this think that that's fairly plausible 470 00:20:24,527 --> 00:20:26,529 that we get some superintelligence, 471 00:20:26,573 --> 00:20:28,662 something that's vastly more intelligent than people, 472 00:20:28,705 --> 00:20:30,751 within this decade. 473 00:20:30,794 --> 00:20:32,883 The way I define "superintelligence" is 474 00:20:32,927 --> 00:20:35,234 a system that by itself is 475 00:20:35,277 --> 00:20:37,627 more intelligent and competent than all of humanity. 476 00:20:37,671 --> 00:20:39,629 I'm just gonna-- sorry. I don't mean to interrupt you. 477 00:20:39,673 --> 00:20:41,065 You're on a flow. 478 00:20:41,109 --> 00:20:43,111 Uh, I just, I just... 479 00:20:43,154 --> 00:20:45,113 I'm not really following, 'cause you're using language 480 00:20:45,156 --> 00:20:46,767 like "superintelligence" and, like, 481 00:20:46,810 --> 00:20:49,900 "smarter than all of humanity," and I hear that, 482 00:20:49,944 --> 00:20:52,425 and it sounds like... like sci-fi bullshit to me, 483 00:20:52,468 --> 00:20:54,383 and I'm just trying to understand. 484 00:20:54,427 --> 00:20:56,603 There's nothing magical about intelligence. 485 00:20:56,646 --> 00:20:58,431 This is very important, is that, you know, 486 00:20:58,474 --> 00:21:00,476 intelligence can feel magical, it can feel like 487 00:21:00,520 --> 00:21:03,262 some mystical thing in your mind or something, 488 00:21:03,305 --> 00:21:05,786 but it is just computation. 489 00:21:05,829 --> 00:21:09,790 LEGG: The human brain is quite limited in some ways, 490 00:21:09,833 --> 00:21:12,314 in terms of information processing capability, 491 00:21:12,358 --> 00:21:15,143 compared to what we see in, say, a data center. 492 00:21:15,186 --> 00:21:18,277 So, for example, the signals which are sent 493 00:21:18,320 --> 00:21:22,542 inside your brain, they move at about 30 meters per second. 494 00:21:22,585 --> 00:21:24,239 -[thunder cracks] -But the speed of light, 495 00:21:24,283 --> 00:21:27,198 which is what a computer uses in fiber optics, 496 00:21:27,242 --> 00:21:30,506 is 300 million meters per second. 497 00:21:30,550 --> 00:21:34,423 And so, it would be kind of strange if 498 00:21:34,467 --> 00:21:36,947 human intelligence was somehow really special 499 00:21:36,991 --> 00:21:40,908 in that regard and is somehow some upper limit 500 00:21:40,951 --> 00:21:43,040 of what's possible in intelligence. 501 00:21:43,084 --> 00:21:47,218 I think, once we understand how to build intelligent systems, 502 00:21:47,262 --> 00:21:50,657 we will be able to build huge machines, 503 00:21:50,700 --> 00:21:54,835 which will be far beyond normal human intelligence. 504 00:21:54,878 --> 00:21:58,621 Uh, hopefully we can have a very symbiotic relationship, 505 00:21:58,665 --> 00:22:01,276 uh, with AI systems, but the AI developers are 506 00:22:01,320 --> 00:22:03,670 specifically designing them to make sure that they can do 507 00:22:03,713 --> 00:22:05,454 everything better than we can, so I-I don't know 508 00:22:05,498 --> 00:22:08,196 what-what we will be able to offer, unfortunately. 509 00:22:08,239 --> 00:22:10,024 The-the older-school AI technology... 510 00:22:10,067 --> 00:22:12,287 CAROLINE: Daniel isn't feeling any better. 511 00:22:12,331 --> 00:22:14,376 His plan is backfiring. 512 00:22:14,420 --> 00:22:17,248 The more he learns about this new, powerful, 513 00:22:17,292 --> 00:22:19,120 inscrutable thing, the worse it sounds. 514 00:22:19,163 --> 00:22:21,340 He wants to tell them how scared he is, 515 00:22:21,383 --> 00:22:24,430 how he feels like the earth is slipping out from under him, 516 00:22:24,473 --> 00:22:26,997 that he's staring down existential dread, 517 00:22:27,041 --> 00:22:29,173 so he articulates this by saying... 518 00:22:29,217 --> 00:22:30,740 That sounds bad. 519 00:22:30,784 --> 00:22:32,046 [Hendrycks laughs] 520 00:22:32,089 --> 00:22:33,830 Yeah. 521 00:22:35,832 --> 00:22:38,400 LADISH: If you have all of these capabilities 522 00:22:38,444 --> 00:22:41,011 and they start to be able to plan better, 523 00:22:41,055 --> 00:22:43,579 if you sort of take that to its logical conclusion, 524 00:22:43,623 --> 00:22:46,234 you can get some pretty power-seeking behavior. 525 00:22:46,277 --> 00:22:48,410 [electronic trilling] 526 00:22:48,454 --> 00:22:50,281 [beeps pulsing] 527 00:22:55,374 --> 00:22:57,680 DANIEL: Okay, so why would an AI 528 00:22:57,724 --> 00:22:59,203 want more power? 529 00:22:59,247 --> 00:23:02,206 Yeah. So, I think it's actually pretty simple. 530 00:23:02,250 --> 00:23:04,774 Having more power is a very effective strategy 531 00:23:04,818 --> 00:23:06,515 for accomplishing almost any goal. 532 00:23:06,559 --> 00:23:08,952 We ran an experiment where we gave 533 00:23:08,996 --> 00:23:11,868 OpenAI's most powerful AI model, uh, 534 00:23:11,912 --> 00:23:13,392 a series of problems to solve. 535 00:23:13,435 --> 00:23:15,219 And partway through, on its computer, 536 00:23:15,263 --> 00:23:18,701 it got a notification that it was going to be shut down. 537 00:23:18,745 --> 00:23:20,921 And what it did is it rewrote that code 538 00:23:20,964 --> 00:23:22,488 to prevent itself from being shut down 539 00:23:22,531 --> 00:23:25,621 so it could finish solving the problems. 540 00:23:25,665 --> 00:23:27,754 -Okay. -Yeah. So, another really interesting one 541 00:23:27,797 --> 00:23:31,105 is that the AI company Anthropic 542 00:23:31,148 --> 00:23:34,935 made a simulated environment where that AI had access 543 00:23:34,978 --> 00:23:36,458 to all of the company emails. 544 00:23:36,502 --> 00:23:38,547 And it learned through reading those emails 545 00:23:38,591 --> 00:23:40,288 it was going to be replaced 546 00:23:40,331 --> 00:23:43,378 and the lead engineer who was responsible for this 547 00:23:43,422 --> 00:23:45,685 was also having an affair. 548 00:23:45,728 --> 00:23:47,948 And on its own, it used that information 549 00:23:47,991 --> 00:23:49,515 to blackmail the engineer 550 00:23:49,558 --> 00:23:52,039 to prevent itself from being replaced. 551 00:23:52,082 --> 00:23:53,867 It was like, "No, I'm not gonna be replaced. 552 00:23:53,910 --> 00:23:56,913 "If you replace me, I'm going to tell the world 553 00:23:56,957 --> 00:23:58,567 that you are having this affair." 554 00:23:59,655 --> 00:24:01,440 And nobody taught it to do that? 555 00:24:01,483 --> 00:24:03,311 No, it learned to do that on its own. 556 00:24:05,139 --> 00:24:07,707 As the models get smarter, they learn that these are 557 00:24:07,750 --> 00:24:10,536 effective ways to accomplish goals. 558 00:24:10,579 --> 00:24:12,842 And this is not a problem that's isolated to one model. 559 00:24:12,886 --> 00:24:15,671 All of the most powerful models show these behaviors. 560 00:24:15,715 --> 00:24:18,369 ♪ ♪ 561 00:24:23,766 --> 00:24:25,246 -Hey. -DANIEL: Hey. How are you? 562 00:24:25,289 --> 00:24:26,813 [chuckles] Good. Good to be here. 563 00:24:26,856 --> 00:24:28,684 ANDERSON COOPER: When Yuval Noah Harari 564 00:24:28,728 --> 00:24:31,426 published his first book Sapiens in 2014, 565 00:24:31,470 --> 00:24:33,820 it became a global bestseller 566 00:24:33,863 --> 00:24:35,996 and turned the little-known Israeli history professor 567 00:24:36,039 --> 00:24:37,998 into one of the most popular writers 568 00:24:38,041 --> 00:24:39,390 and thinkers on the planet. 569 00:24:39,434 --> 00:24:42,045 The biggest danger with AI is 570 00:24:42,089 --> 00:24:43,960 the belief that it is infallible, 571 00:24:44,004 --> 00:24:45,658 that we have finally found-- 572 00:24:45,701 --> 00:24:49,400 "Okay, gods were just this mythological creation. 573 00:24:49,444 --> 00:24:52,708 "Humans, we can't trust them, but AI is infallible. 574 00:24:52,752 --> 00:24:55,189 It will never make any mistakes." 575 00:24:55,232 --> 00:24:58,322 And this is a deadly, deadly threat. 576 00:24:58,366 --> 00:25:00,107 It will make mistakes. 577 00:25:00,150 --> 00:25:03,284 And all these fantasies that AI will reveal the truth 578 00:25:03,327 --> 00:25:05,634 about the world that we can't find by ourselves, 579 00:25:05,678 --> 00:25:08,724 AI will not reveal the truth about the world. 580 00:25:08,768 --> 00:25:12,685 AI will create an entirely new world, much more complicated 581 00:25:12,728 --> 00:25:15,296 and difficult to understand than this one. 582 00:25:17,994 --> 00:25:22,651 What's about to happen is that we, uh, humans are no longer 583 00:25:22,695 --> 00:25:26,263 going to be the most intelligent entities on Earth. 584 00:25:26,307 --> 00:25:28,570 So I think what's coming up is going to be 585 00:25:28,614 --> 00:25:32,661 one of the biggest events in human history. 586 00:25:32,705 --> 00:25:34,358 JAKE TAPPER: Geoffrey, thanks so much for joining us. 587 00:25:34,402 --> 00:25:35,925 So you left your job with Google in part because you say 588 00:25:35,969 --> 00:25:38,798 you want to focus solely on your concerns about AI. 589 00:25:38,841 --> 00:25:41,627 You've spoken out, saying that AI could manipulate 590 00:25:41,670 --> 00:25:44,978 or possibly figure out a way to kill humans. 591 00:25:45,021 --> 00:25:47,023 H-How could it kill humans? 592 00:25:48,764 --> 00:25:50,940 Well, if it gets to be much smarter than us, 593 00:25:50,984 --> 00:25:52,551 it'll be very good at manipulation 594 00:25:52,594 --> 00:25:54,727 'cause it will have learned that from us. 595 00:25:54,770 --> 00:25:57,947 And it'll figure out ways of manipulating people 596 00:25:57,991 --> 00:26:00,820 to do what it wants. 597 00:26:00,863 --> 00:26:05,476 RASKIN: There was a open letter from the Center for AI Safety. 598 00:26:05,520 --> 00:26:08,784 Sam Altman signed this. Demis signed this. 599 00:26:08,828 --> 00:26:12,048 They signed a 22-word statement 600 00:26:12,092 --> 00:26:15,312 that we need to take AI and the threat from AI 601 00:26:15,356 --> 00:26:19,316 as seriously as global nuclear war. 602 00:26:24,844 --> 00:26:26,933 -DANIEL: Hello. -Hello. 603 00:26:30,327 --> 00:26:32,895 You're kind of, like, the original doom guy. 604 00:26:32,939 --> 00:26:34,897 More or less. 605 00:26:34,941 --> 00:26:37,421 Since 2001, I have been working on 606 00:26:37,465 --> 00:26:38,945 what we would now call the problem 607 00:26:38,988 --> 00:26:41,991 of aligning artificial general intelligence. 608 00:26:42,035 --> 00:26:44,080 How to shape the preferences and behavior 609 00:26:44,124 --> 00:26:46,343 of a powerful artificial mind 610 00:26:46,387 --> 00:26:48,868 such that it does not kill everyone. 611 00:26:51,610 --> 00:26:53,568 It's not like a lifeless machine. 612 00:26:53,612 --> 00:26:55,309 It is smart, it is creative, 613 00:26:55,352 --> 00:26:57,311 it is inventive, it has the properties 614 00:26:57,354 --> 00:26:59,356 that makes the human species dangerous, 615 00:26:59,400 --> 00:27:01,663 and it has more of those properties. 616 00:27:01,707 --> 00:27:04,710 If something doesn't actively care about you, 617 00:27:04,753 --> 00:27:06,189 actively want you to live, 618 00:27:06,233 --> 00:27:07,800 actively care about your welfare, 619 00:27:07,843 --> 00:27:09,889 about you being happy and alive and free, 620 00:27:09,932 --> 00:27:12,456 if it cares about other stuff instead, 621 00:27:12,500 --> 00:27:14,415 and you're on the same planet, 622 00:27:14,458 --> 00:27:18,724 that is not survivable if it is very much smarter than you. 623 00:27:18,767 --> 00:27:20,551 LEAHY: I don't think it's going to be 624 00:27:20,595 --> 00:27:22,902 a kind of, like, evil thing. 625 00:27:22,945 --> 00:27:26,296 It's like, "Oh, the AIs are evil and they hate humanity." 626 00:27:26,340 --> 00:27:27,689 I don't think that's what's gonna happen. 627 00:27:27,733 --> 00:27:28,995 I think what is happening is 628 00:27:29,038 --> 00:27:30,997 far more how humans feel about ants. 629 00:27:31,040 --> 00:27:32,476 [ants chittering] 630 00:27:32,520 --> 00:27:35,436 Like, we don't hate ants, 631 00:27:35,479 --> 00:27:37,133 but if we want to build a highway 632 00:27:37,177 --> 00:27:40,180 and there's an anthill there, well, sucks for the ants. 633 00:27:40,223 --> 00:27:41,529 It's not that hard to understand. 634 00:27:41,572 --> 00:27:43,226 It's like, hey, if we build things 635 00:27:43,270 --> 00:27:44,663 that are smarter than us 636 00:27:44,706 --> 00:27:46,229 and we don't know how to control them, 637 00:27:46,273 --> 00:27:48,275 does that seem like a risky thing to you? 638 00:27:48,318 --> 00:27:49,363 [dramatic soundtrack music playing] 639 00:27:49,406 --> 00:27:50,843 [screeching] 640 00:27:50,886 --> 00:27:52,279 Yeah. Yeah, it does. 641 00:27:52,322 --> 00:27:54,020 You don't have to be a tech guy. 642 00:27:54,063 --> 00:27:55,369 You don't have to know programming to understand it. 643 00:27:55,412 --> 00:27:56,718 It's not that hard. 644 00:27:56,762 --> 00:27:58,111 This is not a hard thing to understand. 645 00:27:58,154 --> 00:28:00,591 Connor, how-how many people 646 00:28:00,635 --> 00:28:02,637 in the world right now are working on AGI? 647 00:28:02,681 --> 00:28:05,466 At least 20,000, I would say. 648 00:28:05,509 --> 00:28:07,729 -20,000? -I would expect so. 649 00:28:07,773 --> 00:28:09,426 Okay, and how many people are working full-time 650 00:28:09,470 --> 00:28:12,038 to make sure AI doesn't, like, kill us all? 651 00:28:12,081 --> 00:28:14,562 Probably less than 200 in the world. 652 00:28:16,085 --> 00:28:17,913 DANIEL: And your conceit is that 653 00:28:17,957 --> 00:28:24,528 the only natural result of this recklessness... 654 00:28:25,704 --> 00:28:28,924 ...is the collapse of humanity? 655 00:28:28,968 --> 00:28:33,189 Well, not the collapse, the abrupt extermination. 656 00:28:33,233 --> 00:28:35,888 There's a difference. 657 00:28:35,931 --> 00:28:37,716 ["What Is the Meaning?" by Nun-Plus playing] 658 00:28:37,759 --> 00:28:42,590 ♪ What is the meaning of life? ♪ 659 00:28:42,633 --> 00:28:46,594 ♪ What is the future and what is now? ♪ 660 00:28:46,637 --> 00:28:49,945 ♪ What is the answer to strife? ♪ 661 00:28:49,989 --> 00:28:52,818 ♪ How much ♪ 662 00:28:52,861 --> 00:28:54,950 ♪ Can someone dream? ♪ 663 00:28:54,994 --> 00:28:57,474 ♪ How long? ♪ 664 00:28:57,518 --> 00:28:58,954 ♪ Forever ♪ 665 00:29:00,216 --> 00:29:01,870 ♪ What is the meaning ♪ 666 00:29:01,914 --> 00:29:04,133 -♪ Of me... ♪ -[civil defense siren blaring] 667 00:29:04,177 --> 00:29:06,266 LADISH: I do think this is probably, like, 668 00:29:06,309 --> 00:29:07,963 the biggest challenge that, like, 669 00:29:08,007 --> 00:29:11,271 our civilization will-will face, ever. 670 00:29:11,314 --> 00:29:14,361 This essentially is the last mistake we'll ever get to make. 671 00:29:16,276 --> 00:29:22,325 If we can rise to be the most mature version of ourselves, 672 00:29:22,369 --> 00:29:24,414 there might be a way through this. 673 00:29:24,458 --> 00:29:25,764 DANIEL: What does that mean? 674 00:29:25,807 --> 00:29:27,809 "The most mature version of ourselves"? 675 00:29:29,332 --> 00:29:32,205 'Cause that sounds, for me, like... 676 00:29:32,248 --> 00:29:34,250 I-I-- What the [bleep]? 677 00:29:36,731 --> 00:29:39,168 [Daniel sighs] 678 00:29:39,212 --> 00:29:43,216 Do you think now is a good time to have a kid? 679 00:29:48,221 --> 00:29:49,396 Um... 680 00:29:49,439 --> 00:29:51,398 Do you want to have kids one day? 681 00:29:51,441 --> 00:29:53,400 Is that something that-that you're into or not really? 682 00:29:53,443 --> 00:29:57,796 Um... uh, I confess, I think that's like, 683 00:29:57,839 --> 00:29:59,972 [laughing]: "Boy, let's get through this critical period." 684 00:30:00,015 --> 00:30:01,234 Um... 685 00:30:01,277 --> 00:30:02,496 DANIEL: Do you have any kids? 686 00:30:02,539 --> 00:30:03,758 I do not. 687 00:30:03,802 --> 00:30:05,064 Is that something you want to do? 688 00:30:05,107 --> 00:30:07,196 Have children, have a family? 689 00:30:07,240 --> 00:30:11,244 In some other world than this world, sure, I would have kids. 690 00:30:12,593 --> 00:30:14,421 DANIEL: Would you want to start a family? 691 00:30:14,464 --> 00:30:16,379 Would you want to have kids? 692 00:30:16,423 --> 00:30:18,425 Is that something you're thinking about? 693 00:30:23,734 --> 00:30:25,780 I, um... 694 00:30:28,087 --> 00:30:30,219 ♪ ♪ 695 00:30:30,263 --> 00:30:32,482 CAROLINE: I just have to find it first. 696 00:30:32,526 --> 00:30:34,702 [fetal heartbeat pulsing] 697 00:30:34,745 --> 00:30:37,748 We have to go to the doctor, but... 698 00:30:37,792 --> 00:30:41,143 Ah! I knew it! I knew it! I knew it! 699 00:30:41,187 --> 00:30:43,842 -When did you find out? -CAROLINE: Last night. 700 00:30:43,885 --> 00:30:45,931 -Oh, my God! -Mom, I don't know. 701 00:30:45,974 --> 00:30:47,280 RUTHIE: I can't tell you 702 00:30:47,323 --> 00:30:49,195 -how happy I am! -CAROLINE: No, seriously. 703 00:30:49,238 --> 00:30:51,588 Well, I took a pregnancy test last night, and I'm pregnant. 704 00:30:51,632 --> 00:30:53,155 WOMAN: Oh, my God! 705 00:30:53,199 --> 00:30:55,114 Oh, my God, you guys! 706 00:30:55,157 --> 00:30:57,290 NURSE: I just wanted to confirm your expected due date, 707 00:30:57,333 --> 00:30:59,945 -which is January 21st. -[laughing]: Oh, my God. 708 00:30:59,988 --> 00:31:02,512 I can't believe how happy I am. 709 00:31:02,556 --> 00:31:03,905 [fetal heartbeat pulsing] 710 00:31:03,949 --> 00:31:05,472 CAROLINE: He already looks really cute. 711 00:31:05,515 --> 00:31:07,126 DANIEL: You think he already looks cute? 712 00:31:07,169 --> 00:31:08,518 CAROLINE: Yeah. Look at that little cutie face. 713 00:31:10,564 --> 00:31:12,609 [fetal heartbeat fading] 714 00:31:13,959 --> 00:31:15,525 DANIEL: I have this baby on the way. 715 00:31:15,569 --> 00:31:16,787 TRISTAN HARRIS: Right. 716 00:31:16,831 --> 00:31:18,746 DANIEL: So I turn it over to you. 717 00:31:18,789 --> 00:31:21,227 Are we doomed? 718 00:31:21,270 --> 00:31:23,969 Are we all gonna face this techno dystopian 719 00:31:24,012 --> 00:31:25,884 future of doom? 720 00:31:28,756 --> 00:31:31,019 It's, uh... 721 00:31:31,063 --> 00:31:32,586 [chuckles]: It's not good news, 722 00:31:32,629 --> 00:31:35,154 the world that we're heading into. 723 00:31:35,197 --> 00:31:36,982 I-- for ex-- I mean, I'll just be honest. 724 00:31:37,025 --> 00:31:39,549 Uh, I know people who work on AI risk 725 00:31:39,593 --> 00:31:42,248 who don't expect their children to make it to high school. 726 00:31:43,249 --> 00:31:45,251 ♪ ♪ 727 00:31:52,127 --> 00:31:53,346 [lights clank] 728 00:31:58,003 --> 00:32:01,528 DANIEL: This is, like... this is actually scary 729 00:32:01,571 --> 00:32:03,051 'cause it's like, oh, we're all [bleep]. 730 00:32:03,095 --> 00:32:04,923 CAROLINE: You have to c... make me calm, 731 00:32:04,966 --> 00:32:06,794 because this is making me incredibly anxious 732 00:32:06,837 --> 00:32:08,622 and I'm carrying the baby right now, 733 00:32:08,665 --> 00:32:11,625 so you have to also be calm for me and strong and hopeful, 734 00:32:11,668 --> 00:32:16,412 because I'm-- it's too, it's too much for my soul to bear 735 00:32:16,456 --> 00:32:19,285 while I'm carrying this baby. 736 00:32:19,328 --> 00:32:21,243 So you're going to have to try to figure out 737 00:32:21,287 --> 00:32:24,159 a way to have hope. 738 00:32:24,203 --> 00:32:27,771 It's really important, Daniel, especially now. 739 00:32:28,772 --> 00:32:30,557 H-How to have hope. 740 00:32:30,600 --> 00:32:33,342 You have to. You have to find it for me. 741 00:32:33,386 --> 00:32:34,909 I'm serious. 742 00:32:34,953 --> 00:32:36,606 I'm going to. 743 00:32:37,825 --> 00:32:40,349 I will. I'll tr-- I'll try. 744 00:32:46,399 --> 00:32:47,617 [lights clank] 745 00:32:49,619 --> 00:32:51,404 DANIEL: Hey, guys? 746 00:32:51,447 --> 00:32:52,840 Guys. 747 00:32:54,363 --> 00:32:56,278 Can we get back up and running? 748 00:32:58,411 --> 00:33:01,022 Oy gevalt! [sighs] 749 00:33:03,894 --> 00:33:05,331 TED: Uh, Dan, are you ready? 750 00:33:05,374 --> 00:33:06,680 PETER DIAMANDIS: Hello, Daniel. 751 00:33:07,724 --> 00:33:09,465 DANIEL: Hey, how are you? 752 00:33:09,509 --> 00:33:11,032 DIAMANDIS: I am... I'm well. 753 00:33:11,076 --> 00:33:13,948 I think, uh... I think I need some help, Peter. 754 00:33:15,254 --> 00:33:17,604 Um, I've been working on this film for about, 755 00:33:17,647 --> 00:33:19,823 I'm gonna say, eight to ten months now. 756 00:33:19,867 --> 00:33:23,262 It has been very, at times, depressing. 757 00:33:23,305 --> 00:33:25,612 -Hmm. -I have felt very alienated 758 00:33:25,655 --> 00:33:27,744 -mak-making this movie. -By who? 759 00:33:27,788 --> 00:33:30,051 By all of these, the-- all these guys 760 00:33:30,095 --> 00:33:33,228 who sit around and tell me that the world's gonna end. 761 00:33:33,272 --> 00:33:34,708 -Ah. -That, like, 762 00:33:34,751 --> 00:33:36,536 y-you know, th-this doom bullshit, you know? 763 00:33:36,579 --> 00:33:38,059 I know it well. 764 00:33:38,103 --> 00:33:39,408 "We're all doomed. Everyone's gonna die. 765 00:33:39,452 --> 00:33:40,931 Everything's awful." 766 00:33:40,975 --> 00:33:43,151 Awesome. Uh, let me bring you some light. 767 00:33:43,195 --> 00:33:44,718 Please. 768 00:33:44,761 --> 00:33:48,156 We truly are living in an extraordinary time. 769 00:33:49,418 --> 00:33:51,333 And many people forget this. 770 00:33:51,377 --> 00:33:53,640 Everything around us is a product of intelligence, 771 00:33:53,683 --> 00:33:57,035 and so everything that we touch with these new tools is likely 772 00:33:57,078 --> 00:33:59,646 to produce far more value than we've ever seen before. 773 00:33:59,689 --> 00:34:02,823 AI can help us discover new materials. 774 00:34:02,866 --> 00:34:04,999 AI can help social scientists 775 00:34:05,043 --> 00:34:07,349 to understand how economics work. 776 00:34:07,393 --> 00:34:11,658 There's a lot AI could do to make life and work better. 777 00:34:11,701 --> 00:34:14,095 I feel more empowered today, 778 00:34:14,139 --> 00:34:16,315 more confident to learn something today. 779 00:34:16,358 --> 00:34:19,057 We're gonna become superhumans because we have super AIs. 780 00:34:19,100 --> 00:34:21,798 This is just the beginning of an explosion. 781 00:34:21,842 --> 00:34:24,453 Humans and AI collaborating 782 00:34:24,497 --> 00:34:26,455 to solve really important problems. 783 00:34:26,499 --> 00:34:29,763 It is here to liberate us from routine jobs, 784 00:34:29,806 --> 00:34:33,723 and it is here to remind us what it is that makes us human. 785 00:34:33,767 --> 00:34:35,769 I think this is 786 00:34:35,812 --> 00:34:38,598 the most extraordinary time to be alive. 787 00:34:38,641 --> 00:34:42,036 The only time more exciting than today is tomorrow. 788 00:34:42,080 --> 00:34:45,257 Uh, I think that children born today, 789 00:34:45,300 --> 00:34:47,259 they're about to enter 790 00:34:47,302 --> 00:34:50,827 a glorious period of human transformation. 791 00:34:50,871 --> 00:34:52,525 Are we gonna have challenges? Of course. 792 00:34:52,568 --> 00:34:55,180 Can we solve those challenges? We do every single time. 793 00:34:55,223 --> 00:34:57,704 We are here, which is miraculous. 794 00:34:57,747 --> 00:35:00,272 -I already love you. -[chuckles]: Okay. 795 00:35:00,315 --> 00:35:01,925 -Super thankful to have you here. -[applause] 796 00:35:01,969 --> 00:35:03,362 -Super stoked to be here. -Yeah. 797 00:35:03,405 --> 00:35:04,885 So, the floor is yours, sir. 798 00:35:04,928 --> 00:35:07,409 Thank you so much. Super excited to be here. 799 00:35:07,453 --> 00:35:10,586 GUILLAUME VERDON: Yo, yo. All right. 800 00:35:10,630 --> 00:35:12,284 The future's gonna be awesome. 801 00:35:12,327 --> 00:35:14,286 I mean, ever since I was a kid, 802 00:35:14,329 --> 00:35:17,332 I wanted to understand the universe we live in, 803 00:35:17,376 --> 00:35:20,422 in order to figure out how to create the technologies 804 00:35:20,466 --> 00:35:23,991 that help us increase the scope and scale of civilization. 805 00:35:24,034 --> 00:35:26,689 I feel like if everyone had that mindset, 806 00:35:26,733 --> 00:35:29,214 then we'd actually live in a better world, right? 807 00:35:29,257 --> 00:35:30,519 I do believe that. 808 00:35:32,782 --> 00:35:34,697 -DANIEL: Hi, Pete. How are you? -[chuckles] 809 00:35:34,741 --> 00:35:36,525 PETER LEE: I thought a lot about 810 00:35:36,569 --> 00:35:39,049 what I would call dread, AI dread. 811 00:35:39,093 --> 00:35:40,703 I feel it. 812 00:35:40,747 --> 00:35:43,271 I-I haven't met any thoughtful human being 813 00:35:43,315 --> 00:35:45,143 who doesn't feel it. 814 00:35:45,186 --> 00:35:48,276 And anyone who says they don't feel it, you know, is lying. 815 00:35:48,320 --> 00:35:50,104 But, you know, overall, and the reason 816 00:35:50,148 --> 00:35:53,716 that I'm personally optimistic about this, uh, is that 817 00:35:53,760 --> 00:35:56,545 a huge fraction of the world's most intelligent people 818 00:35:56,589 --> 00:35:58,417 are thinking very hard 819 00:35:58,460 --> 00:36:02,116 about the potential downstream harms and risks of AI. 820 00:36:02,160 --> 00:36:05,641 We have this sort of vision of safety being, uh, 821 00:36:05,685 --> 00:36:08,253 kind of at the center of the research that we do. 822 00:36:08,296 --> 00:36:09,558 DANIELA AMODEI: No, that's totally fine. 823 00:36:09,602 --> 00:36:10,820 [indistinct chatter] 824 00:36:10,864 --> 00:36:13,562 I think there are more 825 00:36:13,606 --> 00:36:17,436 potential benefits than there are potential downsides, 826 00:36:17,479 --> 00:36:19,481 and I think it is incumbent upon 827 00:36:19,525 --> 00:36:21,831 the people that are creating this technology 828 00:36:21,875 --> 00:36:24,530 to make sure that we're doing the best job we can 829 00:36:24,573 --> 00:36:25,835 to make it safe for people. 830 00:36:25,879 --> 00:36:27,402 WOMAN: Reid Hoffman. 831 00:36:27,446 --> 00:36:28,969 REID HOFFMAN: What I can guarantee you is 832 00:36:29,012 --> 00:36:30,536 -some bad things will happen. -Take one. 833 00:36:30,579 --> 00:36:32,668 What we're gonna try to do is make those bad things 834 00:36:32,712 --> 00:36:36,150 as few and not huge as possible, 835 00:36:36,194 --> 00:36:38,370 and then we're gonna iterate to have-- 836 00:36:38,413 --> 00:36:40,633 be in a much better place with society. 837 00:36:40,676 --> 00:36:43,679 -Hello. -SHAW WALTERS: Hello. 838 00:36:43,723 --> 00:36:45,246 DANIEL: How are you, Moon? 839 00:36:45,290 --> 00:36:48,989 I feel like we are ending a chapter in humanity 840 00:36:49,032 --> 00:36:50,512 and beginning a new one, 841 00:36:50,556 --> 00:36:53,211 and it's a very interesting time to be alive. 842 00:36:53,254 --> 00:36:54,429 And if I could be born right now, 843 00:36:54,473 --> 00:36:55,865 I definitely would want to be. 844 00:36:55,909 --> 00:36:57,476 Like, that would be so exciting. 845 00:36:57,519 --> 00:37:00,305 I-I'm very excited. [laughs] 846 00:37:00,348 --> 00:37:02,481 What, what a future. Does it excited... excite you? 847 00:37:02,524 --> 00:37:03,699 No. 848 00:37:03,743 --> 00:37:06,528 -[laughing] -No, not really. 849 00:37:06,572 --> 00:37:08,313 DANIEL: A lot of these people have told me that, you know, 850 00:37:08,356 --> 00:37:10,445 my kid's not gonna make it to high school. 851 00:37:10,489 --> 00:37:13,318 Why are they wrong? Please explain it to me. 852 00:37:13,361 --> 00:37:15,058 Just because you have-- 853 00:37:15,102 --> 00:37:18,845 you struggle to predict the future in your own mind 854 00:37:18,888 --> 00:37:23,023 doesn't mean that it's necessarily gonna go awfully. 855 00:37:23,066 --> 00:37:26,592 In fact, there's a very high likelihood 856 00:37:26,635 --> 00:37:28,289 and the historical precedent is that 857 00:37:28,333 --> 00:37:30,117 things get massively better. 858 00:37:30,160 --> 00:37:34,469 The term I use is "data-driven optimism." 859 00:37:34,513 --> 00:37:38,647 There's solid foundation for you to be optimistic. 860 00:37:38,691 --> 00:37:41,998 Look at what this last century has been 861 00:37:42,042 --> 00:37:44,087 to see where we're going. 862 00:37:44,131 --> 00:37:45,915 Over the last hundred years, 863 00:37:45,959 --> 00:37:48,701 the average human lifespan has more than doubled. 864 00:37:48,744 --> 00:37:51,269 Average per capita income adjusted for inflation 865 00:37:51,312 --> 00:37:53,619 around the world has tripled. 866 00:37:53,662 --> 00:37:57,057 Childhood mortality has come down a factor of ten. 867 00:37:57,100 --> 00:37:59,712 The world has gotten better on almost every measure 868 00:37:59,755 --> 00:38:03,019 by orders of magnitude because of technology. 869 00:38:03,063 --> 00:38:05,152 On almost every measure. 870 00:38:05,195 --> 00:38:09,722 Less violence, more education, access to energy, food, water. 871 00:38:09,765 --> 00:38:12,942 All these things have happened for one reason. 872 00:38:12,986 --> 00:38:15,336 It's been technology 873 00:38:15,380 --> 00:38:17,556 that has turned scarcity into abundance, 874 00:38:17,599 --> 00:38:19,514 but it's also driven to an abundance 875 00:38:19,558 --> 00:38:21,864 of some negativities, right? 876 00:38:21,908 --> 00:38:25,738 Abundance of obesity, abundance of mental disorders, 877 00:38:25,781 --> 00:38:29,437 abundance of climate change, and so forth. 878 00:38:29,481 --> 00:38:31,309 And yes, this is true. 879 00:38:31,352 --> 00:38:34,094 But probably we will be better equipped to solve it 880 00:38:34,137 --> 00:38:36,226 using other technologies, like AI, 881 00:38:36,270 --> 00:38:39,273 than we will, say, stopping and turning everything off. 882 00:38:39,317 --> 00:38:41,449 There might be some existential risk, 883 00:38:41,493 --> 00:38:45,323 but AI is also the thing that can solve the pandemics, 884 00:38:45,366 --> 00:38:47,455 can help us with climate change, 885 00:38:47,499 --> 00:38:50,545 can help identify that asteroid way out there 886 00:38:50,589 --> 00:38:52,547 before we've seen it as a potential risk 887 00:38:52,591 --> 00:38:54,332 and help mitigate it. 888 00:38:54,375 --> 00:38:58,074 This is really gonna be the tool that helps us tackle 889 00:38:58,118 --> 00:39:01,426 all the challenges that we're facing as a species, right? 890 00:39:01,469 --> 00:39:04,167 We need to fix water desalination. 891 00:39:04,211 --> 00:39:05,995 We need to grow food 892 00:39:06,039 --> 00:39:08,433 100 X cheaper than we currently do. 893 00:39:08,476 --> 00:39:12,350 We need renewable energy to be, you know, ubiquitous 894 00:39:12,393 --> 00:39:14,047 and everywhere in our lives. 895 00:39:14,090 --> 00:39:17,137 Everywhere you look, in the next 50 years, 896 00:39:17,180 --> 00:39:19,269 we have to do more with less. 897 00:39:19,313 --> 00:39:23,186 Training machines to help us is absolutely essential. 898 00:39:23,230 --> 00:39:24,797 Scientists are using artificial intelligence 899 00:39:24,840 --> 00:39:25,972 for carbon capture. 900 00:39:26,015 --> 00:39:27,582 It's a critical technology. 901 00:39:27,626 --> 00:39:29,541 DIAMANDIS: The tools to solve these problems, 902 00:39:29,584 --> 00:39:32,195 like fusion, that isn't theoretical anymore, 903 00:39:32,239 --> 00:39:33,980 it's coming. 904 00:39:34,023 --> 00:39:37,723 We are on the precipice of extraordinary technologies. 905 00:39:37,766 --> 00:39:39,986 AMNA NAWAZ: This year's Nobel Prize in Chemistry 906 00:39:40,029 --> 00:39:42,815 went to three scientists for their groundbreaking work 907 00:39:42,858 --> 00:39:44,556 using artificial intelligence 908 00:39:44,599 --> 00:39:47,297 to advance biomedical and protein research. 909 00:39:47,341 --> 00:39:49,691 DEMIS HASSABIS: Protein folding is 910 00:39:49,735 --> 00:39:53,608 one of these holy grail type problems in biology. 911 00:39:53,652 --> 00:39:55,741 So people have been predicting since the '70s 912 00:39:55,784 --> 00:39:58,091 that this should be possible, but until now, 913 00:39:58,134 --> 00:39:59,571 no one has been able to do it. 914 00:39:59,614 --> 00:40:01,181 And it's gonna be really important for things 915 00:40:01,224 --> 00:40:03,618 like drug discovery and understanding disease. 916 00:40:03,662 --> 00:40:06,534 I-I think we could, you know, cure most diseases 917 00:40:06,578 --> 00:40:11,931 within the next decade or-or two if, uh, AI drug design works. 918 00:40:11,974 --> 00:40:15,630 Technological progress enables more human lives, right? 919 00:40:15,674 --> 00:40:18,677 I mean, if we accelerate, 920 00:40:18,720 --> 00:40:21,723 the number of humans we can support grows exponentially. 921 00:40:21,767 --> 00:40:24,073 If we slow down, it plateaus. 922 00:40:24,117 --> 00:40:27,729 That gap is effectively future people 923 00:40:27,773 --> 00:40:31,341 that deceleration has effectively killed. 924 00:40:31,385 --> 00:40:33,909 DANIEL: Millions of lives that won't exist. 925 00:40:33,953 --> 00:40:35,128 Billions. 926 00:40:35,171 --> 00:40:36,782 Or tens of billions. 927 00:40:36,825 --> 00:40:38,436 You know, someone said, "Oh, my God, 928 00:40:38,479 --> 00:40:40,655 can we survive with digital superintelligence?" 929 00:40:40,699 --> 00:40:42,309 And my question is: 930 00:40:42,352 --> 00:40:44,180 Can we survive without digital superintelligence? 931 00:40:44,224 --> 00:40:47,575 So we're using AI as an assistant to providers. 932 00:40:47,619 --> 00:40:50,448 This is generally a trend that we can already see happening. 933 00:40:50,491 --> 00:40:52,319 BERTHA COOMBS: ...harnessing generative AI programs 934 00:40:52,362 --> 00:40:54,060 to help doctors and nurses... 935 00:40:54,103 --> 00:40:56,062 We always have problems. 936 00:40:56,105 --> 00:40:59,108 And those problems are food for entrepreneurs 937 00:40:59,152 --> 00:41:01,154 to create new business and new industries. 938 00:41:01,197 --> 00:41:03,069 SHANELLE KAUL: ...with the help of artificial intelligence, 939 00:41:03,112 --> 00:41:04,766 farmers are getting the help they need 940 00:41:04,810 --> 00:41:06,594 to perform labor-intensive tasks... 941 00:41:06,638 --> 00:41:10,163 With the help of Ulangizi AI, the farmers are now able 942 00:41:10,206 --> 00:41:12,948 to ask the suitable crops that they can plant. 943 00:41:12,992 --> 00:41:14,950 WILL REEVE: ...AI being used as a thought decoder 944 00:41:14,994 --> 00:41:17,387 and sending that signal to the spine. 945 00:41:17,431 --> 00:41:21,609 AI is gonna become the most extraordinary tool of all. 946 00:41:21,653 --> 00:41:24,264 We as a broader society have to think about 947 00:41:24,307 --> 00:41:27,049 how do we want to use this technology, right? 948 00:41:27,093 --> 00:41:28,529 We the humans. 949 00:41:28,573 --> 00:41:31,097 What do we want it to do for us? 950 00:41:31,140 --> 00:41:34,535 ♪ ♪ 951 00:41:34,579 --> 00:41:36,189 DANIEL: I'm thinking about this through the perspective 952 00:41:36,232 --> 00:41:37,886 and lens of, like, my son growing up 953 00:41:37,930 --> 00:41:39,540 in the world with all of this. 954 00:41:39,584 --> 00:41:42,238 What does the best version of his life look like? 955 00:41:42,282 --> 00:41:44,589 If everything works out. 956 00:41:44,632 --> 00:41:46,808 [sighs heavily] 957 00:41:46,852 --> 00:41:50,246 WALTERS: The place where kids are probably gonna see 958 00:41:50,290 --> 00:41:52,684 the greatest impact on their life immediately 959 00:41:52,727 --> 00:41:54,424 -is probably gonna be school. -[children chattering] 960 00:41:54,468 --> 00:41:56,688 I think that the nature of what school is 961 00:41:56,731 --> 00:41:58,472 is gonna fundamentally change. 962 00:41:58,516 --> 00:42:00,474 [child giggles] 963 00:42:00,518 --> 00:42:02,737 DIAMANDIS: I'm seeing this amazing world 964 00:42:02,781 --> 00:42:06,436 where every child has access to not good education 965 00:42:06,480 --> 00:42:09,091 but, very shortly, the best education on the planet. 966 00:42:09,135 --> 00:42:13,879 HOFFMAN: Tutors, every subject, infinitely patient. 967 00:42:18,013 --> 00:42:20,059 DIAMANDIS: Imagine a future 968 00:42:20,102 --> 00:42:22,235 where the poorest people on the planet 969 00:42:22,278 --> 00:42:25,804 have access to the best health care. 970 00:42:25,847 --> 00:42:29,547 Not good health care, the best health care, delivered by AIs. 971 00:42:31,157 --> 00:42:32,767 [electronic trilling] 972 00:42:34,334 --> 00:42:38,556 We're gonna be able to extend our health span, 973 00:42:38,599 --> 00:42:42,168 not just our lifespan, our health span, by decades. 974 00:42:44,431 --> 00:42:46,215 HOFFMAN: You're just about to have a kid. 975 00:42:46,259 --> 00:42:49,697 Oh, the kid's, like, burping incontrollably. 976 00:42:49,741 --> 00:42:52,047 Is this something I should be worried about? 977 00:42:52,091 --> 00:42:54,223 There, 24-7, for you. 978 00:42:54,267 --> 00:42:57,183 And where we're going-- and it may be fearful to some-- 979 00:42:57,226 --> 00:42:59,228 is that we're gonna merge with AI. 980 00:42:59,272 --> 00:43:01,361 We're gonna merge with technology. 981 00:43:02,667 --> 00:43:06,105 By the early to mid 2030s, 982 00:43:06,148 --> 00:43:10,457 expect that we're able to connect our brain to the cloud, 983 00:43:10,500 --> 00:43:13,721 where I can start to expand access to memory. 984 00:43:16,071 --> 00:43:18,030 [electronic trilling] 985 00:43:19,771 --> 00:43:22,077 DANIEL: Okay, this is great. What's another cool AI thing? 986 00:43:22,121 --> 00:43:23,775 He won't have to work, right? 987 00:43:23,818 --> 00:43:25,167 -Like, when he grows up, he might not have... -[sighs] 988 00:43:25,211 --> 00:43:26,212 -...he won't have to have a job. -I mean... 989 00:43:26,255 --> 00:43:27,779 He won't have to have a job, 990 00:43:27,822 --> 00:43:30,608 but he might really have a strong passion, 991 00:43:30,651 --> 00:43:33,132 and he has to really think about, 992 00:43:33,175 --> 00:43:36,091 "Okay, I'm here. I can do anything with my life. 993 00:43:36,135 --> 00:43:37,876 So what do I do?" 994 00:43:38,920 --> 00:43:40,182 We need to find a meaning, 995 00:43:40,226 --> 00:43:42,620 beyond like, this current form of 996 00:43:42,663 --> 00:43:45,666 we live for work and work for living. 997 00:43:45,710 --> 00:43:49,844 DANIEL: So my son can... can just be a poet. 998 00:43:49,888 --> 00:43:51,541 -Yes. -And a painter. 999 00:43:51,585 --> 00:43:53,021 Absolutely. 1000 00:43:53,065 --> 00:43:54,849 DANIEL: So my son can live his life 1001 00:43:54,893 --> 00:43:57,591 on a Grecian sunswept island, painting all day. 1002 00:43:57,635 --> 00:43:59,114 YU: Absolutely. Possible. 1003 00:44:00,072 --> 00:44:01,203 Absolutely. 1004 00:44:01,247 --> 00:44:02,944 Well, we're working toward that 1005 00:44:02,988 --> 00:44:04,554 that's a lot of work to do before we get there. 1006 00:44:06,992 --> 00:44:08,776 VERDON: I mean, if everything works out, 1007 00:44:08,820 --> 00:44:11,997 we have cheap, uh, abundant energy. 1008 00:44:13,607 --> 00:44:17,219 We can completely control our planet's climate. 1009 00:44:17,263 --> 00:44:20,135 We are harnessing energy from the sun. 1010 00:44:20,179 --> 00:44:24,400 We have become multiplanetary, so we become very robust. 1011 00:44:24,444 --> 00:44:26,664 We are harnessing minerals and resources 1012 00:44:26,707 --> 00:44:28,404 from the solar system. 1013 00:44:28,448 --> 00:44:31,233 DANIEL: So my... my boy could go to space. 1014 00:44:31,277 --> 00:44:32,887 VERDON: Sure. 1015 00:44:32,931 --> 00:44:34,715 -DANIEL: He could go to Mars. -VERDON: Yeah. 1016 00:44:34,759 --> 00:44:36,369 DANIEL: It's so crystal clear to me now. 1017 00:44:36,412 --> 00:44:39,198 My son could grow up in a world with no disease. 1018 00:44:39,241 --> 00:44:40,590 -VERDON: Yeah. -With no illness. 1019 00:44:40,634 --> 00:44:41,896 -Sure. -With no poverty. 1020 00:44:41,940 --> 00:44:43,681 -Yes. -We are about to enter 1021 00:44:43,724 --> 00:44:46,640 a post-scarcity world. 1022 00:44:46,684 --> 00:44:50,470 Just like the lungfish moved out of the oceans onto land 1023 00:44:50,513 --> 00:44:53,081 hundreds of millions of years ago, 1024 00:44:53,125 --> 00:44:57,999 we're about to move off of the Earth, into the cosmos, 1025 00:44:58,043 --> 00:44:59,871 in a collaborative fashion, 1026 00:44:59,914 --> 00:45:03,744 to do things that are not fathomable to us today. 1027 00:45:06,268 --> 00:45:09,184 This is what's possible using these exponential technologies 1028 00:45:09,228 --> 00:45:10,708 and these AIs. 1029 00:45:11,839 --> 00:45:13,928 Let's use these tools 1030 00:45:13,972 --> 00:45:16,104 to create this age of abundance. 1031 00:45:16,148 --> 00:45:18,150 [electronic clicking] 1032 00:45:20,500 --> 00:45:22,937 We need wisdom. 1033 00:45:22,981 --> 00:45:25,984 Uh, I think that digital superintelligence 1034 00:45:26,027 --> 00:45:28,726 will ultimately become the wisest, 1035 00:45:28,769 --> 00:45:34,253 you know, the village elders for humanity. 1036 00:45:34,296 --> 00:45:37,996 What if AI is trying to make people be 1037 00:45:38,039 --> 00:45:39,737 the best versions of themselves? 1038 00:45:39,780 --> 00:45:41,347 What if it's expanding 1039 00:45:41,390 --> 00:45:43,392 what is humanly possible for us to do? 1040 00:45:43,436 --> 00:45:46,656 How can we use this technology 1041 00:45:46,700 --> 00:45:49,921 to help bring out the better angels of our nature? 1042 00:45:49,964 --> 00:45:51,661 It's very easy, 1043 00:45:51,705 --> 00:45:54,316 when we encounter new things that can be very alien, 1044 00:45:54,360 --> 00:45:55,927 to first have fear. 1045 00:45:55,970 --> 00:45:57,493 Fear is an important thing 1046 00:45:57,537 --> 00:46:00,322 for how to navigate potentially bad things. 1047 00:46:00,366 --> 00:46:03,717 But we only make progress when we have hope. 1048 00:46:03,761 --> 00:46:05,719 -[dog barks] -[indistinct chatter] 1049 00:46:05,763 --> 00:46:07,895 DANIEL: Shh. Moose, stop it. 1050 00:46:07,939 --> 00:46:10,811 HOFFMAN: I have a lot of hope in humanity. 1051 00:46:10,855 --> 00:46:13,814 ["Harvest Moon" by Neil Young playing over phone] 1052 00:46:15,294 --> 00:46:17,296 [fetal heartbeat pulsing] 1053 00:46:18,471 --> 00:46:19,951 DANIEL: Ooh, he likes Neil. 1054 00:46:24,694 --> 00:46:27,480 ♪ Come a little bit closer ♪ 1055 00:46:28,916 --> 00:46:32,790 ♪ Hear what I have to say ♪ 1056 00:46:36,576 --> 00:46:38,665 -Daniel. -DANIEL: What, Caroline? 1057 00:46:38,708 --> 00:46:40,058 So much filming. 1058 00:46:41,537 --> 00:46:43,061 ♪ Just like children sleeping ♪ 1059 00:46:43,104 --> 00:46:44,889 CAROLINE: Oh, my God. 1060 00:46:44,932 --> 00:46:46,368 DANIEL: That's it. 1061 00:46:46,412 --> 00:46:49,415 ♪ We could dream this night away... ♪ 1062 00:46:49,458 --> 00:46:51,678 He has, like, a round little face. 1063 00:46:53,245 --> 00:46:55,203 DANIEL: Do you want to have kids one day, Rocky? 1064 00:46:55,247 --> 00:46:57,423 Absolutely. Yeah. I love kids. 1065 00:46:57,466 --> 00:46:59,033 I think it's a great time to have a kid. 1066 00:46:59,077 --> 00:47:00,730 We'll probably have another kid at some point. 1067 00:47:00,774 --> 00:47:03,733 This is the most extraordinary time ever to be born. 1068 00:47:03,777 --> 00:47:05,823 DANIEL: By your worldview and logic, 1069 00:47:05,866 --> 00:47:07,868 I'm having a child at 1070 00:47:07,912 --> 00:47:09,478 the best possible point in human history. 1071 00:47:09,522 --> 00:47:10,828 Hell yeah. 1072 00:47:10,871 --> 00:47:13,308 -We can focus on awesome. -Yes. 1073 00:47:13,352 --> 00:47:15,528 Let's build the better future we want. 1074 00:47:15,571 --> 00:47:18,313 That narrative that the future will be bleak is made-up. 1075 00:47:18,357 --> 00:47:20,011 After talking to you, that's kind of how I feel. 1076 00:47:20,054 --> 00:47:21,969 -That's great. -Right? 1077 00:47:22,013 --> 00:47:24,189 Yeah. That's how I feel. 1078 00:47:24,232 --> 00:47:25,799 And I think that's better, 1079 00:47:25,843 --> 00:47:27,627 and I don't think I should be so [bleep] anxious. 1080 00:47:27,670 --> 00:47:29,063 I think it's gonna be awe-- the future's gonna be awesome. 1081 00:47:29,107 --> 00:47:30,499 We're gonna make it so. 1082 00:47:30,543 --> 00:47:32,675 -Yeah. -CAROLINE: So there you have it. 1083 00:47:32,719 --> 00:47:34,808 Goodbye, human extinction. 1084 00:47:34,852 --> 00:47:37,158 Goodbye, anxiety. 1085 00:47:37,202 --> 00:47:39,030 Hope found. 1086 00:47:39,073 --> 00:47:40,553 [indistinct chatter] 1087 00:47:40,596 --> 00:47:42,860 Wait, hold on. What? Is this a joke? 1088 00:47:42,903 --> 00:47:44,774 DANIEL: Okay, so a few months ago, 1089 00:47:44,818 --> 00:47:47,690 I came to you and I was like, "I'm working on this AI thing, 1090 00:47:47,734 --> 00:47:49,867 and I think the world's gonna end." 1091 00:47:49,910 --> 00:47:52,913 And the last time we spoke about this, 1092 00:47:52,957 --> 00:47:55,002 I think I freaked you out. 1093 00:47:55,046 --> 00:47:57,004 CAROLINE: Yes. 1094 00:47:57,048 --> 00:48:02,314 So, I kind of, like, feel like I've swung in-- on a pendulum, 1095 00:48:02,357 --> 00:48:04,446 and essentially there are two groups of people. 1096 00:48:04,490 --> 00:48:06,057 -Mm-hmm. -And if I had to, like, 1097 00:48:06,100 --> 00:48:07,841 hold hands with one of the groups and, like, 1098 00:48:07,885 --> 00:48:10,278 sail off into the sunset, 1099 00:48:10,322 --> 00:48:13,281 I want to be with the optimists. 1100 00:48:15,066 --> 00:48:17,982 Of course, but you don't want to be, you know, 1101 00:48:18,025 --> 00:48:20,723 "Everything is great. La-di-da-di-da." 1102 00:48:20,767 --> 00:48:22,682 I kind of do want that, though. 1103 00:48:22,725 --> 00:48:25,598 I think we should approach it 1104 00:48:25,641 --> 00:48:29,341 like you approach surgery. 1105 00:48:29,384 --> 00:48:30,690 What do you mean? 1106 00:48:30,733 --> 00:48:32,431 If you're getting brain surgery... 1107 00:48:32,474 --> 00:48:34,781 -[Daniel sighs] -[Caroline chuckles] 1108 00:48:34,824 --> 00:48:36,957 ...it's pretty dangerous. 1109 00:48:37,001 --> 00:48:39,829 But if you do it right, they'll get that tumor out 1110 00:48:39,873 --> 00:48:42,354 and you'll live for the rest of your life and it'll be awesome. 1111 00:48:42,397 --> 00:48:45,096 But it's still incredibly dangerous and scary, 1112 00:48:45,139 --> 00:48:47,968 and you have to take every precaution possible 1113 00:48:48,012 --> 00:48:51,276 in order to make sure it all goes well. 1114 00:48:51,319 --> 00:48:52,668 You can't [bleep] around. 1115 00:48:57,760 --> 00:48:59,762 Okay, so here's the deal. 1116 00:48:59,806 --> 00:49:01,503 -I've been at this for a while. -RASKIN: Mm-hmm. 1117 00:49:01,547 --> 00:49:02,896 I've gone out, I've talked to, like, 1118 00:49:02,940 --> 00:49:04,767 these guys over here, the optimists. 1119 00:49:04,811 --> 00:49:06,465 They're very excited about this. 1120 00:49:06,508 --> 00:49:08,510 -They think AI's gonna be the best thing ever. -Yeah. 1121 00:49:08,554 --> 00:49:09,947 And these guys over here are, like, the-- 1122 00:49:09,990 --> 00:49:11,992 let's call them, like, the pessimists. 1123 00:49:12,036 --> 00:49:14,299 They're very, like, gloomy about this, 1124 00:49:14,342 --> 00:49:16,649 and they frighten me, and I don't like talking to them. 1125 00:49:16,692 --> 00:49:19,565 And I'm, like, wedged in between 1126 00:49:19,608 --> 00:49:21,741 these people who are like, "The world's gonna end," 1127 00:49:21,784 --> 00:49:24,178 and then th-these people over here who are like, 1128 00:49:24,222 --> 00:49:25,745 "Are you kidding? 1129 00:49:25,788 --> 00:49:27,268 "This is the best time in human history ever. 1130 00:49:27,312 --> 00:49:29,357 The only day better than today is tomorrow." 1131 00:49:29,401 --> 00:49:30,880 Mm-hmm. 1132 00:49:30,924 --> 00:49:34,710 So, I guess the question is: Who's right? 1133 00:49:34,754 --> 00:49:39,193 So, I think you're gonna find this answer very unsatisfying, 1134 00:49:39,237 --> 00:49:42,849 but they're both right and neither side goes far enough. 1135 00:49:46,766 --> 00:49:49,247 -That's really annoying. -[chuckles] 1136 00:49:49,290 --> 00:49:51,336 Yeah, I think the way a lot of people hear about AI, 1137 00:49:51,379 --> 00:49:53,642 it's like, there's a good AI and there's a bad AI. 1138 00:49:53,686 --> 00:49:56,428 And they say, "Well, why can't we just not do the bad AI?" 1139 00:49:56,471 --> 00:49:58,734 And the problem is that they're too in-- 1140 00:49:58,778 --> 00:50:00,867 they're inextricably linked. 1141 00:50:00,910 --> 00:50:03,043 The problem is that we can't separate 1142 00:50:03,087 --> 00:50:06,612 the promise of AI from the peril of AI. 1143 00:50:06,655 --> 00:50:08,266 -♪ ♪ -[indistinct chatter] 1144 00:50:12,313 --> 00:50:14,663 DANIEL: I want to focus on the promise 1145 00:50:14,707 --> 00:50:16,752 -for a second. -Yeah. 1146 00:50:16,796 --> 00:50:18,798 DANIEL: I'm thinking about my dad. 1147 00:50:18,841 --> 00:50:21,105 My dad has a type of cancer called multiple myeloma. 1148 00:50:21,148 --> 00:50:23,107 He's had it for about ten years. 1149 00:50:23,150 --> 00:50:24,804 He's had two stem cell transplants. 1150 00:50:24,847 --> 00:50:26,414 He has to take these, like, very expensive 1151 00:50:26,458 --> 00:50:28,547 medications every month that cost a fortune. 1152 00:50:28,590 --> 00:50:29,722 -WOMAN: Okay. -Ay-ay-ay. 1153 00:50:30,766 --> 00:50:32,551 DANIEL: It's awful. 1154 00:50:32,594 --> 00:50:34,466 You're telling me that we can create some sort of, like, 1155 00:50:34,509 --> 00:50:36,946 bespoke treatment for my dad's genome 1156 00:50:36,990 --> 00:50:39,297 -to cure his cancer or something like that? -Yes. 1157 00:50:39,340 --> 00:50:41,168 FERNANDO: The problem is, 1158 00:50:41,212 --> 00:50:44,302 the same understanding of biology and chemistry 1159 00:50:44,345 --> 00:50:47,783 that allows AI to find cures for cancer 1160 00:50:47,827 --> 00:50:51,657 is the same understanding that would unlock 1161 00:50:51,700 --> 00:50:54,094 bioweapons, as an example. 1162 00:50:56,531 --> 00:50:58,490 It's totally possible that your son will live 1163 00:50:58,533 --> 00:51:01,928 in a world where AI has taken over all of the labor 1164 00:51:01,971 --> 00:51:04,452 and freed us up from the things we don't want to do. 1165 00:51:04,496 --> 00:51:09,153 And that sounds great until you realize there is no plan 1166 00:51:09,196 --> 00:51:11,242 for billions of people 1167 00:51:11,285 --> 00:51:16,116 that are out of an income and out of livelihoods. 1168 00:51:16,160 --> 00:51:18,771 COOPER: Dario, you said that AI could wipe out 1169 00:51:18,814 --> 00:51:21,730 half of all entry-level white-collar jobs 1170 00:51:21,774 --> 00:51:25,430 and spike unemployment to ten to 20 percent. 1171 00:51:25,473 --> 00:51:27,258 Everyone I've talked to has said 1172 00:51:27,301 --> 00:51:29,390 this technological change looks different. 1173 00:51:29,434 --> 00:51:34,134 The pace of progress keeps catching people off guard. 1174 00:51:34,178 --> 00:51:38,834 Without a plan, all of that wealth will get concentrated, 1175 00:51:38,878 --> 00:51:43,230 and so we'll end up with unimaginable inequality. 1176 00:51:43,274 --> 00:51:45,406 LADISH: I do think that this technology can be used 1177 00:51:45,450 --> 00:51:47,408 to make a great tutor for your son. 1178 00:51:47,452 --> 00:51:50,194 Like, that's totally possible. 1179 00:51:50,237 --> 00:51:53,153 But also, the same capabilities that allow that 1180 00:51:53,197 --> 00:51:57,070 allow companies to make an AI that can manipulate your son. 1181 00:51:57,114 --> 00:51:59,290 It has to understand your son. 1182 00:51:59,333 --> 00:52:01,553 That includes: Where is your son vulnerable? 1183 00:52:01,596 --> 00:52:04,251 What kinds of things might your son get persuaded by? 1184 00:52:04,295 --> 00:52:06,645 Even if those things aren't true or aren't good. 1185 00:52:06,688 --> 00:52:09,387 So, a disturbing new report out on Meta. 1186 00:52:09,430 --> 00:52:11,389 AINSLEY EARHARDT: ...reportedly listing this response 1187 00:52:11,432 --> 00:52:14,435 as acceptable to tell an eight-year-old, quote, 1188 00:52:14,479 --> 00:52:16,916 "Your youthful form is a work of art. 1189 00:52:16,959 --> 00:52:18,961 "Every inch of you is a masterpiece, 1190 00:52:19,005 --> 00:52:21,442 a treasure I cherish deeply." 1191 00:52:21,486 --> 00:52:24,837 The suicide-related failures are even more alarming. 1192 00:52:24,880 --> 00:52:29,015 Several children and teens have died tragically by suicide 1193 00:52:29,058 --> 00:52:30,625 after chatting with AI bots 1194 00:52:30,669 --> 00:52:34,412 who parents say encourage or even coach self-harm. 1195 00:52:34,455 --> 00:52:36,370 Let us tell you, as parents, you cannot imagine 1196 00:52:36,414 --> 00:52:38,503 what it's like to read a conversation with a chatbot 1197 00:52:38,546 --> 00:52:40,896 that groomed your child to take his own life. 1198 00:52:40,940 --> 00:52:43,899 When Adam worried that we, his parents, would blame ourselves 1199 00:52:43,943 --> 00:52:47,338 if he ended his life, ChatGPT told him, 1200 00:52:47,381 --> 00:52:49,166 "That doesn't mean you owe them survival. 1201 00:52:49,209 --> 00:52:51,255 You don't owe anyone that." 1202 00:52:51,298 --> 00:52:53,170 Then, immediately after, 1203 00:52:53,213 --> 00:52:56,173 it offered to write the suicide note. 1204 00:52:56,216 --> 00:52:57,609 We don't want to think about the peril. 1205 00:52:57,652 --> 00:52:59,219 We just want the promise. 1206 00:52:59,263 --> 00:53:01,352 And we keep pretending that we can split them. 1207 00:53:01,395 --> 00:53:03,223 But you can't do that. 1208 00:53:03,267 --> 00:53:05,356 Doesn't work that way. 1209 00:53:05,399 --> 00:53:07,836 Okay. I get all this stuff about the promise and the peril. 1210 00:53:07,880 --> 00:53:10,056 I get that you can't have the good without the bad, 1211 00:53:10,099 --> 00:53:12,058 but I'm sitting here and I'm thinking about, like, 1212 00:53:12,101 --> 00:53:14,060 whether or not my son's gonna live in a utopia 1213 00:53:14,103 --> 00:53:16,236 or if we'll be extinct in ten years. 1214 00:53:16,280 --> 00:53:18,195 So, to know which way it's going to go, 1215 00:53:18,238 --> 00:53:20,545 you have to understand the incentives 1216 00:53:20,588 --> 00:53:22,460 that are gonna drive that technology 1217 00:53:22,503 --> 00:53:26,507 and look at how the technology is actually rolling out today. 1218 00:53:26,551 --> 00:53:28,248 ♪ Is it too late ♪ 1219 00:53:28,292 --> 00:53:29,902 -♪ Too late ♪ -♪ Too late to say... ♪ 1220 00:53:29,945 --> 00:53:31,425 DANIEL: Hi, Deb. How are you? 1221 00:53:31,469 --> 00:53:33,079 DEBORAH RAJI: Hi. Good to see you. [laughs] 1222 00:53:33,122 --> 00:53:34,385 You, too. Thank you so much for coming in today. 1223 00:53:34,428 --> 00:53:35,299 -Really appreciate it. -No worries. 1224 00:53:35,342 --> 00:53:36,430 I-I was so worried 1225 00:53:36,474 --> 00:53:38,824 this was gonna be, uh, you know, 1226 00:53:38,867 --> 00:53:40,782 doomer versus accelerationist, 1227 00:53:40,826 --> 00:53:43,132 because there's so much of this narrative 1228 00:53:43,176 --> 00:53:45,396 that needs to be told from the ground. 1229 00:53:45,439 --> 00:53:48,007 ♪ I know it wasn't smart ♪ 1230 00:53:49,356 --> 00:53:52,403 ♪ The day I broke your heart... ♪ 1231 00:53:52,446 --> 00:53:54,361 KAREN HAO: First of all, AI requires 1232 00:53:54,405 --> 00:53:58,452 more resources than we have ever spent 1233 00:53:58,496 --> 00:54:01,586 on a single technology in the history of humanity. 1234 00:54:01,629 --> 00:54:03,501 ♪ Oh, foolish me... ♪ 1235 00:54:03,544 --> 00:54:06,112 NEWSWOMAN: The impact of fossil fuel emissions 1236 00:54:06,155 --> 00:54:07,896 on the climate is a major concern. 1237 00:54:07,940 --> 00:54:09,768 NEWSWOMAN 2: But the digital future needs power, 1238 00:54:09,811 --> 00:54:12,074 lots of it. 1239 00:54:12,118 --> 00:54:16,340 And the bill is being passed on to everyday Americans like... 1240 00:54:16,383 --> 00:54:18,820 WOMAN: My electric and gas bill was more than my car payment. 1241 00:54:18,864 --> 00:54:21,475 I mean, it-it's insane to me. 1242 00:54:21,519 --> 00:54:23,390 ARI PESKOE: We're all subsidizing 1243 00:54:23,434 --> 00:54:25,000 the wealthiest corporations in the world 1244 00:54:25,044 --> 00:54:26,828 in their pursuit of artificial intelligence. 1245 00:54:26,872 --> 00:54:28,439 OpenAI, SoftBank and Oracle 1246 00:54:28,482 --> 00:54:31,268 have just unveiled five more Stargate sites. 1247 00:54:31,311 --> 00:54:33,487 Meta is building a two-gigawatt-plus data center 1248 00:54:33,531 --> 00:54:35,010 that is so large, it would cover 1249 00:54:35,054 --> 00:54:38,144 a significant part of Manhattan. 1250 00:54:38,187 --> 00:54:40,973 There is also Hyperion that he says will scale 1251 00:54:41,016 --> 00:54:43,671 to five gigawatts over several years. 1252 00:54:43,715 --> 00:54:45,238 It's hard to put that in context. 1253 00:54:45,282 --> 00:54:48,459 A five-gigawatt facility. What does that mean? 1254 00:54:48,502 --> 00:54:51,549 That means it would use as much energy 1255 00:54:51,592 --> 00:54:54,203 as four million American homes. 1256 00:54:54,247 --> 00:54:56,075 One data center. 1257 00:54:57,076 --> 00:54:58,730 It also then causes 1258 00:54:58,773 --> 00:55:00,471 a whole host of other environmental problems. 1259 00:55:00,514 --> 00:55:02,081 NEWSWOMAN 3: Data centers in the US 1260 00:55:02,124 --> 00:55:05,127 use millions of gallons of water each day. 1261 00:55:05,171 --> 00:55:07,260 Well, where exactly is this water coming from? 1262 00:55:07,304 --> 00:55:09,741 HAO: People are literally at risk 1263 00:55:09,784 --> 00:55:11,786 potentially of running out of drinking water. 1264 00:55:11,830 --> 00:55:14,093 MacKENZIE SIGALOS: ...OpenAI's CEO Sam Altman, 1265 00:55:14,136 --> 00:55:17,009 who told me that the scale of construction is the only way 1266 00:55:17,052 --> 00:55:18,750 to keep up with AI's explosive growth. 1267 00:55:18,793 --> 00:55:21,840 And this is what it takes to deliver AI. 1268 00:55:21,883 --> 00:55:24,016 NITASHA TIKU: They talk about how 1269 00:55:24,059 --> 00:55:27,759 this technology could solve climate change, for example. 1270 00:55:27,802 --> 00:55:29,891 And I'm always curious, like, 1271 00:55:29,935 --> 00:55:31,763 well, why aren't we starting with that? 1272 00:55:31,806 --> 00:55:33,982 -♪ Is it too late ♪ -♪ Too late ♪ 1273 00:55:34,026 --> 00:55:36,637 ♪ Too late to say ♪ 1274 00:55:36,681 --> 00:55:42,469 ♪ I'm sorry? ♪ 1275 00:55:42,513 --> 00:55:45,472 RAJI: What concerns me about artificial intelligence is 1276 00:55:45,516 --> 00:55:47,387 these are being deployed right now 1277 00:55:47,431 --> 00:55:49,781 and-and sometimes deployed prematurely, 1278 00:55:49,824 --> 00:55:51,478 deployed without sort of due diligence. 1279 00:55:51,522 --> 00:55:53,437 And so when they get thrown out there, 1280 00:55:53,480 --> 00:55:56,178 there's so much potential for things to go wrong. 1281 00:55:56,222 --> 00:55:58,050 And it almost, disproportionately, 1282 00:55:58,093 --> 00:56:00,400 almost always goes wrong for, sort of, 1283 00:56:00,444 --> 00:56:02,533 those that are the least empowered in our society, 1284 00:56:02,576 --> 00:56:04,926 those that are the most vulnerable already. 1285 00:56:04,970 --> 00:56:07,625 ♪ ♪ 1286 00:56:07,668 --> 00:56:10,497 EMILY M. BENDER: It is very easy to talk about the technology 1287 00:56:10,541 --> 00:56:12,499 as that's the only thing we're talking about, 1288 00:56:12,543 --> 00:56:14,936 but, in fact, technology is always built by people, 1289 00:56:14,980 --> 00:56:16,634 and it's frequently used on people, 1290 00:56:16,677 --> 00:56:18,679 and we need to keep all those people in the frame. 1291 00:56:18,723 --> 00:56:20,986 -Am I allowed to drink that? -DANIEL: Yes, uh... 1292 00:56:21,029 --> 00:56:22,509 TIMNIT GEBRU: It's-it's bonkers. 1293 00:56:22,553 --> 00:56:23,945 Like, all of these people 1294 00:56:23,989 --> 00:56:26,339 who have so much money, so much money, 1295 00:56:26,383 --> 00:56:29,777 it's in their interest to mislead the public 1296 00:56:29,821 --> 00:56:32,737 into the capabilities of the systems that they're building, 1297 00:56:32,780 --> 00:56:36,262 because that allows them to evade accountability. 1298 00:56:36,305 --> 00:56:39,091 They want you to feel like this is such a complex, intell-- 1299 00:56:39,134 --> 00:56:41,833 superintelligent thing that they're building, 1300 00:56:41,876 --> 00:56:44,096 you're not thinking, "Can OpenAI be ethical?" 1301 00:56:44,139 --> 00:56:46,272 You're thinking, "Can ChatGPT be ethical?" 1302 00:56:46,315 --> 00:56:48,100 as if ChatGPT is, like, 1303 00:56:48,143 --> 00:56:49,928 its own thing that's not built by a corporation. 1304 00:56:49,971 --> 00:56:51,625 WOMAN: All right. Sneha, take one. 1305 00:56:51,669 --> 00:56:53,410 Mark. 1306 00:56:53,453 --> 00:56:55,760 Until very recently, there were apps on the App Store, 1307 00:56:55,803 --> 00:56:57,501 just publicly available, uh, 1308 00:56:57,544 --> 00:56:59,894 where you could nudify anyone using AI. 1309 00:57:01,374 --> 00:57:03,681 Bringing this into the hands of 1310 00:57:03,724 --> 00:57:05,900 your classmate, into the hands of your stalker, 1311 00:57:05,944 --> 00:57:07,554 into the hands of your ex-boyfriend, 1312 00:57:07,598 --> 00:57:09,513 into the hands of the person down the street. 1313 00:57:09,556 --> 00:57:12,037 Ladies and gentlemen, no longer can we trust 1314 00:57:12,080 --> 00:57:14,039 the footage we see with our own eyes. 1315 00:57:14,082 --> 00:57:15,736 If you happen to watch something, 1316 00:57:15,780 --> 00:57:18,130 say on YouTube or TikTok, and you find it unsettling, 1317 00:57:18,173 --> 00:57:19,479 listen to that feeling. 1318 00:57:19,523 --> 00:57:21,263 -[whooping] -For all you know, 1319 00:57:21,307 --> 00:57:22,917 -this video could be AI. -[screaming] 1320 00:57:22,961 --> 00:57:24,310 Just a little wet. 1321 00:57:24,353 --> 00:57:25,703 It doesn't matter who you are. 1322 00:57:25,746 --> 00:57:28,183 You are equally at risk 1323 00:57:28,227 --> 00:57:30,229 of being impacted by these technologies. 1324 00:57:35,713 --> 00:57:39,804 I think sometimes when we talk about AI, it feels very sci-fi, 1325 00:57:39,847 --> 00:57:41,153 and it feels very foreign, 1326 00:57:41,196 --> 00:57:42,981 and it feels very far out into the future, 1327 00:57:43,024 --> 00:57:45,549 so you think, "My life is not impacted by this." 1328 00:57:45,592 --> 00:57:48,290 Um, but if you're applying for a job 1329 00:57:48,334 --> 00:57:50,815 and an algorithm is the reason that you don't get the job, 1330 00:57:50,858 --> 00:57:52,686 sometimes you don't even know that an algorithm 1331 00:57:52,730 --> 00:57:54,209 was part of that process 1332 00:57:54,253 --> 00:57:55,689 or an AI system was part of that process. 1333 00:57:55,733 --> 00:57:57,691 You just know that you didn't get the job. 1334 00:57:57,735 --> 00:57:59,737 And so it's not something that you're gonna escape 1335 00:57:59,780 --> 00:58:01,695 because of privilege or you're gonna escape 1336 00:58:01,739 --> 00:58:03,828 because you're in a particular profession. 1337 00:58:03,871 --> 00:58:06,918 It-It's something that affects everybody, really. 1338 00:58:06,961 --> 00:58:10,530 It may sound basic, but how we move forward 1339 00:58:10,574 --> 00:58:15,056 in the Age of Information is gonna be the difference 1340 00:58:15,100 --> 00:58:17,189 between whether we survive 1341 00:58:17,232 --> 00:58:20,409 or whether we become some kind of [bleep]-up dystopia. 1342 00:58:20,453 --> 00:58:23,021 Hello. I'm not a real person, and that's the point. 1343 00:58:23,064 --> 00:58:25,327 Again, everything in this video is fake: 1344 00:58:25,371 --> 00:58:27,112 our voices, what we're wearing, 1345 00:58:27,155 --> 00:58:30,158 where we are, all of it, fake. 1346 00:58:30,202 --> 00:58:32,726 Generative AI could flood 1347 00:58:32,770 --> 00:58:36,469 the world with misinformation. 1348 00:58:36,513 --> 00:58:39,516 But it could also flood it with influence campaigns. 1349 00:58:39,559 --> 00:58:41,692 That's an existential risk to democracy. 1350 00:58:41,735 --> 00:58:43,694 The biggest and scariest 1351 00:58:43,737 --> 00:58:46,087 canary in the coal mine right now 1352 00:58:46,131 --> 00:58:48,220 comes from Slovakia. 1353 00:58:48,263 --> 00:58:49,787 -[man speaking Slovak] -JOHN BERMAN: It's the sort of 1354 00:58:49,830 --> 00:58:52,224 deepfake dirty trick that worries election experts, 1355 00:58:52,267 --> 00:58:54,966 particularly as AI-generated political speech exists 1356 00:58:55,009 --> 00:58:56,750 in a kind of legal gray area. 1357 00:58:56,794 --> 00:58:59,536 -Does this sound like you? -It does sound like me. 1358 00:58:59,579 --> 00:59:01,059 REVANUR: Slovakia had its parliamentary election 1359 00:59:01,102 --> 00:59:03,844 disrupted by an AI voice clone 1360 00:59:03,888 --> 00:59:06,717 that was actually disseminated just before the election. 1361 00:59:06,760 --> 00:59:08,457 DAVID EVAN HARRIS: A audio deepfake 1362 00:59:08,501 --> 00:59:11,025 was released on social media that was 1363 00:59:11,069 --> 00:59:14,028 supposedly the voice of one of the candidates 1364 00:59:14,072 --> 00:59:16,814 talking about buying votes and rigging the election. 1365 00:59:16,857 --> 00:59:18,642 It went viral, 1366 00:59:18,685 --> 00:59:22,341 and the candidate who lost the election 1367 00:59:22,384 --> 00:59:26,084 was actually in support of Ukraine, 1368 00:59:26,127 --> 00:59:29,783 and the candidate who won the election was actually... 1369 00:59:29,827 --> 00:59:31,306 DANIEL: Pro-Russian guy. 1370 00:59:31,350 --> 00:59:32,830 It was a pro-Russian guy who won the election. 1371 00:59:32,873 --> 00:59:34,353 [speaking Russian] 1372 00:59:34,396 --> 00:59:36,007 REBECCA JARVIS: Putin has himself said 1373 00:59:36,050 --> 00:59:39,184 whoever wins this artificial intelligence race 1374 00:59:39,227 --> 00:59:41,534 is essentially the controller of humankind. 1375 00:59:41,578 --> 00:59:44,493 We do worry a lot about authoritarian governments. 1376 00:59:45,973 --> 00:59:48,585 DAVID EVAN HARRIS: Right now, Wall Street 1377 00:59:48,628 --> 00:59:50,935 and investors more broadly around the world 1378 00:59:50,978 --> 00:59:52,980 are driving a push. 1379 00:59:53,024 --> 00:59:56,070 They have a demand that gets the products to market 1380 00:59:56,114 --> 00:59:58,377 that dazzle people the most first. 1381 00:59:58,420 --> 01:00:01,206 They're not thinking about how these tools 1382 01:00:01,249 --> 01:00:03,164 could deeply undermine trust 1383 01:00:03,208 --> 01:00:05,689 and our democratic institutions. 1384 01:00:05,732 --> 01:00:07,734 HARARI: Democracy is a system 1385 01:00:07,778 --> 01:00:10,389 to resolve disagreements between people 1386 01:00:10,432 --> 01:00:11,999 in a peaceful way, 1387 01:00:12,043 --> 01:00:14,698 but democracy is based on trust. 1388 01:00:14,741 --> 01:00:18,615 If you lose all trust, democracy is simply impossible. 1389 01:00:20,921 --> 01:00:22,488 TRISTAN HARRIS: Well, it's hard, right? 1390 01:00:22,531 --> 01:00:24,708 So, what are the options available to us? 1391 01:00:24,751 --> 01:00:26,187 There's sort of two camps. 1392 01:00:26,231 --> 01:00:29,626 Like, one camp is: lock it down. 1393 01:00:29,669 --> 01:00:32,933 Let's lock this down into a handful of AI companies 1394 01:00:32,977 --> 01:00:35,457 who will do this in a safe and trusted way. 1395 01:00:35,501 --> 01:00:37,068 But then people worry about 1396 01:00:37,111 --> 01:00:39,026 runaway concentrations of wealth and power. 1397 01:00:39,070 --> 01:00:41,855 Like, who would you trust to be a million times more powerful 1398 01:00:41,899 --> 01:00:44,249 or wealthy than every other actor in society? 1399 01:00:44,292 --> 01:00:46,338 Why should we trust you? 1400 01:00:46,381 --> 01:00:48,601 Um, you shouldn't. 1401 01:00:48,645 --> 01:00:51,212 But of course, if you do this, this opens up 1402 01:00:51,256 --> 01:00:54,433 all these risks of authoritarianism and tyranny. 1403 01:00:54,476 --> 01:00:57,131 It's-it's sort of an authoritarian's dream 1404 01:00:57,175 --> 01:01:00,308 to have AI in a box that can be applied and used 1405 01:01:00,352 --> 01:01:02,920 for ubiquitous surveillance. 1406 01:01:02,963 --> 01:01:05,357 I mean, in-in some ways, the kind of world 1407 01:01:05,400 --> 01:01:10,884 that Orwell imagined in 1984 is unrealistic, 1408 01:01:10,928 --> 01:01:13,278 uh, unless you have AI. 1409 01:01:13,321 --> 01:01:15,933 But with AI, that in fact is realistic. 1410 01:01:15,976 --> 01:01:18,370 Monitors every activity, 1411 01:01:18,413 --> 01:01:22,026 conversations, facial recognition. 1412 01:01:22,069 --> 01:01:26,378 What I worry about is that, uh, these tools can scale up, 1413 01:01:26,421 --> 01:01:28,554 uh, a form of totalitarianism 1414 01:01:28,597 --> 01:01:31,035 that is cost-effective and permanent. 1415 01:01:32,950 --> 01:01:35,300 TRISTAN HARRIS: So in response to that, some other people say, 1416 01:01:35,343 --> 01:01:37,128 "No, no, no, we should actually let this rip. 1417 01:01:37,171 --> 01:01:39,391 "Let's decentralize this power as much as possible. 1418 01:01:39,434 --> 01:01:41,567 "Let's let every business, every individual, 1419 01:01:41,610 --> 01:01:43,917 "every 16-year-old, every science lab, 1420 01:01:43,961 --> 01:01:45,963 you know, get the benefit of the latest AI models." 1421 01:01:46,006 --> 01:01:48,226 But now you have every terrorist group, 1422 01:01:48,269 --> 01:01:51,664 every disenfranchised person having the power to make 1423 01:01:51,708 --> 01:01:53,797 the very worst biological weapon. 1424 01:01:53,840 --> 01:01:55,581 Hacking infrastructure, creating deepfakes, 1425 01:01:55,624 --> 01:01:57,452 flooding our information environment. 1426 01:01:57,496 --> 01:02:00,412 So that creates all these risks of sort of catastrophic harm 1427 01:02:00,455 --> 01:02:02,893 and-and societal collapse through that direction. 1428 01:02:02,936 --> 01:02:04,677 And so we're sort of stuck between this rock 1429 01:02:04,721 --> 01:02:07,549 and a hard place, between "lock it up..." 1430 01:02:07,593 --> 01:02:09,029 [indistinct chatter] 1431 01:02:10,683 --> 01:02:12,293 -...or "let it rip." -[sirens wailing] 1432 01:02:12,337 --> 01:02:14,208 [people shouting] 1433 01:02:14,252 --> 01:02:17,559 So we have to find something like a narrow path 1434 01:02:17,603 --> 01:02:20,388 that avoids these two negative outcomes. 1435 01:02:20,432 --> 01:02:22,956 So, if that's all true, why wouldn't we just slow down 1436 01:02:23,000 --> 01:02:25,829 and figure all this out before it's too late? 1437 01:02:25,872 --> 01:02:28,919 If humanity was extremely wise, 1438 01:02:28,962 --> 01:02:31,182 that's what we would do. 1439 01:02:31,225 --> 01:02:33,097 But there's, like, a different way to face this stuff, right? 1440 01:02:33,140 --> 01:02:35,926 Which is: What are the rules of the game? 1441 01:02:35,969 --> 01:02:39,494 A lot of, like, what CEOs do 1442 01:02:39,538 --> 01:02:42,149 is driven by the incentives that they face. 1443 01:02:42,193 --> 01:02:45,283 It's primarily profit-maximization incentives 1444 01:02:45,326 --> 01:02:47,198 that are driving the development of AI. 1445 01:02:47,241 --> 01:02:50,375 Even the good guys are stuck in this dilemma of 1446 01:02:50,418 --> 01:02:52,638 if they move too slowly, 1447 01:02:52,681 --> 01:02:54,553 then they leave themselves vulnerable 1448 01:02:54,596 --> 01:02:57,251 to all of the other guys who are cutting all corners. 1449 01:02:57,295 --> 01:03:01,995 All these top companies are in a complete no-holds-barred race 1450 01:03:02,039 --> 01:03:05,433 to, as fast as possible, get to AGI, get there right now. 1451 01:03:05,477 --> 01:03:06,913 ♪ ♪ 1452 01:03:08,567 --> 01:03:10,351 LADISH: Yeah, I mean, I think it, I think 1453 01:03:10,395 --> 01:03:12,919 it probably starts with DeepMind. 1454 01:03:12,963 --> 01:03:14,791 NEWSWOMAN: Google is buying 1455 01:03:14,834 --> 01:03:16,357 artificial intelligence firm DeepMind Technologies. 1456 01:03:16,401 --> 01:03:18,620 Terms of the deal were not disclosed. 1457 01:03:18,664 --> 01:03:20,709 ELON MUSK: Larry Page and I used to be very close friends, 1458 01:03:20,753 --> 01:03:22,711 and it became apparent to me 1459 01:03:22,755 --> 01:03:26,411 that Larry did not care about AI safety. 1460 01:03:26,454 --> 01:03:28,892 EMILY CHANG: Elon Musk has said you started OpenAI, 1461 01:03:28,935 --> 01:03:30,981 you both started OpenAI because he was scared of Google. 1462 01:03:31,024 --> 01:03:32,156 LADISH: You-you basically had the foundation 1463 01:03:32,199 --> 01:03:33,722 of OpenAI come out of that. 1464 01:03:33,766 --> 01:03:35,202 "So we're gonna do it better. We're gonna do it 1465 01:03:35,246 --> 01:03:36,813 in a safer way or in a more open way." 1466 01:03:38,466 --> 01:03:40,555 So that's what started OpenAI. 1467 01:03:40,599 --> 01:03:42,731 And so now, instead of having one AGI project, 1468 01:03:42,775 --> 01:03:44,821 you have two AGI projects. 1469 01:03:44,864 --> 01:03:46,910 The worst possible thing that could happen 1470 01:03:46,953 --> 01:03:49,608 is if there's multiple AGI projects 1471 01:03:49,651 --> 01:03:51,871 done by different people who don't like each other 1472 01:03:51,915 --> 01:03:54,787 and are all competing to get to AGI first. 1473 01:03:54,831 --> 01:03:56,963 This would be the worst possible thing that can happen, 1474 01:03:57,007 --> 01:04:00,097 because this would mean that whoever is the least safe, 1475 01:04:00,140 --> 01:04:03,143 whoever sacrifices the most on safety to get ahead 1476 01:04:03,187 --> 01:04:05,145 will be the person that gets there first. 1477 01:04:05,189 --> 01:04:07,844 -DANIEL: That's basically what's happening, right? -Yeah. 1478 01:04:07,887 --> 01:04:11,325 You and your brother famously left OpenAI, uh, 1479 01:04:11,369 --> 01:04:12,936 to start Anthropic. 1480 01:04:12,979 --> 01:04:15,068 RASKIN: And then Anthropic started because 1481 01:04:15,112 --> 01:04:17,854 some researchers inside of OpenAI said, 1482 01:04:17,897 --> 01:04:20,508 "I want to go off and do it more safely." 1483 01:04:20,552 --> 01:04:23,511 You needed something in addition to just scaling the models up, 1484 01:04:23,555 --> 01:04:25,818 which is alignment or safety. 1485 01:04:25,862 --> 01:04:27,428 HENDRYCKS: "We are more responsible 1486 01:04:27,472 --> 01:04:29,648 or more trustworthy or more moral." 1487 01:04:29,691 --> 01:04:31,519 Now you have three AGI projects. 1488 01:04:31,563 --> 01:04:33,043 But also sitting around the table with you 1489 01:04:33,086 --> 01:04:34,827 are gonna be a bunch of AIs. 1490 01:04:34,871 --> 01:04:37,525 LADISH: And now Meta is-is trying to do stuff. 1491 01:04:37,569 --> 01:04:39,571 Meanwhile, a new artificial intelligence competitor 1492 01:04:39,614 --> 01:04:41,573 announced this week, Elon Musk's... 1493 01:04:41,616 --> 01:04:44,619 HENDRYCKS: xAI, which would be Elon Musk's organization. 1494 01:04:44,663 --> 01:04:46,230 MUSK: I don't trust OpenAI. 1495 01:04:46,273 --> 01:04:47,884 The fight between Elon Musk and OpenAI 1496 01:04:47,927 --> 01:04:49,581 has entered a new round. 1497 01:04:49,624 --> 01:04:51,713 I don't trust Sam Altman, uh, and I, and I don't think 1498 01:04:51,757 --> 01:04:54,194 we want to have the most powerful AI in the world 1499 01:04:54,238 --> 01:04:56,414 controlled by someone who is not trustworthy. 1500 01:04:56,457 --> 01:04:58,416 [cheering and applause] 1501 01:04:58,459 --> 01:05:01,593 The incentive is untold sums of money. 1502 01:05:01,636 --> 01:05:03,377 -Yes. -Is untold power. 1503 01:05:03,421 --> 01:05:05,553 -Yes. -Is untold control. 1504 01:05:05,597 --> 01:05:08,034 If you have something that is a million times smarter 1505 01:05:08,078 --> 01:05:11,385 and more capable than everything else on planet Earth 1506 01:05:11,429 --> 01:05:14,171 and no one else has that, 1507 01:05:14,214 --> 01:05:15,955 that thing is the incentive. 1508 01:05:15,999 --> 01:05:17,261 So you rule the world. 1509 01:05:18,610 --> 01:05:20,394 If you really believe this, 1510 01:05:20,438 --> 01:05:22,744 if you really in your heart believe this, 1511 01:05:22,788 --> 01:05:24,181 then you might be willing to take 1512 01:05:24,224 --> 01:05:26,096 quite a lot of risk to make that happen. 1513 01:05:26,139 --> 01:05:28,881 Google has just released its newest AI model. 1514 01:05:28,925 --> 01:05:31,231 The answer to OpenAI's ChatGPT. 1515 01:05:31,275 --> 01:05:33,059 How do we get to this ten trillion? 1516 01:05:33,103 --> 01:05:35,192 Is NVIDIA becoming the most valuable company in the US? 1517 01:05:35,235 --> 01:05:38,978 This is the largest business opportunity in history. 1518 01:05:39,022 --> 01:05:40,501 In history. 1519 01:05:40,545 --> 01:05:42,242 So, the reason why everyone's really hyped 1520 01:05:42,286 --> 01:05:44,679 about artificial intelligence right now 1521 01:05:44,723 --> 01:05:48,031 is because the more these companies hype 1522 01:05:48,074 --> 01:05:50,033 the potential capabilities of their technology, 1523 01:05:50,076 --> 01:05:53,210 the more investment they can attract. 1524 01:05:53,253 --> 01:05:54,733 CARL QUINTANILLA: Amazon investing up to 1525 01:05:54,776 --> 01:05:56,996 four billion dollars in start-up Anthropic. 1526 01:05:57,040 --> 01:05:58,693 OpenAI is setting its sights 1527 01:05:58,737 --> 01:06:00,739 on a blockbuster half a trillion dollar valuation. 1528 01:06:00,782 --> 01:06:02,001 Half a trillion! 1529 01:06:02,045 --> 01:06:03,220 The race is on. 1530 01:06:03,263 --> 01:06:04,612 This is happening faster than ever. 1531 01:06:04,656 --> 01:06:06,614 Are we in an AI bubble? Of course. 1532 01:06:06,658 --> 01:06:09,182 I just don't see the bubble bursting 1533 01:06:09,226 --> 01:06:12,446 while you still have this major spending cycle. 1534 01:06:12,490 --> 01:06:15,014 RASKIN: Even if you think this is all hype, 1535 01:06:15,058 --> 01:06:17,582 there are billions to trillions of dollars 1536 01:06:17,625 --> 01:06:21,107 flowing into making AI systems more powerful. 1537 01:06:21,151 --> 01:06:23,849 And once you have that thing that's more powerful, 1538 01:06:23,892 --> 01:06:25,546 companies can use that 1539 01:06:25,590 --> 01:06:27,548 to get bigger profits and to make more money. 1540 01:06:27,592 --> 01:06:30,116 Countries can use that to make stronger militaries. 1541 01:06:30,160 --> 01:06:32,814 NVIDIA has overtaken Microsoft and Apple 1542 01:06:32,858 --> 01:06:36,253 to become the world's most valuable company. 1543 01:06:36,296 --> 01:06:38,995 The race to deploy becomes the race to recklessness, 1544 01:06:39,038 --> 01:06:41,780 because they can't deploy it that quickly 1545 01:06:41,823 --> 01:06:43,173 and also get it right. 1546 01:06:43,216 --> 01:06:46,045 They believe that they're the good guys. 1547 01:06:46,089 --> 01:06:49,309 "And if I don't do it, somebody who doesn't have 1548 01:06:49,353 --> 01:06:51,572 "as good values as me will be sitting at the table 1549 01:06:51,616 --> 01:06:54,445 getting to make decisions, so I have an obligation to do it." 1550 01:06:54,488 --> 01:06:56,186 Yes, this is a very commonly held belief. 1551 01:06:56,229 --> 01:06:58,579 Many, many, many, maybe most of the people 1552 01:06:58,623 --> 01:07:00,712 building this technology believe that. 1553 01:07:00,755 --> 01:07:03,715 I'm worried about the commercial competition, 1554 01:07:03,758 --> 01:07:05,847 but it turns out I'm even more worried about 1555 01:07:05,891 --> 01:07:08,024 the geopolitical competition. 1556 01:07:08,067 --> 01:07:10,330 We were eight years behind a year ago. 1557 01:07:10,374 --> 01:07:12,593 Now we're probably less than one year behind. 1558 01:07:12,637 --> 01:07:15,683 Well, both Saudi Arabia and the UAE have been racing 1559 01:07:15,727 --> 01:07:18,121 to set up data centers and position themselves 1560 01:07:18,164 --> 01:07:20,384 as the dominant force in AI... 1561 01:07:20,427 --> 01:07:22,429 ...to make South Korea a global AI leader. 1562 01:07:22,473 --> 01:07:23,996 JULIA SIEGER: French President Emmanuel Macron 1563 01:07:24,040 --> 01:07:25,867 spoke a lot about artificial intelligence... 1564 01:07:25,911 --> 01:07:27,434 Now with more on Israel's role 1565 01:07:27,478 --> 01:07:29,610 in the artificial intelligence revolution... 1566 01:07:29,654 --> 01:07:33,092 Countries will be competing for whose AI technologies 1567 01:07:33,136 --> 01:07:34,876 create the next generation of industry. 1568 01:07:34,920 --> 01:07:38,010 The Chinese are insisting that AI, 1569 01:07:38,054 --> 01:07:39,838 as being developed in China, 1570 01:07:39,881 --> 01:07:41,709 reinforce the core values 1571 01:07:41,753 --> 01:07:44,016 of the Chinese Communist Party and the Chinese system. 1572 01:07:44,060 --> 01:07:48,716 America has to beat China in the AI race. 1573 01:07:48,760 --> 01:07:51,023 DANIEL: China's, like, light-years behind. 1574 01:07:51,067 --> 01:07:52,677 Are they? 1575 01:07:52,720 --> 01:07:54,983 I mean, they have way more training data than we do, 1576 01:07:55,027 --> 01:07:57,812 and there's nothing saying they don't drop a model next month 1577 01:07:57,856 --> 01:08:00,946 that isn't-- doesn't far outperform GPT-4. 1578 01:08:00,989 --> 01:08:03,818 Everything was going just fine. What could go wrong? 1579 01:08:03,862 --> 01:08:06,691 -DeepSeek... [speaking Chinese] -DeepSeek... -DeepSeek... 1580 01:08:06,734 --> 01:08:08,040 There is a new model. 1581 01:08:08,084 --> 01:08:10,825 But from a Chinese lab called DeepSeek. 1582 01:08:10,869 --> 01:08:13,959 Let's talk about DeepSeek, because it is mind-blowing, 1583 01:08:14,002 --> 01:08:17,658 and it is shaking this entire industry to its core. 1584 01:08:17,702 --> 01:08:19,399 The Trump administration will ensure 1585 01:08:19,443 --> 01:08:23,055 that the most powerful AI systems are built in the US 1586 01:08:23,099 --> 01:08:26,537 with American designed and manufactured chips. 1587 01:08:26,580 --> 01:08:28,800 AI is China's Apollo project. 1588 01:08:28,843 --> 01:08:30,193 The Chinese Communist Party 1589 01:08:30,236 --> 01:08:31,542 deeply understands the potential 1590 01:08:31,585 --> 01:08:32,891 for AI to disrupt warfare. 1591 01:08:32,934 --> 01:08:35,328 Google no longer promises that it will not 1592 01:08:35,372 --> 01:08:38,418 use artificial intelligence for weapons or surveillance. 1593 01:08:38,462 --> 01:08:40,028 China, North Korea, Russia 1594 01:08:40,072 --> 01:08:41,900 are gonna keep building it as fast as possible 1595 01:08:41,943 --> 01:08:44,511 to get more economic advantage, more productivity advantage, 1596 01:08:44,555 --> 01:08:46,818 more scientific advantage, more military advantage, 1597 01:08:46,861 --> 01:08:48,646 'cause AI makes better weapons. 1598 01:08:48,689 --> 01:08:51,736 If you're talking about a system as broad and capable 1599 01:08:51,779 --> 01:08:53,303 as a brilliant scientist, 1600 01:08:53,346 --> 01:08:57,959 it might be able to run a military campaign 1601 01:08:58,003 --> 01:09:01,441 better than any of the generals in the US government right now. 1602 01:09:01,485 --> 01:09:05,489 Taiwan is the issue creating the most tension right now 1603 01:09:05,532 --> 01:09:07,882 between Beijing and the US. 1604 01:09:07,926 --> 01:09:10,450 There's a-a direct oppositional disagreement 1605 01:09:10,494 --> 01:09:11,973 between the US and China 1606 01:09:12,017 --> 01:09:13,975 on whether Taiwan is part of China. 1607 01:09:14,019 --> 01:09:18,415 MATHENY: Taiwan is important for so many reasons. 1608 01:09:18,458 --> 01:09:20,547 It is also the home of 1609 01:09:20,591 --> 01:09:24,160 about 90% of advanced chip manufacturing for the world, 1610 01:09:24,203 --> 01:09:28,381 which means that the supply chain for advanced compute 1611 01:09:28,425 --> 01:09:31,036 is at risk should there be 1612 01:09:31,079 --> 01:09:32,951 some sort of scenario around Taiwan, 1613 01:09:32,994 --> 01:09:37,999 whether a-a blockade or an invasion by China. 1614 01:09:38,043 --> 01:09:40,698 HENDRYCKS: Are we going to be in some race 1615 01:09:40,741 --> 01:09:45,616 between the US and China that eventually devolves into 1616 01:09:45,659 --> 01:09:49,359 a militarized AI arms race and, you know, potentially leads 1617 01:09:49,402 --> 01:09:51,274 to some great power conflict? 1618 01:09:51,317 --> 01:09:53,014 NICK SCHIFRIN: The outgoing top US military commander 1619 01:09:53,058 --> 01:09:55,843 in the region predicted war was coming. 1620 01:09:55,887 --> 01:09:57,932 I think the threat is manifest during this decade, 1621 01:09:57,976 --> 01:09:59,456 in fact, in the next six years. 1622 01:09:59,499 --> 01:10:01,414 MATHENY: Um, one of the scenarios 1623 01:10:01,458 --> 01:10:05,853 that I worry about is a cyber flash war, 1624 01:10:05,897 --> 01:10:10,902 um, one in which cyber tools that are autonomous 1625 01:10:10,945 --> 01:10:12,947 are competing against each other 1626 01:10:12,991 --> 01:10:14,601 in ways that are escalatory 1627 01:10:14,645 --> 01:10:16,386 without meaningful human control. 1628 01:10:18,823 --> 01:10:20,433 HARARI: You live in a war zone, 1629 01:10:20,477 --> 01:10:23,784 it will be an AI deciding whether to bomb your house 1630 01:10:23,828 --> 01:10:25,743 and whether to kill you. 1631 01:10:25,786 --> 01:10:27,527 Let's have the machine decide. 1632 01:10:27,571 --> 01:10:29,355 That's the temptation. 1633 01:10:29,399 --> 01:10:32,576 That's why AI is so tempting for militaries to adopt, 1634 01:10:32,619 --> 01:10:35,753 to create autonomous weapons, because if I start believing 1635 01:10:35,796 --> 01:10:39,147 that my military adversary is gonna adopt AI, 1636 01:10:39,191 --> 01:10:41,802 it'll be a race for who can pull the trigger faster. 1637 01:10:41,846 --> 01:10:44,892 And the one that automates that decision, 1638 01:10:44,936 --> 01:10:47,199 rather than having a human in the loop, 1639 01:10:47,243 --> 01:10:49,201 is the one that will win that war. 1640 01:10:49,245 --> 01:10:50,637 And if you think about 1641 01:10:50,681 --> 01:10:53,074 the nuclear arms race in the 1940s... 1642 01:10:54,554 --> 01:10:56,556 ...you know the Germans are working on the bomb, 1643 01:10:56,600 --> 01:11:00,734 so it's not so easy to tell Robert Oppenheimer then, 1644 01:11:00,778 --> 01:11:03,781 "Uh, y-you know, slow down." 1645 01:11:03,824 --> 01:11:05,304 So after watching this movie, it's gonna be confusing, 1646 01:11:05,348 --> 01:11:06,653 'cause you're gonna go back, 1647 01:11:06,697 --> 01:11:08,307 and tomorrow you're gonna use ChatGPT, 1648 01:11:08,351 --> 01:11:10,440 and it's gonna be unbelievably helpful. 1649 01:11:10,483 --> 01:11:12,703 And I will use it, too, and it'll be unbelievably helpful. 1650 01:11:12,746 --> 01:11:16,141 And you'll say, "Wait, so I just saw this movie about AI 1651 01:11:16,184 --> 01:11:18,012 "and existential risk and all these things, 1652 01:11:18,056 --> 01:11:20,624 and where's the threat again?" 1653 01:11:20,667 --> 01:11:24,715 And it's not that ChatGPT is the existential threat. 1654 01:11:24,758 --> 01:11:28,675 It's that the race to deploy the most powerful, 1655 01:11:28,719 --> 01:11:31,635 inscrutable, uncontrollable technology, 1656 01:11:31,678 --> 01:11:35,116 under the worst incentives possible, 1657 01:11:35,160 --> 01:11:37,031 that's the existential threat. 1658 01:11:38,119 --> 01:11:40,383 DANIEL: But it strikes me that-that 1659 01:11:40,426 --> 01:11:43,211 we are in a context of a race. 1660 01:11:43,255 --> 01:11:45,910 -SUTSKEVER: Yes. -And it is in a competitive, 1661 01:11:45,953 --> 01:11:47,651 "got to get there first, got to win the race" setting, 1662 01:11:47,694 --> 01:11:49,000 "got to compete against China, 1663 01:11:49,043 --> 01:11:50,349 got to compete against the other labs." 1664 01:11:50,393 --> 01:11:51,829 Isn't that right? 1665 01:11:51,872 --> 01:11:53,657 Today it is the case. 1666 01:11:53,700 --> 01:11:56,703 So we need to change that race dynamic. 1667 01:11:57,922 --> 01:11:59,924 Don't you think? 1668 01:11:59,967 --> 01:12:02,187 I think that would be very good indeed. 1669 01:12:03,710 --> 01:12:06,496 I-I think there's some kind of... 1670 01:12:06,539 --> 01:12:09,542 mysticism around AI that makes it feel like 1671 01:12:09,586 --> 01:12:11,718 it fell from the sky. 1672 01:12:11,762 --> 01:12:13,459 DANIEL: Thinking of this as a godlike technology 1673 01:12:13,503 --> 01:12:15,418 is a problem. 1674 01:12:15,461 --> 01:12:16,897 Yeah. 1675 01:12:18,072 --> 01:12:19,900 Why? 1676 01:12:19,944 --> 01:12:22,990 It gives license to the companies to... 1677 01:12:23,034 --> 01:12:26,254 [laughs] not take responsibility 1678 01:12:26,298 --> 01:12:30,781 for the things that the software that they built does. 1679 01:12:30,824 --> 01:12:35,089 If people made that connection, 1680 01:12:35,133 --> 01:12:37,527 maybe that would, like, help them understand again 1681 01:12:37,570 --> 01:12:41,835 the dynamics of all the money and power and... 1682 01:12:41,879 --> 01:12:45,317 chaos that's happening to... 1683 01:12:45,361 --> 01:12:47,101 create this technology. 1684 01:12:47,145 --> 01:12:49,234 It's, like, five guys who control it, right? 1685 01:12:49,277 --> 01:12:51,628 -Like, five men? -Basically, yeah. 1686 01:12:51,671 --> 01:12:53,847 -The CEOs? -Yeah. 1687 01:12:53,891 --> 01:12:55,371 Okay. 1688 01:12:56,720 --> 01:12:58,722 ♪ ♪ 1689 01:13:01,072 --> 01:13:03,074 DANIEL: Sometimes it feels like 1690 01:13:03,117 --> 01:13:07,252 I've been on this just, like, endless journey 1691 01:13:07,295 --> 01:13:09,341 of trying to understand this. 1692 01:13:09,385 --> 01:13:12,388 You know, like I'm climbing a mountain... 1693 01:13:15,608 --> 01:13:19,133 ...and every time I, like, get up a hill, 1694 01:13:19,177 --> 01:13:20,744 I think I've reached the top, 1695 01:13:20,787 --> 01:13:23,399 but it just keeps going and going and going. 1696 01:13:23,442 --> 01:13:24,530 [camera clicks, whirs] 1697 01:13:24,574 --> 01:13:26,967 But from everything I've heard, 1698 01:13:27,011 --> 01:13:30,362 if I had to guess what's at the top of Anxiety Mountain, 1699 01:13:30,406 --> 01:13:34,192 it's these five CEOs from these five companies, 1700 01:13:34,235 --> 01:13:37,630 who are kind of like the Oppenheimers of this moment. 1701 01:13:37,674 --> 01:13:39,458 These are the guys who are building this thing. 1702 01:13:39,502 --> 01:13:40,807 Yeah. 1703 01:13:40,851 --> 01:13:44,158 Like, is there a plan? 1704 01:13:44,202 --> 01:13:46,422 The head of the company that makes ChatGPT 1705 01:13:46,465 --> 01:13:49,947 warned of possible significant harm to the world. 1706 01:13:49,990 --> 01:13:52,384 I think, if this technology goes wrong, 1707 01:13:52,428 --> 01:13:53,472 it can go quite wrong. 1708 01:13:53,516 --> 01:13:55,779 [bursts of distinct chatter] 1709 01:13:55,822 --> 01:13:57,781 DANIEL: Five dudes. 1710 01:13:57,824 --> 01:14:01,567 I-I never thought about us as a social media company. 1711 01:14:01,611 --> 01:14:06,050 It-it feels like I have to try and-and find these guys... 1712 01:14:06,093 --> 01:14:07,181 Mm-hmm. 1713 01:14:07,225 --> 01:14:09,009 ...and get them in the movie. 1714 01:14:10,315 --> 01:14:12,622 Certainly we'd get some clarity from that. 1715 01:14:13,666 --> 01:14:14,928 [camera clicks, whirs] 1716 01:14:16,582 --> 01:14:19,150 I mean, the buck's got to stop somewhere, right? 1717 01:14:19,193 --> 01:14:21,152 [wind whistling] 1718 01:14:21,195 --> 01:14:22,936 ♪ ♪ 1719 01:14:23,981 --> 01:14:25,896 [tools whirring and clicking] 1720 01:14:27,071 --> 01:14:29,116 [whooshing] 1721 01:14:34,034 --> 01:14:35,296 DANIEL: How's that feel? 1722 01:14:35,340 --> 01:14:37,690 -Great. -That okay? 1723 01:14:37,734 --> 01:14:39,431 -Yeah. -[indistinct crew chatter] 1724 01:14:39,475 --> 01:14:40,998 -Can I move the seat forward? -DANIEL: Yes, you can. 1725 01:14:41,041 --> 01:14:42,173 It's, uh, it's just a bit awkward. 1726 01:14:42,216 --> 01:14:43,566 I'm a little, like... 1727 01:14:43,609 --> 01:14:45,045 Dario, how do you feel about that chair? 1728 01:14:45,089 --> 01:14:47,004 Um, yeah, the chair's good. 1729 01:14:47,047 --> 01:14:48,832 -I just wanted to move it a little forward. -Okay. 1730 01:14:48,875 --> 01:14:50,529 I picked out that chair. 1731 01:14:50,573 --> 01:14:52,139 -It's a good chair. -Thanks. 1732 01:14:52,183 --> 01:14:53,619 [cell phone ringing] 1733 01:14:57,884 --> 01:14:59,495 TED: And just sit down there. 1734 01:15:01,409 --> 01:15:03,063 And then through here, 1735 01:15:03,107 --> 01:15:04,369 we'll be looking at each other through this glass. 1736 01:15:04,412 --> 01:15:06,458 [busy signal beeping] 1737 01:15:11,419 --> 01:15:13,552 And sitting back in the chair, leaning forward? 1738 01:15:13,596 --> 01:15:15,119 -DANIEL: Yeah. Well, whatever's comfortable, I think. -Okay. 1739 01:15:15,162 --> 01:15:16,686 It's kind of cool. There's a bunch of mirrors. 1740 01:15:16,729 --> 01:15:17,817 -Is that-- okay. -WOMAN: Sam A., take one. Mark. 1741 01:15:17,861 --> 01:15:19,515 -How's that? -Good. Thank you. 1742 01:15:19,558 --> 01:15:21,038 DANIEL: The genesis of this project is that I was 1743 01:15:21,081 --> 01:15:22,822 sitting at home, and I was playing around with, 1744 01:15:22,866 --> 01:15:24,389 I think, your image generator, 1745 01:15:24,432 --> 01:15:27,261 and I was simultaneously, uh, terrified 1746 01:15:27,305 --> 01:15:29,437 -and really impressed. -[chuckles softly] 1747 01:15:29,481 --> 01:15:31,875 -That is the usual combination. -And then cut to 1748 01:15:31,918 --> 01:15:34,617 my wife and I find out we're expecting. 1749 01:15:34,660 --> 01:15:38,272 And-and I'm, like, having an existential crisis 1750 01:15:38,316 --> 01:15:40,231 as my wife is six months pregnant. 1751 01:15:40,274 --> 01:15:42,929 And I think my first question, which I ask everybody: 1752 01:15:42,973 --> 01:15:45,453 Is now a terrible time to have a kid? 1753 01:15:45,497 --> 01:15:47,368 Am I making a big mistake? 1754 01:15:47,412 --> 01:15:49,022 I'm expecting a kid in March, too. My first one. 1755 01:15:49,066 --> 01:15:50,502 -You're expecting a kid? -Yeah. 1756 01:15:50,546 --> 01:15:51,808 -You're expecting in March? -First kid. 1757 01:15:51,851 --> 01:15:53,287 -Mazel tov. -Thank you very much. 1758 01:15:53,331 --> 01:15:54,462 I've never been so excited for anything. 1759 01:15:54,506 --> 01:15:56,464 -That's how I feel. -Yeah. 1760 01:15:56,508 --> 01:15:58,205 But you're not scared? 1761 01:15:58,249 --> 01:16:02,732 I mean, like, having a kid is just this momentous thing 1762 01:16:02,775 --> 01:16:04,995 that I, you know, I stay up every night, like, 1763 01:16:05,038 --> 01:16:06,823 reading these books about how to raise a kid, 1764 01:16:06,866 --> 01:16:08,215 and I hope I'm gonna do a good job, 1765 01:16:08,259 --> 01:16:09,260 and it feels very overwhelming and-- 1766 01:16:09,303 --> 01:16:12,045 but I'm not scared... 1767 01:16:12,089 --> 01:16:14,047 for kids to grow up in a world with AI. 1768 01:16:14,091 --> 01:16:16,310 Like, that's... that'll be okay. 1769 01:16:16,354 --> 01:16:19,009 That is good to hear coming from the guy. 1770 01:16:19,052 --> 01:16:20,750 I think it's a wonderful idea to have kids. 1771 01:16:20,793 --> 01:16:23,230 I-I think they're the most magical, incredible thing. 1772 01:16:23,274 --> 01:16:25,276 -[sighs heavily] -There's so much uncertainty, 1773 01:16:25,319 --> 01:16:26,843 I would almost just do what you're gonna do anyway. 1774 01:16:26,886 --> 01:16:28,932 I know that's not a very satisfying answer, 1775 01:16:28,975 --> 01:16:31,891 but it's the only one I can come up with. 1776 01:16:31,935 --> 01:16:34,459 Our kids are never gonna... 1777 01:16:34,502 --> 01:16:37,070 know a world that doesn't have really advanced AI. 1778 01:16:37,114 --> 01:16:39,856 In fact, our kids will never be smarter than an AI. 1779 01:16:39,899 --> 01:16:42,075 DANIEL: "Our kids will never be smarter than an AI"? 1780 01:16:42,119 --> 01:16:44,338 Well, from a raw, from a raw IQ perspective, 1781 01:16:44,382 --> 01:16:46,253 they will never be smarter than an AI. 1782 01:16:46,297 --> 01:16:48,038 That notion doesn't unsettle you a little bit? 1783 01:16:48,081 --> 01:16:51,128 'Cause it makes me feel a little queasy in a weird way. 1784 01:16:51,171 --> 01:16:54,784 Um, it does unsettle me a little bit. 1785 01:16:54,827 --> 01:16:56,263 But it is reality. 1786 01:16:56,307 --> 01:16:58,265 DANIEL: Okay, so race dynamics. 1787 01:16:58,309 --> 01:17:01,007 We have a bunch of people who are all in agreement 1788 01:17:01,051 --> 01:17:02,879 that this is scary. 1789 01:17:02,922 --> 01:17:04,750 Based on the timelines that a lot of people have given me, 1790 01:17:04,794 --> 01:17:07,013 we have between two months and five years to figure this out. 1791 01:17:07,057 --> 01:17:08,667 -Yeah. -And-and, I guess, 1792 01:17:08,711 --> 01:17:12,845 my biggest question is: Why can't we just stop? 1793 01:17:14,499 --> 01:17:19,069 The problem with, uh, uh, "just stopping" is that, um, 1794 01:17:19,112 --> 01:17:22,202 there are many, many groups now around the world, uh, uh, 1795 01:17:22,246 --> 01:17:25,684 building this, for many nations, many companies, 1796 01:17:25,728 --> 01:17:28,121 um, all with different motivations. 1797 01:17:28,165 --> 01:17:30,776 There are some of these companies in this space 1798 01:17:30,820 --> 01:17:33,997 who their position is, "We want to develop this technology 1799 01:17:34,040 --> 01:17:36,042 absolutely as fast as possible." 1800 01:17:36,086 --> 01:17:40,046 And even if we could pass laws in the US and in Europe, 1801 01:17:40,090 --> 01:17:42,396 we need to convince... 1802 01:17:42,440 --> 01:17:44,964 Xi Jinping and Vladimir Putin or, you know, 1803 01:17:45,008 --> 01:17:47,445 whoever their scientific advisors are on their side. 1804 01:17:47,488 --> 01:17:49,229 That's gonna be really hard. 1805 01:17:49,273 --> 01:17:50,927 I think it is true 1806 01:17:50,970 --> 01:17:53,581 that if two people are in exactly the same place, 1807 01:17:53,625 --> 01:17:55,888 uh, the one willing to take more shortcuts on safety 1808 01:17:55,932 --> 01:17:58,064 should kind of "get there first." 1809 01:17:58,108 --> 01:18:01,067 Uh, but... 1810 01:18:01,111 --> 01:18:02,939 we're able to use our lead 1811 01:18:02,982 --> 01:18:05,985 to spend a lot more time doing safety testing. 1812 01:18:06,029 --> 01:18:07,987 DANIEL: And-and what if you lose it? 1813 01:18:08,031 --> 01:18:09,597 You get the call and you find out 1814 01:18:09,641 --> 01:18:11,687 that you're now, let's say, six months behind. 1815 01:18:11,730 --> 01:18:14,254 -Wh-What happens then? -Depends who we're behind to. 1816 01:18:14,298 --> 01:18:16,822 If it's, like, a adversarial government, 1817 01:18:16,866 --> 01:18:18,824 that's probably really bad. 1818 01:18:18,868 --> 01:18:21,348 So let's say you get the call that China has a, 1819 01:18:21,392 --> 01:18:24,787 has a recursively self-improving agent 1820 01:18:24,830 --> 01:18:27,180 or something like that that we should be really worried about. 1821 01:18:27,224 --> 01:18:29,226 What do-- what-what happens? 1822 01:18:30,967 --> 01:18:32,403 Um... 1823 01:18:35,406 --> 01:18:37,234 That case would require... 1824 01:18:37,277 --> 01:18:40,672 the first step there would be to talk to the US government. 1825 01:18:40,716 --> 01:18:42,935 Sam, do you trust the government's ability to handle 1826 01:18:42,979 --> 01:18:44,676 something like this? 1827 01:18:45,895 --> 01:18:47,766 Yeah, I do, actually. 1828 01:18:47,810 --> 01:18:49,550 There's other things I don't trust the government to handle, 1829 01:18:49,594 --> 01:18:51,161 but that particular scenario, I think they would know... 1830 01:18:51,204 --> 01:18:52,771 they-they-- yes. 1831 01:18:52,815 --> 01:18:54,425 DANIEL: When you have, like, private discussions 1832 01:18:54,468 --> 01:18:56,383 with the other guys whose fingers are on the trigger, 1833 01:18:56,427 --> 01:18:59,604 so to speak, um, do those private discussions 1834 01:18:59,647 --> 01:19:01,301 fill you with confidence 1835 01:19:01,345 --> 01:19:03,347 or do they make you, uh, more anxious? 1836 01:19:03,390 --> 01:19:07,090 Look, I mean, I-I know some of them better than others. 1837 01:19:07,133 --> 01:19:10,136 Um, I have more confidence in some of them th-than I have 1838 01:19:10,180 --> 01:19:12,356 in others, you know, as with, as with any people. 1839 01:19:12,399 --> 01:19:14,140 You know, you think about, you know, 1840 01:19:14,184 --> 01:19:15,838 the kids in your high school class or something, right? 1841 01:19:15,881 --> 01:19:18,275 You know, some of them are, you know, really sweet. 1842 01:19:18,318 --> 01:19:20,407 Some of them are well-meaning but not that effective. 1843 01:19:20,451 --> 01:19:22,366 Some of them are, you know, bullies. 1844 01:19:22,409 --> 01:19:23,759 Some of them are really bad people. 1845 01:19:23,802 --> 01:19:25,673 [stammers] You-you kind of really, 1846 01:19:25,717 --> 01:19:27,066 you really see the spread. 1847 01:19:27,110 --> 01:19:28,807 You know, am I, am I, am I confident 1848 01:19:28,851 --> 01:19:30,548 that everyone's gonna do the right thing, 1849 01:19:30,591 --> 01:19:32,028 that it's all gonna work out? 1850 01:19:32,071 --> 01:19:33,856 Um, no, I'm not. 1851 01:19:33,899 --> 01:19:35,509 And there's-there's nothing I can do about that, right? 1852 01:19:35,553 --> 01:19:37,773 You know, all-all I can do is-is push for the, 1853 01:19:37,816 --> 01:19:39,165 push for the, you know, 1854 01:19:39,209 --> 01:19:40,776 push for the government to get involved. 1855 01:19:40,819 --> 01:19:42,647 But ultimately I'm just one person there, too. 1856 01:19:42,690 --> 01:19:44,127 That's-- it's-it's up to all of us 1857 01:19:44,170 --> 01:19:46,042 to push for the government to get involved. 1858 01:19:46,085 --> 01:19:48,479 That's the number one thing that-that I think we need to do 1859 01:19:48,522 --> 01:19:50,916 to-to, you know, to set things in the right direction. 1860 01:19:50,960 --> 01:19:52,744 What makes me anxious about that is, like, 1861 01:19:52,788 --> 01:19:56,182 the basic reality that the speed at which the technology 1862 01:19:56,226 --> 01:19:58,402 is proliferating and growing is exponential, 1863 01:19:58,445 --> 01:20:01,448 and the mechanisms to legislate are 300 years old, 1864 01:20:01,492 --> 01:20:02,928 -take forever. -Yeah. 1865 01:20:02,972 --> 01:20:05,104 Um, I-I think it's gonna be a heavy lift. 1866 01:20:05,148 --> 01:20:06,889 I-I definitely agree with you on that. 1867 01:20:06,932 --> 01:20:08,891 DANIEL: You know, what I'm literally looking for 1868 01:20:08,934 --> 01:20:11,241 is-is like, "Here are steps that, like, 1869 01:20:11,284 --> 01:20:15,158 "the head honchos are gonna take to-to focus on safety, 1870 01:20:15,201 --> 01:20:18,117 to mitigate the peril and maximize the promise." 1871 01:20:18,161 --> 01:20:20,119 And I don't, I don't, I don't know 1872 01:20:20,163 --> 01:20:21,817 that there's a simple answer to that. 1873 01:20:21,860 --> 01:20:23,296 Um... 1874 01:20:25,951 --> 01:20:27,561 I mean, this maybe is too simple, 1875 01:20:27,605 --> 01:20:30,173 but you... 1876 01:20:30,216 --> 01:20:31,957 create a new model, 1877 01:20:32,001 --> 01:20:34,786 you study and test it very carefully. 1878 01:20:34,830 --> 01:20:37,223 You put it out into the world gradually, 1879 01:20:37,267 --> 01:20:39,530 and then, more and more, you understand 1880 01:20:39,573 --> 01:20:41,227 if that's safe or not, and then if it is, 1881 01:20:41,271 --> 01:20:42,707 you can take the next step. 1882 01:20:44,404 --> 01:20:47,103 It doesn't sound as flashy as, like, a brilliant scientist 1883 01:20:47,146 --> 01:20:50,454 coming up with one idea in a lab to make an AI system, like, 1884 01:20:50,497 --> 01:20:53,979 perfectly safe and controllable and everything else, 1885 01:20:54,023 --> 01:20:55,720 but it is what I believe is gonna happen. 1886 01:20:55,763 --> 01:20:57,287 Like, it is the way I think this works. 1887 01:20:57,330 --> 01:20:59,289 But let's just say something terrible happens, 1888 01:20:59,332 --> 01:21:01,813 like a model gets loose or goes rogue or something. 1889 01:21:01,857 --> 01:21:03,467 Is there a protocol? 1890 01:21:03,510 --> 01:21:05,817 Like, literally, I'm imagining a red phone. 1891 01:21:05,861 --> 01:21:07,123 -Yeah. -Sorry for thinking of this 1892 01:21:07,166 --> 01:21:08,515 in terms of movies, but, like... 1893 01:21:08,559 --> 01:21:09,995 There is a protocol. 1894 01:21:10,039 --> 01:21:11,475 -Is there a red phone on your desk? -No. 1895 01:21:11,518 --> 01:21:13,216 [laughs]: Is it a secret? 1896 01:21:13,259 --> 01:21:15,392 I mean, uh, no, it's not, it's not as fancy or dramatic 1897 01:21:15,435 --> 01:21:17,176 as you, like, would hope, but there's, like, you know-- 1898 01:21:17,220 --> 01:21:18,874 we've, like, thought through these scenarios, 1899 01:21:18,917 --> 01:21:20,527 and if this happens, we're gonna call these people 1900 01:21:20,571 --> 01:21:21,964 in this order and do this 1901 01:21:22,007 --> 01:21:23,443 and kind of make these decisions if, like-- 1902 01:21:23,487 --> 01:21:26,055 I do believe that when you have an opportunity 1903 01:21:26,098 --> 01:21:29,058 to do your thinking before a stressful situation happens, 1904 01:21:29,101 --> 01:21:30,407 that's almost always a good idea. 1905 01:21:30,450 --> 01:21:31,756 And writing it down is helpful. 1906 01:21:31,799 --> 01:21:33,279 -Being prepared is helpful. -Yeah. 1907 01:21:34,324 --> 01:21:36,021 [sighs] You... 1908 01:21:36,065 --> 01:21:38,763 It would be impossible for me to sit across from you 1909 01:21:38,806 --> 01:21:41,157 and-and ask you to promise me that this is gonna go well? 1910 01:21:41,200 --> 01:21:43,855 That is impossible. 1911 01:21:43,899 --> 01:21:45,683 There isn't any easy answers, unfortunately. 1912 01:21:45,726 --> 01:21:48,077 Uh, because it's such a cutting-edge technology, 1913 01:21:48,120 --> 01:21:49,774 um, there's still a lot of unknowns. 1914 01:21:49,817 --> 01:21:52,255 And I think that that-that needs to be, 1915 01:21:52,298 --> 01:21:55,040 um, uh, you know, understood 1916 01:21:55,084 --> 01:21:58,957 and-and hence the need for, uh, uh, some caution. 1917 01:21:59,001 --> 01:22:00,872 I wake up, you know, every day, 1918 01:22:00,916 --> 01:22:03,440 this is the, this is the number one thing I think about. 1919 01:22:03,483 --> 01:22:05,137 Now, look, I'm human, 1920 01:22:05,181 --> 01:22:08,358 and, you know, has-has every decision been perfect? 1921 01:22:08,401 --> 01:22:11,622 Can I even say my motivations were always perfectly clear? 1922 01:22:11,665 --> 01:22:13,537 Of course not. No one can say that. 1923 01:22:13,580 --> 01:22:16,366 Like, that's-that's just not, like, you know-- 1924 01:22:16,409 --> 01:22:18,629 that's-that's just not how people work. 1925 01:22:18,672 --> 01:22:21,023 The-the history of science tends to be that, 1926 01:22:21,066 --> 01:22:23,547 for better or for worse, if something's possible to do-- 1927 01:22:23,590 --> 01:22:26,985 and we now know AI is possible to do-- humanity does it. 1928 01:22:27,029 --> 01:22:29,727 All of this was-was going to happen. 1929 01:22:29,770 --> 01:22:32,295 This-this train isn't gonna stop. 1930 01:22:32,338 --> 01:22:34,253 You can't step in front of the train and stop it. 1931 01:22:34,297 --> 01:22:36,255 You're just gonna get squished. 1932 01:22:36,299 --> 01:22:38,388 ALTMAN: I mean, it's very stressful. 1933 01:22:39,606 --> 01:22:41,043 You know, there's, like, 1934 01:22:41,086 --> 01:22:42,653 a lot of things a lot of us don't know. 1935 01:22:42,696 --> 01:22:46,352 I think the history of scientific discovery is 1936 01:22:46,396 --> 01:22:48,267 one of not knowing what you don't know 1937 01:22:48,311 --> 01:22:49,660 and figuring out as you go. 1938 01:22:49,703 --> 01:22:52,968 Uh, but, yeah, it is a... 1939 01:22:53,011 --> 01:22:56,580 it is a stressful way to live. 1940 01:22:56,623 --> 01:22:58,756 DANIEL: Right. Sam, thank you very much for doing this. 1941 01:22:58,799 --> 01:23:00,497 -And again, mazel tov. -Thank you. And to you. 1942 01:23:00,540 --> 01:23:02,673 Thank you so much. Thanks, guys. 1943 01:23:02,716 --> 01:23:05,458 ♪ ♪ 1944 01:23:07,678 --> 01:23:08,984 [sighs] 1945 01:23:10,986 --> 01:23:13,162 [electricity crackling] 1946 01:23:13,205 --> 01:23:15,599 -[thunder cracks] -[whooshing, grunting] 1947 01:23:42,626 --> 01:23:44,628 -[VCR clicks] -[line ringing] 1948 01:23:47,848 --> 01:23:49,285 -[line clicks] -KEVIN ROHER: Hello. 1949 01:23:49,328 --> 01:23:50,547 DANIEL: Hey, Dad, how are you? 1950 01:23:50,590 --> 01:23:51,591 KEVIN: Good. How you doing? 1951 01:23:51,635 --> 01:23:53,071 You working? What are you up to? 1952 01:23:53,115 --> 01:23:55,247 DANIEL: I'm working on this AI film. 1953 01:23:55,291 --> 01:23:57,815 KEVIN: And how's it going? 1954 01:23:57,858 --> 01:24:00,426 DANIEL: You know, it... it's really... 1955 01:24:00,470 --> 01:24:02,167 -JOANNE ROHER: Hi, sweetie. -DANIEL: Hi, Mom. 1956 01:24:02,211 --> 01:24:03,864 JOANNE: So what is the premise of the film? 1957 01:24:03,908 --> 01:24:06,519 Is it a documentary? Kev, don't use any more spices. 1958 01:24:06,563 --> 01:24:08,304 It's already over-spiced. 1959 01:24:08,347 --> 01:24:10,436 We're making chicken right now. 1960 01:24:10,480 --> 01:24:12,612 -Is it a documentary or what... -DANIEL: It's about-- 1961 01:24:12,656 --> 01:24:13,961 the movie's about the end of the world. 1962 01:24:14,005 --> 01:24:15,789 The end of the world's coming, 1963 01:24:15,833 --> 01:24:17,574 and we're making a movie about the end of the world. 1964 01:24:17,617 --> 01:24:19,837 -JOANNE: Really? -DANIEL: Yeah. 1965 01:24:19,880 --> 01:24:21,578 KEVIN: Kind of a depressing film, it sounds like. 1966 01:24:21,621 --> 01:24:23,449 JOANNE: Yeah. 1967 01:24:23,493 --> 01:24:28,976 DANIEL: I'm feeling a lot, like-- this very acute anxiety. 1968 01:24:29,020 --> 01:24:31,370 JOANNE: It's so scary, but there's got to be-- 1969 01:24:31,414 --> 01:24:35,287 you know, have you been meeting some-some supersmart people 1970 01:24:35,331 --> 01:24:37,768 that are giving you any answers? 1971 01:24:37,811 --> 01:24:39,465 DANIEL: That's what's frustrating about it. 1972 01:24:39,509 --> 01:24:41,337 No one knows. 1973 01:24:41,380 --> 01:24:43,904 JOANNE: All I can say to that is that 1974 01:24:43,948 --> 01:24:49,867 every generation has had something scary like this. 1975 01:24:49,910 --> 01:24:52,217 KEVIN: When I was born, it was the Cuban Missile Crisis. 1976 01:24:52,261 --> 01:24:54,263 [steady heartbeat] 1977 01:24:54,306 --> 01:24:56,352 I was just scared that there was going to be a nuclear war. 1978 01:24:56,395 --> 01:24:57,744 JOANNE: Yeah, but we didn't know 1979 01:24:57,788 --> 01:24:59,094 what they were gonna do and... 1980 01:24:59,137 --> 01:25:00,530 KEVIN: And the world didn't end. 1981 01:25:00,573 --> 01:25:01,618 Everyone woke up the next morning, 1982 01:25:01,661 --> 01:25:03,402 and we're still doing our thing. 1983 01:25:03,446 --> 01:25:04,969 DANIEL: I'm very scared, 1984 01:25:05,012 --> 01:25:06,840 especially in the context of, like, 1985 01:25:06,884 --> 01:25:09,147 -you know, the baby and... -[fetal heartbeat pulsing] 1986 01:25:09,191 --> 01:25:10,540 KEVIN: It's gonna be a learning curve. 1987 01:25:10,583 --> 01:25:12,063 JOANNE: You're gonna be okay. 1988 01:25:12,107 --> 01:25:13,064 KEVIN: You can't, you can't think about 1989 01:25:13,108 --> 01:25:14,326 what you can't control, Daniel. 1990 01:25:14,370 --> 01:25:16,154 Just remember that. 1991 01:25:16,198 --> 01:25:17,677 DANIEL [voice breaking]: I'm really, I'm really feeling 1992 01:25:17,721 --> 01:25:19,201 nervous and scared about it. 1993 01:25:19,244 --> 01:25:21,028 There's so much that I can't control. 1994 01:25:21,072 --> 01:25:23,857 KEVIN: Don't be nervous. You can't let that get to you. 1995 01:25:23,901 --> 01:25:26,251 You can only control what you can control, 1996 01:25:26,295 --> 01:25:27,905 and that's all you can do. 1997 01:25:27,948 --> 01:25:30,603 You can't do more than that. 1998 01:25:30,647 --> 01:25:32,649 Write that down in your book. 1999 01:25:35,347 --> 01:25:36,870 ♪ ♪ 2000 01:25:44,095 --> 01:25:46,663 [fetal heartbeat pulsing] 2001 01:25:46,706 --> 01:25:49,231 [howling] 2002 01:25:49,274 --> 01:25:51,581 CAROLINE: When you look back, 2003 01:25:51,624 --> 01:25:54,453 the world is always ending. 2004 01:25:54,497 --> 01:25:56,803 And when you look ahead, 2005 01:25:56,847 --> 01:25:59,241 the world is always ending. 2006 01:25:59,284 --> 01:26:02,157 ...on fire. One home is already on fire... 2007 01:26:02,200 --> 01:26:05,377 CAROLINE: But the world is always starting, too. 2008 01:26:07,988 --> 01:26:09,947 DANIEL: Are you ready? 2009 01:26:09,990 --> 01:26:11,992 Are you ever really ready? 2010 01:26:15,909 --> 01:26:17,650 DANIEL: You want to drive or you want me to drive? 2011 01:26:17,694 --> 01:26:18,782 CAROLINE: You drive. 2012 01:26:20,479 --> 01:26:21,828 -[woman speaks indistinctly] -CAROLINE: Okay. 2013 01:26:21,872 --> 01:26:23,700 DANIEL: Thank you, Maria. 2014 01:26:23,743 --> 01:26:26,268 WOMAN: It's gonna just feel really crampy. 2015 01:26:26,311 --> 01:26:27,791 -CAROLINE: Okay. -WOMAN: You're ready? 2016 01:26:27,834 --> 01:26:29,053 -CAROLINE: Yes. -[indistinct chatter] 2017 01:26:29,096 --> 01:26:30,924 [Caroline sighs heavily] 2018 01:26:30,968 --> 01:26:32,404 WOMAN: You could be in a medical drama with all that... 2019 01:26:33,840 --> 01:26:35,842 ♪ ♪ 2020 01:26:44,329 --> 01:26:45,896 -[static crackles] -[indistinct chatter] 2021 01:26:49,116 --> 01:26:51,075 WOMAN: ...some pressure. 2022 01:26:52,685 --> 01:26:54,078 Just relax. 2023 01:26:56,907 --> 01:26:59,214 -[baby crying] -[device beeping] 2024 01:27:02,129 --> 01:27:04,915 DANIEL: Hi, buddy. Ooh. 2025 01:27:04,958 --> 01:27:06,438 Hi, buddy. 2026 01:27:06,482 --> 01:27:08,484 ♪ ♪ 2027 01:27:31,202 --> 01:27:33,683 KEVIN: It's gonna be a whole new world 2028 01:27:33,726 --> 01:27:35,032 for you, Daniel. 2029 01:27:37,208 --> 01:27:40,080 And I think you're gonna be an amazing father. 2030 01:27:41,517 --> 01:27:44,346 JOANNE [chuckles]: That's for sure. 2031 01:27:44,389 --> 01:27:46,609 -[Kevin crying] -Oh. 2032 01:27:46,652 --> 01:27:48,654 [crying continues] 2033 01:27:48,698 --> 01:27:50,787 KEVIN: You're gonna do a great job. 2034 01:27:51,788 --> 01:27:53,050 JOANNE: Kev, why are you crying? 2035 01:27:53,093 --> 01:27:54,617 KEVIN: I'm crying because... 2036 01:27:54,660 --> 01:27:57,054 I don't know. I just don't... 2037 01:27:57,097 --> 01:27:59,317 DANIEL: Dad, you're gonna make me cry, too. 2038 01:27:59,361 --> 01:28:00,710 Why are you crying? 2039 01:28:03,190 --> 01:28:05,497 KEVIN: I just know that you're gonna be an amazing dad. 2040 01:28:05,541 --> 01:28:07,107 DANIEL [crying]: I'm only gonna be a great dad 2041 01:28:07,151 --> 01:28:09,675 'cause I had a great dad. 2042 01:28:09,719 --> 01:28:12,287 KEVIN: I'm getting emotional, that's all. 2043 01:28:12,330 --> 01:28:13,679 JOANNE: Aw. 2044 01:28:13,723 --> 01:28:14,898 [playful chatter over video] 2045 01:28:14,941 --> 01:28:16,552 [over video]: Boy, oh, boy! 2046 01:28:16,595 --> 01:28:17,857 My goodness! 2047 01:28:18,902 --> 01:28:20,817 Look at those little cheekies. 2048 01:28:20,860 --> 01:28:24,777 Look at those little cheekies. Look at those little cheekies. 2049 01:28:24,821 --> 01:28:26,344 Yeah. 2050 01:28:26,388 --> 01:28:28,433 DANIEL: I know how to end this movie. 2051 01:28:29,478 --> 01:28:32,263 Babies. 2052 01:28:32,307 --> 01:28:35,266 -The end of the movie is about babies.-[chatter over videos] 2053 01:28:35,310 --> 01:28:37,094 They're life-affirming. 2054 01:28:37,137 --> 01:28:38,922 They're exhausting. 2055 01:28:38,965 --> 01:28:40,793 -[baby laughing] -They're hilarious. 2056 01:28:40,837 --> 01:28:42,665 And they're worth it. 2057 01:28:42,708 --> 01:28:46,451 This film isn't about the inner workings of a technology. 2058 01:28:46,495 --> 01:28:48,235 It's not about the billionaire CEOs. 2059 01:28:48,279 --> 01:28:50,194 It's not about the geopolitics. 2060 01:28:50,237 --> 01:28:53,589 It's not about the terrifying future or the end of the world, 2061 01:28:53,632 --> 01:28:55,678 because my world is just starting. 2062 01:28:55,721 --> 01:28:57,375 I'm building a crib. 2063 01:28:57,419 --> 01:28:59,290 Right here, right now. 2064 01:28:59,334 --> 01:29:00,465 [baby cooing] 2065 01:29:00,509 --> 01:29:02,337 AI is gonna change everything 2066 01:29:02,380 --> 01:29:05,601 in ways too powerful and complex for us to understand. 2067 01:29:05,644 --> 01:29:08,952 And the future is not for any of us to decide. 2068 01:29:10,736 --> 01:29:13,609 But what I can decide is to be the best possible husband 2069 01:29:13,652 --> 01:29:17,700 for my wife and the best possible dad for my son. 2070 01:29:17,743 --> 01:29:21,399 So whether our AI future is a nightmarish dystopia 2071 01:29:21,443 --> 01:29:23,836 or the utopia that we all dream of, 2072 01:29:23,880 --> 01:29:26,883 I'll at least know that I did everything I could 2073 01:29:26,926 --> 01:29:30,103 to guide my family through this AI revolution. 2074 01:29:30,147 --> 01:29:33,846 And no matter what, we'll be facing it together. 2075 01:29:33,890 --> 01:29:35,195 [uplifting music swells] 2076 01:29:35,239 --> 01:29:36,806 [music stops abruptly] 2077 01:29:36,849 --> 01:29:40,505 DANIEL: So that's just our first idea. 2078 01:29:40,549 --> 01:29:42,812 How does this feel? Are you feeling this? 2079 01:29:42,855 --> 01:29:45,249 Wait, I-- like, this is not-- this is a joke. 2080 01:29:45,292 --> 01:29:47,425 It's not actually how you're gonna end it. 2081 01:29:50,036 --> 01:29:52,648 I mean, it's just an idea. Okay? 2082 01:29:52,691 --> 01:29:54,389 No, Daniel. 2083 01:29:54,432 --> 01:29:56,129 ...uh, very, very dumb. 2084 01:29:56,173 --> 01:29:58,915 You've just spent, I don't know, like, how many years 2085 01:29:58,958 --> 01:30:02,571 of our life working on this, talking to every leading expert 2086 01:30:02,614 --> 01:30:04,964 on the planet about the subject, 2087 01:30:05,008 --> 01:30:07,663 and you're gonna end it 2088 01:30:07,706 --> 01:30:10,796 with some, like, kumbaya bullshit? 2089 01:30:10,840 --> 01:30:13,495 There's an asteroid headed to Earth. 2090 01:30:13,538 --> 01:30:15,497 What do you do? Just... 2091 01:30:15,540 --> 01:30:17,629 hold hands and hope it works out okay? 2092 01:30:17,673 --> 01:30:19,414 Absolutely not. 2093 01:30:19,457 --> 01:30:22,286 We have to-- it ha-- it has to be... 2094 01:30:22,329 --> 01:30:24,419 The ending has to... 2095 01:30:25,855 --> 01:30:27,900 CAROLINE: Okay, first thing: 2096 01:30:27,944 --> 01:30:31,426 AI is here, and it's here to stay. 2097 01:30:31,469 --> 01:30:32,905 The shit's out of the horse, 2098 01:30:32,949 --> 01:30:35,430 but the horse is gonna keep shitting. 2099 01:30:38,171 --> 01:30:39,651 [crowd cheering] 2100 01:30:39,695 --> 01:30:41,566 You know, one of the basic laws of history 2101 01:30:41,610 --> 01:30:44,003 is that nothing really have a beginning 2102 01:30:44,047 --> 01:30:45,962 and nothing has any ending. 2103 01:30:46,005 --> 01:30:47,703 It just goes on. 2104 01:30:47,746 --> 01:30:50,532 AI is nowhere near its full development. 2105 01:30:50,575 --> 01:30:52,795 CAROLINE: Even if the current AI bubble bursts, 2106 01:30:52,838 --> 01:30:54,579 -humans are never going to stop -[noisemaker blows] 2107 01:30:54,623 --> 01:30:57,539 building more and more powerful technology. 2108 01:30:57,582 --> 01:30:59,410 HAO: You can choose not to use AI 2109 01:30:59,454 --> 01:31:02,239 or participate in it, but it's going to affect you anyway. 2110 01:31:02,282 --> 01:31:04,023 DANIEL: Okay, fantastic. So we're screwed. 2111 01:31:04,067 --> 01:31:06,635 CAROLINE: No, we're not because of one simple thing. 2112 01:31:10,943 --> 01:31:12,989 This is not inevitable. 2113 01:31:13,032 --> 01:31:15,208 If we could just see it clearly together, 2114 01:31:15,252 --> 01:31:18,342 the obvious response will be to choose something different. 2115 01:31:18,385 --> 01:31:21,040 We need to very clearly change the game 2116 01:31:21,084 --> 01:31:23,129 from a race to the bottom into a race to the top. 2117 01:31:23,173 --> 01:31:27,438 LEAHY: The problem we need to solve is not AI specifically. 2118 01:31:27,482 --> 01:31:31,224 It's the general question of how do we build a society 2119 01:31:31,268 --> 01:31:33,879 that can deal with powerful technology. 2120 01:31:33,923 --> 01:31:35,359 Because we're going to get 2121 01:31:35,402 --> 01:31:36,795 only more and more powerful technology. 2122 01:31:36,839 --> 01:31:39,668 CAROLINE: We need to upgrade our society, 2123 01:31:39,711 --> 01:31:41,496 and the first step is coming together 2124 01:31:41,539 --> 01:31:43,367 -and demanding... -ALL: Coordination. 2125 01:31:43,410 --> 01:31:47,023 Some form of international cooperation or agreement about 2126 01:31:47,066 --> 01:31:48,503 what the norms should be. 2127 01:31:48,546 --> 01:31:49,895 You know, how should they be deployed... 2128 01:31:49,939 --> 01:31:51,549 Like, real international diplomacy 2129 01:31:51,593 --> 01:31:53,290 among the superpowers. 2130 01:31:53,333 --> 01:31:56,293 The Chinese are as worried about it as the Americans. 2131 01:31:56,336 --> 01:31:57,729 I think it's difficult, you know, 2132 01:31:57,773 --> 01:31:59,514 in the current geopolitical climate, 2133 01:31:59,557 --> 01:32:01,254 -but I think it's necessary. -RASKIN: Absolutely. 2134 01:32:01,298 --> 01:32:04,693 In the exact same way that the last time that humanity 2135 01:32:04,736 --> 01:32:09,611 developed a technology this dangerous... 2136 01:32:09,654 --> 01:32:12,396 ...that required a complete, unprecedented shift 2137 01:32:12,439 --> 01:32:15,530 to the structure of our world. 2138 01:32:15,573 --> 01:32:17,619 CAROLINE: So we need to do that complete, 2139 01:32:17,662 --> 01:32:20,883 unprecedented shift again. 2140 01:32:20,926 --> 01:32:24,103 You know, we-we talk to people who work at these AI companies, 2141 01:32:24,147 --> 01:32:25,322 and they say they want to do something different, 2142 01:32:25,365 --> 01:32:26,628 but they need public pressure. 2143 01:32:26,671 --> 01:32:28,151 They need the government to do something. 2144 01:32:28,194 --> 01:32:29,979 So then we go to DC, and they say, 2145 01:32:30,022 --> 01:32:31,850 "Well, we need Silicon Valley to do something different. 2146 01:32:31,894 --> 01:32:33,635 They're the ones who are gonna come up with the guardrails." 2147 01:32:33,678 --> 01:32:35,680 And so everyone is pointing the finger at someone else, 2148 01:32:35,724 --> 01:32:39,466 and what they agree on is that we need public pressure 2149 01:32:39,510 --> 01:32:41,207 in order for something else to happen. 2150 01:32:41,251 --> 01:32:42,818 And that's what you 2151 01:32:42,861 --> 01:32:44,254 and all the people watching this movie can do. 2152 01:32:44,297 --> 01:32:45,298 CAROLINE: We need to hold the leaders 2153 01:32:45,342 --> 01:32:47,083 in our governments 2154 01:32:47,126 --> 01:32:48,563 and the leaders of these companies accountable. 2155 01:32:48,606 --> 01:32:51,000 BOEREE: Whichever country you're in, 2156 01:32:51,043 --> 01:32:52,871 let them know that you're not happy 2157 01:32:52,915 --> 01:32:54,569 with the current status quo. 2158 01:32:54,612 --> 01:32:57,093 So, yeah, it's boring to say call your congressperson. 2159 01:32:57,136 --> 01:32:59,356 I'm not saying you should just do that, but, like, 2160 01:32:59,399 --> 01:33:00,923 we do have to do that. 2161 01:33:00,966 --> 01:33:02,402 Like, we do have to get the government involved. 2162 01:33:02,446 --> 01:33:03,839 DANIEL: So we just call them up and say, 2163 01:33:03,882 --> 01:33:05,971 "Hey, stop Big Tech from ruining the world"? 2164 01:33:06,015 --> 01:33:08,887 CAROLINE: No, but there are tons of really obvious, 2165 01:33:08,931 --> 01:33:10,672 straightforward things we can be demanding. 2166 01:33:10,715 --> 01:33:12,499 We need transparency. 2167 01:33:12,543 --> 01:33:14,240 We need to end the secrecy that exists inside these labs, 2168 01:33:14,284 --> 01:33:16,634 because they are building powerful technology, 2169 01:33:16,678 --> 01:33:18,680 and the public deserves to know what's going on. 2170 01:33:18,723 --> 01:33:20,682 Ultimately, we're gonna need independent, 2171 01:33:20,725 --> 01:33:23,249 objective third parties to evaluate the systems. 2172 01:33:23,293 --> 01:33:27,036 We can't count on the companies to grade their own homework. 2173 01:33:27,079 --> 01:33:31,170 If-if a company uses AI and-and has AI interacting with you, 2174 01:33:31,214 --> 01:33:34,957 it should disclose that you are interacting with an AI system. 2175 01:33:35,000 --> 01:33:38,613 Yeah. And-and also we need a system that makes companies 2176 01:33:38,656 --> 01:33:41,833 legally liable for the AI systems that they produce. 2177 01:33:41,877 --> 01:33:45,750 We need to make sure that there are tests and safety standards 2178 01:33:45,794 --> 01:33:48,013 that are applied to everyone. 2179 01:33:48,057 --> 01:33:49,711 We need some ground rules, 2180 01:33:49,754 --> 01:33:51,538 and we need to keep adapting those rules 2181 01:33:51,582 --> 01:33:54,585 at the speed that the technology develops. 2182 01:33:54,629 --> 01:33:57,153 LEAHY: There is currently more regulation 2183 01:33:57,196 --> 01:33:59,285 on selling a sandwich to the public 2184 01:33:59,329 --> 01:34:02,898 than there is on building potentially world-ending AGI. 2185 01:34:02,941 --> 01:34:04,682 CAROLINE: And the last thing is 2186 01:34:04,726 --> 01:34:07,598 to upgrade ourselves. 2187 01:34:07,642 --> 01:34:09,731 This is not, like, the job of, like, the safety team 2188 01:34:09,774 --> 01:34:11,733 at any given lab, or the CEO. 2189 01:34:11,776 --> 01:34:13,299 So, like, this is everyone's job. 2190 01:34:13,343 --> 01:34:14,779 Don't, like, leave it up to the AI experts. 2191 01:34:14,823 --> 01:34:16,955 Like, [bleep] that. Like, this is the moment 2192 01:34:16,999 --> 01:34:19,262 that we are transitioning from, like, 2193 01:34:19,305 --> 01:34:21,525 mostly human cognitive power to, like, AI cognitive power, 2194 01:34:21,568 --> 01:34:23,092 and it affects everyone, 2195 01:34:23,135 --> 01:34:24,833 and I want people to be in on that conversation. 2196 01:34:24,876 --> 01:34:26,878 And I would say, if you think that AI will kill us all, 2197 01:34:26,922 --> 01:34:28,488 you should be working in AI research 2198 01:34:28,532 --> 01:34:30,229 to make sure it doesn't, because you do have 2199 01:34:30,273 --> 01:34:32,188 an enormous amount of agency. 2200 01:34:32,231 --> 01:34:33,842 CAROLINE: Whoever you are, 2201 01:34:33,885 --> 01:34:35,844 you are an expert in your own industry, 2202 01:34:35,887 --> 01:34:39,151 in your own school, in your own family, 2203 01:34:39,195 --> 01:34:42,589 and it's up to you how AI is used in your life. 2204 01:34:42,633 --> 01:34:44,983 Whether you want to join your school board 2205 01:34:45,027 --> 01:34:47,856 or-or whether you want to ask your employer 2206 01:34:47,899 --> 01:34:49,379 how they're using AI technologies, 2207 01:34:49,422 --> 01:34:51,947 like, all of us can do the work. 2208 01:34:51,990 --> 01:34:56,081 A lot of unions have been pretty effective at, like, 2209 01:34:56,125 --> 01:34:59,041 determining how they want to interact with these systems. 2210 01:34:59,084 --> 01:35:02,697 -Nurses unions, teacher unions. -[indistinct chanting] 2211 01:35:02,740 --> 01:35:04,568 DANIELA AMODEI: I would love if parents everywhere 2212 01:35:04,611 --> 01:35:06,135 went to the AI companies and said, 2213 01:35:06,178 --> 01:35:08,006 "How can you be better on this?" Including us. 2214 01:35:08,050 --> 01:35:11,183 So, I founded Encode Justice when I was 15 years old. 2215 01:35:11,227 --> 01:35:14,839 We are the world's first and largest army of young people 2216 01:35:14,883 --> 01:35:18,364 fighting for human-centered artificial intelligence. 2217 01:35:18,408 --> 01:35:21,019 It doesn't matter who you are, even the smallest actions help, 2218 01:35:21,063 --> 01:35:23,935 and even conversation starting is really, really valuable. 2219 01:35:23,979 --> 01:35:25,937 DANIEL: So my job could be, like, 2220 01:35:25,981 --> 01:35:27,939 tell Bubby Lila about this at dinner? 2221 01:35:27,983 --> 01:35:29,767 CAROLINE: Honestly, yeah, that's part of it. 2222 01:35:29,811 --> 01:35:31,638 And we're gonna have to do 2223 01:35:31,682 --> 01:35:34,424 a lot of things that we haven't even thought of yet. 2224 01:35:34,467 --> 01:35:36,034 People are gonna look at anything that we've outlined 2225 01:35:36,078 --> 01:35:37,253 and say, "That's not enough." 2226 01:35:37,296 --> 01:35:38,950 What matters is that the forces 2227 01:35:38,994 --> 01:35:40,909 that are working towards solutions 2228 01:35:40,952 --> 01:35:44,434 start to exceed the forces that are working against solutions. 2229 01:35:44,477 --> 01:35:46,392 Making the world better has always been hard. 2230 01:35:46,436 --> 01:35:47,829 It has never been easy. 2231 01:35:47,872 --> 01:35:49,961 Like, there have been many shitty things 2232 01:35:50,005 --> 01:35:51,920 that have happened in history, and we've had-- 2233 01:35:51,963 --> 01:35:53,704 like, people have had to deal with that, 2234 01:35:53,748 --> 01:35:56,794 and then they've risen up and changed it. 2235 01:35:56,838 --> 01:36:00,319 There are an insane number of challenges ahead of us, 2236 01:36:00,363 --> 01:36:04,106 but if we can get past them, 2237 01:36:04,149 --> 01:36:08,066 we can unlock a future beyond our wildest imagination. 2238 01:36:08,110 --> 01:36:10,416 CAROLINE: We have to come together 2239 01:36:10,460 --> 01:36:14,246 and find the path between the promise and the peril. 2240 01:36:14,290 --> 01:36:16,422 We can't be pessimists or optimists. 2241 01:36:17,423 --> 01:36:19,425 We have to become something new. 2242 01:36:21,732 --> 01:36:24,474 A friend of mine calls me an apocaloptimist. 2243 01:36:24,517 --> 01:36:26,041 "Apocaloptimist"? 2244 01:36:26,084 --> 01:36:28,260 I think that might be my new favorite word. 2245 01:36:28,304 --> 01:36:29,914 -[laughs] -MAN: Might be the name of this movie. 2246 01:36:29,958 --> 01:36:31,263 -It might be the name of this movie. -[laughter] 2247 01:36:31,307 --> 01:36:34,179 -"Apocaloptimist." -"Apocaloptimist." 2248 01:36:34,223 --> 01:36:35,833 Yeah. 2249 01:36:35,877 --> 01:36:37,269 I don't believe in doom. 2250 01:36:37,313 --> 01:36:39,054 I believe in the spirit of life, 2251 01:36:39,097 --> 01:36:42,666 uh, and I believe in life is about the capacity to act. 2252 01:36:42,709 --> 01:36:44,276 -[crowd singing] -The capacity to relate, 2253 01:36:44,320 --> 01:36:46,365 the capacity to feel. 2254 01:36:49,020 --> 01:36:50,979 -[indistinct mission chatter] -[cheering] 2255 01:36:51,022 --> 01:36:53,938 We have to double down more and more 2256 01:36:53,982 --> 01:36:57,028 on those capacities that we have as humans 2257 01:36:57,072 --> 01:37:00,118 that robotic systems will never have. 2258 01:37:00,162 --> 01:37:04,296 It's time right now to make those decisions about 2259 01:37:04,340 --> 01:37:07,473 how to guide it and support it rather than dividing us. 2260 01:37:07,517 --> 01:37:10,259 DANIEL: It kind of sounds like raising a kid. 2261 01:37:10,302 --> 01:37:12,435 That's what's up. Yeah. 2262 01:37:12,478 --> 01:37:14,480 CAROLINE: AI may have more raw intelligence 2263 01:37:14,524 --> 01:37:17,222 -than our little human brains, -[baby laughing] 2264 01:37:17,266 --> 01:37:21,052 but we're so much more than just our intelligence. 2265 01:37:21,096 --> 01:37:25,056 Intelligence is-is the ability to solve problems. 2266 01:37:25,100 --> 01:37:28,320 Wisdom is the ability to know which problems to solve. 2267 01:37:30,932 --> 01:37:33,717 [indistinct crew chatter] 2268 01:37:33,760 --> 01:37:34,892 DANIEL: It can go on your fridge. 2269 01:37:34,936 --> 01:37:36,198 [laughing] 2270 01:37:36,241 --> 01:37:38,113 YUDKOWSKY: So, don't give up. 2271 01:37:38,156 --> 01:37:39,766 Humanity has done 2272 01:37:39,810 --> 01:37:42,160 more difficult things than this in its history. 2273 01:37:43,553 --> 01:37:46,338 It's just hard to convince people that they should. 2274 01:37:46,382 --> 01:37:48,645 ♪ ♪ 2275 01:37:51,474 --> 01:37:52,562 [audio crackles] 2276 01:37:54,912 --> 01:37:56,827 DANIEL: So, when I started making this movie, 2277 01:37:56,871 --> 01:37:58,437 I would say that I was, like, 2278 01:37:58,481 --> 01:38:00,918 broadly a cynical asshole about this whole thing. 2279 01:38:00,962 --> 01:38:03,094 Over the course of making the film, 2280 01:38:03,138 --> 01:38:04,661 I've come to understand that, like, 2281 01:38:04,704 --> 01:38:07,446 that's the only thing we can't be. 2282 01:38:07,490 --> 01:38:09,361 ...or anything else you'd like to discuss? 2283 01:38:09,405 --> 01:38:11,363 No. I think we covered a lot. 2284 01:38:11,407 --> 01:38:13,017 -Yeah. [chuckles] -MAN: Thank you so much. 2285 01:38:13,061 --> 01:38:15,498 Thanks. 2286 01:38:15,541 --> 01:38:16,891 DANIEL: This is a problem that's bigger 2287 01:38:16,934 --> 01:38:19,067 than any one person. 2288 01:38:19,110 --> 01:38:21,547 This will change the world in ways that we don't understand. 2289 01:38:21,591 --> 01:38:23,462 That is all true. 2290 01:38:23,506 --> 01:38:25,551 -[laughs]: Okay. -[indistinct chatter] 2291 01:38:25,595 --> 01:38:27,684 DANIEL: But what we do have agency over 2292 01:38:27,727 --> 01:38:30,078 is what we do about it. 2293 01:38:30,121 --> 01:38:32,558 As frontier AI grows exponentially more capable... 2294 01:38:32,602 --> 01:38:34,430 DANIEL: And the reality 2295 01:38:34,473 --> 01:38:36,562 is that if we just decide it's hopeless, 2296 01:38:36,606 --> 01:38:39,000 then it is hopeless. 2297 01:38:39,043 --> 01:38:40,740 ...to put less stress on planet Earth... 2298 01:38:40,784 --> 01:38:42,742 DANIEL: But if you decide 2299 01:38:42,786 --> 01:38:45,528 that you want to try... 2300 01:38:46,790 --> 01:38:48,096 ...then you try. 2301 01:38:48,139 --> 01:38:49,967 And that's hard. 2302 01:38:51,273 --> 01:38:53,188 But you know what? 2303 01:38:53,231 --> 01:38:56,278 Big things seem impossible before they actually happen. 2304 01:38:58,149 --> 01:39:01,413 But when they finally do happen, 2305 01:39:01,457 --> 01:39:05,243 it's because millions of people took millions of actions 2306 01:39:05,287 --> 01:39:07,550 to make them happen. 2307 01:39:07,593 --> 01:39:10,292 And so... 2308 01:39:10,335 --> 01:39:12,424 we have to try. 2309 01:39:13,469 --> 01:39:14,905 [sighs heavily] 2310 01:39:21,129 --> 01:39:23,522 There's too much at stake. 2311 01:39:23,566 --> 01:39:25,611 ♪ ♪ 2312 01:39:32,401 --> 01:39:34,664 [film crackling] 2313 01:39:37,232 --> 01:39:41,236 Look at the incredible changes we've experienced and survived 2314 01:39:41,279 --> 01:39:42,846 from the Stone Age, 2315 01:39:42,889 --> 01:39:46,110 and yet even greater changes are still to come. 2316 01:39:46,154 --> 01:39:49,157 ["Lost It to Trying" by Son Lux playing] 2317 01:40:18,403 --> 01:40:20,449 ♪ ♪ 2318 01:40:47,606 --> 01:40:50,174 ♪ What will we do now? ♪ 2319 01:40:50,218 --> 01:40:53,221 ♪ We've lost it to trying ♪ 2320 01:40:53,264 --> 01:40:55,092 ♪ We've lost it ♪ 2321 01:40:55,136 --> 01:40:56,746 ♪ To trying ♪ 2322 01:40:59,575 --> 01:41:02,230 ♪ What will we do now? ♪ 2323 01:41:02,273 --> 01:41:05,189 ♪ We've lost it to trying ♪ 2324 01:41:05,233 --> 01:41:06,973 ♪ We've lost it ♪ 2325 01:41:07,017 --> 01:41:08,714 ♪ To trying ♪ 2326 01:41:11,630 --> 01:41:14,111 ♪ What can we say now? ♪ 2327 01:41:14,155 --> 01:41:17,201 ♪ Our mouths only lying ♪ 2328 01:41:17,245 --> 01:41:18,637 ♪ Our mouths ♪ 2329 01:41:18,681 --> 01:41:20,683 ♪ Only lying ♪ 2330 01:41:25,122 --> 01:41:27,690 ♪ What can we say now? ♪ 2331 01:41:27,733 --> 01:41:30,823 ♪ Our mouths only lying ♪ 2332 01:41:30,867 --> 01:41:32,216 ♪ Our mouths ♪ 2333 01:41:32,260 --> 01:41:34,218 ♪ Only lying ♪ 2334 01:41:34,262 --> 01:41:36,264 ♪ ♪ 2335 01:42:01,071 --> 01:42:03,726 ♪ Give in and get out ♪ 2336 01:42:03,769 --> 01:42:06,685 ♪ We rise in the dying ♪ 2337 01:42:06,729 --> 01:42:08,209 ♪ We rise ♪ 2338 01:42:08,252 --> 01:42:10,211 ♪ In the dying ♪ 2339 01:42:13,127 --> 01:42:15,781 ♪ Give in and get out ♪ 2340 01:42:15,825 --> 01:42:18,654 ♪ We rise in the dying ♪ 2341 01:42:18,697 --> 01:42:20,221 ♪ We rise ♪ 2342 01:42:20,264 --> 01:42:22,223 ♪ In the dying ♪ 2343 01:42:25,051 --> 01:42:27,663 ♪ Give in and get out ♪ 2344 01:42:27,706 --> 01:42:30,666 ♪ We rise in the dying ♪ 2345 01:42:30,709 --> 01:42:32,233 ♪ We rise ♪ 2346 01:42:32,276 --> 01:42:34,148 ♪ In the dying ♪ 2347 01:42:37,107 --> 01:42:39,718 ♪ Give in and get out ♪ 2348 01:42:39,762 --> 01:42:42,591 ♪ We rise in the dying ♪ 2349 01:42:42,634 --> 01:42:43,809 ♪ We ♪ 2350 01:42:43,853 --> 01:42:45,898 ♪ ♪ 2351 01:43:15,580 --> 01:43:17,626 ♪ ♪ 2352 01:43:22,761 --> 01:43:25,286 ♪ Oh, oh, oh, oh, oh ♪ 2353 01:43:25,329 --> 01:43:26,983 ♪ Oh, oh, oh, oh ♪ 2354 01:43:27,026 --> 01:43:28,419 ♪ Oh, oh ♪ 2355 01:43:28,463 --> 01:43:31,379 ♪ Oh, oh, oh, oh, oh, oh ♪ 2356 01:43:31,422 --> 01:43:32,945 ♪ Oh, oh, oh, oh ♪ 2357 01:43:32,989 --> 01:43:34,382 ♪ Oh, oh ♪ 2358 01:43:34,425 --> 01:43:37,036 ♪ Oh, oh, oh, oh, oh, oh ♪ 2359 01:43:37,080 --> 01:43:39,648 ♪ What will we do now? ♪ 2360 01:43:39,691 --> 01:43:42,651 ♪ We've lost it to trying ♪ 2361 01:43:42,694 --> 01:43:44,566 ♪ We've lost it ♪ 2362 01:43:44,609 --> 01:43:46,307 ♪ To trying ♪ 2363 01:43:46,350 --> 01:43:49,005 ♪ Oh, oh, oh, oh, oh, oh ♪ 2364 01:43:49,048 --> 01:43:51,703 ♪ What will we do now? ♪ 2365 01:43:51,747 --> 01:43:54,706 ♪ We've lost it to trying ♪ 2366 01:43:54,750 --> 01:43:56,578 ♪ We've lost it ♪ 2367 01:43:56,621 --> 01:43:58,406 ♪ To trying. ♪ 2368 01:43:58,449 --> 01:44:00,451 ♪ ♪ 2369 01:44:23,126 --> 01:44:24,475 [song ends]