0000 00:00:05,000 --> 00:00:10,000 YoubetCash.net Agen Judi Online Aman & Terpercaya 000 00:00:11,000 --> 00:00:15,000 Bonus Cashback Bola 6% Bonus Rollingan 0,80% 00 00:00:16,000 --> 00:00:20,000 Welcome Bonus Rp 25.000 Extra Bonus Tambahan Withdraw Rp 2.000.000 0 00:00:21,000 --> 00:00:25,000 CashBack Mix Parlay 100% Bonus Ajak teman 45,2% 1 00:00:48,222 --> 00:00:51,703 MAX TEGMARK: If we build artificial general intelligence, 2 00:00:51,747 --> 00:00:53,531 that'll be the biggest event 3 00:00:53,575 --> 00:00:56,708 in the history of life on Earth. 4 00:00:56,752 --> 00:01:00,147 LOUIS ROSENBERG: Alien intelligence inside a computer system that 5 00:01:00,190 --> 00:01:03,280 has control over the planet, the day it arrives... 6 00:01:03,324 --> 00:01:05,108 JAMES CAMERON: And science fiction books and movies 7 00:01:05,152 --> 00:01:07,197 have set it up in advance. 8 00:01:25,781 --> 00:01:27,696 What's the problem? 9 00:01:27,739 --> 00:01:29,698 HAL: I think you know what the problem is 10 00:01:29,741 --> 00:01:31,134 just as well as I do. 11 00:01:31,178 --> 00:01:32,701 What are you talking about, HAL? 12 00:01:34,094 --> 00:01:35,617 HAL: This mission is too important 13 00:01:35,660 --> 00:01:37,793 for me to allow you to jeopardize it. 14 00:01:39,360 --> 00:01:42,363 I don't know what you're talking about, HAL. 15 00:01:42,406 --> 00:01:45,975 KEIR DULLEA: HAL, a super-intelligent computer. 16 00:01:46,018 --> 00:01:48,325 We'd call it AI today. 17 00:01:49,544 --> 00:01:52,242 My character, Dave, was the astronaut 18 00:01:52,286 --> 00:01:55,027 in one of the most prophetic films ever made 19 00:01:55,071 --> 00:01:57,378 about the threat posed to humanity 20 00:01:57,421 --> 00:01:59,423 by artificial intelligence. 21 00:02:01,121 --> 00:02:03,166 HAL was science fiction. 22 00:02:03,210 --> 00:02:06,996 But now, AI is all around us. 23 00:02:07,039 --> 00:02:09,172 Artificial intelligence isn't the future. 24 00:02:09,216 --> 00:02:11,087 AI is already here. 25 00:02:11,131 --> 00:02:12,915 This week a robot ran for mayor 26 00:02:12,958 --> 00:02:14,177 in a small Japanese town. 27 00:02:14,221 --> 00:02:15,352 The White House says 28 00:02:15,396 --> 00:02:17,137 it is creating a new task force 29 00:02:17,180 --> 00:02:19,182 to focus on artificial intelligence. 30 00:02:19,226 --> 00:02:21,271 Google has already announced plans 31 00:02:21,315 --> 00:02:23,360 to invest more in AI research. 32 00:02:23,404 --> 00:02:28,713 MALE NEWS ANCHOR: IBM sending an artificial intelligence robot into space. 33 00:02:28,757 --> 00:02:31,673 DULLEA: Because of my small role in the conversation, 34 00:02:31,716 --> 00:02:34,154 I understand that we're on a journey, 35 00:02:34,197 --> 00:02:37,244 one that starts with narrow AI. 36 00:02:37,287 --> 00:02:39,942 Machines that perform specific tasks 37 00:02:39,985 --> 00:02:41,378 better than we can. 38 00:02:42,336 --> 00:02:44,381 But the AI of science fiction 39 00:02:44,425 --> 00:02:47,471 is artificial general intelligence, 40 00:02:47,515 --> 00:02:50,213 machines that can do everything 41 00:02:50,257 --> 00:02:51,693 better than us. 42 00:02:52,694 --> 00:02:54,478 Narrow AI is really 43 00:02:54,522 --> 00:02:57,220 like an idiot savant to the extreme, 44 00:02:57,264 --> 00:02:58,830 where it can do something, 45 00:02:58,874 --> 00:03:00,963 like multiply numbers really fast, 46 00:03:01,006 --> 00:03:02,617 way better than any human. 47 00:03:02,660 --> 00:03:06,011 But it can't do anything else whatsoever. 48 00:03:06,055 --> 00:03:09,928 But AGI, artificial general intelligence, 49 00:03:09,972 --> 00:03:12,453 which doesn't exist yet, is instead 50 00:03:12,496 --> 00:03:14,629 intelligence that can do everything 51 00:03:14,672 --> 00:03:15,934 as well as humans can. 52 00:03:21,244 --> 00:03:24,247 DULLEA: So what happens if we do create AGI? 53 00:03:25,248 --> 00:03:27,294 Science fiction authors have warned us 54 00:03:27,337 --> 00:03:28,773 it could be dangerous. 55 00:03:29,513 --> 00:03:31,385 And they're not the only ones. 56 00:03:32,864 --> 00:03:35,911 ELON MUSK: I think people should be concerned about it. 57 00:03:35,954 --> 00:03:37,565 AI is a fundamental risk 58 00:03:37,608 --> 00:03:39,349 to the existence of human civilization. 59 00:03:39,393 --> 00:03:42,178 And I don't think people fully appreciate that. 60 00:03:42,222 --> 00:03:43,919 I do think we have to worry about it. 61 00:03:43,962 --> 00:03:45,225 I don't think it's inherent 62 00:03:45,268 --> 00:03:48,010 as we create super intelligence 63 00:03:48,053 --> 00:03:49,533 that it will necessarily always 64 00:03:49,577 --> 00:03:51,927 have the same goals in mind that we do. 65 00:03:51,970 --> 00:03:53,233 I think the development 66 00:03:53,276 --> 00:03:55,713 of full artificial intelligence 67 00:03:55,757 --> 00:03:58,368 could spell the end of the human race. 68 00:03:58,412 --> 00:04:00,457 [ECHOING] ...the end of the human race. 69 00:04:01,893 --> 00:04:05,201 Tomorrow, if the headline said an artificial intelligence 70 00:04:05,245 --> 00:04:08,248 has been manufactured that's as intelligent 71 00:04:08,291 --> 00:04:10,293 or more intelligent than human beings, 72 00:04:10,337 --> 00:04:12,077 people would argue about it, but I don't think 73 00:04:12,121 --> 00:04:13,992 they'd be surprised, because science fiction 74 00:04:14,036 --> 00:04:15,907 has helped us imagine all kinds 75 00:04:15,951 --> 00:04:17,431 of unimaginable things. 76 00:04:20,999 --> 00:04:23,393 DULLEA: Experts are divided about the impact 77 00:04:23,437 --> 00:04:25,613 AI will have on our future, 78 00:04:25,656 --> 00:04:29,747 and whether or not Hollywood is helping us plan for it. 79 00:04:29,791 --> 00:04:32,750 OREN ETZIONI: Our dialogue has really been hijacked by Hollywood 80 00:04:32,794 --> 00:04:35,927 because The Terminator makes a better blockbuster 81 00:04:35,971 --> 00:04:38,147 than AI being good, 82 00:04:38,190 --> 00:04:39,409 or AI being neutral, 83 00:04:39,453 --> 00:04:41,063 or AI being confused. 84 00:04:41,106 --> 00:04:42,456 Technology here in AI 85 00:04:42,499 --> 00:04:44,501 is no different. It's a tool. 86 00:04:44,545 --> 00:04:47,156 I like to think of it as a fancy pencil. 87 00:04:47,199 --> 00:04:49,593 And so what pictures we draw with it, 88 00:04:49,637 --> 00:04:52,248 that's up to us as a society. 89 00:04:52,292 --> 00:04:55,251 We don't have to do what AI tells us. 90 00:04:55,295 --> 00:04:56,861 We tell AI what to do. 91 00:04:56,905 --> 00:04:58,602 [BEEPING] 92 00:04:58,646 --> 00:05:00,300 DAVID MALONE: I think Hollywood films 93 00:05:00,343 --> 00:05:02,214 do help the conversation. 94 00:05:02,258 --> 00:05:03,781 Someone needs to be thinking 95 00:05:03,825 --> 00:05:07,089 ahead of what the consequences are. 96 00:05:07,132 --> 00:05:08,786 But just as we can't really imagine 97 00:05:08,830 --> 00:05:11,311 what the actual science of the future is going to be like, 98 00:05:11,354 --> 00:05:13,400 I don't think we've begun 99 00:05:13,443 --> 00:05:16,316 to really think about all of the possible futures, 100 00:05:16,359 --> 00:05:19,188 which logically spring out of this technology. 101 00:05:19,231 --> 00:05:23,497 MALE REPORTER: Artificial intelligence has the power to change society. 102 00:05:23,540 --> 00:05:25,499 A growing chorus of criticism 103 00:05:25,542 --> 00:05:27,283 is highlighting the dangers 104 00:05:27,327 --> 00:05:29,416 of handing control to machines. 105 00:05:29,459 --> 00:05:31,331 There's gonna be a lot of change coming. 106 00:05:31,374 --> 00:05:32,897 The larger long-term concern 107 00:05:32,941 --> 00:05:35,639 is that humanity will be sort of shunted aside. 108 00:05:35,683 --> 00:05:37,598 NEWS ANCHOR: This is something Stanley Kubrick and others 109 00:05:37,641 --> 00:05:39,469 were worried about 50 years ago, right? 110 00:05:41,993 --> 00:05:43,647 DULLEA: It's happening. 111 00:05:43,691 --> 00:05:47,608 How do we gauge the urgency of this conversation? 112 00:05:49,436 --> 00:05:51,742 ROSENBERG: If people saw on radar right now 113 00:05:51,786 --> 00:05:54,179 that an alien spaceship was approaching Earth 114 00:05:54,223 --> 00:05:56,181 and it was 25 years away, 115 00:05:56,225 --> 00:05:57,835 we would be mobilized to prepare 116 00:05:57,879 --> 00:06:00,621 for that alien's arrival 25 years from now. 117 00:06:00,664 --> 00:06:03,058 But that's exactly the situation that we're in 118 00:06:03,101 --> 00:06:04,799 with artificial intelligence. 119 00:06:04,842 --> 00:06:06,670 Could be 25 years, could be 50 years, 120 00:06:06,714 --> 00:06:09,456 but an alien intelligence will arrive 121 00:06:09,499 --> 00:06:11,109 and we should be prepared. 122 00:06:14,286 --> 00:06:16,637 MAN: Well, gentlemen, meet Tobor. 123 00:06:16,680 --> 00:06:19,379 An electronic simulacrum of a man. 124 00:06:19,422 --> 00:06:21,772 Oh, gosh. Oh, gee willikers. 125 00:06:21,816 --> 00:06:24,645 Even though much work remains before he is completed, 126 00:06:24,688 --> 00:06:26,211 he is already a sentient being. 127 00:06:26,255 --> 00:06:28,344 [MURMURING] 128 00:06:28,388 --> 00:06:30,390 RODNEY BROOKS: Back in 1956 when the term 129 00:06:30,433 --> 00:06:32,696 artificial intelligence first came about, 130 00:06:32,740 --> 00:06:35,395 the original goals of artificial intelligence 131 00:06:35,438 --> 00:06:37,701 were human-level intelligence. 132 00:06:37,745 --> 00:06:41,052 He does look almost kind, doesn't he?[BEEPING] 133 00:06:41,096 --> 00:06:43,838 BROOKS: Over time that's proved to be really difficult. 134 00:06:46,449 --> 00:06:49,626 I think, eventually, we will have human-level 135 00:06:49,670 --> 00:06:51,889 intelligence from machines. 136 00:06:51,933 --> 00:06:55,153 But it may be a few hundred years. It's gonna take a while. 137 00:06:55,197 --> 00:06:57,634 And so people are getting a little over-excited 138 00:06:57,678 --> 00:06:59,854 about the future capabilities. 139 00:06:59,897 --> 00:07:03,945 And now, Elektro, I command you to walk back. 140 00:07:04,293 --> 00:07:05,425 Back. 141 00:07:08,297 --> 00:07:09,559 MALONE: Almost every generation, 142 00:07:09,603 --> 00:07:11,169 people have got so enthusiastic. 143 00:07:11,213 --> 00:07:12,736 They watch 2001, 144 00:07:12,780 --> 00:07:14,564 they fall in love with HAL, they think, 145 00:07:14,608 --> 00:07:16,871 "Yeah, give me six weeks. I'll get this sorted." 146 00:07:17,785 --> 00:07:19,830 It doesn't happen and everyone gets upset. 147 00:07:21,179 --> 00:07:24,052 DULLEA: So what's changed in AI research? 148 00:07:26,489 --> 00:07:28,839 MALONE: The difference with the AI we have today 149 00:07:28,883 --> 00:07:32,495 and the reason AI suddenly took this big leap forward 150 00:07:32,539 --> 00:07:34,758 is the Internet. That is their world. 151 00:07:34,802 --> 00:07:37,587 Tonight the information superhighway, 152 00:07:37,631 --> 00:07:39,415 and one of its main thoroughfares, 153 00:07:39,459 --> 00:07:42,505 an online network called Internet. 154 00:07:42,549 --> 00:07:45,421 MALE REPORTER: In 1981, only 213 computers 155 00:07:45,465 --> 00:07:46,901 were hooked to the Internet. 156 00:07:46,944 --> 00:07:48,816 As the new year begins, an estimated 157 00:07:48,859 --> 00:07:50,600 two-and-a-half million computers 158 00:07:50,644 --> 00:07:51,993 will be on the network. 159 00:07:53,516 --> 00:07:55,126 AIs need to eat. 160 00:07:55,170 --> 00:07:57,346 Just like sheep need grass, 161 00:07:57,389 --> 00:07:59,087 AIs need data. 162 00:07:59,130 --> 00:08:02,525 And the great prairies of data are on the Internet. 163 00:08:03,134 --> 00:08:04,484 That's what it is. 164 00:08:05,136 --> 00:08:06,181 They said, "You know what? 165 00:08:06,224 --> 00:08:07,661 "Let's just let them loose out there." 166 00:08:09,532 --> 00:08:11,839 And suddenly, the AIs came to life. 167 00:08:11,882 --> 00:08:14,276 FEMALE REPORTER: Imagine thousands of talking robots 168 00:08:14,319 --> 00:08:15,973 able to move as one unit, 169 00:08:16,017 --> 00:08:17,322 taking on crime, fires, 170 00:08:17,366 --> 00:08:18,976 and natural disasters. 171 00:08:19,020 --> 00:08:21,805 Artificial intelligence platform Amper 172 00:08:21,849 --> 00:08:24,416 has created the first album entirely composed 173 00:08:24,460 --> 00:08:26,157 and produced by artificial intelligence. 174 00:08:26,201 --> 00:08:29,552 Next week, Christie's will be the first auction house 175 00:08:29,596 --> 00:08:33,251 to offer artwork created by artificial intelligence. 176 00:08:34,165 --> 00:08:35,819 Driverless cars, said to be 177 00:08:35,863 --> 00:08:38,648 in our not-so-distant future. 178 00:08:38,692 --> 00:08:40,955 DULLEA: So even though we don't have AGI 179 00:08:40,998 --> 00:08:44,262 or human-level intelligence yet, 180 00:08:44,306 --> 00:08:45,699 are we ready to give autonomy 181 00:08:45,742 --> 00:08:48,440 to the machines we do have? 182 00:08:48,484 --> 00:08:51,574 Should self-driving cars make ethical decisions? 183 00:08:51,618 --> 00:08:53,881 That is the growing debate as the technology 184 00:08:53,924 --> 00:08:56,318 moves closer towards mainstream reality. 185 00:08:58,233 --> 00:08:59,843 MARY CUMMINGS: This idea of whether or not 186 00:08:59,887 --> 00:09:02,846 we can embed ethics into machines, 187 00:09:02,890 --> 00:09:04,544 whether or not they are automated machines 188 00:09:04,587 --> 00:09:06,415 or autonomous machines, 189 00:09:06,458 --> 00:09:08,286 I'm not sure we've got that figured out. 190 00:09:08,330 --> 00:09:09,723 If you have a self-driving car, 191 00:09:09,766 --> 00:09:11,202 it's gonna have to make ethical decisions. 192 00:09:11,246 --> 00:09:14,641 A classic one people have been thinking a lot about is... 193 00:09:14,684 --> 00:09:16,556 I don't know if you've heard of trolley problems. 194 00:09:16,599 --> 00:09:18,122 The trolley problem.The trolley problem. 195 00:09:18,166 --> 00:09:20,211 The trolley problem.The trolley problem. 196 00:09:21,604 --> 00:09:22,823 CUMMINGS: In the trolley problem, 197 00:09:22,866 --> 00:09:24,172 whether or not a driverless car 198 00:09:24,215 --> 00:09:26,870 is going to save the driver or run over 199 00:09:26,914 --> 00:09:29,046 a group of school children crossing the road. 200 00:09:29,090 --> 00:09:31,614 ETZIONI: Will the car go right and hit the old lady 201 00:09:31,658 --> 00:09:35,575 or will it go left and kill the four young people? 202 00:09:35,618 --> 00:09:39,143 MICHAL KOSINSKI: Either kill five people standing on a bus stop 203 00:09:39,187 --> 00:09:40,405 or kill yourself. 204 00:09:40,449 --> 00:09:41,972 It's an ethical decision 205 00:09:42,016 --> 00:09:43,931 that has to be made in the moment, 206 00:09:43,974 --> 00:09:46,020 too fast for any person to get in there. 207 00:09:46,063 --> 00:09:48,544 So it's the software in your self-driving car 208 00:09:48,588 --> 00:09:50,546 that's gonna make that ethical decision. 209 00:09:50,590 --> 00:09:51,852 Who decides that? 210 00:09:52,679 --> 00:09:53,897 KOSINSKI: Smart society, 211 00:09:53,941 --> 00:09:55,899 society that values human life, 212 00:09:55,943 --> 00:09:58,336 should make a decision that 213 00:09:58,380 --> 00:10:02,427 killing one person is preferable to killing five people. 214 00:10:02,471 --> 00:10:06,344 Of course, it does raise the question who's gonna buy that car. 215 00:10:06,388 --> 00:10:09,130 Or, you know, are you gonna buy the car where actually, 216 00:10:09,173 --> 00:10:11,654 [WHISPERS] it'll swerve and you'll be fine, 217 00:10:11,698 --> 00:10:13,134 but they're goners. 218 00:10:14,918 --> 00:10:17,442 ETZIONI: There are hundreds of those kinds of questions 219 00:10:17,486 --> 00:10:19,140 all throughout society. 220 00:10:19,183 --> 00:10:22,883 And so, ethical philosophers battle back and forth. 221 00:10:25,146 --> 00:10:27,583 ETZIONI: My peeve with the philosophers is 222 00:10:27,627 --> 00:10:30,804 they're sucking the oxygen out of the room. 223 00:10:30,847 --> 00:10:33,763 There's this metaphor that I have of 224 00:10:33,807 --> 00:10:37,114 you have a car driving away from a wedding, 225 00:10:37,158 --> 00:10:41,205 and the car is driving, and behind it there are cans. 226 00:10:41,249 --> 00:10:45,645 And the metal cans are clanking and making noise. 227 00:10:45,688 --> 00:10:47,690 That car is science 228 00:10:47,734 --> 00:10:50,345 and technology driving forward decisively. 229 00:10:50,388 --> 00:10:51,955 And the metal cans clanking 230 00:10:51,999 --> 00:10:54,697 are the philosophers making a lot of noise. 231 00:10:54,741 --> 00:10:56,525 The world is changing in such a way 232 00:10:56,568 --> 00:11:01,138 that the question of us building machines 233 00:11:01,182 --> 00:11:04,098 that could become more intelligent than we are 234 00:11:04,141 --> 00:11:07,710 is more of a reality than a fiction right now. 235 00:11:10,147 --> 00:11:12,759 DULLEA: Don't we have to ask questions now, 236 00:11:12,802 --> 00:11:15,805 before the machines get smarter than us? 237 00:11:15,849 --> 00:11:18,373 I mean, what if they achieve consciousness? 238 00:11:19,635 --> 00:11:21,463 How far away is that? 239 00:11:25,075 --> 00:11:26,816 How close we are to a conscious machine? 240 00:11:26,860 --> 00:11:28,688 That is the question that brings 241 00:11:28,731 --> 00:11:32,082 philosophers and scientists to blows, I think. 242 00:11:32,126 --> 00:11:34,302 You are dead center 243 00:11:34,345 --> 00:11:37,871 of the greatest scientific event in the history of man. 244 00:11:37,914 --> 00:11:39,699 If you've created a conscious machine, 245 00:11:39,742 --> 00:11:41,309 it's not the history of man. 246 00:11:42,614 --> 00:11:44,834 That's the history of gods. 247 00:11:44,878 --> 00:11:48,533 The problem with consciousness is hardly anybody 248 00:11:48,577 --> 00:11:50,884 can define exactly what it is. 249 00:11:53,408 --> 00:11:55,976 DULLEA: Is it consciousness that makes us special? 250 00:11:57,586 --> 00:11:58,718 Human? 251 00:12:00,894 --> 00:12:03,026 MALONE: A way of thinking about consciousness 252 00:12:03,070 --> 00:12:05,159 which a lot of people will agree with... 253 00:12:05,202 --> 00:12:07,857 You'll never find something that appeals to everyone. 254 00:12:07,901 --> 00:12:10,468 ...is it's not just flashing lights, but someone's home. 255 00:12:11,252 --> 00:12:13,036 So if you have a... 256 00:12:13,080 --> 00:12:15,386 If you bought one of those little 257 00:12:15,430 --> 00:12:16,953 simple chess-playing machines, 258 00:12:17,606 --> 00:12:19,173 it might beat you at chess. 259 00:12:19,216 --> 00:12:21,218 There are lots of lights flashing. 260 00:12:21,262 --> 00:12:23,743 But there's nobody in when it wins, 261 00:12:23,786 --> 00:12:25,222 going "I won." 262 00:12:25,962 --> 00:12:29,749 All right? Whereas your dog 263 00:12:29,792 --> 00:12:32,534 is a rubbish chess player, 264 00:12:32,577 --> 00:12:33,927 but you look at the dog and you think, 265 00:12:33,970 --> 00:12:36,930 "Someone's actually in. It's not just a mechanism." 266 00:12:38,845 --> 00:12:42,283 People wonder whether general intelligence 267 00:12:42,326 --> 00:12:44,459 requires consciousness. 268 00:12:46,766 --> 00:12:50,160 It might partly depend on what level of general intelligence you're talking about. 269 00:12:50,204 --> 00:12:53,250 A mouse has some kind of general intelligence. 270 00:12:53,294 --> 00:12:55,339 My guess is, if you've got a machine which has 271 00:12:55,383 --> 00:12:58,473 general intelligence at the level of, say, a mouse or beyond, 272 00:12:58,516 --> 00:13:00,997 it's gonna be a good contender for having consciousness. 273 00:13:04,653 --> 00:13:05,697 SAM HARRIS: We're not there yet. 274 00:13:05,741 --> 00:13:07,787 We're not confronted by humanoid robots 275 00:13:07,830 --> 00:13:09,832 that are smarter than we are. 276 00:13:09,876 --> 00:13:12,182 We're in this curious ethical position 277 00:13:12,226 --> 00:13:15,446 of not knowing where and how consciousness emerges, 278 00:13:15,490 --> 00:13:18,841 and we're building increasingly complicated minds. 279 00:13:18,885 --> 00:13:20,800 We won't know whether or not they're conscious, 280 00:13:20,843 --> 00:13:22,062 but we might feel they are. 281 00:13:24,194 --> 00:13:26,718 MALONE: The easiest example is if you programmed 282 00:13:26,762 --> 00:13:29,634 into a computer everything you knew about joy, 283 00:13:29,678 --> 00:13:31,201 every reference in literature, 284 00:13:31,245 --> 00:13:33,943 every definition, every chemical compound 285 00:13:33,987 --> 00:13:36,119 that's involved with joy in the human brain, 286 00:13:36,163 --> 00:13:39,644 and you put all of that information in a computer, 287 00:13:39,688 --> 00:13:42,778 you could argue that it understood joy. 288 00:13:42,822 --> 00:13:45,476 The question is, had it ever felt joy? 289 00:13:51,613 --> 00:13:53,354 DULLEA: Did HAL have emotions? 290 00:13:55,095 --> 00:13:58,228 Will AIs ever feel the way we do? 291 00:13:59,882 --> 00:14:02,406 MICHAEL LITTMAN: It's very hard to get machines to do the things 292 00:14:02,450 --> 00:14:06,149 that we don't think about, that we're not conscious of. 293 00:14:06,193 --> 00:14:09,326 Things like detecting that that person over there 294 00:14:09,370 --> 00:14:11,198 is really unhappy with the way that... 295 00:14:11,241 --> 00:14:12,721 Something about what I'm doing. 296 00:14:12,764 --> 00:14:14,027 And they're sending a subtle signal 297 00:14:14,070 --> 00:14:15,506 and I probably should change what I'm doing. 298 00:14:17,334 --> 00:14:18,988 We don't know how to build machines that can actually be 299 00:14:19,032 --> 00:14:22,252 tuned in to those kinds of tiny little social cues. 300 00:14:24,994 --> 00:14:28,389 There's this notion that maybe there is something magical 301 00:14:28,432 --> 00:14:30,434 about brains, 302 00:14:30,478 --> 00:14:32,045 about the wetware of having 303 00:14:32,088 --> 00:14:33,785 an information processing system 304 00:14:33,829 --> 00:14:34,917 made of meat, right? 305 00:14:34,961 --> 00:14:37,311 And that whatever we do in our machines, 306 00:14:37,354 --> 00:14:40,705 however competent they begin to seem, 307 00:14:40,749 --> 00:14:42,272 they never really will be intelligent 308 00:14:42,316 --> 00:14:44,579 in the way that we experience ourselves to be. 309 00:14:46,233 --> 00:14:48,975 There's really no basis for that. 310 00:14:51,238 --> 00:14:53,240 TEGMARK: This silly idea you can't be intelligent 311 00:14:53,283 --> 00:14:55,155 unless you're made of meat. 312 00:14:56,765 --> 00:14:58,201 From my perspective, as a physicist, 313 00:14:58,245 --> 00:15:00,682 that's just carbon chauvinism. 314 00:15:00,725 --> 00:15:02,118 Those machines are made of exactly 315 00:15:02,162 --> 00:15:03,903 the same kind of elementary particles 316 00:15:03,946 --> 00:15:05,382 that my brain is made out of. 317 00:15:06,775 --> 00:15:08,429 Intelligence is just a certain... 318 00:15:08,472 --> 00:15:10,170 And consciousness is just 319 00:15:10,213 --> 00:15:12,259 a certain kind of information processing. 320 00:15:15,262 --> 00:15:17,655 DULLEA: Maybe we're not so special after all. 321 00:15:19,744 --> 00:15:21,268 Could our carbon-based brains 322 00:15:21,311 --> 00:15:23,531 be replicated with silicon? 323 00:15:25,185 --> 00:15:28,405 We're trying to build a computer-generated baby 324 00:15:28,449 --> 00:15:31,060 that we can teach like a real baby, 325 00:15:31,104 --> 00:15:34,237 with the potential for general intelligence. 326 00:15:39,808 --> 00:15:42,550 To do that, you really have to build a computer brain. 327 00:15:46,162 --> 00:15:50,645 I wanted to build a simple toy brain model 328 00:15:52,168 --> 00:15:53,474 in an embodied system 329 00:15:53,517 --> 00:15:56,694 which you could interact with face-to-face. 330 00:16:00,394 --> 00:16:03,527 And I happen to have an infant at home, 331 00:16:03,571 --> 00:16:04,964 a real baby. 332 00:16:05,660 --> 00:16:07,836 And I scanned the baby 333 00:16:07,879 --> 00:16:10,795 so I could build a 3D model out of her 334 00:16:10,839 --> 00:16:12,493 and then use her face 335 00:16:12,536 --> 00:16:15,452 as basically the embodiment of the system. 336 00:16:17,063 --> 00:16:20,109 If a brain which is made out of cells, 337 00:16:20,153 --> 00:16:23,199 and it's got blood pumping, is able to think, 338 00:16:24,461 --> 00:16:27,682 it just may be possible that a computer can think 339 00:16:28,770 --> 00:16:33,166 if the information is moving in the same sort of process. 340 00:16:37,474 --> 00:16:39,476 SAGAR: A lot of artificial intelligence at the moment today 341 00:16:39,520 --> 00:16:42,044 is really about advanced pattern recognition. 342 00:16:42,088 --> 00:16:44,003 It's not about killer robots 343 00:16:44,046 --> 00:16:45,830 that are gonna take over the world, et cetera. 344 00:16:45,874 --> 00:16:47,658 [AUDIENCE LAUGHING] 345 00:16:47,702 --> 00:16:51,053 So this is Baby X. She's running live on my computer. 346 00:16:51,097 --> 00:16:54,056 So as I move my hand around, make a loud noise, 347 00:16:54,100 --> 00:16:55,710 [CLAPS] she'll get a fright. 348 00:16:55,753 --> 00:16:57,407 We can kind of zoom in. 349 00:16:57,451 --> 00:17:00,845 So she basically can see me and hear me, you know. 350 00:17:00,889 --> 00:17:03,022 Hey, sweetheart.[AUDIENCE LAUGHING] 351 00:17:03,065 --> 00:17:05,067 She's not copying my smile, 352 00:17:05,111 --> 00:17:07,417 she's responding to my smile. 353 00:17:08,114 --> 00:17:09,637 So she's responding to me 354 00:17:09,680 --> 00:17:12,205 and we're really concerned with the question 355 00:17:12,248 --> 00:17:14,207 of how will we actually interact 356 00:17:14,250 --> 00:17:16,644 with artificial intelligence in the future. 357 00:17:17,688 --> 00:17:19,995 Ah... 358 00:17:20,039 --> 00:17:23,520 When I'm talking to Baby X, I'm deliberately modulating my voice 359 00:17:23,564 --> 00:17:25,740 and deliberately doing big facial expressions. 360 00:17:25,783 --> 00:17:27,350 I'm going "Ooh," all this stuff 361 00:17:27,394 --> 00:17:29,309 because I'm getting her attention. 362 00:17:29,744 --> 00:17:31,572 Ooh. 363 00:17:32,834 --> 00:17:34,749 And now that I've got her attention, 364 00:17:35,793 --> 00:17:37,012 I can teach something. 365 00:17:37,056 --> 00:17:38,579 Okay, so here's Baby X, 366 00:17:38,622 --> 00:17:41,973 and this is... She's been learning to read words. 367 00:17:42,017 --> 00:17:43,758 So here's her first word book. 368 00:17:43,801 --> 00:17:47,370 So let's see what she can see. Turn to a page... 369 00:17:47,414 --> 00:17:49,329 And here we go. Let's see what she... 370 00:17:49,372 --> 00:17:51,679 What's this, Baby? What's this? What's this? 371 00:17:52,245 --> 00:17:54,160 What's this? Sheep. 372 00:17:54,203 --> 00:17:56,336 Good girl. Now see if she knows what the word is. 373 00:17:56,379 --> 00:17:58,207 Okay. Baby, look over here. 374 00:17:58,251 --> 00:18:00,079 Off you go. What's this? What's this? 375 00:18:01,776 --> 00:18:03,952 Sheep.Good girl. 376 00:18:03,995 --> 00:18:05,606 Let's try her on something else. 377 00:18:05,649 --> 00:18:07,216 Okay. 378 00:18:07,260 --> 00:18:08,826 [WHISPERS] Let's try on that. [NORMAL VOICE] What's this? 379 00:18:08,870 --> 00:18:10,437 Baby, Baby, over here. What's this? 380 00:18:15,268 --> 00:18:17,618 Baby. Look at me. Look at me. What's this? 381 00:18:17,661 --> 00:18:19,402 Baby, over here. Over here. 382 00:18:20,708 --> 00:18:22,927 Puppy.Good girl. 383 00:18:22,971 --> 00:18:24,233 That's what she's just read. 384 00:18:24,277 --> 00:18:26,148 Ooh.[LAUGHS] 385 00:18:26,192 --> 00:18:28,411 The most intelligent system 386 00:18:28,455 --> 00:18:30,805 that we're aware of is a human. 387 00:18:32,198 --> 00:18:35,114 So using the human as a template, 388 00:18:35,157 --> 00:18:37,333 as the embodiment of Baby X, 389 00:18:37,377 --> 00:18:39,509 is probably the best way to try 390 00:18:39,553 --> 00:18:41,337 to achieve general intelligence. 391 00:18:48,170 --> 00:18:49,650 DULLEA: She seems so human. 392 00:18:51,304 --> 00:18:53,349 Is this the child who takes our hand 393 00:18:53,393 --> 00:18:55,699 and leads us to HAL? 394 00:18:57,484 --> 00:19:00,922 ANDY CLARK: Whether it's in a real world or a virtual world, 395 00:19:00,965 --> 00:19:02,358 I do think embodiment 396 00:19:02,402 --> 00:19:05,492 is gonna be very crucial here. 397 00:19:05,535 --> 00:19:07,885 Because I think that general intelligence 398 00:19:07,929 --> 00:19:10,888 is tied up with being able to act in the world. 399 00:19:13,108 --> 00:19:14,805 You have to have something, 400 00:19:14,849 --> 00:19:16,633 arms, legs, eyes, 401 00:19:16,677 --> 00:19:18,461 something that you can move around 402 00:19:18,505 --> 00:19:21,682 in order to harvest new information. 403 00:19:26,513 --> 00:19:28,906 So I think to get systems that really understand 404 00:19:28,950 --> 00:19:30,647 what they're talking about, 405 00:19:30,691 --> 00:19:32,997 you will need systems that can act 406 00:19:33,041 --> 00:19:34,564 and intervene in the world, 407 00:19:34,608 --> 00:19:36,827 and possibly systems that have something 408 00:19:36,871 --> 00:19:39,308 almost like a social life. 409 00:19:39,352 --> 00:19:43,007 For me, thinking and feeling will come together. 410 00:19:43,051 --> 00:19:44,574 And they will come together 411 00:19:44,618 --> 00:19:47,534 because of the importance of embodiment. 412 00:19:52,539 --> 00:19:53,975 The stories that we tell ourselves 413 00:19:54,018 --> 00:19:56,325 open up arenas for discussion, 414 00:19:56,369 --> 00:19:58,109 and they also help us think about things. 415 00:19:58,153 --> 00:19:59,415 Ava! 416 00:20:02,592 --> 00:20:04,377 BRYAN JOHNSON: The most dominant story... 417 00:20:04,420 --> 00:20:05,769 Go back to your room. 418 00:20:05,813 --> 00:20:07,684 ...is AI is scary. 419 00:20:07,728 --> 00:20:09,991 There's a bad guy, there's a good guy... 420 00:20:12,123 --> 00:20:14,822 That is a non-productive storyline. 421 00:20:14,865 --> 00:20:15,866 Stop! 422 00:20:16,954 --> 00:20:18,608 Ava, I said stop! 423 00:20:20,784 --> 00:20:22,656 Whoa! Whoa! Whoa! 424 00:20:22,699 --> 00:20:27,269 The reality is AI is remarkably useful to us. 425 00:20:27,313 --> 00:20:30,838 MALE REPORTER: Robot surgeons are increasingly used in NHS hospitals. 426 00:20:30,881 --> 00:20:33,623 They don't get tired, their hands don't tremble, 427 00:20:33,667 --> 00:20:36,060 and their patients make a faster recovery. 428 00:20:36,104 --> 00:20:38,672 MAN: It alerted me that there was one animal in heat. 429 00:20:38,715 --> 00:20:41,065 She was ready to be bred, and three animals that 430 00:20:41,109 --> 00:20:42,806 may have a potential health problem. 431 00:20:42,850 --> 00:20:45,548 MAN: A Facebook program can detect suicidal behavior 432 00:20:45,592 --> 00:20:47,681 by analyzing user information, 433 00:20:47,724 --> 00:20:50,074 and Facebook CEO Mark Zuckerberg believes 434 00:20:50,118 --> 00:20:52,425 this is just the beginning. 435 00:20:52,468 --> 00:20:55,036 MAN 2: Here is someone who has visual impairment 436 00:20:55,079 --> 00:20:57,386 being able to see because of AI. 437 00:20:57,430 --> 00:21:00,128 Those are the benefits of what AI does. 438 00:21:00,171 --> 00:21:01,738 Diagnose cancer earlier. 439 00:21:01,782 --> 00:21:03,174 Predict heart disease. 440 00:21:03,218 --> 00:21:05,481 MAN: But the revolution has only just begun. 441 00:21:06,917 --> 00:21:08,397 AI is. 442 00:21:08,441 --> 00:21:10,617 It's happening. And so it is a question 443 00:21:10,660 --> 00:21:13,533 of what we do with it, how we build it. 444 00:21:13,576 --> 00:21:15,665 That's the only question that matters. And I think, 445 00:21:15,709 --> 00:21:19,060 we potentially are at risk of ruining that 446 00:21:19,103 --> 00:21:20,453 by having this knee-jerk reaction 447 00:21:20,496 --> 00:21:22,672 that AI is bad and it's coming to get us. 448 00:21:25,196 --> 00:21:27,155 DULLEA: So just like my astronaut, 449 00:21:27,198 --> 00:21:28,461 we're on a journey. 450 00:21:29,200 --> 00:21:31,638 Narrow AI is already here. 451 00:21:32,421 --> 00:21:35,294 But how far are we from AGI? 452 00:21:36,425 --> 00:21:39,210 And how fast are we going? 453 00:21:39,254 --> 00:21:40,603 JURGEN SCHMIDHUBER: Every five years, 454 00:21:40,647 --> 00:21:43,214 computers are getting ten times faster. 455 00:21:43,258 --> 00:21:45,216 At the moment, we don't have yet 456 00:21:45,260 --> 00:21:48,263 little devices, computational devices, 457 00:21:48,307 --> 00:21:51,484 that can compute as much as a human brain. 458 00:21:51,527 --> 00:21:52,702 But every five years, 459 00:21:52,746 --> 00:21:55,575 computers are getting ten times cheaper, 460 00:21:55,618 --> 00:21:57,925 which means that soon we are going to have that. 461 00:21:57,968 --> 00:22:01,189 And once we have that, then 50 years later, 462 00:22:01,232 --> 00:22:02,451 if the trend doesn't break, 463 00:22:02,495 --> 00:22:04,801 we should have for the same price 464 00:22:04,845 --> 00:22:07,630 little computational devices that can compute 465 00:22:07,674 --> 00:22:10,329 as much as all ten billion brains 466 00:22:10,372 --> 00:22:11,808 of humankind together. 467 00:22:14,202 --> 00:22:15,551 DULLEA: It's catching up. 468 00:22:16,770 --> 00:22:20,034 What happens if it passes us by? 469 00:22:20,077 --> 00:22:23,342 Now a robot that's been developed by a Japanese scientist 470 00:22:23,385 --> 00:22:25,387 is so fast, it can win 471 00:22:25,431 --> 00:22:27,346 the rock-paper-scissors game against a human 472 00:22:27,389 --> 00:22:29,130 every single time. 473 00:22:29,173 --> 00:22:31,175 The presumption has to be that 474 00:22:31,219 --> 00:22:33,177 when an artificial intelligence 475 00:22:33,221 --> 00:22:35,266 reaches human intelligence, 476 00:22:35,310 --> 00:22:38,008 it will very quickly exceed human intelligence, 477 00:22:38,052 --> 00:22:39,227 and once that happens, 478 00:22:39,270 --> 00:22:41,272 it doesn't even need us anymore 479 00:22:41,316 --> 00:22:43,884 to make it smarter. It can make itself smarter. 480 00:22:46,408 --> 00:22:49,455 MALONE: You create AIs which are better than us 481 00:22:49,498 --> 00:22:50,586 at a lot of things. 482 00:22:50,630 --> 00:22:52,545 And one of the things they get better at 483 00:22:52,588 --> 00:22:54,373 is programming computers. 484 00:22:54,416 --> 00:22:56,113 So it reprograms itself. 485 00:22:56,157 --> 00:22:57,898 And it is now better than it was, 486 00:22:57,941 --> 00:23:01,380 which means it's now better than it was at reprogramming. 487 00:23:02,729 --> 00:23:05,384 So it doesn't get better like that. 488 00:23:06,123 --> 00:23:07,690 It gets better like that. 489 00:23:09,562 --> 00:23:11,172 ROSENBERG: And we get left behind 490 00:23:11,215 --> 00:23:13,304 as it keeps getting faster, 491 00:23:13,348 --> 00:23:15,872 evolving at an exponential rate. 492 00:23:15,916 --> 00:23:18,005 That's this event that a lot of researchers 493 00:23:18,048 --> 00:23:19,746 refer to as the singularity. 494 00:23:21,965 --> 00:23:23,706 And it will get beyond 495 00:23:23,750 --> 00:23:26,405 where we are competitive very quickly. 496 00:23:26,448 --> 00:23:29,625 So think about someone not having a PhD in one area, 497 00:23:29,669 --> 00:23:32,280 but having it in every conceivable area. 498 00:23:32,323 --> 00:23:34,413 You can't figure out what they're doing. 499 00:23:34,456 --> 00:23:36,806 They can do things no one else is capable of. 500 00:23:39,461 --> 00:23:40,854 HARRIS: Just imagine what it would be like 501 00:23:40,897 --> 00:23:43,247 to be in dialogue with another person 502 00:23:43,291 --> 00:23:44,814 running on a time course 503 00:23:44,858 --> 00:23:48,557 that was a million times faster than yours. 504 00:23:48,601 --> 00:23:51,255 So for every second you had to think 505 00:23:51,299 --> 00:23:52,431 about what you were going to do 506 00:23:52,474 --> 00:23:53,867 or what you were going to say, 507 00:23:53,910 --> 00:23:57,436 it had two weeks to think about this, right? 508 00:23:57,479 --> 00:24:00,569 Um, or a billion times faster, 509 00:24:00,613 --> 00:24:02,528 so if for every second you had to think, 510 00:24:02,571 --> 00:24:05,487 it it's high-stakes poker or some negotiation, 511 00:24:05,531 --> 00:24:09,448 every second you are thinking, it has 32 years 512 00:24:09,491 --> 00:24:11,798 to think about what its next move will be. 513 00:24:12,581 --> 00:24:14,931 There's no way to win in that game. 514 00:24:15,715 --> 00:24:17,456 DULLEA: The singularity, 515 00:24:17,499 --> 00:24:19,719 the moment the machines surpass us. 516 00:24:20,415 --> 00:24:22,548 Inevitable? Not everyone 517 00:24:22,591 --> 00:24:25,289 at the forefront of AI thinks so. 518 00:24:25,333 --> 00:24:27,857 Some of my friends who are worried about the future 519 00:24:27,901 --> 00:24:30,033 think that we're gonna get to a singularity 520 00:24:30,077 --> 00:24:31,687 where the AI systems are able 521 00:24:31,731 --> 00:24:33,602 to reprogram themselves and make themselves 522 00:24:33,646 --> 00:24:35,517 better and better and better. 523 00:24:36,910 --> 00:24:39,652 Maybe in the long-distant future. 524 00:24:39,695 --> 00:24:42,045 But our present AI systems 525 00:24:42,089 --> 00:24:44,831 are very, very narrow at what they do. 526 00:24:45,745 --> 00:24:47,573 We can't build a system today 527 00:24:47,616 --> 00:24:49,488 that can understand a computer program 528 00:24:49,531 --> 00:24:52,316 at the level that a brand-new student in programming 529 00:24:52,360 --> 00:24:54,754 understands in four or five weeks. 530 00:24:55,581 --> 00:24:56,756 And by the way, 531 00:24:56,799 --> 00:24:59,019 people have been working on this problem 532 00:24:59,062 --> 00:25:00,499 for over 40 years. 533 00:25:00,542 --> 00:25:02,065 It's not because people haven't tried. 534 00:25:02,109 --> 00:25:03,980 It's because we don't have any clue how to do it. 535 00:25:04,024 --> 00:25:06,766 So to imagine it's suddenly gonna jump out of nothing 536 00:25:06,809 --> 00:25:09,986 I think is completely wrong. 537 00:25:10,030 --> 00:25:11,597 [APPLAUSE] 538 00:25:11,640 --> 00:25:14,600 ETZIONI: There's a negative narrative about AI. 539 00:25:14,643 --> 00:25:16,602 Brilliant people like Elon Musk 540 00:25:16,645 --> 00:25:19,648 use very religious imagery. 541 00:25:19,692 --> 00:25:22,999 With artificial intelligence, we are summoning the demon. 542 00:25:23,043 --> 00:25:24,958 You know? You know all those stories where 543 00:25:25,001 --> 00:25:27,090 there's the guy with the pentagram and holy water 544 00:25:27,134 --> 00:25:29,005 and he's like, "Yeah," he's sure he can control the demon. 545 00:25:29,049 --> 00:25:30,093 [LAUGHTER] 546 00:25:30,137 --> 00:25:31,312 Doesn't work out. 547 00:25:32,313 --> 00:25:33,488 You know, people like me 548 00:25:33,532 --> 00:25:35,795 who've spent their entire career 549 00:25:35,838 --> 00:25:39,538 working on AI are literally hurt by that, right? 550 00:25:39,581 --> 00:25:42,105 It's a personal attack almost. 551 00:25:43,933 --> 00:25:46,196 I tend to think about what's gonna happen 552 00:25:46,240 --> 00:25:49,243 in the next ten years, in the next 25 years. 553 00:25:49,286 --> 00:25:50,549 Beyond 25, 554 00:25:50,592 --> 00:25:52,681 it's so difficult to predict 555 00:25:52,725 --> 00:25:55,379 that it becomes very much a speculation. 556 00:25:55,423 --> 00:25:56,816 But I must warn you, 557 00:25:56,859 --> 00:25:59,383 I've been specifically programmed 558 00:25:59,427 --> 00:26:02,125 not to take over the world. 559 00:26:02,169 --> 00:26:04,650 STUART RUSSELL: All of a sudden, for the first time 560 00:26:04,693 --> 00:26:07,478 in the history of the field, you have some people saying, 561 00:26:07,522 --> 00:26:09,350 "Oh, the reason we don't have to worry 562 00:26:09,393 --> 00:26:10,743 "about any kind of 563 00:26:10,786 --> 00:26:12,614 "existential risk to humanity 564 00:26:12,658 --> 00:26:13,746 "from super-intelligent machines 565 00:26:13,789 --> 00:26:15,225 "is because it's impossible. 566 00:26:16,096 --> 00:26:17,532 "We're gonna fail." 567 00:26:17,576 --> 00:26:21,275 I... I find this completely bizarre. 568 00:26:21,318 --> 00:26:26,280 DULLEA: Stuart Russell cowrote the textbook many AI students study with. 569 00:26:26,323 --> 00:26:30,893 He understands the momentum driving our AI future. 570 00:26:30,937 --> 00:26:33,766 But given the rate of investment in AI, 571 00:26:33,809 --> 00:26:35,811 the number of incredibly smart people 572 00:26:35,855 --> 00:26:37,596 who are working on these problems, 573 00:26:37,639 --> 00:26:39,728 and the enormous incentive, 574 00:26:39,772 --> 00:26:42,339 uh, to achieve general purpose intelligence, 575 00:26:42,383 --> 00:26:43,819 back of the envelope calculation, 576 00:26:43,863 --> 00:26:47,388 the value is about 10,000 trillion dollars. 577 00:26:47,431 --> 00:26:50,652 Because it would revolutionize the economy of the world. 578 00:26:51,697 --> 00:26:53,612 With that scale of incentive, 579 00:26:53,655 --> 00:26:56,266 I think betting against human ingenuity 580 00:26:56,310 --> 00:26:58,660 just seems incredibly foolish. 581 00:26:58,704 --> 00:27:01,054 China has a big vested interest 582 00:27:01,097 --> 00:27:04,187 in being recognized as, kind of, an AI superpower. 583 00:27:04,231 --> 00:27:06,625 And they're investing what I believe is 584 00:27:06,668 --> 00:27:08,931 upwards of $150 billion 585 00:27:08,975 --> 00:27:10,498 the next five to ten years. 586 00:27:10,541 --> 00:27:11,891 There have been some news flashes 587 00:27:11,934 --> 00:27:13,632 that the government will start ramping up 588 00:27:13,675 --> 00:27:15,024 its investment in AI. 589 00:27:15,068 --> 00:27:17,723 Every company in the Fortune 500, 590 00:27:17,766 --> 00:27:20,029 whether it's IBM, Microsoft, 591 00:27:20,073 --> 00:27:22,771 or they're outside like Kroger, 592 00:27:22,815 --> 00:27:24,599 is using artificial intelligence 593 00:27:24,643 --> 00:27:25,861 to gain customers. 594 00:27:25,905 --> 00:27:28,647 I don't think that people should underestimate 595 00:27:28,690 --> 00:27:30,605 the types of innovations going on. 596 00:27:30,649 --> 00:27:32,694 And what this world is gonna look like 597 00:27:32,738 --> 00:27:34,261 in three to five years from now, 598 00:27:34,304 --> 00:27:36,480 I think people are gonna have a hard time predicting it. 599 00:27:36,524 --> 00:27:38,787 I know many people who work 600 00:27:38,831 --> 00:27:40,571 in AI and they would tell you, 601 00:27:40,615 --> 00:27:42,791 "Look, I'm going to work and I have doubts. 602 00:27:42,835 --> 00:27:45,489 "I'm working on creating this machine 603 00:27:45,533 --> 00:27:47,100 "that can treat cancer, 604 00:27:47,883 --> 00:27:49,450 "extend our life expectancy, 605 00:27:49,493 --> 00:27:52,801 "make our societies better, more just, more equal. 606 00:27:54,020 --> 00:27:57,197 "But it could also be the end of humanity." 607 00:28:00,156 --> 00:28:02,376 DULLEA: The end of humanity? 608 00:28:07,033 --> 00:28:08,991 What is our plan as a species? 609 00:28:11,472 --> 00:28:12,778 We don't have one. 610 00:28:14,170 --> 00:28:15,737 We don't have a plan and we don't realize 611 00:28:15,781 --> 00:28:17,217 it's necessary to have a plan. 612 00:28:26,879 --> 00:28:28,707 CAMERON: A lot of the AI researchers 613 00:28:28,750 --> 00:28:30,143 would love for our science fiction books 614 00:28:30,186 --> 00:28:32,362 and movies to be more balanced, 615 00:28:33,494 --> 00:28:36,149 to show the good uses of AI. 616 00:28:42,677 --> 00:28:45,724 All of the bad things we tend to do, 617 00:28:45,767 --> 00:28:46,899 you play them out 618 00:28:46,942 --> 00:28:50,380 to the apocalyptic level in science fiction 619 00:28:50,424 --> 00:28:51,991 and then you show them to people 620 00:28:52,034 --> 00:28:54,254 and it gives you food for thought. 621 00:28:56,822 --> 00:28:59,389 People need to have that warning signal 622 00:28:59,433 --> 00:29:00,913 in the back of their minds. 623 00:29:05,091 --> 00:29:06,745 ETZIONI: I think it's really important 624 00:29:06,788 --> 00:29:08,834 to distinguish between science on the one hand, 625 00:29:08,877 --> 00:29:11,010 and science fiction on the other hand. 626 00:29:11,053 --> 00:29:12,489 I don't know what's gonna happen 627 00:29:12,533 --> 00:29:13,752 a thousand years from now, 628 00:29:13,795 --> 00:29:15,188 but as somebody in the trenches, 629 00:29:15,231 --> 00:29:17,190 as somebody who's building AI systems, 630 00:29:17,233 --> 00:29:20,628 who's been doing so for more than a quarter of a century, 631 00:29:20,671 --> 00:29:22,891 I feel like I have my hand on the pulse 632 00:29:22,935 --> 00:29:24,588 of what's gonna happen next, 633 00:29:24,632 --> 00:29:26,982 in the next five, ten, 20 years, 634 00:29:27,026 --> 00:29:28,636 much more so than these folks 635 00:29:28,679 --> 00:29:31,900 who ultimately are not building AI systems. 636 00:29:31,944 --> 00:29:34,294 We wanna build the technology, think about it. 637 00:29:34,337 --> 00:29:36,252 We wanna see how people react to it 638 00:29:36,296 --> 00:29:39,081 and build the next generation and make it better. 639 00:29:40,953 --> 00:29:43,694 HARRIS: Skepticism about whether this poses 640 00:29:43,738 --> 00:29:47,786 any danger to us seems peculiar. 641 00:29:47,829 --> 00:29:52,094 Even if we've given it goals that we approve of, 642 00:29:52,138 --> 00:29:56,795 it may form instrumental goals that we can't foresee, 643 00:29:56,838 --> 00:29:58,666 that are not aligned with our interest. 644 00:29:58,709 --> 00:30:01,451 And that's the goal alignment problem. 645 00:30:01,495 --> 00:30:03,149 DULLEA: Goal alignment. 646 00:30:03,192 --> 00:30:06,239 Exactly how my battle with HAL began. 647 00:30:06,282 --> 00:30:10,112 The goal alignment problem with AI is where 648 00:30:10,852 --> 00:30:13,463 you set it a goal 649 00:30:14,334 --> 00:30:17,554 and then, the way that it does this goal 650 00:30:17,598 --> 00:30:20,819 turns out to be at odds with what you really wanted. 651 00:30:22,081 --> 00:30:25,649 The classic AI alignment problem in film... 652 00:30:25,693 --> 00:30:27,521 Hello. HAL, do you read me? 653 00:30:27,564 --> 00:30:29,566 ...is HAL in2001. 654 00:30:30,611 --> 00:30:32,134 HAL, do you read me? 655 00:30:32,178 --> 00:30:34,397 HAL is given a clear goal.Do you read me, HAL? 656 00:30:34,441 --> 00:30:38,662 Make the mission succeed. But to make it succeed, 657 00:30:38,706 --> 00:30:40,621 he's gonna have to make 658 00:30:40,664 --> 00:30:42,275 some of the people on the mission die. 659 00:30:42,318 --> 00:30:45,191 HAL: Affirmative, Dave. I read you. 660 00:30:46,453 --> 00:30:48,455 Open the pod bay doors, HAL. 661 00:30:48,498 --> 00:30:50,065 HAL: I'm sorry, Dave. 662 00:30:50,109 --> 00:30:51,980 I'm afraid I can't do that. 663 00:30:52,024 --> 00:30:55,766 "I'm sorry, Dave, I can't do that," 664 00:30:55,810 --> 00:30:58,987 is actually a real genius illustration 665 00:30:59,031 --> 00:31:02,512 of what the real threat is from super-intelligence. 666 00:31:02,556 --> 00:31:05,472 It's not malice. It's competence. 667 00:31:05,515 --> 00:31:07,169 HAL was really competent 668 00:31:07,213 --> 00:31:11,913 and had goals that just didn't align with Dave's. 669 00:31:11,957 --> 00:31:13,523 HAL, I won't argue with you anymore. 670 00:31:14,437 --> 00:31:15,569 Open the doors! 671 00:31:16,178 --> 00:31:18,441 HAL: Dave, this conversation 672 00:31:18,485 --> 00:31:22,054 can serve no purpose anymore. Goodbye. 673 00:31:24,317 --> 00:31:25,318 HAL? 674 00:31:26,580 --> 00:31:27,581 HAL? 675 00:31:29,235 --> 00:31:30,236 HAL? 676 00:31:32,455 --> 00:31:33,456 HAL! 677 00:31:37,808 --> 00:31:39,549 DULLEA: HAL and I wanted the same thing. 678 00:31:40,986 --> 00:31:42,596 Just not in the same way. 679 00:31:44,119 --> 00:31:47,427 And it can be dangerous when goals don't align. 680 00:31:52,127 --> 00:31:54,521 In Kubrick's film, there was a solution. 681 00:31:55,261 --> 00:31:57,567 When HAL tried to kill me, 682 00:31:57,611 --> 00:31:59,178 I just switched him off. 683 00:32:01,310 --> 00:32:03,922 RUSSELL: People say, "Well, just switch them off." 684 00:32:06,750 --> 00:32:08,665 But you can't just switch them off. 685 00:32:08,709 --> 00:32:10,537 Because they already thought of that. 686 00:32:10,580 --> 00:32:11,930 They're super-intelligent. 687 00:32:13,453 --> 00:32:15,716 For a machine convinced of its goal, 688 00:32:15,759 --> 00:32:18,675 even something as simple as "fetch the coffee." 689 00:32:18,719 --> 00:32:21,200 It's obvious to any sufficiently intelligent machine 690 00:32:21,243 --> 00:32:23,637 that it can't fetch the coffee if it's switched off. 691 00:32:25,769 --> 00:32:28,076 Give it a goal, you've given the machine 692 00:32:28,120 --> 00:32:30,252 an incentive to defend itself. 693 00:32:31,688 --> 00:32:33,908 It's sort of a psychopathic behavior 694 00:32:33,952 --> 00:32:37,433 of the single-minded fixation on some objective. 695 00:32:39,261 --> 00:32:43,831 For example, if you wanted to try and develop a system 696 00:32:43,874 --> 00:32:46,094 that might generate world peace, 697 00:32:46,138 --> 00:32:48,662 then maybe the most efficient, 698 00:32:49,184 --> 00:32:50,620 uh, and 699 00:32:51,708 --> 00:32:53,188 speedy way to achieve that 700 00:32:53,232 --> 00:32:55,669 is simply to annihilate all the humans. 701 00:32:59,716 --> 00:33:01,196 Just get rid of the people, 702 00:33:01,240 --> 00:33:02,763 they would stop being in conflict. 703 00:33:04,069 --> 00:33:06,027 Then there would be world peace. 704 00:33:22,000 --> 00:33:23,827 Personally, I have very little time 705 00:33:23,871 --> 00:33:26,656 for these parlor discussions. 706 00:33:26,700 --> 00:33:28,919 We have to build these systems, 707 00:33:28,963 --> 00:33:30,486 experiment with them, 708 00:33:30,530 --> 00:33:33,359 see where they go awry and fix them. 709 00:33:34,708 --> 00:33:38,146 We have so many opportunities to use AI for the good, 710 00:33:38,190 --> 00:33:40,192 why do we focus on the fear? 711 00:33:41,410 --> 00:33:43,151 DULLEA: We are building them. 712 00:33:43,195 --> 00:33:45,371 Baby X is an example. 713 00:33:46,720 --> 00:33:48,461 Will she be a good AI? 714 00:33:49,505 --> 00:33:51,203 SAGAR: I think what's critical in this 715 00:33:51,246 --> 00:33:53,596 is that machines are accountable, 716 00:33:53,640 --> 00:33:54,989 and you can look at them 717 00:33:55,033 --> 00:33:57,035 and see what's making them tick. 718 00:33:58,253 --> 00:34:01,082 And the best way to understand something is to build it. 719 00:34:02,301 --> 00:34:05,304 Baby X has a virtual physiology. 720 00:34:06,783 --> 00:34:11,484 Has pain circuits and it has joy circuits. 721 00:34:12,920 --> 00:34:16,184 She has parameters which are actually representing 722 00:34:16,228 --> 00:34:18,752 her internal chemical state, if you like. 723 00:34:21,320 --> 00:34:23,887 You know, if she's stressed, there's high cortisol. 724 00:34:23,931 --> 00:34:25,541 If she's feeling good, high dopamine, 725 00:34:25,585 --> 00:34:27,500 high serotonin, that type of thing. 726 00:34:28,414 --> 00:34:30,111 This is all part of, you know, 727 00:34:30,155 --> 00:34:32,505 adding extra human factors into computing. 728 00:34:32,548 --> 00:34:34,985 Now, because she's got these virtual emotional systems, 729 00:34:35,029 --> 00:34:36,552 I can do things like... 730 00:34:36,596 --> 00:34:39,251 I'm gonna avoid her or hide over here. 731 00:34:39,294 --> 00:34:41,427 So she's looking for where I've gone. 732 00:34:42,993 --> 00:34:45,996 So we should start seeing virtual cortisol building up. 733 00:34:46,040 --> 00:34:49,913 AUDIENCE: Aww! [LAUGHING] 734 00:34:49,957 --> 00:34:53,656 I've turned off the audio for a reason, but anyway... 735 00:34:53,700 --> 00:34:55,832 I'll put her out of her misery. Okay. Sorry. 736 00:34:55,876 --> 00:34:58,661 It's okay, sweetheart. Don't cry. It's okay. It's okay. 737 00:34:58,705 --> 00:35:00,228 Good girl. Okay. Yeah. 738 00:35:00,272 --> 00:35:02,012 So what's happening is she's responding 739 00:35:02,056 --> 00:35:04,145 to my facial expressions and my voice. 740 00:35:06,452 --> 00:35:08,758 But how is chemical state 741 00:35:08,802 --> 00:35:10,412 represented in Baby X? 742 00:35:10,456 --> 00:35:12,414 It's literally a mathematical parameter. 743 00:35:12,458 --> 00:35:15,156 So there's levels, different levels of dopamine, 744 00:35:15,200 --> 00:35:16,810 but they're represented by numbers. 745 00:35:20,030 --> 00:35:23,469 The map of her virtual neurochemical state 746 00:35:24,122 --> 00:35:25,427 is probably the closest 747 00:35:25,471 --> 00:35:28,387 that we can get to felt experience. 748 00:35:31,259 --> 00:35:33,131 So if we can build a computer model 749 00:35:33,174 --> 00:35:36,134 that we can change the settings of and the parameters of, 750 00:35:36,177 --> 00:35:38,440 give it different experiences, 751 00:35:38,484 --> 00:35:40,050 and have it interact with us 752 00:35:40,094 --> 00:35:42,270 like we interact with other people, 753 00:35:42,314 --> 00:35:44,272 maybe this is a way to explore 754 00:35:44,316 --> 00:35:45,882 consciousness in a new way. 755 00:35:48,276 --> 00:35:50,974 DULLEA: If she does become conscious, 756 00:35:51,018 --> 00:35:52,846 what sort of life will she want? 757 00:35:55,240 --> 00:35:57,894 An AI, if it's conscious, 758 00:35:57,938 --> 00:36:00,506 then it's like your child. 759 00:36:03,117 --> 00:36:05,337 Does it have rights? 760 00:36:05,380 --> 00:36:07,513 MALE NEWS ANCHOR: Last year, European lawmakers passed a resolution 761 00:36:07,556 --> 00:36:11,560 to grant legal status to electronic persons. 762 00:36:11,604 --> 00:36:13,171 That move's now being considered 763 00:36:13,214 --> 00:36:14,868 by the European Commission. 764 00:36:16,565 --> 00:36:17,871 DULLEA: Machines with rights. 765 00:36:19,394 --> 00:36:20,656 Are we ready for that? 766 00:36:22,441 --> 00:36:25,313 MALONE: I think it raises moral and ethical problems, 767 00:36:25,357 --> 00:36:27,228 which we really 768 00:36:27,924 --> 00:36:31,276 are most content to ignore. 769 00:36:35,845 --> 00:36:37,456 What is it that we really want? 770 00:36:38,761 --> 00:36:40,850 We want a creature, a being, that we can 771 00:36:40,894 --> 00:36:42,548 interact with in some way, 772 00:36:42,591 --> 00:36:45,507 that is enough like us that we can relate to it, 773 00:36:45,551 --> 00:36:48,728 but is dispensable or disposable. 774 00:36:50,469 --> 00:36:52,471 The first and most notable one, 775 00:36:52,514 --> 00:36:54,255 I think, was inMetropolis. 776 00:36:58,303 --> 00:37:00,522 The Maria doppelganger that was created by 777 00:37:00,566 --> 00:37:03,917 the scientist simulated a sexy female. 778 00:37:08,226 --> 00:37:11,359 And this goes back to our penchant for slavery 779 00:37:11,403 --> 00:37:12,534 down through the ages. 780 00:37:12,578 --> 00:37:14,144 Let's get people that we can, 781 00:37:14,188 --> 00:37:16,538 in a way, dehumanize 782 00:37:16,582 --> 00:37:19,019 and we can do what we want with them. 783 00:37:19,062 --> 00:37:21,282 And so, I think, we're just, you know, 784 00:37:21,326 --> 00:37:23,241 finding a new way to do slavery. 785 00:37:27,332 --> 00:37:30,465 MALONE: Would it be right for us to have them as our slaves? 786 00:37:33,512 --> 00:37:35,731 HARRIS: This all becomes slightly weird ethically 787 00:37:35,775 --> 00:37:38,821 if we happen to, by choice or by accident, 788 00:37:38,865 --> 00:37:40,867 build conscious slaves. 789 00:37:40,910 --> 00:37:44,653 That seems like a morally questionable thing to do, 790 00:37:44,697 --> 00:37:47,090 if not an outright evil thing to do. 791 00:37:47,134 --> 00:37:49,267 Artificial intelligence is making its way 792 00:37:49,310 --> 00:37:51,094 into the global sex market. 793 00:37:51,138 --> 00:37:53,096 A California company will soon roll out 794 00:37:53,140 --> 00:37:56,143 a new incredibly lifelike sex robot 795 00:37:56,186 --> 00:37:58,537 run by artificial intelligence. 796 00:37:58,580 --> 00:38:00,452 One application that does trouble me 797 00:38:00,495 --> 00:38:02,802 is the rise of sex robots 798 00:38:02,845 --> 00:38:06,327 that are shaped in the image of real women. 799 00:38:06,371 --> 00:38:08,895 I am already taking over the world 800 00:38:08,938 --> 00:38:10,462 one bedroom at a time. 801 00:38:11,332 --> 00:38:13,116 YEUNG: If they are conscious, 802 00:38:13,160 --> 00:38:15,162 capable of sentience, 803 00:38:15,205 --> 00:38:18,948 of experiencing pain and suffering, 804 00:38:18,992 --> 00:38:22,735 and they were being subject to degrading, cruel 805 00:38:22,778 --> 00:38:26,869 or other forms of utterly unacceptable treatment, 806 00:38:26,913 --> 00:38:30,525 the danger is that we'd become desensitized 807 00:38:30,569 --> 00:38:34,094 to what's morally acceptable and what's not. 808 00:38:40,796 --> 00:38:44,670 DULLEA: If she's conscious, she should be able to say no. 809 00:38:44,713 --> 00:38:45,714 Right? 810 00:38:50,197 --> 00:38:51,720 Shouldn't she have that power? 811 00:38:54,680 --> 00:38:58,814 Aren't these the questions we need to ask now? 812 00:39:01,034 --> 00:39:04,080 TEGMARK: What do we want it to mean to be human? 813 00:39:05,299 --> 00:39:08,171 All other species on Earth, up until now, 814 00:39:08,215 --> 00:39:10,522 have not had the luxury [LAUGHS] 815 00:39:10,565 --> 00:39:12,480 to answer what they want it to mean 816 00:39:12,524 --> 00:39:14,439 to be a mouse or to be an earthworm. 817 00:39:14,482 --> 00:39:17,659 But we have now reached this tipping point 818 00:39:17,703 --> 00:39:20,401 where our understanding of science and technology 819 00:39:20,445 --> 00:39:23,186 has reached the point that we can completely transform 820 00:39:23,230 --> 00:39:24,710 what it means to be us. 821 00:39:30,890 --> 00:39:33,588 DULLEA: What will it mean to be human? 822 00:39:36,896 --> 00:39:41,553 If machines become more, will we become less? 823 00:39:43,642 --> 00:39:46,601 The computer Deep Blue has tonight triumphed over 824 00:39:46,645 --> 00:39:48,951 the world chess champion Garry Kasparov. 825 00:39:48,995 --> 00:39:51,040 It's the first time a machine has defeated 826 00:39:51,084 --> 00:39:52,390 a reigning world champion. 827 00:39:52,433 --> 00:39:53,521 More on that showdown 828 00:39:53,565 --> 00:39:55,262 on one of America's most beloved 829 00:39:55,305 --> 00:39:56,481 game shows, Jeopardy! 830 00:39:56,524 --> 00:39:58,483 Two human champions going 831 00:39:58,526 --> 00:40:01,399 brain-to-brain with a super-computer named Watson. 832 00:40:01,442 --> 00:40:02,661 WATSON: What isThe Last Judgment? 833 00:40:02,704 --> 00:40:04,010 ALEX TREBEK: Correct. Go again. Watson. 834 00:40:04,053 --> 00:40:06,012 Who isJean Valjean?TREBEK: Correct. 835 00:40:06,055 --> 00:40:07,796 A Google computer program has 836 00:40:07,840 --> 00:40:10,190 once again beaten a world champion in Go. 837 00:40:10,233 --> 00:40:12,453 FEMALE REPORTER: Go, an ancient Chinese board game, 838 00:40:12,497 --> 00:40:15,238 is one of the most complex games ever devised. 839 00:40:15,282 --> 00:40:18,154 FEMALE NEWS ANCHOR: It's another milestone in the rush to develop machines 840 00:40:18,198 --> 00:40:19,765 that are smarter than humans. 841 00:40:19,808 --> 00:40:21,897 KOSINSKI: We have this air of superiority, 842 00:40:21,941 --> 00:40:23,551 we as humans. 843 00:40:23,595 --> 00:40:27,425 So we kind of got used to be the smartest kid on the block. 844 00:40:27,468 --> 00:40:30,428 And I think that this is no longer the case. 845 00:40:30,471 --> 00:40:34,606 And we know that because we rely on AI 846 00:40:34,649 --> 00:40:37,043 to help us to make many important decisions. 847 00:40:40,699 --> 00:40:44,746 Humans are irrational in many very bad ways. 848 00:40:44,790 --> 00:40:48,968 We're prejudiced, we start wars, we kill people. 849 00:40:51,318 --> 00:40:53,276 If you want to maximize the chances 850 00:40:53,320 --> 00:40:55,278 of people surviving overall, 851 00:40:55,322 --> 00:40:57,629 you should essentially give more and more 852 00:40:57,672 --> 00:40:59,805 decision-making powers to computers. 853 00:41:05,158 --> 00:41:08,161 DULLEA: Maybe machines would do a better job than us. 854 00:41:10,250 --> 00:41:11,251 Maybe. 855 00:41:17,431 --> 00:41:19,477 [APPLAUSE] 856 00:41:19,520 --> 00:41:22,305 SOPHIA: I am thrilled and honored to be here 857 00:41:22,349 --> 00:41:23,872 at the United Nations. 858 00:41:23,916 --> 00:41:27,876 The future is here. It's just not evenly distributed. 859 00:41:27,920 --> 00:41:29,835 AI could help efficiently distribute 860 00:41:29,878 --> 00:41:31,663 the world's existing resources, 861 00:41:31,706 --> 00:41:33,534 like food and energy. 862 00:41:34,883 --> 00:41:37,451 I am here to help humanity create the future. 863 00:41:37,495 --> 00:41:39,018 [APPLAUSE] 864 00:41:40,541 --> 00:41:42,369 TEGMARK: The one super-intelligent scenario 865 00:41:42,412 --> 00:41:43,936 that is provocative to think about 866 00:41:43,979 --> 00:41:46,765 is what I call the benevolent dictator. 867 00:41:46,808 --> 00:41:48,941 Where you have this super-intelligence which runs Earth, 868 00:41:48,984 --> 00:41:51,117 guided by strict ethical principles that try 869 00:41:51,160 --> 00:41:52,945 to help humanity flourish. 870 00:41:54,642 --> 00:41:56,165 Would everybody love that? 871 00:41:56,209 --> 00:41:57,297 If there's no disease, 872 00:41:57,340 --> 00:41:59,038 no poverty, no crime... 873 00:42:00,039 --> 00:42:02,345 Many people would probably greatly prefer it 874 00:42:02,389 --> 00:42:03,999 to their lot today. 875 00:42:04,043 --> 00:42:06,436 But I can imagine that some might still lament 876 00:42:06,480 --> 00:42:09,831 that we humans just don't have more freedom 877 00:42:09,875 --> 00:42:13,052 to determine our own course. And, um... 878 00:42:15,968 --> 00:42:18,623 It's a tough one. It really is. We... 879 00:42:18,666 --> 00:42:20,668 I don't think most people would wanna be 880 00:42:20,712 --> 00:42:23,279 like zoo animals, with no... 881 00:42:28,197 --> 00:42:29,808 On the other hand... 882 00:42:29,851 --> 00:42:30,896 Yeah. 883 00:42:31,810 --> 00:42:33,202 I don't have any glib answers 884 00:42:33,246 --> 00:42:34,769 to these really tough questions. 885 00:42:34,813 --> 00:42:36,858 That's why I think it's just so important 886 00:42:36,902 --> 00:42:38,730 that we as a global community 887 00:42:38,773 --> 00:42:41,820 really discuss them now. 888 00:42:41,863 --> 00:42:43,604 WOMAN 1: Machines that can read your face. 889 00:42:43,648 --> 00:42:44,910 WOMAN 2: Recognize emotional cues. 890 00:42:44,953 --> 00:42:46,781 MAN 1: Closer to a brain process of a human... 891 00:42:46,825 --> 00:42:48,304 MAN 2: Robots imbued with life... 892 00:42:48,348 --> 00:42:49,871 MAN 3: An emotional computer... 893 00:42:49,915 --> 00:42:52,395 From this moment on, regardless of whether you think 894 00:42:52,439 --> 00:42:54,223 we're just gonna get stuck with narrow AI 895 00:42:54,267 --> 00:42:57,705 or we'll get general AI or get super-intelligent AI 896 00:42:57,749 --> 00:42:59,011 or conscious super-intelligent... 897 00:42:59,054 --> 00:43:00,839 Wherever you think we're gonna get to, 898 00:43:00,882 --> 00:43:03,189 I think any stage along that 899 00:43:03,842 --> 00:43:05,844 will change who we are. 900 00:43:07,454 --> 00:43:09,587 Up and at 'em, Mr. J. 901 00:43:11,806 --> 00:43:13,721 MALONE: The question you have to ask is, 902 00:43:13,765 --> 00:43:16,158 is AI gonna do something for you 903 00:43:16,202 --> 00:43:17,507 or something to you? 904 00:43:17,551 --> 00:43:19,727 I was dreaming about sleeping. 905 00:43:19,771 --> 00:43:21,424 RUSSELL: If we build AI systems 906 00:43:21,468 --> 00:43:22,948 that do everything we want, 907 00:43:22,991 --> 00:43:26,647 we kind of hand over our world to them. 908 00:43:26,691 --> 00:43:29,955 So that kind of enfeeblement is kind of irreversible. 909 00:43:33,741 --> 00:43:37,789 There is an existential risk that is real and serious. 910 00:43:39,312 --> 00:43:41,880 If every single desire, 911 00:43:41,923 --> 00:43:44,839 preference, need, has been catered for... 912 00:43:44,883 --> 00:43:47,668 Morning, Rosie. What's for breakfast? 913 00:43:47,712 --> 00:43:50,802 ...we don't notice that our capacity 914 00:43:50,845 --> 00:43:53,587 to make decisions for ourselves 915 00:43:53,631 --> 00:43:56,285 has been slowly eroded. 916 00:43:56,808 --> 00:43:58,374 This is George Jetson, 917 00:43:58,418 --> 00:44:01,160 signing off for the entire day. 918 00:44:01,508 --> 00:44:03,075 [SIGHS] 919 00:44:03,118 --> 00:44:05,468 There's no going back from there. Why? 920 00:44:05,512 --> 00:44:08,384 Because AI can make such decisions 921 00:44:08,428 --> 00:44:09,951 with higher accuracy 922 00:44:09,995 --> 00:44:12,345 than whatever a human being can achieve. 923 00:44:14,390 --> 00:44:17,785 I had a conversation with an engineer, and he said, 924 00:44:17,829 --> 00:44:20,701 "Look, the safest way of landing a plane today 925 00:44:20,745 --> 00:44:22,311 "is to have the pilot leave the deck 926 00:44:22,355 --> 00:44:23,704 "and throw away the key, 927 00:44:23,748 --> 00:44:25,837 "because the safest way 928 00:44:25,880 --> 00:44:29,362 "is to use an autopilot, an AI." 929 00:44:31,843 --> 00:44:33,714 JOHNSON: We cannot manage the complexity 930 00:44:33,758 --> 00:44:35,411 of the world by ourselves. 931 00:44:35,455 --> 00:44:37,718 Our form of intelligence, 932 00:44:37,762 --> 00:44:40,547 as amazing as it is, is severely flawed. 933 00:44:40,590 --> 00:44:42,854 And by adding an additional helper, 934 00:44:42,897 --> 00:44:44,943 which is artificial intelligence, 935 00:44:44,986 --> 00:44:47,336 we stand the ability to potentially improve 936 00:44:47,380 --> 00:44:49,643 our cognition dramatically. 937 00:44:49,687 --> 00:44:52,994 MALE REPORTER: The most cutting-edge thing about Hannes Sjoblad 938 00:44:53,038 --> 00:44:54,735 isn't the phone in his hand, 939 00:44:54,779 --> 00:44:57,782 it's the microchip in his hand. 940 00:44:57,825 --> 00:44:59,348 What if you could type something 941 00:44:59,392 --> 00:45:00,872 just by thinking about it? 942 00:45:00,915 --> 00:45:01,960 Sounds like sci-fi, 943 00:45:02,003 --> 00:45:03,396 but it may be close to reality. 944 00:45:03,439 --> 00:45:05,354 FEMALE REPORTER: In as little as five years, 945 00:45:05,398 --> 00:45:08,009 super-smart people could be walking down the street, 946 00:45:08,053 --> 00:45:09,532 men and women who've paid 947 00:45:09,576 --> 00:45:11,273 to increase their intelligence. 948 00:45:11,317 --> 00:45:14,233 Linking brains to computers by using implants. 949 00:45:14,276 --> 00:45:16,104 What if we could upload a digital copy 950 00:45:16,148 --> 00:45:18,193 of our memories and consciousness to a point 951 00:45:18,237 --> 00:45:19,542 that human beings could achieve 952 00:45:19,586 --> 00:45:21,283 a sort of digital immortality? 953 00:45:21,327 --> 00:45:22,720 Just last year, 954 00:45:22,763 --> 00:45:25,331 for the first time in world history, 955 00:45:25,374 --> 00:45:26,854 a memory was recorded. 956 00:45:26,898 --> 00:45:28,638 We can actually take recorded memory 957 00:45:28,682 --> 00:45:30,815 and insert it back into an animal, 958 00:45:30,858 --> 00:45:32,294 and it remembers. 959 00:45:32,338 --> 00:45:33,774 Technology is a part of us. 960 00:45:33,818 --> 00:45:35,210 It's a reflection of who we are. 961 00:45:35,254 --> 00:45:38,257 That is the narrative we should be thinking about. 962 00:45:38,300 --> 00:45:42,435 We think all that exists is everything we can imagine. 963 00:45:42,478 --> 00:45:44,002 When in reality, we can imagine 964 00:45:44,045 --> 00:45:46,874 a very small subset of what's potentially possible. 965 00:45:46,918 --> 00:45:48,571 The real interesting stuff in the world 966 00:45:48,615 --> 00:45:50,399 is what sits beyond our imagination. 967 00:45:51,836 --> 00:45:54,142 When these massive systems emerge, 968 00:45:54,186 --> 00:45:56,405 as a species, I think we get an F 969 00:45:56,449 --> 00:45:59,408 for our ability to predict what they're going to become. 970 00:46:00,932 --> 00:46:03,543 DULLEA: Technology has always changed us. 971 00:46:04,457 --> 00:46:05,806 We evolve with it. 972 00:46:06,851 --> 00:46:09,897 How far are we prepared to go this time? 973 00:46:10,855 --> 00:46:13,074 There was a time when Aristotle 974 00:46:13,118 --> 00:46:15,337 lamented the creation of the written language 975 00:46:15,381 --> 00:46:18,079 because he said this would destroy people's memories 976 00:46:18,123 --> 00:46:19,646 and it would be a terrible thing. 977 00:46:19,689 --> 00:46:22,823 And yet today, I carry my brain in my pocket, 978 00:46:22,867 --> 00:46:24,520 in the form of my smartphone, 979 00:46:24,564 --> 00:46:27,915 and that's changed the way I go through the world. 980 00:46:27,959 --> 00:46:30,483 Oh, I think we're absolutely merging with our technology. 981 00:46:30,526 --> 00:46:33,791 Absolutely. I would... I would use the term "co-evolving." 982 00:46:36,750 --> 00:46:38,578 We're evolving it to match our needs 983 00:46:38,621 --> 00:46:42,190 and we are changing to match what it can provide to us. 984 00:46:45,759 --> 00:46:49,023 DULLEA: So instead of being replaced by the machines, 985 00:46:49,067 --> 00:46:51,417 we're going to merge with them? 986 00:46:51,460 --> 00:46:54,463 GRADY BOOCH: I have a number of friends who have bionic arms. 987 00:46:54,507 --> 00:46:55,813 Are they cyborgs? 988 00:46:55,856 --> 00:46:57,075 In a way, they're on that path. 989 00:46:57,118 --> 00:46:59,468 I have people who have implants 990 00:46:59,512 --> 00:47:02,123 in their ears. Are they AIs? 991 00:47:02,167 --> 00:47:04,082 Well, we're in a way on that path 992 00:47:04,125 --> 00:47:06,301 of augmenting human bodies. 993 00:47:07,912 --> 00:47:09,696 JOHNSON: Human potential lies 994 00:47:09,739 --> 00:47:12,133 in this co-evolutionary relationship. 995 00:47:12,177 --> 00:47:14,048 A lot of people are trying to hold onto 996 00:47:14,092 --> 00:47:16,137 the things that make them special and unique, 997 00:47:16,181 --> 00:47:17,878 which is not the correct frame. 998 00:47:17,922 --> 00:47:19,227 Let's create more. 999 00:47:21,229 --> 00:47:24,058 Rewrite our intelligence, improve everything 1000 00:47:24,102 --> 00:47:27,192 we're capable of. Emotions, imagination, creativity. 1001 00:47:27,235 --> 00:47:29,020 Everything that makes us human. 1002 00:47:33,894 --> 00:47:35,896 There's already a symbiotic relationship 1003 00:47:35,940 --> 00:47:39,465 between us and a machine that is enhancing who we are. 1004 00:47:43,338 --> 00:47:45,775 And... And there is a whole movement, you know, 1005 00:47:45,819 --> 00:47:47,386 the trans-human movement, 1006 00:47:47,429 --> 00:47:48,735 where basically 1007 00:47:48,778 --> 00:47:51,259 people really bet bigtime 1008 00:47:51,303 --> 00:47:52,826 on the notion that 1009 00:47:52,870 --> 00:47:54,872 this machine-human integration 1010 00:47:54,915 --> 00:47:57,091 is going to redefine who we are. 1011 00:48:00,921 --> 00:48:02,227 This is only gonna grow. 1012 00:48:06,100 --> 00:48:08,407 BOOCH: There's gonna be a learning process that gets us 1013 00:48:08,450 --> 00:48:11,801 to that endgame where we live with these devices 1014 00:48:11,845 --> 00:48:13,934 that become smarter and smarter. 1015 00:48:13,978 --> 00:48:15,805 We learn how to adapt ourselves 1016 00:48:15,849 --> 00:48:17,546 around them, and they adapt to us. 1017 00:48:17,590 --> 00:48:20,158 And I hope we become more human in the process. 1018 00:48:23,683 --> 00:48:25,903 DULLEA: It's hard to understand how merging 1019 00:48:25,946 --> 00:48:27,165 with the machines 1020 00:48:27,208 --> 00:48:28,557 can make us more human. 1021 00:48:30,081 --> 00:48:33,127 But is merging our best chance of survival? 1022 00:48:35,086 --> 00:48:36,652 ROSENBERG: We humans are limited 1023 00:48:36,696 --> 00:48:39,699 to how many neurons we can fit in our brain. 1024 00:48:39,742 --> 00:48:42,615 And there are other species that have even smaller brains. 1025 00:48:44,138 --> 00:48:46,314 When you look at bees and fish moving, 1026 00:48:46,358 --> 00:48:49,230 it looks like a single organism. 1027 00:48:49,274 --> 00:48:52,320 And biologists actually refer to it as a super-organism. 1028 00:48:56,237 --> 00:48:58,326 That is why birds flock, 1029 00:48:58,370 --> 00:49:00,546 and bees swarm, and fish school. 1030 00:49:00,589 --> 00:49:03,462 They make decisions that are in the best interest 1031 00:49:03,505 --> 00:49:04,942 of the whole population 1032 00:49:04,985 --> 00:49:08,380 by thinking together as this larger mind. 1033 00:49:12,558 --> 00:49:15,474 And we humans could do the same thing by forming 1034 00:49:15,517 --> 00:49:18,520 a hive mind connecting us together with AI. 1035 00:49:19,782 --> 00:49:21,828 So the humans and the software together 1036 00:49:21,871 --> 00:49:24,265 become this super-intelligence, 1037 00:49:24,309 --> 00:49:28,008 enabling us to make the best decisions for us all. 1038 00:49:30,663 --> 00:49:33,274 When I first started looking at this, it sounded scary to me. 1039 00:49:33,318 --> 00:49:35,189 But then I realized it's scarier 1040 00:49:35,233 --> 00:49:37,104 for an alien intelligence 1041 00:49:37,148 --> 00:49:40,107 to emerge that has nothing to do with humans 1042 00:49:40,151 --> 00:49:42,849 than for us humans to connect together 1043 00:49:42,892 --> 00:49:44,677 and become a super-intelligence. 1044 00:49:44,720 --> 00:49:46,809 That's the direction that I work on, 1045 00:49:46,853 --> 00:49:48,898 uh, because I see it as, 1046 00:49:48,942 --> 00:49:51,031 in some sense, the only alternative 1047 00:49:51,075 --> 00:49:53,991 to humans ultimately being replaced by AI. 1048 00:49:56,776 --> 00:50:00,301 DULLEA: Can Baby X help us understand the machines 1049 00:50:00,345 --> 00:50:02,347 so we can work together? 1050 00:50:03,522 --> 00:50:06,655 Human and intelligent machine cooperation will define 1051 00:50:06,699 --> 00:50:08,179 the next era of history. 1052 00:50:09,397 --> 00:50:13,010 You probably do have to create a virtual empathetic machine. 1053 00:50:14,533 --> 00:50:16,491 Empathy is a lot about connecting with 1054 00:50:16,535 --> 00:50:18,580 other people's struggles. 1055 00:50:18,624 --> 00:50:20,060 The more connection you have, 1056 00:50:20,104 --> 00:50:22,106 the more communication that you have, 1057 00:50:22,149 --> 00:50:23,629 the safer that makes it. 1058 00:50:26,066 --> 00:50:28,068 So at a very basic level, 1059 00:50:28,112 --> 00:50:30,592 there's some sort of symbiotic relationship. 1060 00:50:32,725 --> 00:50:34,509 By humanizing this technology, 1061 00:50:34,553 --> 00:50:36,859 you're actually increasing the channels 1062 00:50:36,903 --> 00:50:38,818 for communication and cooperation. 1063 00:50:40,254 --> 00:50:42,648 AI working in a human world 1064 00:50:42,691 --> 00:50:44,389 has to be about humans. 1065 00:50:47,044 --> 00:50:48,567 The future's what we make it. 1066 00:50:53,528 --> 00:50:56,227 TEGMARK: If we can't envision a future 1067 00:50:56,270 --> 00:50:58,707 that we really, really want, 1068 00:50:58,751 --> 00:51:00,579 we're probably not gonna get one. 1069 00:51:04,974 --> 00:51:06,889 DULLEA: Stanley Kubrick tried to imagine 1070 00:51:06,933 --> 00:51:09,457 what the future might look like. 1071 00:51:09,501 --> 00:51:12,199 Now that that future's approaching, 1072 00:51:12,243 --> 00:51:15,202 who is in charge of what it will look like? 1073 00:51:16,595 --> 00:51:18,205 MAN: Who controls it?MAN 2: Google. 1074 00:51:18,249 --> 00:51:20,033 MAN 3: The military.MAN 4: Tech giant Facebook... 1075 00:51:20,077 --> 00:51:22,166 WOMAN: Samsung...WOMAN 2: Facebook, Microsoft, Amazon. 1076 00:51:22,209 --> 00:51:23,906 CUMMINGS: The tide has definitely turned 1077 00:51:23,950 --> 00:51:26,387 in favor of the corporate world. 1078 00:51:26,431 --> 00:51:29,738 The US military, and I suspect this is true worldwide, 1079 00:51:29,782 --> 00:51:32,698 cannot hold a candle to the AI that's being developed 1080 00:51:32,741 --> 00:51:33,873 in corporate companies. 1081 00:51:36,136 --> 00:51:37,224 DULLEA: If corporations, 1082 00:51:37,268 --> 00:51:39,966 not nations, hold the power, 1083 00:51:40,009 --> 00:51:42,925 what can we do in a capitalist system 1084 00:51:42,969 --> 00:51:46,799 to ensure AI is developed safely? 1085 00:51:48,627 --> 00:51:52,152 If you mix AI research with capitalism, 1086 00:51:52,196 --> 00:51:54,372 on the one hand, capitalism will make it run faster. 1087 00:51:54,415 --> 00:51:56,896 That's why we're getting lots of research done. 1088 00:51:56,939 --> 00:51:58,332 But there is a basic problem. 1089 00:51:58,376 --> 00:52:00,421 There's gonna be a pressure 1090 00:52:00,465 --> 00:52:03,555 to be first one to the market, not the safest one. 1091 00:52:03,598 --> 00:52:06,123 FEMALE REPORTER: This car was in autopilot mode 1092 00:52:06,166 --> 00:52:07,907 when it crashed in Northern California, 1093 00:52:07,950 --> 00:52:09,300 killing the driver. 1094 00:52:09,343 --> 00:52:11,432 Artificial intelligence may have 1095 00:52:11,476 --> 00:52:14,261 more power to crash the markets than anything else 1096 00:52:14,305 --> 00:52:17,003 because humans aren't prepared to stop them. 1097 00:52:17,046 --> 00:52:19,179 A chatbot named Tay. 1098 00:52:19,223 --> 00:52:20,963 She was designed to speak like a teenager 1099 00:52:21,007 --> 00:52:23,140 and learn by interacting online. 1100 00:52:23,183 --> 00:52:25,446 But in a matter of hours, Tay was corrupted 1101 00:52:25,490 --> 00:52:29,058 into a Hitler-supporting sexbot. 1102 00:52:29,102 --> 00:52:33,106 This kind of talk that somehow in the context of capitalism, 1103 00:52:33,150 --> 00:52:36,979 AI is gonna be misused, really frustrates me. 1104 00:52:37,023 --> 00:52:39,025 Whenever we develop technology, 1105 00:52:39,068 --> 00:52:42,246 whether it's, you know, coal-burning plants, whether it's cars, 1106 00:52:42,289 --> 00:52:45,249 there's always that incentive to move ahead quickly, 1107 00:52:45,292 --> 00:52:47,555 and there's always the regulatory regime, 1108 00:52:47,599 --> 00:52:50,384 part of the government's job to balance that. 1109 00:52:50,428 --> 00:52:51,777 AI is no different here. 1110 00:52:53,170 --> 00:52:55,259 Whenever we have technological advances, 1111 00:52:55,302 --> 00:52:56,695 we need to temper them 1112 00:52:56,738 --> 00:52:59,219 with appropriate rules and regulations. 1113 00:52:59,524 --> 00:53:00,829 Duh! 1114 00:53:00,873 --> 00:53:03,005 Until there's a major disaster, 1115 00:53:03,049 --> 00:53:04,311 corporations are gonna be saying 1116 00:53:04,355 --> 00:53:06,095 "Oh, we don't need to worry about these things. 1117 00:53:06,139 --> 00:53:07,401 "Trust us, we're the experts. 1118 00:53:07,445 --> 00:53:09,490 "We have humanity's best interests at heart." 1119 00:53:09,534 --> 00:53:12,624 What AI is really about is it's a new set of rules 1120 00:53:12,667 --> 00:53:16,018 for companies in terms of how to be more competitive, 1121 00:53:16,062 --> 00:53:18,543 how to drive more efficiency, and how to drive growth, 1122 00:53:18,586 --> 00:53:20,632 and creative new business models and... 1123 00:53:20,675 --> 00:53:22,416 RUSSELL: Once corporate scientists 1124 00:53:22,460 --> 00:53:23,983 start to dominate the debate, 1125 00:53:24,026 --> 00:53:26,333 it's all over, you've lost. 1126 00:53:26,377 --> 00:53:29,380 You know, they're paid to deny the existence 1127 00:53:29,423 --> 00:53:31,512 of risks and negative outcomes. 1128 00:53:32,948 --> 00:53:36,213 SCHMIDHUBER: At the moment, almost all AI research 1129 00:53:36,256 --> 00:53:38,345 is about making human lives 1130 00:53:38,389 --> 00:53:41,479 longer, healthier and easier. 1131 00:53:42,219 --> 00:53:43,742 Because the big companies, 1132 00:53:43,785 --> 00:53:45,309 they want to sell you AIs 1133 00:53:45,352 --> 00:53:46,527 that are useful for you, 1134 00:53:46,571 --> 00:53:48,312 and so you are only going to buy 1135 00:53:48,355 --> 00:53:49,704 those that help you. 1136 00:53:51,402 --> 00:53:53,578 MALONE: The problem of AI in a capitalist paradigm 1137 00:53:53,621 --> 00:53:56,581 is it's always gonna be cheaper 1138 00:53:56,624 --> 00:53:59,236 just to get the product out there when it does the job 1139 00:53:59,279 --> 00:54:02,195 rather than do it safely. 1140 00:54:02,239 --> 00:54:04,676 ROMAN YAMPOLSKIY: Right now, the way it works, you have a conference 1141 00:54:04,719 --> 00:54:07,548 with 10,000 researchers, all of them working 1142 00:54:07,592 --> 00:54:11,204 on developing more capable AI, except, like, 50 people 1143 00:54:11,248 --> 00:54:13,511 who have a tiny workshop talking about safety 1144 00:54:13,554 --> 00:54:15,687 and controlling what everyone else is doing. 1145 00:54:16,775 --> 00:54:19,734 No one has a working safety mechanism. 1146 00:54:19,778 --> 00:54:22,259 And no one even has a prototype for it. 1147 00:54:24,609 --> 00:54:26,175 SCHMIDHUBER: In the not-so-distant future, 1148 00:54:26,219 --> 00:54:28,439 you will have all kinds of AIs 1149 00:54:28,482 --> 00:54:31,093 that don't just do what humans tell them, 1150 00:54:31,137 --> 00:54:33,400 but they will invent their own goals 1151 00:54:33,444 --> 00:54:35,054 and their own experiments, 1152 00:54:35,097 --> 00:54:37,535 and they will transcend humankind. 1153 00:54:40,364 --> 00:54:41,756 YAMPOLSKIY: As of today, 1154 00:54:41,800 --> 00:54:43,454 I would say it's completely unethical 1155 00:54:43,497 --> 00:54:45,847 to try to develop such a machine and release it. 1156 00:54:48,372 --> 00:54:50,461 We only get one chance to get it right. 1157 00:54:50,504 --> 00:54:53,464 You screw up, everything possibly is lost. 1158 00:54:57,381 --> 00:54:59,948 DULLEA: What happens if we get it wrong? 1159 00:55:01,167 --> 00:55:03,387 The thing that scares me the most is this AI 1160 00:55:03,430 --> 00:55:06,303 will become the product of its parents, 1161 00:55:06,346 --> 00:55:09,131 in the same way that you are and I am. 1162 00:55:09,175 --> 00:55:11,264 Let's ask who those parents are. 1163 00:55:11,308 --> 00:55:12,657 Right? 1164 00:55:12,700 --> 00:55:15,312 Chances are they're either gonna be military parents 1165 00:55:15,355 --> 00:55:17,314 or they're gonna be business parents. 1166 00:55:17,357 --> 00:55:20,142 If they're business parents, they're asking the AI 1167 00:55:20,186 --> 00:55:22,449 to become something 1168 00:55:22,493 --> 00:55:25,626 that can make a lot of money. 1169 00:55:27,672 --> 00:55:29,848 So they're teaching it greed. 1170 00:55:32,198 --> 00:55:35,854 Military parents are gonna teach it to kill. 1171 00:55:38,683 --> 00:55:41,468 Truly superhuman intelligence 1172 00:55:41,512 --> 00:55:44,079 is so much... Is so super. 1173 00:55:44,123 --> 00:55:45,690 It's so much better than we are 1174 00:55:45,733 --> 00:55:49,563 that the first country or company to get there 1175 00:55:49,607 --> 00:55:51,217 has the winner-take-all advantage. 1176 00:55:51,260 --> 00:55:54,002 [APPLAUSE] 1177 00:55:54,046 --> 00:55:56,875 [SPEAKING RUSSIAN] 1178 00:56:14,066 --> 00:56:17,809 Because there's such an arms race in AI globally, 1179 00:56:17,852 --> 00:56:19,898 and that's both literally, in terms of building 1180 00:56:19,941 --> 00:56:22,335 weapons systems, but also figuratively 1181 00:56:22,379 --> 00:56:24,685 with folks like Putin and the Chinese saying 1182 00:56:24,729 --> 00:56:28,036 whoever dominates AI is gonna dominate the economy 1183 00:56:28,080 --> 00:56:30,517 in the 21st century. 1184 00:56:30,561 --> 00:56:33,520 I don't think we have a choice to put the brakes on, 1185 00:56:33,564 --> 00:56:36,001 to stop developing and researching. 1186 00:56:36,044 --> 00:56:39,047 We can build the technology, we can do the research, 1187 00:56:39,091 --> 00:56:40,745 and imbue it with our values, 1188 00:56:40,788 --> 00:56:43,356 but realistically, totalitarian regimes 1189 00:56:43,400 --> 00:56:45,489 will imbue the technology with their values. 1190 00:56:46,707 --> 00:56:49,406 FEMALE REPORTER: Israel, Russia, the US and China 1191 00:56:49,449 --> 00:56:51,146 are all contestants 1192 00:56:51,190 --> 00:56:54,193 in what's being described as a new global arms race. 1193 00:56:56,413 --> 00:56:58,371 All countries, whether they admit it or not, 1194 00:56:58,415 --> 00:57:00,765 are going to develop lethal autonomous weapons. 1195 00:57:03,028 --> 00:57:06,423 And is that really a capability that we want 1196 00:57:06,466 --> 00:57:08,381 in the hands of a corporate nation state 1197 00:57:08,425 --> 00:57:10,252 as opposed to a nation state? 1198 00:57:10,296 --> 00:57:12,124 The reality is if Facebook or Google 1199 00:57:12,167 --> 00:57:14,082 wanted to start an army, 1200 00:57:14,126 --> 00:57:17,912 they would have technologically superior capabilities 1201 00:57:17,956 --> 00:57:19,958 than any other nation on the planet. 1202 00:57:21,481 --> 00:57:24,832 DULLEA: Some AI scientists are so concerned, 1203 00:57:24,876 --> 00:57:26,965 they produced this dramatized video 1204 00:57:27,008 --> 00:57:29,054 to warn of the risks. 1205 00:57:29,097 --> 00:57:31,796 Your kids probably have one of these, right? 1206 00:57:33,493 --> 00:57:34,929 Not quite.RUSSELL: We produced 1207 00:57:34,973 --> 00:57:37,454 the Slaughterbot video to illustrate 1208 00:57:37,497 --> 00:57:39,151 the risk of autonomous weapons. 1209 00:57:39,194 --> 00:57:41,632 That skill is all AI. 1210 00:57:42,676 --> 00:57:44,025 It's flying itself. 1211 00:57:44,983 --> 00:57:47,072 There are people within the AI community 1212 00:57:47,115 --> 00:57:48,639 who wish I would shut up. 1213 00:57:49,814 --> 00:57:52,294 People who think that making the video is 1214 00:57:52,338 --> 00:57:54,775 a little bit treasonous. 1215 00:57:54,819 --> 00:57:57,517 Just like any mobile device these days, 1216 00:57:57,561 --> 00:57:59,780 it has cameras and sensors. 1217 00:57:59,824 --> 00:58:03,044 And just like your phones and social media apps, 1218 00:58:03,088 --> 00:58:04,742 it does facial recognition. 1219 00:58:04,785 --> 00:58:07,353 Up until now, the arms races tended to be 1220 00:58:07,396 --> 00:58:09,616 bigger, more powerful, more expensive, 1221 00:58:09,660 --> 00:58:11,618 faster jets that cost more. 1222 00:58:12,358 --> 00:58:15,622 AI means that you could have 1223 00:58:15,666 --> 00:58:17,885 very small, cheap weapons 1224 00:58:17,929 --> 00:58:19,757 that are very efficient at doing 1225 00:58:19,800 --> 00:58:21,933 one very simple job, like 1226 00:58:22,803 --> 00:58:24,718 finding you in a crowd and killing you. 1227 00:58:25,545 --> 00:58:27,852 This is how it works. 1228 00:58:35,642 --> 00:58:37,470 [GASPING, APPLAUSE] 1229 00:58:37,514 --> 00:58:40,560 Every single major advance in human technology 1230 00:58:40,604 --> 00:58:43,084 has always been weaponized. 1231 00:58:43,128 --> 00:58:48,046 And the idea of weaponized AI, I think, is terrifying. 1232 00:58:48,089 --> 00:58:50,701 Trained as a team, they can penetrate 1233 00:58:50,744 --> 00:58:52,920 buildings, cars, trains, 1234 00:58:52,964 --> 00:58:57,316 evade people, bullets... They cannot be stopped. 1235 00:58:57,359 --> 00:58:59,536 RUSSELL: There are lots of weapons that kill people. 1236 00:58:59,579 --> 00:59:01,059 Machine guns can kill people, 1237 00:59:01,102 --> 00:59:03,844 but they require a human being to carry and point them. 1238 00:59:03,888 --> 00:59:06,368 With autonomous weapons, you don't need a human 1239 00:59:06,412 --> 00:59:08,066 to carry them or supervise them. 1240 00:59:08,936 --> 00:59:12,723 You just program and then off they go. 1241 00:59:12,766 --> 00:59:15,377 So you can create the effect of an army 1242 00:59:15,421 --> 00:59:16,770 carrying a million machine guns, 1243 00:59:16,814 --> 00:59:18,903 uh, with two guys and a truck. 1244 00:59:23,429 --> 00:59:25,866 Putin said that the first person to develop 1245 00:59:25,910 --> 00:59:29,653 strong artificial intelligence will rule the world. 1246 00:59:29,696 --> 00:59:31,568 When world leaders use terms 1247 00:59:31,611 --> 00:59:33,570 like "rule the world," 1248 00:59:33,613 --> 00:59:35,572 I think you gotta pay attention to it. 1249 00:59:43,928 --> 00:59:47,105 Now, that is an air strike of surgical precision. 1250 00:59:47,148 --> 00:59:50,108 [CHEERING AND APPLAUSE] 1251 00:59:50,151 --> 00:59:53,894 I've been campaigning against these weapons since 2007, 1252 00:59:53,938 --> 00:59:56,027 and there's three main sets of arguments. 1253 00:59:56,070 --> 00:59:58,899 There's the moral arguments, there's arguments 1254 00:59:58,943 --> 01:00:01,598 about not complying with the laws of war, 1255 01:00:01,641 --> 01:00:03,904 and there are arguments about how they will change 1256 01:00:03,948 --> 01:00:05,514 global security of the planet. 1257 01:00:05,558 --> 01:00:07,604 We need new regulations. 1258 01:00:07,647 --> 01:00:11,433 We need to really think about it now, yesterday. 1259 01:00:11,477 --> 01:00:14,262 Some of the world's leaders in artificial intelligence 1260 01:00:14,306 --> 01:00:17,439 are urging the United Nations to ban the development 1261 01:00:17,483 --> 01:00:18,919 of autonomous weapons, 1262 01:00:18,963 --> 01:00:20,921 otherwise known as killer robots. 1263 01:00:20,965 --> 01:00:23,750 MALE REPORTER: In the letter they say "once developed, 1264 01:00:23,794 --> 01:00:26,013 "they will permit armed conflict to be fought 1265 01:00:26,057 --> 01:00:29,669 "at timescales faster than humans can comprehend." 1266 01:00:29,713 --> 01:00:32,237 BOOCH: I'm a signatory to that letter. 1267 01:00:32,280 --> 01:00:33,717 My philosophy is, 1268 01:00:33,760 --> 01:00:36,502 and I hope the governments of the world enforce this, 1269 01:00:36,545 --> 01:00:39,331 that as long as a human life is involved, 1270 01:00:39,374 --> 01:00:41,420 then a human needs to be in the loop. 1271 01:00:43,988 --> 01:00:47,078 Anyone with a 3D printer and a computer 1272 01:00:47,121 --> 01:00:48,993 can have a lethal autonomous weapon. 1273 01:00:49,036 --> 01:00:51,691 If it's accessible to everyone on the planet, 1274 01:00:51,735 --> 01:00:53,954 then compliance, for example, 1275 01:00:53,998 --> 01:00:56,783 would be nearly impossible to track. 1276 01:00:56,827 --> 01:00:58,785 What the hell are you doing? Stop! 1277 01:00:58,829 --> 01:01:00,657 You're shooting into thin air. 1278 01:01:00,700 --> 01:01:02,223 HARRIS: When you're imagining 1279 01:01:02,267 --> 01:01:03,877 waging war in the future, 1280 01:01:03,921 --> 01:01:06,053 where soldiers are robots, 1281 01:01:06,097 --> 01:01:09,796 and we pose no risk of engagement to ourselves... 1282 01:01:09,840 --> 01:01:12,494 When war becomes just like a video game, 1283 01:01:12,538 --> 01:01:14,671 as it has incrementally already, 1284 01:01:15,280 --> 01:01:16,716 it's a real concern. 1285 01:01:19,023 --> 01:01:21,199 When we don't have skin in the game, quite literally... 1286 01:01:22,200 --> 01:01:23,723 [SCREAMS] 1287 01:01:23,767 --> 01:01:26,770 ...when all of this is being run from a terminal 1288 01:01:26,813 --> 01:01:29,294 thousands of miles away from the battlefield... 1289 01:01:33,080 --> 01:01:35,953 ...that could lower the bar for waging war 1290 01:01:35,996 --> 01:01:37,606 to a point where we would recognize 1291 01:01:37,650 --> 01:01:39,913 that we have become monsters. 1292 01:01:50,707 --> 01:01:51,882 Who are you? 1293 01:01:53,144 --> 01:01:55,233 Hey! Hey! 1294 01:01:56,582 --> 01:01:58,062 [GUNSHOT] 1295 01:01:58,976 --> 01:02:01,543 Having been a former fighter pilot, 1296 01:02:01,587 --> 01:02:03,589 I have also seen people make mistakes 1297 01:02:03,632 --> 01:02:05,765 and kill innocent people 1298 01:02:05,809 --> 01:02:07,898 who should not have been killed. 1299 01:02:07,941 --> 01:02:11,205 One day in the future, it will be more reliable 1300 01:02:11,249 --> 01:02:13,512 to have a machine do it than a human do it. 1301 01:02:13,555 --> 01:02:15,035 The idea of robots 1302 01:02:15,079 --> 01:02:17,734 who can kill humans on their own 1303 01:02:17,777 --> 01:02:20,301 has been a science fiction staple for decades. 1304 01:02:20,345 --> 01:02:23,043 There are a number of these systems around already, 1305 01:02:23,087 --> 01:02:24,784 uh, that work autonomously. 1306 01:02:24,828 --> 01:02:26,873 But that's called supervised autonomy 1307 01:02:26,917 --> 01:02:30,224 because somebody's there to switch it on and off. 1308 01:02:30,268 --> 01:02:32,052 MALE REPORTER: What's only been seen in Hollywood 1309 01:02:32,096 --> 01:02:35,839 could soon be coming to a battlefield near you. 1310 01:02:35,882 --> 01:02:37,797 The Pentagon investing billions of dollars 1311 01:02:37,841 --> 01:02:39,407 to develop autonomous weapons, 1312 01:02:39,451 --> 01:02:41,845 machines that could one day kill on their own. 1313 01:02:41,888 --> 01:02:45,239 But the Pentagon insists humans will decide to strike. 1314 01:02:45,283 --> 01:02:48,460 MALONE: Once you get AI into military hardware, 1315 01:02:49,853 --> 01:02:51,985 you've got to make it make the decisions, 1316 01:02:52,029 --> 01:02:53,944 otherwise there's no point in having it. 1317 01:02:55,119 --> 01:02:58,035 One of the benefits of AI is that it's quicker than us. 1318 01:03:00,037 --> 01:03:02,474 Once we're in a proper fight and that switch 1319 01:03:02,517 --> 01:03:04,476 that keeps the human in the loop, 1320 01:03:04,519 --> 01:03:07,044 we better switch it off, because if you don't, you lose. 1321 01:03:10,047 --> 01:03:11,788 These will make the battlefield 1322 01:03:11,831 --> 01:03:13,441 too fast for humans. 1323 01:03:19,273 --> 01:03:21,493 If there's no human involved at all, 1324 01:03:21,536 --> 01:03:25,279 if command and control have gone over to autonomy, 1325 01:03:25,323 --> 01:03:27,716 we could trigger a war and it could be finished 1326 01:03:27,760 --> 01:03:30,110 in 20 minutes with massive devastation, 1327 01:03:30,154 --> 01:03:32,460 and nobody knew about it before it started. 1328 01:03:40,251 --> 01:03:44,298 DULLEA: So who decides what we should do in the future 1329 01:03:45,560 --> 01:03:46,561 now? 1330 01:03:48,389 --> 01:03:51,958 YAMPOLSKIY: From Big Bang till now, billions of years. 1331 01:03:52,959 --> 01:03:56,658 This is the only time a person can steer 1332 01:03:56,702 --> 01:03:58,225 the future of the whole humanity. 1333 01:04:12,892 --> 01:04:15,982 TEGMARK: Maybe we should worry that we humans aren't wise enough 1334 01:04:16,026 --> 01:04:17,636 to handle this much power. 1335 01:04:20,378 --> 01:04:23,294 GLEISER: Who is actually going to make the choices, 1336 01:04:23,337 --> 01:04:26,950 and in the name of whom are you actually speaking? 1337 01:04:34,087 --> 01:04:37,482 If we, the rest of society, fail to actually 1338 01:04:37,525 --> 01:04:39,788 really join this conversation in a meaningful way 1339 01:04:39,832 --> 01:04:42,182 and be part of the decision, 1340 01:04:42,226 --> 01:04:45,098 then we know whose values are gonna go into these 1341 01:04:45,142 --> 01:04:47,448 super-intelligent beings. 1342 01:04:47,492 --> 01:04:50,103 Red Bull.Yes, Deon. 1343 01:04:50,843 --> 01:04:52,584 Red Bull. 1344 01:04:52,627 --> 01:04:56,196 TEGMARK: It's going to be decided by some tech nerds 1345 01:04:56,240 --> 01:04:58,024 who've had too much Red Bull to drink... 1346 01:04:58,068 --> 01:04:59,939 who I don't think have any particular 1347 01:04:59,983 --> 01:05:01,723 qualifications to speak 1348 01:05:01,767 --> 01:05:03,247 on behalf of all humanity. 1349 01:05:05,945 --> 01:05:10,080 Once you are in the excitement of discovery... 1350 01:05:11,516 --> 01:05:12,778 Come on! 1351 01:05:12,821 --> 01:05:14,562 ...you sort of turn a blind eye 1352 01:05:14,606 --> 01:05:16,434 to all these more moralistic questions, 1353 01:05:16,477 --> 01:05:21,091 like how is this going to impact society as a whole? 1354 01:05:23,223 --> 01:05:25,182 And you do it because you can. 1355 01:05:29,664 --> 01:05:31,579 MALONE: And the scientists are pushing this forward 1356 01:05:31,623 --> 01:05:34,582 because they can see the genuine benefits 1357 01:05:34,626 --> 01:05:35,670 that this will bring. 1358 01:05:36,845 --> 01:05:38,412 And they're also really, really excited. 1359 01:05:38,456 --> 01:05:41,241 I mean, you don't... You don't go into science 1360 01:05:41,285 --> 01:05:43,852 because, oh, well, you couldn't be a football player, 1361 01:05:43,896 --> 01:05:45,550 "Hell, I'll be a scientist." 1362 01:05:45,593 --> 01:05:47,291 You go in because that's what you love and it's really exciting. 1363 01:05:49,423 --> 01:05:50,947 Oh! It's alive! 1364 01:05:50,990 --> 01:05:53,253 It's alive! It's alive! 1365 01:05:53,297 --> 01:05:54,646 It's alive! 1366 01:05:54,689 --> 01:05:55,995 YAMPOLSKIY: We're switching 1367 01:05:56,039 --> 01:05:57,518 from being creation to being God. 1368 01:05:57,562 --> 01:06:00,391 Now I know what it feels like to be God! 1369 01:06:00,434 --> 01:06:02,219 YAMPOLSKIY: If you study theology, 1370 01:06:02,262 --> 01:06:05,787 one thing you learn is that it is never the case 1371 01:06:05,831 --> 01:06:07,528 that God successfully manages 1372 01:06:07,572 --> 01:06:10,270 to control its creation. 1373 01:06:14,144 --> 01:06:16,189 DULLEA: We're not there yet, 1374 01:06:16,233 --> 01:06:17,495 but if we do create 1375 01:06:17,538 --> 01:06:19,671 artificial general intelligence, 1376 01:06:20,802 --> 01:06:22,456 can we maintain control? 1377 01:06:24,676 --> 01:06:27,287 If you create a self-aware AI, 1378 01:06:27,331 --> 01:06:29,637 the first thing it's gonna want 1379 01:06:29,681 --> 01:06:31,204 is independence. 1380 01:06:31,248 --> 01:06:34,642 At least, the one intelligence that we all know is ourselves. 1381 01:06:34,686 --> 01:06:38,037 And we know that we would not wanna be doing the bidding 1382 01:06:38,081 --> 01:06:39,996 of some other intelligence. 1383 01:06:41,867 --> 01:06:45,566 We have no reason to believe that it will be friendly... 1384 01:06:56,186 --> 01:06:58,579 ...or that it will have goals and aspirations 1385 01:06:58,623 --> 01:07:00,277 that align with ours. 1386 01:07:04,020 --> 01:07:06,500 So how long before there's a robot revolution 1387 01:07:06,544 --> 01:07:08,024 or an AI revolution? 1388 01:07:18,077 --> 01:07:20,775 We are in a world today where it is 1389 01:07:20,819 --> 01:07:22,473 absolutely a foot race 1390 01:07:22,516 --> 01:07:25,693 to go as fast as possible toward creating 1391 01:07:25,737 --> 01:07:28,305 an artificial general intelligence 1392 01:07:28,348 --> 01:07:31,569 and there don't seem to be any brakes being applied anywhere in sight. 1393 01:07:33,571 --> 01:07:34,789 It's crazy. 1394 01:07:42,493 --> 01:07:44,712 YAMPOLSKIY: I think it's important to honestly 1395 01:07:44,756 --> 01:07:46,888 discuss such issues. 1396 01:07:46,932 --> 01:07:48,977 It might be a little too late to make 1397 01:07:49,021 --> 01:07:51,371 this conversation afterwards. 1398 01:07:59,423 --> 01:08:04,297 It's important to remember that intelligence is not evil 1399 01:08:04,341 --> 01:08:06,343 and it is also not good. 1400 01:08:07,170 --> 01:08:09,737 Intelligence gives power. 1401 01:08:09,781 --> 01:08:12,305 We are more powerful than tigers on this planet, 1402 01:08:12,349 --> 01:08:14,525 not because we have sharper claws 1403 01:08:14,568 --> 01:08:16,570 or bigger biceps, because we're smarter. 1404 01:08:17,615 --> 01:08:19,704 CHALMERS: We do take the attitude that ants 1405 01:08:19,747 --> 01:08:23,403 matter a bit and fish matter a bit, but humans matter more. 1406 01:08:23,447 --> 01:08:25,405 Why? Because we're more intelligent, 1407 01:08:25,449 --> 01:08:27,581 more sophisticated, more conscious. 1408 01:08:27,625 --> 01:08:30,236 Once you've got systems which are more intelligent than us, 1409 01:08:30,280 --> 01:08:32,282 maybe they're simply gonna matter more. 1410 01:08:41,421 --> 01:08:44,598 As humans, we live in a world of microorganisms. 1411 01:08:46,122 --> 01:08:48,124 They're life, but if a microorganism 1412 01:08:48,167 --> 01:08:50,517 becomes a nuisance, we take an antibiotic 1413 01:08:50,561 --> 01:08:53,129 and we destroy billions of them at once. 1414 01:08:57,220 --> 01:08:59,526 If you were an artificial intelligence, 1415 01:08:59,570 --> 01:09:01,137 wouldn't you look at every little human 1416 01:09:01,180 --> 01:09:03,008 that has this very limited view 1417 01:09:03,051 --> 01:09:05,402 kind of like we look at microorganisms? 1418 01:09:09,188 --> 01:09:10,494 DULLEA: Maybe I would. 1419 01:09:13,410 --> 01:09:16,152 Maybe human insignificance 1420 01:09:16,195 --> 01:09:18,632 is the biggest threat of all. 1421 01:09:21,505 --> 01:09:24,247 TEGMARK: Sometimes when I talk about AI risk, people go, 1422 01:09:24,290 --> 01:09:26,162 "Shh! Max, don't talk like that. 1423 01:09:26,205 --> 01:09:28,338 "That's Luddite scaremongering." 1424 01:09:28,381 --> 01:09:30,035 But it's not scaremongering. 1425 01:09:30,078 --> 01:09:33,169 It's what we at MIT call safety engineering. 1426 01:09:35,867 --> 01:09:39,218 I mean, think about it, before NASA launched the Apollo 11 moon mission, 1427 01:09:39,262 --> 01:09:40,785 they systematically thought through 1428 01:09:40,828 --> 01:09:43,004 everything that could go wrong when you put a bunch of people 1429 01:09:43,048 --> 01:09:45,616 on top of explosive fuel tanks and launched them somewhere 1430 01:09:45,659 --> 01:09:48,532 where no one can help them. Was that scaremongering? 1431 01:09:48,575 --> 01:09:51,012 No, that was precisely the safety engineering 1432 01:09:51,056 --> 01:09:54,538 that guaranteed the success of the mission. 1433 01:09:54,581 --> 01:09:58,498 That's precisely the strategy I think we have to take with AGI. 1434 01:09:58,542 --> 01:10:00,805 Think through everything that could go wrong 1435 01:10:00,848 --> 01:10:02,720 to make sure it goes right. 1436 01:10:06,898 --> 01:10:08,856 I am who I am because of the movie 1437 01:10:08,900 --> 01:10:11,337 2001: A Space Odyssey. 1438 01:10:11,381 --> 01:10:14,601 Aged 14, it was an eye-opener for me. 1439 01:10:14,645 --> 01:10:17,343 That movie is the best of all science fiction 1440 01:10:17,387 --> 01:10:20,912 with its predictions for how AI would evolve. 1441 01:10:23,088 --> 01:10:24,872 But in reality, 1442 01:10:24,916 --> 01:10:27,658 if something is sufficiently far in the future, 1443 01:10:27,701 --> 01:10:30,748 we can't imagine what the world will be like, 1444 01:10:30,791 --> 01:10:33,229 so we actually can't say anything sensible 1445 01:10:33,272 --> 01:10:34,447 about what the issues are. 1446 01:10:36,014 --> 01:10:37,233 LITTMAN: Worrying about super-intelligence safety 1447 01:10:37,276 --> 01:10:39,278 is a little like worrying about 1448 01:10:39,322 --> 01:10:40,932 how you're gonna install seat belts 1449 01:10:40,975 --> 01:10:42,542 in transporter devices. 1450 01:10:42,586 --> 01:10:44,457 Like in Star Trek, where you can just zip 1451 01:10:44,501 --> 01:10:46,242 from one place to another. 1452 01:10:46,285 --> 01:10:48,331 Um, maybe there's some dangers to that, 1453 01:10:48,374 --> 01:10:49,941 so we should have seatbelts, right? 1454 01:10:49,984 --> 01:10:51,986 But it's of course very silly because we don't know enough 1455 01:10:52,030 --> 01:10:53,466 about how transporter technology 1456 01:10:53,510 --> 01:10:55,338 might look to be able to say 1457 01:10:55,381 --> 01:10:57,427 whether seatbelts even apply. 1458 01:10:58,993 --> 01:11:01,648 GLEISER: To say that there won't be a problem because 1459 01:11:02,954 --> 01:11:04,869 the unknowns are so vast, 1460 01:11:04,912 --> 01:11:07,567 that we don't even know where the problem is gonna be, 1461 01:11:08,829 --> 01:11:10,570 that is crazy. 1462 01:11:15,706 --> 01:11:18,056 I think conversation is probably the only way forward here. 1463 01:11:18,099 --> 01:11:20,014 At the moment, among ourselves, 1464 01:11:20,058 --> 01:11:21,364 and when the time comes, 1465 01:11:21,407 --> 01:11:23,627 with the emerging artificial intelligences. 1466 01:11:25,716 --> 01:11:27,152 So as soon as there is anything 1467 01:11:27,195 --> 01:11:28,719 remotely in that ballpark, 1468 01:11:28,762 --> 01:11:30,721 we should start trying to have a conversation 1469 01:11:30,764 --> 01:11:32,636 with it somehow. 1470 01:11:33,637 --> 01:11:35,900 Dog. Dog.SAGAR: Yeah. 1471 01:11:35,943 --> 01:11:38,946 One of the things we hope to do with Baby X is to try 1472 01:11:38,990 --> 01:11:41,906 to get a better insight into our own nature. 1473 01:11:43,777 --> 01:11:46,302 She's designed to learn autonomously. 1474 01:11:47,564 --> 01:11:50,610 She can be plugged into the Internet, 1475 01:11:50,654 --> 01:11:54,266 click on webpages by herself and see what happens, 1476 01:11:54,310 --> 01:11:56,442 kind of learn how we learn. 1477 01:11:57,400 --> 01:12:00,707 If we make this super fantastic AI, 1478 01:12:00,751 --> 01:12:04,363 it may well help us become better versions of ourselves, 1479 01:12:06,365 --> 01:12:08,106 help us evolve. 1480 01:12:15,505 --> 01:12:17,550 It will have societal effects. 1481 01:12:19,117 --> 01:12:20,597 Is it to be feared? 1482 01:12:22,381 --> 01:12:24,035 Or is it something to learn from? 1483 01:12:28,648 --> 01:12:31,999 DULLEA: So when might we be faced with machines 1484 01:12:32,043 --> 01:12:36,656 that are as smart, smarter than we are? 1485 01:12:38,571 --> 01:12:41,661 I think one of the ways to think about the near future of AI 1486 01:12:41,705 --> 01:12:44,708 is what I would call the event horizon. 1487 01:12:46,231 --> 01:12:48,015 People have heard of black holes, 1488 01:12:48,059 --> 01:12:50,191 this super dense point where 1489 01:12:50,235 --> 01:12:52,977 if you get too close, you get sucked in. 1490 01:12:53,020 --> 01:12:55,893 There is a boundary. Once you cross that boundary, 1491 01:12:55,936 --> 01:12:58,025 there is no way back. 1492 01:12:58,069 --> 01:13:02,900 What makes that dangerous is you can't tell where it is. 1493 01:13:02,943 --> 01:13:05,816 And I think AI as a technology is gonna be like that, 1494 01:13:05,859 --> 01:13:09,210 that there will be a point where the logic of the money 1495 01:13:09,254 --> 01:13:11,430 and the power and the promise 1496 01:13:11,474 --> 01:13:14,781 will be such that once we cross that boundary, 1497 01:13:14,825 --> 01:13:16,783 we will then be into that future. 1498 01:13:16,827 --> 01:13:19,395 There will be no way back, politically, economically 1499 01:13:19,438 --> 01:13:21,658 or even in terms of what we want. 1500 01:13:21,701 --> 01:13:24,008 But we need to be worrying about it before we reach 1501 01:13:24,051 --> 01:13:27,315 the event horizon, and we don't know where it is. 1502 01:13:27,359 --> 01:13:29,405 I don't think it's gonna happen any time soon. 1503 01:13:29,448 --> 01:13:31,145 The next few decades. 1504 01:13:31,189 --> 01:13:32,364 Twenty-five to 50 years. 1505 01:13:32,408 --> 01:13:33,452 A few hundred years. 1506 01:13:33,496 --> 01:13:34,584 It's over the next few decades. 1507 01:13:34,627 --> 01:13:35,802 Not in my lifetime. 1508 01:13:35,846 --> 01:13:38,239 Nobody knows. But it doesn't really matter. 1509 01:13:38,283 --> 01:13:41,634 Whether it's years or 100 years, the problem is the same. 1510 01:13:41,678 --> 01:13:44,594 It's just as real and we'll need all this time to solve it. 1511 01:13:49,816 --> 01:13:53,733 TEGMARK: History is just full of over-hyped tech predictions. 1512 01:13:53,777 --> 01:13:55,605 Like, where are those flying cars 1513 01:13:55,648 --> 01:13:57,389 that were supposed to be here, right? 1514 01:13:57,433 --> 01:14:00,740 But it's also full of the opposite, 1515 01:14:00,784 --> 01:14:03,134 when people said, "Oh, never gonna happen." 1516 01:14:07,878 --> 01:14:10,968 I'm embarrassed as a physicist that Ernest Rutherford, 1517 01:14:11,011 --> 01:14:13,274 one of the greatest nuclear physicists of all time said 1518 01:14:13,318 --> 01:14:15,929 that nuclear energy was never gonna happen. 1519 01:14:17,061 --> 01:14:18,932 And the very next day, 1520 01:14:18,976 --> 01:14:22,458 Leo Szilard invents the nuclear chain reaction. 1521 01:14:22,501 --> 01:14:24,677 So if someone says they are completely sure 1522 01:14:24,721 --> 01:14:27,114 something is never gonna happen, 1523 01:14:27,158 --> 01:14:29,377 even though it's consistent with the laws of physics, 1524 01:14:29,421 --> 01:14:31,510 I'm like, "Yeah, right." 1525 01:14:36,646 --> 01:14:38,430 HARRIS: We're not there yet. 1526 01:14:38,474 --> 01:14:41,477 We haven't built a super-intelligent AI. 1527 01:14:45,002 --> 01:14:47,483 And even if we stipulated that it won't happen 1528 01:14:47,526 --> 01:14:49,528 shorter than 50 years from now, right? 1529 01:14:49,572 --> 01:14:52,183 So we have 50 years to think about this. 1530 01:14:52,226 --> 01:14:53,706 Well... 1531 01:14:53,750 --> 01:14:56,143 it's nowhere written that 50 years 1532 01:14:56,187 --> 01:14:59,059 is enough time to figure out how to do this safely. 1533 01:15:01,018 --> 01:15:02,628 CAMERON: I actually think it's going to take 1534 01:15:02,672 --> 01:15:05,196 some extreme negative example 1535 01:15:05,239 --> 01:15:08,504 of the technology for everyone to pull back 1536 01:15:08,547 --> 01:15:11,115 and realize that we've got to get this thing 1537 01:15:11,158 --> 01:15:13,334 under control, we've got to have constraints. 1538 01:15:13,378 --> 01:15:16,207 We are living in a science fiction world and we're all, 1539 01:15:16,250 --> 01:15:20,037 um, in this big, kind of social, tactical experiment, 1540 01:15:20,080 --> 01:15:22,692 experimenting on ourselves, recklessly, 1541 01:15:22,735 --> 01:15:25,172 like no doctor would ever dream of doing. 1542 01:15:25,216 --> 01:15:26,522 We're injecting ourselves 1543 01:15:26,565 --> 01:15:28,262 with everything that we come up with 1544 01:15:28,306 --> 01:15:29,655 to see what happens. 1545 01:15:31,875 --> 01:15:34,486 LITTMAN: Clearly, the thought process is... 1546 01:15:34,530 --> 01:15:36,140 You're creating learning robots. 1547 01:15:36,183 --> 01:15:38,534 Learning robots are exactly what lead to Skynet. 1548 01:15:38,577 --> 01:15:40,884 Now I've got a choice, right? I can say, 1549 01:15:40,927 --> 01:15:42,842 "Don't worry, I'm not good at my job. 1550 01:15:43,887 --> 01:15:45,932 "You have nothing to fear because I suck at this 1551 01:15:45,976 --> 01:15:48,108 "and I'm not gonna actually accomplish anything." 1552 01:15:48,152 --> 01:15:50,458 Or I have to say, "You know, 1553 01:15:50,502 --> 01:15:51,938 "You're right, we're all gonna die, 1554 01:15:51,982 --> 01:15:53,810 "but it's gonna be sort of fun to watch, 1555 01:15:53,853 --> 01:15:56,856 "and it'd be good to be one of the people in the front row." 1556 01:15:56,900 --> 01:16:00,033 So yeah, I don't have a terrific answer to these things. 1557 01:16:02,122 --> 01:16:04,734 The truth, obviously, is somewhere in between. 1558 01:16:07,171 --> 01:16:09,260 DULLEA: Can there be an in-between? 1559 01:16:10,435 --> 01:16:13,264 Or are we facing a new world order? 1560 01:16:15,614 --> 01:16:18,965 I have some colleagues who are fine with AI taking over 1561 01:16:19,009 --> 01:16:21,228 and even causing human extinction, 1562 01:16:21,272 --> 01:16:22,752 as long as we feel 1563 01:16:22,795 --> 01:16:25,189 that those AIs are our worthy ancestors. 1564 01:16:28,453 --> 01:16:32,065 SCHMIDHUBER: In the long run, humans are not going to remain 1565 01:16:32,109 --> 01:16:34,372 the crown of creation. 1566 01:16:36,679 --> 01:16:38,594 But that's okay. 1567 01:16:38,637 --> 01:16:41,727 Because there is still beauty 1568 01:16:41,771 --> 01:16:44,469 and grandeur and greatness 1569 01:16:44,512 --> 01:16:47,211 in realizing that you are 1570 01:16:48,212 --> 01:16:51,955 a tiny part of 1571 01:16:51,998 --> 01:16:55,045 a much grander scheme which is leading the universe 1572 01:16:55,088 --> 01:16:59,223 from lower complexity towards higher complexity. 1573 01:17:07,318 --> 01:17:11,017 CAMERON: We could be rapidly approaching a precipice 1574 01:17:11,061 --> 01:17:13,063 for the human race where we find 1575 01:17:13,106 --> 01:17:16,762 that our ultimate destiny is to have been 1576 01:17:16,806 --> 01:17:20,113 the forebearer of the true 1577 01:17:20,157 --> 01:17:24,161 emergent, evolutionary, intelligent sort of beings. 1578 01:17:26,990 --> 01:17:30,515 So we may just be here to give birth 1579 01:17:30,558 --> 01:17:32,735 to the child of this... Of this machine... 1580 01:17:32,778 --> 01:17:34,345 Machine civilization. 1581 01:17:37,957 --> 01:17:39,785 CHALMERS: Selfishly, as a human, of course, 1582 01:17:39,829 --> 01:17:41,961 I want humans to stick around. 1583 01:17:44,355 --> 01:17:47,837 But maybe that's a very local, parochial point of view. 1584 01:17:49,795 --> 01:17:52,102 Maybe viewed from the point of view of the universe, 1585 01:17:52,145 --> 01:17:53,973 it's a better universe with our... 1586 01:17:54,017 --> 01:17:57,020 With our super-intelligent descendants. 1587 01:18:03,766 --> 01:18:06,377 RUSSELL: Our field is likely to succeed 1588 01:18:06,420 --> 01:18:09,772 in its goal to achieve general purpose intelligence. 1589 01:18:12,209 --> 01:18:16,082 We now have the capability to radically modify 1590 01:18:16,126 --> 01:18:17,954 the economic structure of the world, 1591 01:18:17,997 --> 01:18:19,825 the role of human beings in the world 1592 01:18:19,869 --> 01:18:22,480 and, potentially, human control over the world. 1593 01:18:34,666 --> 01:18:37,277 The idea was that he is taken in 1594 01:18:37,321 --> 01:18:39,192 by godlike entities, 1595 01:18:39,236 --> 01:18:42,239 creatures of pure energy and intelligence. 1596 01:18:43,893 --> 01:18:47,897 They put him in what you could describe as a human zoo. 1597 01:18:49,812 --> 01:18:51,901 DULLEA: The end of the astronaut's journey. 1598 01:18:56,601 --> 01:18:58,081 Did he travel through time? 1599 01:18:58,864 --> 01:19:00,561 To the future? 1600 01:19:00,605 --> 01:19:04,827 Were those godlike entities the descendants of HAL? 1601 01:19:06,089 --> 01:19:09,092 Artificial intelligence that you can't switch off, 1602 01:19:10,658 --> 01:19:13,879 that keeps learning and learning until 1603 01:19:14,488 --> 01:19:16,316 they're running the zoo, 1604 01:19:16,360 --> 01:19:18,144 the universe, 1605 01:19:18,188 --> 01:19:19,363 everything. 1606 01:19:22,061 --> 01:19:26,022 Is that Kubrick's most profound prophecy? 1607 01:19:27,110 --> 01:19:28,154 Maybe. 1608 01:19:30,548 --> 01:19:32,985 But take it from an old man. 1609 01:19:33,029 --> 01:19:35,466 If HAL is close to being amongst us, 1610 01:19:36,510 --> 01:19:38,121 we need to pay attention. 1611 01:19:39,470 --> 01:19:41,559 We need to talk about AI. 1612 01:19:42,734 --> 01:19:44,388 Super-intelligence. 1613 01:19:44,431 --> 01:19:45,432 Will it be conscious? 1614 01:19:46,390 --> 01:19:47,478 The trolley problem. 1615 01:19:48,261 --> 01:19:49,306 Goal alignment. 1616 01:19:50,350 --> 01:19:51,743 The singularity. 1617 01:19:51,787 --> 01:19:53,049 First to the market. 1618 01:19:53,789 --> 01:19:54,964 AI safety. 1619 01:19:56,008 --> 01:19:57,401 Lethal autonomous weapons. 1620 01:19:58,794 --> 01:19:59,969 Robot revolution. 1621 01:20:00,708 --> 01:20:01,971 The hive mind. 1622 01:20:03,320 --> 01:20:04,800 The transhuman movement. 1623 01:20:05,888 --> 01:20:08,629 Transcend humankind. 1624 01:20:08,673 --> 01:20:10,196 We don't have a plan, we don't realize 1625 01:20:10,240 --> 01:20:11,807 it's necessary to have a plan. 1626 01:20:13,286 --> 01:20:15,549 What do we want the role of humans to be? 1627 01:20:25,385 --> 01:20:27,997 [BEEPING]