1 00:00:15,580 --> 00:00:25,547 ♪♪ 2 00:00:25,590 --> 00:00:35,035 ♪♪ 3 00:00:35,078 --> 00:00:41,302 What we're on the brink of is a world of increasingly intense, 4 00:00:41,345 --> 00:00:45,219 sophisticated artificial intelligence. 5 00:00:45,262 --> 00:00:48,396 Man: Technology is evolving so much faster than our society 6 00:00:48,439 --> 00:00:51,181 has the ability to protect us as citizens. 7 00:00:51,486 --> 00:00:55,707 The robots are coming, and they will destroy our livelihoods. 8 00:00:55,751 --> 00:01:01,844 ♪♪ 9 00:01:01,887 --> 00:01:04,238 You have a networked intelligence that watches us, 10 00:01:04,281 --> 00:01:08,590 knows everything about us, and begins to try to change us. 11 00:01:08,633 --> 00:01:12,768 Man #2: Twitter has become the world's number-one news site. 12 00:01:12,811 --> 00:01:15,205 Man #3: Technology is never good or bad. 13 00:01:15,249 --> 00:01:18,948 It's what we do with the technology. 14 00:01:18,991 --> 00:01:22,734 Eventually, millions of people are gonna be thrown out of jobs 15 00:01:22,778 --> 00:01:25,737 because their skills are going to be obsolete. 16 00:01:25,781 --> 00:01:27,435 Woman: Mass unemployment... 17 00:01:27,478 --> 00:01:32,527 greater inequalities, even social unrest. 18 00:01:32,570 --> 00:01:35,530 Man #4: Regardless of whether to be afraid or not afraid, 19 00:01:35,573 --> 00:01:38,185 the change is coming, and nobody can stop it. 20 00:01:38,228 --> 00:01:44,539 ♪♪ 21 00:01:44,582 --> 00:01:46,323 Man #5: We've invested huge amounts of money, 22 00:01:46,367 --> 00:01:49,283 and so it stands to reason that the military, 23 00:01:49,326 --> 00:01:50,893 with their own desires, 24 00:01:50,936 --> 00:01:53,330 are gonna start to use these technologies. 25 00:01:53,374 --> 00:01:55,419 Man #6: Autonomous weapons systems 26 00:01:55,463 --> 00:01:57,552 could lead to a global arms race 27 00:01:57,595 --> 00:02:00,032 to rival the Nuclear Era. 28 00:02:00,076 --> 00:02:02,296 ♪♪ 29 00:02:02,339 --> 00:02:04,036 Man #7: We know what the answer is. 30 00:02:04,080 --> 00:02:05,429 They'll eventually be killing us. 31 00:02:05,473 --> 00:02:10,782 ♪♪ 32 00:02:10,826 --> 00:02:12,349 Man #8: These technology leaps 33 00:02:12,393 --> 00:02:15,874 are gonna yield incredible miracles... 34 00:02:15,918 --> 00:02:18,181 and incredible horrors. 35 00:02:18,225 --> 00:02:24,231 ♪♪ 36 00:02:24,274 --> 00:02:29,323 Man #9: We created it, so I think, as we move forward, 37 00:02:29,366 --> 00:02:33,762 this intelligence will contain parts of us. 38 00:02:33,805 --> 00:02:35,981 And I think the question is -- 39 00:02:36,025 --> 00:02:39,463 Will it contain the good parts... 40 00:02:39,507 --> 00:02:41,378 or the bad parts? 41 00:02:41,422 --> 00:02:47,079 ♪♪ 42 00:02:57,742 --> 00:03:04,793 ♪♪ 43 00:03:04,836 --> 00:03:08,840 Sarah: The survivors called the war "Judgment Day." 44 00:03:08,884 --> 00:03:12,583 They lived only to face a new nightmare -- 45 00:03:12,627 --> 00:03:14,019 the war against the machines. 46 00:03:14,063 --> 00:03:15,412 Aah! 47 00:03:15,456 --> 00:03:18,023 Nolan: I think we've completely fucked this up. 48 00:03:18,067 --> 00:03:21,549 I think Hollywood has managed to inoculate the general public 49 00:03:21,592 --> 00:03:24,247 against this question -- 50 00:03:24,291 --> 00:03:28,251 the idea of machines that will take over the world. 51 00:03:28,295 --> 00:03:30,645 Open the pod bay doors, HAL. 52 00:03:30,688 --> 00:03:33,561 I'm sorry, Dave. 53 00:03:33,604 --> 00:03:35,911 I'm afraid I can't do that. 54 00:03:37,434 --> 00:03:38,696 HAL? 55 00:03:38,740 --> 00:03:40,437 Nolan: We've cried wolf enough times... 56 00:03:40,481 --> 00:03:42,483 HAL? ...that the public has stopped paying attention, 57 00:03:42,526 --> 00:03:43,962 because it feels like science fiction. 58 00:03:44,006 --> 00:03:45,486 Even sitting here talking about it right now, 59 00:03:45,529 --> 00:03:48,228 it feels a little bit silly, a little bit like, 60 00:03:48,271 --> 00:03:51,666 "Oh, this is an artifact of some cheeseball movie." 61 00:03:51,709 --> 00:03:56,584 The WOPR spends all its time thinking about World War III. 62 00:03:56,627 --> 00:03:59,064 But it's not. 63 00:03:59,108 --> 00:04:02,111 The general public is about to get blindsided by this. 64 00:04:02,154 --> 00:04:11,512 ♪♪ 65 00:04:11,555 --> 00:04:13,514 As a society and as individuals, 66 00:04:13,557 --> 00:04:18,954 we're increasingly surrounded by machine intelligence. 67 00:04:18,997 --> 00:04:22,653 We carry this pocket device in the palm of our hand 68 00:04:22,697 --> 00:04:24,829 that we use to make a striking array 69 00:04:24,873 --> 00:04:26,831 of life decisions right now, 70 00:04:26,875 --> 00:04:29,007 aided by a set of distant algorithms 71 00:04:29,051 --> 00:04:30,748 that we have no understanding. 72 00:04:30,792 --> 00:04:34,143 ♪♪ 73 00:04:34,186 --> 00:04:36,537 We're already pretty jaded about the idea 74 00:04:36,580 --> 00:04:37,929 that we can talk to our phone, 75 00:04:37,973 --> 00:04:40,062 and it mostly understands us. 76 00:04:40,105 --> 00:04:42,456 Woman: I found quite a number of action films. 77 00:04:42,499 --> 00:04:44,327 Five years ago -- no way. 78 00:04:44,371 --> 00:04:47,678 Markoff: Robotics. Machines that see and speak... 79 00:04:47,722 --> 00:04:48,897 Woman: Hi, there....and listen. 80 00:04:48,940 --> 00:04:50,202 All that's real now. 81 00:04:50,246 --> 00:04:51,639 And these technologies 82 00:04:51,682 --> 00:04:55,686 are gonna fundamentally change our society. 83 00:04:55,730 --> 00:05:00,212 Thrun: Now we have this great movement of self-driving cars. 84 00:05:00,256 --> 00:05:01,953 Driving a car autonomously 85 00:05:01,997 --> 00:05:06,088 can move people's lives into a better place. 86 00:05:06,131 --> 00:05:07,916 Horvitz: I've lost a number of family members, 87 00:05:07,959 --> 00:05:09,570 including my mother, 88 00:05:09,613 --> 00:05:11,876 my brother and sister-in-law and their kids, 89 00:05:11,920 --> 00:05:14,009 to automobile accidents. 90 00:05:14,052 --> 00:05:18,405 It's pretty clear we could almost eliminate car accidents 91 00:05:18,448 --> 00:05:20,102 with automation. 92 00:05:20,145 --> 00:05:21,843 30,000 lives in the U.S. alone. 93 00:05:21,886 --> 00:05:25,455 About a million around the world per year. 94 00:05:25,499 --> 00:05:27,501 Ferrucci: In healthcare, early indicators 95 00:05:27,544 --> 00:05:29,503 are the name of the game in that space, 96 00:05:29,546 --> 00:05:33,158 so that's another place where it can save somebody's life. 97 00:05:33,202 --> 00:05:35,726 Dr. Herman: Here in the breast-cancer center, 98 00:05:35,770 --> 00:05:38,381 all the things that the radiologist's brain 99 00:05:38,425 --> 00:05:43,386 does in two minutes, the computer does instantaneously. 100 00:05:43,430 --> 00:05:47,303 The computer has looked at 1 billion mammograms, 101 00:05:47,347 --> 00:05:49,261 and it takes that data and applies it 102 00:05:49,305 --> 00:05:51,438 to this image instantaneously, 103 00:05:51,481 --> 00:05:54,441 so the medical application is profound. 104 00:05:56,399 --> 00:05:57,705 Zilis: Another really exciting area 105 00:05:57,748 --> 00:05:59,402 that we're seeing a lot of development in 106 00:05:59,446 --> 00:06:03,275 is actually understanding our genetic code 107 00:06:03,319 --> 00:06:06,104 and using that to both diagnose disease 108 00:06:06,148 --> 00:06:07,758 and create personalized treatments. 109 00:06:07,802 --> 00:06:11,632 ♪♪ 110 00:06:11,675 --> 00:06:14,112 Kurzweil: The primary application of all these machines 111 00:06:14,156 --> 00:06:17,246 will be to extend our own intelligence. 112 00:06:17,289 --> 00:06:19,422 We'll be able to make ourselves smarter, 113 00:06:19,466 --> 00:06:22,643 and we'll be better at solving problems. 114 00:06:22,686 --> 00:06:24,775 We don't have to age. We'll actually understand aging. 115 00:06:24,819 --> 00:06:27,125 We'll be able to stop it. 116 00:06:27,169 --> 00:06:29,519 Man: There's really no limit to what intelligent machines 117 00:06:29,563 --> 00:06:30,868 can do for the human race. 118 00:06:30,912 --> 00:06:36,265 ♪♪ 119 00:06:36,308 --> 00:06:39,399 How could a smarter machine not be a better machine? 120 00:06:42,053 --> 00:06:44,708 It's hard to say exactly when I began to think 121 00:06:44,752 --> 00:06:46,971 that that was a bit naive. 122 00:06:47,015 --> 00:06:56,459 ♪♪ 123 00:06:56,503 --> 00:06:59,288 Stuart Russell, he's basically a god 124 00:06:59,331 --> 00:07:00,898 in the field of artificial intelligence. 125 00:07:00,942 --> 00:07:04,380 He wrote the book that almost every university uses. 126 00:07:04,424 --> 00:07:06,948 Russell: I used to say it's the best-selling AI textbook. 127 00:07:06,991 --> 00:07:10,255 Now I just say "It's the PDF that's stolen most often." 128 00:07:10,299 --> 00:07:13,650 ♪♪ 129 00:07:13,694 --> 00:07:17,306 Artificial intelligence is about making computers smart, 130 00:07:17,349 --> 00:07:19,830 and from the point of view of the public, 131 00:07:19,874 --> 00:07:21,484 what counts as AI is just something 132 00:07:21,528 --> 00:07:23,268 that's surprisingly intelligent 133 00:07:23,312 --> 00:07:25,488 compared to what we thought computers 134 00:07:25,532 --> 00:07:28,404 would typically be able to do. 135 00:07:28,448 --> 00:07:33,801 AI is a field of research to try to basically simulate 136 00:07:33,844 --> 00:07:36,717 all kinds of human capabilities. 137 00:07:36,760 --> 00:07:38,719 We're in the AI era. 138 00:07:38,762 --> 00:07:40,503 Silicon Valley has the ability to focus 139 00:07:40,547 --> 00:07:42,462 on one bright, shiny thing. 140 00:07:42,505 --> 00:07:43,767 It was social networking 141 00:07:43,811 --> 00:07:45,508 and social media over the last decade, 142 00:07:45,552 --> 00:07:48,119 and it's pretty clear that the bit has flipped. 143 00:07:48,163 --> 00:07:50,557 And it starts with machine learning. 144 00:07:50,600 --> 00:07:54,343 Nolan: When we look back at this moment, what was the first AI? 145 00:07:54,386 --> 00:07:55,736 It's not sexy, and it isn't the thing 146 00:07:55,779 --> 00:07:57,389 we could see at the movies, 147 00:07:57,433 --> 00:08:00,741 but you'd make a great case that Google created, 148 00:08:00,784 --> 00:08:03,395 not a search engine, but a godhead. 149 00:08:03,439 --> 00:08:06,486 A way for people to ask any question they wanted 150 00:08:06,529 --> 00:08:08,270 and get the answer they needed. 151 00:08:08,313 --> 00:08:11,273 Russell: Most people are not aware that what Google is doing 152 00:08:11,316 --> 00:08:13,710 is actually a form of artificial intelligence. 153 00:08:13,754 --> 00:08:16,234 They just go there, they type in a thing. 154 00:08:16,278 --> 00:08:18,323 Google gives them the answer. 155 00:08:18,367 --> 00:08:21,544 Musk: With each search, we train it to be better. 156 00:08:21,588 --> 00:08:23,851 Sometimes we're typing a search, and it tell us the answer 157 00:08:23,894 --> 00:08:27,419 before you've finished asking the question. 158 00:08:27,463 --> 00:08:29,944 You know, who is the president of Kazakhstan? 159 00:08:29,987 --> 00:08:31,685 And it'll just tell you. 160 00:08:31,728 --> 00:08:33,600 You don't have to go to the Kazakhstan national website 161 00:08:33,643 --> 00:08:34,818 to find out. 162 00:08:34,862 --> 00:08:37,081 You didn't used to be able to do that. 163 00:08:37,125 --> 00:08:39,475 Nolan: That is artificial intelligence. 164 00:08:39,519 --> 00:08:42,783 Years from now when we try to understand, we will say, 165 00:08:42,826 --> 00:08:44,567 "How did we miss it?" 166 00:08:44,611 --> 00:08:47,527 Markoff: It's one of the striking contradictions 167 00:08:47,570 --> 00:08:48,484 that we're facing. 168 00:08:48,528 --> 00:08:50,051 Google and Facebook, et al, 169 00:08:50,094 --> 00:08:52,053 have built businesses on giving us, 170 00:08:52,096 --> 00:08:54,185 as a society, free stuff. 171 00:08:54,229 --> 00:08:56,013 But it's a Faustian bargain. 172 00:08:56,057 --> 00:09:00,017 They're extracting something from us in exchange, 173 00:09:00,061 --> 00:09:01,628 but we don't know 174 00:09:01,671 --> 00:09:03,760 what code is running on the other side and why. 175 00:09:03,804 --> 00:09:06,546 We have no idea. 176 00:09:06,589 --> 00:09:08,591 It does strike right at the issue 177 00:09:08,635 --> 00:09:11,028 of how much we should trust these machines. 178 00:09:14,162 --> 00:09:18,166 I use computers literally for everything. 179 00:09:18,209 --> 00:09:21,386 There are so many computer advancements now, 180 00:09:21,430 --> 00:09:23,824 and it's become such a big part of our lives. 181 00:09:23,867 --> 00:09:26,174 It's just incredible what a computer can do. 182 00:09:26,217 --> 00:09:29,090 You can actually carry a computer in your purse. 183 00:09:29,133 --> 00:09:31,571 I mean, how awesome is that? 184 00:09:31,614 --> 00:09:35,052 I think most technology is meant to make things easier 185 00:09:35,096 --> 00:09:37,315 and simpler for all of us, 186 00:09:37,359 --> 00:09:40,362 so hopefully that just remains the focus. 187 00:09:40,405 --> 00:09:43,147 I think everybody loves their computers. 188 00:09:44,409 --> 00:09:51,678 ♪♪ 189 00:09:51,721 --> 00:09:53,810 People don't realize they are constantly 190 00:09:53,854 --> 00:09:59,076 being negotiated with by machines, 191 00:09:59,120 --> 00:10:02,993 whether that's the price of products in your Amazon cart, 192 00:10:03,037 --> 00:10:05,517 whether you can get on a particular flight, 193 00:10:05,561 --> 00:10:08,912 whether you can reserve a room at a particular hotel. 194 00:10:08,956 --> 00:10:11,959 What you're experiencing are machine-learning algorithms 195 00:10:12,002 --> 00:10:14,265 that have determined that a person like you 196 00:10:14,309 --> 00:10:15,919 is willing to pay 2 cents more 197 00:10:15,963 --> 00:10:17,791 and is changing the price. 198 00:10:17,834 --> 00:10:21,751 ♪♪ 199 00:10:21,795 --> 00:10:24,014 Kosinski: Now, a computer looks at millions of people 200 00:10:24,058 --> 00:10:28,105 simultaneously for very subtle patterns. 201 00:10:28,149 --> 00:10:31,369 You can take seemingly innocent digital footprints, 202 00:10:31,413 --> 00:10:34,677 such as someone's playlist on Spotify, 203 00:10:34,721 --> 00:10:37,201 or stuff that they bought on Amazon, 204 00:10:37,245 --> 00:10:40,291 and then use algorithms to translate this 205 00:10:40,335 --> 00:10:44,513 into a very detailed and a very accurate, intimate profile. 206 00:10:47,603 --> 00:10:50,911 Kaplan: There is a dossier on each of us that is so extensive 207 00:10:50,954 --> 00:10:52,695 it would be possibly accurate to say 208 00:10:52,739 --> 00:10:55,698 that they know more about you than your mother does. 209 00:10:55,742 --> 00:11:04,054 ♪♪ 210 00:11:04,098 --> 00:11:06,883 Tegmark: The major cause of the recent AI breakthrough 211 00:11:06,927 --> 00:11:08,580 isn't just that some dude 212 00:11:08,624 --> 00:11:11,583 had a brilliant insight all of a sudden, 213 00:11:11,627 --> 00:11:14,325 but simply that we have much bigger data 214 00:11:14,369 --> 00:11:18,242 to train them on and vastly better computers. 215 00:11:18,286 --> 00:11:19,940 el Kaliouby: The magic is in the data. 216 00:11:19,983 --> 00:11:21,463 It's a ton of data. 217 00:11:21,506 --> 00:11:23,726 I mean, it's data that's never existed before. 218 00:11:23,770 --> 00:11:26,686 We've never had this data before. 219 00:11:26,729 --> 00:11:30,733 We've created technologies that allow us to capture 220 00:11:30,777 --> 00:11:33,040 vast amounts of information. 221 00:11:33,083 --> 00:11:35,738 If you think of a billion cellphones on the planet 222 00:11:35,782 --> 00:11:38,393 with gyroscopes and accelerometers 223 00:11:38,436 --> 00:11:39,786 and fingerprint readers... 224 00:11:39,829 --> 00:11:42,005 couple that with the GPS and the photos they take 225 00:11:42,049 --> 00:11:43,964 and the tweets that you send, 226 00:11:44,007 --> 00:11:47,750 we're all giving off huge amounts of data individually. 227 00:11:47,794 --> 00:11:50,274 Cars that drive as the cameras on them suck up information 228 00:11:50,318 --> 00:11:52,059 about the world around them. 229 00:11:52,102 --> 00:11:54,844 The satellites that are now in orbit the size of a toaster. 230 00:11:54,888 --> 00:11:57,629 The infrared about the vegetation on the planet. 231 00:11:57,673 --> 00:11:59,109 The buoys that are out in the oceans 232 00:11:59,153 --> 00:12:01,024 to feed into the climate models. 233 00:12:01,068 --> 00:12:05,028 ♪♪ 234 00:12:05,072 --> 00:12:08,902 And the NSA, the CIA, as they collect information 235 00:12:08,945 --> 00:12:12,644 about the geopolitical situations. 236 00:12:12,688 --> 00:12:15,604 The world today is literally swimming in this data. 237 00:12:15,647 --> 00:12:20,565 ♪♪ 238 00:12:20,609 --> 00:12:22,480 Kosinski: Back in 2012, 239 00:12:22,524 --> 00:12:25,875 IBM estimated that an average human being 240 00:12:25,919 --> 00:12:31,098 leaves 500 megabytes of digital footprints every day. 241 00:12:31,141 --> 00:12:34,841 If you wanted to back up on the one day worth of data 242 00:12:34,884 --> 00:12:36,494 that humanity produces 243 00:12:36,538 --> 00:12:39,062 and imprint it out on a letter-sized paper, 244 00:12:39,106 --> 00:12:43,806 double-sided, font size 12, and you stack it up, 245 00:12:43,850 --> 00:12:46,113 it would reach from the surface of the Earth 246 00:12:46,156 --> 00:12:49,116 to the sun four times over. 247 00:12:49,159 --> 00:12:51,292 That's every day. 248 00:12:51,335 --> 00:12:53,816 Kaplan: The data itself is not good or evil. 249 00:12:53,860 --> 00:12:55,470 It's how it's used. 250 00:12:55,513 --> 00:12:58,342 We're relying, really, on the goodwill of these people 251 00:12:58,386 --> 00:13:01,171 and on the policies of these companies. 252 00:13:01,215 --> 00:13:03,870 There is no legal requirement for how they can 253 00:13:03,913 --> 00:13:06,307 and should use that kind of data. 254 00:13:06,350 --> 00:13:09,266 That, to me, is at the heart of the trust issue. 255 00:13:11,007 --> 00:13:13,793 Barrat: Right now there's a giant race for creating machines 256 00:13:13,836 --> 00:13:15,751 that are as smart as humans. 257 00:13:15,795 --> 00:13:17,971 Google -- They're working on what's really the kind of 258 00:13:18,014 --> 00:13:20,016 Manhattan Project of artificial intelligence. 259 00:13:20,060 --> 00:13:22,671 They've got the most money. They've got the most talent. 260 00:13:22,714 --> 00:13:27,067 They're buying up AI companies and robotics companies. 261 00:13:27,110 --> 00:13:29,069 Urban: People still think of Google as a search engine 262 00:13:29,112 --> 00:13:30,722 and their e-mail provider 263 00:13:30,766 --> 00:13:33,943 and a lot of other things that we use on a daily basis, 264 00:13:33,987 --> 00:13:39,383 but behind that search box are 10 million servers. 265 00:13:39,427 --> 00:13:42,299 That makes Google the most powerful computing platform 266 00:13:42,343 --> 00:13:43,910 in the world. 267 00:13:43,953 --> 00:13:47,217 Google is now working on an AI computing platform 268 00:13:47,261 --> 00:13:50,133 that will have 100 million servers. 269 00:13:52,179 --> 00:13:53,963 So when you're interacting with Google, 270 00:13:54,007 --> 00:13:56,052 we're just seeing the toenail of something 271 00:13:56,096 --> 00:13:58,881 that is a giant beast in the making. 272 00:13:58,925 --> 00:14:00,622 And the truth is, I'm not even sure 273 00:14:00,665 --> 00:14:02,798 that Google knows what it's becoming. 274 00:14:02,842 --> 00:14:11,502 ♪♪ 275 00:14:11,546 --> 00:14:14,114 Phoenix: If you look inside of what algorithms are being used 276 00:14:14,157 --> 00:14:15,811 at Google, 277 00:14:15,855 --> 00:14:20,076 it's technology largely from the '80s. 278 00:14:20,120 --> 00:14:23,863 So these are models that you train by showing them a 1, a 2, 279 00:14:23,906 --> 00:14:27,344 and a 3, and it learns not what a 1 is or what a 2 is -- 280 00:14:27,388 --> 00:14:30,434 It learns what the difference between a 1 and a 2 is. 281 00:14:30,478 --> 00:14:32,436 It's just a computation. 282 00:14:32,480 --> 00:14:35,396 In the last half decade, where we've made this rapid progress, 283 00:14:35,439 --> 00:14:38,268 it has all been in pattern recognition. 284 00:14:38,312 --> 00:14:41,184 Tegmark: Most of the good, old-fashioned AI 285 00:14:41,228 --> 00:14:44,057 was when we would tell our computers 286 00:14:44,100 --> 00:14:46,798 how to play a game like chess... 287 00:14:46,842 --> 00:14:49,584 from the old paradigm where you just tell the computer 288 00:14:49,627 --> 00:14:52,195 exactly what to do. 289 00:14:54,502 --> 00:14:57,505 Announcer: This is "Jeopardy!" 290 00:14:57,548 --> 00:14:59,376 ♪♪ 291 00:14:59,420 --> 00:15:02,510 "The IBM Challenge"! 292 00:15:02,553 --> 00:15:05,730 Ferrucci: No one at the time had thought that a machine 293 00:15:05,774 --> 00:15:08,298 could have the precision and the confidence 294 00:15:08,342 --> 00:15:09,952 and the speed to play "Jeopardy!" 295 00:15:09,996 --> 00:15:11,475 well enough against the best humans. 296 00:15:11,519 --> 00:15:14,609 Let's play "Jeopardy!" 297 00:15:18,569 --> 00:15:20,354 Watson.Watson: What is "shoe"? 298 00:15:20,397 --> 00:15:21,877 You are right. You get to pick. 299 00:15:21,921 --> 00:15:24,836 Literary Character APB for $800. 300 00:15:24,880 --> 00:15:28,014 Answer -- the Daily Double. 301 00:15:28,057 --> 00:15:31,539 Watson actually got its knowledge by reading Wikipedia 302 00:15:31,582 --> 00:15:34,672 and 200 million pages of natural-language documents. 303 00:15:34,716 --> 00:15:36,674 Ferrucci: You can't program every line 304 00:15:36,718 --> 00:15:38,502 of how the world works. 305 00:15:38,546 --> 00:15:40,722 The machine has to learn by reading. 306 00:15:40,765 --> 00:15:42,202 Now we come to Watson. 307 00:15:42,245 --> 00:15:43,986 "Who is Bram Stoker?" 308 00:15:44,030 --> 00:15:45,988 And the wager? 309 00:15:46,032 --> 00:15:49,165 Hello! $17,973. 310 00:15:49,209 --> 00:15:50,993 $41,413. 311 00:15:51,037 --> 00:15:53,343 And a two-day total of $77-- 312 00:15:53,387 --> 00:15:56,694 Phoenix: Watson's trained on huge amounts of text, 313 00:15:56,738 --> 00:15:59,828 but it's not like it understands what it's saying. 314 00:15:59,871 --> 00:16:02,309 It doesn't know that water makes things wet by touching water 315 00:16:02,352 --> 00:16:04,441 and by seeing the way things behave in the world 316 00:16:04,485 --> 00:16:06,182 the way you and I do. 317 00:16:06,226 --> 00:16:10,143 A lot of language AI today is not building logical models 318 00:16:10,186 --> 00:16:11,622 of how the world works. 319 00:16:11,666 --> 00:16:15,365 Rather, it's looking at how the words appear 320 00:16:15,409 --> 00:16:18,238 in the context of other words. 321 00:16:18,281 --> 00:16:20,196 Barrat: David Ferrucci developed IBM's Watson, 322 00:16:20,240 --> 00:16:23,547 and somebody asked him, "Does Watson think?" 323 00:16:23,591 --> 00:16:27,160 And he said, "Does a submarine swim?" 324 00:16:27,203 --> 00:16:29,031 And what they meant was, when they developed submarines, 325 00:16:29,075 --> 00:16:32,992 they borrowed basic principles of swimming from fish. 326 00:16:33,035 --> 00:16:35,037 But a submarine swims farther and faster than fish 327 00:16:35,081 --> 00:16:36,125 and can carry a huge payload. 328 00:16:36,169 --> 00:16:39,911 It out-swims fish. 329 00:16:39,955 --> 00:16:41,870 Ng: Watson winning the game of "Jeopardy!" 330 00:16:41,913 --> 00:16:43,741 will go down in the history of AI 331 00:16:43,785 --> 00:16:46,570 as a significant milestone. 332 00:16:46,614 --> 00:16:49,269 We tend to be amazed when the machine does so well. 333 00:16:49,312 --> 00:16:52,663 I'm even more amazed when the computer beats humans at things 334 00:16:52,707 --> 00:16:55,188 that humans are naturally good at. 335 00:16:55,231 --> 00:16:58,060 This is how we make progress. 336 00:16:58,104 --> 00:17:00,671 In the early days of the Google Brain project, 337 00:17:00,715 --> 00:17:02,804 I gave the team a very simple instruction, 338 00:17:02,847 --> 00:17:05,807 which was, "Build the biggest neural network possible, 339 00:17:05,850 --> 00:17:08,157 like 1,000 computers." 340 00:17:08,201 --> 00:17:09,724 Musk: A neural net is something very close 341 00:17:09,767 --> 00:17:12,161 to a simulation of how the brain works. 342 00:17:12,205 --> 00:17:16,818 It's very probabilistic, but with contextual relevance. 343 00:17:16,861 --> 00:17:18,298 Urban: In your brain, you have long neurons 344 00:17:18,341 --> 00:17:20,256 that connect to thousands of other neurons, 345 00:17:20,300 --> 00:17:22,519 and you have these pathways that are formed and forged 346 00:17:22,563 --> 00:17:24,739 based on what the brain needs to do. 347 00:17:24,782 --> 00:17:28,960 When a baby tries something and it succeeds, there's a reward, 348 00:17:29,004 --> 00:17:32,312 and that pathway that created the success is strengthened. 349 00:17:32,355 --> 00:17:34,662 If it fails at something, the pathway is weakened, 350 00:17:34,705 --> 00:17:36,794 and so, over time, the brain becomes honed 351 00:17:36,838 --> 00:17:40,320 to be good at the environment around it. 352 00:17:40,363 --> 00:17:43,279 Ng: Really, it's just getting machines to learn by themselves. 353 00:17:43,323 --> 00:17:45,238 This is called "deep learning," and "deep learning" 354 00:17:45,281 --> 00:17:48,676 and "neural networks" mean roughly the same thing. 355 00:17:48,719 --> 00:17:52,375 Tegmark: Deep learning is a totally different approach 356 00:17:52,419 --> 00:17:55,161 where the computer learns more like a toddler, 357 00:17:55,204 --> 00:17:56,466 by just getting a lot of data 358 00:17:56,510 --> 00:18:00,340 and eventually figuring stuff out. 359 00:18:00,383 --> 00:18:03,125 The computer just gets smarter and smarter 360 00:18:03,169 --> 00:18:05,997 as it has more experiences. 361 00:18:06,041 --> 00:18:08,130 Ng: So, imagine, if you will, a neural network, you know, 362 00:18:08,174 --> 00:18:09,697 like 1,000 computers. 363 00:18:09,740 --> 00:18:11,438 And it wakes up not knowing anything. 364 00:18:11,481 --> 00:18:14,093 And we made it watch YouTube for a week. 365 00:18:14,136 --> 00:18:16,704 ♪♪ 366 00:18:18,706 --> 00:18:20,360 ♪ Oppan Gangnam style 367 00:18:20,403 --> 00:18:23,189 Ow! 368 00:18:25,408 --> 00:18:28,194 Charlie! That really hurt! 369 00:18:28,237 --> 00:18:30,152 ♪♪ 370 00:18:30,196 --> 00:18:31,327 ♪ Gangnam style 371 00:18:31,371 --> 00:18:33,286 ♪ Op, op, op, op 372 00:18:33,329 --> 00:18:36,202 ♪ Oppan Gangnam style 373 00:18:36,245 --> 00:18:38,508 Ng: And so, after watching YouTube for a week, 374 00:18:38,552 --> 00:18:39,988 what would it learn? 375 00:18:40,031 --> 00:18:41,903 We had a hypothesis that it would learn to detect 376 00:18:41,946 --> 00:18:44,384 commonly occurring objects in videos. 377 00:18:44,427 --> 00:18:47,517 And so, we know that human faces appear a lot in videos, 378 00:18:47,561 --> 00:18:49,302 so we looked, and, lo and behold, 379 00:18:49,345 --> 00:18:51,608 there was a neuron that had learned to detect human faces. 380 00:18:51,652 --> 00:18:56,265 Leave Britney alone! 381 00:18:56,309 --> 00:18:58,354 Well, what else appears in videos a lot? 382 00:19:00,095 --> 00:19:01,792 So, we looked, and to our surprise, 383 00:19:01,836 --> 00:19:04,882 there was actually a neuron that had learned to detect cats. 384 00:19:04,926 --> 00:19:14,849 ♪♪ 385 00:19:14,892 --> 00:19:17,068 I still remember seeing recognition. 386 00:19:17,112 --> 00:19:18,635 "Wow, that's a cat. Okay, cool. 387 00:19:18,679 --> 00:19:20,071 Great." 388 00:19:23,162 --> 00:19:24,859 Barrat: It's all pretty innocuous 389 00:19:24,902 --> 00:19:26,295 when you're thinking about the future. 390 00:19:26,339 --> 00:19:29,733 It all seems kind of harmless and benign. 391 00:19:29,777 --> 00:19:31,605 But we're making cognitive architectures 392 00:19:31,648 --> 00:19:33,520 that will fly farther and faster than us 393 00:19:33,563 --> 00:19:35,086 and carry a bigger payload, 394 00:19:35,130 --> 00:19:37,437 and they won't be warm and fuzzy. 395 00:19:37,480 --> 00:19:39,656 Ferrucci: I think that, in three to five years, 396 00:19:39,700 --> 00:19:41,702 you will see a computer system 397 00:19:41,745 --> 00:19:45,401 that will be able to autonomously learn 398 00:19:45,445 --> 00:19:49,013 how to understand, how to build understanding, 399 00:19:49,057 --> 00:19:51,364 not unlike the way the human mind works. 400 00:19:53,931 --> 00:19:56,891 Whatever that lunch was, it was certainly delicious. 401 00:19:56,934 --> 00:19:59,807 Simply some of Robby's synthetics. 402 00:19:59,850 --> 00:20:01,635 He's your cook, too? 403 00:20:01,678 --> 00:20:04,551 Even manufactures the raw materials. 404 00:20:04,594 --> 00:20:06,944 Come around here, Robby. 405 00:20:06,988 --> 00:20:09,773 I'll show you how this works. 406 00:20:11,122 --> 00:20:13,342 One introduces a sample of human food 407 00:20:13,386 --> 00:20:15,344 through this aperture. 408 00:20:15,388 --> 00:20:17,738 Down here there's a small built-in chemical laboratory, 409 00:20:17,781 --> 00:20:19,218 where he analyzes it. 410 00:20:19,261 --> 00:20:21,263 Later, he can reproduce identical molecules 411 00:20:21,307 --> 00:20:22,482 in any shape or quantity. 412 00:20:22,525 --> 00:20:24,614 Why, it's a housewife's dream. 413 00:20:24,658 --> 00:20:26,834 Announcer: Meet Baxter, 414 00:20:26,877 --> 00:20:29,445 revolutionary new category of robots, 415 00:20:29,489 --> 00:20:30,490 with common sense. 416 00:20:30,533 --> 00:20:31,839 Baxter... 417 00:20:31,882 --> 00:20:33,449 Barrat: Baxter is a really good example 418 00:20:33,493 --> 00:20:36,887 of the kind of competition we face from machines. 419 00:20:36,931 --> 00:20:42,676 Baxter can do almost anything we can do with our hands. 420 00:20:42,719 --> 00:20:45,722 Baxter costs about what a minimum-wage worker 421 00:20:45,766 --> 00:20:47,507 makes in a year. 422 00:20:47,550 --> 00:20:48,769 But Baxter won't be taking the place 423 00:20:48,812 --> 00:20:50,118 of one minimum-wage worker -- 424 00:20:50,161 --> 00:20:51,772 He'll be taking the place of three, 425 00:20:51,815 --> 00:20:55,515 because they never get tired, they never take breaks. 426 00:20:55,558 --> 00:20:57,865 Gourley: That's probably the first thing we're gonna see -- 427 00:20:57,908 --> 00:20:59,475 displacement of jobs. 428 00:20:59,519 --> 00:21:01,651 They're gonna be done quicker, faster, cheaper 429 00:21:01,695 --> 00:21:04,088 by machines. 430 00:21:04,132 --> 00:21:07,657 Our ability to even stay current is so insanely limited 431 00:21:07,701 --> 00:21:10,138 compared to the machines we build. 432 00:21:10,181 --> 00:21:13,446 For example, now we have this great movement of Uber and Lyft 433 00:21:13,489 --> 00:21:15,056 kind of making transportation cheaper 434 00:21:15,099 --> 00:21:16,405 and democratizing transportation, 435 00:21:16,449 --> 00:21:17,711 which is great. 436 00:21:17,754 --> 00:21:19,321 The next step is gonna be 437 00:21:19,365 --> 00:21:21,149 that they're all gonna be replaced by driverless cars, 438 00:21:21,192 --> 00:21:22,411 and then all the Uber and Lyft drivers 439 00:21:22,455 --> 00:21:25,936 have to find something new to do. 440 00:21:25,980 --> 00:21:28,156 Barrat: There are 4 million professional drivers 441 00:21:28,199 --> 00:21:29,723 in the United States. 442 00:21:29,766 --> 00:21:31,638 They're unemployed soon. 443 00:21:31,681 --> 00:21:34,075 7 million people that do data entry. 444 00:21:34,118 --> 00:21:37,339 Those people are gonna be jobless. 445 00:21:37,383 --> 00:21:40,342 A job isn't just about money, right? 446 00:21:40,386 --> 00:21:42,605 On a biological level, it serves a purpose. 447 00:21:42,649 --> 00:21:45,391 It becomes a defining thing. 448 00:21:45,434 --> 00:21:48,350 When the jobs went away in any given civilization, 449 00:21:48,394 --> 00:21:50,787 it doesn't take long until that turns into violence. 450 00:21:53,355 --> 00:21:57,011 ♪♪ 451 00:21:59,622 --> 00:22:02,016 We face a giant divide between rich and poor, 452 00:22:02,059 --> 00:22:05,019 because that's what automation and AI will provoke -- 453 00:22:05,062 --> 00:22:08,588 a greater divide between the haves and the have-nots. 454 00:22:08,631 --> 00:22:10,807 Right now, it's working into the middle class, 455 00:22:10,851 --> 00:22:12,896 into white-collar jobs. 456 00:22:12,940 --> 00:22:15,334 IBM's Watson does business analytics 457 00:22:15,377 --> 00:22:20,600 that we used to pay a business analyst $300 an hour to do. 458 00:22:20,643 --> 00:22:23,037 Gourley: Today, you're going to college to be a doctor, 459 00:22:23,080 --> 00:22:25,082 to be an accountant, to be a journalist. 460 00:22:25,126 --> 00:22:28,608 It's unclear that there's gonna be jobs there for you. 461 00:22:28,651 --> 00:22:32,612 Ng: If someone's planning for a 40-year career in radiology, 462 00:22:32,655 --> 00:22:34,222 just reading images, 463 00:22:34,265 --> 00:22:35,745 I think that could be a challenge 464 00:22:35,789 --> 00:22:36,920 to the new graduates of today. 465 00:22:39,270 --> 00:22:49,193 ♪♪ 466 00:22:50,847 --> 00:22:58,464 ♪♪ 467 00:22:58,507 --> 00:23:02,729 Dr. Herman: The da Vinci robot is currently utilized 468 00:23:02,772 --> 00:23:07,516 by a variety of surgeons for its accuracy and its ability 469 00:23:07,560 --> 00:23:12,303 to avoid the inevitable fluctuations of the human hand. 470 00:23:12,347 --> 00:23:17,787 ♪♪ 471 00:23:17,831 --> 00:23:23,358 ♪♪ 472 00:23:23,402 --> 00:23:28,494 Anybody who watches this feels the amazingness of it. 473 00:23:30,931 --> 00:23:34,674 You look through the scope, and you're seeing the claw hand 474 00:23:34,717 --> 00:23:36,893 holding that woman's ovary. 475 00:23:36,937 --> 00:23:42,638 Humanity was resting right here in the hands of this robot. 476 00:23:42,682 --> 00:23:46,947 People say it's the future, but it's not the future -- 477 00:23:46,990 --> 00:23:50,516 It's the present. 478 00:23:50,559 --> 00:23:52,474 Zilis: If you think about a surgical robot, 479 00:23:52,518 --> 00:23:54,737 there's often not a lot of intelligence in these things, 480 00:23:54,781 --> 00:23:56,783 but over time, as we put more and more intelligence 481 00:23:56,826 --> 00:23:58,567 into these systems, 482 00:23:58,611 --> 00:24:02,441 the surgical robots can actually learn from each robot surgery. 483 00:24:02,484 --> 00:24:04,181 They're tracking the movements, they're understanding 484 00:24:04,225 --> 00:24:05,966 what worked and what didn't work. 485 00:24:06,009 --> 00:24:08,708 And eventually, the robot for routine surgeries 486 00:24:08,751 --> 00:24:12,320 is going to be able to perform that entirely by itself... 487 00:24:12,363 --> 00:24:13,756 or with human supervision. 488 00:24:32,558 --> 00:24:34,995 ♪♪ 489 00:24:35,038 --> 00:24:37,214 Dr. Herman: It seems that we're feeding it and creating it, 490 00:24:37,258 --> 00:24:42,785 but, in a way, we are a slave to the technology, 491 00:24:42,829 --> 00:24:45,701 because we can't go back. 492 00:24:50,053 --> 00:24:52,882 Gourley: The machines are taking bigger and bigger bites 493 00:24:52,926 --> 00:24:57,147 out of our skill set at an ever-increasing speed. 494 00:24:57,191 --> 00:24:59,236 And so we've got to run faster and faster 495 00:24:59,280 --> 00:25:00,890 to keep ahead of the machines. 496 00:25:02,675 --> 00:25:04,677 How do I look? 497 00:25:04,720 --> 00:25:06,374 Good. 498 00:25:10,030 --> 00:25:11,553 Are you attracted to me? 499 00:25:11,597 --> 00:25:14,251 What?Are you attracted to me? 500 00:25:14,295 --> 00:25:17,777 You give me indications that you are. 501 00:25:17,820 --> 00:25:20,562 I do? Yes. 502 00:25:20,606 --> 00:25:22,608 Nolan: This is the future we're headed into. 503 00:25:22,651 --> 00:25:26,046 We want to design our companions. 504 00:25:26,089 --> 00:25:29,266 We're gonna like to see a human face on AI. 505 00:25:29,310 --> 00:25:33,967 Therefore, gaming our emotions will be depressingly easy. 506 00:25:34,010 --> 00:25:35,272 We're not that complicated. 507 00:25:35,316 --> 00:25:38,101 We're simple. Stimulus-response. 508 00:25:38,145 --> 00:25:43,063 I can make you like me basically by smiling at you a lot. 509 00:25:43,106 --> 00:25:45,674 AIs are gonna be fantastic at manipulating us. 510 00:25:45,718 --> 00:25:54,640 ♪♪ 511 00:25:54,683 --> 00:25:56,946 So, you've developed a technology 512 00:25:56,990 --> 00:26:00,036 that can sense what people are feeling. 513 00:26:00,080 --> 00:26:01,472 Right. We've developed technology 514 00:26:01,516 --> 00:26:03,387 that can read your facial expressions 515 00:26:03,431 --> 00:26:06,521 and map that to a number of emotional states. 516 00:26:06,565 --> 00:26:08,697 el Kaliouby: 15 years ago, I had just finished 517 00:26:08,741 --> 00:26:11,482 my undergraduate studies in computer science, 518 00:26:11,526 --> 00:26:15,008 and it struck me that I was spending a lot of time 519 00:26:15,051 --> 00:26:17,793 interacting with my laptops and my devices, 520 00:26:17,837 --> 00:26:23,582 yet these devices had absolutely no clue how I was feeling. 521 00:26:23,625 --> 00:26:26,802 I started thinking, "What if this device could sense 522 00:26:26,846 --> 00:26:29,326 that I was stressed or I was having a bad day? 523 00:26:29,370 --> 00:26:31,067 What would that open up?" 524 00:26:32,721 --> 00:26:34,418 Hi, first-graders! 525 00:26:34,462 --> 00:26:35,855 How are you? 526 00:26:35,898 --> 00:26:37,813 Can I get a hug? 527 00:26:37,857 --> 00:26:40,773 We had kids interact with the technology. 528 00:26:40,816 --> 00:26:42,862 A lot of it is still in development, 529 00:26:42,905 --> 00:26:44,472 but it was just amazing. 530 00:26:44,515 --> 00:26:46,648 Who likes robots? Me! 531 00:26:46,692 --> 00:26:48,911 Who wants to have a robot in their house? 532 00:26:48,955 --> 00:26:51,479 What would you use a robot for, Jack? 533 00:26:51,522 --> 00:26:56,353 I would use it to ask my mom very hard math questions. 534 00:26:56,397 --> 00:26:58,181 Okay. What about you, Theo? 535 00:26:58,225 --> 00:27:02,272 I would use it for scaring people. 536 00:27:02,316 --> 00:27:04,666 All right. So, start by smiling. 537 00:27:04,710 --> 00:27:06,625 Nice. 538 00:27:06,668 --> 00:27:09,018 Brow furrow. 539 00:27:09,062 --> 00:27:10,890 Nice one. Eyebrow raise. 540 00:27:10,933 --> 00:27:12,587 This generation, technology 541 00:27:12,631 --> 00:27:15,068 is just surrounding them all the time. 542 00:27:15,111 --> 00:27:17,853 It's almost like they expect to have robots in their homes, 543 00:27:17,897 --> 00:27:22,336 and they expect these robots to be socially intelligent. 544 00:27:22,379 --> 00:27:25,252 What makes robots smart? 545 00:27:25,295 --> 00:27:29,648 Put them in, like, a math or biology class. 546 00:27:29,691 --> 00:27:32,259 I think you would have to train it. 547 00:27:32,302 --> 00:27:35,218 All right. Let's walk over here. 548 00:27:35,262 --> 00:27:37,394 So, if you smile and you raise your eyebrows, 549 00:27:37,438 --> 00:27:39,005 it's gonna run over to you. 550 00:27:39,048 --> 00:27:40,833 Woman: It's coming over! It's coming over! Look. 551 00:27:43,183 --> 00:27:45,272 But if you look angry, it's gonna run away. 552 00:27:46,534 --> 00:27:48,797 -Awesome! -Oh, that was good. 553 00:27:48,841 --> 00:27:52,366 We're training computers to read and recognize emotions. 554 00:27:52,409 --> 00:27:53,846 Ready? Set? Go! 555 00:27:53,889 --> 00:27:57,414 And the response so far has been really amazing. 556 00:27:57,458 --> 00:27:59,590 People are integrating this into health apps, 557 00:27:59,634 --> 00:28:04,465 meditation apps, robots, cars. 558 00:28:04,508 --> 00:28:06,728 We're gonna see how this unfolds. 559 00:28:06,772 --> 00:28:09,426 ♪♪ 560 00:28:09,470 --> 00:28:11,602 Zilis: Robots can contain AI, 561 00:28:11,646 --> 00:28:14,388 but the robot is just a physical instantiation, 562 00:28:14,431 --> 00:28:16,782 and the artificial intelligence is the brain. 563 00:28:16,825 --> 00:28:19,872 And so brains can exist purely in software-based systems. 564 00:28:19,915 --> 00:28:22,483 They don't need to have a physical form. 565 00:28:22,526 --> 00:28:25,094 Robots can exist without any artificial intelligence. 566 00:28:25,138 --> 00:28:28,097 We have a lot of dumb robots out there. 567 00:28:28,141 --> 00:28:31,753 But a dumb robot can be a smart robot overnight, 568 00:28:31,797 --> 00:28:34,103 given the right software, given the right sensors. 569 00:28:34,147 --> 00:28:38,629 Barrat: We can't help but impute motive into inanimate objects. 570 00:28:38,673 --> 00:28:40,327 We do it with machines. 571 00:28:40,370 --> 00:28:41,502 We'll treat them like children. 572 00:28:41,545 --> 00:28:43,330 We'll treat them like surrogates. 573 00:28:43,373 --> 00:28:45,027 -Goodbye! -Goodbye! 574 00:28:45,071 --> 00:28:48,204 And we'll pay the price. 575 00:28:49,292 --> 00:28:58,998 ♪♪ 576 00:28:59,041 --> 00:29:08,572 ♪♪ 577 00:29:08,616 --> 00:29:10,792 Okay, welcome to ATR. 578 00:29:10,836 --> 00:29:18,060 ♪♪ 579 00:29:25,067 --> 00:29:30,594 ♪♪ 580 00:29:30,638 --> 00:29:36,122 ♪♪ 581 00:29:47,786 --> 00:29:51,485 ♪♪ 582 00:29:51,528 --> 00:29:52,791 Konnichiwa. 583 00:30:24,170 --> 00:30:29,436 ♪♪ 584 00:30:53,677 --> 00:30:56,942 ♪♪ 585 00:30:56,985 --> 00:30:58,682 Gourley: We build artificial intelligence, 586 00:30:58,726 --> 00:31:02,948 and the very first thing we want to do is replicate us. 587 00:31:02,991 --> 00:31:05,341 I think the key point will come 588 00:31:05,385 --> 00:31:09,258 when all the major senses are replicated -- 589 00:31:09,302 --> 00:31:11,130 sight... 590 00:31:11,173 --> 00:31:12,871 touch... 591 00:31:12,914 --> 00:31:14,611 smell. 592 00:31:14,655 --> 00:31:17,919 When we replicate our senses, is that when it become alive? 593 00:31:17,963 --> 00:31:22,010 ♪♪ 594 00:31:24,795 --> 00:31:27,581 ♪♪ 595 00:31:27,624 --> 00:31:29,104 Nolan: So many of our machines 596 00:31:29,148 --> 00:31:31,019 are being built to understand us. 597 00:31:32,847 --> 00:31:34,805 But what happens when an anthropomorphic creature 598 00:31:34,849 --> 00:31:37,417 discovers that they can adjust their loyalty, 599 00:31:37,460 --> 00:31:40,028 adjust their courage, adjust their avarice, 600 00:31:40,072 --> 00:31:42,291 adjust their cunning? 601 00:31:42,335 --> 00:31:44,815 ♪♪ 602 00:31:44,859 --> 00:31:47,166 Musk: The average person, they don't see killer robots 603 00:31:47,209 --> 00:31:48,645 going down the streets. 604 00:31:48,689 --> 00:31:50,996 They're like, "What are you talking about?" 605 00:31:51,039 --> 00:31:53,955 Man, we want to make sure that we don't have killer robots 606 00:31:53,999 --> 00:31:57,045 going down the street. 607 00:31:57,089 --> 00:31:59,439 Once they're going down the street, it is too late. 608 00:31:59,482 --> 00:32:05,010 ♪♪ 609 00:32:05,053 --> 00:32:07,099 Russell: The thing that worries me right now, 610 00:32:07,142 --> 00:32:08,578 that keeps me awake, 611 00:32:08,622 --> 00:32:11,842 is the development of autonomous weapons. 612 00:32:11,886 --> 00:32:19,850 ♪♪ 613 00:32:19,894 --> 00:32:27,771 ♪♪ 614 00:32:27,815 --> 00:32:32,733 Up to now, people have expressed unease about drones, 615 00:32:32,776 --> 00:32:35,127 which are remotely piloted aircraft. 616 00:32:35,170 --> 00:32:39,783 ♪♪ 617 00:32:39,827 --> 00:32:43,309 If you take a drone's camera and feed it into the AI system, 618 00:32:43,352 --> 00:32:47,443 it's a very easy step from here to fully autonomous weapons 619 00:32:47,487 --> 00:32:50,881 that choose their own targets and release their own missiles. 620 00:32:50,925 --> 00:32:58,150 ♪♪ 621 00:32:58,193 --> 00:33:05,374 ♪♪ 622 00:33:05,418 --> 00:33:12,686 ♪♪ 623 00:33:12,729 --> 00:33:15,080 The expected life-span of a human being 624 00:33:15,123 --> 00:33:16,516 in that kind of battle environment 625 00:33:16,559 --> 00:33:20,520 would be measured in seconds. 626 00:33:20,563 --> 00:33:23,740 Singer: At one point, drones were science fiction, 627 00:33:23,784 --> 00:33:28,832 and now they've become the normal thing in war. 628 00:33:28,876 --> 00:33:33,402 There's over 10,000 in U.S. military inventory alone. 629 00:33:33,446 --> 00:33:35,274 But they're not just a U.S. phenomena. 630 00:33:35,317 --> 00:33:39,060 There's more than 80 countries that operate them. 631 00:33:39,104 --> 00:33:41,932 Gourley: It stands to reason that people making some 632 00:33:41,976 --> 00:33:44,587 of the most important and difficult decisions in the world 633 00:33:44,631 --> 00:33:46,328 are gonna start to use and implement 634 00:33:46,372 --> 00:33:48,591 artificial intelligence. 635 00:33:48,635 --> 00:33:50,724 ♪♪ 636 00:33:50,767 --> 00:33:53,596 The Air Force just designed a $400-billion jet program 637 00:33:53,640 --> 00:33:55,555 to put pilots in the sky, 638 00:33:55,598 --> 00:34:01,300 and a $500 AI, designed by a couple of graduate students, 639 00:34:01,343 --> 00:34:03,432 is beating the best human pilots 640 00:34:03,476 --> 00:34:05,782 with a relatively simple algorithm. 641 00:34:05,826 --> 00:34:09,395 ♪♪ 642 00:34:09,438 --> 00:34:13,399 AI will have as big an impact on the military 643 00:34:13,442 --> 00:34:17,490 as the combustion engine had at the turn of the century. 644 00:34:17,533 --> 00:34:18,839 It will literally touch 645 00:34:18,882 --> 00:34:21,233 everything that the military does, 646 00:34:21,276 --> 00:34:25,324 from driverless convoys delivering logistical supplies, 647 00:34:25,367 --> 00:34:27,021 to unmanned drones 648 00:34:27,065 --> 00:34:30,764 delivering medical aid, to computational propaganda, 649 00:34:30,807 --> 00:34:34,246 trying to win the hearts and minds of a population. 650 00:34:34,289 --> 00:34:38,337 And so it stands to reason that whoever has the best AI 651 00:34:38,380 --> 00:34:41,688 will probably achieve dominance on this planet. 652 00:34:45,561 --> 00:34:47,650 At some point in the early 21st century, 653 00:34:47,694 --> 00:34:51,219 all of mankind was united in celebration. 654 00:34:51,263 --> 00:34:53,830 We marveled at our own magnificence 655 00:34:53,874 --> 00:34:56,833 as we gave birth to AI. 656 00:34:56,877 --> 00:34:58,966 AI? 657 00:34:59,009 --> 00:35:00,489 You mean artificial intelligence? 658 00:35:00,533 --> 00:35:01,751 A singular consciousness 659 00:35:01,795 --> 00:35:05,886 that spawned an entire race of machines. 660 00:35:05,929 --> 00:35:09,716 We don't know who struck first -- us or them, 661 00:35:09,759 --> 00:35:12,980 but we know that it was us that scorched the sky. 662 00:35:14,677 --> 00:35:16,766 Singer: There's a long history of science fiction, 663 00:35:16,810 --> 00:35:19,987 not just predicting the future, but shaping the future. 664 00:35:20,030 --> 00:35:26,820 ♪♪ 665 00:35:26,863 --> 00:35:30,389 Arthur Conan Doyle writing before World War I 666 00:35:30,432 --> 00:35:34,393 on the danger of how submarines might be used 667 00:35:34,436 --> 00:35:38,048 to carry out civilian blockades. 668 00:35:38,092 --> 00:35:40,399 At the time he's writing this fiction, 669 00:35:40,442 --> 00:35:43,402 the Royal Navy made fun of Arthur Conan Doyle 670 00:35:43,445 --> 00:35:45,230 for this absurd idea 671 00:35:45,273 --> 00:35:47,623 that submarines could be useful in war. 672 00:35:47,667 --> 00:35:53,412 ♪♪ 673 00:35:53,455 --> 00:35:55,370 One of the things we've seen in history 674 00:35:55,414 --> 00:35:58,243 is that our attitude towards technology, 675 00:35:58,286 --> 00:36:01,942 but also ethics, are very context-dependent. 676 00:36:01,985 --> 00:36:03,726 For example, the submarine... 677 00:36:03,770 --> 00:36:06,468 nations like Great Britain and even the United States 678 00:36:06,512 --> 00:36:09,863 found it horrifying to use the submarine. 679 00:36:09,906 --> 00:36:13,214 In fact, the German use of the submarine to carry out attacks 680 00:36:13,258 --> 00:36:18,480 was the reason why the United States joined World War I. 681 00:36:18,524 --> 00:36:20,613 But move the timeline forward. 682 00:36:20,656 --> 00:36:23,529 Man: The United States of America was suddenly 683 00:36:23,572 --> 00:36:28,403 and deliberately attacked by the empire of Japan. 684 00:36:28,447 --> 00:36:32,190 Five hours after Pearl Harbor, the order goes out 685 00:36:32,233 --> 00:36:36,498 to commit unrestricted submarine warfare against Japan. 686 00:36:39,936 --> 00:36:44,289 So Arthur Conan Doyle turned out to be right. 687 00:36:44,332 --> 00:36:46,856 Nolan: That's the great old line about science fiction -- 688 00:36:46,900 --> 00:36:48,336 It's a lie that tells the truth. 689 00:36:48,380 --> 00:36:51,470 Fellow executives, it gives me great pleasure 690 00:36:51,513 --> 00:36:54,821 to introduce you to the future of law enforcement... 691 00:36:54,864 --> 00:36:56,562 ED-209. 692 00:36:56,605 --> 00:37:03,612 ♪♪ 693 00:37:03,656 --> 00:37:05,919 This isn't just a question of science fiction. 694 00:37:05,962 --> 00:37:09,488 This is about what's next, about what's happening right now. 695 00:37:09,531 --> 00:37:13,927 ♪♪ 696 00:37:13,970 --> 00:37:17,496 The role of intelligent systems is growing very rapidly 697 00:37:17,539 --> 00:37:19,324 in warfare. 698 00:37:19,367 --> 00:37:22,152 Everyone is pushing in the unmanned realm. 699 00:37:22,196 --> 00:37:26,374 ♪♪ 700 00:37:26,418 --> 00:37:28,898 Gourley: Today, the Secretary of Defense is very, very clear -- 701 00:37:28,942 --> 00:37:32,337 We will not create fully autonomous attacking vehicles. 702 00:37:32,380 --> 00:37:34,643 Not everyone is gonna hold themselves 703 00:37:34,687 --> 00:37:36,515 to that same set of values. 704 00:37:36,558 --> 00:37:40,693 And when China and Russia start deploying autonomous vehicles 705 00:37:40,736 --> 00:37:45,611 that can attack and kill, what's the move that we're gonna make? 706 00:37:45,654 --> 00:37:49,963 ♪♪ 707 00:37:50,006 --> 00:37:51,617 Russell: You can't say, "Well, we're gonna use 708 00:37:51,660 --> 00:37:53,967 autonomous weapons for our military dominance, 709 00:37:54,010 --> 00:37:56,796 but no one else is gonna use them." 710 00:37:56,839 --> 00:38:00,495 If you make these weapons, they're gonna be used to attack 711 00:38:00,539 --> 00:38:03,324 human populations in large numbers. 712 00:38:03,368 --> 00:38:12,507 ♪♪ 713 00:38:12,551 --> 00:38:14,596 Autonomous weapons are, by their nature, 714 00:38:14,640 --> 00:38:16,468 weapons of mass destruction, 715 00:38:16,511 --> 00:38:19,862 because it doesn't need a human being to guide it or carry it. 716 00:38:19,906 --> 00:38:22,517 You only need one person, to, you know, 717 00:38:22,561 --> 00:38:25,781 write a little program. 718 00:38:25,825 --> 00:38:30,220 It just captures the complexity of this field. 719 00:38:30,264 --> 00:38:32,571 It is cool. It is important. 720 00:38:32,614 --> 00:38:34,573 It is amazing. 721 00:38:34,616 --> 00:38:37,053 It is also frightening. 722 00:38:37,097 --> 00:38:38,968 And it's all about trust. 723 00:38:42,102 --> 00:38:44,583 It's an open letter about artificial intelligence, 724 00:38:44,626 --> 00:38:47,063 signed by some of the biggest names in science. 725 00:38:47,107 --> 00:38:48,413 What do they want? 726 00:38:48,456 --> 00:38:50,763 Ban the use of autonomous weapons. 727 00:38:50,806 --> 00:38:52,373 Woman: The author stated, 728 00:38:52,417 --> 00:38:54,375 "Autonomous weapons have been described 729 00:38:54,419 --> 00:38:56,595 as the third revolution in warfare." 730 00:38:56,638 --> 00:38:58,553 Woman #2: ...thousand artificial-intelligence specialists 731 00:38:58,597 --> 00:39:01,817 calling for a global ban on killer robots. 732 00:39:01,861 --> 00:39:04,342 Tegmark: This open letter basically says 733 00:39:04,385 --> 00:39:06,344 that we should redefine the goal 734 00:39:06,387 --> 00:39:07,954 of the field of artificial intelligence 735 00:39:07,997 --> 00:39:11,610 away from just creating pure, undirected intelligence, 736 00:39:11,653 --> 00:39:13,655 towards creating beneficial intelligence. 737 00:39:13,699 --> 00:39:16,092 The development of AI is not going to stop. 738 00:39:16,136 --> 00:39:18,094 It is going to continue and get better. 739 00:39:18,138 --> 00:39:19,835 If the international community 740 00:39:19,879 --> 00:39:21,968 isn't putting certain controls on this, 741 00:39:22,011 --> 00:39:24,666 people will develop things that can do anything. 742 00:39:24,710 --> 00:39:27,365 Woman: The letter says that we are years, not decades, 743 00:39:27,408 --> 00:39:28,714 away from these weapons being deployed. 744 00:39:28,757 --> 00:39:30,106 So first of all... 745 00:39:30,150 --> 00:39:32,413 We had 6,000 signatories of that letter, 746 00:39:32,457 --> 00:39:35,155 including many of the major figures in the field. 747 00:39:37,026 --> 00:39:39,942 I'm getting a lot of visits from high-ranking officials 748 00:39:39,986 --> 00:39:42,989 who wish to emphasize that American military dominance 749 00:39:43,032 --> 00:39:45,731 is very important, and autonomous weapons 750 00:39:45,774 --> 00:39:50,083 may be part of the Defense Department's plan. 751 00:39:50,126 --> 00:39:52,433 That's very, very scary, because a value system 752 00:39:52,477 --> 00:39:54,479 of military developers of technology 753 00:39:54,522 --> 00:39:57,307 is not the same as a value system of the human race. 754 00:39:57,351 --> 00:40:00,746 ♪♪ 755 00:40:00,789 --> 00:40:02,922 Markoff: Out of the concerns about the possibility 756 00:40:02,965 --> 00:40:06,665 that this technology might be a threat to human existence, 757 00:40:06,708 --> 00:40:08,144 a number of the technologists 758 00:40:08,188 --> 00:40:09,972 have funded the Future of Life Institute 759 00:40:10,016 --> 00:40:12,192 to try to grapple with these problems. 760 00:40:13,193 --> 00:40:14,847 All of these guys are secretive, 761 00:40:14,890 --> 00:40:16,805 and so it's interesting to me to see them, 762 00:40:16,849 --> 00:40:20,635 you know, all together. 763 00:40:20,679 --> 00:40:24,030 Everything we have is a result of our intelligence. 764 00:40:24,073 --> 00:40:26,641 It's not the result of our big, scary teeth 765 00:40:26,685 --> 00:40:29,470 or our large claws or our enormous muscles. 766 00:40:29,514 --> 00:40:32,473 It's because we're actually relatively intelligent. 767 00:40:32,517 --> 00:40:35,520 And among my generation, we're all having 768 00:40:35,563 --> 00:40:37,086 what we call "holy cow," 769 00:40:37,130 --> 00:40:39,045 or "holy something else" moments, 770 00:40:39,088 --> 00:40:41,003 because we see that the technology 771 00:40:41,047 --> 00:40:44,180 is accelerating faster than we expected. 772 00:40:44,224 --> 00:40:46,705 I remember sitting around the table there 773 00:40:46,748 --> 00:40:50,099 with some of the best and the smartest minds in the world, 774 00:40:50,143 --> 00:40:52,058 and what really struck me was, 775 00:40:52,101 --> 00:40:56,149 maybe the human brain is not able to fully grasp 776 00:40:56,192 --> 00:40:58,673 the complexity of the world that we're confronted with. 777 00:40:58,717 --> 00:41:01,415 Russell: As it's currently constructed, 778 00:41:01,459 --> 00:41:04,766 the road that AI is following heads off a cliff, 779 00:41:04,810 --> 00:41:07,595 and we need to change the direction that we're going 780 00:41:07,639 --> 00:41:10,729 so that we don't take the human race off the cliff. 781 00:41:13,558 --> 00:41:17,126 Musk: Google acquired DeepMind several years ago. 782 00:41:17,170 --> 00:41:18,737 DeepMind operates 783 00:41:18,780 --> 00:41:22,088 as a semi-independent subsidiary of Google. 784 00:41:22,131 --> 00:41:24,960 The thing that makes DeepMind unique 785 00:41:25,004 --> 00:41:26,919 is that DeepMind is absolutely focused 786 00:41:26,962 --> 00:41:30,313 on creating digital superintelligence -- 787 00:41:30,357 --> 00:41:34,056 an AI that is vastly smarter than any human on Earth 788 00:41:34,100 --> 00:41:36,624 and ultimately smarter than all humans on Earth combined. 789 00:41:36,668 --> 00:41:40,715 This is from the DeepMind reinforcement learning system. 790 00:41:40,759 --> 00:41:43,544 Basically wakes up like a newborn baby 791 00:41:43,588 --> 00:41:46,852 and is shown the screen of an Atari video game 792 00:41:46,895 --> 00:41:50,508 and then has to learn to play the video game. 793 00:41:50,551 --> 00:41:55,600 It knows nothing about objects, about motion, about time. 794 00:41:57,602 --> 00:41:59,604 It only knows that there's an image on the screen 795 00:41:59,647 --> 00:42:02,563 and there's a score. 796 00:42:02,607 --> 00:42:06,436 So, if your baby woke up the day it was born 797 00:42:06,480 --> 00:42:08,090 and, by late afternoon, 798 00:42:08,134 --> 00:42:11,093 was playing 40 different Atari video games 799 00:42:11,137 --> 00:42:15,315 at a superhuman level, you would be terrified. 800 00:42:15,358 --> 00:42:19,101 You would say, "My baby is possessed. Send it back." 801 00:42:19,145 --> 00:42:23,584 Musk: The DeepMind system can win at any game. 802 00:42:23,628 --> 00:42:27,588 It can already beat all the original Atari games. 803 00:42:27,632 --> 00:42:29,155 It is superhuman. 804 00:42:29,198 --> 00:42:31,636 It plays the games at superspeed in less than a minute. 805 00:42:35,640 --> 00:42:37,032 ♪♪ 806 00:42:37,076 --> 00:42:38,643 DeepMind turned to another challenge, 807 00:42:38,686 --> 00:42:40,558 and the challenge was the game of Go, 808 00:42:40,601 --> 00:42:42,603 which people have generally argued 809 00:42:42,647 --> 00:42:45,084 has been beyond the power of computers 810 00:42:45,127 --> 00:42:48,304 to play with the best human Go players. 811 00:42:48,348 --> 00:42:51,264 First, they challenged a European Go champion. 812 00:42:53,222 --> 00:42:55,834 Then they challenged a Korean Go champion. 813 00:42:55,877 --> 00:42:57,836 Man: Please start the game. 814 00:42:57,879 --> 00:42:59,838 And they were able to win both times 815 00:42:59,881 --> 00:43:02,797 in kind of striking fashion. 816 00:43:02,841 --> 00:43:05,017 Nolan: You were reading articles in New York Timesyears ago 817 00:43:05,060 --> 00:43:09,761 talking about how Go would take 100 years for us to solve. 818 00:43:09,804 --> 00:43:11,110 Urban: People said, "Well, you know, 819 00:43:11,153 --> 00:43:13,460 but that's still just a board. 820 00:43:13,503 --> 00:43:15,027 Poker is an art. 821 00:43:15,070 --> 00:43:16,419 Poker involves reading people. 822 00:43:16,463 --> 00:43:18,073 Poker involves lying and bluffing. 823 00:43:18,117 --> 00:43:19,553 It's not an exact thing. 824 00:43:19,597 --> 00:43:21,381 That will never be, you know, a computer. 825 00:43:21,424 --> 00:43:22,861 You can't do that." 826 00:43:22,904 --> 00:43:24,732 They took the best poker players in the world, 827 00:43:24,776 --> 00:43:27,387 and it took seven days for the computer 828 00:43:27,430 --> 00:43:30,520 to start demolishing the humans. 829 00:43:30,564 --> 00:43:32,261 So it's the best poker player in the world, 830 00:43:32,305 --> 00:43:34,655 it's the best Go player in the world, and the pattern here 831 00:43:34,699 --> 00:43:37,440 is that AI might take a little while 832 00:43:37,484 --> 00:43:40,443 to wrap its tentacles around a new skill, 833 00:43:40,487 --> 00:43:44,883 but when it does, when it gets it, it is unstoppable. 834 00:43:44,926 --> 00:43:51,977 ♪♪ 835 00:43:52,020 --> 00:43:55,110 DeepMind's AI has administrator-level access 836 00:43:55,154 --> 00:43:57,156 to Google's servers 837 00:43:57,199 --> 00:44:00,768 to optimize energy usage at the data centers. 838 00:44:00,812 --> 00:44:04,816 However, this could be an unintentional Trojan horse. 839 00:44:04,859 --> 00:44:07,253 DeepMind has to have complete control of the data centers, 840 00:44:07,296 --> 00:44:08,950 so with a little software update, 841 00:44:08,994 --> 00:44:10,691 that AI could take complete control 842 00:44:10,735 --> 00:44:12,214 of the whole Google system, 843 00:44:12,258 --> 00:44:13,607 which means they can do anything. 844 00:44:13,651 --> 00:44:14,913 They could look at all your data. 845 00:44:14,956 --> 00:44:16,131 They could do anything. 846 00:44:16,175 --> 00:44:18,917 ♪♪ 847 00:44:20,135 --> 00:44:23,051 We're rapidly heading towards digital superintelligence 848 00:44:23,095 --> 00:44:24,313 that far exceeds any human. 849 00:44:24,357 --> 00:44:26,402 I think it's very obvious. 850 00:44:26,446 --> 00:44:27,708 Barrat: The problem is, we're not gonna 851 00:44:27,752 --> 00:44:29,710 suddenly hit human-level intelligence 852 00:44:29,754 --> 00:44:33,105 and say, "Okay, let's stop research." 853 00:44:33,148 --> 00:44:34,715 It's gonna go beyond human-level intelligence 854 00:44:34,759 --> 00:44:36,195 into what's called "superintelligence," 855 00:44:36,238 --> 00:44:39,459 and that's anything smarter than us. 856 00:44:39,502 --> 00:44:41,287 Tegmark: AI at the superhuman level, 857 00:44:41,330 --> 00:44:42,810 if we succeed with that, will be 858 00:44:42,854 --> 00:44:46,553 by far the most powerful invention we've ever made 859 00:44:46,596 --> 00:44:50,296 and the last invention we ever have to make. 860 00:44:50,339 --> 00:44:53,168 And if we create AI that's smarter than us, 861 00:44:53,212 --> 00:44:54,735 we have to be open to the possibility 862 00:44:54,779 --> 00:44:57,520 that we might actually lose control to them. 863 00:44:57,564 --> 00:45:00,741 ♪♪ 864 00:45:00,785 --> 00:45:02,612 Russell: Let's say you give it some objective, 865 00:45:02,656 --> 00:45:04,745 like curing cancer, and then you discover 866 00:45:04,789 --> 00:45:06,965 that the way it chooses to go about that 867 00:45:07,008 --> 00:45:08,444 is actually in conflict 868 00:45:08,488 --> 00:45:12,405 with a lot of other things you care about. 869 00:45:12,448 --> 00:45:16,496 Musk: AI doesn't have to be evil to destroy humanity. 870 00:45:16,539 --> 00:45:20,674 If AI has a goal, and humanity just happens to be in the way, 871 00:45:20,718 --> 00:45:22,894 it will destroy humanity as a matter of course, 872 00:45:22,937 --> 00:45:25,113 without even thinking about it. No hard feelings. 873 00:45:25,157 --> 00:45:27,072 It's just like if we're building a road 874 00:45:27,115 --> 00:45:29,770 and an anthill happens to be in the way... 875 00:45:29,814 --> 00:45:31,467 We don't hate ants. 876 00:45:31,511 --> 00:45:33,165 We're just building a road. 877 00:45:33,208 --> 00:45:34,557 And so goodbye, anthill. 878 00:45:34,601 --> 00:45:37,952 ♪♪ 879 00:45:37,996 --> 00:45:40,172 It's tempting to dismiss these concerns, 880 00:45:40,215 --> 00:45:42,783 'cause it's, like, something that might happen 881 00:45:42,827 --> 00:45:47,396 in a few decades or 100 years, so why worry? 882 00:45:47,440 --> 00:45:50,704 Russell: But if you go back to September 11, 1933, 883 00:45:50,748 --> 00:45:52,401 Ernest Rutherford, 884 00:45:52,445 --> 00:45:54,795 who was the most well-known nuclear physicist of his time, 885 00:45:54,839 --> 00:45:56,318 said that the possibility 886 00:45:56,362 --> 00:45:58,668 of ever extracting useful amounts of energy 887 00:45:58,712 --> 00:46:00,801 from the transmutation of atoms, as he called it, 888 00:46:00,845 --> 00:46:03,151 was moonshine. 889 00:46:03,195 --> 00:46:04,849 The next morning, Leo Szilard, 890 00:46:04,892 --> 00:46:06,502 who was a much younger physicist, 891 00:46:06,546 --> 00:46:09,984 read this and got really annoyed and figured out 892 00:46:10,028 --> 00:46:11,943 how to make a nuclear chain reaction 893 00:46:11,986 --> 00:46:13,379 just a few months later. 894 00:46:13,422 --> 00:46:20,560 ♪♪ 895 00:46:20,603 --> 00:46:23,693 We have spent more than $2 billion 896 00:46:23,737 --> 00:46:27,523 on the greatest scientific gamble in history. 897 00:46:27,567 --> 00:46:30,222 Russell: So when people say that, "Oh, this is so far off 898 00:46:30,265 --> 00:46:32,528 in the future, we don't have to worry about it," 899 00:46:32,572 --> 00:46:36,271 it might only be three, four breakthroughs of that magnitude 900 00:46:36,315 --> 00:46:40,275 that will get us from here to superintelligent machines. 901 00:46:40,319 --> 00:46:42,974 Tegmark: If it's gonna take 20 years to figure out 902 00:46:43,017 --> 00:46:45,237 how to keep AI beneficial, 903 00:46:45,280 --> 00:46:48,849 then we should start today, not at the last second 904 00:46:48,893 --> 00:46:51,460 when some dudes drinking Red Bull 905 00:46:51,504 --> 00:46:53,332 decide to flip the switch and test the thing. 906 00:46:53,375 --> 00:46:56,770 ♪♪ 907 00:46:56,814 --> 00:46:58,859 Musk: We have five years. 908 00:46:58,903 --> 00:47:00,600 I think digital superintelligence 909 00:47:00,643 --> 00:47:03,864 will happen in my lifetime. 910 00:47:03,908 --> 00:47:05,735 100%. 911 00:47:05,779 --> 00:47:07,215 Barrat: When this happens, 912 00:47:07,259 --> 00:47:09,696 it will be surrounded by a bunch of people 913 00:47:09,739 --> 00:47:13,091 who are really just excited about the technology. 914 00:47:13,134 --> 00:47:15,571 They want to see it succeed, but they're not anticipating 915 00:47:15,615 --> 00:47:16,964 that it can get out of control. 916 00:47:17,008 --> 00:47:24,450 ♪♪ 917 00:47:25,494 --> 00:47:28,584 Oh, my God, I trust my computer so much. 918 00:47:28,628 --> 00:47:30,195 That's an amazing question. 919 00:47:30,238 --> 00:47:31,457 I don't trust my computer. 920 00:47:31,500 --> 00:47:32,937 If it's on, I take it off. 921 00:47:32,980 --> 00:47:34,242 Like, even when it's off, 922 00:47:34,286 --> 00:47:35,896 I still think it's on. Like, you know? 923 00:47:35,940 --> 00:47:37,637 Like, you really cannot tru-- Like, the webcams, 924 00:47:37,680 --> 00:47:39,595 you don't know if, like, someone might turn it... 925 00:47:39,639 --> 00:47:41,249 You don't know, like. 926 00:47:41,293 --> 00:47:42,903 I don't trust my computer. 927 00:47:42,947 --> 00:47:46,907 Like, in my phone, every time they ask me 928 00:47:46,951 --> 00:47:49,475 "Can we send your information to Apple?" 929 00:47:49,518 --> 00:47:50,998 every time, I... 930 00:47:51,042 --> 00:47:53,087 So, I don't trust my phone. 931 00:47:53,131 --> 00:47:56,743 Okay. So, part of it is, yes, I do trust it, 932 00:47:56,786 --> 00:48:00,660 because it would be really hard to get through the day 933 00:48:00,703 --> 00:48:04,011 in the way our world is set up without computers. 934 00:48:04,055 --> 00:48:05,360 ♪♪ 935 00:48:10,975 --> 00:48:13,368 Dr. Herman: Trust is such a human experience. 936 00:48:13,412 --> 00:48:21,246 ♪♪ 937 00:48:21,289 --> 00:48:25,119 I have a patient coming in with an intracranial aneurysm. 938 00:48:25,163 --> 00:48:29,994 ♪♪ 939 00:48:30,037 --> 00:48:31,691 They want to look in my eyes and know 940 00:48:31,734 --> 00:48:34,955 that they can trust this person with their life. 941 00:48:34,999 --> 00:48:39,394 I'm not horribly concerned about anything. 942 00:48:39,438 --> 00:48:40,830 Good. Part of that 943 00:48:40,874 --> 00:48:42,920 is because I have confidence in you. 944 00:48:42,963 --> 00:48:50,710 ♪♪ 945 00:48:50,753 --> 00:48:52,233 This procedure we're doing today 946 00:48:52,277 --> 00:48:57,151 20 years ago was essentially impossible. 947 00:48:57,195 --> 00:49:00,328 We just didn't have the materials and the technologies. 948 00:49:04,202 --> 00:49:13,385 ♪♪ 949 00:49:13,428 --> 00:49:22,655 ♪♪ 950 00:49:22,698 --> 00:49:26,485 So, the coil is barely in there right now. 951 00:49:26,528 --> 00:49:29,923 It's just a feather holding it in. 952 00:49:29,967 --> 00:49:32,012 It's nervous time. 953 00:49:32,056 --> 00:49:36,147 ♪♪ 954 00:49:36,190 --> 00:49:37,626 We're just in purgatory, 955 00:49:37,670 --> 00:49:40,673 intellectual, humanistic purgatory, 956 00:49:40,716 --> 00:49:43,632 and AI might know exactly what to do here. 957 00:49:43,676 --> 00:49:50,596 ♪♪ 958 00:49:50,639 --> 00:49:52,554 We've got the coil into the aneurysm. 959 00:49:52,598 --> 00:49:54,556 But it wasn't in tremendously well 960 00:49:54,600 --> 00:49:56,428 that I knew that it would stay, 961 00:49:56,471 --> 00:50:01,041 so with a maybe 20% risk of a very bad situation, 962 00:50:01,085 --> 00:50:04,436 I elected to just bring her back. 963 00:50:04,479 --> 00:50:05,959 Because of my relationship with her 964 00:50:06,003 --> 00:50:08,222 and knowing the difficulties of coming in 965 00:50:08,266 --> 00:50:11,051 and having the procedure, I consider things, 966 00:50:11,095 --> 00:50:14,272 when I should only consider the safest possible route 967 00:50:14,315 --> 00:50:16,361 to achieve success. 968 00:50:16,404 --> 00:50:19,755 But I had to stand there for 10 minutes agonizing about it. 969 00:50:19,799 --> 00:50:21,757 The computer feels nothing. 970 00:50:21,801 --> 00:50:24,760 The computer just does what it's supposed to do, 971 00:50:24,804 --> 00:50:26,284 better and better. 972 00:50:26,327 --> 00:50:30,288 ♪♪ 973 00:50:30,331 --> 00:50:32,551 I want to be AI in this case. 974 00:50:35,945 --> 00:50:38,861 But can AI be compassionate? 975 00:50:38,905 --> 00:50:43,040 ♪♪ 976 00:50:43,083 --> 00:50:47,827 I mean, it's everybody's question about AI. 977 00:50:47,870 --> 00:50:51,961 We are the sole embodiment of humanity, 978 00:50:52,005 --> 00:50:55,269 and it's a stretch for us to accept that a machine 979 00:50:55,313 --> 00:50:58,794 can be compassionate and loving in that way. 980 00:50:58,838 --> 00:51:05,105 ♪♪ 981 00:51:05,149 --> 00:51:07,281 Part of me doesn't believe in magic, 982 00:51:07,325 --> 00:51:09,805 but part of me has faith that there is something 983 00:51:09,849 --> 00:51:11,546 beyond the sum of the parts, 984 00:51:11,590 --> 00:51:15,637 that there is at least a oneness in our shared ancestry, 985 00:51:15,681 --> 00:51:20,338 our shared biology, our shared history. 986 00:51:20,381 --> 00:51:23,210 Some connection there beyond machine. 987 00:51:23,254 --> 00:51:30,304 ♪♪ 988 00:51:30,348 --> 00:51:32,567 So, then, you have the other side of that, is, 989 00:51:32,611 --> 00:51:34,047 does the computer know it's conscious, 990 00:51:34,091 --> 00:51:37,137 or can it be conscious, or does it care? 991 00:51:37,181 --> 00:51:40,009 Does it need to be conscious? 992 00:51:40,053 --> 00:51:42,011 Does it need to be aware? 993 00:51:42,055 --> 00:51:47,365 ♪♪ 994 00:51:47,408 --> 00:51:52,848 ♪♪ 995 00:51:52,892 --> 00:51:56,417 I do not think that a robot could ever be conscious. 996 00:51:56,461 --> 00:51:58,376 Unless they programmed it that way. 997 00:51:58,419 --> 00:52:00,639 Conscious? No. 998 00:52:00,682 --> 00:52:03,163 No. No. 999 00:52:03,207 --> 00:52:06,035 I mean, think a robot could be programmed to be conscious. 1000 00:52:06,079 --> 00:52:09,648 How are they programmed to do everything else? 1001 00:52:09,691 --> 00:52:12,390 That's another big part of artificial intelligence, 1002 00:52:12,433 --> 00:52:15,741 is to make them conscious and make them feel. 1003 00:52:17,003 --> 00:52:22,400 ♪♪ 1004 00:52:22,443 --> 00:52:26,230 Lipson: Back in 2005, we started trying to build machines 1005 00:52:26,273 --> 00:52:27,709 with self-awareness. 1006 00:52:27,753 --> 00:52:33,062 ♪♪ 1007 00:52:33,106 --> 00:52:37,284 This robot, to begin with, didn't know what it was. 1008 00:52:37,328 --> 00:52:40,244 All it knew was that it needed to do something like walk. 1009 00:52:40,287 --> 00:52:44,073 ♪♪ 1010 00:52:44,117 --> 00:52:45,597 Through trial and error, 1011 00:52:45,640 --> 00:52:49,731 it figured out how to walk using its imagination, 1012 00:52:49,775 --> 00:52:54,040 and then it walked away. 1013 00:52:54,083 --> 00:52:56,390 And then we did something very cruel. 1014 00:52:56,434 --> 00:52:58,653 We chopped off a leg and watched what happened. 1015 00:52:58,697 --> 00:53:03,005 ♪♪ 1016 00:53:03,049 --> 00:53:07,749 At the beginning, it didn't quite know what had happened. 1017 00:53:07,793 --> 00:53:13,233 But over about a period of a day, it then began to limp. 1018 00:53:13,277 --> 00:53:16,845 And then, a year ago, we were training an AI system 1019 00:53:16,889 --> 00:53:20,240 for a live demonstration. 1020 00:53:20,284 --> 00:53:21,763 We wanted to show how we wave 1021 00:53:21,807 --> 00:53:24,113 all these objects in front of the camera 1022 00:53:24,157 --> 00:53:27,334 and the AI could recognize the objects. 1023 00:53:27,378 --> 00:53:29,031 And so, we're preparing this demo, 1024 00:53:29,075 --> 00:53:31,251 and we had on a side screen this ability 1025 00:53:31,295 --> 00:53:36,778 to watch what certain neurons were responding to. 1026 00:53:36,822 --> 00:53:39,041 And suddenly we noticed that one of the neurons 1027 00:53:39,085 --> 00:53:41,087 was tracking faces. 1028 00:53:41,130 --> 00:53:45,483 It was tracking our faces as we were moving around. 1029 00:53:45,526 --> 00:53:48,616 Now, the spooky thing about this is that we never trained 1030 00:53:48,660 --> 00:53:52,490 the system to recognize human faces, 1031 00:53:52,533 --> 00:53:55,710 and yet, somehow, it learned to do that. 1032 00:53:57,973 --> 00:53:59,584 Even though these robots are very simple, 1033 00:53:59,627 --> 00:54:02,500 we can see there's something else going on there. 1034 00:54:02,543 --> 00:54:05,851 It's not just programming. 1035 00:54:05,894 --> 00:54:08,462 So, this is just the beginning. 1036 00:54:10,377 --> 00:54:14,294 Horvitz: I often think about that beach in Kitty Hawk, 1037 00:54:14,338 --> 00:54:18,255 the 1903 flight by Orville and Wilbur Wright. 1038 00:54:21,214 --> 00:54:24,348 It was kind of a canvas plane, and it's wood and iron, 1039 00:54:24,391 --> 00:54:26,828 and it gets off the ground for, what, a minute and 20 seconds, 1040 00:54:26,872 --> 00:54:29,091 on this windy day 1041 00:54:29,135 --> 00:54:31,006 before touching back down again. 1042 00:54:33,270 --> 00:54:37,143 And it was just around 65 summers or so 1043 00:54:37,186 --> 00:54:43,149 after that moment that you have a 747 taking off from JFK... 1044 00:54:43,192 --> 00:54:50,156 ♪♪ 1045 00:54:50,199 --> 00:54:51,984 ...where a major concern of someone on the airplane 1046 00:54:52,027 --> 00:54:55,422 might be whether or not their salt-free diet meal 1047 00:54:55,466 --> 00:54:56,902 is gonna be coming to them or not. 1048 00:54:56,945 --> 00:54:58,469 We have a whole infrastructure, 1049 00:54:58,512 --> 00:55:01,385 with travel agents and tower control, 1050 00:55:01,428 --> 00:55:03,778 and it's all casual, and it's all part of the world. 1051 00:55:03,822 --> 00:55:07,042 ♪♪ 1052 00:55:07,086 --> 00:55:09,523 Right now, as far as we've come with machines 1053 00:55:09,567 --> 00:55:12,134 that think and solve problems, we're at Kitty Hawk now. 1054 00:55:12,178 --> 00:55:13,745 We're in the wind. 1055 00:55:13,788 --> 00:55:17,052 We have our tattered-canvas planes up in the air. 1056 00:55:17,096 --> 00:55:20,882 ♪♪ 1057 00:55:20,926 --> 00:55:23,885 But what happens in 65 summers or so? 1058 00:55:23,929 --> 00:55:27,889 We will have machines that are beyond human control. 1059 00:55:27,933 --> 00:55:30,457 Should we worry about that? 1060 00:55:30,501 --> 00:55:32,590 ♪♪ 1061 00:55:32,633 --> 00:55:34,853 I'm not sure it's going to help. 1062 00:55:40,337 --> 00:55:44,036 Kaplan: Nobody has any idea today what it means for a robot 1063 00:55:44,079 --> 00:55:46,430 to be conscious. 1064 00:55:46,473 --> 00:55:48,649 There is no such thing. 1065 00:55:48,693 --> 00:55:50,172 There are a lot of smart people, 1066 00:55:50,216 --> 00:55:53,088 and I have a great deal of respect for them, 1067 00:55:53,132 --> 00:55:57,528 but the truth is, machines are natural psychopaths. 1068 00:55:57,571 --> 00:55:59,225 Man: Fear came back into the market. 1069 00:55:59,268 --> 00:56:01,706 Man #2: Went down 800, nearly 1,000, in a heartbeat. 1070 00:56:01,749 --> 00:56:03,360 I mean, it is classic capitulation. 1071 00:56:03,403 --> 00:56:04,796 There are some people who are proposing 1072 00:56:04,839 --> 00:56:07,146 it was some kind of fat-finger error. 1073 00:56:07,189 --> 00:56:09,583 Take the Flash Crash of 2010. 1074 00:56:09,627 --> 00:56:13,413 In a matter of minutes, $1 trillion in value 1075 00:56:13,457 --> 00:56:15,415 was lost in the stock market. 1076 00:56:15,459 --> 00:56:18,984 Woman: The Dow dropped nearly 1,000 points in a half-hour. 1077 00:56:19,027 --> 00:56:22,553 Kaplan: So, what went wrong? 1078 00:56:22,596 --> 00:56:26,644 By that point in time, more than 60% of all the trades 1079 00:56:26,687 --> 00:56:29,124 that took place on the stock exchange 1080 00:56:29,168 --> 00:56:32,693 were actually being initiated by computers. 1081 00:56:32,737 --> 00:56:34,216 Man: Panic selling on the way down, 1082 00:56:34,260 --> 00:56:35,783 and all of a sudden it stopped on a dime. 1083 00:56:35,827 --> 00:56:37,611 Man #2: This is all happening in real time, folks. 1084 00:56:37,655 --> 00:56:39,526 Wisz: The short story of what happened in the Flash Crash 1085 00:56:39,570 --> 00:56:42,399 is that algorithms responded to algorithms, 1086 00:56:42,442 --> 00:56:45,358 and it compounded upon itself over and over and over again 1087 00:56:45,402 --> 00:56:47,012 in a matter of minutes. 1088 00:56:47,055 --> 00:56:50,972 Man: At one point, the market fell as if down a well. 1089 00:56:51,016 --> 00:56:54,323 There is no regulatory body that can adapt quickly enough 1090 00:56:54,367 --> 00:56:57,979 to prevent potentially disastrous consequences 1091 00:56:58,023 --> 00:57:01,243 of AI operating in our financial systems. 1092 00:57:01,287 --> 00:57:03,898 They are so prime for manipulation. 1093 00:57:03,942 --> 00:57:05,639 Let's talk about the speed with which 1094 00:57:05,683 --> 00:57:08,076 we are watching this market deteriorate. 1095 00:57:08,120 --> 00:57:11,602 That's the type of AI-run-amuck that scares people. 1096 00:57:11,645 --> 00:57:13,560 Kaplan: When you give them a goal, 1097 00:57:13,604 --> 00:57:17,825 they will relentlessly pursue that goal. 1098 00:57:17,869 --> 00:57:20,393 How many computer programs are there like this? 1099 00:57:20,437 --> 00:57:23,483 Nobody knows. 1100 00:57:23,527 --> 00:57:27,444 Kosinski: One of the fascinating aspects about AI in general 1101 00:57:27,487 --> 00:57:31,970 is that no one really understands how it works. 1102 00:57:32,013 --> 00:57:36,975 Even the people who create AI don't really fully understand. 1103 00:57:37,018 --> 00:57:39,804 Because it has millions of elements, 1104 00:57:39,847 --> 00:57:41,675 it becomes completely impossible 1105 00:57:41,719 --> 00:57:45,113 for a human being to understand what's going on. 1106 00:57:45,157 --> 00:57:52,512 ♪♪ 1107 00:57:52,556 --> 00:57:56,037 Grassegger: Microsoft had set up this artificial intelligence 1108 00:57:56,081 --> 00:57:59,127 called Tay on Twitter, which was a chatbot. 1109 00:58:00,912 --> 00:58:02,696 They started out in the morning, 1110 00:58:02,740 --> 00:58:06,526 and Tay was starting to tweet and learning from stuff 1111 00:58:06,570 --> 00:58:10,835 that was being sent to him from other Twitter people. 1112 00:58:10,878 --> 00:58:13,272 Because some people, like trolls, attacked him, 1113 00:58:13,315 --> 00:58:18,582 within 24 hours, the Microsoft bot became a terrible person. 1114 00:58:18,625 --> 00:58:21,367 They had to literally pull Tay off the Net 1115 00:58:21,410 --> 00:58:24,718 because he had turned into a monster. 1116 00:58:24,762 --> 00:58:30,550 A misanthropic, racist, horrible person you'd never want to meet. 1117 00:58:30,594 --> 00:58:32,857 And nobody had foreseen this. 1118 00:58:35,337 --> 00:58:38,602 The whole idea of AI is that we are not telling it exactly 1119 00:58:38,645 --> 00:58:42,780 how to achieve a given outcome or a goal. 1120 00:58:42,823 --> 00:58:46,435 AI develops on its own. 1121 00:58:46,479 --> 00:58:48,829 Nolan: We're worried about superintelligent AI, 1122 00:58:48,873 --> 00:58:52,790 the master chess player that will outmaneuver us, 1123 00:58:52,833 --> 00:58:55,923 but AI won't have to actually be that smart 1124 00:58:55,967 --> 00:59:00,145 to have massively disruptive effects on human civilization. 1125 00:59:00,188 --> 00:59:01,886 We've seen over the last century 1126 00:59:01,929 --> 00:59:05,150 it doesn't necessarily take a genius to knock history off 1127 00:59:05,193 --> 00:59:06,804 in a particular direction, 1128 00:59:06,847 --> 00:59:09,589 and it won't take a genius AI to do the same thing. 1129 00:59:09,633 --> 00:59:13,158 Bogus election news stories generated more engagement 1130 00:59:13,201 --> 00:59:17,075 on Facebook than top real stories. 1131 00:59:17,118 --> 00:59:21,079 Facebook really is the elephant in the room. 1132 00:59:21,122 --> 00:59:23,777 Kosinski: AI running Facebook news feed -- 1133 00:59:23,821 --> 00:59:28,347 The task for AI is keeping users engaged, 1134 00:59:28,390 --> 00:59:29,827 but no one really understands 1135 00:59:29,870 --> 00:59:34,832 exactly how this AI is achieving this goal. 1136 00:59:34,875 --> 00:59:38,792 Nolan: Facebook is building an elegant mirrored wall around us. 1137 00:59:38,836 --> 00:59:41,665 A mirror that we can ask, "Who's the fairest of them all?" 1138 00:59:41,708 --> 00:59:45,016 and it will answer, "You, you," time and again 1139 00:59:45,059 --> 00:59:48,193 and slowly begin to warp our sense of reality, 1140 00:59:48,236 --> 00:59:53,502 warp our sense of politics, history, global events, 1141 00:59:53,546 --> 00:59:57,028 until determining what's true and what's not true, 1142 00:59:57,071 --> 00:59:58,943 is virtually impossible. 1143 01:00:01,032 --> 01:00:03,861 The problem is that AI doesn't understand that. 1144 01:00:03,904 --> 01:00:08,039 AI just had a mission -- maximize user engagement, 1145 01:00:08,082 --> 01:00:10,041 and it achieved that. 1146 01:00:10,084 --> 01:00:13,653 Nearly 2 billion people spend nearly one hour 1147 01:00:13,697 --> 01:00:17,831 on average a day basically interacting with AI 1148 01:00:17,875 --> 01:00:21,530 that is shaping their experience. 1149 01:00:21,574 --> 01:00:24,664 Even Facebook engineers, they don't like fake news. 1150 01:00:24,708 --> 01:00:26,666 It's very bad business. 1151 01:00:26,710 --> 01:00:28,015 They want to get rid of fake news. 1152 01:00:28,059 --> 01:00:29,974 It's just very difficult to do because, 1153 01:00:30,017 --> 01:00:32,324 how do you recognize news as fake 1154 01:00:32,367 --> 01:00:34,456 if you cannot read all of those news personally? 1155 01:00:34,500 --> 01:00:39,418 There's so much active misinformation 1156 01:00:39,461 --> 01:00:41,115 and it's packaged very well, 1157 01:00:41,159 --> 01:00:44,553 and it looks the same when you see it on a Facebook page 1158 01:00:44,597 --> 01:00:47,426 or you turn on your television. 1159 01:00:47,469 --> 01:00:49,210 Nolan: It's not terribly sophisticated, 1160 01:00:49,254 --> 01:00:51,691 but it is terribly powerful. 1161 01:00:51,735 --> 01:00:54,346 And what it means is that your view of the world, 1162 01:00:54,389 --> 01:00:56,435 which, 20 years ago, was determined, 1163 01:00:56,478 --> 01:01:00,004 if you watched the nightly news, by three different networks, 1164 01:01:00,047 --> 01:01:02,528 the three anchors who endeavored to try to get it right. 1165 01:01:02,571 --> 01:01:04,225 Might have had a little bias one way or the other, 1166 01:01:04,269 --> 01:01:05,923 but, largely speaking, we could all agree 1167 01:01:05,966 --> 01:01:08,273 on an objective reality. 1168 01:01:08,316 --> 01:01:10,754 Well, that objectivity is gone, 1169 01:01:10,797 --> 01:01:13,757 and Facebook has completely annihilated it. 1170 01:01:13,800 --> 01:01:17,064 ♪♪ 1171 01:01:17,108 --> 01:01:19,197 If most of your understanding of how the world works 1172 01:01:19,240 --> 01:01:20,807 is derived from Facebook, 1173 01:01:20,851 --> 01:01:23,418 facilitated by algorithmic software 1174 01:01:23,462 --> 01:01:27,118 that tries to show you the news you want to see, 1175 01:01:27,161 --> 01:01:28,815 that's a terribly dangerous thing. 1176 01:01:28,859 --> 01:01:33,080 And the idea that we have not only set that in motion, 1177 01:01:33,124 --> 01:01:37,258 but allowed bad-faith actors access to that information... 1178 01:01:37,302 --> 01:01:39,565 I mean, this is a recipe for disaster. 1179 01:01:39,608 --> 01:01:43,134 ♪♪ 1180 01:01:43,177 --> 01:01:45,876 Urban: I think that there will definitely be lots of bad actors 1181 01:01:45,919 --> 01:01:48,922 trying to manipulate the world with AI. 1182 01:01:48,966 --> 01:01:52,143 2016 was a perfect example of an election 1183 01:01:52,186 --> 01:01:55,015 where there was lots of AI producing lots of fake news 1184 01:01:55,059 --> 01:01:58,323 and distributing it for a purpose, for a result. 1185 01:01:59,890 --> 01:02:02,283 Ladies and gentlemen, honorable colleagues... 1186 01:02:02,327 --> 01:02:04,546 it's my privilege to speak to you today 1187 01:02:04,590 --> 01:02:07,985 about the power of big data and psychographics 1188 01:02:08,028 --> 01:02:09,682 in the electoral process 1189 01:02:09,726 --> 01:02:12,206 and, specifically, to talk about the work 1190 01:02:12,250 --> 01:02:14,513 that we contributed to Senator Cruz's 1191 01:02:14,556 --> 01:02:16,558 presidential primary campaign. 1192 01:02:16,602 --> 01:02:19,910 Nolan: Cambridge Analytica emerged quietly as a company 1193 01:02:19,953 --> 01:02:21,563 that, according to its own hype, 1194 01:02:21,607 --> 01:02:26,307 has the ability to use this tremendous amount of data 1195 01:02:26,351 --> 01:02:30,137 in order to effect societal change. 1196 01:02:30,181 --> 01:02:33,358 In 2016, they had three major clients. 1197 01:02:33,401 --> 01:02:34,794 Ted Cruz was one of them. 1198 01:02:34,838 --> 01:02:37,884 It's easy to forget that, only 18 months ago, 1199 01:02:37,928 --> 01:02:41,148 Senator Cruz was one of the less popular candidates 1200 01:02:41,192 --> 01:02:42,846 seeking nomination. 1201 01:02:42,889 --> 01:02:47,241 So, what was not possible maybe, like, 10 or 15 years ago, 1202 01:02:47,285 --> 01:02:49,374 was that you can send fake news 1203 01:02:49,417 --> 01:02:52,420 to exactly the people that you want to send it to. 1204 01:02:52,464 --> 01:02:56,685 And then you could actually see how he or she reacts on Facebook 1205 01:02:56,729 --> 01:02:58,905 and then adjust that information 1206 01:02:58,949 --> 01:03:01,778 according to the feedback that you got. 1207 01:03:01,821 --> 01:03:03,257 So you can start developing 1208 01:03:03,301 --> 01:03:06,130 kind of a real-time management of a population. 1209 01:03:06,173 --> 01:03:08,697 In this case, we've zoned in 1210 01:03:08,741 --> 01:03:10,699 on a group we've called "Persuasion." 1211 01:03:10,743 --> 01:03:13,746 These are people who are definitely going to vote, 1212 01:03:13,790 --> 01:03:16,705 to caucus, but they need moving from the center 1213 01:03:16,749 --> 01:03:18,490 a little bit more towards the right. 1214 01:03:18,533 --> 01:03:19,708 in order to support Cruz. 1215 01:03:19,752 --> 01:03:22,059 They need a persuasion message. 1216 01:03:22,102 --> 01:03:23,800 "Gun rights," I've selected. 1217 01:03:23,843 --> 01:03:25,802 That narrows the field slightly more. 1218 01:03:25,845 --> 01:03:29,066 And now we know that we need a message on gun rights, 1219 01:03:29,109 --> 01:03:31,111 it needs to be a persuasion message, 1220 01:03:31,155 --> 01:03:32,591 and it needs to be nuanced 1221 01:03:32,634 --> 01:03:34,201 according to the certain personality 1222 01:03:34,245 --> 01:03:36,029 that we're interested in. 1223 01:03:36,073 --> 01:03:39,946 Through social media, there's an infinite amount of information 1224 01:03:39,990 --> 01:03:42,514 that you can gather about a person. 1225 01:03:42,557 --> 01:03:45,734 We have somewhere close to 4,000 or 5,000 data points 1226 01:03:45,778 --> 01:03:48,563 on every adult in the United States. 1227 01:03:48,607 --> 01:03:51,915 Grassegger: It's about targeting the individual. 1228 01:03:51,958 --> 01:03:54,352 It's like a weapon, which can be used 1229 01:03:54,395 --> 01:03:55,962 in the totally wrong direction. 1230 01:03:56,006 --> 01:03:58,051 That's the problem with all of this data. 1231 01:03:58,095 --> 01:04:02,229 It's almost as if we built the bullet before we built the gun. 1232 01:04:02,273 --> 01:04:04,362 Ted Cruz employed our data, 1233 01:04:04,405 --> 01:04:06,407 our behavioral insights. 1234 01:04:06,451 --> 01:04:09,541 He started from a base of less than 5% 1235 01:04:09,584 --> 01:04:15,590 and had a very slow-and-steady- but-firm rise to above 35%, 1236 01:04:15,634 --> 01:04:17,157 making him, obviously, 1237 01:04:17,201 --> 01:04:20,465 the second most threatening contender in the race. 1238 01:04:20,508 --> 01:04:23,120 Now, clearly, the Cruz campaign is over now, 1239 01:04:23,163 --> 01:04:24,904 but what I can tell you 1240 01:04:24,948 --> 01:04:28,168 is that of the two candidates left in this election, 1241 01:04:28,212 --> 01:04:30,867 one of them is using these technologies. 1242 01:04:32,564 --> 01:04:35,959 I, Donald John Trump, do solemnly swear 1243 01:04:36,002 --> 01:04:38,222 that I will faithfully execute 1244 01:04:38,265 --> 01:04:42,226 the office of President of the United States. 1245 01:04:42,269 --> 01:04:46,273 ♪♪ 1246 01:04:48,275 --> 01:04:50,234 Nolan: Elections are a marginal exercise. 1247 01:04:50,277 --> 01:04:53,237 It doesn't take a very sophisticated AI 1248 01:04:53,280 --> 01:04:57,719 in order to have a disproportionate impact. 1249 01:04:57,763 --> 01:05:02,550 Before Trump, Brexit was another supposed client. 1250 01:05:02,594 --> 01:05:04,726 Well, at 20 minutes to 5:00, 1251 01:05:04,770 --> 01:05:08,730 we can now say the decision taken in 1975 1252 01:05:08,774 --> 01:05:10,950 by this country to join the common market 1253 01:05:10,994 --> 01:05:15,999 has been reversed by this referendum to leave the EU. 1254 01:05:16,042 --> 01:05:19,828 Nolan: Cambridge Analytica allegedly uses AI 1255 01:05:19,872 --> 01:05:23,267 to push through two of the most ground-shaking pieces 1256 01:05:23,310 --> 01:05:27,967 of political change in the last 50 years. 1257 01:05:28,011 --> 01:05:30,709 These are epochal events, and if we believe the hype, 1258 01:05:30,752 --> 01:05:33,755 they are connected directly to a piece of software, 1259 01:05:33,799 --> 01:05:37,194 essentially, created by a professor at Stanford. 1260 01:05:37,237 --> 01:05:41,415 ♪♪ 1261 01:05:41,459 --> 01:05:43,635 Kosinski: Back in 2013, I described 1262 01:05:43,678 --> 01:05:45,593 that what they are doing is possible 1263 01:05:45,637 --> 01:05:49,293 and warned against this happening in the future. 1264 01:05:49,336 --> 01:05:51,382 Grassegger: At the time, Michal Kosinski 1265 01:05:51,425 --> 01:05:52,949 was a young Polish researcher 1266 01:05:52,992 --> 01:05:54,994 working at the Psychometrics Centre. 1267 01:05:55,038 --> 01:06:00,217 So, what Michal had done was to gather the largest-ever data set 1268 01:06:00,260 --> 01:06:03,481 of how people behave on Facebook. 1269 01:06:03,524 --> 01:06:07,789 Kosinski: Psychometrics is trying to measure psychological traits, 1270 01:06:07,833 --> 01:06:09,922 such as personality, intelligence, 1271 01:06:09,966 --> 01:06:11,880 political views, and so on. 1272 01:06:11,924 --> 01:06:15,058 Now, traditionally, those traits were measured 1273 01:06:15,101 --> 01:06:17,712 using tests and questions. 1274 01:06:17,756 --> 01:06:19,410 Nolan: Personality test -- the most benign thing 1275 01:06:19,453 --> 01:06:20,715 you could possibly think of. 1276 01:06:20,759 --> 01:06:22,065 Something that doesn't necessarily have 1277 01:06:22,108 --> 01:06:24,197 a lot of utility, right? 1278 01:06:24,241 --> 01:06:27,331 Kosinski: Our idea was that instead of tests and questions, 1279 01:06:27,374 --> 01:06:30,029 we could simply look at the digital footprints of behaviors 1280 01:06:30,073 --> 01:06:32,553 that we are all leaving behind 1281 01:06:32,597 --> 01:06:34,903 to understand openness, 1282 01:06:34,947 --> 01:06:37,732 conscientiousness, neuroticism. 1283 01:06:37,776 --> 01:06:39,560 Grassegger: You can easily buy personal data, 1284 01:06:39,604 --> 01:06:43,129 such as where you live, what club memberships you've tried, 1285 01:06:43,173 --> 01:06:45,044 which gym you go to. 1286 01:06:45,088 --> 01:06:47,873 There are actually marketplaces for personal data. 1287 01:06:47,916 --> 01:06:49,918 Nolan: It turns out, we can discover an awful lot 1288 01:06:49,962 --> 01:06:51,442 about what you're gonna do 1289 01:06:51,485 --> 01:06:55,750 based on a very, very tiny set of information. 1290 01:06:55,794 --> 01:06:58,275 Kosinski: We are training deep-learning networks 1291 01:06:58,318 --> 01:07:01,278 to infer intimate traits, 1292 01:07:01,321 --> 01:07:04,759 people's political views, personality, 1293 01:07:04,803 --> 01:07:07,806 intelligence, sexual orientation 1294 01:07:07,849 --> 01:07:10,504 just from an image from someone's face. 1295 01:07:10,548 --> 01:07:17,033 ♪♪ 1296 01:07:17,076 --> 01:07:20,645 Now think about countries which are not so free and open-minded. 1297 01:07:20,688 --> 01:07:23,300 If you can reveal people's religious views 1298 01:07:23,343 --> 01:07:25,954 or political views or sexual orientation 1299 01:07:25,998 --> 01:07:28,740 based on only profile pictures, 1300 01:07:28,783 --> 01:07:33,310 this could be literally an issue of life and death. 1301 01:07:33,353 --> 01:07:36,965 ♪♪ 1302 01:07:37,009 --> 01:07:39,751 I think there's no going back. 1303 01:07:42,145 --> 01:07:44,321 Do you know what the Turing test is? 1304 01:07:44,364 --> 01:07:48,977 It's when a human interacts with a computer, 1305 01:07:49,021 --> 01:07:50,805 and if the human doesn't know they're interacting 1306 01:07:50,849 --> 01:07:52,546 with a computer, 1307 01:07:52,590 --> 01:07:54,026 the test is passed. 1308 01:07:54,070 --> 01:07:57,247 And over the next few days, 1309 01:07:57,290 --> 01:07:59,684 you're gonna be the human component in a Turing test. 1310 01:07:59,727 --> 01:08:02,295 Holy shit.Yeah, that's right, Caleb. 1311 01:08:02,339 --> 01:08:04,080 You got it. 1312 01:08:04,123 --> 01:08:06,865 'Cause if that test is passed, 1313 01:08:06,908 --> 01:08:10,825 you are dead center of the greatest scientific event 1314 01:08:10,869 --> 01:08:12,958 in the history of man. 1315 01:08:13,001 --> 01:08:14,612 If you've created a conscious machine, 1316 01:08:14,655 --> 01:08:17,615 it's not the history of man-- 1317 01:08:17,658 --> 01:08:19,356 That's the history of gods. 1318 01:08:19,399 --> 01:08:26,798 ♪♪ 1319 01:08:26,841 --> 01:08:28,452 Nolan: It's almost like technology is a god 1320 01:08:28,495 --> 01:08:29,975 in and of itself. 1321 01:08:30,018 --> 01:08:33,152 ♪♪ 1322 01:08:33,196 --> 01:08:35,241 Like the weather. We can't impact it. 1323 01:08:35,285 --> 01:08:39,593 We can't slow it down. We can't stop it. 1324 01:08:39,637 --> 01:08:43,249 We feel powerless. 1325 01:08:43,293 --> 01:08:44,685 Kurzweil: If we think of God 1326 01:08:44,729 --> 01:08:46,687 as an unlimited amount of intelligence, 1327 01:08:46,731 --> 01:08:48,167 the closest we can get to that 1328 01:08:48,211 --> 01:08:50,474 is by evolving our own intelligence 1329 01:08:50,517 --> 01:08:55,566 by merging with the artificial intelligence we're creating. 1330 01:08:55,609 --> 01:08:58,003 Musk: Today, our computers, phones, 1331 01:08:58,046 --> 01:09:01,615 applications give us superhuman capability. 1332 01:09:01,659 --> 01:09:04,662 So, as the old maxim says, if you can't beat 'em, join 'em. 1333 01:09:06,968 --> 01:09:09,971 el Kaliouby: It's about a human-machine partnership. 1334 01:09:10,015 --> 01:09:11,669 I mean, we already see how, you know, 1335 01:09:11,712 --> 01:09:14,933 our phones, for example, act as memory prosthesis, right? 1336 01:09:14,976 --> 01:09:17,196 I don't have to remember your phone number anymore 1337 01:09:17,240 --> 01:09:19,198 'cause it's on my phone. 1338 01:09:19,242 --> 01:09:22,070 It's about machines augmenting our human abilities, 1339 01:09:22,114 --> 01:09:25,248 as opposed to, like, completely displacing them. 1340 01:09:25,291 --> 01:09:27,380 Nolan: If you look at all the objects that have made the leap 1341 01:09:27,424 --> 01:09:30,122 from analog to digital over the last 20 years... 1342 01:09:30,166 --> 01:09:32,080 it's a lot. 1343 01:09:32,124 --> 01:09:35,388 We're the last analog object in a digital universe. 1344 01:09:35,432 --> 01:09:36,911 And the problem with that, of course, 1345 01:09:36,955 --> 01:09:40,567 is that the data input/output is very limited. 1346 01:09:40,611 --> 01:09:42,613 It's this. It's these. 1347 01:09:42,656 --> 01:09:45,355 Zilis: Our eyes are pretty good. 1348 01:09:45,398 --> 01:09:48,445 We're able to take in a lot of visual information. 1349 01:09:48,488 --> 01:09:52,536 But our information output is very, very, very low. 1350 01:09:52,579 --> 01:09:55,669 The reason this is important -- If we envision a scenario 1351 01:09:55,713 --> 01:09:59,543 where AI's playing a more prominent role in societies, 1352 01:09:59,586 --> 01:10:02,023 we want good ways to interact with this technology 1353 01:10:02,067 --> 01:10:04,983 so that it ends up augmenting us. 1354 01:10:05,026 --> 01:10:07,812 ♪♪ 1355 01:10:07,855 --> 01:10:09,553 Musk: I think it's incredibly important 1356 01:10:09,596 --> 01:10:12,295 that AI not be "other." 1357 01:10:12,338 --> 01:10:14,862 It must be us. 1358 01:10:14,906 --> 01:10:18,605 And I could be wrong about what I'm saying. 1359 01:10:18,649 --> 01:10:20,216 I'm certainly open to ideas 1360 01:10:20,259 --> 01:10:23,915 if anybody can suggest a path that's better. 1361 01:10:23,958 --> 01:10:27,266 But I think we're gonna really have to either merge with AI 1362 01:10:27,310 --> 01:10:28,963 or be left behind. 1363 01:10:29,007 --> 01:10:36,362 ♪♪ 1364 01:10:36,406 --> 01:10:38,756 Gourley: It's hard to kind of think of unplugging a system 1365 01:10:38,799 --> 01:10:41,802 that's distributed everywhere on the planet, 1366 01:10:41,846 --> 01:10:45,806 that's distributed now across the solar system. 1367 01:10:45,850 --> 01:10:49,375 You can't just, you know, shut that off. 1368 01:10:49,419 --> 01:10:51,290 Nolan: We've opened Pandora's box. 1369 01:10:51,334 --> 01:10:55,642 We've unleashed forces that we can't control, we can't stop. 1370 01:10:55,686 --> 01:10:57,296 We're in the midst of essentially creating 1371 01:10:57,340 --> 01:10:59,516 a new life-form on Earth. 1372 01:10:59,559 --> 01:11:05,826 ♪♪ 1373 01:11:05,870 --> 01:11:07,611 Russell: We don't know what happens next. 1374 01:11:07,654 --> 01:11:10,353 We don't know what shape the intellect of a machine 1375 01:11:10,396 --> 01:11:14,531 will be when that intellect is far beyond human capabilities. 1376 01:11:14,574 --> 01:11:17,360 It's just not something that's possible. 1377 01:11:17,403 --> 01:11:24,715 ♪♪ 1378 01:11:24,758 --> 01:11:26,934 The least scary future I can think of is one 1379 01:11:26,978 --> 01:11:29,633 where we have at least democratized AI. 1380 01:11:31,548 --> 01:11:34,159 Because if one company or small group of people 1381 01:11:34,202 --> 01:11:37,031 manages to develop godlike digital superintelligence, 1382 01:11:37,075 --> 01:11:40,339 they can take over the world. 1383 01:11:40,383 --> 01:11:42,210 At least when there's an evil dictator, 1384 01:11:42,254 --> 01:11:44,343 that human is going to die, 1385 01:11:44,387 --> 01:11:46,998 but, for an AI, there would be no death. 1386 01:11:47,041 --> 01:11:49,392 It would live forever. 1387 01:11:49,435 --> 01:11:51,916 And then you have an immortal dictator 1388 01:11:51,959 --> 01:11:53,570 from which we can never escape. 1389 01:11:53,613 --> 01:12:02,100 ♪♪ 1390 01:12:02,143 --> 01:12:10,587 ♪♪ 1391 01:12:10,630 --> 01:12:19,160 ♪♪ 1392 01:12:19,204 --> 01:12:27,647 ♪♪ 1393 01:12:27,691 --> 01:12:36,221 ♪♪ 1394 01:12:36,264 --> 01:12:44,838 ♪♪ 1395 01:12:51,845 --> 01:12:53,717 Woman on P.A.: Alan. Macchiato. 1396 01:13:10,951 --> 01:13:17,610 ♪♪ 1397 01:13:17,654 --> 01:13:24,269 ♪♪ 1398 01:13:24,312 --> 01:13:30,884 ♪♪ 1399 01:13:30,928 --> 01:13:37,543 ♪♪ 1400 01:13:37,587 --> 01:13:44,245 ♪♪ 1401 01:13:44,289 --> 01:13:50,817 ♪♪ 1402 01:13:50,861 --> 01:13:57,520 ♪♪ 1403 01:13:57,563 --> 01:14:04,178 ♪♪ 1404 01:14:04,222 --> 01:14:10,794 ♪♪ 1405 01:14:10,837 --> 01:14:17,496 ♪♪ 1406 01:14:17,540 --> 01:14:24,111 ♪♪ 1407 01:14:24,155 --> 01:14:30,727 ♪♪ 1408 01:14:30,770 --> 01:14:37,473 ♪♪ 1409 01:14:37,516 --> 01:14:44,044 ♪♪ 1410 01:14:44,088 --> 01:14:47,570 Woman: Hello? 1411 01:14:47,613 --> 01:14:54,838 ♪♪ 1412 01:14:54,881 --> 01:15:02,236 ♪♪ 1413 01:15:02,280 --> 01:15:09,548 ♪♪ 1414 01:15:09,592 --> 01:15:16,773 ♪♪ 1415 01:15:16,816 --> 01:15:18,688 ♪ Yeah, yeah 1416 01:15:18,731 --> 01:15:20,080 ♪ Yeah, yeah 1417 01:15:20,124 --> 01:15:27,261 ♪♪ 1418 01:15:27,305 --> 01:15:34,442 ♪♪ 1419 01:15:34,486 --> 01:15:41,580 ♪♪ 1420 01:15:41,624 --> 01:15:43,234 ♪ Yeah, yeah 1421 01:15:43,277 --> 01:15:45,541 ♪ Yeah, yeah 1422 01:15:45,584 --> 01:15:51,764 ♪♪ 1423 01:15:51,808 --> 01:15:57,988 ♪♪ 1424 01:15:58,031 --> 01:16:04,342 ♪♪ 1425 01:16:04,385 --> 01:16:10,609 ♪♪ 1426 01:16:10,653 --> 01:16:13,046 Hello? 1427 01:16:13,090 --> 01:16:22,012 ♪♪ 1428 01:16:22,055 --> 01:16:31,021 ♪♪ 1429 01:16:31,064 --> 01:16:39,986 ♪♪ 1430 01:16:40,030 --> 01:16:48,952 ♪♪ 1431 01:16:48,995 --> 01:16:57,961 ♪♪ 1432 01:16:58,004 --> 01:17:06,926 ♪♪ 1433 01:17:06,970 --> 01:17:15,892 ♪♪ 1434 01:17:15,935 --> 01:17:24,901 ♪♪ 1435 01:17:24,944 --> 01:17:33,910 ♪♪ 1436 01:17:33,953 --> 01:17:40,960 ♪♪