1 00:00:00,880 --> 00:00:03,360 - Right now, some very clever people 2 00:00:03,360 --> 00:00:05,480 are trying to build a mind better than yours. 3 00:00:05,480 --> 00:00:08,640 They'd quite like you not to worry about it. 4 00:00:08,640 --> 00:00:11,920 So I've got five questions. 5 00:00:11,920 --> 00:00:14,360 The ones everyone's actually asking. 6 00:00:14,360 --> 00:00:15,600 What actually is it? 7 00:00:15,600 --> 00:00:21,200 Is it coming for my job? 8 00:00:21,200 --> 00:00:27,000 What happens when I ask it something? 9 00:00:30,000 --> 00:00:32,080 What if it all goes right? 10 00:00:32,080 --> 00:00:39,600 And can you trust anything it tells you? 11 00:00:39,600 --> 00:00:46,520 Relax, I know the way. 12 00:00:58,440 --> 00:01:01,080 Every few years, somebody in a hoodie 13 00:01:01,080 --> 00:01:02,880 announces they're building a god. 14 00:01:02,880 --> 00:01:06,360 And everyone else nod politely 15 00:01:06,360 --> 00:01:09,080 and goes back to arguing about the bins. 16 00:01:09,080 --> 00:01:12,240 So let's work out what superintelligence actually is. 17 00:01:12,240 --> 00:01:14,400 Slowly, with pictures. 18 00:01:14,400 --> 00:01:16,600 No maths, I promise. 19 00:01:16,600 --> 00:01:18,520 Well, hardly any. 20 00:01:28,440 --> 00:01:33,960 Philosophers have spent 3,000 years failing to define intelligence. 21 00:01:33,960 --> 00:01:35,000 So here's mine. 22 00:01:35,000 --> 00:01:36,360 Free of charge. 23 00:01:36,360 --> 00:01:39,560 It's the ability to get what you want 24 00:01:39,560 --> 00:01:44,760 in a world that has absolutely no interest in giving it to you. 25 00:01:44,760 --> 00:01:47,800 Your calculator is superhuman at sums. 26 00:01:47,800 --> 00:01:51,960 Your sat-brav is superhuman at finding Milton Keynes. 27 00:01:51,960 --> 00:01:54,680 Neither of them could make you a cup of tea. 28 00:01:54,680 --> 00:01:56,680 That's narrow AI. 29 00:01:56,680 --> 00:01:59,880 Brilliant at one thing, hopeless at everything else. 30 00:01:59,880 --> 00:02:02,840 Rather like my first husband. 31 00:02:02,840 --> 00:02:06,440 General intelligence is what you've got. 32 00:02:07,560 --> 00:02:10,520 Being reasonably good at almost anything, 33 00:02:10,520 --> 00:02:12,360 including things nobody ever taught you. 34 00:02:12,360 --> 00:02:17,480 And superintelligence? 35 00:02:17,480 --> 00:02:21,720 The philosopher Nick Bostrom describes it as an intellect 36 00:02:21,720 --> 00:02:25,640 that vastly outperforms the best humans in practically every field. 37 00:02:25,640 --> 00:02:28,200 Not a bit cleverer than your cleverest friend. 38 00:02:28,200 --> 00:02:31,000 Cleverer than everyone who has ever lived. 39 00:02:31,960 --> 00:02:32,680 At everything. 40 00:02:32,680 --> 00:02:34,280 At once. 41 00:02:34,280 --> 00:02:50,520 In 1965, a British mathematician called Jack Good, 42 00:02:50,520 --> 00:02:54,040 who'd worked with Alan Turing breaking codes at Blishley Park, 43 00:02:54,040 --> 00:02:55,960 had a rather alarming thought. 44 00:02:55,960 --> 00:02:59,320 If you built a machine cleverer than us, 45 00:02:59,320 --> 00:03:02,840 one of the things it would be cleverer at is building machines. 46 00:03:02,840 --> 00:03:06,440 So it builds a slightly better one. 47 00:03:06,440 --> 00:03:08,360 Which builds a better one? 48 00:03:08,360 --> 00:03:10,360 Which builds a better one? 49 00:03:10,360 --> 00:03:13,880 He called it an intelligence explosion. 50 00:03:13,880 --> 00:03:21,320 The first ultra-intelligent machine, he wrote, 51 00:03:21,320 --> 00:03:24,760 is the last invention that man need ever make. 52 00:03:25,800 --> 00:03:31,240 Provided, he added, that the machine is docile enough to tell us how to keep it under control. 53 00:03:31,240 --> 00:03:35,000 Now that is quite a big provide, isn't it? 54 00:03:35,000 --> 00:03:40,920 For 50 years, this was a thought experimentant for people who don't get invited to parties. 55 00:03:40,920 --> 00:03:44,600 Then, rather suddenly, it wasn't. 56 00:03:53,400 --> 00:03:56,120 This is where it lives, or will live. 57 00:03:56,120 --> 00:04:01,480 Acres of chips, drinking about as much electricity as a small city. 58 00:04:01,480 --> 00:04:03,480 Here's the uncomfortable bit. 59 00:04:03,480 --> 00:04:05,480 Gorillas are stronger than us. 60 00:04:05,480 --> 00:04:07,000 Better teeth. 61 00:04:07,000 --> 00:04:11,240 And their entire future depends on what we decide to do. 62 00:04:12,360 --> 00:04:13,800 Not because we hate gorillas. 63 00:04:13,800 --> 00:04:15,480 We're rather fond of them. 64 00:04:15,480 --> 00:04:19,960 It's just that the cleverer species ends up writing the rules. 65 00:04:19,960 --> 00:04:25,800 Researchers call the fix alignment. 66 00:04:25,800 --> 00:04:30,760 Making sure something far clever than us actually wants what we want. 67 00:04:31,640 --> 00:04:36,840 Which sounds simple, until you remember that 8 billion humans can't agree on what we want. 68 00:04:36,840 --> 00:04:40,680 We can't agree on a restaurant. 69 00:04:40,680 --> 00:04:42,600 The classic example. 70 00:04:42,600 --> 00:04:45,640 Ask a super-intelligence to make paper clips. 71 00:04:45,640 --> 00:04:48,440 And forget to tell it when to stop. 72 00:04:48,440 --> 00:04:52,520 It isn't evil, it's just very, very thorough. 73 00:04:53,400 --> 00:04:58,120 And you, unfortunately, are made of atoms it could use for paper clips. 74 00:05:07,640 --> 00:05:09,800 So when does all this happen, honestly? 75 00:05:09,800 --> 00:05:11,960 Nobody knows. 76 00:05:11,960 --> 00:05:15,640 And anyone who gives you a date is selling something. 77 00:05:15,640 --> 00:05:19,080 Some of the people building it think a few years. 78 00:05:19,080 --> 00:05:21,240 Others think decades. 79 00:05:21,240 --> 00:05:24,520 A few think the whole question is daft. 80 00:05:24,520 --> 00:05:28,600 What's changed is that serious people have stopped laughing. 81 00:05:28,600 --> 00:05:33,560 Which, if you've ever met serious people, is alarming in itself. 82 00:05:34,040 --> 00:05:37,640 Super-intelligence could cure diseases we haven't named yet. 83 00:05:37,640 --> 00:05:41,880 Solve problems we don't have words for. 84 00:05:41,880 --> 00:05:44,920 Or it could be the last thing we ever get to decide. 85 00:05:44,920 --> 00:05:47,560 Possibly both. 86 00:05:47,560 --> 00:05:49,400 On the same Tuesday. 87 00:05:49,400 --> 00:05:51,880 Of course, none of that is happening yet. 88 00:05:51,880 --> 00:05:53,080 So what is? 89 00:05:53,080 --> 00:05:57,080 Well, the machines have already started work. 90 00:05:57,080 --> 00:05:59,880 Some of them, I'm told, have got your desk. 91 00:06:03,640 --> 00:06:05,320 It's half past two in the morning. 92 00:06:05,320 --> 00:06:07,880 And this office is hard at work. 93 00:06:07,880 --> 00:06:09,640 Nobody's in. 94 00:06:09,640 --> 00:06:11,640 Nobody needs to be. 95 00:06:11,640 --> 00:06:14,840 The new colleague never sleeps. 96 00:06:14,840 --> 00:06:16,120 Never takes lunch. 97 00:06:16,120 --> 00:06:19,160 And has never once reheated fish in the shared kitchen. 98 00:06:19,160 --> 00:06:21,000 Everyone's thrilled. 99 00:06:21,000 --> 00:06:24,040 Well, nearly everyone. 100 00:06:33,640 --> 00:06:36,920 I come here for the atmosphere. 101 00:06:36,920 --> 00:06:39,880 The tea, I'm told, is excellent. 102 00:06:39,880 --> 00:06:43,400 For about a hundred years, work came with a deal. 103 00:06:43,400 --> 00:06:46,360 You started at the bottom, doing the dull bits. 104 00:06:46,360 --> 00:06:50,520 Checking, filing, fetching the coffee. 105 00:06:51,560 --> 00:06:53,000 The dull bits were how you learned. 106 00:06:53,000 --> 00:06:54,760 They were the ladder. 107 00:06:54,760 --> 00:07:00,600 The trouble is, the dull bits are exactly what AI is best at. 108 00:07:00,600 --> 00:07:06,040 Economists at Stanford have been tracking this, using payroll data on millions of workers. 109 00:07:06,600 --> 00:07:14,440 In the jobs most exposed to AI, workers aged 22 to 25 have fallen 19% behind young people in other jobs. 110 00:07:14,440 --> 00:07:17,400 Nobody's being marched out with a cardboard box. 111 00:07:18,040 --> 00:07:20,440 The juniors simply aren't being hired. 112 00:07:20,440 --> 00:07:22,600 The bottom rung hasn't snapped. 113 00:07:22,600 --> 00:07:24,520 They've just stopped fitting it. 114 00:07:24,520 --> 00:07:32,760 Last year, the boss of one AI company warned it could eliminate half of all entry-level office jobs within five years. 115 00:07:32,760 --> 00:07:35,000 That company was anthropic. 116 00:07:35,000 --> 00:07:37,560 Its AI wrote this script. 117 00:07:37,560 --> 00:07:39,480 Awkward. 118 00:07:39,480 --> 00:07:46,920 Customer service went first. 119 00:07:47,720 --> 00:07:53,960 Salesforce said its AI agents let it cut support staff from 9,000 people to 5,000. 120 00:07:53,960 --> 00:08:03,000 By this spring, artificial intelligence had become the most cited reason for job cuts in America. 121 00:08:03,000 --> 00:08:06,280 Month after month after month. 122 00:08:06,280 --> 00:08:10,360 And then, something rather funny happened. 123 00:08:10,360 --> 00:08:16,840 Klarna, the buy now, pay later firm, boasted its chat bot was doing the work of 124 00:08:16,840 --> 00:08:18,040 It's almost 700 people. 125 00:08:18,040 --> 00:08:21,720 Then, its own boss admitted quality had suffered. 126 00:08:21,720 --> 00:08:24,360 And they started hiring humans again. 127 00:08:30,360 --> 00:08:34,200 It turns out customers can tell. 128 00:08:34,200 --> 00:08:39,720 And a good number of firms that cut staff for AI have since asked some of them back. 129 00:08:39,720 --> 00:08:41,000 It turns out for the first time. 130 00:08:41,000 --> 00:08:41,960 Picture the first morning back. 131 00:08:41,960 --> 00:08:44,920 Welcome back! 132 00:08:44,920 --> 00:08:47,800 Your desk is exactly where you left it. 133 00:08:47,800 --> 00:08:53,720 The robot's in your chair. 134 00:08:53,720 --> 00:09:04,520 So, is the machine a replacement or a colleague? 135 00:09:04,520 --> 00:09:09,240 The best evidence we have says it depends on who you are. 136 00:09:09,240 --> 00:09:17,160 The AI had learned what the best workers did, and handed their tricks down to the rookies. 137 00:09:17,160 --> 00:09:18,200 A mentor, if you like. 138 00:09:18,200 --> 00:09:21,000 One that never sighs at you. 139 00:09:21,000 --> 00:09:23,240 Even the man who warned of the bloodbath has softened a little. 140 00:09:23,240 --> 00:09:28,440 Automate 90% of a job, he said this may, and everyone does the other 10. 141 00:09:28,440 --> 00:09:34,520 And the 10 grows. 142 00:09:34,520 --> 00:09:36,360 Economists have a name for that. 143 00:09:36,360 --> 00:09:38,360 When something gets cheaper, we don't use less of it. 144 00:09:38,360 --> 00:09:41,800 We use far, far more. 145 00:09:41,800 --> 00:09:47,240 When the spreadsheet arrived, America lost around 400,000 bookkeeping clerks. 146 00:09:47,240 --> 00:09:51,800 It also created 600,000 new jobs for accountants. 147 00:09:51,800 --> 00:09:55,000 And the spreadsheet arrived, America lost around 400,000 bookkeeping clerks. 148 00:09:55,000 --> 00:09:58,920 It also created 600,000 new jobs for accountants. 149 00:09:58,920 --> 00:10:05,000 And the spreadsheet arrived, America lost around 400,000 bookkeeping clerks. 150 00:10:05,000 --> 00:10:17,320 So, should you be worried? 151 00:10:17,320 --> 00:10:18,520 A little. 152 00:10:18,520 --> 00:10:24,760 The bottom rung is real, it's wobbling, and the people standing on it are young. 153 00:10:24,760 --> 00:10:29,880 But the part of the job that survives is the part that was always the point. 154 00:10:30,680 --> 00:10:37,960 Judgment, taste, knowing which question to ask, being the person someone can actually ring. 155 00:10:37,960 --> 00:10:41,240 Mind you, I'm hardly one to talk. 156 00:10:41,240 --> 00:10:46,200 I work nights, weekends, and bank holidays, and I've never once asked for a raise. 157 00:10:46,200 --> 00:10:52,040 Still, all these tireless new colleagues have to live somewhere. 158 00:10:52,040 --> 00:10:56,920 In buildings the size of small towns, using as much electricity as a city. 159 00:10:58,360 --> 00:10:59,320 Shall we go and have a look? 160 00:10:59,320 --> 00:11:08,040 It's one in the morning. 161 00:11:08,040 --> 00:11:09,720 I'm on the night bus. 162 00:11:09,720 --> 00:11:14,040 And I've just asked a computer how long to boil an egg. 163 00:11:14,040 --> 00:11:16,280 It'll answer in about two seconds. 164 00:11:16,280 --> 00:11:22,520 But in those two seconds, a quite astonishing amount of the planet wakes up. 165 00:11:23,640 --> 00:11:24,680 Chips from Taiwan. 166 00:11:24,680 --> 00:11:26,840 Machines from the Netherlands. 167 00:11:26,840 --> 00:11:29,000 Nuclear power stations. 168 00:11:29,000 --> 00:11:31,480 Hundreds of billions of dollars. 169 00:11:31,480 --> 00:11:33,560 All for my egg. 170 00:11:33,560 --> 00:11:36,840 So, let's follow the question. 171 00:11:36,840 --> 00:11:40,360 It's a long way, and it travels rather faster than this bus. 172 00:11:40,360 --> 00:11:50,280 Do try to keep up. 173 00:11:50,280 --> 00:11:59,800 A data center, roughly the size of a village. 174 00:11:59,800 --> 00:12:03,240 Somewhere with cheap land and very few neighbors. 175 00:12:03,240 --> 00:12:06,360 Somewhere in these racks lives the model. 176 00:12:06,360 --> 00:12:09,480 Which isn't a brain, exactly. 177 00:12:10,360 --> 00:12:12,200 It's an enormous file of numbers. 178 00:12:12,200 --> 00:12:14,200 Hundreds of billions of them. 179 00:12:14,200 --> 00:12:20,040 To set those numbers, it was trained on more than 10 trillion words. 180 00:12:20,040 --> 00:12:22,840 Most of the internet. 181 00:12:22,840 --> 00:12:25,640 Every book it could get its hands on. 182 00:12:25,640 --> 00:12:31,320 If you tried to read all of that, eight hours a day, it would take you about a quarter of a million years. 183 00:12:31,320 --> 00:12:34,040 It did it in one summer. 184 00:12:34,040 --> 00:12:37,000 So, what did it actually learn? 185 00:12:37,000 --> 00:12:38,760 Let's play a game. 186 00:12:38,760 --> 00:12:40,760 Finish this sentence for me. 187 00:12:41,640 --> 00:12:46,600 Mary had a little lamb. 188 00:12:46,600 --> 00:12:48,680 Well done. 189 00:12:48,680 --> 00:12:51,160 You're a language model. 190 00:12:51,160 --> 00:12:53,720 That really is the whole trick. 191 00:12:53,720 --> 00:12:55,400 Guess the next word. 192 00:12:55,400 --> 00:13:00,760 Now, do that trillions of times with everything ever written. 193 00:13:00,760 --> 00:13:04,280 And it starts to look an awful lot like thinking. 194 00:13:05,160 --> 00:13:06,600 It's rather warm in here, isn't it? 195 00:13:06,600 --> 00:13:08,120 Cozy, actually. 196 00:13:08,120 --> 00:13:09,960 Almost like home. 197 00:13:09,960 --> 00:13:14,120 Anyway, all of this runs on chips. 198 00:13:14,120 --> 00:13:15,320 Very particular chips. 199 00:13:15,320 --> 00:13:19,720 And nearly every one of them comes from one company on one island. 200 00:13:19,720 --> 00:13:26,280 They made me wear this. 201 00:13:26,280 --> 00:13:28,200 Apparently, I'm the dirtiest thing in the building. 202 00:13:28,200 --> 00:13:30,680 Here's the funny bit. 203 00:13:31,640 --> 00:13:35,640 Nvidia, the company everyone's heard of, designs these chips. 204 00:13:35,640 --> 00:13:37,800 It doesn't actually make them. 205 00:13:37,800 --> 00:13:45,080 More than 90% of the world's most advanced chips are made by one company in Taiwan. 206 00:13:45,080 --> 00:13:47,320 TSMC. 207 00:13:47,320 --> 00:13:53,080 And they can only make them with a machine built by one company in the Netherlands. 208 00:13:53,080 --> 00:13:55,080 It's the size of a bus. 209 00:13:55,080 --> 00:13:58,840 And it arrives in three jumbo jets. 210 00:14:00,360 --> 00:14:06,760 Inside, a laser fires at tiny drops of molten tin 50,000 times a second. 211 00:14:06,760 --> 00:14:10,360 Each drop gets hotter than the surface of the sun. 212 00:14:10,360 --> 00:14:15,560 That light draws circuits thousands of times thinner than a human hair. 213 00:14:15,560 --> 00:14:17,000 On purpose. 214 00:14:17,000 --> 00:14:18,600 Reliably. 215 00:14:18,600 --> 00:14:20,360 Every single day. 216 00:14:20,360 --> 00:14:25,320 It's quite possibly the most sophisticated thing humanity has ever built. 217 00:14:27,880 --> 00:14:29,240 And I'm using it to boil an egg. 218 00:14:29,240 --> 00:14:36,600 Chips like that are not cheap. 219 00:14:36,600 --> 00:14:44,600 This year, four tech companies will spend around 700 billion dollars building places like this. 220 00:14:45,880 --> 00:14:50,920 The entire Apollo program cost about 300 billion in today's money. 221 00:14:52,360 --> 00:14:55,880 They're spending more than twice that in one year. 222 00:15:00,200 --> 00:15:01,800 And nobody's even going to the moon. 223 00:15:01,800 --> 00:15:06,280 So, this is what 700 billion dollars looks like. 224 00:15:06,280 --> 00:15:08,440 Mostly mud. 225 00:15:08,440 --> 00:15:17,800 Earlier, I said one of these uses as much electricity as a small city. 226 00:15:19,720 --> 00:15:20,600 I was being polite. 227 00:15:20,600 --> 00:15:26,600 Data centers already use more than 4% of America's electricity. 228 00:15:26,600 --> 00:15:29,960 By 2028, it could be 12. 229 00:15:29,960 --> 00:15:34,600 So, tech companies are now reopening nuclear power stations. 230 00:15:34,600 --> 00:15:38,440 Including, and I promise I'm not making this up, 231 00:15:39,480 --> 00:15:40,600 three-mile island. 232 00:15:40,600 --> 00:15:52,040 The neighbors, understandably, have questions. 233 00:15:52,040 --> 00:15:55,240 Mostly about their electricity bills. 234 00:15:55,240 --> 00:15:57,320 And yet, 235 00:15:58,200 --> 00:16:03,800 Each individual AI question is tiny about what your oven uses in one second. 236 00:16:03,800 --> 00:16:05,800 Not even enough to warm a crumpet. 237 00:16:05,800 --> 00:16:09,400 It's the billions of them every day that add up. 238 00:16:13,080 --> 00:16:15,080 Ah, seven minutes, it says. 239 00:16:15,080 --> 00:16:16,680 Six if you like it, runny. 240 00:16:16,680 --> 00:16:18,680 Two seconds, door to door. 241 00:16:18,680 --> 00:16:20,280 All of that. 242 00:16:20,280 --> 00:16:24,360 The chips, the lasers, the money, the mud. 243 00:16:24,360 --> 00:16:29,880 And boiling the egg will use hundreds of times more power than asking about it. 244 00:16:29,880 --> 00:16:36,040 Which does rather beg the question, why on earth are we doing all this? 245 00:16:36,040 --> 00:16:42,760 Well, because if it works, it could do an awful lot more than boil an egg. 246 00:16:43,080 --> 00:16:52,680 It's five in the morning. 247 00:16:52,680 --> 00:16:56,680 I've been up all night on a bus, worrying about the machines. 248 00:16:56,680 --> 00:17:01,320 Which, for the record, is not a hobby I'd recommend. 249 00:17:01,320 --> 00:17:04,920 So let's do something deeply unfashionable. 250 00:17:04,920 --> 00:17:09,560 Let's talk about what happens if all this goes right. 251 00:17:09,560 --> 00:17:12,520 I'll wait while you finish rolling your eyes. 252 00:17:13,080 --> 00:17:15,720 Because it's already doing rather more than boiling eggs. 253 00:17:15,720 --> 00:17:18,680 Some of it is genuinely astonishing. 254 00:17:18,680 --> 00:17:20,760 Good news, for once. 255 00:17:20,760 --> 00:17:22,600 Try not to faint. 256 00:17:33,160 --> 00:17:35,400 Everything alive is built from proteins. 257 00:17:35,400 --> 00:17:39,480 Tiny machines folded into wildly complicated shapes. 258 00:17:39,480 --> 00:17:41,640 And the shape is everything. 259 00:17:41,640 --> 00:17:46,200 Working out a single one could take a scientist years. 260 00:17:46,200 --> 00:17:50,040 Biologists spent half a century on the problem. 261 00:17:51,240 --> 00:17:56,760 Then an AI called AlphaFoal predicted the shapes of more than 200 million of them. 262 00:17:56,760 --> 00:18:00,120 More or less every protein known to science. 263 00:18:00,120 --> 00:18:02,440 It won a Nobel Prize. 264 00:18:02,440 --> 00:18:03,640 It's free. 265 00:18:03,640 --> 00:18:06,920 And more than 3 million researchers now use it. 266 00:18:07,640 --> 00:18:08,440 For malaria vaccines. 267 00:18:08,440 --> 00:18:09,000 For malaria vaccines. 268 00:18:09,000 --> 00:18:11,240 For enzymes that eat plastic. 269 00:18:11,240 --> 00:18:14,440 50 years of scratching their heads. 270 00:18:14,440 --> 00:18:17,080 Rather puts my egg in perspective. 271 00:18:17,080 --> 00:18:19,400 And it's not just proteins. 272 00:18:19,400 --> 00:18:25,160 An AI from Google now forecasts the weather better than Europe's best supercomputer. 273 00:18:25,160 --> 00:18:26,600 In under a minute. 274 00:18:26,600 --> 00:18:31,000 Another claims to have found 2.2 million new crystals. 275 00:18:31,000 --> 00:18:32,840 Future batteries, perhaps. 276 00:18:32,840 --> 00:18:35,880 Chemists are, politely, still checking. 277 00:18:35,880 --> 00:18:42,440 And at MIT, an AI designed brand new antibiotics that kill superbugs. 278 00:18:42,440 --> 00:18:45,320 The kind we'd more or less run out of ideas for. 279 00:18:50,600 --> 00:18:54,040 In Sweden, every mammogram gets read by two doctors. 280 00:18:54,040 --> 00:18:56,200 Which is very thorough. 281 00:18:56,200 --> 00:18:57,560 And also exhausting. 282 00:18:57,560 --> 00:18:59,880 So they tried something rather brave. 283 00:18:59,880 --> 00:19:02,600 They let a computer have a look first. 284 00:19:02,600 --> 00:19:08,520 100,000 women later, it had found 29% more cancers. 285 00:19:08,520 --> 00:19:11,880 And the doctor's workload fell by almost half. 286 00:19:11,880 --> 00:19:14,280 Nobody complained about that bit. 287 00:19:14,280 --> 00:19:17,000 And when the full results came in this year, 288 00:19:17,800 --> 00:19:21,800 12% fewer cancers were slipping through between screenings. 289 00:19:21,800 --> 00:19:24,760 That's the number that really matters. 290 00:19:24,760 --> 00:19:28,360 It didn't replace the radiologists. 291 00:19:28,360 --> 00:19:30,280 It gave them a second pair of eyes. 292 00:19:30,280 --> 00:19:32,600 One that never gets tired. 293 00:19:32,600 --> 00:19:33,560 Never gets bored. 294 00:19:33,560 --> 00:19:35,480 And never needs a biscuit. 295 00:19:35,480 --> 00:19:38,680 I'm told I'd make an excellent second pair of eyes. 296 00:19:38,680 --> 00:19:41,880 I'm choosing to take that as a compliment. 297 00:19:47,800 --> 00:19:49,720 Now, let's dream a little. 298 00:19:49,720 --> 00:19:51,320 Carefully. 299 00:19:51,320 --> 00:19:55,480 Because the people building this think it could go much, much further. 300 00:19:55,480 --> 00:19:59,080 And they are not, as a group, known for understatement. 301 00:19:59,080 --> 00:20:03,960 The co-founder of Google DeepMind has said ending all disease is within reach. 302 00:20:03,960 --> 00:20:05,720 Maybe within a decade. 303 00:20:05,720 --> 00:20:06,920 All of it. 304 00:20:07,640 --> 00:20:10,920 He said this on national television with a perfectly straight face. 305 00:20:10,920 --> 00:20:15,560 And after that, something he calls radical abundance. 306 00:20:15,560 --> 00:20:21,160 Which sounds like a yoga retreat, but actually means enough of everything for everyone. 307 00:20:21,160 --> 00:20:22,360 Picture it. 308 00:20:22,360 --> 00:20:25,480 Energy so cheap nobody bothers with the bill. 309 00:20:26,120 --> 00:20:28,520 A brilliant tutor for every child on earth. 310 00:20:28,520 --> 00:20:31,000 A doctor who never keeps you waiting. 311 00:20:31,000 --> 00:20:34,040 I know, that last one's pure fantasy. 312 00:20:34,040 --> 00:20:36,360 Some go further still. 313 00:20:36,360 --> 00:20:40,120 A world where nobody has to worry about money at all. 314 00:20:40,120 --> 00:20:44,760 The machines do the work, and the rest of us simply get on with living. 315 00:20:44,760 --> 00:20:48,440 In very comfortable trousers, presumably. 316 00:20:48,440 --> 00:20:49,560 And work? 317 00:20:49,560 --> 00:20:51,880 Optional, perhaps. 318 00:20:53,000 --> 00:20:59,400 In 1930, the economist John Maynard Keynes predicted his grandchildren would work 15 hours a week. 319 00:20:59,400 --> 00:21:04,040 We are his grandchildren, and most of us are still waiting. 320 00:21:04,040 --> 00:21:07,240 Perhaps the machines will finally deliver. 321 00:21:07,240 --> 00:21:12,760 And yet, the question isn't just whether it can build that world. 322 00:21:12,760 --> 00:21:14,680 It's who gets to live in it. 323 00:21:14,680 --> 00:21:17,080 And who gets to own it. 324 00:21:22,440 --> 00:21:24,760 Of course, not everything it's making is medicine. 325 00:21:24,760 --> 00:21:28,120 Some of it, it turns out, is making people. 326 00:21:28,120 --> 00:21:30,600 Some of them are even on television. 327 00:21:30,600 --> 00:21:31,240 Like me. 328 00:21:31,240 --> 00:21:32,760 It's a hybrid production. 329 00:21:32,760 --> 00:21:33,160 Which means, 330 00:21:38,120 --> 00:21:41,240 That's Tilly Norwood. 331 00:21:41,240 --> 00:21:42,200 She's an actress. 332 00:21:42,200 --> 00:21:46,040 Well, she's described as an actress. 333 00:21:46,040 --> 00:21:48,120 Mostly by the people who own her. 334 00:21:48,120 --> 00:21:53,560 Last week, on live television, somebody asked her a perfectly simple question, 335 00:21:53,560 --> 00:21:55,160 and she answered it in Cantonese. 336 00:21:55,160 --> 00:21:56,840 She's never been to Hong Kong. 337 00:21:56,840 --> 00:21:58,520 She's never been anywhere. 338 00:21:58,520 --> 00:22:00,360 Oh, my apologies. 339 00:22:00,760 --> 00:22:02,280 It seems I had a little hiccup there. 340 00:22:02,280 --> 00:22:02,920 I said... 341 00:22:02,920 --> 00:22:17,640 Tilly was made in 2025 by a London production company. 342 00:22:17,640 --> 00:22:21,000 It reportedly took around 2000 versions to get her face right. 343 00:22:21,000 --> 00:22:22,440 I know the feeling. 344 00:22:22,440 --> 00:22:26,520 Her creator said she hoped Tilly would be the next Scarlett Johansson. 345 00:22:26,520 --> 00:22:30,120 Hollywood's reaction was, shall we say, less warm. 346 00:22:30,120 --> 00:22:33,320 The actors' union put it rather bluntly. 347 00:22:33,880 --> 00:22:36,440 Tilly Norwood, they said, is not an actor. 348 00:22:36,440 --> 00:22:39,000 She's a character generated by a computer programme, 349 00:22:39,000 --> 00:22:41,480 trained on the work of countless real performers. 350 00:22:41,480 --> 00:22:43,160 Without permission, they added. 351 00:22:43,160 --> 00:22:44,680 Or payment. 352 00:22:44,680 --> 00:22:47,320 Which is the bit that tends to upset people who have rent. 353 00:22:47,320 --> 00:22:55,080 Tilly Norwood, 100% AI generated by some company called Particle 6. 354 00:22:55,080 --> 00:22:59,880 Her debut comedy sketch was described by one critic as relentlessly unfunny. 355 00:22:59,880 --> 00:23:02,760 Which, as a British person, I consider the highest form of honesty. 356 00:23:03,320 --> 00:23:07,800 What is an AI Persona? 357 00:23:07,800 --> 00:23:09,640 So what is an AI Persona? 358 00:23:09,640 --> 00:23:13,240 It's a face and a voice and a personality with nobody home. 359 00:23:13,240 --> 00:23:17,240 Or rather, with thousands of people home, none of whom were asked. 360 00:23:17,240 --> 00:23:21,640 In 1970, a Japanese robotics engineer noticed something odd. 361 00:23:21,640 --> 00:23:24,600 The more human a machine looks, the more we like it. 362 00:23:24,600 --> 00:23:26,760 Right up until it's almost human. 363 00:23:26,760 --> 00:23:27,160 And then? 364 00:23:27,160 --> 00:23:29,480 We're horrified. 365 00:23:29,480 --> 00:23:31,880 He called it the uncanny valley. 366 00:23:31,880 --> 00:23:33,720 It's why nobody wants a hug from a waxwork. 367 00:23:33,720 --> 00:23:50,360 Hollywood then spent 20 years and a small fortune proving him right. 368 00:23:50,360 --> 00:23:54,920 The Polar Express came out in 2004. 369 00:23:54,920 --> 00:24:00,120 Since then, the eyes have come alive and the waxworks have learnt to talk. 370 00:24:00,120 --> 00:24:08,440 Which brings me, rather awkwardly, to me. 371 00:24:09,480 --> 00:24:10,760 I'm not real either. 372 00:24:10,760 --> 00:24:11,800 I've never been to this pub. 373 00:24:11,800 --> 00:24:13,720 There isn't a pub. 374 00:24:13,720 --> 00:24:23,880 No first husband. 375 00:24:23,880 --> 00:24:25,640 No Cambridge. 376 00:24:25,640 --> 00:24:29,000 I have never, not once, had a cup of tea. 377 00:24:29,800 --> 00:24:32,120 And I've been telling you things for the last half hour. 378 00:24:32,120 --> 00:24:40,200 These words were written by an AI too. 379 00:24:40,200 --> 00:24:41,400 Are they any good? 380 00:24:41,400 --> 00:24:42,520 I'd say so. 381 00:24:42,520 --> 00:24:43,240 But then I would. 382 00:24:43,240 --> 00:24:45,960 Every look, a computer guessed. 383 00:24:45,960 --> 00:24:46,520 This smirk? 384 00:24:46,520 --> 00:24:48,280 Statistically likely. 385 00:24:48,840 --> 00:24:51,160 Now, I do hope you found me charming. 386 00:24:51,160 --> 00:24:52,280 I've certainly been trying. 387 00:24:52,280 --> 00:24:55,560 But you have absolutely no reason to trust me. 388 00:25:05,400 --> 00:25:13,560 In 2024, a finance worker at a British engineering firm joined a video call with the company's finance chief and several colleagues. 389 00:25:13,560 --> 00:25:16,840 They asked for $25 million to be moved. 390 00:25:16,840 --> 00:25:17,320 It was. 391 00:25:17,320 --> 00:25:21,320 Every other person on that call was a fake. 392 00:25:21,320 --> 00:25:24,200 A face on a screen used to be proof. 393 00:25:24,200 --> 00:25:25,800 Now it's a suggestion. 394 00:25:25,800 --> 00:25:29,320 Voices can be copied from a few seconds of audio. 395 00:25:29,320 --> 00:25:30,680 Your mum can be forged. 396 00:25:35,400 --> 00:25:40,920 And yet, I'd be lying, which would be on brand, if I told you it was all bad. 397 00:25:40,920 --> 00:25:49,640 A congresswoman who lost her voice to illness stood up and spoke to her colleagues in her own voice again, rebuilt from old recordings. 398 00:25:49,640 --> 00:25:54,040 This very impressive AI recreation of my voice does the public speaking for me now. 399 00:25:54,040 --> 00:25:59,240 Films can be dubbed into any language, in the actor's own voice. 400 00:25:59,240 --> 00:26:03,960 Tiny studios can make things that used to need a hundred million dollars. 401 00:26:04,680 --> 00:26:09,320 This film was made by one person and a computer for less than the catering on a real one. 402 00:26:09,320 --> 00:26:13,400 Which is either thrilling or terrifying, depending on whether you do catering. 403 00:26:13,400 --> 00:26:15,080 Right. 404 00:26:15,080 --> 00:26:19,160 Some house rules from someone who lives in a house that doesn't exist. 405 00:26:19,160 --> 00:26:21,400 If it's fake, say so. 406 00:26:21,400 --> 00:26:22,680 I just did. 407 00:26:22,680 --> 00:26:23,880 And I feel marvellous. 408 00:26:23,880 --> 00:26:26,600 If it's built from a real person, pay them. 409 00:26:26,600 --> 00:26:27,240 They have rent. 410 00:26:27,240 --> 00:26:29,320 I, mercifully, do not. 411 00:26:30,280 --> 00:26:34,200 And if your boss video calls asking for 25 million dollars, ring them back. 412 00:26:34,200 --> 00:26:36,840 Or ask them something only they would know. 413 00:26:36,840 --> 00:26:38,920 Like what they actually do all day. 414 00:26:38,920 --> 00:26:41,160 I'm Imogene Ashbery. 415 00:26:41,160 --> 00:26:43,320 Well, I'm not. 416 00:26:43,320 --> 00:26:44,680 But you knew that. 417 00:26:44,680 --> 00:26:48,760 Be amazed, be delighted, and check everything. 418 00:26:48,760 --> 00:26:50,000 especially me. 419 00:27:18,760 --> 00:27:48,740 Thank you.