1 00:00:03,200 --> 00:00:06,560 You might not think about it very often, but we are living 2 00:00:06,560 --> 00:00:08,720 through a golden age of technology. 3 00:00:09,720 --> 00:00:14,840 Our homes and our pockets are full of inventions and gadgets that 20 4 00:00:14,840 --> 00:00:17,840 years ago would have been the stuff of science fiction. 5 00:00:17,840 --> 00:00:20,680 Alexa, how exactly do you work? 6 00:00:20,680 --> 00:00:22,480 Hmm. I don't know that. 7 00:00:22,480 --> 00:00:24,680 That's what we're going to find out. 8 00:00:24,680 --> 00:00:27,400 These things have become such an integral part of our everyday 9 00:00:27,400 --> 00:00:31,160 lives that we've forgotten to remember how clever they are. 10 00:00:31,160 --> 00:00:34,920 I'm Professor Hannah Fry, and I want to open your eyes... 11 00:00:34,920 --> 00:00:37,040 I'm having a great time. 12 00:00:37,040 --> 00:00:40,320 ..to the secret genius of everyday objects. 13 00:00:40,320 --> 00:00:41,680 CARD READER BEEPS 14 00:00:41,680 --> 00:00:44,280 The ones that we take for granted. 15 00:00:44,280 --> 00:00:46,280 Items that owe their existence 16 00:00:46,280 --> 00:00:49,120 to generations of scientific innovation. 17 00:00:49,120 --> 00:00:52,680 THE Edison drove an electric car?! Yeah, this one. 18 00:00:53,840 --> 00:00:57,680 In every episode, I'm going to take one item... 19 00:00:58,920 --> 00:01:00,440 ..and tear it apart... 20 00:01:00,440 --> 00:01:01,960 ..bit... 21 00:01:01,960 --> 00:01:03,160 Oh! 22 00:01:03,160 --> 00:01:04,240 ..by bit... 23 00:01:04,240 --> 00:01:06,400 So now there's some stuff you can see. 24 00:01:06,400 --> 00:01:07,760 ..meet the people behind it... 25 00:01:07,760 --> 00:01:10,280 Frankly, I thought many things were impossible. 26 00:01:10,280 --> 00:01:11,960 ..and unlock the remarkable... 27 00:01:11,960 --> 00:01:14,960 Here is the clandestine listening device. 28 00:01:14,960 --> 00:01:16,720 ..and sometimes hilarious stories... 29 00:01:17,920 --> 00:01:19,840 ..behind its invention. 30 00:01:19,840 --> 00:01:21,560 Hello. Look at you. 31 00:01:24,760 --> 00:01:27,440 This time, it's the virtual assistant. 32 00:01:29,840 --> 00:01:33,360 More and more of us now have devices that allow us to control things 33 00:01:33,360 --> 00:01:36,600 in our home using just our voices. 34 00:01:36,600 --> 00:01:38,920 Alexa, how are you doing? 35 00:01:38,920 --> 00:01:40,520 I'm feeling perplexed. 36 00:01:40,520 --> 00:01:44,400 For some reason, people keep asking if I fancy a cuppa. 37 00:01:44,400 --> 00:01:48,440 It's all thanks to the secret tech used to hunt U-boats... 38 00:01:49,520 --> 00:01:53,040 ..a 1940s star who had a double life as an inventor... 39 00:01:53,040 --> 00:01:55,520 I could promise you so many things. 40 00:01:57,080 --> 00:02:00,160 ..and a 1960s sci-fi villain. 41 00:02:00,160 --> 00:02:02,040 I'm sorry, Dave. 42 00:02:02,040 --> 00:02:04,160 I'm afraid I can't do that. 43 00:02:11,320 --> 00:02:14,320 This is Amazon's swanky London HQ. 44 00:02:15,800 --> 00:02:17,440 It's quite cool in here, isn't it? 45 00:02:19,280 --> 00:02:21,680 I'm here to meet the people behind the world's 46 00:02:21,680 --> 00:02:24,320 best-selling virtual assistant - 47 00:02:24,320 --> 00:02:25,880 the Amazon Alexa. 48 00:02:28,360 --> 00:02:31,480 And this is a story that starts with a sketch. 49 00:02:31,480 --> 00:02:35,880 I think this actually gives you the best view, and it shows 50 00:02:35,880 --> 00:02:38,680 the drawing Jeff drew on the whiteboard. 51 00:02:39,760 --> 00:02:42,320 That's Rohit Prasad, head of Alexa. 52 00:02:45,080 --> 00:02:48,680 So this is Jeff's handwriting? I think so. Yes. 53 00:02:48,680 --> 00:02:50,320 But this is the original sketch. 54 00:02:51,360 --> 00:02:55,920 Alexa was conceived in 2011 by Amazon founder Jeff Bezos. 55 00:02:57,280 --> 00:03:00,040 The hasty sketch on a whiteboard was his vision 56 00:03:00,040 --> 00:03:02,240 of a virtual assistant butler. 57 00:03:03,400 --> 00:03:05,880 They were dreaming about a Star Trek computer. 58 00:03:05,880 --> 00:03:08,000 Isn't "mic" spelt M-I-C? Yeah! 59 00:03:08,000 --> 00:03:09,360 THEY LAUGH 60 00:03:09,360 --> 00:03:12,240 I know. I really love the idea that there are some people in the world 61 00:03:12,240 --> 00:03:15,760 who can just scribble something like that on a whiteboard, 62 00:03:15,760 --> 00:03:17,800 and be like, "I've made the invention. 63 00:03:17,800 --> 00:03:20,400 "Now you go and sort out the details." 64 00:03:20,400 --> 00:03:22,040 But Alexa isn't alone. 65 00:03:22,040 --> 00:03:23,360 Hang on a minute. 66 00:03:23,360 --> 00:03:26,760 There's now a virtual assistant in around half of all UK homes, 67 00:03:26,760 --> 00:03:30,560 and a wide range of different types, like Apple Siri... 68 00:03:30,560 --> 00:03:32,160 Hello, there, viewer. 69 00:03:32,160 --> 00:03:33,600 ..and Google Assistant. 70 00:03:33,600 --> 00:03:34,960 Nice to see you. 71 00:03:38,400 --> 00:03:40,760 I'm going to take apart the virtual assistant 72 00:03:40,760 --> 00:03:42,880 and look at five key components. 73 00:03:43,880 --> 00:03:47,320 From its microphones to its microprocessor. 74 00:03:47,320 --> 00:03:50,600 But first up is the part that is most familiar. 75 00:03:50,600 --> 00:03:53,000 It's where speaker meets software... 76 00:03:54,640 --> 00:03:56,640 ..to create Alexa's iconic voice. 77 00:03:59,960 --> 00:04:01,680 Here it is, then. 78 00:04:01,680 --> 00:04:03,080 An Alexa. 79 00:04:03,080 --> 00:04:07,080 Alexa, do you mind if I use a screwdriver to open you? 80 00:04:08,280 --> 00:04:10,880 Sorry, I'm not sure about that. Hmm. 81 00:04:12,640 --> 00:04:14,320 Well, I'm sorry, Alexa. 82 00:04:14,320 --> 00:04:17,800 It's not your lucky day, because I have a very impressive 83 00:04:17,800 --> 00:04:21,640 array of tools, and I know how to use them. 84 00:04:21,640 --> 00:04:23,560 Oh! There we go. 85 00:04:23,560 --> 00:04:27,240 I want to find out where this voice comes from. 86 00:04:30,040 --> 00:04:33,600 And you can start to see what is within. 87 00:04:33,600 --> 00:04:35,240 And there it is. 88 00:04:36,560 --> 00:04:39,120 This little speaker. At its heart, 89 00:04:39,120 --> 00:04:42,200 this is... This works the same way as a normal speaker does. 90 00:04:42,200 --> 00:04:46,280 It's just got a little membrane, some magnets that pull that membrane 91 00:04:46,280 --> 00:04:50,680 back and forth to vibrate the air, according to whatever electrical 92 00:04:50,680 --> 00:04:53,960 signal is being passed along this wire. 93 00:04:53,960 --> 00:04:58,560 And as for what electrical signal is passed along the wire, 94 00:04:58,560 --> 00:05:01,320 that all comes down to the other 95 00:05:01,320 --> 00:05:03,760 side, and the brains of the device. 96 00:05:03,760 --> 00:05:05,680 This is a microprocessor. 97 00:05:05,680 --> 00:05:08,600 That's the key to its processing power. 98 00:05:08,600 --> 00:05:10,080 We'll do more on that later. 99 00:05:10,080 --> 00:05:13,720 But it still needs a clever computer program to make it speak. 100 00:05:13,720 --> 00:05:17,160 We've moved far more towards giving options, letting people have a male 101 00:05:17,160 --> 00:05:18,400 voice if they want one, 102 00:05:18,400 --> 00:05:20,040 a female voice as well. 103 00:05:18,400 --> 00:05:20,040 EMAIL PINGS 104 00:05:20,040 --> 00:05:22,600 And I've got an email. That's good to know. 105 00:05:22,600 --> 00:05:24,840 Thank you, laptop. 106 00:05:24,840 --> 00:05:26,320 That is Trevor Wood. 107 00:05:26,320 --> 00:05:28,640 He's Alexa's senior speech scientist. 108 00:05:28,640 --> 00:05:32,720 And his day job is to try and make Alexa sound as natural as possible. 109 00:05:34,480 --> 00:05:38,080 When you first wanted to create a voice that could be part 110 00:05:38,080 --> 00:05:40,080 of an assistant like this, 111 00:05:40,080 --> 00:05:41,880 how do you go about that? 112 00:05:41,880 --> 00:05:44,640 We need Alexa to be able to say any possible sentence. 113 00:05:44,640 --> 00:05:47,840 So, to build a voice in the old days, and the old days means 114 00:05:47,840 --> 00:05:50,120 before about 2017, 115 00:05:50,120 --> 00:05:53,400 we used to record lots of data with a voice actor. 116 00:05:53,400 --> 00:05:56,200 They recorded every sound in the English language 117 00:05:56,200 --> 00:05:59,280 so that they could build any word that Alexa might have to say. 118 00:05:59,280 --> 00:06:02,160 So you can think of it as like getting you to read every single 119 00:06:02,160 --> 00:06:03,240 word in the dictionary. 120 00:06:03,240 --> 00:06:05,640 And then when we've got all of the words, we can string them 121 00:06:05,640 --> 00:06:07,440 together to make any sentence we like. 122 00:06:07,440 --> 00:06:10,680 We need tens of hours of data just to be able to build 123 00:06:10,680 --> 00:06:11,800 a single voice. 124 00:06:11,800 --> 00:06:15,480 This time-consuming technique worked in the first iteration of the voice, 125 00:06:15,480 --> 00:06:19,120 but the ups and downs of natural speech weren't quite right. 126 00:06:19,120 --> 00:06:22,240 ROBOTIC FEMALE VOICE: Star Trek is an American media franchise based on 127 00:06:22,240 --> 00:06:26,440 the science fiction television series created by Gene Roddenberry. 128 00:06:26,440 --> 00:06:29,560 That is how Alexa sounded back in 2014. 129 00:06:29,560 --> 00:06:31,240 Compare that to today. 130 00:06:31,240 --> 00:06:34,760 NATURAL FEMALE VOICE: Star Trek is an American media franchise based on 131 00:06:34,760 --> 00:06:38,640 the science fiction television series created by Gene Roddenberry. 132 00:06:38,640 --> 00:06:41,160 That is a pretty impressive improvement 133 00:06:41,160 --> 00:06:42,720 in just the last eight years. 134 00:06:42,720 --> 00:06:46,560 But how do virtual assistants produce such lifelike voices? 135 00:06:46,560 --> 00:06:49,200 The answer lies deep in the past. 136 00:06:50,280 --> 00:06:52,400 The year is 1939. 137 00:06:55,000 --> 00:06:57,400 No, not that 1939. 138 00:06:57,400 --> 00:06:58,920 Not yet, anyway. 139 00:06:58,920 --> 00:07:00,320 THIS one. 140 00:07:02,280 --> 00:07:03,680 3,000 miles away 141 00:07:03,680 --> 00:07:06,960 in New York, with minds free of the threat of war, 142 00:07:06,960 --> 00:07:10,040 people were able to celebrate the technological advancements 143 00:07:10,040 --> 00:07:12,280 of the era with a World Fair. 144 00:07:13,560 --> 00:07:15,880 Isn't it wonderful, Nick? Beautiful. 145 00:07:15,880 --> 00:07:17,240 What do you want to do? 146 00:07:17,240 --> 00:07:20,440 Anything but inspect this temple of capitalism. Oh, Nick! 147 00:07:20,440 --> 00:07:22,840 Some of the exhibits were pretty crazy. 148 00:07:22,840 --> 00:07:25,600 Yes, that's a smoking robot. And no, you can't buy one. 149 00:07:25,600 --> 00:07:28,160 And, folks, he's only two years old. 150 00:07:30,520 --> 00:07:33,400 However, there was one exhibit that was stealing 151 00:07:33,400 --> 00:07:35,120 all of the attention. 152 00:07:35,120 --> 00:07:37,880 It might look like a futuristic church organ, 153 00:07:37,880 --> 00:07:39,920 but it had a very special talent. 154 00:07:39,920 --> 00:07:41,840 It could speak. 155 00:07:41,840 --> 00:07:45,440 Helen, will you have the Voder say "greetings, everybody"? 156 00:07:45,440 --> 00:07:47,640 ROBOTIC VOICE: Greetings, everybody. 157 00:07:47,640 --> 00:07:50,720 It was the first device that could generate continuous human 158 00:07:50,720 --> 00:07:53,120 sounding speech electronically. 159 00:07:53,120 --> 00:07:54,600 ROBOTIC VOICE: Ha, ha, ha. 160 00:07:54,600 --> 00:07:58,800 It was called the Voice Operating Demonstrator, or Voder... 161 00:07:58,800 --> 00:08:00,840 ROBOTIC VOICE: A, B, C... 162 00:08:00,840 --> 00:08:04,600 ..and synthesised speech was born. 163 00:08:04,600 --> 00:08:08,440 Invented by telephone pioneer Homer Dudley of Bell Labs, 164 00:08:08,440 --> 00:08:13,160 it reproduced the basic building blocks of human speech. 165 00:08:13,160 --> 00:08:17,520 The ethereal robotic voice could speak, sing and even change pitch. 166 00:08:18,840 --> 00:08:21,080 HIGH-PITCHED VOICE 167 00:08:21,080 --> 00:08:23,040 VOICE DROPS IN PITCH 168 00:08:24,000 --> 00:08:26,560 Ultimately, Dudley's Voder was too cumbersome 169 00:08:26,560 --> 00:08:28,240 for any useful purpose. 170 00:08:31,600 --> 00:08:34,920 The outbreak of World War II set the stage for the next big 171 00:08:34,920 --> 00:08:37,840 development in synthetic voice technology. 172 00:08:41,680 --> 00:08:44,960 The British government needed a secure way to communicate 173 00:08:44,960 --> 00:08:46,960 with their American allies. 174 00:08:46,960 --> 00:08:51,000 As soon as you want to communicate across the Atlantic, you have to use 175 00:08:51,000 --> 00:08:54,920 something like radio waves, at which point... 176 00:08:54,920 --> 00:08:56,240 ..anyone can listen in. 177 00:08:57,400 --> 00:09:00,440 Obviously, that's not ideal, but the solution would come 178 00:09:00,440 --> 00:09:03,680 from our old pal Homer Dudley and work that he'd done 179 00:09:03,680 --> 00:09:06,240 on undersea telegram cables. 180 00:09:06,240 --> 00:09:10,520 We've got all of these cables that are laid across the Atlantic. 181 00:09:10,520 --> 00:09:14,200 The problem is, is that the amount of information that you can transmit 182 00:09:14,200 --> 00:09:18,440 down one of these cables is absolutely minuscule. 183 00:09:18,440 --> 00:09:21,560 So, maybe enough for a telegram, but definitely not enough for a 184 00:09:21,560 --> 00:09:26,960 phone call, because there is way too much information contained in a 185 00:09:26,960 --> 00:09:29,120 recording of the human voice. 186 00:09:32,240 --> 00:09:35,920 Dudley built a bit of tech called the vocoder. 187 00:09:35,920 --> 00:09:41,080 It compressed the voice enough to transmit by undersea cable. 188 00:09:41,080 --> 00:09:44,160 It wasn't implemented until the Allies realised 189 00:09:44,160 --> 00:09:46,480 that it gave them a way to encrypt voices 190 00:09:46,480 --> 00:09:48,880 and prevent enemy eavesdropping. 191 00:09:48,880 --> 00:09:53,640 Sound is just essentially air wobbling. 192 00:09:53,640 --> 00:09:57,000 When we speak, our vocal cords vibrate, creating thousands 193 00:09:57,000 --> 00:09:59,840 of different wobbles in the air. 194 00:09:59,840 --> 00:10:04,600 So what Homer Dudley realised was that you don't need to hear every 195 00:10:04,600 --> 00:10:06,240 single one of those wobbles. 196 00:10:06,240 --> 00:10:10,960 So he worked out a way to get rid of almost all of them and take 197 00:10:10,960 --> 00:10:14,480 the most important, some high-pitched, some low-pitched, 198 00:10:14,480 --> 00:10:17,640 and only transmit that much smaller 199 00:10:17,640 --> 00:10:21,320 message down the cable, across the Atlantic. 200 00:10:21,320 --> 00:10:25,120 And that is what the electronics of his vocoder did. 201 00:10:25,120 --> 00:10:30,080 Homer Dudley shrunk down the human voice so that it could finally fit. 202 00:10:30,080 --> 00:10:33,400 Now, the thing is, is that you are left with something 203 00:10:33,400 --> 00:10:37,440 that you can understand, but it is something 204 00:10:37,440 --> 00:10:39,840 without any of the nuance of human speech. 205 00:10:41,560 --> 00:10:45,000 ROBOTIC VOICE: Something without any of the nuance of human speech. 206 00:10:46,560 --> 00:10:48,720 Quite robotic, isn't it? 207 00:10:48,720 --> 00:10:50,880 But that didn't bother the allied forces. 208 00:10:50,880 --> 00:10:55,120 The compressed signal could easily be encrypted to keep messages secure 209 00:10:55,120 --> 00:10:57,000 and sent via radio waves. 210 00:10:57,000 --> 00:11:00,480 ARCHIVE: The magic of physics converting sound 211 00:11:00,480 --> 00:11:03,080 into electrical impulses. 212 00:11:03,080 --> 00:11:06,120 The vocoder was codenamed SIGSALY, 213 00:11:06,120 --> 00:11:09,640 and the US Army commissioned Bell Labs to build 12 of them. 214 00:11:09,640 --> 00:11:13,400 The role of the vocoder in securing secret communications 215 00:11:13,400 --> 00:11:17,000 between Churchill and President Roosevelt almost certainly helped 216 00:11:17,000 --> 00:11:18,440 to secure a victory. 217 00:11:20,000 --> 00:11:22,800 ARCHIVE: Now the world was truly ready for peace. 218 00:11:22,800 --> 00:11:25,120 And it was the first time that humans were able 219 00:11:25,120 --> 00:11:29,160 to electronically recreate speech in real time. 220 00:11:29,160 --> 00:11:31,880 By the late 1970s and '80s, the vocoder made 221 00:11:31,880 --> 00:11:33,800 a resurgence once again. 222 00:11:33,800 --> 00:11:37,440 When I go and press one of these notes, then something.. 223 00:11:37,440 --> 00:11:39,280 ROBOTIC VOICE: ..very strange will happen. 224 00:11:42,040 --> 00:11:44,600 You may not realise it, but you've almost certainly listened 225 00:11:44,600 --> 00:11:48,240 to a traditional vocoder, because it is in the music 226 00:11:48,240 --> 00:11:52,680 of New Order, of Phil Collins, if that's your bag, 227 00:11:52,680 --> 00:11:56,000 and my personal favourite, Daft Punk. 228 00:11:56,000 --> 00:11:58,840 MUSIC: In The Air Tonight by Phil Collins 229 00:11:58,840 --> 00:12:00,920 Why have you given me a Phil Collins record? 230 00:12:03,320 --> 00:12:05,720 MUSIC: Around The World by Daft Punk 231 00:12:05,720 --> 00:12:07,560 That's more like it. 232 00:12:07,560 --> 00:12:10,120 But as a method of synthesising human speech, 233 00:12:10,120 --> 00:12:12,160 the vocoder was limited. 234 00:12:12,160 --> 00:12:15,120 It could only convert pre-existing human sentences 235 00:12:15,120 --> 00:12:18,160 into a computer version. It couldn't generate new ones. 236 00:12:19,640 --> 00:12:22,040 So Bell Labs tried a different approach. 237 00:12:22,040 --> 00:12:25,320 They analysed the sound waves in human speech and tried to create 238 00:12:25,320 --> 00:12:28,040 new sentences from scratch. 239 00:12:28,040 --> 00:12:33,760 ROBOTIC VOICE: # Daisy, Daisy Give me your answer, do... # 240 00:12:33,760 --> 00:12:38,160 By the 1970s, speech synthesis had made its way into children's toys, 241 00:12:38,160 --> 00:12:41,920 like the Speak & Spell, which could read entire sentences 242 00:12:41,920 --> 00:12:43,680 of synthesised speech. 243 00:12:43,680 --> 00:12:47,800 ROBOTIC VOICE: Spell "rain". R-A-I-N. That is... 244 00:12:47,800 --> 00:12:50,920 But let's face it, although you know what it's saying, 245 00:12:50,920 --> 00:12:52,400 it hardly sounds human. 246 00:12:54,040 --> 00:12:57,480 For that to happen, we need to fast forward a few years. 247 00:12:57,480 --> 00:12:59,680 Open the pod bay doors, HAL. 248 00:12:59,680 --> 00:13:01,320 I'm sorry, Dave. 249 00:13:01,320 --> 00:13:03,200 I'm afraid I can't do that. 250 00:13:03,200 --> 00:13:06,720 That is HAL from the trailer for the film 2001. 251 00:13:08,400 --> 00:13:12,760 In the actual early 2000s, that film inspired a company called Ivona 252 00:13:12,760 --> 00:13:15,560 to try a new approach to creating speech. 253 00:13:15,560 --> 00:13:17,880 FEMALE VOICE: Hi, I am Ivona. 254 00:13:17,880 --> 00:13:21,720 I speak languages you need with voices you will love. 255 00:13:21,720 --> 00:13:23,400 I will read out loud... 256 00:13:23,400 --> 00:13:26,480 They recorded an actor's voice and split it into fragments 257 00:13:26,480 --> 00:13:27,720 called diphones. 258 00:13:29,000 --> 00:13:32,120 Then their code blended it together into natural-sounding 259 00:13:32,120 --> 00:13:33,360 words and sentences. 260 00:13:33,360 --> 00:13:36,680 My voice is not human. Would you believe? 261 00:13:36,680 --> 00:13:38,880 The results impressed Amazon so much 262 00:13:38,880 --> 00:13:41,600 that they bought the company, and it went on to become 263 00:13:41,600 --> 00:13:43,320 the basis for Alexa. 264 00:13:43,320 --> 00:13:45,760 ALEXA: We've come a long way in 70 years. 265 00:13:50,320 --> 00:13:53,800 Back at Amazon, Trevor thinks that we might be on the verge of creating 266 00:13:53,800 --> 00:13:57,200 synthetic voices that are almost indistinguishable from humans. 267 00:14:00,200 --> 00:14:03,120 So, in the last five years, pretty much everything has been replaced 268 00:14:03,120 --> 00:14:04,520 by neural networks. 269 00:14:04,520 --> 00:14:07,680 And that gives us new kinds of adaptability. 270 00:14:07,680 --> 00:14:09,240 Ah, neural networks. 271 00:14:11,000 --> 00:14:14,560 These are a special set of algorithms that are inspired 272 00:14:14,560 --> 00:14:17,320 by how our brains process information. 273 00:14:18,480 --> 00:14:21,600 This allows Alexa to learn as we do, 274 00:14:21,600 --> 00:14:24,240 listening to speech and working out 275 00:14:24,240 --> 00:14:27,360 the complexity and nuances of language. 276 00:14:27,360 --> 00:14:31,200 ARCHIVE: Human speech, seemingly so simple, 277 00:14:31,200 --> 00:14:33,200 is very complicated. 278 00:14:33,200 --> 00:14:36,240 And that's what lets Alexa produce far more natural patterns 279 00:14:36,240 --> 00:14:38,480 of speech like this... 280 00:14:38,480 --> 00:14:41,240 NATURAL FEMALE VOICE: I wonder what's going to happen in the story. 281 00:14:42,640 --> 00:14:45,000 The neural networks mean that we don't need as much data 282 00:14:45,000 --> 00:14:48,560 any more, so we can teach an old voice new tricks, a new accent, 283 00:14:48,560 --> 00:14:50,680 a new emotion to express. 284 00:14:50,680 --> 00:14:54,560 They can even reproduce your very own Hannah Fry. 285 00:14:54,560 --> 00:14:56,240 We asked you to give us, 286 00:14:56,240 --> 00:14:59,200 it was, I think, 19 sentences, about two minutes of data. 287 00:14:59,200 --> 00:15:02,160 And we've used that to adapt a voice that we think captures something 288 00:15:02,160 --> 00:15:05,720 about how you speak. Wait, hang on. Sorry. 19 sentences? Yeah. 289 00:15:05,720 --> 00:15:09,960 So we've gone from needing tens of hours of data 290 00:15:09,960 --> 00:15:12,080 down to about two minutes. Wait, you can... 291 00:15:12,080 --> 00:15:14,360 You can make me say words that I didn't say 292 00:15:14,360 --> 00:15:15,920 from 19 sentences alone? 293 00:15:15,920 --> 00:15:18,920 I've got you here reading the abstract of Alan Turing's famous 294 00:15:18,920 --> 00:15:22,520 paper on AI, so... OK, go on. Go on, I want to hear this. 295 00:15:22,520 --> 00:15:25,400 IN HANNAH'S VOICE: I propose to consider the question - can machines 296 00:15:25,400 --> 00:15:28,040 think? This should begin with definitions of the meaning 297 00:15:28,040 --> 00:15:30,240 of the terms "machine" and "think". Oh! 298 00:15:30,240 --> 00:15:32,480 The definitions might be framed so as to reflect 299 00:15:32,480 --> 00:15:35,120 so far as possible the normal use of the word. 300 00:15:36,200 --> 00:15:37,520 That was pretty good! 301 00:15:37,520 --> 00:15:39,840 I think it captures something about your voice. 302 00:15:39,840 --> 00:15:41,960 You are more challenging than we were expecting. 303 00:15:41,960 --> 00:15:43,800 I mean, you're welcome. 304 00:15:43,800 --> 00:15:47,560 There is so, so much sitting under this umbrella of making 305 00:15:47,560 --> 00:15:52,480 a natural-sounding voice that goes far beyond just what they're saying. 306 00:15:52,480 --> 00:15:56,320 I think in the past, especially when it was just chopped-up speech, 307 00:15:56,320 --> 00:15:59,200 people used to think of the voice as just, "Well, as long 308 00:15:59,200 --> 00:16:00,840 "as I can understand it, it's fine." 309 00:16:00,840 --> 00:16:03,760 But actually, if you want to have a conversation, 310 00:16:03,760 --> 00:16:07,680 how the person's voice sounds really is part of being a great AI. 311 00:16:07,680 --> 00:16:11,360 We want to get Alexa there to have those human-like attributes, 312 00:16:11,360 --> 00:16:15,400 to be able to have an engaging conversation with you. 313 00:16:15,400 --> 00:16:19,040 So it's all thanks to extraordinary innovations in programming 314 00:16:19,040 --> 00:16:22,040 that the voice coming from our virtual assistant speaker 315 00:16:22,040 --> 00:16:23,760 is so true to life. 316 00:16:27,280 --> 00:16:32,600 But a virtual assistant also needs ears, and that is where our second 317 00:16:32,600 --> 00:16:34,920 component comes in. 318 00:16:34,920 --> 00:16:38,840 It turns the sound of your voice into electrical impulses 319 00:16:38,840 --> 00:16:41,520 that Alexa's brain can understand. 320 00:16:41,520 --> 00:16:43,560 This is the microphone. 321 00:16:45,760 --> 00:16:48,520 I don't know if you've ever played around with an Alexa, sometimes 322 00:16:48,520 --> 00:16:51,160 you'll be standing right next to it and it'll ignore you. 323 00:16:51,160 --> 00:16:53,760 But other times, you'll be, like, three rooms away 324 00:16:53,760 --> 00:16:56,320 and it will light up at the sound of your voice. 325 00:16:56,320 --> 00:16:59,600 And that makes me wonder - what kind of microphones does 326 00:16:59,600 --> 00:17:02,560 it have inside that mean it's able to do that? 327 00:17:02,560 --> 00:17:05,760 And then the follow-up question - does that mean 328 00:17:05,760 --> 00:17:07,920 it's always listening? 329 00:17:07,920 --> 00:17:09,120 Well, let me show you. 330 00:17:09,120 --> 00:17:11,280 Take it apart. 331 00:17:09,120 --> 00:17:11,280 HE LAUGHS 332 00:17:11,280 --> 00:17:14,280 Open-heart surgery on Alexa. They don't normally come apart this easily. 333 00:17:14,280 --> 00:17:15,520 Well, he WOULD say that. 334 00:17:15,520 --> 00:17:19,000 That is Alexa's director of audio, Philip Hilmes. 335 00:17:22,240 --> 00:17:25,480 Given all of the noise that's going on around, how does it know 336 00:17:25,480 --> 00:17:26,680 what to listen to? 337 00:17:26,680 --> 00:17:29,560 So the first thing it does is it tries to determine 338 00:17:29,560 --> 00:17:34,360 is this audio speech or is it the dog barking, something like that. 339 00:17:34,360 --> 00:17:38,200 Inside here, we have six different microphones. 340 00:17:38,200 --> 00:17:42,440 One of the most amazing features of this virtual assistant's microphone 341 00:17:42,440 --> 00:17:45,600 is that it can pick up your voice even when there is lots of 342 00:17:45,600 --> 00:17:47,280 noise around it. 343 00:17:47,280 --> 00:17:49,800 Now, Phil, I know that we wanted to do a bit of a demonstration 344 00:17:49,800 --> 00:17:52,360 in a noisy, bustling cafe. Right. 345 00:17:52,360 --> 00:17:55,120 And we are in what should be a noisy, bustling cafe, 346 00:17:55,120 --> 00:17:57,720 except that we kicked everyone out so we could do some filming! 347 00:17:57,720 --> 00:18:01,000 Yes. We've got an alternative, haven't we, for some cafe noise? 348 00:18:01,000 --> 00:18:03,360 Alexa, play cafe sounds. 349 00:18:05,920 --> 00:18:07,560 Is there something wrong with me? 350 00:18:07,560 --> 00:18:09,360 I mean, I WAS going to buy one. 351 00:18:09,360 --> 00:18:11,040 THEY LAUGH 352 00:18:11,040 --> 00:18:13,280 Alexa, play cafe sounds. 353 00:18:15,400 --> 00:18:17,120 LOUD RUSTLING 354 00:18:15,400 --> 00:18:17,120 OK. 355 00:18:17,120 --> 00:18:20,080 We're going to test whether another Alexa on the other side of the room 356 00:18:20,080 --> 00:18:22,920 can hear us through all of this noise. 357 00:18:22,920 --> 00:18:24,720 There's some people talking over there. 358 00:18:24,720 --> 00:18:28,480 There's the screech of a buggy, of a mum desperate for some 359 00:18:28,480 --> 00:18:30,400 caffeine, I can hear in the corner. 360 00:18:30,400 --> 00:18:32,160 There's lots of stuff going on. 361 00:18:32,160 --> 00:18:37,440 And, yes, even in an environment like this, if I just say to you, 362 00:18:37,440 --> 00:18:39,600 "Alexa, what time is it?" 363 00:18:39,600 --> 00:18:42,600 ALEXA: The time is 2:54pm. 364 00:18:42,600 --> 00:18:44,520 That's right. Picks it up. 365 00:18:46,400 --> 00:18:47,560 Alexa, stop. 366 00:18:47,560 --> 00:18:48,800 CAFE HUBBUB STOPS 367 00:18:48,800 --> 00:18:51,000 That was incredible. I mean, that was quite loud. 368 00:18:51,000 --> 00:18:52,480 It was very loud, yeah. 369 00:18:52,480 --> 00:18:56,960 And yet, all the way over there, that device was able to pick 370 00:18:56,960 --> 00:18:59,880 up on the wake word from here. 371 00:18:59,880 --> 00:19:04,240 It's all thanks to your virtual assistant's multiple microphones. 372 00:19:04,240 --> 00:19:09,960 By having these six microphones, we're able to combine them together, 373 00:19:09,960 --> 00:19:12,200 doing what we call beamforming. 374 00:19:13,360 --> 00:19:16,480 Beamforming is the brilliant bit of tech that allows your virtual 375 00:19:16,480 --> 00:19:20,240 assistant to pick up on your voice in even the noisiest environments. 376 00:19:20,240 --> 00:19:22,880 ARCHIVE: There's a highly-refined and perfect mechanism... 377 00:19:22,880 --> 00:19:26,840 To find out how we got beamforming, we need to go back in time... 378 00:19:26,840 --> 00:19:29,080 ..not to World War II... 379 00:19:29,080 --> 00:19:30,640 ..but to World War I. 380 00:19:34,280 --> 00:19:39,800 By 1914, World War I was in full swing, and the British started 381 00:19:39,800 --> 00:19:42,320 blockading all of the German ports, trying to cut 382 00:19:42,320 --> 00:19:43,640 off their supply lines. 383 00:19:43,640 --> 00:19:47,000 But the Germans, who were a bit penned in by this point, 384 00:19:47,000 --> 00:19:48,800 did have one key advantage. 385 00:19:48,800 --> 00:19:50,080 They had U-boats. 386 00:19:50,080 --> 00:19:53,680 For anyone who hasn't seen Das Boot, U-boats were early submarines. 387 00:19:53,680 --> 00:19:58,520 These are silent killers bristling with torpedoes, and they started 388 00:19:58,520 --> 00:20:02,960 prowling along the coastline, picking off British ships supplying 389 00:20:02,960 --> 00:20:04,560 food to our shores. 390 00:20:06,480 --> 00:20:10,200 To avoid Britain being starved into submission, the Navy had to 391 00:20:10,200 --> 00:20:12,920 find a way to locate and destroy the U-boats. 392 00:20:12,920 --> 00:20:16,560 And there was one key weakness that they could exploit. 393 00:20:16,560 --> 00:20:18,520 Sound. 394 00:20:18,520 --> 00:20:23,280 If you have a clanky U-boat, people on board making 395 00:20:23,280 --> 00:20:24,480 all sorts of noises... 396 00:20:24,480 --> 00:20:27,400 The sound waves will propagate through the water 397 00:20:27,400 --> 00:20:28,960 and can be picked up. 398 00:20:30,320 --> 00:20:33,160 That is Patrick Naylor from Imperial College London, 399 00:20:33,160 --> 00:20:35,360 and he is an acoustic scientist. 400 00:20:37,440 --> 00:20:42,360 So, putting microphones underwater allows you to hear what is going 401 00:20:42,360 --> 00:20:47,080 on below the surface and pick up and detect perhaps enemy vessels. 402 00:20:48,440 --> 00:20:52,560 And that is exactly what the Navy did, using a brand-new 403 00:20:52,560 --> 00:20:54,320 device called a hydrophone, 404 00:20:54,320 --> 00:20:57,000 a microphone designed to work underwater. 405 00:20:59,280 --> 00:21:01,320 They quickly proved effective. 406 00:21:01,320 --> 00:21:04,600 A single microphone could pick up a German U-boat almost 407 00:21:04,600 --> 00:21:05,880 four miles away. 408 00:21:07,360 --> 00:21:11,360 One downside - it couldn't tell you where that U-boat was. 409 00:21:11,360 --> 00:21:14,400 For that, they needed another crucial breakthrough, 410 00:21:14,400 --> 00:21:18,440 led by a maverick sea captain called Cyril Ryan. 411 00:21:18,440 --> 00:21:21,160 And their solution was simple - 412 00:21:21,160 --> 00:21:23,320 use two microphones instead. 413 00:21:24,960 --> 00:21:27,280 Patrick is going to demonstrate. 414 00:21:29,800 --> 00:21:33,240 Here we have the microphone array, which is very much along the same 415 00:21:33,240 --> 00:21:35,600 lines of the two hydrophones. 416 00:21:35,600 --> 00:21:38,000 OK. So, if you have a little jingle over here... 417 00:21:39,320 --> 00:21:44,400 So what we see here is the sound appearing on the left with a peak. 418 00:21:44,400 --> 00:21:47,760 And as the sound moves into the centre, then the peak moves 419 00:21:47,760 --> 00:21:49,200 to the centre. Hmm! 420 00:21:49,200 --> 00:21:52,080 And then as the sound moves slowly to the right, we can see the peak 421 00:21:52,080 --> 00:21:53,840 moves down to the right. 422 00:21:55,120 --> 00:21:57,840 The sound arrives from different directions, 423 00:21:57,840 --> 00:22:02,600 and it will have a different time of arrival at each microphone. 424 00:22:02,600 --> 00:22:05,840 With a clever little bit of maths, you can then use the difference 425 00:22:05,840 --> 00:22:08,920 in arrival time at the two microphones to work out 426 00:22:08,920 --> 00:22:12,000 which direction the sound is coming from. 427 00:22:12,000 --> 00:22:16,120 Using an array of microphones, that helped the Allies uncover a lot 428 00:22:16,120 --> 00:22:17,840 of U-boats. 429 00:22:17,840 --> 00:22:20,920 And today, less life-saving but still pretty helpful, 430 00:22:20,920 --> 00:22:23,680 it allows your virtual assistant to pick out your voice 431 00:22:23,680 --> 00:22:24,920 in a noisy room. 432 00:22:24,920 --> 00:22:28,640 So when we have the microphone array, the microphones work together 433 00:22:28,640 --> 00:22:30,920 and the signals are brought into an algorithm called 434 00:22:30,920 --> 00:22:32,720 a beamforming algorithm. 435 00:22:32,720 --> 00:22:35,400 At last, we are back to beamforming. 436 00:22:35,400 --> 00:22:36,600 Beamforming. 437 00:22:38,000 --> 00:22:41,840 Beamforming allows the array of microphones to focus on and boost 438 00:22:41,840 --> 00:22:45,160 the sound from one direction only. 439 00:22:45,160 --> 00:22:48,840 The beamforming algorithm will combine them in a way such 440 00:22:48,840 --> 00:22:54,040 that from some directions, the sound will add up and get stronger, 441 00:22:54,040 --> 00:22:57,680 whereas from other directions, the sound at the different 442 00:22:57,680 --> 00:22:59,960 microphones will cancel itself out. 443 00:22:59,960 --> 00:23:02,720 OK, so if you've got something really noisy that you don't want 444 00:23:02,720 --> 00:23:05,280 to listen to going on over there, some building work going on, 445 00:23:05,280 --> 00:23:08,320 you can deliberately cancel out all of that noise over there 446 00:23:08,320 --> 00:23:10,640 and boost the noise from this direction. 447 00:23:10,640 --> 00:23:11,800 Exactly. That's right. 448 00:23:11,800 --> 00:23:14,760 I understand that in theory, but what does it actually sound 449 00:23:14,760 --> 00:23:15,880 like when you do it? 450 00:23:15,880 --> 00:23:19,360 Well, we have some bells coming from the left, and we have a 451 00:23:19,360 --> 00:23:21,320 xylophone coming from the right. 452 00:23:21,320 --> 00:23:24,200 First up, let's hear it without beamforming. 453 00:23:24,200 --> 00:23:27,960 What you are hearing on the telly is what I'm hearing in my headphones. 454 00:23:27,960 --> 00:23:29,880 RINGING AND CLANKING 455 00:23:29,880 --> 00:23:32,920 I mean... OK. You two sound like the worst band of all time, 456 00:23:32,920 --> 00:23:34,880 but thank you. OK, great. 457 00:23:34,880 --> 00:23:37,000 The beamformer is now activated. 458 00:23:37,000 --> 00:23:38,800 Focus on... 459 00:23:38,800 --> 00:23:40,160 LOWER VOLUME: ..the bells. 460 00:23:40,160 --> 00:23:41,960 Oh! You've suddenly gone quieter. 461 00:23:41,960 --> 00:23:44,440 Ah, the beamformer's working. 462 00:23:44,440 --> 00:23:46,200 RINGING AND CLANKING 463 00:23:48,680 --> 00:23:51,440 And now, if we switch to the xylophone... 464 00:23:51,440 --> 00:23:52,760 RINGING GETS QUIETER 465 00:23:52,760 --> 00:23:54,360 Oh, that's so interesting! 466 00:23:55,760 --> 00:23:57,080 Go back. 467 00:23:57,080 --> 00:24:01,000 Back to the bells, and you'll hear it come much more into focus. 468 00:24:01,000 --> 00:24:05,480 It is just... It's kind of strange, because it feels very similar 469 00:24:05,480 --> 00:24:09,120 to sort of turning your head and focusing your full attention 470 00:24:09,120 --> 00:24:11,240 on a particular noise. 471 00:24:11,240 --> 00:24:13,760 But it's being done for you by the technology. 472 00:24:13,760 --> 00:24:17,640 This is exactly what would happen in a smart home audio device, 473 00:24:17,640 --> 00:24:20,600 where the device would need to listen in a certain direction 474 00:24:20,600 --> 00:24:24,120 and focus its attention in that particular direction, 475 00:24:24,120 --> 00:24:27,080 whilst suppressing the background noise coming from other directions. 476 00:24:27,080 --> 00:24:29,360 Yeah. That was good. Thanks, guys. 477 00:24:29,360 --> 00:24:31,160 Don't give up your day jobs. 478 00:24:31,160 --> 00:24:33,800 ALEXA: Especially when us voice assistants take over. 479 00:24:34,880 --> 00:24:36,240 Look, I'm a nerd. 480 00:24:36,240 --> 00:24:39,240 And so the technology that lets your virtual assistant hear 481 00:24:39,240 --> 00:24:43,560 you is one of my favourite things, but it does raise some questions. 482 00:24:43,560 --> 00:24:45,240 So let's go back to Phil at Amazon. 483 00:24:45,240 --> 00:24:48,600 Given that we're talking about microphones and we're talking 484 00:24:48,600 --> 00:24:51,920 about Alexa even being able to listen in, even 485 00:24:51,920 --> 00:24:53,360 in a really noisy room, 486 00:24:53,360 --> 00:24:56,240 I guess the natural next question is, does that mean 487 00:24:56,240 --> 00:24:58,360 it's always listening? No, not at all. 488 00:24:58,360 --> 00:25:01,360 What they're doing is they're only looking for that wake word. 489 00:25:01,360 --> 00:25:05,160 And so once we do that, then we turn on the rest of the system. 490 00:25:09,760 --> 00:25:13,120 The thing about these devices is that you are essentially putting 491 00:25:13,120 --> 00:25:16,280 an internet-enabled microphone in your house. 492 00:25:16,280 --> 00:25:19,400 And I know that some people feel really uncomfortable about that, 493 00:25:19,400 --> 00:25:21,720 whereas other people are totally fine with it. 494 00:25:21,720 --> 00:25:26,240 Ultimately, all those concerns about privacy, they come down to, 495 00:25:26,240 --> 00:25:31,600 do you believe that the company is not misusing your data? 496 00:25:33,040 --> 00:25:35,280 Thanks to the microphone array and the genius 497 00:25:35,280 --> 00:25:38,240 of beamforming, your virtual assistant is able to listen 498 00:25:38,240 --> 00:25:40,240 to your questions with clarity. 499 00:25:40,240 --> 00:25:43,240 But on its own, that is not enough for it to make sense 500 00:25:43,240 --> 00:25:47,280 of what you're saying to it and reply with a sensible answer. 501 00:25:49,120 --> 00:25:50,960 For Alexa to understand us, 502 00:25:50,960 --> 00:25:53,960 it needs some sort of brain, 503 00:25:53,960 --> 00:25:56,000 and that is powered by a key piece 504 00:25:56,000 --> 00:25:59,000 of technology that isn't a physical component. 505 00:25:59,000 --> 00:26:01,480 But trust me, it's vital. 506 00:26:01,480 --> 00:26:05,120 It is a complex network of algorithms that form 507 00:26:05,120 --> 00:26:07,640 Alexa's artificial intelligence. 508 00:26:13,840 --> 00:26:17,040 Alexa is pretty good at answering difficult questions. 509 00:26:17,040 --> 00:26:21,440 Alexa, what's the capital of the Solomon Islands? 510 00:26:21,440 --> 00:26:25,400 The capital of the Solomon Islands is Honiara. 511 00:26:25,400 --> 00:26:28,360 Pretty impressive. But let's push it a bit. 512 00:26:28,360 --> 00:26:31,560 Come on. Something a bit more ambiguous. 513 00:26:31,560 --> 00:26:33,680 Alexa, what is love? 514 00:26:33,680 --> 00:26:36,800 Love is sharing your chocolate with someone, even if you want 515 00:26:36,800 --> 00:26:38,480 to scoff it all yourself. 516 00:26:38,480 --> 00:26:41,760 Hmm. We'll agree to disagree on that one. 517 00:26:41,760 --> 00:26:45,880 It's the artificial intelligence that allows the Alexa to hear, 518 00:26:45,880 --> 00:26:49,520 to respond, to answer your questions, to play music when you 519 00:26:49,520 --> 00:26:51,200 want it to. 520 00:26:51,200 --> 00:26:56,120 Its AI has been trained to interact with us in a way that feels human. 521 00:26:56,120 --> 00:26:58,160 But how does AI manage this? 522 00:26:59,480 --> 00:27:01,720 The answer starts 70 years ago. 523 00:27:04,160 --> 00:27:08,320 In a 1950 paper, scientist Alan Turing posed the question - 524 00:27:08,320 --> 00:27:10,040 can machines think? 525 00:27:12,520 --> 00:27:15,080 It was one of the earliest explorations of what would become 526 00:27:15,080 --> 00:27:17,480 known as artificial intelligence. 527 00:27:19,680 --> 00:27:22,800 ARCHIVE: Anyone who urgently wishes to know whether two to the power of 528 00:27:22,800 --> 00:27:27,320 127 minus one is a prime number or not can be given the answer 529 00:27:27,320 --> 00:27:29,000 by an electronic brain. 530 00:27:30,640 --> 00:27:34,360 In 1952, Christopher Strachey programmed a computer to play 531 00:27:34,360 --> 00:27:36,240 a simple game of draughts. 532 00:27:36,240 --> 00:27:41,680 When he fed in his 20 pages of code, his AI-driven draughts game 533 00:27:41,680 --> 00:27:43,960 was challenging everyone in the lab. 534 00:27:45,400 --> 00:27:48,360 It's time now that it got down to serious work. 535 00:27:48,360 --> 00:27:49,920 In the mid 1960s, 536 00:27:49,920 --> 00:27:53,040 AI was used in an interactive device. 537 00:27:53,040 --> 00:27:57,560 ELIZA was the world's first chatbot, and could hold a basic conversation. 538 00:27:57,560 --> 00:27:59,560 FEMALE VOICE: My boyfriend made me come here. 539 00:27:59,560 --> 00:28:01,560 MALE VOICE: Your boyfriend made you come here. 540 00:28:01,560 --> 00:28:05,840 It didn't have much chat, replying with only pre-programmed phrases. 541 00:28:05,840 --> 00:28:08,760 MALE VOICE: I'm sorry to hear that you're depressed. 542 00:28:08,760 --> 00:28:11,080 During the 1970s and early '80s, 543 00:28:11,080 --> 00:28:15,680 AI's progress slowed, funding dried up, and the hardware available 544 00:28:15,680 --> 00:28:17,960 struggled. It's not intelligence at all. 545 00:28:17,960 --> 00:28:20,720 It adds up numbers and subtracts numbers and multiplies numbers 546 00:28:20,720 --> 00:28:22,640 very, very quickly. 547 00:28:22,640 --> 00:28:25,840 That changed with the advent of more powerful computers 548 00:28:25,840 --> 00:28:27,680 in the late 1990s. 549 00:28:29,200 --> 00:28:34,120 By 1997, AI even looked like a match for human intelligence. 550 00:28:34,120 --> 00:28:39,240 IBM's most powerful supercomputer, Deep Blue, stunned the world 551 00:28:39,240 --> 00:28:42,600 and defeated the best chess player on the planet, 552 00:28:42,600 --> 00:28:44,600 grandmaster Garry Kasparov. 553 00:28:46,080 --> 00:28:49,040 ARCHIVE: The game of chess, supposedly a true test of human 554 00:28:49,040 --> 00:28:51,160 intellect, will never be the same again. 555 00:28:51,160 --> 00:28:55,600 This looked as though AI was finally living up to the promise 556 00:28:55,600 --> 00:29:01,720 of the 1950s, but then it proved to be a bit of a false dawn. 557 00:29:01,720 --> 00:29:05,240 Deep Blue's moment in the spotlight wasn't the breakthrough it seemed, 558 00:29:05,240 --> 00:29:09,320 and that is because the computer couldn't actually think for itself. 559 00:29:09,320 --> 00:29:11,840 It was just following a set of rules. 560 00:29:13,280 --> 00:29:17,480 When it comes to chess, you're on a very defined board. 561 00:29:17,480 --> 00:29:21,480 All of the pieces have particular rules that are associated with them. 562 00:29:21,480 --> 00:29:24,720 And all you needed to do, really, was write down those rules 563 00:29:24,720 --> 00:29:27,240 in a way that a computer could understand, and then it 564 00:29:27,240 --> 00:29:30,600 just had to crunch through those calculations. 565 00:29:30,600 --> 00:29:33,720 The problem is, is that when you extend that out beyond the game 566 00:29:33,720 --> 00:29:37,840 of chess into the real world, writing down all the rules becomes 567 00:29:37,840 --> 00:29:40,280 really, really, really hard. 568 00:29:40,280 --> 00:29:42,880 Almost as hard as I'm finding it to win this game. 569 00:29:42,880 --> 00:29:44,560 CHESSBOARD BEEPS 570 00:29:42,880 --> 00:29:44,560 All right, chill out. 571 00:29:44,560 --> 00:29:46,080 I'm actually really competitive! 572 00:29:46,080 --> 00:29:49,560 I'm unwilling to lose this game of chess, even though I know 573 00:29:49,560 --> 00:29:52,360 that I'm actually supposed to be doing my job here. 574 00:29:52,360 --> 00:29:57,240 And so tasks like chess - I won, by the way - that are based on a set 575 00:29:57,240 --> 00:30:00,080 of simply defined rules could be mastered 576 00:30:00,080 --> 00:30:02,280 if you had enough computing power. 577 00:30:02,280 --> 00:30:07,160 But a lot of real-life challenges were REALLY hard for computers. 578 00:30:07,160 --> 00:30:10,920 Let me illustrate with the help of an expert assistant. 579 00:30:10,920 --> 00:30:14,120 Come here. Come sit here. Come here, come here, come here. 580 00:30:14,120 --> 00:30:15,560 This is my daughter. 581 00:30:18,280 --> 00:30:19,960 Say hello. Hello. 582 00:30:19,960 --> 00:30:21,440 HANNAH LAUGHS 583 00:30:21,440 --> 00:30:25,120 And what I thought I'd do is show you some pictures, and you have to 584 00:30:25,120 --> 00:30:27,560 tell me what's in the pictures. OK? OK. 585 00:30:27,560 --> 00:30:29,800 All right, then. So here we go. Let's have a look. 586 00:30:29,800 --> 00:30:32,040 What's that picture? 587 00:30:32,040 --> 00:30:33,840 A dog. A dog! 588 00:30:33,840 --> 00:30:36,480 That's right. And this one? 589 00:30:36,480 --> 00:30:38,280 A dog. A dog! 590 00:30:39,280 --> 00:30:41,240 This is too easy for you. What's that? 591 00:30:41,240 --> 00:30:44,160 A dog. OK, hang on. This one's tough. You ready? 592 00:30:44,160 --> 00:30:45,680 What's this one? 593 00:30:46,840 --> 00:30:49,560 A dog. A dog! Exactly. 594 00:30:49,560 --> 00:30:53,280 So, these are quite different, these pictures. 595 00:30:53,280 --> 00:30:57,440 And yet, even a three-year-old can recognise that every single one 596 00:30:57,440 --> 00:30:59,400 of them is a dog. 597 00:30:59,400 --> 00:31:00,840 Thank you. 598 00:31:00,840 --> 00:31:04,640 That's because we humans quickly learn the features that similar 599 00:31:04,640 --> 00:31:09,440 objects have in common, even when there's a lot of variety. 600 00:31:09,440 --> 00:31:13,560 Until around ten years ago, computers found it surprisingly 601 00:31:13,560 --> 00:31:18,240 difficult to tell the difference between images of dogs and cats. 602 00:31:18,240 --> 00:31:20,880 If only you could get over that hurdle, 603 00:31:20,880 --> 00:31:26,320 then that would be the advent of a whole new era of intelligence. 604 00:31:26,320 --> 00:31:30,520 So, back in 2010, AI scientists launched a competition 605 00:31:30,520 --> 00:31:32,000 called ImageNet. 606 00:31:32,000 --> 00:31:35,320 The best computer scientists pitched their image recognition 607 00:31:35,320 --> 00:31:37,360 software against each other. 608 00:31:37,360 --> 00:31:40,440 All of the entries into this competition... 609 00:31:40,440 --> 00:31:42,480 To put it nicely, they were rubbish. 610 00:31:42,480 --> 00:31:45,480 You'd give them a really easy picture like a ball, 611 00:31:45,480 --> 00:31:48,840 and they would get it right about 40% of the time. 612 00:31:48,840 --> 00:31:53,640 And then, in 2012, there was one entry to the ImageNet competition 613 00:31:53,640 --> 00:31:57,080 which came along and blew everyone away. 614 00:31:59,840 --> 00:32:03,600 That top-dog programme went by the name of AlexNet. 615 00:32:04,920 --> 00:32:10,280 It could successfully recognise 85% of images, a massive improvement 616 00:32:10,280 --> 00:32:12,760 on anything that had come before. 617 00:32:12,760 --> 00:32:16,120 And it was all thanks to a form of artificial intelligence 618 00:32:16,120 --> 00:32:18,040 called deep learning. 619 00:32:21,920 --> 00:32:25,680 Forget about writing down all the rules, just have a clear 620 00:32:25,680 --> 00:32:28,560 objective and let the machine figure out all of the steps 621 00:32:28,560 --> 00:32:30,360 along the way for itself. 622 00:32:30,360 --> 00:32:33,320 Remember those neural networks that learn from experience 623 00:32:33,320 --> 00:32:34,720 just like us? 624 00:32:34,720 --> 00:32:37,360 Well, deep learning is kind of like an advanced version of that. 625 00:32:37,360 --> 00:32:41,600 So the more experience that you give it to learn from, data like pictures 626 00:32:41,600 --> 00:32:44,080 or voice, the better it will become. 627 00:32:44,080 --> 00:32:46,600 This isn't just some niche bit of tech. 628 00:32:46,600 --> 00:32:49,640 Actually, you will find deep learning within most people's 629 00:32:49,640 --> 00:32:53,400 mobile phones. So if I search for "dog" on all of the photos on 630 00:32:53,400 --> 00:32:56,920 my phone, rather than trying to find a picture that 100% matches 631 00:32:56,920 --> 00:32:58,280 a rigid template, 632 00:32:58,280 --> 00:33:02,800 deep learning looks for features, like the shadow that might appear 633 00:33:02,800 --> 00:33:06,400 under a snout or the wispiness that might indicate a tail. 634 00:33:06,400 --> 00:33:09,880 And then if the AI decides there are enough doggie features, 635 00:33:09,880 --> 00:33:12,480 it will classify the image as dog. 636 00:33:12,480 --> 00:33:15,480 Not quite perfect, though, because I'm not quite sure 637 00:33:15,480 --> 00:33:17,840 how I should take this. 638 00:33:17,840 --> 00:33:20,400 It thought that was a picture of a dog. 639 00:33:20,400 --> 00:33:22,600 I'll avoid being offended for now, AI. 640 00:33:22,600 --> 00:33:24,440 But I'm keeping my eye on you. 641 00:33:24,440 --> 00:33:28,640 Deep learning was a giant leap forward in AI. 642 00:33:28,640 --> 00:33:31,240 For virtual assistants to make use of it, they needed to learn 643 00:33:31,240 --> 00:33:33,600 how to interact with humans. 644 00:33:33,600 --> 00:33:37,360 For that, they needed real-world data of what humans say 645 00:33:37,360 --> 00:33:41,000 and how they respond in a myriad of situations. 646 00:33:44,320 --> 00:33:46,000 Back to Rohit. 647 00:33:46,000 --> 00:33:51,040 He and his team needed to expose the prototype Alexa AI to a huge 648 00:33:51,040 --> 00:33:55,320 number of different human voices, asking it a range of questions 649 00:33:55,320 --> 00:33:57,520 so it could learn from them. 650 00:33:57,520 --> 00:34:02,200 There was no precedent of people speaking to these devices, 651 00:34:02,200 --> 00:34:05,560 there was just no data that existed, and we didn't want our 652 00:34:05,560 --> 00:34:07,200 ambitions to leak in this area, 653 00:34:07,200 --> 00:34:10,240 so we had to collect all this data undercover. 654 00:34:10,240 --> 00:34:12,240 To keep their plan secret, 655 00:34:12,240 --> 00:34:15,160 the team hired apartments and hid prototype 656 00:34:15,160 --> 00:34:17,080 virtual assistants everywhere. 657 00:34:17,080 --> 00:34:20,240 They then employed thousands of people with different accents 658 00:34:20,240 --> 00:34:24,200 to walk around the apartments, reading questions off a script... 659 00:34:24,200 --> 00:34:26,280 Play Phil Collins. 660 00:34:26,280 --> 00:34:28,160 I want to hear Phil Collins. 661 00:34:28,160 --> 00:34:30,320 Just give me In The Air Tonight. 662 00:34:30,320 --> 00:34:33,400 ..over and over and over again. 663 00:34:33,400 --> 00:34:36,720 It was one of the most exciting things I've done, where even my 664 00:34:36,720 --> 00:34:38,680 family didn't know what I was doing. 665 00:34:38,680 --> 00:34:42,560 And how did that data look when you tried to sift through it? 666 00:34:42,560 --> 00:34:45,880 For AI scientists, it was exactly the kind of data you would 667 00:34:45,880 --> 00:34:47,560 like your hands on. 668 00:34:47,560 --> 00:34:52,120 They fed the recordings into Alexa's cloud-based AI to find connections 669 00:34:52,120 --> 00:34:57,120 in the data, so it could work out that Phil Collins, along with words 670 00:34:57,120 --> 00:35:01,360 like "play" or "hear", means you want to listen to a song. 671 00:35:01,360 --> 00:35:05,920 By repeatedly making similar connections over and over again, 672 00:35:05,920 --> 00:35:10,840 the AI slowly trained itself to provide the best responses. 673 00:35:10,840 --> 00:35:14,080 By 2014, the AI could understand questions 674 00:35:14,080 --> 00:35:16,720 asked in a variety of different ways. 675 00:35:20,280 --> 00:35:23,040 And Alexa was ready to be launched. 676 00:35:23,040 --> 00:35:24,840 ALEXA: Hello, world. 677 00:35:24,840 --> 00:35:27,440 AI brings virtual assistants to life. 678 00:35:27,440 --> 00:35:30,000 But it couldn't exist without one of the greatest 679 00:35:30,000 --> 00:35:33,000 technological accomplishments of the 20th century. 680 00:35:33,000 --> 00:35:38,360 This is the crucial hardware that runs all the other components. 681 00:35:38,360 --> 00:35:41,400 It's smaller than a postage stamp. 682 00:35:41,400 --> 00:35:44,320 And yet it turns electrical energy into billions 683 00:35:44,320 --> 00:35:46,720 of powerful computer calculations, 684 00:35:46,720 --> 00:35:50,920 capable of operating Alexa's large AI brain. 685 00:35:50,920 --> 00:35:53,080 This is the microprocessor. 686 00:35:55,520 --> 00:35:59,400 You might know this as a microchip, but microprocessor 687 00:35:59,400 --> 00:36:02,600 is what it's called by the computing brains of Silicon Valley. 688 00:36:02,600 --> 00:36:04,920 Just a little bit of heat to move this cover off. 689 00:36:04,920 --> 00:36:07,280 Whichever name you give it, getting a closer look 690 00:36:07,280 --> 00:36:11,280 at the one in my virtual assistant is not exactly straightforward. 691 00:36:11,280 --> 00:36:13,160 They do not make this easy. 692 00:36:15,240 --> 00:36:16,880 It's got this cover on it 693 00:36:16,880 --> 00:36:19,360 so that the microprocessor stays protected. 694 00:36:19,360 --> 00:36:24,080 Presumably...from people like me. 695 00:36:24,080 --> 00:36:26,680 SHE CHUCKLES 696 00:36:24,080 --> 00:36:26,680 Oh! Yeah! 697 00:36:31,240 --> 00:36:32,720 We've got it! 698 00:36:34,560 --> 00:36:38,000 This...this is it! 699 00:36:38,000 --> 00:36:39,720 Ooh! Was it? 700 00:36:39,720 --> 00:36:41,000 SHE LAUGHS 701 00:36:44,120 --> 00:36:47,160 This is it. This is genuinely where all the magic happens, 702 00:36:47,160 --> 00:36:49,080 in this tiny little thing! 703 00:36:49,080 --> 00:36:54,240 This is, without doubt, THE most important part of this entire thing. 704 00:36:54,240 --> 00:36:56,240 This is...this is where the brain is. 705 00:36:56,240 --> 00:36:59,960 The thing is about microprocessors is that at their heart, they're actually quite simple. 706 00:36:59,960 --> 00:37:03,760 All they're doing is just really basic calculations. 707 00:37:03,760 --> 00:37:05,480 Ones and zeros. 708 00:37:05,480 --> 00:37:09,160 But they are doing it at such a level 709 00:37:09,160 --> 00:37:12,560 that you are getting millions, or even billions of calculations 710 00:37:12,560 --> 00:37:14,400 every single second. 711 00:37:14,400 --> 00:37:17,880 Far more than your average mathematician can manage, essentially. 712 00:37:17,880 --> 00:37:21,440 It's the calculating power of the microprocessor 713 00:37:21,440 --> 00:37:26,120 that allows Alexa's version of a brain, its AI, to function. 714 00:37:26,120 --> 00:37:28,520 Because for computers, thinking, 715 00:37:28,520 --> 00:37:31,320 it's basically a lot of mathematical equations. 716 00:37:32,640 --> 00:37:34,840 But how does a microprocessor, 717 00:37:34,840 --> 00:37:36,280 something that on the outside, 718 00:37:36,280 --> 00:37:39,480 just appears to be a lump of metal and plastic... 719 00:37:39,480 --> 00:37:41,000 ALEXA: That's not nice. 720 00:37:41,000 --> 00:37:44,600 ..managed to make billions and billions of calculations a second? 721 00:37:45,720 --> 00:37:49,160 To answer that, I need to wear some funky clothes. 722 00:37:49,160 --> 00:37:52,680 Courtesy of Cambridge University's Electronic Engineering Department, 723 00:37:52,680 --> 00:37:54,840 where they research, design and manufacture 724 00:37:54,840 --> 00:37:57,400 some of the latest high-tech microprocessors. 725 00:37:58,760 --> 00:38:01,440 There is a good reason for us to be dressed like this, isn't there? 726 00:38:01,440 --> 00:38:04,800 There is. We're making things at really small-length scales. 727 00:38:04,800 --> 00:38:06,240 A fraction of a human hair. 728 00:38:06,240 --> 00:38:09,360 So we need to make sure the environment is free of any particulates. 729 00:38:09,360 --> 00:38:11,920 That is Professor Andrew Flewitt. 730 00:38:15,240 --> 00:38:17,840 When you look at a microprocessor like this, 731 00:38:17,840 --> 00:38:22,320 we forget that inside is probably about five-billion transistors. 732 00:38:22,320 --> 00:38:25,080 Transistors are the part of the microprocessor 733 00:38:25,080 --> 00:38:27,520 that actually do the calculations. 734 00:38:27,520 --> 00:38:29,440 In other words, the clever bit. 735 00:38:29,440 --> 00:38:32,840 A single transistor acts as a very simple switch. 736 00:38:32,840 --> 00:38:35,440 One or zero, on or off. 737 00:38:35,440 --> 00:38:37,520 Linking lots of them creates a network 738 00:38:37,520 --> 00:38:40,920 that can make much more complex calculations. 739 00:38:40,920 --> 00:38:45,080 Your Alexa is able to understand your voice in a fraction of a second. 740 00:38:45,080 --> 00:38:47,160 So it needs to be doing calculations 741 00:38:47,160 --> 00:38:49,520 billions and billions of times a second. 742 00:38:49,520 --> 00:38:53,360 So that means we need billions of these transistors. 743 00:38:53,360 --> 00:38:57,200 But how do you squeeze that many transistors 744 00:38:57,200 --> 00:38:59,680 on to a tiny microprocessor? 745 00:39:00,840 --> 00:39:02,840 Well, they start with a wafer. 746 00:39:02,840 --> 00:39:07,760 Which sounds delicious, but sadly, this one is extremely-pure silicon. 747 00:39:07,760 --> 00:39:11,600 We coat that silicon wafer first in a polymer, 748 00:39:11,600 --> 00:39:14,160 which is sensitive to UV light. 749 00:39:15,920 --> 00:39:17,920 Andrew places a stencil above it, 750 00:39:17,920 --> 00:39:20,760 matching the precise pattern of transistors. 751 00:39:23,160 --> 00:39:27,680 It's then put in a machine that focuses UV light through the stencil, 752 00:39:27,680 --> 00:39:31,960 and projects a microscopic version of that pattern on to the wafer. 753 00:39:34,040 --> 00:39:37,360 They now start to etch away the surface of the wafer 754 00:39:37,360 --> 00:39:40,760 in exactly the right places to create the transistors. 755 00:39:42,520 --> 00:39:45,280 So the light only lands in certain regions. 756 00:39:45,280 --> 00:39:47,320 Where it lands, the polymer is weakened, 757 00:39:47,320 --> 00:39:48,840 and we can then wash it away 758 00:39:48,840 --> 00:39:52,240 to make the pattern that we have here on our wafer. 759 00:39:54,400 --> 00:39:56,640 We have to go through this process several times 760 00:39:56,640 --> 00:39:59,960 to build up the complete, complex structure of the transistor. 761 00:40:02,120 --> 00:40:04,280 We end up with something that looks like this. 762 00:40:04,280 --> 00:40:06,800 And this pattern is transistors, is it - 763 00:40:06,800 --> 00:40:09,000 just teeny, teeny, tiny ones? 764 00:40:09,000 --> 00:40:10,480 It is tiny, tiny. Incredible! 765 00:40:10,480 --> 00:40:13,040 This plate has 2,000 transistors, 766 00:40:13,040 --> 00:40:16,600 but the latest versions can have up to 114 billion. 767 00:40:16,600 --> 00:40:18,080 Holy moley! 768 00:40:19,240 --> 00:40:24,280 Once complete, the wafer will be cut into 100 individual microprocessors. 769 00:40:24,280 --> 00:40:27,760 A final protective plastic coating, and then we're done. 770 00:40:28,960 --> 00:40:30,200 Just incredible! 771 00:40:31,880 --> 00:40:35,320 Sadly, Andrew wouldn't let me take this one home to use as a coaster, 772 00:40:35,320 --> 00:40:39,120 but that doesn't stop it being nothing short of a technological miracle. 773 00:40:39,120 --> 00:40:43,120 One based on technology going back over 60 years. 774 00:40:46,880 --> 00:40:51,320 In the 1950s, computer components were handcrafted on to circuit boards. 775 00:40:51,320 --> 00:40:54,040 They contained thousands of giant transistors 776 00:40:54,040 --> 00:40:57,200 and components made from dozens of different materials, 777 00:40:57,200 --> 00:40:59,400 like tin, copper and lead. 778 00:41:00,960 --> 00:41:03,720 This is an example of an old-school circuit board. 779 00:41:03,720 --> 00:41:06,320 And you can see how each of the different components 780 00:41:06,320 --> 00:41:08,440 are soldered on one by one. 781 00:41:08,440 --> 00:41:11,200 Now, this is a pretty simple board, 782 00:41:11,200 --> 00:41:14,200 but if you can imagine you were trying to do something much more complex, 783 00:41:14,200 --> 00:41:16,760 you would end up with this gigantic board 784 00:41:16,760 --> 00:41:20,520 with huge numbers of potential points of failure, 785 00:41:20,520 --> 00:41:24,560 to the point where it's almost guaranteed to fail. 786 00:41:24,560 --> 00:41:29,440 And that is something that they called the "tyranny of numbers". 787 00:41:29,440 --> 00:41:32,800 Computers back then were hot and noisy, 788 00:41:32,800 --> 00:41:37,600 and this took its toll on the many, many soldered connections. 789 00:41:38,720 --> 00:41:41,880 If just one was to fail, the system wouldn't work properly. 790 00:41:43,560 --> 00:41:46,600 Cue a young American engineer, Jack Kilby. 791 00:41:46,600 --> 00:41:50,320 He developed an enterprising solution. 792 00:41:50,320 --> 00:41:53,600 The story goes - in summer 1958, 793 00:41:53,600 --> 00:41:56,560 Jack Kilby had just got a new job at a Texas firm, 794 00:41:56,560 --> 00:41:59,040 and all of his colleagues were off on holiday. 795 00:41:59,040 --> 00:42:01,560 But Jack hadn't accrued enough annual leave, 796 00:42:01,560 --> 00:42:04,200 and so, he was stuck in the office on his own. 797 00:42:04,200 --> 00:42:07,560 And while he was there, he was thinking about the tyranny of numbers. 798 00:42:07,560 --> 00:42:09,400 He quickly hit upon an idea - 799 00:42:09,400 --> 00:42:13,880 What if the circuit boards were made out of one main super material 800 00:42:13,880 --> 00:42:17,240 that could conduct and control the flow of electricity? 801 00:42:17,240 --> 00:42:21,720 Meaning that you didn't need lots of failure-prone soldered connections 802 00:42:21,720 --> 00:42:23,560 between different materials. 803 00:42:23,560 --> 00:42:26,560 So he went around the office and he started sticking things together 804 00:42:26,560 --> 00:42:28,920 to try and make a bit of a prototype. 805 00:42:31,280 --> 00:42:34,160 And he did get a prototype working. This is it here. 806 00:42:34,160 --> 00:42:36,360 It was the world's first integrated circuit. 807 00:42:36,360 --> 00:42:39,960 Meaning the electrical components were made of one single material. 808 00:42:39,960 --> 00:42:42,760 It's basically as crude as you can get with engineering, 809 00:42:42,760 --> 00:42:44,800 just stick some stuff on a plate of glass. 810 00:42:44,800 --> 00:42:47,600 He used something called germanium. 811 00:42:47,600 --> 00:42:52,040 There is a reason why we haven't had a big germanium revolution, 812 00:42:52,040 --> 00:42:54,880 or Germanium Valley in California. 813 00:42:54,880 --> 00:42:56,720 So stay with me for a second. 814 00:42:58,560 --> 00:43:01,400 Because 1,500 miles away in California, 815 00:43:01,400 --> 00:43:05,480 a man called Robert Noyce had come up with an integrated circuit, too. 816 00:43:05,480 --> 00:43:07,680 However, his was made from silicon, 817 00:43:07,680 --> 00:43:11,560 which was much more abundant and easier to mass-produce. 818 00:43:11,560 --> 00:43:14,680 It's here that things get a bit sticky. 819 00:43:14,680 --> 00:43:17,120 Kilby and Noyce's respective companies 820 00:43:17,120 --> 00:43:19,960 both try to patent their circuits. 821 00:43:19,960 --> 00:43:22,640 Each claiming that theirs was first. 822 00:43:22,640 --> 00:43:26,480 A lawsuit ensues that took decades to sort out. 823 00:43:27,560 --> 00:43:30,880 Eventually, they agreed to cross-license these devices, 824 00:43:30,880 --> 00:43:35,000 with both men getting a claim as the inventor of the integrated circuit - 825 00:43:35,000 --> 00:43:39,000 the precursor to today's microprocessors or chips. 826 00:43:40,920 --> 00:43:42,560 Jack went on to have success 827 00:43:42,560 --> 00:43:45,200 with the world's first hand-held calculator. 828 00:43:45,200 --> 00:43:47,080 But Robert Noyce, however, 829 00:43:47,080 --> 00:43:50,600 went on to co-found a little company called Intel. 830 00:43:52,800 --> 00:43:54,120 OK, I'll be honest with you, 831 00:43:54,120 --> 00:43:58,680 the full history of the silicon chip is incredibly technical, 832 00:43:58,680 --> 00:44:02,680 but to me, it's all about just how much power 833 00:44:02,680 --> 00:44:07,480 we can squeeze into something that is really eye-wateringly small. 834 00:44:07,480 --> 00:44:11,040 Because without silicon, we wouldn't be able to do 835 00:44:11,040 --> 00:44:15,440 all of these unimaginable things that we can do now. 836 00:44:16,600 --> 00:44:18,920 The microprocessor and AI technology 837 00:44:18,920 --> 00:44:21,880 are vital for your virtual assistant. 838 00:44:21,880 --> 00:44:24,160 But to get the information you've asked it for, 839 00:44:24,160 --> 00:44:26,840 it has to be able to communicate with the wider world. 840 00:44:29,320 --> 00:44:32,440 And that is where our next component comes in. 841 00:44:35,120 --> 00:44:37,840 A delicate copper wire that uses radio waves 842 00:44:37,840 --> 00:44:43,440 to connect seamlessly and wirelessly to the Internet and other devices. 843 00:44:43,440 --> 00:44:44,960 It's the Wi-Fi antenna! 844 00:44:47,400 --> 00:44:51,120 This quite unassuming corridor is on the 30th floor 845 00:44:51,120 --> 00:44:55,000 of Amazon's international headquarters in Seattle. 846 00:44:56,960 --> 00:45:02,240 Behind that door is a space that's kind of like a laboratory, 847 00:45:02,240 --> 00:45:06,840 but it's also probably one of the smartest apartments in the world. 848 00:45:08,400 --> 00:45:11,120 It all appears to be like any normal house. 849 00:45:11,120 --> 00:45:15,800 Cooker, coffee machine, sink. 850 00:45:15,800 --> 00:45:17,800 Alexa? Lights on. 851 00:45:20,280 --> 00:45:25,280 Ah! So this is where they bring all the new Amazon Alexa products 852 00:45:25,280 --> 00:45:27,960 to be tested before they go out to the market. 853 00:45:27,960 --> 00:45:31,640 So you've got Alexa-enabled washer and drier over there in the corner, 854 00:45:31,640 --> 00:45:34,160 and even some slightly more unusual things, 855 00:45:34,160 --> 00:45:37,120 like these weights right here. So... 856 00:45:38,200 --> 00:45:40,000 Alexa? 857 00:45:40,000 --> 00:45:41,800 SHE LAUGHS 858 00:45:43,480 --> 00:45:46,480 There's...there's a twerking bear! 859 00:45:46,480 --> 00:45:49,400 Alexa? Set weights to 20lbs. 860 00:45:51,440 --> 00:45:54,080 There we go. Right. 861 00:45:54,080 --> 00:45:55,600 Oh! 862 00:45:55,600 --> 00:45:58,200 Yeah, probably needed to build up to that one. 863 00:45:58,200 --> 00:45:59,720 Whoa! Um... 864 00:45:59,720 --> 00:46:04,400 my favourite one, though, in here...is... 865 00:46:04,400 --> 00:46:06,720 Alexa? Turn shower on. 866 00:46:11,320 --> 00:46:12,600 Hey! 867 00:46:13,760 --> 00:46:17,600 The reason why everything in this mega-connected smart home 868 00:46:17,600 --> 00:46:19,040 can talk to each other, 869 00:46:19,040 --> 00:46:22,240 and why Alexa can give you answers to your questions, 870 00:46:22,240 --> 00:46:25,040 is down to the Wi-Fi antenna inside it. 871 00:46:27,560 --> 00:46:29,080 Oh, lovely! 872 00:46:29,080 --> 00:46:30,880 So satisfying. 873 00:46:32,040 --> 00:46:34,440 This little bit here and here, 874 00:46:34,440 --> 00:46:36,960 these are the antenna of the device. 875 00:46:38,760 --> 00:46:41,440 If I ask Alexa a question, 876 00:46:41,440 --> 00:46:44,680 it picks up on the vibrations in the air, 877 00:46:44,680 --> 00:46:47,000 interprets what I'm saying, 878 00:46:47,000 --> 00:46:51,720 and then sends out an invisible signal 879 00:46:51,720 --> 00:46:55,560 across my living room, into my Wi-Fi router in the corner, 880 00:46:55,560 --> 00:46:57,600 which then, in turn, by the way, 881 00:46:57,600 --> 00:47:02,080 sends that signal across a cable under the Atlantic 882 00:47:02,080 --> 00:47:03,440 at the speed of light 883 00:47:03,440 --> 00:47:05,640 to some server in the middle of America, 884 00:47:05,640 --> 00:47:07,400 which works out what the answer is. 885 00:47:07,400 --> 00:47:11,640 Which then returns the signal back on those cables under the Atlantic, 886 00:47:11,640 --> 00:47:14,680 out into my Wi-Fi router in my living room, 887 00:47:14,680 --> 00:47:18,880 which then beams itself invisibly across my living room 888 00:47:18,880 --> 00:47:23,360 back into the smart speaker to hear the response from Alexa. 889 00:47:23,360 --> 00:47:26,000 All within the blink of an eye. 890 00:47:26,000 --> 00:47:28,280 It's just incredible! 891 00:47:28,280 --> 00:47:33,240 That last step of the process, it's all thanks to these little antennae just here, 892 00:47:33,240 --> 00:47:36,240 and the Wi-Fi chip that sits just here. 893 00:47:36,240 --> 00:47:39,120 You can see a little wire that connects them up together. 894 00:47:39,120 --> 00:47:42,320 But how do those tiny bits of metal and the chip 895 00:47:42,320 --> 00:47:44,920 send and receive the invisible information 896 00:47:44,920 --> 00:47:47,960 your virtual assistant needs through the air? 897 00:47:50,600 --> 00:47:54,280 It all began back in 1895 with a chap called Marconi. 898 00:47:55,240 --> 00:47:59,480 And he was trying to exploit a new discovery in physics - radio waves. 899 00:47:59,480 --> 00:48:02,600 Radio waves will travel long distances. 900 00:48:02,600 --> 00:48:07,120 To send and receive them, he worked out that you need an antenna. 901 00:48:09,000 --> 00:48:12,240 The idea was that when radio waves come into contact with it, 902 00:48:12,240 --> 00:48:14,360 a tiny electrical current was created 903 00:48:14,360 --> 00:48:18,280 that could be converted into, say, a sound or a picture. 904 00:48:18,280 --> 00:48:21,480 Essentially, this is how the antenna in your virtual assistant 905 00:48:21,480 --> 00:48:23,680 picks up Wi-Fi. 906 00:48:23,680 --> 00:48:26,160 So, end of story? Not quite. 907 00:48:30,200 --> 00:48:32,560 There was a problem - jamming. 908 00:48:33,720 --> 00:48:37,080 Jamming is when a radio signal on a particular frequency 909 00:48:37,080 --> 00:48:41,080 is drowned out by a stronger signal on that same frequency. 910 00:48:42,360 --> 00:48:45,640 As the 1940s dawned and Europe was at war... 911 00:48:45,640 --> 00:48:48,480 Yes, OK, we're back with war again. 912 00:48:48,480 --> 00:48:52,000 ..it was believed that the Germans would use jamming 913 00:48:52,000 --> 00:48:54,360 to block the radio frequency the Allies used 914 00:48:54,360 --> 00:48:56,480 to guide torpedoes to their target. 915 00:48:56,480 --> 00:48:58,320 Giving Hitler the upper hand. 916 00:49:02,600 --> 00:49:05,560 Enter stage left Hollywood star Hedy Lamarr, 917 00:49:05,560 --> 00:49:07,840 known as Hollywood's most beautiful woman. 918 00:49:07,840 --> 00:49:10,080 I could promise you so many things. 919 00:49:10,080 --> 00:49:13,240 But she had a parallel life as a scientist. 920 00:49:13,240 --> 00:49:15,320 They wouldn't think anything wrong of it. 921 00:49:15,320 --> 00:49:18,080 And she'd come up with a way to prevent the Germans 922 00:49:18,080 --> 00:49:20,440 from jamming Allied radio signals. 923 00:49:20,440 --> 00:49:22,960 She came up with a very elegant idea. 924 00:49:22,960 --> 00:49:26,880 If it's really easy to jam up one frequency, 925 00:49:26,880 --> 00:49:29,120 why not split up the message 926 00:49:29,120 --> 00:49:32,440 and send it across lots of different frequencies? 927 00:49:33,560 --> 00:49:36,320 Her idea, now known as frequency hopping, 928 00:49:36,320 --> 00:49:38,320 came to her through a chance encounter 929 00:49:38,320 --> 00:49:40,600 with pianist George Antheil. 930 00:49:41,560 --> 00:49:43,480 PIANO RECITAL 931 00:49:43,480 --> 00:49:45,600 A bit more scene-setting for you. 932 00:49:47,080 --> 00:49:49,760 Er...unfortunately, I can't actually, 933 00:49:49,760 --> 00:49:51,120 I can't actually play the piano. 934 00:49:51,120 --> 00:49:54,240 This is a player piano, which was a very popular thing 935 00:49:54,240 --> 00:49:58,240 around the time that George and Hedy were sat together. 936 00:49:59,480 --> 00:50:02,480 The story goes that it was the way the player piano worked 937 00:50:02,480 --> 00:50:05,360 that gave Hedy her frequency-hopping brainwave. 938 00:50:07,160 --> 00:50:08,760 Now, the way that these things work 939 00:50:08,760 --> 00:50:10,960 is actually quite simple and quite clever. 940 00:50:12,680 --> 00:50:15,120 The music is stored on a paper drum. 941 00:50:15,120 --> 00:50:19,200 It has 88 columns, one column for each note on the piano. 942 00:50:20,400 --> 00:50:23,360 And the way that it works is, if there's a hole, 943 00:50:23,360 --> 00:50:26,160 it will play the note. No hole, it doesn't play the note. 944 00:50:26,160 --> 00:50:28,600 And it can tell just by using air suction. 945 00:50:28,600 --> 00:50:31,560 Inspired by this, Hedy's idea was to distribute 946 00:50:31,560 --> 00:50:35,320 the torpedo-guidance signals in chunks. 947 00:50:35,320 --> 00:50:38,600 Each signal was turned into a punched-paper sequence. 948 00:50:38,600 --> 00:50:40,320 But instead of different musical notes, 949 00:50:40,320 --> 00:50:42,720 it related to different frequencies. 950 00:50:42,720 --> 00:50:46,720 And if the ship sent it to the torpedo in this form, 951 00:50:46,720 --> 00:50:48,480 well, then the enemy wouldn't know 952 00:50:48,480 --> 00:50:50,720 the unique sequence of frequency hops, 953 00:50:50,720 --> 00:50:52,760 so it couldn't jam the signal. 954 00:50:52,760 --> 00:50:56,240 And in 1942, they applied for a patent 955 00:50:56,240 --> 00:50:59,160 for a secret communication system. 956 00:50:59,160 --> 00:51:01,760 The patent was approved, by the way. 957 00:51:01,760 --> 00:51:03,360 So they sent it off to the Navy, 958 00:51:03,360 --> 00:51:07,480 fully expecting the Navy to just welcome it with open arms. 959 00:51:07,480 --> 00:51:10,720 But actually, the Navy ignored it. 960 00:51:10,720 --> 00:51:13,240 You've been avoiding me. Why? 961 00:51:13,240 --> 00:51:18,680 Now, it's possible that this is a bit of a sign of the times. 962 00:51:18,680 --> 00:51:20,800 That Hedy Lamarr, super famous as she was, 963 00:51:20,800 --> 00:51:23,360 was famous for being a femme fatale. 964 00:51:23,360 --> 00:51:26,520 And surely, a beautiful screen goddess 965 00:51:26,520 --> 00:51:29,400 couldn't also be a brilliant scientist? 966 00:51:29,400 --> 00:51:33,400 On top of that, her invention was quite cumbersome. 967 00:51:33,400 --> 00:51:35,840 And so, it was forgotten about for decades. 968 00:51:37,560 --> 00:51:40,520 Until yet another war. Well, almost. 969 00:51:40,520 --> 00:51:43,880 Ships moved out to establish a quarantine line, through which... 970 00:51:43,880 --> 00:51:46,360 During the Cuban Missile Crisis of 1962, 971 00:51:46,360 --> 00:51:48,760 frequency hopping was finally installed on ships 972 00:51:48,760 --> 00:51:51,400 to prevent Russian signal-jamming. 973 00:51:51,400 --> 00:51:53,960 The democracies had served notice on the Russians that any 974 00:51:53,960 --> 00:51:57,360 aggressive threat in the Western Hemisphere would not be tolerated. 975 00:51:57,360 --> 00:51:59,760 So, how did we get from Hedy's frequency hopping 976 00:51:59,760 --> 00:52:02,880 to your virtual assistant's Wi-Fi capabilities? 977 00:52:04,600 --> 00:52:06,680 Cut to the early 1970s. 978 00:52:11,520 --> 00:52:14,400 On the tropical islands of Hawaii, 979 00:52:14,400 --> 00:52:16,520 far more tranquil than all that war footage... 980 00:52:18,000 --> 00:52:20,320 ..engineers at the university are trying to connect up 981 00:52:20,320 --> 00:52:22,480 the computers in their campuses. 982 00:52:22,480 --> 00:52:24,400 But there is a problem. 983 00:52:25,720 --> 00:52:28,240 The campuses are spread across several islands, 984 00:52:28,240 --> 00:52:31,840 linked up with expensive and error-prone cabling. 985 00:52:31,840 --> 00:52:36,720 And so, they use radio signals sent between antennae and ALOHAnet. 986 00:52:36,720 --> 00:52:39,480 A pioneering wireless computer network is born. 987 00:52:40,480 --> 00:52:42,800 In Hawaii? Yes indeed. 988 00:52:44,680 --> 00:52:46,920 Good job it wasn't invented in Germany. 989 00:52:46,920 --> 00:52:50,040 GUTEN-TAGnet doesn't have quite the same ring to it. 990 00:52:51,920 --> 00:52:54,040 But as Wi-Fi took off, 991 00:52:54,040 --> 00:52:58,200 the narrow range of radio frequencies used became crowded, 992 00:52:58,200 --> 00:53:02,000 and signals from all of the computers began to interfere with each other. 993 00:53:02,000 --> 00:53:06,560 In 1997, 55 years after she'd had the idea, 994 00:53:06,560 --> 00:53:10,240 Hedy's frequency-hopping technique would become the foundation 995 00:53:10,240 --> 00:53:11,840 to solving that problem. 996 00:53:11,840 --> 00:53:14,320 I've loved you since before I knew the meaning of the word! 997 00:53:14,320 --> 00:53:17,320 By spreading the signal between multiple frequencies 998 00:53:17,320 --> 00:53:20,160 to avoid getting clogged up on any one wavelength, 999 00:53:20,160 --> 00:53:22,680 modern Wi-Fi usage became possible. 1000 00:53:27,000 --> 00:53:29,600 Today, we have billions of devices, 1001 00:53:29,600 --> 00:53:33,360 all communicating with the Internet via their Wi-Fi antennas. 1002 00:53:33,360 --> 00:53:38,520 And that is a big deal for Alexa and its multi-device smart home. 1003 00:53:38,520 --> 00:53:43,200 We have over 40 devices in here, and they constantly change. 1004 00:53:44,320 --> 00:53:48,000 That is Marja Koopmans, director of Amazon Smart Home. 1005 00:53:48,000 --> 00:53:51,640 She is responsible for making Alexa connect via Wi-Fi. 1006 00:53:54,040 --> 00:53:56,720 We're quite excited at how far we've come along. 1007 00:53:56,720 --> 00:54:00,400 Over 140,000 different types of devices 1008 00:54:00,400 --> 00:54:03,720 are able to connect with Alexa. 1009 00:54:03,720 --> 00:54:05,800 ALEXA: ..start vacuuming. 1010 00:54:05,800 --> 00:54:08,000 But does having so many of our smart devices 1011 00:54:08,000 --> 00:54:11,400 controlled from a single source, the virtual assistant, 1012 00:54:11,400 --> 00:54:14,000 invite security problems like hacking? 1013 00:54:14,000 --> 00:54:15,400 Alexa, follow me. 1014 00:54:15,400 --> 00:54:17,000 ALEXA: Following is now on. 1015 00:54:17,000 --> 00:54:19,280 Marja doesn't seem to think so. 1016 00:54:19,280 --> 00:54:22,240 Things like privacy and security 1017 00:54:22,240 --> 00:54:25,120 are built in from the get-go. 1018 00:54:25,120 --> 00:54:28,480 We have multiple technology layers in place 1019 00:54:28,480 --> 00:54:31,560 to safeguard for that type of scenario. 1020 00:54:31,560 --> 00:54:33,880 Including encrypted two-ways. 1021 00:54:33,880 --> 00:54:36,160 GIGGLING 1022 00:54:38,320 --> 00:54:40,040 With wireless connectivity, 1023 00:54:40,040 --> 00:54:43,640 the virtual assistant no longer needs to be static. 1024 00:54:43,640 --> 00:54:47,960 Wheels will let it move and cameras will make it see. 1025 00:54:47,960 --> 00:54:50,280 This is the future. 1026 00:54:51,400 --> 00:54:54,400 The future of Alexa is not just the future of Alexa, 1027 00:54:54,400 --> 00:54:56,000 it's the future of robots. 1028 00:54:58,400 --> 00:55:01,880 That's Ken Washington, head of Consumer Robotics. 1029 00:55:03,400 --> 00:55:06,160 We've put Alexa into a robot. 1030 00:55:06,160 --> 00:55:07,640 OK, let me see him, let me see him. 1031 00:55:07,640 --> 00:55:09,360 All right, cool. Let's take a look. 1032 00:55:09,360 --> 00:55:11,600 Astro? Come here. 1033 00:55:11,600 --> 00:55:13,680 Named Astro... 1034 00:55:13,680 --> 00:55:15,440 "I'm coming." Hello, Astro! 1035 00:55:15,440 --> 00:55:19,440 ..this is Amazon's first foray into the world of domestic robots. 1036 00:55:19,440 --> 00:55:20,800 That is actually quite sweet. 1037 00:55:20,800 --> 00:55:23,720 Check out what he looks like. The eyes can convey. 1038 00:55:23,720 --> 00:55:27,920 It can convey, sadness, happiness, joy and, you know, anger. 1039 00:55:29,600 --> 00:55:32,160 Not sure I'd want an angry robot in my house. 1040 00:55:36,200 --> 00:55:39,480 However, Astro does have some other handy features. 1041 00:55:40,880 --> 00:55:42,560 So, Astro? 1042 00:55:44,240 --> 00:55:46,000 Act like a horse. 1043 00:55:46,000 --> 00:55:48,280 ASTRO WHINNIES AND HANNAH LAUGHS 1044 00:55:49,760 --> 00:55:51,880 That's really cute! 1045 00:55:51,880 --> 00:55:55,280 In addition to being cute, it can also be quite useful. 1046 00:55:55,280 --> 00:55:59,800 You can use Astro to monitor your home when you're not home. 1047 00:55:59,800 --> 00:56:03,640 All through an app on your phone and Astro's onboard camera. 1048 00:56:03,640 --> 00:56:05,760 And with this periscope, 1049 00:56:05,760 --> 00:56:07,840 I check on my doors 1050 00:56:07,840 --> 00:56:11,280 to make sure that the doors are closed and that they're locked. 1051 00:56:11,280 --> 00:56:14,280 And then, the obvious question is, stairs? 1052 00:56:19,600 --> 00:56:21,480 You know, it doesn't do stairs. 1053 00:56:21,480 --> 00:56:23,400 Astro is a single-floor robot. 1054 00:56:24,680 --> 00:56:28,800 Giving our voice assistance a body is really the future of this? 1055 00:56:28,800 --> 00:56:33,080 In my head, every home's going to have at least one robot. 1056 00:56:33,080 --> 00:56:35,840 They're going to help you live safer lives. 1057 00:56:35,840 --> 00:56:37,880 This is just the beginning. 1058 00:56:37,880 --> 00:56:39,560 Unless you've got stairs, that is. 1059 00:56:43,120 --> 00:56:46,040 Whether you think of your virtual assistant as being like 1060 00:56:46,040 --> 00:56:47,400 a trusted family member, 1061 00:56:47,400 --> 00:56:51,120 or you worry about it being the ears of a giant corporation, 1062 00:56:51,120 --> 00:56:54,000 there is no doubt that they are having an impact. 1063 00:56:55,240 --> 00:56:59,120 I think that it demonstrates just how quickly ungrateful 1064 00:56:59,120 --> 00:57:01,320 we become about new technology. 1065 00:57:01,320 --> 00:57:02,760 I am as guilty as the next person 1066 00:57:02,760 --> 00:57:04,560 of getting cross with my voice assistant 1067 00:57:04,560 --> 00:57:06,960 when it doesn't do exactly what I want the first time. 1068 00:57:06,960 --> 00:57:08,760 But maybe most importantly of all, 1069 00:57:08,760 --> 00:57:12,320 I think that voice assistants are an example 1070 00:57:12,320 --> 00:57:15,680 of something that we are seeing happen 1071 00:57:15,680 --> 00:57:19,240 where it's the technology that leads the way. 1072 00:57:19,240 --> 00:57:23,080 And it's only later that society asks the question 1073 00:57:23,080 --> 00:57:26,280 of whether it's something that we really want in our lives. 1074 00:57:27,240 --> 00:57:29,920 And that, I think, is a trend 1075 00:57:29,920 --> 00:57:32,600 that we're going to have to deal with a lot more in future. 1076 00:57:34,320 --> 00:57:37,160 Next time, it's the electric car. 1077 00:57:37,160 --> 00:57:40,400 I will find out how frogs and Frankenstein... 1078 00:57:40,400 --> 00:57:41,680 Hello. Look at you! 1079 00:57:41,680 --> 00:57:43,920 ..led to Elon Musk's supercars. 1080 00:57:43,920 --> 00:57:45,400 Oh, my God! 1081 00:57:45,400 --> 00:57:49,520 Why 19th-century meat-packing almost derailed the EV revolution. 1082 00:57:49,520 --> 00:57:51,480 They're not very good escape vehicles. 1083 00:57:51,480 --> 00:57:54,800 And how not to drive a multimillion-dollar 1084 00:57:54,800 --> 00:57:56,520 prototype car of the future. 1085 00:57:56,520 --> 00:57:58,720 Oh, God! Oh, BLEEP! I'm sorry! I'm sorry! 1086 00:58:01,800 --> 00:58:03,600 To explore the interesting dilemmas 1087 00:58:03,600 --> 00:58:06,040 behind some familiar pieces of technology 1088 00:58:06,040 --> 00:58:08,360 and to hear some of my own thoughts, go to... 1089 00:58:11,840 --> 00:58:14,200 ..and follow the links to the Open University.