1 00:00:02,000 --> 00:00:06,280 100 years ago, most lifts were driven by trained operators. 2 00:00:06,280 --> 00:00:08,640 The technology was there to replace them, 3 00:00:08,640 --> 00:00:10,800 but people just didn't feel comfortable 4 00:00:10,800 --> 00:00:12,800 with the idea of automation. 5 00:00:12,800 --> 00:00:14,320 Level one, please. 6 00:00:18,360 --> 00:00:24,280 And then lift designers made some small but ground-breaking changes. 7 00:00:28,640 --> 00:00:31,320 TANNOY: First floor, Christmas lectures. 8 00:00:32,400 --> 00:00:35,400 Add that to a stop button and some relaxing music 9 00:00:35,400 --> 00:00:39,360 and suddenly trust in automated lifts soared. 10 00:00:39,360 --> 00:00:41,080 And here we are today. 11 00:00:41,080 --> 00:00:43,160 But what about today? 12 00:00:43,160 --> 00:00:45,760 Should we trust the machines that surround us 13 00:00:45,760 --> 00:00:48,320 or are we right to be cautious? 14 00:01:13,800 --> 00:01:16,320 CHEERING AND APPLAUSE 15 00:01:31,840 --> 00:01:33,840 Welcome to the Christmas Lectures. 16 00:01:33,840 --> 00:01:36,320 I'm Dr Hannah Fry and tonight we're going to ask 17 00:01:36,320 --> 00:01:38,400 whether we should trust the maths. 18 00:01:38,400 --> 00:01:42,360 Just how far should we be going with our mathematical skills? 19 00:01:42,360 --> 00:01:45,640 And let's demonstrate those mathematical skills first off 20 00:01:45,640 --> 00:01:49,880 because we are joined by Scott Hamlin and your BMX bike. Hello. 21 00:01:49,880 --> 00:01:53,840 And a rather carefully placed ramp just here. 22 00:01:53,840 --> 00:01:57,080 Very careful indeed, to make sure it's absolutely amazing. 23 00:01:57,080 --> 00:02:00,600 And you've been in since 8:00am this morning calculating exactly 24 00:02:00,600 --> 00:02:03,760 where this ramp should be, the shape of the ramp, everything, 25 00:02:03,760 --> 00:02:06,840 because it's quite a short space we've got here. Yes, absolutely. 26 00:02:06,840 --> 00:02:09,760 We need to make sure we can get enough velocity to give us lift 27 00:02:09,760 --> 00:02:12,680 off the ramp and I need to use my personal calculations... 28 00:02:12,680 --> 00:02:15,360 ..OK, instincts, to be able to perform my stunts 29 00:02:15,360 --> 00:02:18,360 and stop before we crash into the wall. 30 00:02:18,360 --> 00:02:21,400 And when you say stunts, Scott, what are you going to do? 31 00:02:21,400 --> 00:02:24,880 Well, it depends. It might be a backflip if people want to see one. 32 00:02:24,880 --> 00:02:26,960 Do you want to see a backflip? ALL: Yes! 33 00:02:26,960 --> 00:02:29,120 Yeah, they want to see a backflip. 34 00:02:29,120 --> 00:02:32,240 All right, Scott, you ready to give this a go? You guys ready? 35 00:02:32,240 --> 00:02:34,560 ALL: Yes. Right, here we go. 36 00:02:34,560 --> 00:02:37,320 Come on, then. We'll give you a countdown, Scott, 37 00:02:37,320 --> 00:02:39,320 when you're in place. Cool. 38 00:02:42,800 --> 00:02:44,320 Here we go. 39 00:02:47,360 --> 00:02:49,920 And then I'm going to get really far out of the way. 40 00:02:49,920 --> 00:02:51,280 LAUGHTER 41 00:02:51,280 --> 00:02:53,320 OK, whenever you're happy. 42 00:02:53,320 --> 00:02:55,400 Right, are we ready? Yeah. 43 00:02:55,400 --> 00:02:58,200 You happy, Scott? I didn't hear the kids. 44 00:02:59,240 --> 00:03:01,880 Are you guys ready for this? ALL: Yes! 45 00:03:01,880 --> 00:03:05,320 All right, we're going to give you a countdown, Scott. Go for it. 46 00:03:05,320 --> 00:03:06,760 Five... 47 00:03:06,760 --> 00:03:11,400 ALL: ..four, three, two, one, go! 48 00:03:13,040 --> 00:03:14,760 Wow! 49 00:03:14,760 --> 00:03:17,080 APPLAUSE 50 00:03:20,640 --> 00:03:22,360 Woo! 51 00:03:24,080 --> 00:03:25,440 I'm alive. 52 00:03:29,720 --> 00:03:31,560 Well done, Scott. 53 00:03:31,560 --> 00:03:35,640 That was some tight calculations going on. Yes, it certainly was. 54 00:03:35,640 --> 00:03:38,760 I'm glad it paid off. Thanks to you guys for making the noise. 55 00:03:38,760 --> 00:03:42,080 Well, no, thank you to you. Scott Hamlin, thank you very much. 56 00:03:42,080 --> 00:03:44,520 CHEERING AND APPLAUSE 57 00:03:47,880 --> 00:03:51,040 Now, OK, we all know that maths is amazing at this kind of stuff. 58 00:03:51,040 --> 00:03:53,280 It's out there in physics and engineering. 59 00:03:53,280 --> 00:03:56,600 It's doing a brilliant job, just as long as you do your sums correctly. 60 00:03:56,600 --> 00:03:58,960 But I want to tell you a little story about a bridge 61 00:03:58,960 --> 00:04:02,640 that I think demonstrates how it's quite a lot easier said than done, 62 00:04:02,640 --> 00:04:06,000 because this here, this is the Millennium Bridge in London. 63 00:04:06,000 --> 00:04:08,640 It had its grand opening in the year 2000. 64 00:04:08,640 --> 00:04:11,560 But you might also know this bridge by its nickname. 65 00:04:11,560 --> 00:04:14,400 This is known as the Wobbly Bridge 66 00:04:14,400 --> 00:04:17,880 because of something that happened very soon after it opened. 67 00:04:17,880 --> 00:04:20,800 Now, all bridges, including this one, they're built to move 68 00:04:20,800 --> 00:04:23,120 left and right just a little bit. It's no biggie. 69 00:04:23,120 --> 00:04:25,720 But there was something about this particular bridge 70 00:04:25,720 --> 00:04:27,800 that the designers hadn't thought of. 71 00:04:27,800 --> 00:04:30,920 They'd missed off something quite important in their equations. 72 00:04:30,920 --> 00:04:33,560 Let me explain what happened here with one of these - 73 00:04:33,560 --> 00:04:35,160 a little metronome. 74 00:04:35,160 --> 00:04:38,400 Now, this thing here ticks and tocks in time. 75 00:04:38,400 --> 00:04:41,600 If I wanted to write a set of equations for this metronome, 76 00:04:41,600 --> 00:04:44,640 it would be quite simple - some very straightforward physics. 77 00:04:44,640 --> 00:04:47,120 And if I wanted to write some equations for a number 78 00:04:47,120 --> 00:04:50,080 of metronomes, I'd just do the same thing over and over again. 79 00:04:50,080 --> 00:04:53,000 It's not like these things can communicate with one another. 80 00:04:53,000 --> 00:04:56,120 I can treat each one as though they're completely individual. 81 00:04:56,120 --> 00:04:59,560 But now let's see what happens when they're on a bridge 82 00:04:59,560 --> 00:05:01,960 with just a little bit of movement, 83 00:05:01,960 --> 00:05:04,360 because something rather intriguing happens. 84 00:05:04,360 --> 00:05:07,360 I don't want that to fall too far. Let's go from there. OK. 85 00:05:07,360 --> 00:05:10,880 Let's see what happens when these things are on a bridge 86 00:05:10,880 --> 00:05:15,640 that can move itself left and right just a little bit. 87 00:05:15,640 --> 00:05:20,640 OK, so now that these things can move left and right, 88 00:05:20,640 --> 00:05:23,480 something a little bit unusual happens, 89 00:05:23,480 --> 00:05:27,080 because every time that a metronome is ticking or tocking 90 00:05:27,080 --> 00:05:28,880 in one direction, 91 00:05:28,880 --> 00:05:32,080 that left and right movement means that the whole bridge 92 00:05:32,080 --> 00:05:35,360 will just be knocked ever so slightly left and right. 93 00:05:35,360 --> 00:05:37,360 What that means is that now 94 00:05:37,360 --> 00:05:40,600 these metronomes can effectively start listening to each other. 95 00:05:40,600 --> 00:05:43,360 They're all now connected by the bridge, 96 00:05:43,360 --> 00:05:46,160 which means that very, very slowly, 97 00:05:46,160 --> 00:05:49,400 you just see it moving very slightly left and right 98 00:05:49,400 --> 00:05:54,120 and very slowly they start to synchronise with one another... 99 00:05:55,120 --> 00:05:57,560 ..like a creepy metronome army. 100 00:05:58,840 --> 00:06:01,400 Now, this is something that my equations 101 00:06:01,400 --> 00:06:03,480 just wouldn't have taken into account. 102 00:06:03,480 --> 00:06:06,520 But humans, like metronomes, actually have to be handled 103 00:06:06,520 --> 00:06:09,600 with quite a lot of care and so for that I'm going to put you 104 00:06:09,600 --> 00:06:12,040 in the very capable hands of my good friend 105 00:06:12,040 --> 00:06:14,040 and mathematician Matt Parker. 106 00:06:14,040 --> 00:06:16,640 Oh, hey. So, thanks, Hannah. 107 00:06:16,640 --> 00:06:19,320 I'm over here in the library at the Royal Institution 108 00:06:19,320 --> 00:06:21,840 where, instead of very small metronomes, 109 00:06:21,840 --> 00:06:25,880 we have a massive wobbly bridge simulator. 110 00:06:25,880 --> 00:06:28,640 We've borrowed this from the University of Cambridge. 111 00:06:28,640 --> 00:06:31,880 They use it to test things like, well, full-size bridges. 112 00:06:31,880 --> 00:06:36,120 It's a very firm structure with a tray attached to it, 113 00:06:36,120 --> 00:06:38,800 which is able to move a little bit side to side. 114 00:06:38,800 --> 00:06:41,880 We've got two treadmills. I'm joined by Dylan here, 115 00:06:41,880 --> 00:06:44,840 who's going to walk on one of these treadmills with me. 116 00:06:44,840 --> 00:06:49,040 If we turn these on, in theory, we'll start walking, 117 00:06:49,040 --> 00:06:51,440 slowly to start with. There we go. 118 00:06:51,440 --> 00:06:55,080 And then, shall we go up to four? Let's try a speed of four. 119 00:06:55,080 --> 00:06:58,560 So now we are trying to walk on a bridge 120 00:06:58,560 --> 00:07:01,400 which is not staying still at all 121 00:07:01,400 --> 00:07:04,640 and it feels a bit like being on a boat, maybe, 122 00:07:04,640 --> 00:07:06,400 where it's moving around 123 00:07:06,400 --> 00:07:09,120 and you're trying to compensate for that movement, 124 00:07:09,120 --> 00:07:12,320 and it means we have perfectly synched up our walking 125 00:07:12,320 --> 00:07:14,320 and that's causing it to move... 126 00:07:14,320 --> 00:07:16,640 ..I'll say a concerning amount. 127 00:07:16,640 --> 00:07:20,080 Yes. You're keeping a brave face but it's terrifying up here. 128 00:07:20,080 --> 00:07:23,480 All our movements is causing it to shift backwards and forwards. 129 00:07:23,480 --> 00:07:27,080 Hannah, you can imagine what would happen if you had loads of people 130 00:07:27,080 --> 00:07:29,760 doing this on a much bigger structure. 131 00:07:29,760 --> 00:07:31,640 Just imagine, indeed. 132 00:07:31,640 --> 00:07:34,360 But it turns out this is precisely what had happened 133 00:07:34,360 --> 00:07:37,120 because the people who designed the Millennium Bridge 134 00:07:37,120 --> 00:07:40,280 hadn't taken into account the fact that people can affect each other 135 00:07:40,280 --> 00:07:42,200 with the way that they're walking. 136 00:07:42,200 --> 00:07:45,400 Those small movements left to right suddenly became a very big deal 137 00:07:45,400 --> 00:07:48,640 and it meant that on the opening day, you had hundreds of people 138 00:07:48,640 --> 00:07:52,840 walking over an £18 million bridge, the beginning of a new millennium, 139 00:07:52,840 --> 00:07:56,120 and every single one of them was hanging on for dear life. 140 00:07:56,120 --> 00:08:00,600 Look at that. That's British ingenuity at its finest, just there. 141 00:08:00,600 --> 00:08:03,080 But there is an important point in all of this, 142 00:08:03,080 --> 00:08:05,080 with wobbly bridges and BMX bikes. 143 00:08:05,080 --> 00:08:08,040 Maths can do an amazing job, but only if your equations 144 00:08:08,040 --> 00:08:11,400 actually match up to the world that you're describing. 145 00:08:11,400 --> 00:08:14,720 Just as long as you've got the right equations for bikes and bridges, 146 00:08:14,720 --> 00:08:17,120 you can be certain of what's going to happen next. 147 00:08:17,120 --> 00:08:20,200 But there are some things that are quite hard to write equations for 148 00:08:20,200 --> 00:08:21,680 in the first place. 149 00:08:21,680 --> 00:08:25,080 OK, let's imagine that it's long into the future and you're trying 150 00:08:25,080 --> 00:08:27,640 to write an algorithm that can help a doctor 151 00:08:27,640 --> 00:08:30,120 work out what's wrong with their patients. 152 00:08:30,120 --> 00:08:33,080 Now, a doctor's job is quite different to that of an engineer 153 00:08:33,080 --> 00:08:36,240 or a physicist because if someone just comes in with a headache, 154 00:08:36,240 --> 00:08:39,040 a doctor can't just take measurements of what's wrong 155 00:08:39,040 --> 00:08:41,120 and end up with an exact answer 156 00:08:41,120 --> 00:08:44,360 because that headache could mean a whole host of different things. 157 00:08:44,360 --> 00:08:47,840 Somehow, the doctor has to use a whole bunch of clues 158 00:08:47,840 --> 00:08:51,120 to build a picture of what might be wrong with you. 159 00:08:51,120 --> 00:08:54,320 Now, this is the difference between calculating the answer 160 00:08:54,320 --> 00:08:58,360 and just making your best possible guess, and no-one had any idea 161 00:08:58,360 --> 00:09:01,840 how to do that in an equation until the 1700s, 162 00:09:01,840 --> 00:09:05,880 when the Reverend Thomas Bayes thought of a very clever game. 163 00:09:05,880 --> 00:09:08,400 Now, we're going to play a version of this game 164 00:09:08,400 --> 00:09:10,360 just with a little bit more fire, 165 00:09:10,360 --> 00:09:14,760 so who would like to come down and volunteer for this? 166 00:09:14,760 --> 00:09:18,240 Erm, let's go... Let's go for you just there. 167 00:09:18,240 --> 00:09:20,600 Round of applause as she comes to the stage. 168 00:09:20,600 --> 00:09:22,280 APPLAUSE 169 00:09:22,280 --> 00:09:24,840 What's your name? Emma. Emma? Yeah. 170 00:09:24,840 --> 00:09:27,840 OK, Emma. Right, what we're going to do is, we're going to play 171 00:09:27,840 --> 00:09:30,600 a version of this game and it involves this red hat, 172 00:09:30,600 --> 00:09:33,120 if you don't mind just popping that on your head. 173 00:09:33,120 --> 00:09:36,400 Now, just so that we can get your view of things. Oh, it's a bit... 174 00:09:36,400 --> 00:09:38,880 Hold on one second. Let me tighten this up. 175 00:09:38,880 --> 00:09:41,120 Now, just so we can get your view of things, 176 00:09:41,120 --> 00:09:44,680 I've also got a version of this red hat over here for this camera 177 00:09:44,680 --> 00:09:46,720 so we'll be able to see what you see. 178 00:09:46,720 --> 00:09:49,680 The other thing that we need for this game is a whole host 179 00:09:49,680 --> 00:09:53,080 of balloons, which is just coming on...just coming on behind you. 180 00:09:53,080 --> 00:09:56,040 If you want to turn around, Emma. Just have a little look... 181 00:09:56,040 --> 00:09:58,320 Just have a little look at these balloons. 182 00:09:58,320 --> 00:10:01,640 What colour are these balloons, Emma? You want to stand over here. 183 00:10:01,640 --> 00:10:04,400 Red. What colour are the balloons? Red. Red. OK. 184 00:10:04,400 --> 00:10:06,720 So if we look through this camera now, 185 00:10:06,720 --> 00:10:10,360 we will be able to see that they do indeed all look red, 186 00:10:10,360 --> 00:10:13,360 and yet... Just step over here. Sorry. 187 00:10:13,360 --> 00:10:17,000 And yet, to everyone in this audience, we can see that, in fact, 188 00:10:17,000 --> 00:10:21,680 what you're looking at are 99 orange balloons and one yellow one. 189 00:10:21,680 --> 00:10:23,960 Now, we all know where the yellow balloon is. 190 00:10:23,960 --> 00:10:25,840 No-one's allowed to give it away. 191 00:10:25,840 --> 00:10:29,600 But your job, Emma, is to try and pop that yellow balloon. 192 00:10:29,600 --> 00:10:33,800 And if you manage it, we're all going to explode in excitement. 193 00:10:33,800 --> 00:10:35,600 And if you fail, 194 00:10:35,600 --> 00:10:38,600 we're going to respond with disappointed silence, OK? 195 00:10:38,600 --> 00:10:42,840 So here we go. You get to pick a balloon at random and pop one. 196 00:10:42,840 --> 00:10:46,360 Which one do you want to pop? You stand here and Matt will do it. 197 00:10:46,360 --> 00:10:48,280 Up? Up? 198 00:10:48,280 --> 00:10:50,280 Down. This one? Yeah. 199 00:10:50,280 --> 00:10:52,280 Perfect. That one there? Yeah. 200 00:10:52,280 --> 00:10:54,080 Ready? Here we go. 201 00:10:54,080 --> 00:10:57,120 That is not the yellow balloon. That's OK. 202 00:10:57,120 --> 00:11:00,040 It's pretty hard in the beginning. I mean, you know... 203 00:11:00,040 --> 00:11:02,360 How can you possibly guess it first time? 204 00:11:02,360 --> 00:11:05,600 What we're going to do is, audience, we're going to help her here. 205 00:11:05,600 --> 00:11:09,600 So, we are going to tell you, on my cue, we're going to tell you, 206 00:11:09,600 --> 00:11:12,080 based on the last balloon that you popped, 207 00:11:12,080 --> 00:11:14,680 whether you should go higher or lower, left or right. 208 00:11:14,680 --> 00:11:16,920 If you think it's higher, say higher. 209 00:11:16,920 --> 00:11:18,960 if you think it's lower, say lower. 210 00:11:18,960 --> 00:11:21,200 If you think it's neither higher nor lower, 211 00:11:21,200 --> 00:11:23,840 I want exactly 50% to say higher, 50% to say lower, 212 00:11:23,840 --> 00:11:26,880 and I'll let you work out between yourselves which one is which. 213 00:11:26,880 --> 00:11:29,800 OK, so, based on her last balloon pop, 214 00:11:29,800 --> 00:11:32,040 should she go higher or lower? 215 00:11:32,040 --> 00:11:33,560 ALL: Lower. 216 00:11:33,560 --> 00:11:36,480 And should she go left or right? 217 00:11:36,480 --> 00:11:39,200 ALL: Left. OK, where do you want to pop? 218 00:11:39,200 --> 00:11:40,640 Erm... 219 00:11:41,680 --> 00:11:43,040 ..lower. 220 00:11:43,040 --> 00:11:44,760 Left a bit. 221 00:11:46,040 --> 00:11:48,160 That one. That one. Ready? 222 00:11:49,200 --> 00:11:51,640 Oh, OK. We'll give her another go. 223 00:11:51,640 --> 00:11:54,120 Do you want to go, based on the last balloon pop, 224 00:11:54,120 --> 00:11:57,320 should she go higher or lower? ALL: Lower. 225 00:11:57,320 --> 00:12:00,360 And should she go left or right? MIXED RESPONSES 226 00:12:00,360 --> 00:12:03,360 Oh, interesting. OK, which way do you want to go? 227 00:12:04,400 --> 00:12:05,680 Down. 228 00:12:06,720 --> 00:12:08,280 Down. That one. 229 00:12:08,280 --> 00:12:09,680 Ready? 230 00:12:10,720 --> 00:12:12,320 EXPLOSION 231 00:12:12,320 --> 00:12:14,000 APPLAUSE 232 00:12:20,280 --> 00:12:22,600 You got there. Thank you so much, Emma. 233 00:12:22,600 --> 00:12:25,680 You got there amazingly quickly. OK, do you want to take this...? 234 00:12:25,680 --> 00:12:29,120 Tell me, what was your strategy in the first place? Just pick one. 235 00:12:29,120 --> 00:12:31,880 Just pick one at random? Yes. You were just guessing at random. 236 00:12:31,880 --> 00:12:33,680 Did our clues help you? Yes. 237 00:12:33,680 --> 00:12:36,360 You knew probably where the balloon was by the end. Yeah. 238 00:12:36,360 --> 00:12:39,560 Were you getting more and more confident? Yes. Yes. Perfect. 239 00:12:39,560 --> 00:12:42,040 All right, Emma, thank you so much. Amazing. 240 00:12:42,040 --> 00:12:44,040 APPLAUSE 241 00:12:47,080 --> 00:12:50,280 What Emma was doing there, she was demonstrating something 242 00:12:50,280 --> 00:12:53,320 that's called Bayesian thinking and, actually, it's something 243 00:12:53,320 --> 00:12:55,120 that all of us do instinctively. 244 00:12:55,120 --> 00:12:57,880 But it wasn't until Bayes thought of a version of that game 245 00:12:57,880 --> 00:13:00,720 that the world realised that you can actually write down 246 00:13:00,720 --> 00:13:03,040 that way of thinking into an equation. 247 00:13:03,040 --> 00:13:06,600 I'm really not exaggerating when I tell you that that Bayes theorem 248 00:13:06,600 --> 00:13:09,400 is one of the most important equations of all time, 249 00:13:09,400 --> 00:13:11,880 because, suddenly, it doesn't matter if you're not 250 00:13:11,880 --> 00:13:13,880 completely sure of the answer. 251 00:13:13,880 --> 00:13:16,560 You can still get a really good sense of the right answer, 252 00:13:16,560 --> 00:13:18,800 even from incomplete pieces of information. 253 00:13:18,800 --> 00:13:21,560 And that is something that is incredibly useful. 254 00:13:21,560 --> 00:13:22,920 Let me show you. 255 00:13:22,920 --> 00:13:27,360 So, OK, let's imagine that you are making a driverless car. 256 00:13:27,360 --> 00:13:30,360 Now, it hasn't got a driver in it so you need to make sure 257 00:13:30,360 --> 00:13:35,040 that you know where you are and, OK, you could use GPS to do that, 258 00:13:35,040 --> 00:13:37,400 but GPS isn't perfect. 259 00:13:37,400 --> 00:13:41,040 So, sometimes, your GPS will get your position out 260 00:13:41,040 --> 00:13:42,880 by about a metre or so. 261 00:13:42,880 --> 00:13:45,160 And if you're a human, that's fine. No big deal. 262 00:13:45,160 --> 00:13:47,040 You can work out where you are. 263 00:13:47,040 --> 00:13:50,640 But if you're a driverless car, the difference of a few metres 264 00:13:50,640 --> 00:13:53,360 can mean the difference between driving on the pavement 265 00:13:53,360 --> 00:13:56,400 and driving into oncoming traffic, which isn't ideal. 266 00:13:56,400 --> 00:13:59,360 So driverless cars, they also have cameras on board. 267 00:13:59,360 --> 00:14:02,400 Now, cameras are pretty good at letting you know where you are, 268 00:14:02,400 --> 00:14:07,400 but, again, they're not perfect because skies look a lot like water 269 00:14:07,400 --> 00:14:11,640 and, you know, lorry tarpaulin looks a lot like a clouded sky. 270 00:14:11,640 --> 00:14:15,640 The point about driverless cars is that you don't just have one thing 271 00:14:15,640 --> 00:14:17,880 that gives you exactly where you are. 272 00:14:17,880 --> 00:14:21,320 You have lots of different things that you use as clues 273 00:14:21,320 --> 00:14:23,400 to indicate where you are. 274 00:14:23,400 --> 00:14:26,360 And that is something that is especially important 275 00:14:26,360 --> 00:14:29,880 when you are driving at 200mph. 276 00:14:29,880 --> 00:14:34,640 So, this thing here, this is the world's fastest autonomous car. 277 00:14:34,640 --> 00:14:37,760 It was built for racing and it's got all kinds of different sensors 278 00:14:37,760 --> 00:14:39,720 to help it work out where it is. 279 00:14:39,720 --> 00:14:41,960 It's got little cameras here, 280 00:14:41,960 --> 00:14:44,920 it's got another kind of camera called LiDAR over here. 281 00:14:44,920 --> 00:14:47,480 it's got radar over here at the back. 282 00:14:47,480 --> 00:14:50,120 All of these are the clues for the car, 283 00:14:50,120 --> 00:14:54,120 and GPS within the computer that's just inside here. 284 00:14:54,120 --> 00:14:57,880 Now, this thing is basically a Bayesian machine. 285 00:14:57,880 --> 00:15:01,360 So, that computer, that is the size of just a lunchbox, 286 00:15:01,360 --> 00:15:04,840 is churning through trillions of calculations every second 287 00:15:04,840 --> 00:15:07,880 to make sure that this car knows where it is 288 00:15:07,880 --> 00:15:10,840 and finishes the race as quickly as possible, 289 00:15:10,840 --> 00:15:14,080 all while avoiding other competitors. 290 00:15:14,080 --> 00:15:17,360 I think that's the thing about how these modern inventions work. 291 00:15:17,360 --> 00:15:20,400 That's how they deal with uncertainty. 292 00:15:20,400 --> 00:15:24,080 They don't just have one sensor, they don't just have two sensors, 293 00:15:24,080 --> 00:15:28,320 they have a whole host of sensors that they use to layer up 294 00:15:28,320 --> 00:15:30,080 and give them information. 295 00:15:30,080 --> 00:15:33,080 That's something that's true of driverless cars, 296 00:15:33,080 --> 00:15:36,840 but it's also true of this little guy here, 297 00:15:36,840 --> 00:15:40,560 who I believe is going to follow me into the studio. 298 00:15:40,560 --> 00:15:42,040 There we go. 299 00:15:42,040 --> 00:15:43,560 Come on, then. 300 00:15:46,320 --> 00:15:48,280 Come on. Come on. 301 00:15:48,280 --> 00:15:50,200 CHUCKLING 302 00:15:57,840 --> 00:16:00,800 Hey! Round of applause for our little drone. 303 00:16:00,800 --> 00:16:02,480 APPLAUSE 304 00:16:13,560 --> 00:16:16,000 Oh, that was a lovely landing. 305 00:16:16,000 --> 00:16:18,760 Right, I want you to join me in welcoming to the stage 306 00:16:18,760 --> 00:16:21,000 Duncan and the Skyports drone. 307 00:16:21,000 --> 00:16:23,000 CHEERING AND APPLAUSE 308 00:16:29,120 --> 00:16:32,080 This is quite some drone, Duncan. This is quite a big one. 309 00:16:32,080 --> 00:16:35,120 This is one of our delivery drones, so we can do medical samples 310 00:16:35,120 --> 00:16:37,240 or e-commerce deliveries with this one. 311 00:16:37,240 --> 00:16:40,360 So what kind of things is this used for? We do blood samples. 312 00:16:40,360 --> 00:16:43,200 We can do them between hospitals and medical facilities. 313 00:16:43,200 --> 00:16:46,120 We can do packages. We were flying them in Finland recently. 314 00:16:46,120 --> 00:16:49,040 Basically, anything you can fit in that box, up to about 5kg, 315 00:16:49,040 --> 00:16:51,880 we can fly it. So how does this thing avoid crashing? 316 00:16:51,880 --> 00:16:55,520 It's got a number of systems on it. It's got 4G, like a mobile phone. 317 00:16:55,520 --> 00:16:59,040 It's got a Wi-Fi network of its own. 318 00:16:59,040 --> 00:17:01,120 And it's also got, if all else fails, 319 00:17:01,120 --> 00:17:03,080 a satellite communications network. 320 00:17:03,080 --> 00:17:05,760 What's this thing over here? That's the fail-safe. 321 00:17:05,760 --> 00:17:08,840 So, if everything goes wrong, that's a parachute and that would deploy. 322 00:17:08,840 --> 00:17:11,160 So, for example, if one of the rotors stops 323 00:17:11,160 --> 00:17:13,800 or one of the motors doesn't work, a big alarm goes off, 324 00:17:13,800 --> 00:17:16,640 that will deploy, and it will come down to Earth very safely. 325 00:17:16,640 --> 00:17:19,680 So it needs all of those different systems running in parallel? 326 00:17:19,680 --> 00:17:22,880 It needs them running in parallel. Hopefully you only ever use one. 327 00:17:22,880 --> 00:17:26,200 The rest we call redundancy. It's there in case something goes wrong. 328 00:17:26,200 --> 00:17:28,400 What's the future of drones like this, then? 329 00:17:28,400 --> 00:17:31,560 This is becoming more and more prevalent. We're flying in Africa. 330 00:17:31,560 --> 00:17:34,120 We're doing snake bite anti-venom. 331 00:17:34,120 --> 00:17:37,640 So very urgent stuff, often very bad road networks. 332 00:17:37,640 --> 00:17:40,360 We're flying in the west coast of Scotland, 333 00:17:40,360 --> 00:17:42,360 doing some medical samples again. 334 00:17:42,360 --> 00:17:44,320 Ultimately, it will come into cities, 335 00:17:44,320 --> 00:17:46,320 much more complex environments, 336 00:17:46,320 --> 00:17:48,720 you've got lots more people, lots more buildings, 337 00:17:48,720 --> 00:17:50,920 lots more things to keep out of the way of. 338 00:17:50,920 --> 00:17:53,120 But the technology is good enough now 339 00:17:53,120 --> 00:17:55,640 that you can fly in pretty much any environment. 340 00:17:55,640 --> 00:17:59,560 Talking of people, could I ever get a personal passenger drone? You can. 341 00:17:59,560 --> 00:18:01,360 In fact, you can already. 342 00:18:01,360 --> 00:18:04,400 So this is your company. Personal passenger drones, are they? 343 00:18:04,400 --> 00:18:07,360 Yes, this is a company called Volocopter, and these are live. 344 00:18:07,360 --> 00:18:09,480 They're going through certification now. 345 00:18:09,480 --> 00:18:12,720 Within two years, everybody will be able to get in one and fly around. 346 00:18:12,720 --> 00:18:15,000 Within two years? Within two years. Goodness me. 347 00:18:15,000 --> 00:18:17,640 Wil our skies be full of them in the future, do you think? 348 00:18:17,640 --> 00:18:20,040 The air space is vast. Cities are now very dense. 349 00:18:20,040 --> 00:18:22,480 It's hard to put more infrastructure into cities. 350 00:18:22,480 --> 00:18:24,920 These can fly in our underutilised air. Amazing. 351 00:18:24,920 --> 00:18:26,920 Duncan, a view of the future there, I think. 352 00:18:26,920 --> 00:18:29,080 A big round of applause, if you can. 353 00:18:29,080 --> 00:18:30,880 APPLAUSE 354 00:18:33,840 --> 00:18:36,840 Duncan was talking a lot there about having backups on backups 355 00:18:36,840 --> 00:18:40,400 on backups, just to make sure that if there's ever a problem, 356 00:18:40,400 --> 00:18:43,840 you know that the drone won't crash, and that is something, actually, 357 00:18:43,840 --> 00:18:46,440 that Matt Parker has been thinking about, too. 358 00:18:46,440 --> 00:18:51,120 Yes, and I've brought a comedy oversized slice of cheese. 359 00:18:51,120 --> 00:18:53,560 A slice of cheese? A slice of cheese. All right. 360 00:18:53,560 --> 00:18:56,840 Because when a lot of people are thinking about things like drones 361 00:18:56,840 --> 00:19:00,000 and trying to avoid disasters, they find it's useful to think of it 362 00:19:00,000 --> 00:19:01,800 in terms of cheese. OK. 363 00:19:01,800 --> 00:19:03,640 Things can go wrong with drones. 364 00:19:03,640 --> 00:19:05,840 You can have... One of the motors might break, 365 00:19:05,840 --> 00:19:07,880 the battery might run out of charge. 366 00:19:07,880 --> 00:19:11,000 When that happens, you don't want it to crash and cause a disaster. 367 00:19:11,000 --> 00:19:14,360 So, you imagine these mistakes, these errors, coming at your system 368 00:19:14,360 --> 00:19:16,000 and you put in barriers... 369 00:19:16,000 --> 00:19:19,720 so, like a slice of cheese, to stop them from making it...bear with me, 370 00:19:19,720 --> 00:19:22,080 making it through and becoming a disaster. 371 00:19:22,080 --> 00:19:25,400 So this could be, for example, like, the GPS system. OK. 372 00:19:25,400 --> 00:19:27,880 It's tracking where it is, anything that goes wrong, 373 00:19:27,880 --> 00:19:30,720 it shouldn't be a disaster. What about the holes, though? 374 00:19:30,720 --> 00:19:33,360 Well spotted. So, GPS, as you know, is not perfect. 375 00:19:33,360 --> 00:19:35,800 You could be on the sidewalk, according to the GPS, 376 00:19:35,800 --> 00:19:38,840 and so it might be inaccurate, it might give you the wrong data. 377 00:19:38,840 --> 00:19:41,600 No one layer to try and stop disasters will be perfect. 378 00:19:41,600 --> 00:19:44,560 So this is trying to block disasters from happening, 379 00:19:44,560 --> 00:19:47,360 mostly it works, but just occasionally it's going to fail. 380 00:19:47,360 --> 00:19:50,440 Occasionally a mistake will slip through and GPS won't be enough. 381 00:19:50,440 --> 00:19:52,800 But over here, we've got more cheese. 382 00:19:52,800 --> 00:19:56,920 If you'd like to take a seat here, under the cheese. It's fine. 383 00:19:56,920 --> 00:19:59,240 Are you sure? You trust the maths. Here we go. 384 00:19:59,240 --> 00:20:02,080 I'm not sure I trust you though, Matt. Wise Very wise. 385 00:20:02,080 --> 00:20:06,400 So, this is another slice of cheese and this one is the parachute. 386 00:20:06,400 --> 00:20:09,320 So, if the big drone fails and the GPS is wrong, 387 00:20:09,320 --> 00:20:11,360 the parachute will deploy. 388 00:20:11,360 --> 00:20:14,400 With two layers together... This has got holes in it as well. 389 00:20:14,400 --> 00:20:17,360 That's true, but what we hope is, the holes in this layer 390 00:20:17,360 --> 00:20:20,040 don't line up with the holes in the other layers. 391 00:20:20,040 --> 00:20:22,040 Actually, we can get another layer. 392 00:20:22,040 --> 00:20:24,880 This one's called rules, which doesn't sound very exciting, 393 00:20:24,880 --> 00:20:26,600 but we all need rules. 394 00:20:26,600 --> 00:20:28,640 So, when we brought the drone in here, 395 00:20:28,640 --> 00:20:31,080 we weren't allowed to fly it above a crowd. 396 00:20:31,080 --> 00:20:33,400 That's one of the rules for using a drone. 397 00:20:33,400 --> 00:20:36,080 So even if an accident happens.. Yeah. 398 00:20:36,080 --> 00:20:38,720 ..even if the parachute fails and the GPS fails, 399 00:20:38,720 --> 00:20:40,680 at least people won't get hurt. 400 00:20:40,680 --> 00:20:43,640 In theory, if you're following the rules, it'll be fine, 401 00:20:43,640 --> 00:20:46,400 although, of course, sometimes people break the rules 402 00:20:46,400 --> 00:20:48,880 and you hope the other layers will help you out. 403 00:20:48,880 --> 00:20:52,040 And so...I'm going to grab a camera so I can show you a point of view. 404 00:20:52,040 --> 00:20:54,000 Thank you. Can I just... 405 00:20:54,000 --> 00:20:57,040 You know what, Matt, I mean, it's not that I don't trust you... 406 00:20:58,400 --> 00:21:02,040 No, it's literally that you don't trust me. It's literally that. OK. 407 00:21:02,040 --> 00:21:05,200 So if you have a look from Hannah's point of view underneath, 408 00:21:05,200 --> 00:21:08,640 you can see up through some holes, but then, almost straight away, 409 00:21:08,640 --> 00:21:11,080 it's blocked by a different slice of cheese. 410 00:21:11,080 --> 00:21:13,880 And the point of view at the top, here you go. 411 00:21:13,880 --> 00:21:17,080 Again, there are some holes that go partway down, but then they stop. 412 00:21:17,080 --> 00:21:20,800 What we're going to do now is rain some errors down on you. 413 00:21:20,800 --> 00:21:22,920 We've got a whole bucket of errors 414 00:21:22,920 --> 00:21:25,880 and, in theory, if we drip them down, 415 00:21:25,880 --> 00:21:29,360 while they will make it through some of the slices of cheese 416 00:21:29,360 --> 00:21:32,600 they will be stopped by... More errors, more errors. 417 00:21:32,600 --> 00:21:35,640 And so, some of them are being stopped by the first layer, 418 00:21:35,640 --> 00:21:37,920 some of them are getting through the first layer 419 00:21:37,920 --> 00:21:40,800 but they're stopped by the next layer, and so, in theory... 420 00:21:40,800 --> 00:21:42,760 OK, I get it, I get it, I get it. 421 00:21:42,760 --> 00:21:45,400 Yeah, overall... Overall, it's fine, right? 422 00:21:45,400 --> 00:21:48,400 All of these different layers are blocking it from happening. 423 00:21:48,400 --> 00:21:50,800 And so even though any one individual layer, 424 00:21:50,800 --> 00:21:53,160 you're like, oh, look at all those holes in it, 425 00:21:53,160 --> 00:21:56,040 if you sit down and you look through the whole lot at once, 426 00:21:56,040 --> 00:21:58,000 you're like, oh, that's amazing. 427 00:21:58,000 --> 00:22:01,320 From here, I can't see any hole which goes the entire way through. 428 00:22:01,320 --> 00:22:03,120 That's good, but then, Matt, 429 00:22:03,120 --> 00:22:05,720 what happens if the holes do go the entire way through? 430 00:22:05,720 --> 00:22:07,800 That's a good point, because every... 431 00:22:07,800 --> 00:22:09,400 LAUGHTER 432 00:22:09,400 --> 00:22:11,200 You make a valid point. 433 00:22:11,200 --> 00:22:13,960 I think we need more errors. Albeit in an interesting way. 434 00:22:13,960 --> 00:22:16,240 More errors. More errors. 435 00:22:16,240 --> 00:22:18,280 CHEERING 436 00:22:28,640 --> 00:22:31,760 I'm trying to make a serious point here. 437 00:22:31,760 --> 00:22:34,840 Occasionally, your cheese holes will line up 438 00:22:34,840 --> 00:22:38,040 and a few mistakes, normally, will make it through 439 00:22:38,040 --> 00:22:39,600 and become a disaster. 440 00:22:39,600 --> 00:22:42,600 And there's nothing you can do about that? Nothing you can do. 441 00:22:42,600 --> 00:22:45,120 Disasters, Matt, are inevitable. Erm, yep. 442 00:22:45,120 --> 00:22:47,480 I'm coming to terms with that as we speak. 443 00:22:47,480 --> 00:22:50,320 Matt Parker. Round of applause. Thank you very much. 444 00:22:50,320 --> 00:22:52,840 CHEERING AND APPLAUSE 445 00:22:57,320 --> 00:23:00,600 So this is the really big downside of uncertainty. 446 00:23:00,600 --> 00:23:03,480 You have to accept that perfection is impossible. 447 00:23:03,480 --> 00:23:06,080 Mistakes are effectively inevitable. 448 00:23:06,080 --> 00:23:09,080 I think that that raises a very big and important question. 449 00:23:09,080 --> 00:23:11,720 If we know for sure that our algorithms 450 00:23:11,720 --> 00:23:13,880 are never going to be perfect, 451 00:23:13,880 --> 00:23:17,320 do we want to put them in charge of making decisions, 452 00:23:17,320 --> 00:23:20,600 especially in situations where people's lives are at stake, 453 00:23:20,600 --> 00:23:22,560 like in the courtroom? 454 00:23:22,560 --> 00:23:27,080 Someone who's thought about this a lot is Professor Katie Atkinson. 455 00:23:27,080 --> 00:23:29,320 APPLAUSE 456 00:23:32,600 --> 00:23:36,120 Katie, the courtroom, it doesn't feel like a natural place 457 00:23:36,120 --> 00:23:39,600 where you would find mathematics. Well, perhaps not, but, actually, 458 00:23:39,600 --> 00:23:43,400 AI and law researchers are working on building models 459 00:23:43,400 --> 00:23:46,640 of legal reasoning using mathematical models 460 00:23:46,640 --> 00:23:49,120 that are then turned into software programmes 461 00:23:49,120 --> 00:23:51,920 that can help judges and lawyers. Why do they need them? 462 00:23:51,920 --> 00:23:54,400 Why can't the judges just do it all themselves? 463 00:23:54,400 --> 00:23:57,440 Well, the point of using these mathematical models 464 00:23:57,440 --> 00:24:00,360 is that we can get consistent, efficient decisions, 465 00:24:00,360 --> 00:24:03,600 and we know that any kind of unconscious biases are stripped out. 466 00:24:03,600 --> 00:24:06,640 So, judges make mistakes, they have unconscious biases, 467 00:24:06,640 --> 00:24:08,960 and the idea is that using algorithms 468 00:24:08,960 --> 00:24:11,720 can help to minimise that? That's absolutely right. 469 00:24:11,720 --> 00:24:14,680 We're hoping these can help. I understand that you've brought 470 00:24:14,680 --> 00:24:17,760 a little friend along to help us understand what's going on here. 471 00:24:17,760 --> 00:24:20,800 Yes, this is Pepper the robot. Pepper the robot. Hello, Pepper. 472 00:24:20,800 --> 00:24:24,840 All rise for the court that will decide the case of Popov v Hayashi. 473 00:24:24,840 --> 00:24:28,400 So this links us to a legal case that was in the United States 474 00:24:28,400 --> 00:24:32,680 and involved a baseball and people catching and dropping a baseball. 475 00:24:32,680 --> 00:24:34,960 So someone hit a baseball into the crowd 476 00:24:34,960 --> 00:24:37,360 and then two people fought over the baseball, 477 00:24:37,360 --> 00:24:40,080 and this baseball was very valuable, wasn't it? Indeed. 478 00:24:40,080 --> 00:24:42,600 It was worth $450,000 in the end. Oh, crikey! 479 00:24:42,600 --> 00:24:45,720 I can understand why two people were particularly upset. Yeah. 480 00:24:45,720 --> 00:24:49,000 So what we have to do is, first of all, work out what the facts 481 00:24:49,000 --> 00:24:52,400 of the case are and then we have to work out what the arguments are 482 00:24:52,400 --> 00:24:55,080 for the two different sides within the legal case. 483 00:24:55,080 --> 00:24:57,960 And that's what Pepper's going to help us with? That's right. 484 00:24:57,960 --> 00:25:02,400 So the facts of the case are that Mr Popov stopped the forward motion 485 00:25:02,400 --> 00:25:04,880 of the ball once it was hit into the crowd. 486 00:25:04,880 --> 00:25:08,360 He tried to get it under control but he was thrown to the ground 487 00:25:08,360 --> 00:25:11,360 by this mob who were also trying to secure the baseball. 488 00:25:11,360 --> 00:25:14,080 It sounds a bit unfair, really, if you're, you know... 489 00:25:14,080 --> 00:25:17,280 You've caught it and then... It's not your fault. That's right. 490 00:25:17,280 --> 00:25:20,280 That's why he felt that he should have been given the opportunity 491 00:25:20,280 --> 00:25:22,160 to complete the catch. 492 00:25:22,160 --> 00:25:25,600 And then Mr Hayashi found the loose ball on the floor 493 00:25:25,600 --> 00:25:29,040 and he picked it up and claimed it as his. OK. 494 00:25:29,040 --> 00:25:32,080 Well, I mean, he was the one who ended up with it at the end. 495 00:25:32,080 --> 00:25:35,160 That kind of seems like he's got a fair claim, too. That's right. 496 00:25:35,160 --> 00:25:37,480 And, in particular, he wasn't part of this mob 497 00:25:37,480 --> 00:25:40,160 that threw Mr Popov to the ground, so he did no wrong either, 498 00:25:40,160 --> 00:25:41,840 so he shouldn't be punished. 499 00:25:41,840 --> 00:25:44,680 And that's really the facts and the arguments of the case. 500 00:25:44,680 --> 00:25:47,520 So you can take all of those facts and arguments over the case, 501 00:25:47,520 --> 00:25:49,760 put them into equations, compare them to cases 502 00:25:49,760 --> 00:25:51,560 that were like this in the past, 503 00:25:51,560 --> 00:25:53,440 and then, ultimately, 504 00:25:53,440 --> 00:25:56,880 can the algorithm give us a verdict? Yes, that's right. 505 00:25:56,880 --> 00:25:59,880 OK, Pepper, what's the verdict in this case, then? 506 00:25:59,880 --> 00:26:04,400 The decision of this court is that the ball should be sold at auction 507 00:26:04,400 --> 00:26:09,400 and the proceeds split evenly between Mr Popov and Mr Hayashi. 508 00:26:09,400 --> 00:26:12,680 OK, that's very clear. We're having lots of fun with a humanoid robot, 509 00:26:12,680 --> 00:26:15,600 but it's not the intention to actually have humanoid robots 510 00:26:15,600 --> 00:26:18,080 in courtrooms, is it? That's absolutely right. 511 00:26:18,080 --> 00:26:20,680 We're aiming at writing these mathematical models 512 00:26:20,680 --> 00:26:24,040 that we're turning into AI tools that will be on the computer 513 00:26:24,040 --> 00:26:25,840 and helping judges and lawyers 514 00:26:25,840 --> 00:26:28,360 who are set there using these for decision support. 515 00:26:28,360 --> 00:26:31,600 It's all the stuff inside Pepper, not Pepper herself. That's right. 516 00:26:31,600 --> 00:26:34,400 I can't imagine us seeing Pepper in a court any time soon. 517 00:26:34,400 --> 00:26:37,280 If she's sending you off to jail, going like that... No. 518 00:26:37,280 --> 00:26:39,720 This example sounds like a very positive thing. 519 00:26:39,720 --> 00:26:42,360 You could get through, I imagine, a big backlog of cases 520 00:26:42,360 --> 00:26:44,400 with something like this on your side. 521 00:26:44,400 --> 00:26:47,520 But algorithms in the courtroom, they're not...they haven't been 522 00:26:47,520 --> 00:26:50,120 universally positive, have they? Yeah, that's right. 523 00:26:50,120 --> 00:26:51,760 There is a big issue of trust 524 00:26:51,760 --> 00:26:54,560 as well as whether the actual technology works in itself. 525 00:26:54,560 --> 00:26:57,760 Because if you're putting all of history into a series of equations, 526 00:26:57,760 --> 00:27:00,640 I mean, history wasn't exactly fair, was it? That's right. 527 00:27:00,640 --> 00:27:03,640 And we want to make these decisions as fair as we possibly can 528 00:27:03,640 --> 00:27:06,400 and get AI technologies to help us do this. I quite agree. 529 00:27:06,400 --> 00:27:09,160 Katie, thank you very much for joining us. You're welcome. 530 00:27:09,160 --> 00:27:10,800 APPLAUSE 531 00:27:15,600 --> 00:27:18,600 I think Katie made a really important point in all that, 532 00:27:18,600 --> 00:27:22,080 because if all you're doing is just chucking in everything that happened 533 00:27:22,080 --> 00:27:25,080 in the past to your equations, then you're going to perpetuate 534 00:27:25,080 --> 00:27:27,880 all of society's biggest imbalances going forward. 535 00:27:27,880 --> 00:27:32,400 And a machine is only ever as good as the data that it's trained on. 536 00:27:32,400 --> 00:27:34,760 Let me show you what I'm talking about here, 537 00:27:34,760 --> 00:27:38,280 because I'd like you to welcome back to the stage Matt Parker 538 00:27:38,280 --> 00:27:40,640 with a very special shoe-detecting machine. 539 00:27:40,640 --> 00:27:43,000 APPLAUSE 540 00:27:45,480 --> 00:27:49,840 Is that a new shirt, Matt? I bring a range of shirts. 541 00:27:49,840 --> 00:27:54,280 So, this is my shoe-detecting device. OK. You can have a go. 542 00:27:54,280 --> 00:27:57,600 I call it Shoe Do You Think You Are? OK. Very nice. 543 00:27:57,600 --> 00:28:01,880 If you point it at something, it can tell you if it's a shoe or not. 544 00:28:01,880 --> 00:28:03,600 All right. Let me give it a go. Try your face. 545 00:28:03,600 --> 00:28:04,880 OK, that is not a shoe. 546 00:28:04,880 --> 00:28:06,720 Not a shoe. Correct. Thank you very much. 547 00:28:06,720 --> 00:28:08,280 Let's try aiming at your shoes. Yep. 548 00:28:08,280 --> 00:28:10,160 Oh, hang on. It says they're not shoes. 549 00:28:10,160 --> 00:28:12,520 The device is fine. We just haven't trained it yet. 550 00:28:12,520 --> 00:28:14,360 We have to put in some training data. 551 00:28:14,360 --> 00:28:16,520 You were here for lecture two. I was. 552 00:28:16,520 --> 00:28:18,560 So we've got to teach it what a shoe looks like 553 00:28:18,560 --> 00:28:20,920 and then it'll be amazing. OK. All right, then. 554 00:28:20,920 --> 00:28:24,800 So let's get a group of people to help me train this shoe. 555 00:28:24,800 --> 00:28:26,760 So let's get some people from over here. 556 00:28:26,760 --> 00:28:28,800 Let's get that little column up there. And up here. 557 00:28:28,800 --> 00:28:31,640 You guys all want to come down here and help me train this machine. 558 00:28:31,640 --> 00:28:33,760 Round of applause as they come to the stage. 559 00:28:33,760 --> 00:28:35,480 APPLAUSE 560 00:28:40,280 --> 00:28:41,880 OK, so here's what we're going to do. 561 00:28:41,880 --> 00:28:43,280 We'll give you 20 seconds, 562 00:28:43,280 --> 00:28:46,200 and I would just like you to draw a picture of a shoe 563 00:28:46,200 --> 00:28:47,640 that I can scan into the machine. 564 00:28:47,640 --> 00:28:49,080 Here we go. 565 00:28:57,200 --> 00:28:59,280 Two, one, stop. 566 00:28:59,280 --> 00:29:02,520 OK. Let's turn those round. Hold them up to the camera. 567 00:29:02,520 --> 00:29:04,200 We'll have a look. Let's scan these. 568 00:29:04,200 --> 00:29:05,800 We've got some lace-y numbers. 569 00:29:05,800 --> 00:29:08,160 Some nice... A top shot of a shoe there. This is great. 570 00:29:08,160 --> 00:29:10,520 Lots of laces. You hare some trainers there. 571 00:29:10,520 --> 00:29:12,680 This is all perfect. OK, great. Perfect. 572 00:29:12,680 --> 00:29:14,680 It's all... I think it's all trained, Matt. 573 00:29:14,680 --> 00:29:16,440 So let's give it a go. 574 00:29:16,440 --> 00:29:18,280 Let's try your shoes there. 575 00:29:18,280 --> 00:29:19,600 OK. Perfect. 576 00:29:19,600 --> 00:29:22,080 Spotted your shoes. Amazing. It's great. It works. 577 00:29:22,080 --> 00:29:24,800 Is it detecting shoes? Yeah, look. Yeah, perfect. 578 00:29:24,800 --> 00:29:27,240 That's amazing. Can I just pop through and get a closer look? 579 00:29:27,240 --> 00:29:29,280 That's incredible. Oh. 580 00:29:29,280 --> 00:29:33,400 And it's not failed yet on any of these near identical shoes? 581 00:29:33,400 --> 00:29:36,040 No. I mean you did all draw very similar shoes. Give these a go. 582 00:29:36,040 --> 00:29:37,280 Let's try this one. 583 00:29:37,280 --> 00:29:40,120 Oh, no. It's saying no. It doesn't detect your shoes at all. 584 00:29:40,120 --> 00:29:42,720 You need a much more diverse range of training. 585 00:29:42,720 --> 00:29:44,960 I'm very disappointed in all of you. 586 00:29:44,960 --> 00:29:46,200 LAUGHTER 587 00:29:46,200 --> 00:29:48,560 I mean, guys, you did just basically draw your own shoes. 588 00:29:48,560 --> 00:29:49,880 So let's try again. 589 00:29:49,880 --> 00:29:51,880 Turn your paper round, try again, 590 00:29:51,880 --> 00:29:55,640 and this time try and think of as wide a range of shoes as possible. 591 00:29:55,640 --> 00:29:56,720 Off you go. 592 00:29:58,560 --> 00:30:01,600 Three, two, one, stop. 593 00:30:01,600 --> 00:30:04,040 OK. Here we go. Let's get this in scanning mode. 594 00:30:04,040 --> 00:30:05,280 We're ready to go. 595 00:30:05,280 --> 00:30:07,440 Turn it around. Hold it up to the camera. That's it. 596 00:30:07,440 --> 00:30:09,560 It looks pretty much like a shoe. This is pretty good. 597 00:30:09,560 --> 00:30:11,160 This is great. Got some high heels. 598 00:30:11,160 --> 00:30:13,000 Lots of high heels. Got flip-flops, got boots. 599 00:30:13,000 --> 00:30:14,680 We've got some wellies there. 600 00:30:14,680 --> 00:30:17,280 This is lovely. Ballerina shoe. That's great. Perfect. 601 00:30:17,280 --> 00:30:19,800 Matt, I think it's good now. I think it's got to be good now. 602 00:30:19,800 --> 00:30:22,320 Surely this works on all shoes now. 603 00:30:22,320 --> 00:30:24,480 Let's have a closer look at it. 604 00:30:24,480 --> 00:30:26,400 Excuse me. MURMURS IN AUDIENCE 605 00:30:28,080 --> 00:30:29,640 Shoe? Shoe. 606 00:30:29,640 --> 00:30:31,560 Not shoe. Not a shoe?! 607 00:30:31,560 --> 00:30:33,120 I mean... Shoe. 608 00:30:33,120 --> 00:30:37,360 ..that's debatable whether that's a shoe. It's definitely footwear. 609 00:30:37,360 --> 00:30:38,640 OK. Let's give it a go. 610 00:30:38,640 --> 00:30:41,040 With the right data it could have detected that as a shoe. 611 00:30:41,040 --> 00:30:42,880 This is true, with the right data. 612 00:30:42,880 --> 00:30:45,720 But I think the point here is that even when you know to draw 613 00:30:45,720 --> 00:30:47,560 as diverse a range of shoes as possible, 614 00:30:47,560 --> 00:30:50,160 it's actually just really hard to think of everything. 615 00:30:50,160 --> 00:30:51,800 Just unbelievable. 616 00:30:54,000 --> 00:30:55,520 Thank you very much, Matt. 617 00:30:55,520 --> 00:30:57,680 A round of applause for our volunteers. 618 00:30:57,680 --> 00:30:59,080 APPLAUSE 619 00:31:01,080 --> 00:31:03,400 Now, there is a really important point in all of this, 620 00:31:03,400 --> 00:31:07,320 because if you don't think through all of the possible situations 621 00:31:07,320 --> 00:31:09,680 that your machine needs to include, 622 00:31:09,680 --> 00:31:12,880 it can end up having very big, real life consequences. 623 00:31:12,880 --> 00:31:14,480 To tell us a little bit more, 624 00:31:14,480 --> 00:31:17,440 please join me in welcoming from the University of York 625 00:31:17,440 --> 00:31:19,640 an expert in image recognition for surveillance, 626 00:31:19,640 --> 00:31:21,360 Dr Kofi Appiah. 627 00:31:21,360 --> 00:31:23,600 APPLAUSE 628 00:31:25,720 --> 00:31:26,960 Hey, Kofi. 629 00:31:32,560 --> 00:31:36,520 Now, Kofi, you've done a lot of work in facial recognition before, right? 630 00:31:36,520 --> 00:31:39,080 That's right. Tell me, how does facial detection work? 631 00:31:39,080 --> 00:31:42,720 So for face detection, all that we need is to be able to pick out 632 00:31:42,720 --> 00:31:46,400 the elements of the face, which is the eye - the key features - 633 00:31:46,400 --> 00:31:48,080 the eye, the nose, the mouth, 634 00:31:48,080 --> 00:31:50,400 and to be able to pick out these features, 635 00:31:50,400 --> 00:31:53,520 there's a big contrast that we can find between the eye 636 00:31:53,520 --> 00:31:56,800 and the eyelashes, the eyelashes and the skin itself. 637 00:31:56,800 --> 00:31:59,400 When it comes to the mouth, the lips, you have an edge there. 638 00:31:59,400 --> 00:32:01,760 So these are the big contrasts that we are able to pick 639 00:32:01,760 --> 00:32:05,160 and as far as we're able to find eyes, nose, mouth, we've got a face. 640 00:32:05,160 --> 00:32:07,800 So is it looking... If I come over here, then, to this one here. 641 00:32:07,800 --> 00:32:09,560 Is it looking for areas of light and dark? 642 00:32:09,560 --> 00:32:11,160 Say, on my nose, for instance? Yes. 643 00:32:11,160 --> 00:32:13,920 It's kind of darker on either side and then lighter down the middle. 644 00:32:13,920 --> 00:32:16,800 That's correct. And if it gets a strip of darker pixels and lighter 645 00:32:16,800 --> 00:32:18,800 pixels, it knows it's found a nose? That's right. 646 00:32:18,800 --> 00:32:20,400 It's looking for these key features. 647 00:32:20,400 --> 00:32:22,880 This is what a face normally will have - the key features. 648 00:32:22,880 --> 00:32:25,440 And that's what we train our systems to use and recognise. 649 00:32:25,440 --> 00:32:28,040 Perfect. Just one thing, though, Kofi, that I'm noticing here. 650 00:32:28,040 --> 00:32:29,960 It's getting my face OK, but... 651 00:32:29,960 --> 00:32:33,520 Right. So it's not able to pick my face, 652 00:32:33,520 --> 00:32:36,840 and it's relating to the data that you just mentioned. 653 00:32:36,840 --> 00:32:39,560 This system has not been trained with enough data. 654 00:32:39,560 --> 00:32:43,680 The contrasting features that you've got between your eyebrows 655 00:32:43,680 --> 00:32:46,760 and the skin texture is different from mine. 656 00:32:46,760 --> 00:32:48,080 And it's not picking it up. 657 00:32:48,080 --> 00:32:50,920 So in this case, I'm going to have this... 658 00:32:52,880 --> 00:32:55,160 It's picking it up as a face because it can pick some 659 00:32:55,160 --> 00:32:57,680 of the salient features that I was talking about. 660 00:32:57,680 --> 00:32:59,760 It's going to put a bounding box around it. 661 00:33:01,280 --> 00:33:03,240 Whereas in my case, no. 662 00:33:03,240 --> 00:33:06,000 But this stuff is actually, you know, it's a really big deal. 663 00:33:06,000 --> 00:33:07,520 It's not just in cameras. 664 00:33:07,520 --> 00:33:10,040 We're now seeing facial detection in all kinds of things. 665 00:33:10,040 --> 00:33:12,440 Passport queues, for instance. That's right, yes. 666 00:33:12,440 --> 00:33:15,120 Obviously, in this case, if you've got a face like me, 667 00:33:15,120 --> 00:33:17,800 unfortunately, it's going to be a long delay for you. 668 00:33:17,800 --> 00:33:20,480 We're using it in law enforcement as well. 669 00:33:20,480 --> 00:33:23,600 An example, if it's not able to pick the right face, 670 00:33:23,600 --> 00:33:26,800 you're going to be prosecuted for something that you've not done. 671 00:33:26,800 --> 00:33:30,240 We're using this facial recognition to be able to unlock phones. 672 00:33:30,240 --> 00:33:32,080 It's like a password now. 673 00:33:32,080 --> 00:33:36,120 So if it's not working right look at the harm that it can do. 674 00:33:36,120 --> 00:33:37,800 Is it improving? 675 00:33:37,800 --> 00:33:39,440 Are the algorithms getting better? 676 00:33:39,440 --> 00:33:41,480 Yes, it's getting better and better. 677 00:33:41,480 --> 00:33:44,360 And what they're trying to do is to make it non-biased 678 00:33:44,360 --> 00:33:47,240 by training with diverse, non-biased data sets. 679 00:33:47,240 --> 00:33:49,560 As you can see, you can pick some of the features, 680 00:33:49,560 --> 00:33:52,960 it can pick mine. And works with a whole different range of faces. 681 00:33:52,960 --> 00:33:54,400 It's trying to fix that. 682 00:33:54,400 --> 00:33:57,720 So it's improving over time and we're getting there. 683 00:33:57,720 --> 00:34:01,360 Obviously, the issues of bias and fairness are incredibly important 684 00:34:01,360 --> 00:34:02,960 when it comes to facial recognition. 685 00:34:02,960 --> 00:34:05,480 But there is another concern that people have when it comes 686 00:34:05,480 --> 00:34:08,680 to this technology, which is that some people don't like the idea 687 00:34:08,680 --> 00:34:11,040 that they can be identified in a crowd 688 00:34:11,040 --> 00:34:12,720 based on their facial features. 689 00:34:12,720 --> 00:34:15,440 Some people don't like the idea of losing their anonymity. 690 00:34:15,440 --> 00:34:18,520 So some people have been experimenting with different ways 691 00:34:18,520 --> 00:34:21,520 to try and trick facial recognition cameras. 692 00:34:21,520 --> 00:34:25,640 And we are lucky to be joined from Glow Up 693 00:34:25,640 --> 00:34:28,720 by make-up artist Tiffany Hunt and Eva. 694 00:34:28,720 --> 00:34:30,480 APPLAUSE 695 00:34:34,440 --> 00:34:36,240 If you come and stand round here. 696 00:34:37,320 --> 00:34:41,000 Now, Eva, you have had, as we can see, 697 00:34:41,000 --> 00:34:44,360 Tiffany has been doing your make-up for a little while backstage. 698 00:34:44,360 --> 00:34:46,240 Looks very remarkable. 699 00:34:46,240 --> 00:34:50,080 Is this the kind of thing that would work to fool facial detection? 700 00:34:50,080 --> 00:34:53,240 Oh, yes. So this kind of system, because it's looking 701 00:34:53,240 --> 00:34:58,200 for that contrasting bit between the eye, the eyelashes, the nose, 702 00:34:58,200 --> 00:35:00,560 but the way the face has being painted now, 703 00:35:00,560 --> 00:35:02,200 it's breaking all the symmetry. 704 00:35:02,200 --> 00:35:04,080 It can't find the nose. 705 00:35:04,080 --> 00:35:06,280 Tiffany, talk us through what you were going for here. 706 00:35:06,280 --> 00:35:08,160 It's quite dramatic! 707 00:35:08,160 --> 00:35:10,160 Just a bit. Just something, like, natural! 708 00:35:10,160 --> 00:35:11,520 Basically, what I was thinking, 709 00:35:11,520 --> 00:35:13,640 was something that was a monochromatic illusion. 710 00:35:13,640 --> 00:35:15,840 I really wanted to break up specific facial features 711 00:35:15,840 --> 00:35:17,920 such as this area in the central area of the face, 712 00:35:17,920 --> 00:35:19,920 which is the area that gets picked up the most. 713 00:35:19,920 --> 00:35:22,720 Also changing the eye shape, the lashes and just really going 714 00:35:22,720 --> 00:35:25,160 for something with colours that are just so different 715 00:35:25,160 --> 00:35:27,840 to your usual skin tone and just changing her face totally, really. 716 00:35:27,840 --> 00:35:29,320 Do you like it? 717 00:35:29,320 --> 00:35:32,360 Yeah. Friday night out. 718 00:35:32,360 --> 00:35:34,800 But the proof is in the pudding. Let's have a look. 719 00:35:34,800 --> 00:35:37,840 So, Tiffany, we're picking you up. If you step to one side. 720 00:35:37,840 --> 00:35:40,360 It's not getting you at all! 721 00:35:40,360 --> 00:35:43,480 I think a big round of applause there for the dazzling make-up. 722 00:35:43,480 --> 00:35:45,160 Very impressive. 723 00:35:45,160 --> 00:35:47,000 To Eva, Tiffany and Kofi. Thank you. 724 00:35:47,000 --> 00:35:48,760 APPLAUSE 725 00:35:55,480 --> 00:35:58,600 We've got some ways now that we know we can avoid being detected 726 00:35:58,600 --> 00:36:01,400 by facial recognition cameras. Might look a little bit crazy 727 00:36:01,400 --> 00:36:03,520 if you're wandering around with that all the time. 728 00:36:03,520 --> 00:36:06,240 But when it comes to protecting our privacy, there are some people 729 00:36:06,240 --> 00:36:08,760 who are worried that even this won't be enough. 730 00:36:08,760 --> 00:36:11,240 There are some people who are worried that with the help 731 00:36:11,240 --> 00:36:14,800 of mathematical algorithms, we are building and processing vast 732 00:36:14,800 --> 00:36:17,400 profiles on each and every one of us, 733 00:36:17,400 --> 00:36:19,040 often without our knowledge. 734 00:36:19,040 --> 00:36:21,960 Now, to explain this, please welcome to the stage 735 00:36:21,960 --> 00:36:25,280 computer scientist extraordinaire, Dr Anne-Marie Imafidon. 736 00:36:25,280 --> 00:36:27,200 APPLAUSE 737 00:36:35,920 --> 00:36:39,680 Anne-Marie, do you think that we're seeing the death of privacy? 738 00:36:39,680 --> 00:36:42,320 In some ways we are, in other ways we aren't. 739 00:36:42,320 --> 00:36:46,080 I know if you go back far enough, we used to communicate with fire 740 00:36:46,080 --> 00:36:48,920 and smoke signals or by yelling things long distance. 741 00:36:48,920 --> 00:36:51,240 Not that long ago, really! Yeah, basically. 742 00:36:51,240 --> 00:36:54,520 But now, with the algorithms that we have and the time 743 00:36:54,520 --> 00:36:58,280 that we spend online, really in-depth profiles are being built up 744 00:36:58,280 --> 00:37:00,120 on all of us continuously. 745 00:37:00,120 --> 00:37:03,520 And it's those profiles that are connecting pieces of information together, right? 746 00:37:03,520 --> 00:37:05,240 That's kind of the big thing. Exactly. 747 00:37:05,240 --> 00:37:07,920 So where those things might not have been private before, 748 00:37:07,920 --> 00:37:10,960 being able to connect them together to build up that picture of someone, 749 00:37:10,960 --> 00:37:13,640 that was harder. Whereas now it's very easy. 750 00:37:13,640 --> 00:37:15,320 And we use things like social media 751 00:37:15,320 --> 00:37:18,240 where we're putting that information out. Voluntarily. Exactly. 752 00:37:18,240 --> 00:37:20,120 So the profiles are pretty complex. 753 00:37:20,120 --> 00:37:23,960 OK. So to give us a sense of these kind of profiles, 754 00:37:23,960 --> 00:37:26,720 I wanted to get, let's say, three volunteers here, 755 00:37:26,720 --> 00:37:29,840 because Anne-Marie's got a little demonstration for us. 756 00:37:29,840 --> 00:37:32,240 OK, perfect. All right. 757 00:37:32,240 --> 00:37:35,120 OK. We'll go for you. You get to come down. 758 00:37:35,120 --> 00:37:37,640 Let's go for... Let's go for you there, in the jumper. 759 00:37:37,640 --> 00:37:39,880 And you there, in the stripy jumper. 760 00:37:39,880 --> 00:37:41,680 APPLAUSE 761 00:37:41,680 --> 00:37:43,720 Let's bring you down to the stage, come on. 762 00:37:46,160 --> 00:37:47,840 What's your name? Natasha. 763 00:37:47,840 --> 00:37:50,440 Natasha, OK, perfect. Natasha, if you want to go there. 764 00:37:50,440 --> 00:37:52,280 What was your name? Kieran. OK, perfect. 765 00:37:52,280 --> 00:37:54,720 Do you want to go there, Kieran? And what was your name? 766 00:37:54,720 --> 00:37:57,320 Caitlin. Perfect. You jump round there. 767 00:37:57,320 --> 00:38:00,440 Anne-Marie is going to talk us through a little demonstration. 768 00:38:00,440 --> 00:38:03,960 Yes. So, each of you have got a website that we've loaded up 769 00:38:03,960 --> 00:38:06,080 for you on your laptop there. 770 00:38:06,080 --> 00:38:08,240 And I want to make sure you've accepted cookies. 771 00:38:08,240 --> 00:38:11,520 And I'd like you to get clicking and browsing on these sites 772 00:38:11,520 --> 00:38:13,400 and doing some cool things. 773 00:38:13,400 --> 00:38:16,800 And as you've accepted cookies, I'm going to give each of you a cookie. 774 00:38:20,680 --> 00:38:23,000 They are rather large cookies. 775 00:38:23,000 --> 00:38:24,760 Thank you very much. 776 00:38:26,480 --> 00:38:29,480 So as your browsing and your clicking around, 777 00:38:29,480 --> 00:38:33,320 you have this cookie in your site, in your laptop, 778 00:38:33,320 --> 00:38:36,560 and there's different data and information that you're putting in. 779 00:38:36,560 --> 00:38:38,200 So I can see you here. 780 00:38:38,200 --> 00:38:40,680 You're about to watch some YouTube videos. 781 00:38:40,680 --> 00:38:42,560 I can see which channel that you're on. 782 00:38:42,560 --> 00:38:45,040 So I'm going to pop a little bit of a chocolate chip there. 783 00:38:45,040 --> 00:38:46,360 That's some data. 784 00:38:46,360 --> 00:38:49,480 I can see the artist that you've got as well. 785 00:38:49,480 --> 00:38:52,040 So I'm going to pop a little bit more data 786 00:38:52,040 --> 00:38:54,200 on your chocolate chip cookie. 787 00:38:54,200 --> 00:38:56,360 I can see you're shopping here. 788 00:38:56,360 --> 00:38:58,080 Brilliant, curtains. 789 00:38:58,080 --> 00:38:59,720 That's a good bit of data. 790 00:38:59,720 --> 00:39:03,120 Just got a new house and she likes blue and white curtains. 791 00:39:03,120 --> 00:39:05,840 So two more bits of data that we've got there. 792 00:39:05,840 --> 00:39:07,280 Fantastic. 793 00:39:07,280 --> 00:39:09,520 I think you've added them to your basket as well, 794 00:39:09,520 --> 00:39:11,320 so we're going to add that into your cookie, 795 00:39:11,320 --> 00:39:13,680 just so when you come back, you don't need to browse blue 796 00:39:13,680 --> 00:39:15,040 and white curtains again. 797 00:39:15,040 --> 00:39:18,160 And just over here, we're looking at your address, 798 00:39:18,160 --> 00:39:20,080 so that's a bit of data. 799 00:39:20,080 --> 00:39:22,320 And I can see you're getting directions to school, 800 00:39:22,320 --> 00:39:24,760 so that's even more data that we've got popped in there. 801 00:39:24,760 --> 00:39:26,720 This is a lot of data now, Anne-Marie. 802 00:39:26,720 --> 00:39:29,680 A lot of data just from browsing and entering information 803 00:39:29,680 --> 00:39:33,120 that is going into these cookies that are stored on their laptops. 804 00:39:33,120 --> 00:39:34,720 Now, if you stop browsing now, 805 00:39:34,720 --> 00:39:38,280 you'll see that we've got quite a lot of data on these cookies. 806 00:39:38,280 --> 00:39:41,400 And these cookies don't just stay within your laptop. 807 00:39:41,400 --> 00:39:43,840 Many of you might have accepted cookies 808 00:39:43,840 --> 00:39:46,240 that have got third party sites as well. 809 00:39:46,240 --> 00:39:48,640 So here's our third party site. What does that mean then? 810 00:39:48,640 --> 00:39:50,960 And you actually get that data as well from their cookies. 811 00:39:50,960 --> 00:39:54,120 So you know it's blue and white curtains that she's looking for. 812 00:39:54,120 --> 00:39:57,200 So if you accept third party cookies, that means another person 813 00:39:57,200 --> 00:40:00,040 can just buy that data and add it to a whole host of other data 814 00:40:00,040 --> 00:40:02,080 that they've got from all loads of other websites 815 00:40:02,080 --> 00:40:04,040 and build this incredibly detailed profile. 816 00:40:04,040 --> 00:40:05,560 That's exactly what's happening. 817 00:40:05,560 --> 00:40:08,240 Do we realise that's what we're really doing when clicking around? 818 00:40:08,240 --> 00:40:09,520 Probably not. OK. 819 00:40:09,520 --> 00:40:11,960 But the thing is, is it's not just the information 820 00:40:11,960 --> 00:40:16,440 that you're clicking on that helps to build this profile of you. 821 00:40:16,440 --> 00:40:19,760 It's also things that you are not clicking on, too. 822 00:40:19,760 --> 00:40:23,800 And so for this, let me welcome to the stage Marc Kerstein. 823 00:40:23,800 --> 00:40:27,520 APPLAUSE 824 00:40:31,960 --> 00:40:34,600 You build websites, don't you? Yes, that's right. 825 00:40:34,600 --> 00:40:36,720 And you in fact built a website for us. 826 00:40:36,720 --> 00:40:39,480 Yes. So, let's have a look at this. Talk us through what you've built. 827 00:40:39,480 --> 00:40:41,800 This is just a web page that I've created that lists 828 00:40:41,800 --> 00:40:44,320 a bunch of animals. OK. Perfect. All right. 829 00:40:44,320 --> 00:40:46,240 So, what I'd like you to do, if it's OK, 830 00:40:46,240 --> 00:40:48,320 is just have a little look through this 831 00:40:48,320 --> 00:40:51,760 and have a little flick through and stop and have a little 832 00:40:51,760 --> 00:40:56,720 read of whichever one you think ends up being particularly interesting. 833 00:40:56,720 --> 00:40:59,720 Happy? Have you stopped on one? Have you picked one? 834 00:40:59,720 --> 00:41:04,600 Yup. OK, perfect. All right. Now, Marc, which one did she pick? 835 00:41:04,600 --> 00:41:08,360 That would be the dog, is that correct? And how did you know that? 836 00:41:08,360 --> 00:41:10,440 Yes, so, it's not just about the information 837 00:41:10,440 --> 00:41:12,680 you're actively putting in, 838 00:41:12,680 --> 00:41:16,080 but it's enough to simply scroll through something to find out 839 00:41:16,080 --> 00:41:19,160 what someone's thinking of, so in this case, this web page, 840 00:41:19,160 --> 00:41:22,280 as you scroll through, you're seeing the different animals that 841 00:41:22,280 --> 00:41:25,560 are being looked at in real time, being transmitted to the server. 842 00:41:25,560 --> 00:41:28,000 So, Anne-Marie, websites aren't just tracking what you're 843 00:41:28,000 --> 00:41:31,240 clicking on, they're tracking on what you're pausing on...? Yeah. 844 00:41:31,240 --> 00:41:32,800 How fast your mouse is moving, 845 00:41:32,800 --> 00:41:35,520 the kind of device that you're using to access the website. 846 00:41:35,520 --> 00:41:38,840 They can even tell the difference between a click and a tap. 847 00:41:38,840 --> 00:41:41,840 But often, there's even more, so things like your IP address, 848 00:41:41,840 --> 00:41:45,120 which might give a clue as to where you are physically browsing from. 849 00:41:45,120 --> 00:41:48,640 Just imagine how much information you could get on someone 850 00:41:48,640 --> 00:41:52,200 if you're a professional company that had been doing it for years. 851 00:41:52,200 --> 00:41:54,520 Exactly, and that's why it's so valuable. 852 00:41:54,520 --> 00:41:57,400 There's so many insights that we can pick up from lots of different 853 00:41:57,400 --> 00:41:59,960 websites. There's so many of us that are using these platforms 854 00:41:59,960 --> 00:42:02,120 and so those cookies are pretty valuable 855 00:42:02,120 --> 00:42:04,960 over time to lots of different people. Incredibly valuable indeed. 856 00:42:04,960 --> 00:42:07,560 Thank you. APPLAUSE 857 00:42:07,560 --> 00:42:08,880 And thank you, Anne-Marie. 858 00:42:13,160 --> 00:42:14,920 Now, with all of this stuff, 859 00:42:14,920 --> 00:42:16,760 we're not saying that it's necessarily bad, 860 00:42:16,760 --> 00:42:19,400 but I think it's important to realise exactly how 861 00:42:19,400 --> 00:42:21,200 much we are giving away 862 00:42:21,200 --> 00:42:24,640 because I think if someone can work out what kind of person 863 00:42:24,640 --> 00:42:27,520 you are, they can use that information to target you 864 00:42:27,520 --> 00:42:30,640 with very precisely tailored messages. 865 00:42:30,640 --> 00:42:33,640 Now, that might be adverts to persuade you to buy something, 866 00:42:33,640 --> 00:42:36,480 but it might also be linked to political messages to 867 00:42:36,480 --> 00:42:39,280 persuade you to vote in a certain way. 868 00:42:39,280 --> 00:42:42,400 And that is something that's been in the news a lot recently. 869 00:42:42,400 --> 00:42:46,920 I think it is important to remember this is kind of how our whole 870 00:42:46,920 --> 00:42:51,240 online world is designed to work, but what makes people 871 00:42:51,240 --> 00:42:56,200 particularly unhappy is when that targeting happens with fake news. 872 00:42:56,200 --> 00:43:00,000 But even there, there is maths hiding behind the scenes. 873 00:43:00,000 --> 00:43:02,480 Because with algorithms on your side, it is 874 00:43:02,480 --> 00:43:06,360 easier to create realistic fake stuff now than it's ever been before 875 00:43:06,360 --> 00:43:10,400 and I want to show you just how easy it can be to create fake stuff. 876 00:43:10,400 --> 00:43:14,400 I want to see if you as an audience are capable of spotting some 877 00:43:14,400 --> 00:43:16,720 fake classical music. 878 00:43:16,720 --> 00:43:18,600 We're going to play a little game that 879 00:43:18,600 --> 00:43:22,120 I like to call Is The Bach Worse Than The Byte? 880 00:43:22,120 --> 00:43:25,880 OK? I was pretty proud of that. Thanks. What they're going to do, 881 00:43:25,880 --> 00:43:29,040 they're going to play you two pieces of classical music. 882 00:43:29,040 --> 00:43:33,960 One of them was composed by the very great Vivaldi and the other 883 00:43:33,960 --> 00:43:39,440 one was composed by a mathematical algorithm in the style of Vivaldi. 884 00:43:39,440 --> 00:43:41,120 What I want you to do is to see 885 00:43:41,120 --> 00:43:43,640 if you can guess which one is which, OK? 886 00:43:43,640 --> 00:43:47,760 So, two pieces of music, your job, spot the real Vivaldi. 887 00:43:47,760 --> 00:43:49,720 OK, here we go, piece of music number one. 888 00:43:53,920 --> 00:43:57,680 THEY PLAY 889 00:44:22,920 --> 00:44:25,720 That was lovely. APPLAUSE 890 00:44:29,360 --> 00:44:31,840 OK. That was piece of music number one. 891 00:44:31,840 --> 00:44:33,520 Here is piece of music number two. 892 00:44:37,040 --> 00:44:40,760 THEY PLAY 893 00:44:59,880 --> 00:45:02,680 APPLAUSE 894 00:45:04,360 --> 00:45:06,760 OK. 895 00:45:06,760 --> 00:45:11,720 But one of those was a fake, OK? So you've got to spot which one. 896 00:45:11,720 --> 00:45:16,080 If you think that the real Vivaldi was the first one, give me a cheer. 897 00:45:16,080 --> 00:45:17,760 THEY CHEER 898 00:45:17,760 --> 00:45:21,000 If you think the second song was the real Vivaldi, give me a cheer. 899 00:45:21,000 --> 00:45:23,040 THEY CHEER 900 00:45:23,040 --> 00:45:25,280 I mean, that's basically 50-50, guys. 901 00:45:25,280 --> 00:45:28,520 Just guessing at random, I see. And the real answer was...? Which one? 902 00:45:28,520 --> 00:45:30,960 Number one. Number one was the real Vivaldi. 903 00:45:30,960 --> 00:45:33,520 Well done, if you got that right. APPLAUSE 904 00:45:38,280 --> 00:45:41,400 Now, OK, that fake Vivaldi was impressive enough to 905 00:45:41,400 --> 00:45:45,040 kind of fool half a room full of people as to which one was real, 906 00:45:45,040 --> 00:45:47,760 but the algorithm itself is actually incredibly simple. 907 00:45:47,760 --> 00:45:51,320 All you do is you take a catalogue of all of the songs that 908 00:45:51,320 --> 00:45:55,880 Vivaldi has ever written and then you give the algorithm a chord 909 00:45:55,880 --> 00:45:59,800 and then it will tell you with what probability the next chord 910 00:45:59,800 --> 00:46:03,360 that was likely to come up in Vivaldi's original music. 911 00:46:03,360 --> 00:46:06,280 So all you do, give it a chord, it gives you a chord back based 912 00:46:06,280 --> 00:46:08,120 on probability, you give it a chord, 913 00:46:08,120 --> 00:46:10,200 it gives you a chord back based on probability, 914 00:46:10,200 --> 00:46:13,200 you chain those chords together, one after the other, 915 00:46:13,200 --> 00:46:16,280 until you end up with something that is entirely original. 916 00:46:16,280 --> 00:46:17,760 But that, I think, 917 00:46:17,760 --> 00:46:21,640 is the real giveaway here about the fact that this is the fake. 918 00:46:21,640 --> 00:46:24,800 Those very simple chord transitions that go on in the background. 919 00:46:24,800 --> 00:46:26,920 And there are other bits though, right? 920 00:46:26,920 --> 00:46:29,400 There was one bit at the end there particularly which 921 00:46:29,400 --> 00:46:32,520 looked like it was quite difficult. I'm afraid there are passages later 922 00:46:32,520 --> 00:46:36,960 on which are impossible to play. Impossible to play? Impossible, how? 923 00:46:36,960 --> 00:46:40,520 At the speed that the artificial intelligence has asked us to play, 924 00:46:40,520 --> 00:46:44,240 you physically can't reach extreme ends of the instrument quick enough. 925 00:46:44,240 --> 00:46:46,120 And I think that's an important point. 926 00:46:46,120 --> 00:46:48,800 Vivaldi, he had knowledge of how to play instruments, 927 00:46:48,800 --> 00:46:52,720 he had knowledge of human hands and human bodies, but the algorithm 928 00:46:52,720 --> 00:46:56,560 doesn't, it just kind of shoves loads of stuff in together. 929 00:46:56,560 --> 00:47:00,000 But I should tell you that it turns out that you can use this 930 00:47:00,000 --> 00:47:02,360 very same idea for lots of other things. 931 00:47:02,360 --> 00:47:05,200 You can use it for music, but you can also use it for lyrics 932 00:47:05,200 --> 00:47:07,640 and because it's Christmas, what I decided to do, 933 00:47:07,640 --> 00:47:11,920 I decided to feed in some classic Christmas carols 934 00:47:11,920 --> 00:47:16,880 and I decided to train my very own algorithm to generate a whole 935 00:47:16,880 --> 00:47:22,000 new carol, and so for this, please welcome to the stage Rob Levey, 936 00:47:22,000 --> 00:47:25,520 and join us for a rendition of a mathematical Christmas carol. 937 00:47:25,520 --> 00:47:27,920 Round of applause for Rob Levey. APPLAUSE 938 00:47:31,880 --> 00:47:34,320 Feel free to sing along. 939 00:48:19,680 --> 00:48:23,560 CHEERS AND APPLAUSE 940 00:48:25,240 --> 00:48:28,600 Rob Levey and the Aeolian String Quartet, everybody! 941 00:48:28,600 --> 00:48:30,200 Thank you very much! 942 00:48:34,360 --> 00:48:37,040 Here's the thing about maths in this very creative way, 943 00:48:37,040 --> 00:48:40,360 it's very impressive way, but it's also not really human 944 00:48:40,360 --> 00:48:43,560 and there is something a little bit uncomfortable about a world 945 00:48:43,560 --> 00:48:45,800 where fakes can fool, 946 00:48:45,800 --> 00:48:49,800 especially when we aren't just mimicking music, but people. 947 00:48:49,800 --> 00:48:52,640 To explain, let's welcome Dr Alex Adam. 948 00:48:52,640 --> 00:48:55,280 APPLAUSE 949 00:49:00,280 --> 00:49:03,240 Now, Alex, you work in an area called deep fakes, right? 950 00:49:03,240 --> 00:49:04,800 That's right. What are they? 951 00:49:04,800 --> 00:49:08,160 So, deep fake algorithms are a special kind of machine 952 00:49:08,160 --> 00:49:12,000 learning algorithm that can take say one person's face, 953 00:49:12,000 --> 00:49:14,840 or their head, or their entire body, or their voice, 954 00:49:14,840 --> 00:49:16,840 and turn it into another person's head. 955 00:49:16,840 --> 00:49:19,400 So, to make people look like they've done stuff in videos that 956 00:49:19,400 --> 00:49:21,080 they've never done. Absolutely. 957 00:49:21,080 --> 00:49:24,240 OK. I mean, that's quite something, right? You can manipulate someone. 958 00:49:24,240 --> 00:49:27,280 Yes, it's quite scary. You can essentially puppet people. 959 00:49:27,280 --> 00:49:28,640 So, how does it work? 960 00:49:28,640 --> 00:49:32,160 So you imagine if I have a video of you and a video of me. 961 00:49:32,160 --> 00:49:34,600 So we take those two videos and we split them 962 00:49:34,600 --> 00:49:36,680 up into their sort of individual frames, 963 00:49:36,680 --> 00:49:38,840 so the images that make up that video, 964 00:49:38,840 --> 00:49:42,000 and we'll show a computer all of those different images of me 965 00:49:42,000 --> 00:49:45,000 and you and what the algorithms will start to learn is they'll start 966 00:49:45,000 --> 00:49:48,320 to learn things like what's the structure of our different 967 00:49:48,320 --> 00:49:51,280 faces and what are the different kinds of expressions that we 968 00:49:51,280 --> 00:49:54,600 make, but critically, they learn how to distinguish the two, 969 00:49:54,600 --> 00:49:56,800 so one of them will know, OK, I can make red hair 970 00:49:56,800 --> 00:49:59,480 and a particular kind of skin tone and the other one will say, 971 00:49:59,480 --> 00:50:01,960 I can make brown hair and a particular kind of skin tone. 972 00:50:01,960 --> 00:50:05,080 But they'll separate out our expressions, so I'll be able 973 00:50:05,080 --> 00:50:10,360 to say, OK, I want to take you smiling and put that smile on to me. 974 00:50:10,360 --> 00:50:14,320 So, you're splitting out what your face looks like 975 00:50:14,320 --> 00:50:17,080 and what your face is doing. That's exactly right. 976 00:50:17,080 --> 00:50:19,880 And then you can put what your face is doing on my face. 977 00:50:19,880 --> 00:50:22,840 Whilst keeping all of your other features the same. OK, all right. 978 00:50:22,840 --> 00:50:25,880 So we've got a couple of puzzles here. So here's a picture of you. 979 00:50:25,880 --> 00:50:30,800 You're doing a little snarl. A bit of an eyebrow raise as well. 980 00:50:30,800 --> 00:50:34,640 OK, so you can make me do this face. Yeah, that's exactly right. 981 00:50:34,640 --> 00:50:37,720 How do you do it? So, if we just flip that over, 982 00:50:37,720 --> 00:50:40,520 so what we can think of is we can think of my expression 983 00:50:40,520 --> 00:50:43,480 as having some kind of abstract mathematical representation, 984 00:50:43,480 --> 00:50:45,800 which I'm sort of thinking of as these puzzle pieces 985 00:50:45,800 --> 00:50:50,120 and what I can do is I can say, well, piece 32 and say 37 over there 986 00:50:50,120 --> 00:50:52,640 correspond to me making this snarl 987 00:50:52,640 --> 00:50:55,720 and what I can do is I can just take those pieces, let's imagine I've got 988 00:50:55,720 --> 00:51:00,440 them over here, and I can just slot them in over here into your face. 989 00:51:00,440 --> 00:51:02,640 You're shuffling my face around, essentially. 990 00:51:02,640 --> 00:51:04,000 Yeah, I'm basically saying, 991 00:51:04,000 --> 00:51:07,240 turn on the bits of your face that will make you make that snarl. 992 00:51:07,240 --> 00:51:09,800 OK, so once you've completed that jigsaw, 993 00:51:09,800 --> 00:51:12,840 if you can flip it over, the idea is... 994 00:51:12,840 --> 00:51:14,560 Ah, there we go. 995 00:51:14,560 --> 00:51:16,720 That's a lovely, snarly face, isn't it? Exactly. 996 00:51:16,720 --> 00:51:18,720 Now we get Hannah with the snarl. 997 00:51:18,720 --> 00:51:22,160 Are there concerns about this kind of technology? 998 00:51:22,160 --> 00:51:25,320 So, like, my day job is a data scientist of faculty 999 00:51:25,320 --> 00:51:29,000 and I work on using artificial intelligence techniques to 1000 00:51:29,000 --> 00:51:31,480 sort of detect these sort of video manipulations. 1001 00:51:31,480 --> 00:51:34,120 Obviously, there's lots of implications for the fact that you 1002 00:51:34,120 --> 00:51:37,080 can just puppet people. I could record a video of myself saying 1003 00:51:37,080 --> 00:51:40,120 something and just transform it into some celebrity, for example, 1004 00:51:40,120 --> 00:51:42,680 and this has implications for say privacy 1005 00:51:42,680 --> 00:51:45,160 but also for say democracy and political disinformation. 1006 00:51:45,160 --> 00:51:47,760 So there are lots of sort of concerns about that, which is 1007 00:51:47,760 --> 00:51:50,600 why it's important to be able to detect this kind of content online. 1008 00:51:50,600 --> 00:51:52,640 There are also many great applications though, 1009 00:51:52,640 --> 00:51:55,080 like dubbing, special effects and things like that. 1010 00:51:55,080 --> 00:51:57,320 Do you have any top tips for spotting deep fakes? 1011 00:51:57,320 --> 00:52:00,160 Yeah, so I think my top tip for sort of spotting deep fakes is 1012 00:52:00,160 --> 00:52:02,600 always just if you're watching a video of something 1013 00:52:02,600 --> 00:52:04,800 and you sort of suspect that it might be a deep fake, 1014 00:52:04,800 --> 00:52:07,320 just ask the question, do you really believe that this person 1015 00:52:07,320 --> 00:52:09,800 would do or say these things that the video is portraying? 1016 00:52:09,800 --> 00:52:11,760 So that's like my number one suggestion. 1017 00:52:11,760 --> 00:52:14,360 My second suggestion would be sort of do some fact checking. 1018 00:52:14,360 --> 00:52:17,600 Sort of try to see if you can find that video on some trusted news 1019 00:52:17,600 --> 00:52:20,840 outlets, particularly if you found it posted on social media or 1020 00:52:20,840 --> 00:52:23,720 something that's harder to verify. And from a technical perspective, 1021 00:52:23,720 --> 00:52:26,680 I think it's useful to look for things like objects in the 1022 00:52:26,680 --> 00:52:29,400 background of a video, so if as I move you notice objects 1023 00:52:29,400 --> 00:52:31,680 in the background of the video moving with me, 1024 00:52:31,680 --> 00:52:34,680 that's a pretty strong indication that the video's been manipulated. 1025 00:52:34,680 --> 00:52:37,520 They, I think, are some excellent tips that will only serve us 1026 00:52:37,520 --> 00:52:40,760 well in the future. Alex, thank you very much indeed. Thank you. 1027 00:52:40,760 --> 00:52:44,160 APPLAUSE 1028 00:52:44,160 --> 00:52:48,480 Now, we really wanted to show you how this deep fake stuff works, 1029 00:52:48,480 --> 00:52:51,320 so what we've done is we've been working with Alex's team 1030 00:52:51,320 --> 00:52:55,000 and we've created a deep fake of someone in the audience. 1031 00:52:55,000 --> 00:52:58,040 We always look to break boundaries here at the Christmas Lectures. 1032 00:52:58,040 --> 00:53:01,600 So, to give this moment the real sense of occasion that it 1033 00:53:01,600 --> 00:53:05,840 deserves, I would like to introduce you to a brand-new talk show 1034 00:53:05,840 --> 00:53:08,440 because tonight, for one night only, 1035 00:53:08,440 --> 00:53:12,960 please welcome your host, it's Matt Parker and This Is Your Face. 1036 00:53:12,960 --> 00:53:15,680 CHEERS AND APPLAUSE 1037 00:53:18,520 --> 00:53:20,640 Thank you. 1038 00:53:20,640 --> 00:53:24,840 Welcome to This Is Your Face. Can you please welcome to the stage 1039 00:53:24,840 --> 00:53:27,320 tonight's face? It's Kaya. 1040 00:53:27,320 --> 00:53:30,520 CHEERS AND APPLAUSE 1041 00:53:32,840 --> 00:53:36,480 Come on down, Kaya. If you'd like to take a seat. Thank you. 1042 00:53:36,480 --> 00:53:40,680 So, Kaya, it's great to meet the person behind the face and to 1043 00:53:40,680 --> 00:53:44,720 get to know you, my first question is, do you have a favourite food? 1044 00:53:44,720 --> 00:53:48,800 Yes, I like pizza and pasta. Pizza and pasta. Yes. 1045 00:53:48,800 --> 00:53:51,480 That's pretty uncontroversial. Pizza and pasta?! No. 1046 00:53:51,480 --> 00:53:54,600 We're not having that. Let's have another go. 1047 00:53:54,600 --> 00:53:57,920 I really like vegetables, 1048 00:53:57,920 --> 00:54:01,600 just plain old vegetables. 1049 00:54:01,600 --> 00:54:04,960 Especially Brussels sprouts, actually. That's number one. 1050 00:54:04,960 --> 00:54:09,160 And often maybe with a side of extra vegetables. 1051 00:54:09,160 --> 00:54:12,240 Well, Kaya, that is your face. 1052 00:54:12,240 --> 00:54:16,040 My next question, do you have a favourite type of animal? Yeah. 1053 00:54:16,040 --> 00:54:19,200 I love cats. You love cats? They're pretty adorable, aren't they? Yeah. 1054 00:54:19,200 --> 00:54:22,920 Have you got like a least favourite creature? Erm, probably insects. 1055 00:54:22,920 --> 00:54:24,560 You don't like insects? No. 1056 00:54:24,560 --> 00:54:27,000 Interesting. Hm, let's see, shall we? 1057 00:54:29,280 --> 00:54:34,080 I just love ants. I think they're amazing. 1058 00:54:34,080 --> 00:54:37,480 I love the idea of them crawling all over me, 1059 00:54:37,480 --> 00:54:41,520 in my hair, up my nose, in my ears. 1060 00:54:41,520 --> 00:54:46,320 Just like a massive swarm of ants everywhere. 1061 00:54:46,320 --> 00:54:48,760 Oh, Kaya, you'd better talk to your face. 1062 00:54:48,760 --> 00:54:51,920 Now, my last question, you've got a younger sister, haven't you? Yes. 1063 00:54:51,920 --> 00:54:54,920 Are they in tonight? Yes, just right there. Over there. Excellent. 1064 00:54:54,920 --> 00:54:56,560 And you get pocket money. Yeah. 1065 00:54:56,560 --> 00:54:58,920 OK, my final question, just hypothetically, 1066 00:54:58,920 --> 00:55:02,560 your pocket money, you wouldn't mind giving all of that to let's 1067 00:55:02,560 --> 00:55:06,000 say your sister for the next roughly five years? 1068 00:55:06,000 --> 00:55:07,360 No! No? 1069 00:55:07,360 --> 00:55:08,880 Hm. 1070 00:55:08,880 --> 00:55:11,600 That's your sister? Yeah. Don't worry, we've got you. 1071 00:55:13,480 --> 00:55:16,520 I've decided to give my sister all of my pocket 1072 00:55:16,520 --> 00:55:21,280 money for the next five, ten, 20 years. 1073 00:55:21,280 --> 00:55:24,320 I think she deserves it, you know, she can have it. 1074 00:55:24,320 --> 00:55:27,760 And she can have all of my desserts too. I don't need those. 1075 00:55:27,760 --> 00:55:31,720 And she can also have my phone and my room, which is 1076 00:55:31,720 --> 00:55:35,440 where she can keep all of my clothes because they are now hers. 1077 00:55:35,440 --> 00:55:37,400 LAUGHTER 1078 00:55:37,400 --> 00:55:41,400 Wow! That is your face, so thank you very much for coming along. 1079 00:55:41,400 --> 00:55:43,240 No problem. No, not you. Your face! 1080 00:55:43,240 --> 00:55:45,360 Thanks, it's been wonderful to be here 1081 00:55:45,360 --> 00:55:48,440 and I completely stand by all of my answers. 1082 00:55:48,440 --> 00:55:52,040 APPLAUSE Thank you very much, Kaya. 1083 00:55:56,400 --> 00:55:59,920 There is a point to all of this 1084 00:55:59,920 --> 00:56:03,440 because whether you're talking about deep fakes or driverless cars 1085 00:56:03,440 --> 00:56:05,920 or facial recognition, under the surface, 1086 00:56:05,920 --> 00:56:09,200 all of these things that have such potential to change our world, 1087 00:56:09,200 --> 00:56:12,520 they're ultimately mathematical creations 1088 00:56:12,520 --> 00:56:15,520 and maths isn't just about bridges and bicycles and buildings. 1089 00:56:15,520 --> 00:56:17,000 Behind the scenes, 1090 00:56:17,000 --> 00:56:20,760 it's mathematical levers that are powering the changes to our society. 1091 00:56:20,760 --> 00:56:23,600 It's got the potential to change everything, what we know, 1092 00:56:23,600 --> 00:56:28,040 how we talk to each other, even how our whole democracy is structured. 1093 00:56:28,040 --> 00:56:29,200 Now, don't get me wrong, 1094 00:56:29,200 --> 00:56:32,040 I don't think this is about being afraid of the advancement 1095 00:56:32,040 --> 00:56:35,040 of machines, but I do think that we need to be honest with 1096 00:56:35,040 --> 00:56:38,360 ourselves about the awesome power of mathematics 1097 00:56:38,360 --> 00:56:40,560 and I think we need to be careful of the very real 1098 00:56:40,560 --> 00:56:44,360 limitations of something that will never be perfect, 1099 00:56:44,360 --> 00:56:46,760 but I think if we can be aware of the pitfalls, 1100 00:56:46,760 --> 00:56:49,000 if we can work our way around the challenges, 1101 00:56:49,000 --> 00:56:52,040 I am optimistic about a future where humans 1102 00:56:52,040 --> 00:56:55,200 and machines can work together, exploiting each other's strengths 1103 00:56:55,200 --> 00:56:57,520 and acknowledging each other's weaknesses, 1104 00:56:57,520 --> 00:57:01,440 a partnership has the real potential to be a force for good. 1105 00:57:01,440 --> 00:57:04,440 And so to finish, let's combine human 1106 00:57:04,440 --> 00:57:08,640 and machine in a performance of collaboration. 1107 00:57:08,640 --> 00:57:11,720 Please welcome the Chineke! Orchestra. 1108 00:57:11,720 --> 00:57:14,000 APPLAUSE 1109 00:57:14,000 --> 00:57:18,720 Your job is to guess which parts are human 1110 00:57:18,720 --> 00:57:21,360 and which were the mathematical algorithm. 1111 00:57:21,360 --> 00:57:23,960 Enjoy. 1112 00:57:23,960 --> 00:57:27,240 THEY PLAY