All language subtitles for Secrets and Lies 2of3 How to Bend the Rules BBC RICL 2019 720p_ENG_Subtitles01

af Afrikaans
ak Akan
sq Albanian
am Amharic
ar Arabic
hy Armenian
az Azerbaijani
eu Basque
be Belarusian
bem Bemba
bn Bengali
bh Bihari
bs Bosnian
br Breton
bg Bulgarian
km Cambodian
ca Catalan
chr Cherokee
ny Chichewa
zh-CN Chinese (Simplified)
zh-TW Chinese (Traditional)
co Corsican
hr Croatian
cs Czech
da Danish
nl Dutch
en English Download
eo Esperanto
et Estonian
ee Ewe
fo Faroese
tl Filipino
fi Finnish
fr French
fy Frisian
gaa Ga
gl Galician
ka Georgian
de German
el Greek
gn Guarani
gu Gujarati
ht Haitian Creole
ha Hausa
haw Hawaiian
iw Hebrew
hi Hindi
hu Hungarian
is Icelandic
ig Igbo
id Indonesian
ia Interlingua
ga Irish
it Italian
ja Japanese
jw Javanese
kn Kannada
kk Kazakh
rw Kinyarwanda
rn Kirundi
kg Kongo
ko Korean
kri Krio (Sierra Leone)
ku Kurdish
ckb Kurdish (Soranî)
ky Kyrgyz
lo Laothian
la Latin
lv Latvian
ln Lingala
lt Lithuanian
loz Lozi
lg Luganda
ach Luo
mk Macedonian
mg Malagasy
ms Malay
ml Malayalam
mt Maltese
mi Maori
mr Marathi
mfe Mauritian Creole
mo Moldavian
mn Mongolian
sr-ME Montenegrin
ne Nepali
pcm Nigerian Pidgin
nso Northern Sotho
no Norwegian
nn Norwegian (Nynorsk)
oc Occitan
or Oriya
om Oromo
ps Pashto
fa Persian
pl Polish
pt-BR Portuguese (Brazil)
pt-PT Portuguese (Portugal)
pa Punjabi
qu Quechua
ro Romanian
rm Romansh
nyn Runyakitara
ru Russian
gd Scots Gaelic
sr Serbian
sh Serbo-Croatian
st Sesotho
tn Setswana
crs Seychellois Creole
sn Shona
sd Sindhi
si Sinhalese
sk Slovak
sl Slovenian
so Somali
es Spanish
es-419 Spanish (Latin American)
su Sundanese
sw Swahili
sv Swedish
tg Tajik
ta Tamil
tt Tatar
te Telugu
th Thai
ti Tigrinya
to Tonga
lua Tshiluba
tum Tumbuka
tr Turkish Download
tk Turkmen
tw Twi
ug Uighur
uk Ukrainian
ur Urdu
uz Uzbek
vi Vietnamese
cy Welsh
wo Wolof
xh Xhosa
yi Yiddish
yo Yoruba
zu Zulu
Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated: 1 00:00:01,000 --> 00:00:05,100 If you want to understand the weird and wonderful world around you, 2 00:00:05,100 --> 00:00:07,660 then maths is your friend. 3 00:00:07,660 --> 00:00:10,980 But the real trick is, that with a little bit of mathematical 4 00:00:10,980 --> 00:00:15,980 thinking, you can take those rules and patterns, bend and twist them, 5 00:00:15,980 --> 00:00:18,500 and fire them straight back at you. 6 00:00:20,700 --> 00:00:25,500 Tonight, we are going to explore just how far we can push those 7 00:00:25,500 --> 00:00:28,180 barriers to achieve amazing things. 8 00:00:51,660 --> 00:00:55,980 CHEERING AND APPLAUSE 9 00:01:03,380 --> 00:01:06,460 Welcome to the Christmas Lectures. I am Dr Hannah Fry 10 00:01:06,460 --> 00:01:08,980 and tonight we are going to be talking about how to give 11 00:01:08,980 --> 00:01:11,380 ourselves superhuman skills. 12 00:01:11,380 --> 00:01:14,580 Now, we're very good at using maths to understand the world around us. 13 00:01:14,580 --> 00:01:17,580 But if we want to bend the world to our will, we're going 14 00:01:17,580 --> 00:01:19,380 to need to be a little bit more inventive. 15 00:01:19,380 --> 00:01:21,780 We're going to need to take this maths up a level here. 16 00:01:21,780 --> 00:01:25,340 So I thought what we'd do is we would start off tonight with 17 00:01:25,340 --> 00:01:26,860 something of a competition. 18 00:01:26,860 --> 00:01:31,140 So I wonder, does anyone know how to solve one of these? 19 00:01:31,140 --> 00:01:33,540 Who doesn't? Oh, actually, quite a few of you. 20 00:01:33,540 --> 00:01:35,180 OK. Who doesn't mind coming down? 21 00:01:35,180 --> 00:01:36,820 Let's go with you, actually, yeah. 22 00:01:36,820 --> 00:01:39,180 If you want to come down. Round of applause, if you can, 23 00:01:39,180 --> 00:01:41,220 as he comes to the stage. 24 00:01:42,740 --> 00:01:44,380 What's your name? George. George. 25 00:01:44,380 --> 00:01:46,220 OK. How good are you at these, George? 26 00:01:46,220 --> 00:01:48,980 Solved it a few times. OK. All right. Let's give you a go. 27 00:01:48,980 --> 00:01:51,100 We're going to give you lots of encouragement as you go. 28 00:01:51,100 --> 00:01:53,660 Can I look at it first, like...? Yeah, you can have a little look, 29 00:01:53,660 --> 00:01:56,060 yeah. Cheers. We'll be patient with you, though. OK. 30 00:01:56,060 --> 00:01:58,900 All right. Yeah. I reckon I'm ready to go. 31 00:01:58,900 --> 00:02:01,460 OK. Come on, then, George. Come on, George! 32 00:02:01,460 --> 00:02:03,940 Hang on, George. Hang on, George. 33 00:02:03,940 --> 00:02:05,460 AUDIENCE MURMURS 34 00:02:07,260 --> 00:02:08,580 George? 35 00:02:08,580 --> 00:02:09,820 AUDIENCE: Whoo! 36 00:02:09,820 --> 00:02:11,500 AUDIENCE WHOOPS 37 00:02:15,460 --> 00:02:17,780 George, you've solved these more than a few times, haven't you? 38 00:02:17,780 --> 00:02:20,900 Yeah, quite a few, actually. Tell us who you actually are, George. 39 00:02:20,900 --> 00:02:23,900 I'm actually the UK champion for solving a Rubik's Cube. 40 00:02:23,900 --> 00:02:25,660 Yeah. You're the best in Britain. 41 00:02:25,660 --> 00:02:27,820 Yeah. Currently reigning UK champion. Yeah. 42 00:02:27,820 --> 00:02:30,180 So tell us, OK, how do you actually solve one of these? 43 00:02:30,180 --> 00:02:33,300 OK, so I guess the first thing to know is that, like, it's got quite 44 00:02:33,300 --> 00:02:35,860 a lot of combinations. So it's not like you just like learn one thing 45 00:02:35,860 --> 00:02:37,580 and then you can solve it straight away. 46 00:02:37,580 --> 00:02:40,300 So, like, when you scramble it, chances are that scramble's 47 00:02:40,300 --> 00:02:42,340 actually never, ever been seen before. It just has so many 48 00:02:42,340 --> 00:02:44,380 combinations - it has 42 quintillion 49 00:02:44,380 --> 00:02:47,060 combinations... That's a big number. 43 followed by 19 zeros. 50 00:02:47,060 --> 00:02:51,980 So, you kind of need to split up into steps and you need to learn 51 00:02:51,980 --> 00:02:54,780 algorithms, like this whole actual... A list of instructions. 52 00:02:54,780 --> 00:02:55,980 Yes. 53 00:02:55,980 --> 00:02:58,700 So, algorithms, when applied to a cube, that kind of means 54 00:02:58,700 --> 00:03:01,780 you're temporarily mixing up the cube, then putting 55 00:03:01,780 --> 00:03:04,100 it back, having switched a few pieces. 56 00:03:04,100 --> 00:03:07,300 So the more algorithms you know, the more efficient you'll be. 57 00:03:07,300 --> 00:03:09,980 If I combine that efficiency with faster turning, it allows me 58 00:03:09,980 --> 00:03:11,340 to be a faster solver. 59 00:03:11,340 --> 00:03:13,700 Now, there's a very key point here in what George is saying, 60 00:03:13,700 --> 00:03:16,660 which is that sometimes maths doesn't look like numbers and 61 00:03:16,660 --> 00:03:19,580 equations, sometimes maths can just look like a list of instructions. 62 00:03:19,580 --> 00:03:22,300 I mean, you can solve this by following a list of instructions. 63 00:03:22,300 --> 00:03:23,940 An algorithm, as you say. 64 00:03:23,940 --> 00:03:27,540 And that means, George, that if I want to beat you at this, all I've 65 00:03:27,540 --> 00:03:30,140 got to do is follow those instructions faster than you, 66 00:03:30,140 --> 00:03:33,780 right? Yeah. Well, thankfully I have just the thing that's going to 67 00:03:33,780 --> 00:03:39,100 help me, and Illius here is the creator of this machine. 68 00:03:39,100 --> 00:03:42,020 So I tell you what, hang on, let me just scramble this one up again. 69 00:03:42,020 --> 00:03:45,020 We're going to have a bit of a competition. We're going to do you 70 00:03:45,020 --> 00:03:46,740 against the machine. OK, so here we go. 71 00:03:46,740 --> 00:03:47,940 Let's scramble this one. 72 00:03:47,940 --> 00:03:49,780 You're not allowed to look at this now. All right. 73 00:03:49,780 --> 00:03:52,020 So I'll tell you when you're allowed to look. 74 00:03:52,020 --> 00:03:53,580 Is it ready to go, Illius? 75 00:03:53,580 --> 00:03:55,940 OK. All right. We're going to count it down. And then you and the 76 00:03:55,940 --> 00:03:58,340 machine can both look. Ready? Three, two, one, go. 77 00:03:59,260 --> 00:04:02,140 Who's going to win? Who's going to win? 78 00:04:02,140 --> 00:04:04,060 Ohhh! 79 00:04:04,060 --> 00:04:07,820 George! You're letting yourself down! I know! 80 00:04:07,820 --> 00:04:10,220 That is absolutely incredible. 81 00:04:10,220 --> 00:04:12,340 APPLAUSE AND CHEERING 82 00:04:15,660 --> 00:04:18,380 I mean, that was amazing. That was amazing. OK. 83 00:04:18,380 --> 00:04:20,980 Let's go over here have a little look back at what we're seeing. 84 00:04:20,980 --> 00:04:23,660 I mean, this machine is doing the same things as you. Exactly, yeah. 85 00:04:23,660 --> 00:04:27,300 It's just, it's, like, seeing when to use those algorithms way more 86 00:04:27,300 --> 00:04:29,380 quickly and it does it, like, all in one go, 87 00:04:29,380 --> 00:04:31,940 so it doesn't have to pause to look to when to use the next 88 00:04:31,940 --> 00:04:34,540 algorithm or anything like that. So we can look at a little slow-mo 89 00:04:34,540 --> 00:04:36,980 replay, I think, of this. 90 00:04:36,980 --> 00:04:39,620 And it is a properly mixed-up cube, there. 91 00:04:39,620 --> 00:04:40,900 Wow! Amazing. 92 00:04:40,900 --> 00:04:42,780 That was absolutely incredible. 93 00:04:42,780 --> 00:04:45,940 So it's not better than you. It's just faster. 94 00:04:45,940 --> 00:04:48,980 I suppose so. Yeah. It'll know more algorithms than I do. 95 00:04:48,980 --> 00:04:51,620 Yeah. Amazing. George and the cube-solving machine, 96 00:04:51,620 --> 00:04:54,740 thank you very much. Thank you. 97 00:04:54,740 --> 00:04:58,220 Thank you. That was amazing. So impressed. 98 00:04:58,220 --> 00:05:00,700 That really, I think, was a great example 99 00:05:00,700 --> 00:05:02,940 of a machine following instructions. 100 00:05:02,940 --> 00:05:06,180 The question is, how do you get those instructions into the machine 101 00:05:06,180 --> 00:05:07,260 in the first place? 102 00:05:07,260 --> 00:05:11,740 Well, one man who knows the answer is coder extraordinaire 103 00:05:11,740 --> 00:05:13,740 Seb Lee-Delisle. 104 00:05:15,180 --> 00:05:16,940 Hey, Seb! 105 00:05:18,740 --> 00:05:20,940 So, Seb, tell us about the kind of things that you code. 106 00:05:20,940 --> 00:05:24,060 Well, I like to make really big light installations 107 00:05:24,060 --> 00:05:25,700 that are interactive. Oh, OK. 108 00:05:25,700 --> 00:05:28,540 And we're seeing some footage of this, here. On the side of 109 00:05:28,540 --> 00:05:29,940 buildings, no less. Yeah. Yeah. 110 00:05:29,940 --> 00:05:33,660 See, my ambition really is to replace fireworks. With lasers. 111 00:05:33,660 --> 00:05:36,460 With lasers. Yeah, cos, like, fireworks are old technology. 112 00:05:36,460 --> 00:05:40,420 Uh-huh. Now we've got technology that's just as bright as fireworks, 113 00:05:40,420 --> 00:05:43,020 lasers, super-bright LEDs. 114 00:05:43,020 --> 00:05:45,660 But all of my technology is controlled by computers. 115 00:05:45,660 --> 00:05:48,060 OK, so talking about computers, then, how do 116 00:05:48,060 --> 00:05:49,300 you control these things? 117 00:05:49,300 --> 00:05:52,380 Well, I'm going to show you how we make something called 118 00:05:52,380 --> 00:05:53,700 a particle system. 119 00:05:53,700 --> 00:05:55,820 So I've got a browser open. 120 00:05:55,820 --> 00:05:58,580 I'm going to do some coding in JavaScript to make some graphics. 121 00:05:58,580 --> 00:06:01,140 You're going to make an algorithm. An algorithm, yeah. A series of 122 00:06:01,140 --> 00:06:02,940 instructions for the machine. Exactly right. 123 00:06:02,940 --> 00:06:05,180 So the first thing I'm going to do is make a single particle, 124 00:06:05,180 --> 00:06:07,860 which is essentially just a shape. And we've got to decide 125 00:06:07,860 --> 00:06:11,140 what colour that shape should be. Any ideas what colour? 126 00:06:11,140 --> 00:06:13,340 AUDIENCE SHOUTS SUGGESTIONS 127 00:06:15,180 --> 00:06:17,980 I think first thing I heard was "pink". 128 00:06:17,980 --> 00:06:20,940 So I'm going to actually use magenta, 129 00:06:20,940 --> 00:06:23,180 which is my favourite type of pink, 130 00:06:23,180 --> 00:06:25,660 and it's a very nice bright pink. 131 00:06:25,660 --> 00:06:29,460 And the next thing I've got to do is draw a filled-in shape, so 132 00:06:29,460 --> 00:06:30,540 let's hear some shapes. 133 00:06:30,540 --> 00:06:31,860 AUDIENCE SHOUTS SUGGESTIONS 134 00:06:31,860 --> 00:06:34,420 Circle. OK, OK! 135 00:06:34,420 --> 00:06:36,900 Square. I think I heard "square". 136 00:06:36,900 --> 00:06:39,980 Definitely. OK. Square. So a square is just a rectangle. 137 00:06:39,980 --> 00:06:42,220 So let's draw a filled-in rectangle. 138 00:06:42,220 --> 00:06:45,860 We're going to draw it at 00, which is the top left of the screen 139 00:06:45,860 --> 00:06:50,140 in X and Y coordinates, and we're going to make it 100 wide and 140 00:06:50,140 --> 00:06:52,540 100 high, so let's look at our square. There it is. 141 00:06:52,540 --> 00:06:54,780 So this X and Y that you were talking about there... Yeah. 142 00:06:54,780 --> 00:06:56,620 ..are you treating this as though it's a graph? 143 00:06:56,620 --> 00:06:59,220 Yeah. So this huge canvas in JavaScript, 144 00:06:59,220 --> 00:07:02,940 it's got pixels and I can identify each pixel by its X and Y 145 00:07:02,940 --> 00:07:06,020 co-ordinate. OK. Yeah. This is just straight-up maths so far. 146 00:07:06,020 --> 00:07:09,340 It's all maths. It's maths all the way down, and I can change this X 147 00:07:09,340 --> 00:07:13,980 value and you can see that the rectangle moves to whatever 148 00:07:13,980 --> 00:07:17,620 co-ordinate I specify for the X position. As you increase X, 149 00:07:17,620 --> 00:07:20,140 this is moving right. Yeah. Exactly. 150 00:07:20,140 --> 00:07:22,140 So let's do the same thing for the Y value. 151 00:07:22,140 --> 00:07:25,180 You can see it moves down and to the right. Ahh! 152 00:07:25,180 --> 00:07:27,300 Right. So this is good. It's the start of our particle. 153 00:07:27,300 --> 00:07:28,380 But it's not moving. 154 00:07:28,380 --> 00:07:29,940 It's completely static. 155 00:07:29,940 --> 00:07:33,300 So in order to get it moving, I'm going to wrap all of this stuff up 156 00:07:33,300 --> 00:07:37,380 in a function, and I'm going to call this function 60 157 00:07:37,380 --> 00:07:38,820 times a seconds. 158 00:07:38,820 --> 00:07:43,380 But now I can do X+=1, which adds one to X every time. 159 00:07:43,380 --> 00:07:46,820 And now, we should have the rectangle in a brand-new place 160 00:07:46,820 --> 00:07:48,660 every time. We've made some animation. Aha! 161 00:07:48,660 --> 00:07:50,740 Doesn't look much like a firework. No. 162 00:07:50,740 --> 00:07:52,820 So we get we're getting there. 163 00:07:52,820 --> 00:07:55,900 So, in particle systems, usually the particles are very small. 164 00:07:55,900 --> 00:07:57,700 That's the first thing we can fix. 165 00:07:57,700 --> 00:07:59,700 Let's just make it really little. Aww! 166 00:07:59,700 --> 00:08:05,220 And now, let's make some new variables for the X velocity 167 00:08:05,220 --> 00:08:07,460 and for the Y velocity. 168 00:08:07,460 --> 00:08:11,900 The best part is when I give those velocity values a random 169 00:08:11,900 --> 00:08:16,140 number. Oh, OK. So you can do that with the command "random". 170 00:08:16,140 --> 00:08:20,140 If I pass in minus ten to plus ten, then we're going to get a random 171 00:08:20,140 --> 00:08:23,100 number in the X velocity between those two numbers 172 00:08:23,100 --> 00:08:25,620 and the same with the Y velocity. 173 00:08:25,620 --> 00:08:30,180 And you'll see now that every time we run it... 174 00:08:30,180 --> 00:08:32,220 Goes in a different direction! ..it goes in a completely 175 00:08:32,220 --> 00:08:34,220 different direction. A random direction every time. 176 00:08:34,220 --> 00:08:36,340 Right. Exactly. So that's kind of fun. 177 00:08:36,340 --> 00:08:38,020 So we've got one particle. 178 00:08:38,020 --> 00:08:39,500 One particle is cute. 179 00:08:39,500 --> 00:08:42,500 But really, the idea of particle systems is that we make a whole 180 00:08:42,500 --> 00:08:44,020 bunch of particles. 181 00:08:44,020 --> 00:08:45,860 So now, in this next bit of code, 182 00:08:45,860 --> 00:08:48,180 I'm doing essentially the same thing, 183 00:08:48,180 --> 00:08:50,260 but I'm creating an array of particles. 184 00:08:50,260 --> 00:08:53,380 And we get an effect like this. Ahh! 185 00:08:53,380 --> 00:08:55,980 That looks like a sparkler! We're getting there. Aren't we? 186 00:08:55,980 --> 00:08:57,220 We're getting to the sparkler. 187 00:08:57,220 --> 00:09:00,540 Now, this is an ordinary projector, so it's not very bright. 188 00:09:00,540 --> 00:09:03,060 If we want to create something that's truly spectacular, 189 00:09:03,060 --> 00:09:06,140 like a real firework, we need to use a laser. 190 00:09:06,140 --> 00:09:08,260 Ah. Wow! They're extremely bright. Right. 191 00:09:08,260 --> 00:09:10,020 They're really, really bright, aren't they? 192 00:09:10,020 --> 00:09:12,340 And so I can adjust all the settings here. 193 00:09:12,340 --> 00:09:14,940 I can change the speed, how fast they are, 194 00:09:14,940 --> 00:09:18,060 make them fade out over time a little bit. 195 00:09:18,060 --> 00:09:23,620 And, yeah, I can add some gravity to make them fall down. Amazing! 196 00:09:23,620 --> 00:09:25,860 But, of course, the stuff that you are really well known 197 00:09:25,860 --> 00:09:29,140 for, Seb, is not just you having control 198 00:09:29,140 --> 00:09:31,940 of where this is on the Y axis, where it is on the graph... 199 00:09:31,940 --> 00:09:34,900 Yeah. ..but letting the audience control it. Exactly. I love doing 200 00:09:34,900 --> 00:09:38,180 interactive things, so what I thought I'd do is actually take 201 00:09:38,180 --> 00:09:42,980 the sound from the microphone in my laptop and apply that level 202 00:09:42,980 --> 00:09:46,940 to the Y origin of the particles and then we could make a game 203 00:09:46,940 --> 00:09:48,980 out of it. So here we've got this. 204 00:09:48,980 --> 00:09:50,820 You can see the sparkler on the left. 205 00:09:50,820 --> 00:09:53,580 Now it should be that the more noise that you make, the higher 206 00:09:53,580 --> 00:09:55,580 this sparkle rises. All right. 207 00:09:55,580 --> 00:09:59,100 Here we go. Three, two, one, go. 208 00:09:59,100 --> 00:10:02,260 AUDIENCE CHEERS RAUCOUSLY 209 00:10:06,180 --> 00:10:09,620 NOISE DROWNS SPEECH 210 00:10:15,620 --> 00:10:17,420 This is it, guys! 211 00:10:20,740 --> 00:10:23,580 It's getting harder. It's getting harder. 212 00:10:25,780 --> 00:10:27,980 NOISE DROWNS SEB'S SPEECH 213 00:10:29,900 --> 00:10:33,300 Ohh! That was amazing! That was such a high score. 214 00:10:33,300 --> 00:10:37,580 Well done, everyone. Amazing, amazing. Seb, thank you 215 00:10:37,580 --> 00:10:40,940 for showing us your all your lasers. Very, very welcome. Thank you. 216 00:10:40,940 --> 00:10:42,300 Thank you very much. Thank you. 217 00:10:42,300 --> 00:10:44,140 CHEERING AND APPLAUSE 218 00:10:45,620 --> 00:10:48,180 Now, Seb's fireworks look very realistic, there, 219 00:10:48,180 --> 00:10:51,380 but I think if you want something that looks as real as possible, 220 00:10:51,380 --> 00:10:54,820 then you really can't do better than looking at Hollywood films - 221 00:10:54,820 --> 00:10:57,340 the visual effects that they use in their movies. 222 00:10:57,340 --> 00:10:59,500 Now, these are the people, the people who create those, 223 00:10:59,500 --> 00:11:02,620 they're the people who, they don't only copy reality, but they really 224 00:11:02,620 --> 00:11:05,060 know how to bend the rules to their will 225 00:11:05,060 --> 00:11:06,860 for our entertainment. 226 00:11:06,860 --> 00:11:10,620 And we are very lucky, because today, we are joined by someone who 227 00:11:10,620 --> 00:11:14,460 does exactly that. From Weta Digital in New Zealand, 228 00:11:14,460 --> 00:11:17,700 please join me in welcoming Anders Langlands. 229 00:11:17,700 --> 00:11:21,060 CHEERING AND APPLAUSE 230 00:11:21,060 --> 00:11:22,820 Hi. Hey, Anders! 231 00:11:24,300 --> 00:11:26,220 So what films have you worked on, Anders? 232 00:11:26,220 --> 00:11:28,660 Well, Weta's got quite a long history of going all the way back 233 00:11:28,660 --> 00:11:32,180 to The Lord Of The Rings, Avatar, Planet Of The Apes and then, most 234 00:11:32,180 --> 00:11:34,540 recently, Avengers this year. Oh, crikey. 235 00:11:34,540 --> 00:11:36,460 I mean, you don't get much bigger than Avengers. No. 236 00:11:36,460 --> 00:11:39,380 How much of the animations that we see - the CGI that we see in films - 237 00:11:39,380 --> 00:11:41,300 how much of it is maths? 238 00:11:41,300 --> 00:11:42,620 Well, it's all maths, really. 239 00:11:42,620 --> 00:11:44,780 It's like a combination between art and science. 240 00:11:44,780 --> 00:11:49,580 And we use algorithms to simulate how things move, to give animators 241 00:11:49,580 --> 00:11:52,980 controls, to make things move according to their desires, 242 00:11:52,980 --> 00:11:56,140 and to render the pictures and make the images through mathematical 243 00:11:56,140 --> 00:11:57,340 equations, as well. 244 00:11:57,340 --> 00:12:00,820 But presumably, if you want to make things look realistic on the screen, 245 00:12:00,820 --> 00:12:03,300 you have to understand how they work in real life. 246 00:12:03,300 --> 00:12:07,220 And that really is what this tower behind us, that has just been built, 247 00:12:07,220 --> 00:12:08,820 is here for. 248 00:12:08,820 --> 00:12:12,580 Because I would like a volunteer who would like to come 249 00:12:12,580 --> 00:12:17,500 down and help me to dismantle this tower. 250 00:12:17,500 --> 00:12:19,780 Let's go for, yeah, do you want to come down? 251 00:12:19,780 --> 00:12:21,740 Round of applause as they come to the stage! 252 00:12:21,740 --> 00:12:24,820 CHEERING 253 00:12:27,220 --> 00:12:28,260 What's your name? 254 00:12:28,260 --> 00:12:29,580 APPLAUSE DROWNS RESPONSE 255 00:12:29,580 --> 00:12:32,300 All right, so we have got this bowling ball. 256 00:12:32,300 --> 00:12:34,060 Now, I want you to roll it very gently. 257 00:12:34,060 --> 00:12:37,580 This is a very old and important building and we will send you a 258 00:12:37,580 --> 00:12:40,180 bill. So very gently and try and smash down this tower. 259 00:12:40,180 --> 00:12:42,780 Can I stand somewhere else? You can stand a bit further back here 260 00:12:42,780 --> 00:12:45,940 if you want to, give you a bit of a run up, but let's count it 261 00:12:45,940 --> 00:12:50,020 down. Ready? OK. Three... ALL: Two, one. 262 00:12:50,020 --> 00:12:51,580 Go! 263 00:12:51,580 --> 00:12:53,260 Oh! 264 00:12:53,260 --> 00:12:54,980 It's pretty good! 265 00:12:54,980 --> 00:12:58,540 Yay! Ha-ha. Well done! 266 00:12:58,540 --> 00:13:00,380 CHEERING 267 00:13:01,740 --> 00:13:05,460 But stuff like that, it isn't just for fun for you guys, right? 268 00:13:05,460 --> 00:13:08,820 No, I mean, we have to do things a little bit more complex than that, 269 00:13:08,820 --> 00:13:12,340 but the basic equations of motion that govern something like that, 270 00:13:12,340 --> 00:13:14,580 as you've just seen, are pretty simple. 271 00:13:14,580 --> 00:13:16,620 Simple enough to solve by hand, in a lot of cases. 272 00:13:16,620 --> 00:13:19,340 And if you've done GCSE maths, then you've probably encountered some 273 00:13:19,340 --> 00:13:21,700 of that, too. Just the physics behind what just happened? 274 00:13:21,700 --> 00:13:24,060 The physics behind what just happened. Of course, we want to 275 00:13:24,060 --> 00:13:25,740 simulate much more complex things, and even 276 00:13:25,740 --> 00:13:28,140 something like this tower falling over, 277 00:13:28,140 --> 00:13:31,060 we need to use computers to do that so we can do it fast enough 278 00:13:31,060 --> 00:13:32,900 to make a moving animation. 279 00:13:32,900 --> 00:13:36,100 And, in fact, talking of moving animations, you have brought one 280 00:13:36,100 --> 00:13:38,860 with you, so let's have a little look over here. 281 00:13:38,860 --> 00:13:40,260 This is our tower, right? 282 00:13:40,260 --> 00:13:43,220 This is kind of the same as our tower? Yes, pretty similar. 283 00:13:43,220 --> 00:13:45,900 But this one is a complete animation. 284 00:13:45,900 --> 00:13:49,740 Yes. So we look at the forces of gravity and the force of the ball 285 00:13:49,740 --> 00:13:52,820 being thrown, analyse the collisions, and then, 286 00:13:52,820 --> 00:13:53,940 every single frame - 287 00:13:53,940 --> 00:13:57,020 so, at least 24 times a second - we update all of that and work out 288 00:13:57,020 --> 00:13:58,860 where the blocks are going to be on the next frame 289 00:13:58,860 --> 00:14:01,260 to create an animation. So let's have a little look at it 290 00:14:01,260 --> 00:14:03,740 as it's animated. It's pretty good. 291 00:14:03,740 --> 00:14:05,940 I mean, you've even got the little bit I had to kick down. 292 00:14:05,940 --> 00:14:07,380 Just the wrong way round. 293 00:14:07,380 --> 00:14:09,980 I haven't got a digital Hannah to kick it down, I'm afraid. OK. 294 00:14:09,980 --> 00:14:13,260 So once you are at this stage, once you have this stuff created, 295 00:14:13,260 --> 00:14:15,580 you understand how the basics work, then what do you do? 296 00:14:15,580 --> 00:14:18,260 Well, then we can have a little fun with it. So because we create the 297 00:14:18,260 --> 00:14:19,860 world, we can do things like change 298 00:14:19,860 --> 00:14:22,500 all the boxes into china so they shatter when it gets hit, 299 00:14:22,500 --> 00:14:24,420 rather than falling over. 300 00:14:24,420 --> 00:14:27,260 We could change all the boxes into balloons and then pump them full 301 00:14:27,260 --> 00:14:28,940 of helium so they float away. 302 00:14:28,940 --> 00:14:31,660 That's nice! Escaping round the side. That's nice. 303 00:14:31,660 --> 00:14:33,900 I like that a lot. It's getting away. Or maybe if we wanted 304 00:14:33,900 --> 00:14:36,020 something a bit more exciting, we could douse them 305 00:14:36,020 --> 00:14:38,460 in petrol and then set them on fire. Oh, wow! 306 00:14:38,460 --> 00:14:39,620 AUDIENCE LAUGHS 307 00:14:39,620 --> 00:14:41,500 That looks very realistic. 308 00:14:41,500 --> 00:14:44,180 Thanks. That's kind of the point! 309 00:14:44,180 --> 00:14:47,020 How do you make that look that realistic, though? 310 00:14:47,020 --> 00:14:48,860 Well, like you said, we have to analyse what's 311 00:14:48,860 --> 00:14:51,540 going on in the real world and use maths to try and simulate it 312 00:14:51,540 --> 00:14:55,500 as closely as possible. So in the case of fire or explosions, for 313 00:14:55,500 --> 00:14:57,700 example, we look at the chemical processes that drive that. 314 00:14:57,700 --> 00:15:02,420 So, for fire, the combustion process is turning a fuel into water 315 00:15:02,420 --> 00:15:04,660 and carbon dioxide and some carbon. 316 00:15:04,660 --> 00:15:08,500 And then the temperature then decides how hot that soot gets, 317 00:15:08,500 --> 00:15:11,580 and that decides what the brightness and the colour of the flame then is. 318 00:15:11,580 --> 00:15:14,780 When you're working with buildings, say, do you take into 319 00:15:14,780 --> 00:15:17,220 account the things that the buildings are made from? 320 00:15:17,220 --> 00:15:19,580 Yeah, we want to analyse all the different material properties 321 00:15:19,580 --> 00:15:21,860 that are in a building, so, like, 322 00:15:21,860 --> 00:15:23,980 how bricks shatter when they get hit. 323 00:15:23,980 --> 00:15:26,980 Mortar turning into dust, the steel bending. 324 00:15:26,980 --> 00:15:30,500 We have to layer all of these things up using either one big simulation 325 00:15:30,500 --> 00:15:33,300 or lots of simulations to get put together. 326 00:15:33,300 --> 00:15:37,060 And, you know, if you're doing a giant bath toy knocking a building 327 00:15:37,060 --> 00:15:39,700 down, you're kind of outside of the realm of real-world physics 328 00:15:39,700 --> 00:15:42,660 at that point. So a lot of what our effects artists do, a lot of their 329 00:15:42,660 --> 00:15:47,940 skill comes into figuring out how to tune the physics of the simulation 330 00:15:47,940 --> 00:15:50,740 to give something that you've never seen before that the director wants, 331 00:15:50,740 --> 00:15:53,020 but also something that the audience can believe is real. 332 00:15:53,020 --> 00:15:56,060 Take reality and bend it. Yeah. So a big part of what our team of 333 00:15:56,060 --> 00:15:58,020 scientists and software engineers 334 00:15:58,020 --> 00:16:01,140 at Weta do is use maths to figure out cool tricks to make the 335 00:16:01,140 --> 00:16:03,460 simulations run faster, while still looking real. 336 00:16:03,460 --> 00:16:05,700 That is amazing. Anders, thank you. very much. 337 00:16:05,700 --> 00:16:07,220 Thank you. 338 00:16:07,220 --> 00:16:09,860 CHEERING 339 00:16:09,860 --> 00:16:14,020 As Anders mentioned there, it takes a huge amount of brainpower 340 00:16:14,020 --> 00:16:18,140 to compute those simulations - takes even machines days to do. 341 00:16:18,140 --> 00:16:20,700 But, of course, it's not just fun and frivolity that's on the table 342 00:16:20,700 --> 00:16:23,500 when you could get machines to do your maths for you, 343 00:16:23,500 --> 00:16:27,340 because working out how to get computers to crunch the numbers 344 00:16:27,340 --> 00:16:29,180 also saves lives. 345 00:16:29,180 --> 00:16:32,340 And something that has had a huge impact from using algorithms 346 00:16:32,340 --> 00:16:34,620 is organ-donor matching. 347 00:16:34,620 --> 00:16:36,980 Let me explain what I am talking about here. 348 00:16:36,980 --> 00:16:39,740 So let's imagine that you're a bit poorly 349 00:16:39,740 --> 00:16:41,900 and you needed a new kidney. 350 00:16:41,900 --> 00:16:46,780 Now, luckily, your friend here has very kindly and generously 351 00:16:46,780 --> 00:16:49,380 offered to donate to you one of hers. 352 00:16:49,380 --> 00:16:52,460 Almost all of us are born with two kidneys and you can live 353 00:16:52,460 --> 00:16:55,860 a very long and happy life with just one healthy kidney. 354 00:16:55,860 --> 00:17:00,260 Now, if you were both good tissue matches or blood matches for each 355 00:17:00,260 --> 00:17:03,620 other, then this would be absolutely fine, the transplant could go ahead. 356 00:17:03,620 --> 00:17:08,300 But if for some reason your blood or tissue type didn't match, 357 00:17:08,300 --> 00:17:10,660 then you would be in a position where you had one person 358 00:17:10,660 --> 00:17:12,220 who needed a kidney, 359 00:17:12,220 --> 00:17:14,660 one person who is willing to donate a kidney, 360 00:17:14,660 --> 00:17:16,780 but nothing that you could do about it. 361 00:17:16,780 --> 00:17:18,420 That was until very recently. 362 00:17:18,420 --> 00:17:21,700 All you could do was really wait around on the transplant list, 363 00:17:21,700 --> 00:17:24,020 hoping for another donation. 364 00:17:24,020 --> 00:17:27,860 That was until quite recently, when someone had a very bright idea. 365 00:17:27,860 --> 00:17:32,540 Because while you two might be stuck in that position, perhaps, 366 00:17:32,540 --> 00:17:35,380 over here, there are another two people 367 00:17:35,380 --> 00:17:37,420 who are in a similar position. 368 00:17:37,420 --> 00:17:39,860 So maybe you need a kidney. 369 00:17:39,860 --> 00:17:43,220 I mean, you have one that you are willing to donate, 370 00:17:43,220 --> 00:17:46,420 but that unfortunately doesn't match your loved one 371 00:17:46,420 --> 00:17:48,700 who you're trying to help. 372 00:17:48,700 --> 00:17:52,740 But what if your kidney, rather than matching your friend here, 373 00:17:52,740 --> 00:17:56,540 what if your kidney was a match for you? 374 00:17:56,540 --> 00:18:00,180 Now, that transplant could go ahead and that would be fine. 375 00:18:00,180 --> 00:18:02,620 But the question is, why would you want to give 376 00:18:02,620 --> 00:18:05,540 up your kidney to someone you've never met before, especially 377 00:18:05,540 --> 00:18:10,100 when your friend is still sick and in need of a transplant? 378 00:18:10,100 --> 00:18:14,180 But what if there was a third pair of people? 379 00:18:14,180 --> 00:18:16,380 What if we could close this loop 380 00:18:16,380 --> 00:18:18,980 with another group who are over here? 381 00:18:18,980 --> 00:18:23,620 So one more person and their loved one who is willing to give 382 00:18:23,620 --> 00:18:26,180 them one, you've got a little loop up here. Perfect. Thank you. 383 00:18:26,180 --> 00:18:27,540 I'll take that down. 384 00:18:27,540 --> 00:18:31,260 Now, in this situation, if this could happen, 385 00:18:31,260 --> 00:18:33,780 if this happened to be a match for you as well, 386 00:18:33,780 --> 00:18:35,340 then everyone's happy, right? 387 00:18:35,340 --> 00:18:39,940 Everyone who needs a transplant has got one and everyone has helped 388 00:18:39,940 --> 00:18:43,100 make sure that their loved ones have the operation that they need. 389 00:18:43,100 --> 00:18:46,940 Now, the only problem with this is that there are often hundreds 390 00:18:46,940 --> 00:18:49,500 of people in the country who will find themselves 391 00:18:49,500 --> 00:18:50,740 in this position. 392 00:18:50,740 --> 00:18:54,340 And finding these matches is incredibly difficult. 393 00:18:54,340 --> 00:18:56,380 The number of possibilities with hundreds of people 394 00:18:56,380 --> 00:18:57,900 is absolutely enormous. 395 00:18:57,900 --> 00:19:00,500 And you have to make sure that you create a closed loop every 396 00:19:00,500 --> 00:19:04,140 single time so that everybody ends up happy. 397 00:19:04,140 --> 00:19:06,700 But this is the kind of thing that computers 398 00:19:06,700 --> 00:19:09,340 are absolutely perfect for. 399 00:19:09,340 --> 00:19:12,900 And, in fact, this is exactly what a team at the University 400 00:19:12,900 --> 00:19:14,740 of Glasgow have done. 401 00:19:14,740 --> 00:19:17,660 So here is the output from that algorithm. 402 00:19:17,660 --> 00:19:21,020 This algorithm runs in seven seconds. 403 00:19:21,020 --> 00:19:23,660 And this is the network that it spits out. 404 00:19:23,660 --> 00:19:28,140 So each one of these dots is a real person, and each one of the lines 405 00:19:28,140 --> 00:19:30,340 is, effectively, a ribbon. 406 00:19:30,340 --> 00:19:34,420 And if we zoom in on this, you can see here just how many 407 00:19:34,420 --> 00:19:38,300 ribbons this algorithm is potentially considering. 408 00:19:38,300 --> 00:19:41,180 Now, clearly, there is no way that any human would be able to weigh 409 00:19:41,180 --> 00:19:43,980 all of this up within their own minds. 410 00:19:43,980 --> 00:19:47,020 But this algorithm, this isn't just something that's theoretical. 411 00:19:47,020 --> 00:19:51,260 This is something that actually does go on to save lives because, 412 00:19:51,260 --> 00:19:55,900 in fact, up there, holding on to the green ribbon is Yuki, 413 00:19:55,900 --> 00:19:58,980 who was part of a real-life chain. 414 00:19:58,980 --> 00:20:02,980 So, Yuki, if you want to join us on stage. Round of applause for Yuki. 415 00:20:02,980 --> 00:20:05,980 CHEERING 416 00:20:10,380 --> 00:20:12,340 Thank you for joining us, Yuki. 417 00:20:12,340 --> 00:20:14,340 So how long ago did you have your operation? 418 00:20:14,340 --> 00:20:16,500 I had it three years ago. 419 00:20:16,500 --> 00:20:19,740 And who was the loved one who donated a kidney on your behalf? 420 00:20:19,740 --> 00:20:22,940 It was my grandma from Japan. From Japan. 421 00:20:22,940 --> 00:20:25,620 I think we have a photograph, actually, here, of you guys 422 00:20:25,620 --> 00:20:27,060 outside Great Ormond Street. 423 00:20:27,060 --> 00:20:31,220 So your grandmother's kidney went to someone else. 424 00:20:31,220 --> 00:20:33,140 Where did your kidney come from? 425 00:20:33,140 --> 00:20:36,060 My kidney came from someone in Birmingham. 426 00:20:36,060 --> 00:20:38,620 And do you know how many people there were in your chain? 427 00:20:38,620 --> 00:20:42,140 I think three. So exactly the same as we have here. 428 00:20:42,140 --> 00:20:44,300 Yeah. It's a really incredible story. 429 00:20:44,300 --> 00:20:48,100 Yuki, thank you so much for joining us and telling your story. 430 00:20:48,100 --> 00:20:51,180 CHEERING 431 00:20:56,020 --> 00:20:58,940 Now, this work is a monumental breakthrough, it's something that's 432 00:20:58,940 --> 00:21:01,900 been a real game-changer for people within the UK. 433 00:21:01,900 --> 00:21:05,940 And the key thing that that algorithm can do, the thing that no 434 00:21:05,940 --> 00:21:07,620 human can do on their own, 435 00:21:07,620 --> 00:21:10,380 is consider this vast number of possibilities. 436 00:21:10,380 --> 00:21:15,180 And that is the superhuman power that you open up when you hand 437 00:21:15,180 --> 00:21:18,980 over your maths to a machine, because suddenly it can use 438 00:21:18,980 --> 00:21:22,180 all of its grunt to tailor things to the individual, to work out 439 00:21:22,180 --> 00:21:24,300 precisely what is right for you, 440 00:21:24,300 --> 00:21:25,740 which kidney you need, 441 00:21:25,740 --> 00:21:29,580 and in a much smaller, but no less profound way, 442 00:21:29,580 --> 00:21:32,980 which song or video you might like. 443 00:21:32,980 --> 00:21:36,500 Now YouTube videos, Netflix, Amazon, BBC iPlayer - all of them 444 00:21:36,500 --> 00:21:40,500 are crunching through vast amounts of information on their websites. 445 00:21:40,500 --> 00:21:44,140 But how do they make sense of the content that is uploaded? 446 00:21:44,140 --> 00:21:47,500 And what kind of process do they use to decide 447 00:21:47,500 --> 00:21:51,060 what you might like to see next? Well, OK, I'll tell you what. 448 00:21:51,060 --> 00:21:54,620 Let's find out here by making our own video. 449 00:21:54,620 --> 00:21:59,860 So I wonder, can I have a volunteer, someone who is happy to help me 450 00:21:59,860 --> 00:22:02,740 commentate on a little YouTube video? 451 00:22:02,740 --> 00:22:04,180 OK. Perfect. Yeah. Let's go there. 452 00:22:04,180 --> 00:22:06,660 Come on down. Round of applause as he comes to the stage. 453 00:22:06,660 --> 00:22:09,140 Now, what's your name? Anthony. 454 00:22:09,140 --> 00:22:10,500 Anthony. OK, perfect. Anthony. 455 00:22:10,500 --> 00:22:13,060 Right. So we're going to make... Have you got your phone? Yeah. 456 00:22:13,060 --> 00:22:15,580 OK. Perfect. Right. We're going to make a video, Anthony. 457 00:22:15,580 --> 00:22:18,140 And we want it to be seen by as many people as possible. 458 00:22:18,140 --> 00:22:21,620 Right? We want it to be picked up by the algorithm. 459 00:22:21,620 --> 00:22:24,220 So we're going to make a great YouTube video here. 460 00:22:24,220 --> 00:22:28,460 Now, thankfully, the world famous Royal Institution demo team are on 461 00:22:28,460 --> 00:22:30,940 hand to help us make our video as exciting as possible. 462 00:22:30,940 --> 00:22:33,980 So if you give me this and I'll do your video for you. 463 00:22:33,980 --> 00:22:36,420 If you take these, you can come and stand over here and help me 464 00:22:36,420 --> 00:22:37,860 commentate on this video. 465 00:22:37,860 --> 00:22:39,220 So I'll take the video. 466 00:22:39,220 --> 00:22:42,900 Here's Gemma. You start off whenever you want to, Anthony. 467 00:22:44,460 --> 00:22:45,780 There we go! 468 00:22:45,780 --> 00:22:47,300 What's up, folks? 469 00:22:47,300 --> 00:22:49,060 It's me, Anthony, again. 470 00:22:49,060 --> 00:22:50,860 Thanks for subscribing. 471 00:22:50,860 --> 00:22:52,660 This is my best... 472 00:22:52,660 --> 00:22:56,780 ..best friend Gemma doing some science. 473 00:22:56,780 --> 00:23:01,020 They have some extremely cold liquid nitrogen. 474 00:23:01,020 --> 00:23:05,340 Whoa! Look at all these balloon dogs! 475 00:23:05,340 --> 00:23:07,500 Where are they coming from? 476 00:23:07,500 --> 00:23:10,300 It's the balloon-dog challenge! 477 00:23:10,300 --> 00:23:15,660 Now, Gemma is going to add hot water to the liquid nitrogen and it 478 00:23:15,660 --> 00:23:17,700 will yeet everywhere! 479 00:23:17,700 --> 00:23:19,660 Let's give her a countdown. 480 00:23:19,660 --> 00:23:21,220 Nice and loud. 481 00:23:21,220 --> 00:23:24,940 ALL: Three, two, one... 482 00:23:24,940 --> 00:23:27,500 AUDIENCE WHOOPS 483 00:23:27,500 --> 00:23:32,500 Now, sure to get views, it's YouTuber Tom Scott! 484 00:23:32,500 --> 00:23:35,940 And behind the cloud, appears 485 00:23:35,940 --> 00:23:38,660 YouTube sensation Tom Scott. 486 00:23:38,660 --> 00:23:41,380 Anthony, thank you so much. There, and you've got your video there. 487 00:23:41,380 --> 00:23:45,020 Thank you very much indeed. Big round of applause if you can! 488 00:23:47,740 --> 00:23:50,780 Do you often appear in clouds of smoke? 489 00:23:50,780 --> 00:23:53,220 No, it's the first time I've been summoned like this. 490 00:23:53,220 --> 00:23:55,700 It's wonderful. Now, I should tell you, for any of you who don't 491 00:23:55,700 --> 00:23:59,740 recognise Tom, Tom is one of Britain's foremost YouTubers. 492 00:23:59,740 --> 00:24:03,380 If there is one person that we can ask about how to make 493 00:24:03,380 --> 00:24:05,540 sure a video gets seen by as many people as possible, 494 00:24:05,540 --> 00:24:07,620 I mean, you are that person. Yeah. 495 00:24:07,620 --> 00:24:11,100 I wish there was some magical way to guarantee it but, I mean, 496 00:24:11,100 --> 00:24:12,540 I've done some pretty cool stuff. 497 00:24:12,540 --> 00:24:17,980 I've been in Zero G, but that video didn't do well compared to, well, 498 00:24:17,980 --> 00:24:21,460 a two-minute shot of me continuously looking at some toasters 499 00:24:21,460 --> 00:24:23,820 and talking about how long it's going to take them to pop up. 500 00:24:23,820 --> 00:24:27,860 That video did better than me flying about in zero G. 501 00:24:27,860 --> 00:24:31,780 But what did better than both of those, to achieve, like, 25 million 502 00:24:31,780 --> 00:24:35,260 views, was me attaching some garlic bread to a helium 503 00:24:35,260 --> 00:24:39,100 balloon and sending it to the edge of space, then bringing it back 504 00:24:39,100 --> 00:24:41,980 down and eating it. Now... 25 million. 25 million. 505 00:24:41,980 --> 00:24:44,860 I kind of figured that one was going to do well. 506 00:24:44,860 --> 00:24:48,460 I didn't think it was going to do THAT well. I have never, ever been 507 00:24:48,460 --> 00:24:52,460 able to find anything that will get these recommendation engines, 508 00:24:52,460 --> 00:24:55,980 these algorithms, to specifically go for a particular video. 509 00:24:55,980 --> 00:24:58,620 Because it is an algorithm that's deciding what gets promoted 510 00:24:58,620 --> 00:25:02,020 on different people's YouTubes, right? Yes. Originally, 511 00:25:02,020 --> 00:25:05,260 years and years ago, it was all down to the title and the thumbnail. 512 00:25:05,260 --> 00:25:06,620 CRACKLING 513 00:25:06,620 --> 00:25:09,180 Is that OK? That's crackling a lot. I'm slightly nervous! 514 00:25:09,180 --> 00:25:11,420 I mean, this is going to make quite a YouTube video 515 00:25:11,420 --> 00:25:14,460 if it explodes on us now, Tom! I'm going to keep going. 516 00:25:14,460 --> 00:25:18,220 Keep going. Originally, when I started 2006, it was all about 517 00:25:18,220 --> 00:25:20,820 the title and the thumbnail and that was it. 518 00:25:20,820 --> 00:25:23,140 So you could trick people into clicking on something 519 00:25:23,140 --> 00:25:25,220 with a lie in the thumbnail, a lie in the title, 520 00:25:25,220 --> 00:25:27,980 and that's still true now, but it doesn't work as well, 521 00:25:27,980 --> 00:25:30,380 because after that, they said it was more about how long people 522 00:25:30,380 --> 00:25:32,060 watched a video for. 523 00:25:32,060 --> 00:25:35,060 So at that point, people put, you know, a little explosion 524 00:25:35,060 --> 00:25:36,380 at the end of a 20-minute video. 525 00:25:36,380 --> 00:25:38,940 So you'd have to watch all the way through. Those long introductions 526 00:25:38,940 --> 00:25:40,500 that were very fashionable for a while. 527 00:25:40,500 --> 00:25:44,060 Because they kept people watching. So now YouTube doesn't give any 528 00:25:44,060 --> 00:25:48,020 advice at all on how to make something please the algorithm. 529 00:25:48,020 --> 00:25:50,620 And as a YouTube creator, all of us just call it 530 00:25:50,620 --> 00:25:54,060 "the algorithm". This one monolithic thing. 531 00:25:54,060 --> 00:25:57,900 They say that if people watch it and people like it, 532 00:25:57,900 --> 00:26:00,820 then the algorithm will also recommend it to other people. 533 00:26:00,820 --> 00:26:04,100 But they will never, ever say what the reasons are, 534 00:26:04,100 --> 00:26:07,460 because the minute they do that, everyone will say, "Oh, yeah, well, 535 00:26:07,460 --> 00:26:08,700 "I'll start doing that." 536 00:26:08,700 --> 00:26:12,100 And suddenly 100,000, 200,000 people are all doing that and 537 00:26:12,100 --> 00:26:15,940 no-one's watching it any more. Yeah. There are 500 hours of video 538 00:26:15,940 --> 00:26:19,260 uploaded to YouTube every minute. Not watched - like, uploaded. 539 00:26:19,260 --> 00:26:22,500 So I'm sure it's a great video on your phone. 540 00:26:22,500 --> 00:26:26,580 Statistically, it's probably not going to go big, but it might, cos 541 00:26:26,580 --> 00:26:29,660 the more times you can roll that dice, the better your chances are. 542 00:26:29,660 --> 00:26:32,860 I'm going to go create a garlic bread video of my own now, I think. 543 00:26:32,860 --> 00:26:34,300 Good luck. 544 00:26:34,300 --> 00:26:36,660 Apparently it does work! 545 00:26:36,660 --> 00:26:39,780 Tom Scott, thank you very much indeed. Thank you. Thank you. 546 00:26:39,780 --> 00:26:43,020 CHEERING 547 00:26:43,020 --> 00:26:47,260 Now, as Tom hinted there, there is actually something slightly 548 00:26:47,260 --> 00:26:49,580 different going on here to the fireworks 549 00:26:49,580 --> 00:26:52,900 or the organ-donor algorithms that we saw earlier. 550 00:26:52,900 --> 00:26:56,620 That YouTube algorithm, it isn't just crunching through a straight 551 00:26:56,620 --> 00:26:58,260 list of instructions any more. 552 00:26:58,260 --> 00:27:00,420 It's actually doing something a little bit different. 553 00:27:00,420 --> 00:27:05,100 I want to explain this to you using a cup of tea and a robot. 554 00:27:05,100 --> 00:27:06,180 Oh, OK. 555 00:27:06,180 --> 00:27:10,580 Maybe not an actual robot, but definitely the next-best thing. 556 00:27:10,580 --> 00:27:14,420 Please join me in welcoming robot Matt Parker. 557 00:27:14,420 --> 00:27:16,540 CHEERING 558 00:27:21,620 --> 00:27:24,460 That's a... Thank you! 559 00:27:26,660 --> 00:27:29,340 That's a great outfit, Matt. Thank you. I've made it myself. 560 00:27:29,340 --> 00:27:30,820 Yeah. Sorry... 561 00:27:30,820 --> 00:27:32,980 ROBOTIC VOICE: I compiled it myself. 562 00:27:32,980 --> 00:27:36,460 OK. Now, robot Matt, like all robots, 563 00:27:36,460 --> 00:27:38,260 he isn't particularly bright. 564 00:27:38,260 --> 00:27:41,940 Oh-oh. Offence registered. 565 00:27:41,940 --> 00:27:44,740 They also take things incredibly literally, 566 00:27:44,740 --> 00:27:48,420 as we are about to discover, because we are going to, together, 567 00:27:48,420 --> 00:27:53,940 try and instruct robot Matt Parker with how to make a cup of tea. 568 00:27:53,940 --> 00:27:55,140 OK, so here we go. 569 00:27:55,140 --> 00:27:57,860 What's the first thing to do if you make a cup of tea? 570 00:27:57,860 --> 00:28:00,220 You heat up the water in the kettle. 571 00:28:00,220 --> 00:28:01,460 Heat up the water in the kettle. 572 00:28:01,460 --> 00:28:03,260 Perfect. Sounds sensible? 573 00:28:03,260 --> 00:28:05,500 There's a kettle. OK, next step. Next step. 574 00:28:05,500 --> 00:28:08,940 You take out the teabags. Take out the teabags. 575 00:28:10,980 --> 00:28:13,420 Ha-ha! That's good. Next step? 576 00:28:13,420 --> 00:28:15,220 Er, get a mug. 577 00:28:15,220 --> 00:28:16,260 Get a mug. 578 00:28:18,300 --> 00:28:21,140 Oh, it's a teeny, tiny one. 579 00:28:21,140 --> 00:28:22,380 What should we do next? 580 00:28:22,380 --> 00:28:24,780 Put the teabag in the mug. 581 00:28:24,780 --> 00:28:26,820 LAUGHTER 582 00:28:30,780 --> 00:28:32,420 Is that the right mug, there? 583 00:28:32,420 --> 00:28:34,420 Well, let's get another. 584 00:28:34,420 --> 00:28:36,660 Put the boiling water in. 585 00:28:36,660 --> 00:28:38,540 LAUGHTER 586 00:28:38,540 --> 00:28:41,940 He takes things very literally! 587 00:28:41,940 --> 00:28:42,980 HANNAH CHUCKLES 588 00:28:42,980 --> 00:28:45,700 Heat registering! 589 00:28:47,220 --> 00:28:50,980 Get a normal-sized teabag and cup. 590 00:28:52,420 --> 00:28:54,340 A bigger cup. 591 00:28:58,940 --> 00:29:01,340 The normal teabag in the mug. 592 00:29:01,340 --> 00:29:03,500 LAUGHTER 593 00:29:03,500 --> 00:29:06,980 I'm going to come round there. Hang on. One second. 594 00:29:06,980 --> 00:29:08,660 Hold on a second. I'm going to come up here. 595 00:29:08,660 --> 00:29:10,340 Here we go. 596 00:29:10,340 --> 00:29:11,700 What's next? What's next? 597 00:29:11,700 --> 00:29:14,100 Stop pouring the water. 598 00:29:16,100 --> 00:29:18,060 Put the teabag in the mug. 599 00:29:18,060 --> 00:29:20,780 Pour the water into the mug. 600 00:29:20,780 --> 00:29:23,300 AUDIENCE MEMBERS CHORTLE 601 00:29:26,940 --> 00:29:28,940 LAUGHTER 602 00:29:28,940 --> 00:29:31,580 Stop pouring the water. 603 00:29:31,580 --> 00:29:34,380 Add milk and sugar to taste. 604 00:29:34,380 --> 00:29:36,340 HANNAH CHUCKLES 605 00:29:37,940 --> 00:29:40,300 LAUGHTER 606 00:29:44,420 --> 00:29:48,380 Pour the milk from the first mug into the mug with the tea. 607 00:29:52,700 --> 00:29:54,260 OK. I think we've got one more. 608 00:29:54,260 --> 00:29:57,420 Stop pouring the milk into the cup. 609 00:29:57,420 --> 00:29:59,660 And the last step? 610 00:29:59,660 --> 00:30:00,940 Drink it. 611 00:30:00,940 --> 00:30:03,100 LAUGHTER 612 00:30:04,860 --> 00:30:07,180 I made this for you. 613 00:30:08,580 --> 00:30:10,020 Cheers, everyone. 614 00:30:10,020 --> 00:30:11,460 Oh, that's... 615 00:30:11,460 --> 00:30:13,460 LAUGHTER AND CHEERING 616 00:30:13,460 --> 00:30:16,380 Thank you very much! 617 00:30:16,380 --> 00:30:19,540 OK. So I think it's quite obvious that our list of instructions there 618 00:30:19,540 --> 00:30:21,620 needs we need to be quite long and that is only for making 619 00:30:21,620 --> 00:30:23,620 a cup of tea - something very easy. 620 00:30:23,620 --> 00:30:27,140 So imagine how much harder it would be to write a list 621 00:30:27,140 --> 00:30:30,100 of instructions for a very literal computer. 622 00:30:30,100 --> 00:30:34,540 OK. How would you explain to a computer how to recognise a picture 623 00:30:34,540 --> 00:30:35,980 of a dog? 624 00:30:35,980 --> 00:30:37,620 What do dogs look like? Shout out. 625 00:30:37,620 --> 00:30:39,500 What are the important things about what dogs look 626 00:30:39,500 --> 00:30:41,220 like that you would need? 627 00:30:41,220 --> 00:30:42,460 AUDIENCE OFFERS SUGGESTIONS 628 00:30:42,460 --> 00:30:43,980 Fluffy, fur, yeah. 629 00:30:43,980 --> 00:30:45,500 AUDIENCE SHOUTS 630 00:30:45,500 --> 00:30:46,980 Four legs. 631 00:30:46,980 --> 00:30:49,780 Er, tail? Yeah. OK. 632 00:30:49,780 --> 00:30:52,460 All right, all right. So we got... We got... 633 00:30:52,460 --> 00:30:56,700 We got fluffy or furry, tail, and four legs. Great. 634 00:30:56,700 --> 00:30:59,580 OK. But what about this? Uhhh... 635 00:30:59,580 --> 00:31:01,060 Slight problem, there. 636 00:31:01,060 --> 00:31:03,020 OK. All right. What else? What else? What about its face? 637 00:31:03,020 --> 00:31:05,540 What's important to recognise about its face? 638 00:31:05,540 --> 00:31:07,100 AUDIENCE OFFERS SUGGESTIONS 639 00:31:07,100 --> 00:31:09,140 Pointy ears. 640 00:31:09,140 --> 00:31:12,380 Pointy snout. OK. OK. OK. Here we go. 641 00:31:12,380 --> 00:31:14,620 All right. So "pointy snout", someone said. 642 00:31:14,620 --> 00:31:15,980 How about this, though? 643 00:31:15,980 --> 00:31:18,060 Ahhh. OK. 644 00:31:18,060 --> 00:31:22,620 Even if you could work out, you know, maybe take into account 645 00:31:22,620 --> 00:31:24,540 the colour, all of that stuff, 646 00:31:24,540 --> 00:31:26,740 didn't you say that a dog had to have four legs? 647 00:31:26,740 --> 00:31:28,700 Because what about this guy? 648 00:31:28,700 --> 00:31:30,820 Still definitely a dog. 649 00:31:30,820 --> 00:31:34,100 Still definitely a dog. But it just goes to show just how hard it is 650 00:31:34,100 --> 00:31:37,700 if you are trying to write a long list of instructions, just how hard 651 00:31:37,700 --> 00:31:41,180 it is to pin down exactly what a dog looks like. 652 00:31:41,180 --> 00:31:43,060 Now, humans, we're amazing at this. 653 00:31:43,060 --> 00:31:46,460 We can do this without even thinking, but explaining how we do 654 00:31:46,460 --> 00:31:49,340 it, especially to a pretty dumb computer that takes things 655 00:31:49,340 --> 00:31:53,180 very literally, ends up being quite a lot harder than you might imagine. 656 00:31:53,180 --> 00:31:56,180 But there is another way around this to explain. 657 00:31:56,180 --> 00:31:59,500 Please join me in welcoming computer scientist extraordinaire 658 00:31:59,500 --> 00:32:01,500 Anne-Marie Imafidon. 659 00:32:01,500 --> 00:32:05,020 CHEERING AND APPLAUSE 660 00:32:07,100 --> 00:32:09,300 There is another way to all of this, Anne-Marie? 661 00:32:09,300 --> 00:32:12,020 There is another way. We need a different type of program 662 00:32:12,020 --> 00:32:15,700 that isn't as specific. Isn't a list of straightforward "do this, 663 00:32:15,700 --> 00:32:17,260 "do this, do this". Exactly. 664 00:32:17,260 --> 00:32:20,780 We need a type of program that has an end point or a goal. 665 00:32:20,780 --> 00:32:23,580 And we let the computer or the machine figure out 666 00:32:23,580 --> 00:32:26,580 its own rules to get to that end goal. Without giving 667 00:32:26,580 --> 00:32:28,540 it straightforward instructions? Without giving 668 00:32:28,540 --> 00:32:31,380 it any instructions other than that's where we're going, 669 00:32:31,380 --> 00:32:34,060 and maybe when it's got it right. OK. 670 00:32:34,060 --> 00:32:35,900 Anne-Marie, this sounds like magic. 671 00:32:35,900 --> 00:32:38,500 Oh, it is computer science more than it's magic! 672 00:32:38,500 --> 00:32:40,380 It's a lot like training an animal. 673 00:32:40,380 --> 00:32:43,780 When you get a dog, you're training it to maybe do a particular trick. 674 00:32:43,780 --> 00:32:46,620 You don't say, "Move this muscle and look that way." 675 00:32:46,620 --> 00:32:49,060 You kind of just give it a treat when it does the right thing. 676 00:32:49,060 --> 00:32:51,260 And you've got a clear idea in your mind of what it is you want 677 00:32:51,260 --> 00:32:52,980 them to do. Exactly. Whether it's sitting 678 00:32:52,980 --> 00:32:55,180 or whether it's anything else. OK. All right. 679 00:32:55,180 --> 00:32:58,580 Well, on that point, then, of training animals, what's this kind 680 00:32:58,580 --> 00:33:01,100 of algorithm called, incidentally, when it's in computer science? 681 00:33:01,100 --> 00:33:03,460 So we call it a reinforcement training algorithm. 682 00:33:03,460 --> 00:33:06,540 And it's one that learns on its own, so it's actually part of machine 683 00:33:06,540 --> 00:33:09,780 learning. OK. All right. So reinforcement learning. 684 00:33:09,780 --> 00:33:12,780 Let's see how this works, because we're going to try some 685 00:33:12,780 --> 00:33:14,460 reinforcement learning on one of you. 686 00:33:14,460 --> 00:33:17,460 So who would like to be our volunteer? 687 00:33:17,460 --> 00:33:20,900 You had your hand up, I think, first. So if you want to come down. 688 00:33:20,900 --> 00:33:23,180 CHEERING 689 00:33:25,820 --> 00:33:28,500 What's your name? Emily. Emily. OK. Perfect. 690 00:33:28,500 --> 00:33:30,220 Emily, if you want to come over here. OK. 691 00:33:30,220 --> 00:33:32,540 So, Emily, what we're going to do, then, is we're going to play 692 00:33:32,540 --> 00:33:34,940 a little game with you. If you just come to the side while they bring 693 00:33:34,940 --> 00:33:38,140 all these things on. What we're going to do is we're going to show 694 00:33:38,140 --> 00:33:40,180 you four pots. You can see all the different rounds 695 00:33:40,180 --> 00:33:43,300 that we've got there. These pots are coloured and they've got sums 696 00:33:43,300 --> 00:33:46,820 on them on the side, and so on. Now, Anne-Marie and I, we've got in 697 00:33:46,820 --> 00:33:49,540 our minds, we've got a very clear idea of which pot 698 00:33:49,540 --> 00:33:50,780 we want you to pick. 699 00:33:50,780 --> 00:33:53,820 So, in each round, all you do is you just touch one pot. 700 00:33:53,820 --> 00:33:56,380 OK. If you are correct, we're going to give you a treat, 701 00:33:56,380 --> 00:33:58,740 we're going to give you a little reward, and I think they might just 702 00:33:58,740 --> 00:34:01,420 be in here. We're going to give you a little reward of a humbug. 703 00:34:01,420 --> 00:34:04,300 And if you're wrong, we're going to move on to the next round. OK? 704 00:34:04,300 --> 00:34:07,220 If you want to just stand in there for me. Now, just to prove 705 00:34:07,220 --> 00:34:10,260 that this really is something that is not specific to humans, 706 00:34:10,260 --> 00:34:13,660 you are going to play this game against a competitor, 707 00:34:13,660 --> 00:34:15,620 by which I mean a pigeon. 708 00:34:15,620 --> 00:34:18,300 If we can welcome onto the stage 709 00:34:18,300 --> 00:34:22,020 Alfie and his trainer, Lloyd. Round of applause. 710 00:34:27,380 --> 00:34:29,860 Now, I have to tell you, Emily, 711 00:34:29,860 --> 00:34:32,660 Alfie has had a bit of a head start on this. Alfie has been trained. 712 00:34:32,660 --> 00:34:34,060 Look how sweet Alfie is! 713 00:34:34,060 --> 00:34:35,740 Very beautiful bird. 714 00:34:35,740 --> 00:34:38,020 You happy? You understand the game? All right, let's go. 715 00:34:38,020 --> 00:34:39,220 Let's go for round one. 716 00:34:39,220 --> 00:34:40,940 So just touch one pot. 717 00:34:43,740 --> 00:34:45,100 Incorrect. 718 00:34:45,100 --> 00:34:47,980 Ah, Alfie got it right, though. He did. He did! 719 00:34:47,980 --> 00:34:50,300 Alfie, one. Human, zero. 720 00:34:50,300 --> 00:34:51,580 LAUGHTER 721 00:34:51,580 --> 00:34:53,940 OK. Here we go. 722 00:34:53,940 --> 00:34:54,940 Round two. 723 00:34:57,340 --> 00:34:58,900 Incorrect. 724 00:34:58,900 --> 00:35:02,540 Ohhh! Alfie, two. Human, zero. 725 00:35:04,180 --> 00:35:05,980 And here we go. 726 00:35:07,820 --> 00:35:09,300 Next one. 727 00:35:11,420 --> 00:35:13,220 Correct. OK! 728 00:35:14,460 --> 00:35:16,100 Next one. 729 00:35:18,060 --> 00:35:20,340 Ohh! Incorrect. 730 00:35:20,340 --> 00:35:22,060 Alfie's got it correct again. 731 00:35:22,060 --> 00:35:24,660 OK. Let's go final round. 732 00:35:24,660 --> 00:35:28,340 Now. OK. The equations on this side are getting harder and harder 733 00:35:28,340 --> 00:35:31,620 and harder, almost like 734 00:35:31,620 --> 00:35:33,740 there's no possible way that you could solve them. 735 00:35:35,940 --> 00:35:37,660 Ohh! Incorrect! 736 00:35:37,660 --> 00:35:39,020 That is tough, that is tough! 737 00:35:39,020 --> 00:35:41,700 OK. Tell us, how did you find it? Tell us what you thought. 738 00:35:41,700 --> 00:35:43,540 It was quite hard. 739 00:35:43,540 --> 00:35:45,940 I just basically picked random ones. 740 00:35:45,940 --> 00:35:48,060 There wasn't really a method. 741 00:35:48,060 --> 00:35:50,180 What were you thinking when you got it right? 742 00:35:50,180 --> 00:35:54,300 I didn't really have a thing. I just sort of picked one, apart from 743 00:35:54,300 --> 00:35:57,140 the one where the yellow pot fell on the floor. 744 00:35:57,140 --> 00:35:59,740 Oh, did you see the yellow pot? That was a big clue. 745 00:35:59,740 --> 00:36:01,020 In fact, actually. 746 00:36:01,020 --> 00:36:03,460 Which pot do you think that Alfie was picking every time? 747 00:36:03,460 --> 00:36:06,100 Shall we tell her? The yellow one? The yellow one. 748 00:36:06,100 --> 00:36:08,540 Exactly right. Exactly right. 749 00:36:08,540 --> 00:36:10,500 APPLAUSE 750 00:36:10,500 --> 00:36:13,420 Now, I know that was quite tough, that was quite tough. 751 00:36:13,420 --> 00:36:15,860 But, Anne-Marie, tell us, how similar is this to the things we see 752 00:36:15,860 --> 00:36:17,060 in computer science? 753 00:36:17,060 --> 00:36:19,780 So this is very similar to what we see in computer science. 754 00:36:19,780 --> 00:36:24,260 But with a computer, it's able to do this thousands, millions, billions 755 00:36:24,260 --> 00:36:25,660 of times in a second, 756 00:36:25,660 --> 00:36:30,580 and so can then pick up that pattern quicker than we can as humans. 757 00:36:30,580 --> 00:36:34,140 And often it can see patterns that we humans can't see and can make 758 00:36:34,140 --> 00:36:35,860 those kind of connections, as well. 759 00:36:35,860 --> 00:36:39,500 And that is a really important point because you only had a few rounds, 760 00:36:39,500 --> 00:36:42,780 but if you were a computer, you'd have had 10,000 in that time. 761 00:36:42,780 --> 00:36:45,100 So there you go, have another treat and a big round of applause. 762 00:36:45,100 --> 00:36:47,020 Thank you! Thank you very much, Emily. Thank you. 763 00:36:47,020 --> 00:36:48,420 CHEERING 764 00:36:50,660 --> 00:36:53,660 Now, the way that Alfie was trained there was just using exactly 765 00:36:53,660 --> 00:36:56,380 that same technique. Alfie just had a lot more goes at it. 766 00:36:56,380 --> 00:37:00,940 But as Anne-Marie said there, if an animal can learn in this way 767 00:37:00,940 --> 00:37:04,300 with very simple rewards, never hearing full instructions, 768 00:37:04,300 --> 00:37:09,020 then so can a machine. Now, that might seem like a bit of leap. 769 00:37:09,020 --> 00:37:12,500 How do computers learn, especially when, as we found out 770 00:37:12,500 --> 00:37:16,020 with that cup of tea, machines are incredibly dumb? 771 00:37:16,020 --> 00:37:18,020 Well, it is possible. 772 00:37:18,020 --> 00:37:21,740 And perhaps unsurprisingly, given that it's in this programme, it is 773 00:37:21,740 --> 00:37:26,820 totally mathematical because it turns out you CAN teach an inanimate 774 00:37:26,820 --> 00:37:28,780 object to learn. 775 00:37:28,780 --> 00:37:31,940 And we have been doing exactly that for the last week. 776 00:37:31,940 --> 00:37:35,380 Please welcome Matt Parker and the Menace Machine. 777 00:37:37,700 --> 00:37:41,380 Have to be honest, Matt. Menace Machine looks quite a lot like a 778 00:37:41,380 --> 00:37:43,540 whole heap of matchboxes. 779 00:37:43,540 --> 00:37:47,060 That's because Menace IS just a heap of matchboxes! 780 00:37:47,060 --> 00:37:50,260 But we've taught these matchboxes to play noughts and crosses. 781 00:37:50,260 --> 00:37:52,660 OK, OK. Talk me through it. How on Earth can matchboxes play 782 00:37:52,660 --> 00:37:55,380 noughts and crosses? Well, the great thing about noughts and crosses is 783 00:37:55,380 --> 00:37:58,140 there aren't that many possible games. There's only nine squares, 784 00:37:58,140 --> 00:37:59,820 it's a nought or a cross. 785 00:37:59,820 --> 00:38:03,420 And on the front of these boxes, every single one is a different 786 00:38:03,420 --> 00:38:05,140 state that the game could be in. 787 00:38:05,140 --> 00:38:09,500 In fact, this is every possible position of every possible game 788 00:38:09,500 --> 00:38:11,420 that Menace could possibly face. 789 00:38:11,420 --> 00:38:12,740 We've got all the first moves, 790 00:38:12,740 --> 00:38:16,220 second moves, the third moves are in blue and the final, fourth 791 00:38:16,220 --> 00:38:18,140 moves are over here in pink. 792 00:38:18,140 --> 00:38:21,220 And inside each box, we've put some coloured beads. 793 00:38:21,220 --> 00:38:24,780 So here's one that we haven't used yet. And you can see there's a real 794 00:38:24,780 --> 00:38:27,100 mixture. There's, like, nine different-coloured beads 795 00:38:27,100 --> 00:38:29,460 and each colour of bead corresponds to a move Menace can make. 796 00:38:29,460 --> 00:38:30,580 Because, of course, 797 00:38:30,580 --> 00:38:32,980 there's only ever nine possible places that you could play for 798 00:38:32,980 --> 00:38:35,420 noughts and crosses. Exactly. 799 00:38:35,420 --> 00:38:38,380 OK, so how do you actually train it, though? Well, initially, there's 800 00:38:38,380 --> 00:38:40,660 just a random collection of beads in every single one. 801 00:38:40,660 --> 00:38:44,060 So actually it's making every possible move equally likely. 802 00:38:44,060 --> 00:38:46,180 And initially, Menace is terrible. 803 00:38:46,180 --> 00:38:49,900 But the reinforcement learning is we track each game, and if Menace 804 00:38:49,900 --> 00:38:53,580 wins, we give it more beads that are the same colour 805 00:38:53,580 --> 00:38:54,940 into the same boxes. 806 00:38:54,940 --> 00:38:57,540 If it loses, we confiscate those beads, which makes 807 00:38:57,540 --> 00:39:00,900 it less likely to do that move again. And to train it, 808 00:39:00,900 --> 00:39:03,420 we've had it play hundreds of games. 809 00:39:03,420 --> 00:39:06,180 And thank you so much, everyone who, before the show tonight, 810 00:39:06,180 --> 00:39:07,500 played against Menace. 811 00:39:07,500 --> 00:39:10,100 Each time we either reward it or we punish it. 812 00:39:10,100 --> 00:39:12,540 And with each game, it gets a little bit better. 813 00:39:12,540 --> 00:39:15,260 And how good is it? Do you think it can beat a human? 814 00:39:15,260 --> 00:39:19,860 I think it will probably not lose to a human. 815 00:39:19,860 --> 00:39:22,820 All right. Well, let's give it a go. 816 00:39:22,820 --> 00:39:26,100 Who would like to come and play with Menace? 817 00:39:26,100 --> 00:39:29,740 Let's get you, there, yeah. Round of applause as she comes down. 818 00:39:31,220 --> 00:39:33,900 So, what's your name? Ali. Ali. OK, perfect. 819 00:39:33,900 --> 00:39:37,100 Ali. Now, Ali, you've got the advantage of being a sentient being, 820 00:39:37,100 --> 00:39:40,020 so I think it's only fair that we let the matchboxes go first. 821 00:39:40,020 --> 00:39:41,540 OK, so Menace is going to go first. 822 00:39:41,540 --> 00:39:44,060 And this is the blank box, where there's nothing 823 00:39:44,060 --> 00:39:45,780 on it because we haven't played at all yet. 824 00:39:45,780 --> 00:39:48,140 I'm going to give it a shake and then I'm going to pull a bit 825 00:39:48,140 --> 00:39:50,340 out at random. So while Menace is the brain, 826 00:39:50,340 --> 00:39:52,580 I've got to do the actual moving around. OK. 827 00:39:52,580 --> 00:39:56,020 So the first bead is green, 828 00:39:56,020 --> 00:39:58,180 which is not surprising because you look in the box - 829 00:39:58,180 --> 00:39:59,860 they're all green! 830 00:39:59,860 --> 00:40:03,100 There's one purple one, because Menace 831 00:40:03,100 --> 00:40:05,220 has learnt very quickly 832 00:40:05,220 --> 00:40:09,780 if it goes in the middle to start with, it's way more likely to win. 833 00:40:09,780 --> 00:40:11,780 Now, there you are, Ali. Where would you like to go? 834 00:40:11,780 --> 00:40:13,860 Put a cross wherever you wish. 835 00:40:16,540 --> 00:40:18,500 Oh, a corner. Bold move. 836 00:40:18,500 --> 00:40:22,260 OK. I've now got to find, in the group of second-move boxes, 837 00:40:22,260 --> 00:40:26,860 this game. OK, there it is. 838 00:40:26,860 --> 00:40:29,940 But if you have a look, Ali, it's the same as this game, 839 00:40:29,940 --> 00:40:31,460 but it's the mirror image. 840 00:40:31,460 --> 00:40:34,180 And that's because we didn't want to use twice as many boxes. 841 00:40:34,180 --> 00:40:35,700 It's not good for the environment. 842 00:40:35,700 --> 00:40:38,220 So if you don't mind me turning the game over, 843 00:40:38,220 --> 00:40:40,220 are you happy that's still the same game? Uh-huh. 844 00:40:40,220 --> 00:40:44,380 But now, it exactly matches what's on the box. So I can give it a shake 845 00:40:44,380 --> 00:40:46,420 and Menace is going to go... 846 00:40:47,820 --> 00:40:50,460 Make sure it's properly random. Yellow. 847 00:40:50,460 --> 00:40:52,900 They're all yellow. OK. So, it's learnt... 848 00:40:52,900 --> 00:40:55,180 I mean, there are other moves it could make, but it happens 849 00:40:55,180 --> 00:40:58,100 to have learnt that there is a good second move. 850 00:40:58,100 --> 00:40:59,420 OK, your go. 851 00:41:02,300 --> 00:41:04,260 Predictable. OK. 852 00:41:04,260 --> 00:41:06,060 You're not letting it win that easily. 853 00:41:06,060 --> 00:41:07,780 That's fine. OK. 854 00:41:07,780 --> 00:41:10,060 Third move. Where are we? Where are we? 855 00:41:10,060 --> 00:41:11,380 There! OK. 856 00:41:11,380 --> 00:41:15,060 So you can see that one there, that matches this state. 857 00:41:15,060 --> 00:41:17,100 Here we go. Give it a shake. 858 00:41:19,100 --> 00:41:20,900 OK. Oh, white. OK. 859 00:41:20,900 --> 00:41:23,900 So the next move is going to be white. 860 00:41:23,900 --> 00:41:25,820 That's... Oh, that's not bad. 861 00:41:25,820 --> 00:41:27,140 There you are. 862 00:41:31,540 --> 00:41:34,940 You're just not letting Menace have any luck, are you? OK. 863 00:41:34,940 --> 00:41:39,820 That's fine. So now I've got to find this in here. 864 00:41:39,820 --> 00:41:41,460 Bear with me. 865 00:41:41,460 --> 00:41:46,140 This is exactly as much fun as real noughts and cross. 866 00:41:46,140 --> 00:41:47,540 HANNAH CHUCKLES 867 00:41:47,540 --> 00:41:51,020 Ah, OK. I think that's it. 868 00:41:51,020 --> 00:41:53,220 Let's have a look. That's... 869 00:41:53,220 --> 00:41:55,740 Yes. OK. Right. 870 00:41:55,740 --> 00:41:57,740 Menace is going to go.... 871 00:41:57,740 --> 00:41:59,660 Oh, there's not many beads left in this one. 872 00:41:59,660 --> 00:42:00,940 Oh, orange. OK. 873 00:42:00,940 --> 00:42:02,980 So I'm going to put an orange one there. 874 00:42:02,980 --> 00:42:05,020 Now, that move. 875 00:42:06,460 --> 00:42:09,180 And now, for the exciting conclusion... 876 00:42:09,180 --> 00:42:11,300 LAUGHTER 877 00:42:13,300 --> 00:42:16,580 Excellent. And now we go over... It's a draw. 878 00:42:16,580 --> 00:42:19,140 And the problem with noughts and crosses is if everyone's playing 879 00:42:19,140 --> 00:42:21,460 perfectly well, it always ends in a draw. 880 00:42:21,460 --> 00:42:26,340 So in this case, I'm going to give Menace one extra of those beads 881 00:42:26,340 --> 00:42:29,260 to say that was fine, but I'm not going to give it as many 882 00:42:29,260 --> 00:42:31,180 as if it had won. Amazing. 883 00:42:31,180 --> 00:42:34,060 Well, there we go. You managed to draw with a bunch of cardboard 884 00:42:34,060 --> 00:42:37,700 boxes. Very impressive! A big round of applause, if you can. Thank you. 885 00:42:40,380 --> 00:42:43,420 The point here is that you don't need to explain the rules 886 00:42:43,420 --> 00:42:46,460 or the instructions. The machine - in that case, the matchboxes - 887 00:42:46,460 --> 00:42:48,780 learns from getting rewards 888 00:42:48,780 --> 00:42:52,260 when it does the right thing and it turns out that you can use this for 889 00:42:52,260 --> 00:42:55,100 all kinds of things, so you can teach matchboxes how to play noughts 890 00:42:55,100 --> 00:42:56,380 and crosses. 891 00:42:56,380 --> 00:43:01,060 And going back to that earlier challenge of recognising dogs - 892 00:43:01,060 --> 00:43:04,020 turns out you can teach machines to do it, too. 893 00:43:04,020 --> 00:43:06,860 Now, for this, I would like a volunteer 894 00:43:06,860 --> 00:43:08,500 to help me demonstrate this. 895 00:43:08,500 --> 00:43:10,860 At the back there. Yeah. 896 00:43:10,860 --> 00:43:13,380 With the red jumper. Round of applause as she comes to the stage. 897 00:43:13,380 --> 00:43:14,860 Thank you. 898 00:43:17,340 --> 00:43:19,420 What's your name? Felicity. 899 00:43:19,420 --> 00:43:22,020 OK, Felicity, right. So let's have a little look over here. 900 00:43:22,020 --> 00:43:26,260 I'm going to demonstrate to you how a machine can learn 901 00:43:26,260 --> 00:43:29,460 to recognise pictures of dogs and not dogs. 902 00:43:29,460 --> 00:43:33,060 Now, the way that this works is, what you do is you give 903 00:43:33,060 --> 00:43:37,220 the machine a bunch of pictures of both dogs and not dogs, 904 00:43:37,220 --> 00:43:38,900 and then you label them for the machine. 905 00:43:38,900 --> 00:43:41,300 So that one goes into the dog pile. Just here. 906 00:43:41,300 --> 00:43:42,500 Perfect. 907 00:43:42,500 --> 00:43:45,780 Now, initially, when you give these pictures to the machine, 908 00:43:45,780 --> 00:43:49,140 it's just going to be guessing at random, like we saw earlier 909 00:43:49,140 --> 00:43:51,460 in the first stages of that pigeon thing. 910 00:43:51,460 --> 00:43:56,260 But what you can do is if you play this over and over and over again 911 00:43:56,260 --> 00:44:01,420 with tens or maybe hundreds of thousands of different images, 912 00:44:01,420 --> 00:44:06,180 and eventually, if you do this enough times, you can end up with 913 00:44:06,180 --> 00:44:09,340 something like this little app over here. 914 00:44:09,340 --> 00:44:11,260 If you want to come and sit over here. 915 00:44:11,260 --> 00:44:14,180 Now, we've got a little algorithm running on this iPod 916 00:44:14,180 --> 00:44:16,860 that has been trained in just the way that we described. 917 00:44:16,860 --> 00:44:20,580 And we are going to show this algorithm a series of objects. 918 00:44:20,580 --> 00:44:22,940 And the algorithm is going to work out whether they are dogs 919 00:44:22,940 --> 00:44:24,780 or not dogs. All right. 920 00:44:24,780 --> 00:44:27,780 Is everyone ready to play dog or not dog? 921 00:44:27,780 --> 00:44:29,940 ALL: Yes! Here we go. 922 00:44:29,940 --> 00:44:33,260 Contestant number one, please. Dog or not dog? 923 00:44:38,220 --> 00:44:40,020 Hello! 924 00:44:40,020 --> 00:44:41,020 DOG BARKS 925 00:44:41,020 --> 00:44:43,180 Ohh! Dog! 926 00:44:43,180 --> 00:44:44,540 Hooray! 927 00:44:44,540 --> 00:44:47,100 Thank you, Luna! Contestant number two! 928 00:44:48,100 --> 00:44:50,340 Dog or not dog? 929 00:44:50,340 --> 00:44:52,780 Not dog! Thank you very much. 930 00:44:52,780 --> 00:44:55,020 Contestant number three. 931 00:44:55,020 --> 00:44:56,380 AUDIENCE: Awww... 932 00:44:56,380 --> 00:44:58,300 You OK, little guy? 933 00:44:58,300 --> 00:45:00,300 Dog! Amazing! 934 00:45:00,300 --> 00:45:02,740 Next contestant, please! 935 00:45:02,740 --> 00:45:04,740 HANNAH CACKLES 936 00:45:04,740 --> 00:45:07,820 Oh, this is very cute! Sit! Sit! 937 00:45:09,140 --> 00:45:11,500 Not dog! Well done. 938 00:45:11,500 --> 00:45:12,980 Haven't got one, mate! 939 00:45:12,980 --> 00:45:15,500 Next contestant, please! 940 00:45:19,380 --> 00:45:21,580 Aww! Look! 941 00:45:21,580 --> 00:45:23,500 Not dog! 942 00:45:23,500 --> 00:45:25,500 And our final contestant. 943 00:45:27,700 --> 00:45:29,980 AUDIENCE: Aww! 944 00:45:29,980 --> 00:45:32,340 Look how beautiful! 945 00:45:32,340 --> 00:45:34,180 Dog! 946 00:45:34,180 --> 00:45:35,780 Thank you very much, 947 00:45:35,780 --> 00:45:38,580 and thank you very much for playing the game with us. 948 00:45:38,580 --> 00:45:41,180 Big round of applause for everyone involved in dog or not dog! 949 00:45:49,500 --> 00:45:52,740 Now, this idea, this kind of algorithm, a reinforcement learning 950 00:45:52,740 --> 00:45:55,780 algorithm, it can actually have profound consequences 951 00:45:55,780 --> 00:45:58,500 because the algorithm doesn't really care what it's looking at. 952 00:45:58,500 --> 00:46:01,060 If it can learn to identify dogs, it can also learn 953 00:46:01,060 --> 00:46:03,220 to identify diseases. 954 00:46:03,220 --> 00:46:06,220 To explain, please welcome an ophthalmologist 955 00:46:06,220 --> 00:46:09,020 from Moorfields Eye Hospital, Pearse Keane. 956 00:46:12,020 --> 00:46:14,820 So, Pearse, tell me what it is that you do. 957 00:46:14,820 --> 00:46:19,100 So I'm a consultant ophthalmologist at Moorfields Eye Hospital 958 00:46:19,100 --> 00:46:22,180 and I specialise in the treatment of retinal diseases. 959 00:46:22,180 --> 00:46:25,420 So, in particular, I specialise in the treatment of a disease 960 00:46:25,420 --> 00:46:27,700 called macular degeneration, 961 00:46:27,700 --> 00:46:30,900 and macular degeneration is the commonest cause of blindness 962 00:46:30,900 --> 00:46:33,540 in the United Kingdom. And is it treatable? 963 00:46:33,540 --> 00:46:36,820 So the thing about macular degeneration is that, if we pick 964 00:46:36,820 --> 00:46:40,660 it up early, we have some good treatments and we can stop 965 00:46:40,660 --> 00:46:42,460 people going blind from it. 966 00:46:42,460 --> 00:46:45,660 The problem that we have is that it affects so many people as they get 967 00:46:45,660 --> 00:46:49,620 older that sometimes we have a challenge to actually find 968 00:46:49,620 --> 00:46:52,460 the people who've developed it and treat them in a timely 969 00:46:52,460 --> 00:46:55,540 fashion. Because there are so many people. Because there's so many. 970 00:46:55,540 --> 00:46:59,060 So we actually get 200 people who develop the blinding forms 971 00:46:59,060 --> 00:47:03,500 of macular degeneration every single day just in the UK. 972 00:47:03,500 --> 00:47:07,940 And so, for us, it seemed like this would be a perfect example 973 00:47:07,940 --> 00:47:12,860 where we could apply artificial intelligence to try to identify 974 00:47:12,860 --> 00:47:15,620 those people with the most sight-threatening diseases at the 975 00:47:15,620 --> 00:47:17,220 earliest point and save their sight. 976 00:47:17,220 --> 00:47:19,060 And I've brought Elaine Manor, 977 00:47:19,060 --> 00:47:21,700 a patient of mine at Moorfields Eye Hospital. 978 00:47:28,300 --> 00:47:30,860 Thank you very much for joining us, Elaine. 979 00:47:30,860 --> 00:47:34,340 For me, Elaine exemplifies actually why this is important, 980 00:47:34,340 --> 00:47:40,220 because the Elaine story is that actually she had lost sight 981 00:47:40,220 --> 00:47:43,660 in one eye from this condition, from macular degeneration, 982 00:47:43,660 --> 00:47:48,100 and then in 2012, she started to develop it in her good eye. 983 00:47:48,100 --> 00:47:50,900 And she went to her high street optician and was told, 984 00:47:50,900 --> 00:47:53,780 "You need to be urgently seen by a retina specialist," 985 00:47:53,780 --> 00:47:58,060 someone like me, but she got an appointment for six weeks later. 986 00:47:58,060 --> 00:48:00,380 So you can imagine a situation if you're losing sight 987 00:48:00,380 --> 00:48:04,140 in your good eye and there is an effective treatment, but you're told 988 00:48:04,140 --> 00:48:05,780 that you have to wait six weeks. 989 00:48:05,780 --> 00:48:07,420 What was that like at the time, Elaine? 990 00:48:07,420 --> 00:48:09,780 I was absolutely terrified. 991 00:48:09,780 --> 00:48:14,500 I felt that, when I went to bed, would I wake up to a world 992 00:48:14,500 --> 00:48:17,740 of darkness? Would I see my family again? 993 00:48:17,740 --> 00:48:19,780 It was worrying. Yeah. 994 00:48:19,780 --> 00:48:22,220 And this is all just because of the sheer volume of patients? 995 00:48:22,220 --> 00:48:25,060 Because of the huge number of patients. 996 00:48:25,060 --> 00:48:27,100 So what is this machine? How does that work? 997 00:48:27,100 --> 00:48:28,900 So this is an eye scanner. 998 00:48:28,900 --> 00:48:30,900 It's something that scans the retina. 999 00:48:30,900 --> 00:48:33,820 It's super-high resolution, a three-dimensional image 1000 00:48:33,820 --> 00:48:34,980 of the back of your eye. 1001 00:48:34,980 --> 00:48:38,100 But it's something that these algorithms can get to work on. 1002 00:48:38,100 --> 00:48:39,540 Well, that's exactly it. 1003 00:48:39,540 --> 00:48:42,380 I mean, if the algorithm can tell dog or not dog, 1004 00:48:42,380 --> 00:48:46,460 then it seems like it could tell macular degeneration or not macular 1005 00:48:46,460 --> 00:48:50,100 degeneration, and help people...help prevent people from going blind. 1006 00:48:50,100 --> 00:48:55,140 So I hope that we can have a look at how this algorithm looks 1007 00:48:55,140 --> 00:48:56,820 when it tries to process an image. 1008 00:48:56,820 --> 00:48:59,420 Can we have a little look? You can see here on the scan, it's called 1009 00:48:59,420 --> 00:49:02,020 out that there is an urgent problem with the eye. 1010 00:49:02,020 --> 00:49:04,420 It's identified choroidal neovascularisation. 1011 00:49:04,420 --> 00:49:07,580 This red area here is some blood vessel growth at the back 1012 00:49:07,580 --> 00:49:11,100 of the eye, which is suggestive of the condition Pearse was describing. 1013 00:49:11,100 --> 00:49:13,260 We've published a research article. 1014 00:49:13,260 --> 00:49:17,420 What we've shown is the proof of concept that the algorithm 1015 00:49:17,420 --> 00:49:22,180 that we've developed is actually as good as me or other consultants 1016 00:49:22,180 --> 00:49:24,860 at Moorfields or other world-leading eye doctors 1017 00:49:24,860 --> 00:49:26,660 at diagnosing these diseases. 1018 00:49:26,660 --> 00:49:29,900 And this stuff is, I mean, it's incredibly important to get 1019 00:49:29,900 --> 00:49:33,100 you to those urgent cases quicker. Yes. 1020 00:49:33,100 --> 00:49:34,340 Yes. Like Elaine, I guess. 1021 00:49:34,340 --> 00:49:36,980 But the great thing was, although it was stressful for Elaine 1022 00:49:36,980 --> 00:49:39,780 at the start, she was able to get the treatment 1023 00:49:39,780 --> 00:49:41,660 that she needed in this eye in time, 1024 00:49:41,660 --> 00:49:43,300 and we have been able to save the sight 1025 00:49:43,300 --> 00:49:45,140 that she has in her good eye. 1026 00:49:45,140 --> 00:49:48,460 Pearse Keane and Elaine, thank you very much indeed for joining me. 1027 00:49:48,460 --> 00:49:49,980 APPLAUSE 1028 00:49:55,420 --> 00:49:56,900 Wow. 1029 00:49:56,900 --> 00:49:59,780 As I think Pearse's work there demonstrates you can see 1030 00:49:59,780 --> 00:50:02,900 just how far you can go with a machine that is capable 1031 00:50:02,900 --> 00:50:05,100 of categorising images. As impressive 1032 00:50:05,100 --> 00:50:06,820 as all of this technology is, 1033 00:50:06,820 --> 00:50:09,060 it's not quite perfect. 1034 00:50:09,060 --> 00:50:11,180 It is important to say that these algorithms, 1035 00:50:11,180 --> 00:50:14,740 they don't really understand the world in the same way that we do. 1036 00:50:14,740 --> 00:50:17,740 They don't understand context, they don't understand nuance, 1037 00:50:17,740 --> 00:50:21,060 and that means that it really can make mistakes. 1038 00:50:21,060 --> 00:50:24,020 In particular, when you take something and you put 1039 00:50:24,020 --> 00:50:27,660 it in a strange situation, like this picture, here. 1040 00:50:27,660 --> 00:50:31,620 So if we take farmyard animals and we put them in strange situations, 1041 00:50:31,620 --> 00:50:34,660 like being cuddled by a child, for instance, 1042 00:50:34,660 --> 00:50:37,380 the algorithm labels as a dog. 1043 00:50:37,380 --> 00:50:40,620 This one I quite like. If you take a farmyard animal and put it in a 1044 00:50:40,620 --> 00:50:43,340 tree - this is a real photograph, by the way - 1045 00:50:43,340 --> 00:50:44,940 the algorithm gets a bit confused! 1046 00:50:44,940 --> 00:50:46,020 LAUGHTER 1047 00:50:46,020 --> 00:50:48,060 Surely it's an orangutan?! 1048 00:50:48,060 --> 00:50:50,660 And my favourite one of all is that, if you leave them back 1049 00:50:50,660 --> 00:50:54,260 in the original context but instead paint them pink, the algorithm 1050 00:50:54,260 --> 00:50:57,860 thinks they must be flowers, which I think is quite sweet! 1051 00:50:57,860 --> 00:50:59,260 LAUGHTER 1052 00:50:59,260 --> 00:51:02,700 This is the consequence of the way that these machines learn. 1053 00:51:02,700 --> 00:51:05,780 As we have seen a couple of times during this show, 1054 00:51:05,780 --> 00:51:08,820 in the beginning they just start off doing loads of completely random 1055 00:51:08,820 --> 00:51:11,980 things, which means that as they settle in to completing 1056 00:51:11,980 --> 00:51:14,540 their task they can end up with some rather 1057 00:51:14,540 --> 00:51:17,780 strange little quirks. Have a look at this little video, here. 1058 00:51:17,780 --> 00:51:19,780 So this is a simulation 1059 00:51:19,780 --> 00:51:24,860 of an algorithm that was given a body, and its objective was to 1060 00:51:24,860 --> 00:51:27,300 work out how to move its body. 1061 00:51:27,300 --> 00:51:30,380 Every time it falls over, it fails and starts again. 1062 00:51:30,380 --> 00:51:32,020 And you can see it in the beginning, 1063 00:51:32,020 --> 00:51:34,980 it's just trying lots of things, lots of random things. 1064 00:51:34,980 --> 00:51:39,460 But the consequence of that is that it really has, like, some quite 1065 00:51:39,460 --> 00:51:43,420 strange...quite strange habits that it ends up picking up 1066 00:51:43,420 --> 00:51:46,980 as it's sort of flailing all of its limbs around at random. 1067 00:51:46,980 --> 00:51:49,740 Now, this is what happens when you put reinforcement learning 1068 00:51:49,740 --> 00:51:53,260 into a body - a simulated one in that particular case - 1069 00:51:53,260 --> 00:51:56,740 but there is nothing stopping you from putting the very same ideas 1070 00:51:56,740 --> 00:52:00,340 into the body of a robot too. So, to explain, 1071 00:52:00,340 --> 00:52:03,180 please join me in welcoming back Anne-Marie Imafidon. 1072 00:52:12,500 --> 00:52:14,740 So, Anne-Marie, robots - talk me through it. 1073 00:52:14,740 --> 00:52:17,060 So, embodied robots, we've got 1074 00:52:17,060 --> 00:52:19,700 lots of them that we're trying to train to do all kinds 1075 00:52:19,700 --> 00:52:22,820 of different tasks, and we've got an example here. 1076 00:52:22,820 --> 00:52:25,660 OK. So this is from Professor Pieter Abbeel 1077 00:52:25,660 --> 00:52:28,460 at the Berkeley AI Research Lab. Exactly. 1078 00:52:28,460 --> 00:52:31,700 So we can see this is a robot that has never kind of known 1079 00:52:31,700 --> 00:52:33,140 how to move its arm before. 1080 00:52:33,140 --> 00:52:35,620 So it's just basically flinging its arm around at random, 1081 00:52:35,620 --> 00:52:37,220 trying to get the stuff in the box? 1082 00:52:37,220 --> 00:52:39,180 Yes, just trying to get the peg into the hole. 1083 00:52:39,180 --> 00:52:40,420 It's not doing very well. 1084 00:52:40,420 --> 00:52:43,580 It's not, because it hasn't had the time to learn what an elbow 1085 00:52:43,580 --> 00:52:46,620 is, kind of how it needs to move around, but also that the shape 1086 00:52:46,620 --> 00:52:48,740 needs to fit the hole that it's got there. 1087 00:52:48,740 --> 00:52:50,580 OK. So this is after five goes. 1088 00:52:50,580 --> 00:52:52,780 We can see here iteration eight as well 1089 00:52:52,780 --> 00:52:55,660 we've jumped to, and it's getting very close. 1090 00:52:55,660 --> 00:52:58,620 Yeah, it's still not managing it. I mean, it's basically just trying 1091 00:52:58,620 --> 00:53:00,380 lots and lots of things at random. 1092 00:53:00,380 --> 00:53:03,780 Yes. This I think, actually, it's something that pretty much every 1093 00:53:03,780 --> 00:53:06,980 parent of a young child will recognise. 1094 00:53:06,980 --> 00:53:11,180 This... This is Juno. Come on, then, 1095 00:53:11,180 --> 00:53:15,060 Juno, do you want to come and have a little play? 1096 00:53:15,060 --> 00:53:17,220 Now, this is something that you really notice, 1097 00:53:17,220 --> 00:53:23,020 is that, when they're very little, they don't necessarily understand 1098 00:53:23,020 --> 00:53:25,500 how to control their limbs. Let's give you a go. 1099 00:53:25,500 --> 00:53:28,340 You can try this, too, Juno, here we go. 1100 00:53:28,340 --> 00:53:30,300 Do you want to play with this one? 1101 00:53:30,300 --> 00:53:33,100 Is it a similar thing going on? Exactly. 1102 00:53:33,100 --> 00:53:36,700 So you learn the kind of motor skills you might need to actually 1103 00:53:36,700 --> 00:53:39,700 complete this task between 12 to 18 months, 1104 00:53:39,700 --> 00:53:41,740 and Juno here is only six months 1105 00:53:41,740 --> 00:53:43,980 so it's gone in her mouth, 1106 00:53:43,980 --> 00:53:47,380 which the robot didn't do, admittedly! 1107 00:53:47,380 --> 00:53:49,860 Different kind of reward! Exactly. 1108 00:53:49,860 --> 00:53:53,940 But it's only over time, by learning through lots and lots of 1109 00:53:53,940 --> 00:53:55,580 random behaviour, initially, 1110 00:53:55,580 --> 00:53:57,380 that we learn how to control our limbs. 1111 00:53:57,380 --> 00:53:59,420 We've all been there. We've all been Juno! 1112 00:53:59,420 --> 00:54:01,700 So the robots are learning in a very similar way, 1113 00:54:01,700 --> 00:54:03,500 just they do have fewer distractions. 1114 00:54:03,500 --> 00:54:06,180 They don't want to put things in their mouths, and their rewards 1115 00:54:06,180 --> 00:54:07,740 are just a little bit different. 1116 00:54:07,740 --> 00:54:11,340 So if the humans take about 18 months to master this task, 1117 00:54:11,340 --> 00:54:13,020 how long did that robot take? 1118 00:54:13,020 --> 00:54:15,780 So this particular robot that we were just looking at learned 1119 00:54:15,780 --> 00:54:18,140 to do that task in under an hour. 1120 00:54:18,140 --> 00:54:21,020 Yeah. Well, not long to go now, Juno. You might not manage it in 1121 00:54:21,020 --> 00:54:22,500 an hour, but give it a few months 1122 00:54:22,500 --> 00:54:24,260 and you'll definitely be doing these. 1123 00:54:24,260 --> 00:54:27,660 Thank you very much to Juno and Anne-Marie! Thank you. 1124 00:54:27,660 --> 00:54:29,540 APPLAUSE 1125 00:54:32,420 --> 00:54:36,980 That is the key idea - it's all about harnessing the power 1126 00:54:36,980 --> 00:54:41,540 of randomness, using trial and error to hone in on something that works. 1127 00:54:41,540 --> 00:54:43,980 Babies do it, robots do it, 1128 00:54:43,980 --> 00:54:46,340 and as an audience now, we're going to do it, too. 1129 00:54:46,340 --> 00:54:51,020 So, as a room, we're all going to be artificial intelligence. 1130 00:54:51,020 --> 00:54:54,580 We're going to let you loose like the machine-learning bots. 1131 00:54:54,580 --> 00:54:56,700 And for that, I'm going to hand over 1132 00:54:56,700 --> 00:54:59,020 to the very wonderful WiFi Wars. 1133 00:54:59,020 --> 00:55:01,380 CHEERING AND APPLAUSE 1134 00:55:04,060 --> 00:55:06,700 OK, Steve, tell us what we're going to do. 1135 00:55:06,700 --> 00:55:10,740 So what we're going to do is we'll get you guys here to represent 1136 00:55:10,740 --> 00:55:12,780 random computer attempts at navigating a maze. 1137 00:55:12,780 --> 00:55:14,540 What we need to do is get your phones out 1138 00:55:14,540 --> 00:55:15,820 and go into your browsers. 1139 00:55:15,820 --> 00:55:18,420 And while you are going back into there, I will show you the maze 1140 00:55:18,420 --> 00:55:20,900 if I can, please, Rob. There we are. So what we're going to do, 1141 00:55:20,900 --> 00:55:23,460 we're going to put all of you in this blue room in the bottom right 1142 00:55:23,460 --> 00:55:25,980 corner, but we're going to incentivise you to go exploring. 1143 00:55:25,980 --> 00:55:28,220 So you will get ten points for each square in the corridor 1144 00:55:28,220 --> 00:55:30,180 you get through, but, far more importantly, 1145 00:55:30,180 --> 00:55:31,860 you'll get 1,000 bonus points for each 1146 00:55:31,860 --> 00:55:33,860 of the coloured rooms you manage to get to. 1147 00:55:33,860 --> 00:55:37,180 So the big prize in this one is get to these three interconnected rooms 1148 00:55:37,180 --> 00:55:39,900 in the top left. If you do that, I will be very impressed. 1149 00:55:39,900 --> 00:55:41,860 Everything we do at WiFi Wars is competitive 1150 00:55:41,860 --> 00:55:43,900 so what we've done is we split you into two halves. 1151 00:55:43,900 --> 00:55:45,660 So you guys on this half of the room, 1152 00:55:45,660 --> 00:55:49,180 I believe, are the blue computer. CHEERING 1153 00:55:49,180 --> 00:55:51,740 And you guys are the much louder red computer! 1154 00:55:51,740 --> 00:55:53,500 CHEERING/BOOING 1155 00:55:53,500 --> 00:55:55,900 That wasn't fair. That wasn't fair. Booing already! 1156 00:55:55,900 --> 00:55:59,180 OK. So I'm going to give you 90 seconds to go through that. 1157 00:55:59,180 --> 00:56:00,780 So, because there's so many people 1158 00:56:00,780 --> 00:56:03,140 and we're going to be looking at that big map... Yeah. 1159 00:56:03,140 --> 00:56:05,740 ..it's really difficult to work out who you are, right? 1160 00:56:05,740 --> 00:56:08,300 So you are essentially each representing a random 1161 00:56:08,300 --> 00:56:10,180 attempt at winning this game. Exactly. 1162 00:56:10,180 --> 00:56:12,580 So if it was one or two you could find your way through, 1163 00:56:12,580 --> 00:56:14,940 but we're going to make that much more complicated. OK. 1164 00:56:14,940 --> 00:56:17,140 Good. Are you ready? ALL: Yeah! 1165 00:56:17,140 --> 00:56:18,860 All right. Let's do a game, then. 1166 00:56:18,860 --> 00:56:21,220 If you send it into their phones now, please, Rob. We'll put 1167 00:56:21,220 --> 00:56:23,660 90 seconds on the clock and hopefully we're going to see... 1168 00:56:23,660 --> 00:56:25,460 Immediately we see you all there appearing. 1169 00:56:25,460 --> 00:56:27,980 We've got a lot of blue and red players from the two computers. 1170 00:56:27,980 --> 00:56:30,620 Already people are getting out - a red person was the first one out. 1171 00:56:30,620 --> 00:56:32,980 My goodness me! Although collectively the blue team... 1172 00:56:32,980 --> 00:56:34,580 I was going to say they're doing better 1173 00:56:34,580 --> 00:56:36,900 but by the time I said it, they're not, so the red computer 1174 00:56:36,900 --> 00:56:39,180 after 15 seconds have 6,000 points and the blues 5,400. 1175 00:56:39,180 --> 00:56:41,180 No-one's found a coloured room yet. 1176 00:56:41,180 --> 00:56:42,540 And, as I've said, if you can get 1177 00:56:42,540 --> 00:56:44,180 to these three in the top left corner... 1178 00:56:44,180 --> 00:56:46,980 Blues doing very well. Yeah, they're all going for it. Going round 1179 00:56:46,980 --> 00:56:49,780 the two main channels. As you would expect, sort of more often than not 1180 00:56:49,780 --> 00:56:52,500 they're picking the main thoroughfares as they hit dead ends. 1181 00:56:52,500 --> 00:56:55,340 And actually we've got a few people who are in the purple room now, 1182 00:56:55,340 --> 00:56:57,420 so well done. I think red was the first so well done. 1183 00:56:57,420 --> 00:56:59,500 For context, if you're wondering what's going on, 1184 00:56:59,500 --> 00:57:02,100 what Rob's created is a thing that allows us - without installing 1185 00:57:02,100 --> 00:57:04,260 anything, any device, browser or operating system - 1186 00:57:04,260 --> 00:57:06,500 to beam games onto your devices. A red one's on the way! 1187 00:57:06,500 --> 00:57:08,740 Wow. All right. Yeah. Good. I don't know if you know... 1188 00:57:08,740 --> 00:57:12,380 If there's a red, if you now are in a green room, cheer. 1189 00:57:12,380 --> 00:57:14,900 Yes, you! Well done to you. You were very, very good. 1190 00:57:14,900 --> 00:57:16,980 You've got 30 seconds left to do something about it. 1191 00:57:16,980 --> 00:57:19,700 Nobody's managed to find the orange or the red yet, 1192 00:57:19,700 --> 00:57:20,900 but we have got a few... 1193 00:57:20,900 --> 00:57:23,700 Well, we've got another red coming for the green, but a lot more blues, 1194 00:57:23,700 --> 00:57:25,380 actually, so the blues might level this up 1195 00:57:25,380 --> 00:57:27,700 if they can all get in there, but with 20 seconds left 1196 00:57:27,700 --> 00:57:29,660 it is 70,000 points to the red computer. 1197 00:57:29,660 --> 00:57:31,100 Blues, you need to do better. 1198 00:57:31,100 --> 00:57:33,300 You've got only 60,000 with 16 seconds left. 1199 00:57:33,300 --> 00:57:35,500 So many people getting stuck in this horrible dead end 1200 00:57:35,500 --> 00:57:37,380 that we cruelly put in the top right corner. 1201 00:57:37,380 --> 00:57:39,740 I think blues will need a miracle. The reds are cheering. 1202 00:57:39,740 --> 00:57:40,900 Ten seconds left. 1203 00:57:40,900 --> 00:57:44,060 Five seconds left now. 88,000 for the red computer. 1204 00:57:44,060 --> 00:57:45,700 82,000 now for the blues, 1205 00:57:45,700 --> 00:57:49,340 so that is a win, I'm afraid, for the red computer! 1206 00:57:49,340 --> 00:57:51,260 Wahey! Well done to you all. 1207 00:57:51,260 --> 00:57:55,140 "Boo!" Polite applause. 1208 00:57:55,140 --> 00:57:57,540 Well done. That was absolutely fantastic. 1209 00:57:57,540 --> 00:57:59,180 OK. Here is the big question 1210 00:57:59,180 --> 00:58:01,740 on top of all of that, because we saw in the last lecture 1211 00:58:01,740 --> 00:58:04,180 the highs and the lows of uncertainty. We saw what can go 1212 00:58:04,180 --> 00:58:07,540 right and what can go wrong when you rely on something 1213 00:58:07,540 --> 00:58:09,780 that has randomness at its core. 1214 00:58:09,780 --> 00:58:14,100 So, OK, if we accept that perfection is impossible - if we acknowledge 1215 00:58:14,100 --> 00:58:17,940 that these things are always going to have errors - we have to ask, 1216 00:58:17,940 --> 00:58:22,300 do we want to put flawed machines in a position of power? 1217 00:58:22,300 --> 00:58:25,180 That, I think, is one for the next lecture. 1218 00:58:25,180 --> 00:58:28,140 But, for now, who'd like to have a rematch of this? 1219 00:58:28,140 --> 00:58:30,300 CHEERING 103883

Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.