1 00:00:03,860 --> 00:00:08,240 - I know the whole House will be very concerned at the extremely serious 2 00:00:08,240 --> 00:00:11,880 incident that has taken place in Southport. 3 00:00:13,400 --> 00:00:16,400 - A 17-year-old male from Banks, in Lancashire, 4 00:00:16,400 --> 00:00:18,600 who is originally from Cardiff, 5 00:00:18,600 --> 00:00:21,200 has been arrested on suspicion of murder. 6 00:00:27,240 --> 00:00:30,960 - As soon as I saw those tweets, 7 00:00:30,960 --> 00:00:34,800 spreading misinformation about the identity of the attacker, 8 00:00:34,800 --> 00:00:38,520 I knew exactly what was going to happen. 9 00:00:47,960 --> 00:00:51,560 - The violence started after false rumours about the suspect... 10 00:00:53,080 --> 00:00:56,760 - Social media platforms are a way that people express 11 00:00:56,760 --> 00:00:58,760 their shock and anger. 12 00:00:58,760 --> 00:01:01,280 - An undocumented migrant decided to go into 13 00:01:01,280 --> 00:01:03,400 a Taylor Swift dance class... 14 00:01:03,400 --> 00:01:07,520 - When toxic opinions go viral, it can have real-world harm. 15 00:01:13,560 --> 00:01:18,840 But Southport was just the latest in a long chain of events. 16 00:01:23,200 --> 00:01:27,360 We have seen social media have so much influence, 17 00:01:27,360 --> 00:01:29,400 over and over again. 18 00:01:31,800 --> 00:01:34,880 And that is down to the algorithm. 19 00:01:34,880 --> 00:01:36,960 An unstoppable flow of content. 20 00:01:38,480 --> 00:01:42,520 And those of us who understand how this works 21 00:01:42,520 --> 00:01:45,440 need to stop that happening. 22 00:01:47,200 --> 00:01:49,720 - Our world is being actively remade. 23 00:01:51,600 --> 00:01:54,200 Falsehood spreads faster than fact, 24 00:01:54,200 --> 00:01:57,120 dividing communities and distorting reality. 25 00:01:58,640 --> 00:02:02,360 And as trust collapses, so does our shared sense of truth. 26 00:02:03,640 --> 00:02:07,680 The defining challenge of the 21st century is not simply 27 00:02:07,680 --> 00:02:10,800 who wields the most powerful technologies, 28 00:02:10,800 --> 00:02:13,920 but who guides them with the greatest wisdom. 29 00:02:13,920 --> 00:02:16,200 - You've created these platforms, 30 00:02:16,200 --> 00:02:18,600 and now they are being misused. 31 00:02:18,600 --> 00:02:20,360 - It's for profits, for money. 32 00:02:20,360 --> 00:02:22,080 - Influence and power. 33 00:02:22,080 --> 00:02:24,960 - People don't understand the beast that they're feeding. 34 00:02:24,960 --> 00:02:27,320 - It becomes our duty to speak out. 35 00:02:38,440 --> 00:02:40,440 - Do you want me to look at the camera, by the way? - Yeah. 36 00:02:40,440 --> 00:02:41,960 - Ready? Shall I clap? 37 00:02:43,480 --> 00:02:46,240 Ever since I can remember, I've loved social media. 38 00:02:48,040 --> 00:02:49,600 I grew up on Facebook. 39 00:02:51,480 --> 00:02:54,520 It was an important part of connecting with my friends. 40 00:02:55,760 --> 00:02:59,120 The worst thing that would happen is you'd find out you hadn't been 41 00:02:59,120 --> 00:03:02,200 invited to a party or a sleepover. 42 00:03:02,200 --> 00:03:04,240 It doesn't feel like that now. 43 00:03:04,240 --> 00:03:05,840 It feels quite different. 44 00:03:07,680 --> 00:03:10,960 My Feed is full of this toxic, angry stuff, 45 00:03:10,960 --> 00:03:13,040 people screaming in my face, 46 00:03:13,040 --> 00:03:15,760 and I can't not look because it's my job. 47 00:03:16,920 --> 00:03:20,400 As the BBC's social media investigations correspondent, 48 00:03:20,400 --> 00:03:24,280 part of what I do is try to understand why there's so much 49 00:03:24,280 --> 00:03:26,320 of this stuff on these platforms. 50 00:03:26,320 --> 00:03:31,240 But it's incredibly hard to get straight answers from the companies. 51 00:03:31,240 --> 00:03:36,560 And the few people who do come forward do so at great risk. 52 00:03:43,040 --> 00:03:46,480 - If you come out as a whistle-blower, and you are named, 53 00:03:46,480 --> 00:03:49,640 you have a scarlet letter on your head. 54 00:03:49,640 --> 00:03:52,680 So, my goal is to not get sued. 55 00:03:54,680 --> 00:04:01,200 By coming public, I know that my career prospects in big tech 56 00:04:01,200 --> 00:04:04,440 are probably next to zero now. 57 00:04:05,440 --> 00:04:07,240 - How does it feel showing me this? 58 00:04:08,560 --> 00:04:10,320 - Sort of terrifying. 59 00:04:10,320 --> 00:04:12,120 Sort of terrifying. 60 00:04:15,160 --> 00:04:17,080 - Oh, wow, you've got so much. 61 00:04:18,440 --> 00:04:21,320 This must have taken you ages to document. 62 00:04:21,320 --> 00:04:23,480 - Yeah, yeah. 63 00:04:23,480 --> 00:04:25,120 A little bit. 64 00:04:25,120 --> 00:04:26,800 I spent four years at Meta, 65 00:04:26,800 --> 00:04:29,000 so I've seen the other side of the curtain 66 00:04:29,000 --> 00:04:33,080 and I know some of the problems more than I wish I did. 67 00:04:34,840 --> 00:04:39,320 I have these kind of higher-level research documents showing, like, 68 00:04:39,320 --> 00:04:42,600 all sorts of harms to users on these platforms. 69 00:04:44,440 --> 00:04:47,360 If these companies are not going to protect people, 70 00:04:47,360 --> 00:04:51,120 it becomes my duty, and people like me - 71 00:04:51,120 --> 00:04:54,680 as people that actually understand how these systems work - 72 00:04:54,680 --> 00:04:56,440 to speak out. 73 00:04:59,960 --> 00:05:03,360 So, a lot of the screenshots that I have here 74 00:05:03,360 --> 00:05:05,720 are how we ranked content. 75 00:05:05,720 --> 00:05:09,880 The algorithm is built to prioritise the stuff that gets 76 00:05:09,880 --> 00:05:13,320 the most engagement, and the political content that gets 77 00:05:13,320 --> 00:05:16,480 the most engagement is typically misinformation. 78 00:05:16,480 --> 00:05:18,200 It's typically very toxic. 79 00:05:18,200 --> 00:05:20,920 So, people yelling and being really mean to each other. 80 00:05:20,920 --> 00:05:24,480 - How worried should we be about the algorithm, in terms of the role 81 00:05:24,480 --> 00:05:26,960 they play in pushing this kind of stuff? 82 00:05:26,960 --> 00:05:29,600 - People don't want to see that content. 83 00:05:29,600 --> 00:05:34,480 But people don't know that they're choosing to push this content 84 00:05:34,480 --> 00:05:37,160 to you, even though you say you don't want it. 85 00:05:37,160 --> 00:05:40,680 And the consequences are really, really bad. 86 00:05:43,080 --> 00:05:46,560 Meta's products are used by north of 3 billion people. 87 00:05:46,560 --> 00:05:49,160 And the more time they can keep you on there, 88 00:05:49,160 --> 00:05:52,560 the more ads they sell and the more money they make. 89 00:05:52,560 --> 00:05:56,160 But it's very important that they get this stuff right, 90 00:05:56,160 --> 00:05:59,280 because when they don't, really bad things happen. 91 00:06:23,000 --> 00:06:26,880 - In 2018, Facebook took Sri Lanka by storm. 92 00:06:26,880 --> 00:06:30,520 This whole road is full of smartphone shops on either side, 93 00:06:30,520 --> 00:06:33,680 and it's just shop after shop after shop. 94 00:06:33,680 --> 00:06:37,000 And people come here from all over the city, actually, 95 00:06:37,000 --> 00:06:38,480 to get a good deal. 96 00:06:40,040 --> 00:06:43,360 Facebook had a programme where, with every phone that was sold, 97 00:06:43,360 --> 00:06:46,800 you got Facebook for free - it didn't use up your data. 98 00:06:46,800 --> 00:06:49,640 So, it kind of became the default platform. 99 00:06:49,640 --> 00:06:52,920 You felt a real sense of community, 100 00:06:52,920 --> 00:06:55,360 a sense of...belonging. 101 00:06:58,120 --> 00:07:02,680 Scratch beneath that veneer, and Facebook became the predominant way 102 00:07:02,680 --> 00:07:08,640 that the worst elements in society used it to inflame hate. 103 00:07:11,040 --> 00:07:14,840 Sri Lanka is marked by a majority Sinhala Buddhist population. 104 00:07:16,000 --> 00:07:19,240 We've had Muslims for centuries. 105 00:07:19,240 --> 00:07:22,960 But there is a small group of Buddhist monks... 106 00:07:25,520 --> 00:07:29,320 ..who used Facebook to directly target the Muslim community 107 00:07:29,320 --> 00:07:30,640 in Sri Lanka. 108 00:07:32,320 --> 00:07:35,280 I was really tuned in to how Facebook was being weaponised. 109 00:07:36,800 --> 00:07:42,120 And there was this video that I saw of this Muslim man. 110 00:07:44,080 --> 00:07:47,680 The people recording the footage are asking him 111 00:07:47,680 --> 00:07:52,000 whether he put infertility pills into the food that they consumed. 112 00:07:59,720 --> 00:08:01,000 And he's confused. 113 00:08:01,000 --> 00:08:03,520 He doesn't clearly understand Sinhala. 114 00:08:03,520 --> 00:08:07,520 In those parts of the country, Tamil is spoken, not Sinhala. 115 00:08:11,920 --> 00:08:15,200 But the voices take it as - this was the evidence, right? 116 00:08:15,200 --> 00:08:18,840 I mean, here was a man who was admitting to the fact 117 00:08:18,840 --> 00:08:24,120 that the Muslim community was making Sinhala Buddhist women infertile. 118 00:08:25,640 --> 00:08:30,200 So, by 48 hours, this was shared in the thousands. 119 00:08:30,200 --> 00:08:32,240 It was like a... 120 00:08:32,240 --> 00:08:34,040 ..a frothing hate. 121 00:08:36,160 --> 00:08:37,760 - TRANSLATION: 122 00:08:56,960 --> 00:08:59,240 - Buddhist mobs set fire to homes and businesses, 123 00:08:59,240 --> 00:09:00,760 killing young Muslims. 124 00:09:02,520 --> 00:09:05,600 - It was quite simply one of the worst anti-Muslim violence 125 00:09:05,600 --> 00:09:07,200 that Sri Lanka had ever seen. 126 00:09:20,760 --> 00:09:23,320 Facebook was the wind to the seeds of racism. 127 00:09:25,240 --> 00:09:28,440 What was on the platform resulted in people believing it 128 00:09:28,440 --> 00:09:30,680 and taking action as a consequence. 129 00:09:32,120 --> 00:09:34,480 And I didn't have any way to reach out. 130 00:09:34,480 --> 00:09:36,240 There was no-one to reach out to. 131 00:09:37,440 --> 00:09:39,440 We were utterly helpless. 132 00:09:43,480 --> 00:09:46,240 - I find it bonkers, when you look at the stuff that Matt shared 133 00:09:46,240 --> 00:09:48,320 from inside Facebook. 134 00:09:49,840 --> 00:09:54,960 I think that this document is one of the most revealing ones, 135 00:09:54,960 --> 00:09:58,720 as it's an internal study led by a data scientist. 136 00:10:00,200 --> 00:10:03,680 It essentially talks about how Facebook rewards outrage 137 00:10:03,680 --> 00:10:06,680 and that, if you get more negative comments, 138 00:10:06,680 --> 00:10:08,760 you're going to get more clicks. 139 00:10:08,760 --> 00:10:11,600 And there's this bit where they talk about the incentives 140 00:10:11,600 --> 00:10:14,360 that the company are creating, essentially saying 141 00:10:14,360 --> 00:10:16,160 "outrage gets attention". 142 00:10:16,160 --> 00:10:19,200 And then laying out and saying, "The current set of financial incentives 143 00:10:19,200 --> 00:10:22,520 "our algorithms create does not appear to be aligned 144 00:10:22,520 --> 00:10:23,920 "with our mission." 145 00:10:23,920 --> 00:10:26,800 Eg, their mission is to do good in the world, 146 00:10:26,800 --> 00:10:29,840 and yet those recommendation systems, those algorithms, 147 00:10:29,840 --> 00:10:34,200 are pushing the kind of content that feeds off negative reaction. 148 00:10:34,200 --> 00:10:37,280 And, therefore, you can see how it's impacting users. 149 00:10:37,280 --> 00:10:39,600 And they essentially lay it out in black and white. 150 00:10:39,600 --> 00:10:40,800 It's not like... 151 00:10:40,800 --> 00:10:43,360 It's not like it's much of a secret, really. We kind of all know it. 152 00:10:43,360 --> 00:10:45,160 But the thing is, THEY know it too. 153 00:10:47,240 --> 00:10:50,040 And Sri Lanka was not a one-off thing. 154 00:10:50,040 --> 00:10:54,040 - Almost as far as the eye could see, a tide of humanity. 155 00:10:54,040 --> 00:10:56,760 They're carrying with them whatever they could salvage 156 00:10:56,760 --> 00:10:58,720 from their villages, their homes that they say they were 157 00:10:58,720 --> 00:11:00,840 burnt out of by the Myanmar military. 158 00:11:00,840 --> 00:11:02,960 Look at that little baby there in a basket. 159 00:11:05,560 --> 00:11:08,800 - Take, for example, what happened in Myanmar. 160 00:11:08,800 --> 00:11:11,360 Facebook was actively pushing hate. 161 00:11:11,360 --> 00:11:13,360 And it's not just these places. 162 00:11:15,400 --> 00:11:19,080 It was part of a pattern that was happening all over the world. 163 00:11:19,080 --> 00:11:22,720 People were acting on the basis of what they were seeing on Facebook. 164 00:11:22,720 --> 00:11:25,400 There wasn't any other information they were responding to. 165 00:11:25,400 --> 00:11:28,120 And I think it's for that reason that Facebook almost had to 166 00:11:28,120 --> 00:11:30,680 come out and apologise. Because there was no way of saying, 167 00:11:30,680 --> 00:11:32,720 "Well, hang on, this wasn't to do with us." 168 00:11:32,720 --> 00:11:34,760 It was entirely connected to Facebook. 169 00:11:34,760 --> 00:11:36,840 - We've got to keep moving, folks. 170 00:11:36,840 --> 00:11:38,640 Watch your steps. 171 00:11:46,840 --> 00:11:50,840 - Recently, UN investigators blamed Facebook for playing a role in 172 00:11:50,840 --> 00:11:54,240 inciting possible genocide in Myanmar. 173 00:11:54,240 --> 00:11:57,320 - What's happening in Myanmar is a terrible tragedy, 174 00:11:57,320 --> 00:11:59,920 and we need to do more... - We all agree with that. 175 00:12:01,760 --> 00:12:05,960 - I was recruited to Meta because the company got a lot of bad PR. 176 00:12:05,960 --> 00:12:08,200 I mean, there was Cambridge Analytica. 177 00:12:08,200 --> 00:12:12,320 There was Russia's involvement in the 2016 election. 178 00:12:12,320 --> 00:12:16,320 And from places like Myanmar or Sri Lanka, 179 00:12:16,320 --> 00:12:18,280 they just weren't investing enough, 180 00:12:18,280 --> 00:12:21,480 and a lot of the hate speech there was just missed. 181 00:12:21,480 --> 00:12:23,200 It wasn't being caught. 182 00:12:23,200 --> 00:12:26,480 - We didn't take a broad enough view of our responsibility, 183 00:12:26,480 --> 00:12:28,720 and that was a big mistake. 184 00:12:28,720 --> 00:12:30,880 And it was my mistake, and I'm sorry. 185 00:12:33,600 --> 00:12:37,200 - So, Mark spent in the billions 186 00:12:37,200 --> 00:12:40,360 to hire a bunch of experts 187 00:12:40,360 --> 00:12:44,320 and create specialised integrity teams. 188 00:12:44,320 --> 00:12:49,560 Our job description was reducing harms to users, at a global scale. 189 00:12:49,560 --> 00:12:52,640 I said, "Oh, that's my dream. 190 00:12:52,640 --> 00:12:54,360 "That is my life's work." 191 00:12:56,880 --> 00:13:00,920 - I had the opportunity to join Facebook, 192 00:13:00,920 --> 00:13:05,440 to actually see it for myself, 193 00:13:05,440 --> 00:13:07,440 from the inside. 194 00:13:07,440 --> 00:13:11,880 And so, I think it was really once-in-a-lifetime opportunity. 195 00:13:11,880 --> 00:13:18,680 There were so many dedicated people that wanted to do the right thing. 196 00:13:20,760 --> 00:13:22,440 - You know, I believe deeply in transparency. 197 00:13:22,440 --> 00:13:25,720 And I think one of the solutions out there is to share more data 198 00:13:25,720 --> 00:13:27,680 with the outside world. 199 00:13:27,680 --> 00:13:29,840 And so, I created CrowdTangle. 200 00:13:31,520 --> 00:13:33,720 An online dashboard you could log into. 201 00:13:33,720 --> 00:13:37,880 What we did was made it really easy to see what stories 202 00:13:37,880 --> 00:13:42,640 and what content was true or false on Facebook that day. 203 00:13:42,640 --> 00:13:44,680 If it weren't for that CrowdTangle dashboard, 204 00:13:44,680 --> 00:13:46,480 you couldn't get any of that. 205 00:13:48,200 --> 00:13:50,880 Mark was so invested in it, and bought it 206 00:13:50,880 --> 00:13:54,120 to make sure Facebook has a positive role on the world. 207 00:13:55,840 --> 00:13:58,320 - What is the new mission of Facebook? 208 00:13:58,320 --> 00:14:02,600 - So, our new mission is to bring the world closer together. 209 00:14:02,600 --> 00:14:05,120 - When Mark announced a major algorithm change, 210 00:14:05,120 --> 00:14:09,440 called meaningful social interactions, my sense, actually, 211 00:14:09,440 --> 00:14:13,480 was that there was a real, genuine interest in trying to build 212 00:14:13,480 --> 00:14:16,040 a better algorithm during those years. 213 00:14:16,040 --> 00:14:18,640 There was a real overhaul to the fundamental way 214 00:14:18,640 --> 00:14:21,400 that content was ranked inside the Feed. 215 00:14:21,400 --> 00:14:24,640 Ranking friends and family over something called civic content, 216 00:14:24,640 --> 00:14:26,760 news and political content. 217 00:14:26,760 --> 00:14:29,720 They said, "Yeah, it turns out people click on the politics, 218 00:14:29,720 --> 00:14:32,480 "but, actually, they end up like feeling worse after a while, 219 00:14:32,480 --> 00:14:34,880 "and so we're going to reduce the amount of politics in the Feed." 220 00:14:36,760 --> 00:14:40,720 - My job entailed running large-scale experiments on 221 00:14:40,720 --> 00:14:43,640 sometimes as many as hundreds of millions of people, 222 00:14:43,640 --> 00:14:47,440 where we would change how content was ranked in Feed. 223 00:14:47,440 --> 00:14:50,800 And these people often had no idea that they were in experiments. 224 00:14:50,800 --> 00:14:54,480 You just might have a different feel to your Feed one day. 225 00:14:54,480 --> 00:14:57,840 So, you log in, and much of the civic content from your Feed 226 00:14:57,840 --> 00:15:01,360 might be missing, or the protections against harmful content 227 00:15:01,360 --> 00:15:02,960 might be turned off. 228 00:15:04,840 --> 00:15:08,560 And what I found was that, you remove the reliable, 229 00:15:08,560 --> 00:15:10,360 trustworthy news sources 230 00:15:10,360 --> 00:15:12,160 and you end up with my uncle, 231 00:15:12,160 --> 00:15:14,560 who likes to post conspiracy theories. 232 00:15:16,480 --> 00:15:20,680 This change that was meant to bring people closer together 233 00:15:20,680 --> 00:15:24,680 and help them connect actually had the opposite effect, 234 00:15:24,680 --> 00:15:28,080 where it was promoting content that was dividing people 235 00:15:28,080 --> 00:15:29,840 and sowing more division 236 00:15:29,840 --> 00:15:31,800 and creating more conflict between people. 237 00:15:31,800 --> 00:15:34,960 So, while it sounded great, we've more than doubled 238 00:15:34,960 --> 00:15:38,800 views on misinformation by demoting civic content. 239 00:15:38,800 --> 00:15:41,400 Is that the goal? No, it's not the goal. 240 00:15:41,400 --> 00:15:42,920 It's the opposite of the goal. 241 00:15:45,280 --> 00:15:50,160 So, we warned against this in a document 242 00:15:50,160 --> 00:15:54,040 that we delivered to the team, 243 00:15:54,040 --> 00:15:57,920 and that was read to Zuck in advance of launching 244 00:15:57,920 --> 00:16:00,880 these broad-scale experiments, and... 245 00:16:02,120 --> 00:16:03,920 ..that was all ignored. 246 00:16:05,560 --> 00:16:10,000 We warned, very sternly, this is not going to go well. 247 00:16:10,000 --> 00:16:11,920 And then, it didn't go well. 248 00:16:17,560 --> 00:16:21,320 - Major cities across the country are looking more like ghost towns. 249 00:16:21,320 --> 00:16:26,240 Shops locked up, normally bustling streets now oddly quiet. 250 00:16:26,240 --> 00:16:29,440 - At first, it seemed like just another event. 251 00:16:30,840 --> 00:16:36,320 And then suddenly we had this massive surge, in terms of volume. 252 00:16:36,320 --> 00:16:39,720 People were spending a large amount of time on online platforms, 253 00:16:39,720 --> 00:16:42,200 where all the information was brand-new, 254 00:16:42,200 --> 00:16:47,080 and we weren't able to clearly identify exactly what the truth was. 255 00:16:47,080 --> 00:16:50,280 - New cases and deaths associated with the coronavirus 256 00:16:50,280 --> 00:16:53,120 continue to climb right around the world. 257 00:16:53,120 --> 00:16:55,840 - What the pandemic created was this perfect storm, 258 00:16:55,840 --> 00:17:00,000 because you've got people stuck at home, spending more time online. 259 00:17:00,000 --> 00:17:02,840 And the algorithms, the recommendation systems, 260 00:17:02,840 --> 00:17:05,040 they were becoming ever more influential. 261 00:17:06,360 --> 00:17:08,920 - They'll kill millions with their vaccines. 262 00:17:08,920 --> 00:17:12,840 There is no vaccine currently on the schedule 263 00:17:12,840 --> 00:17:16,520 for any RNA virus that works. 264 00:17:16,520 --> 00:17:19,080 - I found it very frustrating. 265 00:17:19,080 --> 00:17:22,840 A lot of my time at Meta was spent focused on how to quickly identify 266 00:17:22,840 --> 00:17:25,760 conspiracy theories, conflicting information. 267 00:17:25,760 --> 00:17:28,360 - Right now, you are being lied to! 268 00:17:28,360 --> 00:17:31,800 - It was always difficult to identify which of those would escalate 269 00:17:31,800 --> 00:17:33,800 to rioting or violence. 270 00:17:35,240 --> 00:17:38,440 And that's a hard thing to deal with. 271 00:17:38,440 --> 00:17:41,120 - None of you have reported on the truth since day one. 272 00:17:41,120 --> 00:17:44,360 You're complicit in crimes against humanity. 273 00:17:44,360 --> 00:17:47,720 - Just knowing that there's the possibility that something bad 274 00:17:47,720 --> 00:17:49,720 might have happened and maybe there might have been 275 00:17:49,720 --> 00:17:51,400 an opportunity to prevent it... 276 00:17:52,720 --> 00:17:55,320 ..that's a really jarring thing for me. 277 00:17:57,080 --> 00:18:00,040 The nightmare that kind of sits in the back of my brain. 278 00:18:02,760 --> 00:18:05,520 It's definitely a space which is stressful for me. 279 00:18:06,960 --> 00:18:08,800 - Misinformation running rampant... 280 00:18:08,800 --> 00:18:10,480 - You are being lied to! 281 00:18:11,760 --> 00:18:12,960 - They'll kill millions... 282 00:18:15,120 --> 00:18:16,640 - CHANTING 283 00:18:27,560 --> 00:18:30,480 PRODUCER: - What did it feel like when you were at the riots that night? 284 00:18:39,080 --> 00:18:42,120 - During the lockdown, there was no light in the city. 285 00:18:42,120 --> 00:18:44,400 Everything was closed. 286 00:18:44,400 --> 00:18:47,520 And as a DJ artist, it was devastating. 287 00:18:48,720 --> 00:18:50,320 I felt really lonely. 288 00:18:50,320 --> 00:18:52,520 Sometimes, I even wanted to cry. 289 00:18:52,520 --> 00:18:55,400 Got a little bit more...more frustrated. 290 00:18:57,480 --> 00:19:00,240 We wanted our freedom back and go back to normal. 291 00:19:02,200 --> 00:19:04,720 So, when I saw a post of Facebook group 292 00:19:04,720 --> 00:19:07,640 that was anti-Dutch government, 293 00:19:07,640 --> 00:19:09,760 it intrigued me. 294 00:19:09,760 --> 00:19:11,880 They do for the same cause like I did. 295 00:19:14,520 --> 00:19:18,360 It was my first time in the group. 296 00:19:18,360 --> 00:19:21,840 It felt like I found a new family. 297 00:19:21,840 --> 00:19:25,320 We stayed in contact every day, sometimes even at night. 298 00:19:27,280 --> 00:19:30,160 The whole pandemic, it divided people 299 00:19:30,160 --> 00:19:32,040 into so many different groups. 300 00:19:34,920 --> 00:19:37,920 But then, you always have a few people who have 301 00:19:37,920 --> 00:19:39,960 a lot of influence on other people. 302 00:19:39,960 --> 00:19:42,720 They have power, mental power. 303 00:19:45,040 --> 00:19:47,160 It was almost a daily routine for them 304 00:19:47,160 --> 00:19:49,200 to spread hate and to be negative. 305 00:19:50,840 --> 00:19:53,000 And the tension was growing. 306 00:19:53,000 --> 00:19:55,520 I could feel that something was happening. 307 00:19:57,800 --> 00:20:01,240 Conversations got angrier, more out of control. 308 00:20:01,240 --> 00:20:03,800 And people really wanted to do something about it. 309 00:20:03,800 --> 00:20:06,000 So, to me, fighting for our freedom 310 00:20:06,000 --> 00:20:08,680 was more important than even my own life. 311 00:20:08,680 --> 00:20:09,720 - CHANTING 312 00:20:12,240 --> 00:20:15,720 - Leading up to the protest, I think they were all hyped up, 313 00:20:15,720 --> 00:20:17,640 they were full of adrenaline, 314 00:20:17,640 --> 00:20:20,760 and people already started to incite each other. 315 00:20:21,920 --> 00:20:24,600 In a certain moment, I was really scared, 316 00:20:24,600 --> 00:20:29,000 because the way some people looked, with the hate and aggression, 317 00:20:29,000 --> 00:20:32,520 like they were ready to even kill another man. 318 00:20:32,520 --> 00:20:34,080 - EXCLAIMATION ON VIDEO 319 00:20:34,080 --> 00:20:37,480 - I saw very young boys, maybe age 13 or 14, 320 00:20:37,480 --> 00:20:39,720 throwing rocks at police cars. 321 00:20:40,760 --> 00:20:42,240 The whole city changed, 322 00:20:42,240 --> 00:20:44,800 and it's like people just turned into animals. 323 00:20:49,760 --> 00:20:52,560 - It was a complete war zone in 20 minutes. 324 00:20:54,680 --> 00:20:57,440 They make each other crazy, using social media - 325 00:20:57,440 --> 00:20:58,760 I'm sure about that. 326 00:21:00,600 --> 00:21:03,080 That day, it was the worst day of my life. 327 00:21:05,840 --> 00:21:09,280 This is the place where I used to have my restaurant. 328 00:21:10,760 --> 00:21:13,360 And they burned it with the fireworks. 329 00:21:15,840 --> 00:21:17,720 There was shooting, as well. 330 00:21:17,720 --> 00:21:19,160 - BLASTS ECHO 331 00:21:19,160 --> 00:21:20,680 It must have been so scary. 332 00:21:20,680 --> 00:21:23,560 I mean, even when you left, it must have been scary. - Yes, of course! 333 00:21:23,560 --> 00:21:27,240 I cried, and I thought, tomorrow there's nothing left to work. 334 00:21:31,560 --> 00:21:35,880 - The pandemic was such a stark example that Facebook had 335 00:21:35,880 --> 00:21:39,120 got so much bigger than Mark Zuckerberg. 336 00:21:39,120 --> 00:21:41,400 He'd sort of lost control. 337 00:21:42,960 --> 00:21:45,280 Social media was really dividing people, 338 00:21:45,280 --> 00:21:47,640 particularly in Facebook groups. 339 00:21:47,640 --> 00:21:52,040 People were suddenly talking about hanging doctors and nurses, 340 00:21:52,040 --> 00:21:53,880 and attacking people. 341 00:21:53,880 --> 00:21:55,440 - They're very passionate... 342 00:21:56,800 --> 00:21:59,360 So, I've just been struck in the back of... 343 00:21:59,360 --> 00:22:00,680 ..back of the head by a can. 344 00:22:00,680 --> 00:22:02,760 That really hurt. 345 00:22:02,760 --> 00:22:07,480 - Wanting retribution, wanting justice, in cities across Europe. 346 00:22:11,040 --> 00:22:14,400 The violence, connected to what was happening on social media, 347 00:22:14,400 --> 00:22:16,960 that started in places like Sri Lanka, 348 00:22:16,960 --> 00:22:20,000 was coming closer and closer to home. 349 00:22:25,280 --> 00:22:28,640 - Are you willing to acknowledge that Facebook IS contributing 350 00:22:28,640 --> 00:22:32,440 to society's woes, polarisation and all the rest? 351 00:22:32,440 --> 00:22:35,360 It's pouring gasoline on the burning fire in front of us? 352 00:22:35,360 --> 00:22:37,560 - Well, look, obviously, for a platform which has, what, 353 00:22:37,560 --> 00:22:40,680 a third of the world's population on it, of course you see, you know, 354 00:22:40,680 --> 00:22:42,880 you see the good, the bad and the ugly of humanity 355 00:22:42,880 --> 00:22:44,600 show up on our platform as well. 356 00:22:44,600 --> 00:22:49,280 Our job is to mitigate the bad, reduce it, and amplify the good. 357 00:22:53,800 --> 00:22:58,120 - These user insight documents from inside Facebook come from 358 00:22:58,120 --> 00:23:01,680 the period when the company was using this new argument. 359 00:23:01,680 --> 00:23:03,760 We're not responsible for everything. 360 00:23:03,760 --> 00:23:07,160 We are "a mirror" reflecting back society. 361 00:23:07,160 --> 00:23:11,040 These documents basically talk about the way that users 362 00:23:11,040 --> 00:23:14,520 are having their psychological processes activated 363 00:23:14,520 --> 00:23:16,480 by certain types of content. 364 00:23:18,560 --> 00:23:22,880 In particular, they say this happens when users view sensitive content, 365 00:23:22,880 --> 00:23:26,360 which includes content that has an elevated risk of causing harm 366 00:23:26,360 --> 00:23:28,960 to people and inciting violence. 367 00:23:28,960 --> 00:23:32,360 So, you basically are admitting that the way that you are building 368 00:23:32,360 --> 00:23:34,880 the platforms, and the way that you are using the algorithms, 369 00:23:34,880 --> 00:23:36,480 the recommendation systems, 370 00:23:36,480 --> 00:23:39,800 ultimately result in these harms. 371 00:23:39,800 --> 00:23:42,280 And that can have a real-world consequence for us. 372 00:23:44,360 --> 00:23:46,680 - Generally felt like leadership wasn't sure what the answer was. 373 00:23:46,680 --> 00:23:50,880 But it began to calcify into a sort of defensiveness. 374 00:23:50,880 --> 00:23:55,000 "We're not responsible for all of polarisation in society." 375 00:23:55,000 --> 00:23:58,080 Nobody's saying you're responsible for ALL polarisation. 376 00:23:58,080 --> 00:24:00,560 We're just saying you contribute to it, 377 00:24:00,560 --> 00:24:03,600 and probably in ways where, like, you don't have to. 378 00:24:03,600 --> 00:24:05,440 And if you just made a few changes, 379 00:24:05,440 --> 00:24:07,720 you might not contribute to it as much. 380 00:24:07,720 --> 00:24:10,920 And, yeah, I think that was...that was dispiriting, 381 00:24:10,920 --> 00:24:13,080 because it felt like there was a window in which it was, like, 382 00:24:13,080 --> 00:24:14,520 genuinely introspective. 383 00:24:18,320 --> 00:24:22,280 - If you haven't heard of the TikTok app, you will soon. 384 00:24:22,280 --> 00:24:24,480 - It is among the fastest social networking app 385 00:24:24,480 --> 00:24:28,600 to be able to go from zero to over 200 million in registered users, 386 00:24:28,600 --> 00:24:32,160 60 million daily actives around the world, in a span of three years. 387 00:24:32,160 --> 00:24:33,800 So, that growth is incredible. 388 00:24:36,840 --> 00:24:38,640 - I was very lucky I joined TikTok. 389 00:24:39,920 --> 00:24:43,080 My job was to improve the algorithm, 390 00:24:43,080 --> 00:24:46,720 so you're able to watch 50 videos in a row, 391 00:24:46,720 --> 00:24:48,800 maybe in five or ten minutes. 392 00:24:48,800 --> 00:24:51,760 The TikTok algorithm is able to collect a lot more 393 00:24:51,760 --> 00:24:53,400 information from you. 394 00:24:53,400 --> 00:24:57,120 - Is anyone else, like, a little weirded out about how specific 395 00:24:57,120 --> 00:24:59,920 TikTok's algorithm gets for the For You page? 396 00:25:02,200 --> 00:25:05,360 - Since Facebook became Meta - what, in mid-October - 397 00:25:05,360 --> 00:25:08,160 it's really struggled to find its footing 398 00:25:08,160 --> 00:25:10,240 or gain some traction here. 399 00:25:10,240 --> 00:25:12,480 User growth is stalled or shrinking. 400 00:25:12,480 --> 00:25:14,480 - Meta was down, because TikTok is 401 00:25:14,480 --> 00:25:16,760 literally eating its lunch, right? 402 00:25:16,760 --> 00:25:18,720 Like, there is no doubt in my mind, 403 00:25:18,720 --> 00:25:19,920 Mark Zuckerberg, 404 00:25:19,920 --> 00:25:21,520 for the first time ever, 405 00:25:21,520 --> 00:25:23,200 looked vulnerable last night. 406 00:25:24,800 --> 00:25:28,640 - I actually remember a lot of fear about TikTok. 407 00:25:29,840 --> 00:25:33,120 Mark seemed like he was actually concerned 408 00:25:33,120 --> 00:25:35,360 about them overtaking Meta. 409 00:25:35,360 --> 00:25:38,200 The ethos is very much move fast and break things. 410 00:25:38,200 --> 00:25:44,040 And one way to try to compete with TikTok was to just move fast. 411 00:25:46,080 --> 00:25:47,520 - It seemed to me, at that time, 412 00:25:47,520 --> 00:25:50,240 that it was almost like this engagement arms race. 413 00:25:52,520 --> 00:25:55,400 - Mark is very paranoid about competition. 414 00:25:55,400 --> 00:25:58,520 And so, when he feels like there are potential competitive forces, 415 00:25:58,520 --> 00:26:00,720 there is no amount of money that is too much. 416 00:26:02,400 --> 00:26:05,600 - One way to try to compete with another company that's offering 417 00:26:05,600 --> 00:26:08,920 a unique product is to try to mimic it. 418 00:26:10,320 --> 00:26:12,320 - And I remember a moment when... 419 00:26:12,320 --> 00:26:14,480 ..regular headcount planning process, 420 00:26:14,480 --> 00:26:16,640 and there were some integrity teams and safety teams 421 00:26:16,640 --> 00:26:20,040 who were going to ask for two headcount to work on kid stuff 422 00:26:20,040 --> 00:26:21,640 and ten to work on elections. 423 00:26:21,640 --> 00:26:23,120 There was another team who went, 424 00:26:23,120 --> 00:26:25,960 "Oh, we just got 700 for Instagram Reels." 425 00:26:27,160 --> 00:26:28,880 I was like, "OK." 426 00:26:34,880 --> 00:26:38,520 - Any time you're introducing a new type of product, say, Reels, 427 00:26:38,520 --> 00:26:40,400 there's an elevated risk, 428 00:26:40,400 --> 00:26:43,560 because the infrastructure that existed before, 429 00:26:43,560 --> 00:26:47,200 it's either completely absent or it's very immature. 430 00:26:47,200 --> 00:26:51,920 So it's hard to prepare sufficiently in advance of that launch. 431 00:26:53,360 --> 00:26:58,000 I was frustrated, because there's a common trade off 432 00:26:58,000 --> 00:27:02,840 between, like, protecting people from harmful content and engagement. 433 00:27:07,320 --> 00:27:10,000 - There's another set of documents that speak specifically about 434 00:27:10,000 --> 00:27:12,200 the difference between TikToks and Reels. 435 00:27:12,200 --> 00:27:14,560 Reels are similar in how they work to TikToks. 436 00:27:14,560 --> 00:27:17,320 So, you're getting pushed stuff that you will show an interest in, 437 00:27:17,320 --> 00:27:19,480 not necessarily from people you know. 438 00:27:21,800 --> 00:27:24,120 And this is Meta's own study, 439 00:27:24,120 --> 00:27:26,760 with stats, shared internally, 440 00:27:26,760 --> 00:27:30,720 where they actively acknowledge that they are struggling 441 00:27:30,720 --> 00:27:33,840 to prevent harm when it comes to Reels. 442 00:27:33,840 --> 00:27:35,920 So, stuff like violence and incitement. 443 00:27:37,080 --> 00:27:40,040 So, you can see here they admit that, "Comments on Reels posts 444 00:27:40,040 --> 00:27:42,880 "have a higher Violating Hostile Speech...Prevalence." 445 00:27:42,880 --> 00:27:45,600 So, "75% higher for Bullying and Harassment, 446 00:27:45,600 --> 00:27:47,680 "19% higher for Hate Speech," 447 00:27:47,680 --> 00:27:50,240 and, "7% higher for Violence and Incitement". 448 00:27:50,240 --> 00:27:54,440 And so, essentially, what this document suggests to me 449 00:27:54,440 --> 00:27:57,640 is that they're building these new interfaces, 450 00:27:57,640 --> 00:28:00,600 they're introducing these new products to their sites 451 00:28:00,600 --> 00:28:02,720 to compete with their rivals, like TikTok, 452 00:28:02,720 --> 00:28:05,760 but yet they're not seemingly putting the right safeguards 453 00:28:05,760 --> 00:28:07,960 in place in time. 454 00:28:07,960 --> 00:28:12,080 So, their rush to evolve, to grow, to keep the user engaged 455 00:28:12,080 --> 00:28:15,800 is ultimately leaving users at risk, which they're admitting. 456 00:28:15,800 --> 00:28:17,320 Why is that happening? 457 00:28:18,600 --> 00:28:20,960 - When you're working on these integrity teams, 458 00:28:20,960 --> 00:28:25,000 they are disadvantaged, because in order to launch something 459 00:28:25,000 --> 00:28:28,440 that's going to protect people from some kind of harm 460 00:28:28,440 --> 00:28:30,560 in Reels or in Feed, 461 00:28:30,560 --> 00:28:32,120 you have to convince the team 462 00:28:32,120 --> 00:28:35,640 that owns Feed, or that owns Reels, 463 00:28:35,640 --> 00:28:37,960 to sign off on the product change 464 00:28:37,960 --> 00:28:39,240 that you want. 465 00:28:39,240 --> 00:28:42,640 But there's this power imbalance. 466 00:28:42,640 --> 00:28:46,320 They have incentives to not let those products launch, 467 00:28:46,320 --> 00:28:50,640 because toxic stuff gets more engagement than nontoxic. 468 00:28:52,480 --> 00:28:55,840 - TikTok was really trying to take up the market, 469 00:28:55,840 --> 00:29:00,720 ship out a new version, improve the algorithm, maybe every week. 470 00:29:02,120 --> 00:29:07,840 But I started noticing more and more issues of people using TikTok, 471 00:29:07,840 --> 00:29:11,280 especially more borderline content. 472 00:29:11,280 --> 00:29:14,800 For example, something subtle, like a conspiracy theory, 473 00:29:14,800 --> 00:29:19,080 or problematic content that will only be seen after you browse 474 00:29:19,080 --> 00:29:21,480 for more than a certain time. 475 00:29:21,480 --> 00:29:26,240 As the model becomes more advanced, I think the borderline content issue 476 00:29:26,240 --> 00:29:28,520 become more noticeable. 477 00:29:31,240 --> 00:29:34,640 - In 2022, I remember how social media feeds were flooded 478 00:29:34,640 --> 00:29:38,600 with borderline content, the grey-area stuff. 479 00:29:38,600 --> 00:29:42,480 Some of it, clearly, totally unacceptable. 480 00:29:42,480 --> 00:29:46,080 Maybe that's misinformation or racist hate, 481 00:29:46,080 --> 00:29:47,360 anti-Semitic abuse. 482 00:29:48,960 --> 00:29:50,880 And I was a bit confused, actually, 483 00:29:50,880 --> 00:29:53,000 because it felt like, during the pandemic, 484 00:29:53,000 --> 00:29:55,280 there had been efforts from the companies, 485 00:29:55,280 --> 00:29:58,760 the social media companies, to tackle this borderline stuff - 486 00:29:58,760 --> 00:30:02,840 and it was back again, and back again with a vengeance, really. 487 00:30:02,840 --> 00:30:04,880 Why was that happening? 488 00:30:04,880 --> 00:30:07,000 I'm so happy that you wanted to speak to me 489 00:30:07,000 --> 00:30:08,520 and to do this interview. 490 00:30:18,520 --> 00:30:21,160 So, tell me what you actually did as an engineer. 491 00:30:23,160 --> 00:30:26,600 - I worked at Meta building algorithms or programmes 492 00:30:26,600 --> 00:30:30,360 to reduce the amount of borderline content. 493 00:30:30,360 --> 00:30:33,760 And over time, the business positioning of our team changed 494 00:30:33,760 --> 00:30:37,200 towards, like, allowing more borderline content. 495 00:30:37,200 --> 00:30:40,600 - Do you think the decision was connected to 496 00:30:40,600 --> 00:30:43,480 this engagement race with TikTok? 497 00:30:43,480 --> 00:30:45,880 - Yes, it was definitely connected with that. 498 00:30:45,880 --> 00:30:47,400 You're losing to TikTok, 499 00:30:47,400 --> 00:30:49,520 and therefore your stock price must suffer. 500 00:30:49,520 --> 00:30:52,800 And then, you know, that's when people started becoming, like, 501 00:30:52,800 --> 00:30:55,840 sort of very paranoid and, you know, sort of reactive. 502 00:30:55,840 --> 00:31:00,080 And they were like, "Let's just do whatever we can to catch up. 503 00:31:00,080 --> 00:31:04,320 "Where can we get like 2%, 3% revenue for the next quarter? 504 00:31:04,320 --> 00:31:07,960 "Maybe if this stuff is really bad, then users will stop using the app. 505 00:31:07,960 --> 00:31:10,240 "If they're continuing to use the app, 506 00:31:10,240 --> 00:31:12,200 "then, by definition, it's not bad." 507 00:31:12,200 --> 00:31:14,280 - So, if someone likes harmful content, 508 00:31:14,280 --> 00:31:16,160 they'll get more harmful content? 509 00:31:16,160 --> 00:31:18,760 - Yeah. You don't care as much about it, 510 00:31:18,760 --> 00:31:22,200 because they'll continue to use it in the long term. 511 00:31:22,200 --> 00:31:25,040 - Who ultimately was making those decisions? 512 00:31:25,040 --> 00:31:26,720 - Senior VP. 513 00:31:26,720 --> 00:31:29,320 And I think he reported directly to the CEO. 514 00:31:30,440 --> 00:31:34,280 So, they sort of told us that it's because the stock price is down, 515 00:31:34,280 --> 00:31:36,760 so we might actually come back to this later. 516 00:31:36,760 --> 00:31:39,160 - It can't have been the most reassuring thing to hear, 517 00:31:39,160 --> 00:31:42,160 "Oh, we might come back to it at a later date." 518 00:31:42,160 --> 00:31:44,640 - Yeah, because there's no legal risk, you know? 519 00:31:45,640 --> 00:31:49,080 No-one actually knows what Facebook is doing. 520 00:31:49,080 --> 00:31:52,360 So there's actually no incentive for any of these companies. 521 00:31:56,920 --> 00:31:59,840 - It started with immigration, for me. 522 00:31:59,840 --> 00:32:02,240 I was sort of the perfect candidate. 523 00:32:03,880 --> 00:32:07,320 When I was 13 and 14, I was kind of struggling with a lot of 524 00:32:07,320 --> 00:32:11,440 insecurities about my ability to have friends and relationships. 525 00:32:14,280 --> 00:32:18,240 It was very easy for me to believe that the world was against me. 526 00:32:18,240 --> 00:32:21,120 It made me believe in this idea of decline. 527 00:32:22,440 --> 00:32:26,480 I started, at first, you know, the kind of entry-level, 528 00:32:26,480 --> 00:32:29,280 fact-based video essay type. 529 00:32:29,280 --> 00:32:31,520 - It is time to beef up our borders. 530 00:32:31,520 --> 00:32:34,200 - Like, "We just need to reduce immigration." 531 00:32:34,200 --> 00:32:37,320 Then it goes, slowly but surely, it starts to recommend you 532 00:32:37,320 --> 00:32:39,440 more and more channels about that. 533 00:32:39,440 --> 00:32:43,320 And then you kind of just go into the whole ecosystem at large. 534 00:32:44,920 --> 00:32:48,040 And then it goes to, "Our culture is being changed." 535 00:32:48,040 --> 00:32:52,960 - The sacrificing of our children on the altar of mass migration. 536 00:32:54,360 --> 00:32:57,400 - "The West, as we all knew it, was being replaced." 537 00:32:57,400 --> 00:33:00,600 And then you start getting recommended other fields 538 00:33:00,600 --> 00:33:04,480 in that space, where it then talks about feminism. 539 00:33:04,480 --> 00:33:07,480 And then talks about LGBT rights. 540 00:33:07,480 --> 00:33:10,600 And then, all of a sudden, you hear from a perspective 541 00:33:10,600 --> 00:33:12,600 that sounds reasonable at first 542 00:33:12,600 --> 00:33:14,960 and then, a couple months later, 543 00:33:14,960 --> 00:33:17,920 you believe literally everything in that space. 544 00:33:19,800 --> 00:33:22,000 - Makes some noise for the richest man in the world. 545 00:33:24,080 --> 00:33:29,440 - Breaking news - Twitter has just accepted Elon Musk's offer. 546 00:33:29,440 --> 00:33:31,960 - The social media platform is now in the hands 547 00:33:31,960 --> 00:33:33,640 of the world's richest man. 548 00:33:35,560 --> 00:33:38,720 - I learned on Twitter that Elon Musk was buying Twitter. 549 00:33:38,720 --> 00:33:40,240 It felt like a bizarre move. 550 00:33:40,240 --> 00:33:43,360 Everybody was trying to understand why he was doing that. 551 00:33:43,360 --> 00:33:45,960 - In a tweet, Musk wrote, "Entering Twitter HQ, 552 00:33:45,960 --> 00:33:48,280 "let that sink in." 553 00:33:48,280 --> 00:33:50,800 - I had concerns... 554 00:33:50,800 --> 00:33:52,280 ..because of the amount of 555 00:33:52,280 --> 00:33:54,800 resources Twitter was putting in at that time 556 00:33:54,800 --> 00:33:57,040 to making it a safer platform. 557 00:33:57,040 --> 00:34:00,760 Twitter was never perfect, but it was the top company priority. 558 00:34:00,760 --> 00:34:02,080 It's, you know, 559 00:34:02,080 --> 00:34:04,960 a rage-based business model in a sense. 560 00:34:04,960 --> 00:34:06,560 - If the algorithm is left unchecked, 561 00:34:06,560 --> 00:34:10,280 the algorithm could prioritise the worst possible, 562 00:34:10,280 --> 00:34:12,280 most inflammatory opinions. 563 00:34:12,280 --> 00:34:14,480 Our job was to make sure that, 564 00:34:14,480 --> 00:34:19,440 when you search for something, what you saw was reliable, true. 565 00:34:19,440 --> 00:34:21,920 We all believed in what we were doing so much. 566 00:34:25,280 --> 00:34:28,800 - REPORTER: - Never dull with Elon Musk. Expect lots of ideas, 567 00:34:28,800 --> 00:34:32,040 lots of changes and lots of uncertainty. 568 00:34:32,040 --> 00:34:33,800 - Whenever he spoke to us, 569 00:34:33,800 --> 00:34:36,040 he wanted Twitter to be more like TikTok. 570 00:34:36,040 --> 00:34:38,360 - TikTok has taken America by storm. 571 00:34:38,360 --> 00:34:40,040 - Engagement at TikTok at that time 572 00:34:40,040 --> 00:34:42,120 was just growing phenomenally. 573 00:34:42,120 --> 00:34:46,680 He was focused on making Twitter what he called hardcore. 574 00:34:50,560 --> 00:34:55,920 - I found out on November 15th that my access to my work laptop 575 00:34:55,920 --> 00:34:59,240 and my email account were cut in the middle of the night. 576 00:34:59,240 --> 00:35:00,560 - How did that feel? 577 00:35:00,560 --> 00:35:06,160 - I again was shocked because I thought at least 578 00:35:06,160 --> 00:35:07,800 the company would have some decency 579 00:35:07,800 --> 00:35:10,640 to give an official notice. 580 00:35:10,640 --> 00:35:13,960 Not something that felt so casual. 581 00:35:13,960 --> 00:35:16,240 - TUCKER CARLSON: - What percentage of your staff did you fire at Twitter? 582 00:35:16,240 --> 00:35:19,560 - I think we're about, we're about 20% of the original size. 583 00:35:19,560 --> 00:35:21,040 - So 80% left? 584 00:35:21,040 --> 00:35:22,520 - Uh, yes. 585 00:35:22,520 --> 00:35:26,360 - We had 250 designers about, of the, all design-wide, 586 00:35:26,360 --> 00:35:29,240 and there was probably eight of us left. 587 00:35:29,240 --> 00:35:31,880 The entire Trust and Safety org 588 00:35:31,880 --> 00:35:35,360 almost all was gone at that point, that wasn't a priority. 589 00:35:35,360 --> 00:35:38,720 - Turns out you don't need all that many people to run Twitter. 590 00:35:38,720 --> 00:35:40,800 - But 80%? That's a lot. 591 00:35:40,800 --> 00:35:42,120 - Um, yes. 592 00:35:42,120 --> 00:35:43,880 If you're not trying to run some sort of, 593 00:35:43,880 --> 00:35:47,080 uh, glorified activist organisation... 594 00:35:48,600 --> 00:35:50,840 ..and you don't care that much about censorship, 595 00:35:50,840 --> 00:35:53,840 then you can really let go of a lot of people, it turns out. 596 00:35:53,840 --> 00:35:55,880 - HE LAUGHS 597 00:35:57,640 --> 00:35:59,080 - I don't think any of us saw it 598 00:35:59,080 --> 00:36:02,120 as getting as bad as it got as quickly as it got. 599 00:36:02,120 --> 00:36:04,640 - Come on, get back, get back. - It's good to be back on Twitter. 600 00:36:04,640 --> 00:36:07,200 Don't believe what you read. Don't believe the mainstream. 601 00:36:07,200 --> 00:36:12,240 - Elon reinstated my Twitter account after taking it back. 602 00:36:12,240 --> 00:36:14,760 - Some of the people who'd been banned, 603 00:36:14,760 --> 00:36:18,600 all these names that had caused so much division, allowed back on. 604 00:36:25,360 --> 00:36:27,760 Because he has sole control of that platform, 605 00:36:27,760 --> 00:36:31,160 he can tweak and nudge the algorithm 606 00:36:31,160 --> 00:36:33,360 to boost certain kinds 607 00:36:33,360 --> 00:36:37,200 of opinions and push down others. 608 00:36:37,200 --> 00:36:40,880 It is complete gaming of... 609 00:36:40,880 --> 00:36:42,240 ..freedom of speech. 610 00:36:49,480 --> 00:36:52,120 - In the Internet Referral Unit, we see thousands and thousands 611 00:36:52,120 --> 00:36:55,000 of pieces of content every year being referred to us. 612 00:36:57,400 --> 00:37:00,680 - These are kind of examples that we see on a daily basis 613 00:37:00,680 --> 00:37:02,920 that shows the normalisation of 614 00:37:02,920 --> 00:37:06,680 the graphic content on social media. 615 00:37:06,680 --> 00:37:09,720 - So people spot stuff and then they send it to you guys? 616 00:37:09,720 --> 00:37:12,640 - Yeah. So posts were shared on X... 617 00:37:14,320 --> 00:37:16,920 ..describing Jews as rodents. 618 00:37:18,480 --> 00:37:19,600 - And it's not just that, really. 619 00:37:19,600 --> 00:37:21,040 I mean, the language is really extreme. 620 00:37:21,040 --> 00:37:23,000 - No, exactly, it's extreme anti-Semitism. 621 00:37:23,000 --> 00:37:25,640 It's kind of encouraging further attacks on Jews, 622 00:37:25,640 --> 00:37:28,760 viewed by thousands upon thousands of people. 623 00:37:30,600 --> 00:37:34,640 Yeah, we're certainly seeing extreme right-wing rhetoric. 624 00:37:34,640 --> 00:37:38,840 And the imagery is so graphic and so overtly racist. 625 00:37:40,880 --> 00:37:42,880 And it's kind of increased 626 00:37:42,880 --> 00:37:47,040 to such a level now of this gratuitous violence, really. 627 00:37:47,040 --> 00:37:49,880 - If a user is seeing more and more of this kind of stuff, 628 00:37:49,880 --> 00:37:53,320 you can see how they start to find even more extreme 629 00:37:53,320 --> 00:37:56,160 versions of that imagery. That to me seems akin 630 00:37:56,160 --> 00:37:59,640 to essentially how radicalisation works. 631 00:37:59,640 --> 00:38:03,880 - People are more desensitised to real-world violence, 632 00:38:03,880 --> 00:38:07,880 and they are not afraid to share their views. 633 00:38:14,280 --> 00:38:17,400 It just made me feel perpetually... 634 00:38:17,400 --> 00:38:20,840 They energised me, but not really in a good way. 635 00:38:20,840 --> 00:38:23,640 They just made me very kind of angry. 636 00:38:30,120 --> 00:38:34,960 It very much reflected the way I felt internally, 637 00:38:34,960 --> 00:38:37,080 that I was angry at the people around me. 638 00:38:37,080 --> 00:38:41,040 It was very easy for me to believe that the world was against me, 639 00:38:41,040 --> 00:38:44,440 which I think is completely central to a lot of kind of 640 00:38:44,440 --> 00:38:47,480 hard-right content on YouTube, TikTok. 641 00:38:50,960 --> 00:38:53,320 - NEWSREADER: - Senator Marco Rubio has announced bipartisan 642 00:38:53,320 --> 00:38:57,160 legislation to ban TikTok from operating in the United States. 643 00:38:59,840 --> 00:39:02,000 - Your platform should be banned. 644 00:39:02,000 --> 00:39:04,240 TikTok surveils us all, 645 00:39:04,240 --> 00:39:05,840 and the Chinese Communist Party 646 00:39:05,840 --> 00:39:10,840 is able to use this as a tool to manipulate America as a whole. 647 00:39:10,840 --> 00:39:13,000 - There were threats being made against 648 00:39:13,000 --> 00:39:16,080 TikTok because it's owned by ByteDance - a Chinese company - 649 00:39:16,080 --> 00:39:17,200 by the US government, 650 00:39:17,200 --> 00:39:19,640 that China was going to misuse sensitive data. 651 00:39:19,640 --> 00:39:23,120 ByteDance has always very strongly denied that was the case. 652 00:39:23,120 --> 00:39:26,040 But, actually, the information that TikTok has about us 653 00:39:26,040 --> 00:39:27,760 which is the most powerful 654 00:39:27,760 --> 00:39:30,800 is the way that we linger on a video, the way we scroll. 655 00:39:33,000 --> 00:39:34,080 As an engineer, 656 00:39:34,080 --> 00:39:38,160 how can you build recommendation systems that are inherently safe? 657 00:39:38,160 --> 00:39:39,400 HE EXHALES 658 00:39:39,400 --> 00:39:41,000 - That's, uh... 659 00:39:41,000 --> 00:39:42,320 I don't actually know 660 00:39:42,320 --> 00:39:46,400 the answer to that question, about how to, 661 00:39:46,400 --> 00:39:49,400 you know, build it completely safe 662 00:39:49,400 --> 00:39:54,320 when we have no control on the deep-learning algorithm. 663 00:39:54,320 --> 00:39:59,480 To us, it's still like a very black box, how internally it works. 664 00:40:01,280 --> 00:40:03,880 - Which also is, in and of itself, slightly terrifying, 665 00:40:03,880 --> 00:40:05,920 cos you clearly understand algorithms... 666 00:40:05,920 --> 00:40:08,960 LAUGHS: ..probably better than anyone we've spoken to! - Yeah. 667 00:40:08,960 --> 00:40:10,560 To be honest, we don't 668 00:40:10,560 --> 00:40:14,880 actually pay too much attention to specific contents. 669 00:40:14,880 --> 00:40:17,680 To us, all the content is just a different number, OK? 670 00:40:17,680 --> 00:40:19,400 We're the ones that are responsible 671 00:40:19,400 --> 00:40:21,920 for the recommendation and the content safety team, 672 00:40:21,920 --> 00:40:26,320 they're responsible for eliminating those bad contents. 673 00:40:26,320 --> 00:40:28,240 Like the car manufacturer, right? 674 00:40:28,240 --> 00:40:32,040 There's a team that are responsible for the acceleration, 675 00:40:32,040 --> 00:40:36,040 the engine, right? So we expect the team working on 676 00:40:36,040 --> 00:40:38,520 the braking system was doing a good job. 677 00:40:46,160 --> 00:40:47,880 - I am looking back through my messages 678 00:40:47,880 --> 00:40:51,000 with someone who currently works at TikTok. 679 00:40:51,000 --> 00:40:53,320 And they got back to me earlier this year, actually. 680 00:40:53,320 --> 00:40:56,960 They sent me a message and we've been speaking since then, 681 00:40:56,960 --> 00:40:59,920 trying to work out a time when we can meet each other. 682 00:40:59,920 --> 00:41:01,800 They only wanted to meet me in person. 683 00:41:03,240 --> 00:41:05,920 We've been messaging back and forth on encrypted channels. 684 00:41:07,520 --> 00:41:10,000 They are working in Trust and Safety, 685 00:41:10,000 --> 00:41:12,320 and there's stuff they're concerned about. 686 00:41:12,320 --> 00:41:15,240 They want people to know what's happening on the inside. 687 00:41:16,800 --> 00:41:20,160 It's so hard to get someone to want to show you stuff. 688 00:41:20,160 --> 00:41:22,720 It's a pretty high-risk thing to do when you still work there. 689 00:41:27,560 --> 00:41:28,840 Hello. - Hi. 690 00:41:30,480 --> 00:41:32,400 - So, is this your laptop? - Yeah. 691 00:41:32,400 --> 00:41:33,680 - OK. 692 00:41:33,680 --> 00:41:36,520 Why have you decided that you want to speak out, 693 00:41:36,520 --> 00:41:39,840 that you want to show me some of these documents, conversations? 694 00:41:39,840 --> 00:41:43,040 - If you're feeling guilty on a daily basis because of 695 00:41:43,040 --> 00:41:46,680 what you're instructed to do, at some point, you can decide, 696 00:41:46,680 --> 00:41:47,920 "Should I say something?" 697 00:41:47,920 --> 00:41:49,120 - What are the high-risk issues? 698 00:41:49,120 --> 00:41:51,800 What are the things that most concern you? 699 00:41:51,800 --> 00:41:53,120 - I'll show you. 700 00:41:54,400 --> 00:41:56,960 On the dashboard, the first thing you can see is volume. 701 00:41:58,280 --> 00:42:00,560 - Now that is a lot of cases. 702 00:42:00,560 --> 00:42:01,680 - Yeah. 703 00:42:01,680 --> 00:42:04,800 The platform is being used by children. 704 00:42:04,800 --> 00:42:06,560 So it's upon that platform 705 00:42:06,560 --> 00:42:09,360 to be held at a high level of accountability. 706 00:42:11,360 --> 00:42:14,600 And there's content out there linked to terrorism... 707 00:42:15,760 --> 00:42:19,040 ..sexual violence, physical violence, abuse, 708 00:42:19,040 --> 00:42:20,440 trafficking. 709 00:42:20,440 --> 00:42:21,880 It feels like it's increasing. 710 00:42:23,120 --> 00:42:24,240 And it's quite damaging 711 00:42:24,240 --> 00:42:26,840 because children are much more easily influenced. 712 00:42:26,840 --> 00:42:28,000 - So I guess the risk 713 00:42:28,000 --> 00:42:32,200 of being activated in a negative way... 714 00:42:32,200 --> 00:42:34,520 - Yeah. Especially on a generation that's literally 715 00:42:34,520 --> 00:42:36,360 hooked to this app. 716 00:42:39,120 --> 00:42:40,360 Look at this, for example. 717 00:42:41,520 --> 00:42:45,080 - Yeah. So these P2, P3, P1, 718 00:42:45,080 --> 00:42:47,760 those all show the level of priority, essentially. 719 00:42:47,760 --> 00:42:51,200 - Exactly. If you look at the country where this report comes from, 720 00:42:51,200 --> 00:42:54,120 it's very high-risk because it's a minor and involves, 721 00:42:54,120 --> 00:42:56,040 you know, sexual blackmail. - Yeah. 722 00:42:57,440 --> 00:42:59,280 - And then you can see the priority here. 723 00:42:59,280 --> 00:43:01,960 The urgency is not high at all. - Priority 2. 724 00:43:03,360 --> 00:43:05,680 - You can see all these are reported. 725 00:43:05,680 --> 00:43:07,240 - Oh, yeah. Yeah, yeah, yeah. 726 00:43:07,240 --> 00:43:09,760 - So very high-risk. - Yeah. 727 00:43:09,760 --> 00:43:12,000 - Encouraging people to commit crimes, 728 00:43:12,000 --> 00:43:15,920 terrorism or join terrorist organisations. 729 00:43:15,920 --> 00:43:18,320 And that's... - What, P2? 730 00:43:19,880 --> 00:43:20,960 - And then you have 731 00:43:20,960 --> 00:43:23,480 another case that's classed as P1, 732 00:43:23,480 --> 00:43:25,320 a high-priority case. 733 00:43:26,400 --> 00:43:30,280 - It's a case that relates to a politician, essentially. 734 00:43:30,280 --> 00:43:32,280 - Just to maintain a strong relationship, 735 00:43:32,280 --> 00:43:34,640 not because there's a high risk for the user. 736 00:43:39,520 --> 00:43:40,640 - Looking through all of 737 00:43:40,640 --> 00:43:43,080 these documents from the TikTok whistle-blower, 738 00:43:43,080 --> 00:43:46,120 I'm just struck by how many of them point to the same thing, 739 00:43:46,120 --> 00:43:47,560 which is that politics seems 740 00:43:47,560 --> 00:43:50,440 to matter more than protecting kids or teenagers. 741 00:43:51,560 --> 00:43:54,720 And the case we discussed involving the 16-year-old girl 742 00:43:54,720 --> 00:43:57,480 who reported that a TikTok account was posting 743 00:43:57,480 --> 00:44:00,880 her photo alongside explicit images. 744 00:44:00,880 --> 00:44:02,440 And this is priority number 2. 745 00:44:02,440 --> 00:44:05,520 But this is about an under-18 who is essentially having 746 00:44:05,520 --> 00:44:08,720 a concern relating to explicit images. 747 00:44:08,720 --> 00:44:12,360 And then there's a 17-year-old who is reporting being 748 00:44:12,360 --> 00:44:15,040 the victim of cyber bullying by two accounts. 749 00:44:16,080 --> 00:44:18,800 This is priority number 2. 750 00:44:18,800 --> 00:44:21,120 So take, for example, priority number 1. 751 00:44:21,120 --> 00:44:23,040 "In the picture there is a picture of a chicken..." 752 00:44:23,040 --> 00:44:24,480 And the person is a candidate 753 00:44:24,480 --> 00:44:26,680 of the Iraqi parliamentary election of 2025. 754 00:44:26,680 --> 00:44:28,920 So this is in the same country. 755 00:44:28,920 --> 00:44:30,560 You've got a teenager and then you've got 756 00:44:30,560 --> 00:44:33,240 a politician who's being affected by a picture of a chicken, 757 00:44:33,240 --> 00:44:35,240 and the politician is P1. 758 00:44:35,240 --> 00:44:38,120 The whistle-blower says that these kinds of cases happen all the time, 759 00:44:38,120 --> 00:44:39,960 and they are really uncomfortable with this. 760 00:44:39,960 --> 00:44:42,560 That when kids are at risk, when under-18s are at risk, 761 00:44:42,560 --> 00:44:45,720 it's not the first thing that they are tasked with dealing with. 762 00:44:45,720 --> 00:44:47,600 Instead, they're tasked with dealing with stuff 763 00:44:47,600 --> 00:44:49,880 about politicians or about political figures, 764 00:44:49,880 --> 00:44:52,720 or in places where it's a priority to deal with it. 765 00:44:52,720 --> 00:44:53,920 But this, for example, this is 766 00:44:53,920 --> 00:44:55,640 the next wave of stuff they've got to come to, 767 00:44:55,640 --> 00:44:57,040 and they've already got 768 00:44:57,040 --> 00:45:00,680 loads of examples of cases and harms that they're looking at. 769 00:45:03,200 --> 00:45:05,520 What would you say to a parent 770 00:45:05,520 --> 00:45:08,560 who has a kid or a teenager that is using TikTok? 771 00:45:08,560 --> 00:45:10,280 - Delete it. 772 00:45:10,280 --> 00:45:13,640 Keep them as far away as possible from the app, 773 00:45:13,640 --> 00:45:15,040 for as long as possible. 774 00:45:35,360 --> 00:45:36,760 Hello. - Hi! 775 00:45:38,080 --> 00:45:41,120 How old were you guys when you started using social media? 776 00:45:41,120 --> 00:45:43,680 - Like lockdown, so when I was like nine. 777 00:45:43,680 --> 00:45:45,040 Cos we were at home all the time 778 00:45:45,040 --> 00:45:47,320 and that's the only way I could communicate with my friends, 779 00:45:47,320 --> 00:45:48,640 so I used it a lot more. 780 00:45:48,640 --> 00:45:50,880 - LAUGHS: Who thinks that they spend the most time...? 781 00:45:50,880 --> 00:45:53,240 - It's like five hours a day, maybe. 782 00:45:53,240 --> 00:45:55,160 - I'm doing it, like, before school in the morning, 783 00:45:55,160 --> 00:45:57,120 and then as soon as I get home from school. 784 00:45:57,120 --> 00:45:59,160 It doesn't make me feel great cos I probably should have, 785 00:45:59,160 --> 00:46:02,200 like, socialised a bit more, like gone off my phone a bit more, 786 00:46:02,200 --> 00:46:03,840 but I just get addicted. 787 00:46:03,840 --> 00:46:06,800 So, like, I can't help it and I just won't stop. 788 00:46:06,800 --> 00:46:09,480 - And do you find yourself...? Does it tend to be more 789 00:46:09,480 --> 00:46:11,160 what we'd call like short-form video? 790 00:46:11,160 --> 00:46:13,040 So TikTok, Reels, that kind of thing? 791 00:46:13,040 --> 00:46:15,360 - Yeah, just scrolling there for hours. 792 00:46:15,360 --> 00:46:18,080 THEY CHUCKLE - I also do that. 793 00:46:18,080 --> 00:46:21,160 Are there things that pop up on your feed and you think, 794 00:46:21,160 --> 00:46:23,120 "I don't really want to see more of that"? 795 00:46:23,120 --> 00:46:25,000 - Probably like the same - like violence. 796 00:46:25,000 --> 00:46:27,120 - Like the bullying, the fighting, really, 797 00:46:27,120 --> 00:46:28,840 cos I don't like seeing them. 798 00:46:28,840 --> 00:46:31,360 Like, it just makes people think it's right. 799 00:46:31,360 --> 00:46:33,800 And then you report it and it doesn't come up for like 800 00:46:33,800 --> 00:46:36,640 a day or two, and then it'll just pop back up. 801 00:46:36,640 --> 00:46:38,920 - And then I keep doing it and it just keeps coming back. 802 00:46:38,920 --> 00:46:41,520 - It kind of makes me feel like, "Oh, they're not listening to me." 803 00:46:41,520 --> 00:46:44,280 I'm telling them I don't want to see this kind of stuff. 804 00:46:44,280 --> 00:46:46,440 - How do you feel when you see that kind of content? 805 00:46:46,440 --> 00:46:47,920 - Just really frustrated. Like, I've told them, 806 00:46:47,920 --> 00:46:48,960 I don't want to see this, 807 00:46:48,960 --> 00:46:50,440 I don't want to see this, over and over again, 808 00:46:50,440 --> 00:46:52,400 and they just keep it there again. 809 00:46:52,400 --> 00:46:55,040 Because it's not necessarily something people agree with, 810 00:46:55,040 --> 00:46:57,200 but it has so many views and so many likes 811 00:46:57,200 --> 00:46:59,120 that they keep putting the videos back out there 812 00:46:59,120 --> 00:47:01,960 and back out there because people react with them. 813 00:47:01,960 --> 00:47:03,240 - If I watch too much 814 00:47:03,240 --> 00:47:06,080 and then there's just something that comes up, 815 00:47:06,080 --> 00:47:09,360 some person online just trying to get popularity. 816 00:47:09,360 --> 00:47:11,120 I shouldn't get too annoyed about it, 817 00:47:11,120 --> 00:47:13,040 but it's because I'm in that sort of state 818 00:47:13,040 --> 00:47:14,760 I have a very strong opinion about it, 819 00:47:14,760 --> 00:47:18,160 though it could make me feel quite annoyed and angry. 820 00:47:20,520 --> 00:47:23,080 - What I found at the time was just how often 821 00:47:23,080 --> 00:47:25,480 we were seeing the same content again and again, 822 00:47:25,480 --> 00:47:27,360 that we didn't want to see, 823 00:47:27,360 --> 00:47:29,720 from people we do not know, 824 00:47:29,720 --> 00:47:33,760 way outside of the remit of even what you would have ever searched. 825 00:47:33,760 --> 00:47:36,440 And I think, when it comes to kids and teenagers in particular 826 00:47:36,440 --> 00:47:38,440 that was a serious problem because you're still trying 827 00:47:38,440 --> 00:47:41,680 to work out what's all right and not OK to do. 828 00:47:41,680 --> 00:47:43,480 And your feed is beginning 829 00:47:43,480 --> 00:47:46,800 to dictate the way that you look at the world. 830 00:47:50,440 --> 00:47:51,760 - I've been wanting to talk 831 00:47:51,760 --> 00:47:54,560 about this publicly for over a year now, 832 00:47:54,560 --> 00:47:56,280 and now I've finally worked up the courage 833 00:47:56,280 --> 00:48:00,280 to talk about things I've kept quite close to myself. 834 00:48:02,240 --> 00:48:05,600 I started to abandon my old racist and misogynistic views, 835 00:48:05,600 --> 00:48:08,280 as they no longer made any sense to me in my head. 836 00:48:09,320 --> 00:48:14,480 I think people need to be made aware that the amount of anger 837 00:48:14,480 --> 00:48:19,800 in a lot of hard-right circles was extreme. 838 00:48:19,800 --> 00:48:21,920 If the only thing you're hearing 839 00:48:21,920 --> 00:48:25,680 is about asylum seekers committing really bad offences, 840 00:48:25,680 --> 00:48:29,120 if you're constantly seeing that on your feed, 841 00:48:29,120 --> 00:48:31,040 then it is going to radicalise you. 842 00:48:31,040 --> 00:48:33,400 Like, 100% it is. 843 00:48:39,720 --> 00:48:42,840 - I know the whole House will be very concerned at 844 00:48:42,840 --> 00:48:47,920 the extremely serious incident that has taken place in Southport. 845 00:48:48,960 --> 00:48:51,320 - NEWSREADER: - Two children have been killed in what police called 846 00:48:51,320 --> 00:48:55,200 a ferocious stabbing at a dance class in Southport. 847 00:48:55,200 --> 00:48:58,000 - WOMAN: - And the mother that I was just consoling. 848 00:48:58,000 --> 00:48:59,080 Just the look on her face 849 00:48:59,080 --> 00:49:00,760 cos she knew something had happened to her child. 850 00:49:00,760 --> 00:49:03,520 It's like, oh, my gosh, how do you even comprehend? 851 00:49:03,520 --> 00:49:05,040 There's nothing you can do. 852 00:49:13,080 --> 00:49:15,680 - At Twitter, when we'd have a really 853 00:49:15,680 --> 00:49:19,160 serious news story like that happen, 854 00:49:19,160 --> 00:49:22,360 our job was to make sure that anyone sharing completely 855 00:49:22,360 --> 00:49:27,160 unverified information would be pushed down in the algorithm. 856 00:49:30,800 --> 00:49:32,800 The safeguards were taken away. 857 00:49:35,720 --> 00:49:37,600 Because that is the first thing that Musk did. 858 00:49:39,480 --> 00:49:42,920 There are unsubstantiated opinions speculating about 859 00:49:42,920 --> 00:49:45,600 the identity of the attacker in Southport, 860 00:49:45,600 --> 00:49:47,800 and people are listening to them. 861 00:49:51,800 --> 00:49:54,840 And that is a spider web of misinformation 862 00:49:54,840 --> 00:49:57,560 that grows and grows and grows. 863 00:49:57,560 --> 00:50:01,560 - An undocumented migrant decided to go into a Taylor Swift 864 00:50:01,560 --> 00:50:04,680 dance class today and stab six little girls. 865 00:50:04,680 --> 00:50:07,920 - We knew that people with an agenda could game the algorithm, 866 00:50:07,920 --> 00:50:10,760 could harness this huge amount of power. 867 00:50:11,800 --> 00:50:15,080 I knew exactly what was going to happen. 868 00:50:25,200 --> 00:50:27,920 - I remember seeing that on my video. 869 00:50:27,920 --> 00:50:30,080 Like, within hours, 870 00:50:30,080 --> 00:50:34,040 a whole massive crowd descend on the town. 871 00:50:34,040 --> 00:50:37,400 - A 17-year-old male from Banks in Lancashire, 872 00:50:37,400 --> 00:50:39,320 who is originally from Cardiff, 873 00:50:39,320 --> 00:50:41,760 has been arrested on suspicion of murder. 874 00:50:41,760 --> 00:50:46,160 - The news is stepping in and saying, "We know who did this," 875 00:50:46,160 --> 00:50:47,680 but by then it's too late. 876 00:50:55,800 --> 00:50:59,000 - And then the mosque there is the first thing that gets desecrated. 877 00:50:59,000 --> 00:51:02,200 - CHEERING AND YELLING 878 00:51:07,160 --> 00:51:10,520 - Get back! - Get back now! Get back! 879 00:51:10,520 --> 00:51:14,440 - And all of this is self-enforcing, completely circular. 880 00:51:14,440 --> 00:51:16,320 Because when these things were happening, 881 00:51:16,320 --> 00:51:20,400 they're then being reported on social media platforms, 882 00:51:20,400 --> 00:51:23,320 and it spins faster and faster 883 00:51:23,320 --> 00:51:25,880 and people get angrier and angrier. 884 00:51:25,880 --> 00:51:28,280 - SHOUTING 885 00:51:28,280 --> 00:51:30,800 HELICOPTER OVERHEAD 886 00:51:30,800 --> 00:51:32,320 MAN: - Fucking hell. 887 00:51:33,920 --> 00:51:36,280 - BEEPING, UPROAR 888 00:51:52,280 --> 00:51:55,640 And I remember at that time wanting to get to 889 00:51:55,640 --> 00:51:58,680 the bottom of who was behind these false rumours, 890 00:51:58,680 --> 00:52:01,320 allegations suggesting that the attacker 891 00:52:01,320 --> 00:52:02,520 is an asylum seeker, 892 00:52:02,520 --> 00:52:05,360 that they arrived by boat to the UK, that they're Muslim. 893 00:52:05,360 --> 00:52:06,760 And I think now, if I look... 894 00:52:08,960 --> 00:52:11,840 ..one of the first places that published some of 895 00:52:11,840 --> 00:52:15,120 the speculation was called Channel3 Now. 896 00:52:15,120 --> 00:52:18,120 It was just sort of pumping out stories. 897 00:52:18,120 --> 00:52:19,480 And I tracked down someone, 898 00:52:19,480 --> 00:52:21,560 a man called Farhan in Pakistan, 899 00:52:21,560 --> 00:52:23,520 and I finally got this email back. 900 00:52:24,920 --> 00:52:27,320 We're only just trying to, you know, put news out. 901 00:52:27,320 --> 00:52:29,440 Essentially, we're kind of trying to get clicks 902 00:52:29,440 --> 00:52:30,960 and we didn't mean to get it wrong. 903 00:52:32,400 --> 00:52:34,360 This is clearly a product of the algorithms, 904 00:52:34,360 --> 00:52:35,640 the social media systems, 905 00:52:35,640 --> 00:52:37,760 because Channel3 Now wouldn't have wanted 906 00:52:37,760 --> 00:52:40,520 to post this content had it not known 907 00:52:40,520 --> 00:52:43,200 that it was going to get views and clicks and likes 908 00:52:43,200 --> 00:52:45,640 if it posted it as quickly as it possibly could, 909 00:52:45,640 --> 00:52:47,080 even if it was wrong. And, in fact, 910 00:52:47,080 --> 00:52:49,120 it being wrong meant it got more likes, 911 00:52:49,120 --> 00:52:51,760 more views, more clicks because it got engagement. 912 00:52:51,760 --> 00:52:53,880 It sparked a reaction, it was emotive. 913 00:52:53,880 --> 00:52:56,280 UPROAR 914 00:52:58,880 --> 00:53:00,920 HELICOPTER OVERHEAD 915 00:53:02,280 --> 00:53:04,520 Southport felt like a real shift. 916 00:53:04,520 --> 00:53:08,200 Like the violence connected to what was happening on social media, 917 00:53:08,200 --> 00:53:09,560 it's no longer far away. 918 00:53:15,360 --> 00:53:17,960 - We are the canary in the coal mine. 919 00:53:17,960 --> 00:53:20,480 A lot of what goes on and wrong on social media 920 00:53:20,480 --> 00:53:21,720 was trialled and tested 921 00:53:21,720 --> 00:53:24,200 in countries like Sri Lanka, because nobody cares about us. 922 00:53:27,080 --> 00:53:30,720 All of our markets are active experiments, 923 00:53:30,720 --> 00:53:35,520 and so what is now much more evident in the UK, 924 00:53:35,520 --> 00:53:37,520 in other parts of the world, 925 00:53:37,520 --> 00:53:42,160 was very evident in countries like Sri Lanka years prior. 926 00:53:44,080 --> 00:53:47,640 So what I have warned against, what I have cried about, 927 00:53:47,640 --> 00:53:50,160 suggests that this is not a Sri Lankan problem. 928 00:53:50,160 --> 00:53:52,120 It is a social media problem. 929 00:53:52,120 --> 00:53:54,280 It is a Silicon Valley problem. 930 00:53:54,280 --> 00:53:57,480 It is a governance problem. It is a democracy problem. 931 00:53:57,480 --> 00:53:59,280 It is a problem for all of us. 932 00:54:03,080 --> 00:54:07,400 Algorithms now impact all our lives. 933 00:54:07,400 --> 00:54:10,480 - CROWD CHANT 934 00:54:13,080 --> 00:54:15,600 REPORTER: - Immigration centres targeted nationwide. 935 00:54:16,640 --> 00:54:18,160 Police overwhelmed. 936 00:54:18,160 --> 00:54:20,600 - SIR KEIR STARMER: - There have already been hundreds of arrests. 937 00:54:22,560 --> 00:54:27,160 The criminal law applies online as well as offline. 938 00:54:28,760 --> 00:54:32,000 Around this time at Meta, there was a shift. 939 00:54:32,000 --> 00:54:33,840 - It's time to get back to our roots around 940 00:54:33,840 --> 00:54:36,160 free expression on Facebook and Instagram. 941 00:54:36,160 --> 00:54:38,640 - They announced that they were killing the fact-checking programme, 942 00:54:38,640 --> 00:54:42,160 and they cut back very severely on content moderation. 943 00:54:42,160 --> 00:54:44,280 - We're going to simplify our content policies 944 00:54:44,280 --> 00:54:46,360 and get rid of a bunch of restrictions on topics 945 00:54:46,360 --> 00:54:48,160 like immigration and gender 946 00:54:48,160 --> 00:54:50,840 that are just out of touch with mainstream discourse. 947 00:54:52,080 --> 00:54:55,240 - It felt like they lost the commitment 948 00:54:55,240 --> 00:54:57,120 to combating a lot of 949 00:54:57,120 --> 00:55:00,720 these harms and seeming to just not really care 950 00:55:00,720 --> 00:55:02,720 about actually listening to the data 951 00:55:02,720 --> 00:55:05,320 and actually trying to solve problems. 952 00:55:05,320 --> 00:55:07,320 - My biggest worry is 953 00:55:07,320 --> 00:55:09,400 the politicisation of these platforms. 954 00:55:09,400 --> 00:55:12,440 Mark was generally trying to be neutral, 955 00:55:12,440 --> 00:55:14,160 but now Elon, I think, 956 00:55:14,160 --> 00:55:18,520 specifically treats X as a political outlet for his views. 957 00:55:18,520 --> 00:55:21,120 And I don't think any single company should be allowed 958 00:55:21,120 --> 00:55:26,040 to have this amount of influence and power over 4 billion people. 959 00:55:26,040 --> 00:55:29,120 - Can I get a picture, Charlie? - Charlie! - Charlie. 960 00:55:29,120 --> 00:55:31,600 - I want Trump to be president. Trump. 961 00:55:32,680 --> 00:55:34,480 - It now feels like we're in a world 962 00:55:34,480 --> 00:55:37,840 increasingly inhabited by a new type of individual. 963 00:55:38,960 --> 00:55:40,800 - Would you like to have a conversation? 964 00:55:40,800 --> 00:55:42,360 - Sure. Absolutely. Yeah, yeah. 965 00:55:42,360 --> 00:55:44,640 - OK. - Um, I don't like you. 966 00:55:44,640 --> 00:55:46,680 I think you spread hate, I think you spread bigotry. 967 00:55:46,680 --> 00:55:48,320 I think you piss a lot of people off. 968 00:55:49,520 --> 00:55:51,360 - We record all of it so that we put it on 969 00:55:51,360 --> 00:55:54,000 the internet so people can see these ideas collide. 970 00:55:54,000 --> 00:55:56,480 When people stop talking, that's when you get violence. 971 00:55:59,160 --> 00:56:01,120 - Influencers whose capacity 972 00:56:01,120 --> 00:56:06,080 for generating engagement online is a huge part of their fame. 973 00:56:13,000 --> 00:56:15,040 GUNSHOT 974 00:56:15,040 --> 00:56:16,120 - Oh, my God. 975 00:56:17,080 --> 00:56:18,520 - Oh, my gosh, he was just shot. 976 00:56:18,520 --> 00:56:20,440 He was just shot! 977 00:56:20,440 --> 00:56:22,800 - NEWSREADER: - Charlie Kirk was shot dead while 978 00:56:22,800 --> 00:56:25,240 addressing students on a campus in Utah. 979 00:56:25,240 --> 00:56:27,080 - Videos of his assassination 980 00:56:27,080 --> 00:56:29,280 reached millions across the globe. 981 00:56:29,280 --> 00:56:32,920 - The killer's reasons were initially unclear. 982 00:56:32,920 --> 00:56:34,520 - Right now, federal investigators say 983 00:56:34,520 --> 00:56:37,120 they are talking to this man, Tyler Robinson, 984 00:56:37,120 --> 00:56:40,280 about why he might want to kill Charlie Kirk. 985 00:56:40,280 --> 00:56:44,760 - Conspiracies about who he is and what he believed are flourishing. 986 00:56:44,760 --> 00:56:47,960 - But he'd left some clues at the scene. 987 00:56:47,960 --> 00:56:51,000 - The inscriptions carved onto the bullet casings 988 00:56:51,000 --> 00:56:54,360 that were believed to be owned by the alleged shooter. 989 00:56:54,360 --> 00:56:56,440 - You'd only understand them if you've spent 990 00:56:56,440 --> 00:56:59,160 a lot of time on social media. 991 00:56:59,160 --> 00:57:01,680 - They contain a number of seemingly sarcastic, 992 00:57:01,680 --> 00:57:04,480 irreverent phrases that are common in gaming 993 00:57:04,480 --> 00:57:06,120 and online communities. 994 00:57:07,480 --> 00:57:11,080 - A seemingly ordinary student wrapped up in a dark, 995 00:57:11,080 --> 00:57:13,560 nihilistic meme culture. 996 00:57:13,560 --> 00:57:15,280 Did the killer have one eye 997 00:57:15,280 --> 00:57:18,840 on the online reaction provoked by his violence? 998 00:57:18,840 --> 00:57:23,320 Are we witnessing a new ideology of engagement? 999 00:57:25,640 --> 00:57:29,000 Algorithmic systems shaping politics, 1000 00:57:29,000 --> 00:57:30,640 shaping society, 1001 00:57:30,640 --> 00:57:33,200 shaping how events unfold. 1002 00:57:37,720 --> 00:57:40,120 - If I was parenting young children right now, 1003 00:57:40,120 --> 00:57:45,400 I would encourage them to spend as little time online as possible. 1004 00:57:45,400 --> 00:57:48,080 - We have lost lives to this. 1005 00:57:48,080 --> 00:57:50,520 We know where this goes. 1006 00:57:50,520 --> 00:57:52,560 - Social media platforms have 1007 00:57:52,560 --> 00:57:56,080 an essential responsibility to the truth. 1008 00:57:56,080 --> 00:57:58,680 - If we want this sort of accountability and transparency, 1009 00:57:58,680 --> 00:58:00,680 we have to make it legally required. 1010 00:58:00,680 --> 00:58:02,440 - Problems are going to get worse. 1011 00:58:02,440 --> 00:58:08,600 - I can't continue existing without trying to fight. 1012 00:59:05,800 --> 00:59:08,160 - For more insights from me and the experts 1013 00:59:08,160 --> 00:59:12,520 at the Open University about how to stay alert to algorithms, 1014 00:59:12,520 --> 00:59:14,760 scan the QR code on screen, 1015 00:59:14,760 --> 00:59:20,560 or visit connect.open.ac.uk/ragemachine.