Pełna transkrypcja odcinka Going to the Ballot Bot

voice, and I wanted to start a newsletter. 00:04:10.300 --> 00:04:11.591 I had a journalism background. 00:04:11.591 --> 00:04:12.668 That's what I went to school for. 00:04:12.950 --> 00:04:14.600 and the book was on my mind, too. 00:04:14.617 --> 00:04:18.476 But in talking to people, I realized a couple things in'21. 00:04:18.476 --> 00:04:24.387 One, I needed a little bit more time and distance from everything that had gone on at Facebook to have a, clearer-eyed view of it. 00:04:24.800 --> 00:04:42.973 Two, that all authors kinda need their own platform to sell books, and so I was like,"I should focus on the newsletter maybe first and building that, that up." And then it was like, okay, I want this out before the midterms- In 2026, as we go into the presidential, I knew that technology's always a big story in elections right now. 00:04:42.975 --> 00:04:49.023 I had no idea how big AI was going to, get at first in doing all of it. 00:04:49.047 --> 00:04:51.641 And I wanted to do this for a couple reasons. 00:04:51.641 --> 00:04:53.639 One, one is a selfish reason. 00:04:53.656 --> 00:04:55.446 I didn't want anyone else telling my story. 00:04:55.838 --> 00:04:58.120 There is a lot of reporting out there. 00:04:58.120 --> 00:04:59.889 There's a lot of conversations out there. 00:05:00.108 --> 00:05:04.990 There's a lot of other books out there about what happened during this timeframe, all valid. 00:05:04.999 --> 00:05:10.341 Like, everybody can have their own experiences, but I didn't want anybody else telling my experience. 00:05:10.360 --> 00:05:25.370 And when I talk to friends that are historians and stuff, and I've asked, like,"When you go back 50, 100 years, what do you wish you had?" And they were like,"More stories." I also was hoping that for those of us in the trust and safety field, telling our story can be very fraught. 00:05:25.384 --> 00:05:29.949 You and Mike have talked about this, Yoel Vaishnavhi and others about this. 00:05:29.971 --> 00:05:35.867 And through my newsletter and through the book, I wanted to show just another way that we might be able to tell our story. 00:05:35.882 --> 00:05:39.315 And yes, there's always that, chance that you're gonna have a target on your back. 00:05:39.536 --> 00:05:42.548 and that's always a, you know, a trade-off of doing this. 00:05:42.557 --> 00:05:44.560 But I do think telling our stories... 00:05:44.834 --> 00:05:59.595 And again, I wanted to bring people into the rooms of not what Mark was doing, not what Cheryl was doing, but those of us doing it day to day, how we were trying to make these decisions and how we were trying to navigate these times without a playbook. 00:05:59.625 --> 00:06:12.533 And I dedicate my book to those in the arena, which is a speech by, Teddy Roosevelt that I used a lot when I was at Facebook with our, teams because, you know, it talks about, being in the arena. 00:06:12.541 --> 00:06:17.990 And there's a lot of valor in that and being able to, to be working in this tough environment. 00:06:18.005 --> 00:06:35.233 And so I wanted, I wanted that story of not just mine, but as best as I could describe it to others, what trust and safety teams do inside of these companies, and that there are people who really care about getting this right, but there's also a lot of really hard trade-offs that come with it. 00:06:35.252 --> 00:06:42.454 And then lastly, if I can, and sorry in this long answer, as we go into this AI era, I think there's a lot of things we can learn from the social media era. 00:06:42.735 --> 00:06:48.608 we spent a lot of time with rose-colored glasses on the first 15 years of social media, not talking about the risks. 00:06:48.718 --> 00:06:49.526 Now we are. 00:06:49.766 --> 00:06:51.041 I think we're seeing some of that play out. 00:06:51.041 --> 00:06:53.122 We're gonna talk about that with some of the stories today. 00:06:53.156 --> 00:07:01.682 but I'm hoping that there's some lessons here to be learned from what we got wrong, what we got right, that we can think about as we go into this next stage. 00:07:02.048 --> 00:07:02.358 Yeah. 00:07:02.502 --> 00:07:09.500 And I'm, I'm really glad that you've brought the stories you have today to the podcast and that you're gonna be able to kind of reflect on those with that experience in mind. 00:07:09.505 --> 00:07:22.805 But before we kind of dive into those, you touch on a lot of issues in the book that we talk about on the podcast, the importance of nuance, the trade-offs that happen, you know, in roles like yours, the personal toll that, you suffered as a result. 00:07:22.857 --> 00:07:25.858 you mentioned at one point, you know, you don't feel guilt about it. 00:07:25.858 --> 00:07:28.711 You often get asked how you reflect on, on that era. 00:07:28.855 --> 00:07:31.824 How do you look back on, on that time now? 00:07:32.223 --> 00:07:34.019 I mean, it's mixed, right? 00:07:34.036 --> 00:07:41.324 And like you said, I struggle to find words for it because there's a lot of stuff I'm really proud of. 00:07:41.593 --> 00:07:59.153 There's a lot of stuff, like in the early d- you know, right after 2016, Facebook put a lot of time and money into building the civic integrity team, building the political ad transparency tools, building the election war rooms that we've now seen other companies adopt and, and do that have kind of become commonplace. 00:07:59.161 --> 00:08:04.221 And frankly, I think the political ad transparency tool is an example of self-regulation gone right. 00:08:04.848 --> 00:08:14.331 You know, we often talk about like, not necessarily trusting it, but that is something that, they could have just been like,"No political ads whatsoever, and we're not going to, to do that," and they didn't. 00:08:14.651 --> 00:08:25.781 and I'm really proud of, the teams we built and, like, you can see this Facebook diaspora and the civic integrity diaspora and other companies too, Twitter, et cetera, but, like, as they go out to

other companies, you can really see that work. 00:08:26.093 --> 00:08:39.758 But then there's a lot of things too that, in hindsight, like I really think back to my early political campaigning days where we were doing gimmicks like dressing interns up in dolphin suits, like'cause John Kerry, who was the nominee at the time in'04, was a flip-flopper. 00:08:40.092 --> 00:08:41.302 We just wanted more clicks. 00:08:41.402 --> 00:08:43.038 We thought more clicks were better, right? 00:08:43.057 --> 00:08:44.405 That means more people are watching. 00:08:44.407 --> 00:08:55.442 More is better, so let's keep doing whatever keeps getting us more clicks, not realizing there w- we're the proverbial frog in the boiling water, that over time... 00:08:55.450 --> 00:09:10.528 And I realized this in the book and writing it, like looking at this 25-year span from the 2000 election to the CEOs at the inauguration in'25, how when you only think short-term, you can't necessarily realize what's happening to you. 00:09:10.899 --> 00:09:26.317 And then when you think about long-term about where that sent us, there's a lot of things that I would potentially change, like how we think about measurement and, and all of that, that I'm like,"Okay, AI era, can we not just think about..." I know it's so hard to even know where the world's gonna be like in two days, let 00:09:26.485 --> 00:09:26.869 Yeah 00:09:27.229 --> 00:09:35.129 two, three years, but I'm trying to keep that long-term perspective in mind as well, so we can try to think through some of the unintended consequences. 00:09:35.480 --> 00:09:35.953 For sure. 00:09:36.159 --> 00:09:36.528 For sure. 00:09:36.547 --> 00:09:37.839 I know, and we'll touch on that today. 00:09:38.114 --> 00:09:44.317 I've got to ask as well, you know, other former Meta employees have faced issues when writing books recently. 00:09:44.668 --> 00:09:49.948 Sarah Wynn-Williams is, the obvious example, and I wanted to just ask, like, what has been the reaction from the company? 00:09:50.267 --> 00:09:51.970 Have you had any, pushback? 00:09:52.066 --> 00:09:53.250 what's that looked like? 00:09:53.625 --> 00:09:55.519 I've not heard anything from the company. 00:09:55.548 --> 00:09:58.620 I've not been quiet about the fact that I'm writing this book. 00:09:58.636 --> 00:10:04.625 I've been documenting it on my Substack for years now that I've been doing this. 00:10:04.767 --> 00:10:11.698 I think it is better for both of us that I can honestly say this is my account without company approval, et cetera. 00:10:12.005 --> 00:10:15.100 I also didn't have a non-disparagement like Sarah did. 00:10:15.164 --> 00:10:22.041 and one of the things, there's a nearly 350 citations in my book, and also I put all of those on my Substack as well. 00:10:22.041 --> 00:10:28.910 So if anybody, researchers listening, et cetera, you want links to all these things, I've put it all up on my Substack for you. 00:10:29.166 --> 00:10:44.491 with all the leaks and

court cases and everything, a lot of the things I worked on have become public, and so I really wanted to make sure that any claims or things I was talking about had a public thing that I could point to around all of that. 00:10:44.518 --> 00:10:46.793 And I'm still getting reactions from people. 00:10:46.793 --> 00:10:47.804 The book is not out yet. 00:10:47.804 --> 00:10:49.625 It comes out on Tuesday, October 6th. 00:10:49.922 --> 00:10:52.511 so early reads I'm getting good feedback. 00:10:52.557 --> 00:11:03.235 and so, I'm, excited and a little terrified to be putting this out into the world, but I'm hoping that people can s- I'm hoping that my intentions are seen by people who are reading it. 00:11:03.722 --> 00:11:05.535 Yeah, and, and I'm halfway through. 00:11:05.535 --> 00:11:14.048 I haven't, I haven't finished the book, but it's a fantastic read and it told me lots of stuff that I didn't know, having paid a fair bit of attention over the last 10 years. 00:11:14.253 --> 00:11:18.309 So I think that speaks to your experience and your, your ability to get it down on paper. 00:11:18.311 --> 00:11:21.640 So, we'll, we'll include links in the show notes to where you can get the book. 00:11:21.932 --> 00:11:22.831 It's out on Tuesday. 00:11:22.905 --> 00:11:24.313 listeners, go and buy it. 00:11:24.746 --> 00:11:30.568 for those people who, who also have some, spare change for us, you can become a Patreon supporter of Ctrl Alt Speech. 00:11:30.780 --> 00:11:40.119 you get both a extended episode and earlier episodes if you become either an insider supporter, or a regular supporter of the podcast. 00:11:40.389 --> 00:11:43.451 Uh, you can go and do that at patreon.com/ctrlaltspeech. 00:11:43.456 --> 00:11:48.168 We've had a handful of people over the last week and lots of great reactions to last week's episode. 00:11:48.509 --> 00:11:52.524 and if you just wanna support us in other ways, rate and review us wherever you get your podcasts. 00:11:52.796 --> 00:11:53.822 It's a big help to us. 00:11:54.248 --> 00:12:00.043 Katie, we're gonna jump into a, a story that has been very much on your radar, I know, with your, background in elections. 00:12:00.287 --> 00:12:03.461 that is the Brazilian election, and that's happening on Sunday. 00:12:03.730 --> 00:12:10.764 214 million Brazilians are gonna go to the polls to vote for governors, senators, but also the president. 00:12:10.792 --> 00:12:17.365 And we're kind of, awaiting a very tight election once again between the incumbent President Lula and a Bolsonaro. 00:12:17.663 --> 00:12:21.426 This time it's not Jair Bolsonaro, it's his son Flavio. 00:12:21.803 --> 00:12:34.238 And, you know, I wanted to kind of get your thoughts on, generally the kind of election writ large, and then we can kind of dive into a couple of, you know, sub-threads that you've been paying close attention to, which I think are gonna potentially im- impact the outcome of the election. 00:12:34.677 --> 00:12:37.408 so yeah, what, are you thinking about, about it as a whole? 00:12:37.813 --> 00:12:46.190 Well, one thing to just note for folks, is that, Sunday is the first round, and then there will likely be a second round, I think, on October 25th, if I remember correctly. 00:12:46.509 --> 00:12:53.246 polls are showing an incredibly close race between Lula and Bolsonaro, which doesn't surprise me. 00:12:53.259 --> 00:13:02.730 And thinking back to it was actually between the two rounds that we saw at Facebook some of the more hate speech and stuff pop up. 00:13:02.744 --> 00:13:19.097 And I'll be particularly interested to see in what the US does if the race is super close,'cause they are-- the Trump administration obviously favors Bolsonaro, around all of this and are not shy in getting involved in, international campaigns. 00:13:19.354 --> 00:13:36.796 but I've a- I've been paying a lot of close attention to what they've been doing around AI, AI and the candidates, what the tech platforms can do, how those AI models are responding to this, but then also the role of, like, AI influencers and stuff in this too, because I think that we are seeing... 00:13:37.110 --> 00:13:39.269 You know, in the US, we're about to go through a midterms. 00:13:39.677 --> 00:13:57.230 A midterms is different than a presidential, so one of the reasons I'm watching the Brazil one closely is for many reasons, but another is as the US goes into a presidential and also as we go into French elections next year and all of that, I think we're gonna be seeing some glimmers of what we might be seeing spread across the internet 00:13:57.672 --> 00:13:57.962 Yeah. 00:13:58.124 --> 00:14:00.575 And so talk us through some of those kind of threads then. 00:14:00.575 --> 00:14:06.167 The, the deepfakes one is particularly interesting, I think, because this is not a, a new phenomenon. 00:14:06.168 --> 00:14:16.571 You know, deepfakes and elections have gone hand in hand for a, a number of years now, and, but there is a kind of e- evolution of these AI models which make that even more potentially disruptive this time round, right? 00:14:16.965 --> 00:14:18.735 Yeah, there's two things that I'm really watching. 00:14:18.745 --> 00:14:29.110 So on the deepfakes thing, what was interesting is that-- So Brazil has, a very strong electoral commission, one that is very trusted inside the country. 00:14:29.111 --> 00:14:35.611 They've done a fantastic job, and they do-- they've done memos of, um, MOUs with the tech companies. 00:14:35.701 --> 00:14:40.259 My first election there was 2014, but I think the first MOU was 2018 election. 00:14:40.445 --> 00:14:45.715 but they banned candidate deepfakes, or deepfakes, I think even on behalf of the candidate. 00:14:45.994 --> 00:14:54.644 Now, what was interesting though was that Flavio did a deepfake of his dad because Jair's in jail and can't show up to a party rally. 00:14:55.033 --> 00:15:05.205 And so that was the first, test of this new law, and the

Superior Court decided that that was okay because they weren't putting money behind it and it was a party event. 00:15:05.221 --> 00:15:08.399 And I'm like, immediately they're, twisting themselves into pretzels. 00:15:08.745 --> 00:15:12.748 Like, how are-- Like, you're immediately giving folks a loophole around it all. 00:15:12.758 --> 00:15:17.035 So, like, they were gonna try to find it anyways, but what was the point? 00:15:17.054 --> 00:15:20.782 Like, everybody knew that that deepfake had happened and you're doing this law. 00:15:21.037 --> 00:15:30.476 It just sort of boggled my mind that, like, your first opportunity of trying to enforce this law and you didn't even get into that level of nuance when you were creating it. 00:15:30.835 --> 00:15:37.436 Yeah, and, and that's kind of goes back to something that I often think about, which is that kind of disinformation comes from the top in, in a lot of situations. 00:15:37.436 --> 00:15:43.688 You know, people think that it's, Macedonian teenagers manufacturing this in, in their rooms, and, and that does happen, obviously. 00:15:43.688 --> 00:15:55.057 But you have political figures who are often responsible, as in this case, for, creating a narrative or a deepfake that actually, runs counter to regulation or, authority. 00:15:55.065 --> 00:15:56.706 So, that's really fascinating. 00:15:56.854 --> 00:15:58.698 what was the kind of outcome in that case? 00:15:58.738 --> 00:16:00.448 Or like where did that deepfake go? 00:16:00.928 --> 00:16:03.999 I think they said it was allowed, so, and it had already happened, right? 00:16:04.003 --> 00:16:06.533 So I haven't seen more on that front. 00:16:06.815 --> 00:16:19.753 But the other thing that I thought was super interesting, so this happened before the deepfake law, which was that, I can't remember if it was the Superior Court or who, made this rule, but that it said that the AI platforms could not rank candidates. 00:16:19.996 --> 00:16:23.100 but they didn't clarify about what ranking meant. 00:16:23.274 --> 00:16:28.682 So my guess is their intention was don't tell people who to vote for, right? 00:16:28.774 --> 00:16:35.298 But if you search for something, just who are the candidates running for president, and if you do it in alphabetical order, is that a ranking? 00:16:35.668 --> 00:16:36.409 Is it not? 00:16:36.472 --> 00:16:49.269 And so a lot of the platforms have, and, the Free Systems, Substack just did a great post on this, like right before you and I started recording, kinda doing a deep dive of how the AI platforms are handling election questions in Brazil. 00:16:49.724 --> 00:17:00.933 And a lot of them, the newer models are not even helping them, voters understand who's gonna be on their ballot, where, when, and how to vote, because these platforms are so worried about running afoul of these rules. 00:17:01.201 --> 00:17:04.907 'Cause another thing people might not know is Brazil's a very litigious country. 00:17:05.194 --> 00:17:07.107 People can bring suits very quickly. 00:17:07.509 --> 00:17:16.807 I remember we used to have to hire hundreds of lawyers to be on call to help us to handle all the cases people might be bringing, against Facebook during an election. 00:17:17.126 --> 00:17:18.884 so on one hand, it's understandable. 00:17:18.884 --> 00:17:30.714 The law was meant to try to prevent these AI systems from improperly influencing voters, but what it's actually had the effect of doing is actually disenfranchising them and making it harder for them to get information. 00:17:30.963 --> 00:17:32.765 'Cause most people go to, like, Google. 00:17:32.905 --> 00:17:38.567 They still go to search to find information, and Google, when you search, is ranking things. 00:17:38.733 --> 00:17:40.535 That's the nature of search. 00:17:40.551 --> 00:17:47.615 And so these are some of the questions as we continue to see other countries looking at some of these laws of what folks are doing. 00:17:47.869 --> 00:18:10.058 Again, pay attention to the unintended consequences, and this is where I also oftentimes tell governments, others too, there are plenty of trust and safety people outside of the companies that can help you to think through is what you're creating a law, is your intention matching what the law is that you're passing, and is that how the companies are actually going to respond to it or not? 00:18:10.513 --> 00:18:10.903 Yeah. 00:18:11.065 --> 00:18:11.357 Yeah. 00:18:11.357 --> 00:18:18.544 I mean, it's, it's so often the case that governments are trying to do things for the right reasons and having these unintended consequences that only play out after the fact. 00:18:18.817 --> 00:18:32.128 the other kind of, I guess, subplot here is around ads, on platforms and, and the process by which, you know, Meta approves ads, both of the political candidates, but by other individuals who are looking to influence election. 00:18:32.405 --> 00:18:38.121 there was a period, I think, where President Lula was prevented from, from posting ads on the platform. 00:18:38.444 --> 00:18:53.284 it was reported by a few different outlets and I saw some research from a, nonprofit called Echo, who showed that there was AI-generated ads that was inciting violence and spreading election disinformation that were being posted by other individuals and they were being allowed through. 00:18:53.603 --> 00:18:57.053 this feels like something we kind of, we never get past, is that ability to 00:18:57.053 --> 00:18:57.347 Yeah 00:18:57.673 --> 00:18:59.940 political ads and, and ads that violate policy 00:19:00.153 --> 00:19:03.167 defining it's really, really, really hard. 00:19:03.244 --> 00:19:03.953 Right, right. 00:19:04.125 --> 00:19:15.275 talk about your experience in that regard and, and, you know, c-'cause you would've, been in the room with politicians who wanted to set up campaigns to influence a set of voters on a particular issue or topic 00:19:15.470 --> 00:19:18.469 I was

writing the policy about what it meant to be a political ad. 00:19:18.691 --> 00:19:23.538 so first I want to remind folks, because this also came up with the Elon documentary, here in the States. 00:19:23.557 --> 00:19:26.605 But, people have to remember that when we were... 00:19:26.676 --> 00:19:32.317 Typically, uh, I'll start with America, so I know, forgive me that we started with America, but we did. 00:19:32.550 --> 00:19:37.173 at least with American campaign finance, and I think in most countries, you start with who you are. 00:19:37.173 --> 00:19:42.786 You are a candidate, you are a super PAC, thus your ads must follow these rules and have these disclaimers. 00:19:43.122 --> 00:19:53.400 I think people forget that part of the reason that we created the political ad transparency tools was to catch things like the Russian Internet Research Agency ads that were happening in 2016. 00:19:53.678 --> 00:19:55.340 Most of those were issue-based. 00:19:55.423 --> 00:19:57.903 They were not talking about the candidates themselves. 00:19:58.175 --> 00:20:01.917 So we had to flip it on its head, and we had to start with the advertisement. 00:20:02.165 --> 00:20:03.163 Is it political? 00:20:03.189 --> 00:20:04.458 Is it about an issue? 00:20:04.487 --> 00:20:08.865 And then, doesn't matter who's running it, then you need to go through all these hoops. 00:20:09.220 --> 00:20:13.057 So we had a ton of people who were like,"Wait, we're not political. 00:20:13.151 --> 00:20:18.862 why are you calling us political?" And I'm like,"But you're running an ad about environmental, issues or something like this. 00:20:18.883 --> 00:20:25.055 Like, that's a political..." You know, we've took a very broad definition and then we, we narrowed it down. 00:20:25.309 --> 00:20:29.314 But then we had to go, as we went country by country, every country has different rules. 00:20:29.325 --> 00:20:39.162 So, like in Brazil, they couldn't even do online advertising until the 2018 election, because they had to pass certain laws, in order to allow that. 00:20:39.173 --> 00:20:43.796 They have different rules about what those disclaimers and stuff look like. 00:20:43.796 --> 00:20:47.492 We had to think through the IDs, the verification system, all of that. 00:20:47.511 --> 00:20:57.994 And so, on one hand, these candidates were super excited to finally be able to use Facebook's targeting tools for elections,'cause they had seen the success of it in, in a lot of other places. 00:20:58.430 --> 00:21:02.977 But on the flip side, they were really frustrated all the hoops they had to jump through in order to do that. 00:21:02.996 --> 00:21:09.574 And when these issues pop up, like you mentioned around Lula, my gut is that was a bug, that that was a mistake. 00:21:09.604 --> 00:21:11.295 It was not somebody at the company. 00:21:11.295 --> 00:21:19.849 A lot of times people think there's somebody at the company purposely looking of who can we s- you know, we don't like that politician, so we're gonna do this to them or make this harder for them. 00:21:20.205 --> 00:21:20.522 No. 00:21:20.696 --> 00:21:23.006 Like that is not what was happening at all. 00:21:23.006 --> 00:21:26.383 It was usually some sort of bug, some sort of automated system. 00:21:26.383 --> 00:21:33.397 We try to have We don't need to go into it, but that's what Cross Check was sort of created for, was to try to provide some of those extra, those extra layers. 00:21:33.638 --> 00:21:41.306 And then on the other hand too, when people are doing research and they're like,"These ads are getting through," that in their mind incite violence or hate speech. 00:21:41.612 --> 00:21:52.848 very well, they might be right, but oftentimes too, their definition of that versus what ours might be from a policy perspective and the people that we're reviewing it could very well be very different. 00:21:53.172 --> 00:21:59.469 I could show content to 10 different people, and I could get a very mixed bag of people saying if it's okay or not. 00:21:59.481 --> 00:22:08.057 And so that's one of the challenges when you see that sort of research, is that it is not a super clear definition of what falls under that or not. 00:22:08.422 --> 00:22:09.268 Yeah, of course. 00:22:09.303 --> 00:22:12.186 And, and we talk about that on the podcast a lot, as you know. 00:22:12.660 --> 00:22:18.434 in hindsight, you know, that focus on policy rather than the kind of poster of the ad, do you think that that... 00:22:18.619 --> 00:22:20.207 I mean, has that changed at Meta? 00:22:20.227 --> 00:22:21.846 Do you think that focus should change? 00:22:21.926 --> 00:22:24.792 if you were doing it all again, how would you, how would you set it up? 00:22:25.217 --> 00:22:25.619 Yeah. 00:22:25.840 --> 00:22:39.042 they've been able to narrow it down a bit more in terms of the definitions, particularly as the AI has gotten better, the machine learning algorithms have gotten better in sort of thinking about that, and also what people are okay with or not. 00:22:39.112 --> 00:23:00.530 I still like that because Well, frankly, I like it because I'm still in favor of more transparency versus less, and obviously there's a lot of different people that are influencing our politics beyond just the candidates and these super PACs, and so I would like to make sure that those are available for researchers, journalists, et cetera, to be able to see. 00:23:00.932 --> 00:23:03.788 if we could make it easier, like over time we made it easier. 00:23:03.816 --> 00:23:12.142 Like at first we were emailing peop- or snail mailing people postcards with a code so that we could prove that they lived in the United States or lived in that country. 00:23:12.510 --> 00:23:18.380 We eventually were able to institute thing like if you open a bank or something, it asks you various identifying terms. 00:23:18.463 --> 00:23:33.661 But, there's things around the edges that I would change around the political ad transparency

tools, but I still think overall it's, the direction of it is correct because it helps as we continue to go forward and like we see how influencers and others are impacting this. 00:23:33.673 --> 00:23:57.816 It helps everyone get a sense of like, Okay, where's the money going?" Now, the thing that I think is challenging and would be very interesting to think about how would you create a database around this is the organic content, because the organic content and what's being recommended and stuff like that also plays a really big role in this, and I think we need to stop think, you know, kinda get outside of the old traditional way of thinking about campaigning, that it's all about where the money goes. 00:23:58.138 --> 00:24:02.288 Still super important, but also what are people pushing organically? 00:24:02.720 --> 00:24:02.957 Right. 00:24:02.981 --> 00:24:06.068 and I guess who are the people behind those accounts as well, right? 00:24:06.070 --> 00:24:13.721 And what are their motivations and, and means of revenue that perhaps you don't see as a platform, and which again, we've seen lots of examples. 00:24:13.934 --> 00:24:22.099 Russians paying often kind of right-wing political figures and them posting content that ostensibly was independent, but wasn't 00:24:22.425 --> 00:24:26.259 and the new version, you b- you brought up the Macedonian teenagers earlier. 00:24:26.284 --> 00:24:33.041 Like They were, in 2016, they were running content on Facebook to get clicks to their website where they were, had ads, right? 00:24:33.050 --> 00:24:34.303 And that's how they were making money. 00:24:34.615 --> 00:24:46.250 The new version of that in 2026 going into'27,'28, is gonna be AI influencers that are monetizing off of Facebook, YouTube, TikTok, et cetera. 00:24:46.279 --> 00:24:52.828 And so who's gonna be trying to capitalize on that and using political content in that way? 00:24:53.175 --> 00:25:06.192 And again, it's not paid advertising in the traditional way we think about it, but how we struggled even in 2020, how do you bring transparency to paid influencers when the money is exchanged off platform? 00:25:06.363 --> 00:25:11.130 If it's being monetized by the company, ideally that would be in a transparency database too. 00:25:11.513 --> 00:25:11.898 Yeah. 00:25:11.997 --> 00:25:12.290 Yeah. 00:25:12.321 --> 00:25:12.689 Okay. 00:25:13.093 --> 00:25:23.362 I wanted to ask before we kind of move on to, our eye-catching stories is the extent to which a platform like Meta is set up for a moment like an election in Brazil. 00:25:23.588 --> 00:25:30.776 You know,'cause you talked about you know, these kind of war gaming scenarios, these, these rooms that were set up to prepare for an election. 00:25:31.040 --> 00:25:33.847 but a lot of trust and safety professionals have been laid off. 00:25:33.858 --> 00:25:35.746 Those teams have often become smaller. 00:25:36.103 --> 00:25:38.763 There is a focus on automated moderation increasingly. 00:25:39.029 --> 00:25:50.020 And so, some of the stories you, brought to us today, they also suggest that there's a, lack of readiness in some cases, even though Brazil is a massive market with a huge Facebook, you know, using population. 00:25:50.271 --> 00:25:55.526 What's your sense on how ready they are in, in Brazil, but also in some of those other markets that are having elections next year? 00:25:55.942 --> 00:26:01.724 so like, there's a lot of my elections team that are still there, still doing a war room. 00:26:01.856 --> 00:26:03.226 Brazil is a huge election. 00:26:03.227 --> 00:26:10.757 I know from everyone from OpenAI to Microsoft to Anthropic, like it's one of the big ones that's on their radar with all of this. 00:26:10.758 --> 00:26:13.471 And so I do think there's still a lot of work happening. 00:26:13.481 --> 00:26:21.417 It's just not talked about as much because the companies don't necessarily want that target on their back and getting the criticism and stuff that comes with that. 00:26:21.498 --> 00:26:24.730 so I don't want people thinking like that there's nothing happening. 00:26:25.050 --> 00:26:28.337 The fact of the matter too is there's only so much you can be prepared for. 00:26:28.642 --> 00:26:34.175 Unfortunately, I used to make the joke like,"I can't stop elections from coming," and there's only so much we can do. 00:26:34.200 --> 00:26:35.939 Like it takes time to build these tools. 00:26:35.940 --> 00:26:43.162 It takes time to figure out how these models should respond to these answers, like what these policies should be, et cetera. 00:26:43.186 --> 00:26:48.994 and you don't always know how things are gonna play out, as part of all that, which is the whole reason we created the elections war room. 00:26:49.275 --> 00:26:52.481 So it's never gonna be perfect, when it has it. 00:26:52.482 --> 00:27:00.832 It's just like, do you have the policies and procedures in place so that when the unknown unknowns happen, you have a way of trying to respond to them quickly? 00:27:01.444 --> 00:27:01.824 Okay. 00:27:01.982 --> 00:27:13.669 and so your, your sense just going back to the Brazil election, what is the most likely kind of digital story that's gonna come out of this election based upon how close it is and, and patterns we've seen so far? 00:27:14.184 --> 00:27:17.051 I think we don't know yet,'cause I think it's gonna happen between rounds. 00:27:17.420 --> 00:27:17.761 Right 00:27:17.818 --> 00:27:32.038 it's gonna happen between rounds'cause I think if it is, Like, there are some that's showing it's, like, a half a percentage, So, like, if we are within that, it is gonna get real nasty, with influencers, with others, things that are happening. 00:27:32.316 --> 00:27:44.045 I imagine AI will be a big part of that story, but I don't think we necessarily know what that's gonna be yet, because I do think it'll be between Sunday and October 25th. 00:27:44.282 --> 00:27:44.619 Yeah. 00:27:44.758 --> 00:27:45.115

Okay. 00:27:45.457 --> 00:27:56.459 Well, fingers crossed that the, you know, it all goes smoothly and fairly and that there isn't a repeat of what happened last time, with the, protests and, the putsch that there was, which led obviously to Bolsonaro's jailing. 00:27:56.460 --> 00:27:57.576 So fingers crossed for that. 00:27:57.577 --> 00:27:59.781 We'll be, keeping tabs on it here on Ctrl-Alt-Speech. 00:28:00.128 --> 00:28:08.777 let's go now to those eye-catching stories, Katie, where these are our stories that kind of are worth mentioning, have caught our eye, and are worth our listeners paying attention to as well. 00:28:09.095 --> 00:28:28.910 I wanted to flag a story that kind of happened over the weekend here in the UK, which is a little bit more detail on the UK social media ban, which has obviously been a, big story here in the UK, but has also filtered into this broader narrative of social media bans, across Europe and, and the world, a- actually, as, as we saw in Australia last year. 00:28:29.348 --> 00:28:36.280 the culture secretary here in the UK announced that it was gonna come into force in March 2027, which is a new date. 00:28:36.292 --> 00:28:38.076 That's something that we didn't, fully know. 00:28:38.423 --> 00:29:10.564 but also there's a kind of a bit more detail on the potential approach, which, I gleaned from a, an interview with her in The Guardian, which she said that a kind of strict social media ban for under 16s was a bit like deciding that your town center is dangerous and then making all children stay in their bedrooms, which I thought was a really interesting analogy and is a little bit of a shift from the previous prime minister, Keir Starmer, who, because of lots of parental campaigns, very much was in favor of this under 16s ban by the end. 00:29:10.564 --> 00:29:20.542 So I think we'll see a bit of a shift towards the European model of a feature-based tiered approach that, is more respecting of children's rights as they, graduate through their teens. 00:29:20.811 --> 00:29:23.289 what's your kind of view of, of the ban's wholesale? 00:29:23.432 --> 00:29:27.798 And, and where are you expecting that to go in the last part of this year going into 2027? 00:29:28.212 --> 00:29:30.692 I think you're right about the feature-based stuff. 00:29:30.695 --> 00:29:35.333 I think as, you guys have talked about a ton, it is really hard to verify someone's age. 00:29:35.570 --> 00:29:37.990 it is very hard in which to, do that. 00:29:38.406 --> 00:29:42.432 And two, kids are really clever and they know how to get around bans. 00:29:43.123 --> 00:29:48.786 Like, they know how to, you know, if they're gonna wanna see it, they're gonna figure out a way to get around it. 00:29:48.810 --> 00:29:59.063 And so you're better off thinking about rather than just being like,"You can't be on it at all," how do we think about a way that you can do so in a responsible way, right? 00:29:59.074 --> 00:30:03.895 In a way that helps you to

think about, like,"How much time should I be spending on these screens? 00:30:03.909 --> 00:30:06.357 What is it doing to my attention span? 00:30:06.598 --> 00:30:10.102 What is it doing?" Like, Is the stuff that I'm seeing the stuff I wanna be seeing? 00:30:10.106 --> 00:30:13.571 How do I curate my feed in a way that I want to? 00:30:13.574 --> 00:30:15.627 How do I limit my time?" Like, stuff like that. 00:30:15.636 --> 00:30:23.872 So I think that's the approach I prefer a lot more, than I do about just this wholesale ban because I just don't think it works. 00:30:24.125 --> 00:30:24.484 Yeah. 00:30:24.762 --> 00:30:29.730 I think that's, uh, where a lot of people are landing actually in, in the kind of policy and political sphere right now. 00:30:29.944 --> 00:30:55.210 And, and to your point about kids being smart, people may have seen or been surfaced a video by This American Life this week, which actually talked about a s- a story that Mike and I flagged on the podcast best part of 12 months ago, where kids that have been banned from social media are going to Spotify and going to individual episodes of old NPR podcasts and using that as a forum. 00:30:55.320 --> 00:30:55.839 Chat. 00:30:55.906 --> 00:30:56.548 Incredible. 00:30:56.673 --> 00:30:59.365 Like, whoever's coordinating that is a genius. 00:31:00.182 --> 00:31:08.326 I had a friend of mine, I will not say who, but, she was telling us how her son has been able to watch YouTube videos through Google Maps 00:31:08.726 --> 00:31:09.500 No 00:31:09.792 --> 00:31:12.093 and, like, figured out a workaround to 00:31:12.148 --> 00:31:12.940 Wow, 00:31:13.300 --> 00:31:15.118 to get around, like, the... 00:31:15.329 --> 00:31:15.765 Yeah. 00:31:15.890 --> 00:31:19.000 So again, super smart and can figure out another way 00:31:19.324 --> 00:31:20.207 Yeah, totally. 00:31:20.505 --> 00:31:42.634 you know, these, American Life videos on Instagram and Blue Sky and Threads as well, I think all went viral, all got multiple millions of, of views, which suggests that people are understanding, the creativity of these kids and, and actually probably you know, having to think about social media bans in a new way, that they weren't having to think about six months ago when this was a, a fairly kind of dry policy-led debate. 00:31:43.159 --> 00:31:52.798 Well, and it's been a challenge from-- I mean, I remember in 2011 when I first joined Facebook, there were questions from parents about how were they supposed to talk to their kids about social media. 00:31:52.798 --> 00:31:59.386 And it's always been a challenge for parents, A, to even know what the different tools are and, like, what they can be doing. 00:31:59.387 --> 00:32:02.105 But how do you even talk to your kid about that? 00:32:02.397 --> 00:32:03.538 What does that look like? 00:32:03.538 --> 00:32:05.540 And I think COVID made that even harder. 00:32:05.551 --> 00:32:19.711 I remember my sister kind of being like,"Okay, my nephew," her son,"only way he's able to socialize is playing Minecraft with his friends." So she's like,"I'm not wild at how much screen time he's having, but I also wanna make sure he's getting the socializing. 00:32:20.085 --> 00:32:36.952 So how do I, how do I think about that?" And so, I think the hope is by making it feature-based, can each family be able to make the decisions that are right for their family about what their use of not just social media, but the internet writ large looks like. 00:32:37.231 --> 00:32:38.429 Yeah, I think that's fair. 00:32:38.429 --> 00:32:44.240 I think it's, it's still a tricky line to tread, but I think it's the, probably the best one we have now. 00:32:44.577 --> 00:32:48.215 wanna flip to another eye-catching story that you flagged, Katie. 00:32:48.613 --> 00:32:52.625 I'm quite glad not to be talking at length about AI regulation this week, frankly. 00:32:52.673 --> 00:33:02.509 Um, we had to me- we had to mention it and this is the, famous for lots of reasons, AI Accord that has been signed by some of the big tech CEOs. 00:33:02.938 --> 00:33:05.203 The morally binding AI accord 00:33:05.433 --> 00:33:11.167 yeah, who doesn't love a morally binding accord signed by a bunch of white blokes in suits? 00:33:11.298 --> 00:33:15.701 Um, so talk us through what you found interesting here and without trying to laugh. 00:33:17.728 --> 00:33:25.167 Well, I think-- So the things they signed onto are fine and great, but per usual, the devil's gonna be in the details, right? 00:33:25.217 --> 00:33:29.722 I mean, I think we already knew this was all gonna be, there's gonna be a lot of self-regulation first. 00:33:29.765 --> 00:33:36.912 I still think a lot of this is gonna happen in the States, and we won't see anything at the federal level in the US till 2029 at the earliest. 00:33:37.184 --> 00:33:43.273 What I found interesting, and this is the politics/Facebook in me, is the role that Zuck had in this. 00:33:43.362 --> 00:33:58.439 So it started to come out yesterday afternoon that Mark started seeding this idea at the state dinner with President Xi and Jensen Yang, and then was a big part of like drafting a lot of it and like getting all the other CEOs together in which to sign this. 00:33:58.439 --> 00:34:03.102 And we all know Mark loves a good manifesto and that he lo- you know, around all this. 00:34:03.108 --> 00:34:23.641 But, you know, for somebody who after the'24 election going into the president's inauguration, Elon was the favored CEO du jour, and Mark was kind of seen as catch-up because Trump was mad at him for kicking him off the platform after 2020 and like for a variety of, a variety of other things that we don't have time to get into. 00:34:23.733 --> 00:34:27.789 but now it seems like Mark has brought that full circle, and that influence. 00:34:27.804 --> 00:34:33.164 And what's interesting for me on this is what does this look like as we go into the presidential in'28? 00:34:33.342 --> 00:34:42.713 And

how will these CEOs, you know, they're spending a lot more money in super PACs and on campaigns and trying to like pick their candidates that they think are gonna be more favorable to them. 00:34:43.023 --> 00:34:51.356 Will we see tech CEOs starting to pick favorites as we're seeing different nominees on both the Democratic and Republican side? 00:34:51.565 --> 00:34:54.710 Will they keep their powder dry until we have the nomination? 00:34:54.934 --> 00:34:58.809 what is their long game gonna be in terms of influence? 00:34:58.836 --> 00:35:07.976 Because I think that we have seen with the Trump administration, these CEOs have been in DC more than I think I've ever seen them, i- in influencing that. 00:35:08.010 --> 00:35:12.224 and then how do the different ones start to jockey for, influence and what does that look like? 00:35:12.226 --> 00:35:23.918 So I'm, again, I'm watching a little bit more from the, political angle of all of this because I think that one of the things Mark talked about in a recent profile of him, I disagree with him on this a little bit. 00:35:23.929 --> 00:35:28.380 He, he still thinks the big problem after 2016 is we didn't communicate enough about our good intentions. 00:35:28.695 --> 00:35:33.021 And I write about this in the book about how good intentions are not safeguards and scale is not a strategy. 00:35:33.320 --> 00:35:37.224 And I feel like that's a little bit about what this AI accord is, is like,"We got good intentions. 00:35:37.224 --> 00:35:44.541 Look at us." Like, but it's your actions that are gonna follow that that is what's gonna build up trust with people, and that's what everybody's watching. 00:35:44.541 --> 00:35:45.235 They're like,"Fine. 00:35:45.416 --> 00:35:46.722 You guys can sign your little accord. 00:35:46.722 --> 00:35:49.998 You can't even spell United States right, um, at the bottom of it. 00:35:49.998 --> 00:35:51.878 Let's see what you guys actually do." 00:35:52.329 --> 00:35:52.661 Yeah. 00:35:52.811 --> 00:36:02.739 Listen, so if you, if you haven't seen that,"Unites States" was the way that the, uh, it was kind of written, printed underneath Trump's signature, which, which I guess is, you know, maybe very telling. 00:36:02.820 --> 00:36:06.456 you know, I don't wanna think too much about the poor staffer who had to, had to do that. 00:36:06.679 --> 00:36:12.010 I mean, it's also, Katie, the kind of fact that this isn't the mid-2000s, right? 00:36:12.010 --> 00:36:15.757 This is-- There is a history that these companies are kind of building on. 00:36:15.759 --> 00:36:24.114 And to say right now that we are signing a morally binding accord and that you should trust us because we have good intentions, you know, feels naive at the least. 00:36:24.344 --> 00:36:29.324 do, you know, the CEOs really think that the public are gonna buy that or is it kind of theater? 00:36:29.784 --> 00:36:30.650 I don't know. 00:36:30.744 --> 00:36:31.844 I really don't know. 00:36:31.853 --> 00:36:43.523 They're in such their own little bubble that sometimes I think they convince themselves that Oh, if we just..." Like the number of times I heard from headquarters,"Why aren't you just explaining it better to policymakers about what we're doing? 00:36:43.523 --> 00:36:49.172 Why aren't you just explaining it better?" And I'm like,"It's not a matter of just explaining it better. 00:36:49.175 --> 00:36:58.619 That's not the challenge." So there's a part of me that's like, maybe it's na- I think it's naive on their end that like this is enough, like maybe it'll be enough to like calm things down a little bit. 00:36:58.628 --> 00:37:17.311 I think part of it too is there's this game of hot potato, because the profile I was mentioning about Mark, he talked about like we just didn't communicate enough, but also that he thinks that he took on too much responsibility post-2016, which then allowed Facebook to be a political dumping ground for everyone about it was all Facebook's fault, it's all Facebook's fault, putting it all on them. 00:37:17.311 --> 00:37:41.025 And so you see these CEOs playing more hot potato with the government and others being like, not one of us wants to be entirely in charge in all of this, and so we're gonna try to make sure that we're spreading the responsibility around." Which is like I think what a lot of us want, but because we have low trust in our government, we have low trust in these tech CEOs, it's sort of like, is this the check and balances that we really want? 00:37:41.083 --> 00:37:42.217 what do we want that to look like? 00:37:42.219 --> 00:37:47.778 Like that's a conversation I would love for us to have if I could have a magic wand, but I don't know if we'll have it anytime soon. 00:37:48.498 --> 00:37:49.692 No, no, me neither, unfortunately. 00:37:49.956 --> 00:37:57.130 political theater is a good segue onto our Patreon, segment now, where we're gonna talk a bit about a story that some folks might have seen. 00:37:57.130 --> 00:38:04.054 It's a great investigation by WIRED into changes happening at GIFCT, the Global Internet Forum to Counter Terrorism. 00:38:04.340 --> 00:38:09.376 a really interesting cross-platform body that's been set up to, counter terrorism. 00:38:09.610 --> 00:38:24.139 gonna dive into that, and really what it means, some of these changes mean for not only kind of the platforms themselves and, and their jostling for position and power, but also the potential for terrorist content to, shift, in the next few years. 00:38:24.313 --> 00:38:28.646 And so I, I had a l- a few things to say on that, and I'd love to get your thoughts, Katie, so we'll dive in now 00:38:30.480 --> 00:38:33.760 That's the end of this week's free edition of Ctrl Alt Speech. 00:38:34.141 --> 00:38:41.019 To hear the extended episode with additional coverage, become a Patreon supporter at patreon.com/ctrlaltspeech. 00:38:41.280 --> 00:38:43.820 That's C-T-R-L Alt Speech. 00:38:44.400 --> 00:38:50.820 If your organization is interested in sponsoring the podcast, contact us via our website at ctrlaltspeech.com. 00:38:51.260 --> 00:38:51.980 Thanks for listening