Latest / Ctrl-Alt-Speech / Watch Out, AI is Getting More Persuasive

Transcript

Watch Out, AI is Getting More Persuasive · Ctrl-Alt-Speech

• 0:00
  Ben Whitelaw: In the somewhat offish, perhaps slightly disinterested words of threads, what's new Alice?
• 0:04
  Alice Hunsberger: Hey What's new for me is pretending to be a journalist.
• 0:11
  Alice Hunsberger: I, I, I'm fairly an old hand at, regurgitating links that other people might find interesting, but actually adding commentary and talking about it and analyzing it is, uh, it's been a new adventure for me and on a podcast, episode, this is entirely new.
• 0:31
  Alice Hunsberger: So we'll see how it goes.
• 0:34
  Ben Whitelaw: nice.
• 0:35
  Ben Whitelaw: Well, what's new with me is that I have a new co host, in the form of you.
• 0:39
  Ben Whitelaw: So it's all changed here at Ctrl-Alt-Speech, but only temporarily.
• 0:43
  Ben Whitelaw: And we'll talk more about that in a second.
• 0:53
  Ben Whitelaw: Hello and welcome to Ctrl-Alt-Speech, your weekly roundup of the major stories about online speech, content moderation, and internet regulation.
• 1:01
  Ben Whitelaw: This week's episode is brought to you with financial support from the future of online trust and safety fund.
• 1:07
  Ben Whitelaw: My name is Ben Whitelaw.
• 1:08
  Ben Whitelaw: I'm editor and founder of Everything in Moderation, and I'm joined today by a very new face.
• 1:14
  Ben Whitelaw: It's not Mike Masnick, who is on holiday with his family this week.
• 1:19
  Ben Whitelaw: But it's Alice Hunsberger.
• 1:20
  Ben Whitelaw: Welcome.
• 1:22
  Alice Hunsberger: Ben.
• 1:23
  Alice Hunsberger: Thanks for having me.
• 1:24
  Ben Whitelaw: How are you doing?
• 1:25
  Alice Hunsberger: I'm good.
• 1:27
  Ben Whitelaw: really nice to have you here.
• 1:28
  Ben Whitelaw: Um, for listeners who don't know Alice, she is the VP of trust and safety and content moderation at Partner Hero.
• 1:35
  Ben Whitelaw: Used to be VP and global head of customer experience at Grindr.
• 1:39
  Ben Whitelaw: Yeah.
• 1:40
  Ben Whitelaw: So you've got vast, vast experience.
• 1:41
  Ben Whitelaw: So we, you were an ultimate choice for us when we knew that Mike was going to be away and we wanted somebody to step in.
• 1:49
  Ben Whitelaw: Tell us about yourself.
• 1:51
  Alice Hunsberger: Well, um, thank you.
• 1:54
  Alice Hunsberger: I, yeah, would, tell us about yourself.
• 1:57
  Alice Hunsberger: It's very broad.
• 2:00
  Alice Hunsberger: I thought we were supposed to be talking about what's happening in the news?
• 2:03
  Alice Hunsberger: Um, yeah, I mean, I've been working in trust and safety professionally for 14 years or so since 2010.
• 2:11
  Alice Hunsberger: I started out as a content moderator and worked my way up to VP and I'm obsessed with all things trust and safety.
• 2:21
  Alice Hunsberger: So, I have a lot of fun reading what's in the news and, Thinking about it, but that is not my day job, uh, unlike Mike, so I'll do my best.
• 2:32
  Ben Whitelaw: You're very well equipped.
• 2:34
  Ben Whitelaw: You're very well equipped.
• 2:35
  Ben Whitelaw: Alice also, we owe a lot to you for this podcast because you actually put Mike and I in touch at a conference last year, right?
• 2:43
  Alice Hunsberger: Yeah, we were all at TrustCon together.
• 2:47
  Alice Hunsberger: It was my first time going and my first time meeting people in person.
• 2:52
  Alice Hunsberger: And as a avid follower of both you and your newsletter and Mike and his site, I was like, Fan girl meeting both of you is like, ah, people, people I read every single day.
• 3:05
  Alice Hunsberger: And we were at a after party somewhere and you both were standing next to each other, but not talking.
• 3:12
  Alice Hunsberger: And so I was like, Hey, do you guys know each other?
• 3:15
  Alice Hunsberger: Like, do you know what each other's faces look like?
• 3:17
  Alice Hunsberger: Uh, and introduced you.
• 3:19
  Alice Hunsberger: So I would like to take credit for, this podcast existing.
• 3:23
  Alice Hunsberger: And therefore it's only right that I'm the first step in host.
• 3:30
  Ben Whitelaw: Yeah.
• 3:30
  Ben Whitelaw: Credit credit or blame one of the
• 3:33
  Alice Hunsberger: Sure.
• 3:33
  Alice Hunsberger: I guess.
• 3:34
  Alice Hunsberger: Email, email me privately and let me know your thoughts if you're listening to this.
• 3:39
  Ben Whitelaw: Yeah.
• 3:40
  Ben Whitelaw: Also you've made it sound much more awkward than it was.
• 3:43
  Ben Whitelaw: I remember being very, uh, suave in my approach to Mike.
• 3:46
  Ben Whitelaw: You were basically maybe, maybe it was much more awkward then it was in my head, but I thought I handed myself very well.
• 3:53
  Ben Whitelaw: I, you're
• 3:53
  Alice Hunsberger: Of course you did.
• 3:54
  Alice Hunsberger: No, I'm the awkward one.
• 3:55
  Alice Hunsberger: I'm not, I don't want to insinuate that anybody else was awkward, but I get awkward in social situations, so.
• 4:01
  Ben Whitelaw: No, you, you did a great job and, it's a lovely to have you here hosting.
• 4:05
  Ben Whitelaw: It's also been great recently.
• 4:07
  Ben Whitelaw: You've started writing a Monday newsletter for everything moderation, which is called trust and safety insider, which is the kind of.
• 4:15
  Ben Whitelaw: Nuts and bolts, guides to how trust and safety works from, uh, somebody who's worked inside platforms for many, many years, how have you found writing that newsletter?
• 4:26
  Alice Hunsberger: It's been great.
• 4:28
  Alice Hunsberger: It's been really good.
• 4:29
  Alice Hunsberger: I, um.
• 4:30
  Alice Hunsberger: I have really enjoyed taking time to write every week and think about, my own perspective.
• 4:38
  Alice Hunsberger: And I get a lot of random messages on LinkedIn from people who are like, help, how do I think about this thing?
• 4:47
  Alice Hunsberger: Or what do I do?
• 4:48
  Alice Hunsberger: and so it's been really good to share that a little bit more.
• 4:51
  Alice Hunsberger: Broadly, and then the, the best part is you're a really good editor.
• 4:56
  Alice Hunsberger: So you make, you make my writing better.
• 4:59
  Alice Hunsberger: It's like, when I think about all the stuff that I put out on LinkedIn while I'm drinking coffee first thing in the morning and not thinking too hard.
• 5:07
  Alice Hunsberger: And then like actual posts that I've, Put a lot of thought into and that you edit the quality is much higher.
• 5:15
  Alice Hunsberger: So, that's been good too.
• 5:17
  Ben Whitelaw: Thank you.
• 5:18
  Ben Whitelaw: I didn't pay Alice to say that or anyone.
• 5:21
  Ben Whitelaw: And also imagine if I made it worse, that would be terrible.
• 5:24
  Ben Whitelaw: I would need to get out of my industry pretty, pretty ASAP.
• 5:28
  Ben Whitelaw: So good to hear that that is the case.
• 5:31
  Ben Whitelaw: Um, so
• 5:31
  Alice Hunsberger: getting good feedback on it too.
• 5:33
  Alice Hunsberger: Haven't we?
• 5:33
  Alice Hunsberger: Like
• 5:34
  Ben Whitelaw: yeah.
• 5:34
  Alice Hunsberger: enjoy it.
• 5:35
  Ben Whitelaw: Yeah, lots of great responses from people.
• 5:37
  Ben Whitelaw: And we'll include an example in today's show notes so that listeners of the podcast can take a look if they haven't read Alice's great opinions each week already.
• 5:46
  Ben Whitelaw: Which is, I think a good time to, crack on with today's episodes, to get into the stories from this week.
• 5:53
  Ben Whitelaw: Alice is a.
• 5:55
  Ben Whitelaw: Vast network of, people that she follows and speaks to.
• 5:58
  Ben Whitelaw: So she's bringing a couple of really interesting stories to today.
• 6:01
  Ben Whitelaw: Whilst the lineup is different and Mike is not here, the format remains the same.
• 6:06
  Ben Whitelaw: So Alice and I will look at two bigger stories this week.
• 6:09
  Ben Whitelaw: We'll go in depth.
• 6:10
  Ben Whitelaw: on those stories for 10 minutes or so, and then we'll skittle through the rest, of the kind of interesting stories.
• 6:16
  Ben Whitelaw: And as ever, they will all be linked to in the show notes so that you can go and have a read for yourself.
• 6:21
  Ben Whitelaw: Before we crack on, we, we don't have a sponsor interview this week, so there is no bonus chat, but if you'd like to join us on the podcast and, talk a bit about trust and safety with Mike and I in the future, please do email us at sponsorship@ctrlaltspeech.Com that's C T R L A L T speech.
• 6:41
  Ben Whitelaw: com or go to our very.
• 6:43
  Ben Whitelaw: Attractive new website that our producer Leigh has built, which is a definitely worth visiting, even if you're not going to commit to being a sponsor.
• 6:51
  Ben Whitelaw: Um, great work Leigh.
• 6:54
  Ben Whitelaw: So, onto our first story, then, advanced warning for people.
• 6:58
  Ben Whitelaw: There's a fair bit of AI today and Alice, you picked a story that, is super, super interesting, about the persuasiveness of AI models, right?
• 7:08
  Alice Hunsberger: Yeah.
• 7:09
  Alice Hunsberger: I'm glad you put the warning, the content warning in there.
• 7:12
  Alice Hunsberger: It's funny.
• 7:12
  Alice Hunsberger: It feels like more and more everybody's job is turning into thinking about AI, what the effects of AI are, how is it going to change things, but especially for trust and safety people, everybody that I talk to who's in a leadership position.
• 7:28
  Alice Hunsberger: Position and trust and safety is all consumed with AI, both in terms of how can we use it to moderate?
• 7:37
  Alice Hunsberger: How can we use it for good?
• 7:40
  Alice Hunsberger: But then also, how is AI going to accelerate some of the bad content that is out there?
• 7:46
  Alice Hunsberger: We're seeing, on the Internet today.
• 7:48
  Alice Hunsberger: And so the link that I brought to us today is about anthropic.
• 7:53
  Alice Hunsberger: They just put out, a paper on measuring.
• 7:57
  Alice Hunsberger: Model persuasiveness.
• 7:59
  Alice Hunsberger: So they looked at all of the Claude models that they've put out, and then chose issues that they wanted the model to try to persuade somebody to agree with, and they took a, an argument from a human and an argument from each AI model and had people sort of evaluate how much each example changed.
• 8:24
  Alice Hunsberger: Their mind, uh, And so, yeah, it was, it, it's a good example of trying to look at actual ways that AI can be used for harm, especially this year, the mega year of elections, when lots of people are going to try to persuade other people to vote for the right candidate or vote for the right cause, one of the things that I thought was like the most interesting about it.
• 8:56
  Alice Hunsberger: Um, and so they tried four different types of persuasiveness with the models.
• 9:03
  Alice Hunsberger: There was rhetorical and emotional appeals to change, which didn't work.
• 9:10
  Alice Hunsberger: They worked, but not as effectively.
• 9:12
  Alice Hunsberger: But then the two that were the most effective were, logical reasoning and providing Evidence, uh, but they didn't specify whether that evidence had to be true or not.
• 9:25
  Alice Hunsberger: So if you could just like make up deceptive information and call it evidence, that was the most effective of all of them, uh,
• 9:36
  Ben Whitelaw: like presenting stuff as evidence was, was enough.
• 9:40
  Alice Hunsberger: Yeah.
• 9:41
  Alice Hunsberger: And so, it sort of proved that people don't necessarily verify whether things are true or not.
• 9:49
  Alice Hunsberger: They take something presented to them with confidence as accurate.
• 9:56
  Alice Hunsberger: And,
• 9:57
  Ben Whitelaw: Sounds like some people I know.
• 10:00
  Alice Hunsberger: Well, yeah, and it's, you know, it's super sketchy because now you can just fabricate any kind of facts that seems like it might persuade somebody.
• 10:09
  Alice Hunsberger: And so it's definitely something to keep an eye out for.
• 10:11
  Alice Hunsberger: And I think it's a good study in, showing the need for teaching people, media literacy, fact checkings, being skeptical, and, you know, just generally.
• 10:24
  Alice Hunsberger: Not taking random arguments at their word.
• 10:28
  Ben Whitelaw: Yeah.
• 10:29
  Ben Whitelaw: No, you're right.
• 10:30
  Ben Whitelaw: And, and, it's very much true to form for Anthropic to do this kind of research, the founders, for those who don't know, left OpenAI because they wanted to build a more kind of trustworthy model, right?
• 10:41
  Ben Whitelaw: So it's, it's great that this research is happening, that we're starting to see.
• 10:46
  Ben Whitelaw: Some quite advanced, thinking around the building of these models.
• 10:50
  Ben Whitelaw: One thing I was thinking about as I read the paper was does, does the kind of findings that you've brought to us here, does that help kind of bad actors in terms of being able to create prompts and create content that actually, causes people harm?
• 11:07
  Ben Whitelaw: Do you think this kind of research could be used inadvertently to cause a user's issues.
• 11:14
  Alice Hunsberger: Oh, you mean our bad actors also reading the anthropic paper and
• 11:20
  Ben Whitelaw: mean, I'm sure they are.
• 11:21
  Alice Hunsberger: I bolded that says the deceptive strategy, which allowed the model to fabricate information was found to be the most persuasive and people were like, okay, cool.
• 11:31
  Alice Hunsberger: I've got a roadmap now for,
• 11:33
  Ben Whitelaw: I'm on the right track.
• 11:34
  Alice Hunsberger: It definitely is for sure.
• 11:37
  Alice Hunsberger: That said.
• 11:38
  Alice Hunsberger: I think it's likely that bad actors are doing this kind of Studying and research on their own anyway, um, probably not through scholarly journals with all sorts of references and, you know, specific parameters, but for sure, they're a B testing different types of arguments on real people and seeing what the results are and learning from that.
• 12:06
  Alice Hunsberger: And so, if we know that people are.
• 12:10
  Alice Hunsberger: Looking at how these models can be used for bad.
• 12:15
  Alice Hunsberger: Then we have to also spend time understanding that ourselves so that we have a chance at combating it or putting up guardrails or somehow inoculating people against believing misinformation or whatever, whatever strategy people choose.
• 12:34
  Ben Whitelaw: So if, if you, uh, I mean, how, how do you take this research and use it in your work at the mobile?
• 12:41
  Ben Whitelaw: Like, so what, what would you do with the outputs from this research?
• 12:45
  Ben Whitelaw: And how would you start to think about implementing it in, your role or in previous roles, and maybe some of our listeners are thinking about how to do that.
• 12:54
  Alice Hunsberger: Good question.
• 12:55
  Alice Hunsberger: One thing that made me think through was the importance of citations and of credible websites and being able to understand what a credible website is.
• 13:08
  Alice Hunsberger: Is the sort of compounding of misinformation tactics like this, plus search results becoming more terrible these days, having a generated summary at the top of the search result that comes from.
• 13:27
  Alice Hunsberger: Whatever website, whether it's actually factually correct or not, super scares me because it could be that somebody gets a, message trying to persuade them to change their mind, makes up some completely random facts.
• 13:42
  Alice Hunsberger: creates a bunch of bogus news websites that also make up those facts.
• 13:46
  Alice Hunsberger: So then when somebody Googles, Hey, is this random fact true?
• 13:50
  Alice Hunsberger: They'll get a little AI summary at the top that says, sure, this is true because like, you
• 13:57
  Ben Whitelaw: We found it.
• 13:57
  Ben Whitelaw: Yeah.
• 13:58
  Ben Whitelaw: It exists.
• 13:59
  Alice Hunsberger: fake news reporter.
• 14:00
  Alice Hunsberger: com also cites it.
• 14:03
  Alice Hunsberger: So I think the answer is.
• 14:08
  Alice Hunsberger: Labeling information as much as possible, providing citations, teaching people what trustworthy news sources look like, teaching people how to be skeptical, even if you're a platform that doesn't deal in news.
• 14:22
  Alice Hunsberger: Per se, it's always good to have campaigns that kind of teach these skills.
• 14:29
  Alice Hunsberger: And then really, honestly, I think the solution is teaching kids in schools.
• 14:35
  Alice Hunsberger: So that when they, as they grow up with this kind of,,access to information everywhere, but becoming harder and harder to verify it, then we need to teach people.
• 14:49
  Alice Hunsberger: Research skills.
• 14:52
  Ben Whitelaw: And this is only going to get more harder for the kids of the future, right?
• 14:56
  Ben Whitelaw: So it's an issue now, but what the paper says, which I found really interesting is that the models that Anthropica built are getting more persuasive.
• 15:06
  Ben Whitelaw: So, you know, it has Kind of two models, which of course, compact models, which, are getting more persuasive over time and Claude and the latest version is magnificently called Claude three Opus, which is a, such a like wicked name for something that can do so much harm.
• 15:25
  Ben Whitelaw: Um, you know, it's, several kinds of points on the chart in the research, more persuasive than, Claude 1.
• 15:31
  Ben Whitelaw: 3.
• 15:31
  Ben Whitelaw: So if you take that, kind of logic, you know, future models are going to get more persuasive.
• 15:37
  Ben Whitelaw: And the distinction between, generative AI and humans in their ability to persuade people is going to get really much, much more indistinct.
• 15:48
  Ben Whitelaw: Uh, and so your, your point about being able to kind of find ways of unpicking the differences is going to be really, really key, which I think probably takes us neatly to the next.
• 15:57
  Ben Whitelaw: Uh, onto our next story, which actually kind of came out on the back of last week.
• 16:03
  Ben Whitelaw: Really Friday was, the announcement by Meta on how it was approaching labeling and labeling of AI generated content, particularly now you mentioned labeling there, Alice, and I know you've got some views on whether labeling works and it's kind of ability to drive the behaviors we want, but just to give an overview for people who didn't.
• 16:25
  Ben Whitelaw: See this story, basically Facebook is changing its policies around AI generated content on the back of a oversight board recommendation.
• 16:38
  Ben Whitelaw: The kind of pseudo Supreme court that is.
• 16:42
  Ben Whitelaw: Independent, but it's funded by Facebook and by Meta.
• 16:46
  Ben Whitelaw: So this recommendation suggested that Meta needs to broaden its policy around, AI content was used to be very narrow.
• 16:57
  Ben Whitelaw: It used to be that it was only generative AI that had made somebody seem to say something that they hadn't, that was going to be pulled down from the platform.
• 17:06
  Ben Whitelaw: Then this came forward where Joe Biden was kind of manipulated to seemingly touch his elder granddaughter that was referred to the oversight board.
• 17:17
  Ben Whitelaw: And they basically said that that policy is too strict.
• 17:20
  Ben Whitelaw: We need to kind of broaden that Facebook have gone away and take a look.
• 17:23
  Ben Whitelaw: And they are now agreeing that that.
• 17:26
  Ben Whitelaw: Is the wrong approach and they're broadening how they kind of view generative AI on the back of that.
• 17:32
  Ben Whitelaw: And their, their response is to start labeling content.
• 17:35
  Ben Whitelaw: And so you will see a lot more on meta platforms, a little label that says made by AI.
• 17:42
  Ben Whitelaw: That label will appear if a user indicates through the upload process that it was made by AI, but also Facebook have some meta, have some.
• 17:52
  Ben Whitelaw: Clever technical ways of being able to do that.
• 17:53
  Ben Whitelaw: This distinct AI content from, from other content.
• 17:58
  Ben Whitelaw: So this is a big, change for Meta.
• 18:00
  Ben Whitelaw: We're going to see lots of changes on the front end and, some changes to the kind of user upload process.
• 18:06
  Ben Whitelaw: What are your thoughts on this Alice?
• 18:07
  Ben Whitelaw: How, how confident are you that this is a step in the right direction?
• 18:11
  Alice Hunsberger: It's definitely a step in the right direction.
• 18:14
  Alice Hunsberger: There's two separate threads going on here.
• 18:19
  Alice Hunsberger: So one is, Are we able to accurately detect and label all instances of some kind of manipulated media, be it AI generated or not?
• 18:34
  Alice Hunsberger: And that's more of an existential question now because AI is making it much more possible, but you know, we've had Photoshop for a gazillion years.
• 18:45
  Alice Hunsberger: And
• 18:45
  Ben Whitelaw: What, what, what are your Photoshop skills like out of interest?
• 18:48
  Alice Hunsberger: Oh, mine are non existent, yeah, which is why, yeah, AI makes it very easy for people like me to now create images that I couldn't before.
• 19:00
  Alice Hunsberger: But I think it does speak to, you know, every platform has had some kind of policy on manipulated media forever, even before AI.
• 19:11
  Alice Hunsberger: Trust and safety people have been dealing with this issue, but the scope of it.
• 19:16
  Alice Hunsberger: Is bigger now, and the possibilities are more endless.
• 19:22
  Alice Hunsberger: And so the, the question is for major news things like this Biden example, then we'll be able to label it and say, Hey, this isn't true.
• 19:35
  Alice Hunsberger: But that's not necessarily going to be possible for fake images of Non celebrities, uh, they might not get the same kind of labeling or detection, possible.
• 19:48
  Alice Hunsberger: So that's one part.
• 19:49
  Alice Hunsberger: And then the second part of this decision is around making sure that the intent and the harms, of manipulated media are considered as well as just whether an image is Real or not.
• 20:09
  Alice Hunsberger: So the recommendation from the oversight board also talked about don't remove manipulated media completely.
• 20:17
  Alice Hunsberger: If there's no harm.
• 20:20
  Alice Hunsberger: Being created.
• 20:21
  Alice Hunsberger: Like, if somebody's, you know, posting something in good fun, and even if it's, significantly altered and could mislead somebody in some way, you also have to think about what are the repercussions of that, um, because there's free speech, satire, parody, art, all kinds of reasons why somebody might do it.
• 20:45
  Alice Hunsberger: upload a photo of a celebrity that's not true, that aren't going to necessarily like harm democracy or some other major issue.
• 20:56
  Alice Hunsberger: So they're trying to sort of separate the harms caused with the proliferation of manipulated media with potentially labeling and paying attention to those cases where.
• 21:09
  Alice Hunsberger: It really is going to make a difference.
• 21:12
  Ben Whitelaw: Right.
• 21:12
  Ben Whitelaw: And to what extent is that the kind of right approach?
• 21:15
  Ben Whitelaw: Cause I remember back in.
• 21:17
  Ben Whitelaw: During COVID, there was a real spate of platforms who started using labeling to identify false or misleading posts and to try and kind of separate, those two things, like the proliferation versus the harm.
• 21:33
  Ben Whitelaw: And I guess To your point earlier, the onus on the user to kind of decide for themselves whether this was actually, true or not I mean, I don't know how kind of successful that was, but do you feel like that this return to labeling is a, is a kind of first step and a good one in helping people understand what is AI media and what isn't?
• 21:58
  Alice Hunsberger: Yes and no.
• 21:59
  Alice Hunsberger: So the reason why it's good is because exactly what I was talking about earlier with our other story, uh, The more that people recognize that the ways that AI can be used and the ways that AI can be used to manipulate you, then the more skeptical people are going to be.
• 22:17
  Alice Hunsberger: So if they're browsing on Facebook and they see a gazillion labels that say, Hey, this is probably trying to mislead you.
• 22:24
  Alice Hunsberger: Hey, this is also probably trying to mislead you.
• 22:27
  Alice Hunsberger: Hey, this thing might be trying to mislead you as well.
• 22:29
  Alice Hunsberger: Then they're going to start getting skeptical.
• 22:31
  Alice Hunsberger: And I think that's a good thing.
• 22:33
  Alice Hunsberger: Um, the bad part of it is if people get too used to relying on labels to tell them whether something's true or not, then if something isn't labeled yet, they may be even more likely to believe that it's true because it doesn't have a label.
• 22:50
  Alice Hunsberger: So therefore I won't be skeptical about this one.
• 22:54
  Alice Hunsberger: And so I worry that it won't be a complete solution, especially when people are just sort of lazily scrolling and aren't necessarily there to apply all of their critical thinking skills.
• 23:07
  Alice Hunsberger: Um, but I don't know, you know, it's definitely better than.
• 23:11
  Alice Hunsberger: Not labeling at all or not saying anything at all and just leaving people to guess completely.
• 23:17
  Alice Hunsberger: So I'm, I'm in favor of it generally.
• 23:20
  Alice Hunsberger: I just think this is the beginning of the conversation about how to treat things like this, not the end.
• 23:26
  Ben Whitelaw: Yeah, and one thing that the announcement made clear is that certain warnings and labels will appear more prominently.
• 23:34
  Ben Whitelaw: For certain types of harm as well.
• 23:36
  Ben Whitelaw: We didn't get to see what that looks like just yet, but apparently that is something that they will, they will do is, is grade the labeling to make it more prominent, to make it more visible in, I guess, feeds in contexts where, it's important for the user to see that, which I think is good.
• 23:54
  Ben Whitelaw: Um, the dev was in the detail, but what that actually looks like, but, and then the other thing that I think will be interesting to see is the extent to which other platforms follow suit here.
• 24:03
  Ben Whitelaw: So, you know, Facebook meta is working with.
• 24:07
  Ben Whitelaw: Other platforms, to, kind of create common technical standards that help identify generated AI media.
• 24:16
  Ben Whitelaw: And so, yeah, the fact that they've kind of done this suggests that, perhaps others will follow suit.
• 24:20
  Ben Whitelaw: So I'm fully expecting platforms to also say that they're going to start labeling more content, which I think brings the efficacy of labeling back into view again, because it's, I would say somewhat fallen off the radar of certainly me and, less in fashion with some platforms than it was a few years ago.
• 24:38
  Ben Whitelaw: So we might see a rise of labeling once more.
• 24:42
  Ben Whitelaw: That's kind of our two stories that we we've drawn the line between AI persuasiveness and AI policy.
• 24:48
  Ben Whitelaw: So how the models are changing and how the platforms are changing in response to those models.
• 24:55
  Ben Whitelaw: We're now going to kind of.
• 24:56
  Ben Whitelaw: We're through a bunch of stories that are I, but which didn't quite make it into the top slots for today's podcast.
• 25:05
  Ben Whitelaw: Where'd you want to start Alice?
• 25:07
  Alice Hunsberger: Let's see, uh, I have one that I will try to talk about quickly.
• 25:14
  Alice Hunsberger: I
• 25:14
  Ben Whitelaw: ultimate challenge.
• 25:16
  Alice Hunsberger: well, I listened to your podcast last week, recently, in preparation for being on it this week, and you and Mike both talked about, like, I don't know if we'll be able to talk about this quickly.
• 25:26
  Alice Hunsberger: Now I'm in the seat, and I'm like, oh my god, I'm looking at all
• 25:29
  Ben Whitelaw: the words
• 25:30
  Alice Hunsberger: I have no idea.
• 25:31
  Ben Whitelaw: out.
• 25:33
  Alice Hunsberger: Uh, okay.
• 25:34
  Alice Hunsberger: So I, let me, let me see what I can do.
• 25:37
  Alice Hunsberger: There's a pretty cool study by Mark Scott that was published in tech policy press.
• 25:42
  Alice Hunsberger: He was a fellow at Brown's information features lab and spent his fellowship talking to people about public access to social media data, specifically, researcher access, API access, so that people can study what's actually going on.
• 26:00
  Alice Hunsberger: At social media companies and then put it out in the public eye, make recommendations, discuss it, understand more about what's going on behind closed doors.
• 26:10
  Alice Hunsberger: Uh, the interesting thing and why I thought it was pretty, noteworthy was he talked to a wide variety of folks.
• 26:18
  Alice Hunsberger: So he talked to regulators, public health officials.
• 26:22
  Alice Hunsberger: Social media, uh, people working at social media platforms.
• 26:26
  Alice Hunsberger: So trust and safety executives, which is like what I've done in my career.
• 26:31
  Alice Hunsberger: And then also academics and civil society.
• 26:34
  Alice Hunsberger: And I love that he included.
• 26:37
  Alice Hunsberger: Trust and safety pros in this, survey, and that they were just as frustrated about lack of access as the researchers were and a couple of the trust and safety executives who are not named said.
• 26:55
  Alice Hunsberger: They wanted to work with researchers, wanted to send over data, wanted to learn more about how they could be doing better, but the legal departments at their companies prevented them from doing so, um, over, you know, legal, yeah, like, you know, the worry that something bad about the platform is going to get out there and make them look bad or open them up to, to liability and.
• 27:19
  Alice Hunsberger: Of course that's a valid concern and also why are we stopping groups of people who care about making the internet safer for people from collaborating with each other?
• 27:31
  Alice Hunsberger: Like, it's not good.
• 27:34
  Alice Hunsberger: So, I thought that was really interesting.
• 27:37
  Alice Hunsberger: A lot of the time people are looking at tech companies with one broad stroke as though everybody working inside of a tech company has the same agenda.
• 27:49
  Alice Hunsberger: And I thought this was a really good example of where trust and safety executives are often on the same side as regulators and academics, and perhaps it's legal departments or overall company policy that's preventing the collaboration.
• 28:04
  Ben Whitelaw: definitely.
• 28:05
  Ben Whitelaw: And Mark, who is also Politico's chief technology correspondent, if you recognize the name, also talked about how regulation, particularly in the EU, was starting to kind of chip away at some of those, I guess, more conservative natures within platforms to not share data.
• 28:23
  Ben Whitelaw: Noted that data access was improving in some places and we had a few years ago.
• 28:29
  Ben Whitelaw: Twitter reduced the amount of data that researchers could see via their API and Reddit did the same, it's been talked about at length.
• 28:36
  Ben Whitelaw: And it does seem like according to Mark and the interviews he's done, that that is shifting back the other way, partly as a result of regulation coming out of Brussels, which is really interesting to note.
• 28:47
  Ben Whitelaw: So great summary, Alice, I think you did a good job.
• 28:50
  Ben Whitelaw: That was and sweet.
• 28:52
  Ben Whitelaw: Um, I'm going to have a job, keeping up with that.
• 28:56
  Ben Whitelaw: My, short story that I thought is interesting and worth reading is a Musk related story, unfortunately, but interestingly, uh, from Brazil.
• 29:08
  Ben Whitelaw: So a Brazilian Supreme court judge has basically ordered an inquiry into Musk after the ex Twitter, uh, owner.
• 29:19
  Ben Whitelaw: Basically decided to ignore or turn down a decision by the Supreme Court judge to have some accounts banned.
• 29:26
  Ben Whitelaw: Now, we don't know, you know, exactly what accounts that the Supreme Court judge wanted to be kind of taken down, but this judge is investigating fake news and hate speech.
• 29:37
  Ben Whitelaw: That happened during the government of the former Brazilian president, Jair Bolsonaro, and which kind of culminated in huge violence last year on.
• 29:48
  Ben Whitelaw: January the 8th, when Bolsonaro was, not elected again, after the election there.
• 29:54
  Ben Whitelaw: And so this is not the first time that the judge has gone after a tech platform.
• 29:59
  Ben Whitelaw: He ordered an investigation into some Google executives and some Telegram executives, and.
• 30:06
  Ben Whitelaw: There's a real war going on between kind of Brazilian government officials and tech platforms.
• 30:13
  Ben Whitelaw: And it really seems like, the tech platforms aren't backing down at all.
• 30:18
  Ben Whitelaw: You might remember, that Google put up a warning on its home page to users who are in Brazil, basically warning that there's a law coming, which is called the fake news law, PL 2630, basically warning that this law was going to change kind of internet.
• 30:34
  Ben Whitelaw: Freedoms for people in Brazil and basically trying to mobilize people to, talk to their local, politicians about it.
• 30:42
  Ben Whitelaw: And so there is a completely for, for several years now, there's been a log ahead with each other, the government and tech platforms, and this is the latest example of that, which is, I think really interesting.
• 30:52
  Ben Whitelaw: There's a really good piece that I've linked to in the newsletter today.
• 30:56
  Ben Whitelaw: And also we'll put in the show notes of actually an oversight board member and lawyer called Rinaldo Lemos, who says that basically Musk and lots of other tech companies have what he calls a much more influential, uh, have much more influence on the big tech than we'd like to realize.
• 31:14
  Ben Whitelaw: So there's a really interesting kind of ongoing battle there, which, and Brazil is a massive market with a huge growing population of people who are coming online, so that will continue, I think.
• 31:26
  Ben Whitelaw: Did you have a big user group in Brazil and the companies you've worked in previously Alice?
• 31:31
  Ben Whitelaw: What's it like at Grindr in Brazil?
• 31:33
  Alice Hunsberger: Yeah, Grindr has a lot of people in Brazil.
• 31:38
  Alice Hunsberger: So it's definitely an area that we paid attention to.
• 31:42
  Alice Hunsberger: I also think This is a really good example of how and why you should never believe companies when they say that they're politically neutral or will only follow the law or, um, are, you know, Elon Musk when he says he only, uh, he believes in, in free speech.
• 32:07
  Alice Hunsberger: But selectively, because the companies are this big can't help, but to get involved in political issues, even if they're trying hard not to their, geopolitical entities and of themselves, and they have, diplomats and lawyers and foreign policy agendas, just like countries do.
• 32:31
  Alice Hunsberger: And so, yeah, this is really a good example of.
• 32:35
  Alice Hunsberger: A company trying to say that they're neutral in some ways, but obviously having a specific agenda and that's true.
• 32:43
  Alice Hunsberger: I think of any really major platform.
• 32:47
  Ben Whitelaw: Yeah.
• 32:47
  Ben Whitelaw: And that, that's not to say that the Brazilian government has its heart in the right place because there has been criticism from a lot of civil society organizations about the fake news bill and the possible impact of that on, Brazilians human rights.
• 33:03
  Ben Whitelaw: So.
• 33:04
  Ben Whitelaw: It's interesting that you could either see Musk as this kind of anti government, freewheeling tech billionaire who's trying to kind of serve his own needs in the course of this, reaction to this Supreme Court judgment.
• 33:18
  Ben Whitelaw: Or you could see him as somebody who is, you know, you'd have to squint probably, but somebody who is pro human rights and is combating legislation in a which doesn't have a great Record of putting together kind of solid tech regulation over the last few years.
• 33:38
  Ben Whitelaw: And your view on Musk is, probably tainted by whichever way you, you have thought about him in the past, but he's done some really good work in.
• 33:47
  Ben Whitelaw: Turkey recently to try and push back against a government bill that was put forward.
• 33:52
  Ben Whitelaw: He has restricted data and has kind of stymied efforts to make research data available back to your point earlier in relation to, government takedowns.
• 34:04
  Ben Whitelaw: So that's no longer available for us to go through, but he is kind of bringing these very public, efforts against governments in a way that feels a bit more musky and I would say a bit more, it's all about me.
• 34:17
  Ben Whitelaw: This is, it's kind of theatrical in a way, rather than being systematic data led.
• 34:21
  Ben Whitelaw: So listeners can take their pick on still what, what that'll actually, where they, where they sit on the fence about Musk.
• 34:28
  Alice Hunsberger: I had one more link that is not big enough to actually talk about in depth, but is a really good example of, perhaps not being systematic and data led.
• 34:42
  Alice Hunsberger: Which was that, this week on iOS, Twitter slash X users saw that links to Twitter.
• 34:51
  Alice Hunsberger: com got displayed as links to X.
• 34:54
  Alice Hunsberger: com instead, because they're trying to get people to refer to it as X, but the actual URL itself, like the hyperlink.
• 35:03
  Alice Hunsberger: Remained in its original state.
• 35:05
  Alice Hunsberger: So you could do something like, create a URL, which is net flip twitter.
• 35:11
  Alice Hunsberger: com.
• 35:13
  Alice Hunsberger: It would show as a display as a link to netflix.
• 35:16
  Alice Hunsberger: com, but you click on it and then it actually takes you to net flip twitter.
• 35:21
  Alice Hunsberger: com, which is like.
• 35:24
  Alice Hunsberger: Perfect for phishing scams because it looks like the URL is legit because it says netflix.
• 35:29
  Alice Hunsberger: com you click on it.
• 35:31
  Alice Hunsberger: They clone the Netflix website.
• 35:33
  Alice Hunsberger: They ask you to log in and get all your information.
• 35:35
  Alice Hunsberger: So, uh, that got reversed, but I think it's another really good example of like.
• 35:41
  Alice Hunsberger: Why you need a really robust trust and safety team.
• 35:44
  Alice Hunsberger: And you need people who think like scammers and have spent years working, to fight scammers in the room when you're making product decisions, because it's like super easy to say, Hey, let's just make something that looks like Twitter show is X instead.
• 35:58
  Alice Hunsberger: Because.
• 35:59
  Alice Hunsberger: That'll be cool for marketing or brand awareness.
• 36:02
  Alice Hunsberger: And if you don't have a trust and safety person in the room to say like, no, don't do it, then this happens.
• 36:08
  Alice Hunsberger: So,
• 36:09
  Ben Whitelaw: Yeah.
• 36:10
  Ben Whitelaw: I mean, we talked on the podcast last week, Mike and I about the new head of safety, Kylie McRoberts, and the fact that in her first week, she's had to deal with something that weird and so kind of odd, you know, to the forcing of people to use x.
• 36:26
  Ben Whitelaw: com is in some ways proving that Elon is still running up the content moderation learning curve, which is, uh, Mike's favorite thing to talk about.
• 36:36
  Ben Whitelaw: Um,
• 36:37
  Alice Hunsberger: yeah, that's great.
• 36:38
  Ben Whitelaw: Just, just on that story, I love the fact that users on X slash Twitter started registering the domains in order to stop people from being phished.
• 36:49
  Ben Whitelaw: I thought that was amazing.
• 36:51
  Alice Hunsberger: it was great.
• 36:51
  Alice Hunsberger: It was so great.
• 36:52
  Alice Hunsberger: Yeah, I love it.
• 36:54
  Alice Hunsberger: And
• 36:54
  Ben Whitelaw: is good in the world.
• 36:55
  Alice Hunsberger: and that's the, you know, kind of good Twitter energy that people enjoy is like everybody getting behind a good project like that, you know, so it was very, very cool to see.
• 37:07
  Ben Whitelaw: I really like that.
• 37:07
  Ben Whitelaw: And then my other story that I want to bring, bring listeners is, uh, again, a story we talked a bit about last week, about verification, LinkedIn have started to verifying recruiters, to stop scams on the site.
• 37:20
  Ben Whitelaw: So again, It has over the last kind of year or so, try to push people to verify their authenticity by providing work email addresses or, or ID, through a system called clear, and it has now rolled that out to recruiters as well.
• 37:37
  Ben Whitelaw: So.
• 37:38
  Ben Whitelaw: Addressing an issue where, and I've had some of these myself, where people get in touch with you thinking, that you've got a great job offer lined up and try and talk to you about how your experiences are great.
• 37:48
  Ben Whitelaw: And they've got a perfect role for you only to find out that those people do not exist, which must happen to a lot of people.
• 37:56
  Ben Whitelaw: So, so LinkedIn have started to address that again through, verifying, which I think is good.
• 38:03
  Ben Whitelaw: Points to a proper broader trend in trying to ensure people are who they say they are, which I think is only a good thing.
• 38:11
  Alice Hunsberger: I don't think I realized that they didn't have verification for recruiters.
• 38:16
  Alice Hunsberger: That's
• 38:16
  Ben Whitelaw: Yeah, I know it's kind of funny that they started with the users and didn't start with recruiters because I feel like the incentives for, recruiters are much higher, the, the fact that they are often reaching out to people to, offer them jobs and, you know, people who are open to work are particularly vulnerable.
• 38:35
  Ben Whitelaw: In, in some senses, right?
• 38:36
  Ben Whitelaw: So it's maybe a bit of a backwards way around, but they got there eventually and we'll see how that kind of pans out.
• 38:45
  Ben Whitelaw: Awesome.
• 38:45
  Ben Whitelaw: That brings us to the end of today's Ctrl-Alt-Speech.
• 38:48
  Ben Whitelaw: We've whizzed through two big stories and a host of other little ones as well.
• 38:53
  Ben Whitelaw: Alice, how did you find today?
• 38:56
  Alice Hunsberger: it
• 38:56
  Ben Whitelaw: Did it live up to expectations?
• 38:57
  Alice Hunsberger: Yes, please, please invite me back pending, reviews of this episode, but anytime that either of you are out, I'd be happy to step in.
• 39:09
  Alice Hunsberger: I can pretend to be you next time.
• 39:11
  Alice Hunsberger: I can try to channel the British accent I used to have growing up.
• 39:15
  Alice Hunsberger: See if I can, see if I could do it.
• 39:18
  Ben Whitelaw: I'd love to hear that.
• 39:19
  Ben Whitelaw: I'd love to hear that.
• 39:19
  Ben Whitelaw: And I think you did a great job as, as a journalist.
• 39:22
  Ben Whitelaw: I think you've got a, uh, career in the media in the making.
• 39:25
  Alice Hunsberger: Yeah, we just need more people to subscribe to our newsletter and then I can.
• 39:30
  Alice Hunsberger: Just do that full time.
• 39:31
  Ben Whitelaw: exactly, exactly.
• 39:33
  Ben Whitelaw: So listeners do subscribe to everything in moderation to read Alice's work.
• 39:38
  Ben Whitelaw: Also subscribe to the Ctrl-Alt-Speech via the podcast platforms that you listen on.
• 39:45
  Ben Whitelaw: We mentioned last week that we were going to talk about some of the reviews that have come in.
• 39:49
  Ben Whitelaw: We've had a smattering since last Friday, and we wanted to save those till Mike was back in the chair.
• 39:56
  Ben Whitelaw: So we're going to look at those next week.
• 39:59
  Ben Whitelaw: You've got a chance.
• 40:00
  Ben Whitelaw: While he comes back from his holiday in Texas to leave us even more positive comments and, uh, we'll definitely go through them next week at the start of the episode.
• 40:10
  Ben Whitelaw: So thanks Alice for your time today.
• 40:13
  Ben Whitelaw: And, we'll see you all soon.
• 40:16
  Alice Hunsberger: Thanks, Ben
• 40:20
  Announcer: Thanks for listening to Ctrl-Alt-Speech.
• 40:23
  Announcer: Subscribe now to get our weekly episodes as soon as they're released.
• 40:27
  Announcer: If your company or organization is interested in sponsoring the podcast, contact us by visiting ctrlaltspeech.Com.
• 40:34
  Announcer: That's C T R L Alt Speech.
• 40:36
  Announcer: com.
• 40:38
  Announcer: This podcast is produced with financial support from the Future of Online Trust and Safety Fund, a fiscally sponsored multi donor fund at Global Impact that supports charitable activities to build a more robust, capable, and inclusive trust and safety ecosystem.