00:06:10 · Katie Prescott, Yoel Roth
[00:06:10] S2: But different in the fact that, I guess, in social media, you've got maybe more hate speech, illegal content perhaps that you're facing. Whereas it sounds like what you're saying is actually this is about people just proving that they are who they say they are. [00:06:22] S3: Yeah, that's right. I think in the context of especially public social media and social media that's focused on politics, I think interpersonal hostility and hate is much more of a consideration. And the nature of the technology incentivizes that. Unlike social media, you can't go viral on a dating app. And so the behaviors and incentives look quite different. But safety and respect are also very much a consideration on dating apps as well. Particularly with women, we see that being disrespected, abused, harassed, even overly sexualized is very much a factor that would lead to people being disengaged from dating apps. And so we work on a lot of the same questions as social media does, but with a bit of a different emphasis. Dating apps aren't free speech platforms. We are platforms that are focused on making one on one in person connections, and so we place a much higher value on safety and respect than perhaps some social media sites would who are prioritizing freedom of speech. [00:07:29] S2: AI must make your job so much harder.
00:07:29 · Katie Prescott, Yoel Roth
[00:07:29] S2: AI must make your job so much harder. [00:07:31] S3: You know, like any technology, I think AI is a double edged sword. On one hand, AI is empowering bad actors to move faster, but on the other hand, it's also empowering us to move faster in detecting them and mitigating the harmful behaviors that they engage in. To give you just one example of this, across our dating apps here in The UK, we've rolled out an AI powered feature called face check. Face check is a simple video selfie that users are asked to perform when they sign up for our apps. And on the basis of that video selfie, we're able to do a wide range of things from detecting duplicate registrations to keeping known bad actors off of the apps, all the way to giving you a verification tick if your profile photos match the video selfie. This is something that's only possible using advanced AI, and the impacts have been quite profound. We've seen a more than 60% reduction in people's exposure to bad actors on Tinder and Hinge as we've rolled out this product, and a more than 40% reduction in reports that we receive. So this is, I think, one of the really positive uses of AI in what we do. Equally, without a doubt, AI is empowering scammers, fraudsters, and other types of bad actors to move faster and more effectively. And so it's always about striking that balance between the positive uses of the technology and the possibilities of malicious abuse.
00:09:01 · Katie Prescott
[00:09:01] S2: Yeah, using an AI tool sounds a lot more effective than the Facebook group, Are We Dating the Same Guy? Which, where people upload pictures pictures of people that they're dating. I guess the challenges with dating apps as well is people on there are in some ways vulnerable. And that's what, you know, it's a horrible way to put it, but people are being preyed on because they are opening their hearts to finding someone. And, you know, it's kind of awful, but it does also mean that it's ripe for kind of bad actors to try and abuse that.
00:09:33 · Yoel Roth
[00:09:33] S3: Yeah. I mean, dating is one of the most intimate and personal things that you can do. You are literally putting yourself out there to people that you don't know and are seeking that most fundamental of things, which is human connection. And especially in a world where we're more and more connected to other people, but find it harder and harder to actually translate that into friendship or relationships or in person connection, I think you're right that there's a real vulnerability there. And without a doubt, scammers are exploiting that vulnerability. We see that AI powered scams have been accelerated by things like large language models and chatbots, which make it easier than ever for scammers to write persuasive sounding messages that can build that trust and confidence. It's no longer as easy to spot a scam because, you know, the language doesn't quite make sense or it seems like it was written by somebody who's not a native speaker. All of these things are streamlined through the use of AI. And so we place such a high value on investment and safety precisely because people come to our apps at that moment of greatest vulnerability, and we believe we really have a responsibility as a company to protect them using all of the resources and technology at our disposal.
00:10:53 · Katie Prescott
[00:10:53] S2: I'd love to ask you about AI more broadly in the online misinformation and disinformation speech. As I said at the start, mean, you're such a deep thinker on this subject and someone governments turn to for your advice. How do you think we are tackling it? And are we doing a good enough job broadly around the world to think about the rise of misinformation and disinformation?
00:11:17 · Yoel Roth
[00:11:17] S3: You know, my personal reflection on the state of AI discourse and especially regulatory discourse is that we sometimes are a bit distracted by the existential risk conversations and can be underfocused on some of the day to day challenges that emerge around this technology. It's certainly worth giving a bit of thought to the possibility of killer robots that will end humanity. I don't want that either, to be clear, but there are very practical challenges that we're already seeing when it comes to things like election security or even AI powered scams. We see in the research that AI language is more persuasive than ordinary human conversation, and that supercharges disinformation campaigns. We see that AI powered imagery makes it easier for scammers to create convincing social media profiles. We see that agentic AI makes it faster for bad actors to automate multiple accounts. Those are the types of issues that I believe the tech sector and regulators need to be focused on. And certainly existential risk needs to be part of the conversation, but it's not the one that I would prioritize as highly as the day to day challenges we're already seeing. So when I look at a problem like the future of AI, there's both the technical side of it. How do we identify bad actors? How do we prevent misuse? How do we build things responsibly? But then also the normative side, is like, as humans, have we figured out what it is that we collectively want from these technologies? About two thirds of singles in a recent study we conducted in The UK say that they've used AI in their dating lives, but used can mean a lot of very, very different things from brainstorming creative ideas for a first date to writing messages to deepfaking your profile photos. And perception of the acceptability of AI usage varies dramatically across these uses. We need to be specific when we're talking about AI so that we can think about the right interventions.