Welcome to Tech Talk by CT. Team.

So hi. I'm Alia Bhatia, and I'm excited to host this episode of Tech Talk. Today, we're talking about age verification and what it means for users. To understand how users interact with these systems and perceive them, I'm happy experts who, amongst other things, have conducted research on users and user perception of age verification. My colleague, Michal Luria, and our special guest, Ziv. Thanks so much, Michal and Ziv, for joining us. So before we dive in, I want us to do a quick note on terminology. You know, we're talking about age verification, which is part of this larger umbrella of age assurance methods. Age verification typically refers to methods that require hard identifiers to prove age, such as users showing ID to prove their age or maybe some approaches have also used collection of PII, personally identifiable information, or biometric data such as face scanning despite these approaches being more estimatory. Obviously, there are age estimation methods as well that have been proposed such as using machine learning or inference methods to estimate a user's age based on face scanning or voice analysis or analyzing user behavior such as search history. So, Ziv, I'll start off with you. Your recent research has been concerned with how users, specifically in Louisiana, have changed behavior about what they access online in response to a Louisiana age verification law. Can you walk us through the study, what you were interested in studying, and what types of age verification

this is a thread I really wanna pull on in our conversation as we, continue it in terms of, like, why it's really important to understand sort of the efficacy of interventions or perceptions of interventions. Michal, I wanna loop you in. You've also been looking at how users think about or perceive age verification systems. Most recently, you put out a brief, think you know, which covers how parents and teenagers think about different age verification approaches. Tell us a little bit more about this research and what age verification approaches you included in this study. Absolutely. And, glad to be here today.

So this work that we did on age verification was part of a broader research study where we wanted to better understand how young people and their parents perceive a range of child safety policy proposals. We've seen that policy making and research many times occurs on these two different planes, and we wanted to bring that together. So we looked into all kinds of things that are being proposed in the in the policy space. And for age verification, we talk to people about several approaches, ID based verification, face scanning for age estimation, device based and app based verification, and parental verification. And so these different approaches, we talk to people because we used, a human centered design research approach, and it's not always well understood as, differently different from technical or usability evaluations. So unlike technical evaluations that focus on feasibility or usability testing that emphasizes, efficiency and ease of use, Design research allows people to explore a wide range of possibilities, for future interactions. And when we talk to them, we talk to them within a particular context, and we try to talk through what a particular scenario might look like. And so when we engage with teens and their parents in these structured conversations, we were trying to capture what would work, not on a technical level, but what aligns or misaligns with their lived experiences

Were there ones that were preferred to others? Yeah. I think, one common thread that we've seen across all the different, situations and scenarios that we presented was that people were highly concerned about their privacy and we're very sensitive to uploading any kind of verification or providing any type of verification that requires identifiable information, ID, even syncing with someone's phone. People perceive a phone as a highly private possession and that has a lot of information on it. And so anything like that requires a lot of transparency and really making sure the users walk through exactly what is being done. And so this is kind of a thing that kept coming back and we kept coming back to with the

with people. There seems to be this gap also between how users perceive systems or, like, what they, you know, how they trust these systems and then also how they work? Like, why why does perception matter so much in terms of, like, the actual implementation of these systems? Well, it matters because

at the end, people are going to be encountering these technologies, and they will be responding in various ways. It's not about what can or cannot be done technically. Of course, it's a it's an important component. But what we look at when we do human centered research is we try to understand how people are going to encounter these things. And that's why research, like the research that Ziv did, is so important because it actually looks into real reactions and real things that are happening, and that really can determine whether some kind of intervention succeeds succeeds or not. And so, Ziv, I'm curious for you to talk a little more also about how you think that implies back to policy. Is there any are there any tweaks that could make this better, or are you mostly looking at what others should or shouldn't do? Yeah. So I think this is a really important point. Right? Like,

circumvent or find the content that they're looking for or find things that they're looking for? Like, how comfortable, I guess, do you fee you know, how do you perceive the findings in this study or this pertaining to the Louisiana law to extrapolate into other instances? Like, are there

in places like Mississippi, so that's probably not the goal.

Hopefully, not the goal. Right. Yeah. There is this, like, sort of question of, like, proportionality. Right? And, like, also, I think a little bit of, like, disaggregation that needs to happen in terms of, like, what problems we're trying to solve. Like, I think sometimes age verification is, like, you know, if we're trying to propose age verification approaches because some of the tech has advanced, that sort of neither here nor there if the problem we're trying to solve is more nuanced. And and, Mikala, I think in some of your work, including this sort of, study where you're trying to understand teens and parents' perception of proposed policy and or technical interventions, but in also your other work on, like, how do users, specifically child users, act when they encounter sort of negative messages on messaging platforms. One of the threads that you've pulled across all of this work is, like, parents and teens together, but also separately have really sophisticated tools to keep children safe online. Can you talk a little bit about, yeah, what are the other approaches that

you found have helped? What are parents doing in this, like, new sort of complex landscape of the big bad web? Yeah. That was really interesting. And as a parent to two toddlers, I am very I I got a bunch of perspectives on what to do with teens, so I I feel prepared. But, I think one thing that has been really interesting for me is that parents really take a range of approaches and families adjust to their needs, which seems like an obvious statement. But in this context, we've seen that parents and kids work through what works for them in the context of technology use, in the context of platforms and social media. And so one thing that we've seen that was particularly promising was parental age verification so that parents have a a way of verifying their child's age. And one reason why that was particularly promising for people was that it allowed to keep the decision making within the household, and it allowed some flexibility that is needed. Because as we know and lots of research has said again and again that children are different, their maturity is different, how they approach platforms is different. And so having this option allowed parents, to a, get a better sense of where their where their kids are going online, what platforms they want to join. So it also gives them a way to kind of consent to what is happening. But the other part of it is that it gives them flexibility to decide together to lie about the the child's age, which I found really interesting because it's something that kept coming like, it wasn't only one or two parents who mentioned it. This was something that happens that parents and teens decide together. Okay. You're not at the right age yet according to, you know, social media's, recommendation, but you are mature enough, and we think that you're ready to go on to this platform. And the and vice versa. We've also seen cases in which a platform, Snapchat, let's say, is from a particular age and the child is older than that, but the parents and teens decide together that maybe it's too soon for them to go on that platform. And so this approach allows for more flexibility for parents and teens to have a conversation and to decide together. I will say this wouldn't work for all households. There are some kids who don't have that kind of support and communication system with their parents. And so that leads me to another important finding that we are highlighting in this work is that we need a variety of approaches and a variety of options because, of course, there was no no consensus on all the different age verification methods that we talked about. And different people feel differently and feel a range of comfort around different age verification methods. And so giving users a little menu of possibilities would really help, especially as this is kind of becoming a standard to get a sense of what they can do or or choose something that they feel personally comfortable with. Yeah. And it it's it's interesting how far we are away from

Yeah. It seems like there's a lot of work that can be also done at, like, the standards level here of, like, trying to create a floor of, like, what privacy preserving or what anonymity, anonymity preserving age verification could look like. I know that the Louisiana law is actually, like, singular in mentioning even anonymity in a way that, like, the other state laws aren't. But, you know, our colleague Nick Doty has been thinking a lot about, like, what are existing, privacy standards that can be used to create alternative approaches to enable, parents, for example, to say, okay. Sites that are already labeled, adult content, like, how can labeling and signaling methods try to lead to more of a robust browser level set of controls where parents can say, okay. No adult content can be accessed on, like, safe mode here. So there are other approaches, but I think instead, in addition to what you're saying of, like, the incentives that these bills are creating, you know, a lot of these bills use, like, language like, use commercially viable methods to age verify, which to me seems like a silver platter delivery to the sort of cottage industry of age verification providers regardless of, like, actually what, you know, their data collection retention policies are, exactly what data you know, how accurate their systems are, etcetera. And in what we what we saw in The UK, like, a lot of these age verification mandates are just incredibly burdensome and, like, cost prohibitive and are, you know, to Michal's point, like blue sky leaving Mississippi or footy club saying, actually, we're not going to have new accounts register anymore because we can't verify those accounts. So you can, like, consume the content a little bit, but we have to, like, limit access to a lot of it as well because, you know, we know football conversations also, like, go awry, I guess. So I I think there's a lot of factors that sort of, like, poke holes at the claim that age verification is, like, a perfect child safety size Band Aid error. I guess, like, you know, I I really I have a couple of interesting I'm really interested in your, like, the larger argument of why empirical research here is needed. But before that, I think one final question about the findings, like, I think what you know, if we wanted to package some of these findings to policymakers, like, what would we be saying? Like, what does age verification mean for users and and what they will and will not,

if I can add one more complaint, I think there and this is also a classic researcher move, so apologies for that. But we need more human centered research, guys. I mean, if there's just not enough research here and I'm so glad that we're having this conversation and that we do have actors, and organizations who are having this research done, but it's not enough. And I think a lot of the conversation has just been about the technical feasibility of approaches, and that's not the whole conversation. Like I said, it's an important part of it, but it can't be tested technically alone. It has to be tested and piloted and shaped with the people who are going to be impacted and who will be using these platforms.

part two of the converse tech talk conversation with you and Michal. So, but it's it's certainly something we're thinking a lot about And this sort of platform, like, very narrow platform view of kids' safety legislation, I think, is 100% accurate but also scary because, like, now we're gonna see this, like, whole new flurry of bills that are, like, prematurely trying to be, like, let's, you know, stop everything, halt the horses. I'm just gonna take moderator privilege also to say, you know, beyond, to to answer that question of, like, I think what policymakers need to know, which is, you know, not only the sort of socio technical context in which these age verification laws are being implemented, the the importance of sort of doing this human centered research and understanding user perception as the proxy for effectiveness of these laws. But, also, to poke holes a little bit at the claims that age verification systems are technically feasible all the time. You know, a lot of our research, in the policy front at CDT has been focused on the fact that these so called, you know, 100% accurate, verification and estimatory methods work across all users. You know, a lot of the research we found is that, users with disabilities, users of color are routinely misclassified by these more machine learning or inference systems. Verification systems also, fall short, for a lot of users who don't have ID, don't have valid ID. And I always like to sort of quote Kate Ruan, the director of our free expression team at CDT, who says, you know, even a single small you know, single digit error rate when it accounts for the scale at which people are using these services is gonna be a big number. Right? Like, a fraction of a a large number is gonna be a large number. So I think what we're talking about is, like, thousands, if not tens of thousands of people who are gonna be misclassified and are going to be stuck in this dead end of accessing the web, whether to provide even more sensitive information to these already data rich tech companies or for go service or, as the viewer studies have shown, like, access sort of, like, more murky territory on the Internet. So all to say that I think I think this conversation has been fascinating, has sort of raised a lot of questions that I think I hope policymakers, industry experts, researchers are sort of penning down and and pursuing in their own sort of Internet research holes. And we could go on for so much longer, but thank you so much for your time. Thank you. Thanks for having us.