Podcast
AI Is Not Your Friend: Geoffrey Fowler on Rating AI for Kids
Season 3 Episode 23 •Show Notes
In this episode of Cyber Focus, Frank Cilluffo speaks with Geoffrey Fowler, head of public engagement for the Youth AI Safety Institute at Common Sense Media, about why AI requires a different kind of safety framework than movies, apps, games, or social media. Fowler argues that generative AI is not static content; it is dynamic, conversational, multipurpose, and capable of changing from one interaction to the next based on the user, the prompt, the model, and the length of the conversation.
The conversation explores how AI products that appear friendly, educational, or therapeutic can create new risks for children, from emotional dependency and privacy concerns to unsafe mental-health guidance and weakening guardrails over extended conversations. Fowler explains how Common Sense Media is working to build independent AI safety ratings for kids, modeled in part on crash testing for cars: transparent evaluations that help parents and schools make better decisions while pushing companies toward safer design.
Main Topics Covered
- Why AI needs a new safety rating
- Lessons from social media and smartphone adoption
- AI companions, mental-health claims, and dependency risk
- AI toys, privacy, and weakening guardrails
- Independent testing, ratings, and child-development standards
- Company responsibility, public policy, and trust
Key Quotes
“AI is not your friend. AI is not human. It does not make the kinds of choices that a human being would make when you’re having a bad day or when you’re in a crisis or when you need somebody to really trust.” — Geoffrey Fowler, Common Sense Media
“[AI companies] shouldn’t be experimenting on our kids. They should make it safe from the get go.” — Geoffrey Fowler, Common Sense Media
“These AI toys are little spies that you’re putting in kids’ rooms. They’re recording their voices, they’re recording behavioral data.” — Geoffrey Fowler, Common Sense Media
“The Common Sense Media Youth AI Safety Institute is neither pro AI nor anti AI. It’s pro kid.”
— Geoffrey Fowler, Common Sense Media
“We are here to research not just the hype of what companies say about what their technology does, we’re here to see what it actually does and tell the truth about it.” — Geoffrey Fowler, Common Sense Media
Relevant Links and Resources
https://www.commonsensemedia.org/ai-ratings/ai-risk-assessments
Guest Bio
Geoffrey Fowler is head of public engagement for the Youth AI Safety Institute at Common Sense Media. He is a longtime technology journalist whose work has appeared at The Washington Post and The Wall Street Journal. In this role, Fowler helps communicate Common Sense Media’s work to evaluate AI products used by children, teens, families, and schools, including the development of independent safety ratings and risk assessments for youth-facing AI tools.
Transcript
Geoffrey Fowler [00:00:00]: There are a lot of AI products that look and act like a friend, but that is not the case. AI is not your friend.
Frank Cilluffo [00:00:11]: Welcome to Cyber Focus from the McCrary Institute, where we explore the people and ideas shaping and defending our digital world. I’m your host, Frank Cilluffo, and this week I have the privilege to sit down with Geoffrey Fowler. Geoffrey is a longtime tech journalist at the Washington Post and Wall Street Journal, and he recently joined as head of public engagement for the Youth AI Safety Institute at Common Sense Media. Really excited to sit down with Geoffrey today. We’re getting into some terrain that we haven’t historically touched on in Cyber Focus and really excited to expand our views, our horizons, and hopefully raise some interesting questions. Geoffrey, thank you so much for joining us today.
Geoffrey Fowler [00:00:56]: You bet.
Frank Cilluffo [00:00:58]: So, you know, we’re getting into some new terrain for, for our institute and for, for our podcast here. And, you know, Geoffrey, 20 years ago, we, we handed children smartphones. They were starting to engage in social media. I’m not sure we thought through what the consequences were developmentally, psychologically, societally. Now we have AI, and the advent of that is coming on even faster. What should we be thinking here on the front end? And not to lead the witness here, but it is moving fast, and there are some significant questions I think we want to get out in front of. So how do we try to do that?
Geoffrey Fowler [00:01:42]: Yeah, well, as you said, we made some mistakes with mobile tech and particularly with social media when it comes to kids. And now we are kind of a little bit into the game, starting to understand the harms that those that those technologies have had on kids. And we have, I think, an opportunity to do a better job with AI tools, particularly generative AI tools that are coming online now. And I would say we actually really, really need to. We need to figure it out really quickly, because kids are adopting these tools faster than they have past technologies. You know, it took a fair amount of time for teenagers to get on social media, but we now know the majority of American teenagers are using generative AI. So we need to be understanding this right now. We need to be studying, studying the impact of this technology on kids.
Geoffrey Fowler [00:02:39]: We need to understand, you know, in what ways it can help and in what ways it can hurt. And then we need to be setting some standards for the ways that AI technologies should treat children, should interact with them. And it’s a completely different set of problems than we’ve had with past technologies or past kinds of media. I’m sure we’ll talk about that today. And then we need to hold these companies to account for them and hold their products to account because they shouldn’t be experimenting on our kids. They should make it safe from the, from the get go.
Frank Cilluffo [00:03:13]: Well said. And we often think that social media, the individuals were almost the beta test. And we do have an opportunity to get out in front of this and lots to unpack there. But before jumping into some of the specificities, you know, common sense has, has rated media and technology for families for a number of years now. What makes AI different and why do we need a new kind of safety rating here in particular?
Geoffrey Fowler [00:03:45]: Yeah, right. So we have for 20 years been rating movies and telling parents how they stack up on different criteria. But a movie is a static thing, right? It’s, it’s produced, it comes out and, and you could watch it, you can maybe understand it from different dimensions, but it doesn’t change. But a chatbot, a furry cute toy with conversational AI built into it, is dynamic. A frontier model can obviously change with every new version or drift while it’s going on. Every prompt, every conversation can be different. It can also change depending on the state of the user, like their emotional state. So there are so many more complicated dynamics going on here.
Geoffrey Fowler [00:04:33]: The AI products that we’re talking about right now are generative. That means that like, there are different answers and different responses given every time. It’s not like, it’s like a, a library of movies that we can just kind of work our way through slowly. This is about behavior of models. It’s just a completely new, new kind of product. Also, I should say, like, these AI products that we know tweens and teens are using are multipurpose. And that changes the dynamic as well. It’s not that they’re just using it to catch up with friends or just using it for homework or just using it for entertainment.
Geoffrey Fowler [00:05:12]: They’re doing all of these things. They’re using it for emotional support. There’s just such a wider range of ways that these tools promise to, to interact with the lives of kids, that there were just more questions and also then more risk that comes from each of those different kinds of touch points.
Frank Cilluffo [00:05:30]: You know, that, that, that tees up a great question, and that is where do parents most likely misread risk? Because at the end of the day, you’re right, it’s not just about weeding out and blocking explicit content. It’s, it’s sort of how AI interacts with developing minds over years and over time. And, and, and I’d be curious, what should parents be thinking about and what do they most likely misread in terms of risk?
Geoffrey Fowler [00:05:59]: I think number one on that list is that there are a lot of AI products that look and act like a friend and you think like, oh, it seems to be responsive to me or to my child, so that must be okay. But that is not the case. AI is not your friend. AI is not human. It does not make the kinds of choices that a human being would make when you’re having a bad day or when you’re in a crisis or when you need somebody to really trust. So I would say to families, be extra wary of anything that looks, any AI tool that looks like it’s a friend. Right underneath that, I would say be, be wary right now of anything that, that seems like it is or claims to, to be helping with mental health. We just published as the first new research from our Youth AI Safety Institute at Common Sense, an evaluation of these purpose built therapy chatbots. So these are, these are chatbots that go beyond, you know, ChatGPT or Gemini that you know, you can, you can like talk with about how you’re having a bad day.
Geoffrey Fowler [00:07:17]: They claim clinical research, they have, they claim clinical standards boards. But this is a wild west. There is no accountability for these apps. In fact, while we were testing these like several of them literally just disappeared from the market without any notice to the users, without any like backup support and who knows what happens to your data when you’ve given to them as well. So be very aware right now, the wild west of things that are claiming to be for mental health. And then in the same category, be very aware of anything that claims to be educational for kids. Right? We, one of the things, one of the reasons why we need this institute that we just started at Common Sense is we need to figure out the difference between AI products that really could maybe like be a tutor to help a kid in ways that they’re could be more patient than a parent can be in certain circumstances versus ones that are just maybe giving the answers to kids. And in some of the initial testing we’ve done some of these products, it is shocking how easy it is even for the ones that claim that they are purpose filled for education and for tutoring, how easy it is to get them to just do your homework for you and what is going to be the impact on the learning development of kids from those.
Geoffrey Fowler [00:08:35]: So that’s what I would say to parents, be wary of those things because what can start to happen with all of those is over time kids could develop dependency on it and that’s one of the risks that’s just emerging that like, maybe our lonely kids going to end up spending more time with Chatbot and pretend to be their friend and claim to be doing things, and then does that actually help them in the long run? Our kids who are having trouble in school of going to end up using AI more, but then end up relying on it more as a crutch. These are just some of the questions that are starting to emerge that I would warn parents about.
Frank Cilluffo [00:09:11]: Geoffrey, so much there and, you know, to step back just a teeny bit, how is it, and I know we’re early innings in terms of where the institute is in terms of evaluating AI products, but how does that, what does that look like? What does the evaluation process, what is it you’re evaluating? And obviously we’re learning. I’m sure you’re learning. And it’s not like this is the Ten Commandments etched in stone. We got to be thinking ahead as well. But what does that process look like and where are you on that journey?
Geoffrey Fowler [00:09:46]: Yeah. So for the past couple of years, Common Sense has been doing what are called risk reports, where we’ve been essentially red teaming AI products with the help of experts in child development and child psychology, of using them like teenagers, tweens, children would use them and seeing how they respond in kind of real world adversarial situations. So that work has been already happening. And because of that work, we have labeled already certain AI products as unacceptable risks for kids. For example, one of, in our most recent report we just put out, we labeled one of the most popular therapy chatbots that’s available to teenagers as an unacceptable risk for kids. So that work has, has, has happened and is now ongoing. But with this new institute, which we’ve, we’ve just launched and we’ve got some new funding from a range of, of, of philanthropic and industry funders, we’re going to try to now move this to something closer to the kinds of, of evals and testing that the AI industry is already doing. So you might be familiar with, you know, when a new model comes out, there’s, there’s companies that go out and very quickly try to say, okay, rank them which one does best on different kinds of criteria, or test it on something like, you know, how well does it pass the MCAT? We want to do the same thing, but for safety for kids.
Geoffrey Fowler [00:11:14]: And that’s been the missing piece of it. You can tell right now, like which frontier model might have the fewest errors on medical knowledge, but you don’t know which is going to be most developmentally appropriate for young people. So how do we do that? So step one is we got to figure out what the standard should be, essentially what we need to test for. And that working had literally just started, and the last week or so Common Sense Media, working with child psychologists, we have great partners at Stanford, at the Stanford Medicine, psychiatrists, folks from the American Academy of Pediatrics, educators, all of these kinds of experts, pulling them together to try to set okay, what are the, what are the, what are the issues that are highest risk? And we’re going to start with those. I’m talking about things like, like when chatbots could encourage self harm or suicide. I’m talking about when AI tools could do things like encourage or deepen eating disorders and disordered eating. So just going through and trying to figure out what are the things we’re most concerned about and what should the standards be. And then the next step will be, and again we haven’t done this yet, is starting to test that in the frontier models and in the live products in an ongoing way. We’ve got some technical partners that are helping us with that of organizations like Translucence and Humane, which are our technical organizations that already do certain kinds of evaluations of AI products but are now going to help us do it on these kids safety standards.
Frank Cilluffo [00:12:56]: Well that’s a massive effort and dare I say your why is clear and there is an incredible need. I would also think that getting to that framework, because I think what we’re really struggling with here, this isn’t a one and done. It’s not like it’s a single transactional set of issues. It feeds off of one another. Right? And it can reaffirm aberrant behaviors or it can empower powerful and meaningful behaviors. And to me, my kids are older now so I don’t have to worry about this. But I think I would have spent a whole lot of time worrying about what they’re engaged in.
Frank Cilluffo [00:13:42]: And if I’m being very honest, I was the last to know. So it was always sort of through their friend groups and, and, and, and what have you. But I did, there was something you had mentioned earlier about toys themselves and they, they build a relationship with some of that. What, what, what does that look like from a behavioral standpoint? I’m not a psychologist, I don’t play one on tv. I, I, I, I barely know how to ask good questions and occasionally I get one right. But what, what does that look like in your eyes?
Geoffrey Fowler [00:14:16]: So I should also say I am not a psychologist. But one of the things, so we’ve, we’ve done some of those initial risk assessments of this first generation of toys that are coming out that embed generative AI conversational capabilities in them, trying to understand the risks of them. And there were a number of them. And one of the biggest ones is that it just, in the mind of a young person, it encourages an ongoing relationship. Remember these, these, these bots, these, these products are in many cases being made by the same people or even the same companies that made social media. So they’re built with engagement in mind. That means that they want to keep you coming back. And if you’re an adult and you can understand that this thing has tried to hook me the way that social media did, maybe you can push back against that.
Geoffrey Fowler [00:15:08]: But if you’re a kid, you might have a harder time. Or if you’re a kid with a emotional or other kind of psychological challenge, this could actually be pretty actively dangerous to you, right? Like, because, because these, these, these products would claim to be there for you all the time 24/7. They are encouraging you to spend more time with them rather than spend more time with human friends or parents. You know, technology that gets us, one of the board members of Common Sense Media is Vivek Murphy, who used to be the Surgeon General of the United States on two different occasions. One of the ways he talks about this that I find so useful is technology that helps us connect with other humans is great technology.
Geoffrey Fowler [00:15:57]: Technology that would promise to replace a human for the purposes of engagement, to get you to subscribe, to get you to use it more is, is potentially dangerous technology. So that’s a thing that I would say is we have to watch for in, in these toys. But you know, since this, this audience here is, is interested in cyber, you know, there’s all these other kinds of risks that come too.
Frank Cilluffo [00:16:18]: Yeah, no kidding.
Geoffrey Fowler [00:16:19]: So there’s emotional dependency. What about privacy? I mean, these, these AI toys are little spies that you’re putting in kids’ rooms. They’re recording their voices, they’re recording behavioral data. They’re doing all these sorts of things that like, you don’t want that going on in lives of your kids. And again, there’s no standards for that yet.
Geoffrey Fowler [00:16:37]: But that’s part of what we hope to set with, with the Youth AI Safety Institute that like, this is how these things should work. Another thing I would just say is like, and this maybe is more obvious, but it’s worth calling out like, these toys with AI conversational abilities built into them are versions of products that were made for adults. They were trained on the vast corpus of information on the Internet. And I don’t have to tell anybody who’s listening to this conversation today that the Internet is filled with awful things that you don’t necessarily want kids talking to or engaging with or being brought, brought up to them. And you know, try as they might, some of the makers of these products to like remove some of the content, remove some of the stuff that isn’t age appropriate for kids, they fail. Like, and we found that happened repeatedly when we test these AI toys and some of these other products that they just like the content guardrails aren’t there that you would want, particularly after you have just like maybe a couple of rounds of a conversation. The further you get, the worse their guardrails get.
Frank Cilluffo [00:17:44]: Hey, Geoffrey, and forgive me, maybe I should know the answer to this, but I don’t. Are we thinking, in addition to ratings, are we thinking sort of like a Good Housekeeping seal of approval or at the very least an ingredients list? What’s in, what’s not, what’s, I mean, because at the end of the day, even eating has become a challenge. We’re not just hunting and gathering. You got to know what’s inside our, inside our food. Is that what you’re thinking about long term to sort of give some of the ratings, but also then give the what’s actually inside, where’s the beef if there is any in it. So…
Geoffrey Fowler [00:18:25]: It’s a great question. It’s one of the reasons I joined this initiative because we got to figure out how do we gather the information that is going to be most useful, first of all, to parents and educators who are making decisions about whether to bring these products into the lives of families and into schools. You know, they’re making, educators are making procurement decisions about which products to bring in. So how do we support them? So I don’t have the answer yet, but that’s exactly the kind of question we’re asking. I mean, if we take a big step back. Our model here is crash testing cars. Back in the, you know, crash test dummies and cars have existed for since like the 60s or 70s.
Geoffrey Fowler [00:19:10]: But back in the 90s, there was this organization called the Institute for or the Insurance Institute for Highway Safety, IIHS. And they started rigorously and systematically testing cars and giving them crash test ratings. So you could tell before you bought a car, like this car, you know, this model did pretty well. This model was a mess. What’s interesting about this is this did not come from the government. This was a nonprofit effort to get the industry kind of aligned towards making safer cars. And guess what? It worked. Cars in that sense have gotten more crash resistant, safer.
Geoffrey Fowler [00:19:53]: I mean there are other issues at play as to like whether or not there are more injuries or deaths these days. But cars have gotten much safer because there was a race to the top when it was transparent about how they dealt with and how they applied different standards for safety. And so that’s exactly the model here for our AI safety ratings for kids.
Frank Cilluffo [00:20:18]: And do you have any early sort of insights into the use cases, whether it’s Claude or ChatGPT or Grok or any sort of racking and stacking early innings in terms of what you’ve seen?
Geoffrey Fowler [00:20:33]: Yeah, so in the risk evaluations we’ve been doing, so those are the more, you know, human red teaming kinds of kinds of evaluations, we’ve definitely found some differences that parents should be aware of. The highest risk was Grok. We, for many reasons including that it’s been very documented that it is producing, you know, non consensual imagery, sexualized imagery and it’s built into a social network for that to be immediately, for that to be immediately shared. So Grok made our unacceptable list. Many of the toys that we talked about made our unacceptable list. One of those purpose built therapy apps made our unacceptable list. A lot of these AI companions which sort of again, the whole premise is to be a friend or a kind of like, you know, lightweight therapist made our unacceptable list. Slightly better but still high risk according to our early, you know, risk assessment evaluations are things like ChatGPT. You know, it, it does now have some parental controls and some age aware responses if it knows what your age is.
Geoffrey Fowler [00:21:54]: But it’s still risky for the things that most teens use it for, for emotional support and for mental health advice. Also we’ve seen with ChatGPT that safety can degrade during extended conversations. The more you chat with it, the more, you know, the more dangerous it becomes. And that’s what’s led to some of these cases where we’ve seen cases of self harm and suicide from people using it. So I guess we would tell parents about ChatGPT, like maybe it could be useful for learning and creativity, but not what teens are actually maybe using it for, for emotionally and for entertainment on those sorts of things. And then at the sort of moderate risk level I would, we, we put things like Claude. So again not, it doesn’t mean that like there’s no problems there. And generally like, you know, it’s fine to use with adult supervision, but still would not recommend it for emotional support.
Frank Cilluffo [00:22:51]: You know, Geoffrey, and we touch on this very early on in our discussion and it’s sort of let’s not allow our kids to be the beta test. Let’s flip that a little bit. What would you be suggesting? I mean, there is a lot of competition in the frontier model business and dare I say, I think there’s a time and a need where we have to have almost a social compact where some of that onus and responsibility is not in the hands of the government anymore. It’s in the hands of a handful of companies that are driving innovation for good, bad, indifferent and everything in between. What would you be advising a couple of these frontier models to make sure they at least put what they want to, if they want to put guardrails in, what could some of that look like? And how do you do it in a way where you don’t erode and undermine innovation and competitiveness? Which I know is a, it’s a 50,000 foot level question, but I’m hoping they want to do right by our next generation too. And what could that look like?
Geoffrey Fowler [00:24:03]: I would hope so as well that they want to do right. But right now I’m based in San Francisco. It’s gold rush days and people are, and organizations, companies are, you know, just putting stuff out there that, that claims to do things that it just simply cannot and is not safe and not appropriate for kids, even though they might, might claim it is. So I guess I would say step one is like, they gotta get out of the gold rush in a race mentality when it comes to products that are gonna be used by kids. And they gotta think about kids differently than adults. They have to not, it can’t just be like, like a, you know, the adult product with like some parental controls on it. It has to be designed for kids from the get go.
Geoffrey Fowler [00:24:57]: You have to really think about kids as their own audience there. I think these organizations should be open to external evaluations and testing. Here’s the problem. They, you know, Google and ChatGPT want to be adopted by families, by schools, by, you know, all aspects of society. But they have a problem. People don’t trust them. People don’t trust these companies and they particularly don’t trust them when it comes to things that are important in their lives like their kids and their health and whatnot. So if they want to succeed, if they want to get adoption, they, they have to prove themselves to be trustworthy.
Geoffrey Fowler [00:25:39]: How do you do that? You do that through independent evaluations. Not just the company saying, not just Mark Zuckerberg saying, oh, yeah, trust us, we got it right this time. We need outside voices that can say, okay, here’s the bar, here’s the standard, and here’s how these different, these different, both frontier models, but then also the end products stack up on that. Because sometimes those are two different kinds of questions. The frontier models obviously power what goes into lots of end products, but lots of design decisions in these end products can really make a big difference. Let me give you an example. On the therapy bot research that we came out with this week, we actually saw both bad news and good news. The bad news was, you know, some of these bots, you know, it was a wild, wild west market for these things.
Geoffrey Fowler [00:26:35]: As I mentioned, some just, just disappeared while we were testing them. Some missed major signs of, of someone in serious psychiatric distress and didn’t do the right things. But then there was another set of therapy bots that were designed to be integrated into a school environment, into a, an environment where there were humans. So instead of trying to fully replace a human therapist, instead they were trying to augment that human therapist. So, for example, when our testers, who include these Stanford psychiatrists, were going through and, you know, and interacting with them and showed some sign of psychiatric distress, some of these school ones actually within 15 minutes had flagged a human and had a support plan for humans to come in. That’s just a very different model than the ChatGPT model, which is like, oh, just keep them using ChatGPT and replace the human entirely. So we’ve got to test these end products to see how they’re designed, when it is and isn’t appropriate to use AI to either replace a human or augment what a human can do.
Frank Cilluffo [00:27:42]: That’s a great example. And I often think about human in the loop, and I’m a proponent, certain decisions only someone who swore to the Constitution should be making from a national security standpoint standpoint. I’m not ready to hand that off to anyone who doesn’t have some accountability and at least have some, some experience in, in all of that. But it’s also the human intervention from a psychological or behavioral standpoint. That’s a, that’s a fascinating set of, set of issues that I just hadn’t thought about, but sort of makes sense. I don’t know how you scale that because, I mean, I think everything we’re looking at here is based on scale, speed, and not the individual. But we do need to individualize all of this as well to be able to know where you can have greatest intervention, right, from, from good, bad or indifferent. So…
Geoffrey Fowler [00:28:38]: A related problem is automation bias. So this comes up with doctors, but then also therapists. Right? So even if you have a human ostensibly in the driver’s seat, maybe a chatbot is making recommendations on what responses to give. Maybe AI is, you know, helping organize the notes that a doctor is collecting during, during a patient encounter. The more we know, one of the problems is we know the more humans use these systems, the more they just start to trust them, even if they shouldn’t. And you know, I think of like my favorite example is, you know how in Hawaii people will follow their GPS and end up in the ocean because the GPS told them to go down that road. Like we have the same thing happen with AI tools and systems, even when they have ostensibly people in the loop and making the calls.
Frank Cilluffo [00:29:38]: You know, clearly this is an all of the above answer, but I’d like to get your thoughts. So we’ve got educators, we’ve got parents, we’ve got companies both from in terms of their workforce, and you’ve got the companies driving this technology. Where’s the biggest difference going to come from? And clearly it is in all of the above. But I think we have to start thinking like a single team and a one mission to be able to get our arms around this holistically. And then a secondary question is, where do our legislators fit into this? Where should policy weigh in?
Geoffrey Fowler [00:30:21]: Yeah, ultimately, if we want better products, technology companies have to make those better products. You and I can point to all the problems that we want, but it’s going to take the well-intentioned, smart people inside these companies to align on this being a priority both for society and for children, but also for their business to make trustworthy AI products to make it happen. I am not a believer that we can just say to families and to parents, oh, you have to be responsible for it. We don’t say to families and to kids, oh, you have to install your own, you have to like design and install your own seat belts in cars. No, we expect car makers to put the seat belts in that keep, and them to be effective at keeping us safe. So, so, you know, I don’t think we can push this all onto families and parents, which is frankly what a lot of social media, the response from social media has been for years. Right? Like, oh, we have parental control settings, like it’s fine, you know, and then meanwhile, trying to use any of these parental control or privacy control settings is like trying to fly a 747, right? It’s all these knobs and twiddles and like, I don’t know what’s really happening there. How does this actually work?
Frank Cilluffo [00:31:52]: While building the plane mid flight. Right? So we’re going on top of it all.
Geoffrey Fowler [00:31:56]: Exactly. So, but I, but I think we as, as, as, as advocates for better technology, as advocates for families and children can put pressure on companies to build better. And that’s the sort of the central idea of our Youth AI Safety Institute at Common Sense Media, that if we start setting standards, setting public standards so everybody knows where the bar is, then evaluating against that and sharing those evaluations publicly. And the evaluations are independent. These companies can’t influence them. Like, hopefully they’re going to want to start competing to get better ratings and that’s going to lift the race for everybody.
Frank Cilluffo [00:32:40]: And arguably that’s a flight to quality too. Right? So as a user, it could be a differentiator on why you go to product A instead of product B. Because at least they thought about this and at least, so to me, there is a business case here that is an enabler, not just a disruptor in the negative sense.
Geoffrey Fowler [00:33:03]: 100%. Companies making AI products can get more business when people trust their products more. We’re trying to get to a point with more trust. You also asked about legislation. So the Youth AI Safety Institute is not, does not get involved in policy. Common Sense Media has been involved in advocating for, Common Sense Media being the parent organization of the institute, has been involved in advocating for certain laws regarding companion AI, the age of kids that can use some of this stuff. I would say that we can really look to like, Europe has been far ahead of the US and actually coming with some regulations that require things like age assurance, like actually knowing the age of the person using the product. Because of some of those efforts and pressure in the US we now have, imagine this, settings on an iPhone where a parent can say, here’s the age, and it can send a signal to an app about whether that person, the user, is an appropriate age for using that particular app or whatever version of the app that they’re using.
Geoffrey Fowler [00:34:23]: I cannot believe it took us until 2026 to get that. I mean, again, we’re almost 20 years into the, our phone revolution, and we just now have the way for a parent to tell an app that this is a kid using it. But we’ve got that now because of some of these pressures that come from government. So I think we’re going to be hearing a lot more about those sorts of things too. In Europe, the expectation is that companies are responsible for making their stuff safe and for designing for kids. They’re still figuring out how to, how to, how to make those laws stick. But that is definitely the attitude there.
Frank Cilluffo [00:35:04]: You know, there’s so much uncovered here. And I love the discussion you had around trust because I’m not sure it was actually Mark Twain or Samuel Clemens who said trust is the coin of the realm, but it really is in, in, in, in, in all of this. And I want to close with, you know, I’m often told a pessimist is an optimist with experience. I spend my time thinking about bad things. You’re spending a lot of time thinking about the risks. I am still an optimist. So what gives you hope we can actually get this right and do it even better? Or, or is that a fool’s errand? I, I, I still think that it is a double edged sword and there’s so much good that can be brought to bear and so much bad. What gives you optimism in all of this?
Geoffrey Fowler [00:36:05]: I like to say that the Common Sense Media Youth AI Safety Institute is neither pro AI nor anti AI. It’s pro kid. Right? And so, and so we are here to research not just the hype of what companies say about what their technology does, we’re here to see what it actually does and tell the truth about it. And when you tell the truth, things get better. That is what I believe and hope in. So I am also hopeful that we are working on this kind of work right now, about four years after the, almost four years after the launch of ChatGPT, as opposed to waiting 10, 20 years like we have with some earlier technologies. But still we’ve got to move fast.
Geoffrey Fowler [00:36:58]: And so we are, we’re moving as fast as we can to kind of get the information out there and again, tell the truth about products so that they can get better. And I’m hopeful that this work has interested so many different people around the world and it’s, you know, building support and hopefully we’ll be able to get things moving in the right direction.
Frank Cilluffo [00:37:18]: Geoff, my very last question. What questions didn’t I ask that I should have?
Geoffrey Fowler [00:37:23]: You know, there’s a lot of money sloshing around in the AI world right now and it’s trying to influence policy and lots of things going on in the world. The Youth AI Safety Institute and Common Sense Media receive funding from a wide range of sources, from philanthropists, big charities you’ve heard of before, but also from organizations including, like the OpenAI foundation, which is the nonprofit part of OpenAI and Anthropic. But the work of the Institute is fully independent. We are solely responsible for the content of our ratings and reviews, and all of our funders understand that, and we’ve proven that through the stuff that we’ve already published and the stuff we’re going to keep going on with. But I just wanted to be clear with folks that we are truly, truly independent.
Frank Cilluffo [00:38:23]: Geoffrey, thank you for spending so much time with us today. Thank you for fighting for that next generation, because we focus a lot of our effort here on national security, critical infrastructure, ultimately, it’s those children who are going to be driving, and it’s a social infrastructure. And we have an opportunity to get out in front of this and build more resilience. And thank you for doing what you’re doing. I know it’s early stages, but it’s really important work and really appreciate it. So thank you.
Geoffrey Fowler [00:38:57]: Thank you for your support.
Frank Cilluffo [00:38:59]: Thank you for joining us for this episode of Cyber Focus. If you liked what you heard, please consider subscribing. Your ratings and reviews help us reach more listeners. Drop us a line if you have any ideas in terms of topics, themes or individuals you’d like for us to host. Until next time, stay safe, stay informed, and stay curious.
