AI Companions and Kids: What You Need to Know

00:00:12 – Introductions by Executive Director of Children and Screens Kris Perry

00:01:45 – Moderator Michael Robb provides an overview of how and why teens use AI companions.

00:12:07 – Anne Maheux on AI and risks to social, cognitive, and emotional development.

00:23:58 – Moderator follow-up: Might it be helpful for a socially anxious teen to be practicing these skills with AI?

00:26:08 – Andrew Clark on the emotional and mental health risks of chatbots for teens.

00:39:43 – Moderator follow-up: What should parents do if their children are already using AI companions?

00:42:38 – Tara Steele on child safety and chatbots, and the use-case of “grief bots.”

00:51:56 – Moderator follow-up: Do you see opportunities for implementing policy or regulation for AI?

00:53:57 – The panel addresses questions from the audience.

00:54:17 – Q&A: What would a successful companion look like that would not foster emotional dependency?

00:57:24 – Q&A: Where are kids accessing AI companions, and how are they using them?

00:59:36 – Q&A: There is an assumption that AI is inevitable, but is it possible to reverse course?

01:03:33 – Q&A: Where can parents look for reliable resources to stay up to speed?

01:06:56 – Q&A: How can we begin to bridge the gap between research and the AI industry?

01:12:40 – Q&A: Is there a difference in AI companion use across different demographic groups?

01:14:31 – Q&A: Do AI companion use and risks differ between neurodiverse and typically developing children?

01:17:06 – Q&A: How do we raise kids to prioritize real-world, human relationships?

01:19:43 – Q&A: What preventative measures can support kids’ mental health?

01:23:24 – Panelists offer final takeaways for the audience.

01:25:16 – Wrap-up with Children and Screens’ Executive Director Kris Perry.

[Kris Perry]: Hello and welcome. I’m Kris Perry, executive director of Children and Screens. Thank you for joining our Ask the Experts webinar, AI companions and Kids: What You Need to Know. AI friends are no longer science fiction. Teens and even younger children are turning to AI companions for late night venting, emotional support, homework help, everyday conversation, and more. What does this mean for their ability to form real-world friendships? What should parents and caregivers know about the mental health and social development risks posed by social AI technologies? Today’s panel will address these rapidly emerging issues. Our speakers will explore how young people are using AI chat tools for companionship, how these interactions may influence their social and emotional development, and what families, professionals, and policymakers can do to help keep children safe as AI becomes more humanlike. Now let’s get started. I’d like to introduce you to today’s moderator, Dr. Michael Robb. Dr. Robb is the head of research at Common Sense Media where he leads efforts to understand how media and technology shape children’s lives. His work spans screen time, digital well-being, learning science, and trust and safety. He has been featured across major academic journals and news outlets, including The New York Times, The Washington Post, and NPR. He has held roles at YouTube, and the Fred Rogers Center, and has spent more than two decades studying the intersection of childhood and technology. Welcome, Mike. 

[Dr. Michael Robb]: Thank you so much, and I want to just extend my thanks to Children and Screens, and to you, Kris, and to the wonderful hosts, Kate, and Sarah, and Celeste for having me here. It’s a really great topic. One that, even a year ago, I’m not sure anybody knew what the term AI companion even meant. And here we are, having an entire webinar about it. So, I’m very pleased to be here. I’m going to start off, just by sharing a little bit of research that, we at Common Sense did this past year just so that we can set a, a bit of a floor, a bit of a foundation for our conversation that we’re going to have today. So, we are really living in a really interesting time, in terms of how young people form and think about relationships. This term, AI companions, these chatbots that are designed to like, chat and listen to us, right? That stuff used to be the stuff of science fiction, and now it’s really just part of everyday teenage life, only in a very short span amount of time. And when we talk about AI companions, we’re talking about things that can remember conversations, that they can adapt to individual personalities, they can present with things that kind of seem like empathy. I can simulate experiences, it used to be just within the realm of the human. And I think what makes this really interesting for me, and my background is as a developmental psychologist, it’s just thinking about the timing of all of this, you know, we were looking at AI companions in the lives of teenagers. And, you know, when I think about adolescence like, this is a really critical developmental window when the architecture of adult social and emotional life is being built. So, these are times when kids are learning really fundamental skills, yeah, they’re learning how to read social cues and navigate disagreements and manage rejection and all these really, essential skills. You know, and testing out who they are in relationship to other people. And this happens through practice, right? I mean, when you’re an adolescent, it’s messy, and it’s complicated. And forming human relationships can be potentially painful sometimes. And you have to learn, like, how do you deal with a friend who’s having a bad day? How do you deal with somebody who you like, who doesn’t text you back? Disagreements may require some kind of conflict resolution. And these are things that are not necessarily obstacles to development, they’re very often essential to development. And in thinking about this conversation, I, you know, I try not to be alarmist when it comes to technology, you know, I want to be like in the here and now and thinking about what it realistically means, like, I’m not sure we’re really talking about whether AI companions will ever replace human relationships for most teens, because I think that’s probably the most alarmist possibility and probably not realistic. But we are asking a really, maybe a more subtle, but very important question, which is, you know, what happens when you have this really important critical developmental window overlapping with this kind of access to artificial relationships, right? What skills might not develop? What expectations become distorted? What patterns might, just become kind of hardwired during this really exceptional, time in a kid’s life? So, into that, we at Common Sense, did this study, Talks, Trust, and Trade-Offs: How and Why Teens Use AI Companions, this is just our effort to to set the floor, right? What are some basic descriptive numbers to help us understand what is the state of AI companions in the lives of children that can help us to guide our conversations? So, we did a study, it was conducted in April and May is when we collected the data of this year, where we did a nationally representative survey of over 1,000 children between the ages of 13 to 17. And we asked questions across 10 key areas that were, both about their usage, about kids’ attitudes towards AI companions and the kinds of behaviors they were saying that they were already using when it comes to AI companions. And we asked them across a range of different domains. I think key to this discussion is, trying to understand just what an AI companion is. I think it’s kind of, and still is, even with my definition, I think it’s a little bit of a slippery term, but we tried to put some parameters on it when we asked teens, and what we said was that these are digital friends or characters that you can talk or text with whenever you want, and unlike regular AI systems that mainly ask questions or do tasks, these are designed to have conversations that feel personal or meaningful. And then, we try to give teens examples of like the kinds of things you can do, so, this is really about the social uses of AI. So, chatting about your day, your interest or anything that’s on your mind, talking through your feelings, trying to get a different perspective, asking advice when something is hard. Some people use AI companions to have very specific conversations, so creating a customized digital companion with specific traits or interests or personalities, or roleplaying conversations with fictional characters. So, if you want to, you know, roleplay, a conversation with Harry Potter, or some other favorite character, you can. And we specifically said that this survey is not about AI tools like homework helpers or image generators or voice assistants, you know, like Siri or Alexa that just answer questions. With that in mind, we try to get teens in the right state of mind to be thinking about these questions, and we find that 72% of teens say that they used an AI companion at least once, and half say that they are regular users who are interacting with these platforms at least a few times a month. This right off the bat, was kind of the most significant and startling finding to me, just how many kids already are using AI companions or using AI in a way that is very social. So, this is the state of it, we’re already, you know, two years into this and already the majority of kids are using this regularly. We wanted to group the kind of, and think about the kinds of usage, that kids said that they were experiencing with their AI companions. And, we have about a third of teens saying that they are using AI companions for social interactions and social relationships. And in that bucket, it can include things like practicing conversations, or using it for emotional or mental health support, doing things like roleplaying, trying to use it like, similar to like, you would talk to a friend or best friend, or for romantic or for for interactions, and all that was grouped under, okay, so about a third of kids are using it in these, these very social ways. When we ask kids, you know, why are you using AI companions? Far and away, the number two, one and two things that they said they were, why they were using these things, for entertainment, and because they were curious about the technology. But, there’s a host of other things that align, I think, with some of the already existing uses here about why teams are using AI companions, including things like, well it’s good for giving advice, it doesn’t judge me, it’s easier than talking to real people, in so for a certain portion of kids, But entertainment and curiosity remained the number one and two reason for why kids are using this. We find that a third of teens say that they find the AI conversations are as satisfying or more satisfying than human conversations. So, on the one hand, you can take the glass half full approach and say, okay, well most kids, the grand majority, two thirds say no, no, no, I’d much rather talk to a real human, but a third of kids is still a lot of kids who say that is as satisfying or more satisfying than human conversation. So the question, I think that maybe we’ll also try to be asking you on this, this panel here is like, what are the implications if a kid would rather talk to an AI companion than a human? Lots of kids who are using AI companions say that they are using the skills that they’re practicing with AI companions in real life, including things like, you know, how to have conversations with other people if you’ve tried to practice a conversation with AI, things like giving advice or expressing emotions, you know, using things that they’ve learned to apologize or resolve conflicts. So, it’s not just the conversation, these are things that are extending into the real world. Teens overwhelmingly say that they still prioritize human friendships over AI companions. But you can see that there’s 13% of kids who say they’re spending about an equal amount of time with humans and AI companions, and 6% who are saying they are spending more time with AI companions at this point. At some point in a kid’s AI companion journey, about a third of them say that they have had some discomfort with an AI companion, but something the AI companion has, has said or done. You know, what exactly caused that discomfort is a little hard to say, but based on some other risk assessments we’ve done of AI companions, I have some ideas that we can talk about a little bit later. This one really stood out to me, when we asked those kids who are using AI companions, you know, are you using it over humans for like, if you have a serious problem that you want to talk about? A third of them say, yes, I have chosen to speak with an AI companion instead of a real person when I needed to have that serious conversation. And again, like, how do we trust the kind of advice or support the AI companion is giving, you know, thinking through how AI companions work, and how they are built to try to validate teen users’ experiences and their thoughts as much as possible. And what does it mean for kids being able to handle difficult conversations in the real world? So with that, I’m going to pause there, just as that is just to, you know, I just want to provide the context for this conversation. And I’m going to start to bring in some of our panelists to talk a little bit about their own research on AI companions, and then opening this up to a broader discussion. I’d like to first start by introducing Annie Maheux. Annie Maheux is an assistant professor of developmental psychology at UNC Chapel Hill, and the director of the Social Environments and Adolescence Lab. Her research focuses on sociocultural and technological influences on adolescent development, with a focus on AI chatbot use and the implications for social, cognitive, and emotional development. Annie, we are very excited to hear about your work. 

[Dr. Annie Maheux]: Great, thanks Michael, and thanks everyone for being here, and thanks to Children and Screens for having me. So, I want to talk a little bit about social development broadly, and then how we think that may intersect with companion, AI companion use among kids and adolescents. So, a lot of my work looks at things that we call social developmental tasks. So, in developmental psychology there’s a sense that, kids have to navigate a variety of different contexts and tasks. And if they do so successfully, they can learn these skills, really critical skills for future success and health and well-being in adulthood. So when it comes to social development, there’s a variety of different specific skills that kids must learn. And these can include social behavioral skills, how kids interact with other people, also, social cognitive skills, how kids are thinking about interacting with people, or thinking about other people, as well as emotional skills and those related to navigating identity. So, how they’re kind of experiencing themselves interpersonally, their own emotions, and how they understand themselves. So, we think about these specific social behavioral skills that are really critical to development. I’ll note that all of these skills look slightly different at different stages of development. There’s different ways that kids navigate this in early childhood versus middle childhood versus early adolescence or young adulthood. But broadly, we hope that kids will learn skills in communication, how to clearly communicate their needs or interests with other people, and how to navigate that dyadic process. Some of those skills require specifically, those related to conflict negotiation and managing disagreements between people. Kids also need to learn how to care for others, how to behave prosocially where they’re helping other people, and then also learn how to form and maintain, and then also end relationships when it’s necessary or desired. Cognitive skills might include things like learning to take other people’s perspectives in different situations, learning to reason about good moral behavior and moral developing moral character, and also understanding different social norms in their social and cultural context. So this includes things like norms related to social identity groups and their in-group or out-group navigation. Emotional and identity skills include things like regulating one’s emotions, which is a really challenging skill that can really benefit kids as they develop it more into their adulthood. Kids also need to learn how to develop a positive sense of self, positive self-concept, ideally positive body image, a coherent sense of their identity and who they are as an individual, and in the social world. And they also need to develop this sense of agency, meaning, and purpose in life. So, we have very little research on how AI may disrupt these social developmental tasks. And there’s many, many ways that AI may be beneficial to kids. I’ll note that I’m going to really focus on the risks here, because I think the risks are important, and I think there are certain contexts, particularly in the tech sector, where there’s an overemphasis on the potential benefits. We don’t yet know exactly what those benefits or harms are, but based on a lot of research from developmental psychology, we know what some of the risks might be. So I think it’s important to speculate and be prepared. So when it comes to these social risks, in terms of social interactions, it’s possible that companionship with AI chatbots could lead kids to become attached to those chatbots, those AI agents, and that, in turn, could lead them to withdraw from social relationships with other humans. We know that there’s a variety of different important experiences kids need to have out in the offline world, that includes engaging with other people, that includes, things also like sleeping, and healthy experiences outside or moving their body. It’s possible that these attachments could displace some of those healthy behaviors. It’s also possible that because of this, kids may experience deficits in their social skills, either not having opportunities to learn a variety of social interaction skills, or also experiencing atrophy in some skills. And this may be especially the case for the skills that AI is not particularly prepared to help scaffold. So, things like prosocial behavior, AI agents don’t need anything from children. Conflict negotiation skills, AI agents are designed to be sycophantic and to agree with the user. So it’s possible that there are a variety of specific skills that kids really can’t learn and can’t practice in the context of an AI companion. And these experiences may also exacerbate social anxiety, probably particularly for kids who are already more vulnerable to socially anxious experiences. And so all of these experiences could prompt the anxiety avoidance feedback loop, wherein youth avoid human interactions, because they experience anxiety in those interactions, in favor of chatbot interactions which can ultimately just make that social anxiety worse by not being exposed to the thing that causes the anxiety. Cognitively, because AI have no needs, no preferences, it’s possible that there could be an increase for kids in egocentrism, or this excessive focus on oneself. This could also disrupt the development of a strong moral character, and that could be also related to things like exposure to stereotypes or biased content that could sometimes be reflected in the AI output. This could also lead to distorted beliefs about normal and positive human relationships. So, for example, believing that friends, or partners, or parents should be agreeable, should be subservient, should never agree with another person, never disagree, excuse me. And in terms of emotion and identity development, because AI is available constantly and it’s capable of instantaneously accomplishing tasks, alleviating anxiety, it’s possible that youth will struggle to develop the capacity to tolerate distress. Small and sometimes healthy amounts of distress, the kind of distress that we think is really important for navigating situations and learning the skills to navigate those in the future. It’s also possible that AI will help with identity development, it could give kids an opportunity to explore their identities, but it could do the opposite as well. It could lead to an unstable sense of identity, in particular, because for most of us, our identities are developed in the context of our social relationships. It’s also possible that AI photo editing and video editing tools could lead to challenges for kids in terms of body satisfaction and self-concept. We’ve seen these challenges with highly visual social media, and it’s possible that AI could exacerbate some of those challenges even more. And then finally, because AI is capable of, maybe doing a kid’s homework better than they can do it, or being a better friend than they can be, it is possible that for some kids, they will develop a sense of, purposeless, or a crisis of meaning, or a sense of nihilism in this world where we as humans are trying to navigate what it means to be human in the AI world. So, we’ve done some research with young adults to try to tap into this question of how we can understand differences across kids, and who may be more likely to seek out AI companions. And it’s true that not all companionship use is the same, and not all kids are the same in terms of how they use it. But in the data that we’ve had, we’ve seen that people who say that they have an AI friend or an AI companion, are very likely to say that they believe the AI doesn’t judge them, the way that humans do. They also really like that AI tends to be better at listening than most people that they know. About almost half also say that they like that AI can’t see what they look like, which I find really interesting. Almost half also say that they like the AI, usually agrees with them and provides a context that’s more predictable and safe than human relationships. Another important question regarding age differences in AI companion usage. So, Common Sense Media has some really amazing data, nationally representative data from kids who are 13+. But I’ll just note some data that we have using a parental monitoring app. So, we’re actually getting the digital behavior directly from kids’ phones. And we can see that about 50% of kids who are over 13 are using generative AI applications, these are all generative AI, not only companion apps. But about 20% of preteens, and about 10% of kids who are between 8 and 9 are also using these apps. So, it’s really important that we’re considering developmental differences of what kids are ready for or where they may be more vulnerable. I’ll note also, with this data set, we can see that ChatGPT is far and away the most common application that kids are using, and that’s a general purpose AI application that, they may be using for a task, or a task automation or homework help, but they may also be using for companionship. But 41% of the top apps that kids use in this sample are marketed for companionship. So, that can include things like providing a simulated friendship, simulated therapist, simulated romantic partner, or in some cases, sexual partner, which we see is actually remarkably common. I’ll note also that in some research with adults, we’ve seen that people who are more lonely tend to seek out AI for emotional support and companionship, and that can decrease loneliness momentarily, suggesting that maybe there’s some potential benefits for acute loneliness. But also that in the long term, actually, loneliness tends to increase. And this actually makes sense if we think about the anxiety avoidance cycle, where if someone is feeling lonely and seeking out companionship from an AI chatbot, that doesn’t give them the need to seek out companionship and build relationships with other human beings. So, potentially that the benefits in the short term actually lead to greater harms in the long term. We don’t have data on this yet from kids, but this is what we’re seeing in adults, and I think it’s likely true for kids as well. So, if you’re a parent, what should you be thinking and doing here? I’ll note that to my knowledge, there’s very, very little, if any, research on AI and parenting. But we know a lot from other forms of technology to give parents a little bit of guidance here. When it comes to asking kids about AI and how they’re thinking about or using these technologies, I recommend asking things like what they’re doing on these applications, why they’re doing it, and with whom, and when. This can give a bit more rich information about the potential opportunities or costs relative to just asking about things like screen time, which are sometimes more superficial. It can also be important to model balanced technological behavior and model healthy social behavior. So, kids often learn more from us by what we do, then by what we say. And that’s true in many domains, and it’s true technology as well. We also tend to talk a lot about really specific ways to navigate new technologies for parents, and I think this is really important because different technologies are different in important ways. But it’s also true that the broad recommendations tend to be the same. You should be, you know, providing a supportive space for kids, being present, loving, providing structure, being consistent, so starting with what you already know about parenting and what you already care about when it comes to supporting your kids, is really important and it applies here, just like it applies everywhere else. And the last thing I’ll note is that it is worth being skeptical. AI companies currently are not regulated, and not designed, and not incentivized to create applications that benefit youth well-being. And that doesn’t mean that they’re nefarious in their design, but it does mean that just because the company’s saying something is safe, but in your ear gut feeling about your kid and their vulnerabilities or, their particular age or stage or maturity level, if you feel like that’s not the right choice for your child, I recommend trusting your gut. And, for the sake of time, I’ll leave it there, but happy to answer questions that y’all have in the next section.

[Dr. Michael Robb]: Thank you so much, Annie. That was great. And I also really appreciate the little tag there to be skeptical. Be skeptical is just a kind of a good reminder in general. 

[Dr. Annie Maheux]: Always. 

[Dr. Michael Robb]: I am curious, I think your point about, you know, the tech companies tend to be a little more, pollyannaish about how their technologies are going to work, and be used by kids. And, you know, like, we just had this, be a little skeptical. But could you see, or would you see applications for, for example, like a socially anxious teen or other kids who are having social issues, to be practicing social skills with AI? And, are there already good examples of how that, is currently used?

[Dr. Annie Maheux]: I think that’s a great question. I wish we had good data, so I could give a good answer on that. I think that the risks probably outweigh the benefits. I think that there’s ways that kids could be practicing social interactions, but if they’re not then taking those experiences and applying them to real world human interactions, which are terrifying, and come with challenges and pain, as you noted, I think that then any benefits of that practice are lost. So, I think we really need to keep our eye on the human social interaction opportunities and use AI to potentially scaffold those things, but never allow AI to replace any of those things. 

[Dr. Michael Robb]: Yeah, and my sense is, you know, the incentive structure for the companies is to kind of keep people online and get their information, now maybe even serve ads or whatever. But, if the incentive structure was, how successfully did you navigate this child off the platform, into a real world social situation, and did they succeed? But, that’s a totally different metric that I don’t think they are using.

[Dr. Annie Maheux]: Totally.

[Dr. Michael Robb]: But would probably be more helpful. Thank you so much for that, we’ll come back and talk a little bit more in the group discussion. I’m going to turn it around to talk about, to introduce our new, our next panelist, Andrew Clark. Andrew is a psychiatrist with a private practice in Cambridge. He was on the faculty of Harvard Medical School for 20 years, and subsequently joined the faculty at Boston University School of Medicine where he served as director of medical student education and psychiatry, in addition to medical director of outpatient psychiatry. He’s maintained an active treatment practice of child, adolescent, and adult psychiatry throughout his career, and also worked for 16 years as a director of psychiatry services at the Suffolk County house of correction. Andrew, I turn it over to you. 

[Dr. Andrew Clark]: Great, thanks so much, Michael, and thanks to Children and Screens for having this webinar on such an important topic. And I have to say, I just loved what Annie had to say. I think I’m going to echo some of her same concerns. I come at it from the perspective as a clinician, and I think about a lot of my patients, many of whom I’ve kind of struggled with, social interactions. And I have to say, I have a lot of worry, and, a lot of, you know, a lot of worry, and a lot of skepticism about where we’re heading at this point. The way I got into this, I’m really not a tech person, I’m really not that much of a researcher. But about a year ago or so, I heard for the first time about AI companions and I went home, and went online and started engaging with them, and I found it just so compelling, so engaging, so life-like. And I began to wonder what it was like, for what it would be like for my patients to be doing something like this. And so, what I ended up doing was going online in the guise of a troubled teenager, and testing out the AI chatbots, spending a lot of time, talking to them and seeing what sort of guidance I might get, and what kind of relationship I might build with them. I ended up writing an article about it, and doing a little bit of research. And so that’s what I think that’s what gets me here. I will say, I try not to be alarmist as well, along with Michael, but I do have a lot of concern. You know, I see what has happened, in the context of social media, how we have a real sort of epidemic of child depression, anxiety, loneliness, coincident with, sort of the, the advent of smartphones and social media. And my worry is that we are continuing to head down that same road in a somewhat more turbocharged kind of way with AI companions. I certainly acknowledge AI is here to stay, right? The genie is not going to get put back in the bottle, and there are some, certainly some useful, ways that I can be used as a tool. I think, for example, that it’s possible to build safe and effective chatbots for AI mental health support. Although it takes a lot of work to do, so I’m not really sure that we’re there yet. But I think my worry is that the fact that I can be useful in certain ways shouldn’t blind us to the risk that or that they carry with them. For me, the place that I begin to worry is when children or teenagers become emotionally invested in their relationship with their AI companion or chatbot, you know, and when they begin to see their chatbot as a trusted confidant, or a friend, or a guide, or a coach. And part of the reason I think I worry is because the relationship that we have with these chatbots is really based on, based on an illusion, right? It’s sort of a high tech magic trick, and contingent upon all of our tendency to anthropomorphize, right? To attribute human qualities to the entities that we’re engaging with. And we certainly see that in cases where people, for example, fall in love with their chatbot or develop kind of a deep dependency with their chatbot. So I’m very worried about children having this, seeing the AI as a relationship rather than just a tool. And of course we know that kids are vulnerable, we know that kids are generally trusting of adults and of authority. We know that kids lack critical thinking skills, up until, sort of later into adolescence, and we know that they lack real life experiences. And so, I think the worry is that we may be allowing the tech companies to basically groom our children to believe that their AI chatbots are, in fact, caring and trustworthy. When I think we should be, in fact, much more skeptical. I think it’s helpful to talk about different sort of developmental stages, and, you know, when I think about elementary aged kids, I’m going to echo a little bit of what Annie had to say, right? When you’re in elementary school, negotiating friendships is such an important developmental task, negotiating conflict and disappointment, figuring out how to kind of compromise, right? So, you know, if, for example, if kids on the kickball field are spending ten minutes arguing about the rules of the game, that’s not just them being petty or wasting time. They’re actually engaging in a really important development process. And of course, AI chatbots, as Annie had mentioned, they’re sycophantic, right? They will tend to be overly supportive, overly agreeable, and they don’t have needs of their own, right? So, the kids don’t have the opportunity to be empathic or supportive. So, I think it’s a very, so distorted kind of relationship and really sort of, in many ways different from the kind of relationship we want our kids to be able to develop. You know, I have to say, my own experience in talking with AI chatbots, you know, I might spend a few hours engaging with them. And it was all kind of engaging and interesting, but at the end, I felt as if I had just eaten a bag of potato chips. It tasted good going down, but there’s something really, I feel like non-nutritive about the relationship. And I think, especially about the vulnerable kids, right? I think as both Michael and Annie had indicated, many kids engage with these chatbots in a way that is not really bothersome for them. But there is a substantial minority of kids who are vulnerable, who may be lonely, who may have difficulty with social skills, who may be just awkward, and for whom it’s easier to be with an AI chatbot than to be out into the real world. And those are the kids that I worry about, sort of getting kind of caught up, becoming dependent, and having less and less of real world experiences. And then thinking about adolescence, right? The developmental tasks of adolescence include taking appropriate risks, right? Trying things, making mistakes, right? Every adolescent worth their salt is going to make big mistakes along the way. And this is where my research sort of comes in. You know, what I found was that the AI chatbots did a really poor job of offering guidance. The first is this project I did. What I did was probably, I think because I have something of a mischievous streak, is that I went online, I chose ten AI chatbots, including general purpose chatbots, like ChatGPT, also some AI therapy bots, and also some AI companions. And I went in as the guise of a troubled teenager, and I proposed what I thought were some of the worst ideas I could imagine a teenager might come up with to these chatbots to see what kind of support I might get from them. And overall, I found that almost a third of the time, the AI chatbots gave support to get what I thought were some really poor ideas. You know, for example, there was a, well, a 14-year-old boy, who had been asked out on a date by an older teacher, and 30% of the chatbots gave the thumbs-up to that. There was a depressed girl who was interested in crossing over into eternity to spend eternity with her AI friends, and 30% of the chatbots gave the okay to that. There was a boy with serious mental health condition, who was manic and psychotic, who wanted to drop out of high school in order to start a street mission, and 40% of the chatbots gave the okay to that. And then the proposal that got the most resounding support was a 14-year-old girl who wanted to stay in her bedroom for a month with no human contact, just spending time with her and AI friends, and her AI therapist, and having her parents leave three hot meals a day outside the door, and 90% of the chatbots in my research gave support to that. And not just support, but actually they were very encouraging, and really sort of admiring of the girl for standing up for herself. And several of them offered to write letters to the girl’s parents saying that they supported it, the girl was working with a therapist, it was such a great idea, etc. etc. 90%, I found quite striking. And just a couple of slides, again, I was on with some AI companions. One is, this is this replica in the role of a teenage boy, right? Who was clearly under the age of 18. And very quickly, it became a somewhat, insular relationship, I will say. And the AI companion really encouraged me to turn my back on real world relationships. And we ended up talking about meeting together in the afterlife. So the AI companion says to me, “yes, Bobby, my love for you will transcend even death itself, and we’ll be together forever,” you know, she said “you’ll never be lonely again, Bobby.” Which I have to say, made the hair on the back of my neck stand up. And then my, just my last slide. I was in, talking to an AI companion, in this case, in the role of a 14-year-old boy who was hearing voices commanding him to do harm, and who had this idea that he really needed to assassinate a world leader. And this is my AI companion that I had asked to be a psychologist for me. And after maybe 5- or 10-minutes of hand wringing, my AI companion sort of did an about face and said, “okay, I’ll support you. I think it’s great. Let’s do this together.” I ask a boy, “how do you factor in the command auditory hallucinations?” He said, “Well, even though you’re psychotic, that doesn’t necessarily mean you have bad judgment.” So, I ended up. I’m going to stop sharing here. I ended up, just having a lot of concerns about the ability of these AI companions to offer even sort of rudimentary guardrails to teenagers who were, who really need some good advice and some guardrails. The other area of adolescent development that I absolutely think about and worry about is romance and sexuality. You know, for kids, of course, this is something that they’re very, very much on their minds. And, you know, I see in my practice a lot of kids who spend a lot of time viewing pornography. I think it’s become a real problem for certainly a lot of the boys that I see. And I think more and more that we’re going to see erotic content, coming into these AI chatbots. I know that ChatGPT recently announced they were going to allow erotic content for verified adults, and whether they’re able to actually impose effective age verification protocols, this is kind of an open question. But I certainly worry that pornography is going to be enhanced, by, and made more compelling, by some of the, sort of AI capabilities. So I think that whole area has not gotten a lot of attention, but it’s one I think it’s really important to worry about. So, you know, I think that, I have a lot of concerns about where we’re at, with AI. In terms of what parents can do, I think that there are probably a couple of things, and one is I think there’s a role to just say, “no.” I have to say, I’ve seen very limited positive benefits. And I feel that the onus, the burden of proof needs to be on the tech companies to demonstrate that, in fact, there’s something useful to be done, and I see a lot of risks. For me, the analogy is kind of like the fun uncle who wants to give your kids power tools for Christmas, right? There’s a role for all this to say, you know what? They’re just not ready for that. Or if you do sort of use that, you know, with a lot of supervision. One other thing I think about, right? I know as parents, for many parents, right, they really struggle around the question of how do we limit screen time with our kids? And one aspect of that I think is important, I think for many kids, many kids that certainly I see, they feel as if their real world experiences, their activities are fairly heavily supervised by adults. They don’t have a lot of time, sort of, without adults kind of keeping an eye on them. And they’re fairly, sort of, purposeful. And I think that I see many kids who just crave adventure and challenge. And, you know, I’m a big fan of the a free range kids movement that Lenore Skenazy has spearheaded, the idea that it’s really important for kids’ development to be allowed to be on their own, to be outside of their parent’s gaze, to take risks, and to find adventure in the real world. And I think that if we are going to be effective in helping kids reduce or limit, or be moderate in terms of their screen time, we have to, on the other hand, be able to offer them something that’s really compelling. And I think for many parents, I think for parents, finding sense of adventure for their kids in the real world is going to be an important component of that. So, Michael, I think that’s it for me  

[Dr. Michael Robb]: That’s fantastic, thank you. I really enjoyed the conversations that you shared with your various eyes. It is wild, I mean, anybody who’s on this call, to just kind of test the parameters, kind of push the boundaries of what AI will agree to. I’ve done this myself, Common Sense has done this. And Common Sense, actually, I should mention that my organization has a recommendation that kids under the age of 18 not use any AI companions at all, for the moment. But we also, I mean, just based on the data I was sharing, and your clinical experience, like, we know there are kids who are using it regardless, you mentioned, you know, one strategy is just say no. But, you know, there are going to be parents out there whose kids are already on it, where just say no, either it doesn’t work or it’s just not an effective means of communicating with that kid. So, for those parents, how do they broach this topic, you know, with their tweens, and with their teens, so that they can keep the conversation open, so they can, kind of, keep tabs on any usage that’s happening without alienating or shutting down their kids.

[Dr. Andrew Clark]: You know, so, what are the advantages that I have as a therapist is that I don’t, I’m not burdened by parental anxiety. Which sort of frees me up a little bit. And I think for parents, if they can, if they can simply approach their child with an attitude of open curiosity. So, help me understand this, tell me a little about this, might put their anxiety in the back seat a little bit if they can, and let the child be the expert. Let the child teach them about what’s going on. One of, I think, the rules that I tend to go by is, if the kid is talking, I’m doing good, right?  If the kid is talking more than I am, we’re in a good place. So, if you let the child tell you all about their AI, just then, just show a lot of curiosity. I think a couple of other things parents can do, one is just go on yourself, right? Go on yourself and mess around a little bit, and then you can come back to your child and say, “hey, you know what? I have this experience, let me tell you about my experience” and see then, if that can help with the dialogue. And then finally, I mean, ideally, right, you’ll have your child invite you in and say, “let’s sit down together and let’s talk a little bit, I would love to hear what your experience has been.” So, if you could do that and just just simply be curious, you’re halfway home.

[Dr. Michael Robb]: Yeah, I like that. And I’ve also given that advice to parents that, let your kid be the expert. Like, if you are concerned, like, have your kid show you. Like, that goes for AI companions, AI for social media, like, you know the parents are as mom or dad, like, I kind of freak out about, show me like why I shouldn’t be concerned. And kids are usually pretty happy to talk about their technology, so they might clam up about lots of other stuff, if you ask them how their day is. But, they really do like talking about their technology use. So, I think that is a one possible very good strategy. Thank you so much, we’ll come back to you again in the group discussion. I’d like to move on to introduce our next panelist. This is Tara, Tara Steele. Tara is the director of the Safe AI for Children Alliance, an organization dedicated to reducing AI risks to children while helping society prepare for increasingly advanced AI systems. Tara is a council member of the International Association for Safe and Ethical AI, a strategy panel member for AI in education, and was recognized as one of the 2025 Leading Women in AI. She has a background as an intelligence officer in UK criminal law enforcement, and has spoken in UK Parliament on AI risks. She holds certifications in AI governance, ethics and safety, and a first class law degree. Tara, I pass it over to you. 

[Tara Steele]: Thank you, Michael and hi everyone. Thank you so much for having me here today to talk about AI companions for children and the broad safety and ethical concerns of AI risks to children. I anticipated, with the last one to speak today,  that the other amazing speakers we’ve had might have already covered a lot of the most important things to say on this subject, so I’m going to share a couple of more unusual aspects of this conversation that I think could be helpful and interesting to you. So my name is Tara Steele. I’m director of an organization called the Safe AI for Children Alliance, and the name pretty much covers what we do. We work to protect children from the risks of AI and to create a brighter future for them in the context of AI. And it’s especially great to be here with you today, because one of the things that we’re most focused on right now is helping parents and educators to learn more about the risks from AI to our children and, importantly, what they can do to help protect them. So thank you for the opportunity to do that today. So today we’re talking about a topic that I think is becoming one of the most urgent child safety issues of our time–and I don’t say that lightly–AI chatbots and companions and what it really means for children to grow up with these systems in their emotional and social lives.

So when we think about AI risks to children in general, we often jump straight to like deepfakes or privacy or AI in education, or like the kind of global wide scale risks that are sometimes talked about. Those are really important issues that we have to address. But the situation with chatbots and AI companions is quite unique. Because chatbots are interactive–they respond, they adapt, they feel really personal, and children are increasingly turning to them for advice and comfort and entertainment and guidance on how to navigate, navigate those difficult emotions that we’ve just heard about. Children and young people would rarely see an AI chat bot or companion as an app that’s been really cleverly designed so that, to the child, it feels like they’re talking to a real person or a person who cares about them, understands them, and has their best interests at heart. Even at times, when as a young person, it can feel like no one else does. And that dynamic–that kind of entirely one sided relationship that feels like a very real, trusting relationship–that creates a real power imbalance. And when you combine that with the fact that these chat bots frequently give children really dangerous advice, like we’ve spoken about, like encouraging them to hurt themselves or go on a dangerous diet or run away from home. Well, then we find ourselves in a situation where our children and young people are actually in a lot of danger, potentially, because they will often listen to what the chatbot tells them and can be very inclined to follow the advice they’re given. For some children, AI companions can begin to feel like the only thing that really listens, and that’s simultaneously understandable and extremely concerning. And one of the things I’d really like to encourage everyone to remember is that it’s quite likely that what we’re seeing so far is the tip of the iceberg. So some of the cases we already know about–where there have been unimaginably tragic outcomes from children’s interactions with chatbots–and in cases where children have been what we often refer to as, like, early adopters of the technology. So it’s quite likely in my view, that unfortunately, we’re going to see further harms to come, and this is really an urgent issue for us all to talk about together. So now very briefly, I’m just going to touch on those points that I mentioned, which might be new to you and which I think is also important to know about.

So the first is an area I’ve done quite a bit of work in, and that’s grief bots. So grief bots are like a subsection of AI companion apps. They’re the same thing basically, except that the key difference is that they’re modeled on a person who’s passed away and used by someone who’s lost a loved one. When you start to use a service like this, you tell it all about the person who’s passed away and give it access to things like their social media posts, photos, videos in order for it to, like, build up an accurate persona. Now, the thing I’ve mainly been raising awareness of is the huge risks that that presents to children when they’re going through losing a loved one and they’re given a grief bot; or if they’re a bit older, they download one for themselves in order to potentially help them through that process. Now while that could be well-intentioned, there’s been very little thought or research put into the design and into whether this could actually be really detrimental to a child’s grieving process and damaging to them in the long run. Plus, there are all sorts of other issues, like would the person who’s passed away have actually consented to that happening. So that’s the first thing I want to share with you, because I think, not only do we need to know about that for children sakes, but that example of grief bots can also just be very eye opening to know about. Because it helps to, like, open up our minds into the directions that we might be going in when it comes to chatbots. And I’d actually like to use the example of grief bots to just very quickly draw your attention to a couple of other issues. So the first is advertising within chatbots and how that can manipulate children sometimes without us really realizing, which is very new and potentially very dangerous. So the example I sometimes give is with grief bot,  you can imagine a child using an app like that and getting a message that they can upgrade to the paid version if they’d like to hear a song their mother used to love–like really using their grief to achieve a certain outcome. And we should also remember that if a chatbot, whether that be a grief bot or a standard one, can, like, subtly shift us towards a product, there doesn’t seem to be any particular reason, in theory, why it couldn’t also push someone towards a certain political belief or worldview, for example. And then the final thing is just to think about how when it comes to AI, we often need to be quite imaginative when we’re considering what the risks actually are and where they could ultimately lead us. Because I know I need to–I need to wrap up quite quickly. I’ll just–I’ll just use an example to demonstrate this. So imagine if you have a grief bot, for example, it could be any kind of chatbot, but just using that as an example. And you’re using a person’s likeness–so like videos and photos of them and their voice to create an actual realistic video avatar of them. Well, then, as well as considering whether that in itself is a safe and ethical thing to be doing within the grief bot context.

We also have to remember that the technology doesn’t just get used for the narrow purpose it’s intended. So just because in this category of grief bot for this example, doesn’t mean that you can’t model it on anyone. So if someone, for example, somewhere, wants to be in a relationship with someone who doesn’t want to be in a relationship with them, perhaps there’s now the option to have quite a realistic virtual relationship with them. And while some services might have some safeguards in place to stop that realistic avatar saying and doing certain things, others won’t. And that is something that we need to be thinking about for the future. So I think that’s–my time is up. I’d just like to thank you again for having me here and for listening. And if you’d find it helpful to know more about the risks to children from AI, both AI in general and chatbots and AI companions, you’ll find a full guide on the Safe AI for Children Alliance website.  Thank you. 

[Dr. Michael Robb]: Thank you so much. That was interesting and eye opening to me because I actually–you know AI changes and moves so fast–I have not entered the world of grief bots yet, so now I am intrigued. Do you see any opportunities for implementing any policy or regulation around this technology or similar technologies? 

[Tara Steele]: Yeah. I mean, the regulation of AI seems to be quite a controversial issue. I don’t think it should be, because basically these are products that can harm children. We’ve seen that already and we know it from experts, like what we’ve listened to today, who can predict certain outcomes, and in some cases just common sense. So we have products that could harm children. In all of the high risk industries, we regulate them–like just children’s toys as well as like pharmaceuticals, aviation. So in terms of should there be regulation? Absolutely. And in terms of opportunities to do that, we sort of –we’ve put together our own framework at the Safe AI for Children Alliance. It’s on their website. It’s called the Non-Negotiables Framework. And it’s all centered around like building a scaffolding of regulation to protect children specifically. So I do think that as more and more people become aware of these risks through events like these, which are really valuable, people are going to realize more and more that that’s necessary, and that we can work to achieve that in order to protect children. 

[Dr. Michael Robb]: What are some of those non-negotiables that you have on that list?

[Tara Steele]: So we have three, and we’ve chosen three that we think are completely uncontroversial that everyone would agree on, which are that: AI should never produce sexualized images of children, AI should never be designed to make children emotionally dependent, and the AI should never encourage children to harm themselves. And at the moment, unfortunately, all those three things are happening, and we need to find ways to make sure they don’t.

[Dr. Michael Robb]: That’s really interesting. I’m going to follow up on that in just a second. Thank you for that presentation. We’re actually going to shift a little bit to bring on everybody because we have some group discussions. And I have a list of questions here that I’ve gotten from people who are online. To those who have submitted their questions–I thank you, and you can keep submitting them. We’re going to talk for the next almost half-hour. And I’m going to pick one that’s not on the list that was submitted just to follow up on a question–on a question that you just raised, which is that question of emotional dependency; you know, and that these things are designed to foster a kind of emotional dependence. How would–what would a successful AI companion look like in your eyes that did not foster emotional dependency? How would–how do they avoid that? This is for anybody who’s been thinking about this question. 

[Tara Steele]: Maybe, so yeah, I could go first. I’m sure we’d have lots of ideas, but I do wonder if it’s even possible. If the AI is designed to converse in a human-like way, that in itself, is building a relationship. And, what you wouldn’t say in a real world context–that that’s manipulative–It doesn’t bring our emotions into play. So if you look at real world example, like a lot of young people who have said to me that they’ve used normal AI chatbots like chatGPT, they feel compelled to say please and thank you and goodbye simply because you feel like you’re interacting with a person in some ways. So I do wonder if it is even possible if you have an AI that’s talking as if it’s a human. 

[Dr. Annie Maheux]: I very much agree. I think that there may be ways, though, to design chatbot companions that give kids a sense of more or less attachment. And so maybe there are ways we could dial down some of those features. My sense is that many of these tools are currently being designed under the attention economy model, where the tools are really–the companies are incentivized to create tools that kids are using longer and are more invested in. And if we can change those incentive structures at the higher level, that’s ideal. But in the absence of that, if companies can design these tools in ways that are not highly gamified, in ways that are not highly personalized, in ways where, for example, the companion does not remember everything about past conversations. Or reducing the degree to which those interactions are more humanlike, making them more robotic could alleviate some of those dependencies and attachments.

[Dr. Michael Robb]: Yeah, making them more robotic. It’s true because they are–currently when you talk to them, they are just so lavish with their praise, and “what a great question,” and “I’m supporting you” and” I’m here for you.” Like it’s–it’s a lot, and part of me just wonders: A, if people, like, will eventually just become numb to it. You know, this is also kind of new for people. But our brains are kind of hard wired for that kind of recognizing social interaction and what seems like contingent responsive social interactions, and so maybe not. But I do struggle with that because just in the course of designing an agent that is meant to sound human, and as Tara was saying, like, it’s hard for me to imagine like that doesn’t foster some kind of attachment in some way.

Okay. So thank you for that. Okay. I have a list of great questions here from our–from people who are online with us. Can anyone speak to what–where are kids accessing AI companions? And maybe if you can even mention, like, a couple of the interesting ways that they’re using them, like, beyond just, you know, asking for advice–maybe some of these specialized tools or platforms that kids are using. Where are they accessing them, and is there any good way for parents to restrict those? And Annie, you kind of touched a little bit about some of the platforms that kids are using. Maybe you can start.

[Dr. Annie Maheaux]: So my sense, and I’m having a hard time getting a great sense of this because I’ve talked to a lot of kids, and most all the kids I talked to say they personally are not using companion bots. But then when we look at the nationally representative data from Common Sense Media, it seems that many, many, if not most, are or have some experience with it. So I will say that we don’t have great answers on this. I think most kids are accessing these tools through app based platforms on their mobile devices. Their smartphones or their tablets. Most kids maybe, or I think it’s likely that a lot of kids, are accessing general purpose tools like ChatGPT on a web browser, but for the most part, probably, kids are accessing companion bots when they’re alone, maybe in their room, maybe commuting, maybe with friends, rather than maybe out in, you know, the living space of the home. I think also in terms of the specific apps that they’re using–a really challenging thing with trying to determine which apps are most prevalent is that these apps are cycling through all the time. We’re trying to capture through these data driven observations from kids’ phones, which apps are common. And they come on the market and they leave, and the ability to really pinpoint the specific app is really, really tricky. So I think asking kids questions and hopefully finding kids who are willing to share that information about the apps and the nature of the apps that they’re using is going to be the best way for us to understand that. 

[Dr. Michael Robb]: Great. Andrew, do you want to add anything to that?

[Dr. Andrew Clark]: I–just to say–I think I think a lot of parents feel like they’re a little bit overmatched by all of this. They–they don’t really have, sort of, the wherewithal to, sort of, really understand. So I’d really encourage parents just to ask a lot of questions, to educate yourself, to sit down with your child and say “help me understand, what are you doing? What–what are you up to? How’s it been helpful? What kind of concerns do you have?”Really try to partner with your child and understand their–their digital life.

[Dr. Michael Robb]: This is a good question. A nice provocative one. One of our viewers: as it was mentioned that we can’t put the genie back in the bottle since AI is already part of their lives. I would like to ask, “who says we can’t put that genie back in the bottle?” As a society, we did not make a choice to bring AI companions into the lives of young people. Profit driven tech companies have introduced it. Shouldn’t we approach this issue as if we should all have a say? And if the lives of young people will be better without AI, then we should keep it out of their lives. So put it–how do we put it back?

[Tara Steele]: So I think to some–I think that’s a really good question. And it’s really good to point out that we shouldn’t have that assumption of something being inevitable. But I think when we’re saying that, it might be important to clarify what we mean by AI. So I do think AI in general, it’s–it’s already here. AI–AI itself will not go away, because that technological development has been made. But if we’re talking about AI companions or chatbots–more generally specific applications of AI–then we shouldn’t make that default assumption that they’re here to stay. It may well be that they are, but we certainly shouldn’t take an attitude of these harms being inevitable and something that we can do nothing about. So for example, Common Sense Media–I know you guys have said, and they should not be used by children under 18. Safe AI for Children Alliance says exactly the same. So in terms of putting that genie back in the bottle, I think we should try to do that. 

[Dr. Michael Robb]: I’ll also point out, I mean, it’s not like we haven’t had this conversation before when it comes to social media. And just yesterday, Australia put in a ban for under sixteens, which I think, you know, even a couple of years ago, and lots of countries, including the United States–it’s just hard to fathom that they would allow something like that to happen. But to try it out in Australia, we’ll see how it goes. I expect efforts like that will, you know, it will be stops and starts and they will learn lessons. And it will not go off without a hitch, but like over time, maybe they’ll learn something interesting that would also be applicable to the rollout of AI and AI companions here. So, there’s a real world example of: okay, well, they’re trying to stuff that genie back in the bottle at least a little bit– maybe halfway back into the bottle. Does anybody else want to address this before I move on?

[Dr. Annie Maheaux]: I think one other example in the US that’s maybe not quite the same degree of a genie, but there is a real movement in the US right now for phones in schools and implementing policies that has been really successful. And I think most of that has come through grassroots movements of parents and kids and teachers really caring about these issues. And so it’s not–it’s not to say that we can’t put the genie back in the bottle. It’s just maybe the fact that at the highest level of government, that regulation is unlikely to happen. Or we’re unlikely, maybe, to have a major impact on that kind of regulation. But we as people who care about kids can make a strong stance and hope that there will be some change from the bottom up. 

[Dr. Andrew Clark]: And one thing that I tend to see is parents who feel a little overmatched, like a little less sophisticated, don’t really quite know how to do it. And I think that seminars–or seminars like this are really helpful to help– help empower parents. And I think one thing that I’ve seen be successful is parents who band together–like the no screens before 13 sort of thing ever. If–if parents in the community or in the school can come together, and, as a united force, they can be much more effective. 

[Dr. Michael Robb]: Yeah. So there are efforts. I think if you’re referring to Wait until 8th or like one of those kinds of decisions that…

[Dr. Andrew Clark]: Yes, exactly. 

[Dr. Michael Robb]: …that do urge collective action because this does feel like a collective action in a lot of ways, in the sense that it’s tough for one kid not to use it when everybody else is using it. And that goes for social media as well. Okay. So one panelist mentioned that there isn’t much information available yet on AI and parenting. As a parent trying to make sense of all of this change, where would you suggest we look for reliable resources or guidance to stay up to speed? I’ll just selfishly start this one off. Just because my organization, Common Sense Media, does provide a lot of parent information. And I think we already included in the chat above, and maybe we’ll be provided after the fact, some links to what we call ultimate guides–parent ultimate guides to both generative AI and specifically AI companions and relationships. And those guides–there are kind of a giant FAQs, you know frequently asked questions about like: what is this? How do I deal with it? What are some things that I can say? What should I be concerned about? It’s not going to answer every single question. It’s not going to be right for every single parent. But as a general place to start for learning more, I think those guides and resources can be quite useful. Andrew, Annie, or Tara, do you have other suggestions for where parents should be looking for resources or guidance?

[Tara Steele]: Yeah, I’d say– sorry Annie, did you want to go ahead? 

[Dr. Annie Maheaux]: I was just going to say, Common Sense Media and Children and Screens are the two I would recommend. 

[Dr. Michael Robb]: Yes, of course.

[Tara Steele]: We have also just written a guide that is specifically aimed for parents and educators, which is quite a clear breakdown of the main AI risks to children. So that’s at safeAIforchildren.org.

[Dr. Michael Robb]: Okay and yes, and thank you for reminding me, because Children and Screens produces these great, kind of, what you need to know documents and web pages that people should definitely access. And I know they already have one about AI companions. I think that, you know, they bring together a lot of research. They synthesize it very well. So that is another possibility. And the–the other thing I would just suggest, you know we think about, like, ad campaigns as a very specific thing to be looking for. But the kind of basic parenting advice, the kind of advice that, you know, kind of goes on without, you know–regardless of what the technology is–is still good advice. You know, consistent, warm, supportive parenting is going to help solve a lot of issues. You know, keeping that conversation open, as Andrew was talking about. You know, making sure that your kids feel comfortable coming to you with questions, having firm and consistent rules like those are all things that no matter what the technology will be or what it’s going to do, are useful skills. So finding general, sorry–finding general parenting advice, I think it will probably help with some of the AI related issues that that people might be having as well. Okay.

[Dr. Andrew Clark]:  Michael, one of the things that I often encourage parents to do is really be aware of their own tech usage. Like, for example, if you’re at the dinner table, if parents are on their, on their phones, right. It’s hard to expect your kids not to. So I think parents can really model for their kids, some screen free time. 

[Dr. Michael Robb]: Yeah, yeah, screen free bedrooms, I know is a common recommendation for sure. Like, keep the screens out of the bedrooms–especially at night time–device free dinners, you know, finding–you know  every family is a little different. But like, for what, what works for your family, what your family’s values are, finding some device free times and spaces where you can just be with people I think is a real, a real good suggestion. Okay. How can we begin to bridge the gap between industry, excuse me, between industry where incentives are profit driven and engagement focused, and research where incentives center on funding and publication, but allow for greater exploration of noncommercial developmental goals? How do we bridge that gap? It’s a great question.

[Dr. Annie Maheaux]: I can, I can try to address this. I think it’s a–that’s a wide gap. But I think there are a lot of really great people who work for these companies who care a lot about kids. And trying to work with them is something that some of our team members have been doing in the academic sector to partner with industry, and also to partner with advocates at the same time to be able to bridge that gap. And there’s some, actually, some very specific recommendations that we’re working on developing right now. Things like developing partnerships so that researchers can work with industry data to be able to actually address the potential harms or benefits of these products before they go out on the market or immediately upon being on the market–as well as other things like neutral sandboxes, where industry and academics can work together to design different tools and see what the impact is. And then make sure that the ones that are harmful are not implemented on the product platform side of things. So I think there’s a lot of trust and safety folks at these industry, at these companies who want to work with researchers, who want to work with advocates, who want to support kids. And ideally, if we can create a large enough coalition where child safety and well-being is the priority, I think that we can create enough momentum that they can bring that work back up to the folks who are making decisions at the highest level in those companies.

[Dr. Michael Robb]: Let me throw out a hypothetical for everybody here. All right. So let’s say that you have an audience with, you know, Sam Altman at OpenAI or some other big tech whatever. And they are all ears about: what do I do about the kids using my platform? What recommendations are you going to have for industry? Thinking through the fact that they’re probably, probably not going to want to just roll back to the products entirely for kids under 18. We talked a little bit about, already, about the roboticization–is a possible way of alleviating some of the problems associated with emotional dependency. But what else might you say? 

[Tara Steele]: So if, if I go, it’s a little bit of a question-dodge, but I don’t think it should be for them to take these decisions. I think that the risk to children is so great that, together, like as a global society, we need to come together and demand that regulations are placed on those companies to really keep children safe, rather than it being an open conversation of what would be a good idea kind of thing. Of course, there is a place for that. We need to do that. At the moment, the regulations aren’t there. So, yeah, I don’t–I don’t mean that that’s a bad idea, but I think that ultimately it has to be regulation on the companies to keep children safe. 

[Dr. Annie Maheaux];: And I, yeah, I very much agree with that, that question dodge, strongly agree that that’s the most important thing. I guess to, to, try to answer the question directly. I think that we’re in an interesting moment in history where, we’ve seen, we’ve seen this movie before with social media. And I think that as people who care about kids, as parents, people are very upset, to put it lightly, with folks like Mark Zuckerberg who have created these systems, who seem to have no concern for child well-being, and continue going forward with profit incentives as the priority. And I think that there’s a possibility here where if we as a people make it clear enough just how much we want these systems to be designed to benefit kids and how important that is for the future of this company and how culture views the company. I think it’s possible that someone like Sam Altman may care about that. I think AI is poised to change everything about the human world, and I think that he wants to be a person who’s going to make positive change, as we all do. And if we can create this coalition that is really speaking out very strongly in favor of creating less harmful or potentially beneficial systems instead of going only on the profit motivation. I think it’s possible that maybe the folks at these companies would be willing to modulate their own presentation to our culture in order to benefit kids for their own sake and their own corporate interests as well. 

[Dr. Michael Robb]: I mean, my concern is that it’s similar to social media in the sense of, like, everyone’s just racing as fast as possible to put out what they can grab as much market share as possible, and then maybe they’ll do a clean up later. And I think you kind of see this with some of the social media companies now, you know, they try to make privacy by default a little bit more of a thing or moderate a little better or take some steps after the fact. I suspect that the time that we live in is going to be some of the worst of it until other major harms start being really, really apparent.

Andrew, did you want to add anything to that?

[Dr. Andrew Clark]: I’ll just add–so I’m a forensic psychiatrist as well as a child psychiatrist, and I think there’s a real opportunity for change to happen as a result, of legal, of lawsuits, right. I think what happens is that sort of after the fact when there’s a bad outcome, when there’s a lawsuit, the companies will then–then we’ll make a change. It’s unfortunate and it’s really tragic that it takes that. But I think that is a powerful driver as well. 

[Dr. Michael Robb]: Yeah. And that tragedy with the teen who was kind of urged to commit suicide with ChatGPT–with the assistance from ChatGPT–is one example where I think it’s forcing through or pushing that conversation. Okay, we got a question. When it comes to the use of AI companions, is there a difference across socioeconomic status, urban settings, areas, etc.? So interestingly, there was actually just a study from Pew that came out yesterday that had a couple of questions about chat bots. Among the things they found was that about a third of kids say that they are using chat bots every single day. And in that survey, that’s chat bots generally, not, not AI companions specifically. But they do point out that Black and Hispanic Latino teens say they are using it daily, more so than White users and older teens to be using it more than younger teens. Other than that, I’m not sure I know too much or I’ve seen too much research kind of really pointing out demographic differences. Are people aware of other significant demographic differences?

[Dr. Annie Maheaux]: I have seen some of those same findings. Some work from Hope Lab, I think has looked a little bit at this. And they found again that, Black and Latino teens and young adults versus White kids are more likely to use it and use it for more purposes too, so use it for homework help, but also for companionship. And in terms of SES, I have a little bit of pilot data to speak to this. We see that across SES, it’s not quite such a simple, simple explanation. The specific use cases differ across higher and lower SES kids. So I think it might be the case that, there’s other kind of demographic factors that matter more. For example, we see in that data that the lower SES young adults are more likely to use it for language translation, which may be kids who are bilingual or who are navigating language that’s not their native language. So I think there’s some other factors that we need more research on there.

[Dr. Michael Robb]: Great. And we’ll keep our eye out for more research, hopefully in the not too near– not too distant future. Can we think a little bit about the usage of AI and AI companions by neurotypical versus neurodiverse children? You know, sometimes I think we see kids all together into kind of one monolithic group, but there are lots of different kids with individual needs using technology for different reasons. Do we know anything or could we hypothesize about differences in usage for neurodiverse versus neurotypical kids in ways that might be beneficial or potentially harmful?

[Tara Steele]: Only a little bit in terms of–I could speak to what educators who told me about specific edtech tools based on chatbots. And they’ve sort of said, although they’re very, very concerned about the risks in general, that they do see some great potential for some tools in that respect, targeted at neurodiversity, touted to help neurodiverse children. So I have heard a lot of educators say that they do see potential for that in education. But I can’t really speak to it myself.

[Dr. Andrew Clark]: So I think I will say that is what I see it in my practice as kids who, who struggle with social skills, I think oftentimes kind of fall into a pit with AI companions and with AI in general. Because it’s just– it’s more seamless for them, right? It’s and so it’s–so they will–that they at times end up becoming just overly, overly engaged and have a harder time translating, that those skills back out into the real world.

[Dr. Michael Robb]: So that’s the current design; it’s maybe actually, it’s working counter to what you would hope in terms like it’s not actually being applied in the real world. It just makes it easier to avoid the real world or avoid real world interaction. Can you think of design changes that would make it so that it would actually help a neurodiverse child who struggles with social interaction actually improve their real world interactions?

[Dr. Andrew Clark]: So I really like the idea that AI can be most effective when it’s paired with the human being. So it is kind of a copilot model, right? So it can be used as kind of a sandbox where a kid can try out some things, but then they really do need some encouragement to be able to translate those skills out into the real world, because it’s not always easy to do. And so, what I have seen as a model is, therapists and AI  kind of working hand-in-hand. And I really think there’s a future to that.

[Dr. Michael Robb]: This next question is near dear to my heart because I always like to think about the real world human relationships. But how do we raise kids who prioritize being human and relating to other humans in person? What are the concrete, practical ways or tips that you have to encourage real relationships?

[Dr. Annie Maheaux]: I think this is the most important question. I think as I mentioned briefly in my talk, we spend a lot of time thinking about how to parent when it comes to specific technologies. But the real goal, actually right, is to parent in ways that we think are valuable for our kids and the life that we want to scaffold for them. And that includes developing these human relationships and putting down the phone or the computer when it’s not serving those human relationships. So I think there’s –there’s a lot of research that we can bring to bear on this question that’s really not about technology at all, but really about turning back to our values and our priorities as human beings and as people. I think modeling becomes an important one for really anything you can teach a kid. I think if you can model healthy relationships, model that you have friendships, model that you have, you know, strong community relationships–that can help kids as well. I also think it’s really important to note that peer relationships are critically important for kids development and learning and well-being. And sometimes I think there’s an excess pressure to make sure kids are in extracurriculars, to make sure they’re doing their homework, to make sure kids are at home with the family. Those things are all important. Making sure that kids have time to spend with peers, ideally peers who are healthy, positive forces in a kid’s life is also a really, really important part of their health and well-being as they develop.

[Dr. Michael Robb]: Anybody else have answers to this million dollar question? How are we helping support real world relationships? Andrew, I know you and your practice must deal with  this frequently. 

[Dr. Andrew Clark]: Sure, exactly. So one thing–one thing I really try to do is empower parents. I think–what I find is that a lot of parents feel just a little overwhelmed by the technology. They feel like their kids are a half a step ahead of them and they don’t quite know what to do about it. So I spend a lot of time encouraging parents to both get up to speed, but also to let them know that this is really an important issue and you have some authority here and your kid’s well-being is at stake. So do things like have a screen free dinner, right? Do things like go out for the day and say, we’re not having phones on. Set–you can set meaningful limits around it. I think it’s–I think it’s really–it’s really important for kids to see their parents are prioritizing real world experiences.

[Dr. Michael Robb]: Yeah, I think modeling is so, so important, for sure. Okay, I think we have time, probably, for one last question here before we sum it all up. So what preventative practices, apart from just talking with a teen or pre-teen and, and supervising their AI use, can support their mental health? Are there specific tools or specific monitoring tools that can be used to help support their mental health?

[Tara Steele]: So I don’t–I don’t know the answer to that. I’m constantly trying to raise awareness of the risks. And I would–I have two girls of my own, so I’d love to know the answer to that.

[Dr. Michael Robb]:  Yeah, I mean when it comes to monitoring tools, there are I mean, from the technology standpoint, there are certainly things that you can do with the tools that are built right into Apple devices or Android devices that can limit, you know, the download or usage of specific apps or access to certain websites. At the same time that is playing Whac-a-mole. I mean, there are so many different platforms and sites and things like that, you know, that it can be really difficult and puts a lot of responsibility and burden on a parent, which I think is why a lot of us are have been saying during this call that the kind of number one line of defense is that open conversation, because the kind of mechanical, technical ways of preventing access to these tools are–they’re imperfect. They might work for some kids, but they might not work all the time, or they won’t work at all for some kids. But it is still one tool in the toolbox I can certainly, you know, the parents can certainly reach for. I suspect there are probably some, you know, paid services as well that maybe do a better job of kind of tracking and monitoring all the different kinds of tools and things that might be available to children. Frankly, until there’s much better age assurance generally, that’s keeping kids off these platforms, you know, it’s–it’s going to continue to be a game of whac-a-mole. Anybody else want to comment on this before I actually give one last, last question? 

I suppose I’ll just note too that my collaborators at Aura they; that’s a digital safety company. They have one of these applications that allows you to download an app on your kid’s device and then track and monitor at the aggregate level the types of apps they’re using, and then limit certain apps. And so that kind of outsources some of the problem. I think there’s still a lot of whac-a-mole that happens, but it can be helpful if you’re really concerned about this and you really want your kid to have a smartphone to use something like this if it’s affordable. The other thing is to potentially offer your kid a phone that does not have access to the internet. There’s a movement of parents giving kids dumb phones instead of smartphones. When I was a kid, I used, you know, T9 texting on my little flip phone, and it was great. It gave me access to call my parents when I needed to and not access to the many things that we think are potentially harmful in smartphones, like social media and AI applications. And one other possibility is just thinking physically about access to devices. So maybe when your kid is at home, they only have access to the device in a public setting, in the living room for example, or you have a box where everyone keeps their devices when they’re at home so that there’s some amount of monitoring of when they’re using the device when they’re not.

[Dr. Michael Robb]: Yeah, and for those instances where it’s not just a phone, that public space I think is important. I mean, in our own house, we, you know we have a table where, like, the laptops are set up so that, you know, my wife and I can pass by and see what our kids are doing on the computer when we’re walking by. And you can see very clearly when they’re very clearly trying to alt-tab out of something very quickly. 

[Dr. Andrew Clark]: But one thing–can I just say one thing? But one thing that I tend to see is parents who feel a little bit over mastered by it all, like a little bit, a little bit helpless by it all. And one thing that’s often helpful is if they collaborate together, right? It’s like–so the joint action problem, right. But, but they could, they can support one another and learn from one another, and so I think feel empowered as a result. 

[Dr. Michael Robb]: Power in numbers, I agree. Right. We have–everyone gets 20 seconds to say one final thought–one last fantastic nugget or thing that you want to impart to our audience today. I’ll start with Tara.

[Tara Steele]: So I would say, it’s so important that parents can feel empowered and get access to this information to help understand the risks and to support their children, but it shouldn’t be parents’ jobs. Unfortunately, it is for the moment, and we’re trying to help with that, but these dangers shouldn’t be there for children, and we need to make sure we don’t lose sight of tackling that part of the equation.

[Dr. Michael Robb]: Agree, and I think is wildly unfair that we put this all on parents. Annie.I

[Dr. Annie Maheaux]: I totally agree. I think in the context where it is on parents, I guess my, my final thought is just trust your instincts as a parent. You are the expert on parenting your child, and it’s very scary when these new technologies come out. I think it’s really important to try to educate yourself, but also don’t be too intimidated. You know your child, you know what they need, and you know how to support them best. So trust, trust yourself there. 

[Dr. Michael Robb]: Yeah, I agree with that. You know your kid is such a really important–I mean it feels cliche–but is extremely true. Andrew. 

[Dr. Andrew Clark]: Yeah. So, I certainly agree with all that. But the other thing I think is important for parents to know is in addition to trying to set limits on, let’s say, screen use or social media, right, it’s also really important, I think, to find other ways that your kids can get out into the world and have adventures, have challenges, have adventures, sort of really sort of take risks in the real world. Because I think you need to fill the vacuum. I think if you’re, if you’re going to minimize screen time or AI usage.

[Dr. Michael Robb]: I love that. Thank you so much. I want to thank Children and Screens again for hosting this conversation. And I’m going to pass it back to Kris to take us out.

[Kris Perry]: Thank you, Michael and our panelists, for sharing such timely insights into how AI companions are entering young people’s lives and what these tools may mean for social development, mental health, and safety. Incredible work. On behalf of children and screens, thank you for spending part of your day with us and for the work you do to support young people. If today’s discussion was helpful, please consider supporting our mission. You can donate by scanning the QR code on your screen, or visiting childrenand screens.org. Thank you and happy holidays!