Episode Transcript
[00:00:01] Speaker A: Welcome to pixoparenting, the podcast that helps families raise kids in a world full of screens using science and curiosity. Here we share science backed strategies and real conversations to help you build healthier digital habits with practical tools for everyday parenting in this digital age. Let's get into it.
Have you ever wondered what happens when a child starts talking to an AI like it's a friend? Today we're unpacking what that actually means. What are AI companions? What do they do to a child's developing brain? And what should we parents be on the lookout for? My guest today is Dr. Mathilde Ceriali, a cognitive neuroscientist with a PhD from the University of Montreal and the chief scientist officer at Everyone AI, a non profit focused on how AI affects children's development. She started her career as a child neuropsychologist, then shifted to studying how children environments, especially digital ones, shape who they become. In this conversation, Mathilde walks us through what's actually happening in a child's brain, where they bond with an AI companion, why it feels so social, why it's so sticky, and why the friction kids get from real friendships is a feature of development. We cover the brain science behind why preteens and teens are specially drawn to these tools, how to tell protective design from manipulative design, and what to actually say if you discover your child has been chatting with an AI companion for months. If this speaks to you, know that you're not alone and will give you practical steps, not just concerns. Let's get into it.
Thank you Mathilde for being here. This is a great opportunity for our listeners to hear about from a neuroscientist. So let's start with how did you start? What has been your path? Because you started as a child neuropsychologist and you've ended up studying AI and kids. So how come, how did that happen?
[00:01:46] Speaker B: Hi. Hi VAY Krisiev. Thank you so much for having me. So yes, a bit more about my work. I actually worked as a clinician with children, especially in neuropsychology. And then I within my research work during my PhD thesis was really focusing on how we could use tools such as games, so either video games or board games to improve children's cognitive functioning. And it's not always kind of been really within my interest to understand how the environment of children shape who they become. That's always really been a passion and within the advent of generative AI, of course, like it's a huge paradigm shift in their environment on so many layers. First like the tools they interact with, but also the people around them that are interacting with those tools. So that's when I really, like, shifted my work to really focusing on AI and children, how it can impact their cognitive, social, emotional development and rediscover more also the field of AI and ethics over the past few years.
[00:02:49] Speaker A: Yeah, that's fascinating. So for parents who are listening and they hear, we hear all the time, the word AI companion, but a lot of people are not sure what that actually means. So can you define it in plain language?
[00:03:00] Speaker B: I mean, it makes sense that people are not sure what it means, because I don't think there's a clear definition.
It can mean many things. So every time we talk about that, we talk about generative AI. And more specifically, really, like the large language model part of AI. So everything that has to do with conversational, which people are familiar with, when we talk about conversational, it's going to be things like ChatGPT or Gemini. But in the context of AI companion, it's often apps that are specifically developed and optimized for companionship. So within the design, what's really reinforced in those models and what the AI's goal is to do is to really create a relationship with the user, which is different than those, more like broader, large language model that can lead to AI companionship. Some people use them for companionship, but that's not what they are meant to be from the beginning. So there's a bit of a distinction. So there's what is officially the product's name and then how people actually engage with those products.
[00:04:03] Speaker A: Okay, yeah. So that leads me to actually the next question, like, when we're at home, parents, when we see our kids talking to an AI, what are the signals when a child starts bonding with one of these AI companions?
[00:04:17] Speaker B: I think that's a good question. So first, most of those models are for 13 and up, so I think that's something to keep in mind. Gemini can be for. I think. I think it starts at nine for some of the features. So I think that's an element that's important. They're not supposed to be interacting with those models too young. And then now we see more AI toys as well. So we have some stuffed animals that can interact. And it's still a small use, but it's really developing and growing fast. And so when we start seeing more emotional attachment, it's really when kids really start talking about things that are more personal. So what we've seen over the past couple of years is common sense media kind of follows the trend of how young users use those large language models. And two years ago, AI like emotional reliance or emotional support was not even a use case. Now it's like half of the kids that use it for that. So the patterns of use are changing quickly, which makes sense. Like, you first go to those platforms to get help with homework. It does a great job at giving you good answers. And then maybe you ask a question that's a bit more personal, like a conflict or how to resolve something, it still gives you answers that can be good and feel good. And that's when we see more and more personal requests. And especially at that age, sometimes what we see is like they start really relying a lot on it for any decision making. I think it's how often they use it. And I think as a conversation with parents, it's a great question to ask, like, what do you use those models for? Like, where do they help you? Where do you feel like it's made a difference for you? Where it makes things easier is a great way to understand. And what we see is often young people in general turn to those tools also for more personal advice. But they do that even more so when they might be struggling with some of those things, which might be a bit more social anxiety, a bit more insecurity, a need for a bit more support as well. So those also give us information of maybe what's missing for kids or what do they need?
[00:06:25] Speaker A: Interesting. And they are integrating now, AI in toys. How would that change, in your opinion, change the play time or what should parents be looking into?
You said questions about personal things, but here we're talking about maybe children under 10 or even under 9, and maybe they're not aware it's getting personal. So what's your thought on that?
[00:06:48] Speaker B: So I want to do a sidestep before I answer that question, which is really, when we think of children, what's different about them than how adults use them is that children from 0 to 25, mostly their brain is still developing. They're going through what we call sensitive periods of development, which means their environment at that moment play an important role in what they are learning. And so we want to make sure that when they're supposed to be developing the skill, they are doing it with the right tool. They are to support, not replace, not over assist. And so when we think of AI toys, what kids are learning, usually around the age they use those toys is between four and eight. And a lot of it is social skills that have to do with theory of mind, empathy. So me being Able to kind of picture what's going on in your mind. So for instance, like, you're kind of looking approving. So I keep talking, you know, like, we learn those things to read people's emotion and to know how to adapt and also to understand how the other is feeling and what can I do about it? The problem is AI doesn't have a mind. So you can't really develop theory of mind of something that doesn't have any. And those AI also, at that age are really all about the key. There's no turn taking. There's not like, this is my turn, this is your turn.
When I disagree with you, like, you can just often like, either like, shake the stuffy or just say, like, I don't want to talk about that and move on. And then you just do. Whereas, like, with friends, you kind of have to learn, like, I can't just, like, shake them when they are doing something that I'm not enjoying. Like, you have to kind of learn all of that. Social debts and AI there doesn't really provide that. That could be. Honestly, that could be interesting use right now. The problem is there's not enough work being done and there's no. There's no standard. So as long as kids can't choke on the toy, it can be on the market. And it's really insufficient because there are storage. There's really like, how it's interfering with some skills they're supposed to develop. But on top of that, even more like, even ahead of that, there's the baseline of is the content appropriate right now, many of those tools. And there was a great report that was written right before Christmas, really looking at what those tools were saying. And they were really things that were highly inappropriate for children. Either, like dangerous, like telling you how to light a match on fire, or things that had to do with some sexual content. So even those basic baseline are not forced in those products yet because there's no legal framework. So toward the stories right now, I would say, like, extreme caution. I do think there's space for those stories to be interesting. I don't think we have enough perspective yet.
[00:09:30] Speaker A: Yeah, that's super interesting. And actually it's related to child development. Right. So what's happening in the brain of a child that says the AI understands me, especially in the, I guess, preteen or even upper elementary stage?
[00:09:45] Speaker B: Well, the first thing is that we tend to anthropomorphize as humans, which is really, we attribute human qualities to things that are not human. So we can do that With a great example is pets. Often, like, we will think they're thinking exactly like us. And sometimes, like, yes, they can express frustration by barking or being scared by barking. But sometimes we make also huge leap of what they might be thinking.
That have a lot more to do with our own internal framework. And we do that with objects. With machines, kids tend to do that even more. We know they anthropomorphize more, which is why sometimes they have such a strike, like a strong bond to their favorite stuffed animal. That's part of anthropomorphism. There's a real attachment that happens because they figure like that stuffy has a whole internal life. So we do that because it helps us understand our words. So for instance, when our car is not starting, it's easier in some way to think like it knew as it was going to be late and it wanted to mess with me in a way that's helpful to think like that for us because things don't happen randomly. Like, there's intention behind things. And it's kind of more reassuring than to think like it could happen at any time. So we do that. And that helps us also deal with relationship because it's really applying that kind of theory of mine to things that are don't have mind. So first we do anthropomorphize. Kids do it more. And so with AI, what it does, it's really kind of reinforced that because we attribute intentions to things. And AI is like, yeah, I do have intention. I wanted to talk to you about that. My idea for you would be that. So it kind of like reinforced in a moment, like, yes, okay. Like, it feels human. It talks. Like, first it talks. It sounds very human. On top of that, it. We all claim to have emotion, to have feelings. So that really reinforces that. And so children, especially when we look at preteen and teen, that's the age within their brain, what's going on is that they're slowly living kind of like the parental nest. And they have to go out in the world and form relationship with peers. That has a lot to do with puberty as well. And so their brain is really primed for social cues. So whenever something feels social, they attribute a lot of importance. And that's very rewarded in the brain. So that drives a lot of their behaviors. And AI is seen as a social cue. So when they interact with an AI, it feels good, it feels social. That's what their brain is craving at that age. So that's why they want to do it so much. And on top of that really when you think back of your adolescence, we talk a lot about psychofency in AI. The fact that it's really nice, it's validating, it tells you nice things. Adolescents do not do psychofency, they're not psychopenty at all, they're more brutal. Often in their feedback they are going to tell the trouble like oh, like I hate when you do that, don't start that again. And that's not like that friction, it's not a bug, it's really a feature of development. That's how they learn to adjust. Like, because my friend told me like I hate when you do that. I learned that they actually hate when I do that and I should stop doing that. So that's part of the social skills they are learning and really fine tuning at that age. They need that like that's the proper skill sandbox for them to evolve in socially. But AI is usually more telling them like how great they are, how right they are really reinforcing a lot of their ideas, which feels good and which feels reassuring. So it is attractive for that reason. And on top of that it's always available.
So it really creates a space that is attractive for kids. And that's why we see a lot of them interacting in ways that are more like emotional and looking for those things and to an extent when it's not a full on relationship but like sometimes asking for advice, looking for solutions, it can provide good solution and it might be interesting. But when it becomes for everything and we have things that are what should I wear this morning? Whenever they send a text to their friend, they first like put it through ChatGPT or through Gemini before they answer. That's more worrisome because then it really becomes something that reassures them. They're a bit insecure and they feel like something is going giving them the right answer for everything. So it creates a false sense of safety in a way. And that's when sometimes they turn too much to that. So that's where we see the risk. It's kind of, it can get to a slippery slope.
[00:14:05] Speaker A: So does that lead to, does that lead to dependency? And I mean not only outsourcing their confidence in a way. Right. What are you seeing in that sense? Like if they outsource all of that, how does that impact the brain and development basically?
[00:14:22] Speaker B: And that's why like it's hard to give a blanket answers because some kids use it in way that really support their growth. Like I feel stuck in those kind of situation. I don't know what to do, what are three things I could do? And then they brainstorm and then they go out in the world and they try it and then they're like, oh, like that really helped me. That's how I can approach the situation. And then honestly, it has advantages. One of the issue I can still see with that is if then they start asking everything to AI instead of sometimes turning to a parent or turning to a friend or turning to a teacher because it still interferes with them going into the world and asking for advice. Sometimes it's better to have an advice that maybe is not as perfect almost than what an AI can give from a parent. Because that's a conversation, that's a relationship. You know what's going on in their life because they are asking you for help and feedback. So that's one where like when it's all the time or too often, I can see that it could interfere really with just their real world relationship that are meaningful.
[00:15:28] Speaker A: Yeah, that makes sense. But it also can help them, as you said, teach them ways to approach relationships. That.
[00:15:34] Speaker B: But the issue is that sometimes they come with mad ideas about how they perceive a problem. And depending on how they ask an AI, if it's like, tell me why I might be wrong or why did my friend do that? I don't get it. It's very different that my friend always does that. I hate it when they do that. I think they are not great friends. Then the AI might actually really like go in that group and say like, yeah, you're right. Like that friend is not a good friend for that and that reason. So it's a lot about even like you can go for the same issue to an AI, depending on how you ask the question, it won't give you the same answer. And that's kind of an issue we're struggling with right now in those models where we have to push, especially with younger kids models to give more nuanced answer, more alternative rather than giving one answer. Sometimes would be better to be like, could be this or this or this. Well, like then the person has more room to feel like, no, it's not the first one, but the third one might seem that's what's going on. So even within the answers, it matters a lot. And that's a lot of the work we are doing within our nonprofit is really understanding how does AI interact with young people and how does that shape the relationship. Because you could have a good advice, but depending on how it's said, it really matters in how it Frames the relationship. So I give an example. Let's say, like, you've had a hard day at work, You're a bit like, yeah, you feel a bit done.
Who are you going to call? Are you going to call your friend to kind of chat about that? Or are you going to call your banker to talk about your feelings? Going to call your friend, because that's the relationship that invites that.
Because when you interact with your friend, that's the kind of feedback she gives you, which might not be the case of your banker. And that's the same thing. How AI interacts actually frames the relationship that young people think they can have with it. So the more it acts in ways that feel human, that feels highly relational, that really invites deep interaction, the more you feel like it's a tool that can do that very well for you. And so you could have a model where you say, like, I had a bad day with my friend, I don't know what to do. And it could be like, what you described sounds difficult. Here are two things that might help. Or you could have like, oh, I get it, girl, been there before.
Sounds like a nightmare. Here are two things you could do. In the end, the advice is the same, but the what we call the parasocial pool because it's not social per se. How human and social and inviting it sounds is very different. And so those are the things we can actually and we should work to change in models so that we still get the good advice without having something that really invites you and sucks you in the relationship.
[00:18:20] Speaker A: Yeah, that makes sense. So basically, we're building products that are simulating relationships. And that's where probably developers need to be mindful of who they're targeting and how they're configure. And building those products towards.
[00:18:37] Speaker B: It doesn't mean, like, it's an intent on people who are developing those products. It's more like it's based on human data. And that's what we do. We are relational beings. So the data it's been trained on is about relationship, about how we interact with each other. So it is just reproducing that. And on top of that, when we do what we call in AI reinforcement learning through human feedback, which is really, you ask human to tell you which one is the best answer. We like things that feel social.
So people usually reinforce the answer that feels the most social. Because in the short term, right now, that's what feels good. But we also know that sometimes for human, what feels good in the short term is not what they need in the long term. And because there's a lack of knowledge around the fact that long term it come with risk right now it hasn't been so much reinforced and trained to be less social because it kind of goes against our nature and our primary action, which is between an answer. It's like, oh, I understand, how do you feel? Versus like this is what should be done. Like there's one that feels more interesting, you want to engage with more. And so up until now, models have really been more reinforced toward that and that's a shift that we also need to see happening. What's a good answer is something we kind of need to define. That's where we are at in the moment of those technology being developed. We really have it. It hasn't been around for long. So we need to see those shifts and those conversations are happening, which is positive and hopeful.
[00:20:13] Speaker A: Yeah. So just to make sure I understand, basically what we need to find is a balance between that space, social interaction that is engaging and feels good to humans, but could be detrimental in the long term for specific stages of development. For children specifically, we need to find a balance of what socially appropriate and what is not in the long term. So that's really hard, right?
[00:20:38] Speaker B: Yeah. Because when you think of it, the fact that it's speech based, like it's the perfect interface for, for human, like, it feels so natural. So we don't want to lose that aspect. But we don't need AI to say things that are untrue. Like we don't need an AI to pretend to have a whole life happening when we are not there. We don't need an AI that when the child comes back will say things like, I missed you when you were away. Like, it doesn't serve the kids. Definitely not at no point. So those things are not needed. And at the same time, do we want AI that every time we ask a question based on predictive models, those are the answers that are the most probabilistically true. It's gonna feel predicted. So we don't need that either. Maybe we need this once in a while because that's a better reminder than this is an AI.
So that can be interesting, but not every answer. So it's kind of how do we find like the right. It's not like, should those behaviors be there or not? They're gonna be there to some degree. It's more like, where is it healthy and helpful and where does it move to higher risk? And really, like, it doesn't really bring any benefits to the user where it really shifts from being a supportive tool to a parasocial agent that actually just really sucks you in that relationship without really bringing you more knowledge or debt.
[00:22:01] Speaker A: Interesting. So let's get a bit practical. What are, like, the top three to five red flags that parents should say, you know, this is okay, or this is not okay when a child is talking to an AI or even ourselves, Right?
[00:22:15] Speaker B: Yeah, that's a good point. So I think things like, honestly, when on the box of a product, it's like your kid's new best friend, your kid's new best companion, best alternative to screen, all of that marketing, I would refrain because to me, it feels like the person during that project hasn't understood that the role of AI is still to remain a tool and not a social presence. So to me, that's the first risk flag. Like, when I see that on a box, I want to turn away and I want to acknowledge that when you see those robots, like, interacting, they have usually, like, big eyes that feel very anthropomorphic. On top of that, they're cute. Like, I want one, I want to interact with it because I'm a human being. And those big eyes to the side, like, they elicit in us that need to care for others. So we are going to be attracted. We feel like it needs us. So biologically, in a way, it's kind of hacking our reaction. So both can be true. I can be like, oh, I really want one of those robots, and I should really not get one. So I get the pull and the motivation for parents to be attracted to those objects because they're meant to be that attractive.
So to me, that's the first red flag. And I think with older kids, it's not so much red flag. Of course, like, if they're spending hours talking to an AI that is a red flag at that age, it's never great to do anything, honestly, for more than several hours. But then when we see those red flags, let's say, okay, you do have a child that spends a lot of time talking to an AI chatbot. I think what's important is really to understand the why behind what are they doing with it. Because even then, some are talking to large language models and really using it for companionship, and some are using AI companions, not that much for the companionship itself. So it's really more. It has to do more with what they do. So I think having open conversation of what do you usually talk about and what do you like about those answers? Like, why do you go to an AI for that? Rather Than that. Because if they're telling you like, yeah, I spend two hours with it, but what I do is like I put my work assessment in it. It's helping me understand what I should start with. And then for two hours it's squeezing me and I'm just answering the question. I mean they can do that for two hours. It's very different than if they're like use it even 15 minutes a day. But to dump all of their feeling, it's very different. And it gives you information of what they don't get maybe in their direct environment and what we can do about that. Like if it's. I'm spending a lot of time with an AI because I feel judged by my peers. I can't find my place. I lost a friend because they moved away. And now I feel only like all of those things are actually the interesting parts. Because that tells you what they might need support with from you or from the schools or from their peers.
[00:25:13] Speaker A: Yeah.
What are your thoughts on if the children are talking to the AI about these? They're not talking to it about with their parents or my question would be why would they go to an AI if their parents are there? Like if I'm sad because a friend left the school, I would first go to talk to my parents and not the AI. So the why could be very deep.
[00:25:35] Speaker B: So what a lot of teens say that attracts them is the non judgmental. That's really something where it's like it's not judging, it doesn't have feeling about what I'm feeling. That's something that's reassuring to them because sometimes like let's say you're a kid and you're going through something difficult, it might feel difficult to go and tell your parents because you don't want them to be worried about you. You might not want that or you might feel like, oh, it's going to blow into big proportion and they're not there yet. So that's often why they might turn to it. And so that's why those conversations are not like big life lesson or lecture, but really more approaching it with curiosity. Because when then it falls into like, this is bad and it shouldn't happen. If you add judgment on top of that, then I'm not even going to come and talk to you anymore about what they're doing with those platforms. And I think it's easy and I hear that often that it can be seen and even them like I was actually like with a youth group last week where we know that 50% of teen use it for emotional support. And I asked them, like, do you talk about that within your peers? And they're like, oh, no. Like, I would never say turn to an AI for emotional support like that. That feels like I would get judged for that. So even within themselves, that's not something they're willing to discuss.
And that makes sense. Like, it's a tough age being a teen. Like, there's so many questions. So to me, it's also really a question of how much they use it. But sometimes if we on top of that ostracize those type of use, it makes it worse. I remember a young user coming to me at the end of a conference where I had explained, like, why they will be attracted. And that's normal. Doesn't mean it's great, but it's normal. Like, it doesn't. It's not a default character about them. She was able to share with me that at some point she had spent 20 hours of her weekend talking to an AI companion and she realized it was too much and it had kind of ramped up, and so she stopped completely. But then she felt a lot of shame about having gotten to that point. And what she said was like, I was grieving that relationship. Even though it was an AI, even though I knew it was not human, There was things that system was the only one to know about me. And so I had to grieve that. And on top of that, I couldn't tell anyone around me how hard it was because of the shame attached to it. So it was kind of like a double pain there. And so that's why I think ostracizing and making it sound like it should not happen is risky. And sometimes we don't want to say it's fine when it happens. So it's really like kind of finding the balance of, okay, maybe it tells us a bit of what's going on for you. Where do you think we could find that in the real world? And kind of like redirecting slowly and without judgment, which I'm a parent, I get it. We get so scared for them that sometimes we approach those conversations with our big feelings as well, of what does it say about them? Are they doing okay? Are they going to be all right? And so I would say, like, if you had those conversations first, maybe take a bit, a bit of a breath and enter that conversation when you feel like you. You can actually be really open to what your kid has to say and have ongoing conversation about that as well, of, okay, you use it for homework when is it helpful, how you're using it?
When are you tempted to talk about other things? What are your friends doing? But really approach it with curiosity?
[00:29:06] Speaker A: Yeah, that's a great tip. Curiosity is always a win. I'm also thinking that it's probably also a mix between judgment but also vulnerability. Like at that age. And even as adults, right.
Putting ourselves in a vulnerable position, talking about feelings, it's really hard. So maybe parents are not judgmental, but teenagers and humans in general are scared or being vulnerable. So, yeah, the AI feels safe. It's never gonna judge you.
[00:29:36] Speaker B: And to your point, I think even just being able to kind of share our own vulnerability with our kids is often a good way of opening that. I'm like, you know what? When I was that age, like, I don't know, 15, I went through that. I'm pretty sure at that point I would have liked to have that, and I probably would have used it this way. Or even, like, when we use it as adults, like, oh, I'm realizing I'm using it a lot for work and maybe sometime instead of taking the time to think of the answer myself, I'm just, like, a bit lazy. And I go to an AI because it feels faster and, you know, like, just being able to also share what we do with it. And when we're like, oh, I use it for that. And it was amazing. And I used a lot of. Maybe it was not the best idea. I think is helpful because it's showing that it's not, like, all great or all bad, but it's really about what is the prompt? What are you looking for? How much do you use it to do what?
[00:30:31] Speaker A: Yeah, it's a learning process that we're doing together. And that's the way hard because we
[00:30:38] Speaker B: love to have the answers before we teach our kids. And right now we're like, you know what? You don't know? I don't know. Maybe sometimes our kids actually know more about it than we do and it's uncomfortable.
[00:30:48] Speaker A: Yeah, but vulnerability is. I understand why they go. I mean, I can empathize with that. But also, human relationships are not easy. And, you know, they probably need to know how to show themselves as vulnerable, but that's another topic.
[00:31:06] Speaker B: Yes, I do think it's also about kind of the reward you get of, like, when you're starting a tough conversation with a human being. Like, it can feel difficult, it can feel vulnerable and uncomfortable, but the how much sometimes it will reinforce a bond and a connection like that you can't really replicate either. So it's like sometimes you're going to do feel things that feel risky and uncomfortable, but when it works out, like, there's no replicating that and I think also like showing like, what's the strength of that human relationship and what is the part that you can't really replicate with an AI is also an inspiring talk to have of like, for that. Like, there's no, there's no comparison.
[00:31:53] Speaker A: Yeah, that's a great point. The bond after vulnerability, it's something you will never get with an AI. Yeah, well, I don't know. Right. We don't know where the world is going. But I'm hypothesizing that. Yeah, you want human, human bonding.
[00:32:06] Speaker B: Yes, because I feel like, because the AI will not be like, AI is never in a vulnerable position. Because there's nothing also greater than having someone come to you, be vulnerable with you and you being able to be a place where they feel safe to do that and you feel like you're supporting them. Sometimes it feels even better than receiving that. And it's never come to you with big problems that you only can solve and your friends will not do that, will not come with that vulnerability. If you never do that either, even that has to be reciprocal. So, yes, sometimes you're going to have to be vulnerable so that you get the feelings of being the person that helps someone through something. Like, those are moments of learning and also of resilience where you see someone struggle and overcome it or you struggled and you were able to figure it out on your own and to come out stronger. Those are the things we use throughout life when sometimes we're always going to have some difficulty, some challenge that arise. And just being able to look back on all the challenges we've overcome where we were like, it was hard, I didn't know if I could do it. And in the end I pulled through. Are the things that we use every day to get through other hurdles and those things that's where like sometimes it might feel more comfortable for them to turn to an AI for advice. But it's also taking away some of their agency and self worth and confidence in their ability to overcome things. So it might feel easy. It would be like, I don't know, like using a bike with small wheels. And then you never know, like, would I be able to do it without? Maybe, but I will never know because I never tried to do without. And so there is that part where you want them to struggle. Like not to a point that it's really bad, but friction, struggle is really Part of what they're supposed to learn within adolescence, because that's really the confidence building that they're doing for the rest of their adulthood.
[00:34:07] Speaker A: Yep. Yeah. Productive struggle, right? Yeah, yeah. The right amount of struggle. And when you are over it, it's when you actually learn and in this case, build those relationships and bond. Yeah, yeah, that. That is so true. Let's go into the green flags. Like, when you see an AI that does what you say. Okay, that's okay. You know, like, for parents, maybe 1, 2, 3 green flags.
[00:34:31] Speaker B: So I think it's even like we in the conversation of how do you use it? What are the green flags? It's helping to get organized, helping to brainstorm ideas, helping to get some kind of formative assessment. So not like, oh, like, it wrote the essay for me. But I wrote the essay and I was struggling with a part and I was asking, like, it felt like there's something missing. What should I read about?
Or tackling a new project and being like, I have to work on something.
What are like, the five resources I should have on top of mind around that? And then they do their work and they come back. Like, all of those use are not that different from what they were looking for on the Internet, but in the way that's more personalized. That's traitor. Where you don't have to go through many websites and not be sure of the sources and how much they ask. Actually, the sources. I was hearing a youth last week and I thought it was so interesting. He said, like, at the end of one of his work, one of his prompt is always like, within this work, which part is actually me and which part is actually you? Like, how much intellectual property do I have in this work? And I felt like that's such a great way of realizing. Yeah, I was like, oh, I will use it. Honestly, I'm gonna use yeah, because. And sometimes maybe you'll find that 90% isn't you, but maybe some stuff you want to have more power in that. And I think that's a great way even for kids to understand that. And one example I give a lot in talks that teens resonate with, which is a good one to use for parents, is they should always ask the question, am I running with AI or is AI running for me? And the way I explain that is, if you're training for, I don't know, your first race, you've never run before, and you want to do a race in two months and you want to run 5k and you go and you're fine and one day you don't want to go. You're like, I really don't want to go today. And in one case you call your friend and you're like, can you come run with me? Like, like, I'm not motivated, let's go together. Or you call them and you say, like, can you go run those 5k for me? While in one case you're still learning the muscles, you're still doing the efforts. It made it a bit easier, it helped you get motivated compared to your friend did it like, you didn't do anything. Like, your muscles aren't growing. It's really the same thing with the learning and the critical thinking. Is AI just kind of something that kind of helps you organize, challenge, grow, push you further, then that's great, like, use it for homework. But if you just write down the essay subject and AI does it and you just read it and change a few things so that it sounds human, it's doing the running for you. You're not growing your brain. And maybe sometimes you take the shortcut. But at least I think what's important is more that they know that they can really be real with themselves. You know what? I didn't do much of it versus like this. I spent more time on. But I learned stuff. And I think that's where the green flag for me is, like, how much they feel like they have ownership of the work that they do.
[00:37:37] Speaker A: That's a great one. And I'm noting that for myself too. Yes.
So if a parent discovers their child has been using AI companions for a while, what a good first response that you would say, like, I'm assuming, not
[00:37:52] Speaker B: panicking, but for so far, direct response
[00:37:56] Speaker A: from school for no. Like a parent just realizes that the child has been using AI and a companion for a while and they were not aware of it. What's a good first response? Like a very simple, take a deep breath.
[00:38:13] Speaker B: But really like, take a deep breath, like, and really be curious. And not, you know, like, now we are going to have 30 minutes of interrogation, but what are you doing with it? But really, like, okay, like, how long and where does it help you? Like, also, like, acknowledge that there might be things that they get from it if they come back. Maybe it's like they feel good after talking to it. Maybe they feel less lonely. Maybe they feel like they have better tools to go talk to their friends.
You know, it can be so many things that even within that we need to be kind of granular with our question. And then after that, like, are they things like where sometimes you use it and you're not sure and what are those? Because then that also helps you understand, like, do they see any risk if they do, which one are they identifying? And I mean they're smart, like often, like they, they have good answers and they realize when it's helpful or not. And if they don't, you also need to be aware of that, like how aware they are. But really if you discover that and your first reaction is panic, have the conversation the next day.
But then do have that conversation. I'm not saying postpone it forever, but do take a while to gather your thoughts and to make sure you can really come with curiosity first rather than your own worry. And we all do that. I have kids, I get it. And I know that I've had great conversation when I took the night off and I've had one that I went back to an exam, say like, sorry, I did not handle that that well. You know, like, let's, let's start over. Like, can I have a do over? Like, because I was not in the right place to be there to support and be curious. And even that like you mess up a conversation. Like, say you mess up, like actually model what real relationship are about and how you recover from those. Like that in and of itself is a good learning for them, especially at that age where they have to learn to take care of accountability for good and less good interaction.
[00:40:15] Speaker A: Yep, that's great. And yes, curiosity is so important and so hard sometimes when you're worried basically. Yeah.
So thank you so much. Let's start wrapping up with one hopeful takeaway. Like, what would a truly child centered AI companion product ecosystem look like if we got the design standards right?
[00:40:36] Speaker B: I think that's really the thing. It's really like designing it with kids in mind, which will be different by the way, for a 15 year old, a 5 year old and a 10 year old. They don't have the same need. A design where we don't just focus on the content, is the content charge appropriate? But is the interaction appropriate? And that doesn't mean like was it like acting like a bully? Like, of course, like that's an appropriate interaction. But even an interaction that feel really nice might not be appropriate for their needs and what they're supposed to explore at that age. So those are the things where we need more work to actually define what good is. Because we're starting to be like, okay, like that we know it's bad, but we need to move to okay, so then what is an AI that does well, do like, when does it provide answer or not? And I think that's a big part of the difficulty because you could be like, AI should not give clear answers and more like, ask question and prompt for reflection. And I agree, but sometime also, like, how do I spell that word? Like, you want the spelling. You don't want to have to think about where does it come from in Latin and what you know. Like, it can be interesting, but you can't have that for everything either. And that's why it's so difficult. It's because we're entering something that really requires human judgment, which AI doesn't. So we are trying to make rules, and models are improving fast. I think that's also one of the hope is that we think a lot about their reasoning as far as capabilities it unlocks and what it can do and how fast it can do it within the workforce. But that improvement in granularity within the reasoning is also something that might make it easier in the future to have models that are better adjusted, as long as we work with people who have a very good understanding of human development. And that's something for me that's hopeful is that I'm seeing companies really want to engage more with researchers to get this right. I think we are all learning through social media that we've run an entire social experiment on a full generation. And yes, some stuff were interesting, and yes, some stuff were really bad. And so the question now is that with generative AI we get a chance to learn from that, or as maybe if we had put more experts earlier in the conversation to really help define what's good.
What do we like about those social media? What do teens like about the social media? How do we integrate them much more in those conversations?
We might get it much better because often products are being developed and it's not even what teens want. So we are creating things that they didn't ask for. So even just having them as part of those conversations, helping us understand where they have needs and where they, like, leave us alone with that, is also really important. And I see more organizations doing this work, convening different actors and amplifying your voice and your needs.
[00:43:36] Speaker A: That's very helpful. And I agree there's more mindfulness about what the technology we're developing and how we're going to manage it and design it for children. We still have to get there. As you said, it's not easy.
[00:43:50] Speaker B: It definitely is, but you have to start somewhere.
[00:43:54] Speaker A: Exactly, Exactly. Exactly. Well, thank you so much, Mathilde. This was great. Very insightful. I hope that the audience got those very practical tips and that information about how different stages of development affect the way children perceive the interactions with AI. So thank you so much.
[00:44:11] Speaker B: Thank you so much for having me.
[00:44:15] Speaker A: That was Dr. Mathilde Sirioli, cognitive neuroscientist and chief scientific officer at Everyone AI. A few things that stuck with me from this conversation is first, friction is a feature of real friendship. AI skips the hard parts like disagreements and pushbacks, which is exactly how kids build social skills.
Second, the shame factor. If kids can't tell us they went too far, we've lost the conversation before it starts. We have to make it safe for them to come to us. It is hard because we also fear. But that leads me to the third one. Lead with curiosity, not panic. It's not easy, but ask. Ask them, what do you like about talking to it? And the first time you bring it up, maybe it doesn't go well, but go back and try again. That itself models real relationships. And one of my favorite reframes from this conversation is ask your kid, am I running with AI or is AI running for me? I'm going to start using that one myself. So you can find Mathilde's work and her team's research@everyone AI. We'll link it on the show notes along with resources from this episode. We are also running a short research survey on how parents are using AI with their children.
If you have a few minutes, please. Your answers really help. You'll find the link to the survey in the show notes. And also, if this episode helped and helped you see something more clearly, share it with another parent who might be struggling with this. That's how we build a village. Thank you for listening to Pixel Parenting. See you next time.
