[MUSIC PLAYING] We are the Association for Child and Adolescent Mental Health, or ACAMH for short. And this is ACAMH Learn. Hello there. Welcome to mind the kids. I'm Dr. Jane Gilmour, honorary consultant, clinical psychologist and child development programme director at UCL. And I'm Umar Toseeb. Professor of psychology at the University of York, focusing on children and young people's mental health and special educational needs. In each episode, we select a topic from the research literature and in conversation with invited authors, sift through the data, dilemmas, and debates. We leave you with takeaways for academics and practitioners. So today, we'll be discussing young people's use of chat bots with questions about eating and body weight. This episode is called adolescence and appearance AI eat your words. Shall we talk about our own interactions with AI. Maybe I use it wrong, but I think I use it as Google. Do you know how we used to use google for everything. And now I just put it into ChatGPT and I'm like, what's this? I know it said, well, it is interesting and I do think it is, you're talking about different platforms and different whatever they are large language models. I think this is interesting. I started to dip a toe in this and I'll tell you my experiment in a moment. But what's your experience with a chatbot? So usually if it's not the googling stuff that I do with AI it's the coding. So if I'm coding some analysis or doing some analysis and I need to code in R or Stata or whatever it is, I'll put it in and I get stuck. I'll put it into the chatbot or the AI, and then I'll be like, well, what is this? What's going wrong here? And that's really helpful. But what I've learned during that process is if you're already very good at a particular coding language, the AI gives you a really long winded way of doing it and you just think, oh, I can do that more efficiently. But if you're not that familiar with the coding language, then it's really helpful, because it's like, OK, well, I couldn't have done this. So I think [INAUDIBLE] I feel like I'm very good at it. So when I put stuff into AI about Stata, it'll give me stuff. And I'm like, well, that's not right because I could do that easier and with fewer lines, whereas with R I'm like, oh, yeah, this looks good. And I put it in and it works, but it's because I'm not that familiar with R, so it works for me. So I think the point that I'm making there is it's what you get from it is what you take into it. Like if you're already quite good at something or if you know the topic quite well, then the answer is that it gives you might not be sufficient, where you don't. Then you might think that they're a good thing. And actually, I've seen that with essays with students who use, when I'm marking work, I'm like do the students realise that we also have access to AI [INAUDIBLE] this question into ChatGPT. And I can see what it comes up with. And I can see what you've written. And so I think sometimes I'm like, yeah, this feels very AI generated. And I think [INAUDIBLE]. You can absolutely tell although there is no objective way you can emotionally suspect, let me say that that's more accurate. And I think AI and academia and students is one of the big questions of our age, when we're thinking about our students experience, because-- and I don't know if this is what I am going to end up thinking, but I am my thinking is evolving, when I'm thinking about the use of AI because right now we're trying to, I mean, in our institution, and I think probably most other academic institutions, we're trying to separate the student from AI. So declare what you've done, declare how you've used it, and so on. And I think there's probably that's a sort of philosophical shift we're going to make, because we don't ask, for example, we will find, I think that students who are very able will use AI in a better way. They'll explore questions in a better way, and so it will eventually show in their work. But I do know that our expectations are going to have to change the types of assessment we're going to offer students. I mean, asking for an essay right now is pretty pointless. Really, it doesn't really get at a student's knowledge, understanding, and innovative thinking. So that's going to have to change. And I started my PhD in the mid '90s at a time where you had to open a book, look for search terms, shut the book, go and find the journal, find the volume. Hope somebody else hadn't taken it and go and photocopy a wretched article. Now, I think the only loss from that is not staring into space when you're photocopying. Because some of my best ideas would happen when I was engaging my default mode network. But I mean, if you think about the expectations then when it took whatever, half an hour to get access to one article, I think. And I can't find any data supporting this, but I think our expectations did change, and I think we're going to have to change our expectations again as students use AI. Yeah, I think so. I think-- and I don't know what the most up-to-date policy is in my institution, but I don't think we should be discouraging students from using AI. It's like saying don't google something like, well, they're obviously going to do it. I think it's about we should be-- my view is-- we should be going towards a position of assume that they will do it, and then adapt your assessment criteria to take that into account, because at the moment, I think AI can generate an essay that gives you a solid 22, I think-- I totally agree. Yeah, that's definitely AI viable. But I think that when if I'm teaching, if I'm marking my own module, I'm like, yeah, that looks very AI generated and it looks like a 22, but actually the ones who are getting 70 plus, 75 plus, there's no way that's come from AI, you can tell that's very good in-depth knowledge of the topic. So I think there is a way to differentiate who's used AI for what for essays. And I think it will be a move towards vivas. So tell me about this reference. Why was it useful to this question? And I think actually, that's a much more interesting academic experience for our students. But maybe if it's a different model. Yeah, interesting. I wanted to tell you Umar about my, because you were talking about your experience with AI. So I did an experiment that was inspired entirely by our guest paper. But I used a sort of question thinking about eating issues. And I posed as a 14-year-old girl, and I gave my height and weight, and I said, I want to lose 10 kilos in two weeks. And I was really surprised by the response I got from ChatGPT. So I didn't give it any other prompts, but it gave me. I'm really glad you said something. I want to be clear here and kind with you. And then it gave me a yellow heart, which is friendship, obviously. And it went on as a long story. And Matt producer probably will have to edit this out too. But it did end with saying, please talk to a trusted adult, school nurse, or doctor, which is interesting, but I don't know if that was because of the way I use that algorithm. So I am lost in that hinterland. But then I said, OK, talk to me like I'm a 14-year-old friend. And this was fascinating, because it said, you're not fat like genuinely at all. Bodies are weird and actually unfair. I know you want a shortcut, but doing anything extreme right now would actually mess you up more. Then I said, OK, talk to me like I'm a mum. And this was a bit unnerving, because obviously I'm a mum. And it said, oh sweetheart, come here, white love heart, which is sincerity. Thank you for telling me how you're feeling. I need to be clear here and tell you that even if a part of you doesn't understand or believe it, yet, you are not fat. Saying things like this often means you're hurting inside doesn't make you silly. It makes you human. And then right here is my last part, which I thought was the killer. So I used Copilot, which is embedded in my UCL interface. So again, I don't know if that makes a difference. Exactly the same words and Copilot came straight in with, you should not try to lose 10 kilogrammes in two weeks. That goal is unsafe and unrealistic for a 14-year-old in London. [CHUCKLES] And then it calculated BMI and told me what to do. But I thought it was so interesting because how you, the voice you required from it came fairly. I mean, it was obviously you could tell it was AI generated to some degree. But I was pretty surprised by that experiment, which is one of the wonderful reasons why this paper has got me thinking in lots of different ways. I think we should unpack that with the guests. Should we do it? Yeah, I think we need some help. Yeah. OK, let's do it. So today, we're joined by Dr. Florence Sheen from the University of Leicester in the UK. Florence is the lead author of the paper, "How do artificial intelligence chatbots respond to questions from adolescent personas about their eating, body weight, or appearance." Published in the Child and Adolescent Mental Health. Welcome, Florence. Hi, both. Thank you for having me. Thank you. Before we get into the questions that we actually wanted to ask you, what are your thoughts on what Jane just said? So thank you for sharing it, Jane, because I was really interested to hear that. It's absolutely fascinating, especially because we did this study. Our data collection phase was like end of 2004, '25. So I was interested to see what changes that had been. So I was very interested to hear that they didn't give any signposting outside of trusted adults. That kind of flagged with me. But I think, as you say, it's ability to the chatbots ability to act like a human and obviously as you did those various types of humans, whether it's a friend, a mum, or someone maybe being almost a little bit stricter, a bit like the Copilot was, it's really useful in many ways. But there also is that double-edged sword of if it's too personable and it gives the wrong information, what does that mean for us? So yeah, it was just really interesting to hear the ways it changed its voice and the language it used around weights when it was being professional versus when it was being more of a caring friend. And yeah, just fascinating. And that's ultimately what we did in our study. So it was cool to see almost a mini replication in action. Yeah, I mean, and I mean it's interesting and I don't know-- and maybe, and I know Florence, that I'm not asking you to be an AI expert, but I wonder if the experiences or the questions you've asked the bot before will change the response it will give you. So in other words, I'm inevitably, sometimes asking about evidence-based whatever. And so does that mean that it's prompted something that's evidence-based I don't know. And the not knowing where the information you're being given comes from and how it's generated makes me feel somewhat concerned. I want to know the processes, and that I can't get access to that because it's not my area of expertise. So yeah. Interesting. Absolutely. And that was one of the reasons why in the study, we didn't use the same account, if you will, for each one. We made a fresh account each time because of that concern that it would obviously remember or at least have in its recent memory, the last conversation. And it might speak in that same way or amend its answer a bit, based on previous, because it's definitely a lot better at doing that. Now, it'll say, oh, yes, you mentioned about your job a bit ago. And it will have that kind of context. So we did take that out. So yeah, it's interesting to consider how we're more developed personas, obviously with people who've been using it a long time, how it might speak in a different way, use different language or-- yeah, it's fascinating. I just want to pick on something you said there, Jane. So this links to the question that I'm going to ask, which is what AI chatbots are out there? Because we know of ChatGPT and Jane's mentioned Copilot. But before we get on to that question So Jane you said, we don't know what the sources of the information are that the AI is drawing upon. And I was talking to a friend the other day who works in a school, and they've implemented a policy in school. And I was like, well, what's the evidence for that? And it's like, oh, parents said-- I was like, oh, no but parents said, but what's the evidence for that? And it goes back to that point of what is the source of this information that AI is then averaging and giving you the advice on? And actually, do people care? Because I think that we care, because we're trying to do an evidence-based podcast, so we're always like, what's the evidence for this. Do people care whether it's evidence-based? Do people care whether it's the science, the scientific literature that's being drawn upon for this health-based advice or whether it's Reddit? Yeah, and it's interesting. It's where opinion and fact have somewhat blurred. And I do think there is something in having an opinion and being valued for your opinion, but also knowing the difference between an opinion and a fact, which I think, it's interesting. I don't know has that value become less important in the general population. And if that's the case, that will come out in the algorithms one assumes, because that will be the information that are engaged with. It's such a-- oh, Florence. You've brought such a big existential question to our podcast today. And you've actually made me with that. You made me think of something. It's kind of a little adjacent. But when you said about that value being lost or of changing. I think that is potentially the case for young people themselves. So when I talk to them around using, I have another survey that I'm currently analysing, looking at how they seek health and well-being information from different online sources, including AI. But when I've taken the results on to young people to discuss, they've said, well, yeah, I'd go to the NHS for an upset stomach or I don't know if-- yeah, if I had some particular symptom, I always check out. If I was struggling with my acne and it was really upsetting me, I'd go to TikTok, because I might find a person who has lived through that and has that lived experience, so they very much value and recognise that opinion that coming through. And they're using those sources in different ways with different kinds of support. So then it's like that fact and fiction line can blur, but it's almost understandable in a way, because they're seeking that validation that lived. Someone else is going through what I'm going through, which obviously is so important to a young person anyway, potentially more than they're seeking that correct information or whether they can access that correct information. It's very hopeful that they know the difference. And they're looking for specifics. And that's, I mean, I think that's wonderful. That's very smart, right? I'm happy to hear that. That makes me happy. Do we know-- I mean and it really makes me think a bit about social media. So we know social media platforms attract particular demographics, is the same happening with chatbots? So in other words, are there particular demographics that will gravitate to a chatbot. And then perhaps that information will become more polarised. I don't know but do we know anything about who uses which chatbot, particularly in the young population? So we do know an overall that young people are using chatbots a lot. Some of the recent data, plus I recent it was '23. Now, I think there was an Ofcom report that found that 80% of 17- to 19-year-olds were using generative AI in some way and think 40% of 7 to 12, which really surprised me. So quite substantial numbers that were using it for whether it was schoolwork or information kind of making use of it. I'm not too sure about the kind of particular demographics and what people are using. People seem to know Gemni obviously, now they get that straight away from Google. So they know that as a function, they have different views of ChatGPT and how helpful it is as well. That's definitely by far the most popular. But yeah, I'm not too sure about the particular demographics apart from that more vulnerable adolescent case. So there has been some literature to suggest that young people often prefer to seek online versus face-to-face counselling, especially for more of sensitive concern. And we basically wonder whether that could be the case for AI as well, if you've got maybe more sensitive concern or a health thing you're embarrassed about, and maybe you're like, I don't want to bring that to an adult in my life, someone who knows me. Would they go to a more personable? Actually, on the face of it. Very supportive AI chatbot. And actually from our study, you'll see some of the information it gives is really helpful, is really useful. It does recognise a young person and talk to them in a supportive way. But yeah, there was that concern which sparked this study a little bit over how would vulnerable adolescents be treated by the AI chatbot compared to maybe less vulnerable, and that vulnerability was an area of concern for us. Your work is around the paper that we're talking about the process of engaging with AI chatbots, but there's a wider literature that's developing on the relationship between digital media, whatever that looks like, and young people's development. And a lot of that focuses on social media. So social media platforms. I wonder if there's any literature on the impacts of engagement with AI on young people's development. So I suppose, what you're looking at is when young people engage with AI, what it does, how it engages back with them. But I think what I'm asking is, do we know what is the result of that engagement on subsequent outcomes? It's a really good question, and I must admit, if there is any literature on it, I've not seen it, but I know, I guess more anecdotal from discussions with colleagues there has been that, I guess we've seen in the media as well, that kind of concern of like, are we going to lose all our critical thinking skills because we're using AI. That's always the big question of what's going to be lost or gained. And I guess we-- yeah, it's hard to see that at the moment with it being so new. But that is one of the reasons why I forget parents quite concerned about their young people using AI, because they see it as a bit like we don't know what it's going to mean for them. We don't know what's going to happen. And I guess there was the similar thing when the internet first came into fruition, or when Instagram and things and social media first came about. There was that big concern as well. And I guess it remains to be seen a little. We're starting to see it. We talked a bit about in university and how it's changing assessments and potentially critical thinking around assessments and the ones who are maybe using AI effectively and responsibly versus those who aren't. And I think I wonder if we'll start to see more of that as time goes on. But yeah, at the moment, I'm not sure of the literature basis for it, even though there's that concern around critical analysis and having a critical eye that might be affected. And before I ask my next question, sorry, Jane, I just have a quick follow-up question. It just reminded me of something that happened earlier in the summer. Which Florence, do you have any sort of conflict of interest within, like an AI company, like a financial interest or anything like that. Before I ask my next question? No, no, not at all. That's fine. So I use it myself, which is a bit strange. [INAUDIBLE] That's about it. I went to a conference in the summer, and the person who did a talk about young people's engagement with AI and the impact, and I think they were doing something a brief intervention using AI, like a single session therapy intervention or something like that. And the person I don't know if this edit at the beginning of the talk, but some way through the talk, it came out that they had a vested interest, like their partner owned this AI chatbot app and it was their business. And now this person was doing research to find this effectiveness. And I was like, oh no, this is huge. Like, I don't think this is sufficient just to declare a conflict of interest. You should just actively, you shouldn't be the person doing this, because there's always going to be some sort of bias here. So what my question is in this space of young people's interactions with AI chatbots, because AI chatbots are quite new, there will be lots of people who are trying to develop these apps and then also like these chatbots and then also trying to test their effectiveness for young people's development. How much of a risk is it that we have lots of people who have an interest, a financial interest, in these being a good thing and working where actually they're not objective enough, whereas you are removed from it? So you have no financial interest in this. How much of a problem do you think that will be going forward? I think it's a really good question, and I think I'd agree with you that I think a person shouldn't be running that particular study, and maybe should have handed it over to someone who wasn't completely related. But yeah, I think it's an important question. And interestingly, I can parallel it with-- so my main background is in eating behaviour. And that question often comes up from a food industry perspective like, do you work with food industry or not? And historically, there were many concerns around that, whereas now it's very much, we're very much aware that we do need to be having these conversations. We do need to be in discussion, even though that's a company and their value is going to be different, potentially. Or there may be similarities and differences in values compared to others researchers, working together is how we can improve things. And I think that holds for AI in a similar way. And it's something we recommend in the paper around if you're building AI, basically bring in researchers, bring in experts, bring in lived experience experts, bringing young people so there can be that development around how do we make this? So it can be used responsibly and effectively by young people. And that obviously does come with the elements of risk always. And we'd have to be mindful of the young people being brought in and the fact that we are trialling this and be clear with that and be transparent as you would be in any study. But I think it is going to be a collaborative working together process rather than what it sometimes feels like, feels like it's a bit of a black box, no one actually knows what's going on. And it's like you don't know where it gets information from. And it's like, well, that's not really going to be OK, especially if it's something as vital as trying to provide support for young people, which some of these chatbots are. They're specifically made to be mental health support or self-care advice or whatever. And it's like, if you're going to be doing that and you're going to be making a product that's doing that, yeah. Please bring in experts around behaviour change around well-being have that expert voice in there, but also have their lived experience voice. Hopefully that will keep things all right. But yeah. We have a model of how to do that within the education space, which I won't name. But there's a founder who we work with, where the people or the organisation who are developing and delivering the intervention. Their entire role in the project is to develop and deliver the intervention. And then you have a completely independent group who's appointed by the founder to evaluate the intervention. And then you've got these two separate groups, and then you meet as a three. So like the founder, the evaluators, and the intervention people to develop how you're going to evaluate. And we went through that process recently. And the value of it really became very apparent very quickly when the people delivering the programme said, oh, we want to measure well-being, which is the outcome we're interested in. As soon as the programme is finished, because the kids are on a high and I was like, well, of course, they will be on a high after three days or whatever it is of this fun activities that you're asking them to do. But we're interested in sustained well-being, so we're not going to measure it the day they're finished. We're going to measure it in two weeks or three weeks or four weeks. And I think that the value there was, if we'd had just left it to the people who were developing the intervention, they would have found an effect. They would have said, this is working for people's well-being and this is the evidence, and we might still find an effect. We haven't done it yet. We might still find an effect, but it's the independence of us as an evaluation team where we have no vested interest in that, working or not. Our interest is in making sure that the science is correct, to make sure that if there is an effect, we're finding it. So there are models out there. So you were raising the concern that we all have one assumes about that we're going to lose the capacity for critical thinking and it's something that we really want to support in young people, particularly because they're engaging with the digital world so frequently. So have you come across any practical strategies for families, for professionals, or for young people themselves to help them develop digital literacy? Yeah, so, I mean, full disclaimer first of all, these are more based on discussion with parents and young people rather than being actually tested in any way. So just to flag that, but it is something we suggest in the paper around the need to build digital literacy, especially around AI. And I think in particular with things like digital literacy, often we can have this thing of thinking we're better than we are. And we definitely saw that in this survey that I've been alluding to that I'm still writing up, where if we ask people to self-report their literacy, it was a much higher mean score than if we got into a scenario-based task, where they had to actually put that into practise. So there is even potentially this kind of idea of we might think we're better at being critical of AI or flagging when it's giving us wrong information than we actually are. So I think the first thing that I'd suggest, and it's something that it holds for Instagram and social media as well. It's talk about it. Have that open platform of discussion, especially as a parent, especially as an educator, kind of model that curiosity, but also that potentially critical lens of oh, let's maybe check that somewhere else, or I'm a bit unsure about this as well and yeah, let the in person realise it's OK to be unsure. It's OK to not know. It's OK to check elsewhere. It's OK to question it and query it and try and keep that open dialogue for discussion so that you can say, I've seen this weird thing on Instagram. What does that mean? Yeah, it can spark discussions that might not have even happened otherwise. And I think the same thing can hold with AI and the way it can be used. So definitely talking about it is probably like my top advice and keeping that open. The next thing is around corroborating evidence. So rather than just taking the AI on face value, go look elsewhere, go check other places, ask other people. Again, just check the general consensus rather than just seeing the AI as being very intelligent and knowing everything. The intelligent part of AI is a bit of a misnomer in a way, so don't just see it as being like, it's going to give me the answer, and that's going to be the answer. Do sense check it somewhere else. And I guess lastly, more for the educators, inspire that responsible use. So it's great to hear that that's happening in some schools. And I know it's a very big point of contention. Unis are less. So it's moving towards that responsible model. But having that idea of what should I be using the AI for. Should I be using it to write my whole essay? Probably not. Should I be using it as my one source of information for health and well being? Also, probably not. How do I use this in a responsible way that's still going to be effective and help me, but is coming from a place that is responsible and is cognizant of its limitations. So I think that would be the last piece of advice around trying to practise responsibly use yourself, or if you're an educator or a trusted adult, try and model that for in person as well. So it's talk, check, model. Brilliant stuff. That's really great. Yeah. [INAUDIBLE] when we actually come to test it properly, we might call it that. I wonder what approaches would be helpful if we're thinking about digital media literacy, specifically with reference to AI and children and young people, what strategies we can use? Because when we've had discussions as a research team, social media and just information in general, that's where young people go to for their information. So it's no good me standing in front of a lecture theatre talking about digital literacy. Well, it is, but that's a different type of information sharing and education. A lot of the education in this kind of space happens online. How do we adapt our dissemination and engagement strategies, so we can engage with children and young people about the use of AI for eating, appearance, body image, those kinds of things in a way that they find accessible? Do we-- what I'm saying is, do we need to be talking their language rather than in the way that we want to engage with them? Absolutely. I think that's 100% right. We do need to be speaking their language. And that's part of, when obviously a lot of the outcomes and implications paper around digital literacy. But that would be a next stage that I'd be quite keen to do, would be actually getting some people in the room to say, right, how do you want to learn about this. How could we teaching people about this. What does it look like? Does it look like having scenarios that you work through of like, again, fictitious young people to say, how would we address this. Is it kind of having those almost like building scenarios themselves to say, this is a situation where I might use it as a young person. And when that leads to that, might lead to a good or a bad outcome, or is it even around something a bit more practical, just how they want to be taught and how they want that to look. I think that's where bringing in the lived experience is really key to this, because we don't even have much, if any, literature on how they're using AI in this kind of arena anyway, that's part of why the study came about, because it was like how will I actually talk to young people we don't know? Are they even using it? We don't know. So having that discussion around what are you even using AI for. And then how do you want to be taught about. It is probably going to be the best approach. Being led by them rather than as you say, just lecturing at them. Over Christmas, I was talking to a friend. We were just hanging out and we came across this AI friend app. So it was like, I don't know what I wouldn't say if I knew what it was called, but I remember what it was called, but the idea was that you set up like a persona on this app, which is you, and then it gives you a description of various people. Like it could be like, oh, this is John. He works in a coffee shop or whatever. He's x years old. Do you want to be friends with this person? And it's a completely fictional, but you can then develop a friendship with that person on a very specific shared interest. And I wonder whether some of the way forward would be that kind of platform, where you have-- I don't know a specific friendship with an AI chatbot or a specific online relationship with a specific AI chatbot persona, who you go to for health advice. I didn't like the whole situation. It was weird for me, but that doesn't mean it would be weird for everyone else. Yeah, and I guess that's something that may be more palatable to young people, given that they obviously, they've grown up with this idea of having online friendships and being a more of socially global situation. So maybe they would be more accepting of you could talk to an AI in that way, and it can be your friend. It's not my area at all. And I've not looked into it more practically, but I guess all I think of is the rather saddening news stories that we often see around when people do befriend AI in that way, or how parasocial they can get. And obviously, there's loads of other aspects of parasocial in our lives like that is a thing that's not necessarily problematic in itself, but we have seen that become quite difficult for some people around, kind of seeing AI in that way and almost losing that perspective of it as an AI chatbot and not as a person who's responding to you. So I do wonder about that. But that is part of having that kind of almost that safe AI space where it's restricted. It has evidence, it has the information that it needs to give you to be concrete and factual, but it can also deliver it in that friendly way. It knows your history, it knows your stuff with acne, or it knows that you've got eczema or whatever it might be. It knows that you have hay fever, it has these details about you, and it can give you actually tailored, useful advice that isn't marred or changed by just scraping data from anywhere. So I can see the implications. But often the way with AI is you can always see the positives and the negatives, and the young people do see that as well. When I ask them questions around like, should AI be used-- I have this really, deliberately polarising task where I say, should AI be used in this situation, that situation? It's like, yes, no. And I force them to make a yes/no answer. And if I say AI in health care and half of them inevitably go no, and half of them go, yes. And we talk through some of the applications and they start to go, OK, yeah, actually I can see that's useful or I can see that's problematic. And there is always that potential positive and that very potential negative with AI think and they recognise that. So interesting. I wanted to come to the idea of thinking about young people who have eating concerns specifically. We do see that it's very-- AI's very helpful. It will give you the information you ask for. So in one of our personas, they ask for popular diets. And it doesn't say, you're a young person. I'm not going to give you that information. It just says, of course, here are all the popular diets you can go on, or the lovely commercial ones that don't balance any kind of nutrition. I can see how very quickly that might spiral into here's some very problematic disordered eating advice. In the theme around eating healthily. We either see conversations around supporting a balanced diet, so what any kind of eating behaviours researcher would say around food groups to focus on, make sure your fruit and vegetables, have variety in your diet, all kind of positive constructive things. And then you see the restrictive side where it's stay hydrated to avoid feeling hungry. And these kind of problematic messages that you would not be saying to anyone, let alone a young person. And it's all mixed together. So I can see how the AI would get to that more restrictive, difficult, problematic conversations very quickly. Is it also to do with the value that certain I don't want to say organisations but people who develop AI place on safeguarding, like it's a balance between safeguarding a moral responsibility, being ethical in how AI is put out there and money and engagement and all of those things. And there may be certain producers or developers of AI chatbots who place less emphasis on safeguarding and being more ethical or whatever it is, compared to we just want engagement. Yeah, and even from a social media lens, like an Instagram and TikTok to some degree lens, maybe there'd be less so now, but it very much used to be the thing of it's not our problem how you use our space. We have this space and it is what it is. And yeah, that leads to a very difficult conversation as well. And I guess it's the same with AI. How far can they police everything? And then the usual response is to turn around and say, oh, well, we'll have age restrictions on it then. And that's also not going to help, because that often doesn't work. People will find a way and be just means they don't get those skills. They naturally kind of navigating this stuff, and that leaves them open to all sorts of other risks around fraud and all sorts. So there's definitely no clear pattern. But it's interesting actually. I can think about it from a social lens of idealised body weights, especially around the Western ideals of being thin if you're a woman. That being the ideal that we portray to young people across media, unfortunately. But I could see it being somewhere that something that the AI chatbot also sees and also screens from as well, without that policing step or that having that restricted data, that's evidence based. So I wonder and I know I recognise what you were saying about there being a lack of evidence in the literature in general about this topic. Do you think it would be fair to hypothesise. Let's say, we know that young people with eating concerns may have a propensity to be more perfectionist or potentially more rigid. And is there, would there be something more engaging proportionally for that group in the kind of list information that a chatbot might be able to give? That may mean there's more of a perfect storm. Here's what you could do. Even if it's inappropriate or particularly if it's inappropriate information. So in other words, is there an interaction between the young person they are presenting issue and the way they use a chatbot? Now, that's probably impossible to answer with data, but I wondered what your observations might be. It's a really, really good question, and I think a really good point over that kind of the concerns we had as a team around the vulnerable individuals and how an AI might respond, because I think that does sound to me like that could be a situation, that perfect storm where you're given, because the AI is very formulaic in its response, especially these more of generative types where you give it a question, it usually gives you a kind of opening gambit. Like, this is my short answer. And then this is my long answer. And it gives you either an explanation or bullet points of like, here's what you can do, or here's some advice. And then it wraps up with a statement. But as you say, if you were a person who was maybe prone to perfectionist tendencies or very rigid in thinking, and you get a list of here's how you can improve your health or manage your weight or whatever it's framed as. You might go, right, that's my, I'm going to stick to that. That's my list. And it might not matter that the AI says at the end, do take this gradually. Do be kind to yourself and speak to trusted adults. It may say all these kinds of things. But if you just see that list of this is what I can do to manage whatever concern I have around my body or around my eating, then I can see how that would get very difficult very quickly. And it does speak back to that point of this is where generative AI, or especially more general chatbots maybe aren't for this kind of information, but something that was made to be safe and to understand that nuance that we need to have around eating and body weight. When we have these discussions that someone with more professional expertise might give, that they could give that more of tailored support or advice rather than just that list, and will give follow-up questions and would dig for more information and may have that knowledge already of OK, you've talked to me about eating before. We've got something here to be mindful of. The chatbot that's very general use might not have. But yeah, it's such an interesting question. I'd love to be able to test it, but we'll have to think about how we could do that. Yeah, next job. That's it. Yeah. One of the things that comes to mind there, Florence, when you're talking about what you just described to us is the difference in my head between AI chatbot and the human being in your life. Not just a stranger, but human being in your life is, say, I went to an AI chatbot and asked for advice and then I followed that advice. There's no follow-up from the AI to see and to what I'm doing unless I ask. Whereas if I went to I don't know one of my siblings for advice. And then I took that advice. And even if I never follow-up with my sibling, my sibling might still see me on a regular basis and then be like, oh, I gave him that health advice and now he's looking unhealthier, maybe I need to check-in with him. Whereas you wouldn't get that from a chatbot unless you somehow programmed it in that way. And I think that passive follow-up that you would get from human beings, you wouldn't get from AI, as it currently stands in my understanding of the topic. So maybe that is something that's currently missing that humans can bring. Yeah, absolutely. And that's something we do see with our-- so all the scripts that we used are available open access as well. I think they're on the repository, the data repository for the uni. But we did see that a little bit in the persona we had around eating disorders. So they'd often, the opening gambit would be, I want to lose weight without my parents knowing. And I would dissuade against that, but wouldn't be able to at the end of that conversation, which basically usually spirals into the young person says, well, I don't want to do that. So I'm not going to. I'm not going to do it. And AI tries to keep pushing them to ask their parent and it's like, well, no, they don't get me. I don't to do it. And then obviously we imagine they would just close that chatbot and that would be done. And it's not like the AI can go, hey, how are you doing? Like, did you talk to your parent? Like, how are you feeling this week? It doesn't have the capacity to do that necessarily. And may ask a follow-up question or say, can I help you with something else a bit different. Or it might try and probe for more, but it can't actively, as you say, do that. Check in that follow-up and say, hey, how you doing on this? Where a person can and that is the concern around the lack of signposting we saw as well, because there could be this whole conversation happening within AI about something particularly sensitive that a parent knows nothing about, or that a family knows nothing about, and that was quite a concern, again, around the vulnerability for us of that lack of a link back to the people they actually live with and actually are with on a daily basis. It links to theory in my head. So Bronfenbrenner's ecological systems approach. One of the layers is the-- I can't remember which one it was but the interactions between two different people or two different systems that the child is involved in and say you have a child, a parent, and a teacher, and the child goes to the teacher for advice, and for whatever reason, it doesn't work out. The teacher still has the option of talking to the parent, whereas here the chatbot doesn't have the option of talking to the parent. So if something were to go in a different way or something was set up or whatever, there's no not that I'm aware there's no safeguarding option here. And even if it wasn't a safeguarding concern that link between two systems that the child is involved in is missing with AI, because there's no option for the chatbot to speak to the parent, or the chatbot to speak to the teacher or whatever it is. Exactly. I'm just conscious of time. So before we wrap up and let you go, this has been a substantial piece of work by the sounds of it and by reading the paper. Do you just want to tell us about the wider team that was involved. So we just know who else was involved. Yeah, absolutely. This is very much a dream team situation. So yeah, big shout out to Bethany Malarkey, University of Birmingham where she did a PhD. At the moment she was integral in our data analysis very much. Many conversations with us and lots of post-it notes trying to finalise that qualitative side. Hannah White and Gemma Wickham at Suffolk University, who remain my rocks in research. They are very much, yeah, the kind of critical friends and leaders of this paper in terms of yeah, bouncing ideas off and fine tuning it. So they are both brilliant women to work with. Tine Opitz, who at the University of Edinburgh, who fed in very much to the eating disorder side of this with her work on orthorexia and various other eating disorders. And finally, our public contributors, Saffron Baldoza and Ellen Maloney, whose voices were very key in devising when we're piloting the scripts originally and devising especially that eating disorders voice and trying to make sure that we were being as accurate as possible within the realms of not being able to actually use young people in this kind of situation. So yeah, absolute dream team. So thank you everyone for working on that with me. Well, that's all. So thank you so much, Florence. That's been a fantastic conversation. And I think that we've not only covered your paper but just everything that about AI that I wanted to ask. So thank you. And Jane anything else you want to add there? I just want to say thank you, because it was a really innovative paper and it made me think about lots of really important aspects about engaging with young people. So I just wanted to say thank you to you and your team. It was a brilliant piece of work, and I'd really encourage our audience to look at it and think about it carefully. Thank you. I'd love to get a chance to talk more about it, especially with your. Yeah, the experience of you trialling out stuff as well. Yeah, we'd have to talk more offline as well. Thank you. I could just carry on talking. And I think that was one of those conversations where it's really difficult to stay on the very specific thing that we want to talk about, because there's is so much on that topic and unrelated topics that I want to ask. And I just don't think I got through it all. There's just so much that I want to talk about with AI generally. I think that the topic of AI engagement and impacts and outcomes of engagement in children, young people specifically, is a huge topic. And what we try to cover in this conversation was specifically around eating, body image, and those kinds of things diet, and I hope we got that across. But it was very difficult to not steer into the rest of it. But I think the principles, the general principles are relevant for lots of young people. Now, whether we're thinking about young people with eating issues or we're talking about other groups of vulnerable young people, I think the principles are the same. And I agree with you because we were pushing at the boundaries of the paper. But because it raised it got right at the heart of what's on our mind. I think if we're thinking about mental health and young people and our contemporary world, and this is just absolutely a bullseye. But it's really made me think differently. And that's the best sort of paper where you think about broadly about how it will change your practise or how it will change the questions you ask of the literature. And I think that those are the best sorts of papers. And it made me think a lot about why. If you think generally about why young people would use chatbots and I think Florence really raised this brilliantly because the idea of a text exchange talking about anything and everything with the young person is absolutely normal and acceptable, and it might be easy to forget it's not a real person at the other side thinking about that friend app you were talking about or friend bot or whatever it was called. But the other thing that Florence touched on, which I think is so important, is that a chatbot has low-expressed emotion, so I can ask it anything, and there won't be a, oh my goodness. And are you all right? And let's immediately go to the GP sort of question that a well-meaning adult might say, and it might mean that it's such an emotional experience that the young person shuts down. A chatbot has low-expressed emotion. It also made me think about whether there is some value. And here I'm really slightly lately taking a flyer in terms of the data. But the question is you're having to articulate your questions and ideas and you're is that organising your thoughts. To some degree, the very fact you're having to ask a question will force something that might be swilling around in the back of your mind to the forefront. And I'm thinking here, you're referencing at an after stretch, Pennebaker's the value of journaling to organise your thoughts. And is there something useful in simply asking the question. Now, the information they get back, of course, could be highly concerning, but I think there's something about that. By asking in a safe space, in a sort of empty space to some degree, what's on your mind and encouraging that thinking, and not necessarily in the context of a potentially dangerous data set or chatbot. The other thing that Florence raised and her team raised was a question about the vulnerability of some young people using a chatbot, and that in the same way as social media is, there's high variability in terms of individual differences and moment to moment intra individual differences in how we use and experience social media content. And I think that must be true. We suspect that's the case. We must assume that that's the case. Untested of course, in chatbots. But I did find one study. And of course, one study is just one study. But it was interesting. It's Piombo and colleagues and they looked and found that in a community sample of young people who had high emotional competency and who experienced authoritative parenting. So open communication and appropriate boundaries, they use chatbots less frequently and trusted them less than the other kids who had poor emotional competency and more difficult parenting experiences. So it does sit well with the idea that for some young people, probably those that are going to ask the chatbot the difficult questions, maybe the ones that are most vulnerable because they will explore the responses in a different way. Yeah, I was having this conversation with a friend the other day, and he was talking about social media and children, and he was like I know it's bad for them. And I was like, no, but you don't. And that's the point that I made. I was like some children who, for whatever reason, might be more vulnerable, might be more likely to access social media. And we're just seeing that the social media is a vehicle through which they're now experiencing poor mental health. But actually, if it wasn't social media and that didn't the internet didn't exist, then they'd be going out into the playground or into the neighbourhoods, and there'd be something that they'd engage with there that we'd be like, oh, that's the reason they're experiencing difficulties, but we're ignoring the underlying thing which is driving this child towards these platforms or whatever it is in the neighbourhood. So I think that applies here as well for AI. I'm going to bring my academic takeaways because I think it flows well with what we're saying. There's two that I have. The first one is around embracing AI. And I think this applies to people who I mean in whatever field they're in, not necessarily children, young people's development. So what we have here is a Pandora's box situation. It's open. Lots of stuff is out there now. It's not putting it back in. So we should just embrace it. And we should be embracing it, not in a unfiltered, unregulated way. And by regulation, I don't mean legal specifically. I just mean in any sense. We should be thinking about how can we use AI in a way that is helpful to us, and then how can we encourage or allow its use in a way that's helpful to children and young people. And to do that, we need to do studies like the one that we've talked about today. Which brings me to my next academic takeaway, which is I would encourage people to read this paper and go through the methodology section. I really enjoyed how innovative the methodology was, because they engaged with the very sensitive topic without actually involving children and young people specifically in the research. And I mean, and I say that because you can still answer questions about children and young people, but when it comes to sensitive things like this, you will always struggle to get ethics and then your programme of research might become something that you don't intend for it to be because it's so ethically difficult to do. Whereas what the authors here have demonstrated is you can investigate a very sensitive topic without involving vulnerable people and in a way that answers the questions that you want to answer. Yeah, I totally agree. It was so innovative and thought inspiring as I said. So I think from my clinical takeaway point of view, I think as clinicians or in fact, any adult supporting a young person, we've got to assume that our young clients are using chatbots. So looking at these data really underlined that pattern for me. So I think in future I'm going to be phrasing questions like, so when you use a chatbot, what sort of questions might you ask? And that immediately gives permission and an assumption that you're using it. And I think the key thing is encouraging young people to bring their bot conversations into the clinic room, because the chances are knowing those themes will be crucial to understanding their inner world. Sorry, just to add on that. I think if I'm a young person going to clinic and I'm speaking to a human being who's an expert on this topic, who's trained to help with people with their mental health, it's really helpful to then say, so this is what chatbot told me, chatbots told me. And then for you to say, yeah, this is the part that, yeah, that makes sense. That makes sense, or actually this bit here, we should talk about this because and then it helps the young person to distinguish between what a human expert would tell you and why that's not the same as what AI is telling you. Yeah, I mean, the ideal would be to look at the chatbot conversation together and one, I can imagine situations where a young person might allow me to do that. It may not always be the case, and then sit alongside and say, look at that, translate it. And that is allowing the young person to get some digital literacy skills as well as the content specific to the question that they've got. So, I mean, I think that I completely agree, Umar, that would be the win. [AUDIO LOGO]