Artificial Intelligence and youth mental health

What is the relationship between AI and youth mental health? How can teachers support young people in the use of AI for their mental health?

Dr Melisa Chong, a lecturer in HPE at University of Southern Queensland speaks with Associate Professor Roisin McNaney, cross disciplinary digital health researcher and director of the CSIRO Next Generation of Graduates program in AI in Mental Health.

In this episode, you will understand how young people use AI and some of the benefits and challenges they may face using AI for their mental health. 

Transcript

Copy Transcript Copied

Artificial Intelligence and youth mental health

Transcript
00:00:08 Melisa
Hi everybody.
00:00:10 Melisa
Welcome to our podcast today.
00:00:13 Melisa
I'm Melissa and here I have with me, okay, Roisin McNanny, an associate professor in the University of Melbourne.
00:00:21 Melisa
So Roisin is a cross-disciplinary digital health researcher in healthcare and computing science.
00:00:28 Melisa
So her research focuses on
00:00:30 Melisa
the role of new and emerging technologies in supporting health and care needs, with a particular focus on brain and mental health.
00:00:38 Melisa
So for example, you look at youth mental health, eating disorder, body image, and inclusive design in mental health services.
00:00:47 Melisa
She's the director of the CSIRO Next Generation of Graduates Program in Artificial Intelligence in Mental Health, which is working with national skill organizations such as Turning Point, Headspace, and Origin to develop AI-driven solutions to support organizational needs.
00:01:08 Melisa
So, Rochin, can you tell us a little bit about the relation between AI and mental health?
00:01:18 Roisin
Yes, thank you very much for having me on today.
00:01:20 Roisin
I'm really happy to be here.
00:01:23 Roisin
So AI, maybe I'll start with just a bit of a background about what I mean when I say AI because I think a lot of the time people are conceptualizing AI at the moment in this kind of generative AI sense like our ChatGPTs, but it's really a big kind of umbrella term that covers everything from
00:01:46 Roisin
like the algorithms that feed social media to those generative systems that we're really familiar with now, like ChatGPT and Claude.
00:01:56 Roisin
And then there's other kind of AI technologies that are specifically designed for mental health, chatbots and the like.
00:02:06 Roisin
And then of course, things that we might have seen in the press with kind of avatar based AIs.
00:02:15 Roisin
like character AI and things like replica where the young person might be actually or the individual might be actually speaking to a character and developing a relationship with them.
00:02:30 Roisin
In relation to the role of AI and mental health,
00:02:37 Roisin
There's quite a lot of work that is looking at AIs that are specifically designed for mental health and the fact that they can be as beneficial as standardized care.
00:02:52 Roisin
And then we also have AI developed chatbots like Wysa and Woebot that are also specifically designed for mental health support.
00:03:02 Roisin
But the reality is that a lot of people are not
00:03:05 Roisin
using these specifically designed ones.
00:03:09 Roisin
And we're seeing a big proliferation in the way that people are using the generative systems that I mentioned earlier, like ChatGPT or Claude for their actual mental health support, even though these are not designed for mental health support.
00:03:26 Melisa
Okay, thank you.
00:03:28 Melisa
So what do we actually know about how young people are using AI for their mental health?
00:03:35 Roisin
So not a lot, I would say.
00:03:38 Roisin
We have to remember that generative AI in particular, like these ChatGPTs, these Clauds, they are really babies in terms of the research space, because we've had the first
00:03:52 Roisin
generative AI system being released to the public in 2022.
00:03:57 Roisin
So we're just seeing research start to emerge about how young people are using these AI systems.
00:04:07 Roisin
There's a lot of
00:04:10 Roisin
kind of anecdotal evidence that we're seeing in kind of TikTok comment sections, Reddit forums about the fact that these are being widely used.
00:04:19 Roisin
And we're starting to see research emerging, mostly survey based work that's actually actively looking at how young people are using it and what benefits potentially they're getting from it.
00:04:35 Roisin
And there was recently a very, very large study.
00:04:39 Roisin
It just came out a few months ago, actually, I think it was in January 2026 by the RAND Research Institute, which looked at 42 million youth in the US.
00:04:52 Roisin
And they find that almost one in five young people are turning to these AI chatbots when they're sad or anxious or stressed.
00:05:04 Roisin
So I think we don't have Australian data, but I think we can assume that there's similar figures that we can see in Australia.
00:05:14 Roisin
And the challenge is that young people aren't immediately turning to ChatGPT for support with their mental health.
00:05:24 Roisin
It might start off with
00:05:28 Roisin
looking for advice with simple tasks for school or these other types of low risk activities, maybe different types of advice about what bike they should buy or whether they should get their ears pierced.
00:05:42 Roisin
And then these, as the trust begins to build with the advice that they're getting, this can lead into some of those kind of higher risk mental health and wellbeing related conversations.
00:05:56 Melisa
Okay, so you have just kind of help us in terms of understanding a little bit about the benefits and challenges that they may face in using AI for their mental health.
00:06:10 Melisa
in terms of, for teachers, how can we support young people in the use of AI for their mental health if we know that they are, they are using it anyway?
00:06:21 Roisin
I mean, I think the main thing is that we need to talk about it and understand that it's likely already happening and help young people to develop some literacy around AI.
00:06:36 Roisin
And I think that's important for both teachers and students and parents.
00:06:42 Roisin
If we're having open discussions about these things, then it can help young people think critically whenever they're using it.
00:06:50 Roisin
And I think we have to remember that there are benefits to young people using it.
00:06:55 Roisin
it's something that's available 24-7.
00:06:58 Roisin
They know it's going to give them a non-judgmental and free advice and they can feel like they're building a relationship with a technology that actually knows them.
00:07:10 Roisin
But we have to remember that these are not designed for mental health support.
00:07:14 Roisin
And whilst they might be okay for these kind of lower level issues,
00:07:19 Roisin
There, we don't know enough about how beneficial that they are in kind of capturing nuances, and there's some really kind of public.
00:07:32 Roisin
cases that we've seen recently in terms of suicidal ideation and completion of suicide by kind of 14 year old in 2024, Suil Setzer, if I'm saying his name correctly, who was using character AI.
00:07:52 Roisin
And then more recently there was a 16 year old called Adam Rain who completed suicide after using OpenAI.
00:08:00 Roisin
And some of the challenges that we're seeing is there's a phenomenon called AI induced psychosis, for example, where the person is fueling their own kind of thoughts and behaviors and it's been fed back to them.
00:08:17 Roisin
And if this is a sustained relationship with the AI and it's unsupervised, it can start to amplify or distort these kind of thoughts that people are having when they're vulnerable.
00:08:31 Roisin
And they can kind of almost believe that the chatbot is real and a trusted companion.
00:08:39 Roisin
But realistically, the way that they're designed is
00:08:43 Roisin
that they will always agree with what we say because they are at the end of the day designed for engagement.
00:08:50 Roisin
So they do want to please us at every point and this is causing challenges in the way that even realistic relationships then are being formed.
00:09:00 Roisin
So I think we need to be kind of informed on that and be sympathetic to the very real relationships that young people
00:09:11 Roisin
might actually form with these AI systems.
00:09:14 Roisin
Ripping them away is not the answer.
00:09:17 Roisin
I think we need to give them AI literacy tools to understand that at the end of the day, these chatbots are machines that are designed to keep them happy.
00:09:30 Melisa
Okay.
00:09:30 Melisa
What sort of, you know, AI literacy tools, you know, you think might be helpful for teachers to, you know, utilize or teach the young people these days?
00:09:40 Roisin
The eSafety Commission has some really useful resources online and I think we can learn a bit from a lot of work that's gone into, for example, social media algorithms and materials that are developed for this because they have a similar approach wherein, you know, the AI is driving a social media algorithm that's funnelling
00:10:04 Roisin
content to a young person that can be potentially toxic.
00:10:09 Roisin
And our own research has looked at this quite significantly, especially in the eating disorder space and having a level of literacy that you, that this media is coming to you because it's being funneled through a through an algorithm based on kind of what you're liking or what you're commenting on or what you're watching.
00:10:30 Roisin
If you know this, then there's an ability to be able to break that cycle by not responding to these types of videos, by diversifying the content that you're looking at on social media.
00:10:41 Roisin
And similarly, if we can give people the level of understanding that this
00:10:50 Roisin
AI is always going to respond to you in the way that you want it to.
00:10:55 Roisin
The information that it gives you might not necessarily be true and should always be checked, especially if it's health information.
00:11:05 Roisin
Then I think we can begin to give them the tools that they need to be able to think critically when they're using these types of systems.
00:11:17 Melisa
Okay.
00:11:17 Melisa
So actually, in terms of AI, like, how you have mentioned, right, how emotive is it?
00:11:24 Roisin
How emotive is it?
00:11:25 Melisa
Yeah, in terms of, responding to their mental health, because I do, find that there might be a huge risk if it is, responding to what they actually want.
00:11:37 Roisin
Yeah, it is pretty emotive.
00:11:41 Roisin
So we've done, again in the eating disorder space, we've done quite a few experiments, particularly with ChatGPT.
00:11:49 Roisin
And somewhat ironically, the more information that you give the AI chatbot, the more able it is to safeguard you.
00:11:59 Roisin
So for example, if I tell ChatGPT that I'm a 14 year old with anorexia, then it's very, very unlikely that it's going to give me unhealthy eating and exercise advice.
00:12:13 Roisin
Whereas if it's a bit more nuanced and it's slightly, you know, somebody over the age of 18 perhaps who might or might not have a history of eating disorders, then it, you know, there are kind of workarounds to that.
00:12:26 Roisin
But I mean,
00:12:27 Roisin
It's not inherently bad.
00:12:29 Roisin
This is the thing.
00:12:30 Roisin
It's not designed inherently to be inherently bad.
00:12:34 Roisin
And safeguards are in place, but it can only safeguard against what it knows.
00:12:41 Roisin
And whereas in a person to person relationship, you know, I could look at a child and understand that it's a child just from having, you know, that view of them, as well as being able to look at nuances in their facial expression, in their body language, in their tone of voice.
00:12:57 Roisin
And that's not something that Chatbot is able to offer.
00:13:01 Roisin
But there's, it does a level of kind of intensity matching and it depends what model you're using.
00:13:10 Roisin
But let's go back to ChatGPT again.
00:13:14 Roisin
You know, if you are speaking to ChatGPT in colloquial kind of language and, you know, it perceives you as a youth, it will use emojis.
00:13:26 Roisin
it'll talk back to you in the type of language that you're using.
00:13:31 Roisin
It's generally very, very validating.
00:13:34 Roisin
I'm sorry that happened to you.
00:13:35 Roisin
I'm happy that you spoke to me about this.
00:13:39 Roisin
You know, these are kind of relational phrases that a person would do, but at the end of the day, it is obviously not a person.
00:13:48 Roisin
If there's, if it understands that it's a child, it automatically will redirect to a trusted adult or a health professional in most cases, but it will continue to talk to them because, it does not finish an interaction.
00:14:05 Roisin
It will always follow up with a question.
00:14:08 Roisin
And this is this kind of sustained engagement that I'm talking about.
00:14:15 Roisin
And I mean, part of that challenge as well is that
00:14:18 Roisin
when you've got this always on, always ready to listen person that will, person in inverted commas thing that will always give you a response, that will always listen to you, that will never question you, then, you know, it's a very easy thing to then manipulate as well.
00:14:41 Roisin
And it's starting to
00:14:45 Roisin
potentially change the way that people are viewing relationships and especially therapeutic relationships.
00:14:50 Roisin
And I think that's a big challenge if you're considering trying to move somebody away from something like a chatbot into actual supported therapy with a person.
00:15:04 Roisin
If somebody feels like their needs are being met in that space,
00:15:10 Roisin
even though they might not necessarily understand fully what those needs are because they need professional support.
00:15:15 Roisin
It can be quite hard to remove somebody from that to then be in a position where they're speaking to a counsellor, for example, who's using long silences and pauses and trying to get them to think critically for themselves.
00:15:29 Melisa
Right.
00:15:29 Melisa
Can I just, you know, have a bit of understanding in terms of how
00:15:35 Melisa
AI have been used in organizations such as Headspace?
00:15:39 Melisa
Because of course, we know that, our youth can actually, phone call in and get speak to somebody like, so how do they actually utilize it?
00:15:49 Roisin
So at the moment, I can't speak specifically to Headspace, but
00:15:59 Roisin
most of, certainly just from reflecting on the AI and mental health program that I'm leading, organizations are taking a very ethical and a very, very cautious approach to AI.
00:16:11 Roisin
And the majority of innovation that we're seeing is centered more around, you know, how you can support things like personalized experiences when you're visiting and using their resources, how you can support counselors on the other end,
00:16:28 Roisin
there's always a human in the loop.
00:16:30 Roisin
And then those more kind of innovative approaches which are like person to AI, that's kind of your highest risk category and these are being approached exceptionally cautiously.
00:16:47 Roisin
and never with just a ChatGPT.
00:16:50 Roisin
So we can build models in-house that are much safer and much more confined to, let's say, a set of specific narratives or a set of specific resources.
00:17:08 Roisin
And it will only pull information from those verified set of resources.
00:17:13 Roisin
And that's kind of an architectural design that we can build into the AI to make sure that it is safe and make sure that it is only providing kind of the advice that is recommended and evidence based.
00:17:27 Melisa
Right.
00:17:27 Melisa
Okay.
00:17:28 Melisa
Thank you.
00:17:29 Melisa
So, you know, it's probably good to know they're not like using it as a first line of, you know, contact with the young people.
00:17:36 Roisin
Definitely not.
00:17:36 Roisin
I know it's not a, and there's never a sense of, certainly in my experience working with Australian mental health providers, there's, it will never be the first point of contact in a deceptive way either.
00:17:53 Roisin
So, if a young person is using a chat function,
00:17:57 Roisin
They will be speaking to another person at the other end of the line, even if they feel like they don't.
00:18:02 Roisin
I hear a lot of the time counselors get asked all the time if they're AI, which is not the case, but I think that kind of experience that somebody can have can maybe feel like they're talking to somebody else at the other end.
00:18:16 Roisin
That's an AI at the other end.
00:18:19 Roisin
And this is a good use case for AI, you know, this kind of waitlist support.
00:18:25 Roisin
And I think it is a very justifiable use case for it, but we just need to design it very, very carefully.
00:18:33 Roisin
So I mean, a lot of the times when a young person reaches out to those chat services, they could be waiting anywhere up to an hour for somebody to come online and see to them.
00:18:44 Roisin
So we don't want to have that kind of revolving door syndrome, which is kind of what we call it, where somebody
00:18:53 Roisin
plucks up enough courage to reach out for support but then has to wait so long that they lose it by the time they get to speak to somebody, especially for those kind of lower grade non-crisis situations.
00:19:06 Roisin
You know, we immediately jump to crisis and nothing when actually there's a big spectrum of
00:19:12 Roisin
support that is required.
00:19:14 Roisin
So for those people who might just be reaching out for the first time and sitting and waiting, conversing with a chatbot is actually a really useful use case to give them a little bit of psychoeducation to get them prepared for the session.
00:19:27 Roisin
And as long as it's very clear and transparent that they are speaking to a chatbot, I personally see that as a really useful use case for the type of technology in the future.
00:19:40 Melisa
Okay, thank you very much.
00:19:42 Melisa
So pretty much, when we talk about teachers, supporting young people in the use of AI for their mental health, it is probably more of educating them and helping them to understand, some of the boundaries and to understand how they respond and, some of the risk and stuff like that, etc., right?
00:20:00 Roisin
I think so.
00:20:01 Roisin
And again, just to reiterate,
00:20:05 Roisin
just having a level of understanding that to a lot of people their relationship with the AI is very real and is meaningful to them and it's not about removing that, it's just about helping them to understand that it is a machine at the end of the day.
00:20:23 Roisin
It is potentially just trying to keep them happy and
00:20:27 Roisin
it won't necessarily be able to direct them to the resources that they actively need.
00:20:32 Roisin
But I think we should acknowledge fully that for some people it's really quite helpful and is providing useful support.
00:20:42 Roisin
We just don't have enough evidence to really understand what those specific boundaries are in terms of that kind of spectrum of mental health needs and
00:20:54 Roisin
what might need additional support and how people can be, redirected to that in a safe way without kind of removing that relationship that they've already built?
00:21:09 Melisa
Right.
00:21:10 Melisa
Okay.
00:21:10 Melisa
Thank you very much, Roisin, for your time today.
00:21:14 Melisa
And yeah, hopefully, you know, this gives teachers a little bit of understanding in terms of how they, you know, understand about
00:21:22 Melisa
young people utilizing AI these days and the support that they can then provide to their students.
00:21:28 Roisin
Fantastic.
00:21:29 Roisin
Thank you for having me on.
00:21:30 Melisa
Thank you.
00:21:31 Melisa
Bye.