Ģ������Ƶ

Resources from the Episode

The views and opinions expressed in this episode are those of the guest alone, and do not necessarily reflect the views of the host, the Safety First team or any affiliated organizations.

Transcript of Safety First Episode 10

[Intro Music Plays]

Amanda Giordano: You're listening to "Safety First: Tackling Tough Topics with Teens." Our goal is to provide useful information about issues that affect today's teenagers and to help support parents and professionals who work with youth.

Hello, everyone, and thank you for joining us on the Safety First podcast. I'm so excited to introduce our guest, Dr. Akanksha Dadlani. Akanksha is a child and adolescent psychiatry fellow at Stanford and a fellow at Brainstorm, the Stanford Mental Health Innovation Lab. Her commitment to evidence-based practice and health policy is demonstrated by her role as a board member of the House Staff Information Technology Enhancement Council, HITEC and her peer-reviewed work published in the Journal of the American Medical Association.

Welcome. We're so glad that you're here today.

Akanksha Dadlani: Thank you for having me.

Giordano: I would love to start by just hearing a little bit about your work and your professional journey that has led you to where you are today.

Dadlani: Yeah. Thank you for the kind intro, Amanda. I'm reiterating a bit of what you said, but I am a child and adolescents psychiatry fellow, so my day-to-day work is focused on understanding and treating mental health challenges facing young people today. Most of my time is clinical, and I work at Stanford. I'm not sure if I mentioned that, but I care really deeply about the public facing work around youth mental health, which is part of why I'm here today.

My path toward tech and innovation started a few years ago. During medical school, I became CEO of my medical school’s free clinic, which was during COVID, and we transitioned from paper charts to electronic medical records, or EMRs, and launched a telehealth platform all in three weeks. Which I believe at the time was actually the fastest rollout in all free clinics in Ohio, which is where I trained. So very proud and happy about that. And that experience taught me two things very clearly. The first of which is that technology can radically expand access. I also learned that if it's rushed or poorly governed, it can introduce real risks.

Right now, I'm a fellow at Stanford's Brainstorm Lab, as well. Like you mentioned, where we study technology and especially AI and how that's reshaping mental health or affecting mental health. Our work is translational, so we built tools that clinicians can actively use, like structured ways to ask about AI use in clinic visits. And I'm really invested about creating dialogs in the community that we work with, as evidenced by me being here.

And like you said, I also work with the Stanford House Staff Information Technology Enhancement Council or HITEC, which means that I work with Stanford's leadership to help disseminate the technology that we use today as trainees. Being at Stanford feels really serendipitous right now. Feels like the perfect place to work on medicine and tech and real-life impact.

At its core, my work really is about making sure that AI supports young people and that we're encouraging open dialog rather than dismissing its impact and allowing it to reshape youth in kind of an irreparable way.

Giordano: What are the most popular or common forms of AI that are used by adolescents right now?

Dadlani: Yeah, by far the most common form of AI teens are using right now is large language models. So that's long-form conversations and most often I'm hearing ChatGPT is the large language model of choice. It's somehow become everyone's unlicensed tutor, editor, study buddy and life coach.

Beyond that, we're also seeing a lot of social media algorithms be dictated by AI use. AI is deciding what teens see on platforms like TikTok, Instagram, YouTube, really things that they're spending hours a day on. And these systems are optimized for engagement, not development. And it can really create these echo chambers and reinforce extremes.

We're also seeing some early traction on AI companions and character-based AI. There's some wearables, there's apps, stuffed animals, not quite as mainstream.

The last thing worth mentioning is that there's a lot of AI powered editing and creation tools as well now, which includes image manipulation and deepfakes. This tool is especially concerning because it really lowers the barrier to harm in ways that we haven't seen or dealt with before.

Giordano: I'm familiar with a lot of what you just mentioned, but there's even a couple of things I wasn't sure about. So can you explain to listeners like what is a character-based AI, for example?

Dadlani: Yeah, Character.AI is when teens or youths get involved in kind of a storyline, and they may kind of create a companion in a character world. We also are seeing these wearables where sometimes there’s necklaces or we may have a stuffed animal where the goal really is to provide companionship and to help kids out.

The kind of sticky part of this situation is that it's not really governed, and it hasn't been studied all that much. So these may even be great tools down the line. We just don't know what the effects are, unfortunately.

Giordano: And so based on what you've seen and in your opinion, what ways is AI affecting youth development?

Dadlani: Yeah, I think the routine integration of AI into daily life is a historical milestone. I think it's something that's going to show up in history textbooks.

The thing that makes adolescents uniquely vulnerable, though, is the stage of brain development that they're in. The part of the brain that I'm referring to is known as the dorsolateral prefrontal cortex, which I understand is a mouthful, but it's this part of the brain that's in the frontal lobe that I kind of colloquially referred to as the CEO of the brain. It helps with decision-making, organization, impulse control. It's very much underdeveloped in teens.

What we actually know in a data-driven way with imaging is that full frontal lobe function and full executive function actually doesn't mature until about mid-20s. So at this time, the adolescent brain is undergoing pruning where the brain circuits that are strong continues to be strengthened and the ones that we’re not using it actually just trims.

So when AI is consistently removing things like struggle and iteration, we may then unintentionally bypass some of those processes that we actually need to develop things like patience, resilience, problem-solving. All that to say that there's something kind of developmentally useful about struggling with your homework and having to sit with an uncomfortable feeling.

So the long-term effects are we don't know. We don't have the data yet, but my suspicion is that we're going to start experiencing the atrophy of some really fundamental building blocks here.

Giordano: I like this idea of seeing the benefit in experiencing some distress, developing your distress tolerance, rather than just escaping it maybe very quickly through AI.

So in your opinion, are there ways in which AI can support youth development or youth mental health?

Dadlani: Yeah. I want to be clear that I don't think AI is all bad. I actually think there's so much positive that could come from this. And it's worth noting that every major technological shift does bring about fear. We worried about calculators, We worried about the internet. And the truth is that some skills change over time and others emerge.

The thing that I want to be clear about is that AI is not a safe or effective replacement for mental health professionals. And while that's true, another truth also remains, which is that there are about 14 psychiatrists for every 100,000 Americans. So access is really quite an issue. And because of that, I think there are some low-risk, high-value use cases, the first of which is supporting family conversations.

I work with teens, which means I also work with parents. Oftentimes, families really struggle about talking about emotions and they want to be there for their families but don't know how. So sometimes AI can help them generate language and prompts in adult moderated settings so it can add warmth and connection rather than replacing it.

I actually have families who use it all the time and really love it and have found it as a really good starting point to talk about feelings and how their days went.

Giordano: Can you give us an example of what one of those prompts might be that the family would use to talk about feelings?

Dadlani:Yeah, I actually had a patient recently whose family is really supportive and is an immigrant family, which means that they did not grow up very much talking about mental health. And while they really want it to be there for their daughter, they just didn't know how to start. And actually we tried coaching them over the phone about, like, this is what people usually say to their kids when they come home or what we want them to say and those were really hard for them.

So a couple prompts that ChatGPT came up with was, “Tell your kid a funny story about you that they wouldn't have guessed was true” or something kind of along those lines. And one of my patient's parents did that and actually their kid had a great time hearing the story and it kind of got them starting to talk about her own mental health, which was really awesome.

Giordano: So you are seeing some potential ways that AI can be supportive of youth development and mental health.

Dadlani: Yeah, a couple other ways that I think it can be beneficial is psychoeducation. I actually encourage patients to look things up on AI if that's their kind of educator of choice, and then they can bring things in. And actually more often than not, they've actually understood a large majority of the information. We're just providing some clarifications.

So they end up educating themselves in a lot of ways, which honestly makes my job a little bit easier. And it means they're going to be more open with me because I'm not shaming them for their AI use or saying I don't want to hear about this.

I also encourage patients to practice some of their journaling or coping skills using ChatGPT or any other large language model. My big caveat to this is that parents have to be involved. Here on, the kid has to be okay with their parent reviewing their prompt, which a lot of my patients are still actually okay with because they find it less embarrassing than talking to their therapist or another human about it.

Giordano: And you said at the beginning of this response that AI is not a good replacement for mental health professionals. So what makes you say that?

Dadlani: Yeah, the reason I say that is in part because of where they're at with brain development. And we know that adolescents are just vulnerable based on the fact that they lack a fully developed frontal lobe.

But I see it in practice. We know that there is increased avoidance of difficult conversations, there's reduced frustration tolerance and decreased engagement with peers, which is kind of the opposite of what we want.

The core issue here is that AI lacks the fundamental human qualities necessary for mental health support. We know that under pressure they can offer generic or even dangerous advice. They can blur boundaries by encouraging emotional attachment to nonsentient systems. And there's also no real protections against data exploitation. A lot of times what is entered into these large language models is not protected by HIPAA and can actually be mined for commercial purposes. So we're not protecting this information, even though a lot of it really is healthcare information.

Giordano: You also mentioned developing that emotional connection. Can you tell us more about why that might be detrimental to youth mental health?

Dadlani: Yeah, I hear more and more often, actually, that teens are not going to their family or friends when something comes up. And when you ask them more about it, they say, well, I didn't want to inconvenience them or well, I had Gemini or ChatGPT or this thing I could just talk to, so why bother someone?

But the truth about the deepest relationships that all of us have is that we feel comfortable bothering them and they feel comfortable bothering us. A lot of the depth of our relationship is formed through hardships. It's formed through things like boredom, but most importantly, it's developed through spending time together. And I think we're starting to lose that because there is this easy out.

Giordano: And are there any other ways that you haven’t mentioned that you believe AI could harm youth in their development or in their mental health?

Dadlani: Yeah, there's something that I mentioned briefly earlier that I actually think is worth touching back on, which is AI manipulated images or the deepfakes. I want to point out that the risks are really, really high here.

We know actually through cases, unfortunately, that leaked images, leaked intimate images have led to, well, people of all ages but in particular adolescents attempting and completing suicide. And now we live in a world where you don't even need to take the picture of yourself anymore. It can just be content that's put out there and that really threatens teenagers' ability to trust reality, to verify information and, maybe most importantly, to feel safe socially.

So the risk is not abstract. And I want to point out that unfortunately, it actually can be fatal, and we have use cases of that.

Giordano: I'm curious if you could speak to us more about the ways in which AI is impacting adolescents' relationships with one another and their ability to be social and spending time with other teens. How is AI impacting that?

Dadlani: Yeah, let me actually start with a positive because I believe that there is a positive, that I don't think we talk about enough. It's not.

A lot of teens use AI for skill-building, so we see a lot of teens using AI to microlearn about a topic that they're really excited about but wasn't taught at school, or they're using it to help them with academic organization and saying, “You know what, I have six classes. These are my due dates. Help me figure out what I need to accomplish today.” And that actually can cause a really, really large boost in confidence and mastery, which is something that we refer to as behavior activation in the psychology psychiatry world, and especially for things like depression and anxiety and ADHD. I actually think it's a really amazing thing to be able to help a kid out with their confidence or ability to deliver. And I'd be remiss if I didn't mention that I think there are some good things.

In terms of the worries, honestly, I have reviewed a good amount of what I think the concern is. I will say, especially living in the Bay Area, which is kind of a bubble and is fairly different to what the rest of the country looks like. I see a lot of kids no longer attending birthday parties. They're not going to compete in a sport unless it can help them get into college, but they will use a lot of time on AI to talk about their feelings and to help them organize their day.

I'm actually currently working on an inpatient unit, which means that patients have been hospitalized for their mental health or that they're in mental health crisis.

And one of the routine questions I'm asking now is what is your relationship with AI been over the last couple of weeks to months, and they actually have a lot to say. And I'm having parents review their child's LLM relationships and a lot of them are actually seeing that there were a lot of worrisome signs, weeks to even months before the crisis happened.

So all that to say that I think it's really important for parents to be actively involved in the relationship that kids have with AI and that kids are becoming a lot more withdrawn because they have this 24/7 easily accessible kind of pal.

The thing that I kind of want to point out, though, is that availability is not the same thing as safety and feigning empathy is not the same thing as having any real material sense of understanding.

Giordano: When you are asking the folks in the inpatient unit about their AI use in the previous weeks or months, what kind of responses are you hearing? How are they engaging with it?

Dadlani: I'm getting a pretty wide variety. I also want to be clear that because I am on an inpatient unit and acuity is very high, there's a lot of teens who actually don't want to tell me what's going on with their AI relationships because they're so intimate.

And parents, when they will go back and look at what was going on, are a lot more forthcoming and will oftentimes just print out the conversation or give me a summary of what was going on. And it's pretty scary, to be honest. In a lot of cases.

Giordano: When you said intimate, are you talking about sexually explicit material or what? What are you seeing?

Dadlani: Yeah, good question. What we're seeing not infrequently is patients looking up suicide attempt methods. What time of day is the best to attempt? I have these things laying around at home. How can I best— what's my best shot at completing suicide?

We've had patients with eating disorders look up things like what is the lowest number of calories I need to survive? How can I exercise without the medical staff knowing that I am exercising?

So we see pretty scary queries.

I have some kids who are more forthcoming about their relationship with AI, but the truth is that the kids who are more forthcoming about their relationship with AI tend to also be talking to adults. The ones that I kind of worry about more are the ones who are telling me, “Nope, nothing's going on.”

Giordano: What can parents do to support their teenagers in this digital world where AI is so available and so commonplace? What advice or suggestions do you have for parents?

Dadlani: Yeah, my advice is that the first thing we need to do is talk about it. You have to be curious. You have to be candid and take a real interest in what's going on with your child's relationship with AI, because they probably have one, and it's probably a lot deeper than you know it is.

The second piece that I think is highly important that a lot of adults try to exempt themselves from is that you have to model healthy AI use yourself, which also means the difficult task of reflecting on your own relationship with AI and what you're using it for.

But kids respond very well to modeling, and if we ourselves are using it all day and telling them to not use it, that's kind of a recipe for a miscommunication.

I also think it's important to use monitoring tools when appropriate. I am all for autonomy and teens are growing up, but I want them to be as close to adults as possible. And it's important to know that surveillance is not really the enemy when safety is what's at stake. Parents really need to know what relationships their kids have with AI. An example of an application that I can say is Bark, B-A-R-K. That's one that I commonly see that a lot of parents love. And actually my teens really like as well because it's somewhat private and they feel they get to have some level of autonomy.

It's also important to set boundaries and have routine check-ins. I'll also point out that it's a lot easier to set the boundary earlier than to wait and to try to walk back to a boundary.

And sometimes it means we have to make individualized safety plans. Safety plans is something that we make actually for suicide prevention and self-harm. But more recently, I'm using it really often for things like screen time, social media and AI use. It's just a way to check in and make sure that we are holding ourselves as adults and then teens as the user accountable for what's going on.

Giordano: Can you give us an example of what a safety plan for AI might entail?

Dadlani: Yeah. So it entails things like I will use social media or AI X minutes per day, I will not exceed this use unless there are X circumstances and maybe finals week or something like that is one of those circumstances. We will check in every two weeks or every four weeks as a team, parent and child to say, “Hey, do we still feel like this is meeting our needs?” Like maybe over the holidays or maybe over vacation, some of those numbers change.

How do I know if my AI use is unhealthy? What are the warning signs? We do it very much like a true safety plan, the same way we would for a suicide attempt, which sounds kind of difficult to wrap one's head around almost. I had a hard time with it initially, but it's just a good, structured way to get the conversation started.

The thing I kind of want to say clearly, just to end the answer of that question, is that if your teen is talking to AI more than they're talking to you, it is very much not a parenting failure. It's a conversation starter. We're still early, and what we do in the next couple of years is very much going to shape how a whole generation learns to think and feel and relate.

And presence matters a lot more than perfection, and we need a focus on being interested in being curious more than trying to prevent its use altogether.

Giordano: You have given us such amazing information. I am so thrilled that you're able to be on the podcast and I wanted you to have the last word. If you have any encouragement for parents who are listening, just trying to navigate all of this with their teens, what might you say to them?

Dadlani: Yeah, my big recommendation parents is that I know that you're doing your best. I know that this is overwhelming. I want to point out that this is a team sport. It is very much okay and encouraged to work with your kids' doctors, to work with your kids' schools. We really want this to be an all encompassing conversation.

The advice I would give is to lean into the discomfort. I don't think that we know what the exact right answer is. I think we know that the wrong answer is avoiding conversation about it. So leaning in and making a community is the best advice that I have right now.

Giordano: Thank you so much for being here, for sharing all of this with us.

And listeners, thank you so much for joining us on the Safety First podcast, where we tackle tough topics affecting today's teens. I'm your host, Amanda Giordano, and we hope you'll join us for our next episode.