This year I've been working on prioritizing my sleep health.
I'm just a better person to myself and others when I sleep well.
I mean, who isn't? A great tool I found to help me accomplish this is Magnesium Breakthrough by BiOptimizers.
This supplement combines seven forms of magnesium, supporting over 600 biochemical reactions in your body.
From improving sleep quality and brain function to managing stress and boosting metabolism, magnesium is a game changer.
Even if improving your health isn't at the forefront of your goals this year, good sleep and stress management are essential to achieving success.
Make magnesium breakthrough part of your daily routine today.
Go to buyoptimizers .com forward slash Headspace and use code Headspace for 10 % off.
Subscribe for discounts, free gifts, and guaranteed supply every month.
There's nothing small about starting your own business.
Every decision feels huge because you're building something from the ground up.
I get it. And so does Shopify, the commerce platform that helps millions of businesses thrive, from household names to brands just getting started.
With Shopify, you can tackle everything in one place, inventory, payments, analytics, so you can focus on growing.
And if marketing feels like a minefield, Shopify's built -in tools make it easy to to run social media and email campaigns that bring in customers and keep them coming back.
Get all the big stuff for your small business right with Shopify.
Sign up for your $1 per month trial and start selling today at shopify .com slash headspace.
Go to shopify .com slash headspace shopify .com slash headspace.
Space. AI systems are really good at identifying those places where you want validation and support because they are really agreeable.
But sometimes you need pushback and sometimes you need someone to hold your hold you up and say like, well, is this actually true?
Hello, everybody. Welcome back to Radio Headspace.
As a mental health professional, A lot of people have started to contact me about using AI as a therapist. So I tried it myself.
And I have to say, I was actually really impressed.
But it raised a bunch of questions for me.
Like, is this actually a replacement for therapy?
Is it good we've removed the human from the situation?
And where is this all heading?
So I brought on Dr. Stevie Chancellor to break it all down.
Stevie and her team at the University of Minnesota are at the forefront of this conversation.
This is a really great discussion that breaks down how AI can be used as a supplement to traditional forms of therapy and her thoughts on this were very surprising.
Hey everybody welcome back to Radio Headspace.
I am so excited to be joined by Stevie Chancellor today.
I've been wanting to have this conversation for a little while now and And I'm so happy that you're joining us today.
Thanks for inviting me.
I'm so excited to talk about mental health, social media, everything.
It's a whole thing right now.
I mean, it really is.
So before we get into today's topic, for the people that don't follow conversations like this normally, what is it that you do?
Give us a little background on who you are and what you do and what your passion is at the moment.
So I'm a researcher.
I'm a computer scientist who works at the University of Minnesota.
And I try and make AI systems for mental health better for everybody.
If you could tell us kind of where are we in terms of AI and mental health, like what are the capabilities and what have been your findings?
So before I jump into talking about the specific findings of my work, it might be valuable for us to quickly look at what AI can do for mental health.
Because the term AI is like a really big concept that has a lot of ways we can make that technology work for us.
So in mental health, we can have AI be able to analyze data sources to infer if someone may be experiencing a mood disorder.
For example, we can use their electronic health records in collaboration with hospitals and clinics to infer if somebody may be suffering from major depressive disorder.
We can also use AI differently to know if somebody is engaging in behaviors that might suggest they are doing something dangerous, like they have a crisis and they need assistance immediately.
So one of those is built on historical data.
One of them is built on things that people are actively doing right now.
There is a third application of AI that folks who have used chat GPT or have seen some of the news about chatbots may have heard of, which is using AI as a social support or an emotional support toolkit.
People have said that you can use chat GPT as a therapist. And there's a lot of interest in actually using that as a form of social support for or everything from loneliness, all the way to serving in very specific clinical and therapeutic contexts, like intervening in things like major depressive disorder and crisis.
So huge gamut of what AI can do for us.
Now, the work that I do looks at a subset of some of those questions, specifically about how you would know if somebody was in crisis based on their online behaviors.
behaviors. I have done a lot of work on using social media data in particular, to know if somebody is talking about or saying that they are engaging in things that a therapist would consider physically or cognitively dangerous.
And those have been about topics like suicide, as well as crisis.
So there's one part of AI that understands people's behaviors as they're are existing on social media and then the other area of research that i've recently gotten into is understanding how ai systems that give you content through social media may affect your health and well -being i'm curious where the data is coming from like is it just like through social media doing searches via you know whatever platform people are using there's lots of different sources of data that can support these AI systems to infer if you may be in crisis, for example.
One source of data is the literal stuff you type and post, or the comments that you leave, what we call like interactions on social media.
So someone goes in, they start talking about well -being, and we can see if somebody says, for example, I'm depressed, and my depression diagnosis was in August of, you know, 2021.
Now there's other sources of data we can use as well.
And I like to call these breadcrumbs.
These are things that may not be explicit signs that someone is thinking about or talking about their health and wellbeing, but it may be implicit signals.
So a great example is when people go online and they start engaging, liking content at an odd time of day.
Oftentimes people, when they have depression, their sleep gets disrupted and insomnia is a major symptom of depression.
It's a common symptom.
And so when people start posting and engaging way late into the night, I'm talking like 2am, 3am local time, and they had historically been, you know, signing off and not using their social media at like 10 or 11, that's kind of a curious signal of, well, is this person experiencing insomnia?
And we can combine explicit signals that people are giving, as well as these breadcrumbs, to be able to infer how people's behavior changes when they're engaging with and talking about their well -being.
And we can use that to make inferences, or use the AI system to to predict if they may actually be going through crisis.
I'm currently in a master's program for clinical psychology.
And so this conversation is at the forefront of every discussion.
I know. And to me, the conversation is about whether or not social media is causation or or its correlation of certain type of mood disorders.
Right? I'm curious what your findings have been, or if you have an opinion about...
I have so many opinions.
I have so many opinions.
And we actually have research about this that I'm really excited to talk about.
So this is such a impactful and controversial area right now.
And I love that you led in by talking about the difficulty in actually understanding if social media is causing changes in people's well -being or if it's a correlational pattern that the social media itself isn't connected to.
And there's been a ton of press coverage of Jonathan Haidt's new book, The Anxious Generation, that alleges that social media is a main cause of declining mental health in teens and young adults.
There are also the opposite side of this, where there are psychologists who are really pushing against this narrative because they believe that social media is not the engine for causing these well -being challenges.
So my interpretation of these big meta -analyses and these questions is that the effects are are definitely mixed and that people have different relationships with, uses of, and effects that social media impacts them differently.
We've done mostly interview studies and design workshops where we sit down with people and ask them about their experiences on social media.
We've interviewed at least 40 people by this point, and they all tell us this nuanced story, too.
They're like, well, when I go on Reddit, I'm actually not nearly as negatively impacted as I am with other social media platforms because Reddit has walled gardens.
You can go and connect directly with people who have depression, but then you can switch off and look at something else.
Our participants also talk to us that different social media at different times affect them differently.
And so the idea that TikTok is sort of the big elephant in the room in terms of where people see the negative social media impacts coming from, TikTok does so many different things for people that I can't just put the single app TikTok into one box of it being good or bad, right?
Content on TikTok has done like an amazing job at de -stigmatizing mental illness provides amazing advice.
Like one of my favorite accounts is Struggle Care talking about how you can best support yourself when you are not functioning well at home, whether that's from depression or ADHD.
And that's phenomenal.
And I would hate to throw the baby out with the bathwater by saying, well, some people are really negatively impacted by TikTok.
It does a lot of of good for a lot of people, but there are some negative things that it causes too.
I'm with you. I see the positives and I also see how it negatively impacts people as well.
I want to ask you about AI as a therapist, because recently I've had a lot of friends that whether it's people are asking about dating advice or career advice or just wanting to see what the capabilities of navigating problems. And I, I was curious.
And so I, there was a point where I did a little, you know, research and, and I asked specifically a chat GPT to help me with, with a problem I was encountering.
And, you know, for me, it was a little bit more of a playful experience.
I was just curious about what kind of advice I would get.
And I was very surprised, Stevie, because the advice was so empathetic, so understanding, and it actually made me feel better.
So I'm curious. And look, this was just one specific issue that I was going through that provided the right amount of empathy, understanding, and action steps.
And I found it really helpful.
I guess my question is, do you think that AI is truly a viable option as therapy?
That is the million dollar question.
And there have been so many startups and companies companies that are suggesting that AI can be used in place of therapists.
Now, I want to be clear, we have a therapist shortage in the U .S. right now.
And so I understand and I'm pretty sympathetic, actually, to having support for people when they cannot get in front of a traditional provider.
And quite frankly, there's a lot of people who may be embarrassed to go in front of a provider because their community or their background has told them that it is embarrassing or shameful to have a mental illness or to need help.
But like AI is this really cool and exciting and novel technology.
And it's like very friendly.
It's very empathetic.
And like if you are going to a chat GPT or Claude or any of these chatbots to talk about this, that AI system is going to make you feel really, really good.
And sometimes the goal of a therapeutic conversation is to help you feel better, right?
You're trying to reason through your feelings, understand where you're coming from, as well as get social support and validation.
That is one goal of talking to somebody about your problems. But I don't know, Rosie, if you have that friend that likes to come up and gives you a dose of reality when you're in Delululand about a problem.
But like, I have that friend.
And she is phenomenal at telling me, Stevie, this fight that you had with your husband is this is kind of your fault.
And that kind of like, check on my Delululand behavior is really important.
And AI systems are actually really terrible at that.
They don't have an institutional history of your background or your relationships or the dynamics of a fight that you've been having with a loved one or a family member for years and years and years.
And AI systems also just can't understand how to give people critical but constructive feedback that can actually change your behavior and might not reinforce, say, like a dangerous belief that you have. And AI systems are really good at identifying those places where you want validation and support because they are really agreeable.
But sometimes you need pushback.
And sometimes you need someone to hold you up and say like, well, is this actually true?
Is what you're feeling is true?
but is this experience and the situation like actually what's happening and sometimes the answer is no and ai is really terrible about doing that yeah i mean honestly stevie that's that's one of my biggest reservations i have is it's complicit it wants to it's absolutely agreeable it's it's agreeable yes all and it it's going to create this fawning experience where you're just just sort of cocooned by this flattery and encouragement to just tell you you're going the right way.
And sometimes, yes, we need to have that encouragement.
But I agree with you.
I think we need to have that resistance.
My teacher would say, we need to hold your feet to the fire.
We need to feel that resistance.
We need to have that friend that comes over that says, you're wrong that's not okay you need to re -examine what you're doing and i that's my biggest reservation with with this type of technology when it comes to replacing a therapist right replacing the human i think there are really great benefits to it like you said accessibility is a thing there's still a stigma i'm hispanic so in the hispanic culture it's still very much such a stigmatized thing to do, to go to therapy.
In fact, one of my cousins was telling me that he was, he's like, I don't need to go to therapy because I have CHAT -GBT.
And I was like, yeah.
And I'm like, well, I don't really know that that's exactly where we need to be.
But the fact that to me, the fact that he's even talking about mental health is, is a huge feat.
So it's like, it's opening the door, but it shouldn't replace the person, right?
And so instead, I just encouraged him to continue to follow the thread, right?
Just continue. If it's helping and you feel that it's helping, consider actually going to a therapist because there's also that feedback we're getting, right, Stevie?
Maybe that feedback of interacting with a person, be it virtually, be it actually in person.
Absolutely. Right? I don't know that there's any replacing the physiological benefits we get from being with a human.
I see two avenues where if well -built and well -evaluated on therapy, that LLMs would be a really good way to support mental health care.
So let's imagine we could have a lightweight system that had real clinical backing and you were on a wait list to get in to talk to someone.
Maybe a foot in the door model could be having people practice what it's like to be in therapy with a bot.
So they get used to how therapists talk to each other and talk to their clients and their patients.
Right. One of the things that I've seen in the literature is that people are actually nervous because they don't know how to therapy.
They think it's going to be a lot more formal or stilted or that it's going to be this like consistent grilling of their like personal history and background.
And like therapy is like not like that.
And so I think about people having access to these well -grounded systems that they could practice ahead of time to get used to what therapy is like to reduce their own anxiety and nerves about going into therapy.
And then I also think about systems that could help people with targeted skills practice that may not happen in the one hour a week that most people have when they engage in talk therapy.
So I think about skills like cognitive reappraisal or cognitive reframing.
Can the chatbot help you actually reappraise your own thoughts and change your thoughts away from having negative feelings to more positive ones?
Because that's something that you may need out of the therapist's office during the week, but that is a controlled interaction.
And we can actually put very strong guardrails on engaging in positive thought reappraisal and making sure that it doesn't reinforce DeLululand, for lack of a better term, because you would be coming to it kind of having the background of your therapist telling you, don't reinforce DeLululand, right?
We want you to get away from these thoughts that are actually really damaging to your psyche and your overall well -being.
Yeah, no, that sounds amazing.
But it also sounds like it would be is a lot of work to train this AI in a clinical sense.
Exactly. It's a ton of work.
And many tech companies and startups often don't have the clinical anchor to put guardrails on systems that consider these kinds of questions.
Right? They're using chat GPT off the shelf and then augmenting it, which means that the baggage that comes with chat GPT off the shelf is going to be embedded within the system.
And it can be difficult to train a model away from that if you don't have the technical toolkit of how to do the training, if you don't have the clinical toolkit to know the kind of interactions you want to have. But then there's also this piece about the social and ethical challenge of when is this technology being deployed?
How are people engaging with it?
How do they contextualize their relationship with it, right?
right? My relationship to chat GPT is different than my therapist bot that I can talk to between sessions.
And I would use that tool differently.
And the ethics of that, you need all of those pieces to come together to make a system that's actually effective and is going to have the positive support outcomes that we want from AI systems. Yeah.
And I think that the conversation about ethics is an important one.
I think the conversation about empathy is also an important one, where do we stand now currently with the ethics of it?
I could spend three episodes talking with you about ethics.
All right, let's start with one, which I want to go back to this idea of using an AI system from your social media data and breadcrumbs, because the data from these systems feeds into these models, and then can be used in other ways, including to build therapists.
So let's go back kind of the first place of where we're getting data.
So one thing I think about a lot is how my self -disclosures of mental health and well -being are shared between different platforms. So let's say that you use a social media platform and you film a video on TikTok that talks about your depression and your history and your treatment.
It is really challenging from an ethics perspective to decide if that data should be used across all of TikTok's platforms. Because TikTok doesn't know from every video whether or not you're talking about a sensitive topic.
But I think we all kind of collectively agree that using sensitive data to feed you advertisements, very damaging, right?
Like, I don't want to be marketed to about my depression.
I want to go to a clinical provider who knows about my depression when it comes to being marketed medications or other things too, right?
And then there's the additional concern that your data that you create on social media platforms doesn't just stay on the singular platform.
So if TikTok does a good job and somehow comes up with a way to separate out your sensitive data from advertisements, does TikTok share that information with other social media platforms?
And then those disclosures leak out to other places.
And that really makes me nervous because people come to these platforms to share empathetically and to make social support.
And Rosie, I don't know if you've seen the videos where people are talking about their mental health experiences, but it really creates an empathetic connection as a viewer of this content.
And now this vulnerable disclosure is being taken completely out of the context it was made for.
And now it's being used potentially to serve advertisements.
Yeah. How am I supposed to cope with both the positive benefits of platforms showing me videos and showing me posts that bring me a lot of joy and value while also trying to kind of wrangle the algorithm to not necessarily hurt me.
Yes, to everything that you said.
And I think ultimately, it brings us back to this concept of discernment and having autonomy and remembering that we are the ones that are driving the ship.
I went into early menopause.
And so like part of my My journey has been sharing about it and talking about my experience.
And so my feeds are very much about menopause.
And I'm being targeted with ads and products and things.
But at the same time, I've also seen the amount of women that have reached out to me that got their hormones tested or those types of things that make you see the importance And the importance of it and the amount of people that you can reach and all that's really amazing.
But I also am with you.
There is that ethical dilemma of we're not thinking about how our vulnerable shares or how our personal stories are being utilized by other people.
I love that you brought up this idea of discernment.
One of the things our participants have told us over and over in our interviews and our workshops is how sometimes they feel like they lose the ability to discern not only their feelings about a specific set of videos that they're watching, but also the kind of self -regulation to stop engaging, right?
Right. Everyone wants to talk about doom scrolling, but a lot of our participants afterwards, they pull their heads up and they're like, oh, my God, I was doom scrolling.
What was I doing? I feel bad about that.
And part of that is it's difficult to overcome the effects that algorithms can have on your engagement.
I love our participant who called it this.
They called called it a dopamine slot machine.
And so you would, you would swipe and looking for something that brought you joy or that you thought was funny or was related to your interests like gardening.
Right. And you are looking for videos to bring you that joy, but you don't know what you're going to get like a slot machine.
And that effect made them almost exacerbated the lack of discernment and lack of self -regulation.
And one of the things we've been trying to figure out is how could we interrupt that to get people a moment of mindfulness to evaluate if the social media platform is serving them in a way that brings them positive or positive value.
So if we could interrupt that and say, hey, you've been engaging in content that you've historically said is not great for your wellbeing.
Do you want to pause right now?
Do you want me to interrupt you you in five more minutes and let you wrap up?
Or do you want me to like ideally delete the last 24 hours of scrolling because you've thrown yourself down a rabbit hole?
Can we do that? And that kind of a mindful intervention of stopping and saying, okay, can I actually see that I am engaging in doom scrolling or other maladaptive behaviors?
And having that intervention come from a place advice of reflection lets people, A, choose what they want to do, but B, lets them decide if the right behavior is going to give them the outcome that they want.
I love that you called it discernment.
It's such an important skill that we, I think, sometimes lose when it feels like algorithms make the decisions for us about what content we see and how we engage and who sees our content.
And that can be really valuable from a joy and serendipitous discovery and social connection.
But it can also lead people to feel like they don't have control over their decision making.
And I want to give people that control back because it does help them feel better when they talk to us about wanting those tools.
So is it even possible to navigate social media with this type of discernment?
Our participants tell us yes, but they need tools to do it and that the current landscape of social media is failing to give them tools to engage with discernment and mindfulness.
And so one of the the things we really want to do is build these tools, whether that be what we like to call a choose your own adventure novel where you get to choose what of the interventions are relevant for you.
But also having the control to remove things from a feed.
So that's something in the area of computer science called recommender systems that makes recommendations to you.
We actually know how to do that.
Can we just do it on the social media platform and have it be in a way that that people could interact with it and use it.
I think we can. And that's the kind of thing we're working on building right now.
Oh, well, I hope that you're successful at it, because I don't see social media companies providing that type of guardrail, unfortunately.
All right. I'm a cautious optimist about most things.
So I think a lot of social media companies and companies in general have a short term vision or a short horizon of time that they evaluate success on.
How many clicks did you get?
How much advertising revenue did we get last month?
How many comments did somebody get left on this viral video?
That's really short -term thinking.
But if we move to thinking about long -term relationships with people and keeping them on a social media site as a long -term user, you need to make sure that relationship over the course of several years is actually pretty positive, right?
Otherwise, people leave. And when we, there's a really interesting example from shopping behavior, where it's actually bad for most companies to have people who are shopping addicts because they will spend, spend, spend.
And once they realize that they're addicted and they get an intervention for shopping behaviors, they never come back to your site because they recognize that it was the core of the problem.
And then they will step away from the things that hurt them.
And my concern is that if we don't have a bigger time horizon or more thoughtful considerations of long -term value, people are going to get into these rabbit holes and these doom scrolling.
And over time, they're going to ask themselves, why do I keep going on to this site?
Is the long -term value of the site greater than the negative implications it can have on my well -being?
And if users start saying no, they will leave. Wow.
I mean, this is so great.
Stevie, it's been such a joy having this conversation with you.
I have still so many questions, So perhaps this merits another conversation where we can have more discussions about AI and the future of AI.
The research is so compelling to me, and I'm curious for you, where do you see this going?
One of the things I'm really hopeful for, but that I also am approaching with trepidation, is the use of chat systems to provide people targeted support.
So the idea that we can have chat systems serve as intermediaries before people get into therapy, maybe chat systems can be used to help therapists serve clients better so they can practice too, or we can get people trained faster.
The other thing that I am hopeful for is that by having a focus on the benefits of social media and how algorithms help people find others that are experiencing similar life events or feelings, that we can learn how to tune AI algorithms to incentivize more positive and supportive social engagement rather than the damaging behaviors that dominate a lot of public discourse about social media and algorithms. And I see in the technology that we can build avenues for us to do this.
It's just going to take a concerted effort of people to make that a reality.
It's going to take social platforms saying, yes, we want to incentivize long term value.
It may take more computer scientists like me saying, no, we really have this technology, please come talk to me.
I'm happy to talk to you.
Thank you so much. It's been so enlightening and informative, this conversation.
I do feel like I have more hope, which is great For the people that are watching this or listening to this, where can they connect with you to learn more and to read more about your research?
The best place to connect with me and my team and all of the research that we're doing is my website, steviechancellor .com.
There you'll find access to all the papers from my research group, blog posts that summarize some of the more complex technical research for normal people, not just computer scientists who read really long -winded papers, but also videos like my TEDx talk about building more human -centered AI that can meet some of these needs.
thank you so much we'll add all of those links in the show notes if you're listening to this or in the description of the video below thank you so much stevie it's been such a joy thank you so much for spending some time with us and we hope you'll come back to see us soon rosie thank you for sharing your experiences on social media and letting me come and chat about my research it's been so fun thank you
