2026 AI for Mental Health (AI4MH) Symposium: Keynote Panel

Our keynote panel um brings together leaders from across industry, academia, policy and lived experience to help frame the future of AI for mental

health. The discussion will focus on

health. The discussion will focus on responsible deployment and defining clear um object uh actionable priorities for the field. Our moderator for this

panel is Dr. Carolyn Rodriguez, who's professor of psychiatry and behavioral science sciences and an associate dean for academic affairs here at Stanford University in the school of medicine.

She's also co-director of our AI for mental health initiative. I'd like to welcome um up to the stage Dr. Rodriguez as well as the um folks who will be

joining her for this pan this uh keynote. So Dr. Adelli, um Mr. Brendan

keynote. So Dr. Adelli, um Mr. Brendan Staglin and also Dr. Val Wright.

Come on up.

All right. I'm very excited to kick off and uh this keynote panel. Um just to set the stage, we have three speakers.

Um and each is going to talk for about five minutes and then we're really excited to get your questions and comments. Um we're going to have a slide

comments. Um we're going to have a slide ready and there'll be a QR code that you can uh submit questions. Um so one of the things that is uh incredibly

heartbreaking and the World Health Organization has um published uh new updates is that there is a incredible shortage of mental health care workers

and um on median uh the the massive treatment gap is uh the median health care is 13 workers per 100,000 um and in

low and middle inome uh countries it's less than two per 100,000 and yet llm MS are used active users are in the over uh

500 million uh range and worldwide individuals um use by individuals is in the billions and so presents um an an exciting moment where technology is

available how can we use it responsibly and our speakers today are going to give us some nice examples and kick us off um from across academia foundations and

professional organizations. So Dr.

professional organizations. So Dr. Adelaide is a assistant professor of psychiatry and behavioral sciences and by courtesy u biomedical data science and computer science at Stanford. He

directs the Stanford translational artificial intelligence or stay lab and co-directs the Stanford AI for mental health initiative. And his research

health initiative. And his research bridges computer vision and computational neuroscience with an emphasis on connecting human motion and behavior to neural mechanisms that

regulate movement. Recently his group

regulate movement. Recently his group has developed ambient intelligence technologies to detect and delineate neuroscsychiatric and be behavioral systems at home and his team has also

developed multimodal large language models to evaluate and align clinical fidelity and safety and therapy chatbots. Uh Brandon Stangland, he is a

chatbots. Uh Brandon Stangland, he is a one mind co-founder um chief advocacy and engagement officer. He channels his deep experience in leadership, advocacy,

and his personal schizophrenia recovery to drive mental health innovation and community engagement to heal lives. With

years of successful advocacy for science-driven advances in early intervention for youth facing severe illness, he has emerged as a champion for the elevation of lived experience-based guidelines to shape the

future of mental health systems. His current work focuses on assisting employers to successfully hire, retain, and benefit from the unique skills of workers with lived mental health

experience. And finally, Vil Wright as

experience. And finally, Vil Wright as the inaugural senior director for healthcare innovation at the American Psychological Association, where she leads national efforts to advance the

responsible use of technology and data in mental health care. A licensed

psychologist and nationally recognized expert, Dr. Wright develops strategies to expand access to care, strengthen quality measurement, and modernize mental health delivery, health service

delivery. She's a trusted voice in

delivery. She's a trusted voice in federal health policy with research published and leading peer-reviewed journals and is a frequent contributor to the national media. She earned her

PhD from the University of Illinois at Urbana Champagne and is licensed to practice psychology in the District Columbia. Without further ado, Asan.

Columbia. Without further ado, Asan.

Thank you. And maybe

Oh yes. Um and so you can scan this QR code um and uh we'll see your questions.

Perfect. And we also have running mics in the u in the room as well. Okay. Uh

awesome. So um uh thank you everyone.

Thanks. Thanks Caroline for the introduction and also um thanks everyone for being here this early in the morning and maybe for many of us uh even before

finishing our first copy of a cup of coffee. Anyways, um so

coffee. Anyways, um so I'm sure everyone now knows that AI is everywhere, specifically LLMs, and we

are using those for our um every for many everyday uh purposes.

Even some uh people consult AI before or sometimes instead of seeing a doctor and um uh this is uh true for physical

health uh and even more so for mental health. Uh some of the some studies

health. Uh some of the some studies found therapy and companionship um is now one of the most pre prevalent use

cases of LLMs. And uh I want to highlight our very own HAI just published their 20 uh 26 AI index report

a couple of weeks ago in which um they explained experts predict 10% of use of US uh adults will use AI for

companionship at least once per day by 2027 and uh this is predicted to rise to

30% uh by 2030 2040 the uh top quartile of experts actually forecast that this is going to

be more than uh 40%. And again this is daily use every single day and and uh I think these numbers are staggering.

Another report hot off the press is the AI safety report by Stanford's very own um center for digital health. It was

published just last Friday, yesterday of the working days and uh in which um experts outlined what harms uh chatbots

cause, how they can be measured and how these um measurements can help improve safety. So AI is being used for

safety. So AI is being used for companionship and even entering therapy.

But clinical quality is more than sounding empathic. A model can sound

sounding empathic. A model can sound warm but still fail clinically. Good

therapy requires agenda setting uh guided uh discovery, risk recognition, escalation, and appropriate safety

boundaries. That's what motivated us to

boundaries. That's what motivated us to develop therapy gym, a new paradigm for building clinically grounded cognitive behavioral therapy chatbots. We first

built a multi-turn realistic benchmark annotated by expert CBT practitioners across nine clinical skill u domains and four safety measures. From the benchmark

we drive a a structured rubric for evaluating uh conversations against therapeutic skills and safety

standards and trained a judge model.

Then we uh fine-tuned open-source chatbots using reinfor reinforcement learning injecting therapy skills and um

safety awareness directly into the model and the resulting chatbot uh chatbot showed measurably improved clinical communication skills and safety

consensus uh behavior. Just the showcase of how these systems could actually work. Zooming a little bit out. Um,

work. Zooming a little bit out. Um,

while popular culture tends to equate AI with chatbots and LLMs, it's worth uh emphasizing that AI is a much broader

field. LLMs gave us linguistic

field. LLMs gave us linguistic intelligence.

But language is only one channel. Mental

health is expressed through voice, face, movement behavior sleep physiology environment, and brain activity to just

name a few. Multimodel AI will uh shift the lens from what people say to how people feel, behave, and change over

time. I believe the next frontier of AI

time. I believe the next frontier of AI will be multimodal behavioral special and and physical AI. We've been uh building systems in collaboration with

our very own Stanford uh adult hospital creating these physical intelligence and behavioral intelligence models and we've tested those for detecting

uh and um understanding neuroscychiatric symptoms in senior living facilities over the course of a month. So NPS or behavioral psychological changes such s

such as uh depression, anxiety and others that you see on the slide. Um

our system by by uh passively monitoring daily activities in participants homes.

The it it learns to identify subtle changes in behavior and physical function as well as movements of breakdown. The ultimate goal is to

breakdown. The ultimate goal is to enable early preventative interventions for individuals at risk of dementia and uh related conditions. This type of

technology creates many opportunities by enabling more objective measurements between uh visits, building personalized and adaptive interventions and providing

better support for clinicians and caregivers. Our early results do show

caregivers. Our early results do show high efficacy in detecting the moments of severity for each symptom automatically, passively and more uh

over extended periods of time. And with

that uh I just uh want to conclude by reiterating that I believe AI in mental health will be moving from chatbots to

multimodal uh clinical intelligence. And

uh when it comes to LLMs alone, I do I don't think it the question is can I sound like a therapist which most of the models are trying to do right now. The

real question is can I support safer, better and more equitable mental health care. Thank you

care. Thank you Brandon. Love to hear from you. Yeah.

Brandon. Love to hear from you. Yeah.

Yeah. Thank you Carolyn. Thank you Islam for kicking us off. Is my light working?

Can you hear me?

Yes.

Awesome.

Yes. Yeah. Good.

So, with recent advances in AI, the world of mental health and the world of communication is looking more and more like science fiction.

In fact, there are scientists now that are currently developing AI that can potentially even become conscious within a few years time. As I'm reading about Michael's most recent book, so truth can

be stranger than fiction, even science fiction. Who here is a science fiction

fiction. Who here is a science fiction fan?

All right, fair number of people. It's

okay. Yeah, it's totally acceptable. Um,

I'm a science fiction fan. Uh, who here remembers the great science fiction writer Isaac Asimov of the golden age of science fiction? Yeah, fantastic. Um,

science fiction? Yeah, fantastic. Um,

one of my favorite quotes from Isaac Aszimov, who is, by the way, my teenage hero. I wrote a term paper on him in

hero. I wrote a term paper on him in high school, is never let your sense of morals prevent you from doing what is right.

What does this mean? It seems like a challenge, like a riddle.

Well, I'll give you a hint. So, in this age of increasing complexity of our society and the mental health fallout that results, lives are at stake and technology can be a double-edged sword

in that battle.

So, I'm going to illustrate the point that I'll bring you to in terms of interpreting what that means uh with a personal story.

So I lived with schizophrenia for 36 years and uh one of the worst times in my experience with schizophrenia was right after my second episode of schizophrenia in 1996. I was about 26 at

the time.

I had been working as an engineer here in the Silicon Valley at that time for about four years building satellites and I had a relapse. I had gotten into graduate school. I knew that I had to

graduate school. I knew that I had to sleep less than nine hours per night to succeed in graduate school. So, I went off my schizophrenia medication and it was a devastating choice. I had

to resign from that job. I I I couldn't work and I had to go back into the psychiatric hospital and my morale was

crushed. My confidence was shot and my

crushed. My confidence was shot and my faith in myself was in in the tank and my social skills really suffered as a

result.

So, I had to move to San Francisco to live close to my psychiatrist to be able to see him twice a week for therapy sessions. And living alone in an

sessions. And living alone in an apartment in a big city, not knowing anybody in the in the environment there, not being able to make friends was really difficult. I was so lonely in

really difficult. I was so lonely in that time.

And I learned from that experience that human relationships, deep in person human connections are essential for mental well-being.

At that time, technology presented itself to me uh with two opportunities.

One of them was the advent of masculine multiplayer online role- playinging games, which can be a way to socialize, but and I had been a a gaming fan at the

time, but I was wary of them because the socialization that takes place online is fundamentally thinner and less rewarding than that takes place in person. so much more rich

than inerson interactions. So I wanted to develop deep inerson relationships and being sucked into the world of role playing games would not afford me that.

So I avoided that technology. The other

opportunity was cognitive remediation training which was a new innovation at the time for schizophrenia treatment.

I participated in an experimental study of cognitive remediation training through UCSF for about two months. And

in so doing, I used a computer program that helped remold my neural pathways in my brain to rebuild my social abilities.

Within six months of doing that training, I was back at work, back enjoying time with friends again. It was

a breakthrough. And the first novel that I had the concentration to read again was an Isaac Asimoth novel, by the way.

It was a great relief.

So technology is a double-edged sword.

How can we ensure that technologies like AI cut in the right direction?

So it's essential that we make these technologies in a way that of course as has been said ethical, safe, responsible, but we must not let overregulation

slow down the pace unnecessarily because lives are at stake. So how can we navigate this conundrum sa effectively?

Well, science is essential to do this and scientists all around us are doing this and you're doing this as scientists, but it's essential also to listen to people's lived experience of mental health conditions

and we're doing that at one mind right now my organization. So one mind is a program called the one mind accelerator.

There are two members of our accelerator team here today. Ally Alfred seated over there and thank you Alli. Sarah

Schneider who's also here the senior director uh seated somewhere here. Um

yeah, great to see you Sarah. Thank you.

Um but this program is helping entrepreneurs that develop technologies such as AI based technologies to help people with mental illnesses in a variety of ways. It's been a very successful program but we're guiding

them in part to our lived experience initiative which I lead at one mind connecting them with people with deep lived mental illness experience as well as deep experience in fields related to

mental health professionally.

So we're guiding them to product uh target their products and services at real human needs based on our perspective on the ground. So far, we've guided 14 companies through one of my

accelerator cohorts over the last three years to to do this. We're guiding five more to form their own lived experience councils coming this year to propagate our model outward. And thank you Alli

for helping us do that.

And one of I'll give some examples. So

last year, one of our cohort members with Slingshot AI, uh Derek Holler from Slingshot spoke here in April and we're very proud of what they're doing.

uh in part with the accelerator's guidance and our guidance uh to target their chatbots um focus in helping people toward developing agency and human

relationships not just relying on the AI. Uh another example is Biomia, a

AI. Uh another example is Biomia, a company we're helping former lived experience council this year developing medications for schizophrenia using AI to interpret organic molecule

structures. So we're excited to develop

structures. So we're excited to develop this work further. If you'd like to learn more about what we do and engage with the lived experience community to help develop your technologies, please contact me either during the symposium

or after. So, never let your sense of

or after. So, never let your sense of morals get in the way of doing what is right. Hope we have a better sense of

right. Hope we have a better sense of what that means and conviction to do what is right for people struggling with mental illnesses. Thank you.

mental illnesses. Thank you.

Well, thank you so much. there's a

moment where it's gonna happen.

Uh good morning everybody and thank you Brandon for for sharing and as I said in the prep call these are very difficult people to follow.

Um and so I will do my best but just to sort of recap I mean we've heard about the fact that we are experiencing a global mental health crisis. So it's not just here in the US it's across the world. We've also heard about the

world. We've also heard about the workforce shortage and if you really want to see the numbers and striking graphs go to HERSA. They actually have mapped out the workforce across all of

behavioral health that you can see that the demand for services and the supply of providers. The two lines never meet.

of providers. The two lines never meet.

We've also heard about really incredible advances from scientists like here at Stanford on the promise that AI can have to help address this crisis. Uh and I'm

here to tell you that we have not met that yet.

And I think that the opportunity that we all crave and hope that exists I worry may never happen in part because

while we don't want a regulatory system to impede innovation, we cannot have healthc care advances without a

modernized regulatory system that engenders trust in these products among providers, among patients and almost more importantly among payers. ers

because we truly want equity in our way of addressing the mental health crisis. We have to be able to pay

health crisis. We have to be able to pay for it and we can't pay for it if we don't have tools that payers will pay for and

that's the reality. So how do we get there? What do we do? I think we need to

there? What do we do? I think we need to partner, right? We need to partner

partner, right? We need to partner between payers and regulator regulators and legislators and scientists and

people like Brandon and myself to help create tools that will actually move the needle and have impact.

The other thing I worry about is, you know, this the challenge with the mental health crisis is so much bigger than AI.

Yet AI is taking up almost all the air in the room. And so what I worry is that we're not talking about all the other solutions that we need to implement in order to address the systemic problems.

And instead we're spending a lot of our time in some ways spinning our wheels talking about AI.

But here's the reality.

This is what we're talking about. In

fact, there's probably people in the room right now on their phones. There's

certain people online on their phones. I

was on a panel the other day and somebody was talking too long. I was a speaker. I was reaching for my phone.

speaker. I was reaching for my phone.

So what does it mean when our number one relationship is with a device?

When instead of purpose fitted tools being available and scalable, people are turning to general purpose LLMs to fill the gap in their emotional

needs. In fact, I was sitting on a plane

needs. In fact, I was sitting on a plane the other day and somebody was asking an LLM whether they should buy a house.

These are the things we used to talk to our friends about.

So, I think that yes, AI has promise.

Yes, we need to figure out how we partner to create better technology.

But I also think there's an existential question here as well.

How do we put the value ad back into humanto human relationships?

How do we mindfully think about our relationship with our devices, with AI?

And how do we really realign our experiences with our values in a way that reminds us that human

relationships are what make us us? We

absolutely do not exist without them because this is what I see right now.

So, I'm looking forward to having a discussion about it. Thanks.

All right. So, now we're going to open it up for questions and we'll put up the QR code again. But just to uh start us off um several world leaders have

recently and very prominently um uh talked about how human dignity can be preserved um with AI. Would love to hear

from from each of you how we can center um the human experience um in AI and um and and live up to that potential.

Whoever would like to wait and maybe start you.

Oh, sure. Sure. Um, yeah, I mean I I think the fact that you've seen, you know, for example, Poplio come out and talk about AI speaks to the global nature of this, right? Even though many of the frontier models are built here in

the US, it's clearly a larger global issue. Um, and

issue. Um, and I think that part of how we keep the human dignity or the humanness in it is not just

human in the loop on the user end, but it's really keeping humans throughout the process um throughout the entire developmental life cycle. And I think that in order for us to really

understand technology that's built by humans for humans, you have to have the experts on humans at the table. um and

that is behavioral health scientists right and having them not just again as sort of your end user but really throughout the entire process I meet with a lot of um you know startups in in

my role and you know often they are very well-intentioned in what they want to do and it's in part because everybody has a mental health story of some right either yourself or a family member or a

colleague um and so they want to build with that in mind and I asked them well who are your clinicalmemes and they say well we don't have Right? Um, we were going to figure that

Right? Um, we were going to figure that part out later.

There's no later. It has to be at the front end. It has to be at the start. It

front end. It has to be at the start. It

has to be your subject matter experts, your individual lived experiences. Has

to be not just your technologists.

Thank you, Veil. That's a great answer.

And I'll I'll add on to that by um building on what you said about involving live experience in that as well. So when you involve people with

well. So when you involve people with lived experience in co-developing solutions for techn technological mental health, you are bringing in insights that

clinicians do not generally see if they don't have their own lived experience.

So I'll give an example. So in one mind's accelerator program, one of our very first cohort members was a company called Motif Neuroch. It's a company developing minimally invasive neurosimulation devices that are

implantable on the skull to treat using like TMS for depression.

And to charge that small device implanted on your skull requires the use of a a battery and a charger for the battery. And the the charger is

battery. And the the charger is interfaces with the battery through a cap that you put on your head. And when

the company developing motif was developing it, they were very fortunate to have the guidance of a lived experience council which my co- uh which my friend and and uh fellow member of

one of my lived experience council member John Nelson uh was serving as a director of that lived experience council for Motif.

And because of his guidance and that of the other members of their council, they were able to understand that the cap should not be something worn at night because people are afraid that it'll be like pushed off when somebody's asleep and they turn over and they won't get

the charge they need for their neuros stimulation device, which would be a disaster for their mental well-being, but rather something they wear during the day. So, they decided to make the

the day. So, they decided to make the cat very stylish, something that people would be proud to wear and be seen with.

uh this diverted the the course of the development of that product because of a lived experience-based insight that could not have be seen any other way. So

involving people with lived experience as you mentioned will be a critical component of AI development just as it is for any mental health solution development.

Yeah, thank you so much there. Um there were several questions that you answered about that about lived experience and questions specifically for you on giving some examples. So thank you so much. Um,

some examples. So thank you so much. Um,

another question uh for for you Assan or others is um how how can models themselves be evaluated for safety?

Yeah, that's a that's a great question and um something I wanted to very briefly also add. I I

think it was a great question answer uh to the previous question but I think the the idea of codees is uh something that everybody's uh looking into right now in

the AI for mental health space and in the next uh series of talks that we have today. I think we have I've I've looked

today. I think we have I've I've looked at the agenda. We have very good u uh talks around the very topic when it

comes to safety and evaluating and um understanding the different measures concerns with respect to either um

chatbots or any of these AI models. One

of the things that always comes uh to mind is uh how can we annotate or label or uh

see if if actually one conversation is safe or not or or one of the items that is is coming uh up in the chat is a safe

uh direction instruction or not. So

there are so many different ways of doing this in the AI world.

Even at Stanford we have this AI safety center center for AI uh safety where they are developing formal methods for

evaluating uh AI but the thing is that when it comes to psychiatry and mental health

many of these um tools become uh not that useful because we don't have formal ways of measuring for um annotating

safety when it when it's a subjective measure. There are lots of uh challenges

measure. There are lots of uh challenges in building those former methods for evaluation. So we have to be creative.

evaluation. So we have to be creative.

We have to build uh things similar to what I uh present at therapy gym. Um

trying to annotate um data and see if we can um make sense of data with respect to those subjective measures or not. And

there was also a great discussion that Kian was talking about bringing all of these different stakeholders uh together to be able to define some of these uh

measures. So anyways, with respect to

measures. So anyways, with respect to safety, I think it does require uh bringing in individuals from both AI and

and clinical sites, defining the measures that we can use some of those more formal methods of evaluation.

But as I said, we have to be smart and and and and uh try to uh build systems that can also evaluate

um safety and other skill uh concerns with all of these uh subjectivity that we have in the measures.

Thank you. All right, another question from the audience. Uh AI should augment and not replace scare scarce clinicians, but the economic pressure pushes the other way. What's the actual guardrail

other way. What's the actual guardrail that keeps augmentation from quietly sliding into substitution, especially for underserved populations who already have the least access to a human

um so I would question that question um and state that there may not be a better

solution to reach people who have no access to a clinician at this point. Um,

but you can build in guard rails into AI based tools that can keep people safe.

So, I'll give an example from a paper I recently read. Um, AI based tools, chat

recently read. Um, AI based tools, chat bots for example, can sometimes cause people to dissociate, to lose touch with reality if people have deep conversations with them on an ongoing

basis for months or years.

And if people are already at risk for psychosis, then it's a it's a double risk at that point.

So if you can build into an AI based tool something like a psychiatric advanced directive, so like when you start to use

an AI based tool, have the person define what they want the tool to do if they begin to dissociate.

Perhaps contact a friend, contact a family member, contact a clinician or alert them they're beginning to dissociate and remind them to introspect

and compare how they're feeling and what they're how they're behaving with how they did before. Um, this can be a method to prevent people from the danger that could otherwise result with an AI

based tool. Another factor would be

based tool. Another factor would be similar to what I described earlier uh using the example of the ash ash chatbot where it doesn't prioritize engagement

with the app itself but rather forming those human relationships outside of use of the app and giving people the tools and skills to do that effectively.

So um that that those are two examples of what how to address that I think. Any

other thoughts?

Sure. Yeah. I I I used to say um AI is not going to replace therapists.

I now think that AI will replace therapists for those who are never going to seek out a therapist in the first place.

There will always be a room for therapists. And I think that the reality

therapists. And I think that the reality is we need to absolutely be rethinking what our system of care is. that we

can't just expect everybody to go to weekly 45 minute psychotherapy.

Um because while that's what everybody deserves, if that's what they want, that's not necessarily what everybody needs. And so we really need to be

needs. And so we really need to be thinking about how do we actually meet the needs of the individual who's either in front of us or again never going to walk through our door but is still suffering and needs something. Now

unfortunately I would argue the technology is not good enough yet and we have to make the technology better. Um, but I but I do think that

better. Um, but I but I do think that there is a role for it to to in some ways go beyond augmentation. And I'll

give an example, a personal example. I

learned um recently that my stepson who now just finished his third year of university, but in his first year was trying to find a therapist and couldn't find one.

I would have given anything for him to have a digital option that would have been successful and impactful and helped him because the reality is he just suffered on his own. And that's what

we're talking about here. And when I give talks to other psychologists who say, "Well, can't we just find them a person?" I said, "Don't you don't

person?" I said, "Don't you don't understand there aren't enough people."

So, we do have to figure out what are other solutions. And again, as I said at

other solutions. And again, as I said at the beginning, technology is not the only solution. It's one of many. And so

only solution. It's one of many. And so

my hope is that since we're having so many disc conversations around AI post pandemic and more focus on mental health than ever before, that what this really does is crack open a conversation about

the system itself and how we impact that.

Me, chime in. Thank you. Um that's a great point about how we need to reform the system and look at it in new ways.

Um we need to augment what the system can do including 3A tools but in other ways too. So for example peer support is

ways too. So for example peer support is an incredibly effective part of mental health care and even people who are not

trained peers can learn skills to listen effectively. Listening can be a really

effectively. Listening can be a really powerful healing tool if it is done in the right way. If it is done non-judgmentally, if it is done um supportively, if it's done

appreciatively and with referral to some sort form of mental health care when that's needed.

I had I can give an example of a time when a a peer listened to me and it redefined my concept of myself. I've

I've been struggling with a experience I had had in a creative writing class about a memoir that I wrote very vulnerable memoir and having it trashed

by my fellow students and that that really crushed my confidence, my ability to write and my interpretation of the experience that I wrote about which had been a very formative experience for me

when a a peer who was actually my mentor somebody else with schizophrenia his name was Robin Cunningham he passed away about four years ago amazing person um He listened to me telling that story to

him and in so doing he just listened and in so doing I was able to redefine the meaning of that experience and reclaim the value of that initial experience for

myself and rebuild the confidence that that experience initially engendered. So

um listening can be really powerful and peer peers can be an essential component of that. And just to add one quick great

of that. And just to add one quick great answers and very good examples just uh to add one more um of the examples of

use cases of AI in this space. I want to um highlight the fact that we could um use AI as uh some of my colleagues in the create center and also in the

computer science department are building and uh developing AI systems that can help clinicians by simulating for example patients uh instead of just

simulating the therapist. So there are so many different other use cases that could be uh considered here. I'll stop

here because there was a great discussion on this topic and we can probably uh wonderful and um and actually maybe um Esan you know how can we think about

AI in mental health beyond chat bots and LLM do you great yes examples thank you yeah so um I I I tried to show

a few slides on this topic because um um I needed those visuals showing how with

AI tools Um we could um go beyond language as I said um chatbots focusing on language bringing

in the linguistic intelligence that is a very good uh front door to to understanding people's state of the mind

but um what we argue is that mental health is is bigger than that. Again,

it's it's a lot of um human behavior, interaction with others, interaction with the environments that defines um uh

mental health. At home, for example, the

mental health. At home, for example, the moments of breakdown are are are those that are very good uh indicators and symptoms of uh mental health disorders,

which we call them neuroscychiatric symptoms. So other types of sensory data, sensory um uh tools like uh

variables or even uh ambient contactless sensors can help us understand the uh behavior, have better ways of measuring

behavior in a in a more objective manner. These are tools that could be

manner. These are tools that could be used over extended periods of time noticing changes in behavior and having

the information ready for the clinicians when uh it's the time of annual visits of individuals for example because in in annual visits we often are always given

these forms of uh have you had any changes in your behavior, mental status and so on. We always tend to select no all of us because even ourselves or

those who are living with us they don't very much um measure or understand the changes. These changes are so cra

changes. These changes are so cra gradual that are often ignored and sometimes even they are stagmatized stigmatized very much similar very much

in in like especially for older adults in the case of older adults because um that's associated with age and often not

reported. So these tools, these uh AI

reported. So these tools, these uh AI models paired with proper sensory uh pairing

could uh measure some of these symptoms uh passively. I'll stop there. There are

uh passively. I'll stop there. There are

so many other examples, but that's that's the part that was related to my own research.

Well, um okay. uh for any of you. What

kind of public health systematic educational approaches might our society lean into to ensure people develop healthy skills for interacting with chatbots?

Yeah, I'll start. Um so

I was just down in Santa Barbara yesterday doing a walk, a fundraising walk for my organization, One Mind. Uh I

met some amazing people down there. One

of them was a gentleman named Dino Ambrosie who's a professor at UC Berkeley who as an undergraduate in college had a life-defining experience

with phone addiction. He had just gone to a new um college away from his home and and a kind of a culture shock for him and he had difficulty making friends in the new environment. So he gradually

turned more and more to social media, to feeds um and and to search on his phone to find something to do and a sense of

companionship or engagement. And he

realized after about a year of constant phone use that he was addicted to it. So

he googled phone addiction. There was

nothing on it back in 2017, no studies.

So he took it upon himself to start to study the phenomenon and develop ways to address that for himself that can be translated for other people. uh since

then there have been studies that have been done on and he's leveraged those as well to develop a curriculum uh at UC Berkeley which he taught as an undergraduate through a program they have for undergraduates there called

decal to his fellow students and now he has a program there as a faculty member called project reboot where he works with young people trains them to go into

schools in the community and talk with young people young people young people excuse me about healthy technology use including social media media including

AI. They're developing the curriculum

AI. They're developing the curriculum around that now and to give them the healthy skills they need to maintain their relationships, their agency and so

forth as they engage with technology.

Another example is broadscale public health educational systems. So my niece and nephew go to a a school in in my community in Napa called it's a Waldorf

school. Who's heard of that model of

school. Who's heard of that model of education? The Waldorf model. Yeah.

education? The Waldorf model. Yeah.

Yeah. So they it they they um prevent the kids from using phones as they're growing. They they don't allow

they're growing. They they don't allow that anytime during the week, but um they teach them through experiential learning, through hands-on projects,

team- based with their peers. And from

this, they develop skills to interact with other people successfully and and healthfully. And and this can give them

healthfully. And and this can give them what they need as they start to interact with technology down the line. them like

well suited for that and ready. Um my

niece and nephew are are great kids.

They have great manners. When they

visited my wife and me about a month ago, my nephew was saying when when we served lunch, he was saying, "No, no, uh Tia and Nancy, that's my wife. You eat

first. You you you get the food first."

So he he was eight. So I was really impressed with that and the kind of education he's getting. Those are two examples.

Wonderful.

Yeah. At APA, we've been putting out some um health advisories geared towards uh mostly the public, but they also have recommendations for policy makers, for researchers, and others. Um our most

recent health advisory was on the use of generative AI and wellness chat bots to try to start this, you know, campaign about trying of helping individuals

understand not that chat bots are bad, right?

Because they're there and saying that they're bad and does not use them is not effective. Um but to help people

effective. Um but to help people understand what AI is good at and what AI is not great at. Um I I I do think the problem is much broader than that.

So we at APA we also did a study of university professors asking them if they thought it was their responsibility to help their students understand how to use AI and they universally said no that

it was not their job. Right? And so I think we need a mind shift, right? And

we need to start thinking about, you know, what is our responsibility to ensure that future generations know how to use this technology successfully given that it's not likely going away,

right? You do have still calls for sort

right? You do have still calls for sort of, you know, just it to disappear, but I don't think that's going to happen.

Um, so how do we think about our job and and you know what happens ethically if we don't help future generations think about how this technology could be

helpful or not helpful. So I I think there's a lot of work ahead of us to be done. Um, and I'm not sure who's going

done. Um, and I'm not sure who's going to step into that, but I I love those examples of of what you have. I think

that um that probably that grassroots approach is going to be how we start, but I don't think that can be I don't think that that's sufficient. I think we need something broader.

And just to double check, does anybody in the audience have any questions? Feel

free to raise your hand. And um as well, give folks a moment. Yeah. Go ahead.

Thank you for sharing and uh all the stories. Amazing. Uh question for Dr.

stories. Amazing. Uh question for Dr. Wright. You mentioned today's technology

Wright. You mentioned today's technology is not good enough. Curious, what do you see missing and what does it look like when it's good enough?

Thank you.

Question. Well, I I want to start by saying I'm not a questionist, so let's preface it with that. Um I I

mean I I I think the fact that um individuals turn to uh general purpose LLMs over purpose fitted options

suggests that something about the purpose fitted options whether it's cost or just lack of accessibility or something else means it's not meeting whatever need

they need. Right? Like they're they're

they need. Right? Like they're they're choosing to to use a no-named chat a general chatbot uh instead of ash right which is built by

clinicians for clinical purposes and yet people are not using it right so I I I think and even in those examples right where you

have some promising research right for ash for um theot for some others um I I just I I there's a reason people

aren't turning to them, right? And so

that to me suggests that the technology just isn't where we need it to be. Now,

that being said, the elephant in the room is therapists aren't as good as they need to be either, right? So

either, right? So that's sort of the conundrum. The the

real conundrum is quality at the end of the day.

The conversation we probably really need to be having is the quality conversation.

How do we have better quality technology? How do we have better

technology? How do we have better quality humanto human therapists?

And part of that comes down to evaluation and measurement in ways that we're not doing currently.

And so that's my longabout way to not answer your question.

I think you did a pretty good job.

Anybody else? Yeah, go ahead. in the

center there.

Thank you so much for sharing your experiences and I'm going to lead into this question with a self-disclosure.

Um, I have anxiety and I also build mental health uh for AI and so I'm really grateful for hearing from people who have experience lived experiences on this panel. So, one of the things I've

this panel. So, one of the things I've been struggling with the most is understanding how to handle AI refusal when I'm building my technology and making sure that I'm actually getting

people who refuse to use AI for a variety of reasons in my research. One

in five Americans use AI. Most of them don't, or at least they're not admitting on surveys that they're using it, and they're probably not counting the summary at the top of Google as using AI. And the thing I worry about is we've

AI. And the thing I worry about is we've got a lot of plausible reasons why people don't want to use AI. They

believe it supplants human relationships. They're worried about

relationships. They're worried about water use or data consumption or giving their information to a chatbot provider that they can't even control these. And

while I'm very positive about the ability for AI to transform mental health care, I also don't know myself how to integrate this concern that might be coming from actually a majority of

people who want to see more ethical and safe and like pro-social development of AI fundamentally before it can even touch mental health. And so I'm curious

for the whole panel, how do you think about working within people who are more skeptical of AI use? And how would you integrate those concerns into the

development of AI systems so we could actually reach that 80% of Americans who say that they don't use AI?

Not start.

I'll start it. Uh yeah, thank you. Great

question. So,

I'll go back to the answer I given before, but in a different way. So, when

you when you co-create products with people from a community that's going to use the product, they're more likely to trust it than if it's created by people in an ivory tower or in a corporation or it's perceived that it's created that

way. So if if companies developing AI

way. So if if companies developing AI like your startup or or even large AI developers um would partner with the

community that they aim to serve uh in very visible and and very real ways that will help build the trust and

people will see that um the the solutions are relevant for them and and have their best interest in mind. So

mind. So that is one way to start to address that that I think could could be effective.

I mean I think if we're talking about AI for health care, right? So not AI for just right for anything else but for healthare. I still think it honestly

healthare. I still think it honestly comes back to regulation. We would not just say here's a new SSRI. Let's test

it out in the public before actually having sure that it actually is effective. Right? Yet we've done that

effective. Right? Yet we've done that with AI. And so what we have now is is a

with AI. And so what we have now is is a culture where where AI companies have made real mistakes in real time and now we're trying to course correct.

Um and I think that's fundamentally the challenge here. Um and so I do think

challenge here. Um and so I do think that um our regulatory system is clearly flawed.

It needs to be modernized as I said at the beginning, but it is what we have.

And so I'd rather us I'd rather see us try to work with the system that we have in place and improve it than to continue to circumvent it the way we've been doing because I think that has caused a

lot of the lack of trust um and the concerns that people have.

All right. Thank you so much. Um I know we have a lot of questions. I didn't

even touch the ones on Slido. Um there's

going to be time for a lot of discussion in between panels, but what I'd like to do now is uh maybe just ask a theoretical question and have each person uh give their either uh you know answer the question or maybe a parting

thought that they would like people to get away from our session. Um so the question is if you had a magic wand and you could uh magically change one thing to improve AI for mental health, what

would it be or any parting thoughts?

That's on I would I would make it a little bit more objective so we can build algorithms. Good.

It's an easy question, right?

All right. Yeah. No, I I would I would make sure that it's co-developed with people with live experience and and guides people toward their peers to to help them uh through that those real

relationships.

Um I I I think similarly what I would want is for AI technologies to bring us back to humanity and to human

relationships um as opposed to bring them bring us away from them.

All right. Thank you all so much. Thank

you for your wonderful questions. Thank

you to our panelists.