How AI can help parents, with Happypillar’s Sam Gardner

0:00Starting a business comes with its share of ups and downs, which is why staying true to your vision is essential. a non-negotiable for Romeo and Milka Bregali, Capital One business customers and co-owners of Ra's plant-based restaurant in New York. Romeo and Milka took a leap of faith when starting their own restaurant, gutting an empty space and building it from the ground up. Every pipe, every wall, every detail. But building from scratch came with a heavy financial burden, which is when they turned to their Capital One business card. With the flexibility of the card's no preset spending limit, they were able to spend more and earn more rewards while bringing their vision to life.

0:36Today, Raz's success is proof that with passion and the right support, it's possible to make your dreams a reality. Learn more at CapitalOne.com slash business cards. Pioneers of AI is made possible with support from Inflection AI. It's not just enterprise AI. It's your enterprise AI.

1:02It was December of 2021. I was really stressed out. So Winston is my firstborn. He was four then, and he was losing it all the time. Every food I offered was wrong. Every toy was frustrating. Everything made him mad. His whining would turn to tantrums, and I wasn't handling it very well. That's Sam Altman Gardner, the co-founder and CEO of Happy Pillar, an AI-based parenting app that's focused on the mental health of young children. I was erupting right back at him. We're screaming at each other. It's like we're both toddlers. And every night I would think to myself, wow, I really, I messed this kid up.

1:50He used to be so good and sweet and I'm a bad mom. I've ruined him. The experience that Sam is sharing is something that all parents can relate to. As a mom myself, helping your child work through difficult behaviors and mental health challenges can be quite daunting. But luckily, my pediatrician recommended therapy, even for a kid that young, four. And it was really effective. But most parents aren't so lucky. In the United States, specialized mental health services for children can mean mountains of medical bills. There's also a critical shortage of mental health professionals, and that can mean months of delayed treatment.

2:34I also noticed how a therapist analyzed how I was speaking to my child was really similar to how language AI analyzes how humans speak. Exactly. The AI can analyze human speech similar to how a human therapist would. So what if AI could be a mental health resource for parents? Could this allow parents access to affordable mental health care right at their fingertips? Today, Sam Altman-Gardner and I are talking about how AI can become a bridge in our health care system. And how this cutting-edge technology can democratize access to mental health services. I'm Rana El-Khalyubi, and this is Pioneers of AI.

3:24A podcast taking you behind the scenes of the AI revolution.

3:42I've known Sam for years, and she was early in seeing the power of AI to transform a key part of life for many people. Parenting. I believed in her work so much, I became an early investor in Happy Pillar. The name is a reference to caterpillars, by the way, a favorite motif of childhood, and of course, the precursor to butterflies. And here's a fun fact about Sam. Her given name is actually Sam Altman, same as the founder of OpenAI, whose name you see in so many headlines today. Okay, let's get going with her. Hi, Sam. Thanks so much for joining us today. Thanks for having me. So I have to start with this question because I bet it's on everybody's mind.

4:28What does it feel like to have your name be Sam Altman? It is the weirdest sensation almost all the time. Every time something happens with other Sam Altman, I'll say in my mind, I'm original Sam Altman, even though I think he is definitely older than I am.

4:48Sam has two kids who are six and three. I also have two kids at a very different stage of their lives. They're now 21 and 15. But honestly, it feels just like yesterday that I was in Sam's shoes, managing toddlers in early elementary school. Her story about feeling like an inadequate parent brought some memories for me. I remember the growing pains that come with those years. In fact, there's one story that still haunts me today. Jenna, my oldest, was only one at the time. I told Sam about what happened next. We were in Cambridge during my PhD, and it was just she and I. And I was trying to get her to sleep in her crib.

5:32And I read somewhere that you should really just like plop them in the crib and let them cry it out every night. So I did that. And I would put her in, and she'd cry and cry and cry. She wouldn't stop crying. And I would give in, and I'd take her out, put her in my bed. And then one day I was like, okay, I'm going to tough it out. I'm just going to let her cry. And I sat there and she cried maybe like for 20 minutes or something. And then she threw up. And that was it. I was like, we're not trying this again. And she continued to sleep in my bed until she turned five. I don't know. Makes sense to me.

6:06I would go that direction too if my kid threw up after crying. This is what Sam is like. She's kind, empathetic, and a fierce advocate for parents. After seeing the child therapist, her relationship with her kid got so much better. And she wanted other parents to experience that. She saw an opportunity. I was working at a conversational AI company at the time, and I turned to my friend Maddie Mantha, an AI engineer and product and marketing leader at that job. And I said, can we rebuild this kind of therapist observation and coaching protocol using AI? Because it really worked for me. And she said, yeah, we absolutely can do that.

6:49And that is how Happy Pillar started. And Maddie is my co-founder. And we founded Happy Pillar to make evidence-based mental health treatment accessible to everyone. That is amazing. One of the things that I'm super passionate about is this idea of leveraging AI to help with mental health. You know, when you walk into a doctor's office today, they don't say, oh, Sam, what's your blood pressure or temperature? They just measure it. But in mental health, it's still a questionnaire. They say, oh, scale from one to 10. Like, how do you rate, you know, your kid's anxiety level or your depression level or whatever?

7:22And AI provides an opportunity to really change that. And I guess that's what you saw, right? Definitely. And I think we also saw that a clinician can often give a patient feedback either in a physical or mental setting, doing physical therapy, seeing a doctor. They can really observe you and give clinical feedback based on their training. But there are only so many clinicians in the world and only so much time for them to see patients. If AI can spread the expertise of clinicians to a broader community, then that community of people can get the kind of clinical expertise and clinical feedback that sometimes only the wealthiest and most privileged can get.

8:03And that is really, really special, a special thing about AI. Yeah, I love this idea of democratizing access to mental health and mental health care and support. Love the mission you guys are on. And I just think it's super impactful. It's an amazing use case of AI. So take us to kind of the core science or the core therapy around which Happy Pillar is built. So when you went to that human therapist, what was that therapy session like? So I saw a therapist who had a handful of certifications in various modalities, but the ones that were most interesting for my personal case were PCIT, which stands for Parent-Child Interaction Therapy.

8:44In that therapy, a therapist trains and coaches a parent on how to do a daily therapeutic session with their kid where they get down on the floor and play together. In that session every day, the parent has some verbal goals, some things that they are trained to say to their kid, or some ways of speaking you're trained to say that are really helpful and effective in repairing relationships, building resilience. Modality of therapy, PCIT, has over 40 years of evidence behind it. If you're seeing a human therapist, you're trained on how to do that five-minute interaction. You go home and you do it every day for a week.

9:19And then once a week, the therapist watches you do it on Zoom and gives you really personal pointed feedback on how you're doing it, how your child is developing, how you're changing, things like that. PCIT often brings in elements of play therapy. So a session can look as simple as, say, sitting down with your child as they draw a picture. A clinician will then coach the parent on how they can affirm their child through play. Sam took these therapeutic principles and brought them to an app. With Happy Pillar, we teach parents the same skills that you would learn in PCIT therapy or in forms of CBT or parent-child play therapy.

9:58Parents learn that from short lessons and videos that feel kind of like TikTok plus Duolingo. And then once a day, just like in real human therapy, you sit down with your child and play and talk to them. and Happy Pillar observes that parent-child interaction and gives exactly the same kind of feedback a therapist would give. In fact, all of the feedback is written and designed by licensed clinical therapists. Extra great about using AI to deliver that is feedback that in a human session, you'd only get once a week. You can get every single day and immediately. And we're hoping that that shows a lot faster outcomes for parents and kids.

10:37We're already seeing it with our users now And we're starting to design studies to get our own clinical evidence basis on that. I love that. So set the scene for us. Say, you know, parent and a child. And what kind of situation would they need to be in to find themselves looking for your app? And then walk us through this play session or a scenario where they're using Happy Pillar. So there are a lot of goals that you can work on in Happy Pillar. And when you start the app, you fill out kind of what feels a little bit like an intake survey if you were going into therapy. and you outline some of the things you might want to work on.

11:10So some of them are things that might be causing parents some stress, like temper tantrums, nightmares, low mood, anxiety. Some might be situational, a move, a new baby, things that just cause stress to a child and you want to prepare for them, maybe going to kindergarten. And some are just beneficial things that you want to strengthen, like fun new ways of playing with your kid, teaching them mindfulness and resilience, building connection, learning developmental parenting. and evidence-based techniques. So you can choose all sorts of goals and we have custom programs based on those goals and custom lessons.

11:45So you get yourself set up and then on your own as a parent, you do our short lessons. They're between three and five minutes a day. Watch some videos, answer some questions, really learn some of the things that a therapist would teach you. And then once a day, you sit down on the floor with your child with some toys and you're taught to play with them and to really let them lead the conversation. And there are some verbal ways that we help you lead the conversation. Some of them are suggesting that a parent narrates what their kid is doing. And we recommend that parents give specific praise or celebrate the behavior they're seeing that they really like.

12:21Like, I love how quietly you're sitting and I love how gently you're playing with your cars. Thank you for sharing. And we record that five-minute play session where the parents are encouraged to say some specific things. And then immediately we're able to give feedback on where parents are with those skills, how the play session went. We also collect data every week from parents based on evidence-based psychometric assessments so that we can show parents how things are moving, how things are tracking. What's an example of that? Yeah, so there are a handful of psychometric assessments out there that are evidence-based to measure progress in mental health care.

12:58How many tantrums happened like in the last few days or something? Exactly. So measurable things. And we customize what you're reporting back based on what you're wanting to work on. So you can see progress on what you're wanting to work on. And then we use another form of AI. So machine learning to design your therapeutic program based on your speed and progress and what you want to work on. So it is not just a video lesson like you could get from an Instagram therapist. And it's not just a book that is one size fits all. It's really customized and individualized. That's somewhere I think AI is really going to help expand healthcare and science, that we can really individualize things at an efficient and high-scaling capacity.

13:40There are so many factors we need to consider when making individualized, AI-powered healthcare tools. Of course, these tools need to be research-backed and evidence-based. But we also need to consider culture. Happy Pillar coaches parents on how to interact with their kids. It evaluates how we praise them and how we build up their confidence. But that kind of behavior can look really different depending on where you're coming from. For example, in my family, straight A's are expected. And there are certainly no participation trophies. So something that Happy Pillar teaches and coaches parents on doing is giving praise for the kinds of behaviors that you want your kid to be doing.

14:25and we know from decades of research in PCIT and other modalities that praise is a constructive tool and that when used in a specific and focused way can really affect the outcomes for your kid. But we've also learned that with different generations, different cultures, different accents, different idioms, praise can look really different for different people and the way we talk about praise and the way we coach parents on doing praise needs to be done in different ways for different folks. And for us, it's important to do two things to make sure that the coaching and training in Happy Pillar is culturally informed and sensitive and aware of how we all think about praise and the kind of praise we're encouraging parents to do and ways in which a certain sentence could be interpreted as praise in a different culture and figuring out the best way to do that in a way that gets the best clinical outcomes for kids.

15:19Yeah, absolutely. So what kind of insights would a parent give after the session? And this is not happening in real time, right? It's not like coaching the parent in real time. It's waiting after the five minutes are done. It's about 10 seconds after the five minutes are done and the parent can go review that on their own. So while you're playing with your kid, your phone is tucked away. The screen is off. It's listening, but you're not interacting with the phone. You're really just interacting with your kid and focusing on them. The app will give you the clinical feedback later. The kind of feedback you get is a lot of things.

15:51So definitely some encouragement. We pull out a top moment. So a moment where we heard connection between you and the child. It's great. Oh, love that. And then we have a little gallery of top moments. So you can really see all the great connection you're having. And then some of the skills that I mentioned, we count how many times you were able to do those. So how many times did you narrate what they were doing? How many times did you give them a labeled praise or a celebration? One of the things that you're taught to do in therapy is to echo back what your child says. So if they say, I'm building a tower, you say you are building a tower.

16:24It doesn't have to be an exact echo, but you're trying to reflect back what they're saying. It really builds their confidence and it encourages them to expand more. And there's a clinical saturation point at which it really has an effect on their future mood and anxiety and behavior. And we make sure you're getting really close to that number so that you can really have beneficial outcomes. Happy Pillar didn't exist when my kids were younger, but I'm reminded of a conversation I had with my son, Adam, a couple of months ago, because I think a lot of parents, especially today, were distracted, right?

16:55So I was in a conversation with him and he was trying to remind me of something and I just couldn't remember it. And I was like, I must have like dementia, right? And he was like, no, mom, you just never pay attention enough to any one thing. You're always like half present. And it was such a both sad, to be honest, moment and also a powerful moment. And it just made me realize that in a lot of our conversations, you know, I might be on my phone, I might be thinking about like, what I, you know, the next meeting that's coming up. And I kind of wish that I had something that just nudges me as a parent to be more present.

17:33Although I do want to point out that if he is comfortable enough saying that to you in that way, that shows that you've really done a good job, that he respects you, understands you, connects with you. So pat yourself on the back for that. Okay, I appreciate the support. But I think you're absolutely right. So the therapy itself taught me presence and focus and mindfulness. And also it subconsciously taught me different ways of communicating, not just with my child, but with my partner, my other children, my co-founder, my colleagues. So some of it is subconscious learning. And then some of it, you're right, is taking the time to be present and focus and connect with them.

18:11But you're focused on the early child target audience, correct? Right now, definitely, especially because there is very little evidence-based, self-guided options for that age group out there. This therapy that I did exists, but there are only so many clinicians. It is not cheap. It's kind of hard to access. But what we've built is a system that can analyze any two-person conversation. And we really can apply this to all sorts of mental health modalities for other ages, specifically the ages that are not well served as far as technological advancements go. So what's really extra important about focusing on the early childhood age group first is that we're really seeding mental health intervention and prevention and comfort in young kids.

18:59So that as they grow older, they believe that mental health care is just something we all do, like brushing our teeth. And ideally, we'll be preventing a lot of future crises or mental health challenges. And then also, this is a population of children who are born in and around a global pandemic. So they need some support. their parents need some support. So that's a big reason we focus on that age group right now. There's a real shortage of mental health care professionals, which means that families are not getting the support they need. And this is where AI can help. But putting your trust in AI when it comes to your child's health can be tricky.

19:38So how do platforms like Happy Pillar earn and keep the confidence of families? That's in a minute.

20:02If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's Teamwork Collection. It has planning in Jira, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build. Learn more at Atlassian.com slash TeamChanger. that's a-t-l-a-s-s-i-a-n dot com slash team changer

20:51so you've kind of walked us through how the user experience with happy pillar looks like let's go behind the scenes and talk about the ai that's powering all of this so let's start with that what kind of ai slash machine learning technologies are you using for happy pillar So it starts with us recording a parent-child interaction. And then we do something called diarize the speakers. That means identify who's the parent and who's the child. And after that, we convert the speech to text so that it can be analyzed while retaining some of the characteristics of the recorded speech, like uptalk or tone.

21:32So it's not just the words. You're not just analyzing the words. you're also analyzing the vocal intonations and other aspects of the speech characteristics. That's absolutely right. So one of the things parents are asked to avoid just during happy times, the child is in control. They're asked to avoid questions so that the child is leading the whole conversation. Questions are great tools for 23 hours and 55 minutes of the day. But in these five minutes, the child's in control. So we ask parents to avoid questions because it's clinically effective. But you could say you're building a car and that is perceived by a child as a question and often interrupts their thought pattern and what they might have said next.

22:13So if we hear a question based on tone, we'll mark that as a question and coach parents on how to say things as statements or as echoes or reflections. So cool. Are there any phrases or things that happen that are particularly challenging for the AI to understand? I think one of the bigger challenges at the beginning was actually the diarization, which I just mentioned. Often the biological parent of a child, if it's the mom and it's their own child, a female adult voice and her own child's child voice can sound really similar. And Maddie and the engineering team had to figure out how you identify who's the child and who's the adult.

22:57Because the child can say whatever they want during happy time. We don't analyze what they're saying. We throw away any recording of what the child is saying. So we really need to identify speakers accurately. And at first, that was pretty hard. Unless it was a dad, dads have deeper voices than their kids, but moms don't. So Maddie's team discovered that adult and child speech each have different formant frequencies. So they are using AI to identify the formant frequency of each speaker and figure out which one would typically belong to a kid and which one would belong to an adult. Well, that actually brings us to a really important question.

23:31Anytime we're talking about AI, the fuel of AI is, of course, the data, the training data that you're using to train these models or these algorithms. Do you need any human labelers or annotators? And just for our listeners, that's when the data needs to be kind of labeled by actual human experts who can say, okay, this was a great interaction, or this was maybe a really lousy interaction, or this was a question, this was not. Do you need a team of human annotators at all? Our data is annotated by licensed clinical therapists. So not by engineers or coders. It is annotated by clinicians. So that is something we do to ensure that our model and our data set is being treated with the most clinical efficacy and that it's the safest.

24:18I want to pause on that, by the way, because that's a really important decision that you've made as an entrepreneur and as a founding team. Because you didn't put this data on Amazon Mechanical Turk, right? MTurk, which is this like platform on Amazon where you can hire labelers for like, you know, all around the world, very, very cost effectively. But they're not experts. They have no idea what they're annotating. And you went almost like the opposite extreme where you're hiring these licensed clinicians to annotate the data. We also want to make sure that we are delivering a therapeutic intervention that is actually effective.

24:52So maybe we could have done things faster or more cheaply from a business perspective by going with something like MTurk. But there's no guarantee that that would have beneficial outcomes for kids and parents. You know, I also think data can be a true competitive moat when it comes to AI companies. And I believe you're amassing like the first ever dyadic data set between child and parent. Tell us more about this data. What do you plan to do with it? Have you mined it? So we started Happy Pillar a little over two years ago with just an idea. And there was no data set out there at the time that would have satisfied what we were needing.