​Dr. Daniel Knoepflmacher (Host): Welcome to On the Mind, the official podcast of the Weill Cornell Medicine Department of Psychiatry. I'm your host, Dr. Daniel Knoepflmacher. In each episode, I speak with experts in various aspects of psychiatry, psychology, research, and other important topics on the mind.

As we've discussed on this podcast before, we continue to face a mental health crisis in our country and around the world with a shortage of qualified clinicians to provide the treatments people need, both biological and psychotherapeutic. With decades of advancement in neuroscientific research and psychiatry, one fundamental truth has persisted, psychotherapy remains a fundamental component of effective mental health care across diagnoses and conditions.

Alongside biomedical breakthroughs, there has been a rich tradition of research into psychotherapy, establishing the efficacy of different modalities and improving the practice and delivery of meaningful care. Efforts to create precision psychiatry have focused on the goal of matching individuals with the right personalized treatment early in the course of their illness. This same question extends to psychotherapy. With half of those receiving psychotherapy remaining symptomatic after treatment, how can we do a better job tailoring the right modality for the specific problem affecting a specific individual? And what role might artificial intelligence play in all of this.

Today, we will address these questions and many more with our guest, an accomplished psychotherapy

Dr Daniel Knoepflmacher: researcher

Dr. Daniel Knoepflmacher (Host): Dr. Nili Solomonov. Dr. Solomonov is an Associate Professor of Psychology in the Weill Cornell Medicine Department of Psychiatry. Nili, thank you so much for joining me today on the podcast.

Dr. Nili Solomonov: Absolutely. Thanks for having me. Such an honor to be here.

Dr. Daniel Knoepflmacher: Well, my favorite part of this show typically is this first question, which I ask everybody, and that's about hearing your story. So, I'm wondering what in your life, what choices did you make that led to your path becoming a psychotherapy researcher?

Dr. Nili Solomonov: Yeah, it's a great question. It's a moment to reflect. I was raised in a multicultural home in a small town of immigrants in Israel. And my father fled as a child with this family from Morocco, and my mother was the daughter of Holocaust survivors in Poland. And so, we had a quite a multicultural home. And social justice, open-mindedness, humility, those were core aspects in our household. And so, my father actually later became one of the first North African immigrants to graduate from an Ivy League School at Columbia at New York City, which is where I was born

So, my parents always encouraged this love for science and critical thinking and questioning social norms. So, no question was ever off the table. So, we've often debate topics at dinner time. And so, from an early age, I think I really felt that my voice mattered. And thinking back, I realized that the choice specifically in focusing on late-life depression in the last few years is really kind of coming from my relationship with my maternal grandparents and their home was the center of the family, especially with my grandmother with whom I had such a close relationship, and she was the first to expose me to psychological theory. She was a really special person. She was one of the founders of the Adler Institute in Israel, and one of the first women to really kind of make some significant breakthroughs in education in that field. And so, growing up, really, I observed them leading a life of graceful aging and deep commitment to their community, and especially to the vulnerable. And I've always appreciated and valued the role that older adults have in our communities. And I think that that has been the driving forces of my passion to focus on mental health and live life. And so, this is sort of my personal background.

And professionally, I've always been interested in how people change. And my interest in clinical research started in my undergrad. I attended Ben-Gurion University in Israel and worked with Dr. Yoav Bar‑Anan, and he was really focused on social cognition. So, we learned to develop these elegant experiments to understand fundamental questions about the human mind. And that was a really enriching experience. But I realized that I was really interested in learning how people change over time and what makes us change. And so, during that time, I was also volunteering in a community residential facility with youth with chronic mental conditions. And so, I was seeing this lack of evidence-based interventions available to those folks who are really kind of deeply lacking access to good mental health care. So, I was determined to study psychotherapy research, and I was accepted to work with Dr. Jacques Barber, who was-- still is-- one of the world renowned experts in psychotherapy research. And he had just moved to Adelphi university to be the dean.

So, I had this big dream to move across the ocean. And we did that. Me and my husband, with his great support, we moved to New York City. And so, we got to New York City and I started my PhD. And the PhD, the doctoral work with Dr. Jacques Barber and others focused on identifying specific interventions that target mechanisms of action and psychotherapy. A mechanism of action is basically sort of like the secret sauce, what makes psychotherapy work. And I developed this deep fascination with this idea that we can understand what makes psychotherapy work and hence improve it. And I also became fascinated with how computational approaches can help us get there.

And so, I spent hours and hours learning how to code and how to identify these hidden patterns in psychotherapy process and outcome. In a series of studies, we developed a short measure to assess the use of psychotherapy interventions. And we also identified what interventions were most beneficial in driving change in psychotherapy. And so, I was going through this and really becoming extremely passionate about clinical research and becoming an academic. And during that time, I also was pursuing my clinical training. And I sought out clinical training in community hospitals here in New York City. And I trained in Mount Sinai Beth Israel, back then in Jacobi Medical Center. And we worked with deeply underserved communities, both in inpatient and outpatient. And specifically, my training in geriatric inpatient and outpatient units sparked my interest in geriatric psychiatry. We're seeing this pervasive social isolation.

It's very interesting that New York City is so dense, yet people describe this exceptional sense of isolation despite being surrounded by other people. And so, at the same time, I was learning that psychotherapy research really focused mostly on pediatric and youth and adult depression and very few studies focused on late-life depression.

So, I looked for an opportunity to bridge this gap and bring some of the computational methods that we were developing into geriatrics. And fortunately, met Dr. George Alexopoulos, Dr. Faith Gunning, and Dr. Jo Anne Sirey, and was accepted to the NIH-funded T32 fellowship here at the Geriatric Institute. And that was deeply life-changing. And so, I know we'll talk more about the work, but that's sort of how I got here.

Dr. Daniel Knoepflmacher: Well, you were fortunate, but we were too, to have you come here and really take all that generational influence, that interest in the mechanisms and deep experience with the computational approaches and put that all together into the work you've done here. As I alluded to earlier, we are facing a crisis in mental health, and this means that depression, anxiety, and suicide rates have been increasing over the past several years. And these major issues that we're facing in our field, they really dovetail with the focus of your research, including the shortage of clinicians who are trained in evidence-based models of psychotherapy. So, how does your research address that core problem?

Dr. Nili Solomonov: Yeah. So, this line of work really started with my work with George Alexopoulos, who is really focused on this exact problem that you're describing, that we have these phenomenal evidence-based models, but they're rarely used in community settings. So, I'm very passionate about this work, because I know that very few can access these types of gold standard psychotherapies that are delivered in academic centers such as the one that we're in and many others.

And so, I really think that we can leverage some of these novel tools, these computational tools, to develop effective and scalable therapies to the patients who need it the most. And in context to your question, we know that most of the folks who actually need mental health are treated in these community settings, public insurance-based mental health clinics, community centers, public hospitals. And in the U.S., these settings typically include very few highly trained PhD level psychologists. And most of the patients are seen by master's level social workers and mental health counselors, which often are absolutely wonderful and phenomenal and have great skills, but they really don't have a lot of access to training and evidence-based psychotherapies, and they don't have the resources. They're seeing a tremendous amount of patients, long waitlists, high caseloads, and they often are just not able to keep up with the science to have these complex therapies like the packages that we know work, interpersonal therapy, cognitive behavioral therapy under their belt. Our work focuses on leveraging computational methods like AI to develop simple therapies that can be delivered in the community by community clinicians and even sometimes non-clinicians to address this public health imperative.

And I really believe-- we believe-- that by identifying what makes psychotherapy work, like what is the mechanisms, what is this secret sauce, right? We can identify the interventions that are driving change in these mechanisms and grab those interventions into simpler therapies that are likely to be delivered successfully with adherence by community therapists. And so, everything that we're going to talk about is with great deep respect to the evidence-based psychotherapies and also the understanding that we need models that can be used quickly and effectively by folks who may not have time to learn the richness of the evidence-based models that are out there.

Dr. Daniel Knoepflmacher: So you're really identifying the fundamentals and then building up accessible interventions based on those mechanisms.

Dr. Nili Solomonov: Absolutely. Yeah, exactly.

Dr. Daniel Knoepflmacher: All right. We keep mentioning how so few people are getting the gold standard treatments, and I think at least 50% are not receiving the right therapy for what they're seeking therapy for. You've talked about your experience with computational methods. How can AI and machine learning help to tackle these issues?

Dr. Nili Solomonov: That's such a great question. And I want to take a step back. And before we talk about how to fix our broken therapies, I want to start with the positive that we have such a rich psychotherapy research data base at this point of everything that we've learned. And we know that therapy works. You started the podcast with that. And I feel an obligation to say that to our listeners, because I work a lot in community spaces and I hear many, many folks repeating back, "This is not going to work. Talk therapy doesn't work for this, or wouldn't work for me." And I want to say it works for more than half of the people who get it, and for some treatments, that the rates are even higher. And that's huge. And so, we're really thinking about the question of why it works for some people and not others.

And so, for folks who are, you know, in the 50, 60, sometimes even 70% where psychotherapy is effective, that's wonderful and fantastic, and we're not so worried about those. We're worried about the folks who receive these really great therapies, but are still suffering. And our research has shown us and many, many other people's research that one of the core reasons that psychotherapy may not work for so many people is that it's traditionally taken this one-size-fits-all approach. And we've talked about precision. That's exactly that. It's this notion that if CBT is effective, we should give every depressed patient that walks through the door CBT. But we know that patients vary in their clinical presentation, right? And so, two patients with depression can look completely different.

So, patients vary not only on their depressive symptom and how their depression looks, and I know you've talked about that in prior podcasts, it's also in their social functioning and their sleep habits and their activity levels and their brain functioning, which is something that we're really focused on, and many more aspects. And so, kind of going back to your question, this what we call heterogeneity of depression, the reasons why people are so different is where computational methods like AI machine learning can make a big difference. And they're powerful because they can learn fast and they can sort through large amounts of complex multidimensional data, like what I've described, these multiple domains of functioning. And they can identify hidden patterns that are not necessarily discoverable with our conventional methods. And so, these methods are quite clever. Because they adapt based on new information and we're all experiencing that. What we're going to talk more, I know, later about people, folks, is interactions with AI. But the more training a model has, the more patients it has seen with depression that looks different. The more examples it's seen, the better it would get at prediction. And so, for example, a machine learning model can tell us out of a hundreds of factors that lead to treatment success or failure based on hundreds of examples of patients who succeeded or failed in treatment, which are the most important ones? And that's where we really start to move the needle, I think, I hope, right? Because then, we say, okay, the top ones, the top predictors that are going to make the biggest difference, those are the ones we should assess and those are the ones we should intervene on. And that's a key goal of the work that we're doing.

Dr. Daniel Knoepflmacher: Well, you use the word prediction, and I think that's very important here, because I think when anyone hears the word AI, maybe people who are listening, they think of chatbots or LLMs that respond to questions you have. But what we're talking about in this context is AI being used to analyze data that comes from clinical interactions, and then using that to determine what are the right modalities or interventions to meet the fingerprint of that data that emerges. Is that accurate?

Dr. Nili Solomonov: Yeah, absolutely. And I think that's such a great point, because we're in this AI evolution period now, that AI is kind of taking the forefront in so many aspects of our lives. And the main interaction that people have with AI is with generative AI, AI like ChatGPT, right? These models are as indicated by the name, they generate new content, they're generating new information.

So, it's a very important aspect of AI. And we'll talk more about that later in the podcast. But there is also a key very powerful aspect of AI, which is the prediction piece. So, models that are able to use the data to predict and explain outcomes. And this is when we talk about many machine learning models, for example. These are powerful tools that we're using in our research. So, over the past last five years or six years, our lab has done a series of studies in leveraging mostly machine learning techniques to identify what makes psychotherapy work. And as I've mentioned, the question is really what are the mechanisms, what are the active ingredients and why does it work for some people and not others?

And so, just to give an example, we've built and trained a model to detect early risk of non-response. And this is really important, right? Because I think a lot of people and probably many of our listeners can identify with the experience of going to therapy and having a sense that it's not working after three or four sessions, right? And then, the therapist may say, "Well, let's just keep going. Let's just keep doing what we're doing. It will work." Sometimes that's true, but often it isn't, right? And so, we've asked that question with these models. We've asked a model to identify what patients are not going to respond early on to treatment. And then, we've asked, "Well, is the fact that they didn't respond early predicts their response at the end of treatment?" And the answer was yes. And this is work that I've done with Jihui Lee. And it was really kind of surprising to see. We found that in our data, all of the patients who didn't show response by half of the treatment did not respond at the end, meaning they were still depressed. And so, we really kind of need to do a better job at identifying folks who are not responding early and doing something about it. And that's where we hit the predictors question, right? How do we know who doesn't respond?

And so, in this study, we've looked at tens of predictors, somebody's disability profile and the type of treatment that they received and their cognitive profile and anything basically that you can think of that would explain what this person is suffering from. And the number one predictor was actually perceived social support, how much you felt supported by others, more than what treatment you got, more than what your cognitive, your social, your disability profile and so forth. And this was really good news to us because perceived social support is modifiable. We can do something about it. We can help folks feel more supported. We can't do anything about your demographics or about most of the time, the disability, and other factors. But this is something that we can intervene on.

Dr. Daniel Knoepflmacher: So, that's at the key of everything, is that there's a deficit in social support, and you've really used all of this technology to identify that as one of the aspects of the problems that people face when they're not responding to psychotherapy. I know that you've created a simple, scalable psychotherapy model based off of this research that you've designed, and it's called Engage and Connect. It's something I'll say that I'm particularly excited about because I was just talking with some of our residents who I just yesterday spoke to them about it, that they learned this from you, or they heard an introductory lecture on this, and they're all very excited about how they can adapt this into their practice. So, can you describe this form of psychotherapy to our listeners?

Dr. Nili Solomonov: Yeah, absolutely. And it was so great just to meet the residents. They really challenged me with some interesting questions about how to bring this into a psychiatry space, this therapy, which I haven't thought about because mostly we do our work with social workers and community therapist. I can talk a little bit more about that later.

But I want to kind of say, you know, the cliche of sitting on the shoulders of giants, Engage and Connect is a modified version of Engage Behavioral Activation Psychotherapy that George Alexopoulos and Patricia Areán developed. I was so fortunate to work with them and to have access to all of their data and their brilliant perspectives on psychotherapy.

And so, taking a step back, Engage was designed with this based on neuroscience evidence that's showing that the brain's reward system is severely impaired during the depressive episodes. And that basically means that our brain is not anticipating rewarding experiences and not enjoying rewarding experiences when they occur, right? So, let's say, Daniel, if you and I decided after this podcast to go and grab a cup of coffee, then I would even though I'm nervous about the podcast and talking about my research, I would say, you know, at least have the cup of coffee later. We're going to talk about everything. We're going to, share our experiences and it's going to be very rewarding, that experience, right? And then, I would really enjoy having that experience, even though we haven't had it yet, right? The anticipation is rewarding in and of itself. And then, when we have the cup of coffee, of course, I'm going to enjoy your company and we're going to have fun, and I'm going to have joy and pleasure while we're experiencing that.

And so, what we've found in research, and we found this particularly with the Engage samples, but this is based on a very longstanding neuroscience research that's showing that, in depression, those two aspects of reward are impaired. So in my example, I would be dreading coffee. I wouldn't be experiencing joy and pleasure, anticipation. And while we're having it, I wouldn't enjoy it either. I would kind of just say, "I can't be myself," or I wouldn't say anything, or I would feel very negative, or I would have negative thoughts about you. And so, we're really kind of thinking that by increasing exposure to these rewarding experiences, we can rescue the reward system and in turn reduce depression.

And so, in these studies that I've mentioned earlier, we've also found that engagement and meaningful social experiences with significant others was particularly beneficial. And folks who engaged in more socially rewarding experiences with significant others did better in Engage. And so, we've developed Engage and Connect with a sort of a social twist to it. And the idea is that social reward engagement is the key intervention. It's extremely simple. We teach people how to do it within five to seven hours, even if they have minimal or to no clinical experience, and they can do it reliably. And the idea is that through this increased social engagement, we can reactivate the reward system and reduce depression. And so far, we've successfully piloted it in the lab and in RCT space. And we also have an implementation study where we're looking at how this works in the community.

And I think what's most important and encouraging to us is that we're seeing that what is really driving change in the mechanism for us is the reward system of the brain that we're trying to activate seems to be the socially rewarding experiences.

Dr. Daniel Knoepflmacher: It's so important at the basis for geriatric depression. And I think, in general, you mentioned in New York City, a crowded city, but across the country and if not the world, there's a lot of talk about increasing disconnection in this social realm. And so, this is especially impressive as something that is addressing that individually in therapy. But if it was scaled out, then on a more wide basis. I'm curious though about the challenges that come when you're translating a psychotherapy designed in this more ideal research setting and then taking that and putting it into practice in the messiness of the real world. What data do you have about the application of Engage and Connect in real world practice settings?

Dr. Nili Solomonov: Yeah, I think you're raising a really important point here that one of the main gaps of psychotherapy research is that often the greatest innovations, the things that are most exciting to us as researchers don't actually reach the real world, especially underserved communities that don't have access to medical centers. And so, we've seen this explosive progress with AI and treatment innovations and psychotherapy research. But very few of these advancements are actually used by community therapists. And this is certainly something that we think about a lot.

And so, specifically with Engage and Connect, about five years ago together with  Jo Anne Sirey, who's a close collaborator and a mentor, we launched a project that's supported by New York City Department for the Aging. And we implemented Engage and Connect in the community across 20 senior centers. And I have to say this was a specially humbling experience for me as a researcher, because I've learned that there are so many blind spots that I have when I'm testing a therapy in the lab and actually are wonderful community therapists or patients or community partners, directors of senior centers, supportive staff. Those were the folks who sometimes had the best insights on how to improve treatment. I'm saying that also because I think one of the key missions of my work has been bridging those gaps, not just in research, but also in thinking, in conceptualizing.

And so, in the community, we see folks with housing, food insecurity, complex clinical presentations that are not necessarily simply first episode of depression, high rates of trauma and crime, and bringing Engage and Connect that was designed originally for depression and suicidality was really a big part of it, was simplicity of this model. And so, our work with computational approaches early on helped us hone in on what are the interventions that we can actually bring out, and I can talk for hours about how we've changed things and how we've adapted them to these settings. But certainly, we are still actively in senior centers. If folks who are listening are attending senior centers, you will see our social workers and community therapists doing a phenomenal job. And we're really, really interested in this collaboration continuing long term.

Dr. Daniel Knoepflmacher: Well, that's for those who have the community therapists and have access to them, which fortunately is many people, but there are many people who can't access any therapist. Is there a way in which this treatment would be able to reach them?

Dr. Nili Solomonov: Yeah, that's a great question, Daniel. I think one that keeps myself and many clinicians and researchers up at night, especially building a career as a psychotherapy researcher and focusing really specifically on therapist-delivered interventions. There was a point where I was really seeing that, even with the best simple and the simplest psychotherapies, we really can't reach everyone.

And I think digital apps and hybrid interventions are really promising here. And I can say specifically for Engage and Connect, in a collaboration with Dr. Samprit Banerjee, we've designed an app that delivers this therapy with a simple guided platform. And again, going back to the mechanisms question, that's where we pulled from our knowledge and from our data about what interventions were most effective and targeting the mechanism and bringing those to a digital app.

And what's unique about this specific app that we're very much in progress is that it leverages some of the prior machine learning studies that we've done to predict early risk and offer personalized interventions. And we're also developing in parallel computational models using methods that we've discussed earlier to hopefully match patients with the right treatment, to hopefully be able to early on, right at the gate, at the assessment stage, assign patients to a digital intervention, or to a self-guided hybrid intervention versus the full nine yards of the therapist-delivered intervention. I think this approach can be really transformative, because it can increase access to more patients and also reduce burden of therapists.

Dr. Daniel Knoepflmacher: Well, you've been doing a lot of your work and where you started was in the older adult population. Can you say a little bit about this application to other populations?

Dr. Nili Solomonov: Absolutely. That's a great point actually. About three years ago, I was approached by Benda and Megan Reading from Public Health. And they were designing this risk assessment algorithm to identify which women are going to experience postpartum depression. And so, we talked about interventions for this population, and we sort of reached this point where we're saying, "Well, Engage and Connect is focusing on social isolation." We know that during the postpartum period, there is socially rewarding deficits. You know, if you've treated women with postpartum depression, you often see this lack of reward response to the infant that is quite distressing for the mother. And so, conceptually based on neuroscience evidence, it made total sense. And so, we've adapted Engage and Connect postpartum depression. We've tested it out, our pilot. We've published a pilot. It had really promising results with extremely high response rates. And what was most interesting is that we found that this adapted version of Engage and Connect to put the postpartum period that was really tailored for women in the postpartum period showed a signal on where we thought the mechanism was.

And so, we found that the increase in socially reward responsivity, meaning your response to socially rewarding experiences explained the reduction in depression. We also found that Engage and Connect in this small sample led to improvement in quality of attachment with the infant, which I think is really important, because we always think about what are the long-term benefits of a treatment. What are people left with when we end our therapy with them. And so, we've done the postpartum work. We're extremely excited about that. I think this is a key direction that we're going to go in.

And we also are taking a transdiagnostic approach in our

collaboration with the community. We're thinking about folks, for example, with housing and food insecurity, right? It really is not right for someone who is experiencing food insecurity-- which I think is very sort of salient to us right now because we are facing a food insecurity crisis right now in New York-- to say, "Why don't you go for coffee with your friend?" right? Or, "Why don't you, like, host a party?" That's not aligned with their experience. And I think that's where Engage and Connect has been really modified in different directions where we say, "Well, why don't you have a conversation with-- let's identify a significant other that you can have a conversation with about your insecurity," "Who in your life do you feel that is supportive of you, that is bringing the best self in you?" or "How can we help you get connected with food sources with the support of others?" And so, I think, often it's not very fancy. It's actually extremely practical. And when we face problems in our lives that are as significant as insecurities like this, we need others. And we see patients, especially with depression, that are furthering and furthering their isolation because of things like shame and guilt and stigma. And we say, let's do the opposite. Let's find someone you can connect with.

Dr. Daniel Knoepflmacher: Well, that adaptability is really impressive and I'm really excited to hear about the different directions that you will take this in. You just don't have enough hours in the day, I'm sure. But that's wonderful. At the same time, I am guessing that engage and connect is not going to work for everyone and for every psychiatric or psychological problem. In fact, I mean, I think you and I were talking about this earlier, but I think about personality disorders, they're complex and can really impact social connection. There may be real difficulty with social connection in that population. So, how do you decide what individualized treatment to give each patient?

Dr. Nili Solomonov: Yeah, I think that's a great point. And it's sort of like the $1 million question for us, right? How do we see this strong data coming from Engage and Connect studies, which I'm really excited about that. I certainly don't think that it's the right treatment for every patient because no treatment is right for everyone, right? And that's what we've learned from these decades of psychotherapy research.

And so, one example that I have in mind is a collaborative study that we've done together with George Alexopoulos and Faith Gunning and a big interdisciplinary team where we recently published a machine learning algorithm that's called the Treat and Decision Rule, we've based off of really excellent work that has been going on in the field of precision psychiatry. And this algorithm takes into account all aspects of the patient's profile, like we talked about social, cognitive functioning, symptoms, disability, and other factors. And it can automatically assign them to the right treatment by taking into account these multifaceted variabilities that people have. And so, we hope that this is quite promising, because it addresses some of the core problems that we're facing as a field. It reduces the assessment time, right? Because we know that therapists spend a tremendous amount of time assessing a patient, and that is a huge burden for both the patient and the therapist, especially for patients who are going through multiple rounds of treatment and have to explain their story over and over again.

And so, we're assessing only the key factors that we know based on our model training that are predictive of outcome. We assign patients automatically, and that means that we give every patient a fair opportunity to receive the best treatment. And this is something that I'm particularly excited about, because we know from prior work that clinicians, even with best intentions, we're all human, right? We're all humans, and we often and intentionally are impacted by racial and ethnic biases. And we see this bias in assigning patients from minority backgrounds. And so, that takes care of that by the automization. And it also enhances cost effectiveness, because it reduces the clinician assessment time, but it also assigns patients to the briefest treatment that they can benefit from.

So, we adapted this algorithm specifically based on our experience with community settings. And we've added this modular aspect of it where you can essentially tell the algorithm, I can only afford to treat 20% of my patients that are coming through the door. And if you're then hiring one or two or four clinicians, you can say, I can only afford to treat 40% of the patients who are walking through the door. And the algorithm would be able to adapt and still assign patients based on the optimal predicted benefit. And so, in our initial study, we found that this algorithm increased depression reduction by 30%, compared to random assignment, non-optimal assignment. So, we're hoping that this can make a real difference.

Dr. Daniel Knoepflmacher: That's really impressive. And you're integrating into that model the different modalities that exist for the profiles that emerge based on that. So obviously, it's not just Engage and Connect, there's other choices that get integrated into the system.

Dr. Nili Solomonov: Yeah, it's a great question. So with our model, we trained it on simple psychotherapies, like Engage and Connect versus care as usual, which was often case management. And I think it's such a good point, right? We're now starting to expand this into other spaces, and particularly we're interested in youth depression because it's such a big health imperative there. And we're hoping to adapt the assignment based on what's available in that setting, right?

But I think what's important here is that our algorithm was able to identify patients who would benefit from case management and not psychotherapy. And I think that that's really important because I think that as clinicians, we feel like we should give every patient the whole nine yards or best longest therapy. And what our results are showing from these models is that that's not actually accurate, right? There is a matching aspect that not every patient would benefit from that. So for example, we talked about apps for example. Some patients may benefit more from an app first, from a low touch intervention, that's not the whole comprehensive a hundred percent of what you can offer and then maybe later need therapy or not, right? And so, I think this is where we think there could be a big difference, right? If we can assign the folks who can benefit from a digital intervention-- quickly to a digital intervention, we clear out therapist time to treat the patients who really deeply need therapist-delivered interventions. And we can do this. We're hoping to learn over time how this works with different modalities.

Dr. Daniel Knoepflmacher: This really gets to something I want to discuss with you, thinking about AI. So, we've been talking about prediction, machine learning, predictive AI. What about generative AI to where the public is maybe more familiar with, which there's a lot of controversy about this for good reason. And whether human therapists would get replaced by digital artificially intelligent therapists. Can you just speak to generative AI and where that may fit into the future of psychotherapy?

Dr. Nili Solomonov: Daniel, I think that's such a great point. We're seeing this explosive increase in the use of AI chat bots, and particularly for therapy or for companionship. We can talk more about that. And I think most people don't realize that AI can be very many things, right? And they also, I think, are not aware that these AI bots, especially bots like ChatGPT are meant to be general, meaning be able to generate content, good content about pretty much anything and everything that you ask them. And if you've used this, and I would be surprised if our listeners haven't tried it because it's so, so popular right now, is that we know that AI is designed by definition to always give you an answer, even if it doesn't know. AI often would simply give you a wrong answer, right? Which is not a big deal if you're asking for a restaurant recommendation, you'll just have a bad meal maybe. But it's a big deal if you're relying on this tool for your own safety and your own psychological health. These models can experience what we call the technical term is hallucinations, right? Where an AI bot would give you a very confident but very wrong advice.

And so, with these strategies that have happened with youth suicide that we've seen in the news, there's clear instances of that, and the safeguards is a huge issue too, and I'll stop after that. But I think that's one that many of our listeners are probably concerned about, that the therapist is obligated by law to keep you safe and abide to very clear guidelines on when to escalate and when to take the next step of calling transdiagnostic.

Now, as a clinician, and I'm sure many clinicians who are listening to us like me hate to do this, right? We don't want to call 911. We don't want to do these things, but we're 100% going to do it if we need to, right? And we're going to make sure that a patient is safe. And we don't have regulation on this with AI bots, right? There are no consequences or there's very little consequences if they don't. And we don't know what happens if a patient is suicidal and an AI bot says call 911. Does the person call? Do they not call, right? If we have a patient who's experiencing this, we follow up. We make sure that help has arrived, right? And so, I think the follow up piece is very, very troubling, because we see these experiences, examples where safety guard rails are not working and dangerous advice is being given and there is no follow up on that.

And so, these are all challenges we need to address in the upcoming years in the field. And how can these technologies become less dangerous? What policies do we need in place to protect ourselves, our children, our vulnerable members of the community? And, you know, I think this is not all to say that AI is bad or shouldn't be used. It's here. It's going to be used. It is in the mental health space. It is going to be an intervention that millions of people are already using it. But we do need to make sure that it's safe, and we do need to be proactive about it as a society, as a community. And I think that a lot of us clinicians and researchers are really kind of focused on that.

Dr. Daniel Knoepflmacher: Absolutely. And I'm glad we're having a chance to talk about this. The vulnerable population of youth is something that you alluded to, that key developmental stage. I mean, there's even work to think about chatbots, not in terms of psychotherapy, but as toys that could start at the very earliest age and be these very immersive, attention-drawing digital companions, which could have really scary impacts on attachment, on attention, on all of these kind of social relatability, developmental steps that have to happen.

And for that reason, I know already there is bipartisan legislation that's being discussed to really limit this. But the train has left the station, and I think people really need to know about this and need to be organized and aware so that we can think about how this is going to develop in the future. So, thinking about that vulnerable stage of psychological development, what do you think are some ways we can protect the future generation from these risks?

Dr. Nili Solomonov: Yeah. Those are such great points. And not even just as a clinician and as a researcher, but also as a parent, I think about these things a lot, right? And I think the short answer is that we need better education for our communities. And there's a cyclic issue, right? Because people are becoming so accustomed to going to AI for information. They're asking AI for information about AI, right? And so, I think that we need clear regulations. We need better education. There are sources that are unbiased and science-based.

You and I were corresponding about the latest American Psychological Association report on artificial intelligence and machine learning in mental health. And I really think that every parent, every person should read that. It's really well done. And we need government. We need government together with scientific and clinical communities to step up and set policies and guidelines around AI. And we need ways to validate who's using AI. We need to protect minors. We need to have better safety guard rails for these models. We need to have better transparency about how these models are training, what data they're training on. We need better education for patients and especially parents for what these tools are doing.

And I would strongly recommend, if you're listening to this podcast and you're a parent of a child or youth that has access to AI and that has a smartphone, get involved. Have an open and honest conversation with your children about this. Be aware of what they're doing with AI, what they're using it for. Be proactive about this. We're seeing especially with youth and children that the information that they're entering into these bots is exceptionally sensitive. And I always kind of tell my patients, if you're putting information into the internet, you have to be okay with it living there forever, right? And I think that, as adults, and we talk with older adults about this a lot actually from a different direction, right? Because they're more trusting because sometimes, they didn't grow up in the digital world and often they would just put things on the internet that they're not thinking twice about. And we have to be very educated consumers. And with youth, it's especially an issue, right? Because you can't monitor every single thing that they're using their phone for. But I think this is where it's really critical to have parent involvement. We see this, we talked a lot in this podcast about social reward. All of the data that I've seen is pointing clearly to the direction that we need socially rewarding connections in the real world with real people, right? And this shift that we see in youth and using AI as a replacement for that, because AI would not bully you AI would not reject you, AI would not tell you you're a bad friend. I'm giving extreme examples, but this sensitivity to social approval is really playing a huge role here. And so, I think that this reduced preference for human connection is something that we really need to be aware of and work on.

Dr. Daniel Knoepflmacher: There are tremendous risks and you mentioned that statement from the APA and we'll put that in our transcript for this episode so that people listening can link to that. Researchers, clinicians, we all need to have a voice to ensure the safety of patients. At the same time, we can't ignore the fact that generative AI is here to stay and we need to think about how it can be used responsibly in the world in general, and maybe in the therapeutic space as well. So, any vision for the responsible integration of generative AI? We talked about the risks, but what about perhaps potential benefits that could come?

Dr. Nili Solomonov: I think that's such a great point. And as I've said earlier, this is coming, this is happening, it's here. And we need to become part of the conversation if we are not already part of the conversation, community partners, educators, clinicians, academics, parents, we need to get involved. And just anecdotally here at Cornell with my collaborator, Logan Grosenick, we're thinking and we're working on harnessing these technologies and training models on real patient's data with responsibility, with privacy, with data protection. And I think scientists have a huge role here, not just developing things in academia, but also co-partnering with industry and collaborating and making sure that these models are safe. And our mission is not to have a product that patients would use more and more and more here, right? Our mission is to have a product that is safest, most efficacious, most accessible to patients. And I think that's key, right? And we really can have a lot of power in impacting the direction of this. And so, I think passive recipients of this is not the right call here.

I think we really need to push the development and to offer criticisms in a constructive way, offer alternatives. And similar to what has happened with digital mental health interventions, where the industry is absolutely flooded with digital mental health interventions, but most of them don't work, right? But we do have data-based academic-driven or academically developed interventions that are highly effective. And I think, hopefully, we're going to see a similar direction here.

Dr. Daniel Knoepflmacher: It gets into the market, I guess, because there's competition between academically driven things and things that are driven by investors that may not have that rigor. And it's a whole nother topic that we're not going to get into today. But as we're talking about AI and how it dovetails with your work, you just think about these tremendous risks, but also this great potential. So, let's end on a positive note.

Dr. Nili Solomonov: Yeah, always.

Dr. Daniel Knoepflmacher: Yes. Yeah. I don't know if that's part of Engage and Connect, going to part this episode. What's your vision for the future?

Dr. Nili Solomonov: Yeah. It's a great question. It's probably going to change tomorrow. I'll give you the vision today. I think this is a really exciting time to be in this space. And we have things that I would spend like 60 hours coding two years ago. Now, I can do in 30 seconds, right? And so, I think we have faster and better tools to answer the questions that matter the most to us. And I think we have serious problems to tackle. The world is becoming more and more complex and more and more isolating, speaking of socially rewarding experiences and more difficult to navigate. And things are rapidly changing every single day, right? And we're going to continue to see detrimental rates of depression and suicidality. And when we talk to patients, we hear these challenges being echoed in their experience. They're coming to us, because they're so isolated, because they feel left out of the game because they can't be part of the digital world, or they feel like, everyone has a better life than them because of social media. So, we see even clinical psychiatry, even clinical psychology changing in its nature as the world is changing so rapidly.

And I do hope that's what makes me optimistic, that the research that we're doing to improve, to simplify psychotherapy, to harness these technologies in a good way and a beneficial way, in the way that contribute society to understand how to tailor them to individual patients, to work with communities, to bring therapies to them. I really hope that that these are the things that are going to make a difference. And I want to just end by saying that I spoke a lot about my work as an individual person. It's the cliche of a village, right? Does take a village and I have an incredible team. And one of the things and incredible collaborators here at Cornell, and in the world and in the U.S. And I think one of the things that make me most hopeful, like my brightest days is when my mentees, folks in my lab, and people that I work with share their ideas and their passions and their missions for the future. And I'm just seeing, you know, we talk a lot about dangers of young people and youth, but I'm just seeing so many incredible examples of grit and passion and commitment and brilliance. And those are the things that make me hopeful, and I'm very committed to helping train those brilliant people to surpass whatever we're trying to do here.

Dr. Daniel Knoepflmacher: Well, I mean, those social connections are what are so generative and have made your work so impressive and so impactful. I mean, Nili, it was an absolute pleasure to have you here today and to talk about all of this amazing work and this really important more general topic about AI. Thank you so much for coming and speaking with me.

Dr. Nili Solomonov: Yeah, absolutely. Thank you for having me. It was great being here and I'm excited to continue to talk after this over coffee.

Dr. Daniel Knoepflmacher: Sounds good. And so, our social reward.

Dr. Nili Solomonov: Exactly. Now, we've earned it.

Dr. Daniel Knoepflmacher: Yes, yes.

Dr. Nili Solomonov: Thank you, Daniel.

Dr. Daniel Knoepflmacher: Thank you. And I really am truly in awe of what you've accomplished as a researcher. I mean, in telling your story, I can see how your humanism is married with this tremendous, creative, scientific brain, it's just so impressive to see how that just emanates through in your work. So, I think we may have to have you back again because I anticipate many more breakthroughs in your work.

Dr. Nili Solomonov: Happy to be back. Thank you.

Dr. Daniel Knoepflmacher: Thank you so much. I really appreciate it. I know you're busy, so thank you. I also want to thank all of you out there who've listened to this episode of On The Mind. This, as you know, is the official podcast of the Weill Cornell Medicine Department of Psychiatry. Our podcast is available on all major audio streaming platforms. Now, that includes things like Spotify, Apple Podcast, YouTube, you name it, it's out there. If you like what you heard today, tell your friends. Please give us a rating, subscribe and stay up to date with all of our latest episodes. We'll be back soon with another one for you.

Back to Health promo: Back To Health is your source for the latest in health, wellness, and medical care for the whole family. Our team of world-renowned physicians at Weill Cornell Medicine are having in-depth conversations covering trending health topics, wellness tips, and medical breakthroughs. With the spotlight on our collaborative approach to patient care, the series will present cutting-edge treatments, innovative therapies, as well as real-life stories that will answer common questions for both patients and their caregivers. Subscribe wherever you listen to podcasts. Also, don't forget to rate us five stars.

disclaimer: All information contained in this podcast is intended for informational and educational purposes. The information is not intended nor suited to be a replacement or substitute for professional medical treatment or for professional medical advice relative to a specific medical question or condition. We urge you to always seek the advice of your physician or medical professional with respect to your medical condition or questions. Weill Cornell Medicine makes no warranty, guarantee, or representation as to the accuracy or sufficiency of the information featured in this podcast, and any reliance on such information is done at your own risk.

Participants may have consulting, equity, board membership, or other relationships with pharmaceutical, biotech, or device companies unrelated to their role in this podcast. No payments have been made by any company to endorse any treatments, devices, or procedures. And Weill Cornell Medicine does not endorse, approve, or recommend any product, service, or entity mentioned in this podcast.

Opinions expressed in this podcast are those of the speaker and do not represent the perspectives of Weill Cornell Medicine as an institution.