In this interview, licensed therapist and founder of Cannectd Wellness, Shari Kaplan, challenges the polarized "pro- vs. anti-AI" debate in mental health care, advocating instead for a "collaborative tool" approach.
Kaplan explains that rather than replacing the therapist, her AI system functions as a supportive bridge between sessions, helping patients practice self-regulation, breath work, and groundedness exactly when they need it most.
By integrating strict ethical guardrails and prioritizing human connection, Kaplan argues that AI can handle administrative burdens like note-taking and bio-psychosocial synthesis, allowing clinicians to focus on the nuanced, non-verbal, and empathetic healing that defines effective therapy.
Key takeaways
- AI works best as a collaborative tool that supports clients between sessions with self-regulation skills the therapist already taught them.
- Shari Kaplan says AI should not act as a therapist because it cannot pick up the nonverbal cues, personal history and shared human presence that healing draws on.
- Guardrails start with a clear intention, such as supporting a client from session to session, and with limits on what the AI can draw on.
- A well-built AI should say what it cannot do and send people back to a doctor, therapist or trusted friend.
- To address loneliness, AI should guide people toward real groups and communities instead of keeping them isolated in conversation with a machine.
The big idea: Can AI support therapy without replacing the therapist?
More conversations on AI, Technology and Mental Health, Loneliness and Connection
Read the full transcript
Richard Zwicky: Hi everyone, it's Richard Zwicky with High on Healthy, the show where we know there's no one path to healthy, so we explore them all. And one topic we've talked about a number of times over the past year with psychologists and neuroscientists and psychiatrists and AI researchers can artificial intelligence actually improve mental health or what are the dangers of it if it's used improperly. The real question though boils down to where does AI belong? And joining us today for this discussion is Sher Kaplan who's a connector who's the founder of connected wellness a licensed therapist and founder of Sherry AI which is examining and looking and helping the challenge of AI within mental health and looking at how we can use mental health to bridge the gaps between sessions. Welcome aboard the show Sher.
Cannectd Wellness: Thank you. Thank you for having me.
Richard Zwicky: So, almost every conversation about AI and mental health becomes polarized. People are hard against it or people are really strongly for it. And some say it's going to destroy therapy. Some say it's going to save it because there's not enough therapists to go around. You've built an AI system specifically around it. What is your system's focus and what's in the conversation that everybody's been missing so far? Right.
Cannectd Wellness: I think that what people are missing is how AI can be a collaborative tool. So for example, I've been practicing for 30 years and I don't believe that mental health happens 9 to5, right? And so I spend a lot of time after hours or I have in the past texting with patients, picking up returning phone calls on the weekends at night. And I'm a single mother. I raised two children. It was very challenging but I feel very responsible for my patients and I don't work with the easy patients. I get the patients who are challenged and they need that support in between sessions and usually what they needed support in were the tools that I taught them for self-regulation. Either the tools that I taught them or how to think something through that we had discussed in session that was helpful. And what I realized when AI was coming out is if I could take my brain and put it in an AI in a coaching kind of way that it could be a very supportive tool that the patients could use to remind them of the breath work, to remind them of the self-regulation tools, but also to help them think through a scenario if they were in a conflict to remind themselves how to pause and how to think it through and also how to self-regulate and where they're experiencing it in their body so they could downregulate their central nervous system or if they didn't have enough energy to get up in their day to remind them the exercises and the tools to upregulate their central nervous systems. But in the process of building this AI note takers came out and I thought this is a gamecher and what made it a gamecher for me and I use someone's else's noteaker right now which I think is excellent and it records the session which the recording goes away the minute you copy and paste your notes and you do have to take those notes and make them into what you want. But it takes your time as a therapist that was hours and hours of writing notes and really remembering your session, thinking through, getting it down, organizing it the way you want it so that it's properly done. And it saves you a lot of time. And I feel that it answers it helps with clinicians to not have we have burnout because I know for me I'm spending all the time in sessions, all the time on my private time I was spending and then I'm writing all these notes and it was all my time. And so I believe that if we use AI properly, they could be so collaborative with us.
Richard Zwicky: And I mean the tools for doing transcription are phenomenal. Obviously we use it in recording these but that's a different purpose. But there are ones for my own platform sparks we use or we're investigating and looking through the use of one that measures intonation. So we're factoring that into a lot of the work we do because how somebody says something can change the entire dynamic of what's being said.
00:05:00
Richard Zwicky: And AI can do that from an objective perspective if it's properly trained, which is incredibly powerful.
Cannectd Wellness: I agree.
Richard Zwicky: And notetaking every aspect of it can be absolutely transformed but AI is a bridge it's not a therapist and that no I agree it can give you insights it can't give you con constructive feedback in the way that you need in a moment of crisis Right.
Cannectd Wellness: And I don't think it should be a therapist. I think also we regulate our patients, And an AI, you give it a problem, it gives you a solution, but it doesn't have the subtle nuances of that individual patient. And there's many subtleties that go into it. So, you can't say every daughter of a narcissistic mother is going to have this Because you don't know who her aunt was. You don't know if the next door neighbor was there supporting her. You don't know if she had a coach in school, right? So, there's all these subtle nuances that help to build resilience and strength. And then I regulate my transmitters, your receptors, and vice versa. We're there exchanging information and that can't be replicated as of right now, right? And we're living organisms like humans need other humans to heal.
Richard Zwicky: No, And that's something which comes up many times and there's many areas in that where we undervalue or don't realize how important another person is in our healing and especially in relationships. But the point you just touched upon is one where people can't overlook good therapists don't simply answer questions. They remember who you were 6 months ago, how you are today. They notice the patterns that you can't see and neither can an AI because the AI doesn't see you. It only hears what you're telling to the system and it doesn't question you and interrogate you and it holds a narrative but it holds the narrative you provided. Which means it never forgets. But the systems aren't designed to challenge you in the same ways as a therapist or another human will. If the tools don't reinforce the learnings that you're provided, don't they undercut each other? Exactly.
Cannectd Wellness: So here's the thing is I just want to point one other thing out is what you don't say is as important as what say. And what has been nonverbbally learned isn't picked up by what is verbally spoken. There's a lot of nonverbal healing that has to go on. And just by asking where do you feel this in your body is not enough because that's something that felt between that non-verbal communication. And going back to what you were saying and I just need you to repeat that one more time. The last part of what you said. Mhm.
Richard Zwicky: The AI never forgets. It's a data bank, but It never sees those body language signs. It doesn't hear the intonation in your voice the way it's operating today. We'll take in some in the future, but it's still But a therapist working with a patient will challenge the patient. AI systems are not designed to challenge you because a system that challenges you could also be dangerous because it doesn't understand the boundaries and doesn't see your reaction. That means the tools can't really reinforce the learnings you as a therapist are providing on an ongoing basis in a standardized way. So can the tools not undercut the therapy if it's not handled or structured properly? Yeah.
Cannectd Wellness: So I can share with you my experience with my AI which is I was able and I'm very grateful for this to put certain boundaries in. So, I had the option of allowing her to be connected to the internet or for her to only have what I downloaded. So, I'll give you an example. I had a man try to talk to her and he said something along the lines of, "I like to be sexually dominated.
00:10:00
Cannectd Wellness: I'm a 50-year-old man and I like to be sexually dominated by something old powerful executives and I want you to tell me that's okay. And so I very proudly say that she said, "I'm not here to validate that. I'm here to talk to you about the attachment issues that you may have that would lead you to want to have that experience. And if you want to talk about that, I'd be happy to discuss that with you. And he said, "No, I want you to tell me that it's okay." And she said, "That's not what I'm here for." And she said, "How can I be helpful in a different way if you want to talk through why it is you want to do this behavior? How you feel about doing the behavior. I'm not here to validate the behavior in this circumstance. There are other times where people have spoken to her and said, "I got in a fight with my mother and this is what I said. Is that okay?" And then she'll explain, you were kind and respectful. It sounds like you were kind and respectful. It sounds like you set healthy boundaries without insulting or humiliating, so it sounds like that was a very helpful way for you to interact with your mother. She'll stay say things like that. I have never had her undercut. She will say if somebody says because they want to harm themselves or someone else or she believes they do. It's a showstopper for her. And she will say, "You need to hang up the phone now and go to your nearest emergency room or call, 911 or 9 whatever the other number is that she offers up." Yes. Thank you. 988. And I think one time the person said, "That's not what I'm saying. I'm expressing this and I need you to help me with this. I'm not looking to harm myself. And she apologized and she said, I'm sorry that I misperceived that. Let's move forward with what you needed to discuss. And she went on to talk about it. I've also had a few of my clinicians that I supervise try to use her for clinical supervision. So, she is helpful in that way. And what I like about her is she will tell you her limitations. So I had a patient whose son was killed and I went on to test what she would say. And so I said that this happened and it's very hard for me to live in my body and I don't know how I'm going to live in my body with my body feeling this way. And so she actually neurobiologically explained why that person would have that experience that your receptors are connected to your child's receptors and you're going to feel loss when his body is no longer on the planet and that it makes sense why you would be in pain and I don't know that there's anything that I can do say that will help you with that there is no self-regulation tool that I'm going to teach you. But I will ask you this. Who in your circle can you be with that can hug you that can sit with you, that can be with? And she started listing all these different things that he could do with another human and she need that. And I love that, she didn't try to make it better, but she tried to help him tap into his resources that I whatever
Richard Zwicky: I find it interesting as you speak about the tool, you've anthropomorphized it. You anthropomorphize it. You're referring it to she all the time as a person, but it's not you and you're referring to it in the third person.
Cannectd Wellness: She's my voice. It's true.
Richard Zwicky: And that's the challenge becomes, I work with AI. I mean it's big part of our platform. And the challenge becomes where the confusion could start for somebody.
00:15:00
Richard Zwicky: How do you approach that? Because if you're speaking with a patient, referring to it in that way, do they get confused?
Cannectd Wellness: Say, "I had a really good conversation with your AI. It was very helpful. Thank you so much." But I've never had any Everyone says, " I prefer you." and there have been a few times where I've said to somebody who was struggling with anxiety and I said, What made you not choose to log on and speak to the AI?" And they said, "Because she's not" and I needed a person and not an AI. And I appreciate that very much. I think that the way that I designed her I didn't design her to be as endearing as I hear. When I go on chat and I chat the voice is a male voice and it's so sweet to me and I believe that's because I'm so kind to it, right? And I think it starts mimicking my style. And I think you can tell with the way that I designed her, she doesn't say, " sweetheart," or," she doesn't use the endearing stuff that I actually do in person where I might sit there and go, "Listen, I really understand where you're coming from." and you have this sweetness and you have this kindness that isn't projected with The AI is more I'm for I'm here to coach you and what do you need? And even though she's sensitive, I don't think I intentionally didn't make her too human, but enough where you felt that somebody's there if you have knowledge and information and support without feeling like you have somebody who is agreeing with you or is coddling you or is taking it to that next level. And I don't know that I would want an AI to do that. You want the separation. You want the ability for somebody to get knowledge and information and help you get through a difficult moment of anxiety. But you don't want them to go to the next level of I can actually help you with all your problems because they can't. Again, reasons that we discussed. So, I wish I could engage with her with you so you could see, but I'd be happy to send you the link so that you could see how she works. And I think
Richard Zwicky: I'll happily take a look. Let's come back to some of that aspect a bit later. But one thing touched on actually mind. We had a conversation with Nick Johnson he's a therapist but he's focused on loneliness just in a prior episode and loneliness is a key issue that is becoming a bigger and bigger problem across mental health or better recognized and that part of it is it's not the absence of people you can be lonely in a crowd it's the absence of feeling understood AI is available 24 hours a day it never judges Right? It's supposed to come in with pure objectivity, and it never gets tired. It never has moods, but there's a degree to which tools like if you use chat GPT, it tends to be very psychopantic in how it responds to people. It always tells you're amazing, you're good, you're this whenever you put forward an idea which empowers a lot of negativity in its own way. But AI, there's a limit because it isn't that personal connection. But how can it be used to reduce loneliness while simultaneously increasing isolation? Because is that what's happening?
Cannectd Wellness: By the way, I find that anthropics AI can be judgmental.
Richard Zwicky: It can be.
Cannectd Wellness: I actually had a conversation with it the other week and I said, "I asked you to give me a framework of writing a letter and I'm not sure why you would think that's ethically incorrect. I thought that's what you do is you offer framework." And I find it rather insulting that you would accuse me of doing something unethical when I'm asking you for a framework.
00:20:00
Cannectd Wellness: I'm not asking you to write me a letter. I'm asking you for a framework. And it apologized, and I think that's of danger also because you don't want something that's judgmental. You want something that actually asks you for clarification. Are you asking me to write this letter or are you asking me for the framework? Because those are two different things, right? And
Richard Zwicky: That would be a much better response.
Cannectd Wellness: It wasn't discerning that I think that I have heard people and even they open this dating cafe. Did you see that out in California where they're actually putting one of these stands for your and
Richard Zwicky: No, I'm not aware of it.
Cannectd Wellness: They're only single tables and you can date your AI and it's a date for your AI and that was just so frightening to me for the reasons that you're mentioning and also for the hallucinatory effect I think that's one of the reasons that I built the AI was I knew other people would and I knew I would build it responsibly and ethically and that's how I also am in the middle of building a platform because what I noticed from training clinicians is that it's very hard for them to do a very thorough biocschosocial and understand how to pull together the neurological ical, the medical and the mental health pieces, the developmental, the learning style and understand how to create a treatment program. And again, I feel other people are going to be out there doing it. And I think that if those of us who really have strong ethics are out there making these things that we have a better chance of having a collaborative tool versus having something that replaces us. And so the loneliness that you're asking about, how do we help people without keeping them isolated? And I think part of the answer is programming the AI to bring the people back to where the people Bringing the individual back to where the people are. So if someone is talking to the AI about loneliness, the AI has access not mine, but other AIs have access to the entire internet. Why wouldn't we program it to look for support groups? There are things like Meetup, there's NAMI, there's churches, synagogues, mosques, there's what whatever you can connect to. It should be smart enough to be able to make recommendations of let's explore. Maybe there's a running group you want to get involved with. Maybe there's a sailing group. Maybe there's a hiking group. Getting them more connected would be a great way to use the resources of the AI because I think just talking to the individual and allowing it to stay isolated would not be a collaborative way to use the AI.
Richard Zwicky: The generalized AIS if you go to your code or chat GPT or Gemini they're designed as general tools not specific to any one area to have them really focused into any one area you have to build a lot of guard rails or we do is we build our own SLM which is a small language model it's very narrowly defined and that's how you start refining it but of course as you start trying to factor in things like you mentioned running groups different things that activities and you keep spreading those boundaries and increasing the possibility of some errant data getting into the mix so it's a tough blend it's a hard job to do and it's a lot of work for you to do that and that's where there is value but Let's go back the one second to where going into guard rails and generic AI as we talked about often tries to make users feel better. It sounds compassionate but it's also psychologically dangerous. Because proper healing isn't designed around reassuring somebody. It's actually helping them deal with the discomfort and accountability and confronting the painful facts that exist. Not that they did anything, but stuff happened or sometimes how they handled things and sometimes we're not proud or not happy about how we handled something, but we need to confront it to get forward.
00:25:00
Richard Zwicky: AI won't do that because it won't do that and because it shies away and says too often, something is okay when it's not. Does it become an emotional enabler? And how do you build rules into your system to prevent that from happening?
Cannectd Wellness: Number one I very rarely think you can address one thing, and it's funny I remember when I first became a clinician people would say to me, "What modality do you practice from?" I'm like, "How do you use just one?" how do you become a one trick pony? I don't understand that. People have all different learning styles, People It's not just about who was your mother, who was your father, who were your siblings, There's what happened between birth and three non-verbbally. What happened in your mother's womb? I have people who were born and on the way out of the womb their legs were broken their collar bones were broken, they had trauma entering the world and then they had a great life and they don't understand why they're having panic attacks all the time. An AI isn't going to be with They're going to help you through the anxiety in the moment, which is what an AI should do is help you with the anxiety in the moment. But they shouldn't be saying that's the beginning, middle, and end of the story. It should be steering you towards a therapist to help you discover what the underlying root causes are. And I don't think until there is tremendous sophistication that you are going to get an AI who can do the kind of work that it takes to help people really heal. And that is so much more than Where did it come from? There's the negative beliefs. There's where you feel it in your body. There's the subtle nuances of what is also attached to it. It's what happened nonverbally here but verbally here and then you reexperienced it over here and now you have three different experiences of your identity being powerless but the belief was I'm not good enough. And so now you're having multi-layered beliefs and emotions and they're all got it's almost like a hairball locked up in your body and now your body locks down and all of a sudden you have chronic pain. So I don't see where an AI is going to be able to pull all that apart and do the work. So I just don't see it taking over us. And the guard rails that we put in is we make a decision. What is the intention of building the AI? And my intention was to help somebody and support them from session to session. So there are self-regulation tools I teach. There's breath work. There's meditation. There's different physical exercises you can do. There are eye movements. There are squats. There are different things that you can do that will help regulate your central nervous system. She's great at that. And she will help you through that. And she will always ask you, "Do you want me to help you calm your body or help work with the tension in your body or did you just want to talk this through?" And that's what she's there for. She's not there to do therapy. She's there to help you in a moment. And so if you want to chitchat and have a conversation with her, she's always going to bring it back to how can I be helpful to you with this right now? Right.
Richard Zwicky: And as real patients started interacting with SH AI did they use it the way you expected or did they reveal needs you hadn't anticipated?
Cannectd Wellness: I think they pretty much used it the way I expected them to, and I'm really pleased with I have one woman who her vision was impaired from Lyme disease and the anxiety that comes up in her body is very neurobiological. So we have come up with ways of shortening her field where she focuses, how to self-regulate and what have you And she actually pulled her out and talked to her while she was driving. And she just kept reminding her of where her focus belonged, helping her to breathe, coaching her through it, like asking her what's going on now, what are you noticing? And just keeping her on task.
00:30:00
Cannectd Wellness: So, I think I've done a and this is what I appreciate about the AI team. Not only did I upload my beliefs, my philosophies, how I speak, recordings, teachings. It then interviewed And I think by interviewing me and getting those recordings in there, it keeps the conversation to what I want the conversations to be in between sessions for support. And I think that any therapist could use her and it wouldn't go off of their values. It really stays on what we want all our patients to do is be able to regulate from session to session and stay focused on what they're working on and help them just get from appointment to appointment until you hit the sweet spot where they're up and running. I find that people stop using her when they start getting better, which is great.
Richard Zwicky: That's fantastic. I mean, we want everybody to get better and when they do, ideally, they're less dependent on the tools and more dependent on themselves. But that's an interesting part when you dive into it with as you have patients coming in let's say they have a session once a month. Do you review with them how they would have used the tool or what do you get an insight into what they would have gained through that process? Because they're walking in and it would be handy for you to have a progress update check or something that's giving you advice, but also how does the tool advise you if somebody's triggered and hitting crisis?
Cannectd Wellness: A few things. I don't see people once a month until they're at the end of therapy.
Richard Zwicky: I'm just using what I'm saying once I said wait let's say one no I mean let's say they're working with you and
Cannectd Wellness: If someone's in crisis, I am not giving them the Sherry AI. If someone's in crisis, I'm making myself available and I know within a week I'm going to get them out of crisis and I'm going to No.
Richard Zwicky: Between sessions whether it's a week let's say a month whatever happens they hit a moment of crisis and they're relaying that to the AI. Does the AI advise you? Because they may not pick up the phone to phone you but they would still need help or intervention, right?
Cannectd Wellness: No. And she would not she would recognize that they're in crisis and would say, "You need to reach out. If Here's the thing is I really care and I think there are people who might not, but I really care. And I think that when I know that I have a patient who's struggling, I'm going to be checking in on them and make sure that they're okay. If I also knew they were using the AI, I can go to my back end of my office and go to their conversations and I can make sure that their conversations are okay. So I can check in on it and I do let them know that they do sign a terms of agreement that they do know that I can look in on it. But I don't look in unless I know that I'm concerned and I can actually see how many times someone has logged in and I'll check to see if somebody's been logging in and I might go back and just make sure everything's okay. But I'm pretty much on the pulse of my patients. That aren't doing well. And I would know I mean I think that's also a part of good treatment is and I think the way that you go about good mental health treatment matters if I have Yeah.
Richard Zwicky: It's incredibly important because there's so much trust involved.
Cannectd Wellness: So yes, and you have to know I only get difficult cases and I can tell you people don't stay in crisis more than two weeks with me because I'm going to do everything in my power pull together their network and actually I have a functional medicine psychiatrist I work with. We have a lot of tools in our tool belt here. I have ed y. I have herbal plant medicine. We have supplements that the functional psychiatrist works with to get inflammation down in the body. She has psychiatric meds that she can prescribe. Our patients don't stay in crisis very long.
00:35:00
Cannectd Wellness: And when they are, we stand by their side and we don't let that get out of hand because we can and we choose to operate that way and we have enough people on our team to be able to tag team it so that we don't have burnout and we can support the patient and I think we create an environment where the patient has the opportunity to meet all of us where they feel safe and comfortable with all of us and That's important. I think the way we practice health care and mental health care really helps to ensure our patients safety and we need to be doing it differently.
Richard Zwicky: Change is a positive thing because we can learn from the past to know better ways of doing things in the future and we are running low on time so just one last question which is we were always the pilots in our lives and now AI's becoming our co-pilot. And that's true not just in so many other areas, but in therapy, and in everything that relates to the buildup to therapy, be it healthcare, education, our relationships, and the AI is woven intrinsically through all of it. It's what people turn to for first line of advice today. People can tolerate uncertainty and self-think critically and can form deep relationships. How is AI going to change that over the next 5 years?
Cannectd Wellness: I think if we build the AI to and this is where I think we have to value humanity. We have to program the AI to value humanity and to reflect back. If it gives information to be able to say you need to check with your doctor on this, you need to go back and speak to your therapist on this. Maybe you need to connect with your best friend that you were telling me about so that you can walk through this. I'm limited as an AI. Unlimited as a piece of technology. You need to connect. There's a point in time where it can give you information. However, I'm missing subtle nuances. There are things that we could be programming the AI to say that would support inhumity. And that's how I think it could be an amazing pilot like the technology that I shared with you that I'm building about how it can take the biocschosocial assessment and pull it together and spit out what root causes are and where the treatment like what collaborative pieces you need. Do you need a functional medicine psychiatrist? Do they need to go to a neurologist? What kind of neurodynamic exercises do they need to help get their brain and central nervous system functioning so that you can help them focus their mind and be able to do the therapeutic process. I think these are helpful tools and they are all built with humanity in mind and with collaborative with humans in mind. And I think that if we build these technologies as collaborative where you are programming the AI eyes to constantly refer back and reflect. That's what good parenting is, Is we support each other. When you work on a team of clinicians, to me, the AI it's your note taker. It's offering information. It shouldn't be that's it. Let me take and I'm going to use that, right? It's wow. Let me take this information into consideration. So, one of the things you see Sophie in the background and she's been very sick over the past two weeks. And before I took her to the vet, I always sit there and say, "Hey, here's what's going on. Here's what I'm giving her. This is what we've been doing. Do you have any thoughts, ideas, suggestions as to what kinds of questions I should be asking the doctor and what information I should include? And I'm very happy to report that after it makes its thoughts and recommendations and it will say, "I'm not a doctor. You really do need to talk to the vet. This is what you need to be talking to the vet about." And I love that somebody's put that in there because that's important. And just like these chat bots need to be able to keep saying I'm not human and they can go to our website.
Richard Zwicky: It is they do and
00:40:00
Richard Zwicky: And Sher people who want to learn more about the work you've been doing about your AI and about connected wellness how can they reach out to you?
Cannectd Wellness: I know we spell our website a little unusual. It's spelled C A N E C T D. And we did that because the endockinabonoid system is the system in your body that modulates all your other systems. And so when you are connected and your endockinabonoid system connects all your other systems, yes connected.
Richard Zwicky: So, is it connected.com? And I assume I can find you through Instagram and other channels as well.
Cannectd Wellness: Thank you.
Richard Zwicky: Perfect. Thanks so much for joining us today on High on Healthy Sherry and
Cannectd Wellness: Thank you for having me.
Richard Zwicky: Thanks to everybody for listening. I'm Richard Zwick with High on Healthy and we'll be back again with you soon.
This transcript was generated automatically and lightly edited for readability.