Episode 7: MindBridge. The Human Role in AI Therapy with Dr. Binyamin Klempner MSW, PhD

Sabrina talks with Dr. Binyamin Klempner MSW, PhD from Tel Aviv about using artificial intelligence in therapy without losing clinical judgment, privacy, or the human relationship. They break down how MindBridge uses structured prompts, therapist oversight, and safety flags to support clients between sessions. Learn about the following:

• Binyamin's path from hypnotherapy influences to building a therapeutic AI tool 
• Turning a structured clinical interview into an AI-guided experience 
• Why “human in the loop” matters more than clever output 
• Two models for use: subscription support vs therapist-integrated check-ins 
• Safety flags for concerning language and the limits of algorithmic detection 
• The gap between what clients tell AI and what they tell therapists 
• Encryption, confidentiality, and sharing only what clinicians need 
• A sacrosanct journal that stays private aside from AI 
• Optional image-based reflections for self-development prompts 
• Between-session continuity, DBT-style coaching possibilities, and shifting client expectations 
• Customizing AI tone to match a therapist’s modality and approach 
Go on to themindbridge.app, or email at admin@themindbridge.app

Transcript

SPEAKER_00 2:06

Hi, welcome to the Therapist Lounge. I'm your host, Sabrina John, licensed clinical social worker. And today I have with me Vinyamin Klempner, social worker over in Tel Aviv Israel. Thanks so much for joining me today, Vinyamin.

SPEAKER_01 2:22

Thank you, Sabrina, for having me.

SPEAKER_00 2:24

I just wanted to discuss such an important topic for clinicians around the use of AI, artificial intelligence in therapy. There's lots of use right now for note-taking and assessment. And with you developing AI software, just so excited to have you on to talk about this topic, wherever you'd like to start, maybe your journey, what got you here and what you're working on.

From Hypnotherapy To An App

Why Human Intuition Still Matters

The Interview That Felt Shockingly Accurate

Two MindBridge Use Models

SPEAKER_01 2:51

Yeah. So so if I could just start at the beginning of the journey, how I sort of develop how I got into therapeutic AI and take it from there. So yeah, so I my training is in Ericksonian hypnotherapy. And I'm a big fan of Milton Erickson and Carl Jung and also Joseph Campbell, who really wasn't a therapist, but I think has been very influential to many therapists. Joseph Campbell's been influential to many different disciplines, including therapy, filmmaking. And what happened was I began putting together a clinical interview. And based on the works of Milton Erickson and Carl Jung and Joseph Campbell and Carl Rogers as well. And then I when I finished, and the the interview was a tool that I was going to use one-on-one with clients. And then I had an idea. Well, I wonder what it would look like as an app. I wonder what this interview would look like if it was, if I was to create an app out of it. So I went into Claude. Claude is an AI like ChatGPT. And I I uploaded the interview questions onto Claude. And the first thing I did was I had it analyze the interview. And then after having it analyze the interview, I asked what would this look like as an app. And it gave me like um a Figma style drawing of what what it what it might look like as an app, what some of the user interface, user experience might look like, how it might feel. And I thought, well, that that's sort of cool. I wonder if I could just build an app. I wonder how easy it would be just to build this into an app. And and I started doing, and I started building it into an app. And as I was doing it, I realized, well, you know, there's something key, there's something crucial, a few crucial things missing here. But one of the crucial, maybe the most crucial element missing here is the human in the loop, is the professional with human intuition, not algorithmic, if we could call it intuition. Algorithm is great. AI uses algorithm, and algorithms are are powerful. And they're effective, but it's not the same as intuition. It's not the same as is uh is a human felt sense. And then I started to think to wonder, well, how can we how can we integrate a human in the loop, a professional, into this app, which I began calling the mind bridge. Because it bridges it bridges the we have a conscious mind and an unconscious mind, and it makes that bridge, it builds a bridge between our unconscious and our conscious minds. And back again from our unconscious minds to our conscious minds. It's that we're constantly going back and forth on that bridge from the conscious to the unconscious and back again. And so I began to build this app with a human in the loop. And the way that that it just sort of sort of it sort of built itself. I mean, as much as I built it, it was sort of like a sculpture. You have, you know, the sculpture, the the the sculptor starts out with just uh a piece of clay. Start this way and that way and this way and that way until something forms. And it's it's this dynamic, you know, this dynamic activity between the artist and the clay. So that's sort of the approach I took building this app. Sort of an artistic approach, almost like like a jam band, you know, or just sort of listening to the other musicians and then giving your own input and then listening to what they do at the input, and then so on and so forth until you have this sound sculpture. So that's sort of the approach I took. It's sort of also the approach I take with clients. And so what developed was a system where you have in, you have, and I I myself, I took this interview. So that's an important, that's important to mention that I plugged it into AI. I I put the interview into ChatGPT and I gave it a prompt. Pretend you're the therapist and I'm the client, and you're asking me these questions, and you're not, you're sticking to the script. You're not going off and asking other questions. One by one, you're asking me these questions, giving me a little, giving me your feedback regarding my answer to each question before moving on to the next question. And then at the end, give me a summary. And it was so powerful, it was a powerful experience. I mean, what can I say? It was like like when I was 16, I took LSD and it was a powerful experience. It was a powerful experience. I showed showed it to my wife after I finished, and she started to cry because she was so moved by it. Wasn't that AI was has any wisdom of its own. It's really based on in large part on the work of Milton Erickson. The client, the person who's answering the questions, they have the inner wisdom and the prompts, the questions just bring out that wisdom. And AI is extremely good at reflection and analysis. That's really what it's good at analysis. So I answered the questions and then it analyzed my answers so profoundly accurately that it like it was like mind opening. And but the problem is, is that there's no human in the loop. So then I decided the the best way to work this, also like economically, it needed to be economical and it needed to be implementable. And that is to have, after this initial interview, to have the AI to allow the user to you to have two 30-minute sessions a week. Or in the case of a therapist who are using it with their clients between sessions, between sessions 10 minutes a day, just to give their client a little extra and then be able to read and comment on what their client is saying.

SPEAKER_00 9:37

Wow. So it's almost like helping to use as a check-in and to use the tools.

Safety Flags And Real Risk

SPEAKER_01 9:44

Exactly. Well, so there's the version that's independent of therapists, in which case the client has 30 minutes twice a week, and one of our therapists then reads what they say once a month or twice a month, and then writes them uh a report, a long letter about their experience having read the client's AI conversations. And then there's the the therapist model, which are the the therapist version of Mind Bridge, which is which is a check-in. And then obviously during their sessions, the therapist is able to discuss with with their client something that might have come up during the AI generated check-in. But another thing I realized is that there needs to be flags. So if the client says something that's concerning, the therapist, their therapist will get flagged regarding what the client said that might be concerning. But you know what? I I took it myself, you know, just to test it out. And uh I was having a difficult day. And I said something like, you know, I feel like I'm in closet, it's pitch black, and I can't even see my hand in front of me. And that got flagged. That got flagged as being concerned. So that yeah, as being a a concern, you know, the the client expresses feeling trapped in a dark closet. This might indicate some depression, something to keep an eye on. About five or six months ago, you probably remember there was a young woman who committed suicide. Uh af uh uh she was having conversations with Chat GPT. That's very tragic. It's very tragic. What one of the things that really struck me about that was that she had a therapist who she was speaking to. And it's possible that she said things to ChatGPT that she wasn't saying to her therapist. So this is about bridging that gap. There are things that people are willing to say to Chat GPT or to AI in general, that there's too much um there's too many, and they're too, there's too many inhibitions to say to their therapist. So this is about bridging that gap and letting the therapist know.

Encryption And Confidentiality Choices

SPEAKER_00 12:10

That's great because that's really missing right now. We see people who are using it this way to seek support, guidance, advice, a problem solve, even for couples, right? And we don't have someone there to get this feedback. Like you said, be the human in the loop to be able to use this information to help identify risks and safety. I know lots of therapists worry about confidentiality and its use or you know, information getting hacked. What are your thoughts on researching this and for your own software?

SPEAKER_01 12:45

For my own software, we've implemented a lot of layers. So the therapist receives the information they need to receive about their client. If like if there's a flag, the therapist needs to know that. But everything else is encrypted. So I don't know how the technology works, but it's that I guess it's sort of like like a tunnel vision, right? Where the therapist gets a clear vision, and then everything outside of that tunnel is blurred. So the client's identity remains confidential between them and the therapist, and everything else is just blurry.

SPEAKER_00 13:28

Right. So it identifies or summarizes maybe the risks or concerns, but it's not bringing up anything else that's talked about. So it I when as you're talking, I imagine it's almost being used like journal writing or, you know, somebody just putting down their thoughts. Um, maybe something we wouldn't even share with anybody else. And again, with intent therapist has access to whatever themes, comments, things that are helpful within the session to pull out.

SPEAKER_01 13:57

What it's really interesting that you bring up journaling because journaling is a whole separate topic. And part of what we've done is create a journal. So part of part of Mindbridge has a journal.

SPEAKER_00 14:10

Amazing.

Art Uploads For Self-Development Insight

SPEAKER_01 14:12

But it is totally it's not even, it's more than confident, it's it's anonymous, meaning the therapist doesn't even get to see the client's journal. And the journal also doesn't get flagged. So if they say something in the journal that is it's it's high risk, AI isn't, we we haven't joined AI. We have, and we're not, we're not going to to join AI with the journal. And the the the reason we we we decided that is because and I'm I'm not I don't have the best memory for things. So I can't like some people can like quote you studies and I'll read something, but then I won't remember where I read it. So I'm not great at quoting studies, but um, but you know, I mean, your listeners can go online and Google, you know what, what I'm talking about and they'll find it. That that studies have shown that journaling needs to be sacrosanct. If a client wants to share their journal, they can do so, but it needs to be sacrosanct. And that means it's not AI, it's not an AI chat. We have AI chat, but the journal that we've created as part of it is not AI chat. There's no AI analysis or summary. What we did do when I had a crazy idea after going on a really great hike in the desert, I was playing around with AI in one of the photos that I took to sort of make it like a George O'Keefe. And I thought, you know, I wonder, I wonder what AI would say about me in relation to this photo that I sort of played with and toyed with and turned into a George O'Keefe style based on like what does it say about my personal self-development? So I plugged it into AI and got a really cool sort of summary about where I am in life based on this uploaded image. And I thought, well, that would be a sort of a cool feature to add in the journaling area. So if a person, if a person uploads a piece of art or something like that, they'll get an AI assessment of where they are in self-development. But if they write words into the into the journal, they won't get anything AI because that's it's beyond words. It's beyond, it's it's beyond commenting on. Somebody's journal is beyond some somebody or something else's comment.

SPEAKER_00 16:51

I think it's such a great part of the app to have because so many people wonder what to write or what to journal about or where to start, even. So having these prompts, whether it's uploading an image or having other ideas to kind of process or to reflect on things, that's great.

Between-Session Continuity For Clients

SPEAKER_01 17:09

Yeah, yeah, it's pretty cool. I mean, it's it's just it's just fun. I don't know. I just I just sort of play around with like I have no idea where I'm gonna go with it always. It's just, you know, I don't know, I'll sort of see something missing and and add it and and or yeah, it's just it's a sculpture.

SPEAKER_00 17:27

So you mentioned it being like a a a sculptor and helpful in in session. What are some ways? What do you envision therapists being able to do with this tool? And then maybe we can go into what clients will be able to benefit from in using AI and MindBridge.

SPEAKER_01 17:42

Well, it's interesting, you know, just with with my own personal experience, not as a therapist, but as a client, I when I started working with my therapist, and I and I've done the same thing with my clients, but when I started working with my therapist, when we started working together, you know, it was a session was once a week, and it was 60 minutes. And as a therapist sessions, my son my sessions tend to be 45 minutes once a week. And what happened with me was I started to to to just ask my my personal therapist things, and he would respond back, just you know, a short paragraph. And it was really meaningful to me. And then when my clients would ask me something, or I encourage them, knowing how it felt as a client, I would encourage my clients, feel free to message me during the week. Don't do it, don't take advantage, but once a week you have something that comes up, I'll I'll send you a one paragraph answer. And that sense of between session continuity was it just it made all the difference in the in the therapeutic relationship. So in this case, it could be every day. The client is able to have just a 10-minute session with AI, the therapist gets a summary. They don't get the exact transcript, they get a summary that's about one paragraph long, so that they stay up to date on a daily basis with with their client if if they want. But if there's something that's really terribly wrong that they're missing in session, like this case that we discussed earlier with this young woman who is uh using Chat CPT, and maybe she missed something, maybe the therapist missed something, maybe something wasn't expressed in therapy that should have been. But this is a way to allow the therapist to catch what might either be being missed or might not be being expressed.

SPEAKER_00 19:49

Right. When you're discussing the use of this tool, I was thinking about how it could apply to different models of intervention, say in the future, and for DBT coaching, where a therapist is there and available by text or phone for support for the client. And they see this being a helpful tool as part of that as well. Yeah, very that would be very interesting.

Adapting AI To Each Therapist Style

SPEAKER_01 20:12

A very helpful tool. You know, it's it's it's about it's about, you know, and and and and also, you know, if we're if we're working with with older clients, they might not have those expectations of uh they might not share the same expectations that younger clients share. A younger client has expect the a younger client who's growing up with AI and has grown up with WhatsApp and all of these technologies, they have a different expectation. They expect to have this sort of continuous conversation. You know, it's pulsating conversation. It's an ebb and flow conversation. Well, and I guess to that extent, so is a weekly therapy session, an ebb and flow conversation that's continuous. But the older, like a client, when I say older, I mean a client that's like around, you know, 50 plus or even 45 plus has the expectation of, you know, it's an ongoing conversation that happens once a week. Whereas a younger therapy, you know, a younger client might have an expectation of, yeah, it might be once a week that I talk to this person, but it's every day that I talk to their, their AI representative, their AI assistant. And the AI is also trained to speak sort of, I don't mean speak in a in a in an actual voice, but to approach the client in the that um in the way that the therapist who's using Mindbridge with her clients would approach their client. So, say a therapist who is greatly influenced by Carl Rogers. might be very gentle and human-centered, person-centered. Whereas a clo a therapist who's been very influenced by um Alfred Ellis might be very confrontational. So it's really, it's it's very, very flexible. It really the the AI is able to take on the therapeutic modality of the of the therapist.

SPEAKER_00 22:26

Right. I see this a step above, let's say an app that provides tools that somebody might, you know, be clicking and scrolling and and and reading reviewing too. It's it's it's a dialogue, it's a summary back, it's it's, you know, more interactive.

Voice And Video Replies Next

SPEAKER_01 22:43

Yeah, it's more interactive. And one thing that I haven't added that I I would like to add soon is so right now the the therapist can respond to the to the client. They they can do it through their own channel that they speak to the client however that is phone, email, whatever. But I would like to include also a feature within the Mind Bridge app. And right now it's a web app. It's it's it's not a it's mobile friendly, but it's it's a web app. I'd like to include a feature where they can record either a voice recording, a voice memo, or a video memo.

SPEAKER_00 23:20

That would be great.

SPEAKER_01 23:22

Yeah.

Structured Prompts Plus Human Oversight

SPEAKER_00 23:22

To use with it. Definitely. And it's so there's so many benefits to help support between sessions to when we're thinking about rewiring the brain, right, with repetition and having this additional tool for support and therapy. How have you yeah how would it benefit a person to use this app, let's say versus an a coping tools app or you know AI separately. What stands out let's say for a a client or a user, a consumer?

SPEAKER_01 23:57

Yeah, yeah. So I think that there's there's a few things. One is um well the the main thing is the professional human oversight. That even if they were well a few things. One is is the prompt engineering. One thing is the prompt engineering that this isn't ChatGPT where they can just I I don't want to call it a free for all, but it is a free for all. This is this is structured. This is the prompts are are engineered to have a certain structure and a certain therapeutic it's to have a a therapeutic feeling environment energy to it. And and we were very careful in our structuring and engineering the prompts. That's one thing. And then the other thing is of course the human oversight that whether it's a therapist using Mindbridge with their own clients or it's somebody who's using Mindbridge as a um as a subscription product for their own their own self-development there's always a human looking over what's happening. Even if it's not under a microscope even if they're not getting transcripts if there is an issue if there's if something's being flagged there always needs to be a human that is aware of that that has contact information for the client who can call the client and make sure the check in with the client are you okay and to have a a telephone number to have the client's telephone number because through the telephone number can be hopefully if the client is in really dire straits hope you know help found tracked through the telephone number. So that might that might sound invasive, but if somebody wants to commit suicide we have to do what we can to prevent that.

Beta Testing And How To Connect

SPEAKER_00 25:52

It's really great to know and I think it's really helpful for clinicians out there to hear there's that support attached to it or people who want to use it. So you had shared with me uh prior just it's open for beta testing um if people are interested or are counselors in their practice.

SPEAKER_01 26:09

Absolutely yeah we're we're open to starting with with 20 beta testers who are therapists as well as we'd like to start beta testing Mindbridge for HR so if there's anybody listening who is in HR, we're also interested in testing and beta testing HR.

SPEAKER_00 26:32

Wow yeah so if anybody wants further information around the use of of your tool how it could support ways to use it in their practice what would be the best way to reach you and and to connect or to continue to learn learn more information about your journey and get updates.

SPEAKER_01 26:49

Yeah for sure the best way is to go on to themindbridge dot app, themindbridge dot app, or email me at admin at the mindbridge dot app.

SPEAKER_00 27:02

Thank you so much for sharing such a great useful AI software tool and and I think it's so helpful that you know it's created to find safety flags and and have that human experience and and connection in the loop as you said. And thanks so much for sharing um and I love I appreciate your time today.

SPEAKER_01 27:21

Thank you so much for having me and giving me the time