Preparing for the AI Shift in Therapy with Dr. Jordan Harris

welcome to Mental Health Evolution, a podcast about what's changing in mental health and why it matters. I'm your host, Rachel Harrison, inviting you into honest conversations with people from all perspectives in the field. Clinicians, tech founders, investors, insurance companies and all the folks in between. Let's explore what's working, what's not, and what's next. Welcome back everyone to the Mental Health Evolution podcast where we are talking about how the landscape is quickly evolving in the mental health industry. Today we are taking a look at none other than AI. It is a hot topic really everywhere, but especially in the mental health world. And Our guest today, Dr. Jordan Harris, has researched and written about what is happening with AI. He is is also a therapist and he coaches therapy practices in their growth and just figuring out what to do with artificial intelligence in general. As always, I want to give a little background here to set the stage for our conversation so you have some information about the topics we are addressing. Here are a couple things that have been in the News. First, from telehealth.org we have the article about Amazon Health Services and Talkspace, partner for Virtual therapy. So this is a corporate partnership between these two companies and the article outlines the the elements of that partnership as well as the implication that these two companies plan to use recorded sessions and written interactions with clients and therapists in order to train AI to take over the therapist role. They're not making this a secret. That's kind of something that's happening in a lot of forums and this is just one of those. Then I also want to reference an article from Fortune magazine. This was pretty well distributed article and it the title says in a showdown of psychotherapists versus chatgpt the latter wins. New Study finds this study was interesting. It set out to tell if people could tell the difference between ChatGPT, that therapist and their therapist in which they preferred. So they found that participants had difficulty identifying whether responses were written by ChatGPT or by therapists. And they correctly guessed that the therapist was the author about 56% of the time. And they correctly guessed that ChatGPT was the author about 51% of the time. So kind of a 5050 split there whether they could tell if it was ChatGPT or the therapist. And then further, the article goes on to say that in most cases that the clients preferred ChatGPT's take on the matter at hand based on how they understood the speaker showed empathy. What they said was Appropriate and what a good therapist would say. Of note, this was specifically couples therapy and this was written responses between either a therapist or ChatGPT. So not sort of the full AI presentation here. And lastly, I want to reference the article that kind of was the springboard for this episode of the podcast. Podcast. And that's Dr. Jordan's article. And he wrote an article called AI is Reshaping Psychotherapy. We have two years to get ready. He outlines in this article that there will be four big players in the AI therapy space. Insurance companies, model based apps, big box provider platforms, and general large language models. He also talks through the idea of autonomous AI being a co pilot in therapy and potentially helping us with process coding, meaning that we would receive real time feedback to enhance attunement and effectiveness in the therapy room. His article is robust and there were a lot more pieces. That's just me pulling out a couple of highlights there so that our listeners kind of have some background with what it is that we're referencing and what we're going to be talking about. So with that I would like to open up the conversation with our guest today and talk through some of these events as well as looking at how to make mean meaning of all of this from the perspective of clients, clinicians and businesses in this space. So first I want to say welcome, Jordan. Thanks for being here.

Hi. Thanks for having me. Glad to be here.

Yeah. So let's look at the overall patterns happening with AI in the mental health space right now. What are you seeing so far?

Yeah, I mean there's sort of two things that I. That I'm seeing. One is. Well, one is that no one knows what's happening. This thing is moving so fast. Every week there's a new thing happening, particularly in the mental health space. People are really trying to figure out like what are the limits of what we should allow it to do.

Right.

So it can do a lot, but like should we allow it to record sessions? Should we not? We don't. Those are all questions that we're trying to figure out. And so with, with that, a lot of what I'm also seeing is the first thing that a lot of people are testing out is this X ray note taking. Right. Can this thing do my notes beyond that, I mean, I think there's a few different ways that the field is sort of unfolding, but we're trying to figure out as a field and I think as a society, like what is this thing and what's it going to do?

Yeah. Do you have a particular stance on this whole idea of AI taking notes for us in session.

Yeah. So, you know, a lot of people think that it's unethical, that it violates privacy and confidentiality sort of things. There's, there's, there's two things to that in my mind. One is that I don't think that it does. Right. Like you have a duty to do informed consent, but in as long as you're doing that, there's no reason why you couldn't have it record your sessions unless there's some sort of law specific to your state. If you think about like, like for me, I went to college and all of our sessions were recorded and we review them with our supervisors. Now, this is, this is back in the olden days, old school. Met in person. You met in person to me.

Yeah. We didn't store it to the cloud, we just destroyed it. Right on the. Yeah, right.

So like. But we had it in our informed consent. People knew them. When it came to the university clinic where I was learning that they were going to be recorded and watched, and that's what happened. The other thing is, and this is, I think, the thing that's important because the field is moving so fast, I don't think regulatory bodies, which are usually like AAmft or the APA or the state boards can move fast enough to tell people what to do. We might see some like, standards of care type stuff, but like, you know, my state board is struggling to get us updated onto a digital platform for license renewal. So I don't think they're going to be telling me about how to use AI anytime soon. Like, it's just not going to happen.

No, I think you're right about that.

Yeah.

But I, I am one of those people that. I do think that there are ethical risks to doing this because once that our, once that recording exists, we don't have control at that point. I hear your client consent piece, but I also sort of wonder if clients know fully what they're consenting to from the perspective of, I've talked to several different AI companies about, like, what happens. Right. And they all will say, well, some of them destroy recording, sure, but all of them data mine, the recording. So there's this whole nuance between, okay, maybe the actual recording is destroyed, but what about the data mine. And my additional concern is what if policies change and they decide it doesn't benefit them as a company to destroy these anymore? What happens if attorneys in a case get wind that there's information they can access? You know, there are all These protections may be there to just think about. I think for me, rolling out anything like this, we need ethics in our field. Privacy and confidentiality is the whole reason why people are safe enough to come and talk to us. Right? So that's just for me. That's just one piece that I wonder about with AI note taking.

When I'm thinking about ethics, I'm thinking about, like, what are the actual ethical codes I have to adhere to and what are my state legislators saying. And so to, to my mind from this, this question, it's. There's nothing in the ethical code that I have to adhere to or the state legislator now, like morally, what's maybe best for the practice for my clients? I think your questions are totally valid. The only thing I would say, the only thing I would say is most of our clients are already giving away their data and they don't care. So if I have a client who says this wigs me out, I don't want to do it, then I'm not going to do it. Right. But, you know, like, even in the article that I wrote, they surveyed about 500 clients and like 98% of them are already using ChatGPT or some other sort of large language model. People, I mean, if, if you're on social media, like if you're on X or LinkedIn, like, they're already mining your data. So I don't think people, I think we care a lot about that for other people, but for ourselves, most of us just give our data away anyway.

I mean, that's a, that's a fair point. I think that is happening everywhere, all the time, but maybe not with the content that people are sharing in therapy.

I mean, that's a good point. The other thing, and this is my bias, you know, my background is really in clinical expertise. How do we become a better therapist? And one of the things I think we see across fields, there's two big ways to become better at any field. And one of those ways is through the data. Like, if you get better analytics, whether you're a pilot in an airplane or a baseball player, or even comedians, I've seen comedians, like, track the amount of flash they get. Like, they just get better at the field. So personally, I want to help people. And again, if my client says that this is not for me, I'm not, I'm not going to force this on them at all. But I will take what tools I can help. I can get to help people better.

Yeah, I hear that. That's understandable. I would love to talk about this whole idea. I mentioned it in one of the articles about that AI or ChatGPT might be a better therapist to a human therap. And I would love to hear some of your thoughts about that.

Yeah. So, I mean, this goes back two things to my mind. So last year, I. That's my. My whole push in my actual academic career is expertise, clinical expertise. And so a buddy of mine who has written a ton of books on clinical expertise recommended me to an article by two researchers, Henny Westra and Alyssa DeBartolomeo, about clinical expertise. And basically what they found is they specialize in studying resistance. And what they mean by resistance is any moment where the client says, this is not a good fit for me therapist. And what they basically found is that most therapists, not only do they misrespond, they don't notice those moments are actually happening. They just don't see them. And so I actually. I mean, this is kind of extreme, but I paid Alyssa to coach me for about a year so that I could see this dynamic. And it's everywhere. It happens a lot. So when we're talking about AI and therapy, one of the big things that I see online is people saying things like, man, this AI, it's sort of too nice to other people. And on the one hand, if someone's, like, in a real sort of delusion, right? If they're really in a place where they're really at risk for themselves and for other people, like, that's not good. And I have friends who are working on this problem now of how do we actually make these systems safe? And a lot of systems, not. Not all large language models are safe, but some of them are actually pretty safe already. But the other side of this is, is that this thing that AI does, where it is consistently validating, is actually the antidote to a lot of resistance. Now, you have to be careful that you don't validate poor behavior, but validating positive intent, validating underlying emotions, and being relentless in it. I mean, from the research that I've done and the coaching that I've gotten, we don't do enough of that. And I think that's one of the key reasons why this. These. These large language models are so persuasive, because they are consistently validating. I mean, even back to the old. To me, this all is like, sort of a weird conversation to have because, like, 60 years ago, you had Eliza, which was the first chatbot, and all it did was Rogerian reflection. And people loved it. Like, they lost their minds about how amazing it was, right?

Yeah.

So now we have that, but like a million X. Like yes, it's really good because when you're working with people in these really tender spots, they need a lot of support. And so, and especially when, when most therapists don't really see that they're not giving their clients as much support as the client would really like, it's no surprise to me that these models are doing really, really well.

Yeah, but there, I mean, I guess there's also the other piece of that I've heard it called psychophantic bias. Right. That, that chat a yes man. That AI is a yes man. That's kind of what you're saying, right. And that if you ask it to, if you say, I don't like that response, give me a different one, they'll give you a 180 response. I think that's fascinating. So we're almost, there's, I guess for me it takes away that human element in the room of psychotherapy where we are talking and putting out who we are and there is another human being taking that in and feeding that back and some way, right. Basic attachment, kind of human interaction. And maybe we lose that if we only get exactly what we're wanting in response.

You're, you're hitting on I think a few different points that I think are really important. The first is will the models of tomorrow be as agreeable? Right. So like on the, on the level of reflecting, validating your intentions and your emotions, I think that most clients actually need a lot more of that than we're actually getting them in terms of like the behaviors. No, not every behavior that you have is valid. And my sense is, is that the models that are trained to be therapy models won't have that problem. Like that'll be part of the difference. And so that's my first sense. The second thing is, and I think this is a really important question and I think that we're going to find out is I 100% agree with you. Like there's something that we lose in the person to person interaction. But I'm not sure that most people will actually care. My model for this is the old Harry Harlow studies.

Oh yeah, the monkey studies.

The monkey studies basically showed us that mammals are cue based. And so if this, if this, if this wire monkey doesn't have the cues that we need for softness, right. We don't perceive it as being warm, but if this terry cloth mother does, we perceive it as being one and we'll attach that thing. And so that's my. And I think that this really goes back to your question of what did we lose? And that's like the question, but will most people be like, man, this terry cloth mother feels really good. I think a lot of people will. I think a lot of people will, you know, and I saw something today that like 20 of people who use AI are using it already feel like romantic partners.

So I have seen that. I know, I know. Oh, yeah, that's. I mean, that's a whole other topic. We could go out of it all.

It's a whole other topic.

But let me dive into another thing that you say, even in your title. So I feel like it's an important piece. But do you kind of frame the next two years as a critical window for adaptation? So I'm curious from your thoughts, what needs to happen during the next two years and kind of what happens if we miss it?

That two year window comes from a few different things. One, it's from a interview between Ezra Klein and the chief sort of AI officer under Joe Biden. And in the interview, they basically said that Biden was well aware that within the next two years we'd have autonomous AI. And then everything flipped because Trump would sort of elect the president. And they didn't see that coming. So they're like, it's coming in the next few years. The other thing is that there's a report out called AI 2027. I don't know if you've, if you've read this report, it's really fascinating.

Okay.

And basically they're also saying two years from now, we're going to have AGI, right, super AI, which I don't know if that's going to happen in two years. I don't know if anybody knows. But I do think the next years are going to be like a crazy period. And part of why that happens is we're in a cold war. Who gets AI first? So we're getting a lot of funding in it. I mean, as soon as Trump was elected, he called. All the tech leaders were in, like, literally behind him. Right. Like up on the, up on the, at the inauguration. And he gave Sin mom and like $500 billion or something. I don't remember the number. Some astronomical amount that he partnered with. So we're in a cold war with the government basically saying, give as much money as we can toward this. And the first thing that all the AI companies are trying to do is use AI to code. They're trying to use it in a cursive process to make it better at making more AI. So that's where all the energy is going. So that's why I think the next years are like we're going to go through a time where in my opinion, I think, I think what, what this means for us is. And this goes back to the business side, right? So as, as much as I'm academically interested in this question of expertise, I, and I have a practice where I do all these things. I also have a business where I help people launch the practice and run practices. And the biggest issue in most practices is retention. I saw a study last week, I posted this on my, on my website that went through and categorized different types of therapists based on outcomes. And in this study, 25% of therapists were inefficient and ineffective. On surveys of hey, how much do you like your counselor? Clients? And I, I don't like them very much. And then on outcome surveys of how much better do you feel? They said I don't really feel better. Another 40ish percent were good for some people, but for the rest of their clients, for whatever reason, the clients dropped out really early in therapy. What this means is we have somewhere between probably 25 to 60% of counselors who struggle significantly, at least with some portion of their, of their caseload. So here's what I think we have to do. I think we kind of got to button up, right? Like, like I am, I am very pro people. I don't think, like I am very much of the belief that other people might choose to use AI as a therapist. I don't think it's a field that we should be. I don't think that that's best. I believe in the human connection, but I don't want to deny what I think other people will do. And so we want to really be people who are standing in the gap for our patients, for our clients. We have to really, I think up level in our skills that we, you know, if, if a quarter of therapists are not doing great in their practice, like we got to help them people help those people to really meet clients where they are so that clients can actually get what they need from a, from a person. So that's the first big thing is I think we got to upskill. And then the second thing is we have to be using these tools now so that we're familiar with them. And I mean they change weekly. So pay the 25 bucks, get the pro subscription and like start playing with it, start using it. Maybe not for your notes, maybe for Your marketing materials. Right. Maybe for your logos, maybe for your newsletters, like maybe for your blogs, like, use it to develop those. And then I think the third thing that we need to be doing is not fighting it, but really pushing for any way we can to use it. Right. I think if we fight against it, then the house is against us. If we really join this thing and say, hey, let's use this as a. As a co pilot. Let's use this as a librarian, let's use this as a assistant to us, we can use it and we can use it as a superpower to really help people. So that's. Those are my, like, three big things that we have to do.

I like that. I think for that last one, to me, there's also this, maybe this duty for us to ethically use it. Ethically. Right. Figure out totally how we can make sure it does no harm to clients. We know that's already happened. And use it in a way that can be of the greatest benefit. Yeah. To people. I like that.

I will say. And you know, so I also work at adjunct and do some consultation with a university called Cintio, and they're in the forefront of this stuff. So we're getting training on this in our faculty meetings on how to use it ethically. So maybe I'm less afraid of that because I'm getting training in it. And they have a class that they're putting out in September. I think it is. And I don't. I mean, I'm not getting any such kickback from this, but I think it's helpful. They're giving a certification on working with using AI in your. In your practice. So I. I'm very much a fan of what you're talking about, and maybe I'm a little bit less worried because I'm getting training in it. Maybe people. Other people aren't.

Yeah, I think that's a unique thing. I have not heard of another class like that. Not that they're not out there. Just hasn't crossed my path. So that's very cool. I like that. All right, well, with that, we are about out of time. Jordan, thank you so much for being here and digging into this discussion. I feel like there were even more things that we could have gotten into. And I encourage everyone listening to read his article and to dig in more and more to what's happening so that we can be aware, so that we can be smart. Thanks, everybody, for listening, and I hope you will join us next time for another look, another view at the chain changes in the mental health industry.