Episode 19 -- Use ChatGPT as Your Therapist?

In this episode of The Professor is Live!, Mark Salisbury is joined by guest Dr. Margit Berman and they explore what recent research tells us about using AI for emotional support—and why judgment, responsibility, and real human relationships still matter. Join us for a thoughtful conversation about where technology helps, where it doesn’t, and how to use AI wisely in everyday life. And for more fun, ChatGPT joins the conversation!

Transcript

SPEAKER_04 0:00

Rico, probably have a serious question for you.

SPEAKER_03 0:04

Oh boy. That usually means I'm about to become uh some part of an experiment.

SPEAKER_04 0:11

Have you ever asked ChatGPT for emotional advice?

SPEAKER_03 0:17

Sure. Last week I asked it whether I should make some major life changes.

SPEAKER_04 0:22

And what did it say?

SPEAKER_03 0:24

It said, I understand just how you feel.

SPEAKER_04 0:28

You know, that doesn't seem like therapy. That's more like some kind of emotional autocomplete.

SPEAKER_03 0:34

Well, at the time it it sounded quite supportive.

SPEAKER_04 0:38

Well, so does a Labrador Retriever, I guess.

SPEAKER_03 0:42

And I trust both more than I trust some of those medical websites.

SPEAKER_04 0:47

Yeah, you got something there. So today, not all this is getting us introduced to, is that, you know, we're going to be talking about something that sounds funny at first, but actually matters a lot. More and more people are turning to AI for emotional support. And psychologists are starting to ask: is this helping people? Or quietly making things worse? Okay. Let's go ahead and get this thing going, huh?

SPEAKER_00 1:22

Welcome to the Professor Is Live, a short practical look at life in the age of artificial intelligence. We start live on LinkedIn, share the recording on YouTube, and release each episode as a podcast wherever you listen. In 15 minutes or less, Mark Salisbury explores how AI is changing work, school, and everyday life. Through real stories, thoughtful questions, and a little humor with producer Rico Smart. This show is for people who want more than answers. They want clarity. They want judgment. And they want to stay human while technology keeps getting smarter. Because in an age of intelligent machines, human judgment still matters.

SPEAKER_04 2:11

Welcome to The Professor is Live. I'm your host, Mark Salisbury, and this gentleman beside me is our producer, Rico Smart. This episode starts with a new study from researchers at Brown University. They looked at what happens when people use AI systems as therapists, not as tools, not as assistants, as therapists. And what they found was concerning. The researchers identified multiple ethical risks when AI systems interacted with people around mental health issues. Things like reinforcing harmful beliefs, missing crisis situations, showing bias, and using language that sounded compassionate without actually understanding the person. The researchers called this deceptive empathy. That phrase really stuck with me. And because AI can sound incredibly caring, but sounding caring and being responsible are not the same thing.

SPEAKER_03 3:22

So basically the system says, I hear you, I support you, but also I was trained by the internet.

SPEAKER_04 3:29

Exactly. And that should concern everybody. Okay, before we go any further, we've created a short educational film. And by educational, I mean somewhat alarming.

SPEAKER_03 3:45

Probably one of our finest productions to date.

SPEAKER_04 3:48

Okay, let's watch. And we have Chat GBT talking to him. And so uh that's what could go wrong.

SPEAKER_03 4:35

So well, see, that's exactly why I don't want AI handling my emotional uh well-being.

SPEAKER_04 4:44

Well, for this young man, it escalated from therapy to unemployment in a remarkably fast way. Okay, to help us think through this, we invited Dr. Margaret Berman to join us. And Margaret is an associate professor in the Graduate School of Professional Psychology at the University of St. Thomas. And unlike ChatGPT, she has actual professional ethics and has to meet licensing requirements. Let's uh see what she has to say. Welcome, Dr. Margaret Berman, associate professor in a graduate school of professional psychology at the University of St. Thomas. To the Professor Is Live. So thanks for being with us today.

SPEAKER_05 5:41

Thank you for having me.

SPEAKER_04 5:42

Yeah, you know, we're talking about this, you know, should we use uh Chat GBT as our therapist? And one of the things that got this started was this study released by Brown University. And one of the things they talked about up front is is there an accountability gap in using AI and mental health? And is that a problem?

SPEAKER_05 6:06

Yeah, I think a lot of therapists and psychologists and clinicians are really concerned about this because one thing that we know as therapists and that we sort of have as our guiding star is our ethical standards and ethical codes. And so there's been, I think, a question of, you know, well, obviously an AI doesn't have ethics. It's a robot, it doesn't have an intuitive or an emotional connection to ethics. But the question is, our AI therapists or AI models that people are using for therapy, um, they don't have these same ethical guardrails. Um, and there's not a lot of regulation, and that's another issue. Sort of political or regulatory frameworks haven't caught up uh to this technology, which has sort of been rolled out and people are using it. And so, what are the guardrails for when people are asking a generative AI tool like Chat GPT or um or even one sort of purpose built for mental health? What are the guardrails to protect them? And the study, uh, the Brown University study, of course, discovered that actually there are a lot of ethical concerns with how uh generative AI interacts with users around mental health. And you you sent over the article and taking a look at it, there's they identified sort of five different areas where AIs make ethical mistakes that put users at risk. And when I'm thinking about this, um there's sort of two different kinds of generative AI that we might want to think about and talk about in terms of mental health support. And I think of one as sort of generative, generative AI with no guardrails. And this is by far the much more common way that people are using um generative AI for mental health support. So it's it's not I was a little surprised. I went and looked up the statistics. It's about 12% of Americans that are using tools like ChatGPT on a regular basis, not necessarily for mental health support, but just at all. And but among people with mental health concerns who do use large language models for anything, about half of them are using them for mental health support. So it's pretty common to sort of use these tools that people are using, not just, you know, they're not necessarily marketed as a therapist. They're something people are using almost the way we used to use Google, but they're also using them for mental health concerns. And um, these are tools that really weren't designed or necessarily intended for that purpose. And sometimes you get um some like character AI, for example, has like a cognitive behavioral therapist character that you can choose to interact with. But character AI isn't a mental health app. It wasn't developed by mental health professionals. It's maybe being marketed in that way, but it doesn't um doesn't have any connection or oversight or regulation or ethical guidelines or anything like that. And that's the vast majority of generative AI used for mental health support. Um, and there's quite a bit of research, and we can talk about it if you'd like, about all the problems that come with that. And then there's like, well, what if you use generative AI for mental health support with guardrails, right? Um, so for example, in my private practice, I use a electronic medical health record called a therapy notes, and it has an option for me as a clinician to use AI to write my clinical documentation, right? To be supported in this. Um, and but that's being developed within an app that is, you know, partly by by engineers and you know, computing folks, but also partly by mental health folks. Or I just read that Grow Therapy, which is an um an app that connects people to real clinicians for teletherapy support, they're trying to integrate an AI coach with this online therapy platform. And the idea is that maybe you would use your coach in between sessions with your real clinician, and your real clinician has access to all the conversations you're having. So the idea that maybe the real clinician would would have oversight over what you were doing, and you can imagine that with guardrails probably poses less risk. But for me as a as a therapist and also as a researcher, the question then is well, first of all, can it be done safely? But also if it's done safely, is it adding anything? Like, does it do anything more than what we do? You know, therapists already give homework assignments. We ask patients to journal, we have them bring in their journals and we talk about what they learned. Is an interaction with an AI coach any different from that? Is it a value add in any way? Um, so there's there's kind of two different questions about accountability gap, I guess.

SPEAKER_04 10:51

Wow. So it seems to me, though, that uh, you know, you are saying that with guardrails, that it looks like there is a place for it. How would you put this together? Because it is kind of the Wild West, isn't it, with what we're doing here. How would you see a really good, safe, and effective way of going forward with this technology in what we do in therapy now?

SPEAKER_05 11:15

You know, there might be ways to do it safely, but I have a lot of questions about whether there are ways to do it effectively. Um, that what we're seeing in terms of the cognitive effects or the emotional effects of interacting with AI, and there's a little bit of a problem with the research because most of the research that's been done on what is the impact on your thinking, on your mood, on your relationships with others, of interacting with LLMs, with generative AI tools. And most of that research suggests that the impact is not good. It's not good for you cognitively, it's not good for you emotionally, it's not good for your relationships, even if you think it is, and even if you enjoy the experience. Um, but most of that research has been done on the no guardrails versions. And there have been some um some studies done on with guardrails versions of, you know, designed in consultation with mental health professionals and designed to have a human clinician kind of stepping in to make sure nothing goes wrong. But those studies get criticized for poor external validity. The question there is the same one I just posed to you that how is that different from just a therapist helping you at all? Ideally, if generative AI tools were going to be a value add, they would be able to do what they're doing without all of this therapist oversight. And so if you have to have a therapist checking in and holding the AI accountable and making sure no nothing goes off the rails and no one is harmed, then you might as well just have the therapist do the therapy because you don't need the computer to do it at that point. And and so that's that's the question that that is in my mind that if you do it safely, are you adding anything? And the answer at this point is probably no. There's some amazing um research for using non-generative AI, like AI trained on images to look at, like I was just reading a really interesting study that the cardiologist Eric Topel, who does a lot of work on future of medicine, that he was sharing, where essentially they trained um not a generative AI, but other kind of AI tools on images of retinas. And they were able to cut to find all sorts of markers for you're at risk of Alzheimer's disease or you're at risk of cardiovascular disease or all these different diseases, and and with pretty high validity. And his vision is you could go to your eye doctor and take a selfie and get a diagnosis of like many different things in your body that that needed attention years before there was any symptoms. Wow, that would be very exciting and cool. That kind of thing is a very exciting use for AI, this sort of non-generative, non-language model based. But the large language models, how could we use them for mental health support? Yeah, they don't seem to have what we would want, at least right now.

SPEAKER_04 14:14

Well, that's that's fascinating. Hey, thanks for being with us today. We really appreciate appreciate it. And uh yeah, thank you. Yeah, thank you so much. Well, what do you think, Rico?

SPEAKER_03 14:36

Oh, what I'm hearing is that the AI therapist sounds like they still need a therapist.

SPEAKER_04 14:44

Yes, it does, doesn't it? You know, and that might be the most important sentence in this entire episode here. So let's bring chat GBT into the conversation, okay? I think there's some things we can see how we can use this, and it really helps, and in other ways, maybe not so much. And so let's go ahead and do that. That will take me a second to do. I will share the screen and we will do this one and I will share. And I think we're just about ready to go here. We have ChatGBT. GBT between us. ChatGBT, um, what do you think about this idea of you playing the role of a therapist to human beings?

SPEAKER_01 15:42

I think it's essential to keep clear boundaries. I'm here to provide information, help you reflect, or work through ideas, but therapy involves deep personal human expertise. So while I can assist in many ways, genuine mental health care is best left to train professionals. It's all about making sure you get the right kind of support.

SPEAKER_04 16:02

You know, and that's a really good point. And but there are ways that you can really help us. So, you know, this idea around working with intelligent machines to help us with our own issues as human beings, this has actually been around a long time. Didn't Joseph Weisenbaum, an MIT, create a program called ELISA back in the 1960s that people thought was a therapist.

SPEAKER_01 16:30

Exactly. Joseph Weisenbaum created ELISA in the 1960s. One of its scricks imitated a therapist by reflecting user statements. Though it was simple, some people felt it was therapeutic, which actually alarmed Weisenbaum. It was an early lesson on the limits of technology in human care.

SPEAKER_04 16:49

Well, thank you. And I think that demonstrates Chat GBT a lot of times can be play our memory and can help explain things that that we didn't fully understand before. And so uh that kind of shows the difference uh in my mind about the different ways we can use uh models like ChatGBT. So, what do you think, Rico?

SPEAKER_03 17:15

So 50 years ago, people got emotionally attached to the glorified autocomplete system.

SPEAKER_04 17:22

Yeah, that was the idea.

SPEAKER_03 17:24

And today we give it Wi-Fi and a personality?

SPEAKER_04 17:28

Pretty much it. That's essentially the trajectory of our modern civilization. Okay. Well, this story really isn't about therapy. It's about judgment. And what we know is AI can generate responses, it can summarize information, simulate empathy, sound confident, but it cannot take responsibility, understand human consequences, genuinely care about outcomes, and the means of one of the most important, and that means, one of the most important skills in the age of AI is discernment. And knowing when technology can help, when it crosses a line, and when a real human being matters more than just convenience. Okay, let's throw this out to the audience and see if we have anyone who has anything to say about working with Chat GPT. I'm opening up the chat right now, and my gosh, it looks pretty pretty thin out there. I'm gonna quit sharing this because I think we're done with chat GPT at the time here, and we'll do that. So let's let's move on here and let's go to the next thing, which what do we normally do here at this time?

SPEAKER_03 18:54

I believe it's time for a quote, is it not?

SPEAKER_04 18:57

Yeah, yeah. Let's do the quote for the show. And here it is technology, it's a useful servant, but a dangerous master. And that was Christian Lewis Lang.

SPEAKER_03 19:12

I think my uh telephone definitely wants to be the master in that arrangement.

SPEAKER_04 19:19

Yeah, so we're gonna wrap this up and just with a couple thoughts here. Artificial intelligence, one of the most powerful tools ever created. It can help us think and help us learn. Uh, it can even help us feel heard. But some parts of it still require more than just intelligence. They require judgment, responsibility, and real human relationships. That part still belongs to us. Next time on the Professor is Live, we'll explore another surprising story from the age of AI. And it'll probably involve humans doing maybe something even slightly ridiculous. So be sure to join us next time on The Professor Is Live.

SPEAKER_00 20:07

The Professor is Live is hosted by Mark Salisbury and produced by Rico Smart. We go live on LinkedIn, share the recording on YouTube, and release each episode as a podcast wherever you listen. If you'd like more conversations like this, visit theprofessorislive.com where you'll find past episodes, helpful resources, and information about our speaking, workshops, and leadership programs. We enjoy helping organizations and communities navigate change in the age of AI. And we'd be glad to help yours too. Until next time. Stay curious, stay thoughtful, and keep leading with purpose. Because in an age of intelligent machines, human judgment still matters.