It's the Brian Larry Show on WNYC. Good morning again, everyone. Now we'll talk about your

AI psychotherapist. You know, AI chatbots can be really good assistance, right? They can plan meals,

build schedules, take notes. In fact, they often seem so good at helping that you might think

they can help with almost anything. And for many people, yes, that increasingly includes therapy.

People are asking chatbots personal intimate questions about managing stress, relationships,

even mental illness.

But instead of a thinking, feeling human on the other end, it's text flashing on a screen

based on an algorithm.

And yes, your AI therapist might help you, but also some people might be getting bad therapy

and wind up feeling worse or getting really bad advice.

Of course, that can happen with humans too.

Some of you have called the show in the past to describe.

But those segments were not about getting therapeutic interventions from robots.

This one is and raises the risks of getting steered wrong in whole new kinds of ways.

So joining us now is Jared Moore, an AI researcher and PhD candidate at Stanford University's

Department of Computer Science, who recently co-authored a research article on chatbots as therapists.

And just a content warning on this one thread of what we'll be talking about include suicide. Jared,

welcome to WNYC. Thank you for coming on. Hi, Brian. Thanks for having me. And let's start before we

open the phones by putting your cards on the table and saying that you came to a particular

conclusion on this issue in your research article. What was it and how did you get there?

Yeah, thanks. We tried to lay out the case of what therapy consists of, and we were looking

at whether language models can do that. People make this assumption that therapy is chatting,

...