Go back

The AI Therapist: When People Turn to Chatbots for Mental Health

from WSJ What’s News

20m 58s

The AI Therapist: When People Turn to Chatbots for Mental Health

At UT MD Anderson Cancer Center, the mission is to cure cancer through precision medicine and advanced care, emphasizing treatment over symptom management. Meanwhile, the use of AI mental health tools—such as chatbots like Ash—is rising rapidly as a response to the difficulty of accessing traditional therapy. Many users find these tools helpful for managing stress, anxiety, and daily emotional challenges, with some reporting improvements like quitting vaping. However, experts caution that current AI tools are not substitutes for licensed therapists and lack robust, peer-reviewed studies to prove their effectiveness. Significant risks remain, including the potential for harmful advice, especially in cases of severe mental illness or self-harm, and the absence of legal obligations like mandatory reporting that apply to human therapists. Unlike human professionals, AI tools do not have ethical or legal duties to intervene in crisis situations. While companies like Slingshot and OpenAI offer safety features and user guidance, concerns persist over transparency, bias, and oversight. The lack of regulation and independent research means the true impact and safety of these tools remain uncertain. As AI mental health tools gain adoption, the need for clear standards, ethical safeguards, and rigorous scientific evaluation becomes increasingly urgent.

Transcription

3518 Words, 20013 Characters

English
At UT MD Anderson Cancer Center, our goal is to cure cancer, not simply treat the symptoms. We go beyond standard diagnostics and conduct in-depth tests to ensure our patients receive an accurate diagnosis and the right treatment plan from the start. This deeper level of precision gives our patients the best hope to overcome cancer. At UT MD Anderson, we know cancer inside and out. Choose us first at MD Anderson dot org. It was tough to leave in New York. I was there 13 years. It took me a while to build the social circle that I have there. But here in DC, I only know a couple people. It was definitely a bit of a scary jump. Here at BNMA is a producer here at the Wall Street Journal. He recently moved from New York City to Washington, DC for a new job at the journal. To cope with the stress of the move, he tried the usual things that make him feel relaxed and centered, like going to the gym, playing piano, and spending time with friends. But he also tried something new, talking to an AI chatbot designed for mental health. Hey Pierre, great to meet you. I heard you're making a big move from New York to DC next week. That's a lot of boxes and logistics to juggle. How's the stress showing up for you right now? I've skeptical and uneasy about AI overall and thinking about its risks and the resources it consumes and all that. But as a journalist and as someone who is just moving and stressed out by it, I thought I would give it a shot. I'm Alex O'Soleff for the Wall Street Journal. On this special episode of what's new Sunday, we'll be talking about the growing number of people, like Pierre, who are trying out AI tools for mental health and the growing universe of companies creating chatbots. We'll be digging into the promise of these apps for helping people in moments of need and the dangers that could come with them. This is the AI Therapist Part One. Some people say that everyone would probably benefit from therapy. That includes people dealing with life's normal stresses, like moving to a new city. And also, people coping with mental illness. The National Institutes of Health estimates that about a fifth of Americans live with mental illness. That ranges all the way from mild conditions to disorders that make it hard to live a normal life. But being able to see a mental health professional when you need one isn't always easy in this country. The American Mental Health System is quite fractured. That's Karthik Sarma. He's a physician and researcher at the University of California, San Francisco. He's a psychiatrist who sees patients and studies how AI can deliver mental health care. The amount of difficulty that you have to access care, if you don't have access to substantial funds, it's huge. That really leads to huge inequities and what the outcomes are for people who are struggling with mental health. Sarma sees a lot of potential for AI tools. My dream would be that every patient could simply log on to a chatbot on their phone and be able to get that therapy right away. Sarma works with a company called Slingshot AI as an advisor. Slingshot is the company that makes ash, the chatbot Pierre used. It's currently free and unlike most real therapists, it's always available. Getting mental health support to people whenever they needed was a big reason why Neil Perique co-founded Slingshot in 2023. You have a problem, you can show up and Ash is there to help you think through that. Perique's sort of a serial entrepreneur. Since he dropped out of medical school, he's founded a handful of companies. One of them is Casper, a company that made its name delivering mattresses in boxes to people's homes. It was meant to upend the mattress industry and it had a lot of slick marketing. It was 25 helping run a thousand person company had no idea what I was doing. I had lots of crazy thoughts going through my mind. And my friend said one day you should go see a therapist. I remember feeling insulted because I'm thinking like, "What? Why would you tell me to go see a therapist?" Because nobody in my family ever talked about mental health. Perique took his friend's advice and did start working with a therapist. He says it changed his life. That growing interest in mental health bled over into his work too. He joined the board of a company called Psych Hub which helps therapists learn new skills. And he says that this problem of access to therapy kept coming out. It takes people months to years, sometimes to find somebody. There's still a lot of stigma and shame around it, and I knew we could do something better. And then came the rise of AI, specifically open AI. In 2022, the company released chat GPT, powered by the model GPT 3.5. The natural way the model handled conversations made Perique think AI's therapy moment had arrived. I kept thinking like, "Oh, somebody's going to build an AI version of therapy." And I originally wanted to go invest in the company and then I traveled the world trying to find somebody working on it. And I couldn't. And so, Perique, along with his co-founder, Daniel Khan, created Slingshot. Their chatbot Ash went live last year. I feel like the AI app is really great because it's an immediate response from somebody, a computer, that you can work through your problems with. That's Ashley Jones. She's 37 and lives in Savannah, Georgia. She's a divorced single mom and works six days a week, managing two sushi restaurants. I have to deal with people constantly needing things from me. That's my job, making sure the customers are happy and everything's running smoothly. And at the end of the day, you know, it's really hard to shut your brain down after going on your feet all day long. So, Jones says she turned to Ash to help her handle her work stress. I talked to Ash in the morning about what my date is going to look like, how it's supposed to unfold. And then if it changes and it bothers me, I can open up the app and be like, "This is how I wanted it to go and it's not going this way." And we work through it. It'll tell me what's one thing that you can do right now to pull you into your body and ground you so that you can process what you're going through and a calmer state. Coming up, lots of people are turning to general AI chatbots like ChatGPT to help with their mental health. Sometimes it goes fine and sometimes it really doesn't. That's after the break. At UTMD Anderson Cancer Center, our goal is to cure cancer, not simply treat the symptoms. At UTMD Anderson, we know cancer inside and out. Choose us first at mdanderson.org. Having support that's always just a few taps away is part of the reason people have started using general AI chatbots like ChatGPT or Gemini for mental health, even if they're not designed for that. And it's pretty common. A survey published in June by the American Psychological Association found that three-quarters of the 1200 therapists they asked said their patients told them they'd used AI for things like emotional support. And a third said patients are using them to assist their treatment. It initially started with ChatGPT, mental health, and also just questions about health generally. Jennifer Millier is a 45-year-old attorney living in Los Angeles. I had a, I'll just go out and say it, was diagnosed with breast cancer last year. And so obviously had a lot of questions, questions about treatment, questions about outcomes. And so I use ChatGPT a lot in that process. She had worked with a therapist before, but felt she had outgrown her most recent one. It was just a lot, you know, undertake finding a new therapist and go into her all that where I had been kind of using ChatGPT already as a resource. So why not go ahead and give it a try for something else? But families of some people who turned to ChatGPT and other Chatbots said it led to some pretty devastating outcomes. A tough turn now to a teenager's death by suicide encouraged by artificial intelligence. Families are suing tech giant OpenAI claiming that it's ChatGPT program drove people to suicide. Character AI has agreed to settle multiple lawsuits. After it was accused of contributing to mental health crises among young people. And earlier this summer, Florida Attorney General James Uthmeyer filed a lawsuit against OpenAI and its CEO Sam Aldman. For endangering our kids and deceiving parents into believing that this application is safe for use. It's clearly not. People are getting hurt, parents are getting deceived, and they need to pay for it. An OpenAI spokesman says the company has strengthened how ChatGPT responds in sensitive and acute situations with input from mental health experts. He said ChatGPT is not a substitute for medical or mental health care, and that the safeguards the company puts in place are designed to help users in distress get real world help. And we should also note here that News Corp owner of the Wall Street Journal has a content licensing partnership with OpenAI. Carthic Sarma, the California psychiatrist and researcher who works with slingshot, says he's worried that these general AI chatbots could hurt vulnerable people, and that the companies behind the AI might not know how to stop it. There's a reason why therapists are licensed, they have to be educated, and anyone who's providing these services has to be educated because there are real risks. Like OpenAI, Anthropic says its AI chatbot, Claude, isn't designed or intended to act as a mental health expert. health professional, and then it makes that clear to its users. Anthropics said the chatbot encourages people to seek guidance from licensed professionals when appropriate. A Google spokesperson referred me to a publicly available blog. It says the company has been committed to helping people in moments of crisis and has been working on providing better access and better responses to people in distress with Gemini. There aren't a lot of laws about what these AI companies have to do if users are at risk. To get a little wonky for a second, human therapists are what are called mandated reporters. So for certain kinds of scenarios like suspected child abuse, therapists are almost always required to report it to authorities. If someone says they're planning to hurt themselves or someone else, sometimes that means reporting it to authorities, but sometimes not. In the US, each state has different rules about what needs to be reported. But an AI tool isn't a human therapist, so those rules don't apply. Many makers of chatbots specifically designed for mental health say they do this voluntarily, that they have ways to escalate worrisome conversations to a human. And you'll hear more about how that works and what the psychology community thinks about it in the next episode of our series. So the regulations here may be a gray area, but companies aren't waiting for laws on the books. Coming up, we'll take a closer look at the rise of AI tools specifically designed for mental health and what it's like to use one. Is it helpful? Is it a bit strange? More? Next. At UTMD Anderson Cancer Center, our goal is to cure cancer. Not simply treat the symptoms. The AI therapy industry is exploding. John Torres, a psychiatrist at Boston's Beth Israel Deaconess Medical Center, estimates that there are at least 100 mental health-focused AI bots. There's a huge range. Some are coming from big companies specializing in mental health like headspace and talkspace. Sometimes we really want someone to talk to, and it can't wait. Meet T, an AI mental health guide that's safe, private. Some of them like Spring Health and WoBOT have tools only available through certain employers. And then there are dozens of standalone apps. To name a few, Earkick, Harmony, Alomia, Weissa, Meat Weissa, your mental health buddy. And of course, Ash. One more month. And here's what I can tell you. Ash AI therapy works. Each of these tools claims something a little different. They might focus on a particular type of therapy or say they can help connect you to a human therapist if needed. And investors are seeing dollar signs. I think AI therapy can be like a $10 billion plus, a single company can be a $10 billion plus company. Amy Martin is an investor at Menlo Ventures, a longstanding venture capital firm that has backed companies ranging from Amthropic to Uber to syntax like Betterment. In 2024, it also invested in Slingshot, the company behind Ash. Overall, Slingshot says it's raised about $93 million. Right now, Ash is free to use. But the company says it plans to start charging a subscription fee in the next few months. Today, people are used to talking to a human therapist. Tomorrow, are way more people used to talking to an AI therapist. We're betting that people will get up that adoption curve. And when Martin says talking to, in the case of Ash, she needs it literally as our producer Pierre experienced. And yeah, it's a chatbot. You can choose its voice. It can sound a few different ways. Welcome. Welcome. Welcome. Welcome. Ash is your space to talk through whatever's on your mind. You can type to it and read the answers back or you can speak to it and hear the answers back. This is kind of interesting, even though you're not a human. I'll take that as a high compliment. You don't need me to be human for this to work. You just need a place to think out loud without being judged or rushed. If it helps you sort your thoughts, that's enough. But the way Ash communicates can also come with some quirks. One thing Pierre noticed was a delay between when he said something and Ash's response. He even asked Ash about it. And that delay you're about to hear, it's not your podcast app glitching. Maybe that's something that'll improve in the future. Could it be that you would kind of respond more quickly the way a person would? That pause you're noticing, totally fair observation. Right now, generating a thoughtful response takes computation time. I'm processing what you said, predicting language, checking guardrails, shaping tone. All of that happens in sequence. It's not instant. Slingshot told us that sometimes the pauses are intentional because the conversation benefits from silence. But sometimes they're a technical lag. They're working on reducing that. Ash also has a way of speaking that Pierre told me feels totally characteristic of AI. Some of the lines it would give were quite poetic as if it was kind of trying to speak beautifully. Here's what Ash was telling me when I was telling it about how I'm trying to figure out my new commute by bicycle. It's funny how a city stays abstract until you've burned a specific route into your legs. Suddenly the grid isn't just lines on a map. It's the smell of exhaust at 8 a.m. or the way the light hits a certain block. So talking to Ash sometimes reminds people that yes, that's an AI. But the real question is, is Ash effective? Can it help people feel better? In June, Slingshot put out a study, not peer-reviewed, to assess this. Slingshot's Neil Parique. We measure a bunch of different outcomes. PhD-9 is movement on a depression scale, GAD-7 is movement on an anxiety scale. And what we saw is the people that we're using Ash saw similar movements on those two scales as it is what you would get in traditional therapy. Translation. Ash helped people showing signs of depression and anxiety feel better as much as a real therapist would. But how reliable is a study like this, really? There's a lot of pre-prints, there's companies showing grown research, and that's fantastic. They should do a test or early steps. But we don't have that type of high-quality evidence to say a mental health AI is actually going to be better. John Torres, who I mentioned earlier, directs the Digital Psychiatry Division at Beth Israel Deaconess Medical Center. He worked on a systematic review that's also not peer-reviewed, looking at how companies are evaluating their tools. And he and his co-authors found that there were 178 proposed rating scales for mental health AI. 178, that actually may be more kind of evaluation scales and frameworks than there probably are tools that people are using for mental health for AI. And what we found is a lot of these evaluation schemes are being written by companies who are low and behold selling you an AI. Torres is working to develop a more independent evaluation for these AI mental health tools. The American Psychological Association, the professional organization for psychology, says to watch out for AI therapists. Bill Wright is the Senior Director of Healthcare Innovation at APA. And she's one of the people who created the Groups Advisory that warned against AI therapy last year. Currently, what we know from the research is that these general purpose LLMs and really even the wellness apps on the market are not sufficient replacements for mental health therapy with a licensed professional. But users we spoke with say they found AI mental health chatbots pretty helpful. Ashley Jones, the restaurant manager from Georgia, says, "Ash helped her quit vaping." I used to smoke for 17 years and I quit this past year. Not like saying the ash helped me quit, but it did actually. It comes up with a plan for you. I didn't listen to its plan. I did my own thing and I just went cold turkey. But the support and it asking me regularly if I was still quitting, that was really nice. As for Pierre, there was one particular moment with ash that stayed with him because of how it made him feel. At one point, I felt the familiar feeling that I felt in the little experience I have with therapists with a human, which was a bit of tension in myself before talking about something and then some relief after talking about it. I was telling you about one of my friendships, a bit of a tricky friendship. At one point, it came back to me with this. Resentment usually shows up when we're giving more than we really want to give or more than we have to give. And no one has quite acknowledged the cost of that. Sure, that's the kind of thing that I might be able to read at a blog post somewhere from a therapist or something like that or a website about friendship or relationships. But I thought it was, it's just stuck with me as kind of insightful. So the promise is that these AI tools like ash can help support people who need it when they need it. And for many of the users we spoke with, it mostly seemed to be doing that. But what about severe mental illness? How would these apps handle someone who had, say, paranoid delusions, who was talking about hurting themselves or hurting someone else? I asked slingshots Neil Parique. Are there patients who should not use ash? Yeah, I think the most important thing is like if you don't think you should be using it, you shouldn't use it. We'll get more into what could go wrong with an AI therapist and how startups are trying to show that they really are safe in the next episode in our series. That's coming out next Sunday. Heads up that we've made a correction to this episode, an earlier version of this podcast incorrectly stated that human therapists are required to report to authorities when someone says they want to hurt themselves or hurt someone else. The requirements for reporting vary. depending on how serious and immediate the threat is. And that's what's news for August 9th. Today's show is produced and mixed by Pierre Bienne-Mais with supervising producer Tally R. Bell. Michael LaValle created our music. Additional editorial support came from Chris Sinsley. And I'm Alex Oselef for The Wall Street Journal. We'll be back with our regular show on Monday morning. Thanks for listening. At UT MD Anderson Cancer Center, we're raising the standard in cancer care. Our world-renowned researchers and sub-specialized physicians work together to turn discoveries into life-saving treatments. Through the largest cancer clinical trials program, we bring the most innovative treatment options directly to our patients. Choose us first at MD Anderson.org.

Podcast Summary

Key Points:

  1. UT MD Anderson Cancer Center focuses on curing cancer, not just managing symptoms, using advanced diagnostics for accurate diagnoses and personalized treatment.
  2. AI mental health chatbots like Ash are growing in popularity, offering immediate, accessible support—especially for people facing stress, anxiety, or mental health challenges.
  3. Users report benefits such as emotional grounding, stress management, and improved mental clarity, with some, like Ashley Jones, noting tangible outcomes like quitting vaping.
  4. Despite positive user experiences, experts warn that AI tools are not substitutes for licensed therapy and lack rigorous, peer-reviewed evidence of long-term effectiveness.
  5. Significant ethical and safety concerns exist, including risks of harmful advice, lack of mandatory reporting obligations in distress scenarios, and potential harm to vulnerable populations.
  6. Companies like Slingshot and OpenAI offer AI mental health tools, with some including safety features to escalate to human professionals, though implementation and oversight vary.
  7. Regulatory gaps exist because AI tools aren’t bound by the same legal duties as human mental health professionals, such as mandated reporting of self-harm threats.
  8. There is a rising investment in AI mental health, with companies expecting massive growth, but independent evaluation remains limited due to bias and lack of transparency.

Summary:

At UT MD Anderson Cancer Center, the mission is to cure cancer through precision medicine and advanced care, emphasizing treatment over symptom management. Meanwhile, the use of AI mental health tools—such as chatbots like Ash—is rising rapidly as a response to the difficulty of accessing traditional therapy. Many users find these tools helpful for managing stress, anxiety, and daily emotional challenges, with some reporting improvements like quitting vaping.

However, experts caution that current AI tools are not substitutes for licensed therapists and lack robust, peer-reviewed studies to prove their effectiveness. Significant risks remain, including the potential for harmful advice, especially in cases of severe mental illness or self-harm, and the absence of legal obligations like mandatory reporting that apply to human therapists. Unlike human professionals, AI tools do not have ethical or legal duties to intervene in crisis situations.

While companies like Slingshot and OpenAI offer safety features and user guidance, concerns persist over transparency, bias, and oversight. The lack of regulation and independent research means the true impact and safety of these tools remain uncertain. As AI mental health tools gain adoption, the need for clear standards, ethical safeguards, and rigorous scientific evaluation becomes increasingly urgent.

FAQs

UT MD Anderson Cancer Center's primary goal is to cure cancer, not just treat symptoms. They use in-depth diagnostics and specialized care to ensure accurate diagnoses and effective treatment plans.

The center combines expert research, sub-specialized physicians, and large-scale clinical trials to deliver innovative, evidence-based treatments that are directly available to patients.

A study by Slingshot found that users of Ash showed similar improvements in depression and anxiety symptoms as those seen in traditional therapy, suggesting potential effectiveness for mild to moderate cases.

No, current research and professional organizations like the American Psychological Association state that AI chatbots are not sufficient replacements for licensed mental health professionals.

Risks include potential misdiagnosis, lack of emotional understanding, and in rare cases, contributing to harmful outcomes such as suicide, especially if users are vulnerable or in crisis.

Tools like Ash are currently free to ensure accessibility, but companies plan to introduce subscription models to fund development and support ongoing services.

Chat with AI

Loading...

Pro features

Go deeper with this episode

Unlock creator-grade tools that turn any transcript into show notes and subtitle files.