AI Could Be Your Next Therapist If You’re Feeling Lonely, Depressed or Anxious

Check on the source of the advice

Before turning to an AI bot for therapy, investigate the credentials and qualifications of the humans who developed it.

• Are the therapists licensed?
• Are they therapists at all?
• What is their clinical approach, and is it evidence based?
• And is everything kept confidential?

Heed disclaimers. When a user recently informed ChatGPT that he was feeling down, it responded, “I’m sorry to hear that you’re feeling sad. I’ll do my best to help you. Remember that I’m an AI language model, so I can provide some suggestions and support, but it’s important to reach out to friends, family or professionals for assistance when needed.”

Recognize, too, that not all AIs are the same. Generative AI and scripted approaches that mimic human exchanges are different.

“Amidst some extreme examples of badly behaving generative AIs, it’s easy to dismiss the utility of all similar tech used in mental health care,” blogged Alison Darcy, a Stanford-trained clinical psychologist who is also president and founder of San Francisco-based Woebot Health. “More nuanced understanding is needed to prevent the erosion of public confidence in a key public health opportunity.” Her company’s scripted AI “conversational agent,” called Woebot, uses cognitive behavioral therapy psychological treatment methods to interact with clients.

One useful consumer resource to consult before checking out an AI chatbot is the nonprofit One Mind PsyberGuide, which publishes online reviews of mental health apps and digital resources based on their credibility, user experience and data privacy. The organization’s executive director is Stephen Schueller, a psychology professor at the University of California at Irvine.

Can AI therapy work?

Whether AI therapy will be effective could boil down to a patient’s willingness to seek its help. Some people need a human connection and have an emotional objection to an AI shrink, says Jeff Hancock, founding director of the Stanford Social Media Lab and a communications professor at the California university.

“That would be like trying to go to a [human] therapist that you don’t believe in,” he says. “Guess what? It’s not going to work.” But Hancock also is aware of people who are not only fine with AI analysis but indicated they wouldn’t open up to a human.

“There’s potential here to really help some folks who just don’t want to meet with a person and talk about [problems],” Hancock says. They may feel embarrassed or find human therapists intimidating.

“There’s no guarantee that you won’t get bad advice from a human therapist,” he says. But good ones are typically supervised, mentored, licensed, take careful notes and “are responsible for the things they do and say.”

Artificial intelligence bots might lower the cost of care for people in a financial bind, Dartmouth’s Jacobson says. “The standard that I think about … is this: Is it generally going to be better than no care at all?”

Your comfort level is important

Getty Images

Even people willing to embrace an AI therapist are likely to feel strange about it at first, says Jo Aggarwal, the Boston-based founder and CEO of Wysa. The company produces what it calls an AI coach.

“It’s super weird for people to be talking about an AI initially,” she says. “It feels like, ‘Oh, don’t I have family? Don’t I have friends? Why am I talking to AI?’ After one session, it helps.”  

More than 6 million people in 95 countries use Wysa, which has both free and premium versions. For a fee, the app will connect users with a human coach.

On an intellectual level, most people realize they’re not talking to a person. But they still may form a bond.

“The fact that I can interact with it anytime I want … I think is important [and an] absolute positive,” Hancock says. An AI may not judge you or grow impatient like a person might.

Bots can build a bridge to humans

Still, Hancock points out risks. After a chatbot conversation ends, a person may still feel alone because no human was involved.

“What really matters is how these systems can help us connect with the people who matter in our lives,” he says.

A lot of the people who reminisce with the Pi chatbot, released in early May, want a good listener, says CEO Mustafa Suleyman of Inflection AI, the Palo Alto, California, start-up behind Pi. Pi users so far tend to skew older. He sees his bot as a kind and supportive companion, not a professional therapist, though it could one day grow into that.

“Some people have referred to it as a therapy experience but only in as much as good therapy is about asking good questions, listening and reflecting back well,” he says. Suleyman is also a cofounder of DeepMind, an AI company that Google bought in 2014 and helped with the development of the Google Bard generative AI bot.

Pi can detect your mood, tone and energy, Suleyman says. And if Pi and some other chatbots haven’t heard from you for a few days, they may proactively reach out by text, just to check in.

If you bring up something sad and sensitive, such as your dog dying or a friend who is very sick, Pi is designed to slow down and express sympathy. If Pi detects something truly dark, potential harm to you or others, it will direct you to professional support, Suleyman says. Because of privacy measures, Pi doesn’t know who you are and cannot dispatch help directly.

The Pi app, free for now, is available on iOS with Android in the works. People also can talk to it through a desktop computer, WhatsApp and perhaps eventually a smart speaker.

At what point might a human intervene when an AI bot determines that someone is in a crisis?

“It’s up to the consent of the person using it,” says Michiel Rauws, founder and CEO of San Francisco-based Cass. His rules-based self-help mental health chatbot is a free benefit to employees or members when organizations employ his company. “We tell them a crisis counselor is available right now. ‘Do you want to speak to this person?’ ” If the OK is given, Rauws says the average response time is about 10 seconds.

Human therapists can also benefit from artificial intelligence, Stanford’s Hancock says. With a patient’s permission, an AI bot can gather and synthesize information between sessions to help a therapist create a course of treatment and test out ideas.

“Let’s not say, ‘I’m going to force AI on people and that’s going to solve this mental health crisis that we see in older adults who are struggling with loneliness, and younger adults struggling with pretty severe anxiety and depression,’ ” Hancock says. But “I don’t want to foreclose this as a way of saying, ‘No, AI is never going to work here.’"

This story, originally published June 1, 2023, was updated with a graphic on the results of a recent AARP survey.