Talkspace Wants to Build a Better Therapist With AI Listening In

(Bloomberg Businessweek) -- In the midst of the pandemic, more people are turning to online therapy to deal with anxiety, depression, and relationship strife. Talkspace, which hired gold medal Olympic swimmer Michael Phelps as a spokesman, says web traffic doubled in just three months.

The New York-based company is a pioneer in the online therapy business, which links patients with a licensed counselor by video chat and text, encouraging them to open up in the privacy of their own home.

While the details of most in-person sessions never leave the room, Talkspace says it’s trying something different to advance the field of talk therapy and give itself a competitive advantage. It’s using tools such as machine learning and artificial intelligence to analyze transcripts of therapy sessions, which have been scrubbed of any identifying data. The company hopes its sophisticated algorithms can pick up on trends in a patient’s behavior and speech patterns that can lead to better treatment options or provide early warning signs.

“It’s pretty surprising that we use language in a very similar manner when we deteriorate,” says Oren Frank, a former advertising executive who founded Talkspace with his wife, Roni. The company says 1.5 million people have used the app so far.

Talkspace raised $50 million from investors last year to fund, among other things, software tools to identify what makes therapy successful and provide better feedback for counselors. The company says this aspect of the business is clearly disclosed in privacy terms, which state that Talkspace uses “non-identifying and aggregate information” in research and trend analysis. The way Talkspace handles the transcripts also complies with the Health Insurance Portability and Accountability Act, which governs confidentiality and information-sharing in health care, the company says.

The mental health community is in the early stages of understanding if machine learning software can be used to inform treatment, and some of the early research has shown the limitations of technology. “Human language is complex, and context matters so much,” says Dr. John Torous, director of digital psychiatry at Beth Israel Deaconess Medical Center. He pointed to a JAMA Psychiatry study that looked at algorithms used in large-scale studies that tried to predict suicide risk. “Their accuracy of predicting a future event is near 0,” according to the study, which surveyed more than 7,000 abstracts with more than 14 million participants.

Talkspace says its programs are more sophisticated than others because it looks at the stage of therapy a person is in and their level of engagement with a counselor to detect a potential crisis. Dr. Joshua Gordon, director of the National Institute for Mental Health, believes there’s a lot of potential in machine learning. “Imagine if we had all of the collective words that psychotherapists have used with their patients—and all of the words that those patients have responded to—and we could do a data-driven analysis,” he says.

He cautioned that research needs to protect patient identities and be used for medical purposes and not marketing. Given how precipitously mental health conditions can deteriorate, he warns, there isn’t any room for error. “The consequences are much greater here, and that’s why we argue for careful clinical trials and the kind of regulation to make sure these things work properly,” Gordon says.

©2020 Bloomberg L.P.