‘It’s All About Relationships’: Why AI Won’t — and Can’t — Replace Humans in Behavioral Health

The power of human relationships and liability concerns make replacing staff with artificial intelligence unlikely and ill-advised for the future.

Industry insiders say the broader implementation of artificial intelligence in the behavioral health sector will likely magnify the work of staff and make the work experience better.

“I’m a physician by training: I’m not really anxious to remove the human element from any sort of interaction,” John Cohen, CEO of Talkspace Inc. (Nasdaq: TALK), said during a Behavioral Health Business webinar.


Some websites and apps already attempt to deliver mental health support via direct interactions with AI-based chatbots. But for Cohen — who leads one of the largest virtual-only therapy and psychiatry companies in the U.S. — such AI tools are out of the question for him and his business.

“We have no intention of using AI to provide care as an AI bot,” Cohen said of therapy chatbots. “We’re taking that off the table.”

When asked if he expected the wider behavioral health sector to use AI to replace staff, Cohen added, “It’s highly doubtful.”


Talkspace employs 4,300 mental health providers and provides virtual therapy and other mental health services direct-to-consumer (D2C) and business-to-business (B2B), with the latter making up the bulk of its business.

Talkspace has had AI embedded into its business for years, both on the administrative and clinical sides.

The company’s AI can generate clinical notes over a few days’ worth of patient chats and live sessions. Before entering Talkspace’s in-house electronic health record (EHR), clinicians review these summaries to track care.

In doing so, the company lessens the clinical burden on its providers.

“We’re still playing with the whole issue about how it gets integrated into our business operations,” Cohen said, adding that Talkspace is beginning to explore the use of AI in revenue cycle management and other business processes.

Similarly, Gaudenzia, a Norristown, Pennsylvania-based behavioral health and human services system, uses AI tools from Eleos to create notes for and grade individual therapy sessions.

Dale Klatzker, CEO of Gaudenzia, said that Eleos’ tools helped standardize notetaking and input into the EHR it rolled out, starting five years ago. It also gave his organization a potent compliance and clinical quality assurance tool.

“When I was trained, my supervisor sat [behind] a one-way mirror observing me,” said Klatzker, whose early career was as a clinician trained in social work. “This (Eleos) is like that on steroids.”

Klatzker echoed Cohen’s sentiments about lessening administrative burdens on providers and added that the AI tools it uses to increase his staff’s ability to provide clinical supervision and adhere to evidence-based practices.

“In an environment where we are all workforce challenges … one thing that works against us is people in our industry burning out or being overwhelmed by useless paperwork,” Klatzker said. “So this kind of intelligent automation is a force multiplier; it allows people to do what they were trained to do. It streamlines lots and lots of functions.”

By improving the work experience, Klatzker hopes to make Gaudenzia more competitive in the labor market.

Looking forward, Gaudenzia hopes to find ways to use AI to enhance its call center operations and revenue cycle management.

For Talkspace, it built its AI tools based on a closed data ecosystem from its own work. These include 110 million therapist-patient messages that encompass 6 billion words. While the organization is not eager to share its data and insights with other organizations, Cohen said Talkspace is open to processing the data of other health care organizations with its AI tools.

“You can go to the big engines — and we all know who they are — but they’re not able to deliver, yet, the mental health sophistication or learnings that our engine has because we are so specialized,” Cohen said.

Companies featured in this article:

, ,