AI software created to help psychotherapists enhance skills
A team employs machine learning to assess the effectiveness of cognitive behavioral therapy sessions and to maintain counselors' competence.
This country’s mental healthcare needs are so overwhelming, relative to psychotherapists’ availability, that they obscure another distressing truth: The training and effectiveness of counselors in the community can vary widely.
“The lack of performance-based feedback and quality indicators reverberates throughout behavioral healthcare,” said psychologist David Atkins. He’s a research professor of psychiatry and behavioral sciences at the University of Washington School of Medicine.
“Healthcare systems are aware of this problem; it’s kind of hiding in plain sight,” he said. “About 80 million Americans need treatment for mental illness or addiction, and only about half get any treatment. So the focus is on access to care versus quality, but increasing access to highly variable care is itself problematic.”
For a decade, Atkins has worked with artificial intelligence and machine-learning scientists to create software that will help behavioral health trainees learn how to counsel effectively and help licensed psychotherapists keep skills sharp.
The researchers’ newest findings about their automated evaluation software were published last week in PLOS ONE.
Atkins offered this background:
“When psychotherapists are in training, we will often record counseling sessions, which get shared with a supervisor, who will review them and provide performance-based feedback. When we finish training, that feedback ends, but the need to learn and grow doesn’t. Mental health professionals are practicing without feedback on how effective a session was or how to improve their skills.”
The research and AI software was done in collaboration with colleagues at the University of Southern California and the University of Pennsylvania. The software analyzes cognitive behavioral therapy (CBT) sessions, where algorithms were trained to discern—on par with the judgment of a human evaluator—the effectiveness of CBT skills.
The software’s task is enormously complex.
From an audio recording of a therapy session, it first must identify who is counselor and who is client. Then it must accurately generate an accurate text transcript from the audio recording, which can be affected by the recording quality itself as well as by the participants’ accented speech.
Finally, the software analyzes the transcript for the same 11 quality indicators that a human would use to assess the therapist’s competence—for instance, in terms of “interpersonal effectiveness” or “focusing on key cognitions and behaviors.”
“Counseling and psychotherapy are skills,” Atkins said. “Consider that professional musicians practice scales, the basic building blocks of music, to help them to maintain high performance overall. We are working to create that kind of skill-maintenance instrument for psychotherapists.”
Atkins and his research colleagues have founded a spinout company, Lyssn, which aims to commercialize the technology.