Kevin Cowley remembers many things when it comes to April 15, 1989. He had taken the bus to the Hillsborough football stadium in Sheffield, England, to see the semi-finals of the championship between Nottingham Forest and Liverpool. He was 17 years old at the time and it was a beautiful, sunny afternoon. The fans filled the stands in the stands.
Cowley remembers that he was pressed so tightly between the people that he couldn’t take his hands out of his pockets. He remembers the collapse of the security barrier, which collapsed behind him when his team almost scored a goal and the crowd went wild.
Hundreds of people suddenly fell like dominoes – knocked over by those trapped next to them. Cowley was drawn into the crowd like a lake. He remembers waking up suddenly among the dead and dying, compressed under the weight of their bodies. He remembers the smell of urine and sweat, the screams of these people. He remembers looking in the eyes of the man struggling to survive next to him and then standing on top of him to save himself. He still wonders today if this man was one of the 94 people who died at the stadium that day.
Flashbacks for half a lifetime
All of these memories haunted Cowley throughout his adult life. He suffered from flashbacks and insomnia for 30 years. He had difficulty working but was too ashamed to speak to his wife about it. He suppressed the worst thoughts by drinking. In 2004 a doctor referred him to a young therapist in training, but he did not help and he stopped the therapy after a few sessions.
But two years ago, Cowley saw advertisements for a therapy provider on the Internet – and he decided to try again. After dozens of regular sessions of texting his therapist, Cowley, now 49, has finally recovered from his severe post-traumatic stress disorder. “It’s amazing how a few words can change a life,” says Andrew Blackwell, chief scientist at Ieso, a UK psychiatric clinic that treats Cowley.
The key was to hear the right words at the right time. Blackwell and his colleagues at Ieso are pioneering a new approach to mental health care that uses an AI system to analyze the language used in therapy sessions. The idea is to use natural language processing (NLP) to determine which parts of a therapist-patient conversation – what types of utterances and what precise linguistic exchange of emotions – are most effective in treating various disorders.
More quality in treatment
The aim is to give therapists better insight into their work in order to help experienced clinicians maintain a high standard of care – and to help trained psychologists get better. Given the global mental health shortage, an automated form of quality control could be critical in helping existing facilities finally meet demand.
Ultimately, this approach could shed light on how psychotherapy works in the first place – something that, surprisingly, practitioners and researchers are still largely in the dark about. A new understanding of the effectiveness of talk therapy could open the door to personalized psychiatric care, allowing therapists to tailor psychiatric treatments to specific clients, much like they are increasingly doing when prescribing medication.
Researchers have been trying to study talk therapy for years to find out why some therapists get better results than others. It’s an art, but also a science. Success is often based on the experience and gut feeling of qualified therapists. So far, it has been virtually impossible to fully quantify what works in therapy and why. Zac Imel, a psychotherapy researcher at the University of Utah, recalls trying to hand-analyze transcripts of therapy sessions himself. “It takes forever – and the sample size is embarrassingly small,” he says. “And so we haven’t learned much in the decades we’ve already done it.”
The computer can help
AI changes this calculation. The machine learning technique that does the automatic processing can analyze large amounts of language quickly. This gives researchers access to an inexhaustible, untapped source of data: the language that therapists use. The scientists believe they can use the insights from this data to give talk therapy a long overdue boost. The result could be that more people feel better again – and that this condition remains.
Blackwell and his colleagues aren’t the only ones pursuing this vision. A US company, Lyssn, is developing a similar technique. Lyssn was co-founded by Imel and CEO David Atkins, who studies psychology and machine learning at the University of Washington. As mentioned, the AI systems are trained with transcripts from therapy sessions. To feed the NLP models, a few hundred transcripts are hand-annotated to highlight the role that the therapist’s and patient’s words play at this point in the session. For example, a session can begin with the therapist greeting the person concerned and then moving on to discussing the patient’s current mood. In a later conversation the therapist can empathize with the patient’s problems and ask him whether he has performed the exercises discussed in the previous session. All of this goes into greater detail.
The technology works in a similar way to an algorithm for analyzing moods, which can recognize whether film reviews are positive or negative – or like a translation program that learns to combine English and Chinese. In this case, however, the AI translates the natural language into a kind of barcode or fingerprint of a therapy session, which reveals the role of the various utterances.
Fingerprint from therapy sessions
A session’s fingerprint can show how much time was spent on constructive therapy and how much was spent on general chatter. This display can help therapists focus more on constructive conversation in future sessions, says Stephen Freer, Ieso’s chief clinical officer, who guides the clinic’s 650 or so therapists.
