–  A few years ago, I was approached by Dr. Izabela Stefaniak, a psychiatrist and therapist who at the time was working at the Institute of Psychiatry and Neurology in Warsaw. She had an idea for a study, but she needed technical support – says Prof. Artur Janicki from the Faculty of Electronics and Information Technology.

Visualising inner voices

The study focused on the treatment of vocal hallucinations, which often accompany schizophrenia, but also occur in other diseases. They hinder everyday functioning and lower the self-esteem of people who hear them.  To improve the patients’ quality of life, psychiatrists usually use pharmacological treatments. However, other methods are increasingly being used.

– In our case, it was a computer avatar (‘a talking head') controlled by a therapist. During the first two sessions, the computer-generated character repeated negative sentences that the patient ‘heard’ in their head. For the next four sessions these phrases were switched for content with positive connotations – explains Prof. Janicki. – A doctor supported the patients throughout the process, teaching them how to deal with voices and how to defy them.

Patients responded very well to the new method. Indicators on both the psychotic symptom rating scale and the hallucination strength measure decreased significantly. These decreased ratings persisted even after three months. The study also revealed that people undergoing therapy were interested in this kind of work with computers.

– The study included 23 patients. The results were published in 2019 in the 'Schizophrenia Research' journal. They encouraged us to go a step further – says Prof. Janicki.

My name is Terabot. I'd like to help you with your therapy

A team from the Faculty of Electronics and Information Technology developed a dialogue system called Terabot, which has been piloted in cognitive-behavioural therapy (CBT).

– Unlike the previous project, the new system was automatic, not controlled by a psychiatrist. Terabot had a conversation with the patient, recognised what they were saying and what their emotional state was. Depending on the answers given, the system adjusted its response – explains Prof. Janicki.

The aim of the study was to train patients in how to cope with emotions and to develop mindfulness. The three exercises developed by Dr. Stefaniak concerned the emotions that patients face most often – fear, shame and anger.

– The patient agreed with the doctor what they wanted to focus on. The first part of the session was a conversation with Terabot that included elements of CBT therapy. In the second part, the system suggested a relaxation exercise – says Prof. Janicki.

The Terabot interface

Waiting for results

The study involved 38 patients and was carried out as part of the “Terabot – a dialogue system for cognitive-behavioural therapy" project. Studies are currently underway to determine how much the system has helped the patients.

– During the project, we conducted surveys and we have a lot of feedback. Many patients responded very well. It turned out that they were bolder in front of the computer and talked more than they would to a doctor – says Prof. Janicki.

There are areas that Prof. Janicki and his team would like to work on and develop. – Terabot doesn’t always understand the patient’s intention. We’d like Terabot to be more and more empathetic, for example be able to easily recognise that the patient has finished speaking and not wait too long with an answer, or – the contrary – to not interrupt the patient. We’d also like to have our system more widely available – stressed Prof. Janicki.

The Terabot dialogue system diagram

The “Terabot – a dialogue system for cognitive-behavioural therapy” project ran from January 2021 until June 2023 and was financed as part of the “Excellence Initiative – Research University” programme implemented at WUT. It was among the winners of the “SzIR-2” research grant competition in the Artificial Intelligence and Robotics Priority Research Area.

Prof. Artur Janicki, Dr. Marek Kozłowski, Dr. Izabela Stefaniak, Artur Zygadło, Karolina Gabor-Siatkowska, Marcin Sowański, Rafał Rzatkiewicz and Adam Serodziński participated in the project.