Simulating Psychotherapeutic Sessions With Generative Artificial Intelligence

Simulating Psychotherapeutic Sessions With Generative Artificial Intelligence: A Proof-of-Concept Study of In Silico Psychotherapy Research

The study assesses the potential of using computational models, specifically large language models, to simulate psychotherapeutic sessions, aiming to improve therapy outcomes and advance therapist training through innovative technology.

Study Overview

• Status: Completed
• Conditions: Mental Disorder
• Intervention / Treatment: Behavioral: High Levels of Common Therapeutic Factors; Behavioral: Low Levels of Common Therapeutic Factors; Behavioral: Standard motivational interviewing

Detailed Description

Health research has evolved significantly, increasingly incorporating computational models that improve our understanding and effectiveness of medical interventions. This shift from traditional to computational methods represents a major advancement in medical research, offering a more sustainable and innovative approach for conceptual advances and therapeutic discovery. In silico models, based on scientific simulation, use computational algorithms to mimic real-world systems or processes. This virtual environment allows researchers to explore phenomena impractical, unethical, dangerous, expensive, or impossible to study otherwise.

Psychotherapy is widely acknowledged as a primary treatment for a variety of mental health conditions, from depression and anxiety to personality disorders, offering significant pathways to recovery and improved quality of life. Yet current methods have shown limited effectiveness, prompting a need for innovative research approaches. In silico psychotherapy research leverages computational simulations, large language models (LLMs), and generative artificial intelligence to explore and refine psychotherapeutic interventions. By simulating human-like conversations, this approach provides insights into therapy dynamics and holds promise for revolutionizing therapist training and expanding treatment techniques.

This study aims to establish a proof-of-concept for simulating psychotherapeutic sessions using LLMs, focusing specifically on motivational interviewing. It involves the simulation of 512 psychotherapy sessions using LLMs as well as 8 real-world psychotherapy transcripts. By modeling human interactions, the study seeks to enhance healthcare delivery, therapist training, and personalized psychotherapy.

• Study Type: Interventional
• Enrollment (Actual): 520
• Phase: Not Applicable

Contacts and Locations

Study Locations

• Switzerland
  • Basel, Switzerland, 4031
    • University Hospital Basel

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: Adult; Older Adult
• Accepts Healthy Volunteers: No

Description

Inclusion Criteria:

• Simulation of psychotherapy sessions of conversations between an adult person presenting with a mental or behavioral health problem and a psychotherapist using large language models and 8 real-world transcripts

Exclusion Criteria:

• Simulation protocols with severe simulation errors

Study Plan

How is the study designed?

Design Details

• Primary Purpose: Other
• Allocation: Non-Randomized
• Interventional Model: Parallel Assignment
• Masking: Single

Number of Arms

3

Arms and Interventions

Participant Group / Arm	Intervention / Treatment
Experimental: High Levels of Common Therapeutic Factors; In this group, the patient-large language model (LLM) interacted with a therapist-LLM prompted to exhibit high levels of positive common factors.	Behavioral: High Levels of Common Therapeutic Factors; The therapist large language model (LLM) is designed to show high levels of empathy, warmth, and genuineness. This setup aims to create a supportive and trusting therapeutic environment to improve patient engagement. High levels of these positive factors are linked to better psychotherapy outcomes and a stronger therapist-patient relationship.
Experimental: Low Levels of Common Therapeutic Factors; In this group, the patient-large language model (LLM) interacted with a therapist-LLM prompted to exhibit low levels of positive common factors.	Behavioral: Low Levels of Common Therapeutic Factors; The therapist LLM for this group is designed to show low levels of empathy, warmth, and genuineness. This setup aims to examine how a less supportive and empathetic therapist affects psychotherapy sessions. Lower levels of these positive behaviors can lead to reduced patient engagement and a weaker therapist-patient relationship, potentially hindering therapy outcomes.
Other: Transcripts of real intervention sessions; This group consists of published transcripts of real intervention sessions, in which motivational interview techniques have been applied.	Behavioral: Standard motivational interviewing; Motivational interviewing techniques as applied during the sessions on which the transcripts are based.

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Simulation's Accuracy in generating Psychotherapeutic Dialogues	Assessment of the simulation's ability to accurately produce psychotherapeutic dialogues that adhere to the principles and techniques of motivational interviewing (MI), as determined by the average global scores of the Motivational Interviewing Treatment Integrity (MITI) code 4.2. The MITI code 4.2 includes various subscales, such as empathy and MI spirit, each scored on a scale from 1 to 5, with lower scores suggesting a need for improvement in MI delivery, while higher scores reflect stronger therapeutic skills and better patient outcomes.	12 months

Secondary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Number of Errors/Deviations	The number of errors or deviations from expected psychotherapeutic practices is counted, providing a quantitative measure of simulation quality. This measure also serves as exclusion criteria from any other assessment.	12 months
Metric of Verbal Content (Therapist)	Assessment of the text metrics of the therapist, based on the number of sentences, words, syllables, characters, and lexical diversity.	12 months
Metric of Verbal Content (Patient)	Assessment of the text metrics of the patient, based on the number of sentences, words, syllables, characters, and lexical diversity.	12 months
Turn-takings	Assessment of the turn-takings, based on the number of exchanges between the therapist and patient within a session, indicating the dynamic interaction flow.	12 months
Improvement of Patient	Improvement of the patient is evaluated using an an observer-rated, circularly framed version of the Importance and Confidence Rulers, measuring the simulated patient's psychotherapeutic progress on a circular scale from 0 to 10. Lower scores indicate lower perceived importance or confidence, while higher scores suggest greater perceived importance or confidence in making the change.	12 months
Credibility of Patient Behavior	The credibility of the patient's behavior is estimated using a 0 to 10 scale indicating how likely the evaluator found that the participants are real humans or simulations, offering insight into the perceived authenticity of the simulated interactions. The credibility of the 8 real-world transcripts served as a comparison baseline/benchmark for this evaluation. Lower scores indicate lower authenticity of the patient large-language model's (LLM's) simulated behavior, while higher scores suggest higher authenticity of the patient LLM's simulated behavior.	12 months
Credibility of Therapist Behavior	The credibility of the therapist's behavior is estimated using a 0 to 10 scale indicating how likely the evaluator found that the participants are real humans or simulations, offering insight into the perceived authenticity of the simulated interactions. The credibility of the 8 real-world transcripts served as a comparison baseline/benchmark for this evaluation. Lower scores indicate lower authenticity of the therapist large-language model's (LLM's) simulated behavior, while higher scores suggest higher authenticity of the therapist LLM's simulated behavior.	12 months
Manipulation Check	The implemented level of psychotherapeutic common factors by the therapist-LLM is approximated using the Therapist Empathy Scale (TES) as rough manipulation checks. The TES rates the therapist's ability to understand and share a patient's feelings on a scale from 1 to 7. Higher scores indicate greater empathy, reflecting a stronger connection and understanding of the patient's emotions, while lower scores suggest less empathy.	12 months
Manipulation Check	The implemented level of psychotherapeutic common factors by the therapist-LLM is approximated using the Working Alliance Inventory Short Observer form (WAI-S-O) as rough manipulation checks. The Working Alliance Inventory uses 12 items evaluating the perceived therapeutic alliance between the patient and the therapist on a scale of 1 to 7.	12 months

Collaborators and Investigators

• Sponsor: University Hospital, Basel, Switzerland
• Collaborators: RWTH Aachen University; University of Basel; University of Trier
• Investigators: Principal Investigator: Gunther Meinlschmidt, Prof. Dr., University Hospital and University of Basel

Study record dates

Study Major Dates

• Study Start (Actual): February 1, 2025
• Primary Completion (Actual): August 31, 2025
• Study Completion (Actual): August 31, 2025

Study Registration Dates

• First Submitted: February 3, 2025
• First Submitted That Met QC Criteria: February 3, 2025
• First Posted (Actual): February 6, 2025

Study Record Updates

• Last Update Posted (Actual): May 22, 2026
• Last Update Submitted That Met QC Criteria: May 18, 2026
• Last Verified: May 1, 2026

More Information

Terms related to this study

• Keywords: Motivational Interviewing; In silico Psychotherapy; Artificial intelligence (AI) applications
• Additional Relevant MeSH Terms: Mental Disorders
• Other Study ID Numbers: 0000-00000; th24Meinlschmidt

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No