How can we use large language models to make mental health chatbots smarter?

  • Problem Description

    • Treatment resources for mental disorders, such as psychotherapy, are scarce with long waiting lists. Chatbots can deliver low-threshold, flexible, scalable and cost-effective interventions aiming at promoting psychological well-being.
    • Over the year, we have developed a mobile app in which a chatbot teaches users the techniques of cognitive behavioural therapy to improve their mental health. However, the chatbot's natural language capabilities are low and users mostly interact through buttons. On the one hand, this limits the risks of harmful responses. On the other hand, these robotic conversations can frustrate users, which limits treatment success.
    • Over the past months, large language models have become increasingly popular - offering a solution to make chatbots smarter and better equipped to help with individual issues. Yet, in the context of mental disorders, the risks of wrong responses are harmful. Therefore, the integration of large language models needs to be approached carefully and with limited scope.

     

    Goal of the Thesis

    • Explore the current state of the art in using large language models in mental health applications
    • Review our chatbot-based mobile app and other chatbot apps for mental health
    • Conceptualise and explore how to integrate large language models into our app

     

    Requirements

    • Interest in large language models and mental health
    • Willingness and ability to tackle a wide space of potential solutions
    • Programming skills (e.g. Python, JavaScript, Flutter, Rust)
    • Knowledge of large language models and how to contextualise them (e.g. prompt engineering, fine-tuning)

     

    Contact

    If you are interested and want to apply for this topic, please contact Florian Onur Kuhlmeier(florian.kuhlmeier@kit.edu) with a short motivation statement, your CV, and a current transcript of records. Feel free to reach out beforehand if you have any questions.

     

    References

    • Bendig, Eileen, Benjamin Erb, Lea Schulze-Thuesing, and Harald Baumeister. “Die nächste Generation: Chatbots in der klinischen Psychologie und Psychotherapie zur Förderung mentaler Gesundheit – Ein Scoping-Review.” Verhaltenstherapie 29, no. 4 (2019): 266–80. https://doi.org/10.1159/000499492.
    • Lim, Shi Min, Chyi Wey Claudine Shiau, Ling Jie Cheng, and Ying Lau. “Chatbot-Delivered Psychotherapy for Adults With Depressive and Anxiety Symptoms: A Systematic Review and Meta-Regression.” Behavior Therapy, Oktober 2021. https://doi.org/10.1016/j.beth.2021.09.007.
    • Fitzpatrick, Kathleen Kara, Alison Darcy, and Molly Vierhile. “Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial.” JMIR Mental Health 4, no. 2 (June 6, 2017): e19. https://doi.org/10.2196/mental.7785.
    • Ahmed, Arfan, Asmaa Hassan, Sarah Aziz, Alaa A Abd-alrazaq, Nashva Ali, Mahmood Alzubaidi, Dena Al-Thani, et al. “Chatbot Features for Anxiety and Depression: A Scoping Review.” Health Informatics Journal 29, no. 1 (January 1, 2023): 14604582221146720. https://doi.org/10.1177/14604582221146719.
    • Kuhlmeier, Florian Onur, Ulrich Gnewuch, Stefan Lüttke, Eva-Lotta Brakemeier, and Alexander Mädche. “A Personalized Conversational Agent to Treat Depression in Youth and Young Adults – A Transdisciplinary Design Science Research Project.” In The Transdisciplinary Reach of Design Science Research, edited by Andreas Drechsler, Aurona Gerber, and Alan Hevner, 30–41. Lecture Notes in Computer Science. Cham: Springer International Publishing, 2022. https://doi.org/10.1007/978-3-031-06516-3_3.
    • Kumar, Harsh, Ilya Musabirov, Jiakai Shi, Adele Lauzon, Kwan Kiu Choy, Ofek Gross, Dana Kulzhabayeva, and Joseph Jay Williams. “Exploring The Design of Prompts For Applying GPT-3 Based Chatbots: A Mental Wellbeing Case Study on Mechanical Turk.” arXiv, September 22, 2022. http://arxiv.org/abs/2209.11344.