Post by Else Kröner Fresenius Center for Digital Health

How can we better understand mental health conditions and develop new therapeutic approaches by using large language models? In a publication in The Lancet Digital Health, an interdisciplinary research team at EKFZ for Digital Health shows that LLMs can serve as experimental systems for psychological research. The researchers induced seven different affective states such as anxiety, sadness, or stress systematically in six different LLMs. They analyzed the reactions and were able to down-regulate them using an established mindfulness-based method. The models also exhibited cognitive patterns known from depression research, such as negativity bias. LLMs can serve as valuable addition to biological and clinical studies: ▫️ Experiments can be conducted under controlled, reproducible and scalable conditions. ▫️Hypotheses can be refined prior to further studies. ▫️New therapeutic approaches can be explored in silico before clinical trials. At the same time, the authors emphasize that these systems do not have real emotions. Reactions are based on learned patterns. LLMs are not a substitute for traditional research method, but a powerful complementary tool. The modelling study is a joint effort of the following researchers: Dr. Magdalena Wekenborg (née Kanthak), Elizabeth A. M. Michels, Georg Kurze, Matti Lasse Kropp, Fabian Wolf, Josi Harzbecker, Isabella Wiest, Jakob Nikolas Kather. It highlights how interdisciplinary collaboration at the intersection of psychology, medicine, and computer science enables new research approaches and innovative solutions. Technische Universität Dresden TUD | Faculty of Medicine Universitätsklinikum Carl Gustav Carus Dresden Read full publication ➡️ https://lnkd.in/dYj3dBhg