Promoting empathy in decision-making by turning agent-based models into stories using large-language models
Philippe J. Giabbanelli, Cedric Daumas, Noé Y. Flandre, Ashutosh Pitkar, Jessica Vazquez-Estrada · Journal of Simulation · 2025
This study explores how empathy can be integrated into decision-making within three Agent-Based Models (ABMs) of disasters and migrations. We design, implement, and evaluate methods to translate the experiences of simulated agents into empathetic narratives through Large Language Models (LLMs), using GPT-4 as a guiding example. We compare two approaches: a direct method that prompts to create empathetic stories (thus leaving it to GPT’s interpretation of empathy), and an indirect method using style transfer by adopting the voice of well-known empathetic figures. Using a Design of Experiments framework, we evaluate the impact of factors including text length and temperature on readability, sentence accuracy, and character authenticity. Results show the indirect method yields better quality narratives. Human readers generally agree that our generated stories show genuine emotions, although full empathy is limited by the extreme scenarios simulated. Our work is provided open-source to support researchers in transforming their ABMs into narratives.