Abstract for: Exploring the use of Large Language Models with Chain-of-thought Prompting as an Aid in Causal Loop Diagram Development
The rapid development of Generative Artificial Intelligence (GenAI) has sparked interest in its application to modeling and simulation (M&S). Large language models (LLMs) in the domain of GenAI have shown remarkable progress in replicating human learning and thinking processes with structured querying. This study explores how LLMs with chain-of-thought (COT) prompting can improve the effectiveness of collaborative group model building in the early stages of the systems modeling process. A human-in-the-loop approach was employed to incorporate LLMs in the group modeling process(GMB). The study focused on key tasks in the early stages of system modelling(causal loop diagramming) using three LLMs (ChatGPT, Claude AI, and DeepSeek). Few-shot COT prompting based on the scaffolding paradigm in education theory was utilized to enhance the accuracy of LLM-generated results. Results were validated against the ground truth established through the GMB process. The final CLD specification included 12 feedback loops (6 reinforcing and 6 balancing) for ChatGPT and Claude AI, while DeepSeek identified only 6 reinforcing loops. Betweenness centrality analysis highlighted chronic disease burden and frailty burden as the most influential nodes, aligning with modelers' expectations. Precision, recall, and F1-score demonstrated high validity. The scaffolded COT prompting methodology proved to be an effective method for improving LLMs output. This study demonstrates the feasibility of an augmented intelligence approach to CLD generation leveraging on COT prompting. While promising, further research is needed to evaluate different prompting techniques, alternative LLMs, and additional use cases. Within the methodology, LLMs were used in the process