Abstract for: Evaluation and extension of models using artificial intelligence

Artificial intelligence use in system dynamics has primarily been applied prior to modeling and model construction. We use AI to comment of the quality of a system dynamics model and perform an experiment using multiple system dynamics tools. We also speak to the applicability of AI in this endeavor and recommendations for its use. During the verification of a public health system dynamics model we attempted to use AI to evaluate the quality of the model instruction. Motivation for this came from a desire to provide excellent modeling and the relative inexperience of one team member in the system dynamics method. This was accomplished by providing a list of all the models equations to ChatGPT o1. Across the four implementations, three of which are translations from the original DYNAMO, we find that ChatGPT is relatively consistent across tools. Except for DYNAMO, the other three tools used a combined DYNAMO and standard naming convention for variables.(Lai et al. 2001). We assume that this aided our experimental design. Based upon the simple experiment (one model, one software implementation) and more complex (but incomplete) expanded tests; we strongly recommend that AI be used for completed or under construction models of system dynamics models. The recommendation is qualified. Examination of the output from ChatGPT shows that the source code format and organization have an impact on the AI engine’s analysis. Best practice in model construction that includes naming, dimensional correctness and modularity will return clearer results from an AI engine. Extensive repeated use of ChatGPT for complete model codes.