Abstract for: Building and Learning with Models Using AI

Introduction: Artificial Intelligence, especially large language models, offers System Dynamics a transformative opportunity. LLMs can reduce learning barriers, and enhance productivity, without altering System Dynamics’ core methodology. By treating LLMs as tools for organizing and interpreting end-user input, the sd-ai platform enables natural language interaction with models to build and analyze them. This research produced two engines: quantitative for model construction and Seldon for model analysis, exploration, and ideation. Approach: With information-dense schemas for passing content to LLMs for the purposes of model construction and basic prompting schemes that target the transformation of end-user provided information into models and model related content, the engines leverage the information transformation properties of LLMs, reducing the incidence of hallucination and allowing LLMs to build and analyze models. Results: The quantitative engine and Seldon using current day LLMs are able to build and explain models with minimal instruction or correction from the end-user. This frees the end-user to be concerned about validity and problem solving rather than implementation or analysis details. Discussion: System Dynamics faces a turning point: present-day LLMs can now perform modeling and analysis with remarkable skill, threatening to absorb the field into broader information sciences. Yet this shift is a chance to evolve. By embracing LLMs, we can preserve System Dynamics’ core insights on feedback and delays while making modeling more accessible than ever before.