Application and scenario simulation of multimodal GPT in circular economy transformation: A case study of Taiwan's material flow data

Abstract

Despite growing interest in AI-driven environmental research, the use of multimodal GPT in circular economy transformation remains underexplored. This study bridges this gap by demonstrating how GPT interprets circular economy system diagrams—such as stock-flow and causal loop diagrams—and translates them into executable software models. By integrating textual descriptions with mathematical equations, GPT establishes dynamic relationships among software objects, capturing interactions between industrial activities, pollutant emissions, material flow indicators, and system stocks. Additionally, GPT enhances system visualization, enabling multi-level analysis and key factor identification. Beyond traditional modeling, GPT improves scenario simulation by evaluating parameter variations and optimizing decision-making, supporting evidence-based policy formulation. Using Taiwan’s material flow data (2013–2022), this study develops system dynamics models, designs future scenarios, and assesses circular economy policies’ potential impacts by 2030. The findings present an AI-assisted approach for policymakers to evaluate and accelerate circular economic transformation.