Supporting systems thinking in environmental science education with large language model chatbots

Understanding the climate system requires systems thinking to consider intersecting components and interrelationships within the atmosphere, geosphere, hydrosphere, and biosphere. Developing systems thinking is challenging, as students often identify simple, linear relationships instead of more complex causal ideas. Modelling practices, including developing, evaluating, and refining models, can enhance systems thinking. While research on modelling has focused on ecological and physical processes, we designed chatbots powered by large language models (LLMs) to highlight perspectives about biodiversity and climate change’s impact. We piloted the chatbots in a two-week high school environmental science curriculum. Students created system models in groups and engaged in open-ended interactions with the chatbots to revise models. Data sources included students’ group models before and after chatbot interactions, written reflections, and chat logs. We qualitatively coded the data for shifts in systems thinking (indicated by components and causal relationships in models) and perspective taking (how students articulated sociocultural and economic factors, scientific facts, and system complexity). Analyses revealed that students incorporated more system components and complex causal ideas following chatbot interactions. We found shifts in perspective taking regarding socioeconomic and scientific climate issues. We discuss how science instruction can scaffold complex causal ideas and effective interactions with LLMs-based tools.