2025: Motivating Climate Action Using Human-AI Dialogues
Climate change skepticism and inaction remain major barriers to collective climate action, with traditional static messaging showing little impact on either beliefs or behaviors. Building on evidence that adaptive large language model (LLM) dialogues can durably reduce conspiracy beliefs and climate doubts, this project proposes a new wave of experiments to understand and improve their effectiveness. We will examine why LLM dialogues shift some climate doubts more effectively than others, and why effects on specific concerns surpass broader beliefs and policy support. The research will integrate deeper elicitation of participants’ mental models of climate change, enabling tailored dialogues that target underlying doubts. We will also harness LLM adaptability to create individualized measures of climate action (e.g., contacting elected representatives), allowing direct assessment of behavioral impact. In parallel, we will test strategies for promoting public uptake of the intervention, including survey and field experiments with social media and search ads.
Investigators: David Rand, Ann S. Bowers College of Computing and Information Science; Gordon Pennycook, Arts & Sciences