Misrepresentation or inclusion: promises of generative artificial intelligence in climate change education

Generative Artificial Intelligence (AI) technologies, including large language models (LLMs) that can generate novel text output, present promise for creating tailored science communication for broad audiences. However, LLMs might reflect inaccuracies and social biases from their training sources. In this work, we examine the promises and challenges of using LLMs to depict climate issues from intersectional perspectives. We prompt an LLM (GPT-4) to generate content about localized climate issues and simulate different communication mediums and intersectional identities. We conduct content analysis of the responses, drawing from Intersectional Climate Justice and Culturally Sustaining Pedagogies frameworks. Findings suggest that the LLM-created responses can restate climate justice principles in the prompts and do not frequently show inaccuracy. However, they may lack elaboration, show deficit framing, and overlook identity aspects. We discuss suggestions from critical education research, to question the assumptions underlying AI technologies and explore ways to promote inclusive climate education.