The role of artificial intelligence in climate change scientific assessments

Abstract

Climate change scientific assessments prepared by the Intergovernmental Panel on Climate Change (IPCC) face interconnected dual challenges: the exponential growth of literature, hindering synthesis efficiency, and the increasing length of its reports, impeding accessibility. Building upon the emerging discussion of adopting artificial intelligence (AI) tools in scientific assessments, this essay develops specific operational and governance frameworks to guide the IPCC’s integration of these tools. It makes three distinct contributions. First, it develops a systematic framework for AI-augmented evidence synthesis, detailing how machine learning (ML) can be integrated into each stage of the assessment workflow. Second, it provides a critical analysis of Large Language Models' (LLMs) use for reports communication through the lens of ‘addressable’ versus ‘inherent’ limitations, clarifying which risks require technical solutions versus those that demand robust governance. Finally, it proposes a novel governance structure for the IPCC based on two institutional roles, the ‘producer’ and the ‘assessor’ of AI products, to ensure scientific integrity is maintained. This essay provides a clear path for the responsible, expert-led integration of AI, ensuring it serves to augment, not replace, human expertise.