Anticipating the challenges of AI in climate governance: an urgent dilemma for democracies

There is increasing interest in AI as a means of accelerating climate policy interventions. While undoubtedly promising, AI's recent history in other fields demonstrates the risk of significant unintended consequences that widen social inequalities or reduce democratic engagement. In this perspective, we review recent developments in climate governance and in AI governance and anticipate several potential problems when the two are combined. In particular, we highlight potential democratic challenges for the application of AI in climate governance through narrowing the range of policy options, narrowing the range of experts and publics that can contribute to climate governance, and how the implementation of AI may run counter to norms of democratic accountability. These challenges represent an urgent dilemma for climate governance as ignoring these issues will erode democratic oversight, lead to unpopular unintended consequences, and could reverse recent positive trends in diversity and participation within climate science and policy. In contrast, engaging with them could strengthen democracy and increase the successful social uptake of the technologies. By way of mitigating these risks, we introduce four principles for a bounded application of climate AI technologies that recognizes and enhances understanding of the political and contested nature of environmental decision‐making. First, situating AI within expert and lay public debates. Second, valuing non‐quantifiable knowledge. Third, expanding deliberation within AI decision‐making. Fourth, developing domain‐specific AI applications.