AI Language Models Could Both Help and Harm Equity in Marine Policymaking: The Case Study of the BBNJ Question-Answering Bot

AI LLMs like ChatGPT are set to reshape some aspects of policymaking processes. Policy practitioners are already using ChatGPT for help with a variety of tasks: from drafting statements, submissions, and presentations, to conducting background research. We are cautiously hopeful that LLMs could be used to promote a marginally more balanced footing among decision makers in policy negotiations by assisting with certain tedious work, particularly benefiting developing countries who face capacity constraints that put them at a disadvantage in negotiations. However, the risks are particularly concerning for environmental and marine policy uses, due to the urgency of crises like climate change, high uncertainty, and trans-boundary impact. To explore the realistic potentials, limitations, and equity risks for LLMs in marine policymaking, we present a case study of an AI chatbot for the recently adopted Biodiversity Beyond National Jurisdiction Agreement (BBNJ), and critique its answers to key policy questions. Our case study demonstrates the dangers of LLMs in marine policymaking via their potential bias towards generating text that favors the perspectives of mainly Western economic centers of power, while neglecting developing countries' viewpoints. We describe several ways these biases can enter the system, including: (1) biases in the underlying foundational language models; (2) biases arising from the chatbot's connection to UN negotiation documents, and (3) biases arising from the application design. We urge caution in the use of generative AI in ocean policy processes and call for more research on its equity and fairness implications. Our work also underscores the need for developing countries' policymakers to develop the technical capacity to engage with AI on their own terms.