Our speaker is Angel Hsu. Angel is an Associate Professor of Public Policy and the Environment\, Ecology and Energy Program\, E3P. She is also Founder and Director of the Data-Driven EnviroLab\, an interdisciplinary research group that innovates and applies quantitative approaches to pressing environmental issues. Angel’s research explores the intersection of science and policy and the use of data-driven approaches to understand environmental sustainability\, particularly in the areas of climate change and energy\, urbanisation and air quality. Her research projects apply large-scale datasets derived from satellite remote sensing and other spatially-explicit sources to evaluate environmental policy performance. Focusing particularly on China and the Global South\, Dr. Hsu has provided expert testimony to the US-China Economic Security and Review Commission\, and is a member of the National Committee on US-China Relations\, and a Public Intellectual Program Fellow.
Angel will be presenting:
Calibrating Climate AI: Can AI be used Credibly for Net-Zero Governance?
Generative AI is increasingly being used to interpret climate information\, summarize evidence\, and support decisions about decarbonization and net-zero action. But what happens when these systems produce outputs that sound credible while remaining generic\, weakly evidenced\, or poorly aligned with the policy context? This talk introduces the concept of “greenslop” to describe that problem and explores how it relates to hallucination\, bias\, and other failure modes in climate AI. It also asks a harder question often overlooked in discussions of AI for sustainability: can Climate AI justify its own energy use? Drawing on recent research comparing generic LLMs with domain-specific climate chatbots\, the talk argues that trustworthy climate AI will require more than better models\, including benchmarking\, validation\, transparency\, and human oversight robust enough to make these tools genuinely useful for decarbonization and net-zero governance.
One of the preprints for the talk can be accessed on the following link 'The Energy Footprint of LLM-Based Environmental Analysis: LLMs and Domain Products'.
Angel also recently wrote this op-ed on ‘right-sizing AI' in the Business Times.
URL;VALUE=URI:https://www.lse.ac.uk/granthaminstitute/events/calibrating-climate-ai-can-ai-be-used-credibly-for-net-zero-governance-angel-hsu/ SUMMARY:Calibrating climate AI: Can AI be used credibly for net-zero governance? | Angel Hsu DTSTART;TZID=Europe/London:20260624T140000 END:VEVENT END:VCALENDAR