Second Green AI Summit at Harvard and Boston University Successfully Convened
Session Date: April 25, 2025 Location: Harvard University Gutman Conference Center
Overview:
This pivotal session on Day 1 showcased concrete initiatives aimed at scaling sustainable AI practices across infrastructure, policy development, and research. Presenters from leading research institutions and policy centers introduced collaborative efforts designed to tackle the environmental footprint of AI and harness its potential responsibly.
Launched Initiatives:
EPRI’s Open Power AI Consortium: Presented by Jeremy Renshaw, Executive Director of AI and Quantum at the Electric Power Research Institute (EPRI), this global collaboration aims to provide higher-performing, smaller, and more efficient AI models specifically for the energy sector. The consortium focuses on three workstreams: curating open-source data, models, and tools tailored for power industry needs; developing AI sandboxes for testing and validation in realistic, secure environments; and facilitating real-world implementation with feedback loops to continuously improve models. Renshaw emphasized building domain-specific models by fine-tuning existing open-source models with energy-specific data (both public and private/anonymized) and invited broad participation from utilities, academia, and tech partners to de-risk deployment and share learnings globally. He also mentioned EPRI's parallel work on data center flexibility.
The Power and AI Initiative (PAI) at Harvard SEAS: Introduced by Professor Le Xie, Gordon McKay Professor of Electrical Engineering at Harvard's School of Engineering and Applied Sciences (SEAS), the PAI initiative focuses on the critical two-way street between the power sector and AI. Recognizing the compute-driven power surge and the potential for AI to optimize grid operations, PAI aims to foster research and education at this intersection. Professor Xie shared research examples, including analyzing the grid flexibility offered by large compute loads (like crypto data centers) and evaluating the capabilities and limitations of foundation models for power system applications, such as wildfire prediction and understanding scaling laws for fine-tuning. The initiative includes an upcoming symposium and short course designed to bridge the knowledge gap between the power and AI communities.
Dynamic Governance Model for AI Policy: Paulo Carvao, Senior Fellow at the Harvard Kennedy School's Mossavar-Rahmani Center for Business and Government, discussed the need for policy innovation alongside technological advancements. Based on research involving industry leaders and US Congress members, he highlighted findings such as concerns about policy fragmentation, a desire to maintain US AI leadership, and a lack of trust between sectors. Carvao proposed a "Dynamic Governance Model" as a framework for US AI policy. This adaptive, incremental approach involves government setting policy goals (e.g., data center efficiency), followed by public-private partnerships creating standards, market-driven solutions for compliance and auditing (including regulatory sandboxes), and finally, systems for accountability and liability, creating a feedback loop as technology evolves.
Speakers:
Jeremy Renshaw: Executive Director, AI and Quantum, Electric Power Research Institute (EPRI)
Le Xie: Gordon McKay Professor, Electrical Engineering, Harvard John A. Paulson School of Engineering And Applied Sciences
Paulo Carvao: Senior Fellow, Mossavar-Rahmani Center for Business and Government, Harvard Kennedy School
This session provided concrete examples of how collaboration across institutions and sectors is being mobilized to create the frameworks, tools, and knowledge needed to guide AI towards a more sustainable trajectory.