MIT Climate & Sustainability Consortium Project Highlights Introduction and Update Jeremy Gregory Executive Director, MIT Climate & Sustainability Consortium
The Climate and Sustainability Implications of Generative AI Noman Bashir Computing & Climate Impact Fellow, MIT Climate & Sustainability Consortium
The rapid expansion of generative artificial intelligence (Gen-AI) neglects consideration of negative effects alongside expected benefits. This incomplete cost calculation promotes unchecked growth and a risk of unjustified techno-optimism with potential environmental consequences, including expanding demand for computing power, larger carbon footprints, and an accelerated depletion of natural resources. The current siloed focus on efficiency improvements results instead in increased adoption without fundamentally considering the vast sustainability implications of Gen-AI.
In this talk, I will propose that responsible development of Gen-AI requires a focus on sustainability beyond only efficiency improvements and necessitates benefit-cost evaluation frameworks that encourage (or require) Gen-AI to develop in ways that support social and environmental sustainability goals alongside economic opportunity. However, a comprehensive value consideration is complex and requires detailed analysis, coordination, innovation, and adoption across diverse stakeholders. Engaging stakeholders, including technical and sociotechnical experts, corporate entities, policymakers, and civil society, in a benefit-cost analysis would foster development in the most urgent and impactful directions while reducing unsustainable practices. More details are in our white paper, which is accessible at MIT Gen-AI Sustainability White Paper.
A Cautionary Tale about Deep Learning-based Climate Emulators Björn Lütjens Postdoctoral Associate, MIT Department of Earth, Atmospheric, and Planetary Sciences
Climate models are computationally very expensive for exploring the impacts of climate policies. For example, simulating the impacts of a single policy emission scenario can take multiple weeks and cost hundreds of thousands of USD in computing. Compellingly, deep learning models can now forecast the weather in seconds rather than hours in comparison to conventional weather models and are being proposed to achieve similar reductions by approximating climate models. Climate approximations or emulators, however, have already been developed since the 1990s and I will present how we implemented a linear regression-based emulator that outperforms a novel 100M-parameter transformer-based deep learning emulator on the most common climate emulation benchmark. I will use our results to discuss more nuanced insights highlighting how chaotic dynamics influence emulator performance and use cases where deep-learning emulators can improve existing linear emulators.
Collaborative Development of an Interactive Decision Support Tool for Trucking Fleet Decarbonization Danika MacDonell Impact Fellow, MIT Climate & Sustainability Consortium
This presentation shares the journey of creating an interactive geospatial decision support tool in close collaboration with industry and academic partners of the MIT Climate & Sustainability Consortium. The tool leverages comprehensive public data on freight flows, costs, emissions, infrastructure, and regulatory incentives. Integrating key insights and methodologies from our partners, it aims to assist trucking industry stakeholders in identifying and assessing strategies to transition fleets to low-carbon energy carriers.