Abstract
The NSF Workshop on Sustainable Computing for Sustainability (NSF WSCS 2024) established a strategic roadmap for the intersection of computer science and environmental stewardship. This report outlines the dual challenge of reducing the environmental footprint of computing while maximizing its potential to solve global sustainability crises.
Introduction
Computing is a double-edged sword for the environment. While it is a significant source of carbon emissions and electronic waste, it remains our most potent tool for addressing the climate crisis. The NSF WSCS 2024 workshop aimed to bridge the gap between computer science and domain-specific fields like agriculture and materials science to create a unified framework for interdisciplinary action.
Problem Statement
The research community faces two interconnected challenges. First, Sustainable Computing addresses the growing environmental footprint of the AI boom, data center energy consumption, and the "embodied carbon" of hardware manufacturing. Second, Computing for Sustainability focuses on the transition of technical solutions—such as climate modeling or grid optimization—from theoretical research to operational status. Currently, these solutions often fail due to a lack of local granularity, neglect of human behavior, or data silos.
Methodology
The workshop employed a multi-phase community engagement process to identify core technical themes and facilitate collaboration:
- Request for Inputs (RFI): Over 160 responses identified core technical themes.
- Team Science Integration: Sessions focused on psychological safety and team charters to facilitate interdisciplinary collaboration.
- Hybrid Deliberation: 250 participants engaged in panels across six sub-themes, including grid integration, climate modeling, and smart building materials.
Key Insights
The Hidden Cost of Embodied Carbon
While operational energy (electricity used during runtime) is a common focus, the workshop highlighted that the carbon cost of manufacturing, shipping, and disposing of hardware—embodied carbon—is equally significant and requires deeper investigation.
The Human Factor
Sustainability is not merely a mathematical optimization problem. Technological adoption depends on accounting for human perception of risk and occupant comfort in "smart" systems. If a climate model doesn't account for how people perceive risk, the technology will not be adopted.
Local vs. Global Modeling
There is a critical need to "downscale" global climate models to provide actionable data for local stakeholders, such as farmers and city planners. Global models are sophisticated but often lack the granularity needed for specific local impacts.
Bio-inspired Hardware and Grid Integration
To combat the e-waste crisis, the report suggests moving beyond silicon toward bio-degradable sensors. Furthermore, data centers should evolve from energy sinks into active grid participants, providing frequency control and reusing waste heat for local agriculture or heating.
Recommendations and Results
The workshop's findings are categorized into three pillars:
- General: Promotion of open data models and standardized ontologies to break down proprietary silos in agriculture and building sectors.
- Sustainable Computing: Incentivizing hardware modularity for easier repair and establishing sustainable computing as a distinct academic discipline.
- Computing for Sustainability: Developing "digital twins" for infrastructure and "FarmGPT"-style AI assistants for precision agriculture to guide water and fertilizer use.
Limitations and Future Outlook
The report identifies a significant gap between industry and academia, particularly in large-scale renewable energy deployment. Future work must address funding silos that hinder cross-layer projects and ensure the long-term maintenance of data repositories essential for climate science. Academic researchers need better access to real-world industrial data to remain relevant in this rapidly evolving field.
Summary updated