Researchers at the University of Washington have built an AI system that can estimate the carbon footprint of electronic devices in minutes instead of weeks. The system uses two AI agents that work together: one acts as an analyst to define what data is needed, and the other acts as an engineer to pull component-level data from public sources like FCC databases and iFixit teardown archives. The results have an error margin of 5% to 19%, which is comparable to traditional expert analysis. The system handles missing data by using a nearest neighbors technique. It groups similar products by specs like screen size and processor type, then computes weighted averages to estimate carbon footprints for devices with incomplete data. This cut average estimation errors from 143% down to 23%. The approach could help electronics manufacturers get faster sustainability assessments during product design, especially for new materials where LCA data does not exist yet.
