Researchers at the University of Washington have developed a multiagent AI system capable of calculating the carbon footprint of digital devices in under one minute, aiming to make environmental impact data as accessible as a nutrition label. The tool, described in a study published June 12, uses two AI agents that analyze text and image data from public sources, including product descriptions, manufacturer guides and existing life cycle assessment databases, to estimate emissions associated with a device’s full life cycle, from raw material extraction to disposal.
Zhihan Zhang, a Ph.D. student involved in the research, said the system was designed to simplify a traditionally time-consuming manual process requiring data from multiple manufacturers and suppliers. According to the university, the tool’s error rate ranges from 5% to 19%, comparable to human-conducted assessments, and performs significantly better than humans when evaluating materials not included in existing databases, averaging a 23% error rate compared to 143% for manual calculations.
Vikram Iyer, an assistant professor who supervised the project, noted that each device’s carbon footprint only needs to be calculated once, minimizing the AI system’s own energy use. The research team has released the tool’s browser extension code as open source, though it is not yet in wide use.
