Energy and emissions tracking for dynamic ML systems

Develop methods to accurately track energy consumption and carbon emissions across numerous dynamic training and hosting instances of AI systems to enable environmental impact assessments and policy interventions.

Background

Environmental governance of AI requires credible accounting of energy use and emissions across the lifecycle, but real deployments are distributed and fast-changing.

Solving tracking logistics at scale would enable evidence-based policies, ratings, and incentives for more sustainable AI development and deployment.