DARPA is addressing the challenge of measuring energy consumption in AI models with its new program, dubbed Mapping Machine Learning to Physics (ML2P). While AI development traditionally focuses on optimizing performance, this initiative seeks to align AI efficiency with the principles of physics by monitoring energy use in joules. Bernard McShea, the ML2P program manager, emphasized the need to account for energy use alongside performance optimization. Targeting the battlefield applications of AI technology, DARPA is interested in systems that manage power consumption effectively due to the battery-operated nature of field equipment. This emphasis ensures that soldiers have reliable AI systems in various operational scenarios. Additionally, the initiative mandates that all participants release their methodologies as open-source resources, aiming for widespread utility and adoption in the research community while setting a new benchmark for AI model power efficiency analysis. McShea foresees potential enhancements in AI hardware by understanding model performance in realistic computational conditions. Proposals for this $5.9 million program, involving various disciplines, are due by December 8.