An important aspect of the U.S. Department of Energy’s (DOE) mission is to ensure the nation’s energy independence and security both in the short and long term. Key to meeting this challenge are continued advancements in artificial intelligence (AI), especially in the context of applied energy. As an initial step toward addressing these challenges, a group of about 100 experts on AI/machine learning and applied energy convened at Argonne National Laboratory in December 2023 to map out future needs related to utilizing AI. The goal was to detail pressing technical challenges and propose AI-assisted solutions.

DOE is ideally positioned to address challenges associated with energy independence and security due to its unique set of assets, including a highly skilled workforce with relevant domain expertise and an array of world-leading experimental facilities for making advances in materials, chemistry, and more. By integrating these resources with other AI capabilities outlined in the previous AI for Science, Energy, and Security report, the DOE can leverage AI to stay at the forefront of the rapidly evolving landscape.

The applied energy focus described in this report centers on five areas vital to the energy future of the U.S., as well as underscores the critical role that AI can play in shaping our world—highlighting the urgency and importance of being leaders in AI to ensure impactful solutions to global energy needs. These areas include Nuclear Power, Power Grid, Carbon Management, Energy Storage, and Energy Materials. It will be essential to integrate these together and with other efforts in AI for science and technology. Complexity, the large-scale effort involved, real-time decision making required, robustness of systems, and safety implications all pose extra challenges. The Grand Challenges described in this report span multiple disciplines and have not been solved by conventional methods. The power of AI for solving such problems lies in its capacity to simultaneously handle multiple system characteristics while incorporating both data and specific domain models, and to do so on a scale and at a complexity otherwise not possible.