DOE Prepares Scientific Challenges for Genesis Mission

Update as of 2:18 pm Feb. 12: DOE has published its list of 26 Science and Technology Challenges.

The Department of Energy will soon publish a list of scientific “lighthouse challenges” that its national labs will seek to solve using AI as part of the Genesis Mission.

Under Secretary for Science Darío Gil, who is leading Genesis, elaborated on the agency’s vision for the mission during a visit to Brookhaven National Lab last week. The initiative, launched via executive order in November, aims to “double the productivity and impact of American science and engineering within a decade” using AI.

“Ultimately, we’re doing this not because we’re AI lovers,” Gil said. “It begins and ends with, we want to do better science. We want to help our scientists ask even better questions and solve better scientific challenges, right? AI is a tool, and computing is a tool in that context,” he added.

The executive order directed DOE to identify, by late January, at least 20 science and technology challenges in the areas of advanced manufacturing, biotechnology, critical materials, nuclear fission and fusion energy, quantum information science, and semiconductors and microelectronics that the mission may be able to address.

The Genesis Mission also aims to build the American Science and Security Platform, an “integrated AI platform to harness federal scientific datasets” that will be used to “train scientific foundation models and create AI agents to test new hypotheses, automate research workflows, and accelerate scientific breakthroughs.” This includes robotic automation and improvements in user facility operations, Gil said. The department announced $320 million for projects to begin building the platform in December.

The platform “will have a direct impact over time with every scientist and engineer and member of the technical staff that participate in our endeavors,” Gil said.

Not every scientific problem may be well-suited to be addressed using AI right now, Gil added. Some may lack appropriate data sets, and others may require more advanced AI methodologies than currently exist.

A well-suited problem likely has “good data scaling laws,” Gil said, meaning the performance of a model addressing that problem will improve as it is trained on more data and as the size of the model increases. In those cases, Gil said, the team working on such a model should continue scaling it up and be given access to “a lot more computation, a lot more access to data, a lot more expertise.”

Importance of collaboration

On Monday, DOE announced the creation of a consortium to serve as a single access point for Genesis-related information, events, working groups, and potential partnerships. Gil emphasized the importance of collaboration within each national lab and with the private sector during the town hall.

Gil pointed to DOE’s Exascale Computing Project, which produced the first U.S. exascale computer in 2022, as an example of a program that, though successful, was overly centered on the agency’s computing office, with less focus on how the exascale capabilities would apply to other fields, such as physics, chemistry, and biology.

The current computational capacity at the labs is insufficient for the Genesis Mission, and DOE needs to “work with all the sectors” to scale it up, Gil said. “The gap in the appetite of how much of you, the communities of the national laboratories and the university ecosystem, could consume, versus what we have, is massive,” he said.

DOE is “driving interesting, new, novel partnership models” with the private sector to bring greater computational capacity and “record speeds” into the national labs, he added. The agency announced initial agreements in December with 24 private companies interested in collaborating on the mission.

Within the lab

DOE must identify all federal computing, storage, and networking resources available to support the Genesis Mission, as well as potential resources from industry partners, by late February, according to the executive order.

All 17 national labs are involved in the Genesis Mission, Gil said. For instance, Brookhaven is currently using data from its particle accelerators to build a prototype of the full American Science and Security Platform that will be ready in the next few months, said John Hill, the lab’s interim director. The prototype will test “the fundamentals of how to handle these very complicated experiments,” Hill said.

Researchers at Brookhaven presented their Foundation Model for Nuclear and Particle Physics to Gil, who said their results matched the scaling laws the department is looking for in promising models. The study used 300 petabytes of data, researcher Yeonju Go said, which came from the Relativistic Heavy Ion Collider (RHIC), one of the lab’s flagship scientific instruments. RHIC ended its last run last week.

Regarding Genesis, RHIC scientist Alex Jentsch said, “Brookhaven is definitely all in on it, because it’s very clearly the way things are going to move in the future.”

Jentsch added that the initiative has supported researchers at Brookhaven in exploring non-obvious applications of AI to their work.

“You’re always trying to find the best use of your time to maximize your scientific output and your career potential long term,” Jentsch said. “Sometimes pursuing a new avenue can be a risk, and it’s one of the reasons why we kind of get stuck doing things a certain way for a long time,” he added.