Harnessing artificial intelligence to develop mission-focused solutions

Artificial intelligence (AI) is fueling innovations that affect every facet of our lives, transforming how we live, work, and solve problems. At a March LLNL event called aiEDGE for Innovation Day involving keynote talks, tool demonstrations, and workshop sessions, nearly 3,200 employees explored how AI could accelerate everything from scholarly research to experimental design to project management to IT support, all in the service of advancing national security. This collaborative event underscored the transformative nature of AI technology.

AI is far from new to our researchers, but the pace of AI-enabled breakthroughs at our Lab is accelerating. What’s making this possible?

A major factor in our ability to achieve these breakthroughs is our access to some of the world’s most powerful supercomputers, including the fastest exascale system currently in operation. These computational platforms allow us to perform complex simulations and rapidly process massive datasets. However, access to these computational tools is just one factor in our success. The true engine behind this progress is LLNL’s team science approach. In our highly collaborative research environment, domain scientists and AI experts work side-by-side to solve national security challenges.

One area where this multidisciplinary research approach is making a significant impact is the design of high explosives and energetic materials, which are essential to our mission. Researchers start with the desired outcome in mind, and then use AI and machine learning to rapidly explore design pathways to achieve this aim. In addition, the Lab’s sophisticated additive manufacturing capabilities make it possible to produce these tailored designs for materials and devices used in extreme conditions.

In another example, LLNL is using AI to help rapidly respond to biosecurity threats, including efforts to identify promising new vaccines and therapeutics to combat pathogens. With support from the U.S. Department of Defense, and in collaboration with other U.S.-based research institutions, LLNL teams use AI-assisted supercomputing to substantially narrow the list of possible antibody candidates and then move forward with lab testing—reducing the time needed to develop effective medical countermeasures from years to weeks. These efforts are supported by LLNL’s bioscience teams, computational scientists, and specialized LLNL facilities, like the Center for Predictive Bioresilience and the Rapid Response Lab.

We’re also pioneering AI-enabled nuclear science research with national security applications, such as nuclear reactor operations, nuclear forensics, nonproliferation, and stockpile science. A cross-directorate research team is using a “de-multiplier” approach to develop an AI “supermodel” that learns from existing theoretical and experimental data to study nuclear reactions and make accurate predictions where data does not exist. The model can also estimate the uncertainties of these predictions, which is essential for a research area where the amount of available experimental data is relatively small. For instance, these models can provide realistic uncertainty estimates for prototype small modular reactors and fusion systems, helping reduce design costs.

As we continue exploring ways to leverage LLNL’s unique research environment and powerful computing tools, we’ve already seen how the next generation of Lab scientists is eager to pursue AI-enabled solutions. With mentorship from senior scientists and access to a collaborative AI research community, they’re working on exciting new projects. Examples include:

• Developing advanced models of magnetic fusion devices, enabling scientists to study fusion boundary plasma physics, including the search for alternative solutions for plasma exhaust management, which plays a key role in harnessing the power of fusion energy.
• Developing simulations that can predict the properties of atomic nuclei, including novel quantum computing algorithms that can be used to simulate nuclear reactions.
• Developing new AI-driven diffusion models that can predict 3D atomic structures of highly complex material features. The models can analyze existing structures using spectroscopic data, and help scientists engineer materials with tailored atomic structures.
• Building autonomous labs powered by AI to predict, screen, and tailor polymer and composite materials for diverse applications, including efforts to produce safe, stable batteries and improve the lifetime of membranes for water electrolysis.

Throughout our history, our Lab has been home to world-class innovators. With sophisticated AI tools now part of their toolkit, I look forward to seeing how their future discoveries will continue advancing national security.

– Glenn Fox, PLS Principal Associate Director