Bringing artificial intelligence to bear on issues relating to nuclear weapons might sound like the stuff of a scary sci-fi movie — but at the Department of Energy’s Pacific Northwest National Laboratory, it’s just one of the items on the to-do list.
One of PNNL’s research priorities is to identify and combat complex threats to national security, and AI can help meet that priority by detecting attempts to acquire nuclear weapons or associated technology.
Nuclear proliferation detection is one of the potential applications that could get an assist from the Center for AI @PNNL, a newly announced effort to coordinate research that makes use of AI tools — including the generative AI tools that have captured the attention of the tech world over the past year or two.
“For decades we’ve been doing artificial intelligence,” center director Court Corley, PNNL’s chief scientist for AI, told GeekWire in a recent interview. “What we’re seeing now, though, is an exceptional phase shift in where AI is being used and how it’s being used.”
Historically, PNNL’s researchers have made use of data analytics in specific domains to provide deeper insights into special types of challenges, “But in the past 10 years since we started using modern AI, which includes deep learning and now generative AI and deep reinforcement learning, the number of our scientists and engineers that are using it in their day-to-day work has just continued to grow,” Corley said.
Corley said one-fifth of PNNL’s thousands of scientists and engineers are using AI. “These are power systems engineers working on electricity grid resilience, and scientists working to decode the biomolecular universe and understand biological function, and so many domains in between — where previously, the application of AI was challenged because either it wasn’t robust or it wasn’t resilient, but now it’s really beginning to work well,” he said.
The Center for AI @PNNL is meant to optimize how AI tools are employed at the national laboratory, which is based in Richland, Wash., but also has researchers working in Seattle and Sequim in Washington state, plus Portland, Ore., and even College Park, Md. Corley and his staff have the job of coordinating AI projects and championing technological innovations across the organization.
Among the AI-centric projects PNNL’s researchers have been working on are initiatives to speed up materials development for energy storage systems; to engineer microbes to produce next-generation biofuels; and to make use of a technique called autonomous experimentation to create new types of alloys.
“There are pockets of AI research and AI engineering all across the laboratory, and we’re at a point now to be able to aggregate our collective expertise and really champion our work,” he said. “It makes sense to bring everybody together in a virtual research hub. That way, we can have an entity that’s greater than the sum of its parts. We can better prioritize infrastructure, prioritize resources, prioritize tooling, upskilling, training and so forth.”
Corley is well-suited for his new role as the center’s director: He received his Ph.D. in computer science and engineering from the University of North Texas in 2009 and has been focusing on data analytics and AI research at PNNL ever since. Here’s a selections of quotes from our interview, edited for brevity and clarity.
On efforts to ensure the safe and secure use of AI for national security applications: “There’s a huge call-out in the executive order about AI that was issued by the president recently. It delves into how we as a nation can protect against revealing dangerous information — for example, recognizing the ability to extract dangerous information from these systems as to how they’re used, and building a response to that. Right now, the government is on a very short timetable to respond to those concerns. It’s a whole-of-government effort with many organizations involved, not just national labs like PNNL. We need to answer those specific questions: What are the safeguards? What are the mechanisms by which we’ll protect our infrastructure and high-consequence systems?”
On working with tech partners to make AI safer: “Tech companies are going to develop their own roadmap. But the folks from the tech companies are very clear. They say, ‘We shouldn’t be the one only ones developing safe and secure methods for these systems. We really need the government support.’ And you see that in Washington now. They’re having conversations with many organizations about partnerships to make sure these systems are developed in a safe, secure manner.”
On how generative AI being used at PNNL: “We’re in an exploratory phase right now. We’re trying to see where generative AI will have the most impact in the missions that we support. We have a new effort, using some of the tools provided by our hyperscalers, like Microsoft or Amazon, to see whether those tools are sufficient to do both computational chemistry and AI workloads. We have other efforts that we are just starting with the Department of Energy — for example, to accelerate permitting processes. The process for getting environmental permits for different types of electricity use is a very paper-heavy process. And so we have an exploratory project right now to see if can we accelerate that process, to remove some of the barriers to accessing government resources.”
On using AI for research in Earth science and biological sciences: “There are two applications there. One is speeding up the complexity of our climate models and our atmospheric models to get a better understanding of how climate change is affecting the United States. In the biological sciences, we’re thinking about how we can understand biological function in a way that we’ve never been able to before, by having an ‘-omics’ view that goes beyond genomics — by looking at the proteomics, metabolomics and all these other -omics, and integrating them in a way that hasn’t been possible before because of the sheer complexity. This is another area where we’re trying to bring AI into the fold.”
On areas of research where AI and quantum computing can work together: “One of the areas is in materials. If we can design better materials that can operate in the regimes that are needed for quantum computing, I think that’ll be game-changing. And we’re seeing that machine learning can be used in quantum chemistry. So, I would say the No. 1 area where AI is probably going to have an impact is in that materials chemistry space — which is where quantum computing is, because of the temperatures at which it needs to run and the types of materials it needs for its design.”
On the ways in which PNNL is improving how AI is done: “One example that we have worked on is in an area called scientific machine learning. It’s where we incorporate physics into deep learning itself, so you can constrain it or provide it with explainability and interpretability by saying, ‘Here are known governing laws about the physical universe.’ By constraining our models based on that, we add a layer of reducing the cost of training and computation time and data. We’re also increasing how interpretable AI is for the end users, because now it’s not an unbounded system, but it’s bounded by the laws of physics.”
On whether autonomous experimentation will ever replace human researchers: “The only thing that AI is going to do is accelerate the process. We still absolutely have to have human ingenuity in the loop for a lot of the designs and experiments. To have a deep theoretical understanding of the domain in addition to the ability to partner or use a co-pilot of some sort, I think is really going to be the game-changer for this autonomous experimentation.
“The only ones who will be left behind are those who don’t adopt the use of AI in some way. We’re still exploring where it should be used. But there are many applications that go beyond automation, where you’re just getting a repetitive process that’s done, maybe not in an intelligent way, and then you add some intelligence on top of it that says, ‘OK, now you could perform this next experiment based on the hypothesis that you just tested.’”
On the AI applications from PNNL that could be the first to change daily life: “In the next five to 10 years, we’re going to see a complete transformation in energy generation because of the ability to bring on new types of resources. The only way to do that in a reliable manner is through the use of AI, because now we’re talking about generation sources that are variable. They’re not consistent. They’re not dispatched easily. There’s a lot of operational control that needs to happen. AI is probably one of the solutions that will be used to help us get there.”
