AI could help scientists decode nuclear events faster, saving time when every second counts

]]>

Nic Uhnak To give you a simple definition of what nuclear forensics is: It is an investigation to characterize nuclear materials. That is either an interdicted material or device, or an examination of post-detonation debris, with the primary aim of assisting law enforcement and the U.S. government to prevent, mitigate, or attribute that material.

Terry Gerton So we don’t hear a lot about nuclear explosions and I think that’s a good thing, but how often does it happen and why do you need to move kind of quickly with it?

Nic Uhnak Well, it is a high-consequence event. It doesn’t happen that often, thank goodness. This is an interesting job I have. It is one that I genuinely don’t necessarily want to be applicable at all times, but we need to go as quickly as possible so that the decision-makers can make decisions quickly.

Terry Gerton And so this has been kind of a process that takes a lot of time and a lot of computing power, right?

Nic Uhnak In the past, I would say that it’s taken almost no computing power at the end of the day. Generally speaking, a nuclear forensics investigation requires a tremendous amount of tedious laboratory work. So I primarily exist in the post-detonation world, where we look at that complex nuclear debris after an explosion. I’m going to set the table if you don’t mind real quick. After this explosion has happens, you end up with a highly complex matrix of material that contains a bulk of the periodic table and is highly radioactive. The difficulty in that is that many of those radioactive elements that comprise that material provide a piece of the puzzle that allows us to understand that event better. And isolating and combining those individual pieces to build the puzzle requires a lot of effort and a lot chemistry.

Terry Gerton You’ve used the analogy of trying to figure out the ingredients of a cake after it’s baked. So maybe walk us through that analogy compared to what you just described.

Nic Uhnak This nuclear debris is like a baked cake. It’s got frosting, it’s got flour, eggs, sugar, all of those ingredients combined. And I don’t necessarily know where each of those ingredients came from, or how they were mixed, or the proportions that they were mix. But through these tedious investigations, we can start to take those pieces apart and say, hey, those eggs were an organic egg from a specific place, and this flour was cake flour versus bread flour. And that information is passed on to the decision-makers that actually decide.

]]>

Terry Gerton So if I could then extrapolate it to a real-world example and you could tell me whether this is right or wrong: If we had been totally successful in our attack on Iran’s nuclear facilities and there might’ve been an explosion, would you have gone in then to look at the debris from that and figure out where they got the bits and pieces from?

Nic Uhnak Not necessarily. That one is an interesting one. We theoretically could have received material, but that sort of detonation is a little different than a nuclear detonation. So yes, we could tell you what it was made of. We could probably tell you all sorts of information that would allow decision-makers to make better decisions with regards to how they respond to Iran. But outside of that, no ma’am.

Terry Gerton I’m speaking with Dr. Nic Uhnak. He’s a radio chemist at the National Security Directorate at Pacific Northwest National Laboratory. All right, well then we’ll go back to your example with the cake.

Nic Uhnak The thing that I want to kind of reiterate is that it does take time, and that time is of the essence. So in this case, our primary aim is to decrease the time from start to finish, and this where our research has kind of come into play.

Terry Gerton And so now you’re using artificial intelligence to speed up that determination process. Tell us how you’re doing that.

Nic Uhnak Our general approach here is to take a few baby steps, and in this case, this is a baby step. We’re combining generative AI and computational chemistry experience, as well as nuclear material composition understanding and expertise, to start at the very simplest scale of interaction. So interaction between a molecule and an ion. And these things actually matter. These are very specific to the nuclear forensics investigation. So we’re looking at these fundamental interactions and we’re building up complexity. The ultimate goal is to use these interactions between these molecules, ions, etc., at different scales to provide optimized and targeted chemistry for our actual chemical separations. We’re streamlining the process so that we can get at those individual pieces of the cake, or the puzzle, more rapidly.

Terry Gerton And what are the practical benefits of solving this problem more quickly? Does it help first responders? Does it help investigators? What do people do with the information you provide them?

Nic Uhnak So what PNNL does is that we actually provide the data. There’s a whole separate group of people that end up putting that puzzle together after we’ve provided them the pieces. Those individuals within the U.S. government or some other organization, they end up talking to the decision-makers. And the faster we provide that information, the faster we can make those decisions. And the better we provide that information, the higher fidelity, that improves the decision that we can make.

Terry Gerton So you mentioned that this first project using the AI is a baby step. Once this baby starts to walk, where does it go? Does this mean that you can get the testing capacity out into the field as opposed to keeping it in a lab?

]]>

Nic Uhnak Generally speaking, the laboratory setting, or what we call a fixed lab, is the area that we are going to say is the most important. That is where you’re going have your best analytical tools. You’re going to be able to do the chemical separations as best as possible, again, to get the highest fidelity information as quickly as possible. The aim is that we’ll build a tool that allows us to do predictive chemical separations to make that incredibly fast, incredibly targeted, so that we can get the materials, those radioactive elements, that are the highest impact faster.

Terry Gerton And are there other areas where you think that this analytical tool or process could be applied?

Nic Uhnak Absolutely. The overall approach is not unique to nuclear forensics. Nuclear forensics just happens to be like the test case, or it just happens to my field of expertise. But generally speaking, this is widely applicable to chemical separations, and it has a very easy hop to medical isotope production.

Terry Gerton So what’s your next step as you build out the research capability?

Nic Uhnak Our next step is to continue with these baby steps. We’re currently building more tools within our larger tool that allows us to do better interactions between larger bulk molecules, with the aim of continuing to build up until we have a working generative AI tool that can take input that we can gather from, say, from the elemental or radiation signatures from that sample.