Image
Q&A: Stephen Baek on AI, Extreme Physics, and the Future of Innovation
When most of us think of physics, we picture falling apples or planets in orbit. But for Stephen Baek, associate professor of data science at the University of Virginia, some of the most fascinating and urgent questions lie in what he calls extreme physics—moments when physical forces push the limits of speed, pressure, and possibility.
From rocket launches to baseball pitches, from airbags to advanced materials, Baek and his colleagues are applying artificial intelligence to understand, predict, and even design systems where failure is not an option. We sat down with Baek to talk about what extreme physics means, how AI is changing the game, and why this work matters for everything from national defense to everyday life.
Q: To start, what exactly is “extreme physics”?
Baek: We live in a physical world, so physics is everywhere. But extreme physics focuses on phenomena involving unusually large values—large forces, velocities, accelerations, or abrupt changes.
Think of a baseball pitcher like Shohei Ohtani. The way he throws pushes human biomechanics to an extreme—arm rotation, speed, energy. That’s extreme physics in action.
Explosives are another example. We usually think of them in a military context, but they’re everywhere in daily life too. They are, for example, crucial for building tunnels and mining natural resources. Also, a fun fact: if you drive a car, you have at least two “mini bombs” inside—your airbags. Those explosives must be engineered so they only trigger when needed, under wildly changing conditions of heat, cold, or moisture. So, knowing what triggers their extreme reaction is of critical importance.
Or consider space exploration: propellants to fire rockets, or hypersonic combustion engines to let us fly 5 times or more faster than the speed of sound. Extreme physics underpins all of it. What fascinates me is that these events aren’t just interesting from a physics point of view—they’re also fascinating data science problems as many of these extreme values are statistical outliers. They’re rare, hard to measure, and difficult to model. That makes them scientifically and computationally challenging, which is where data science and AI come in.
Q: Why do these extremes pose such a challenge for data science?
Baek: Most machine learning models are designed to follow the average—to capture general trends. For example, predicting tomorrow’s weather: if you say “it will be sunny and nothing will happen,” you’ll be right 90 percent of the time. But that prediction is useless, despite having a whopping 90% accuracy, because it misses the rare tornado or hurricane, which are more important to predict.
The problem with extreme physics is twofold: the events are rare, and the data are scarce. We have large, outlier values but only a small number of examples. That makes prediction extremely difficult. Yet these are precisely the events—extreme weather, explosions, structural failures—that matter most.
Q: How do you define artificial intelligence, and how does it apply here?
Baek: AI is a broad term and sometimes misused. At its core, it’s about building machines capable of prediction, decision-making, and problem-solving in ways that resemble human thinking. Machine learning is a subfield of AI—it designs algorithms that let computers learn patterns from data.
In the context of extreme physics, AI can help us study events that are too fast, too dangerous, or too costly to observe directly. For example, predicting the risk of injury for an athlete. As a coach, you’d want to know: if a pitcher lowers their arm angle by just one degree, what happens to ball speed? To shoulder injury risk? You can’t run those experiments ethically on real players.
Similarly, with explosives, you can’t just test them repeatedly—it’s dangerous and expensive. Traditional physics simulations are possible, but they require solving very complex equations and can take days on a supercomputer. That makes design cycles painfully slow.
AI changes that. By training algorithms on existing data, we can build models that predict these extreme events in seconds—even on a standard GPU-enabled laptop—where before it took days of simulation.
Q: That sounds like a huge leap. How do you ensure AI predictions still respect the laws of physics?
Baek: Great question. Neural networks, by themselves, only look for statistical correlations between inputs and outputs. But physics is governed by very specific equations—like Newton’s famous F = ma.
So what we do is embed those governing equations into the AI models. We essentially add “guardrails” for prediction. That way, even as the algorithm learns from data, it cannot violate the fundamental laws of physics. This approach, sometimes called physics-informed machine learning (PIML), gives us the best of both worlds: the speed and adaptability of AI with the reliability of physical laws.
Q: Will AI eventually help us design materials that don’t even exist yet?
Baek: It’s already happening. About a decade ago, the White House launched the Materials Genome Initiative to accelerate the discovery of new materials. My team participates in this effort, which involves building massive databases of material properties—what happens when you combine copper with aluminum, or test new alloys under different conditions.
As these databases grow, AI gets smarter at predicting outcomes. Pharmaceutical companies already use AI to simulate what happens when molecules combine, which accelerates drug discovery. Materials companies like 3M or DuPont are doing the same for composites, polymers, and metals.
What used to take decades of trial-and-error experimentation is now happening in years—or less. AI is rapidly shortening the runway from scientific idea to commercial product.
Q: Looking ahead, what kinds of breakthroughs do you expect in the next decade?
Baek: I think long-standing problems in material science and mechanics will be solved in surprisingly short timeframes. The pace is accelerating because so many research groups worldwide are pushing in the same direction, and the outcomes are already visible.
One example I came across recently: a golf club designed with AI. At first, it might sound trivial, but the science is serious. Club manufacturers used to rely on physics simulations—expensive, slow computations—to optimize distance, forgiveness, and material properties. Now they’re running AI models that simulate thousands of design combinations at lightning speed.
The result is more sophisticated products and faster innovation cycles. That’s just one sector. Similar advances are happening in aerospace, energy, medicine, and sports science. Everywhere you look, AI is reshaping design and discovery.
Q: Bringing it back to everyday life, why should the public care about extreme physics and AI?
Baek: Because it touches all of us. Airbags in cars, medical treatments, athletic performance, safer buildings, faster computers, sustainable energy systems—the list goes on. Extreme physics underlies how we survive in dangerous moments and how we thrive in competitive ones.
AI is the key that lets us study these extremes safely, affordably, and quickly. By merging physics with data science, we’re not just crunching numbers—we’re expanding what’s possible.
And the most exciting part? This is just the beginning. Ten years from now, I think we’ll look back and realize how much of the world around us was shaped by AI-assisted extreme physics. It’s happening now, and it’s happening fast.
Closing Thoughts
Extreme physics may sound like something reserved for rocket scientists or elite athletes, but as Stephen Baek makes clear, it’s all around us. From the safety of our cars to the future of our materials, the fusion of AI and physics is opening doors to discoveries once thought impossible.
For Baek, that’s what makes this work so rewarding: not just understanding the extremes, but using them to imagine—and build—a future that’s safer, faster, and more resilient.
Read the corresponding article in UVA Today.
Interview conducted Sept. 2025. Edited for clarity and length.
