Research & Discovery

AI research at UVA tackles life’s most extreme moments

When Stephen Baek, associate professor at the University of Virginia’s School of Data Science, talks about extreme physics, he isn’t referring to an obscure corner of science. He’s describing the powerful forces that touch everyday life – from a car’s airbags to rockets launching humans into space.

“These are the moments when physics pushes the limits,” Baek said. “Large forces, high velocities, sudden impacts – phenomena that happen too quickly, too dangerously or too rarely for us to study easily.”

Extreme physics shows up everywhere: in the explosive power of rocket fuel, the whiplash torque of a baseball pitcher’s arm or the delicate reliability of a car’s airbag. The challenge, Baek says, is these events are statistical outliers. They don’t happen often, and when they do, they are difficult to measure, making them notoriously hard to predict with traditional methods.

Stephen Baek, associate professor at the University of Virginia’s School of Data Science, explains that while machine learning finds patterns in big datasets, it often misses rare, high-stakes events. “If I predict tomorrow will be sunny, I’ll be right 99% of the time,” he said. “But I’ll miss the tornado.”

That is where artificial intelligence comes in.

From outliers to insights

In most cases, machine learning excels at finding patterns in large datasets like common weather patterns, typical disease progression and standard voter trends. But averages aren’t useful when the stakes are in the extremes. “If I predict tomorrow will be sunny, I’ll be right 99% of the time,” Baek said. “But I’ll miss the tornado.”

Baek and his colleagues are using AI to compensate for limited data, building algorithms that respect the laws of nature and predict the unpredictable.

“These models let us move past averages and study the rare, high-impact events,” Baek said. “They give us speed, precision and safety.”

Safer cars, faster rockets, stronger athletes

The applications extend from defense systems to daily commutes. Consider the explosive charges that inflate car airbags. They must reliably detonate on demand during a crash, but never in response to heat, cold or moisture. Ensuring that balance requires careful modeling of materials under extreme conditions.

Similar challenges face engineers designing rockets or using controlled blasts. “You want a guarantee that these materials are stable during transport and handling, but responsive when triggered,” Baek said.

Sports science provides other striking examples. Elite athletes like baseball’s Shohei Ohtani or Aaron Judge push human limits. Teams create digital “twins” of players – computer models that mimic an athlete’s movements – to replicate their biomechanics. By simulating subtle changes – an altered arm angle, a different stride – coaches can optimize performance and prevent injuries without compromising the player’s health.

Discovery and Innovation: Peanut allergy immunotherapy reduces parent worry

“In many cases, you simply cannot run the experiment in real life,” Baek noted. “It’s too dangerous, too expensive or too fast to capture. AI simulations give us that window.”

Replacing supercomputers with laptops

Traditionally, scientists turned to computer simulations, which rely on solving complex physics equations. But such models can take days to run on supercomputers, slowing the pace of innovation.

Baek’s research team, with support from government agencies, has shown AI can cut that time dramatically. “We’ve developed algorithms that predict extreme phenomena – like the initiation of energetic materials or the airflow over a hypersonic flight – in seconds on a GPU-enabled laptop,” he said. “Previously, that would have required days on a supercomputer.”

Inventing the materials of tomorrow

Perhaps the most revolutionary application lies in material science.

“AI is already being used to imagine materials that don’t yet exist,” Baek said.

Pharmaceutical companies run AI simulations to combine molecules and test potential treatments. Manufacturers like 3M and DuPont are applying similar techniques to design new composites and polymers.

What once required decades of trial-and-error experimentation can now happen in a fraction of the time. “We’re seeing the discovery cycle for new materials shrink dramatically,” Baek said. “AI is shortening the runway from idea to product.”

A glimpse into the future

Baek believes the next decade will see breakthroughs across multiple industries. He points to a golf club, designed with the help of AI, he recently encountered. While it might seem like a niche example, the science illustrates a broader truth.

“Club makers used to rely on expensive physics simulations to optimize materials and geometry,” he said. “Now they’re running AI models that can test thousands of combinations quickly, leading to more sophisticated designs. Multiply that across aerospace, energy, medicine and sports, and you see the scale of what’s coming.”

For Baek, the lesson is clear: extreme physics is not an abstract concept. It’s the hidden science behind safety, exploration and human performance. With AI as a partner, he believes researchers are entering an era where impossible becomes routine.

“The most exciting part,” Baek said, “is that 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.”