DARPA awards UB engineers $1 million to ‘teach’ physics to AI systems

Intimidated by physics?

You’re not alone.

Artificial intelligence (AI) systems — the uber-smart machines that many predict will someday outpace human intelligence — struggle with the subject, too.

To address this issue, UB engineers have been awarded a $1 million Defense Advanced Research Projects Agency (DARPA) grant to combine physics-based models with conventional, data-driven AI methods.

The goal is to provide AI systems, which work within specific frameworks and lack tools to explain their reasoning process, with a broader foundation of knowledge through physics. In theory, this will allow for more streamlined, efficient and adaptable AI systems — ideal traits for defense systems, such as unmanned aerial vehicles (UAVs), which operate in uncontrolled environments.

“Unmanned aerial vehicles are trained in collision avoidance. For example, they spot another UAV or a bird and take an action, such as slowing down, to avoid striking that object,” says Rahul Rai, the grant’s principal investigator and associate professor of mechanical and aerospace engineering, School of Engineering and Applied Sciences.

“What we’re proposing would give that UAV an understanding into the physics of things like how birds fly,” he continued. “This information, combined with weather and data that other sensors are processing, will provide the UAV with better collision-avoidance mechanisms.”

To make that possible, Rai and his team will integrate physics-based models — these are math-based formulas that explain the world around us, such as Einstein’s E=MC² — into the algorithms that guide machine learning, deep learning and other data-driven AI systems.

“In a sense, we’re teaching physics to AI systems,” he says.

Because these combined models will provide AI systems with a greater understanding of their surroundings, Rai says it should reduce the amount of data that purely data-driven AI systems require. In turn, that will lead to more efficient and less costly systems, he says.

“The goal is to create hybrid systems that generalize well, which means they’re good at adapting to foreign environments where data may not be readily available,” he explains.

Rai is a member of the UB Artificial Intelligence Institute, which was launched last September. The institute, which focuses on health, medicine and autonomous systems, advances core AI technologies that optimize human-machine partnerships, and provides complementary tools and skills to understand the societal impact of these technologies.