$2M USAF Award Aims to Bridge Gap Between Brain, AI

For people, learning can be almost instantaneous: Touch a hot stove and you know not to do it again. But that’s not the case even for advanced artificial intelligence systems, which don’t yet possess the “one-and-done” capability the human brain has.

“Typically, they require enormous amounts of data and computing power to learn new tasks through numerous rounds of trial and error,” said physics Professor Wolfgang Losert, who’s leading a project supported by a new $2 million grant from the U.S. Air Force Office of Scientific Research that aims to understand and recreate the brain’s unique capacity for learning and adapting quickly.

The team—including cell biology and molecular genetics Professor Kan Cao, chemistry and biochemistry Professor John Fourkas, and electrical and computer engineering Associate Professor Cheng Gong as well as other academic and industry partners—will build a test bed to study the functioning of neural networks and develop new approaches to quantum computing and sensing inspired by the living brain.

While traditional computers process information through individual components working in sequence, the brain distributes information across many networks of cells working in parallel. This fundamentally different approach allows for faster learning and adaptivity but consumes far less energy than a computer.

A key focus is on astrocytes, a type of cell that makes up more than half of the cells in the human brain. Long considered mere support cells for neurons, astrocytes are now recognized as crucial to how the brain crunches data. By engineering laboratory-based systems that incorporate both neurons and astrocytes, Losert’s team will closely observe how the two types of cells form living neural networks sense and react to various types of stress like ultrasound or electrical fields.

Recent discoveries by the neuroscientist on Losert’s team, Assistant Research Scientist Kate O’Neill, and other researchers have already shown that astrocytes participate in brain signaling and may be essential to the brain’s ability to both learn and adapt to new situations quickly. Further observations could provide insights into how the brain maintains its performance under different conditions and may lead to more resilient forms of artificial intelligence (AI).

The variety of signal types in biological neural networks—electromagnetic, chemical and mechanical—opens up another exciting aspect of the team’s work.

“We can use living neural networks to test and improve quantum sensors for a range of biomedical applications,” said Losert, who is also an MPower Professor and interim associate dean for research in the College of Computer, Mathematical, and Natural Sciences with a joint appointment in the Institute for Physical Science and Technology.

Quantum sensors have the potential to measure minute physical changes like the presence of magnetic fields or electrochemical activity in cells in minimally invasive ways. Novel non-invasive biosensors could allow scientists and health care professionals to observe brain processes in patients that they couldn’t see before, potentially leading to better medical treatments and a more nuanced perspective on brain performance.

“By understanding and replicating how brain cells work together, we hope to create more efficient and adaptable computing systems,” Losert said. “This project represents the start of a new paradigm in biocomputing that may help shape the future of both AI and neuroscience.”