The next generation of scientific discovery will not come from artificial intelligence (AI) replacing scientists. It will come from AI helping them ask better questions, solve more complex problems and reach reliable answers faster.
researchers will advance that goal through two projects selected for the U.S. Department of Energy’s inaugural Genesis Mission, an effort to create what the department calls “the world’s most powerful integrated science discovery platform.”
Working with the Thomas Jefferson National Accelerator Facility, known as the Jefferson Lab, Pacific Northwest National Laboratory and other national laboratory partners, researchers will develop AI technologies addressing two fundamental challenges: helping scientists operate complex research facilities more efficiently and ensuring they can trust the AI-supported results those facilities produce.
One project will create a virtual model scientists can use to test equipment adjustments before applying them in the physical world. The other will develop safeguards to detect cyberattacks, manipulated data and unreliable AI results before they compromise scientific research.
“This recognition reflects ’s growing role in addressing some of the nation’s most ambitious scientific challenges,” said Kenneth J. Fridley, Ph.D., vice president for research and economic development. “Our researchers are helping shape the future of AI for scientific discovery, while strengthening Virginia’s position as a leader in research, innovation and economic growth.”
Giving Scientists More Time for Discovery
One project, led by Monika Yadav, Ph.D., assistant professor of physics and faculty member in the School of Data Science, will explore how AI can help scientists operate particle accelerators more efficiently.
Before an experiment can begin, experts must carefully tune a particle injector by adjusting thousands of interconnected settings. A change to one setting can affect many others, making the process complex and time-consuming.
Dr. Yadav’s team will develop an AI-enabled digital twin, a virtual version of the injector that behaves like the physical equipment. Much like a simulator allows someone to test a decision before applying it in the real world, the digital twin would allow accelerator operators to see how the injector is likely to respond before changing the actual equipment.
The system will combine physics-based simulations, operational data and expert knowledge to help scientists identify promising adjustments more quickly. If successful, the technology could reduce the time spent configuring equipment, improve accelerator performance and give researchers more time to conduct experiments.
During the first phase, the team will develop and demonstrate the technology and evaluate whether it can accurately predict system behavior, help guide experiments and accelerate the research process.
“When we use a digital twin to test injector settings before changing the physical equipment, we can identify promising adjustments without repeatedly tuning the accelerator itself,” Dr. Yadav said. “Better tuning can produce a more stable, precisely controlled particle beam, giving scientists higher-quality experimental conditions and more time to collect the data they need to understand matter at its most fundamental levels.”
Protecting the Reliability of Scientific Results
As scientists rely more heavily on AI, they also need ways to recognize when a system, dataset or result has been compromised.
A second -led project will develop safeguards designed to detect cyberattacks, data manipulation and other problems that could undermine AI-assisted scientific research.
The nine-month feasibility study will be led by Jiang Li, Ph.D., professor in the Department of Electrical and Computer Engineering in the Batten College of Engineering and Technology, in collaboration with Jefferson Lab.
Conventional cybersecurity tools can identify familiar signs of malicious activity, but scientific systems present an additional challenge: Researchers must determine whether an unexpected result represents a genuine discovery, an equipment problem or deliberate interference.
Dr. Li’s team will combine AI with the laws of physics to help make that distinction. If manipulated data, a compromised AI model or another threat produces a result that is physically inconsistent, the system would be designed to detect the discrepancy, identify the potential source and help researchers recover reliable results.
The goal is to build security into AI-enabled research from the beginning, protecting not only the technology but also scientists’ confidence in the findings it produces.
“When an AI-generated result conflicts with the underlying laws of physics, that can be a sign that something has gone wrong in the data, the model or the system itself,” Dr. Li said. “Our physics-aware guardrails are designed to catch those inconsistencies, help researchers understand what happened and recover a reliable result so scientists can use AI with greater confidence in the discoveries that follow.”
Advancing AI That Is Both Capable and Trustworthy
Together, the projects address two sides of the same challenge. Dr. Yadav’s team will study how AI can help scientists use complex facilities more efficiently, while Dr. Li’s team will focus on protecting the reliability of those AI-supported operations.
The work brings together researchers from the University’s School of Data Science; the College of Sciences’ Center for Accelerator Science, Department of Physics and Department of Computer Science; and the Batten College of Engineering and Technology’s Department of Electrical and Computer Engineering.
“This is an exciting opportunity to see how bringing experts from across disciplines come together to explore the complex questions of today,” said Dr. Khan Iftekharuddin, dean of the Interdisciplinary Schools, professor of electrical and computer engineering and Batten Endowed Chair in Machine Learning. “By pooling knowledge and experience from different fields, we can advance the interdisciplinary fields of AI in scientific discovery and make way for innovation.”
is also contributing to a third Genesis Mission project led by the University of Hawaiʻi at Mānoa in partnership with Argonne National Laboratory.
Rui Ning, Ph.D., assistant professor in the School of Data Science, serves as the University’s project lead for Security and Trust Runtime Architecture for Time-Critical Operational Science, or STRATOS. The project will develop technology capable of detecting and responding to cyberattacks against AI-enabled scientific systems in real time.
Dr. Ning will lead the University’s development of efficient detection methods and contribute to approaches for stopping or limiting attacks as they occur.
Through these projects, University researchers will help determine how national scientific facilities can use AI to conduct experiments more efficiently without sacrificing the security and reliability on which credible discovery depends.