DOE Lab Uses AI to Help Scientists Determine What to Test Next

An AI system that can evaluate experimental results and recommend what scientists should test next is at the center of a new Energy Department project led by SLAC National Accelerator Laboratory. 

The project aims to connect AI reasoning directly to live experiments, allowing researchers to adjust their approach as they learn rather than waiting until an experiment is complete to analyze the results.  

Through the catalysis project, researchers hope to establish a more efficient way to investigate how catalysts work and ultimately help develop materials that are both effective and durable enough for practical energy and chemical production. 

“Catalysis underpins roughly a third of the global economy, from fuels and fertilizers to plastics and medicines,” Principal Investigator Dimosthenis Sokaras, director of the Chemistry and Catalysis Division at SLAC, told GovCIO Media & Research. “Predictive understanding lets us design catalysts and catalytic systems that are active, selective and durable much faster, turning the feedstocks we have on hand into the products we need in ways that are energy-efficient, economically viable and scalable. That also strengthens our ability to make critical chemicals reliably here at home.” 

The Phase I project, titled “Closed-Loop Autonomous Discovery of Mechanistic Activity-Selectivity-Stability Rules in Catalysis,” is part of the DOE’s Genesis Mission, which seeks to apply advanced AI capabilities to scientific and technological challenges. 

The catalysis project builds on SLAC’s existing Integrated Scientific Agentic AI for Catalysis (ISAAC) initiative. ISAAC connects experimental data from multiple DOE facilities with theoretical calculations and published scientific literature. The new effort adds a critical capability: using that knowledge to guide the next experiment while research is underway. 

“Every measurement updates the AI’s understanding, and the AI chooses the most informative next one,” Sokaras said. “The aim is to make this a routine way of running such type of experimental campaigns, reaching mechanistic answers in far fewer measurements, and then to extend it to complementary methods at other national labs, research institutions and universities.” 

At Stanford Synchrotron Radiation Lightsource, a division within SLAC, the team will combine real-time X-ray measurements with electrochemistry, computer models and existing scientific knowledge. The AI system will compare competing explanations for how catalysts work, identify gaps in the evidence and recommend measurements that could provide the most useful information. 

In addition, SLAC will also participate in eight other newly announced Genesis Mission projects led by national laboratories, universities and industry. The collaborations span research areas including enzyme design, scientific computing, quantum materials, microelectronics, accelerator science, fusion energy and RNA biology. 

“Together, these nine new projects will leverage SI combined with SLAC expertise to accelerate our scientific and technological impact,” SLAC Laboratory Director John Sarrao said in the laboratory’s announcement.

The Genesis Mission awards have different development goals. Phase I projects, including SLAC’s catalysis effort, are intended to demonstrate promising research workflows and evaluate whether AI can accelerate discovery, improve predictions or generate new scientific insights.  

A potential Phase II would expand the closed-loop system to additional laboratories, incorporating complementary measurement techniques. That expansion would depend on the Phase I project’s success and a subsequent funding decision, Sokaras said.