What Battery Recycling Could Tell Us About AI’s Role in Experimentation

The Energy Department is testing whether artificial intelligence can change how scientists conduct research, using critical mineral recovery as an early example of how the Genesis Mission could accelerate scientific discovery.

Researchers at the Stanford University-led SLAC National Accelerator Laboratory are using a team of AI agents to identify how to recover nickel, cobalt and other metals from used lithium-ion batteries. Recovering those materials could help strengthen the domestic supply of critical minerals used in batteries and other technologies.

The project is part of the Genesis Mission’s effort to build an integrated scientific platform that connects supercomputers, experimental facilities, AI systems and specialized datasets across the research ecosystem. DOE has identified securing critical mineral supplies as one of the Genesis Mission’s national science and technology challenges.

For SLAC, the battery project provides a test of what real-world issues this AI model could impact.

“As the market for batteries grow, there will be increasingly a larger number of batteries that have reached end of life,” Steve Eglash, director of SLAC’s Applied Energy Division, told GovCIO Media & Research. “We don’t want those in landfills. We want to recycle them, and having the best recycling technologies can provide an important source of materials for the next generation of batteries.” 

Turning AI Into a Research Partner

The challenge is not simply extracting metals from a battery. Researchers need to determine how to separate metals such as nickel and cobalt efficiently and at a high level of purity.

One approach, called solvent extraction, uses different chemicals and conditions to separate the metals. Finding an effective combination can require extensive trial and error, which requires a lot of time and money.

This is where AI agents will assist in the research. Rather than relying on researchers to manually work through thousands of possible experiments, the lab is looking at AI to help identify the experiments that are most likely to produce useful results, said Ahamed Irshad Maniyanganam, a SLAC associate scientist and lead researcher on the project.

“It’s different from our original approach where we do thousands of experiments. We, researchers, will do a limited number because other experiments were done by the AI agent,” Maniyanganam said. “We’ll conduct the experiments, give feedback to the AI agent, and it will refine the process again until we get accurate results.”

The AI agents will have different roles such as searching scientific literature and other datasets like those from the American Science Cloud, and developing hypotheses. Researchers then choose which experiments to conduct and feed results back to the system to allow the AI to refine its recommendations.

That ability of the agents to search across disciplines is one of the potential advantages, Eglash said.

“The AI agent can look much more broadly across fields of biology and other aspects of the physical sciences,” he said. “That ability to look across different disciplines strikes me as very powerful.”

A Test of Genesis Model

The battery project is an example of a larger goal of the Genesis Mission of using AI to analyze scientific information and connect data in a continuous research workflow.

SLAC has already been developing AI-enabled workflows. Under Genesis, the lab is working to build AI-ready experimental ecosystems that can support more rapid and autonomous research across areas including energy materials, fusion energy, chemistry and physics.

The lab noted that if AI can reduce the number of experiments for this use case, the same approach could be applied to other scientific problems where researchers face enormous amounts of data and a large number of possible experiments.

Maniyanganam said this effort also demonstrates how to bring together expertise that historically has been siloed.

“When we collaborate, we’re not all AI or machine-learning experts. I’m an electrochemistry and battery material scientist who knows the battery problem, and my collaborator is really good at doing AI and theoretical modeling,” said Maniyanganam. “We are coming with a different expertise and collaborating with this new, modern tool to solve real-world problems.”