A national team led by the Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab) is developing FORUM-AI, an advanced AI research assistant designed to compress the decades-long timeline of materials discovery into years. Backed by a $10 million, four-year grant, this “agentic AI” platform aims to autonomously manage every step of the research process—from generating hypotheses to running supercomputer simulations and robotic lab experiments—to accelerate breakthroughs in batteries, semiconductors, and other critical energy technologies.
Imagine if the lengthy, arduous journey to create something like the lithium-ion battery could be shortcut through intelligent automation. That’s the ambitious goal of a new national project poised to reshape how we discover the materials that power our world. The mission is to build a proactive AI partner that doesn’t just answer questions but actively plans and executes complex scientific research.
The core problem FORUM-AI aims to solve is the traditional, linear bottleneck of scientific discovery. Developing a new battery material can take 20 years, moving painstakingly from idea to simulation to lab synthesis and testing. This new system seeks to parallelize and automate that workflow, allowing hundreds of hypotheses to be evaluated simultaneously. “FORUM-AI aims to be the first full-stack, agentic AI system for materials science research and discovery,” said the project’s principal investigator, Dr. Anubhav Jain, a staff scientist in Berkeley Lab’s Energy Technologies Area.
READ ALSO: https://modernmechanics24.com/post/boeing-mq28-ghost-bat-indo-pacific/
The system’s basic function is to act as an orchestrator for the entire research lifecycle. A scientist could present a goal, like finding a new solid-state electrolyte for faster-charging batteries. FORUM-AI would then generate multiple research plans, deploy agentic models to run simulations on leadership supercomputers at facilities like NERSC or Oak Ridge National Laboratory, and even command robotic labs like Berkeley Lab’s A-Lab to synthesize promising candidates. It uses three classes of AI: generative models for creating text and images, reasoning models for planning and interpretation, and agentic models to perform actions.
The visionary driving this effort is innovator and principal investigator Anubhav Jain, who also leads the famed Materials Project database. The engineering and development is a massive cross-institutional collaboration, uniting experts from Berkeley Lab, Oak Ridge National Laboratory, Argonne National Laboratory, MIT, and The Ohio State University. This team was selected under the DOE’s SciDAC program to co-lead the project.
However, a significant and honest limitation of such advanced AI in science is the risk of inaccuracy or “hallucination,” where models generate plausible but incorrect information. Jain acknowledges this head-on. To ensure reliability, FORUM-AI is designed to ground its decisions in verified databases like the Materials Project, use transparent and inspectable research plans so scientists can see the “thought process,” and rely on well-benchmarked, physics-based simulation tools instead of black-box predictions.
WATCH ALSO: https://modernmechanics24.com/post/china-humanoid-robot-combat-moves/
The overall value and summary of this project is transformative efficiency. By acting as a force multiplier for researchers, it aims to accelerate solutions to urgent energy challenges. Furthermore, the team is committed to energy efficiency in AI itself. They will use a technique called “distillation” to create smaller, less energy-intensive models from larger ones, enabling the AI to run on a laptop or directly attached to lab equipment, vastly expanding its accessibility.
The project is a testament to the foundational work done over decades at the national labs. “The national labs are an ideal place to develop AI-assisted materials research because for the last few decades they’ve been laying the foundation for it,” Jain explained. This includes creating vast materials databases, developing software to automate complex simulations, and building automated robotic labs—all essential infrastructure for an AI research agent.
Looking ahead, the four-year project hopes to demonstrate a fully autonomous loop: the AI proposes a novel, complex material, simulates its properties, instructs the A-Lab to synthesize it, analyzes the results, and then designs the next, improved iteration. Future visions include connecting FORUM-AI to major experimental facilities like light sources, where it could prepare experiments for visiting scientists.
READ ALSO: https://modernmechanics24.com/post/nasa-artemis-spacesuit-astronaut/
In a field where time is the most precious resource, FORUM-AI represents a paradigm shift. It’s not just another tool, but a potential collaborative partner, designed to leverage the nation’s most powerful supercomputers and lab infrastructure to usher in a new era of accelerated discovery for energy security and beyond.
