AIBRIEF
RAND says self driving lab experiments keep arriving from U.S. groups while the shared benchmarks, data standards, and ten year institutional homes to compound them still don't exist.
AI can propose a new battery cathode, alloy, or superconductor candidate in days. A new RAND perspective paper argues the U.S. is not yet building the shared plumbing that lets those wins compound.
The paper, "Governing The Accelerator," appears in Advanced Materials Technologies and identifies three governance gaps: data and infrastructure, algorithmic intelligence, and autonomous experimentation. Its authors, researchers at federally funded R&D centers, propose a two-tier model: S&T labs run the science, while S&A centers handle policy and cost-benefit work.
The build question is more concrete than the taxonomy. Open benchmarks, shared data formats, autonomous-lab interfaces, and a ten-year institutional home for the field all still run at hobbyist speed. Without them, one AI-driven result rarely compounds into the next. The paper also notes that foreign competitors "pursue centralized national materials programs of their own," the authors' framing and not something independently sourced here.
Because the paper is a perspective rather than a finding of fact, a full feature would still need one named U.S. AI-materials program and one foreign comparator with documented capability. For now the read is plain: the U.S. keeps inventing new materials with AI, and the shared layer that would let the next one come cheaper has not been built.