AI scientist autonomously runs X-ray experiments at DOE facility

Researchers at Northeastern University have developed an artificial intelligence system capable of conducting X-ray experiments autonomously, including the ability to adjust its approach when equipment malfunctions or unexpected conditions arise.

The system, described in the July 2026 issue of Nature Machine Intelligence, represents a significant step toward fully automated scientific laboratories. Unlike conventional AI agents that follow rigid instruction sets, this "AI scientist" reasons through experimental problems and modifies its strategy based on real-time observations.

How the system works

The AI scientist operates at the Stanford Synchrotron Radiation Lightsource (SSRL), a Department of Energy facility at SLAC National Accelerator Laboratory. The facility uses a circular particle accelerator roughly 768 feet in circumference to generate X-rays for studying quantum materials.

Traditionally, X-ray scattering experiments require extensive manual setup. Scientists must align tiny crystal samples with extreme precision, position detectors, and continuously adjust equipment—tasks that consume substantial portions of expensive beam time at oversubscribed facilities.

The AI system handles these preparatory steps independently. Built on large language model architecture similar to ChatGPT, it can physically interact with laboratory equipment, send commands to diffractometers, analyze scattering patterns, and make mechanical adjustments.

During testing at SLAC, the system successfully determined a crystal's spatial orientation and managed the complete experimental workflow. When a motor malfunction occurred during sample positioning, the AI detected the problem, adapted its approach, and applied that knowledge to subsequent steps.

Why it matters

X-ray facilities like SSRL face intense demand and high operational costs. Beam time is limited, and much of it gets consumed by setup and alignment rather than data collection. Automating these tasks could dramatically increase research throughput at facilities where scientists currently compete for access.

The technology also addresses a practical constraint: existing AI models can handle the work without expensive custom training. By demonstrating success with off-the-shelf language models, the research team showed that autonomous laboratory systems could be deployed more widely without prohibitive development costs.

Broader context

The Northeastern project is part of a larger initiative at the university's Quantum Materials and Sensing Institute, directed by Arun Bansil, to develop fully self-driving laboratories. The work was funded by the U.S. Department of Energy in collaboration with SLAC.

Self-driving labs are already operational in other domains. Atinary Technologies, a chemistry lab in Boston's Seaport district, reportedly generates as much experimental data weekly as a typical Ph.D. student produces over an entire degree program. Rigoberto Advincula at Oak Ridge National Laboratory noted that combining machine learning with robotics has proven valuable in reaction chemistry, drug discovery, and materials science.

The Northeastern team deliberately chose the term "scientist" rather than "agent" to emphasize the system's reasoning capabilities. According to Bansil, the distinction matters: generic AI agents execute prescribed instructions, while this system receives broad guidance and then operates independently, making decisions based on experimental conditions.

The development allows human researchers to focus on experimental design and scientific interpretation rather than equipment operation and troubleshooting, according to the research team.

Details of the system and experimental results were first reported by Northeastern Global News.