AI models can now help run physical science experiments
Anthropic · 2026-08-27 · official · 417,625 views
What's in the video
Description written by Gemini, which watched and listened to the whole video.
Summary
Anthropic presents "Model Hardware Standard" (MHS), an open protocol designed to allow AI models like Claude to directly interface with and control physical laboratory hardware and scientific instrumentation. Anthropic technical staff members Alek Kemeny and Gagan Bhat document real-world tests and collaborations with researchers at HHMI Janelia Research Campus, Leica Microsystems (Danaher Corporation), and Genentech across neuroscience, robotic manipulation, live microscopy, and automated drug discovery.
What is shown
- [01:10 - 02:30] Dr. Arco Bast at HHMI Janelia Research Campus demonstrates his custom-built multiphoton laser-scanning microscope used for live brain imaging, highlighting the challenge of synchronizing diverse hardware components.
- [02:35 - 03:05] The Anthropic team collaborates with Janelia to establish the initial Model Hardware Standard communication layer, testing remote stage control and laser activation.
- [03:10 - 04:30] In Anthropic's office, Gagan Bhat connects Claude via MHS to a multi-axis robotic arm, establishing a 3D safety bounding box ("Safety Range Visualizer") that blocks out-of-bounds motions before commanding Claude to locate and grasp a beverage can.
- [04:35 - 06:14] At Danaher/Leica Microsystems, engineers connect Claude to a Leica research microscope; Claude navigates the sample, focuses, and interprets stained botanical cell wall structures (differentiating lignified xylem vessels from parenchymal cells).
- [06:40 - 07:49] Claude generates a Python script and a live user interface to autonomously track a swimming micro-organism (diatom) in real time under the microscope for several minutes.
- [08:22 - 10:25] At Genentech, researchers connect Claude to high-throughput liquid-handling platforms; Claude detects air bubbles inside 96-well microplates and adjusts pipetting parameters in a closed-loop sequence to reduce volume transfer errors.
Claims & numbers
- Time spent on experimental setup: Alek Kemeny states that building experiments, setting up devices, and debugging hardware/software consumes "maybe 80% of a scientist's time" [00:23].
- Setup efficiency for PhD researchers: A Danaher team member claims that setting up such dynamic systems typically takes a PhD researcher "two years to get it running," whereas with this prototyping framework "he only needs two months" [07:58].
- High-throughput screening scale: Margaret Porter Scott notes that Genentech tests "thousands, or even hundreds of thousands, or even millions of molecules to find the right molecule" [08:44].
Notable quotes
- [02:23] Alek Kemeny: "This idea could be used to have AI run any science experiment in the world."
- [04:18] Gagan Bhat: "The mere fact that I was able to build this from scratch today, and it achieved it in a matter of minutes—that's insane."
- [06:19] Luciano Guerreiro Lucas: "Claude walked in, we told him nothing, and it was just trying to figure it out."
Assessment
This is an official demonstration documentary by Anthropic illustrating early practical integrations of Claude with lab automation and scientific instruments. The trials depict real laboratory interactions—including terminal execution logs, UI development, and mechanical safety intercepts—presented through a professionally edited promotional narrative highlighting successful test runs.
Described by gemini-3.8-flash on 2026-09-29 from the video's audio and frames.