Anthropic Releases Interface to Help AI Agents Operate Machines

Anthropic introduced an interface to make it easier for AI agents to operate machines in the real world, as part of the AI lab’s continued foray into physical AI.

The generative AI vendor unveiled a preview of the Model Hardware Standard (MHS) last week, with initial deployments set for scientific research and advanced manufacturing tools, including microscopes, robotic arms and other lab equipment.

Anthropic developed the standard as part of a collaboration with private biomedical research institute HHMI.

In addition to operating instruments, Anthropic said MHS enables agents to perform complex tasks, including drug discovery and laser calibration on a quantum computer.

Automating these tasks typically requires custom coding, but with the MHS, Anthropic says it can standardize the process and reduce integration time to hours or even minutes. The standard is also model agnostic, meaning users are not locked into Anthropic’s Claude models.

MHS can also provide agents with information about hardware that may not be apparent from code alone, such as a robotic arm's weight or safety limits. Users can provide this information in natural language, with MHS then creating a reference file that tells the agent what a device can measure, what can be adjusted and how it can be operated safely.

“By incorporating AI into these tools, MHS helps researchers and engineers more readily orchestrate autonomous, round-the-clock experiments and workflows,” Anthropic said in a blog post. “Agents [are] able to reason through each step in an experiment, update parameters in real time, and, in some cases, recover from hardware errors without intervention.”

Anthropic plans to make MHS available under an open source license, meaning it will be available to any device manufacturer across any industry. In addition to the current scientific applications, MHS is expected to be rolled out in manufacturing, with AWS and Hugging Face already working to integrate the standard into their workflows.