Anthropic Tests New Way for Claude to Work With Robots and Scientific Lab Tools

Anthropic PBC announced a new software standard that will allow artificial intelligence assistants like Claude to work better with robots and science and manufacturing hardware.

The AI startup is releasing its Model Hardware Standard, or MHS, under a research preview, so developers in the scientific, robotics and manufacturing fields can test how Claude interacts with the physical world and build in necessary safeguards before releasing it broadly.

Essentially, MHS stores information for AI assistants to understand how a piece of hardware works, without having to refer to paper manuals or tacit knowledge that only resides with a few experts. For example, a factory robot arm vendor could specify in an MHS how AI can manipulate a heavy arm safely by limiting the speed or angles with which it moves.

There is growing business and investor interest in how AI’s advancements can be applied in the physical world, such as for manufacturing, robotics and science experiments. A Barclays report from earlier this year forecast that AI-powered robots and autonomous machines have the potential to grow into a trillion-dollar market by 2035. Meanwhile, Alphabet Inc.’s Google, OpenAI and Nvidia Corp. are also developing AI models and software for robots.

Industry stakeholders can join a waitlist to trial MHS, Anthropic said in a statement on Thursday. It intends to open-source the framework after the preview period so that any hardware manufacturer can build their own MHS specifications, but it declined to share specific timelines.

“In many cases, the science doesn’t happen because you can’t use the equipment or it’s too technically challenging,” said Jonah Cool, a head of partnerships and deployment at the life sciences arm of Anthropic. “With MHS, Claude can be that kind of enabler for scientists as they can use the right equipment in an expert manner.”

Anthropic has similarly already introduced and popularized a standard for software applications called the Model Context Protocol, or MCP, so that AI assistants can more easily interact with third-party apps. That has enabled Claude and chatbot rival ChatGPT to connect with Gmail, Google Calendar and Slack, for example.

“What MCP did for software, MHS will do for the hardware world,” said Alek Kemeny, a member of technical staff at Anthropic. “Where it provides value is in science labs, where you have dozens or hundreds of devices.”

Anthropic has already been testing the hardware framework with a number of industry stakeholders, including Amazon Web Services, Danaher, Hugging Face Inc. and Raspberry Pi. In a promotional video from Anthropic, the company said a scientist from biotech firm Genentech Inc. was able to send Claude a PDF document of an experiment they’d designed, and the assistant autonomously executed the experiment with hardware that was equipped with the MHS specification.