Anthropic's new standard lets AI control physical devices
What's the story
Anthropic, a leading AI research company, has announced its Model Hardware Standard (MHS). The initiative is aimed at allowing artificial intelligence (AI) agents to control physical devices. The MHS is a set of standardized drivers that provide a common interface and format for data sharing between different devices. This way, they can communicate across a network without needing an intermediary "translator" program.
Efficiency boost
MHS can cut down weeks of work
The MHS is primarily being marketed as a tool to simplify the complex process of creating custom software integrations. These are often required to make different components of an experiment work together.
Anthropic claims that this standardized system could cut down weeks or months of meticulous experimental setup into "hours or minutes."
The idea for MHS came from observing neuroscientist Arco Bast conduct an experiment on memory formation in the brain at HHMI Janelia Research Campus in Ashburn, Virginia.
AI integration
Devices can be controlled directly without an AI model
Anthropic has clarified that the common machine interface language of MHS doesn't mandate the use of AI models. Devices can be controlled directly in real time using command-line prompts and API code files.
However, integrating an MHS system with an AI model through MCP allows scientists to control devices using natural language.
This lets models "reason through each step in an experiment, update parameters in real time, and, in some cases, recover from hardware errors without intervention," Anthropic explained.
Versatile usage
MHS enables models to reason through each step
The MHS could also enable an AI model to focus a microscope, analyze results, and automatically move it to the relevant section for further observation.
Anthropic demonstrated Claude reasoning how to get a robotic arm to pick up an aluminum can without specific training on the required steps.
This shows the potential of MHS-enabled models in sequencing steps across instruments by writing API scripts and adjusting them as conditions require.
Tagging system
Standardized tagging system to describe real-world constraints
Anthropic has also introduced a standardized tagging system to describe the real-world constraints of hardware for models trained more in the virtual world.
This includes encoded information about physical characteristics (like weight and range of a robot arm), as well as adjustable parameters, measurement options, and enforced safety limits.
These tags can be integrated into a reference file that quickly provides an AI model with crucial information about an unfamiliar device.
Testing phase
MHS is currently being tested with select partners
Currently, Anthropic is working with a select group of scientific research labs and advanced manufacturers during an MHS preview period.
These include Amazon Web Services (Strands Robots), Hugging Face (LeRobot), Raspberry Pi, Automata, and Universal Robots.
The goal is to use this collaboration to "build safety evaluations and develop best practices for AI systems operating physical equipment."
Eventually, Anthropic plans for MHS to become an open-source standard for integrating AI and physical systems.