Anthropic has introduced the Model Hardware Standard (MHS), a set of standardized drivers designed to enable AI agents to interface with and control various physical devices. The initiative, described as a 'research preview,' is intended to help scientists streamline the process of creating custom software integrations for experiments. MHS provides a common interface for data sharing between devices, allowing them to communicate across a network without requiring a bespoke 'translator' program. Anthropic claims the standardized system could cut experimental setup time from weeks or months to 'hours or minutes.'

The MHS effort was inspired by observing a neuroscientist at the HHMI Janelia Research Campus in Ashburn, Virginia, who developed a common interface to coordinate components of an experiment. According to Anthropic, the idea could enable AI to run any science experiment globally. While MHS does not require AI models, it allows for real-time control via command-line prompts and API code files. Integrating MHS with AI models through the Model Context Protocol enables natural language interaction with devices, allowing models to 'reason through each step in an experiment, update parameters in real time, and, in some cases, recover from hardware errors without intervention.' Anthropic provided an example of a model like Claude adjusting a laser, checking results via a camera, and repeating the process to automatically calibrate a system.

Anthropic also highlighted that MHS includes a standardized tagging system to describe hardware constraints for models trained primarily in virtual environments. These tags encode information about physical characteristics, adjustable parameters, measurement options, and safety limits. The tags can be integrated into a reference file, providing AI models with crucial information about unfamiliar devices. Currently, Anthropic is collaborating with a group of scientific research labs and advanced manufacturers, including Amazon Web Services, Hugging Face, Raspberry Pi, Automata, and Universal Robots, to build safety evaluations and develop best practices for AI systems operating physical equipment. Source: arstechnica