Anthropic's new AI tool can conduct scientific experiments: How it works

Anthropic has opened a research preview of its Model Hardware Standard, allowing AI agents such as Claude to control scientific equipment, automate experiments and manage workflows across labs

Anthropic has opened a research preview of a system that allows artificial intelligence (AI) assistants such as Claude to control scientific and engineering equipment and conduct experiments, marking the company’s first tool designed to operate in the physical world.

 

Called the Model Hardware Standard (MHS), the technology enables AI agents to autonomously operate a wide range of devices that can be programmed using computer code. These include complex microscopes, liquid handlers, lasers and robotic arms.

 

The research preview will initially be available to a first group of scientific research labs and advanced manufacturers, Anthropic said. The development comes as the Claude chatbot maker seeks to expand its enterprise business and boost revenues with new applications.

 

 

How does Anthropic’s new hardware standard work?

 

Setting up and integrating hardware in a laboratory or manufacturing facility can take weeks or even months, Anthropic said. Many devices do not communicate with one another, requiring specialists to create bespoke software integrations for each setup.

 

According to Anthropic, MHS is designed to reduce this integration work to hours or minutes by providing a standard way for computers and AI agents to communicate with different hardware.

 

At the core of MHS is a standardised driver, software that acts as an interface between a computer’s operating system and a hardware device. The driver uses a simple set of commands, or primitives, such as “read” to retrieve information such as temperature and “write” to change a setting such as temperature. Each device is also made discoverable in a standard format, allowing hardware and AI agents to find and communicate with one another across networks without requiring a separate translator for each device.

 

How can AI agents use MHS to conduct experiments?

 

MHS is designed to work with a wide range of programmable scientific and engineering devices. These include advanced microscopes, liquid handlers, lasers and robotic arms.

 

Anthropic said AI agents can reason through individual steps in an experiment, change parameters in real time and, in some cases, recover from hardware errors without intervention. This could allow researchers and engineers to run autonomous experiments and workflows around the clock. The company said Claude can also interact with hardware in an exploratory manner, similar to how a scientist might approach an unfamiliar experiment.

 

Jonah Cool, who leads partnerships and deployment for science at Anthropic, told the Financial Times that advanced microscopes can contain hundreds of mirrors, motors and lasers that are normally operated by specialist engineers. With MHS, Claude can communicate directly with individual components of such equipment, potentially allowing researchers to automate complex processes that would traditionally require specialist intervention.

 

“The conceptual leap here is going from Claude agents working in the digital world [of] analysis, hypothesis generation to do experiments in the physical world, in the lab,” Cool said, according to the Financial Times.

 

Where has Anthropic tested the technology?

 

Anthropic has been testing MHS with a small group of laboratories and hardware manufacturers working in biotechnology, robotics and quantum computing.

 

The participants include the Howard Hughes Medical Institute, Carnegie Mellon University, Genentech and QuEra, a company developing quantum computers.

 

Anthropic said scientists have observed AI agents tracking, adjusting and repeating experiments in ways similar to human researchers. In one neuroscience experiment, an Anthropic scientist observed Claude examine live brain tissue and identify the alveus, a specific part of the brain, according to the Financial Times.

 

Anthropic plans to open-source technology

Anthropic plans to open-source MHS, but said it will first work on safeguards to reduce the risk of errors when AI agents interact with physical equipment, reported Financial Times. 

 

The launch follows Anthropic’s introduction in June of Claude Science, its first product dedicated to scientists. The product includes applications such as rendering three-dimensional protein structures and assisting with drug discovery. MHS takes that push a step further by allowing AI agents to move from analysing scientific information and generating hypotheses to interacting with the physical equipment used to test them.

 

Anthropic is also competing with major technology companies developing AI systems for robotics and physical-world applications. Alphabet’s Google, OpenAI and Nvidia are among companies working on technologies designed to enable AI systems to interact with and control robots.