“It’s a better use for a PhD student than waiting in the lab until 3 in the morning to run an experiment to the end,” says Ant Rostron, ARIA’s chief technology officer.
ARIA selected 12 projects for funding out of 245 proposals, doubling the amount of funding to be allocated due to the large number and high quality of submissions. Half the teams are from the UK; The rest are from America and Europe. Some teams are from universities, some from industry. Each will receive approximately £500,000 (about $675,000) for 9 months of work. At the end of that time, they must be able to demonstrate that their AI scientist was able to draw new conclusions.
The winning teams include Leela Sciences, a US company that is building AI NanoScientist, a system that will design and run experiments to discover the best ways to compose and process quantum dots, which are nanometer-scale semiconductor particles used in medical imaging, solar panels and QLED TVs.
“We’re using the money and time to prove a point,” says Rafa Gomez-Bombarelli at Leela Sciences: “The grant lets us design a real AI robotics loop around a focused scientific problem, generate evidence that it works, and document the playbook so that others can reproduce and extend it.”
Another team at the University of Liverpool, UK, is building a robot chemist that runs multiple experiments simultaneously and uses a vision language model to help the robot troubleshoot if it makes an error.
And Humanis AI, a London-based startup, is developing an AI scientist called Thetaworld, which is using LLMs to design experiments to study the physical and chemical interactions that are critical to battery performance. The experiments will then be run in an automated laboratory by Sandia National Laboratories in the US.
take temperature
Compared to the £5 million projects over 2-3 years that ARIA typically funds, £500,000 is small change. But Rostron says that was the idea: It’s also an experiment on ARIA’s part. By funding a series of projects over a short period of time, the agency is taking the temperature of the cutting edge to determine how the way of science is changing, and how fast. What it learns will become the baseline for funding future large-scale projects.
Rostron admits there is a lot of hype, especially now that the teams at most top AI companies are focused on the science. When results are shared by press release and not peer review, it can be hard to know what the technology can and cannot do. “It’s always a challenge for a research agency trying to get funding at the frontier,” he says. “To work at the frontier we have to know what the frontier is.”
