Making it A-Lab: Artificial Intelligence for the Prediction and Synthesis of Chemically-Bound and Unusual Materials

The A-Lab, located at LBNL, uses state-of- the-art robotic equipment to mix and heat powdered ingredients, and then analyse them to see whether the procedure worked. The set-up took 18 months to build. But the biggest challenge lay in using AI to make the system truly autonomous, so that it could plan experiments, interpret data and make decisions about how to improve a synthesis. The innovation is really under the hood, says Ceder, but the robot is great fun to watch.

Predicting the existence of a material is one thing, but making it in the lab is quite another. That is where the A-lab can help. Gerbrand Ceder is a materials scientist at LBNL and the University of California, Berkeley, and leader of the A-Lab team.

Andy Cooper, the director of the Materials Innovation Factory at the University of Liverpool, UK, says that systems like G No ME can make a lot of more predictions than a lab could keep up with. “What we really need is computation that tells us what to make,” Cooper says. More of the predicted material’s physical and chemical properties will be accurately calculated by the systems.

The problem, according to Persson, is bias: Over time, that collective knowledge has come to favor certain familiar structures and elements. TheEdison effect is a reference to the rapid trial-and- error quest to deliver a lightbulb, testing thousands of types of carbon before arriving at a variety derived from bamboo. It took another decade for a Hungarian group to come up with tungsten. “He was limited by his knowledge,” Persson says. He was convinced that he was biased.

In 17 days the A-lab created 41 new materials, 9 of which were created only after active learning improved the synthesis. Some materials were eventually synthesized, but only after humans interceded, because the majority of the 17 materials that the A- Lab failed to make were due to experimental difficulties.

The targets from the Materials Project database were cross-checked with the GNo ME database, and handed over to the A-Lab.

Several tactics are used by GNoMe to predict more materials than before. For example, rather than changing all of the calcium ions in a material to magnesium, it might substitute only half of them, or try a wider range of unusual atom swaps. It’s no problem if these tweaks don’t work out, because the system weeds out anything that isn’t stable, and learns from its mistakes. This is the kind of discovery that is called CHAT-GPT for materials discovery.

“Scientific discovery is the next frontier for AI,” says Carla Gomes, co-director of the Cornell University AI for Science Institute in Ithaca, New York, who was not involved in the research. “That’s why I find this so exciting.”