Computers have suggested around 150,000 promising new materials that could improve everything from battery electrodes to fabrics. 

However, those computers have left it to humans to figure out the right balance of ingredients for each of those suggestions and then to test those notions, one by one, to see which actually work.

As a result, scientists have been able to try only a small fraction of those possibilities and found even fewer that make sense.

At the Lawrence Berkeley National Laboratory, humans now have help: a robotic lab, enlivened by artificial intelligence that knows state-of-the-art chemistry, working 24 hours, seven days a week, to test those predictions and see which are worth pursuing.

For each predicted material, the AI guesses what the right proportion of materials will be. 

It then directs robotic arms to take small scoops from among about 200 different powdered ingredients in compartments around the lab. Once the materials are mixed, a different robot dollops the materials into crucibles which are set into furnaces where the mixtures can be treated with various gases.

The AI determines the temperature at which to bake the materials, how long to leave them in the furnace, how long to let them dry or cool, and so on.

Next, a dispenser drops a ball bearing into each crucible. The crucibles are shaken to grind the new material back into a powder. 

Yet another robot then sprinkles a sample of the new material onto a glass slide and slips it into an x-ray machine or similar device to be analyzed. 

The result is entered into a database, which records the sample’s structure and properties. If the outcome is unsatisfactory, the AI can adjust its recipe and try again.

The need for human action is limited to making sure the lab is supplied with materials, to be on hand in case anything breaks down or otherwise goes wrong, and to review the results.

The lab is testing 100 times more new materials every day than humans could do in the same lab, the project’s managers say. So far, it has produced about 40 new materials.

Similar robotic AI labs are at work discovering new drugs. Another set up by Samsung is exploring possible new electronic materials.

“I have made more new compounds in the last six weeks than in my whole career,” Gerbrand Ceder, a scientist involved with the project, which is known as the A-Lab, told Science magazine.

TRENDPOST: Artificial intelligence combined with robotics is already revolutionizing materials science, which, in turn, will yield materials that will revolutionize everything from medical therapy to electric vehicles.

The result of speedier discovery should allow industry to leapfrog entire generations of products that would have been made without the discoveries the new labs are producing.

TRENDPOST: Cedar claims to “have made more new compounds in the last six weeks than in my whole career.” Actually, Ceder didn’t make any new materials. He co-created a lab where AI made new materials.  

His comment raises the broader issue of the transfer of human abilities to machines and computers: if machines take over more and more complex work, why would humans bother to do the hard work to learn those skills ourselves—and as we surrender our abilities to machines, how does that affect the evolution of human intelligence?

Lawrence Berkeley National Lab’s A-Lab.

Photo:  Lawrence Berkeley National Laboratory.

Skip to content