Materials Research | Now artificial intelligence is cooking up a huge number of new materials

Search company Belongs to Google artificial intelligence Deepmind has been worked to find new materials.

It was quite a surprise when Deepmind produced almost 400,000 recipes for new substances.

Now their first batch has been “cooked” in the laboratory, which the name is A-lab. It is located at the University of Berkeley in California, USA.

A-lab especially designs materials that can be used in batteries or solar cells. But a lot more has been added to the list.

A-lab studies compounds and tries to make them real with robots, without human help.

To the same in time, another part of the artificial intelligence examines hundreds of thousands of new materials and finds out if they remain stable.

The work provides A-Lab with plenty of new candidates for materials.

Chemists and other researchers believe that artificial brains like Deepmind will dramatically accelerate the production of new materials.

The new materials have, for example, clean energy technologies, such as new generation electronics, says the materials researcher Even Dogus Cubuk science journal in Nature.

Cubuk leads a team that searches for materials at Google’s Deepmind center in London. He was involved in two studies that Nature published at the end of November.

Over the centuries, people and scientists have developed thousands and thousands of inorganic compounds.

They are materials that are not based on chains of carbon atoms and compounds. Carbon atoms are typical for organic chemistry.

Artificial intelligence Deepmind’s search and findings suggest that even billions of materials are still waiting to be discovered. Many of them are built quite simply.

In laboratories in the past, attempts have been made to shorten the time it takes to search for new materials.

Researchers have computationally modeled new inorganic materials. For example, they calculate how different atoms would pack together in the same crystal.

These tests recently produced around 48,000 pieces of material. Lawrence Berkeley National Laboratory (LBNL) in Berkeley was involved.

“

The biggest challenge was to make the artificial intelligence so independent that it can design experiments, interpret data and make decisions.

Google Deepmind extended these methods with what AI researchers call Graph Networks for Material Exploration (GNoME).

It calculated whether the newly discovered materials would be stable. It also predicts their crystal structures.

In this way, it produced the combinations of no less than 381,000 new inorganic substances. Now they have been added to the Material projects database.

GNoME effectively weeded out materials that are not stable. It is different to predict a material than to do it in a laboratory.

A-Lab comes along. A-Lab uses robots to mix powdered solid ingredients and heat them. Thus, it still tests the product.

A-Lab’s construction in Berkeley took 18 months.

The biggest challenge was to make the artificial intelligence so independent that it can plan experiments, interpret data and make decisions about what could be improved in the experiments.

“Robots are fun to look at, but the real discovery is under the hood,” says the materials researcher Gerbrand Ceder. He led the research in A-lab.

Ceder’s team identified 58 compounds from the project’s database that were predicted to be stable. He then fed these candidates to A-Lab’s machine learning models.

A-lab suggested the ingredients needed for the preparation and the reaction temperatures. Then the device itself picked the ingredients from the racks. A-lab cooks the product and examines the final result.

If less than half of the product is the targeted substance, the program tries to come up with a better procedure using “active learning”. And then the tireless robot starts all over again.

in A-Lab it took 17 days to produce 41 new inorganic materials. 9 of them were born only after the program improved its preparation.

However, A-Lab failed to make 17 new materials. In the end, a person was also needed to grind the mixture again.

“With the help of artificial intelligence, we get a map of the ability of ordinary substances to react with each other. It changes the world – not A-Lab itself, but the knowledge and information it produces,” says Ceder.