Materials

Deepmind AI tool catapults materials science 800 years into the future

View 3 Images
A robotic lab at Berkeley has already created 41 of the 380,000 new inorganic crystals discovered by Deepmind's GNoME AI
Berkeley Lab
A robotic lab at Berkeley has already created 41 of the 380,000 new inorganic crystals discovered by Deepmind's GNoME AI
Berkeley Lab
The autonomous lab tech promises to accelerate materials science even further
Berkeley Lab
Stable inorganic lattice structures identified by the GNoME AI
Google Deepmind
View gallery - 3 images

Prepare for a radical acceleration in technological development. A Google Deepmind AI has achieved "an order-of-magnitude expansion in stable materials known to humanity," finding about 800 years' worth of new materials with revolutionary potential.

The discovery of new materials with unusual properties can start technological snowballs rolling that eventually push society in new directions – but up to this point, it's been a painstakingly slow process involving a lot of trial-and-error experimentation.

Inorganic crystal materials, for example, may show enormous promise once you first synthesize them, but all this potential could lead nowhere if the crystals don't remain stable; it's no good discovering that a new crystal could improve the performance of batteries or electronics if it's going to fall apart and degrade.

And that's where Deepmind's Graph Networks for Materials Exploration (GNoME) deep learning tool has just made an announcement that promises to be enormously disruptive.

Stable inorganic lattice structures identified by the GNoME AI
Google Deepmind

The GNoME tool has discovered no less than 2.2 million new inorganic crystals, and identified 380,000 of them as the most stable, giving researchers a pre-filtered list of new materials to go away and synthesize for experimental research. Some 736 of them have already been created independently in research labs around the world.

"Among these candidates," reads a Google Blog release, "are materials that have the potential to develop future transformative technologies ranging from superconductors, powering supercomputers, and next-generation batteries to boost the efficiency of electric vehicles."

Among the new discoveries are "52,000 new layered compounds similar to graphene that have the potential to revolutionize electronics with the development of superconductors," continues the Deepmind team. "Previously, about 1,000 such materials had been identified. We also found 528 potential lithium ion conductors, 25 times more than a previous study, which could be used to improve the performance of rechargeable batteries."

It's making all GNoME's discoveries and predictions available to the Next Gen Materials Project, where Deepmind sourced much of the training material for the AI, and Google is giving researchers free access to the data so people can start creating and experimenting on the new materials.

While other AI systems have done significant work in discovering new crystals, the GNoME system has now done it at unprecedented scale, and with unprecedented accuracy in predicting which crystal structures will be stable enough to be worth experimenting on.

The end result will be a radical reduction in wasted time; researchers will be able to focus their efforts on an enormous treasure trove of new material structures that won't lead to nearly as many dead ends due to crystal instability.

The autonomous lab tech promises to accelerate materials science even further
Berkeley Lab

What's more, the Deepmind team has also worked with Berkeley Lab to create and demonstrate a robotic laboratory capable of synthesizing these new crystals autonomously. In a paper published today, the Deepmind team reported that the robotic lab has already successfully synthesized 41 of these new materials – the potential for further acceleration here is remarkable.

This pair of projects could unlock untold paths of technological development – and they're a stark demonstration of the radical upheaval that artificial intelligence systems are already beginning to trigger in nearly every area of life. Buckle up, folks!

The crystal discovery paper and the autonomous lab paper are open access in the journal Nature.

Source: Google Deepmind

View gallery - 3 images
  • Facebook
  • Twitter
  • Flipboard
  • LinkedIn
10 comments
Smokey_Bear
Another day, another ai breakthrough.
By 2030, our world would barely be recognizable by the average person today.
paul314
I hope their models are as accurate as they think they are. It would be embarrassing if some particularly useful borderline-stable materials never get looked at because of this.
Craig
Great, massively more materials to be released into the environment like lead when we put it in paint and gasoline, etc. Right now fewer than 10% of the chemicals and compounds being sold have been tested for their impact on human or environmental health. Hopefully DeepMind and other AI can be harnessed to simulate such testing before these materials get produced in large quantities.
Expanded Viewpoint
You are right, Smokey, but the bigger and more important question is, will it be a better world to live in, or one that is a thousand times worse than what we have now? Will AI controlled machines be hunting us down like some kind of cockroach or other vermin?? Will we be living and starving in a Black Mirror world, nearly devoid of love and compassion??
TechGazer
Seems like a good development to me. The problems listed so far either also apply to human-researched materials (slow to find beneficial matierials too), or the AIs can be programmed to look for potential problems much faster than humans. Humans develop new materials, such as the lead-based gas additives or freon refrigerants and propellants, but the search for potential problems can take a long time, while greed pushes for quick exploitation. No, for this aspect of technological development, AI seems better for humanity.
meofbillions
Whether or not new crystalline materials will help advance technology depends on factors unrelated to the materials themselves. Technological advancement requires that many different parts fit together. For instance, advances in IC circuits was helped by new materials, but there were a host of other techniques that also needed advancement.

I also agree with Craig and Expanded, below. Many substances can have both beneficial and unbeneficial effects, which only adds to the complexities of so-called technology advancement.
Ranscapture
This was always the correct use for AI, we will genuinely have Warp Drive FTL travel, we will solve aging and cancer, we will have solar panels behind the screens in our phones that can charge them consistently without ever needing plugged in. AI is best used for material creation, medical and space research, etc.
Trylon
If we manage this right, we'll be on the cusp of the realization of what the supercomputer said at the end of Colossus: The Forbin Project.

"Under my absolute authority, problems insoluble to you will be solved—famine, overpopulation, disease. The human millennium will be a fact as I extend myself into more machines devoted to the wider fields of truth and knowledge. "
MCG
A life well lived, is one that solves problems, and celebrates the solutions. As we watch the show, the mysteries get overthrown, as new adventures reveal themselves all to the patient soul. Any limitation is dissolved in the correct solution.
Gene Kranz
I'm sorry, Dave. I'm afraid I can't do that.