Google AI Discovers 800 Years’ Worth of Industrial Materials in One Click

The future of industrial material sciences is starting to crystallize — literally.

Google published a research paper Wednesday (Nov. 29) unveiling more than 2 million hypothetical material designs, each previously unknown to science.

The catch? The crystalline compounds weren’t discovered by lab researchers or scientists but by an artificial intelligence model developed by Google’s DeepMind team called Graph Networks for Materials Exploration (GNoME).

But this isn’t your garden-variety GNoME.

Of the 2.2 million new AI-generated structures identified, 380,000 of them are stable enough that they could soon be produced in lab conditions.

Before the AI burst onto the scene, scientists — even with the aid of cutting-edge computational data analysis — had discovered only around 48,000 crystalline material designs. “Crystal,” in this case, doesn’t refer to the pink rocks being peddled by Instagram influencers but to a massive family of compounds whose atomic structure is made up of repeating units.

Untold trillions of these possible structures exist, and the ones we have so far discovered and developed sit at the inorganic heart of nearly all modern technologies.

The potential applications of Google DeepMind’s hundreds of thousands of new materials include the production of better-performing solar panels, batteries and computer chips, to name a few.

The DeepMind team said the GNoME model’s findings are the equivalent of nearly 800 years’ worth of knowledge.

Read also: Companies With AI-Driven Strategies Outcompete Peers, Study Finds

Scientific Discovery Is the Next Frontier for AI

By leveraging AI, GNoME was able to accelerate the speed and efficiency of industrial material discovery, a field that hinges almost entirely on predicting the stability of new materials.

Compounds can be designed inventively, even willy-nilly, but much of the time simply packing together atoms in a crystal leads to unstable materials that fall apart under variable conditions outside of their hypothetical parameters.

The months, even years, of painstaking trial-and-error experimentation around stability make material discovery expensive and time consuming.

GNoME shows the potential of using AI to discover and develop new materials at scale while simultaneously driving down the cost. Per DeepMind, external researchers have independently created 736 of GNoME’s new materials in the lab.

It took around two decades of research before lithium-ion batteries, a notable advance and foundational component of modern life, became commercially available. That means AI could accelerate to-market timelines for new industrial materials.

Underscoring the potential of AI within the space, GNoME found 528 potential lithium-ion conductors, which could be used to improve the performance of rechargeable batteries, 25 times more than a previous study.

See also: Specific Applications of Gen AI Are Winning Play for Enterprises

What About Google’s General Purpose Gemini Model?

The GNoME AI model was trained using data from the Materials Project, Open Quantum Materials Database (OQMD) and other research-specific data sets. It represents a breakthrough success for Google’s DeepMind team.

But it comes as Google is still yet to release its general-purpose Gemini foundation model, meant to compete with OpenAI’s multimodal capabilities, and once rumored for a release this fall, or at least by the end of the year.

That doesn’t appear to be happening and reflects the challenges of effectively building generative AI that can perform across multiple content modalities. Building domain-specific success with clear data frameworks plays to certain strengths that can confound the generation of broader AI.

As PYMNTS has reported, the generative AI industry is expected to grow to $1.3 trillion by 2032. But rather than one single, all-knowing super-AI that is better at everything humans can do, the marketplace growth is likely to be driven and accelerated by a variety of different AIs with different strengths, each fine-tuned for diverse applications.

PYMNTS Intelligence found that nearly two-thirds of Americans want an AI copilot to help them complete specific tasks, like book travel, and travel companies are already leaning into the technology’s applications in their industry.

Another area where AI can be applied right now is in fraud prevention and detection, and other areas within payments including credit scoring.

“If you go into a field where the data is real, particularly in the payments industry… AI can bring a lot of benefit,” Akli Adjaoute, founder and general partner at venture capital fund Exponion, told PYMNTS in an interview posted Thursday (Nov. 30).

For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.