Google DeepMind’s AI Finds 2 Million New Building Blocks

Google DeepMind's AI discovers 2 million new building blocks with artificial intelligence

In a groundbreaking development, Google DeepMind researchers have harnessed the power of artificial intelligence (AI) to uncover an unprecedented 2.2 million crystal structures, opening the door to advancements in renewable energy, advanced computation, and beyond. The discovery, made using the AI tool GNoME, surpasses the cumulative number of such substances discovered in the entire history of science by a staggering 45 times, marking a significant leap forward in material science.

The findings, detailed in a paper published in Nature, underscore the transformative potential of AI in expediting scientific discovery and innovation. The researchers plan to share 381,000 of the most promising structures with fellow scientists, facilitating experimentation across various fields, from solar cells to superconductors.

Ekin Dogus Cubuk, a co-author of the paper, highlighted the profound impact of improved materials on technology, stating, “Materials science to me is basically where abstract thought meets the physical universe. It's difficult to think of any technology that wouldn't advance with improved components.

The research objective was to identify new crystals to augment the existing 48,000 known structures. Employing machine learning, the DeepMind team generated candidate structures and assessed their likely stability. The magnitude of substances discovered is equivalent to almost 800 years of previously experimentally acquired knowledge, as estimated by DeepMind based on the discovery of 28,000 stable materials in the past decade.

The Nature paper states, “From microchips to batteries and photovoltaics, discovery of inorganic crystals has been bottlenecked by expensive trial-and-error approaches. Our work represents an order-of-magnitude expansion in the human knowledge of stable materials.

Two potential applications of the newly discovered compounds include the invention of versatile layered materials and the development of neuromorphic computing, a technology that mimics the workings of the human brain using chips.

Researchers from the University of California, Berkeley, and the Lawrence Berkeley National Laboratory utilized the findings in experimental efforts to create new materials. Their success rate was impressive, with more than 70% of the target list of 58 compounds synthesized by the autonomous laboratory, known as the A-lab.

Gerbrand Ceder, co-author of the paper and a professor at the university, highlighted the surprising success ratio, noting that the integration of AI techniques with existing sources, such as a large dataset of past synthesis reactions, played a key role. Ceder emphasized that the real innovation lies in the intelligent combination of various sources of knowledge and data with the A-lab to drive synthesis effectively.

“While the robotics of the A-lab is cool, the real innovation is the integration of various sources of knowledge and data with A-lab in order to intelligently drive synthesis,” said Ceder.

Bilge Yildiz, a Massachusetts Institute of Technology professor not involved in the research, lauded the techniques outlined in the two Nature papers, stating that they enable the identification of new materials at speeds necessary to address global challenges. Yildiz expressed optimism about the expansive database of inorganic crystals, describing it as a potential source of valuable discoveries to advance solutions in clean energy and environmental challenges.

The findings from Google DeepMind represent a monumental leap in the quest to obtain materials at speeds surpassing traditional empirical synthesis approaches. The impact of AI in revolutionizing material science is increasingly evident, with this breakthrough poised to shape the trajectory of innovation in diverse technological domains.