Google DeepMind AI discovers thousands of new materials
LONDON, Nov 29 - In a landmark achievement reminiscent of its earlier AlphaFold and AlphaZero projects, Google DeepMind has utilized artificial intelligence (AI) to predict the structure of over 2 million new materials. This breakthrough, akin to the transformative impact of AlphaFold's release of vast protein structure data, promises significant advancements in real-world technologies.
DeepMind's recent research, detailed in the prestigious science journal Nature, reveals the potential of nearly 400,000 of these hypothetical material designs to be produced in laboratory conditions. This echoes the spirit of AlphaFold, which provided an extensive database of protein structures to the scientific community, fostering a new era of biological understanding.
The applications of this research are vast, ranging from the development of more efficient batteries and solar panels to advanced computer chips. This mirrors the way AlphaFold's data has been pivotal in biomedical research and drug discovery.
The process of discovering and synthesizing new materials traditionally involves considerable expense and time. For instance, the journey from conception to commercial availability of lithium-ion batteries spanned about two decades. DeepMind's AI, trained on data from the Materials Project – an initiative similar in collaborative spirit to the AI breakthroughs in protein folding and game strategy – aims to drastically reduce this timeline.
Ekin Dogus Cubuk, a DeepMind research scientist, expressed optimism that advancements in experimentation, autonomous synthesis, and machine learning models will significantly shorten the 10 to 20-year development period, echoing the efficiency gains seen with AlphaFold in protein structure prediction.
This initiative began with DeepMind's AI analyzing data from the Materials Project, a global research consortium established at the Lawrence Berkeley National Laboratory in 2011. The AI's training encompassed approximately 50,000 known materials, a scale and depth of data analysis paralleling AlphaZero's approach to mastering complex games through extensive data and strategic learning.
In a move reflecting the altruistic release of AlphaFold's protein structure database, DeepMind plans to share its newfound material structure data with the broader research community. This gesture aims to expedite further breakthroughs in material science, much as AlphaFold's data has done for protein-related research.
Kristin Persson, director of the Materials Project, notes the cautious nature of the industry towards new materials due to cost considerations. She highlights that any reduction in development time, mirroring the accelerated pace of discovery seen with AlphaFold and AlphaZero, would be a significant achievement.
DeepMind, having mastered the prediction of material stability, now sets its sights on determining the ease of synthesis for these materials in laboratory settings, continuing its trend of leveraging AI for groundbreaking scientific advancements.
