Google DeepMind researchers have discovered 2.2 million new materials with the help of AI

Scientists at Google DeepMind have used artificial intelligence technology to search for 2.2 million previously unknown crystal structures that could find applications in fields ranging from renewable energy to quantum computers.

Here's What We Know

According to the study, the number of theoretically stable, but not yet synthesised in the lab, compounds discovered using the GNoME AI tool exceeds all previously known materials by a factor of 45. This is equivalent to nearly 800 years of previous experimental discoveries in the field.

Next, the scientists plan to test the viability of 381,000 of the most promising structures in the production of solar cells, superconductors and other technologies. In this way, DeepMind wants to demonstrate the potential of AI to accelerate scientific progress and create useful materials.

According to co-author Ekin Dogus Cubuk, it's hard to find any industry that wouldn't benefit from better materials. For example, the new compounds could help develop versatile layered substances or neuromorphic chips that mimic brain function.

The DeepMind team used machine learning to generate and then evaluate the stability of candidate structures. This allowed them to find many more compounds compared to the expensive trial and error method used previously.

Scientists from the University of California at Berkeley have already tried some of the predicted compounds in practice. They were able to experimentally synthesise 41 out of 58 of a given list of target materials using computation and an automated chemistry laboratory.

According to independent experts, this combined approach will dramatically accelerate the discovery of new promising materials for solving global problems. DeepMind's database of inorganic crystals should become an invaluable source of innovation in clean energy and environmental protection.

Source: Financial Times