by Myra Malhotra (’26) | February 2, 2024
We often hear that artificial intelligence (AI) is going to be faster and bigger than the Industrial Revolution. While we consume and interact with some AI technology on our devices in everyday life, many features that have been made using AI remain out of sight. This January, a research team at Microsoft developed a variant of a lithium ion battery with improvements that would have theoretically taken decades to complete without the help of AI. Their research demonstrates how AI is rapidly driving technological advancements.
Lithium ion batteries are a staple of everyday life, but acquiring lithium is both expensive and damaging to the environment. In batteries, electrically charged atoms, or ions, are transferred through electrodes via a material called an electrolyte. In lithium ion batteries, the electrolyte is a hazardous liquid that can leak and cause fires.
These researchers specifically aimed to build a lithium battery with a solid electrolyte. A list of candidate materials was built using a mix-and-match scheme where elements in crystal structures of known electrolyte materials were substituted with different elements. However, with 32 million candidates, the list would have taken decades to evaluate using traditional calculations. Instead, by using machine learning models to learn patterns of existing materials and their suitability, the team was able to complete the calculations in a mere eighty hours.
First, the researchers used AI to narrow down the 32 million candidates to 600,000 by checking whether a material was likely to exist and be accessible in a real world setting. Then, another iteration of AI analyses was applied to determine whether a candidate had the electrical and chemical properties necessary for batteries. The remaining candidates were subsequently evaluated using traditional computational methods to ensure suitability. After filtering out rare, toxic, and expensive candidates the researchers were left with twenty-three options. Five of these candidates were already known, so the researchers picked a candidate material from the remaining new candidates that had good stability, conductivity, and could also be synthesized in a lab. Over the course of six months, the researchers created a working prototype using the new material.
The new electrolyte used in the prototype was very similar to an existing material containing lithium, yttrium, and chlorine. However, it swapped some lithium for sodium, making it more cost-effective. Experts in material technology noted that they would not have mixed lithium and sodium together as they would have expected the combination to perform worse. So, in this scenario, the collaboration between scientists and artificial intelligence led to a solution that may not have been reached by human experts alone.
The swift development of a solid electrolyte lithium battery, facilitated by AI, exemplifies the remarkable potential of AI in driving rapid advancements. This collaborative effort not only shortened the evaluative process from decades to hours, but also resulted in a cost-effective solution that may not have been previously discovered. As science and AI continue to merge, a future where AI consistently outpaces human expertise becomes increasingly tangible, promising groundbreaking discoveries and transformative advancements.
