( Invited ) Searching for Ideal Electrolytes in the Molecular Universe via AI

In the past three decades since the invention of lithium-ion batteries, extensive efforts have been made to seek ideal electrolyte components that can resolve the many challenges faced by emerging battery chemistries. However, scientists have been confined to a rather narrow chemical space of 10 2 , most of which were either discoveries of serendipity, or from human-designs in empirical instead of rational manners. Leveraging the rapid development of artificial intelligence, we seek to map the entire molecular universe of small organic compounds computationally, and rapidly identify those most promising candidates for the next-generation batteries. In this talk I will summarize my recent adventure at SES AI, where we built immense databases of battery literature, chemical structures, and properties, mapped the molecular as well as ionic universes of cosmic scales, and trained an omniscient large-language-model to explore the large maps. Some of the AI-discovered/-designed molecules have already been synthesized, tested and outperformed human-designed electrolyte materials.