Rafael Gómez-Bombarelli, a newly tenured MIT professor, is leading a shift toward "general scientific intelligence." By combining traditional physics-based simulations with modern generative AI, his lab is bypassing the physical limitations of manual laboratory work. This approach allows researchers to run hundreds of thousands of digital experiments simultaneously—a process known as high-throughput simulation—to identify promising molecules for everything from advanced batteries to organic LEDs.

The field is currently at a critical second inflection point. While the first wave in 2015 focused on teaching computers to recognize chemical structures, the current era is merging multiple data types (multimodal AI). This means models can now reason about natural language research papers, 3D material structures, and complex synthesis recipes all at once. This capability allows for a more holistic approach to scientific "thinking" that mirrors human reasoning but operates at a much larger scale.

Gómez-Bombarelli’s latest venture, Lila Sciences, aims to build a scientific superintelligence platform. This initiative reflects a broader industry trend where the goal is to transform AI from a simple assistant into a core driver of scientific discovery. By creating "virtuous cycles" where physics simulations generate high-quality data to train better AI models, the time required to bring a new material from concept to reality is drastically reduced.

Rafael Gómez-Bombarelli, a newly tenured MIT professor, is leading a shift toward "general scientific intelligence." By combining traditional physics-based simulations with modern generative AI, his lab is bypassing the physical limitations of manual laboratory work. This approach allows researchers to run hundreds of thousands of digital experiments simultaneously—a process known as high-throughput simulation—to identify promising molecules for everything from advanced batteries to organic LEDs.

The field is currently at a critical second inflection point. While the first wave in 2015 focused on teaching computers to recognize chemical structures, the current era is merging multiple data types (multimodal AI). This means models can now reason about natural language research papers, 3D material structures, and complex synthesis recipes all at once. This capability allows for a more holistic approach to scientific "thinking" that mirrors human reasoning but operates at a much larger scale.

Gómez-Bombarelli’s latest venture, Lila Sciences, aims to build a scientific superintelligence platform. This initiative reflects a broader industry trend where the goal is to transform AI from a simple assistant into a core driver of scientific discovery. By creating "virtuous cycles" where physics simulations generate high-quality data to train better AI models, the time required to bring a new material from concept to reality is drastically reduced.