Autonomous Scientific Discovery
Automating and steering scientific experiments, observational instruments, and field sites with computing facilities using advanced algorithms and simulations. Continuously analyzing, archiving, and curating incoming data.
The generation of scientific hypotheses and data has traditionally been dependent on laborious human effort. However, the ongoing revolution in artificial intelligence (AI) and robotics, coupled with advanced networking and computing hardware, brings with it the promise of automated scientific discovery. Advanced algorithms, simulations coupled with experiments, and next-generation networking and computing infrastructure are required for the automation of scientific discovery across DOE mission space. At Berkeley Lab, our work spans AI-based algorithms for real-time steering of beamlines at DOE light sources, specification of targets for telescopes, and design of molecules with desired properties. Additionally, we are building data-management workflows that enable seamless integration and access throughout the data life cycle.
It is becoming critical to connect experimental infrastructure to high-performance computing. The superfacility concept is a blueprint for seamlessly integrating experimental, computational, and networking resources to support reproducible science across the DOE and beyond. This integrated infrastructure provides a foundation for considering self-driving and self-guiding facilities and facilities for autonomous scientific discovery. Berkeley Lab’s work in this area is designed to define an automated architectural model to support streaming data from experimental facilities, improved resilience, and integrated tools for sharing, searching, and analyzing data for more productive, reproducible science.
Together, these advances will enable the automation of diverse DOE science experiments and increase the pace and efficiency of scientific discovery.