Computational methods have become a cornerstone of scientific progress, with nearly every scientific discipline relying on computers for data collection, data analysis, and simulation. As a result, significant opportunities exist to accelerate scientific discovery by accelerating the development and execution of scientific software.
The first priority research direction of the U.S. Department of Energy’s (DOE) report on extreme heterogeneity and the DOE community’s AI for Science report highlight the need for research in AI-driven methods for creating scientific software. This workshop expanded on these reports by exploring how program synthesis, an approach to automatically generating software programs based on some user intent—along with other high-level, AI-integrated programming methods—can be applied to scientific applications in order to accelerate scientific discovery.
The report reviews the relevant areas of program synthesis work, discusses successes to date, and outlines opportunities for future work.