UC Berkeley

Critical research advancements require usable, reliable software products—including AI tooling—for high performance data and computing work. To improve AI systems we must understand scientists’ and engineers’ needs and define design best practices for building interfaces that facilitate collaboration between scientists and AI tools across ecosystems. Ensuring the long-term usability and utility of scientific software products and AI systems for communities protects critical investments, supports high quality research, and reduces end user challenges. This talk will present findings from our semi-structured interview study investigating how scientists and engineers use AI tools for their research work. We recruited from universities and national labs, seeking perspectives from multiple disciplines. Looking beyond common use of AI tools for coding tasks, we sought to understand the varied ways AI is or could be adopted in scientific settings. Our interviews investigated how users determine the utility of AI tools, the concerns they have about adoption, and expectations for agentic tooling in research infrastructures. In this talk, we discuss opportunities for AI tooling in science and engineering beyond coding support, providing a lens into the development work the US-RSE community may take on in the near future and the socio-technical circumstances driving it. Our results inform development priorities, strategies for issue mitigation, and long-term stewardship in scientific facilities and work. We anticipate our insights helping to improve the usability and stewardship of scientific software, especially for high performance data and computing work.