Panel recording: The Role of Human Understanding in the Future of Scientific Discovery (Stanford HAI, May 2026)
Organiser: Stanford Institute for Human-Centered Artificial Intelligence (HAI), as the closing panel of the conference *AI + Science: Accelerating Discovery*. Recorded 5 May 2026 on the Stanford campus. Speaker(s): Risa Wechsler (Stanford, moderator), Surya Ganguli (Stanford, moderator), Angèle Christin (Stanford), James Evans (University of Chicago), James Zou (Stanford) and Eun-Ah Kim (Cornell University).
In this panel, researchers from physics, sociology, communication and biomedical data science debate what AI is doing to science as a practice. They look at the tension between automation and understanding, between efficiency and creativity, and between industrial and academic values. The findings they discuss include evidence that AI makes scientists more productive while pushing them toward a "big data" monoculture that shrinks the range of questions being asked. They also discuss how frontier LLMs still struggle with physics problems that need mathematical rigour and conceptual reasoning, and how AI agents can now design experiments, write papers and review one another's work, yet still cannot recognise what they don't know or ask good questions as they go.
For academics, the recording is useful because it moves past tool demos to the institutional questions: whether cost-cutting logic will start to replace PhD students, whether researchers are trading understanding for speed, and how to avoid skill atrophy while adopting automation. Hearing a sociologist of science like James Evans alongside James Zou, who leads the Stanford AI for Science Lab, makes it a balanced hour on where AI already helps research and where human judgement still matters.
Source: Stanford HAI on YouTube.
Comments