Confronting Style in AI for Science

AI developers, some scientists, and policy makers aspire to augment or replace human expertise and accelerate scientific progress with complex AI systems. Those aspirations lean heavily on the myth of a multi-disciplinary/omni-standpoint, or standpoint-free system---one that will perform scientific tasks better or faster than human experts. But standpoints run deep: AI systems rely on the outputs of scientific practices. The very scientific data with which systems are built, trained, and queried possess inextricable \textit{style} shaped by scientists' standpoints. Questions remain: which standpoints? How do we know? What does those standpoints mean for the future of AI for science?