NEJM AI paper: let AI list options, humans decide care

Shimizu is described as having a wide range of interests and a habit of quickly engaging with new fields. Before this paper, he appeared in a special edition of Nikkei Medical published on May 20, 2026, in a lead interview about AI. He has also been working with the ARISE Network, an international research team founded by members from Stanford and Harvard, on evaluating and verifying the safety and effectiveness of clinical AI.

The paper's argument rests on a division of labor: AI is good at divergent work like listing many differential diagnoses, while humans are better at weighing multiple circumstances to decide. Shimizu gives home care as the example — AI may offer a broad set of ideas, but the patient, family and clinicians make the call. He says the paper drew wide attention, and his impression is that few players combine a user perspective with technical expertise.

The conversation also touches how doctors are trained. Suzuki, a psychologist and psychosomatic physician, argues clinicians have long been trained to give fast answers where a correct answer exists, and are less used to thinking where none does. He said doctors often label something as a weakness after two failed attempts, and that "weakness" is better described as not yet knowing the right method. Whether hospitals adopt these ideas may hinge on how far they let AI generate options while keeping the decision with the care team.