Biomedicine / Report
AMIE reaches a real clinic. What does “90%” measure?
Pichai said early on October 9 Beijing time that Google and BIDMC’s AMIE study had appeared in The Lancet. Patients chat with this research system before appointments. We read the March 9 preprint, not the journal revision; the findings below refer to that earlier version.
In a single-center, single-arm feasibility study, 100 adults used computer text chats up to five days before urgent primary-care visits in Boston. Patients needing emergency care, pregnant patients and mental-health chief complaints were excluded. This was not autonomous emergency medicine.
The final diagnosis, established through eight-week chart follow-up, appeared somewhere in AMIE’s candidate list in 90% of cases. That is not first-choice accuracy. Overall diagnostic-list and management-plan quality did not differ significantly from primary-care doctors; practicality and cost-effectiveness favored doctors.
Physicians monitored every conversation live, with zero safety stops observed. A small, supervised, single-site study cannot establish unsupervised safety. Alphabet funded the research. The journal revision, wider populations and health outcomes remain to be checked.
Sources and further reading
Edited report
Selected through a verified followed account: @sundarpichai.