← All posts
This post is written in English.
Google has published a post on a clinical study of AMIE, its research AI for medical conversations. Participants could chat with AMIE before their next doctor's appointment. It took their history and sent a summary to their physician.
What Google reports
A trained physician oversaw every conversation, and "they never had to interrupt the conversation or intervene." Doctors said it helps them "shift from gathering information to collaborating more with patients."
Google plans a larger study across the US, with more patients and controlled comparisons.
What's missing from the post
The post doesn't give the number of patients, name the clinical sites, or report numerical outcomes. It's a company article, and the underlying research is linked but not described in detail, so I haven't assessed the published paper itself.
A supervised study with no interventions needed is encouraging. But if a physician is watching every chat, it can't tell us how the system behaves without one. That's the setting people are really curious about.
My read
History-taking is a sensible place to start. It's time-consuming, it's lower risk than diagnosis, and a doctor still sees the patient afterwards. It also makes a good test of the idea that AI should give clinicians more time with people, not replace them.
The planned controlled study is the part to wait for. "Doctors liked it" and "patients got better care" are very different claims, and only the second one justifies wide deployment.
