Google DeepMind has been working with Harvard Medical School, Stanford Medicine, and many other medical centers to develop what they are calling AI co-clinician.
They are hoping to build a better and more sustainable model for healthcare and, also, to address the health worker shortage (both current and the World Health Organization’s prediction of a 10+ million health worker shortfall by 2030).
The model that they are proposing involves the patient (of course), the physician (and their team), and the AI co-clinician. The AI co-clinician’s role would be as a teammate to amplify how clinicians care for patients.
One of the many things that they are testing is the NOHARM framework (Numerous Options Harm Assessment for Risk in Medicine).
This tests the AI tool for incorrect information (errors of commission) and failing to surface critical information (errors of omission).
So far in head-to-head blind evaluations, physicians consistently preferred the AI co-clinician’s responses. The critical error rate has also been very low (zero actually) over 97 cases.
There is a lot more information about this, as well as cases that demonstrate the abilities of the AI co-clinician (no real patients, just research).
Here is the link to the DeepMind page about this - https://deepmind.google/blog/ai-co-clinician/