1. Rapid advancement in image classification AI models allow potential for potential identification of varying visible symptoms.
2. Access to a large data set would enable correlations with symptoms relative to the demographics of the patient profile.
3. Access to third-party labs (E.g. Quest Diagnostics) allows for blood/urine samples to be tested and analyzed using trained models.
Essentially this would be designed to conclusively limit people from seeing physicians if their condition was considered benign (E.g. common cold), and direct those with potentially serious conditions to the appropriate medical care.
In the US, I would be wary of lawsuits.
I can easily imagine a grieving family member suing for the wrongful death of a loved one because they got roboadvisory advice that turned out wrong.
There is a reason why doctors in the US have medical malpractice insurance and it is very expensive.