Having worked tangentially on AI that approved payments for outpatient ops for one of the companies involved in handling German health system... The very basic

Having worked tangentially on AI that approved payments for outpatient ops for one of the companies involved in handling German health system... The very basic design goal was that the AI system could only approve, not reject. The goal was that obvious cases should be auto-approved, and anything where the AI returned below a certain confidence value was booted back to human to make a decision (same as pre-AI). So at least you get that.