The model is not the hard part
A melanoma AI that learned surgical ink, a cost label that missed need, an alert that is right 16% of the time and a sepsis model that drifted. Four questions anyone can ask about a medical AI tool, and why they are clinical questions.
