93. An AI risk score conflicts with your observations. How would you discuss it?
HardPatient safety
Case
Show model answer
I would examine the score's inputs and timing and communicate conflicting observations to the responsible clinician. A prediction is limited support, not a diagnosis or reason to disregard a patient's report or changing condition.
I would use approved review and escalation arrangements and document the discrepancy. Any treatment decision must fit my actual competence and authorized scope; disagreement alone does not prove model failure without reviewing inputs and validation.
What interviewers listen for
- Treat predictions as limited support
- Communicate conflicting observations
- Retain qualified clinical judgment