The problem isn't in using AI to find things that shouldn't be. The problem is in trusting the results. Current "AI" is really just a glorified autocomplete a

The problem isn't in using AI to find things that shouldn't be. The problem is in trusting the results. Current "AI" is really just a glorified autocomplete and thus it's going to return the common answer. Fine if it's a common situation, but it should be *expected* to get it wrong when it's a sufficiently rare situation. 1) Modify the system so the doctors can add notes as to why this isn't the usual situation. Don't make the person doing the approving figure it out. 2) Use AI to find suspect things but have a human review them--and any rejection must explicitly address anything in the notes about how it's unusual. 3) Rejecting a request is treated as a medical decision, requiring the person making it to be qualified to make it and subject to malpractice as if they had failed to recommend it to the patient when seeing them.