An AI tool flagged my father's scan result as low risk and the human radiologist disagreed, who do you trust?

My father had a lung scan six weeks ago. The hospital uses an AI-assisted radiology tool that provides an initial assessment before a radiologist reviews it. The AI flagged the result as low probability of concern. The radiologist looked at the same scan and recommended a follow-up biopsy. The biopsy came back with an early-stage finding that is now being treated.

I do not want to be alarmist about this because I know one case is not data. Maybe the AI was right about the probability and the radiologist was being cautious. Maybe the AI was wrong and the radiologist caught something it missed. I genuinely do not know.

What I want to understand is how these tools are supposed to work in practice. Are they meant to be a safety net below the radiologist or a pre-filter above them? What happens institutionally when the AI and the human disagree? And is there transparency about how often each one is wrong?