Well medicine's come with error bars, typically in the form of a doctor who determines whether the promised benefit plus risk is worthwhile for you. Often this is in conjunction with a consultation with the patient. So the general point that we don't give control to other entities that can make mistakes, I thnk still holds.

But you're right in the general sense: there are other areas where we leave the interpretation of the AI (as it were) to experts and have no say in it ourselves. Road bridges would be the simplest example I can think of (given that the safety simulations are akin to a simple form of AI).