Apple devices’ new collision feature triggering false alarms for emergency help

So this kind of analysis is always troublesome.

It isn't very hard to find instances of false positives when the sample size is so large. For those false positives, the cost and effort is no doubt a waste and potentially could distract from a real instance.

But what is the big picture here and the overall cost benefit? Can we find true positives where the crash victim is better off because of the alert? Are there more or less of these than the false positives? Have any of the true positives saved a life? How do you measure that? The algorithm no doubt does not have six sigma accuracy, but it that the bar that must be set before we save some lives?