Algorithms that are 80% good aren’t scary.

Leaving aside the difference between sensitivity and specificity: if people know the machine is sometimes wrong, it’s not so bad.  Things get scary when the machine is 99% or 99.9% accurate, and you’re caught on the wrong end of a presumption.

I’ve often wondered, for example, how many people get picked up on warrants for someone else.  I think it’s important that we have systems for review of the totality of the circumstances by a human.