This begs the question: what data is going to be used as a training set? How do we make sure our black boxes don't become just echo chambers giving us answers w

This begs the question: what data is going to be used as a training set? How do we make sure our black boxes don't become just echo chambers giving us answers we already know? At the end of the day it is always: garbage in, garbage out.