There is the filter bubble paradox - the stuff that we need to learn is what we have never encountered before, but the ML filters can only learn from our past opinions.