How AI Learning Will Look Like Human Learning Within the Lifetime

The author says there are basically two ways in which humans have values. One is they have priors over their generative model, and the other is they have like discriminative preferences over plans. People thinking about whether how to get someone's mother out of a burning building are going to be at first significantly biased towards action sequences that respect human values. So all of these of course have analogies in what we can do in deep learning.