Doshi-Velez: Many of the most important improvements in the final 5 years are actually how effectively AIs now conduct in significant data regimes on precise forms of duties.  We have seen [DeepMind’s] AlphaZero turn out to be the most beneficial Go participant fully via self-Perform, and