Rather than generating potential outcomes based on historical data, deep reinforcement learning teaches AI agents and machines with the time-tested "carrot and stick" method.