I'm reminded of the robot that goes wrong in Robocop. Except, with Reinforcement Learning, we'll have a robot that executes hundreds flawlessly until stopped. This is the kind of advancement that should be regulated, IMHO.