Stuart Russell, a professor of Computer Science at the University of California at Berkeley, and Director of the Center for Human-Compatible AI explains how we've been thinking about AI incorrectly and how we can course correct to create safer AI that is less biased and successfully performs amidst uncertainty.