Google DeepMind executives outlined an approach to artificial general intelligence safety, warning of "severe harm" that can ...