3 principles for creating safer AI

TED main-stage talk proposing three principles for human-compatible AI: altruism toward human values, humility (uncertainty about those values), and learning values from human behavior.