1 paper
Tim Franzmeyer, Edith Elkind, Philip Torr +2
AI agents are commonly trained with large datasets of demonstrations of human behavior. However, not all behaviors are equally safe or desirable. Desired characteristics for an AI…