The theory of close calls suggests that humans often miscategorize events where something almost went wrong, but didn't. This concept can inform the development of robots and machine learning.