The more we discuss the dangers of training AI on only small sets of data and narrow ideas about identity, the better prepared we will be for the future.