People are more likely to trust AI when the system makes its confidence and limitations understandable.