Assigning uncertainty scores to model predictions to identify outputs that may need human verification.