A training approach where humans iteratively provide feedback or examples to guide model improvement based on current model behavior.