Weak learners are models that perform slightly better than random guessing, often used in ensemble methods to build stronger predictive models. They are fundamental in algorithms like boosting, where multiple weak learners are combined to improve overall accuracy and robustness.