Non-negative weights are used in various algorithms and models to ensure that the contributions of different components are additive and do not detract from the overall value. This constraint is particularly important in fields like machine learning and optimization, where it helps maintain interpretability and stability of the model outputs.