The need for qualitative reasoning in fuzzy modeling: robustness and interpretability issues