Climbing the Ladder of Interpretability with Counterfactual Concept Bottleneck Models