Concept Bottleneck Model

A neural network architecture that first predicts human-understandable concepts from the input, then uses those concepts to make the final prediction, enabling interpretability.

An AI that first identifies understandable concepts (e.g., "has stripes," "has four legs") before making its final prediction ("it's a zebra"). This makes it easier to understand and correct.