Deep nets have layers too. We need to assign names to the nodes of the layers. That's the challenge of interpretability in neural nets.