Rethinking Aleatoric and Epistemic Uncertainty