Probabilistic Consistency in Machine Learning and Its Connection to Uncertainty Quantification