Toward a taxonomy of trust for probabilistic machine learnin