Fairness and Randomness in Machine Learning: Statistical Independence and Relativization