Testing Relative Fairness in Human Decisions With Machine Learning