Sources of unfairness in machine learning