Is fairness in AI a practical possibility?