Towards the Right Kind of Fairness in AI