The use of Artificial Intelligence (AI) to assign individuals into classes has stimulated concerns. One major issue is about the fairness of an automated classification algorithm. Consider a fictitious case that the Taitai Bank developed a predictive model for predicting future loan default. The higher score indicates that the individual has a higher probability of defaulting a loan. Taitai Bank has a policy of being fair to all its customers. Their main concern is that the risk score should be fair to customers in different occupations. Internal discussions reveal three definitions of fairness. To simplify our discussion, a "positive case" means that the individual will default in the future and a "negative case" means that the individual will not default in the future. The three different definitions of fairness are: (1) The risk score should be well-calibrated across different occupational groups: if the algorithm identifies a set of people as having a probability of z of constituting positive cases, then approximately a z fraction of this set should indeed be positive instances. This conditional should hold when applied separately in each group. (2) The average score received by people constituting positive instances should be the same in each occupational group. (3) The average score received by people constituting negative instances should be the same in each occupational group. Answer the following questions.