> In a hypothetical scenario, if it turns out that immigrants from a certain country are in fact high-risk for certain kinds of fraud, and the AI automatically

> In a hypothetical scenario, if it turns out that immigrants from a certain country are in fact high-risk for certain kinds of fraud, and the AI automatically picks this up, is a human intervention countering its bias an erroneous one? > But assuming the objective truth does lead to some form of selection discrimination, how does one counter this? The basis of non-discrimination is to judge the individual. Even if 99% of people from some country are known to be fraudsters, you treat an individual from that country the same as an individual from another country. Anything else would be discrimination by definition. If somebody prefers to apply statistical results to the individual, there is no way to do that without discrimination because that is discrimination.