Asking from a cultural standpoint, but curious about the technical approach to an “audit” when the same people who biased the training data are ostensibly influencing the audit. Basically, who decides what’s “just” and how could that be reflected in a change to the algorithm?
