What I think is the most interesting thing in this area, though, is that there may be times when it is possible to show at the level of a group that a given AI model creates some sort of harm (e.g., an unlawful bias), but difficult to show that at the individual level. 4/
