The best-practice seems to indicate assigning people randomly to groups so they can examine their own biases. This instance at Google was to assign people to their own stereotypes so they can wallow in it? Not the same thing, is it?
If you assign people randomly to labels that are supposed to be disjoint from their attributes (but you use the pejorative names of real attributes), there's a non-zero chance you'll randomly assign someone to the group named by the pejorative version of an attribute they have.
