Comment by drawkward

>Much as they tried to pretend there was no conflict between diversity and quality. But you can't simultaneously optimize for two things that aren't identical. What diversity actually means, judging from the way the term is used, is proportional representation, and unless you're selecting a group whose purpose is to be representative, like poll respondents, optimizing for proportional representation has to come at the expense of quality. This is not because of anything about representation; it's the nature of optimization; optimizing for x has to come at the expense of y unless x and y are identical.

Eh, if x and y are correlated, you can optimize for x to a point and still get y gains.