I agree! I think trying to apply a high level value will result in more discrimination, however. i.e, if we see X group of society under/overrepresented by Y%, we do not have enough understanding to draw any meaningful conclusions from this. Large enough discrepancies should be investigated as possibly a canary for discrimination, but in some cases there may not be any (NFL lineman will likely always be men).

I think the correct answer is to evaluate people as individuals on their merits. However, this is expensive (in many ways) and still flawed. My view is that we are much close to coming up with a "fair interview" than we are with understanding the massive amounts of complexity explaining discrepancies in human populations.

> A simple first order approximation is "no preference".

That's a reasonable first-order approximation, but it seems not to be reflected in the data, which show that as societies become more equal, gender disparities in field choice increase.

See: "Countries with greater gender equality have a lower percentage of female STEM graduates" (https://www.sciencedaily.com/releases/2018/02/180214150132.h...)