No, affirmative action is this: when two candidates are approximately equally qualified but one was raised black in the inner city and the other white, then it's best practice to choose the black candidate.
"When comparing two candidates of otherwise approximately equal qualification, apply historical socio-economic advantages/disadvantages as modifiers to get their probable scores in the absence of those modifiers, then re-evaluate accordingly."
Compare it to testing two sets of children, one that was kept up all night and one that got to sleep after studying, and extrapolate writ-large.
In a similar vein, we can't make useful assessments about alcoholism susceptibility in Native American populations without first accounting for all of the historical socio-economic pressures that would likely drive a disproportionate amount of any given population to drink.
There's lots of literature backing this up - when you have a MORE diverse population in your company/college/whatever, it performs by almost any measure you can come up with, better as a whole.