"Double that then subtract the square" just doesn’t have the same ring, huh?
Squaring alone would work if you looked at the odds of both people being nice. If 90% of the population is nice, squaring it would give you 81% of pairings being nice, which matches the above example.
Just have to take the glass half full perspective.
1-(1-x)^2 to be pedantic :) (i.e. if 10% of the population is mean, 19% (=100%-(100%-10%)^2) of random pairings have at least one mean person, not 1% (=10% squared))
I don't know of a simple term to describe that relation though; is there one?