The net result of a biased filter is that the victims of the bias must perform at a higher level to pass through the filter.
I.e., suppose the filter were set to exclude students of quality Q < 10, but were biased against black students to the tune of 2 points. Then we'd expect non-black students to have a quality distribution bounded below by 10, and black students to have a quality distribution bounded below by 12.
If we gave black students a bonus of 2 points, their quality distribution would again be bounded below by 10.
Of course, if our original filter was unbiased, then we expect the quality of black students to drop to 8. What do we see in the real world?
A point system itself is discriminatory against minorities.
By definition, a point system which ignores race cannot be discriminatory. It literally lacks the ability to discriminate.
admitStudent :: Student -> Boolean
admitStudent student = -- code does not refer to student.race
...
If you disagree, you should be able to provide a pure function for which this test fails:
admitStudent ( student { race = Asian}) == admitStudent (student {race = Black})
(In Haskell, x { a=b} takes a record x, makes a copy, and sets field a equal to b.)