Well, two major issues: The first is technical: how does one guarantee that the algorithm is actually fair? And that of course hinges on how one defines 'fair'

Well, two major issues: The first is technical: how does one guarantee that the algorithm is actually fair? And that of course hinges on how one defines 'fair'. Many of the measures that appear objective aren't. There is a growing amount of evidence that the method in which a test is administered can have a statistically significant impact on how different races and sexes perform [1][2]. On top of that, there is evidence that similar biases creep into how teachers grade essays and even math problems [3]. The second is more social science-y: switching to an algorithm doesn't fix the problem that affirmative action attempts to solve, which is that even after correcting for any test bias certain minorities or subgroups will still perform statistically worse than others, not due to inherent differences in intelligence, but due to the cumulative effects of the inequalities of opportunities in society, eg lesser performing schools, less time/energy for parents to help with homework, less money for extracurricular activities, &c. &c. [4]. Yes, socioeconomic status accounts for a large amount of this, but it does not 100% account for it. There are still statistically significant amounts that are due to things like race, sex, or religion. [1]: http://mitadmissions.org/blogs/entry/picture-yourself-as-a-stereotypical-male http://mitadmissions.org/blogs/entry/picture-yourself-as-a-s... [2]: https://www.insidehighered.com/news/2010/06/21/sat https://www.insidehighered.com/news/2010/06/21/sat [3]: http://www.slate.com/blogs/xx_factor/2015/02/10/teacher_bias_in_math_new_study_finds_teachers_grade_boys_more_generously.html http://www.slate.com/blogs/xx_factor/2015/02/10/teacher_bias... [4]: http://thewireless.co.nz/articles/the-pencilsword-on-a-plate http://thewireless.co.nz/articles/the-pencilsword-on-a-plate