Typically, affirmative action takes the form of preferential treatment based on socially salient group membership such as race or sex. A common way of criticizing them is pointing out the mismatch between the aims they aspire to and the means they employ. If the aim of gender quotas is to address obstacles women typically suffer, it goes, why not directly consider the fact of candidates, whether female or male, engaging in caregiving, of their being “effeminate” and vulnerable to gender harassment, instead of (or in addition to) their sex? Considering all the relevant characteristics when implementing affirmative action has widely been assumed infeasible. However, the idea of “personalized law” may change this intellectual spectrum. Personalized law is a name characterizing the legal practices where a decision for an individual is made based on all the information available about this person relevant to the aims of the law, with the aid of big data and AI algorithms. In this presentation, I examine whether “personalized” affirmative action can be a morally better alternative to its conventional counterpart, that uses only race or sex. I classify different justifications for affirmative action into two categories: directly-benefitting-type and non-directly-benefitting type, and contemplate upon whether the personalized version can meaningfully be pursued under these two categories respectively. A tricky case arises when a number of minorities (irrationally) regard as better role models even those members of their group hostile to racial or gender justice than those members of majority promoting justice. I argue that this case shares similarity with the one where the stigma or anti-balkanization objection to affirmative action is made; in both cases, the prospect of irrational reactions is taken as a given. This insight leads us to a further question whether there is any moral asymmetry between the two.