A theory of discrimination [5]: Rationality and unconscious bias

My theory so far has been largely normative. The thought is that it depicts which cases of discrimination would be permitted in an idealised society and provides a general algorithm (Base Omega) approach to attain said ideals. This idealisation has certain weaknesses however; it depicts each individual as acting perfectly in accordance to the theory. It happens that in reality, people lack the same degree of perfection. This is of course common with normative theories, the idea being that we should attempt to act in the idealised way. For example in epistemology we say that rational agents will not believe in contradictions and their credence functions ought to be probabilistic. The issue is that our theory of discrimination has considerable moral and political impacts. In this sense we have additional reasons against complete abstraction. We must recognise that agents are not ideal and in fact I suggest this should be incorporated into the theory.

According to our theory, individual agents ought not discriminate based on protected characteristics unless there is a significant reason / it is mandated in relation to the context. (A clarification point: base Omega says innocent until proven guilty, unless in the case of protected characteristics where we apply base Delta concepts of guilty until innocent) The scope of our theory is meant mostly to cover action, additional moral and epistemic argument would need to be established for the theory to have impositions on the thoughts and beliefs of ideal agents. That is, the theory tells us how to act, it doesn't in itself tell us that we should actually believe it. Thus it allows for agents who for example hate German people but do not act upon their beliefs. In practice however, this allowance is only available to ideal and rational agents. This is because only they are able to fully separate belief from action, metaphysically speaking. Ordinary agents are imperfect in seperating their beliefs and their actions. Their actions inevitably are based around their beliefs. Furthermore, sometimes this is unconscious.

Unconscious bias involves making quick judgements and assessments on entities without our conscious realisation. These biases can be influenced by our background, the cultural context as well as our personal experiences. Notably, these biases can influence our behaviour; we may for example be more intimidated at night by a black man than white woman. I leave it neutral at the moment as to whether this is justified. The main thing is that idealised agents would not be susceptible to these biases. They are consciously aware of all of their actions and can act perfectly in accordance to our normative theory. However since humans are non ideal, unconscious bias is a topic that should be accounted for.

One way to do this is by separating the theory into conscious and unconscious cases. In some sense this is quite intuitive; there seems to be thought that conscious discrimination is (if impermissible) worse than unconscious discrimination. The idea presumably is that conscious discrimination bridges one's beliefs and one's actions. If the individual consciously discriminates against black people, then this suggests that the individual believes / is in favour / supports some kind of discriminatory practices. On the other hand, an individual can be unconsciously discriminating in action whilst having non discriminative beliefs. For example some data suggests that science faculties may be more inclined to hire male candidates over female candidates. Our theory of course states that this is impermissible, and it is likely many recruiters in said faculties agree with the theory. Regardless, they may still be discriminative due to unconscious biases. It seems that faculties that behave as such unconsciously are morally better (although still wrong) than those that consciously do so, for example by actively giving credit on the basis of gender.

In terms of action guidance, it seems considerably easier to deal with conscious discrimination. The thought here is that consciousness discrimination is at least easy to notice. An entity is explicit in their discrimination, which if impermissible should be dealt with. Unconscious biases and the resulting discrimination are harder since it is not so explicit. Standard justifications of normative theories tell our imperfect agents to mimic our idealised agents. Thus we want to say that individuals should behave away from their unconscious biases. This unfortunately is not particularly helpful; it is not useful to tell me not to be fooled by things I can't actively control. In terms of guidance however we can still appeal to things such as education or ways for us to recognise our unconscious biases. The discussion here however ceases to be normative. The best way to deal with unconscious bias will thus seem to be a matter of empirical investigation.

A theory of discrimination [4]: Symmetry arguments

In this post I'm going to discuss symmetry arguments. These arguments typically arise as objections to discrimination against 'privileged groups'. For example say we didn't allow white men into a film showing. Someone might object that this is unacceptable because if you replaced white men with black women it would be viewed as utterly immoral. These arguments are typically based around some idea of positive discrimination. I'm going to argue that they are interesting and have some force but unfortunately they cannot apply. The symmetry doesn't work here.

So far the theory I have developed has the following kind of structure:

1) Discrimination is not inherently bad: Base Omega theory says that discrimination is permissible unless there is significant reason of the contrary.

2) Whether an instance of discrimination is permissible depends on both the context and the factor being discriminated against.

3) Principle of distributional equality: Across protected characteristics such as race, gender, sexuality all groups have the same talent distribution.

4) One can permissibly discriminate based on a factor if it is significantly relevant for the fundamental goal of the context.

Conclusion:

5) By 3 and 4, in talent related contexts such as jobs, education etc, it is is impermissible to discriminate on the basis of protected characteristics.

In particular, this means that it is impermissible to discriminate on the base of for example, race, in the context of education. This seems rather plausible, it would seem immoral for example to prevent black students from entering higher education on the basis of their race. What however about the contrary? What about 'positive discrimination'?

Something my theory has so far maintained is a kind of neutrality. Whilst it discusses discrimination on the basis of race, it takes individual races as abstract entities that can be plugged into a formula. The thought here is that discrimination on the basis of race is wrong in itself, and every instance of it regardless of which race it is targeted towards is morally impermissible. On some level this is very much intuitive. On another level, it feels like it is too abstract that it misses elements of reality. There is the thought perhaps that discrimination against certain races may be morally worse than discrimination against other races. Whilst I find this discussion rather uncomfortable I think it is important to address.

First things first: in an ideal or equilibrium state, it seems obvious that discrimination against all races is of equal moral weight. In an ideal state we would have no notions of racism in the sense of power dynamics, or history. There would be no impact of things like slavery, war and politics in the case of discrimination. These ideal states allow for perfect execution of base omega theory, we just systematically follow our method to discuss which cases of discrimination are good or bad. It seems harder to apply to this to our current reality. Whilst our base omega theory has some good applications, it is in some sense too discrete: whilst it tells us that some things may be permissible or not, it doesn't tell us the scope. It is not clear that our action guidance here is deontic: they may just provide moral reasons of different strength. There is the thought that whilst discriminating against white and black people is both wrong,  perhaps it is worse to do so against black people in our current society. 

Let's lay neutral to that conclusion for now. In our current reality, there are unfortunately things such as history that have impacts, conscious or subconscious on how we treat individuals. Whilst my theory largely covers explicit discrimination; we also need to recognise implicit cases where the subjects are unaware of some things they may do. Now one might consider the example of a black student who is unable to get a certain grade to get into a university. Perhaps being black is unfortunately correlated with a number of things such as going to a state school, being more likely to be part of a gang or so on. It seems inevitable that even if we followed our theory to not discriminate on the basis of race, there would still be very unevent treatment, despite our principal of distributional equality. Recall that the principal is meant to capture the notion that without discrimination, we expect equal distributions since all races are meant to be equal. In our real case we can clearly see a problem with our principal: its aims and the outcomes do not align.

This opens the door to positive discrimination arguments. For example universities may offer students from state schools lower offers on the grounds that they have achieved less support. This is of course an instance of discrimination, and perhaps one that my system would deem immoral. Yet this idea does not seem so fundamentally problematic. It perhaps sounds worse if we give black students lower offers, but upon statistical justification and restatement of our theory, even if can seem not so awful after all.

This sort of argument can be repeated in many cases, leading to arguments that for example it is permissible to discriminate against men or be racist against white people. Permissible not in the strict sense of course; but that relatively speaking it is less bad, and that the action may overall be morally okay if it helps achieve a good outcome. For example the labour party enforced 50/50 women politicians. This is of course discriminative, but they are motivated by claims that women are otherwise discriminated against / correlated discrimination with factors such as social pressures.

The important thing to note is that the base omega theory cannot be applied directly to imperfect or current situations. It serves really as an ideal. In our current situation, there is a lack of symmetry. In particular, it means that discrimination against for example races A and B are not morally equivalent, and perhaps additional analysis is required. This is not to say that I condone for example positive discrimination, I very much dislike it and will address this in a later post. I simply want to rebutt assymetry arguments. In particular there may exist arguments that propose discrimination against for example white men can often meet backlash claims of sexism or racism. They are reminded that it would horrific if you for example replaced white men with black women. Whilst I dislike both cases, this argument fails because the substitution is unjustified. Unfortunately in our current society, in most contexts black women and white men do not occupy the same kind of status that would justify these moral claims about permissibility of discrimination. Whilst they are both persons which means their interests should be taken into account, the factor of history and the subconscious impacts it has makes direct substitution unjust. Many theorists would of course appeal to notions of power dynamics, cases of subjugation between races, genders and so on. I don't necessarily go down that route, but I can see the argument. I certainly accept that these symmetry arguments are faulty in some sense.