Google has actually released a paper on codifying what discrimination looks like for lending (https://arxiv.org/abs/1610.02413 + blog post: https://research.googleblog.com/2016/10/equality-of-opportun...). It can be trivially extended to discuss hiring.

Basically a fair hiring process as defined by the above would be:

P(hire | features) = P(hire | race=1, features) = P(hire | race=2, features)

This may sound a bit silly or unnecessary, but unless this formula is specifically enforced, if a race were correlated with expected profit of a hire, a hiring algorithm might accidentally infer race from the other features. You can see the paper for an extended discussion.

With that background out of the way, let's consider hiring based on quotas. Since hiring takes place continuously across time, in order to maintain a quota across time we would need to change P(hire | race=1) based on the current composition of the company. Therefore P(hire | race=1) != P(hire | race=2) at least some of the time. Therefore it's discriminatory according to the above definition.

Of course, you may not like the definition. I would be very interested to hear an alternate formalization of non-discrimination!

Unfortunately this isn't enough. Discriminatory rules such as "don't lend to any zip code that's over 75% black" will pass this test if you include this variable as a feature. The same logic applies to less obvious cases such as purchasing behavior.

A particularly hairy case is where you use variables as features which are themselves generated through other potentially-discriminatory processes. For example, including a "has committed a felony" feature seems like a no-brainer for a hiring or lending application, but now any racial or other discrimination present in the criminal justice system has now "infected" your hiring process, so that the outcomes of your hiring process are now racially biased even though your process itself was not.

But at some level, doesn't that just mean reality is biased, and since we don't live in utopia, trying to model one in the small is a sandcastle?

What do you mean by "reality"? Society is pervasively biased, yes, that's the point. That makes it difficult to fight against that bias, but it is necessary if you want to live in a more equal world.

> What do you mean by "reality"?

The thing scientists (especially physicists) are trying to model.

You're not the person I was asking, and I didn't ask for a definition.

In case you missed my point, we aren't talking about physics here, but about society, which is a reality we create.

Appealing to "reality" in a discussion about bias amounts to throwing up your hands and dismissing the problem as just "the way things are".

It's like if you described a complicated social problem you've observed and I shrugged and said "physics, eh?".

> Appealing to "reality" in a discussion about bias amounts to throwing up your hands and dismissing the problem as just "the way things are".

Yes, that was how I read GGP's comment. I think you are overstating the degree of control that we have over society.

For example, I have often heard the case that society influences young women into roles that eventually prevent them from becoming engineers. Gendered children's toys are often used as an example of this. However, gendered toy preference exists before socialization has occurred (it has been demonstrated in 3-8 month old infants[1]). The same gendered toy preference that exists in humans has also been demonstrated in vervet monkeys[2] and rhesus monkeys[3].

[1] http://link.springer.com/article/10.1007/s10508-008-9430-1

[2] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2643016/

[3] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2583786/

With those references now provided, I think it's safe to say gendered toy preference in humans is much bigger than simply a question of which toys we encourage children to buy -- there seems to be a considerable amount of genetics involved.

Hopefully now you are understanding what I meant when I defined reality for you. As I said I believe this is also what GGP was referring to.

So with that background out of the way:

> In case you missed my point, we aren't talking about physics here, but about society, which is a reality we create.

> It's like if you described a complicated social problem you've observed and I shrugged and said "physics, eh?".

A. How do you know it's a problem with the "reality we create" and not a problem with the reality we are stuck with? (Again, this is how I read GGP)

B. If we can trace the problem to something akin to male vervet monkeys preferring to play with Tonka trucks... what are we to do about it beyond ensuring that P(loan | race=1) = P(loan | race=2)?

I ask B because the idea of forcing my life choices onto someone else makes me feel ill. It reminds me of being forced to join the basketball team in high school, which I hated (though others seemed to love). The sick feeling compounds when I consider doing it purely on the basis of their race or gender in contexts where their race or gender is causal... and that's usually the course of action people on your side of this discussion recommend.

There is plenty of evidence that people have conscious and unconscious biases. There is ample evidence that the American criminal justice system, for example, is biased at several levels. I see where you're coming from: at any discussion of bias, you would rather blame it on genetics, which means we as a society are not responsible for doing anything about it.

I'm sure the people that have been forced into high-paying fields against their natural inclinations appreciate your feeling ill on their behalf, though. /s

> There is plenty of evidence that people have conscious and unconscious biases.

Agreed. I didn't say otherwise, I just also pointed out that there's evidence some of those biases are genetic.

> There is ample evidence that the American criminal justice system, for example, is biased at several levels.

I believe this should be fixed! I'm the person who linked the google formalism for modifying a model to be non-racist. This whole thread came from that.

> I see where you're coming from: at any discussion of bias, you would rather blame it on genetics, which means we as a society are not responsible for doing anything about it.

No. My position is that I want bankers to use a race-blinded models when deciding on loans. In general, where there is bias (especially where there is genetic/non-correctable bias), I believe the biased human should be replaced by a computer programmed to not have a bias.

I want this because I think the problem is real, I would like it solved, and I believe the other methods people propose (bias training, etc) are unlikely to work.

I also believe replacing as many humans as possible will not fix all the problems I imagine you are currently attributing to societal bias (ie I believe that some life decisions may well have a large genetic component). I do not view those "problems" as things to be solved, so much as deeply disturbing indications that free will does not exist. That in some sense you and I might be discussing optimization methods for player happiness in a zero player game. That in as much as happiness exists, I worry it is primarily found by embracing our genetic biases.

> I'm sure the people that have been forced into high-paying fields against their natural inclinations appreciate your feeling ill on their behalf, though. /s

People used to make arguments like this in defense of arranged marriages you know. Letting the heart decide isn't stupid.

While I agree that `P(loan | race=1, black_zip) = P(loan | race=2, black_zip) = 0` does indeed satisfy the condition, I think the kind of discrimination it creates is in some sense out of scope.

This is because if P(loan | black_zip) is profitable > 0, any bank primarily motivated by profit will approve such loans. [this is the statement I believe you meant]

If P(loan | black_zip) isn't profitable > 0 after correcting for race, this would mean is that the neighborhood itself signaled something about the person's likelihood of paying back the loan. Perhaps theft is very common, employment is very sparse or seasonal, vandalism/arson of property is common, etc... because we already corrected for race it amounts to saying "don't approve housing loans for neighborhoods where people regularly burn down houses" or similar. This doesn't seem discriminatory to me.

So while I do agree there are discriminatory issues not addressed by the formalism, your example seems to be handled by using the normal economic models for what decisions a bank should make given a model P(loan).

Of course if all banks do that, how is society to deal with this pit/failed neighborhood?

Through products and services not offered by banks?

Ensuring model safety is indeed a massive pain and is completely non-trivial.

I can think of many situation outside of employee distributions where quota would be discriminatory and wrong, so maybe a better question is what make hiring process so unique?

To give a few examples where quota would not work, lets talk about lotto winner. Why should winners not be on a quota so that winners are proportional to the population demographics? How is it fair that winner demographics don't match the population in race, gender and location?

Or what about having quota on roads in order to create equal road maintenance based on population. You drive through a unpopulated area and suddenly only one car per hour can drive through.

Or say that the police would use a quota system when deciding which crimes to investigate. Would it be fair to base how many murder investigations should be done by cities proportional population to the country? What about police budgets?

It would be nice if we could talk about what other systems that we want quota on and see what commonality there is between those and what distinguish them from the others.

Your examples are interesting:

Lottos: are known to give each ticket a fair chance based on a published set of rules. I think if that existed for hiring then many of the issues we see here would be gone. Unfortunately for many jobs it is difficult to give a test which is a fair sole determiner of suitability for work. And if I understand correctly also illegal in some jurisdictions, more or less because it can be abused.

Roads: maintenance does in fact vary by the wealth of the supporting area. Try driving from Nebraska to Kansas on US77 for a stark example. In addition, roads are networks and thus it makes sense to subsidize maintenance of connections to increase overall network value.

Police: some people do indeed advocate a form of this because they say police already relatively over-investigate crimes of minorities. That is, a white person has a higher chance to get away with any one given crime (for some categories)

While each ticket has a fair chance there isn't a 1:1 relationship between tickets and people. I think what's being asked is whether we should correct the outcomes of lotteries in response to opportunity inequality: force a 1:1 relationship.

While road funding may be distributed based on usage data after the fact or density projection before construction, should we throttle traffic flow so that roads wear evenly?

But should a murderer go free? And what of the scenarios where a given minority group collectively _does_ commit more crimes? It may in fact be nurture not nature (although it doesn't really matter) that yields such an outcome yet we would ignore such behavior because it feels overbearing to subject the group to "more" scrutiny relative to other groups that requires less.

I think the point is that it is confusing to entertain such diversity policy when applied outside of a socially hot topic such as the tech labor market. But also a challenge to find a scenario where equality of outcome is objectively desirable and draw parallels.

Sure people with the right skills are being discriminated against if they aren't in one of the favored groups. There is also no quota for white people or men. In my company these groups seem pretty underrepresented in some departments. Maybe they should do blind tests like government often does.

White people are underrepresented at most large tech companies[1]. So is almost every other race.

The exception is Asians. Asians are vastly overrepresented. I don't think this is a bad thing. It is meritocracy in action. But it violates the common notion that meritocracy and equality are the same thing.

[1] http://www.unz.com/gnxp/silicon-valley-has-an-asian-people-p...

You're saying that Asians are more meritorious of positions at large tech companies than people of other ethnicities. That's an extraordinary claim.

By the way, if Asians comprise about 30% of the workforce in the Valley, I wouldn't say they're over-represented. They're probably under-represented, in fact.

I think this is similar to the "equality" vs "equity" argument. Making sure a company is forced to hire a percentage of every single demographic is not fair even though it seems fair, and it also is discriminatory in the sense that instead of making it to where race / gender / etc don't factor into whether you get the job, it completely factors in.

"We can't hire any more black people, we have too many as is, if your white friend was to apply we would hire him in a heartbeat we have trouble getting enough of those"

Seems pretty discriminatory to me.

Or worse yet: "Sorry, a group of our employees recently went back to Mexico to start their own business there, we can only hire Hispanic for the foreseeable future."

Bonus question: should the HR department be allowed to ask for ethnicity ahead of time or would noneligible applicants have to sacrifice precious time in futile interviews?

Every quota is also a limit. If you say "there must be at least 30% women", it is equivalent to saying "there must be at most 70% men".

Rephrase every demand for quota as a limit, let's see how that would go down.

Example: The NBA institutes a rule requiring 50% of players to be non-Hispanic and white. This is well below 62%, the proportion of the population that is non-Hispanic and white. This rule discriminates against every black athlete who would have made the cut but didn't because a less-qualified white player was hired instead.

Discrimination means to select one thing over another because of some criterion. It is precisely the correct word for this. Racial discrimination means selecting one person over another using race as the criterion. Racial discrimination is always immoral and is, thankfully, illegal.