Hacker News new | ask | show | jobs
by 50656E6973 2704 days ago
Statistical differences in outcome cannot be automatically attributed to discrimination, though they frequently are
2 comments

Yes they can, because that's what "discrimination" means: differences in outcome due to uncontrollable irrelevant factors like race. Discrimination is an effect, not an intent. Racist intent is called "prejudice". Discrimination can be caused by prejudice, or it can be caused by something else, like a poorly considered algorithm.
No, that's pseudo-social-science because it falsely assumes outcomes are solely determined by external variables and is completely ignorant of internal variables, such as differing cultural values.

Some cultures value family life over money, others value money over family life, for example.

I'm sorry, this reply doesn't make sense to me. What does "different cultures are different" have to do with whether computer systems can be discriminatory or not?
The FICO example from another thread shows a different outcome based on race without any data in the computer system about race.

If the computer system doesn't have access to race, like in the FICO example, then can it be discriminating based on race (racist)?

Yes, if the data contains proxies for race and/or unexamined racialized bias, then yes it can. Police presence has often been kept higher in zipcodes with mostly non-white residents. This results in higher arrest rates in an area because of increased eyes, not necessarily because of increased rates of law-breaking. If an algorithm were to be designed to determine where to allocate police officers using this data, it could contain no racial data, but it would dictate that more officers should be placed in the zipcode with more non-white residents, as those areas have more arrests. Perhaps you don't want to call this racist because you have a narrow definition of that word, but I don't know what other word to use.
My apologies for not linking to the other thread, but as far as I can tell, there are no proxies for race used to calculate FICO scores.
I agree, of course computers extend and amplify our prejudices.

But as the article states in it's conclusion:

>Just because something is expressed in numbers doesn’t make it right

Many statistical social differences are often automatically attributed to discrimination based solely on "numbers".

you should ignore that comment, it's an incredibly disingenuous claim intended to mask the racist sentiment that 'non-white people are over-represented in american prison populations because they don't value not being in prison as much as white people do'.
It is racist if the data is built on 100s of years of racism and the ML is trained on a dataset poisoned by racism.