AI will be racist for as long as the Internet is racist. were you to curate datasets, you'd just bias towards whoever curated it chose it to be. what's unbiased? nothing. no one.
When does it stop being referred to as racial differences and instead referred to as cultural differences(regardless of which racial sub-culture the current culture was borne out of)?
To clarify, "racial differences" generally meant "genetic differences". So a better way to put it is not race vs. culture, but genetics vs. culture, as ethnic groups (or more broadly speaking, races) can have both different genetics and different cultures. In other words, nature vs. nurture. You take people of different races, look at the differences between them and then study how much of the variations are due to genetic or environmental causes. "Racial differences" I guess could mean both genetics+culture.
When America catches up with the rest of the developed world and stops lumping individuals with complex multifaceted genetic backgrounds and mixtures into ridiculous, segregated, monolithic groups?
Who is defining race, racism, and whether treatment is detrimental?
If an AI sends more police to areas with more crime, and those areas are poorer and have more black residents than average then:
* is the data good or are we just sending more police where more police already got sent (and made arrests and thus appeared to have a higher crime rate)?
* is this a good thing because law and order are important or bad thing?
* is it discrimination or does the higher crime rate justify it?
* if the same system sends fewer police officers to an Asian or Latin or other neighbourhood that has low crime numbers, is that good or bad? Is it racist if white communities get fewer police officers and crime goes up?
* if it is discrimination then is it because of race or poverty?
People are very excited to should that AI (or machine learning or whatever) is racist. But no one wants to have the hard discussion about what any of that means in any practical case. As long as that's true were just going to rehearse the same tired arguments we've had for 50 years but with a computer making decisions instead of a person.
Personally I think we could make more progress by making local police accountable to local leaders. Then if communities want more/less action on a given type of crime, they can get it. And no one feels they're being oppressed from afar.
14 comments
[ 3.2 ms ] story [ 46.0 ms ] threadHow would you define race vs. culture?
If an AI sends more police to areas with more crime, and those areas are poorer and have more black residents than average then:
* is the data good or are we just sending more police where more police already got sent (and made arrests and thus appeared to have a higher crime rate)?
* is this a good thing because law and order are important or bad thing?
* is it discrimination or does the higher crime rate justify it?
* if the same system sends fewer police officers to an Asian or Latin or other neighbourhood that has low crime numbers, is that good or bad? Is it racist if white communities get fewer police officers and crime goes up?
* if it is discrimination then is it because of race or poverty?
People are very excited to should that AI (or machine learning or whatever) is racist. But no one wants to have the hard discussion about what any of that means in any practical case. As long as that's true were just going to rehearse the same tired arguments we've had for 50 years but with a computer making decisions instead of a person.
Personally I think we could make more progress by making local police accountable to local leaders. Then if communities want more/less action on a given type of crime, they can get it. And no one feels they're being oppressed from afar.
Here's a wayback machine link: https://web.archive.org/web/20210319191554/https://www.nytim...