Current Events > Cornell creates hate speech detector. Decides it's wrong when minorities...

(+) Yeah! (+)

https://news.cornell.edu/stories/2019/08/study-finds-racial-bias-tweets-flagged-hate-speech

The tl;dr is their AI found specific minorities to be in some cases twice as likely to use hate speech, sexism, abusive language, etc on Twitter, and their conclusion was that the data must be biased because people marking a tweet as hateful didn't have enough training.

It's an interesting study with perhaps unexpected results, but also highlights the difficulty that AI systems are going to have if they don't come to "the right" conclusions.

To perform their analysis, they selected five datasets one of which Davidson helped develop at Cornell consisting of a combined 270,000 Twitter posts. All five had been annotated by humans to flag abusive language or hate speech.

For each dataset, the researchers trained a machine learning model to predict hateful or offensive speech.

In all five cases, the algorithms classified likely African American tweets as sexism, hate speech, harassment or abuse at much higher rates than those tweets believed to be written by whites in some cases, more than twice as frequently.

The researchers believe the disparity has two causes: an oversampling of African Americans tweets when databases are created; and inadequate training for the people annotating tweets for potential hateful content.

---

Or they need to feed the ai the correct data to get the correct bias they want
---

The AI needs to learn to be less racist.

---

Well its good that theyre doing their due diligence in making sure the data is kosher.

Will be interesting to see where the conversation goes with this.
---

Anyway this isn't anything special. The entire point of experiments is to repeat it until you get the results you want. Them failing a million times won't matter
---

RdVEHfJqAvUPIbk posted...

Anyway this isn't anything special. The entire point of experiments is to repeat it until you get the results you want. Them failing a million times won't matter

Something tells me that if the results showed that whites were twice as likely to use hateful speech, they wouldn't be double checking the data.

---

The Admiral posted...

RdVEHfJqAvUPIbk posted...

Anyway this isn't anything special. The entire point of experiments is to repeat it until you get the results you want. Them failing a million times won't matter

Something tells me that if the results showed that whites were twice as likely to use hateful speech, they wouldn't be double checking the data.

Shows how much you know about data science
---

Do you have guys have like zero experience with minorities or something?

---

Seems like alot of work just to statistically prove that most Twitter users are fucking ignorant assholes.
---

This is quite funny because their results are pretty much 100% truth, but then again telling the truth is the worst thing you can do these days.
---

you can't expect some ai to understand context when even the great minds of ce struggle with it
---

Sounds like the AI doesn't understand that it's only wrong when white people do it and that it doesn't count as hate speech when it's directed at white people. That's the problem with machines, they function objectively and aren't able to comprehend that orange man bad.
---

AlisLandale posted...

Well its good that theyre doing their due diligence in making sure the data is kosher.

Will be interesting to see where the conversation goes with this.

Five independent data sets used in Cross validation with a total of 270,000 data points is a pretty good experimental setup. Unless they go in and intentionally bias the data manually, I don't know how the results would change very much.

But I agree, unexpected results should always be double-checked. The important thing is whether they publish or announce the result of their secondary analysis, or whether they bury this project because they can't get the results they want.
---

Can't let the data hurt the narrative guys. Better fudge it.
---

Squall28 posted...

Can't let the data hurt the narrative guys. Better fudge it.

"Why algorithms are actually white supremacy" -Vox
---

wouldn't this just be because so many people talk shit about white people now?
---

DifferentialEquation posted...

Sounds like the AI doesn't understand that it's only wrong when white people do it and that it doesn't count as hate speech when it's directed at white people. That's the problem with machines, they function objectively and aren't able to comprehend that orange man bad.

thronedfire2 posted...

wouldn't this just be because so many people talk shit about white people now?

---

DoctorVader posted...

Do you have guys have like zero experience with minorities or something?

Careful, I just got modded for saying bout the same thing

thronedfire2 posted...

wouldn't this just be because so many people talk shit about white people now?

In my experience, which is a lot, it's hardly ever about white people.

I can't give examples or say anything else or I'll get modded again but DoctorVader seems to know what he's talking about

And it's not some awful bad thing. It's fucking words. Some people just express their shit in different ways.
---

gatorsPENSbucs posted...

DoctorVader posted...

Do you have guys have like zero experience with minorities or something?

Careful, I just got modded for saying bout the same thing

thronedfire2 posted...

wouldn't this just be because so many people talk shit about white people now?

In my experience, which is a lot, it's hardly ever about white people.

I can't give examples or say anything else or I'll get modded again but DoctorVader seems to know what he's talking about

And it's not some awful bad thing. It's fucking words. Some people just express their shit in different ways.

I legit typed up a post about the racism I experienced working at a Chinese restaurant, but chose to delete it because I'm not sure if the narrative police will warn me for it.
---

gatorsPENSbucs posted...

In my experience, which is a lot, it's hardly ever about white people.

I can't give examples or say anything else or I'll get modded again but DoctorVader seems to know what he's talking about

And it's not some awful bad thing. It's fucking words. Some people just express their shit in different ways.

Yup

---

DifferentialEquation posted...

"Why algorithms are actually white supremacy" -Vox

That was an actual article lmao
---

thronedfire2 posted...

wouldn't this just be because so many people talk shit about white people now?

Minorities can't be racist. Only white people are racist. It is known.

*Nods sagely*

/s

---

These results aren't at all surprising.
---

Surely minoritys cant be racist towards whites! Rolls eyes.

---

I don't know why people are surprised when they learn that all humans have the potential for racism. It could very well be the twice as likely statistic is incorrect, but I'm sure they'd fine that all humans have about the same level of innate potential for prejudice.

This magnificent new invention will end racism forever!

/machine proceeds to do nothing about centuries of systemic racism

I don't know what happened.
---

The AI needs to learn its not racist if black people use the n-word.
---

AmericaTheBrave posted...

The AI needs to learn its not racist if black people use the n-word.

Yeah I wonder if this is a huge portion of it as well

---

Ok that is stupid

Talk about selective bias
---

Maybe they should put in a "is the person white?" filter, and just toss out all the ones that goes in the "No" pile.
That's sure to give the desired effect.
---

ROBANN_88 posted...

Maybe they should put in a "is the person white?" filter, and just toss out all the ones that goes in the "No" pile.
That's sure to give the desired effect.

Sadly most AI algorithms like in photo recognition have explicit racial filters
---

This is a prime example of what's fucked in this country and I legit had a talk with a dude about this the other day.

You all see black and white. It's a lot more than that. A whole lot more. I wish I could give examples but I guarantee you the vast majority of these words and phrases probably don't have the slightest thing to do with white people.

Yet, everyone immediately goes to that. Go hang out with people, experience shit.

s0nicfan posted...

sexism, abusive language, etc

No one sees that
---

Shablagoo posted...

Yeah I wonder if this is a huge portion of it as well

It probably would only be that way if the people who flagged the data set flagged black people using the n-word as racist, or the AI is not robust enough to make simple deliniations in what constitutes hate speech between different races. Both possible, but would represent a pretty major failing of the people conducting the research.

---

Shablagoo posted...

Yeah I wonder if this is a huge portion of it as well

Could be, but the AI showed much higher rates for sexism, harassment, and abuse as well.

---

gatorsPENSbucs posted...

You all see black and white. It's a lot more than that. A whole lot more. I wish I could give examples but I guarantee you the vast majority of these words and phrases probably don't have the slightest thing to do with white people.

Yeah it bugs me that the immediate thought of everyone is that minority is just black people, and the racism is only towards white people.
---

ROBANN_88 posted...

Maybe they should put in a "is the person white?" filter, and just toss out all the ones that goes in the "No" pile.
That's sure to give the desired effect.

It's a lot more complex than that. You have to determine where the person falls on the oppression hierarchy as determined by the left, as well as the target of their of their insults. For example, a non-white Hispanic person using racial slurs against white a person would not be racist, where as if they used slurs towards a black person it would be racist. On the other hand, a black person can use slurs against most groups without being a racist. But, sometimes, it can be a toss up.

For example, what if a black man uses slurs against a Native American? That's hard to say, you'd have to look at other traits like gender. A black man using slurs towards a Native American woman is definitely racist. But if it's towards another man, that's not so easy to settle. You'd really need to know a lot of other traits so you can accurately calculate an oppression index for each person, only then can you really determine if it's racist when that person uses a racial slur. And even then, you're going to have ties. In that case you would probably just want to blame it on white people in general (white males, preferably).
---

If you treat someone different based on race then its racism. Its that simple.

---

gatorsPENSbucs posted...

And it's not some awful bad thing. It's fucking words. Some people just express their shit in different ways.

that's basically the argument that the authors in the paper make fwiw


---

There's actually a logical explanation too: white people tend to keep their racism private or specifically among other whites, in private communities or anonymously if publicly.

Black folks don't really censor themselves online at all.

The levels of racism are likely very similar, with one or the other taking the lead depending on so many different criteria.
---

RdVEHfJqAvUPIbk posted...

Or they need to feed the ai the correct data to get the correct bias they want

Works for the climate change whackos
---

Squall28 posted...

gatorsPENSbucs posted...

You all see black and white. It's a lot more than that. A whole lot more. I wish I could give examples but I guarantee you the vast majority of these words and phrases probably don't have the slightest thing to do with white people.

Yeah it bugs me that the immediate thought of everyone is that minority is just black people, and the racism is only towards white people.

Can you blame them? America's biggest racial conflict is white vs. black* so obviously many discussions about racism is in that context.

*You can make the argument that it's white vs. native but natives are like 1% of the population, so there's not much attention there.
---

I wonder if this has to do with the greater amount of the religious within the black community, religion and backwards ideas tend to go hand in hand.
---

sevihaimerej posted...

I wonder if this has to do with the greater amount of the religious within the black community, religion and backwards ideas tend to go hand in hand.

Considering the Baptist church is one of the most anti-gay churches in the country. (Baptists aren't mostly black or anything, but black folks tend to mostly be Baptist).

I grew up Baptist, it's bad. Real bad. My youth group leader, while staring directly at teen girls, said with full sincerity that girls who get abortions go to hell no matter what and deserve to be crucified.
---

White people get more scrutinized for such speech.

If you are less scrutinized for the speech, yoi are less likely to care to not do it.

Voila

Jagr_68 posted...

Seems like alot of work just to statistically prove that most Twitter users are fucking ignorant assholes.

A lot* Not sure which is worse Twitter or Youtube comments.

---

The Admiral posted...

Shablagoo posted...

Yeah I wonder if this is a huge portion of it as well

Could be, but the AI showed much higher rates for sexism, harassment, and abuse as well.

Interesting, I guess wed have to see the structure of the algorithm to know for sure. Like, could someone making a post Look what this dumbass bitch did!!! along with a video of some lady doing something stupid be tagged for those things?
---

Interesting what happens when you hold everybody up to equal standards.

Annihilated posted...

Interesting what happens when you hold everybody up to equal standards.

part of the argument they're making (not all of it, but one element here) is that the same word used by different people isn't the same. like the n-word is not slanderous if a black person says it but it is if a white person does

---

Balrog0 posted...

Annihilated posted...

Interesting what happens when you hold everybody up to equal standards.

part of the argument they're making (not all of it, but one element here) is that the same word used by different people isn't the same. like the n-word is not slanderous if a black person says it but it is if a white person does

Have you spent any time with black people? It quite often IS slanderous. Also I see no reason why it can't be used both ways by both people.