

Microsoft Researchers Say NLP Bias Studies Must Consider Role of Social Hierarchies Like Racism (venturebeat.com) 150
As the recently released GPT-3 and several recent studies demonstrate, racial bias, as well as bias based on gender, occupation, and religion, can be found in popular NLP language models. But a team of AI researchers wants the NLP bias research community to more closely examine and explore relationships between language, power, and social hierarchies like racism in their work. That's one of three major recommendations for NLP bias researchers a recent study makes. From a report: Published last week, the work, which includes analysis of 146 NLP bias research papers, also concludes that the research field generally lacks clear descriptions of bias and fails to explain how, why, and to whom that bias is harmful. "Although these papers have laid vital groundwork by illustrating some of the ways that NLP systems can be harmful, the majority of them fail to engage critically with what constitutes 'bias' in the first place," the paper reads. "We argue that such work should examine the relationships between language and social hierarchies; we call on researchers and practitioners conducting such work to articulate their conceptualizations of 'bias' in order to enable conversations about what kinds of system behaviors are harmful, in what ways, to whom, and why; and we recommend deeper engagements between technologists and communities affected by NLP systems."
Shitty smelly parasites H1B hindu-chimps (Score:1, Informative)
Re: (Score:3)
Damn straight, if everyone can come here and there's more competition for limited resources, we might end up with capitalism!
How do you show you eliminated bias? (Score:4, Insightful)
Is it only because a person believes the results are free of bias? Must results always be exactly equal to distribution of race, creed, gender? Such that immigrants of color would be penalized because they tend to outperform native-born people of color on tests?
This sounds like some folks are afraid that using a completely dispassionate method of reaching conclusions (smart systems/AI) may, in fact, show that there ARE differences amongst groups of individuals, and that would bring the walls of the SJW castle crashing down...
Re:How do you show you eliminated bias? (Score:5, Informative)
Must results always be exactly equal to distribution of race, creed, gender?
It depends on what the system is used for.
The system that triggered this debate was a ML system that looked at historical data, compared it to a current defendant's case, and then made a recommendation whether the defendant should be released on bail or held in pre-trial confinement.
It turned out to be very "racist", denying bail far more often to black defendants.
But it wasn't "wrong" because race is actually very strongly correlated with jumping bail. Blacks are significantly more likely to fail to show up for trial.
But as a society, we have decided that is unacceptable. Everyone should be treated as an individual, not as a "black guy" or "white guy".
So they tried and failed to fix the system by tweaking the inputs. If race was excluded, the system was still nearly as racist because it found correlations between the defendant's zip code, type of offense (blacks use crack cocaine, whites prefer powder cocaine), and even the defendant's name (Deshaun is more likely to be black than Travis). No matter what they did, the outputs were still "racist".
So if the result is socially unacceptable, and tweaking the inputs doesn't work, the obvious solution is to tweak the outputs.
Just let the system do it's work and then apply a "race filter" on the output to select the same percentage of whites and blacks for bail. This will result in slightly more no-shows, but I think that is an acceptable tradeoff.
Re:How do you show you eliminated bias? (Score:5, Informative)
The original system (COMPAS) never included race as an input. It had 137 features, but none of them was race (nor name). This is hard to analyze. Some researchers [researchgate.net] made a similar predictor with only two features -- age, and number of convictions. It performed about as well, and was about as "racist".
Yeah, but that gives away the game.
Re: (Score:2, Flamebait)
Some researchers made a similar predictor with only two features -- age, and number of convictions. It performed about as well, and was about as "racist".
Probably because the cops are more likely to give a young white guy a stern talking to and take his weed away, where as a young black guy will end up getting charged.
Re: How do you show you eliminated bias? (Score:2)
Totally. That must also be why young black men are convicted of murder many many orders of magnitude more often than old white women. Because cops just give the old white women murderers a stern talking to.
Re: How do you show you eliminated bias? (Score:2)
If you adjust for relative levels of poverty and deprivation then yes, you do find that white people are treated less harshly when someone has been killed. Lesser charges, e.g. manslaughter instead of murder, lesser interrogation and lower rates of coerced false confessions, juries more likely to believe mitigating circumstances like self defence or accidental death etc.
Lol. Sure. You've found some way to tell the difference between "juries more likely to believe mitigating circumstances like self defence" and "actually more cases of self defense". Pull the other one.
The super handy thing about that is you can use it an excuse to ignore mitigating factors instead of controlling for them, which of course gives you exactly the result you want.
Re: (Score:2)
Taiwan #1
Re: (Score:2)
Re: (Score:2)
What's the point of using the system if you're just going to ignore the results.
They shouldn't ignore the results. But they shouldn't just use them blindly either.
If it's showing disparities for something that should not matter
Often something that shouldn't matter actually matters.
I'd wager that if you fed it even more data, particular socio-economic information, you'd find out that being black has very little to do with it
I would gladly accept that wager because that is what they did, and they didn't get the results you are predicting. Political correctness is not the same as correctness.
Re: (Score:2)
> Ignoring the results here is like doing an analysis that tells you that CO2 contributes to global warming, but deciding to build more coal power plants just because phasing them out would make people uncomfortable.
<Cries in Australian>
Re: (Score:2)
How do you correlate
with
are blacks slightly more, or strongly more likely to jump bail?
Re: How do you show you eliminated bias? (Score:2)
But as a society, we have decided that is unacceptable
We haven't; a vocal subset of society has decided that it is unacceptable, and the people in charge have decided that they really really don't want to be called racist even if it's completely unfounded. The vast majority of people, on the other hand, are perfectly fine with the idea that high risk individuals should be treated as high risk, even if they happen to be an inconvenient colour.
Re: (Score:2)
I'm horrified.
Re: (Score:2)
It turned out to be very "racist", denying bail far more often to black defendants.
But it wasn't "wrong" because race is actually very strongly correlated with jumping bail. Blacks are significantly more likely to fail to show up for trial.
So you are saying that the system is biased against black people, and that black people are less likely to participate in the system that is biased against them so it's not "wrong"?
Seems that if you are white and you know you will probably get off with a slap on the wrist and a fine you are much more likely to not skip bail than if you are black and expect to do significant time, doesn't it?
Re: How do you show you eliminated bias? (Score:2)
Seems that if you are white and you know you will probably get off with a slap on the wrist and a fine you are much more likely to not skip bail than if you are black and expect to do significant time, doesn't it?
Sure, if you're a moron. Skipping bail guarantees you won't get off with a slap on the wrist. It's a self-fulfilling prophecy.
Re: (Score:2)
The term you are looking for is "vicious circle."
Re: How do you show you eliminated bias? (Score:2)
Call it what you will, it's fucking stupid. Either way it doesn't change the fact that since blacks are more likely to skip bail, you can't expect them to be granted bail at the same rate as whites.
Re: (Score:2)
Either way it doesn't change the fact that since blacks are more likely to skip bail, you can't expect them to be granted bail at the same rate as whites.
That's textbook racism. Someone is black so they can't get bail because other people with a similar skin tone have skipped bail.
Justice is supposed to be blind, i.e. treat everyone fairly and the same.
Re: How do you show you eliminated bias? (Score:2)
That's textbook racism.
It's not racism at all, let alone textbook. The ever increasing abuse of the word over the last two decades has rendered such accusations completely meaningless.
Someone is black so they can't get bail because other people with a similar skin tone have skipped bail.
No, that's not what anyone is suggesting. You just made it up.
Justice is supposed to be blind, i.e. treat everyone fairly and the same.
Yes, and when we make it completely blind (as in the originally referenced AI model) we find that blacks are denied bail at higher rates than whites.
You don't want justice to be blind; you want justice to see clearly, and then discriminate in a way that makes you feel like theres no rac
Re: (Score:2)
Dude, you are very, very wrong. If you are treating someone differently on the basis of their skin color, guess what? That's racism.
Yes, the word has been abused and people have tried to change it. This isn't it, though.
Re: How do you show you eliminated bias? (Score:2)
Dude, you are very, very wrong. If you are treating someone differently on the basis of their skin color, guess what? That's racism.
And, again, nobody is suggesting doing that. What you seem to be unable to comprehend is that a disparity in outcomes does not have to be driven by "treating someone differently". Different outcomes for different race/sex/age/culture groups arise even when the system is completely blind to those characteristics. I know that this seems impossible to cultural relativists and biology-deniers, but it is true nonetheless.
Re: (Score:2)
You wrote this:
Either way it doesn't change the fact that since blacks are more likely to skip bail, you can't expect them to be granted bail at the same rate as whites.
That is racism. It doesn't matter if 10% or 90% of black people skip bail; you don't deny bail based on what other people with the same skin tone have or haven't done.
There is no control on how people treat each other; that is the outcome where there are different results and it won't be equal across race/sex/age/culture. That is MUCH different from a legal system that says "others with your skin tone have done this, so bail denied."
Re: How do you show you eliminated bias? (Score:2)
That is racism. It doesn't matter if 10% or 90% of black people skip bail; you don't deny bail based on what other people with the same skin tone have or haven't done.
No, it isn't; your reading comprehension skills are abysmal. The selection which you quoted in no way states or implies that anyone should be denying bail based on skin tone.
Re: (Score:2)
Again, you wrote this:
Either way it doesn't change the fact that since blacks are more likely to skip bail, you can't expect them to be granted bail at the same rate as whites.
Re: How do you show you eliminated bias? (Score:2)
Again, you wrote this:
Either way it doesn't change the fact that since blacks are more likely to skip bail, you can't expect them to be granted bail at the same rate as whites.
Yes, I did. And, again, it neither states nor implies the idea that anyone should be denied bail based on their skin tone.
Is English your second language? Are you still having trouble understanding how differences in outcome can arise between groups even when factors like race aren't used as inputs? What exactly is the misunderstanding here?
Re: (Score:2)
Yes, I did. And, again, it neither states nor implies the idea that anyone should be denied bail based on their skin tone.
Holy shit. You don't even understand what you type. I'll try one more time. You wrote:
you can't expect them to be granted bail at the same rate as whites.
You're trying to say that "neither states nor implies the idea that anyone should be denied bail based on their skin tone", but it DOES.
Re: How do you show you eliminated bias? (Score:2)
ESL then?
If I tell you that blue people steal 20 times more often than red people, therefore you can't expect them to be arrested at the same rate, does that imply we need to go out and start arresting all the blue people? Or does it mean that they're going to get arrested at a higher rate even if the person doing the arresting is colour blind?
If you can answer that question then maybe we can figure out whether your issue is with language or with logic.
Re: (Score:2)
Just let the system do it's work and then apply a "race filter" on the output to select the same percentage of whites and blacks for bail. This will result in slightly more no-shows, but I think that is an acceptable tradeoff.
I don't find that acceptable. Treating everyone exactly the same regardless of race is the only way to actually avoid racism, whether you like it or not.
Re: (Score:3)
Commit more crimes, get treated as if it's less of an issue.
Reality isn't that simple.
Blacks are about four times more likely to be convicted of a crime.
But studies have shown that blacks are more likely to be prosecuted for offenses where a "nice" white kid would get a warning or a slap-on-the-wrist [wikipedia.org].
So the conviction rate isn't the same as the crime rate.
African Americans are also 50% more likely to be wrongfully convicted [msu.edu].
Re: How do you show you eliminated bias? (Score:2)
But studies have shown that blacks are more likely to be prosecuted for offenses where a "nice" white kid would get a warning or a slap-on-the-wrist.
Yes, primarily because they are more likely to already have a record, and partly due to other aggravating factors. When you control for that, most of the difference goes away.
So the conviction rate isn't the same as the crime rate.
Obviously not; a lot of them are never caught. The crime rate is much higher.
African Americans are also 50% more likely to be wrongfully convicted.
No, as your link shows they are about 50% more likely to be exonerated on the charge of murder. Those are two different things. Given the number of activists fighting to overturn the convictions of blacks convicted of murder, and the distinct lack of such
Re: (Score:2)
Well, you've just shown your unsubstantiated bias.
Re: How do you show you eliminated bias? (Score:2)
Well, you've just shown your unsubstantiated bias.
It's not a bias, it's a claim, based on personal observation. One which I've already demonstrated I'm willing to part with since I could be wrong and it makes very little difference to the issue at hand.
You've just shown your dishonesty by focusing on that one claim and ignoring everything else.
Re: (Score:2)
"Your police"
Fair question... why do police in predominantly Democrat strongholds have that problem?
I don't hear all that much of it happening in say... North Dakota.
Re: (Score:2)
Re: (Score:2)
95% of police murders are against men. Should we not be examining why this country is so sexist?
As a man, can I now not trust the police - and apparently be justified for whatever shit I do when they try to arrest me ?
Re: (Score:2)
What they are against is treating different groups differently in the law and in common social environments. They are against treating the individual based on the group they are lumped with.
What possible reason could you have to argue that we should be biased, and jail working black people who use marijuana, and let white college students smoke all they want?
That there are differences doesn't mean we should act is if to keep perpetuating some of the negative diff
Re: (Score:2, Interesting)
No one denies there are differences. What they are against is treating different groups differently in the law and in common social environments. They are against treating the individual based on the group they are lumped with.
Is that what is happening here? Or are the results - with ZERO input on race - coming out as you don't like, and thus must be racist?
What possible reason could you have to argue that we should be biased, and jail working black people who use marijuana, and let white college students smoke all they want?
In this case, the study didn't even include race - but the results said "deny bail" more times to black defendants than white (as a percentage of population) and thus must still be racist. Even though the "AI" system didn't even know about race.
That there are differences doesn't mean we should act is if to keep perpetuating some of the negative differences. You can't get an ought from an is.
Likewise, we should not penalize someone for being of a different group - even if that group is "better" somehow. Stop using race a
Re: (Score:2)
In this case, the study didn't even include race - but the results said "deny bail" more times to black defendants than white (as a percentage of population) and thus must still be racist. Even though the "AI" system didn't even know about race.
So you train your AI on racist data and get a racist AI...
Who knew...
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
I just want to know what rules you plan to twist to get your way.
Tell ya what: you first tell me how racism only applies to 3rd+ generation blacks, and then we can talk about if there is racism in the first place.
Re: (Score:2)
Tell ya what: you first tell me how racism only applies to 3rd+ generation blacks, and then we can talk about if there is racism in the first place.
Easy, it doesn't.
Your turn:
Re: (Score:2)
Re: (Score:2)
GOTO 10 [slashdot.org]
Already asked and answered.
You're setting up a strawman anyway. Being disadvantaged doesn't mean you cant succeed.
Still your turn:
Also already explained to you here [slashdot.org]
Re: (Score:2)
Do you get paper cuts? Sure. Do you have a problem with it, where it is a constant issue that hinders your life? No.
In the same vein, does racism exist? Sure. Does pedophilia exist? Absolutely. Does the US have a problem with either where it is a constant issue endemic throughout society? No.
People motivated enough to move to another country for a better opportunity are more likely to take advantage of that opportunity. They not only benefit financially, making it easier for their kids. But instill that kind of work ethic in their offspring.
So the problem isn't that racism exists, but that somehow people lose their drive to succeed and take advantages of the opportunity that this country affords. I believe - if you were actually honest - you'd agr
Re: (Score:2)
95% of LEO killings are of men. Does America have a sexism problem based on this data?
The answer will be the same for your question and mine. Hit it.
Re: (Score:2)
Tell a black person racism is just like a paper cut. I'm sure that will go down well.
You are claiming Black people 'lose their drive' more than white people.
Isn't that racist of you?
Re: (Score:2)
I don't think you know what "gassing people" means.
Protests and riots are what define America. For better or worse. Africans and Uighurs in China WISH they could protest and riot.
Eventually good will come of it all, probably.
Re: (Score:2)
This Google search for racist AI [google.com] will help get you started
Re: (Score:2)
Fail.
Now, if you had added that you suffered from hereditary baldness you would win. But since you're obviously a hair-ist - you are a Nazi and should be punched in the face.
Re: (Score:2)
Re: (Score:2)
So the solution is to be even LESS colorblind?
If by 'solution' you mean keep doing the same things and hope the problem goes away.
Re: (Score:2)
You'd only pointlessly argue my definition. I'll let you take a swing.
How do you define it?
How do you claim it doesn't exist when you aren't willing to define what it is?
why do 1st/2nd generation black immigrants vastly out-perform 3rd+ generation native born blacks.
Selection bias.
Immigrants of all kinds outperform the comparative native born people for any group.
People motivated enough to move to another country for a better opportunity are more likely to take advantage of that opportunity. They not only benefit financially, making it easier for their kids. But instill that kind of work ethic in the
Re: How do you show you eliminated bias? (Score:2)
How do you claim it doesn't exist when you aren't willing to define what it is?
The same way I claim god doesn't exist. Because nobody has ever defined one for which there is any evidence of existence. I don't need to define it myself; those who claim that it exists have to define it. You could always claim that the Sun is god, and then I would agree that your god exists, but we would probably find many other points of disagreement given the uselessness of your definition.
Re: (Score:2)
but we would probably find many other points of disagreement given the uselessness of your definition.
slow clap
It's almost as if you'd need your own definition then isn't it...
You either accept the other persons definition, and that becomes your working definition. Or you define it yourself. There is no other way. Both ways you have a definition to argue against.
I can't meaningfully argue 'skogjsb' doesn't exist if nobody knows what it is.
Re: How do you show you eliminated bias? (Score:2)
It's almost as if you'd need your own definition then isn't it...
Not at all.
You either accept the other persons definition, and that becomes your working definition.
That's exactly what I did. Calling the sun your god is your definition. Hooray. The reason it is useless isn't because I have my own definition; it's useless because you're just renaming something I already have a word for.
I can't meaningfully argue 'skogjsb' doesn't exist if nobody knows what it is.
No, but apparently you think that you can argue that "there is a problem" without being able to define what the problem is. And you think that you can insist that others define your problem for you and then prove it doesn't exist.
Go figure.
Re: How do you show you eliminated bias? (Score:2)
Did you even think about what you wrote?
Yes. Didn't think anyone would be stupid enough to misunderstand it.
Re: (Score:2)
listening to many on the SJW spectrum the answer is to be "color brave".
ie explicitly racist in an anti-white way.
Re: (Score:2)
This sounds like some folks are afraid that using a completely dispassionate method of reaching conclusions (smart systems/AI) may, in fact, show that there ARE differences amongst groups of individuals, and that would bring the walls of the SJW castle crashing down...
This sounds like someone who doesn't understand that dispassionately training (smart systems/AI) on racist data will result in a racist AI.
That would bring the walls crashing down on everyone pretending there isn't any racism in the original data.
How are you explain the racism getting in there? What are you planning to do about it?
Or are you just happy to expand racism into future technology/generations because you pretend you can't see it?
Re: (Score:2)
How are you explain the racism getting in there?
How do you know it's racist? What data do you have to draw that conclusion?
What are you planning to do about it?
Nothing, it's it's not racism - which it does not seem to be.
Can you explain why 1st and 2nd generation African immigrants vastly out-perform [qz.com] 3rd+ generation native-born blacks in scholastic and financial success? Why they have a dramatically lower incarceration rates [nber.org] - meaning they commit fewer crimes? Are we "racist" only against some blacks - but not others?
Re: (Score:2)
Are we "racist" only against some blacks - but not others?
Yes. Not every place in America is equally racist.
Can some black be successful despite the racism? Yes.
Did you have a point?
Does America still have a racism problem, or is it all solved? yes/no ?
Re: (Score:2)
Re: (Score:2)
So 1st and 2nd generation blacks only settle into places without racism. That's your answer?
No that's your conclusion.
Are you claiming they spread out perfectly around the country?
Is that what your data says? I'd be happy to read it if you can show it.
Or are you instead claiming all parts of America are exactly equally as racist as the rest.
And you still didn't answer this:
Re: (Score:2)
No, that's what the data shows. Look at the academic and economic success of the two groups: one is quite on-par with most immigrants, and indeed very close to whites; indeed, 2nd generation immigrant blacks outperform whites educationally, and are very close economically. Then by the 3rd+ generation, it starts to fall apart.
If your premise is true - that America is fundamentally a racist nation - then how does it "allow" those 1st and 2nd generation blacks to succeed so well? Perhaps because the US does
Re: (Score:2)
If your premise is true - that America is fundamentally a racist nation - then how does it "allow" those 1st and 2nd generation blacks to succeed so well?
That's your usual strawman.
Just because some blacks succeed (As I've already showed you a few times now [slashdot.org]) doesn't mean racism doesn't exist.
It's either fundamentally dishonest or profoundly stupid for you to claim that Lynnwood.
Re: (Score:2)
NLP stands for "Natural Language Processing", i.e. they are trying to make their system understand normal human speech patterns. So there is no need for belief, you can simply test it to see if it scores as well with various different cultural speech patterns and accents.
Try running this through Siri: https://youtu.be/tFdUmLEJqtY [youtu.be]
Re: (Score:2)
It is hard to show that all bias is eliminated, especially if you are unaware of your own subjective biases, but it is not impossible to create objective criteria. These studies, though, show that it is harder than previously thought.
Obviously not.
Re: (Score:2)
Re: (Score:2)
Thankfully, George Floyd's killer has been arrested and charged with murder. We'll have to wait for the trial, but he probably won't go free.
Just like another Minneapolis police officer, Mohamed Noor, who shot and killed Justine Damond [wikipedia.org] without even a reason to arrest - let alone kill - the young woman.
Re: (Score:2)
I agree, American police are over militarized and undercontrolled. But Office Chauvin and his accomplices would not have been charged if not for the uproar and video tape. Ahmaud Arbery's killers would not have been charged if not for the uproar and video tape. Jemel Roberson's killer is loose, and the NRA does nothing. There are a few patterns, and excessive force by law enforcement is one, law enforcement covering up their own crimes is another, and disproportionate violence inflicted on blacks without c
Re: How do you show you eliminated bias? (Score:2)
Ahmaud Arbery's killers would not have been charged if not for the uproar and video tape
And they shouldn't have been. All of the available evidence so far shows that he attacked the guy who shot him. Fairly clear cut self defense. There is literally zero evidence to suggest that the shooting wasn't justified.
The charges against them are the most blatant example of politically motivated prosecution that I've seen in ages. There's a 99% chance they walk, just like Zimmerman did.
Re: (Score:2)
Re: How do you show you eliminated bias? (Score:2)
They tried to grab him.
There is zero evidence of this. The video shows him running towards and attacking them, not the other way around.
He fought back in what his supporters would call self defense.
If they had tried to grab him, his actions certainly would constitute (a foolish attempt at) self defense, and they would be guilty of murder. But, again, there is no evidence that they tried to grab him.
Then they shot him in what you call self-defense. That isn't "literally zero evidence to suggest that the shooting wasn't justified".
It's literally less than zero; all the evidence points to the contrary.
There is always a possibility that they did something which justified him attacking them, but if they did, it occurred off
Re: (Score:2)
Re: How do you show you eliminated bias? (Score:2)
*RIM SHOT!*
Lol, I'll be here all week, folks!
No (Score:2)
Is there absolutely nothing in this world that these folks don't insist on mucking up and slowing down?
If you want to improve the world by working on a bias issue that matters, turn on the news this week. Any channel, it doesn't matter.
What is NLP? (Score:5, Insightful)
If you're going to use an acronym that much in your post, please define it. After a quick search it seems OP is talking about Natural Language Processing.
Re:What is NLP? (Score:5, Informative)
I agree, but it is also theoretically possible to be:
Neuro-Linguistic Programming.
The confusion exists because the original source at Cornell was incompetent and despite including an acronym in the title of the paper did not describe that acronym in the synopsis.
Re: What is NLP? (Score:2)
Putting words in people's mouths (Score:5, Insightful)
This is basically just a complicated way of putting words in people's mouths. Suddenly making a neutral statement using the vocabulary you grew up with becomes a "micro-aggression" or even frankly x-ist.
Followed closely by creeping meanings. New really for-real neutral words get insisted upon. Those words promptly gain all of the connotations of the old 'unacceptable' words. Because in final analysis, it's not the words themselves but the intended meaning behind them that makes the difference.
Newspeak was just taking it to the extreme of not actually having words that might have a negative connotation. That is, restricting the ability to express oneself until it becomes impossible to express anything that isn't officially approved. But even then, you can still feel it.
Re: (Score:3)
On the plus-side, attempts to force you to change your vocabulary in this way are just plain aggression, no "micro" about it. It does allow for easy identification of the fuckups that have nothing actually worthwhile to contribute but want to dominate the conversation anyways.
Re: (Score:2)
...making a neutral statement using the vocabulary you grew up with...
My grandparents grew up using the words "colored boys" to refer to black teenagers, which they continued to use in their grandparent years. They didn't mean anything by it (as least I'm pretty sure they didn't), it's just what they were used to using.
Vocabulary changes. Deal with it.
Re: (Score:2)
My grandmother also used "colored". That was the proper term to be used in her day. It was the term used by groups advocating for racial equality in her day.
Re: (Score:2)
And still is. Pop quiz: What does the "C" in "NAACP" stand for?
Re: (Score:2)
What does growing up with something have to do with it? My mum had a gollywog growing up, doesn't make them any less racist now and doesn't make her a racist since she discarded it when she came to understand why it was a problem.
Anyway, this has nothing to do with people, it's about computers understanding what is said to them. Natural Language Processing, i.e. what Siri does to make sense of your request.
Re: (Score:2)
TFA wandered well beyond computers understanding speech. TFA did, in fact include as an implicit assumption that NLP doesn't work well for (some) African-Americans because software engineers are racially biased.
Pretty research-speak (Score:2)
Microsoft intersectionality© (Score:2)
And water is wet. If you read a book by a German or South American author then the text has to reflect the background of the writer. What do you expect. Why is this deemed in need of correcting.
I just wonder is anything like this “research” is being investigated in places like China or Sa
TLA (Score:3)
ARGH. God I hate TLAs. Undefined three letter acronyms... what the hell? Who writes this stuff? And who is it for? All of the uber nerds in the world in a giant circle jerk? I thought for sure this was referring to neurolinguistic programming, but that didn't make any sense. Define your terms unless you don't care to communicate with anyone outside your little club of shibboleth-slinging jargon monsters.
Note to poster (Score:2)
As the recently released GPT-3 and several recent studies demonstrate, racial bias,
as well as bias based on gender, occupation, and religion,
can be found in popular NLP language models.
Making a summary by cutting and pasting from the first paragraph of a paper, written
in cryptic abbreviations, is not at all helpful to readers.
Re: (Score:2)
I agree with the sentiment but not the example. I found that very readable.
"Articulate their conceptualization of 'bias' " is really the only mouthful; that probably could have been written, perhaps as "explain their models of 'bias' "
Re: (Score:2)
or the models predict the poor behaviour of people who are not white and we are going to assume that's becasue the models are racist, not because they are accurately predicting behaviour.
Typical for academics who see "social injustice" everywhere, such as "black people are much more likely to go to prison" and their first response is to come up with "well, that's because cops are racist"
Re: (Score:2)
Or the models are not predicting behavior at all.
They are trying to parse and emulate human language in a way understandable to humans, and therefore tend to copy people's biased wording. Not at all surprising, but not a good thing to teach your bots if you want them to get along in the world.
Re: (Score:2)
I would have written something like this:
Bias is systematic under- or over- estimation of a quantity of interest. Bias cannot be averaged out or estimated statistically, but requires comparison of model estimates with the true value. Since the true value is usually unknown, and because systematic error can have important real-world consequences, we recommend researchers clearly state their assumptions, and discuss possible implications of their definition of truth in terms of any proposed applications.
Re: (Score:2)
Still, when I was a kid, my cousin, a graduate student at the University of Chicago, was staying with us for a few days. When he flipped a light switch and the light didn't turn on, he said something like "It appears that the light is not functioning properly." Me and my brothers thought that was hilarious.
Re: (Score:2)
Lol. You realize that since Slashdot banned random anonymous cowards, posting as an AC tells everybody that you specifically and purposely did so because you don't have the guts to attach your pseudonym to what you've said.
The point of an academic paper is to communicate. Some jargon is often unavoidable. But the more clearly and simply you write, the wider an audience, academic and otherwise, will be able to understand.
Academics writing in a needlessly inaccessible style is a recognized issue. I didn't jus