This has been a real problem with facial recognition. Faces used to train engines are typically Caucasian, so the ability to distinguish other ethnicities is quite sub-par.
Isn't there a problem with shadow contrast which makes features more difficult to pick out on darker skinned faces? I vaguely recall an East Asian developer giving that reason for why he developed on Western data-sets.
Yes, but then you need special equipment rather than standard cameras (that usually filter IR); which would prove the point that it's about physical limitations rather than racism.
That's not racism, that's laziness. Or perhaps just being a bad engineer, if you want to be more cynical about it.
It's not that hard to find both conscious and subconscious examples of systemic or individual racism. Engineers taking the easy path with webcam facial recognition is probably not a good one and serves only to give more fuel to those who claim people jump to the racism cry too quickly.
If you ship a product because it only works for white people and that's good enough, that is definitely racism. Just as it would be sexism if you shipped a voice recognition product that only worked for men.
Laziness and racism aren't mutually exclusive. Historically in America, white people haven't considered black people as fully human. You can read the various declarations of secession or the 3/5ths compromise for that. Or the long post-Reconstruction history of slightly more subtle ways.
Sure, laziness was involved here. But deciding a product was good enough to ship without caring that it worked for black people requires the effective belief that black people didn't really count as people. At least, not people that mattered. Imagine the reverse: if the product didn't work on white men, would it have been shipped? Of course not.
When laziness just happens to have a blatantly racist outcome in a place where there is a centuries-long history of racism, Occam's Razor suggests the explanation is racism. If laziness could not have caused the bad outcome to happen for white people, then it's pretty clear that pure laziness is not the real cause. It's instead white people being differentially lazy when it comes to black people. That's clearly racism.
As a confirmatory example, look at the American justice system. In a lot of places and times, the same nominal laws applied to white and black people. But they were enforced very differently. Serious crimes against black people were ignored. Minor infractions by black people were enforced vigorously. [1] Were the cops lazy? Sure, everybody's lazy sometimes. Was that why there was a racially different outcome? Definitely not.
[1] Examples of this are all over Loewen's "Sundown Towns" for example.
I think it's important to start talking about, and acknowledging, that nearly every second of the average person's life is thinking about something that is within their own universe of their life's experiences.
So plausibly, this really was an innocent and understandable mistake. Maybe they grew up in a town full of white people, and maybe all of their friends and coworkers are white. I understand what they did is technically racist, but that's because of that particular combination of people. Let's use _this_ 'experience' of "racist", but not it's exact definition. For example:
- It was racist you forgot your mother in law's birthday (who you see once a year.)
- It was racists you didn't lift the toilet seat. (Edit: I grew up in a culture where it was offensive to not raise the seat)
Definition: Any unintentional side effect from not thinking about someone* outside of everything you've experienced, and presumably causing harm to that outsider.
* Usually this person is disadvantaged in some way. But almost everyone is disadvantaged in some way, thus neutralizing this particular point IMO.
Does that sound rational? I don't think it does.
The world is full of suffering, and I'm advocating people help and love the people they are surrounded by. If many people did that, it would be easier to recognize each other, where ever you come from. Why? Because hate will push away everything it is unfamiliar with, but love will accept everything it is unfamiliar with.
(Now meta argument, do I expect people to change? No. So I try accepting them instead.)
Again, something can be racism and something else.
> Maybe they grew up in a town full of white people, and maybe all of their friends and coworkers are white.
This is not an accident. All-white contexts in America are the result of personal and systemic racism. People who grow up in those contexts are, unsurprisingly, more likely to be personally biased and will regardless do things that further systemic racism.
People in that condition have a choice: they can either overcome their upbringing or they can continue to support white supremacy. If they do the latter, well, then they've made a choice. You can accept that if you want. I don't.
I don't think it's likely that all the folks in Vermont are racist or that the system of governance there is supremely racist - yet Vermont has an absolutely tiny percentage of non-white folks (3.1%)
Less than 15% of the U.S. is African American so it's not at all surprising that there are areas where there are very few black folks as a result. That's nobody's fault and shouldn't be described as racist because, well, it isn't.
3.1% is about 1 in 30 people. That's not that small. And this isn't 100 years ago, I'm pretty sure people in Vermont have seen a black face on TV/film.
This is historically ignorant. I don't blame you personally, because the American education system is very bad at covering America's racist history. But you're still basing your estimation of "likely" not on real data, but on what you'd like to be true.
If you'd like to learn more, start with Loewen's "Sundown Towns", which describes how hundreds and possibly thousands of towns across the US were turned and kept white:
It's also not necessary for a lot of white people to be actively racist to preserve white supremacy. As is clear from Loewen's book, all it takes is a small number of people willing to be active (e.g., making threats, burning crosses) to drive off new black arrivals. The majority can be passively accepting of a racist system without lifting a finger.
But that doesn't mean they won't end up biased. White people who don't know many black people are more likely to be biased, and all-white towns produce people that are more likely to have racist views. (For more, see Loewen's Ch 11, "The effects of sundown towns on whites".)
This is neat, in a conversation about the technical aspects of image recognition with webcams in low lighting conditions, you're talking about secession and the 3/5th compromise. Let's try to stay on topic.
> But deciding a product was good enough to ship without caring that it worked for black people requires the effective belief that black people didn't really count as people.
Nobody ever said "oh it doesn't work with black people? Who cares LOL." I was going to make a crack about mental gymnastics but this isn't even that. You're just making shit up at this point.
> When laziness just happens to have a blatantly racist outcome in a place where there is a centuries-long history of racism, Occam's Razor suggests the explanation is racism.
No, even using your sentence here you admit the cause is laziness but then try to reframe it to racism because that fits your world view. Just like the fact that we're talking about webcams and you're talking about the 3/5th compromise and the American justice system.
Is there institutional/structural racism in the US? Absolutely. Webcams aren't a good example of it, sorry.
I do not admit that "the cause" is laziness. Laziness is part of the causal chain, sure. Single-cause thinking is a poor way to do a failure retrospective.
We don't know whether or not they knew it didn't work with black people before they shipped it. Maybe they did and dismissed it with technical hogwash, as you are here. Maybe they did and decided it was an ok flaw. Maybe they didn't know because the company doesn't have any black employees and nobody though to test it on black people because the people in charge don't know any. Or maybe they do and just don't care. There are many possible paths to this decision, but any of them demonstrates systemic racism.
And this is on topic, because the broader topic is social implications of technology.
As an aside, your attempt to narrowly define the topic so racism can't even be spoken of is a pretty typical move from people experiencing white fragility: http://libjournal.uncg.edu/ijcp/article/view/249
Seeing someone interact with others after drinking enough kool-aid to fill a pool is actually pretty entertaining. Thanks, I needed a break from work.
Speaking of work, have you tried testing your definition of racism on non-white subjects and products, such as Black Entertainment Television? It doesn't hold up very well.
My definition of racism is pretty standard among people who study it. I'm talking about a societal power system, like capitalism or communism. The US was undeniably racist to start out; black people were property, and only white men could vote. The level of white supremacy has declined some since then, but hasn't been eliminated. And it has been going back up in the last few years.
You could mean two things by "testing your definition of racism on non-white subjects": in the US or elsewhere. Either way, yes, of course I have. When going elsewhere it's necessary to speak of "the dominant racial group", and that can be helpful here too. So taking BET, we can ask questions like, "What effect are they having on non-dominant racial groups?" And, "Do their actions tend to increase or decrease the racial power imbalance?"
My first impression is that they are having positive effects on a non-dominant racial group, and are acting to decrease the racial power imbalance. Therefore, no, they are not racist.
Since you are anonymous person with a new account turning up to be a dick about the topic of racism, I presume your question was entirely insincere. But just in case, I hope you find the answer useful.
> That's not racism, that's laziness. Or perhaps just being a bad engineer, if you want to be more cynical about it.
It can be both. An attitude that considers facial recognition software acceptable if it only recognizes light-skinned people is both lazy and racist. Only choosing to test on light-skinned people is racist because doing so assumes dark skinned people are an exception or an outlier, rather than an equally valid part of the set of "human faces."
True, but that doesn't really contradict the premise of my comment.
The assumption of ones own race being the default color for humans (rather than an arbitrary value) that leads to facial recognition software being trained on a data set so narrow that the resulting software can't even recognize skin colors that diverge from that norm is as much about implicit, albeit unintentional, racial bias as it is the particular technical problems involved and the need for engineers to cut costs and meet deadlines.
It's not entirely racist, but racism is a component.
> That's not racism, that's laziness. Or perhaps just being a bad engineer, if you want to be more cynical about it.
I wasn't ready to admit the genius of the show before, but I am now: what you said is quite close to a line from the same episode (said by the sociopathic boss, Veronica) when told by the protagonist that the sensors were effectively racist "[...] it's actually the opposite of racist, because it's not targeting black people. It's just ignoring them. [company] insist the worst people can call it is "indifferent.""
I agree. It's a well known problem where the training set isn't representative of the underlying population. While it can certainly be argued that the engineers should have recognized this deficiency and taken corrective action, I really don't understand why all the respondents to your post are so quick to assert racist intent based on clickbait headlines from Forbes.
It's not an active racism on the part of the engineers. It's way more subtle than that.
There could be a hidden assumption on the part of the engineering team that light-skin is "normal" and anything else is a special case. Nobody is saying "I hate black people" or anything of the kind.
Or, as happened with a voice recognition system a former employer used, it was tested on the engineering staff, who happened to be all male. As a result it didn't work well for most women who tried to use it. There was no intentional exclusion of women from the test data, and I'd argue, no intentional exclusion of women from the engineering teams. But it is reasonable to say that systemic sexism that excludes women from engineering careers helped this system fail.
These kinds of problems are difficult to solve, because they aren't active decisions on anyone's part. They evolve out of pervasive conditions, and unconscious biases. At root, the source is still racism, or sexism (or another form of discrimination).
Because it is racism. It is ignoring a large swath of humanity based on the color of their skin. You may not want to think of it as racism because it's not the burn a cross on their lawn type of racism, but it's still systemic racism.
> It is ignoring a large swath of humanity based on the color of their skin.
No, it's an insufficiently sensitive contrast filter combine with too narrow a training set. Screaming racism at everything that is even marginally approaching the topic of race detracts from real racism and ignores the actual issue here - shitty software.
But laziness around something that dispropotionately impacts people based on their race is racist. If you know your system has a harder time training against non-white faces, and you choose to train it only against white faces to be lazy, that's still explicitly racist. This isn't even a case of negligent racism ("oh well I didn't know my system wouldn't work for people with darker skin because only my white coworkers tried it out"). The example here is a case of the engineer explicitly deciding to avoid the harder case of darker skinned individuals, knowing that the results would be poorer for those individuals, and thinking it doesn't matter if the system doesn't work well for people with darker skin. That is explicit racism.
Skin color is not synonymous with race. Every race has people that span a wide range of skin tones, and skin tone fades with age, so you may as well be arguing it's ageism.
Actually, even if it is accidental or subconscious, I'd say this is a perfect example of systemic racism--racist behavior (I'm referring to an algorithm/device specifically here, which you can safely call objectively racist) which is normally within society's acceptance level which, when magnified to a societal scale, is no longer acceptable. Another symptom of this illness is that you might not even notice as a member of this system the system is broken if you're white, but you would if you're black.
Personally, I think it could be understandable people didn't consider race when developing facial recognition technology, especially when we've only had mainstream awareness of this for under a decade and many people live in racially homogeneous or dominated cultures. However, I don't think it's acceptable for organizations, and the time when you can safely say you didn't understand biased learning data will be over soon. There are considerations you need to make scaling your tech from personal project to something the public will consume.
Also, the day will come when computers can point out racist stuff better than the average human can now, albeit with a high false positive rate. I say this because it's relatively easy; even if you only count a subset of tweets talking about racism as not trolling, that's still a shockingly high number of meaningful things about the world many people aren't seeing.
It is racism to realize the negative aspects to other races, nevertheless pitch it as universally functional, and push the cost/pain to other races. Laziness is some of it, some of it is the realization that only other races bear the brunt of the cost.
Most cheap cameras actually don't filter out IR; it often shows up as blue or purple. You can use this fact to see if an IR remote is working -- just shine it at your cellphone camera or webcam. Whether there's enough IR sensitivity to be able to illuminate a face is another question.
I find this phenomenon of protecting oneself from being called racist fascinating. Your opinion is not exactly racist, but you had to bring up another minority, in case someone accuses you?
https://www.theatlantic.com/technology/archive/2016/04/the-u...