Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
Google Privacy

Google Contractors Reportedly Targeted Homeless People For Pixel 4 Facial Recognition (theverge.com) 55

Posted by msmash from the stranger-things dept.
From a report: In July, Google admitted it has employees pounding the pavement in a variety of US cities, looking for people willing to sell their facial data for a $5 gift certificate to help improve the Pixel 4's face unlock system. But the New York Daily News reports that a Google contractor may be using some questionable methods to get those facial scans, including targeting groups of homeless people and tricking college students who didn't know they were being recorded. According to several sources who allegedly worked on the project, a contracting agency named Randstad sent teams to Atlanta explicitly to target homeless people and those with dark skin, often without saying they were working for Google, and without letting on that they were actually recording people's faces. Google wasn't necessarily aware that Randstad was going after homeless people, but a Google manager reportedly did instruct the group to target people with darker skin, one source told the Daily News.
This discussion has been archived. No new comments can be posted.

Google Contractors Reportedly Targeted Homeless People For Pixel 4 Facial Recognition More

Google Contractors Reportedly Targeted Homeless People For Pixel 4 Facial Recognition

Comments Filter:

  • Comment removed (Score:5, Insightful)

    by account_deleted ( 4530225 ) on Thursday October 03, 2019 @07:55PM (#59268194)
    Comment removed based on user account deletion

    • Not for databases. The moment I read that I knew exactly what it was for and that people were going to immediately cry racism, except it's probably being done to address complaints of racism.

      People have long complained about facial recognition tech (especially cameras that are supposed to follow your face as it moves in the cameras field of view) and how it doesn't work for black people while it does for white people. Well, this would help it work for all. Google probably doesn't want to sell a phone with a

      • Well, the article also points out that there hasn't been enough data recognition from people with darker skin, so I don't think there's going to be as many misinformed cries of racism as you think.

        The outrage is the contracted company that captured a bunch of this data allegedly misrepresented themselves a lot.
        > Some were told to gather the face data by characterizing the scan as a “selfie game” similar to Snapchat, they said.
        > One said workers were told to say things like:
        > “Just

      • Seems to me you're racist if you try to do something about it and racist if you don't. Why even bother?
        • Or they could have started out doing a representative sampling then nobody would be saying it. While it is true that this is an attempt to correct for an initial bias it wouldn't be happening if there was no initial bias, so it is still indicative of a problem rooted in racial disparity.

          • Or they could have started out doing a representative sampling then nobody would be saying it.

            Fantasy. They would still say it.

            • "Fantasy. They would still say it."

              I have to hear this reasoning. If they didn't need to do this they wouldn't be doing it, but your claim is that even though they aren't doing it people would still say the reason they are doing it is racist? Surely you can see where your reasoning went horribly wrong.

          • Or they could have started out doing a representative sampling then nobody would be saying it.

            They don't need a representative sample, they need an adequate sample. Part of what makes black faces harder for algorithms to recognize is lower contrast, which may well mean that you need overrepresentation of black faces to provide uniform performance across all faces.

            • I'll just say this [writingexplained.org].

              • I'll just say this [writingexplained.org].

                Are you saying that you have only a little knowledge, and that you're therefore dangerous? I agree.

                • You don't have a basic understanding of how systems work, and it hasn't occurred to you that if the system isn't tuned to compensate for differences in contrast the system itself needs to be fixed. I realize now that you aren't informed enough to understand how truly ignorant you are so hopefully this helps you realize how clueless you are so that you can begin to learn basic principles on the subject matter of which you have mistakenly convinced yourself you are an expert.

        • Seems to me you're racist if you try to do something about it and racist if you don't. Why even bother?

          Because they want to build a product that works for everyone, regardless of what some people may shout at them.

    • Re: (Score:2)

      by AHuxley ( 892839 )
      Why the rush to do it? Why not just go to such nations and ask any university to help?
      A few images, some money for the time, next person...
      Its for math, science and computers...
      Happy support staff, happy people, a database full of the needed images.

    • Dark skinned people are often underrepresented in these kind of databases

      Well that's less insulting than using gorillas photos to make up the numbers : https://eu.usatoday.com/story/... [usatoday.com]

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      I don't know why you were modded funny, this is absolutely true. The issue is that they were deceptive about gathering the data, but should be commended for trying to include people of different races and skin tones.

      • Deceptive contractor(s), virtuous Google.

        The issue here is spin, if they said that google found a way to improve their facial recognition AI AND hand out $5 each to countless homeless residents, it would seem like a win-win.

        Pretty sure there were no names stored with the images, they were just used to train AI to better handle dark skinned people.

        There have been countless articles about facial recognition being 'racist' because it sucks at processing dark skin faces, Google is trying to correct that bias.

        Th

        • Re: (Score:2)

          by AmiMoJo ( 196126 )

          If Google had set up a stall at the mall offering $5 to test its facial recognition technology would people have signed up? Seems like the targeted homeless people because most people want more than $5 to participate in such a test, so they had to go after the desperate ones for whom $5 is very significant.

          That would be exploiting the homeless.

      • Being more likely to remain unidentified by facial recognition software strikes me as perhaps the only advantage to having dark skin in America.

    • I recall when working for a large OEM that dark skinned people were not detected properly by facial recognition in the early days. An episode of Better Off Ted even made fun of this... and was a most excellent reference to all of us who very well knew of the problem.

      It could be a problem of gaining enough data... or it could be targeted to fix a similar problem (i'm thinking this) where the camera doesn't "see" certain features of dark skinned individuals.

    • Imagine if the article was "Google only pays millionaires for their studies."

  • You have to ask yourself, how else will Google be able to provide useable software to facially ID the future for-profit prisons and concentration camps, such as the millions in China and the US that Google intends to profit off of?

    The software has to be able to identify darker skinned males, so they use homeless people to load the databases.

    Who do you think are in those prisons? It's not white males or females.

    • Who do you think are in those prisons? It's not white males or females.

      OK, let's be fair.

      Many prison populations have "mostly" darker skins individuals. (and we needn't explore the reasons here and now)

      But to say "It's not white males or females." is patently false.

      As someone else alluded to facial recognition has been primarily developed using images on persons with lighter skin tones.

      Which doesn't surprise me at all considering "Black" represents something like 12% of the US population.

      But yeah, they need to focus in on the niche now, hammer it out, get it sorted.

      • Keep saying that to yourself.

        History will judge our actions from the viewpoint of a world that is not what we perceive ourselves to be today, and it will not be in a favorable light.

        • Keep saying that to yourself.

          History will judge our actions from the viewpoint of a world that is not what we perceive ourselves to be today, and it will not be in a favorable light.

          I "feel" like this was a direct reply to me , but i can't be sure.

          It's too vague.

          What do i need to keep telling myself, which part specifically?

        • Keep saying that to yourself.

          History will judge our actions from the viewpoint of a world that is not what we perceive ourselves to be today, and it will not be in a favorable light.

          I'm sorry. I didn't notice the "title" of your post when i originally posted.

          "How else can they fill the for-profit prisons

          ?"

          I think capitalism is fairly efficient.

          But I do not think it should ever be applied to :

          1) Schools
          2) Prisons
          3) Healthcare

          I blasphemously refer to it as the Holy Trinity (don't judge , let "him" do that)

          I know that makes a lot of people upset. (start running numbers in your head)

          It also makes a lot of people happy. (and healthier)(yep all 3)

          If you sleep on it, and take a little bit of time to consider it, I think you might find yo

      • > As someone else alluded to facial recognition has been primarily developed using images on persons with lighter skin tones.
        > Which doesn't surprise me at all considering "Black" represents something like 12% of the US population.

        Not only that. Any photographer can tell you why. Unlike cartoons, real humans don't have lines drawn around their noses or other features. The shape of the face is inferred by the shadow and highlights. Not surprisingly, shadows show up a lot stronger on a light backgroun

        • Re: (Score:2)

          by Mal-2 ( 675116 )

          The average "white" skin tone reflects about 12% of the light that falls on it. By that measure, every single one of us except the Insane Clown Posse and KISS is closer to black than to white, and both of the named groups do half and half.

          • That's interesting; thank you. I'm curious where you got that figure because measurements with the NIST Spectral Tri-function Automated Reference Reflectometer (STARR) give much higher values. See this chart, which goes well into the ultraviolet (the left side of the figure is visible light).
            https://www.nist.gov/programs-... [nist.gov]

            Also interesting would be why does the average photo of african American people, in a color space with values from 0-255, has skin tones typically higher than 127? One possibl

            • Re: (Score:2)

              by Mal-2 ( 675116 )

              First off, it's actually 18%, not 12% -- which is why 18% gray cards are what they are. Sorry about that, I was simply misremembering. They're the best neutral density approximation of "white people". Second, I understand what you mean about this still being well above the midpoint in a log scale, but CCDs don't actually count in a log scale -- if you pull RAW data off the sensor, it's going to be linear. This puts 18% well below the median. We use log encoding because it uses the bits of color depth more e

              • Corbell explains pretty well why 18% gray is used.

                https://www.tonycorbell.com/bl... [tonycorbell.com]

                It's exactly halfway between black and white, in terms of optical brightness (which is, as I mentioned, logarithmic, not linear). It's the geometric mean, the average value of 0% - 100% (by product, not sum).

                So anyway nothing to do "white people", it's simply mathematically halfway between complete black and stark white.

                • Re: (Score:2)

                  by Mal-2 ( 675116 )

                  Yup. But using optical brightness is an artifact of both the way film used to work (still would if used), and how our own retinas work. The CCD data is just a photon count, until it saturates. In that domain, halfway is half brightness, one stop down from the maximum intensity, or 50% gray. Humans are way below 50%, all of them, unless they're wearing white makeup or zinc sunscreen or something. IR and UV are irrelevant for this.

                  Besides, the whitest white skin isn't neutral, it's rather biased toward the pi

                  • > just a photon count, until it saturates. In that domain, halfway is half brightness, one stop down from the maximum intensity, or 50% gray. Humans are way below 50%

                    Yeah, they'd be very roughly around 20% by photon count. (Very roughly)

                    > So "white" is wrong on that count too

                    Which is of course the entire point. Since there are no black people and there are no white people, can we all shut the fuck up all that already?

                    My daughter doesn't know the terms "black" and "white" as applied to people just yet,

                    • Stupid autocorrect. Her favorite color is teal.

                      Someday soon I'm going to have to explain to kiddo why some people call her "black", others call her "white" and both act like that's somehow important - when it's plain as day that she's pinkish tan.

                    • Re: (Score:2)

                      by Mal-2 ( 675116 )

                      Heh. I thought maybe she was seeing colorless green dreams sleeping furiously, and decided she prefers real colors.

  • How else to good 3D data on people who will not co (Score:3)

    by io333 ( 574963 ) on Thursday October 03, 2019 @08:06PM (#59268236)

    Many people refuse to utilize FaceID freely. At some point, in order to facilitate perfect tracking and total control of the populace, the few remaining non-compliant, such as homeless, anarchists and libertarians, will need to be forced to submit to the database one way or another.

    • Constitutionally Protected Don't Tread On Me (Score:5, Funny)

      by Anonymous Coward on Thursday October 03, 2019 @08:08PM (#59268246)
      You can have my face when you pry it from my cold, dead head.

    • Many people refuse to utilize FaceID freely. At some point, in order to facilitate perfect tracking and total control of the populace, the few remaining non-compliant, such as homeless, anarchists and libertarians, will need to be forced to submit to the database one way or another.

      Yeah so the whole whole facial recognition thing.

      There hasn't really been any grass-roots efforts to circumvent this that I've read about, outside of the people of Hong-Kong wearing masks. Which seems easy enough. Until they made it illegal to wear masks. ( and they never explicitly stated it was to get around facial recognition, i just assumed)

      But what if there was something even more subtle that someone could do to "break" the tech?

      Think - The covers you put on licence plates to prevent image capture.

    • Re: (Score:2)

      by AHuxley ( 892839 )
      Any 4th, 3rd and second world nation will have an ethics export pass the "math"/"art" project.
      Use a lot of fine print (important) about the results international's use, math use, art use, gov use, mil use, science use... but its all non identifying (important).
      1. Bring the equipment needed to their nation.
      2. Hire local experts and staff to share the results with (important).
      3. Take images of people who want to be part of the project. Say thank you (important) and pay a small amount of cash for the p

    • Re: (Score:2)

      by AmiMoJo ( 196126 )

      The fingerprint sensor is a major loss for the Pixel 4. I'd have been happy with an under-screen one. Face ID only is a massive downgrade.

      Why do companies downgrade their phones like this? Previously Google took away wireless charging and then brought it back. The headphone socket is gone from most high end models now.

    • Don't worry, it won't recognize you without your tinfoil hat.

  • try South America, Africa, Asia, the EU?
    Their nations "ethics" oversight would have allowed it.
    People would have taken part under a study project in their own nations for the same costs.
    Why only stay in the USA? When it all could have been done as a "study" in most parts of the world?

  • Google can simply download all the photos from Mugshots.com

    • And soon they'll be able to get their first sampling of a man with orange skin color from there! ...and don't even try to tell us you aren't a Trumptard after making that comment.

  • I thought we were seeing about one tech story per day about how facial recognition AI was biased toward recognizing white faces. That white faces were over-represented in the training datasets.

    Now Google tries to do something about that, and you are complaining?

  • These billionaires are so fucking cheap, giving a lousy 5 dollar gift certificate to improve their system which will make them more millions.

    They think the homeless are stupid and real desperate, and they don't care what kind of damage this will do in the future (wait until and employer uses this system and finds out somebody was homeless. More often then not, people get rejected on this basis alone).

    This is not much different then the "bumfights" videos where homeless people were given a c

  • We can consider two facts. Homeless men usually have heavy beards. They do this as they often have arrest warrants outstanding. It is difficult to identify people who have grown heavy beards. Next, we have the simple fact that homeless people tend to get way too much sunlight and also wind and cold burns. That means their skin is darker than the typical persons. Therefore, if you want information on the homeless or transient populations, you must take a harder look at people with darker skins. Keep in m
    • Often these warrants are for unpaid drinking tickets or tresspassing (which can mean loitering on the sidewalks). There are more of these than felony warrants. This ensures they have a criminal record which makes it much harder to get out of their situation.

  • "a contracting agency named Randstad sent teams to Atlanta explicitly to target homeless people and those with dark skin, often without saying they were working for Google, and without letting on that they were actually recording people's faces. "

    So why weren't they up front about their intent? It's because they knew what what they were doing was wrong, and most of these people would have objected. When a company does something like this, they are up to no good and they know it

    I hope Google

Slashdot Top Deals

Where there's a will, there's an Inheritance Tax.

Close