Follow Slashdot blog updates by subscribing to our blog RSS feed

 


Forgot your password?
Close
typodupeerror
×
Google Technology

Google Searches for New Measure of Skin Tones To Curb Bias in Products (reuters.com) 72

Posted by msmash from the moving-forward dept.
Google is developing an alternative to the industry standard method for classifying skin tones, which a growing chorus of technology researchers and dermatologists says is inadequate for assessing whether products are biased against people of color. From a report: At issue is a six-color scale known as Fitzpatrick Skin Type (FST), which dermatologists have used since the 1970s. Tech companies now rely on it to categorize people and measure whether products such as facial recognition systems or smartwatch heart-rate sensors perform equally well across skin tones.

Critics say FST, which includes four categories for "white" skin and one apiece for "black" and "brown," disregards diversity among people of color. Researchers at the U.S. Department of Homeland Security, during a federal technology standards conference last October, recommended abandoning FST for evaluating facial recognition because it poorly represents color range in diverse populations. In response to Reuters' questions about FST, Google, for the first time and ahead of peers, said that it has been quietly pursuing better measures.
This discussion has been archived. No new comments can be posted.

Google Searches for New Measure of Skin Tones To Curb Bias in Products More

Google Searches for New Measure of Skin Tones To Curb Bias in Products

Comments Filter:

  • He he he. (Score:2, Insightful)

    by gurps_npc ( 621217 )

    What idiot thought that 6 skin colors should be 4 shades of White plus Brown and Black. Cause we all know there are four different kinds of white people and only one kind of black people and one kind of brown. American Indians are the same as Asians and Hispanics, right?

    • Good luck telling the difference between Hispanic and Filipino by skin color.

      • Re: (Score:2)

        by fazig ( 2909523 )
        TFS says nothing about distinguishing between ethnicity though. So maybe that was never the goal to begin with.
        Still, a neutral network might actually be able to distinguish between ethnicity based on skin tone if it's been given enough data to train with, if that was the goal.
        To be perfectly honest, I doubt that it's going to be accurate, but I've been amazed by what patterns machine learning has discovered (even though still obscured to us) a number of times so far.

        But in general yeah, I'd suggest to u

    • Re:He he he. (Score:5, Informative)

      by K. S. Kyosuke ( 729550 ) on Friday June 18, 2021 @11:36PM (#61500620)

      What idiot thought that 6 skin colors should be 4 shades of White plus Brown and Black.

      Somebody named Fitzpatrick. I guess it's kind of obvious why.

      Although considering that the scale was apparently developed for assessing UV therapy doses, it's also kind of obvious *why* there are four shades of white in particular.

    • In Minneapolis the police are well trained and don't go in for this BS. All perps will get totally unbiased, even handed treatment.
    • Google's original color model: eggshell white, ivory white, off-white, passing-white, pat them down brown, and send them back black.

    • Re:He he he. (Score:5, Informative)

      by bsolar ( 1176767 ) on Saturday June 19, 2021 @06:04AM (#61501054)

      These are the 6 types in the FST [wikipedia.org]:

      * Type I (scores 0–6) always burns, never tans (palest; freckles)
      * Type II (scores 7–13) usually burns, tans minimally (light colored but darker than fair)
      * Type III (scores 14–20) sometimes mild burn, tans uniformly (golden honey or olive)
      * Type IV (scores 21–27) burns minimally, always tans well (moderate brown)
      * Type V (scores 28–34) very rarely burns, tans very easily (dark brown)
      * Type VI (scores 35–36) never burns (deeply pigmented dark brown to darkest brown)

      First of all, only the first two types can be considered "white". Furthermore, the main distinction is not the skin color itself, but how the skin reacts to sun exposure. The scale is mainly intended for use in dermatology to estimate the response of the skin to the sun and in that context is still a very useful tool.

      Basically, that "idiot" as you call him was a dermatologist and developed a very useful tool in the field of dermatology. The reason you are calling him an "idiot" is due to your ignorance and gullibility.

      You didn't know the field, the context, have taken a misleading article at face value, failed to double-check its credibility and jumped at a misguided, shortsighted conclusion.

      Basically before calling someone an idiot make sure the idiot is not you and before attacking something you think it's racist make sure racism is not in the eye of the beholder first.

      • First of all, only the first two types can be considered "white". Furthermore, the main distinction is not the skin color itself, but how the skin reacts to sun exposure. The scale is mainly intended for use in dermatology to estimate the response of the skin to the sun and in that context is still a very useful tool.

        Thank you for adding information to this. Indeed that summary wasn't very good. I wonder if the summary was conflating "whiteness" with "European / Caucasian -ness". Type III / Type IV could potentially be skin types found in Mediterranean countries (particularly Greece, Cyprus, Turkey) who are generally clumped in with European nations more than Asian or African nations. If you clump those in that way you could claim 4 shades of "white" skin.

        I wouldn't defend that idea, but it is one possible expl

      • "Type VI (scores 35–36) never burns (deeply pigmented dark brown to darkest brown)"

        And yet I'm seeing more "type VI" people in sunscreen commercials. Interesting.

      • First of all, only the first two types can be considered "white".

        This statement makes no sense. Lots of white people tan well, especially in the southern regions.

    • Isn't it more of a continuum and they've just arbitrarily said "white starts here"?

    • Re:He he he. (Score:4, Insightful)

      by Geoffrey.landis ( 926948 ) on Saturday June 19, 2021 @07:36AM (#61501224) Homepage

      What idiot thought that 6 skin colors should be 4 shades of White plus Brown and Black. Cause we all know there are four different kinds of white people and only one kind of black people and one kind of brown. American Indians are the same as Asians and Hispanics, right?

      Reading the summary, the scale was devised by dermatologists. Presumably their interest was in how much pigment was in the skin to protect against ultraviolet. They didn't intend it to distinguish what "kind" of black or what "kind" of brown the skin tone was, only how sensitive the skin was to UV, and six levels was enough for that determination.

      There's a problem when you use a measurement intended for one purpose for something completely different.

    • Yeah, as TFS presents it the scale sounds pretty stupid. From bsolar's clarification it makes more sense. The real problem then is twofold: (1) using a dermatological sunburn scale for cultural emblems such as emoji is bound to introduce bias (there's a greater sun exposure difference between shades of what we culturally consider "White", resulting in the appearance of more "White" than "Non-white" shades); (2) limitation of conceptual treatment of skin shades to mere UV exposure reactions. Clearly skin sh

    • Re: (Score:1)

      by Anonymous Coward

      If you are looking to divide skin tones by brightness, there's way more diversity (and therefore then need for way more divisions) in lighter color skin than darker.

      The article is wrong, of course- there are three for light skin (pale white, fair skin, darker white skin) and three for darker skin (light brown, brown, black).

      Obviously, you could subdivide it more, and probably people will. But there is legitimately much less difference between, say, an Asian and a South American in skin tone, than there is

    • Just looking at Wikipedia, it looks more like two shades of white plus four shades progressing from light brown to darkest black.

      I totally agree that this is inadequate, but it's not quite as bad as the article suggests.

  • So what am I black in summer and white in winter and classified by who the fuck as that. My skin is a colour, generally speaking from brown to white, my face to my arse, I am not a big on nudity (skin cancer vs vitamin D). If you classify people by race based upon skin colour, you are the fucking racist, fuck off with your skin colour bullshit.

    So wheres the fucking catergory for people who tan easy, who are we just classified as brown in summer and white in winter, yeah, go fuck yourselves arseholes. My sk

    • So wheres the fucking catergory for people who tan easy, who are we just classified as brown in summer and white in winter, yeah, go fuck yourselves arseholes. My skin is a colour and it varies, I am not the colour of my skin.

      So you are somewhere around type 3 or type 4 on the Fitzpatrick scale. Are you too lazy to Google the subject of the article?

      • So wheres the fucking catergory for people who tan easy, who are we just classified as brown in summer and white in winter, yeah, go fuck yourselves arseholes. My skin is a colour and it varies, I am not the colour of my skin.

        So you are somewhere around type 3 or type 4 on the Fitzpatrick scale. Are you too lazy to Google the subject of the article?

        Yes. This is /. We don't do that here.

    • Re:I Tan Easy (Score:5, Informative)

      by jonbryce ( 703250 ) on Saturday June 19, 2021 @02:00AM (#61500768) Homepage

      That is exactly what the Fitzpatric scale does

      I am type 1, don't tan at all, just burn, found in about 10% of the population of Ireland, 8% of Scotland, and small percentages everywhere, including in groups with the darkest skin colours.

      Most people native to Northern Europe are type 2 so they tan a bit.

      Most people native to Southern Europe and the mediterranean areas are type 3, and so on.

      You don't change skin when you are tanned vs not tanned.

    • Re: (Score:1)

      by fazig ( 2909523 )
      That's part of the problem with the old Fitzpatrick scale being garbage that needs to be replaced.

      Also, this is not that much about race, even though a lot of people around here seem to think so.
      In the article they talk about smart watches measuring heart rate. And they try to do that with an optical sensor provides data to a algorithm that was trained to correlate pictures of skin with heart rate. I presume it does so by correlating the contrast data of the previous pictures with the current picture fro

      • >> I tan Easy

        > That's part of the problem with the old Fitzpatrick scale being garbage that needs to be replaced.

        So you're saying the Fitzpatrick scale is garbage that needs to be replaced because it doesn't recognize that some people tan easily. Is that correct?

        This is the Fitzpatrick scale:

        Type I (scores 0â"6) always burns, never tans (palest; freckles)
        Type II (scores 7â"13) usually burns, tans minimally (light colored but darker than fair)
        Type III (scores 14â"20) sometimes mild b

        • Re: (Score:2)

          by fazig ( 2909523 )
          The Fitzpatrick scale is a good indicator for what kind of UV protection you'll need if you want to expose your skin to sunlight for prolonged times without damaging your skin.
          But for what google intends to do with it according to what the article says, it is about as a good tool as a hammer for tightening a screw.
          If you try real hard, yeah you'll kinda get some results. But as far as choosing the right tool goes, you chose garbage. And you'd better replace that garbage with a tool that is better suited t

    • Whether you identify as black or white is up to you. Nobody has the right to change your pigmentation identity. No one has the right to stop you using the reparations restroom.

  • Given how shitty the lighting and color balance are in your typical snapshot, GIGO might apply here. I imagine it's inherently more difficult to analyze color than the shapes and distances between facial elements.

    I'm not sure why they would care about this particular scale either. Are the algorithms mapping the photos to one of the 6 grades on the scale? If so, why? They can't be quantizing other elements of the face that coarsely.

    • the whole thing is stupid

      the color has a hue, a continuous numeric value, a saturation, a continuous numeric value, and a brightness, a continuous numeric value...

      So after they quantize the very continuous state space that is skin color, to a whole 5 values, they protest about how not useful the quantized value is.... WHAT THE FUCK?

      • So after they quantize the very continuous state space that is skin color, to a whole 5 values, they protest about how not useful the quantized value is.... WHAT THE FUCK?

        Five buckets is all anybody needs, says dumbass that doesn't know a histogram from an Instagram.

      • Re: (Score:2)

        by fazig ( 2909523 )
        The whole thing could be a bit more complicated than it's apparent on the surface.

        Hues with a low 'value' are indeed a problem both for digital sensors, which have to deal with a lot of noise phenomenon when the amount of photons hitting sensors are low and also on how we digitally represent the hues.

        Youtuber Tom Scott has done a pretty easily comprehensible (my opinion) breakdown of the phenomenon: https://www.youtube.com/watch?... [youtube.com]

        And having worked with 3D art myself during the last 2 years, I can o
  • And if you do, avoid "Octa" and "Quad" as prefaces. https://www.youtube.com/watch?... [youtube.com]
  • Outside of some lighting and or contrast issues with current cameras, the main problem seems to be just basic training sets. Here’s a crazy thought that must never have been thunk before - why not train and test against diverse groups of people in the first place? There might be some excuses for medical equipment, but facial recognition? Doubly so when it’s a fairly large chunk of the user base.

  • I had no idea they used such basic groups for skin colour. I assumed they had a whole colour range to pick from.

    • At first they trained neural nets with no concern for bias but recently there are activist groups who are shaming them on bias related issues. Now they are looking into making a test dataset in order to evaluate and debias their models. The 5 color skin tone was just an old system that's useless today.

  • Wrong technology (Score:5, Interesting)

    by NotEmmanuelGoldstein ( 6423622 ) on Saturday June 19, 2021 @02:30AM (#61500808)
    I heard of the Von Luschan's chromatic scale with its 36 tones, 10 years ago and it's 100 years old. The Fitzpatrick scale is for measuring sunburn.
  • With all that our society has learned in the approximately 50 years since the FST was introduced, it didn’t occur to Google that the entire [expletive] proposition is flawed?

    Newsflash, Google: if you cannot develop a solution that treats every human being equally, your solution is racist and must not be used.

    There, problem solved.

    The trap you fell in to, Google, is the idea that you must have a working solution, no matter what. But the problem is, as we are learning, that when you develop any
    • TG we can trust judges to reach fair and unbiased sentencing.

    • Google are essentially an advertising agency. Their bread & butter is discriminating consumers into demographic groups so that they can sell 'targeted advertising' to their clients. Therefore, racism, sexism, misogyny, classism, etc., are in their very nature & impossible to mitigate.

      Have you ever had a conversation with anyone who works in advertising or ever had to listen to them pontificate about what they know about 'people'? They're despicable human beings.

  • Lying for political points? (Score:5, Interesting)

    by bradley13 ( 1118935 ) on Saturday June 19, 2021 @05:13AM (#61500972) Homepage
    Google "Fitzpatrick Skin Type" and you find that there are not four white types, one brown and one black. The six types are:
    • - Pale white skin
    • - White skin
    • - Light brown skin
    • - Moderate brown skin
    • - Dark brown skin
    • - Deeply pigmented dark brown to black skin

    Starting with the "light brown", sample pictures include people of apparent Mediterranean or Hispanic descent. Your Northern Europeans are pretty much confined to the first two categories. There may be reasons to replace this scale - one would have to ask the experts in the area. But TFA is simply lying. The interesting question is: why?

    • Re:Lying for political points? (Score:4, Insightful)

      by Geoffrey.landis ( 926948 ) on Saturday June 19, 2021 @07:44AM (#61501236) Homepage

      Google "Fitzpatrick Skin Type" and you find that there are not four white types, one brown and one black. The six types are:

      I just did that, and I find that it is a scale to assess how sensitive skin is to the sun, and whether it burns or tans.

      Which seems reasonable for a scale developed by dermatologists. Not so reasonable for facial recognition.

      ...

    • Re: (Score:3)

      by AmiMoJo ( 196126 )

      Because people from Southern Europe are considered white in some contexts, like if they were banned from sitting at the front of the bus at one point.

      Even scales with more tones tend to be biased towards lighter skin. If you look at cosmetics most of the range available in the West is useless for most Black people, they have very little choice.

    • From the article:

      "Former baseball great Derek Jeter as a type IV, model Tyra Banks a V and rapper 50 Cent a VI."

      Most of us probably know what Tyra Banks looks like.
      Jeter's mom has a light complexion, his dad is closer to 50 Cent, and he's right in the middle.

      So on the scale that measures sensitivity to UV light (which is mostly controlled by melatonin levels), there are two that would probably be called "white people" - pale freckled redheads from Ireland vs tanner Spanish people. One is light brown / olive

    • But TFA is simply lying. The interesting question is: why?

      The answer is usually either "because the reporter is a dumb idiot who didn't research the topic of his article correctly" or "hidden agenda, possibly paid anonymously by a shadow group of said hidden agenda".

  • Can't we all just get along without splitting hairs to the Planck length? Jesus was middle brown BTW.

    • I love pointing that out to religious folks. Not many fair skinned blue eyed blonde middle easterners walking around. Jesus was certainly black.

  • Phrenology (Score:2, Interesting)

    by VeryFluffyBunny ( 5037285 )
    Let's face it. Google are using shitty, racist, mysoginist data to build their models from, i.e. from shitty, racist, mysogynist culture in the USA. Remember the shitty, racist, mysoginist AI chatbots they & MS tried not so long ago? What this amounts to is polishing a turd. AI & data-based models of humanity are essentially cultural mirrors that show us who we really are as a society. Don't like what you see? Then change. The models aren't lying.
  • So they're looking for a way to distinguish skin tone, so they can make assumptions about people and stereotype them based on that skin tone.
    They want to be racist.
    Because they have so far failed at developing AIs that give the racial outcomes they desire.
    This won't end well.

    • The best way to categorize people is by the type of food they eat. If they eat junk food, they're normal. If they eat fancy crap, they're hipsters.

  • How about shove your racist terms up your ass and stop defining people in obviously racist ways, and instead use "bright" and "dark" skinned if that's how you want to refer to someone according to their skin color? Generally there's no reason to even do that in the first place.

  • ...to make this kind of research illegal.

    Of course it can be used for all sorts of good-ish things. That's nice. It'll obviously be used for far more terrible things. The obvious one being to simply stir-up trouble and chaos.

    I'd support making it illegal to have any computerized system ever distinguish between humans by anything more than height and weight (for roller-coasters, bungee jumping, and parachutes).

    Just another fight waiting to happen.

  • My skin is kind of a pink-ish, almond-like hue, you insensitive clod!

  • Maybe they could find what they are looking for!

Slashdot Top Deals

Time is the most valuable thing a man can spend. -- Theophrastus

Close