Home >> Featured >> Google is looking for a new way to categorise skin tones in order to reduce product bias.

Google is looking for a new way to categorise skin tones in order to reduce product bias.


Fitzpatrick Skin Type (FST) is a six-colour scale used by IT firms to categorise individuals and determine if goods function equally well across skin tones.

Google, owned by Alphabet Inc., told Reuters this week that it is working on a replacement for the industry-standard method for classifying skin tones, which a growing chorus of technology researchers and dermatologists say is insufficient for determining whether products are biassed against people of colour.

Fitzpatrick Skin Type (FST), a six-color scale used by dermatologists since the 1970s, is at dispute. It is currently used by tech firms to classify individuals and determine whether things like face recognition systems or wristwatch heart-rate monitors work equally effectively across skin tones.

FST, which has four categories for “white” skin and one each for “black” and “brown,” according to critics, ignores variety among persons of colour. During a federal technology standard meeting last October, researchers from the US Department of Homeland Security advised ditching FST for assessing face recognition because it fails to reflect the colour range in different populations.

For the first time and ahead of rivals, Google responded to Reuters’ queries regarding FST by saying that it has been discreetly seeking improved safeguards.

“We’re working on alternative, more inclusive metrics that may be helpful in the development of our goods,” the business said, refusing to provide specifics. “We’ll engage with scientific and medical professionals, as well as organisations working with communities of colour,” the company added.

The debate is part of a broader conversation about racism and diversity in the IT business, which has a predominantly white workforce than other industries like banking. As new technologies, frequently driven by artificial intelligence (AI), expand into sensitive and regulated sectors like healthcare and law enforcement, ensuring that technology functions effectively for all skin colours, ages, and genders is becoming more important.

Companies are well aware that their goods may be defective among populations that are underrepresented in research and testing data. FST’s restricted scale for darker skin is a source of worry since it may lead to technology that, for example, works for golden brown skin but not for espresso red tones.

Palettes much richer than FST are available in a variety of goods. Crayola released 24 skin tone crayons last year, while Mattel Inc’s Barbie Fashionistas dolls cover nine tones this year.

For Google, the problem is far from academic. When the firm revealed in February that certain Android phones’ cameras could detect pulse rates with a fingertip, it claimed that readings would be off by 1.8 percent on average, regardless of whether users had light or dark skin.

Skin type will not significantly influence the results of a function for filtering backdrops on Meet video conferences, nor of a forthcoming online service for detecting skin problems, unofficially called Derm Assist, according to the firm.

The six-tone FST was used to get such findings.


The scale was developed by Dr. Thomas Fitzpatrick, a dermatologist at Harvard University, to customise UV radiation therapy for psoriasis, an inflammatory skin disease. He assigned Roman numbers I to IV to the skin of “white” individuals based on how much sunburn or tan they got after spending time in the sun.

A decade later, type V for “brown” complexion and VI for “black” skin were introduced. The scale is still used to evaluate sunscreen products in the United States, and it is a prominent dermatological benchmark for determining cancer risk and other factors. Some dermatologists believe the scale is a poor and overused measure of care, and that race and ethnicity are often confused with it.

Dr. Susan Taylor, a University of Pennsylvania dermatologist who established Skin of Color Society in 2004 to encourage research on disadvantaged groups, stated, “Many people would think I have skin type V, which seldom to never burns, yet I burn.” “To claim I am type V based on my skin tone does me a disservice.”

Until recently, technology firms were unconcerned. In 2014, Unicode, the industry organisation in charge of emojis, cited FST as the rationale for adding five skin tones beyond yellow, claiming that the scale had “no negative connotations.”

The use of FST for assessing AI was popularised by a 2018 research named “Gender Shades,” which showed that face analysis algorithms often misgendered individuals with darker complexion. FST was characterised in the study as a “beginning point,” but scientists from subsequent studies told Reuters that they utilised the scale to remain consistent.

“It fulfils its goal to assist us detect red flags as a first measure for a very young industry,” said Inioluwa Deborah Raji, a Mozilla fellow focusing on AI auditing.

FST “obviously does not capture the variety among brown and black skin tones,” Facebook Inc researchers said in an April study evaluating AI for identifying deepfakes. Despite this, they provided recordings of 3,000 people to be used in AI system evaluations, with FST tags added based on the ratings of eight human raters.

The raters’ judgement is critical. AnyVision, a facial recognition software company, provided raters famous examples last year, classifying former baseball great Derek Jeter as a type IV, model Tyra Banks as a type V, and rapper 50 Cent as a type VI. AnyVision told Reuters that it agreed with Google’s intention to reconsider its usage of FST, while Facebook said that it is open to better methods. When working on health-related sensors, Microsoft Corp, Apple Inc, and Garmin Ltd use FST as a reference.

However, doctors from the University of California San Diego, inspired by the Black Lives Matter social equality movement, argued in the journal Sleep last year that the use of FST may be feeding “false promises” regarding heart rate readings from smartwatches on darker skin. FST’s flaws were recognised by Microsoft. Apple claims to test on people of different skin tones using a variety of methods, with FST being one of them only on rare occasions. Garmin claims that the results are accurate as a result of extensive testing.

Victor Casale, the founder of Mob Beauty and a consultant for Crayola on the new crayons, said he created 40 foundation colours, each differing by approximately 3%, or enough for most people to tell the difference. “You can’t simply have six,” he said, adding that colour accuracy on electronics suggests that standards should have 12 to 18 tones.

Leave a Reply

Your email address will not be published. Required fields are marked *



Check Also

The Sleekest 14-inch Chromebook is Asus Flip C436

  The newest 10th Gen Intel Core CPU powers Chromebook Flip and customers may select ...

Apple FaceTime is now available on Android and Windows devices.

  FaceTime calls will be accessible to Android and Windows users through the web browser, ...