How 10 skin tones will change Google’s approach to artificial intelligence

- Advertisement -


Over the years technology companies relied on what’s called the Fitzpatrick scale to classify skin tones for their computer vision algorithms. Originally developed for dermatologists in the 1970s, the system only includes six skin tones, possibly contributing to Well-documented failures of AI in identifying people of color. Google is now starting to introduce a 10-tone skin tone standard called Monk Skin Tone into its products.MST), from Google Search images to Google Photos, and more. Development can reduce bias in the datasets used to train AI in everything from healthcare to content moderation.

- Advertisement -

Google first announced plans to go beyond the Fitzpatrick scale last year; Internally, the project traces back to a summer 2020 effort to make AI “work better for people of color,” according to Thread on Twitter from Xango Eyeé, responsible artificial intelligence product manager at the company. Today Google I/O Conference, the company detailed just how far-reaching the impact the new system could have on many of its products. Google will also make MST open source, which means it can replace Fitzpatrick as the industry standard for evaluating the integrity of cameras and computer vision systems.

- Advertisement -

“Think about where human face images are being used, where we need to check the algorithm for fairness,” Aye says.

The Monk Skin Tone Scale is named after Ellis Monk, a sociologist at Harvard University who conducted decades a study of the impact of colorism on black lives in the United States. Monk created the scale in 2019 and worked with Google engineers and researchers to incorporate it into the company’s product development.

- Advertisement -

“The reality is that life chances, opportunities, all of these things are very much related to your phenotypic makeup,” Monk said in prepared comments in a video shown at I/O. “We can weed out these biases in our technology at a very early stage and make sure our technology works equally well for all skin tones. I think this is a huge step forward.”

An initial analysis by Monk and Google scientists last year of more than 3,000 participants found that people felt better represented by MST than by Fitzpatrick. The scale achieved representative results on par with skin tone scales of more than 40 shades, such as the one used by Rihanna’s cosmetics company Fenty Beauty. Google is continuing to work on checking the monk’s skin tone in countries such as Brazil, India, Mexico and Nigeria, according to a source familiar with the matter. More details are expected soon in a scientific research article.

Now the company will expand the use of MST. Google Images will offer the ability to sort makeup-related search results by skin tone based on a scale, and filters for people with more melanin will roll out to Google Photos later this month. If Google adopts a 10 skin tone scale across its product lines, it could affect the fair evaluation of the algorithms used in Google search results, Pixel smartphones, YouTube classification algorithms, Waymo self-driving cars, and more.

Colorism coded in technology can lead to unworthy results for people with dark skin, such as Google Photos mislabeling images of blacks like gorillas, racist soap dispensersand automatically generated stereotypical images. The algorithm that Google developed there was a lack of inclusion for people with dark skin to detect lesions. Autonomous driving systems have been found identifying dark-skinned people is much less reliable than white-skinned people. Most famous, 2018 research co-authored with former co-lead of the Ethical AI team, Timnit Gebru, concluded that facial recognition algorithms developed by large companies perform worse on dark-skinned women, detailed in the documentary. Coded offset.

Following Google dismissal of Gebru at the end of 2020Black in AI and Queer in AI groups committed to no longer receive funds from Googleand the company’s 2021 Diversity Report. found that his dropout rates are the highest among blacks and Native Americans.

Aye says further research is needed to confirm results indicating Monk’s preference for Fitzpatrick, or whether Monk’s approach leads to a fairer algorithms for dermatologists. But early results, especially for groups poorly represented in computer vision datasets, are promising.

.


Credit: www.wired.com /

- Advertisement -

Stay on top - Get the daily news in your inbox

DMCA / Correction Notice

Recent Articles

Related Stories

Stay on top - Get the daily news in your inbox