Blog
How many skin tones do we need to build unbiased AI and machine learning models for facial recognition and fashion?
How many skin tones do we need to build unbiased AI and machine learning models for facial recognition and fashion?
For the longest time, since 1975, Fitzpatrick skin tone scale has been used by AI researchers to represent a broad range of skin tones for AI development. Now, Google says that the Fitzpatrick scale could be biased in the favor of people with fair skin tones. 4 out of the 6 skin tones in the Fitzpatrick scale are fair skin tones. The new Monk skin tones have a broader range of skin tones - 12 in total - that represent a fuller spectrum of skin tones around the world. The claim is it is more unbiased and equitable for facial recognition and fashion machine learning and AI.