Skin Tone Bias and the Monk Skin Tone Scale

Google and Harvard sociologist Ellis Monk released the Monk Skin Tone Scale in 2022. It provides ten visual shades intended to represent a wider range of human skin tones for technology research and product development. The scale can support dataset annotation, image search evaluation and subgroup analysis where teams previously used coarse categories or the Fitzpatrick Skin Type scale as a proxy for visible skin color. The scale is openly available for research and product work; it is not a government standard or legal requirement. The MST addresses a measurement problem, not all causes of racial or color bias. A 2024 study comparing Fitzpatrick, Fenty and Monk scales surveyed 2,214 people in the United States and found the Fitzpatrick scale was perceived as less inclusive than the other two, especially by darker-skinned and historically marginalized participants. Google’s annotation research found that skin-tone labeling can be subjective and studied how annotator experience affects agreement. Skin tone in an image also depends on lighting, camera processing, display characteristics and the image itself. A ten-shade scale therefore provides useful common reference points, not perfect measurement or a substitute for self-identification. Researchers should distinguish skin tone from race, ethnicity, ancestry and medical response to ultraviolet exposure. The Fitzpatrick scale was originally designed around burning and tanning response, not as a universal image-skin-tone metric. Using it as a skin-color proxy can cause mismatch. The MST is likewise not designed to diagnose skin disease or assign a person to a racial group. When evaluating AI, teams should state what the label measures, who selected it, how image capture was standardized and how annotation uncertainty is handled.