Fitzpatrick Scale
The Fitzpatrick scale is a clinical six-band classification of human skin types developed in 1975. The bands are defined by response to UV exposure:
- I: Very fair. Burns easily, never tans.
- II: Fair. Burns easily, tans minimally.
- III: Medium. Burns moderately, tans gradually.
- IV: Olive. Burns minimally, tans well.
- V: Brown. Rarely burns, tans easily.
- VI: Deep. Rarely burns; deeply pigmented baseline.
In AI personalization, the Fitzpatrick scale is the primary axis for disaggregated bias evaluation. Skin diagnostic accuracy is reported per-band; meaningful AI personalization commitments include floor-level performance guarantees per band (e.g., "no band below 85% accuracy on the diagnostic task").
Historic ML systems have under-performed on Fitzpatrick V-VI due to under-represented training data and lighting-handling assumptions optimized for lighter skin. Closing those gaps is an ongoing commitment for any vendor that takes responsible AI seriously.
Related: Skin Diagnostics, Computer Vision Skin Analysis.