Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification

Our summary

The landmark audit that measured commercial classifiers across skin-type and gender subgroups, finding error rates up to 34.7% for darker-skinned women versus 0.8% for lighter-skinned men.

Why it matters

The model for bias auditing as a population-level statistic with subgroup breakdowns — disparity evidence that holds up where a single anecdote would not.

Cited by these methods

Cite this