L8 · Safety, Alignment & Society
Bias You Can Measure
Split a score by group, read what the split shows, and say what it cannot show.
In 2018 Joy Buolamwini and Timnit Gebru tested three commercial systems that guessed gender from a face. On lighter-skinned men, IBM's was wrong 0.3 percent of the time. On darker-skinned women, the same system was wrong 34.7 percent of the time.