Author: Joy Buolamwini, Timnit Gebru
Publisher: MIT Media Lab
Publication Year: 2018
Summary: The following video discusses the Gender Shades project which evaluates the bias of 3 commercial artificial intelligence (AI) gender classification products (Microsoft, Face++, and IBM). Their focus and motivations are to show the need for increased transparency in the performance of any AI products/services that focus on human subjects. They found that the 3 companies had high accuracy overall but there were notable differences in error rates for specific groups. The AI products perform better on males and lighter subjects. They saw that all the products perform worst on darker females. IBM had the largest gap in accuracy for that comparison. The team published their results and IBM’s leaders responded within a day with a statement that they would make changes to their AI visual recognition.