Communications of the ACM
Twitter's Photo-Cropping Algorithm Confirms Inherent Biases
The winning entry used a GAN to generate faces that varied by skin tone, width, and masculine versus feminine features.
Twitter's open competition challenging users to find algorithmic bias in its photo-cropping system is complete, and the winning entries confirm experiments by Twitter users last year that suggested the algorithm favors white faces over black faces.
The first-place entry showed the algorithm favoring faces that are "slim, young, of light or warm skin color and smooth skin texture, and with stereotypically feminine facial traits." Second- and third-place entries showed that the system is biased against people with white or grey hair, suggesting age discrimination, and favors English over Arabic script in images.
"AI and machine learning are just the Wild West, no matter how skilled you think your data science team is," said Patrick Hall, a judge in the competition. "If you're not finding your bugs, or bug bounties aren't finding your bugs, then who is finding your bugs? Because you definitely have bugs."
From The Verge
View Full Article
No entries found