An algorithm Twitter uses to decide how photos are cropped in people’s timelines appears to be automatically choosing to display the faces of white people over people with darker skin. The bias was discovered by Twitter users in recent days who posted photos of people with light and dark faces who found that priority was given to white faces.
JFC @jack https://t.co/Xm3D9qOgv5
— Marco Rogers (@polotek) September 19, 2020
As a result of the apparent bias, a Twitter spokesperson said the company plans to open source an evaluation of the algorithm.
Twitter scrapped its face detection algorithm for a saliency detection algorithm to predict the most important part of an image in 2017. A Twitter spokesperson said today that no race or gender bias was found in evaluation of the algorithm before it was deployed “but it’s clear we have more analysis to do.”
Twitter engineer Zehan Wang tweeted that bias was detected in 2017 before the algorithm was deployed but not in significant levels. VentureBeat reached out to Twitter for additional details about the 2017 evaluation and what steps will be taken to reassess the algorithm. This story will be updated when we hear back.
Algorithmic bias researcher Vinay Prabhu created a methodology for assessing the algorithm. Results will be shared via the recently created Twitter account Cropping Bias.
I wonder if Twitter does this to fictional characters too.
Lenny Carl pic.twitter.com/fmJMWkkYEf
— Jordan Simonovski (@_jsimonovski) September 20, 2020