Last month saw a number of users tweet and complain about how Twitter’s image cropping tech seemed to be biased towards fair-skinned people. These Twitter users posted several examples of how, in an image featuring both a black and white person, the preview on the timeline more frequently displayed the white person.
Trying a horrible experiment…
Which will the Twitter algorithm pick: Mitch McConnell or Barack Obama? pic.twitter.com/bR1GRyCkia
— Tony “Abolish (Pol)ICE” Arcieri 🦀 (@bascule) September 19, 2020
In response to this, Twitter stated that while its algorithm ad been tested for bias before, further investigation was going to be carried out.
This was then followed by a statement yesterday with the company saying it is working on how to give users more control over how the final image will look like.
“We are prioritizing work to decrease our reliance on ML-based image cropping by giving people more visibility and control over what their images will look like in a Tweet. We’ve started exploring different options to see what will work best across the wide range of images people Tweet every day.”
Twitter went further to explain that its machine learning algorithm relies on saliency. This is a measure that predicts where people might look first in an image. Apparently, the system was tested for bias between a different range of demographic groups, after which the team found no bias.
However, the company stated that the way it automatically crops photos means there is a “potential for harm”.
The company didn’t exactly explain the way it intends to implement the new image cropping system as it is exploring various ways. The algorithm will be redesigned in such a way that what you see in the image composer is what you get to view in the final posted image on the timeline.