Twitter is working to fix its biased automated image cropping

Twitter is fixing its automated image cropping issue since they say there’s ‘potential for harm’ though they have not found evidence of bias yet. A recent viral tweet showed how the cropping consistently picked lighter-skinned people regardless of the framing of the original image.

Recently, a number of users called out the image-cropping algorithm after a tweet pointed out that the cropping was consistently picking out lighter-skinned people regardless of the framing of the original image. (Pixabay)

Twitter is working on changing its automated image cropping feature after users complained that it was biased. The platform announced in a post titled ‘Transparency around image cropping and changes to come' that they were working on decreasing their reliance on automatic cropping so users can control how images appear in their tweets.

Twitter said that they are hoping by giving people more choices for image cropping and allowing them to preview what they will look like in the Tweet composer may reduce the risk of harm. “Going forward, we are committed to following the ‘what you see is what you get',” the platform explained.

This is currently a work in progress and it's not clear when these changes might be implemented.

The micro-blogging platform had first introduced the image cropping feature in 2018 and had said that it was brought in to highlight some “salient” areas of the image by using neural networks to predict where someone is more “likely to look”.

Recently, a number of users called out the image-cropping algorithm after a tweet pointed out that the cropping was consistently picking out lighter-skinned people regardless of the framing of the original image.

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

After the tweet went viral, many others started tweeting their own versions of the experiment including images of cartoon characters and even white and black dogs.

Twitter wrote in its blog that they had conducted their own tests and results have not shown any racial or gender bias though they did recognise “that the way we automatically crop photos means there is a potential for harm”. They added that they plan to run additional tests and explore ways to open-source their analyses so others can help keep them accountable.