Some features are preferred by Twitter's image editing algorithms

The AI of Twitter's image editing algorithm prefers young and thin women.

Bogdan Kulynych, a doctoral student at the École polytechnique fédérale de Lausanne (EPFL) Laboratory of Security and Privacy Engineering, has developed a new method to prove that Twitter's image editing algorithm prefers certain facial features. She found that the software used by the microblog favours young and thin women with light, warm skin tones when creating preview photos. This confirms that the software used by the social platform follows a certain ideal of beauty. For this discovery, Kulynych received a reward of $3500 USD.

Image source:

He noticed the phenomenon after another PhD student, Colin Madlan, reported it. He noticed that Twitter's automatic photo-editing software had automatically put his face in the foreground of a photo of him and his colleague because he was white. And the part of his colleague's face was removed completely. Comments from other Twitter members confirmed that it was not a mistake and that it was a one-off incident.

The microblog indicated that it would investigate the phenomenon and improve its algorithm. After a short while, the company shut down the solution, and announced a contest offering a $3500 USD prize to anyone, who could find new ways to prove that its AI was disadvantaging certain groups. Kulynych was finally awarded the prize at the DEF CON IT security conference in Las Vegas.

3 columns
2 columns
1 column
Join the conversation now