Twitter’s image capture algorithm favors facial features of slim and younger people

Twitter’s image capture algorithm favors facial features of slim and younger people

New York, Aug. 11, 2021

Twitter’s AI algorithm that captures images favors people who are younger, slimmer and have lighter and/or warmer skin tones, according to a researcher at the EPFL School of Engineering.

Twitter had launched the contest last month, asking hackers and researchers to find biases in the „saliency algorithm“ designed to automatically display the most relevant section of images when scrolling through tweets. The contest is part of a broader effort in the tech industry aimed at developing artificial intelligence technologies that act ethically.

Bogdan Kulynych won the $3,500 grand prize for discovering bias in the algorithm during Twitter’s recent Bias Bounty competition.

Kulynych found that the algorithm favors lighter-skinned, slimmer and younger versions of an original image, as well as those with „stereotypically feminine facial features.“ This could have the effect of excluding minorities and perpetuating stereotypical beauty standards, he noted.

Other participants noted a bias against white-haired people and a preference for Latin over Arabic script.

Twitter had only recently admitted that its auto-cropping algorithm discriminated against men and black people.