Twitter has announced the result of an open competition regarding the photo-cropping system. The algorithmic bias of Twitter prefers young, light-skinned, and beautiful faces associated with the photo-cropping system.
Precisely, Twitter disabled the automatic photo-cropping system back in March last year. This is because Twitter users the experiments suggested that Twitter favored white faces over Black ones. After that, Twitter launched an algorithmic bug bounty to analyze and fix the issue easily.
The competition announced by Twitter has confirmed these earlier findings. The top placed entry showed that Twitter’s cropping algorithm was more biased and favored the faces that are slim, young, of light or warm skin color.
Smooth skin texture and a few feminine traits add to the point mentioned above. The second and third entries concentrated on the biases against people with white or grey hair. This suggests age discrimination and favors English scripts in the images over Arabic scripts.
What exactly is the algorithmic bias associated with the photo-cropping system of Twitter is?
In a presentation of these results at DEF CON 29, Rumman Chowdhury, the director of the META team praised the competitors for showing the real-life effects of such algorithmic bias. META Team is the team that studies Machine Learning Ethics, its transparency, and accountability.
The competition’s first place was bagged by Bogdan Kulynych, who won a huge prize of $3500. Undoubtedly, the approach used by Bogdan Kulynych is appreciable. Furthermore, she adds that it’s not only about the academic and experimental aspects when they think of bias.
Twitter’s photo-cropping algorithm prefers young, beautiful, and light-skinned faces https://t.co/IYHS7ckgL1 pic.twitter.com/GgLSbhNKfd
— The Verge (@verge) August 10, 2021
He is a graduate student at EPFL, a research university in Switzerland. He used StyleGAN2 as the AI program to figure out this algorithmic bias. This software generates a huge amount of realistic faces.
The bottom line of the content
He varied the skin color, distinguished between feminine and masculine traits, and slimness of it. He then fused these facial structures in Twitter’s photo cropping to find out what it preferred. Another special mention was given to one of the contenders. Vincenzo Di Cicco is his name, which is appreciated for his innovative approach.
He derived was that Twitter photo-cropping system didn’t favor the darker emojis over lighter skin tones. The third-place entry, which Roya Pakzad bagged, derived that the photo-cropping system of Twitter is also concentrated towards written features.
Pakzad’s work used the mechanism of English and Arabic scripts. She figured out that the system cropped the image in the meme and only highlighted the English text.
Image courtesy of Science and tech news/YouTube