Twitter’s image-cropping algorithm favors young and light-skinned faces, says study
Twitter has been rolling out features like testing alerts to stop the spread of fake news to prompts that appear before users post a potentially harmful reply. In another effort to uncover other harmful biases, the platform launched its first bounty challenge in July. On Sunday, it awarded the contest winner on findings in its photo-cropping algorithm.
Bogdan Kulynych, a graduate student from Switzerland, landed first place and will bring home $3,500 (approximately Php. 176,515) for his findings. His study showed that the “saliency” of an image could be more noticeable by “making the persons’ skin lighter or warmer and smoother; and quite often changing the appearance to that of a younger, more slim, and more stereotypically feminine person.”
Kulynch demonstrated how he uncovered the algorithmic biases by creating artificial faces and operating them through Twitter’s cropping system to discover which qualities the program was gravitating to.
1st place goes to @hiddenmarkov whose submission showcased how applying beauty filters could game the algorithm’s internal scoring model. This shows how algorithmic models amplify real-world biases and societal expectations of beauty.
— Twitter Engineering (@TwitterEng) August 9, 2021
The other contestants who placed second and third are AI group HALT and tech policy researcher Roya Pakzad, respectively. The former found out that the platform’s saliency system perpetuated marginalization, while the latter used bilingual memes to show linguistic diversity online.
Twitter also gave out two special awards – Vincenzo di Cicco received the most innovative award for exploring Emoji-based communication to see bias in the algorithm. On the other hand, an anonymous entrant’s work was most generalizable. They were able to alter the system’s preferences.
Software Engineering Director Rumman Chowdhury said, ‘The ability of folks entering a competition like this to deep dive into a particular type of harm or bias is something that teams in corporations don’t have the luxury to do.’
Banner: (L) Unsplash/stereophototyp, (R) GitHub/bogdan-kulynych
The post <b> Twitter’s image-cropping algorithm favors young and light-skinned faces, says study </b> appeared first on WE THE PVBLIC.
Source: we the pvblic
No comments