A student researcher has found that Twitter’s image-cropping algorithm prefers faces that are slim, young and light-skinned
A graduate student at Switzerland’s EFPL university has discovered a bias in Twitter’s image-cropping ‘saliency’ algorithm.
Bogdan Kulynyc proved that the algorithm preferred faces that are light-skinned, slim and young. Twitter’s saliency algorithm decides the most interesting part of an image to crop for preview.
The researcher tested how the software responded to AI-generated faces
Kulynyc found that by he could manipulate the algorithm to be prefer faces by “making the person’s skin lighter or warmer and smoother; and quite often changing the appearance to that of a younger, more slim, and more stereotypically feminine person”.
He achieved this by using an AI face generator to create artificial people with varying features. He was then able to run the images through the algorithm to see which faces the software preferred.
“We should not forget that algorithmic bias is only a part of a bigger picture. Addressing bias in general and in competitions like this should not end the conversation about the tech being harmful in other ways, or by design, or by fact of existing,” said Kulynyc.
“A lot of harmful tech is harmful not because of accidents, unintended mistakes, but rather by design”
Bogdan Kulynyc
“This shows how algorithmic models amplify real-world biases and societal expectations of beauty”
Twitter’s director of software engineering and head of AI Ethics Rumman Chowdhury says the findings “showcased how applying beauty filters could game the algorithm’s internal scoring model.
“We create these filters because we think that’s what ‘beautiful’ is, and that ends up training our models and driving these unrealistic notions of what it means to be attractive.”
We should not forget that "algorithmic bias" is only a part of a bigger picture. Addressing bias in general and in competitions like this should not end the conversation about the tech being harmful in other ways, or by design, or by fact of existing.
— Bogdan Kulynych (@hiddenmarkov) August 9, 2021
Twitter’s “algorithmic bug bounty”
The findings mark the conclusion of Twitter’s first “algorithmic bug bounty”. The event was part of an in-house competition at the DEF CON security conference in LA.
Twitter rewarded the student $3500 for his efforts.
Last year, Twitter came under fire for cropping out Black faces
This comes after and incident last year, where the tech giant found that the preview crop was more likely to hide Black faces.
Twitter’s director of software engineering Rumman Chowdhury said the findings illustrated that “how to crop an image is a decision best made by people”.
