The study by three of its machine learning researchers was conducted
after user criticism last year about image previews in posts excluding black
people's faces.
It found an 8% difference from demographic parity in favour
of women, and a 4% favour towards white individuals.
The paper cited several possible reasons, including issues
with image backgrounds and eye colour, but said none were an excuse.
“Machine learning based cropping is fundamentally flawed
because it removes user agency and restricts user's expression of their own
identity and values, instead imposing a normative gaze about which part of the
image is considered the most interesting,” the researchers wrote.
To counter the problem, Twitter recently started showing
standard aspect ratio photos in full — without any crop — on its mobile apps
and is trying to expand that effort.
The researchers also assessed whether crops favoured women's
bodies over heads, reflecting what is known as the “male gaze,” but found that
does not appear to be the case.
The findings are another example of the disparate impact
from artificial intelligence systems including demographic biases identified in
facial recognition and text analysis, the paper said.
Work by researchers at Microsoft Corp and the Massachusetts
Institute of Technology in 2018 and a later US government study found that
facial analysis systems misidentify people of colour more often than white
people.
Amazon Inc in 2018 scrapped an AI recruiting tool that
showed bias against women.