Twitters Photo Preview Generator Shows Racial Bias

Photo: David Paul Morris/Bloomberg/Getty Images.

Over the weekend, you may have noticed a lot of white faces on your Twitter feed. Like, more than usual. That's because many users discovered an issue with the platform's photo preview generator. As in so many areas of tech and media, Twitter's photo preview generator appears to have a racial bias problem.

Twitter users demonstrated that if you tweet a photo that features a white person and a Black person, no matter how the photos are laid out, the platform's photo preview generator favors the white face, which means that previews more often show the white person instead of the Black person. Here are a couple of example tweets. Click on the photos to see them in full, and you'll understand the issue.

AdvertisementADVERTISEMENT

pic.twitter.com/QNZ8xCsEKj
— sina rawayama (@sina_rawayama) September 20, 2020

Trying a horrible experiment...

Which will the Twitter algorithm pick: Mitch McConnell or Barack Obama? pic.twitter.com/bR1GRyCkia
— Tony “Abolish (Pol)ICE” Arcieri 🦀 (@bascule) September 19, 2020

Interestingly, the issue first came to light when a Twitter user posted about facial recognition bias on another tech platform, The Verge reports. @colinmadland tweeted about Zoom not showing his Black colleague's face on video calls. The screenshots he shared on Twitter as evidence of the Zoom issue automatically populated his white face in the preview instead of his colleague's.

any guesses? pic.twitter.com/9aIZY4rSCX
— Colin Madland (@colinmadland) September 19, 2020

Geez...any guesses why @Twitter defaulted to show only the right side of the picture on mobile? pic.twitter.com/UYL7N3XG9k
— Colin Madland (@colinmadland) September 19, 2020

According to a statement from Twitter, the platform was previously unaware of the bias issue. "Our team did test for bias before shipping the model and did not find evidence of racial or gender bias in our testing," a Twitter spokesperson tells Refinery29. "But it’s clear from these examples that we’ve got more analysis to do. We'll continue to share what we learn, what actions we take, and will open source our analysis so others can review and replicate." Additionally, Twitter's chief data officer and chief technology officer have been interacting with users conducting various tests on the platform, and an expert at Carnegie Mellon has shared his independent analysis. We'll have to wait and see what changes come of the experiments.

Based on some experiments I tried, I think @colinmadland's facial hair is affecting the model because of the contrast with his skin. I removed his facial hair and the Black man shows in the preview for me. Our team did test for racial bias before shipping the model. pic.twitter.com/Gk33NQlGgB
— Dantley 🔥✊🏾💙 (@dantley) September 19, 2020

Here's another example of what I've experimented with. It's not a scientific test as it's an isolated example, but it points to some variables that we need to look into. Both men now have the same suits and I covered their hands. We're still investigating the NN. pic.twitter.com/06BhFgDkyA
— Dantley 🔥✊🏾💙 (@dantley) September 20, 2020

This is a very important question. To address it, we did analysis on our model when we shipped it, but needs continuous improvement.

Love this public, open, and rigorous test — and eager to learn from this. https://t.co/E8Y71qSLXa
— Parag Agrawal (@paraga) September 20, 2020

Results:
Face stimuli: Neutral
Categories: White-male / Black-male
White-to-Black ratio: 40:52 (92 images)
Code used: https://t.co/qkd9WpTxbK
Final annotation: https://t.co/OviLl80Eye @colinmadland @JeffDean @ChrSzegedy @AnimaAnandkumar
(Repost: last tweet had a typo) pic.twitter.com/k4x3zCZhxT
— Vinay Prabhu (@vinayprabhu) September 20, 2020