Twitter says it will investigate after users find its preview of a photo, with a Black person and a white person, more frequently displayed the white person
The first look a Twitter user gets at a tweet might be an unintentionally racially biased one. — Twitter said Sunday …
MashableRachel Kraus
Context & Ripple Effects
Users discovered that when a tweet contained photos of both a Black person and a white person, Twitter's automatic preview crop more often framed the white person — meaning an algorithm was silently deciding whose face represents a tweet before anyone reads it. The announcement that Twitter will investigate is the opening move in what related coverage shows became a multi-stage reckoning for the company's image pipeline.
What makes this worth tracking rather than a one-day story: Twitter's own subsequent disclosures turned a user report into confirmed findings — the company later shared results showing the cropping algorithm favored white people and women, and ultimately retreated from automated cropping altogether on mobile and the web.
First-order effects
Twitter's image-cropping model — built to surface 'salient' regions of a photo — comes under internal review, and the company faces immediate pressure to explain how a neutral-looking optimization produced racially skewed output.
Second-order effects
Under sustained criticism, Twitter pivots from defending the automation to constraining it: it commits to giving users more control over previews, then escalates to paying outsiders — including a bug bounty of up to $3,500 — to hunt bias the company's own testing missed.
Third-order effects
The endgame visible in the coverage is structural retreat: Twitter stops cropping photo previews on the web after doing the same on iOS and Android, suggesting platforms conclude that invisible automated defaults are harder to defend than simply showing users the whole image.
The trend: Opaque machine-learning defaults that rank and frame content are being forced toward user control, outside auditing, or outright removal as their hidden biases surface.
thanks to everyone who raised this. we tested for bias before shipping the model and didn't find evidence of racial or gender bias in our testing, but it's clear that we've got more analysis to do. we'll open source our work so others can review and replicate. https://twitter.com…
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://twitter.com/...
I care a lot more about how people and organizations *react* to a crisis than whether they got everything right from the start. We should hold Twitter accountable, but this is a good first reaction. https://twitter.com/...
Twitter's cropping algorithm is showing why instagram insisting on square images for its first few years was actually good user experience for both posting and viewing.
curious what “tested for bias” means. what were the tests, what were the metrics used, etc. i'd be pretty surprised if *any* company out there right now had rigorous bias testing in production. this is a pretty extraordinary claim so i look forward to the open sourcing https://tw…
This is a great response from Twitter - just a clear, simple statement and a commitment to check further. Feels very human. Look forward to seeing a follow-up https://twitter.com/...
i don't understand why Twitter has an algorithm to decide which parts of a photo to display in a tweet in the first place. Just display the full photo. https://twitter.com/...
when i tell you i am disgusted but not the least bit surprised. when live stand up comedy was a thing i was constantly cognizant of how my posts were treated by the algorithm https://twitter.com/...
“When attaching photos of Barack Obama and Mitch McConnell to tweets, Twitter seemed to exclusively highlight McConnell's face — Obama only popped up when Arcieri inverted the colors, making skin color a non-issue.” https://www.engadget.com/...
So even though @twitter actively builds campaigns and advertises on Black people It's design is PROGRAMATICALLY UNABLE to not subordinate Black people . https://twitter.com/...
I really, really want to build a test image dataset for QAing detecting blatant algorithmic biases before production (in a vein similar to the Big List of Naughty Strings) https://twitter.com/...
This looks really bad. It appears Twitter's photo preview algorithm focuses on the white man's face over the Black man's face despite their placement on the photos. This is what algorithmic bias looks like. https://twitter.com/...