Twitter accused of bias over Barack Obama, Mitch McConnell photo

A simple experiment involving Mitch McConnell and Barack Obama has exposed a built-in bias in a key Twitter algorithm.

Twitter is trying to fix the algorithm that automatically crops image previews on the site after several experiments exposed an apparent bias.

The social media platform introduced a new algorithm to automatically detect what it thinks will be the most interesting part of a picture in order to display that part of the image in preview thumbnails.

Funnily enough, a Twitter user who sought to draw attention to the poor facial recognition of a different company highlighted the issue last week.

any guesses? pic.twitter.com/9aIZY4rSCX

— Colin Madland (@colinmadland) September 19, 2020

Colin Madland shared an image of videoconferencing app Zoom editing out the head of his black colleague because it didn’t recognise his face as a face.

When he sought to show Zoom’s bias, he also revealed Twitter’s, noting how the image preview defaulted to his white face when cropping the image preview.

This led several other users to test it out themselves, including in one “horrible experiment” involving two US politicians and the issue of who will become the next Supreme Court judge.

A 2009 photo of Ruth Bader Ginsburg greeting then-President Barack Obama. Picture: Saul Loeb / AFP

RELATED: Trump approves TikTok sale to avoid ban

RELATED: Kim K dumps Facebook for ‘spreading of hate’

The “experiment” involved US Senate majority leader Mitch McConnell, who became the subject of controversy following the death of US Supreme Court justice Ruth Bader Ginsburg last week.

In 2016, after the death of justice Antonin Scalia at the start of 2016, Mr McConnell prevented then-President Barack Obama’s nomination for successor being voted on until Mr Obama had left office, where the Donald Trump-nominated Neil Gorsuch was appointed.

Ms Bader Ginsburg died much closer to the election than Mr Scalia, but that hasn’t stopped Mr McConnell vowing to have her replaced before the election is held.

This has led to accusations of hypocrisy and protests outside Mr McConnell’s house.

Programmer Tony Arcieri decided to use Mr McConnell and former President Obama for his experiment to test Twitter’s algorithm, placing photos of the pair at opposite ends of an otherwise blank image to see which one Twitter’s algorithm picked.

RELATED: Reason Facebook staffer turned down $87k

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

RELATED: Why TikTok suicide video is hard to stop

RELATED: China’s fury over porn ‘rumour’

He also tried replicating the experiment but swapped what colour tie the two wore.

Twitter still favoured the picture of Mr McConnell.

It only selected an image of Mr Obama when the colours of the image had been inverted, and only did so for one of the images.

Twitter has promised to improve its image cropping algorithm. Picture: Olivier Douliery / AFP

The issue of algorithms struggling with facial recognition when it comes to differences in race and gender has long been known, but Twitter’s image-cropping algorithm doesn’t use facial recognition anymore. A Twitter spokeswoman said the company found no evidence of the new algorithm having issues before it was rolled out.

“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,” Twitter communications staffer Liz Kelley said on the platform.

“We’ll open source our work so others can review and replicate,” she promised.

“We did analysis on our model when we shipped it, but (it) needs continuous improvement,” Twitter chief technology officer Parag Agrawal said in a separate tweet.

In January 2018 Twitter announced it was moving away from facial recognition to an algorithm that automatically crops images based on what the most “salient” part of the image is.

“A region having high saliency means that a person is likely to look at it when freely viewing the image,” the company said when announcing the change.