The tech giant uses a system called neural network to automatically crop photo previews before you can click on them to view the full image.
This focuses on the area identified as the “salient” image region, where it is likely a person would look when freely viewing an entire photo.
Trying a horrible experiment...
Which will the Twitter algorithm pick: Mitch McConnell or Barack Obama? pic.twitter.com/bR1GRyCkia— Tony "Abolish ICE" Arcieri 🦀🌹 (@bascule) September 19, 2020
But tests by a number of people on the platform suggest that the technology may treat white faces as the focal point more frequently than black faces.
One example posted online shows American politician Mitch McConnell and Barack Obama, with the system favouring Mr McConnell in its preview over the former US president.
Meanwhile, another person tried with Simpson cartoon characters Lenny and Carl – the latter who is black – with Lenny appearing to take preference.
A third user even tried with dogs, resulting in a white dog in the prime preview position over a black dog.
Twitter chief design officer Dantley Davis tweeted that the firm is investigating, saying that there are “some variables that we need to look into”.
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
Advertisement— Dantley Davis (@dantley) September 20, 2020
Mr Davis posted an image of a black man and a white man in a rough test of his own, in which the black man was favoured in this case but only once both men’s hands had been cropped out of view.
“It’s not a scientific test as it’s an isolated example, but it points to some variables that we need to look into,” he said.
“Both men now have the same suits and I covered their hands. We’re still investigating the NN.”
Chief technology officer Parag Agrawal also tweeted about the matter, saying that developers did analyse the model before rolling it out but said it “needs continuous improvement” and that he is “eager to learn from this”.
