Twitter Image Thumbnail Racial Bias Controversy

Overview

Twitter Image Thumbnail Racial Bias Controversy refers to a controversy surrounding the Twitter image preview algorithm apparently selecting faces of Caucasian people over faces of people of other races for the thumbnail preview when the full image is too long or wide to be shown. Following the discovery of the apparent bias, multiple users posted tweets in which they tested the assumption.

Background

On September 18th, 2020, Twitter^[1] user @colinmadland posted a tweet in which he shared a story that indicated an apparent bias towards Caucasian people in face recognition technology utilized by the video conferencing service Zoom (tweets shown below, left). In the thread, after attempting to upload evidence of the algorithm bias, @colinmadland pointed out^[2] that the algorithm used by Twitter also showed the part of the image containing his photograph rather his colleague's as the thumbnail (tweet shown below, right). The tweet received overt 1,800 retweets and 18,800 likes in three days.

Developments

In the same thread, Twitter^[3] user @NotAFile tested the assumption, with the Twitter algorithm selecting a photograph of a white man over a photograph of a black man three times out four attempted. One of the tweets^[4] by @NotAFile received over 3,100 retweets and 8,000 likes (original image and tweet shown below, left). In the same thread, Twitter^[5] Chief Design Offer Dantley Davis wrote "it's 100 percent our fault. No one should say otherwise. Now the next step is fixing it" (tweet shown below, right).

On September 19th, 2020, Twitter^[6] user @bascule posted a two long images containing photographs of former US president Barack Obama and the US Senator Mitch McConnell, with the Twitter algorithm selecting the photograph of McConnel for the preview both times (original images and tweet shown below). The tweet received over 62,200 retweets and 192,900 likes in two days.

On the same day, Twitter^[7] user @himhajria tweeted that stated that the algorithm used by Twitter is based on targeting parts of the image with the greatest brightness and color density (tweet shown below, left). The same was confirmed Twitter^[8] CDO Dantley Davis (tweets shown below, right).

Parodies

The September 19th viral tweet by @bascule inspired multiple users to test the assumption, with users testing the algorithm with images of people, animals and fictional characters. For example, on September 20th, Twitter^[9] user @_jsimonovski tested the assumption with images of The Simpsons characters Carl and Lenny. The tweet (shown below, left) received over 9,100 upvotes and 46,300 likes in one day. On the same day, Twitter^[10] user @MarkEMarkAU made a test with images of a white and a black labrador, with the tweet gaining over 1,200 retweet and 6,800 likes in one days (shown below, right).

Search Interest

External References