lswcap
On May 19 (local time), Twitter released a survey finding that the ability to automatically trim photos tended to focus on whites first. Accordingly, the company has established a policy to phase out the automatic cropping function algorithm and leave the user decision on trimming.
Twitter introduced automatic image cropping optimization by AI technology in 2018. With the introduction of the algorithm, if a person is reflected in an image, it has been improved to include the person in the center as much as possible, but some users have voices expressing doubts whether there is a race or sexually transmitted disease bias as a target for trimming.
Therefore, as a result of Twitter verifying whether the algorithm has racial prejudice, it was found that the algorithm favored white female faces over black women at a rate of 7%, and white male faces over black women at 2%. Overall, the percentage of algorithms favoring whites was 4%.
In addition, as a result of examining the differences between sexes, it was found that there was a tendency to prefer women to men at an 8% rate, but it is said that there was no tendency to prioritize the chest or legs over the face.
According to the findings, Twitter confirmed the balance between the speed of the auto-cropping function and the potential risk of racism revealed in the findings, and announced that it has reached the conclusion that it is best for humans to determine how to crop the image.
Prior to this announcement, the application was changed to make the photo visible on the timeline in early May 2021 on Twitter, but the purpose was to display the image as large as possible and to minimize the effect of automatic cropping by the algorithm. In addition, when trimming is required, a preview is displayed so that the user decides the area to be cropped. Related information can be found here.
Add comment