[ad_1]
Last fall, Canadian Student Colin Madland noticed this Twitterthe automatic cutting algorithm constantly selected her face — not a dark-skinned colleague — to display in tweets from the couple’s photos. The the section sparked accusations according to the bias, they posted long photos as a wave of Twitter users to see if the AI would choose the face of a white person in front of a black person or put women’s boxes on their faces.
At the time, a Twitter spokesperson said the assessment algorithm It was discovered before it was launched in 2018 no evidence of racial or gender bias. Now, he has done the biggest analysis of AI so far has found the opposite: Twitter’s algorithm favors whites over blacks. According to this assessment, the most interesting part of the picture is that AI does not focus on women’s bodies.
Previous tests Hand in hand with Twitter and researcher Vinay Prabhu it involves a hundred images or less. The analysis released by Twitter research scientists on Wednesday is based on 10,000 image pairs from different demographic groups to test who the algorithm helps.
The researchers found a bias in the algorithm when photos of people from two demographic groups appear. In the end, the algorithm selects a person whose face will appear on Twitter timelines, and some groups are better represented on the platform than others. When researchers introduced a picture of a black man and a white woman into the system, the algorithm chose to display a 64 percent time off white woman, while only 36 percent of the black man was the largest gap among all demographic groups. analysis. For the images of the white woman and the white man, the algorithm showed 62 percent of the woman’s time. In terms of images of a black and white woman, the algorithm showed the white of the woman 57% of the time.
On May 5, when Twitter removed the cropping of images for single photos posted using its Twitter phone app, Dantley Davis appeared in favor of Twitter’s design manager’s approach. since the discussion of the algorithm exploded last fall. The change led to people being sent high photos and indicated the end “Open for a surprise” tweets.
The so-called verification algorithm is still used on Twitter.com, as well as for cutting multi-image tweets and creating small images. A Twitter spokesperson said high or wide photos are now too cropped, and the company plans to end the use of the algorithm on its Twitter website. Sales algorithms are trained by following what people look at when they see an image.
Other sites, including Facebook and Instagram, they use AI-based automated cutting. Facebook did not respond to a request for comment.
Accusations of gender and racial bias in computer vision systems are quite common, unfortunately. Google has recently determined that Android cameras will improve the performance of people with dark skin. Last week the team Algorithm Watch found it depicts cartoons of people with dark skin used by iPhone to label the image as “animals”. An Apple spokesman declined to comment.
Regardless of the results of fairness measurements, Twitter researchers say that making algorithmic decisions can deprive users of the opportunity and have a wide impact, especially on excluded groups of people.
In a recently published study, Twitter researchers said they found no evidence that the photo-cutting algorithm favors women’s bodies in their faces. To determine this, they were given a randomly selected image algorithm of 100 people identified as women, and found that only three bodies were centered on their faces. Researchers suggest that there are badge or jersey numbers on people’s boxes. For the research, the researchers used photos from the WikiCeleb database; the identities of the people in the photos are taken from Wikidata.
[ad_2]
Source link
