Twitter Highlights White-Skinned People in Shares

If you are an avid Twitter user, you may notice that tweets containing more than one image appear in your feed. in the form of a small gallery You may have noticed that it appears. In this gallery view, a maximum of 4 of the images in the posts are listed in a reduced size, and the focal point of the images is placed in the thumbnails. 

For example, if there is a person on the far right of a panoramic landscape photo, Twitter will show you the person on the far right in the small gallery view, not the landscape section in the middle. When you click, the real-size version of the photo is loaded. From this automation process Twitter's artificial intelligence is responsible. in the image human, text, smiling face is selected and highlighted in the thumbnail. 

However, Twitter's artificial intelligence fueled racism debates with the choices it made. Try opening photos from these posts, which seem completely normal at first glance:

The first post includes US politicians Mitch McConnell and Barack Obama. One of these people is light-skinned and the other is dark-skinned. However, in the thumbnail gallery view, only white-skinned Mitch McConnell is featured. When we click, we see that Barack Obama is also in the image. Moreover, in the later stages of the experiment, Obama and McConnell Even if the location changes, the result does not changeTwitter's algorithm continues to insist.

In the second post, the tie colors are changed, and even though there are shifts in the thumbnail, the photo of the white-skinned person is still highlighted:

When the images are made negative, that is, the colors are reversed, the result changes:

Twitter's algorithms, based on machine learning, are based on comments made by humans on the platform. By analyzing millions of shares reaches this conclusion. In other words, it makes its choices by synthesizing people's sharing behaviors. 

Security analyst Utku Şen bases this scandalous structure of Twitter on the most fundamental mistake of artificial intelligence applications: 

Since the ethical problems of societies are reflected in social media, they can be shared on Twitter or any other medium, in any structure where artificial intelligence is developed with the support of machine learning and the materials are human behavior. making “humanoid” mistakes is welcomed naturally. 

But something like Twitter in the new generation media This error should be at the top of the list of problems that the platform should recognize and prevent. On the other hand, Twitter can eliminate this situation with an update at any time. 

You can take a look at another experiment related to the news below: