Pages

Researcher reveals that Twitter’s image cropping algorithm is slightly more favorable towards younger age, light skin and feminine features

Technology and the internet is advancing with leaps and bounds, newer developments in the field of technology are now allowing companies to create more and more complex AIs (Artificial intelligence). Companies are developing better algorithms and are using them in their product thus enhancing their user experience. Almost every company based on technology works on developing machine learning models, especially one in the social media industry.

Twitter is also one platform that recently developed an interesting image cropping algorithm. The image cropping algorithm was to be used for automatically cropping photos on Twitter. The algorithm uses different variables to determine which part of a photo is useful and which is not, then moves forward to cut out the unneeded part.

Twitter’s new photo cropping algorithm was working quite well for them so the company decided to roll out the algorithm for developers and researchers to recheck it and report if they find any potential harms in the algorithm. Things went the other way when a researcher Bogdan Kulynych found out that Twitter’s algorithm is more favorable towards people with a fairer skin, thin body and younger face which means that these people having these features are less likely to be cropped in a photo than those who are dark skinned, fat or old. This surely was offensive and made a lot of people mad at Twitter.

After the algorithm was found to be offensive, a spokesperson from Twitter came out and said that Twitter’s team before the release experimented with the algorithm, but did not found it to be racial or gender bias. He added to this saying that it is clear that we have got more analysis to do.

Twitter prior to releasing the algorithm held a competition to identify mistakes in the algorithm. According to the Twitter the winner Bogdan Kulynych (A privacy, security and artificial intelligence student) found the algorithm to have an eight percent favor towards women and four percent towards white individuals.

Twitter faced a situation similar to this just last year when its algorithm was tested between a photo of President Obama and senator Mitch McConnell, the algorithm was found to be focusing on the face of Mitch McConnell thus proving that it favored white faces more than black faces.


Photo: Dado Ruvic / reuters

Read next: Twitter is working on a new feature that will allow users to eliminate any followers that they don't like

No comments:

Post a Comment