Twitter has launched a new initiative called “Responsible Machine Learning”. With this initiative, the micro blogging service aims to evaluate the “unintentional” damages caused by its algorithms. Building a team of engineers, researchers and data science people, Twitter will examine the effects of machine learning methods on creating algorithmic biases that can negatively affect users.
One of the first tasks of this initiative will be to assess racial and gender biases of Twitter’s image cropping algorithm. Users of the microblogging service have shown that white faces are highlighted compared to Black faces in the preview of previously cropped photos. Last month, the company started testing showing full-size images instead of cropped previews.
The team will also look at how timeline recommendations differ between racial subgroups. In addition, we will examine how content recommendations will be analyzed across political subgroups in different countries. In the statement made by Twitter, it was stated that a close cooperation will be carried out with third-party researchers and the analysis results will be shared for feedback.
It is not easy to say anything about how the results will have an impact. Microblogging service underlines that the results achieved may not always lead to visible product changes. The company stated that its main purpose is to increase the difference and start important discussions about the use of machine learning.
Twitter is not the first social network to analyze its algorithms for bias. Other social platforms, including Facebook, had also set up special teams to analyze their algorithms.