It's a great day to be either a mainstream political conservative party or a right-leaning news source on Twitter. The results of an algorithmic amplification study on political content on Twitter have been released by the company. This confirms what many had suspected: The political right is doing well on Twitter.
Advertisement
The company's Machine Learning Ethics Transparency and Accountability team (or META) reviewed millions of tweets from elected officials in seven countries: Canada, France and Germany. They also looked at hundreds of millions more tweets that contained links to news articles. The company found that tweets by the political right were amplified in all but Germany.
The same thing happened with news outlets. (The company analysed links to news outlets' content, not tweets from the news outlets). The algorithmic amplification of right-leaning news outlets was higher than that of left-leaning outlets. Twitter did not classify news outlets according to its own criteria but instead used third-party researchers' classifications.
According to the study, certain political messages are amplified via Twitter. However, the study left one important question unanswered: Why is this?
Rumman Chowdhury (director of the META Team) told Protocol that some of the amplification could have been user-driven and related to people's actions on the platform.
G/O Media could get $100 off the Apple iPad Pro 12.9 Latest and Greatest iPad
Gizmodo calls your tablet "basically a perfect piece hardware." Amazon: Buy it for $999
We can't predict what will happen when algorithms are released into the world and how people interact with them. She explained to the outlet that we cannot model how individuals or groups will use Twitter and what will happen in the real world to impact how people use Twitter.
Chowdhury, Luca Belli (machine learning researcher) wrote in a tweet that the META team aimed at examining these issues and reducing any injustice they might be causing. Chowdhury and Luca Belli stated that algorithmic amplifying is not a problem by default.
Advertisement
They wrote that algorithmic amplification can be problematic when there is preferential treatment based on how the algorithm was constructed and the interactions people have. To reduce the negative effects of our Home timeline algorithm, further root cause analysis will be required.
