A Minor Adjustment to the “For You” Algorithm Can Rapidly Foster Political Polarization.

Studies indicate that altering the tone of posts on X can escalate political polarization within just a week, a shift that traditionally would have taken about three years.

An innovative study examining the impact of Elon Musk’s social media platforms on political polarization discovered that even minor increases in posts featuring anti-democratic sentiments or partisan aggression led to a marked rise in negative sentiments toward the opposing political faction among Democrats and Republicans.

The level of division, termed “emotional polarization,” reached in just one week due to the modifications made to the feeds of a specific number of X users equated to what would typically take an average of three years from 1978 to 2020.

Most of the over 1,000 participants in the experiment during the 2024 U.S. presidential election remained unaware of the changes in the tone of their feeds.

The campaign featured divisive viral content on X, including a fake image of Kamala Harris with Jeffrey Epstein and an AI-generated depiction from an image Musk posted showing Harris as a communist dictator, which garnered 84 million views.

Researchers observed that consistent exposure to posts reflecting anti-democratic views or partisan animosity significantly affected users’ feelings towards polarization, inducing heightened emotions of sadness and anger.

Musk acquired Twitter in 2022, rebranded it as X, and introduced a “for you” feed that presented content aimed at maximizing user engagement rather than just displaying posts from accounts that users actively follow.

The finding that increasing anti-democratic content heightens hostility towards political adversaries underscores the “power of algorithms,” noted Martin Savesky, an assistant professor at the University of Washington’s School of Information and a co-author of the study alongside colleagues from Stanford University, Johns Hopkins University, and Northeastern University. This research is published in Science magazine.

“While the adjustments in users’ feeds were subtle, they reported marked changes in their sentiments toward others,” explained Tiziano Picardi, an assistant professor in the Johns Hopkins University School of Computer Science and co-author of the study. “These shifts align with approximately three years of polarization trends seen in the U.S.”

The study also indicated that even slight alterations in users’ feed content could substantially diminish political hostility between Republicans and Democrats, implying that X could foster political unity if Musk opts to implement such changes.

“The intriguing aspect of these findings is that platforms can implement measures to mitigate polarization,” added Savesky. “This offers a new perspective for algorithm design.”

Mr. X was reached out for comment.

According to Pew Research, eight in ten American adults believe there’s an inability among Republicans and Democrats to agree on not only policies, but also on fundamental facts. Additionally, over half the British population perceives political differences as dangerously divisive, as revealed by a recent Ipsos poll.

The evolution of political polarization caused by exposure to posts on X was evaluated using an innovative methodology. Initially, researchers utilized AI to analyze posts in X’s “for you” feed in real time. The findings indicated that some groups were exposed to more divisive content while others faced less, demonstrating X’s predominant influence. Divisive posts included support for undemocratic practices, partisan violence, a lack of bipartisan consensus, and skewed interpretations of politicized facts.

After a week of reading these subtly modified feeds, researchers prompted users to evaluate their political opponents’ warmth or coldness, favorability or unfavorability. Changes in “emotional deflection” were rated at two degrees or higher on a scale from 0 to 100 on a “feeling thermometer.” This level of increase in polarization matched the typical trend observed in the U.S. over the past four decades leading to 2020. Conversely, reducing posts with anti-democratic views and partisan hostility led to a corresponding decline in political polarization.

Social media platforms have long faced criticism for amplifying divisive content to boost user engagement and thereby increase advertising revenue. Nevertheless, the study revealed that when divisive posts were deprioritized, users tended to like and share more frequently, despite a slight decrease in overall engagement in terms of time spent on the platform and posts viewed.

“The effectiveness of this approach illustrates its potential for integration into social media AI, aimed at mitigating detrimental personal and societal impacts,” the authors argue. “Simultaneously, our engagement analysis indicates a notable trade-off; implementing such measures could decrease short-term engagement levels, posing challenges to engagement-driven business models, supporting the idea that content that elicits strong reactions tends to generate more engagement.”