Elon Musk's Opinion Machine: X Algorithm Influences Political Attitudes

Contents

The recommendation algorithm of the platform X (formerly Twitter) can measurably change the political attitudes of its users. This is the conclusion of a study published in the journal Nature.

Social media is suspected of influencing elections by amplifying extremist content. However, it is very difficult to actually measure the concrete effects of recommendation algorithms in social networks. Germain Gauthies and his team have now succeeded in doing so for the first time.

Small, but measurable results

The researchers randomly divided around 5,000 active US users of X, recruited for the study, into two groups in the summer of 2023: One of the two groups used the "For You" feed for seven weeks, the other the purely chronological "Following" feed. Political attitudes, priorities, and perceptions of current events were surveyed before and after the experiment.

Users who switched from the chronological to the algorithmic feed showed a small but statistically significant shift towards right-wing positions afterwards. This became clear, for example, when asked how important the topics of migration and crime were to them and how they judged the investigations against Donald Trump. The measured shift was in the range of around 0.1 standard deviations – a common but clearly demonstrable effect in political attitude research.

Asymmetrical effect

However, it is remarkable that the effect did not occur in the opposite case. While switching on the algorithm had measurable effects, switching back to the chronological feed did not lead to a corresponding counter-movement. The researchers attribute this to the fact that users continued to follow the previously recommended accounts even after the recommendations were switched off. The result is consistent with a 2023 study on Facebook and Instagram, which also showed no effect when the recommendation algorithm was switched off.

More activism, less news

Using a browser plug-in, the researchers were also able to analyze how the content displayed in the feeds changes. It turned out that the algorithmic feed plays out content from right-wing influencers and activists more frequently and displays posts from classic news media less often than the chronological feed. At the same time, the algorithm significantly increases engagement: posts in the "For You" feed received many times more likes, comments, and shares.

Evaluation and classification

The authors emphasize that their results are specific to active US users of X and only apply to the period under investigation. This may not apply to other platforms. However, in general, it can be stated that "the feed algorithms of social media play an important role in the development of political orientation and online behavior."

What this specifically means for election campaigns and especially for the situation in Europe cannot be deduced from this investigation alone. "In the German context, it is still important to consider that only five percent of the population uses X for news consumption," said Judith Möller from the University of Hamburg and the Leibniz Institute for Media Research in a statement for the Science Media Center.

"For Germany, it would first have to be checked whether there is a bias in favor of conservative topics and attitudes." On the other hand, she considers "especially the downranking of news organizations" by the feed to be problematic. "This means that the algorithmic group not only saw more conservative content but also less content that had been verified and contextualized by journalistic means."

"To be able to say something about long-term effects and real-world impacts, we would need significantly larger-scale field experiments," says Philipp Lorenz-Spreen from the Technical University of Dresden. "For this, we would either need cooperation with existing platforms – also legally required. However, I am not very hopeful about that. Or we would need larger-scale, independent experimental platforms for research."

This article was first published on t3n.de.

(jle)