X Algorithm Amplifies Content Against Users' Values

Arguing with people on X may lead to more posts that users dislike, according to a study published in the Proceedings of the National Academy of Sciences. The research indicates that X's 'For You' algorithm amplifies content that conflicts with users' stated values, especially among Democrats. Researchers from MIT, the University of Michigan, Johns Hopkins, Stanford, and Yale analyzed 715 active U.S. X users in September and October 2024. Participants used a browser extension to track their 'For You' and Following feeds. The study found that Democrats were shown more content opposing their values, with a correlation of -0.229, while Republicans had a weaker correlation of -0.041. Overall, amplified content scored -0.150, indicating misalignment with users' values. The researchers noted that engagement behaviors may misrepresent users' true preferences, suggesting the algorithm promotes content misaligned with explicit values. A replication study showed similar partisan results, with -0.199 for Democrats and 0.0372 for Republicans. The study also suggested that replies to conflicting content might signal to X a desire for more such posts. The findings highlight the tension between personal values and engagement-driven content amplification. The role of bots in this dynamic remains unclear, as the study did not determine whether accounts behind the posts were human or automated. This research follows X's decision to open-source its Grok-powered 'For You' algorithm in January, which Musk acknowledged requires significant improvements.