Summary
- A study involving 715 users in the U.S. revealed that X promoted posts that did not align with users' declared values.
- The alignment correlation for Democrats was -0.229, while for Republicans it was -0.041.
- Researchers suggest that responding to posts opposing one's values may lead X to recommend similar content.
Engaging in disputes on X could be inadvertently teaching the platform to deliver more content that users dislike.
A recent study published on Monday in the Proceedings of the National Academy of Sciences indicates that X's algorithm for the "For You" feed tends to amplify posts that conflict with users' stated beliefs, especially among Democrats.
Myriad: What will Elon Musk's net worth be by August 31? Click to make your prediction.The researchers stated, "Social media feed algorithms are designed to prioritize content that users are thought to prefer, yet the engagement behaviors influencing these algorithms often serve as indirect indicators of users' explicitly stated values."
This investigation, conducted by teams from MIT, the University of Michigan, Johns Hopkins, Stanford, and Yale, analyzed the behavior of 715 active X users in the U.S. during September and October 2024. Participants utilized a browser extension that tracked their "For You" and Following feeds.
Findings indicated that X exposed Democrats to more content contrary to their expressed values, while the effect was significantly less pronounced for Republicans. The correlation for Democrats was -0.229, and for Republicans, it was -0.041, with lower values indicating a greater disconnect. On average, the amplified content had a score of -0.150, whereas posts from followed accounts generally aligned with users' values.
Researchers noted, "Despite this alignment, we found that participants' values were negatively correlated with the likelihood of content being amplified," suggesting that the algorithm tends to promote content misaligned with users' explicit values.
A repeat of the study showed a similar partisan difference, with correlations of -0.199 for Democrats and 0.0372 for Republicans.
The study posits that replies to opposing content may contribute to this discrepancy. Users responding to posts that contradict their values might unintentionally indicate to X that they desire more such content.
The researchers concluded, "These findings illuminate the conflict between individuals' values and those that elicit reactions, demonstrating how these tensions can lead to misaligned algorithmic outcomes."
Importantly, the study did not investigate whether the accounts behind the posts that elicited value-misaligned replies were human or automated, leaving the influence of bots unaddressed.
This report follows X's decision to open-source its Grok-powered "For You" algorithm in January, which revealed a recommendation system that ranks posts based on predicted user interactions, including likes and replies. Elon Musk indicated that the algorithm "requires significant improvements."
Musk remarked on Twitter after the release, "We recognize the algorithm is flawed and needs significant enhancements, but at least you can observe our efforts to improve it transparently and in real-time. No other social media company does this."