Small changes to 'for you' feed on X can rapidly increase political polarisation (www.theguardian.com)

Researchers running a field experiment on X (formerly Twitter) found that barely perceptible tweaks to the platform’s “for you” feed can rapidly amplify political polarization. Over one week during the 2024 US election, an AI system analyzed users’ feeds in real time and selectively boosted posts expressing anti‑democratic attitudes and partisan animosity for one cohort while down‑ranking them for another. Among more than 1,000 participants (most of whom didn’t notice the changes), those exposed to more divisive content reported an increase in “affective polarization” of just over two points on a 0–100 feeling thermometer—an effect the authors equate to roughly three years of historical U.S. polarization trends. Technically, divisive posts were defined to include support for undemocratic practices, partisan violence, opposition to bipartisan consensus, and biased takes on politicized facts. The intervention also produced more sadness and anger. Crucially, down‑ranking such content produced symmetric reductions in animosity, showing platforms could algorithmically reduce polarization. The study highlights a practical trade‑off: reducing divisive content slightly lowered aggregate time and posts viewed but increased propensity to like or repost, posing challenges for engagement‑driven business models. The findings demonstrate the strong causal power of recommender algorithms and underscore policy and design levers for mitigating harm.