After the 2024 U.S. presidential election, a large number of users turned to the decentralized social media Bluesky due to dissatisfaction with the partisan offensive remarks of the X platform (formerly Twitter). Bluesky does not use algorithms to recommend content, and it did reduce extreme speech and misinformation in the initial stage. But within a year, the platform still had an echo chamber effect.
Researchers from the University of Amsterdam found through AI simulation experiments that even without complex algorithms, social media will still naturally split into closed circles and form a polarization phenomenon with only basic functions such as posting, forwarding, and following. They built a simplified platform with 500 virtual users, set user characteristics based on US National Election Research data, and used large language models such as ChatGPT, Llama and DeepSeek to simulate user behavior. After the experiment ran for 10,000 interaction cycles, three major problems emerged on the platform: echo chambers, concentration of influence, and extreme speech.
The researchers tried six interventions, such as showing posts in chronological order or pushing opposing viewpoints, but none completely solved the problem, and some even exacerbated the negative phenomena. Academics at the University of Washington believe that this result reflects "a resonance between human nature and the dynamics of social media attention," and that even if the algorithm is removed, the system architecture may still retain toxicity.
Experts from Indiana University have questioned that the AI training data itself may contain social media toxicity, causing the experimental results to be inherently biased towards polarization. Researchers at New York University believe that although there is no simple solution, users posting more neutral content may help alleviate polarization.
This study reveals the underlying mechanism of social media polarization and shows that it is difficult to eradicate the problem through technical adjustments alone and requires a combination of multi-party strategies to deal with it.