330 | Petter Törnberg on the Dynamics of (Mis)Information

What It Covers

Petter Törnberg presents research using agent-based models and large language models to simulate social media dynamics, revealing how echo chambers, attention inequality, and polarization emerge naturally from platform structures rather than algorithms alone.

Key Questions Answered

• •Schelling Segregation Online: Törnberg adapted Thomas Schelling's 1969 checkerboard model to digital communities, finding segregation emerges even stronger online than spatially. Users who tolerate 70-80% different-minded neighbors still create near-complete echo chambers through cascading moves, making integrated states fundamentally unstable.
• •LLM Social Simulation: Using 500 large language model agents with personas from American National Election Survey data, researchers created a bare-bones social network. Without engagement algorithms, three problematic outcomes emerged automatically: political echo chambers, power-law attention distributions, and amplification of extreme voices through sharing dynamics.
• •Retweet Feedback Loop: Emotional, reactive sharing creates a structural feedback effect where polarized users gain more followers. This mechanism operates independently of algorithmic curation, suggesting platform architecture itself drives polarization. The sharing behavior shapes network formation, not just content visibility, creating self-reinforcing attention inequality.
• •Intervention Failure: Testing solutions like chronological timelines, hiding user biographies, and bridging-based content ranking failed to fix emergent problems. Some interventions worsened outcomes—chronological feeds increased extreme user attention. The robustness of these negative patterns suggests fundamental platform structure redesign is necessary, not cosmetic algorithm changes.
• •Political Misinformation Strategy: Cross-country analysis of politicians' Twitter posts over five-six years shows radical right populist parties specifically drive misinformation spread. Social media incentives for attention-gaining become intertwined with political movements, making misinformation a deliberate competitive strategy rather than random information quality degradation.