Social media as we know it is facing a structural crisis. According to Petter Törnberg, a researcher at the University of Amsterdam, the problems that plague platforms—echo chambers, attention inequality, and the amplification of extreme voices—are not the result of flawed algorithms or human bias. Instead, they are built into the very architecture of social media. Traditional fixes like feed tweaks or moderation policies fail because they don't address these deep-seated mechanisms. In a series of new studies, Törnberg uses AI-powered simulations to explore why social media is broken and what might replace it. The future, he warns, is messy but not hopeless.
What makes social media inherently toxic?
Törnberg’s research shows that social media’s worst features are not caused by algorithms, chronological feeds, or human negativity bias. Instead, the core issue is structural: the platform’s architecture creates feedback loops that reward extreme content and concentrate influence among a few elite users. Unlike real-world interactions, online networks lack the physical constraints that naturally moderate behavior. This leads to attention inequality, where a tiny minority dominates discussions, and echo chambers, where users only encounter reinforcing viewpoints. These dynamics are so deeply embedded that superficial changes—like tweaking recommendation systems—cannot fix them. The system is designed to amplify the loudest, most divisive voices, making toxicity an output of the platform’s DNA.
Why can’t platform-level interventions fix these problems?
Many proposed solutions—such as content moderation, deplatforming, or algorithm adjustments—target symptoms rather than causes. Törnberg’s work demonstrates that interventions at the platform level are ineffective because the underlying structural incentives remain unchanged. For example, changing the feed from chronological to algorithmic might alter how content is distributed, but it doesn’t break the concentration of influence or the incentive to post extreme material. The system is self-reinforcing: extreme content generates engagement, which amplifies that content, which attracts more users to extreme positions. Until we redesign the fundamental architecture of social media—not just patch it—these cycles will persist. As Törnberg puts it, we’re stuck in a toxic feedback loop unless someone invents a completely new paradigm.
How did Törnberg study these dynamics?
To explore why social media behaves as it does, Törnberg combined agent-based modeling with large language models (LLMs). He created thousands of AI-driven personas that simulate real user behavior online. These “agents” interacted in a controlled virtual environment, allowing researchers to observe how echo chambers and attention inequality emerge naturally from the platform’s rules. The first paper, published in PLoS ONE, focused specifically on the echo chamber effect. By adjusting parameters like network structure and content visibility, the team could test which architectural changes might reduce polarization. This method offers a powerful way to experiment with fixes before deploying them in the real world.
What did the new papers reveal about echo chambers?
Törnberg’s first new paper confirmed that echo chambers are not just a result of user preferences or recommendation algorithms. Instead, they emerge from the network topology of social media itself. When users connect freely, they naturally cluster around shared interests. But because online platforms lack the physical cues that normally bridge differences (like accidental encounters), these clusters become isolated. The AI simulations showed that even when users are exposed to diverse content, they tend to interpret it through their existing beliefs, reinforcing old views. The only way to break this cycle is to redesign how content is discovered and how connections form—a much harder task than tweaking feeds.
What comes after social media’s collapse?
Törnberg predicts that the current model of social media is unsustainable. The structural flaws lead to toxicity, user burnout, and regulatory backlash. But what replaces it won’t be a single platform; instead, we’ll likely see a fragmented ecosystem of smaller, purpose-built networks. These might include community-owned platforms, interest-based forums, or decentralized protocols that don’t rely on engagement-maximizing algorithms. Some could mimic real-world social dynamics more closely, using constraints like limited reach or opt-in filtering. The transition will be messy, with many failed experiments, but it offers hope for healthier online spaces. The key is learning from Törnberg’s insights: build architecture that rewards nuance, not outrage.
Can social media ever be fixed, or is it doomed?
Törnberg is not entirely pessimistic. He believes a fundamental redesign is possible, but it requires moving beyond band-aid solutions. Platforms must rewire their core mechanics—for example, by introducing friction that slows down viral spread, or by designing algorithms that prioritize context and credibility over raw engagement. Experiments with federated models (like Mastodon) and topic-based networks offer clues. However, the economic incentive to maximize attention remains a huge obstacle. Without regulatory pressure or a shift in user demand, meaningful change is unlikely. So social media as we know it may indeed be doomed, but a healthier, messier successor could emerge from the ashes if we’re willing to rethink the rules of the game.