Social Media’s Structural Collapse: Why Tweaks Won’t Work and What’s Next

Introduction

Social media, once hailed as a democratizing force, has become a breeding ground for echo chambers, attention inequality, and amplified extremism. Petter Törnberg, a researcher at the University of Amsterdam, has spent years dissecting the underlying mechanisms that drive these toxic outcomes. His conclusion is stark: the problems are not bugs but features—deeply embedded in the very architecture of these platforms. Recent papers from Törnberg’s team reinforce this view, showing that incremental fixes are futile and that the future is likely to be chaotic and fragmented.

The Architecture of Toxicity

Unlike physical-world interactions, social media operates on a fundamentally different structural logic. Törnberg’s research identifies three core dynamics that arise from this digital architecture:

• Partisan echo chambers – Users are algorithmically funneled into homogeneous information bubbles.
• Attention inequality – A tiny elite captures the vast majority of engagement, drowning out diverse voices.
• Extremism amplification – The most divisive content is systematically boosted because it generates the strongest reactions.

These phenomena are not accidental. They are emergent properties of a system designed to maximize engagement, not social health. In his earlier work, Törnberg argued that no amount of tweaking algorithms or feeds can fix them—the problem lies in the structural dynamics themselves.

Echo Chambers and AI Personas

The first of his new papers, published in PLoS ONE, dives deep into echo chambers. Törnberg’s team combined standard agent-based modeling with large language models (LLMs) to create AI personas that simulate online behavior. These digital avatars allowed the researchers to test how different platform structures affect the formation of echo chambers (see more below).

New Research: Using LLMs to Model Social Media

Törnberg has been prolific since the original interview. Two additional papers and one preprint build on his core thesis: that social media’s information architecture is the root cause of its dysfunctions. The second paper extends the agent-based modeling to explore attention inequality, while the preprint examines cross-platform dynamics.

Key Findings from the PLoS ONE Study

• Standard algorithmic interventions (e.g., chronological feeds) do not reduce echo chambers.
• Even when users are forced to encounter diverse viewpoints, the structural pull toward homophily reasserts itself.
• The only effective strategies involve fundamental redesign—for example, altering how information is spread or how users form connections.

In other words, the mess we see is not a bug that can be patched; it is the default operating state of a system built on engagement metrics and network effects.

Why Incremental Fixes Fail

From fact-checking to content moderation to toxic-behavior penalties, platforms have tried a laundry list of interventions. Yet Törnberg’s models consistently show that these address symptoms, not root causes. The dynamical system of social media has attractor states that lead to polarization and inequality, regardless of surface-level changes. As he puts it, “We’re probably doomed to endless toxic feedback loops unless someone hits upon a brilliant fundamental redesign.”

This conclusion challenges both industry optimism and regulatory faith. It suggests that simply tweaking algorithms or adding accountability features will not reverse the damage. Instead, we may need entirely new platforms built on different social and informational premises—something that would imply a disruptive, messy transition.

What Comes Next: A Messy Transition

If social media as we know it is structurally doomed, what will replace it? Törnberg’s work points to a future that is far from clean. The collapse of the current model could lead to:

1. Fragmented niche communities – Smaller, purpose-built platforms that prioritize specific values over virality.
2. Decentralized networks – Blockcahin-based or federated systems that resist central control but struggle with moderation.
3. Increased regulation – Governments stepping in to impose structural changes, such as algorithmic transparency or user-ownership models.

None of these paths is straightforward. Each carries trade-offs between freedom, safety, and usability. The transition will be messy, as legacy platforms resist change and new experiments fail or get co-opted. Törnberg’s research (backed by empirical modeling) suggests that we have only begun to grapple with the depth of the problem.

Implications for Users and Society

For individual users, the takeaway is sobering: your online experience is shaped more by platform structure than by personal choices or content moderation. For policymakers, the message is clear: stop tinkering and start thinking about fundamental redesign. Törnberg’s work provides a roadmap of what not to do, but the path forward remains uncertain.

In the end, the RIP social media sentiment may be premature, but it captures the right direction. The era of monolithic, engagement-driven platforms is ending. What comes next is indeed messy—but it is also an opportunity to build something better, if we are willing to face the structural truth Törnberg has uncovered.