The Structural Collapse of Social Media: Why We Are Moving Toward a 'Messy' Future
For years, the public discourse around social media has focused on the "evil algorithm" or the innate negativity of human nature as the primary drivers of online toxicity. However, recent research by Petter Törnberg of the University of Amsterdam suggests a more systemic problem: the very architecture of social media is designed in a way that makes echo chambers and polarization inevitable.
As legacy platforms like Facebook and X (formerly Twitter) see a decline in genuine human engagement, we are witnessing a transition from "social networking" to a fragmented landscape of algorithmic broadcasting, private group chats, and AI-driven interactions. Understanding this shift is critical for anyone trying to envision a healthier digital public square.
The Architecture of Polarization
Törnberg’s research utilizes agent-based modeling combined with Large Language Models (LLMs) to simulate how users interact in online communities. The findings challenge the common narrative that "filter bubbles"—the algorithmic curation of content—are the sole cause of echo chambers.
Surprisingly, the research indicates that echo chambers emerge naturally even when users want to be in diverse spaces. When a user encounters too much disagreement, they tend to leave the community. This creates a feedback loop: as the most dissenting voices exit, the remaining community becomes more homogenous and extreme. This "boiling the frog" effect gradually pushes the remaining users toward more radical positions.
The Paradox of the Filter Bubble
In a counterintuitive twist, Törnberg found that a mild form of the filter bubble might actually stabilize a community. If a user is guaranteed to see at least 10% of content they agree with, they are more likely to tolerate diverse and contradicting opinions. Without this small anchor of validation, the system tips over into extreme homogeneity.
The "Botification" and Decline of Legacy Platforms
Data from the American National Election Studies (2020 and 2024) reveals a stark trend: human posting activity on legacy platforms like Facebook, YouTube, and X has dropped significantly—in some cases by as much as 50%. However, the volume of posts has not necessarily declined.
This discrepancy points to the "botification" of social media. As humans disengage, AI and LLMs are filling the void, generating content and simulating engagement. This shift fundamentally changes the nature of these platforms; they are no longer tools for human connection but engines for automated content delivery.
The Political Shift of X
Törnberg noted a dramatic shift in engagement behavior on X, describing a "72 percentage point shift to the right." While other platforms remained Democratic-leaning on balance, X has evolved into a space where high activity is now strongly correlated with a dislike for Democrats and a preference for Republicans, effectively mirroring the ideological goals of its current ownership.
What Comes After Social Media?
If the traditional model of the "social network" is dying, what is replacing it? Törnberg identifies three emerging categories of online interaction:
- Private/Semi-Private Silos: Migration to WhatsApp groups or protected communities like Substack. While these avoid some public toxicity, they lack the geographical constraints of a physical "coffee house," meaning they can become extreme echo chambers even more easily than public platforms.
- Algorithmic Broadcasting: Platforms like TikTok and Instagram Reels, where the "social" element is secondary to a fully algorithmic feed of short-form video. These are media platforms, not social networks.
- AI Companionship: A growing trend where users interact with AI chatbots for sociality rather than posting to a public feed. Data suggests twice as many people are now talking to chatbots as are posting on traditional social media.
Searching for a Solution
Is there a way back to a healthy digital public square? Törnberg argues that simply fleeing to private groups is not a solution, as it lacks the "functional scaffolding" and democratic systems required to maintain prosocial values.
Potential "pivot points" for improvement include:
- Granular Control: Tools like BlueSky’s advanced blocking mechanisms.
- Cross-Partisan Bridging: Features like X’s Community Notes, which can provide factual corrections that bridge ideological divides.
Ultimately, the goal is to redesign the fundamental rules of interaction to ensure users find enough common ground to remain tolerant of diversity, without falling into the trap of total homogeneity. Until such a redesign occurs, the transition from social media to its successor promises to be, in Törnberg's words, "messy."