Digital fragmentation accelerates as social media evolves into private chats and AI broadcasting

Research by Petter Törnberg of the University of Amsterdam indicates that the digital landscape is undergoing a fundamental transformation, splintering from traditional public networks into private group chats, protected communities, and AI-driven interactions. This fragmentation marks a departure from the user-generated social networks of the past, with Törnberg's latest studies suggesting that echo chambers emerge naturally from platform architecture regardless of specific algorithms. The findings warn that current digital structures risk creating toxic feedback loops without the functional democratic scaffolding required to sustain healthy public discourse.

Törnberg's new agent-based modelling study, published in PLoS ONE, demonstrates that echo chambers form naturally when users leave communities if disagreement exceeds a certain threshold. This mechanism creates a feedback loop that leads to extreme homogeneity or polarization without needing algorithmic nudges or filter bubbles. The research highlights that even if users prefer diverse spaces, the basic structure of interaction drives them toward segregated environments where opinions become increasingly extreme as dissenting voices depart.

Concurrent with these structural shifts, legacy platforms like Facebook and X are experiencing marked declines in human activity and increased political polarization. Analysis of the 2020 and 2024 American National Election Studies surveys reveals that while human posting activity on these platforms has dropped by approximately 50 per cent, AI and large language models may be sustaining post volume through a process of botification. On X, engagement behaviour has shifted 72 percentage points to the right, whereas Facebook activity is now driven almost exclusively by the most partisan users, effectively silencing casual voices.

A new preprint identifies three emerging media categories that define this new era: private or semi-private group chats such as WhatsApp, protected communities led by influencers like Substack, and direct interaction with AI chatbots. Data suggests that chatbot usage is currently double that of traditional social media posting, indicating a significant move toward algorithmic broadcasting. This transition challenges the traditional definition of social media as a platform for connecting users, as the infrastructure increasingly allows entities to generate their own content and automate interactions without human participation.

Törnberg cautions that fleeing to smaller private spaces does not guarantee diversity or prevent polarisation. Non-local digital groups can still tip into toxic echo chambers without the geographical constraints that once forced interaction between diverse groups in physical settings. The research suggests that abandoning public squares for private enclaves may exacerbate isolation, as these new formats lack the organic mixing of viewpoints found in local community spaces.

The study concludes that while reorganising social media spaces is possible, it requires a fundamental redesign of system rules to manage complex feedback effects. Törnberg notes that most users prefer pleasant online communities over those rife with toxic waste, but creating spaces that are both engaging and resistant to dark feedback loops remains a difficult challenge for policymakers and platform designers.