The messy future of social media after its decline
▼ Summary
– Petter Törnberg’s research identifies social media’s structural architecture, not algorithms or user behavior, as the root cause of echo chambers, attention inequality, and extreme voices.
– He argues that most platform-level intervention strategies are ineffective because the toxic dynamics are embedded in social media’s fundamental design.
– Törnberg’s new study used agent-based modeling with LLMs to simulate online behavior, finding that echo chambers emerge naturally from basic platform architecture.
– In simulations, users left communities when the proportion of disagreeing members exceeded a set threshold, leading to segregated spaces.
– A surprising result was that filter bubbles, typically blamed for homogeneity, can actually help reduce echo chamber formation.
Last autumn, we sat down with Petter Törnberg of the University of Amsterdam, a researcher who dissects the underlying mechanics of social media that produce its most troubling features: partisan echo chambers, attention inequality among a tiny elite of users, and the relentless amplification of extreme, divisive voices. At the time, his outlook on the future of these platforms was grim.
Törnberg’s earlier work demonstrated that most proposed platform-level interventions are doomed to fail. The problem, he argued, isn’t our much-maligned algorithms, non-chronological feeds, or even our natural attraction to negativity. Instead, the very architecture of social media is structurally wired to generate these toxic outcomes. Without a radical redesign that fundamentally alters these dynamics, we appear trapped in endless feedback loops.
Since that interview, Törnberg has been prolific, releasing two new papers and a preprint that explore how social media’s structure differs from the physical world, producing unexpected consequences. The first study, published in PLoS ONE, zeroes in on the echo chamber effect. Using a novel method that combines standard agent-based modeling with large language models (LLMs) , the team created small AI personas to simulate online behavior.
In the simulation, these digital agents were randomly assigned an opinion or its opposite and then interacted with randomly selected members of a virtual community. When the proportion of disagreeing members exceeded a preset threshold, the agents “left” to join a more agreeable group.
Here is the twist: filter bubbles are not the culprit. In fact, they may be part of the solution.
Consistent with Törnberg’s earlier findings, echo chambers emerge naturally from the basic design of social media platforms, even without algorithmic nudges. “One surprising finding is that we get echo chambers even without any filter bubbles, even if people really love being in diverse spaces,” Törnberg explained. “You don’t need an algorithmic nudge. You can still get these highly segregated spaces. The other surprising finding is that filter bubbles, which have been blamed for homogeneity, can be a cure.”
(Source: Ars Technica)
