Why We Split: Simple Cognitive Rules Drive Opinion Polarization in New Model

Opinion polarization has become one of the defining puzzles of the digital age, and a new study argues that the answer may lie not in algorithms or echo chambers, but in two of the most basic tendencies of the human mind: the wish to stand out from the crowd and the urge to simplify a complicated world. Writing in PLOS Complex Systems, researchers Alina Dubovskaya, David J. P. O’Sullivan and Michael Quayle present an agent-based model in which virtual individuals balance these competing drives, and show that the interplay between them is enough to generate the sharply divided opinion clusters that characterize real societies. The work, published on September 15, 2026, offers a rare bridge between cognitive psychology, sociology and the mathematics of complex systems.

The model rests on a well-established idea from social psychology known as optimal distinctiveness theory. People, the theory holds, are caught between two opposing needs: the desire to belong to a group and the desire to remain unique within it. Lean too far toward belonging and you lose your individuality; lean too far toward uniqueness and you lose your community. Most previous models of opinion dynamics have captured conformity, the pull toward agreement, but far fewer have built in this simultaneous push toward difference. The new work makes that push explicit, giving each simulated agent a reason to seek diversity among the people it interacts with most closely.

The second ingredient is what the authors call cognitive compression. Human beings are limited information processors, and one strategy for coping with complexity is to compress it: to reduce a sprawling landscape of nuanced positions into a handful of recognizable categories. In the model, agents evaluate how complicated the overall opinion environment is and prefer configurations that are simpler to represent. Technically, both the local diversity of an agent’s small social circle and the global simplicity of the entire opinion landscape are measured using Shannon entropy, the standard information-theoretic quantity that quantifies how much uncertainty or variety a distribution contains. High local entropy means a rich mix of opinions among one’s close contacts; low global entropy means the society as a whole is easy to summarize.

Each simulated interaction follows the same basic logic. Two agents meet, compare opinions, and decide whether to adopt each other’s views by weighing two objectives that pull in opposite directions. Adopting a different opinion can increase the diversity within the agent’s local group, satisfying the drive for distinctiveness, but it can also make the global opinion landscape more complex, violating the drive for compression. The agent’s decision therefore depends on a trade-off between maximizing local complexity and minimizing global complexity, both scored with the same entropy measure. It is a deceptively simple rule, yet when many such agents interact over many rounds, the collective behavior becomes strikingly rich.

The headline result is that polarization emerges naturally from this trade-off. Rather than converging to a single consensus or dissolving into random noise, the population organizes itself into distinct, heterogeneous opinion clusters, mirroring the fragmented but structured opinion groups observed in real survey data. This is significant because many existing models achieve polarization only by imposing assumptions that essentially build the outcome in, such as bounded confidence thresholds or pre-defined social networks. Here, the clusters arise from the bottom up, as an unintended consequence of individuals pursuing psychological goals that have nothing to do with division itself.

Perhaps the most surprising finding concerns what happens after the clusters form. In many classical opinion models, once the population settles into groups, opinions freeze: the dynamics effectively stop, and each cluster becomes a static bloc. In the compression-based model, individuals continue to adjust their opinions long after clusters have emerged. The result is ongoing variation both within and between opinion groups, so that the clusters remain alive and internally diverse rather than ossifying into monolithic camps. This continuous churn looks far more like real societies, where people drift within broad ideological families and the boundaries between groups shift over time, than like the frozen end-states of earlier simulations.

The computational experiments also reveal a sharp sensitivity to group size. Polarization, the authors find, appears when local group sizes are moderate, a regime the paper connects to Dunbar’s number, the famous estimate of roughly 150 stable social relationships that humans can maintain. When agents’ local groups are smaller than this moderate range, the population fragments into many small, weakly structured factions. When the groups are larger, distinct clusters struggle to form at all, and the opinion landscape blurs. The implication is provocative: the scale of our natural social circles, shaped by cognitive limits on relationships, may itself be a precondition for the kind of large-scale polarization that worries observers of modern politics.

The strength of cognitive compression acts as a second control knob on the collective outcome. When agents compress aggressively, simplifying the world strongly, the system becomes more unpredictable, with opinion trajectories that are harder to anticipate and group structures that shift in less regular ways. When compression is weak, the opposite occurs: the population settles into more consistent, stable group structures. In other words, how much individuals simplify the world around them changes not just whether polarization happens, but what kind of polarization it is, from rigid and predictable to volatile and erratic. This parameter could plausibly differ across cultures, media environments or historical moments, offering a lever for understanding why some societies polarize more violently than others.

Methodologically, the study is a showcase of what agent-based modeling does best. Polarization is a phenomenon that spans levels of description: it is produced by individual cognition, mediated by social interaction and visible only in population-scale patterns. No single discipline can capture all three at once, but an agent-based model can, by encoding psychological rules inside simulated individuals and letting sociological structure emerge from their interactions. The use of Shannon entropy as a common currency for both local distinctiveness and global simplicity is the technical keystone, allowing two very different psychological drives to be compared and traded off within a single decision rule. It is an elegant piece of mathematical bookkeeping that turns a vague psychological tension into a computable objective.

The broader lesson is that complex, realistic social behavior does not require complicated agents. Two simple rules, seek some diversity among your close peers and prefer a world you can easily summarize, are sufficient to produce opinion clusters, ongoing internal churn and a sensitive dependence on group size that echoes human cognitive limits. For researchers, the model provides a new tool for probing how changes in social scale or information environments might push societies toward fragmentation or consensus. For everyone else, it is a reminder that polarization may not be a bug introduced by technology, but an emergent property of minds trying, quite reasonably, to be both individuals and members of groups in a world that is too complicated to take in all at once.

Subject of Research: Agent-based modeling of opinion polarization driven by optimal distinctiveness and cognitive compression

Article Title: Opinion polarization from compression-based decision making where agents optimize local complexity and global simplicity

Article References: Dubovskaya, A., O’Sullivan, D. J. P., & Quayle, M. (2026). Opinion polarization from compression-based decision making where agents optimize local complexity and global simplicity. PLOS Complex Systems, 3(9), e0000115. https://doi.org/10.1371/journal.pcsy.0000115

Image Credits: AI Generated

DOI: 10.1371/journal.pcsy.0000115

Keywords: opinion polarization, agent-based modeling, cognitive compression, optimal distinctiveness theory, Shannon entropy, complex systems, Dunbar's number, social psychology, opinion dynamics, PLOS Complex Systems, emergence, computational social science