Source-linked AI summary
The Emergence of Consensus: A Primer
Andrea Baronchelli
TL;DR
The paper addresses how population-scale consensus emerges across a fragmented interdisciplinary literature, especially when no centralised institution coordinates agents. It provides a compact review using game theory, evolutionary and complex-systems approaches, illustrative contagion and network models, and recent experiments. The review finds that microscopic rules, network structure, and post-consensus mechanisms shape agreement, while empirical work is increasingly clarifying these processes.
Problem
Consensus research is broad and scattered across disciplines, complicating navigation and obscuring similarities between explanations of coordination.
Method
The paper gives a brief overview of spontaneous consensus using game theory, evolutionary and agent-based approaches, illustrative contagion models, network analysis, and empirical studies.
Results
Recent theoretical and empirical advances have clarified the micro-macro connection in many consensus situations, while models show that interaction rules and networks affect collective agreement.
Takeaways & Limitations
Consensus formation is shaped by microscopic behavioural contagion, social networks, mechanisms that alter or prevent agreement, and the conditions under which alternatives are selected.
Abstract
from arXiv · showhide
The origin of population-scale coordination has puzzled philosophers and scientists for centuries. Recently, game theory, evolutionary approaches and complex systems science have provided quantitative insights on the mechanisms of social consensus. However, the literature is vast and widely scattered across fields, making it hard for the single researcher to navigate it. This short review aims to provide a compact overview of the main dimensions over which the debate has unfolded and to discuss some representative examples. It focuses on those situations in which consensus emerges 'spontaneously' in absence of centralised institutions and covers topic that include the macroscopic consequences of the different microscopic rules of behavioural contagion, the role of social networks, and the mechanisms that prevent the formation of a consensus or alter it after it has emerged. Special attention is devoted to the recent wave of experiments on the emergence of consensus in social systems.
I. INTRODUCTION
Consensus underpins shared conventions and coordinated social life, yet its emergence is puzzling when several equilibria are possible. This review offers a brief, cross-disciplinary map of explanations, emphasizing spontaneous coordination and illustrative examples.
- Shared acceptance of money, language, dress codes, decorum, and fairness creates expectations that help societies operate.
- Consensus research spans Social Sciences, Biology, Physics, Ethology, Artificial Intelligence, and Robotics.
- The interdisciplinary literature is rapidly expanding but remains fragmented by differing jargons and disconnected research communities.
- The review briefly surveys consensus mechanisms, spontaneous emergence, behavioural contagion, social networks, consensus disruption, and recent experiments.
- Conventions are customary, expected, self-enforcing behaviours maintained because unilateral deviation makes everyone worse off.
Space of individuals
Consensus can be produced by centralised institutions or, without them, through agent interactions and predefined behaviours. Relevant mechanisms include authority, leadership, broadcasting, incentives, feedback, punishment, externalities, conformity, and quorum sensing.
- Centralised institutions: Centralised institutions can impose order through authority, leadership, broadcasting, explicit incentives, or population-level informational feedback.
- Decentralised processes: Without centralised institutions, consensus can arise through agent interactions or predefined individual behaviour.
- Spontaneous emergence: Spontaneous consensus is produced by self-interested individuals who do not intentionally seek global coordination, while dynamics select the equilibrium.
- Decentralised mechanisms: Peer punishment promotes consensus when collective benefits exceed punishment costs or the cost of being punished is sufficiently large.
- Decentralised mechanisms: Positive payoff externalities, conformity bias, and quorum sensing provide additional decentralised routes toward shared behaviour.
Space of alternatives - Equilibrium selection
When multiple equilibria exist, populations may select among them through rational assessment, shared psychological biases, or interaction-driven dynamics that choose by chance. Only the last scenario requires communication between individuals.
- Rational individuals may select the equilibrium they assess as most advantageous and act to maximise their benefit.
- Shared psychological biases may select one alternative even when individual choice is not rational.
- When alternatives are equivalent, interactions among learning individuals can select an equilibrium by chance.
- Communication or interaction is necessary for consensus only when learning dynamics select among equivalent alternatives.
IV. APPROACHES TO THE STUDY OF SPONTANEOUS CONSENSUS
The review studies spontaneous consensus by connecting microscopic interactions to macroscopic outcomes through game theory, evolutionary approaches, and agent-based models. Illustrative Moran-process and naming-game models show that consensus can emerge and persist, while interaction rules and networks shape how alternatives are selected.
- Spontaneous-consensus analysis asks how microscopic behaviours produce macroscopic outcomes and uses game theory alongside evolutionary or dynamic approaches.
- Traditional coordination-game theory does not explain equilibrium selection when multiple equivalent Nash equilibria exist without strong information-processing assumptions.
- Evolutionary approaches replace strong rationality assumptions with anticipation, learning, experience, and adaptive choice adjustment.
- Agent-based models represent interacting agents whose state changes follow rules, including simple contagion after one exposure and complex contagion requiring multiple sources.
- In finite populations, both the Moran process and naming game reach persistent consensus, but their mechanisms for selecting alternatives differ qualitatively.
- The illustrative analysis is space-limited because factors beyond exposure number, including social strength, immediacy, and group size, also matter.
1. Homogeneously mixing populations
In homogeneously mixing populations, the Moran process removes alternatives progressively through symmetric interactions, whereas the naming game undergoes symmetry breaking in which the larger faction imposes its consensus.
- Moran process: The Moran process uses symmetric interactions, so consensus requires fluctuations that progressively eliminate alternative states.With two states, either state wins with probability 1/2 after an interaction; with multiple states, alternatives disappear progressively.
- Naming game: In the binary naming game, agents hold A, B, or both names, and contacts between A–AB or B–AB increase the corresponding single-name faction with probability p = 3/4.The state inventory contains only A, only B, or both A and B.
- Comparison: Figure 1 compares surviving states and success rates across homogeneous mixing, lattices, and random networks for Moran and naming-game dynamics.The plotted population uses N = 10,000 individuals initially assigned M = N different states; the lattice and random network have coordination number k = 4.
- Naming game: The naming game’s difference dynamics favor the larger faction, which imposes its consensus in large populations.The difference nA − nB evolves proportionally to nA − nB.
- Naming game: With unrestricted states, naming-game dynamics begin with competition between names and enter a winner-take-all symmetry-breaking regime.Consensus time is proportional to log N for the binary model and N interactions for the unrestricted model.
2. Spatial networks
Spatial structure changes the route to consensus: the naming game forms compact local-consensus clusters, while the Moran process remains fluctuation-dominated. Network heterogeneity can also alter consensus speed and outcomes, and consensus may remain vulnerable to committed minorities or shocks.
- Spatial networks: On two-dimensional regular lattices, consensus time scales as t_consensus ∼ ln N for Moran and t_consensus ∼ N for the naming game.Despite this timing contrast, the two models can appear similar in aggregate plots on lattices.
- Spatial networks: The naming game rapidly forms compact local-consensus clusters that compete at boundaries between regions using different conventions.The Moran process does not form comparable compact clusters because simple contagion leaves same-colour regions fragmented.
- Heterogeneous networks: In clustered networks, evolutionary forces can determine coordination outcomes, whereas fully connected graphs may converge slowly to the risk-dominant strategy and depend strongly on initial conditions.These results show that alternative interaction structures can produce outcomes opposite to those of the contagion models discussed above.
- Heterogeneous networks: On heterogeneous networks, topology and agent roles become entangled because interaction partners are sampled through degree-dependent neighborhood structure.Small-world and broad-connectivity networks can recover homogeneous-mixing behavior for both models, while scale-free hubs slow Moran consensus and favor voter-model consensus.
- Fragile consensus and committed minorities: Consensus can shift between equilibria after small shocks or weak forces, and committed minorities can sometimes drive such changes.In the naming game, committed agents supporting B can overturn consensus on A when they exceed roughly 10% of the population.
VII. OBSTACLES TO SPONTANEOUS CONSENSUS AND COEXISTENCE OF DIFFERENT STATES
Consensus can be slowed or prevented by network topology, interaction rules, homophily, and bounded confidence, while experiments show how communication, incentives, and influential users shape coordination.
- Strong community structure can slow or prevent consensus in complex contagion models, while nonlinear transition rules and irresolute agents can sustain coexistence.These effects can occur on lattices and fully connected graphs, not only in fragmented networks.
- Topology and homophily can generate isolated online communities that maintain polarisation and hinder agreement about conventions and scientific evidence.The paper links these echo-chambers to consequences for public debate.
- Communication experiments found that participants spontaneously developed distinct, parsimonious codes, while unstable environments could facilitate compositional coordination.The studies used restricted communication systems and pairs of participants.
- Groups of up to N = 10 reached more stable codes and more successful interactions than pairs, although larger groups initially agreed more slowly.This supports a distinction between the speed of initial agreement and the stability of the resulting convention.
- Experiments also examined consensus games with groups up to N = 96, networked coordination among N = 36 individuals, complex contagion, and incentive effects.Higher stakes increased pressure to establish and adhere to shared expectations that persisted across rounds.
- On Twitter, the conventions ‘RT’ and ‘via’ became dominant after initially being proposed and adopted by active, well-connected users.The finding connects convention success with status, influence, and connectedness in the community.
IX. CONCLUDING REMARKS AND OUTLOOK
The review argues that theory, computation, and empirical observation have clarified many mechanisms linking individual interactions to collective consensus, while important questions remain open. It identifies synergy among these approaches as a likely source of further insight.
- Theoretical advances in game theory and complexity science, aided by computational power, have clarified the micro-macro connection in many consensus settings.
- Empirical studies using experiments, social-media activity, and wearable sensors have begun to illuminate mechanisms of consensus formation in human societies.
- Important questions remain open, and the review expects further insights from combining theoretical, computational, and empirical approaches.
Appendix
The appendix defines the paper’s core terminology for dynamical systems, networks, contagion, and strategic interaction, using concise operational descriptions.
- Contagion is transmission of a disease, idea, or behaviour, with simple contagion using independent exposures and complex contagion depending nonlinearly on exposure count.
- A Nash equilibrium is a stable state in which no player benefits by changing strategy while the other players keep theirs unchanged.
- Self-organisation is a system’s ability to acquire functional, spatial, or temporal structure without specific outside interference.The paper also identifies it with ‘spontaneous’ order in the Social Sciences.
- Spontaneous symmetry breaking is symmetry reduction driven by arbitrarily small fluctuations, producing an asymmetrical final state from symmetric laws.
- A network consists of nodes representing system components and links representing possible interactions between them; communities contain nodes more tightly connected internally than externally.
- Network structure is further described through degree, degree distribution, scale-free and homogeneous connectivity, shortest-path distance, and the small-world property.Scale-free networks have heavy-tailed degree distributions and hubs, whereas homogeneous networks lack hubs; small-world networks have short average path lengths.