Scale-Dependent Collective Adaptation in Self-Amending LLM Societies: A Cross-Family Study of Emergent Governance

2026-05-17

AI Summary

This study investigates how the scale of large language models (LLMs) influences collective decision-making and adaptive governance in artificial societies where rules can be collectively revised. Building upon the Nomic self-amending game framework, we conduct cross-scale experiments across two LLM families, integrating multi-scale agent simulations, temperature perturbations, shifts between voting mechanisms (unanimity versus majority rule), and linear probe analyses to systematically examine consensus formation, rule adoption, and behavioral diversity. We find that collective adaptability exhibits a non-monotonic relationship with model scale, with medium-sized models achieving the best balance between rule adoption and consensus stability. Smaller models tend toward rigidity, larger models favor restrictive voting behaviors, and heterogeneous mixed groups are prone to veto deadlocks. These results reveal a complex decoupling between representational differences and emergent collective behavior, offering novel insights for AI-driven adaptive governance.

Abstract

We study group decision-making in artificial societies where the rules of play are themselves subject to collective amendment. Using the self-amending game Nomic, we compare multiple scales across two LLM families and find that collective adaptation does not improve monotonically with model size. Instead, both families exhibit a narrow mid-scale regime that supports sustained rule adoption, diverse amendments, and balanced consensus. Smaller models tend to remain rule-inert, whereas larger models often converge on restrictive voting patterns, and heterogeneous mixed-size groups collapse into veto-driven gridlock. These cross-scale contrasts persist under temperature perturbations and under a shift from unanimity to majority voting, although latent-state structure varies by family and scale. Hidden-state divergence alone does not explain collective performance: high representational divergence can coincide with poor behavioural outcomes. Linear probes reveal regime-selective coupling between latent vote-predictive signals and collective behaviour, but decodability is necessary rather than sufficient for adaptive play. Overall, the recurring regularity is non-monotonicity, not the particular scale at which the optimum appears. Self-amending games therefore provide a controlled testbed for studying collective adaptation in artificial societies beyond raw model scale.

Problem

• collective adaptation
• self-amending
• LLM societies
• scale-dependence
• emergent governance

Innovation

• collective adaptation
• self-amending systems
• scale dependence
• emergent governance
• large language model societies