Emergent social conventions and collective bias in LLM populations

Summary

Experimental demonstration that decentralized populations of LLM agents spontaneously converge on universally adopted social conventions, and that strong collective biases emerge during this process even when agents exhibit no bias individually. It also shows committed minorities of adversarial LLM agents can impose alternative conventions on the larger population. The emergent-bias-from-unbiased-individuals result is the population-level analogue of the deliberation pathologies the paper's consensus condition should exhibit, and it carries the visibility an IS audience expects.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-08-29 literature sweep as a high-priority candidate for the Architecture as Epistemology (ISR) paper (angle: correlated errors / homogeneity across LLM agents; organization theory applied to AI agents). Verified against: Opened https://arxiv.org/abs/2410.08948 (arXiv abstract page, v2 29 May 2025; journal-ref: Science Advances 11, eadu9368, 2025). Publisher page at science.org returned HTTP 403.

Critique / open questions

Queued — not yet assessed.

Key quotes

Social conventions are the backbone of social coordination, shaping how individuals form a group. (abstract)

As growing populations of artificial intelligence (AI) agents communicate through natural language, a fundamental question is whether they can bootstrap the foundations of a society. (abstract)