Emergent social conventions and collective bias in LLM populations

Ariel Flint Ashery, Luca Maria Aiello, Andrea Baronchelli · 2025 · Science Advances 11(20)

Read the source

Populations of language-model agents playing a naming game settle on shared conventions through purely local pairwise interactions. Strong collective biases appear even when no single agent holds them, and committed minorities can overturn a convention, with tipping points that vary widely by model.

Questions it bears on

• Cumulative culture or only conventions
• From individual psychology to collective behavior