H Human–AI Coevolution

Entry

Discovering Differences in Strategic Behavior Between Humans and LLMs

Caroline Wang, Daniel Kasenberg, Kim Stachenfeld, Pablo Samuel Castro

Synopsis

Uses AlphaEvolve to discover interpretable models of human and LLM strategic behavior from data; on iterated rock-paper-scissors finds frontier LLMs (Gemini 2.5 Pro) display deeper strategic behavior than humans.

Keywords

·AlphaEvolve ·strategic behavior ·RPS ·behavioral game theory

Open paper ↗ arXiv ↗ Report issue ↗