In-context learning in multi-agent games mostly follows patterns not reasoning
Recursive Reasoning or Statistical Extrapolation? In-Context Learning in Multi-Agent Interdependent Decision-Making
Artificial Intelligence
Summary
Deciding whether large language models (LLMs) truly think through problems or just notice patterns is tricky. This paper looks at LLMs playing games where players must guess each other's thoughts to succeed. The researchers found that if the usual patterns in the game's history are scrambled, the model's decision skill drops to simple guesses, suggesting the model relies more on spotting patterns than on deep strategic thinking. This helps us understand how these models learn from past interactions in complex settings.
What this means in practice
- •For game developers: Improve AI opponents in strategy games by understanding when LLMs rely on patterns rather than planning.
- •For chatbot designers: Design conversational agents that better anticipate user strategies by clarifying limits of LLM reasoning.
Authors
Yu Liu, Wenwen Li, Yifan Dou, Guangnan Ye
Abstract
In-context learning (ICL) enables large language model (LLM) agents to improve decisions using interaction history, yet it remains unclear whether such improvement reflects refined internal reasoning or mere extrapolation of statistical patterns. To disentangle these mechanisms, we study LLM agents in multi-agent incomplete-information games that require recursive belief reasoning. By constructing a public goods game and manipulating the statistical structure of historical feedback, we evaluate decision quality against a history-independent rational expectations equilibrium (REE) benchmark. Our experiments reveal that when historical statistical patterns are disrupted, the benefits of longer context largely vanish, degrading decision quality to the no-context baseline in a way sharply amplified by stronger strategic interdependence. These results suggest that, in such strategic environments, ICL behavior is more consistent with statistical extrapolation than with strategic reasoning. Our work extends the mechanistic study of ICL to strategic multi-agent settings, introduces REE as a diagnostic tool for distinguishing reasoning from extrapolation, and provides a reusable framework for probing the boundaries of LLM reasoning in recursive belief tasks.