Code-Space Response Oracles: Generating Interpretable Multi-Agent Policies with Large Language Models

TL;DR

This paper introduces Code-Space Response Oracles (CSRO), a framework that uses large language models to generate interpretable, human-readable multi-agent policies, replacing opaque neural network approaches in multi-agent reinforcement learning.

Contribution

CSRO replaces neural network oracles with LLMs for policy generation, enabling interpretable, human-readable strategies in multi-agent reinforcement learning.

Findings

CSRO achieves competitive performance with baseline methods.

CSRO produces diverse, explainable policies.

The approach leverages LLMs' pretrained knowledge for complex strategy discovery.

Abstract

Recent advances in multi-agent reinforcement learning, particularly Policy-Space Response Oracles (PSRO), have enabled the computation of approximate game-theoretic equilibria in increasingly complex domains. However, these methods rely on deep reinforcement learning oracles that produce `black-box' neural network policies, making them difficult to interpret, trust or debug. We introduce Code-Space Response Oracles (CSRO), a novel framework that addresses this challenge by replacing RL oracles with Large Language Models (LLMs). CSRO reframes the best response computation as a code generation task, prompting an LLM to generate policies directly as human-readable code. This approach not only yields inherently interpretable policies but also leverages the LLM's pretrained knowledge to discover complex, human-like strategies. We explore multiple ways to construct and enhance an LLM-based oracle: zero-shot prompting, iterative refinement and \emph{AlphaEvolve}, a distributed LLM-based evolutionary system. We demonstrate that CSRO achieves performance competitive with baselines while producing a diverse set of explainable policies. Our work presents a new perspective on multi-agent learning, shifting the focus from optimizing opaque policy parameters to synthesizing interpretable algorithmic behavior.