Multiple LLM Agents Debate for Equitable Cultural Alignment
This addresses the need for equitable cultural alignment in LLMs to benefit diverse global communities, representing an incremental improvement over single-LLM approaches.
The paper tackles the problem of adapting LLMs to diverse cultural contexts by proposing a Multi-Agent Debate framework where two LLM agents debate over cultural scenarios, improving overall accuracy and cultural group parity on the NormAd-ETI benchmark across 75 countries, with smaller models (7-9B parameters) achieving accuracies comparable to a larger model (27B parameters).
Large Language Models (LLMs) need to adapt their predictions to diverse cultural contexts to benefit diverse communities across the world. While previous efforts have focused on single-LLM, single-turn approaches, we propose to exploit the complementary strengths of multiple LLMs to promote cultural adaptability. We introduce a Multi-Agent Debate framework, where two LLM-based agents debate over a cultural scenario and collaboratively reach a final decision. We propose two variants: one where either LLM agents exclusively debate and another where they dynamically choose between self-reflection and debate during their turns. We evaluate these approaches on 7 open-weight LLMs (and 21 LLM combinations) using the NormAd-ETI benchmark for social etiquette norms in 75 countries. Experiments show that debate improves both overall accuracy and cultural group parity over single-LLM baselines. Notably, multi-agent debate enables relatively small LLMs (7-9B) to achieve accuracies comparable to that of a much larger model (27B parameters).