How can multi-agent LLMs improve educational AI inclusivity?
Toward Inclusive Educational AI: Auditing Frontier LLMs through a Multiplexity Lens
January 8, 2025
https://arxiv.org/pdf/2501.03259This paper explores how to make large language models (LLMs) used in education less biased towards certain cultures (primarily Western). It introduces a framework called "Multiplexity" to measure and mitigate this bias, focusing on representing diverse cultural viewpoints. A key point for LLM-based multi-agent systems is the introduction of a multi-agent approach where individual agents, each representing a specific culture, contribute to a final response synthesized by a central "Multiplex Agent". This approach, implemented using the Camel AI framework, proved highly effective in creating more culturally inclusive and balanced LLM outputs for educational purposes.