Aligning Language Models with Indigenous Community Cosmovision through Constitutional AI

Key points

• Large language models (LLMs) are predominantly trained on corpora that reflect Western, anglophone worldviews.1
• This asymmetry is not neutral: when a model answers questions about territory, community authority, or the relationship with nature, it does so from assumptions that may be alien to the communities using the system.1
• On this basis, we implemented a Constitutional AI (CAI) pipeline in two phases: supervised fine-tuning (SFT) and direct preference optimization (DPO).1
• The system was deployed in a conversational interface with retrieval-augmented generation (RAG) that allows communities to feed the knowledge base, turning alignment into a participatory cycle.1

Sources (1)

• [1]Aligning Language Models with Indigenous Community Cosmovision through Constitutional AI
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 18, 02:31 PM

  This paper presents a system for aligning language models with the ethical principles of Indigenous communities in Latin America.

  Large language models (LLMs) are predominantly trained on corpora that reflect Western, anglophone worldviews.

Extractive summary: sentences quoted from the sources.