Harnessing Generative AI to preserve and promote Indigenous languages

Generative artificial intelligence (AI) presents a powerful opportunity to mitigate the digital isolation of Indigenous communities and enhance the visibility of their peoples and cultures. That is one of the central conclusions of the report “Performance of Artificial Intelligence in the Use of Indigenous Languages in the Americas”, developed by LLYC, a global Marketing and Corporate Affairs consulting firm, in collaboration with BID Lab, the innovation and venture capital arm of the Inter-American Development Bank Group, and Microsoft. The report highlights the current state of play and offers a series of practical recommendations designed to foster a more inclusive and culturally representative approach to AI.

To help bring Indigenous languages into the digital world, the report suggests 21 strategies aimed at increasing the availability of data in these languages and enhancing the technologies that support them. These efforts involve working closely with the communities to encourage digital conversations in Indigenous languages, highlighting the voices of Indigenous influencers, preserving platforms and digital archives of traditions, and creating translation and voice technologies. Taken together, these strategies could help train AI models in ways that significantly improve their performance in these languages.

AI has the potential to become a powerful tool for preserving, sharing, and revitalizing cultural and linguistic traditions, and for narrowing gaps created by illiteracy or monolingualism in remote areas. For Indigenous communities to truly benefit from the development and job opportunities that AI can provide, it’s crucial to enhance its effectiveness in their native languages. Without that progress, digital and social divides could deepen.

Currently, the most well-known generative AI models exhibit uneven performance when interacting in Indigenous languages. In 54% of the cases analyzed, questions received partially accurate responses. However, those responses tended to be up to four times shorter than equivalent ones in Spanish and were rated significantly lower in expression (2.4 out of 10) and comprehension (2.3 out of 10). The report also found considerable cultural bias in generative AI systems, which often default to Western references, even when prompted in Indigenous languages.

The report highlights the importance of collaboration among governments, Big Tech initiatives, NGOs, and consumer brands to enhance AI performance in Indigenous languages. It also confirms a strong correlation (91%) between the volume of digital content available in a language and the quality of AI responses.

“For artificial intelligence to genuinely include everyone around the world, it needs to grasp and adjust to the different linguistic and cultural contexts.” said Adolfo Corujo, Partner & Marketing Solutions CEO at LLYC. “This study is an important first step in ensuring that Indigenous languages are better represented in future technologies.”

“Clients around the world are demanding greater linguistic and cultural relevance in the AI-powered products and services we provide,” said Daniel Korn, director of AI Policy and Innovation for Microsoft Americas. “At Microsoft, our goal is to put people first. We are committed to addressing these disparities by working alongside governments, academia, civil society, and multilateral organizations like the IDB.”

“We promote the development of AI solutions that reflect the real-world contexts of Latin America and the Caribbean through our fAIr LAC program,” said César Buenadicha, acting chief of the Ecosystem Building and Acceleration Division at IDB Lab. “This study helps us identify challenges and opportunities to build more accessible and relevant technologies for our communities.”

Read the full report here.