This article critically reviews recent initiatives to employ artificial intelligence (AI), particularly large language models (LLMs), for the revitalization of Indigenous languages. Structured by geographical contexts, the analysis includes Irish Gaelic (Europe), Māori (Aotearoa/New Zealand, Oceania), Guaraní (Paraguay/Bolivia, South America), and Inuktitut (Canada, North America). Applying a theoretical framework grounded in data colonialism and Indigenous data sovereignty, the article examines the key achievements in different regional endeavors, as well as investigate how government-led projects and Big Tech collaborations across these diverse contexts navigate (or fail to navigate) issues of data extraction, community consent, cultural representation, and ownership. Through this lens, the article identifies specific ethical pitfalls as well as commendable practices that either reproduce colonial dynamics or empower Indigenous communities. This critique emphasizes regional and contextual nuances, arguing that authentic community agency and rigorous adherence to Indigenous data sovereignty principles are vital to ensuring ethical AI practices and meaningful linguistic revitalization.

As AI changes the way the world gathers information, some critics say that it is perpetuating stereotypes and erasing cultural nuances for Indigenous groups and people of color. Most mainstream models are trained on the work of Western writers—particularly white men—and regularly mimic those values, writing styles, viewpoints, and biases. Some critics say the data grab is a new form of colonialism, where information gathering replaces Imperial-era land seizures while the AI companies—rather than a conquering nation—reap profits from marginalized groups. Data collection from these groups is often done without their consent or any verification that the information is accurate.

Jorge Perez

Little attention is instead given to what values LLMs may reflect and what behaviours they may assume beyond those relating to safety. How well do the values, beliefs and behaviours of these models align with those of their users?

This paper is concerned with issues of trust and bias in generative AI in
general, and chatbots based on large language models in particular (e.g. ChatGPT).
The discussion argues that intercultural communication scholars must do more to
better understand generative AI and more specifically large language models, as
such technologies produce and circulate discourse in an ostensibly impartial way,
reinforcing the widespread assumption that machines are objective resources for
societies to learn about important intercultural issues, such as racism and discrim
ination. Consequently, there is an urgent need to understand how trust and bias
factor into the ways in which such technologies deal with topics and themes central
to intercultural communication. It is also important to scrutinize the ways in which
societies make use of AI and large language models to carry out important social
actions and practices, such as teaching and learning about historical or political
issues

Are AI systems truly representing human values, or merely averaging across them? Our study suggests a concerning reality: Large Language Models (LLMs) fail to represent diverse cultural moral frameworks despite their linguistic capabilities. We expose significant gaps between AI-generated and human moral intuitions by applying the Moral Foundations Questionnaire across 19 cultural contexts. Comparing multiple state-of-the-art LLMs’ origins against human baseline data, we find these models systematically homogenize moral diversity. Surprisingly, increased model size doesn’t consistently improve cultural representation fidelity. Our findings challenge the growing use of LLMs as synthetic populations in social science research and highlight a fundamental limitation in current AI alignment approaches. Without data-driven alignment beyond prompting, these systems cannot capture the nuanced, culturally-specific moral intuitions. Our results call for more grounded alignment objectives and evaluation metrics to ensure AI systems represent diverse human values rather than flattening the moral landscape.

Abstract

Culture fundamentally shapes people’s reasoning, behavior, and communication. As people increasingly use generative artificial intelligence (AI) to expedite and automate personal and professional tasks, cultural values embedded in AI models may bias people’s authentic expression and contribute to the dominance of certain cultures. We conduct a disaggregated evaluation of cultural bias for five widely used large language models (OpenAI’s GPT-4o/4-turbo/4/3.5-turbo/3) by comparing the models’ responses to nationally representative survey data. All models exhibit cultural values resembling English-speaking and Protestant European countries. We test cultural prompting as a control strategy to increase cultural alignment for each country/territory. For later models (GPT-4, 4-turbo, 4o), this improves the cultural alignment of the models’ output for 71–81% of countries and territories. We suggest using cultural prompting and ongoing evaluation to reduce cultural bias in the output of generative AI.

Elba Ramirez
Senior lecturer and programme leader for international studies at the School of Social Sciences and Public Policy (Te Wānanga Aronui o Tāmaki Makau Rau), Auckland University of Technology, New Zealand.