±¬ÁÏÍø

Today, artificial intelligence (AI) tools are at our fingertips. But ask an AI to generate an image of Indigenous Peoples, and it will likely return a picture of figures in feathered headpieces and beadwork, possibly resembling a travel brochure. This raises important questions. What kind of images has AI learned to associate with Indigenous Peoples? Whose data and knowledge shaped those associations? Who gets to decide how they are used?

Diversity is important in the context of AI. If Indigenous Peoples and languages are poorly represented in the data used to build these systems, then AI may flatten the diversity of thousands of distinct communities into a handful of stereotypical representations. The United Nations reports that there are more than 476 million Indigenous Peoples across the world, making up around 6.2 per cent of the global population. They represent thousands of communities and speak many of the world¡¯s estimated 7,000 languages.

Indigenous Peoples serve many roles, from environmental caretakers and scientists to keepers of history and memories. Every 9 August ¨C the International Day of the World¡¯s Indigenous Peoples ¨C we are invited to recognize the lives and futures of Indigenous communities, and ask whether global policies are accountable to them.

AI can support education, productivity and governance, and even safeguard minority languages and endangered cultures. But it also has environmental costs, and risks extracting data without consent and reducing Indigenous knowledge into information to be processed and monetized.

Many people speak of ¡°responsible AI¡±, ¡°human-centered AI¡± and ¡°inclusive AI¡±, but how responsible, human-centered or inclusive can it be when Indigenous Peoples don¡¯t have a say in AI policies?

This absence matters because AI systems are built on data. In the podcast series ¨C produced by the United Nations University Institute for the Advanced Study of Sustainability (±¬ÁÏÍøU-IAS) ¨C we heard how data is often shaped by those in power, including in an featuring United Nations Chief Information Technology Officer Bernardo Mariano Junior. For Indigenous Peoples, whose lands, knowledge, languages and bodies have long been studied and governed by outsiders, AI risks reproducing old colonial patterns in technological form.

This is why Indigenous data sovereignty (IDS) must be central to how AI is developed and regulated. With the rise of data-driven AI, IDS has become even more urgent as a way to protect the right of Indigenous Peoples to govern how data about their communities and knowledge systems is collected, owned, accessed, interpreted and used. It also reinforces their inherent right to self-determination, as reaffirmed in the .

In practice, IDS can already be seen in emerging efforts around Indigenous language and AI. Led by members of First Nations communities in Canada, seeks to build Indigenous Voice AI around data sovereignty and linguistic self¨Cdetermination. Mozilla¡¯s shows how Formosan language communities and volunteers can help shape speech technologies through community-led data creation.

Reimagining AI governance

IDS can influence how we reimagine AI. At the heart of this approach is the recognition of Indigenous Peoples¡¯ meaningful authority over how AI systems ¨C their data, languages, knowledge or cultural materials ¨C are developed and used. Governments, international organizations, technology companies and research institutions have a responsibility to embed this authority into AI development and regulation.

First, AI policies must treat IDS as a legal and ethical requirement. This means that Indigenous communities must be guaranteed meaningful participation in governing any system that uses Indigenous languages, cultural materials, environmental knowledge or educational data ¨C and from the start, not only after the system has been designed.

Second, data disaggregation must follow Indigenous-led protocols. Institutions involved in using, developing and regulating AI must ask who defines the categories, who stores the data, who interprets the results, who benefits from the analysis and who has the power to say no.

Third, AI policies should be grounded in free, prior and informed consent, as affirmed in ±¬ÁÏÍøDRIP. Consent must be built into the entire life cycle of AI design, deployment, evaluation and revision.

Fourth, Indigenous Peoples must be included as decision-makers in AI governance. They must go beyond ¡°consultees¡± or ¡°validators¡± and become part of the bodies that set standards, audit systems, review risks and decide what kinds of AI are acceptable in Indigenous contexts.

Finally, AI systems must make room for knowledge that cannot be reduced to datasets, prompts, metrics or predictive models. A genuinely inclusive AI future must recognize that some knowledge should be protected, some shared only under community protocols, and some not digitized at all.

Counting (on) Indigenous Peoples

As the world continues to debate the ethical potential of AI, stakeholders involved in AI development and regulation must also learn from Indigenous principles of responsibility, reciprocity, relationality and respect. The question has gone beyond how Indigenous Peoples can be included in AI to how AI can be shaped by Indigenous knowledge and rights.

As diverse as the images we may have of Indigenous Peoples are, their knowledge systems and rights are even more varied and deeply historical. As we just celebrated this year's International Day of the World¡¯s Indigenous Peoples, we must ensure that Indigenous Peoples are counted (on) without reducing them to data points. We must count them as rights-holders who are co-creating the AI ecosystem. We have a responsibility to build a governance framework for truly inclusive AI.

The views expressed in this article are those of the authors and do not necessarily reflect the views of the United Nations University.