How a Māori Radio Station Became an AI Pioneer While Protecting Ancient Knowledge
<aside> <img src="notion://custom_emoji/3b7b79b0-95af-4500-931c-e5c63e5df242/133c6f79-9a33-809e-b97a-007a409367c7" alt="notion://custom_emoji/3b7b79b0-95af-4500-931c-e5c63e5df242/133c6f79-9a33-809e-b97a-007a409367c7" width="40px" />
Peter Lucas shares the journey of preserving Māori language and culture through technology and community engagement.
</aside>
The next wave of AI innovation might not come from Silicon Valley's attempt to replicate human consciousness. It might come from a small radio station in New Zealand that dared to imagine a different future - one where artificial intelligence helps preserve ancient wisdom rather than commodify it.
Inside Vancouver's NeurIPS 2024 conference, between dense talks of transformer architectures and scaling laws, Peter Lucas Jones opens with "He kupu tuku iho Mo te nei reanga." The traditional Māori greeting silences the room.
Standing at the podium isn't another Silicon Valley exec preaching about AI safety - it's a leader from Te Hiku Media, a tribal radio station that's rewritten the rules of ethical AI development while running a GPU cluster in one of the most remote parts of New Zealand.
"In the late 80s our grandmothers and aunts had a dream that our language would not be siloed into an academic experience," Jones explains. That dream materialized as a community radio station in 1990, but the real transformation came in 2014. Using Raspberry Pis, mobile phones, and sheer ingenuity, they built a digital platform streaming everything from current affairs to dance competitions. Their target audience? A displaced community where 80% live outside their traditional territory, disconnected from their language and culture.
NeurIPS Poster Fit for our purpose, not yours: Benchmark for a low-resource, Indigenous language
Their archives held thirty years of irreplaceable knowledge: elders sharing detailed information about every maunga (mountain), every awa (river), traditional medicine, and celestial navigation. "Why do you pick the bark from a tree that faces the sun, and not the bark that does not face the sun?" Jones asks, revealing the depth of indigenous knowledge in their recordings. They knew big pharma would salivate over their traditional medicine data, especially regarding marine resources. But Te Hiku Media chose sovereignty over quick profit.
By 2018, they'd developed something unprecedented: the first indigenous language automatic speech recognition system. NVIDIA provided GPU support, but the game-changer was location - their cluster runs in Kaitaia, processing their data on their ancestral lands. Their Papareo API includes not just speech recognition but a pronunciation model designed specifically for language reclamation.