WORDS BY MEGHIE RODRIGUES
photograph by christian cassiel
Last March when Marina Gatto went to Rio de Janeiro for the first time, she was mystified by Parque Lage—a park at the base of the hill topped by Christ the Redeemer. Born and raised in the Brazilian Amazon, Gatto was in Rio for a game development competition called GameJamPlus.
The park was beautiful, but Gatto, a young Indigenous Mura leader from Amazonas, couldn’t find any information about the European-style architecture and gardening that adorned it. As many in the year 2024 would, she turned to an artificially intelligent chatbot, only it wasn’t ChatGPT, Gemini, or any other large language model produced by American Big Tech. It was Tainá—an AI assistant developed in her native Inhaã-bé village in Manaus.
“The park was pretty crowded that day, and there were no guides available. So I asked Tainá if she knew any of the plants I was seeing—and she recognized them, even if I was in a different biome,” Gatto said. She uploaded pictures of an aqueduct, and Tainá provided the backstory for it, too.
Gatto was surprised that Tainá knew about the park because the chatbot was initially developed to preserve information about Mura, Sateré-Mawé, and Tikuna culture from her homeland of Inhambé in Manaus. It was developed to store traditional knowledge digitally and in perpetuity. The cultural memory vault works like a Telegram chatbot. Indigenous community members send voice recordings, images, and videos, which Tainá uses as training data—though now, users can input data from elsewhere, which is likely how Tainá knew of Parque Lage. Users can then ask the chatbot questions in English, Spanish, Portuguese, Bahasa, and Swahili. In addition to, apparently, public parks in Rio, its current knowledge base includes traditional wisdom of basketry techniques, botanical and animal knowledge, and oral stories—some of which only Indigenous elders remember.
“It is great to have technology that is tailored for us, Indigenous people, and know we can rely on it whenever we travel to distant places,” Gatto said.
Community ownership is central to Tainá. The data are stored locally—not on the cloud—and the community decides which parts get shared with the world. It’s a far cry from the status quo of “Big Tech lords deciding what to do with our data,” explained Gatto. Gatto is also a member of GainForest, a Zurich-based science nonprofit that started Tainá and has worked with Inhaã-bé members for a year.
GainForest is developing data collection technology for biodiversity study and conservation with the help of Indigenous communities in the Amazon. In addition to Tainá, that proposal included drones and a mobile environmental DNA lab that could fit in a backpack. Their team won a $250,000 prize from a competition called XPrize Rainforest in 2024—and Indigenous scientists representing their communities, like Gatto, have the final say on where those winnings will be used. Now, GainForest works with 30 Indigenous groups across South America, Africa, and Asia.
“It is great to have technology that is tailored for us, Indigenous people, and know we can rely on it whenever we travel to distant places.”
“Our main idea for the XPrize was to engage these communities in systematizing AI technology to their own needs, in their own terms. They convey information in ways [people in the West] don’t, but technology should serve them, too,” said Gabriel Nunes, GainForest’s science lead.
GainForest approached the Inhaã-bé community to carry out technology education workshops in early 2024. “At first the idea of tinkering with AI was not well received because everyone thought this technology would be extractive of our knowledge and invasive to our culture,” Gatto said. But after the workshops, the community noticed AI tools could be useful to safeguard their wisdom and decided to give them a try.
Gatto was one such convert. “When we got involved in the project, I had some previous [technology] knowledge, as I spent some time away from the community to study in the city. But I wasn’t interested, because I believed technology would put an end to our village and ways of life. But then I got involved with GainForest, dug deeper into that knowledge, and realized technology could be more useful than I had realized at first,” she said. Now a member of GainForest, Gatto helps her community vet information that gets inputted into Tainá and takes part in deciding what to do with the data.
Tainá has energized Inhaã-bé’s young people, said Samuel Munduruku, a young Indigenous leader in Parque das Tribos—Brazil’s largest urban Indigenous settlement, also located in Manaus. “When we first learned of the project, everyone got excited and wanted to get involved—children in particular. The project spurred a passion for technology in them,” he said.
“Teenagers now think of a different future, and they know they want to use technology to help them get there. They want to develop new things by themselves,” Gatto added.
Even elders have gotten involved. That’s critical because in many cases, they’re the last remaining recordkeepers of oral traditions and knowledge. “The elderly are connecting more with our younger generations by telling their stories to Tainá,” Gatto said. “It was a trust-building process.”
To Nunes, who grew up in a traditional riverside community in Amazonas himself, this type of technological training is a game-changer. “We used to have few options for the future. We either traded our community for the city or stayed, facing harsh conditions. As urban settings are highly appealing, most of the young usually left—and as they left, stories and ways of life would be lost,” he said.
Younger generations want to use technology to improve their quality of life, according to Gatto and Nunes. That goes beyond safeguarding their history—and with the help of GainForest, many of these efforts are already underway.
For example, the riverine community of Santa Helena do Inglês, which sits on the Rio Negro about 60 kilometers away from Manaus, used drone images to train an AI model that helped them determine whether nuts were ripe for harvest. The model also helped the community identify the exact nut trees to pick. “By training that model, we shortened the harvest season from over a week to a few days,” Nunes said.
“The elderly are connecting more with our younger generations by telling their stories to Tainá. It was a trust-building process.”
Drone technology is also helping the Inhaã-bé community identify plants and animals they weren’t fully aware of. And, it’s helped at least one community brace itself for drought. Over the last two years, the Amazon experienced the worst drought since records began in 1950. The community saw from the skies that a nearby tributary of the Negro River was drying faster than usual. “Inhaã-bé members knew about the drought three weeks before the media started reporting on it last year,” said David Dao, lead scientist at GainForest.
The community began reinforcing their water and food supplies weeks before the worst of the drought came. It also organized a fundraising campaign for emergencies. “We warned other Indigenous villages and mobilized them to take women and children to the city and other places so they wouldn’t suffer so much with water shortages. We thought of strategies to protect ourselves against malaria and dengue,” Gatto said.
Not every community has bought into the idea—especially when it comes to AI. But Dao is confident that enthusiasm will grow as technology advances. As these models get smarter, they require less programming to build and apply. “You can explain [their parameters] with better language,” Dao said.
Equally important to buy-in is that AI must embody community values, Dao said. His team is now experimenting with constitutional AI, which involves writing a “constitution” of principles and values that the AI is trained on. GainForest hopes to help create local Tainás for different communities aligned to unique constitutions. “What do you care about? How should this AI behave? What should it prioritize? What should it not do? These questions are leading the second phase of Tainá’s development,” Dao said.
The Inhaã-bé decided some of Tainá’s stored knowledge would be shared with the world. GainForest built a sister language model called Poli, which includes information they want to share publicly along with several United Nations documents involved in climate negotiations. By translating intricate UN data to native tongues and thinking of local examples to illustrate the impact of certain policies, “Poli helps young Global South negotiators improve their arguments in climate negotiations,” Dao said.
The Youth Negotiators Academy tested the technology at COP28 and COP29. “Global South negotiators were more confident by using Poli—this is a smart way to level the field,” Dao said.
The future of these developments depends on further training and better infrastructure in Indigenous communities—reliable internet connection, powerful laptops, and programming skills are essential. Nunes said his team has planned a few AI workshops for 2025.
If all of that pans out, Nunes believes the visibility and opportunity that technology provides can change things for the better. “The most important of it all is we are being able to systematize the nature-protective work these communities do into visible data. These technologies shine light on work that is rarely seen,” he said. “Talent might be evenly distributed, but opportunities are not.”
Join our membership community. Support our work, receive a complimentary subscription to Atmos Magazine, and more.
Indigenous Groups Are Safeguarding Culture with Their Own ChatGPT
