News
New report and guidelines for indigenous data sovereignty in artificial intelligence developments
UNESCO launched a pioneering report to foster culturally sensitive technological development and information dissemination, ensure the data sovereignty of over 800 Indigenous Peoples in Latin America and the Caribbean, and prevent inappropriate cultural adoption within the accelerated development of artificial intelligence (AI).
The report, "Indigenous People-Centered Artificial Intelligence: Perspectives from Latin America and the Caribbean," was discussed for the first time at the Third Global Forum against Racism and Discrimination, recently held in Sao Paulo, Brazil. It urges the participatory inclusion of local and indigenous communities, an appropriate data operation respecting their autonomy, proposes public policies to integrate indigenous peoples' perspectives in all phases of AI development, and explores best practices, including five from Mexico.
The report by UNESCO offices in Mexico and Montevideo reaches the region where:
- more than 10% of the world's indigenous population resides,
- where nearly 30% of the population lives in extreme poverty, and
- only 40% have basic computer skills.
Mariana Lazos, Head of Projects and Operations at Pit Policy Lab, an independent C Minds spin-out that collaborated on the study, said in the first public discussion of the report at the Third World Forum of UNESCO Against Racism and Discrimination that providing necessary tools to indigenous individuals enables and encourages their participation in AI developments. For example, a pilot AI project in Mexico to streamline learning and fortify the promotion of Tu'un Savi, the third most spoken national indigenous language.
The autonomous digitization of indigenous data is crucial for recording, transmitting, and revitalizing their heritage, especially among younger generations. However, their direct participation is often limited. Therefore, these guidelines stress the determinations of indigenous people and the proper management of their data to respond to the countries' commitments to cultural and linguistic diversity, the right to education sensitive to their reality, and the pursuit of sustainable development.
AI is reshaping our societies, especially Generative AI, so its advances must be ethical and respond to the level of the unique characteristics of the indigenous data and to the rights of indigenous peoples to collect, interpret, and use their voices, images, representations, knowledge, techniques, symbolic and linguistic systems. Case in point, to countering misappropriation of aesthetical and sacred textile patterns through technological developments.
The digitization of indigenous data must guarantee their right to self-determination and to govern their data; to use them under their values and common interests; the free, prior and informed consent for their collaborative participation; and ensure their privacy and intellectual property rights.
The practices taken from Mexico and tackled more extensively in the report are:
Researchers from the Autonomous Metropolitan University created a text-strings and AI proposal to catalogue and identify form, aesthetic and iconographic elements of indigenous garments of the Altos de Chiapas region, capable of identifying variations or plagiarisms.
A research team at the Technological Institute of Coatzacoalcos in Veracruz developed a bot using Natural Language Processing (NLP) to assess the correct pronunciation of words in indigenous languages, collecting phonetic symbols from voice models.
Researchers from the Technological Institute of Oaxaca generated an app for interactive learning of the Tu'un Savi language, using visual computing in an AI-trained model to provide information on pronunciation and writing associated with objects captured by mobile device cameras.
A group from the Institute of Applied Mathematics and Systems (IIMAS) at the National Autonomous University of Mexico (UNAM) developed an automatic translation system between 11 families of indigenous languages in Mexico using machine learning and deep learning, feeding the computer with examples to learn to generate translations. Their proposal includes translation from/to Wixarika, Náhuatl, Yorem nokki, P'urhepecha, Mexicanero (Náhuatl variant) and Spanish.
The Feminist Research Network in Artificial Intelligence (f<A+i>r), with the Indigenous Professional Advisory, Defense, and Translation Center (CEPIADET), and the Human Rights Defender of the People of Oaxaca coordinated a project to assess existing techniques for the automatic translation of 10 indigenous languages in America. A part of that included the design of a conversational agent centred on the experiences and knowledge of interpreters of indigenous languages in Mexico. The proposed system was inadequate because it didn't respond to interpreters' realities according to the evaluation.
