Artificial intelligence is changing how people search for information, learn new skills, access services, and communicate with one another. But not every language is being supported by the AI technologies people are increasingly relying on. Many risk being left behind entirely due to a lack of language data.

The Minimum Viable Dataset Working Group was convened as part of CODI's first research initiative focused on advancing multilingual AI. This diverse group of experts in artificial intelligence, linguistics, digital access, Indigenous data governance, and cultural preservation worked for months to answer one critical question: What does a language need to participate meaningfully in the age of AI?

The resulting framework introduces recommendations, guidelines, and practical tools to determine what makes a dataset culturally relevant and ethically sourced for use in generative AI models. Rather than dictating what communities should build, the report offers a flexible framework they can use to define their own goals.