The Pulse
60 Organizations Set Five-Year Goal for AI in 3.4 Billion People’s Languages
Sixty organizations have committed to a five-year effort to help an estimated 3.4 billion people use artificial intelligence in their own language and voice.
AI.info Team ·
“It’s important that AI tools work for everyone in the languages they actually speak,” said Bill Gates, chair of the Gates Foundation. “Right now, even some of the better-supported African languages have significant gaps. When AI tools can't understand somebody’s natural language, the cost of a mistake can be serious. I'm optimistic that within a short period of time, and with the right leadership, we can solve this problem.”
Bill Gates, chair of the Gates Foundation
Sixty organizations from technology, government, research, philanthropy, and civil society have committed to a five-year effort aimed at helping an estimated 3.4 billion people use artificial intelligence in languages and voices that current systems often handle poorly.
The coalition, announced by the Gates Foundation in New York on September 21, 2026, includes Amazon, Anthropic, Google, Microsoft, Mistral, NVIDIA, OpenAI Foundation, UNICEF, the World Bank Group, and organizations focused on African, South Asian, and other underrepresented languages. The initiative does not announce a pooled budget or a dollar commitment. Instead, signatories say they will contribute according to their expertise and resources.
Bill Gates Warns of Serious Costs From Language Errors
The coalition’s case rests on a gap between the roughly 7,000 languages spoken worldwide and the much smaller set supported well by leading AI models. Languages described as low-resource often have less digital data, fewer evaluation tools, and limited representation in the datasets used to train and test models.
That gap affects more than translation quality. Dialects, slang, idioms, and cultural context can change the meaning of a request or response. In health, education, agriculture, financial services, and public administration, a system that misunderstands a speaker may provide advice that is less useful or unsafe.
Four Workstreams Form the Coalition’s Starting Point
The organizations plan to concentrate on four areas. The first is an open language layer: shared data infrastructure that developers can use under open licenses. The second is a set of assessments and benchmarks intended to show whether systems improve in real-world language tasks rather than simply adding languages to a product list.
The third area involves turning language data into models and applications that developers with fewer resources can use. The fourth is deployment, with the coalition calling for privacy protections, consent, and data sovereignty as organizations collect speech and text from local communities.
The detailed structure and governance of the coalition will be developed collaboratively during the next year.
Voice Matters Where Text Interfaces Fall Short
The announcement places particular emphasis on speech. In communities where typing is difficult, literacy is uneven, or internet access is limited, voice may offer a more practical way to interact with AI. The coalition’s goals therefore cover both written language and spoken interaction.
Several signatories are already working on those components. ElevenLabs says its models can speak and listen in more than 90 languages, while Sunbird AI says it is building open text and speech technology for 67 low-resource African languages. DigitalGreen says its FarmerChat service has reached 2.2 million farmers across six countries and supports 16 languages, with nearly two-thirds of queries arriving through voice or images rather than typed text.
Those figures come from statements supplied by the organizations themselves and illustrate the varied starting points inside the coalition. Some members build models, some collect or curate datasets, and others connect language technology to health, agriculture, education, or government services.
Local Ownership Is Central to the Plan
The commitment repeatedly assigns responsibility to the communities whose languages are being added to AI systems. Its accompanying statement calls for investment in local research, capacity building, and ownership so that language resources do not become extractive inputs controlled entirely by outside companies.
That concern appears in comments from Masakhane, which works on African languages, and Karya, which says speakers of underrepresented languages are the people who can supply much of the missing data. The coalition also includes AI4Bharat, Sarvam, BharatGen, Lahore University of Management Sciences, the Ethiopian Artificial Intelligence Institute, and the Ministry of Telecommunications and Digital Affairs of Senegal.
The signatory list spans major technology companies and smaller regional groups, but the announcement does not set a specific number of languages to be supported or define a common technical threshold for success. Its stated target measures people served, while the benchmarks and reporting systems needed to assess that target are still to be built.
A Public-Interest Goal Without a Shared Budget
The five-year pledge gives the organizations a common objective, but leaves implementation details open. The Gates Foundation says partners will develop the coalition’s governance and workstreams over the coming year, meaning decisions about funding, data licensing, evaluation, and accountability remain ahead.
That structure reflects the problem the coalition is trying to address: no single lab controls the necessary language data, community relationships, models, and delivery channels. The practical test will be whether open resources reach working tools that people can use in their own communities, rather than stopping at datasets, pilot projects, or declarations.
For now, the commitment establishes its measure of ambition: within five years, 3.4 billion people who speak languages underrepresented in current AI models should be able to use AI in their own language and voice.