Artificial intelligence: language, an issue of equality
The advent of artificial intelligence (AI) could help reverse trends in unequal development around the world. The latest Goalkeepers report from the Gates Foundation, published on Tuesday September 15, notably proposes adapting AI to the different languages of the world in order to considerably reduce these disparities.
Released on Tuesday, September 15, the Gates Foundation’s Goalkeepers Report 2026, titled “Let’s Change the Game: AI and Equity, The Urgency of a Choice,” focuses on the inclusiveness of artificial intelligence. The global context requires us to put AI at the service of all in order to significantly reduce inequalities, but also to facilitate access to the poorest and most vulnerable populations.
It is also about reducing the gap between rich and poor countries and contributing to achieving the United Nations Sustainable Development Goals (SDGs). As it is currently conceived, the development of artificial intelligence benefits populations in Northern countries more than those in Southern countries, which could accentuate development inequalities.
However, to achieve this, the Goalkeepers report emphasizes three priorities. The first is to adapt AI to the world’s languages and ensure that artificial intelligence tools work in all spoken languages. “Most of the world is invisible to today’s AI models. These AI systems were trained primarily on what exists on the Internet; and the Internet is a reflection of a profoundly unequal world. More than 90% of the data used to train the first Llm (Large language models) came from English-speaking sources,” reports the document.
Today, the report notes, the communities that stand to gain the most from AI are largely absent from the knowledge base on which these tools are built. As a result, they are not only poorly served, but remain largely isolated. “Every month, AI becomes more precise and more efficient for the billion people, mainly English-speaking, who already have access to it, while it remains unchanged for the other seven billion inhabitants of the planet. In English, the best AI voice recognition systems make errors in less than 6% of cases. In Yoruba, this same system is wrong more than 60% of the time. It’s not just about everyday vocabulary,” the report underlines.
Rely on local data
As proof, Goalkeepers insists, if AI is to help improve health on a continent with more than 2,000 languages, it must understand what people actually say and how they express themselves, including their dialects, accents and slang.
However, this “requires investments in local language datasets and assessments tailored to language use in the regions these tools are intended to serve.”
Then, the report recalls the need for AI tools to rely on local data and adapt to the realities of different territories. Because “developing AI that works in all contexts requires collecting the right local data (medical records, agricultural yields, weather trends, student learning styles) from the populations and in the places that these tools are intended to serve.”
In doing so, Goalkeepers emphasizes, “countries must be able to decide how best to manage and protect this data, including how it is stored, under what conditions and with whom it is shared. Finally, the report urges investing in human skills and improving access to tools in order to put AI at the service of all.
Ibrahima BA
