Microsoft has unveiled an inclusive AI initiative in Kenya that seeks to make advanced technology more relevant to communities that are often underserved in global innovation efforts.
The programme, known as Project Gecko, involves collaboration among Microsoft Research Africa based in Nairobi, Microsoft Research India, the Microsoft Research Accelerator team in the United States, Digital Green as an implementation partner, and several academic and philanthropic supporters.
The goal is to design AI that reflects real world needs in emerging markets rather than relying on systems that are primarily created for English speaking regions with strong connectivity and digital literacy.
A key innovation in this project is a model referred to as the Multi Modal Critical Thinking Agent (MMCTAgent). It can interpret a variety of forms including speech, images, and video so that people can interact using natural communication methods rather than advanced technical inputs.
The technology has been published through open source platforms to encourage improvement and adoption by global researchers. One target sector for the rollout is agriculture, an industry that represents the livelihood of millions in Kenya and the wider continent.
In rural communities many people rely on oral learning in local languages and limited internet access, which often makes mainstream AI tools difficult to use.
READ: How Microsoft’s AI for Good Lab Is Helping Save Giraffes from Extinction
Project Gecko builds on an existing farmer support platform operated by Digital Green. That platform uses voice driven assistance and an extensive knowledge library containing thousands of training videos recorded in multiple local languages.
Until now, it has been difficult for users to easily search and extract relevant knowledge from this large library.
With the new system, farmers are able to ask questions in their own language and then receive answers through audio, text, or video, including direct time stamps that take them to the exact instructions they require.
The initiative already supports several widely spoken Kenyan languages including Swahili, Kikuyu, Kalenjin, Dholuo, Maa, and Somali. Thousands of hours of speech data collected from Kenyan users have helped improve accuracy.
More than one hundred farmers have actively tested the system and offered feedback that influenced new features including a clearer question and answer flow and practical guidance that mirrors local farming conditions.
The long term ambition is to publish a detailed guide that can help developers create inclusive AI systems that work for all communities, regardless of geography, language or connectivity.
