Microsoft has launched a new research initiative aimed at making artificial intelligence (AI) systems more inclusive for communities currently underrepresented online; and it’s starting in Kenya.
Called Project Gecko, Microsoft says that the initiative is a global effort to build more inclusive AI systems that cater to the “world's majority”, particularly communities whose languages and cultural contexts are currently underrepresented in AI models.
The initial deployment – a collaboration with digital Green – centres on agriculture in Kenya, a critical economic sector where millions of smallholder farmers rely on hyperlocal, community-based knowledge. Project Gecko enhances tools like Digital Green’s Farmer.Chat, a speech-first assistant. With the new system, a Kenyan farmer, for example, can ask a complex question verbally in their native language, such as Kikuyu, and receive an actionable, response via text, audio, and a video clip that auto-jumps to the precise instruction.
If the initiative – led by a collaborative team from Microsoft Research Africa (Nairobi), Microsoft Research India, and the Microsoft Research Accelerator in the U.S. – is to have an impact, there needs to be “a fundamental rethinking of how AI is localised, evaluated, and deployed”, notes Microsoft.
Instead of adapting existing, English-centric large models, Project Gecko focuses on creating cost-effective, adaptable AI that leverages local languages, oral knowledge, and multimodal communication, such as speech and video.
“Building AI systems from the ground up shaped by the knowledge, languages, and modalities of the global majority yields more innovative, useful solutions for a great number of people,” said Ashley Llorens, corporate VP and MD, Microsoft Research Accelerator.
Also central to this project is the MultiModal Critical Thinking Agent (MMCTAgent), an AI system developed to analyse and reason across speech, images, and video to generate locally grounded answers. MMCTAgent uses small language models to ensure it can run efficiently on the low-cost, low-bandwidth devices common in rural areas.
To overcome the deficit of training data for African languages, Microsoft said the Project Gecko team built new automatic speech recognition and text-to-speech tools from scratch. The initiative currently supports widely spoken Kenyan languages, including Swahili, Kikuyu, Kalenjin, Dholuo, Maa, and Somali, drawing on a dataset of over 3,000 hours of crowdsourced Kenyan speech.
“If we want to build AI for everyone everywhere, we need to develop new methods of human-centred AI. This involves forging new and deeper connections among disciplines such as machine learning, linguistics, and the social sciences, as well as the communities the AI is to serve,” said Jacki O’Neill, lab director Microsoft Research Africa.
According to Microsoft, early field studies in Kenya have demonstrated significant improvements in response quality, usability, and user trust compared to generic, state-of-the-art AI systems.
This success highlights a key finding: for the global majority, local relevance and cultural grounding are more powerful drivers of AI impact than sheer model size.
Looking ahead, Microsoft plans to expand Project Gecko's application into other sectors, such as healthcare, education, and retail.
Share
