Harnessing Gender-Responsive Artificial Intelligence to Deliver Inclusive, Equitable, and Human-Centered Agricultural Advisories that Bridge Gender Gaps in Digital Climate Services for Smallholder Farmers in Kenya and Beyond

This innovation introduces a gender-responsive artificial intelligence (AI) framework for agricultural advisory services in Kenya that ensures digital climate information and farming advice are inclusive, context-specific, and equitable. Developed by the Alliance of Bioversity International and CIAT under the AICCRA project, in collaboration with i…

This innovation introduces a gender-responsive artificial intelligence (AI) framework for agricultural advisory services in Kenya that ensures digital climate information and farming advice are inclusive, context-specific, and equitable. Developed by the Alliance of Bioversity International and CIAT under the AICCRA project, in collaboration with iShamba, it addresses one of the most pressing challenges in digital agriculture today — that AI and digital advisory systems often overlook the needs of women farmers, reinforcing existing gender and social inequities. Agricultural extension systems in Africa have traditionally struggled to reach women, who make up the majority of the agricultural workforce but face barriers such as limited phone ownership, lower literacy levels, language constraints, and social norms restricting access to information. Data from iShamba, a national SMS and WhatsApp advisory platform, revealed that only about 15% of users were women despite women’s central role in food production. The innovation directly responds to this gap by creating AI-driven tools that actively recognize and mitigate the biases that exclude women from benefitting equally from digital agricultural services. The approach combines machine learning techniques with human-centered design (HCD) and gender analysis to make AI-powered advisories fair, relevant, and trusted. It begins with analyzing gender-disaggregated data from over 9,000 farmer queries submitted through iShamba’s platform between 2020 and 2025. Researchers identified key biases — including language bias (English-dominant responses), referral bias (women’s queries more often escalated), and brevity bias (shorter, less contextual queries from women receiving weaker replies). These findings informed the design of new training datasets, prompt structures, and evaluation metrics that teach AI models to recognize and correct for such inequities. To operationalize these insights, the team refined a human-in-the-loop AI framework where agricultural experts review and improve AI-generated responses using real farmer data. Through iterative feedback and reinforcement learning, the AI model was trained to provide accurate, locally grounded, and gender-responsive advice — for example, by offering affordable alternatives, using simple language, and acknowledging both women’s and men’s farming roles. Testing showed that after refinement, the AI’s overall accuracy and inclusivity improved significantly, with fully satisfactory responses increasing from 62% to 83%, and instances of bias reduced. This innovation not only strengthens the quality and inclusivity of AI-generated farming advice but also demonstrates a replicable process for bias detection and mitigation in other digital platforms. Future phases aim to integrate these gender-responsive algorithms into national agricultural information systems and expand multilingual support (English, Kiswahili, and local languages). Ultimately, this innovation benefits smallholder women farmers by delivering more relevant, trusted, and accessible information, enabling them to make informed decisions about planting, livestock, and climate risk management. It also serves digital advisory providers, extension officers, and policymakers, offering a tested framework for ensuring fairness and equity in the rapidly growing field of AI-enabled agricultural services.