Integrating AI into humanitarian data systems can increase speed and efficiency, but it also introduces risks for data accuracy, bias, and contextual nuance. Purely automated workflows often fail to capture complex crisis realities or incorporate local knowledge. This session focuses on the importance of human-in-the-loop systems for ethical AI, from responsible design to crowdsourced analysis. We will focus specifically on how KoboToolbox and the GANNET SituationHub use human-centered AI to strengthen the humanitarian data ecosystem.

KoboToolbox: Intelligent Design for Primary Data

We will discuss Kobo’s approach to designing AI-powered tools for the entire primary data lifecycle, from collecting to analyzing data. Our approach begins at survey design, where KoboToolbox’s AI-enabled Formbuilder democratizes expertise, allowing non-specialists to create ethical, context-aware surveys in minutes. The tool acts as an intelligent assistant, suggesting improvements for clarity and bias reduction while embedding safeguards for consent and data protection. Crucially, it augments rather than replaces human judgment: users maintain full control and decision-making authority through a review and validation workflow.

This human-centered philosophy extends to analysis. For qualitative data, AI-powered transcription and translation make it possible to collect rich audio data in multiple languages, and AI-assisted analysis helps identify key themes. This process is always subject to human verification, combining the speed of automation with the irreplaceable nuance of human interpretation. This allows organizations to rapidly surface insights from thousands of community voices that would otherwise remain unheard.

GANNET AI: AI/Human-in-the-Loop Analysis

The GANNET tools are AI-powered platforms for humanitarian analysis that keep humans firmly in the loop throughout automated processes. This session traces the evolution of these tools and explores how participatory design and analysis can transform humanitarian response.

We will share how the foundation began with DEEP, a platform where humanitarian analysts manually annotated and structured crisis data, and the evolution from manual annotation to AI-assisted analysis. We will demonstrate how human expertise can scale, and our journey in true human-in-the-loop design, where machines augment rather than replace human judgment.

Today, the GANNET SituationHub operates through a comprehensive content management system that embeds human verification at critical stages of automated analysis. We will dive into how this framework ensures local knowledge, cultural sensitivities, and ethical judgment shape every analytical output, combining the scalability of digital platforms with the nuanced understanding only humans provide and how in the future we aim to democratize humanitarian analysis through crowdsourced workflows where distributed networks of trained volunteers process diverse datasets building on participatory models that build local analytical capacity, respects data sovereignty, and enables communities to contribute to their own protection.

While significant progress has been made, full integration of primary data collection remains an ongoing challenge. This session explores both achievements and gaps as we work toward truly participatory humanitarian intelligence systems.