Exploring the Usability of Retrieval Augmented Generation and Artificial Intelligence in Government : Toward High-Usability AI Systems in governance

Sammanfattning

Recent advances of Artificial Intelligence (AI) and Retrieval-Augmented Generation (RAG) have facilitated interfacing with textual data in unprecedented ways. However, usability studies identifying factors for high-usability AI systems in governance and the employee perspective are lacking. The purpose of the study was to bridge the gap between artificial intelligence system capabilities and governance needs from the perspective of its end users, contributing to a foundation for designing generative AI and RAG systems with high usability in governance. The study had a two-pronged approach to data collection. A think aloud protocol and semi structured interview were used together. A RAG-model based on over 7000 documents from open government data was created for the study, served as the basis for data collection and represents a contribution in how RAG can be designed for the context of government and usability research methodology. Seven government employees from Region Skåne participated in the study: politicians, committee secretaries and a political secretary. Data was analyzed using thematic analysis. Results indicate several key factors for designing high usability AI systems in governance and exemplify many facets of usability that were valuable to the participants. Five overarching themes pertaining to the study purpose were revealed: areas of application, application handling, process of integration, quality and relevance and dangers of AI. The study provides guidance for developers and government stakeholders when designing AI and RAG systems and demonstrates that usability in the context of governance extends beyond interface design and requires considerations regarding safety and contextual fit.