Socio-Technical Design of AI-based Systems: Addressing Context, Data, Performance, and Human Factors with Generative AI Support

Abstract

The rapid integration of AI-based systems into organizational contexts has surfaced a unique type of development challenges that traditional software engineering approaches are not equipped to address. This thesis examines how socio-technical design principles, combined with emerging generative AI capabilities, can structure the requirements engineering phase of AI-based system development to better account for the complexity these systems introduce. We synthesize findings across four key areas of literature: AI-specific requirements engineering challenges, socio-technical systems (STS) theory, human-centered AI principles, and generative AI tools for design support. Following a Design Science Research Methodology, we then translate these findings into a concrete research artefact: the Socio-Technical AI Systems Design Framework (STASDF). The framework comprises four integrated components such as a four-dimension challenge model, a GenAI-supported RE process, socio-technical requirements templates, and multi-level design guidelines that are organized around the challenge dimensions of contextual definition, data requirements, performance metrics, and human factors. Our analysis consistently points to the need for joint optimization of both the technical and social subsystems, with particular emphasis on augmenting human judgment rather than replacing it, and on maintaining human oversight throughout system design and operation.