Towards a Framework for Interdisciplinary Studies in Explainable Artificial Intelligence

In this interdisciplinary paper, we present the SocioTechXAI Integration Framework (STXIF), a novel approach that aims to seamlessly integrate technical advances with social science methodologies for a nuanced understanding of Explainable Artificial Intelligence (XAI) tailored to specific use cases. We begin with an overview of related work exploring XAI in both the social sciences and computer science. The focus is on the presentation of the STXIF, which includes the XAI Compass and socioscientific analysis of stakeholder perspectives. The perspectives of these stakeholders, classified by the XAI Compass as Model Breakers, Model Builders, and Model Consumers, are investigated using qualitative content analysis. By examining the EU AI Act as a Model Breaker and conducting scenario-based focus group discussions with Model Consumers (medical professionals) and Model Builders (developers) in a real medical diagnostic use case, we demonstrate the specific insights gained through the STXIF application and its adaptability in real scenarios. The discussion section addresses the complex relationships between different XAI goals and their implications for a flexible and adaptive development approach. The practical implications extend to the concrete development and implementation of XAI in real-world applications, in line with the thematic focus of the Human Computer Interaction International Conference, which emphasizes human-centered design and usability in interactive systems. Emphasizing nuanced interactions with XAI and its practical applications establishes a foundational framework for future interdisciplinary research and application in the evolving landscape of human-computer interaction (HCI).