Zeng Li
The adoption of Artificial Intelligence (AI) in high-stakes domains like healthcare faces complex organisational challenges that hinder its potential. AI systems are defined by dynamism, autonomy, and algorithmic inscrutability, which intensify socio-technical complexity in their adoption. A key limitation in understanding these challenges lies within the existing Information Systems (IS) literature, which often treats 'adoption barriers' as a uniform category. This lack of conceptual differentiation obscures that different types of challenges and contradictions, from simple technical problems to organisational paradoxes, have different natures and demand different strategies. Consequently, current research and practice lack a comprehensive framework to explore and respond to the challenges and contradictions that impede AI adoption. To address this gap, this thesis develops a framework to systematically connect the challenges, tensions, and paradoxes that arise during medical AI adoption. This thesis then investigates the diverse organisational responses tailored to these contradictions and discovers their impact on tensions and paradoxes in return. The primary qualitative data for this study were collected from the departments of Orthopaedics, Gastroenterology, and Radiology across three public hospitals in China, comprising 48 semi-structured interviews with diverse stakeholders, including doctors, technology vendors, and hospital managers. Firstly, the thesis provides an integrated framework for tension genesis in the medical AI adoption, theorising an escalated pathway from challenges to tensions to enduring paradoxes. This framework unifies these three contradictions that have been inadequately studied and clearly defined. Furthermore, by examining the characteristics of AI, the thesis introduces the concept of technological tensions and extends the existing typology of paradox theory in the context of AI adoption. Moreover, this framework integrates temporal and contextual impacts into tension analysis. It demonstrates that the manifestation of tensions is contingent upon the research, development and use stages in AI adoption, and the departmental requirements and culture in the hospital. Secondly, the thesis specifies six response strategies to tensions and three response strategies to paradoxes. The thesis analyses that the mitigating, neutralising, or exacerbating impacts of these strategies on tensions are contingent upon the actor's power and the nature of the contradiction. This analysis clarifies the different effectiveness of response from multi-level in the context of AI adoption. This thesis makes two main theoretical contributions. First, this thesis extends paradox theory by introducing a process-oriented framework that maps the escalation of contradictions, from initial challenges to tensions and persistent paradoxes. A key element of this framework is the concept of technological tensions, which integrates technological factors of AI systems into paradox theory, which centred on social and organisational contradictions. Second, the thesis challenges a convention in technology adoption literature: the treatment of adoption barriers as a uniform category. This thesis provides a different perspective that argues for distinguishing between natures of contradictions, as their distinct natures demand different management strategies. In summary, this study offers a fresh perspective, contending that researchers should distinguish between the essential natures of different contradictions, given that varied contradiction profiles require distinct management strategies. Overall, this thesis offers a unified, systemic framework for scholars to analyse technology adoption in healthcare organisations, and provides actionable insights to organisation management, vendors, and policymakers for managing the persistent contradictions in the responsible adoption of medical AI systems.