Sociotechnical Misalignments in Hospital AI System Implementation: Qualitative Case Study.

Background: Implementing AI into real-world health care settings is known to be challenging, particularly regarding how well AI is embedded into the existing knowledge, practices, and procedures of a context. Understanding this process is critical for maximizing the successful implementation of AI tools and planning the time, costs, and resources needed for their successful implementation. Objective: This study sought to examine the contextual challenges of implementing an AI chatbot, ChatAI (which provided quick access to clinical and operational information without relying on the intranet), in a large tertiary government hospital by analyzing the impact of sociotechnical factors on its sustained use. Methods: We used an instrumental case study approach, utilizing interviews and meeting minutes. A total of 16 semistructured interviews were conducted with the implementation team and hospital staff who interacted with ChatAI. Interviews were audio-recorded and transcribed. Sociotechnical systems (STS) theory, specifically Davis et al's (2014) framework, was adopted to examine ChatAI's implementation and use. Results: Multiple misalignments among 5 of Davis et al's sociotechnical elements (goals, people, processes, technology, and infrastructure) limited ChatAI's user adoption and sustainability. Although the hospital's innovation center team attempted to address these initial misalignments, contextual changes such as new regulatory mandates, infrastructure changes, and evolving stakeholder practices introduced further misalignments between ChatAI and the hospital-eventually leading to its discontinuation. Conclusions: This study highlights how sociotechnical misalignments can undermine the use and sustainability of large-scale implementation of AI systems. These findings will inform future efforts to implement AI tools in real-world health care settings, increasing awareness of the need to align sociotechnical dimensions of goals, people, processes, technology, and infrastructure. It highlights the particularly challenging aspect of aligning continually evolving infrastructure with regulatory requirements. Future research should focus on how infrastructure and infrastructure changes, as well as external regulatory requirements, influence AI implementation and use.